btr0cur.c 92.7 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14
/******************************************************
The index tree cursor

All changes that row operations make to a B-tree or the records
there must go through this module! Undo log records are written here
of every modify or insert of a clustered index record.

			NOTE!!!
To make sure we do not run out of disk space during a pessimistic
insert or update, we have to reserve 2 x the height of the index tree
many pages in the tablespace before we start the operation, because
if leaf splitting has been started, it is difficult to undo, except
by crashing the database and doing a roll-forward.

15
(c) 1994-2001 Innobase Oy
16 17 18 19 20 21 22 23 24 25 26 27

Created 10/16/1994 Heikki Tuuri
*******************************************************/

#include "btr0cur.h"

#ifdef UNIV_NONINL
#include "btr0cur.ic"
#endif

#include "page0page.h"
#include "rem0rec.h"
28
#include "rem0cmp.h"
29 30 31 32 33 34 35 36 37 38
#include "btr0btr.h"
#include "btr0sea.h"
#include "row0upd.h"
#include "trx0rec.h"
#include "que0que.h"
#include "row0row.h"
#include "srv0srv.h"
#include "ibuf0ibuf.h"
#include "lock0lock.h"

unknown's avatar
unknown committed
39 40 41 42
/* If the following is set to TRUE, this module prints a lot of
trace information of individual record operations */
ibool	btr_cur_print_record_ops = FALSE;

43 44 45
ulint	btr_cur_rnd	= 0;

ulint	btr_cur_n_non_sea	= 0;
unknown's avatar
unknown committed
46
ulint	btr_cur_n_sea		= 0;
unknown's avatar
unknown committed
47 48
ulint	btr_cur_n_non_sea_old	= 0;
ulint	btr_cur_n_sea_old	= 0;
49 50 51 52 53 54 55 56 57 58

/* In the optimistic insert, if the insert does not fit, but this much space
can be released by page reorganize, then it is reorganized */

#define BTR_CUR_PAGE_REORGANIZE_LIMIT	(UNIV_PAGE_SIZE / 32)

/* When estimating number of different kay values in an index sample
this many index pages */
#define BTR_KEY_VAL_ESTIMATE_N_PAGES	8

59 60 61 62 63 64 65 66 67
/* The structure of a BLOB part header */
/*--------------------------------------*/
#define BTR_BLOB_HDR_PART_LEN		0	/* BLOB part len on this
						page */
#define BTR_BLOB_HDR_NEXT_PAGE_NO	4	/* next BLOB part page no,
						FIL_NULL if none */
/*--------------------------------------*/
#define BTR_BLOB_HDR_SIZE		8

68 69 70 71 72 73 74 75 76 77
/***********************************************************************
Marks all extern fields in a record as owned by the record. This function
should be called if the delete mark of a record is removed: a not delete
marked record always owns all its extern fields. */
static
void
btr_cur_unmark_extern_fields(
/*=========================*/
	rec_t*	rec,	/* in: record in a clustered index */
	mtr_t*	mtr);	/* in: mtr */
78 79 80 81 82 83 84 85 86 87 88
/***********************************************************************
Adds path information to the cursor for the current page, for which
the binary search has been performed. */
static
void
btr_cur_add_path_info(
/*==================*/
	btr_cur_t*	cursor,		/* in: cursor positioned on a page */
	ulint		height,		/* in: height of the page in tree;
					0 means leaf node */
	ulint		root_height);	/* in: root node height in tree */
89 90 91 92 93 94 95 96 97 98 99
/***************************************************************
Frees the externally stored fields for a record, if the field is mentioned
in the update vector. */
static
void
btr_rec_free_updated_extern_fields(
/*===============================*/
	dict_index_t*	index,	/* in: index of rec; the index tree MUST be
				X-latched */
	rec_t*		rec,	/* in: record */
	upd_t*		update,	/* in: update vector */
100 101 102
	ibool		do_not_free_inherited,/* in: TRUE if called in a
				rollback and we do not want to free
				inherited fields */
103 104
	mtr_t*		mtr);	/* in: mini-transaction handle which contains
				an X-latch to record page and to the tree */
unknown's avatar
unknown committed
105 106 107 108 109 110 111 112 113
/***************************************************************
Gets the externally stored size of a record, in units of a database page. */
static
ulint
btr_rec_get_externally_stored_len(
/*==============================*/
			/* out: externally stored part, in units of a
			database page */
	rec_t*	rec);	/* in: record */
114 115 116 117 118 119 120 121 122

/*==================== B-TREE SEARCH =========================*/
	
/************************************************************************
Latches the leaf page or pages requested. */
static
void
btr_cur_latch_leaves(
/*=================*/
unknown's avatar
unknown committed
123
	dict_tree_t*	tree __attribute__((unused)),	/* in: index tree */
124 125 126 127 128 129 130 131 132 133
	page_t*		page,		/* in: leaf page where the search
					converged */
	ulint		space,		/* in: space id */
	ulint		page_no,	/* in: page number of the leaf */
	ulint		latch_mode,	/* in: BTR_SEARCH_LEAF, ... */
	btr_cur_t*	cursor, 	/* in: cursor */
	mtr_t*		mtr)		/* in: mtr */
{
	ulint	left_page_no;
	ulint	right_page_no;
unknown's avatar
unknown committed
134 135
	page_t*	get_page;
	
136 137 138 139
	ut_ad(tree && page && mtr);

	if (latch_mode == BTR_SEARCH_LEAF) {
	
unknown's avatar
unknown committed
140 141
		get_page = btr_page_get(space, page_no, RW_S_LATCH, mtr);
		buf_block_align(get_page)->check_index_page_at_flush = TRUE;
142 143 144

	} else if (latch_mode == BTR_MODIFY_LEAF) {

unknown's avatar
unknown committed
145 146
		get_page = btr_page_get(space, page_no, RW_X_LATCH, mtr);
		buf_block_align(get_page)->check_index_page_at_flush = TRUE;
147 148 149 150 151 152 153

	} else if (latch_mode == BTR_MODIFY_TREE) {

		/* x-latch also brothers from left to right */
		left_page_no = btr_page_get_prev(page, mtr);

		if (left_page_no != FIL_NULL) {
unknown's avatar
unknown committed
154 155 156 157
			get_page = btr_page_get(space, left_page_no,
							RW_X_LATCH, mtr);
			buf_block_align(get_page)->check_index_page_at_flush =
									TRUE;
158 159
		}
				
unknown's avatar
unknown committed
160 161
		get_page = btr_page_get(space, page_no, RW_X_LATCH, mtr);
		buf_block_align(get_page)->check_index_page_at_flush = TRUE;
162 163 164 165

		right_page_no = btr_page_get_next(page, mtr);

		if (right_page_no != FIL_NULL) {
unknown's avatar
unknown committed
166 167 168 169
			get_page = btr_page_get(space, right_page_no,
							RW_X_LATCH, mtr);
			buf_block_align(get_page)->check_index_page_at_flush =
									TRUE;
170 171 172 173 174 175 176 177 178 179
		}

	} else if (latch_mode == BTR_SEARCH_PREV) {

		/* s-latch also left brother */
		left_page_no = btr_page_get_prev(page, mtr);

		if (left_page_no != FIL_NULL) {
			cursor->left_page = btr_page_get(space, left_page_no,
							RW_S_LATCH, mtr);
unknown's avatar
unknown committed
180 181
			buf_block_align(
			cursor->left_page)->check_index_page_at_flush = TRUE;
182 183
		}

unknown's avatar
unknown committed
184 185
		get_page = btr_page_get(space, page_no, RW_S_LATCH, mtr);
		buf_block_align(get_page)->check_index_page_at_flush = TRUE;
186 187 188 189 190 191 192 193 194

	} else if (latch_mode == BTR_MODIFY_PREV) {

		/* x-latch also left brother */
		left_page_no = btr_page_get_prev(page, mtr);

		if (left_page_no != FIL_NULL) {
			cursor->left_page = btr_page_get(space, left_page_no,
							RW_X_LATCH, mtr);
unknown's avatar
unknown committed
195 196
			buf_block_align(
			cursor->left_page)->check_index_page_at_flush = TRUE;
197 198
		}

unknown's avatar
unknown committed
199 200
		get_page = btr_page_get(space, page_no, RW_X_LATCH, mtr);
		buf_block_align(get_page)->check_index_page_at_flush = TRUE;
201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222
	} else {
		ut_error;
	}
}

/************************************************************************
Searches an index tree and positions a tree cursor on a given level.
NOTE: n_fields_cmp in tuple must be set so that it cannot be compared
to node pointer page number fields on the upper levels of the tree!
Note that if mode is PAGE_CUR_LE, which is used in inserts, then
cursor->up_match and cursor->low_match both will have sensible values.
If mode is PAGE_CUR_GE, then up_match will a have a sensible value. */

void
btr_cur_search_to_nth_level(
/*========================*/
	dict_index_t*	index,	/* in: index */
	ulint		level,	/* in: the tree level of search */
	dtuple_t*	tuple,	/* in: data tuple; NOTE: n_fields_cmp in
				tuple must be set so that it cannot get
				compared to the node ptr page number field! */
	ulint		mode,	/* in: PAGE_CUR_L, ...;
unknown's avatar
unknown committed
223
				Inserts should always be made using
224 225 226 227 228
				PAGE_CUR_LE to search the position! */
	ulint		latch_mode, /* in: BTR_SEARCH_LEAF, ..., ORed with
				BTR_INSERT and BTR_ESTIMATE;
				cursor->left_page is used to store a pointer
				to the left neighbor page, in the cases
unknown's avatar
unknown committed
229 230 231 232 233 234
				BTR_SEARCH_PREV and BTR_MODIFY_PREV;
				NOTE that if has_search_latch
				is != 0, we maybe do not have a latch set
				on the cursor page, we assume
				the caller uses his search latch
				to protect the record! */
235
	btr_cur_t*	cursor, /* in/out: tree cursor; the cursor page is
unknown's avatar
unknown committed
236
				s- or x-latched, but see also above! */
237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259
	ulint		has_search_latch,/* in: info on the latch mode the
				caller currently has on btr_search_latch:
				RW_S_LATCH, or 0 */
	mtr_t*		mtr)	/* in: mtr */
{
	dict_tree_t*	tree;
	page_cur_t*	page_cursor;
	page_t*		page;
	page_t*		guess;
	rec_t*		node_ptr;
	ulint		page_no;
	ulint		space;
	ulint		up_match;
	ulint		up_bytes;
	ulint		low_match;
	ulint 		low_bytes;
	ulint		height;
	ulint		savepoint;
	ulint		rw_latch;
	ulint		page_mode;
	ulint		insert_planned;
	ulint		buf_mode;
	ulint		estimate;
unknown's avatar
unknown committed
260
	ulint		ignore_sec_unique;
unknown's avatar
unknown committed
261
	ulint		root_height = 0; /* remove warning */
262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278
#ifdef BTR_CUR_ADAPT
	btr_search_t*	info;
#endif
	/* Currently, PAGE_CUR_LE is the only search mode used for searches
	ending to upper levels */

	ut_ad(level == 0 || mode == PAGE_CUR_LE);
	ut_ad(dict_tree_check_search_tuple(index->tree, tuple));
	ut_ad(!(index->type & DICT_IBUF) || ibuf_inside());
	ut_ad(dtuple_check_typed(tuple));

#ifdef UNIV_DEBUG
	cursor->up_match = ULINT_UNDEFINED;
	cursor->low_match = ULINT_UNDEFINED;
#endif	
	insert_planned = latch_mode & BTR_INSERT;
	estimate = latch_mode & BTR_ESTIMATE;
unknown's avatar
unknown committed
279 280 281
	ignore_sec_unique = latch_mode & BTR_IGNORE_SEC_UNIQUE;
	latch_mode = latch_mode & ~(BTR_INSERT | BTR_ESTIMATE
					| BTR_IGNORE_SEC_UNIQUE);
282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299

	ut_ad(!insert_planned || (mode == PAGE_CUR_LE));
	
	cursor->flag = BTR_CUR_BINARY;
	cursor->index = index;

#ifndef BTR_CUR_ADAPT
	guess = NULL;
#else
	info = btr_search_get_info(index);

	guess = info->root_guess;

#ifdef BTR_CUR_HASH_ADAPT

#ifdef UNIV_SEARCH_PERF_STAT
	info->n_searches++;
#endif	
unknown's avatar
unknown committed
300
	if (btr_search_latch.writer == RW_LOCK_NOT_LOCKED
301
		&& latch_mode <= BTR_MODIFY_LEAF && info->last_hash_succ
302
		&& !estimate
unknown's avatar
unknown committed
303
		&& mode != PAGE_CUR_LE_OR_EXTENDS
unknown's avatar
unknown committed
304
		&& srv_use_adaptive_hash_indexes
305 306 307 308 309 310 311 312 313 314 315 316
	        && btr_search_guess_on_hash(index, info, tuple, mode,
						latch_mode, cursor,
						has_search_latch, mtr)) {

		/* Search using the hash index succeeded */

		ut_ad(cursor->up_match != ULINT_UNDEFINED
					|| mode != PAGE_CUR_GE);
		ut_ad(cursor->up_match != ULINT_UNDEFINED
					|| mode != PAGE_CUR_LE);
		ut_ad(cursor->low_match != ULINT_UNDEFINED
					|| mode != PAGE_CUR_LE);
unknown's avatar
unknown committed
317 318
		btr_cur_n_sea++;

319 320 321 322 323
	        return;
	}
#endif
#endif
	btr_cur_n_non_sea++;
unknown's avatar
unknown committed
324

325 326 327 328 329 330 331 332
	/* If the hash search did not succeed, do binary search down the
	tree */

	if (has_search_latch) {
		/* Release possible search latch to obey latching order */
		rw_lock_s_unlock(&btr_search_latch);
	}

unknown's avatar
unknown committed
333 334 335
	/* Store the position of the tree latch we push to mtr so that we
	know how to release it when we have latched leaf node(s) */

336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364
	savepoint = mtr_set_savepoint(mtr);

	tree = index->tree;
	
	if (latch_mode == BTR_MODIFY_TREE) {
		mtr_x_lock(dict_tree_get_lock(tree), mtr);

	} else if (latch_mode == BTR_CONT_MODIFY_TREE) {
		/* Do nothing */
		ut_ad(mtr_memo_contains(mtr, dict_tree_get_lock(tree),
							MTR_MEMO_X_LOCK));
	} else {
		mtr_s_lock(dict_tree_get_lock(tree), mtr);
	}
	
	page_cursor = btr_cur_get_page_cur(cursor);

	space = dict_tree_get_space(tree);
	page_no = dict_tree_get_page(tree);

	up_match = 0;
	up_bytes = 0;
	low_match = 0;
	low_bytes = 0;

	height = ULINT_UNDEFINED;
	rw_latch = RW_NO_LATCH;
	buf_mode = BUF_GET;

unknown's avatar
unknown committed
365 366 367 368
	/* We use these modified search modes on non-leaf levels of the
	B-tree. These let us end up in the right B-tree leaf. In that leaf
	we use the original search mode. */

369 370 371 372 373 374
	if (mode == PAGE_CUR_GE) {
		page_mode = PAGE_CUR_L;
	} else if (mode == PAGE_CUR_G) {
		page_mode = PAGE_CUR_LE;
	} else if (mode == PAGE_CUR_LE) {
		page_mode = PAGE_CUR_LE;
unknown's avatar
unknown committed
375 376
	} else if (mode == PAGE_CUR_LE_OR_EXTENDS) {
		page_mode = PAGE_CUR_LE_OR_EXTENDS;
377 378 379 380 381 382 383 384 385 386 387 388
	} else {
		ut_ad(mode == PAGE_CUR_L);
		page_mode = PAGE_CUR_L;
	}
			
	/* Loop and search until we arrive at the desired level */

	for (;;) {
		if ((height == 0) && (latch_mode <= BTR_MODIFY_LEAF)) {

			rw_latch = latch_mode;

unknown's avatar
unknown committed
389 390
			if (insert_planned && ibuf_should_try(index,
							ignore_sec_unique)) {
391 392 393 394 395 396 397 398 399 400
				
				/* Try insert to the insert buffer if the
				page is not in the buffer pool */

				buf_mode = BUF_GET_IF_IN_POOL;
			}
		}
retry_page_get:		
		page = buf_page_get_gen(space, page_no, rw_latch, guess,
					buf_mode,
401
					IB__FILE__, __LINE__,
402 403 404 405 406 407 408 409 410
					mtr);
		if (page == NULL) {
			/* This must be a search to perform an insert;
			try insert to the insert buffer */

			ut_ad(buf_mode == BUF_GET_IF_IN_POOL);
			ut_ad(insert_planned);
			ut_ad(cursor->thr);

unknown's avatar
unknown committed
411
			if (ibuf_should_try(index, ignore_sec_unique) &&
412 413 414 415 416 417 418 419 420 421 422 423 424 425 426
				ibuf_insert(tuple, index, space, page_no,
							cursor->thr)) {
				/* Insertion to the insert buffer succeeded */
				cursor->flag = BTR_CUR_INSERT_TO_IBUF;

				return;
			}

			/* Insert to the insert buffer did not succeed:
			retry page get */

			buf_mode = BUF_GET;

			goto retry_page_get;
		}
unknown's avatar
unknown committed
427 428

		buf_block_align(page)->check_index_page_at_flush = TRUE;
429 430 431 432 433 434 435
			
#ifdef UNIV_SYNC_DEBUG					
		if (rw_latch != RW_NO_LATCH) {
			buf_page_dbg_add_level(page, SYNC_TREE_NODE);
		}
#endif
		ut_ad(0 == ut_dulint_cmp(tree->id,
436
					btr_page_get_index_id(page)));
437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508

		if (height == ULINT_UNDEFINED) {
			/* We are in the root node */

			height = btr_page_get_level(page, mtr);
			root_height = height;
			cursor->tree_height = root_height + 1;
#ifdef BTR_CUR_ADAPT
			if (page != guess) {
				info->root_guess = page;
			}	
#endif
		}
	
		if (height == 0) {
			if (rw_latch == RW_NO_LATCH) {

				btr_cur_latch_leaves(tree, page, space,
						page_no, latch_mode, cursor,
						mtr);
			}

			if ((latch_mode != BTR_MODIFY_TREE)
			    && (latch_mode != BTR_CONT_MODIFY_TREE)) {

				/* Release the tree s-latch */

				mtr_release_s_latch_at_savepoint(
						mtr, savepoint,
						dict_tree_get_lock(tree));
			}

			page_mode = mode;
		}

		page_cur_search_with_match(page, tuple, page_mode, &up_match,
					&up_bytes, &low_match, &low_bytes,
					page_cursor);
		if (estimate) {
			btr_cur_add_path_info(cursor, height, root_height);
		}	

		/* If this is the desired level, leave the loop */

		if (level == height) {

			if (level > 0) {
				/* x-latch the page */
				btr_page_get(space, page_no, RW_X_LATCH, mtr);
			}

			break;
		}

		ut_ad(height > 0);

		height--;
		guess = NULL;

		node_ptr = page_cur_get_rec(page_cursor);
		
		/* Go to the child node */
		page_no = btr_node_ptr_get_child_page_no(node_ptr);
	}

	if (level == 0) {
		cursor->low_match = low_match;
		cursor->low_bytes = low_bytes;
		cursor->up_match = up_match;
		cursor->up_bytes = up_bytes;

#ifdef BTR_CUR_ADAPT		
unknown's avatar
unknown committed
509
		if (srv_use_adaptive_hash_indexes) {
510

unknown's avatar
unknown committed
511 512 513
			btr_search_info_update(index, cursor);
		}
#endif
514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546
		ut_ad(cursor->up_match != ULINT_UNDEFINED
						|| mode != PAGE_CUR_GE);
		ut_ad(cursor->up_match != ULINT_UNDEFINED
						|| mode != PAGE_CUR_LE);
		ut_ad(cursor->low_match != ULINT_UNDEFINED
						|| mode != PAGE_CUR_LE);
	}

	if (has_search_latch) {
		
		rw_lock_s_lock(&btr_search_latch);
	}
}

/*********************************************************************
Opens a cursor at either end of an index. */

void
btr_cur_open_at_index_side(
/*=======================*/
	ibool		from_left,	/* in: TRUE if open to the low end,
					FALSE if to the high end */
	dict_index_t*	index,		/* in: index */
	ulint		latch_mode,	/* in: latch mode */
	btr_cur_t*	cursor,		/* in: cursor */
	mtr_t*		mtr)		/* in: mtr */
{
	page_cur_t*	page_cursor;
	dict_tree_t*	tree;
	page_t*		page;
	ulint		page_no;
	ulint		space;
	ulint		height;
unknown's avatar
unknown committed
547
	ulint		root_height = 0; /* remove warning */
548 549
	rec_t*		node_ptr;
	ulint		estimate;
unknown's avatar
unknown committed
550
	ulint           savepoint;
551 552 553 554 555 556

	estimate = latch_mode & BTR_ESTIMATE;
	latch_mode = latch_mode & ~BTR_ESTIMATE;
	
	tree = index->tree;
	
unknown's avatar
unknown committed
557 558 559 560 561
	/* Store the position of the tree latch we push to mtr so that we
	know how to release it when we have latched the leaf node */

	savepoint = mtr_set_savepoint(mtr);

562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578
	if (latch_mode == BTR_MODIFY_TREE) {
		mtr_x_lock(dict_tree_get_lock(tree), mtr);
	} else {
		mtr_s_lock(dict_tree_get_lock(tree), mtr);
	}
	
	page_cursor = btr_cur_get_page_cur(cursor);
	cursor->index = index;

	space = dict_tree_get_space(tree);
	page_no = dict_tree_get_page(tree);

	height = ULINT_UNDEFINED;
	
	for (;;) {
		page = buf_page_get_gen(space, page_no, RW_NO_LATCH, NULL,
					BUF_GET,
579
					IB__FILE__, __LINE__,
580 581 582 583
					mtr);
		ut_ad(0 == ut_dulint_cmp(tree->id,
						btr_page_get_index_id(page)));

unknown's avatar
unknown committed
584 585
		buf_block_align(page)->check_index_page_at_flush = TRUE;

586 587 588 589 590 591 592 593 594 595
		if (height == ULINT_UNDEFINED) {
			/* We are in the root node */

			height = btr_page_get_level(page, mtr);
			root_height = height;
		}

		if (height == 0) {
			btr_cur_latch_leaves(tree, page, space, page_no,
						latch_mode, cursor, mtr);
unknown's avatar
unknown committed
596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611

			/* In versions <= 3.23.52 we had forgotten to
			release the tree latch here. If in an index scan
			we had to scan far to find a record visible to the
			current transaction, that could starve others
			waiting for the tree latch. */
 
			if ((latch_mode != BTR_MODIFY_TREE)
			    && (latch_mode != BTR_CONT_MODIFY_TREE)) {

				/* Release the tree s-latch */

				mtr_release_s_latch_at_savepoint(
						mtr, savepoint,
						dict_tree_get_lock(tree));
			}
612 613 614 615 616 617 618 619 620
		}
		
		if (from_left) {
			page_cur_set_before_first(page, page_cursor);
		} else {
			page_cur_set_after_last(page, page_cursor);
		}

		if (height == 0) {
unknown's avatar
unknown committed
621 622 623 624
		        if (estimate) {
			        btr_cur_add_path_info(cursor, height,
						      root_height);
		        }
625 626 627 628 629 630 631 632 633 634 635 636

			break;
		}

		ut_ad(height > 0);

		if (from_left) {
			page_cur_move_to_next(page_cursor);
		} else {
			page_cur_move_to_prev(page_cursor);
		}

unknown's avatar
unknown committed
637 638 639 640
		if (estimate) {
			btr_cur_add_path_info(cursor, height, root_height);
		}

641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687
		height--;

		node_ptr = page_cur_get_rec(page_cursor);
		
		/* Go to the child node */
		page_no = btr_node_ptr_get_child_page_no(node_ptr);
	}
}
	
/**************************************************************************
Positions a cursor at a randomly chosen position within a B-tree. */

void
btr_cur_open_at_rnd_pos(
/*====================*/
	dict_index_t*	index,		/* in: index */
	ulint		latch_mode,	/* in: BTR_SEARCH_LEAF, ... */
	btr_cur_t*	cursor,		/* in/out: B-tree cursor */
	mtr_t*		mtr)		/* in: mtr */
{
	page_cur_t*	page_cursor;
	dict_tree_t*	tree;
	page_t*		page;
	ulint		page_no;
	ulint		space;
	ulint		height;
	rec_t*		node_ptr;

	tree = index->tree;
	
	if (latch_mode == BTR_MODIFY_TREE) {
		mtr_x_lock(dict_tree_get_lock(tree), mtr);
	} else {
		mtr_s_lock(dict_tree_get_lock(tree), mtr);
	}
	
	page_cursor = btr_cur_get_page_cur(cursor);
	cursor->index = index;

	space = dict_tree_get_space(tree);
	page_no = dict_tree_get_page(tree);

	height = ULINT_UNDEFINED;
	
	for (;;) {
		page = buf_page_get_gen(space, page_no, RW_NO_LATCH, NULL,
					BUF_GET,
688
					IB__FILE__, __LINE__,
689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851
					mtr);
		ut_ad(0 == ut_dulint_cmp(tree->id,
						btr_page_get_index_id(page)));

		if (height == ULINT_UNDEFINED) {
			/* We are in the root node */

			height = btr_page_get_level(page, mtr);
		}

		if (height == 0) {
			btr_cur_latch_leaves(tree, page, space, page_no,
						latch_mode, cursor, mtr);
		}

		page_cur_open_on_rnd_user_rec(page, page_cursor);	

		if (height == 0) {

			break;
		}

		ut_ad(height > 0);

		height--;

		node_ptr = page_cur_get_rec(page_cursor);
		
		/* Go to the child node */
		page_no = btr_node_ptr_get_child_page_no(node_ptr);
	}
}	

/*==================== B-TREE INSERT =========================*/

/*****************************************************************
Inserts a record if there is enough space, or if enough space can
be freed by reorganizing. Differs from _optimistic_insert because
no heuristics is applied to whether it pays to use CPU time for
reorganizing the page or not. */
static
rec_t*
btr_cur_insert_if_possible(
/*=======================*/
				/* out: pointer to inserted record if succeed,
				else NULL */
	btr_cur_t*	cursor,	/* in: cursor on page after which to insert;
				cursor stays valid */
	dtuple_t*	tuple,	/* in: tuple to insert; the size info need not
				have been stored to tuple */
	ibool*		reorg,	/* out: TRUE if reorganization occurred */
	mtr_t*		mtr)	/* in: mtr */
{
	page_cur_t*	page_cursor;
	page_t*		page;
	rec_t*		rec;

	ut_ad(dtuple_check_typed(tuple));
	
	*reorg = FALSE;

	page = btr_cur_get_page(cursor);

	ut_ad(mtr_memo_contains(mtr, buf_block_align(page),
							MTR_MEMO_PAGE_X_FIX));
	page_cursor = btr_cur_get_page_cur(cursor);
	
	/* Now, try the insert */
	rec = page_cur_tuple_insert(page_cursor, tuple, mtr);	

	if (!rec) {
		/* If record did not fit, reorganize */

		btr_page_reorganize(page, mtr);

		*reorg = TRUE;

		page_cur_search(page, tuple, PAGE_CUR_LE, page_cursor);

		rec = page_cur_tuple_insert(page_cursor, tuple, mtr);
	}

	return(rec);
}

/*****************************************************************
For an insert, checks the locks and does the undo logging if desired. */
UNIV_INLINE
ulint
btr_cur_ins_lock_and_undo(
/*======================*/
				/* out: DB_SUCCESS, DB_WAIT_LOCK,
				DB_FAIL, or error number */
	ulint		flags,	/* in: undo logging and locking flags: if
				not zero, the parameters index and thr
				should be specified */
	btr_cur_t*	cursor,	/* in: cursor on page after which to insert */
	dtuple_t*	entry,	/* in: entry to insert */
	que_thr_t*	thr,	/* in: query thread or NULL */
	ibool*		inherit)/* out: TRUE if the inserted new record maybe
				should inherit LOCK_GAP type locks from the
				successor record */
{
	dict_index_t*	index;
	ulint		err;
	rec_t*		rec;
	dulint		roll_ptr;

	/* Check if we have to wait for a lock: enqueue an explicit lock
	request if yes */

	rec = btr_cur_get_rec(cursor);
	index = cursor->index;
	
	err = lock_rec_insert_check_and_lock(flags, rec, index, thr, inherit);
	
	if (err != DB_SUCCESS) {

		return(err);
	}

	if ((index->type & DICT_CLUSTERED) && !(index->type & DICT_IBUF)) {

		err = trx_undo_report_row_operation(flags, TRX_UNDO_INSERT_OP,
					thr, index, entry, NULL, 0, NULL,
					&roll_ptr);
		if (err != DB_SUCCESS) {

			return(err);
		}

		/* Now we can fill in the roll ptr field in entry */

		if (!(flags & BTR_KEEP_SYS_FLAG)) {

			row_upd_index_entry_sys_field(entry, index,
						DATA_ROLL_PTR, roll_ptr);
		}
	}

	return(DB_SUCCESS);
}

/*****************************************************************
Tries to perform an insert to a page in an index tree, next to cursor.
It is assumed that mtr holds an x-latch on the page. The operation does
not succeed if there is too little space on the page. If there is just
one record on the page, the insert will always succeed; this is to
prevent trying to split a page with just one record. */

ulint
btr_cur_optimistic_insert(
/*======================*/
				/* out: DB_SUCCESS, DB_WAIT_LOCK,
				DB_FAIL, or error number */
	ulint		flags,	/* in: undo logging and locking flags: if not
				zero, the parameters index and thr should be
				specified */
	btr_cur_t*	cursor,	/* in: cursor on page after which to insert;
				cursor stays valid */
	dtuple_t*	entry,	/* in: entry to insert */
	rec_t**		rec,	/* out: pointer to inserted record if
				succeed */
852 853 854
	big_rec_t**	big_rec,/* out: big rec vector whose fields have to
				be stored externally by the caller, or
				NULL */
855 856 857
	que_thr_t*	thr,	/* in: query thread or NULL */
	mtr_t*		mtr)	/* in: mtr */
{
858
	big_rec_t*	big_rec_vec	= NULL;
859 860 861 862 863 864 865 866 867 868 869 870
	dict_index_t*	index;
	page_cur_t*	page_cursor;
	page_t*		page;
	ulint		max_size;
	rec_t*		dummy_rec;
	ulint		level;
	ibool		reorg;
	ibool		inherit;
	ulint		rec_size;
	ulint		data_size;
	ulint		extra_size;
	ulint		type;
unknown's avatar
unknown committed
871
	ulint		err;	
872

873 874
	*big_rec = NULL;

875 876 877
	page = btr_cur_get_page(cursor);
	index = cursor->index;

unknown's avatar
unknown committed
878 879
	if (!dtuple_check_typed_no_assert(entry)) {
		fprintf(stderr,
unknown's avatar
unknown committed
880
"InnoDB: Error in a tuple to insert into table %s index %s\n",
unknown's avatar
unknown committed
881 882 883 884 885 886
					index->table_name, index->name);
	}
	
	if (btr_cur_print_record_ops && thr) {
		printf(
	"Trx with id %lu %lu going to insert to table %s index %s\n",
887 888
		(unsigned long) ut_dulint_get_high(thr_get_trx(thr)->id),
		(unsigned long) ut_dulint_get_low(thr_get_trx(thr)->id),
unknown's avatar
unknown committed
889 890 891 892
		index->table_name, index->name);
		dtuple_print(entry);
	}
	
893 894 895 896 897
	ut_ad(mtr_memo_contains(mtr, buf_block_align(page),
							MTR_MEMO_PAGE_X_FIX));
	max_size = page_get_max_insert_size_after_reorganize(page, 1);
	level = btr_page_get_level(page, mtr);

898
calculate_sizes_again:
899 900 901 902 903 904
	/* Calculate the record size when entry is converted to a record */
	data_size = dtuple_get_data_size(entry);
	extra_size = rec_get_converted_extra_size(data_size,
						dtuple_get_n_fields(entry));
	rec_size = data_size + extra_size;

905 906 907 908 909 910
	if ((rec_size >= page_get_free_space_of_empty() / 2)
	    || (rec_size >= REC_MAX_DATA_SIZE)) {

		/* The record is so big that we have to store some fields
		externally on separate database pages */
		
911
                big_rec_vec = dtuple_convert_big_rec(index, entry, NULL, 0);
912 913 914 915 916

		if (big_rec_vec == NULL) {
		
			return(DB_TOO_BIG_RECORD);
		}
917

918
		goto calculate_sizes_again;
919 920 921 922 923 924 925 926 927 928 929 930 931 932
	}

	/* If there have been many consecutive inserts, and we are on the leaf
	level, check if we have to split the page to reserve enough free space
	for future updates of records. */

	type = index->type;
	
	if ((type & DICT_CLUSTERED)
	    && (dict_tree_get_space_reserve(index->tree) + rec_size > max_size)
	    && (page_get_n_recs(page) >= 2)
	    && (0 == level)
	    && (btr_page_get_split_rec_to_right(cursor, &dummy_rec)
	        || btr_page_get_split_rec_to_left(cursor, &dummy_rec))) {
933 934 935 936 937

	        if (big_rec_vec) {
			dtuple_convert_back_big_rec(index, entry, big_rec_vec);
		}

938 939 940 941 942 943 944 945
		return(DB_FAIL);
	}
	
	if (!(((max_size >= rec_size)
	       && (max_size >= BTR_CUR_PAGE_REORGANIZE_LIMIT))
	      || (page_get_max_insert_size(page, 1) >= rec_size)
	      || (page_get_n_recs(page) <= 1))) {

946 947 948
	        if (big_rec_vec) {
			dtuple_convert_back_big_rec(index, entry, big_rec_vec);
		}
949 950 951 952 953 954 955 956
		return(DB_FAIL);
	}

        /* Check locks and write to the undo log, if specified */
        err = btr_cur_ins_lock_and_undo(flags, cursor, entry, thr, &inherit);

	if (err != DB_SUCCESS) {

957 958 959
	        if (big_rec_vec) {
			dtuple_convert_back_big_rec(index, entry, big_rec_vec);
		}
960 961 962 963 964 965 966 967 968 969
		return(err);
	}

	page_cursor = btr_cur_get_page_cur(cursor);

	reorg = FALSE;

	/* Now, try the insert */

	*rec = page_cur_insert_rec_low(page_cursor, entry, data_size,
unknown's avatar
unknown committed
970
								NULL, mtr);
971 972 973 974 975 976 977 978 979 980 981 982
	if (!(*rec)) {
		/* If the record did not fit, reorganize */
		btr_page_reorganize(page, mtr);

		ut_ad(page_get_max_insert_size(page, 1) == max_size);
		
		reorg = TRUE;

		page_cur_search(page, entry, PAGE_CUR_LE, page_cursor);

		*rec = page_cur_tuple_insert(page_cursor, entry, mtr);

983 984 985 986 987 988 989 990
		if (!(*rec)) {
			char* err_buf = mem_alloc(1000);

			dtuple_sprintf(err_buf, 900, entry);
			
			fprintf(stderr,
	"InnoDB: Error: cannot insert tuple %s to index %s of table %s\n"
	"InnoDB: max insert size %lu\n",
991 992
			err_buf, index->name, index->table->name,
			(unsigned long) max_size);
993 994 995 996

			mem_free(err_buf);
		}
		
997 998 999 1000 1001 1002 1003 1004 1005 1006
		ut_a(*rec); /* <- We calculated above the record would fit */
	}

#ifdef BTR_CUR_HASH_ADAPT
	if (!reorg && (0 == level) && (cursor->flag == BTR_CUR_HASH)) {
		btr_search_update_hash_node_on_insert(cursor);
	} else {
		btr_search_update_hash_on_insert(cursor);
	}
#endif
1007

1008 1009 1010 1011 1012 1013 1014 1015 1016
	if (!(flags & BTR_NO_LOCKING_FLAG) && inherit) {

		lock_update_insert(*rec);
	}

/*	printf("Insert to page %lu, max ins size %lu, rec %lu ind type %lu\n",
			buf_frame_get_page_no(page), max_size,
					rec_size + PAGE_DIR_SLOT_SIZE, type);
*/	
unknown's avatar
unknown committed
1017
	if (!(type & DICT_CLUSTERED)) {
1018 1019 1020 1021 1022
		/* We have added a record to page: update its free bits */
		ibuf_update_free_bits_if_full(cursor->index, page, max_size,
					rec_size + PAGE_DIR_SLOT_SIZE);
	}

1023 1024
	*big_rec = big_rec_vec;

1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048
	return(DB_SUCCESS);
}

/*****************************************************************
Performs an insert on a page of an index tree. It is assumed that mtr
holds an x-latch on the tree and on the cursor page. If the insert is
made on the leaf level, to avoid deadlocks, mtr must also own x-latches
to brothers of page, if those brothers exist. */

ulint
btr_cur_pessimistic_insert(
/*=======================*/
				/* out: DB_SUCCESS or error number */
	ulint		flags,	/* in: undo logging and locking flags: if not
				zero, the parameter thr should be
				specified; if no undo logging is specified,
				then the caller must have reserved enough
				free extents in the file space so that the
				insertion will certainly succeed */
	btr_cur_t*	cursor,	/* in: cursor after which to insert;
				cursor stays valid */
	dtuple_t*	entry,	/* in: entry to insert */
	rec_t**		rec,	/* out: pointer to inserted record if
				succeed */
1049 1050 1051
	big_rec_t**	big_rec,/* out: big rec vector whose fields have to
				be stored externally by the caller, or
				NULL */
1052 1053 1054
	que_thr_t*	thr,	/* in: query thread or NULL */
	mtr_t*		mtr)	/* in: mtr */
{
1055 1056 1057 1058 1059 1060 1061
	dict_index_t*	index		= cursor->index;
	big_rec_t*	big_rec_vec	= NULL;
	page_t*		page;
	ulint		err;
	ibool		dummy_inh;
	ibool		success;
	ulint		n_extents	= 0;
unknown's avatar
unknown committed
1062
	ulint		n_reserved;
1063 1064 1065
	
	ut_ad(dtuple_check_typed(entry));

1066 1067
	*big_rec = NULL;

1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080
	page = btr_cur_get_page(cursor);

	ut_ad(mtr_memo_contains(mtr,
				dict_tree_get_lock(btr_cur_get_tree(cursor)),
							MTR_MEMO_X_LOCK));
	ut_ad(mtr_memo_contains(mtr, buf_block_align(page),
							MTR_MEMO_PAGE_X_FIX));

	/* Try first an optimistic insert; reset the cursor flag: we do not
	assume anything of how it was positioned */

	cursor->flag = BTR_CUR_BINARY;

1081
	err = btr_cur_optimistic_insert(flags, cursor, entry, rec, big_rec,
unknown's avatar
unknown committed
1082
								thr, mtr);
1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104
	if (err != DB_FAIL) {

		return(err);
	}

	/* Retry with a pessimistic insert. Check locks and write to undo log,
	if specified */

	err = btr_cur_ins_lock_and_undo(flags, cursor, entry, thr, &dummy_inh);

	if (err != DB_SUCCESS) {

		return(err);
	}

	if (!(flags & BTR_NO_UNDO_LOG_FLAG)) {	
		/* First reserve enough free space for the file segments
		of the index tree, so that the insert will not fail because
		of lack of space */

		n_extents = cursor->tree_height / 16 + 3;

unknown's avatar
unknown committed
1105
		success = fsp_reserve_free_extents(&n_reserved, index->space,
1106 1107 1108 1109 1110 1111 1112 1113
						n_extents, FSP_NORMAL, mtr);
		if (!success) {
			err = DB_OUT_OF_FILE_SPACE;

			return(err);
		}
	}

1114 1115 1116 1117 1118 1119 1120
	if ((rec_get_converted_size(entry)
				>= page_get_free_space_of_empty() / 2)
	    || (rec_get_converted_size(entry) >= REC_MAX_DATA_SIZE)) {

		/* The record is so big that we have to store some fields
		externally on separate database pages */
		
1121
                big_rec_vec = dtuple_convert_big_rec(index, entry, NULL, 0);
1122 1123 1124

		if (big_rec_vec == NULL) {
		
unknown's avatar
unknown committed
1125 1126
			if (n_extents > 0) {
			        fil_space_release_free_extents(index->space,
unknown's avatar
unknown committed
1127
								n_reserved);
unknown's avatar
unknown committed
1128
			}
1129 1130 1131 1132 1133
			return(DB_TOO_BIG_RECORD);
		}
	}

	if (dict_tree_get_page(index->tree)
1134 1135 1136 1137 1138 1139 1140 1141
					== buf_frame_get_page_no(page)) {

		/* The page is the root page */
		*rec = btr_root_raise_and_insert(cursor, entry, mtr);
	} else {
		*rec = btr_page_split_and_insert(cursor, entry, mtr);
	}

1142
	btr_cur_position(index, page_rec_get_prev(*rec), cursor);	
1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154

#ifdef BTR_CUR_ADAPT
	btr_search_update_hash_on_insert(cursor);
#endif
	if (!(flags & BTR_NO_LOCKING_FLAG)) {

		lock_update_insert(*rec);
	}

	err = DB_SUCCESS;

	if (n_extents > 0) {
unknown's avatar
unknown committed
1155
		fil_space_release_free_extents(index->space, n_reserved);
1156
	}
1157 1158 1159

	*big_rec = big_rec_vec;

1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189
	return(err);
}

/*==================== B-TREE UPDATE =========================*/

/*****************************************************************
For an update, checks the locks and does the undo logging. */
UNIV_INLINE
ulint
btr_cur_upd_lock_and_undo(
/*======================*/
				/* out: DB_SUCCESS, DB_WAIT_LOCK, or error
				number */
	ulint		flags,	/* in: undo logging and locking flags */
	btr_cur_t*	cursor,	/* in: cursor on record to update */
	upd_t*		update,	/* in: update vector */
	ulint		cmpl_info,/* in: compiler info on secondary index
				updates */
	que_thr_t*	thr,	/* in: query thread */
	dulint*		roll_ptr)/* out: roll pointer */
{
	dict_index_t*	index;
	rec_t*		rec;
	ulint		err;
	
	ut_ad(cursor && update && thr && roll_ptr);

	rec = btr_cur_get_rec(cursor);
	index = cursor->index;
	
unknown's avatar
unknown committed
1190 1191 1192
	if (!(index->type & DICT_CLUSTERED)) {
		/* We do undo logging only when we update a clustered index
		record */
unknown's avatar
unknown committed
1193 1194
		return(lock_sec_rec_modify_check_and_lock(flags, rec, index,
									thr));
unknown's avatar
unknown committed
1195 1196
	}

1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242
	/* Check if we have to wait for a lock: enqueue an explicit lock
	request if yes */

	err = DB_SUCCESS;

	if (!(flags & BTR_NO_LOCKING_FLAG)) {
		err = lock_clust_rec_modify_check_and_lock(flags, rec, index,
									thr);
		if (err != DB_SUCCESS) {

			return(err);
		}
	}

	/* Append the info about the update in the undo log */

	err = trx_undo_report_row_operation(flags, TRX_UNDO_MODIFY_OP, thr,
						index, NULL, update,
						cmpl_info, rec, roll_ptr);
	return(err);
}

/***************************************************************
Writes a redo log record of updating a record in-place. */
UNIV_INLINE
void
btr_cur_update_in_place_log(
/*========================*/
	ulint		flags,		/* in: flags */
	rec_t*		rec,		/* in: record */
	dict_index_t*	index,		/* in: index where cursor positioned */
	upd_t*		update,		/* in: update vector */
	trx_t*		trx,		/* in: transaction */
	dulint		roll_ptr,	/* in: roll ptr */
	mtr_t*		mtr)		/* in: mtr */
{
	byte*	log_ptr;

	log_ptr = mlog_open(mtr, 30 + MLOG_BUF_MARGIN);

	log_ptr = mlog_write_initial_log_record_fast(rec,
				MLOG_REC_UPDATE_IN_PLACE, log_ptr, mtr);

	mach_write_to_1(log_ptr, flags);
	log_ptr++;

unknown's avatar
unknown committed
1243 1244 1245 1246 1247 1248 1249
	/* The code below assumes index is a clustered index: change index to
	the clustered index if we are updating a secondary index record (or we
	could as well skip writing the sys col values to the log in this case
	because they are not needed for a secondary index record update) */

	index = dict_table_get_first_index(index->table);

1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300
	log_ptr = row_upd_write_sys_vals_to_log(index, trx, roll_ptr, log_ptr,
									mtr);
	mach_write_to_2(log_ptr, rec - buf_frame_align(rec));
	log_ptr += 2;

	row_upd_index_write_log(update, log_ptr, mtr);
}	

/***************************************************************
Parses a redo log record of updating a record in-place. */

byte*
btr_cur_parse_update_in_place(
/*==========================*/
			/* out: end of log record or NULL */
	byte*	ptr,	/* in: buffer */
	byte*	end_ptr,/* in: buffer end */
	page_t*	page)	/* in: page or NULL */
{
	ulint	flags;
	rec_t*	rec;
	upd_t*	update;
	ulint	pos;
	dulint	trx_id;
	dulint	roll_ptr;
	ulint	rec_offset;
	mem_heap_t* heap;

	if (end_ptr < ptr + 1) {

		return(NULL);
	}
	
	flags = mach_read_from_1(ptr);
	ptr++;

	ptr = row_upd_parse_sys_vals(ptr, end_ptr, &pos, &trx_id, &roll_ptr);

	if (ptr == NULL) {

		return(NULL);
	}

	if (end_ptr < ptr + 2) {

		return(NULL);
	}

	rec_offset = mach_read_from_2(ptr);
	ptr += 2;

unknown's avatar
unknown committed
1301 1302
	ut_a(rec_offset <= UNIV_PAGE_SIZE);

1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335
	heap = mem_heap_create(256);
	
	ptr = row_upd_index_parse(ptr, end_ptr, heap, &update);

	if (ptr == NULL) {
		mem_heap_free(heap);
		
		return(NULL);
	}

	if (!page) {
		mem_heap_free(heap);

		return(ptr);
	}
	
	rec = page + rec_offset;
	
	/* We do not need to reserve btr_search_latch, as the page is only
	being recovered, and there cannot be a hash index to it. */

	if (!(flags & BTR_KEEP_SYS_FLAG)) {
		row_upd_rec_sys_fields_in_recovery(rec, pos, trx_id, roll_ptr);
	}

	row_upd_rec_in_place(rec, update);

	mem_heap_free(heap);

	return(ptr);
}

/*****************************************************************
unknown's avatar
unknown committed
1336 1337
Updates a record when the update causes no size changes in its fields.
We assume here that the ordering fields of the record do not change. */
1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356

ulint
btr_cur_update_in_place(
/*====================*/
				/* out: DB_SUCCESS or error number */
	ulint		flags,	/* in: undo logging and locking flags */
	btr_cur_t*	cursor,	/* in: cursor on the record to update;
				cursor stays valid and positioned on the
				same record */
	upd_t*		update,	/* in: update vector */
	ulint		cmpl_info,/* in: compiler info on secondary index
				updates */
	que_thr_t*	thr,	/* in: query thread */
	mtr_t*		mtr)	/* in: mtr */
{
	dict_index_t*	index;
	buf_block_t*	block;
	ulint		err;
	rec_t*		rec;
unknown's avatar
unknown committed
1357
	dulint		roll_ptr	= ut_dulint_zero;
1358
	trx_t*		trx;
1359
	ibool		was_delete_marked;
1360 1361 1362 1363 1364

	rec = btr_cur_get_rec(cursor);
	index = cursor->index;
	trx = thr_get_trx(thr);
	
unknown's avatar
unknown committed
1365 1366 1367
	if (btr_cur_print_record_ops && thr) {
		printf(
	"Trx with id %lu %lu going to update table %s index %s\n",
1368 1369
		(unsigned long) ut_dulint_get_high(thr_get_trx(thr)->id),
		(unsigned long) ut_dulint_get_low(thr_get_trx(thr)->id),
unknown's avatar
unknown committed
1370 1371 1372 1373
		index->table_name, index->name);
		rec_print(rec);
	}

1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384
	/* Do lock checking and undo logging */
	err = btr_cur_upd_lock_and_undo(flags, cursor, update, cmpl_info,
							thr, &roll_ptr);
	if (err != DB_SUCCESS) {

		return(err);
	}

	block = buf_block_align(rec);

	if (block->is_hashed) {
unknown's avatar
unknown committed
1385 1386 1387 1388 1389 1390
		/* The function row_upd_changes_ord_field_binary works only
		if the update vector was built for a clustered index, we must
		NOT call it if index is secondary */

	        if (!(index->type & DICT_CLUSTERED)
		    || row_upd_changes_ord_field_binary(NULL, index, update)) {
unknown's avatar
unknown committed
1391 1392 1393 1394 1395

		        /* Remove possible hash index pointer to this record */
	                btr_search_update_hash_on_delete(cursor);
	        }

1396 1397 1398 1399 1400 1401 1402 1403 1404
		rw_lock_x_lock(&btr_search_latch);
	}

	if (!(flags & BTR_KEEP_SYS_FLAG)) {
		row_upd_rec_sys_fields(rec, index, trx, roll_ptr);
	}

	/* FIXME: in a mixed tree, all records may not have enough ordering
	fields for btr search: */
1405 1406

	was_delete_marked = rec_get_deleted_flag(rec);
1407 1408 1409 1410 1411 1412 1413 1414 1415
	
	row_upd_rec_in_place(rec, update);

	if (block->is_hashed) {
		rw_lock_x_unlock(&btr_search_latch);
	}

	btr_cur_update_in_place_log(flags, rec, index, update, trx, roll_ptr,
									mtr);
1416 1417 1418 1419 1420 1421 1422
	if (was_delete_marked && !rec_get_deleted_flag(rec)) {
		/* The new updated record owns its possible externally
		stored fields */

		btr_cur_unmark_extern_fields(rec, mtr);
	}

1423 1424 1425 1426 1427 1428 1429
	return(DB_SUCCESS);
}

/*****************************************************************
Tries to update a record on a page in an index tree. It is assumed that mtr
holds an x-latch on the page. The operation does not succeed if there is too
little space on the page or if the update would result in too empty a page,
unknown's avatar
unknown committed
1430 1431
so that tree compression is recommended. We assume here that the ordering
fields of the record do not change. */
1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461

ulint
btr_cur_optimistic_update(
/*======================*/
				/* out: DB_SUCCESS, or DB_OVERFLOW if the
				updated record does not fit, DB_UNDERFLOW
				if the page would become too empty */
	ulint		flags,	/* in: undo logging and locking flags */
	btr_cur_t*	cursor,	/* in: cursor on the record to update;
				cursor stays valid and positioned on the
				same record */
	upd_t*		update,	/* in: update vector; this must also
				contain trx id and roll ptr fields */
	ulint		cmpl_info,/* in: compiler info on secondary index
				updates */
	que_thr_t*	thr,	/* in: query thread */
	mtr_t*		mtr)	/* in: mtr */
{
	dict_index_t*	index;
	page_cur_t*	page_cursor;
	ulint		err;
	page_t*		page;
	rec_t*		rec;
	ulint		max_size;
	ulint		new_rec_size;
	ulint		old_rec_size;
	dtuple_t*	new_entry;
	dulint		roll_ptr;
	trx_t*		trx;
	mem_heap_t*	heap;
1462 1463
	ibool		reorganized	= FALSE;
	ulint		i;
1464 1465 1466 1467 1468
	
	page = btr_cur_get_page(cursor);
	rec = btr_cur_get_rec(cursor);
	index = cursor->index;
	
unknown's avatar
unknown committed
1469 1470 1471
	if (btr_cur_print_record_ops && thr) {
		printf(
	"Trx with id %lu %lu going to update table %s index %s\n",
1472 1473
		(unsigned long) ut_dulint_get_high(thr_get_trx(thr)->id),
		(unsigned long) ut_dulint_get_low(thr_get_trx(thr)->id),
unknown's avatar
unknown committed
1474 1475 1476 1477
		index->table_name, index->name);
		rec_print(rec);
	}

1478 1479
	ut_ad(mtr_memo_contains(mtr, buf_block_align(page),
							MTR_MEMO_PAGE_X_FIX));
unknown's avatar
unknown committed
1480
	if (!row_upd_changes_field_size_or_external(rec, index, update)) {
1481

unknown's avatar
unknown committed
1482 1483 1484
		/* The simplest and the most common case: the update does not
		change the size of any field and none of the updated fields is
		externally stored in rec or update */
1485 1486 1487 1488 1489

		return(btr_cur_update_in_place(flags, cursor, update,
							cmpl_info, thr, mtr));
	}

1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506
	for (i = 0; i < upd_get_n_fields(update); i++) {
		if (upd_get_nth_field(update, i)->extern_storage) {

			/* Externally stored fields are treated in pessimistic
			update */

			return(DB_OVERFLOW);
		}
	}

	if (rec_contains_externally_stored_field(btr_cur_get_rec(cursor))) {
		/* Externally stored fields are treated in pessimistic
		update */

		return(DB_OVERFLOW);
	}
	
1507 1508 1509 1510 1511 1512
	page_cursor = btr_cur_get_page_cur(cursor);
	
	heap = mem_heap_create(1024);
	
	new_entry = row_rec_to_index_entry(ROW_COPY_DATA, index, rec, heap);

unknown's avatar
unknown committed
1513 1514
	row_upd_index_replace_new_col_vals_index_pos(new_entry, index, update,
									NULL);
1515 1516 1517 1518 1519
	old_rec_size = rec_get_size(rec);
	new_rec_size = rec_get_converted_size(new_entry);
	
	if (new_rec_size >= page_get_free_space_of_empty() / 2) {

1520
		mem_heap_free(heap);		
1521

1522
		return(DB_OVERFLOW);
1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585
	}

	max_size = old_rec_size
			+ page_get_max_insert_size_after_reorganize(page, 1);

	if (page_get_data_size(page) - old_rec_size + new_rec_size
					< BTR_CUR_PAGE_COMPRESS_LIMIT) {

		/* The page would become too empty */

		mem_heap_free(heap);

		return(DB_UNDERFLOW);
	}

	if (!(((max_size >= BTR_CUR_PAGE_REORGANIZE_LIMIT)
	       				&& (max_size >= new_rec_size))
	      || (page_get_n_recs(page) <= 1))) {

		/* There was not enough space, or it did not pay to
		reorganize: for simplicity, we decide what to do assuming a
		reorganization is needed, though it might not be necessary */

		mem_heap_free(heap);		

		return(DB_OVERFLOW);
	}

	/* Do lock checking and undo logging */
	err = btr_cur_upd_lock_and_undo(flags, cursor, update, cmpl_info, thr,
								&roll_ptr);
	if (err != DB_SUCCESS) {

		mem_heap_free(heap);

		return(err);
	}
        
        /* Ok, we may do the replacement. Store on the page infimum the
	explicit locks on rec, before deleting rec (see the comment in
	.._pessimistic_update). */

	lock_rec_store_on_page_infimum(rec);

	btr_search_update_hash_on_delete(cursor);

        page_cur_delete_rec(page_cursor, mtr);

	page_cur_move_to_prev(page_cursor);
        
	trx = thr_get_trx(thr);

	if (!(flags & BTR_KEEP_SYS_FLAG)) {
		row_upd_index_entry_sys_field(new_entry, index, DATA_ROLL_PTR,
								roll_ptr);
		row_upd_index_entry_sys_field(new_entry, index, DATA_TRX_ID,
								trx->id);
	}

	rec = btr_cur_insert_if_possible(cursor, new_entry, &reorganized, mtr);

	ut_a(rec); /* <- We calculated above the insert would fit */

1586 1587 1588 1589 1590 1591 1592
	if (!rec_get_deleted_flag(rec)) {
		/* The new inserted record owns its possible externally
		stored fields */

		btr_cur_unmark_extern_fields(rec, mtr);
	}

1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642
	/* Restore the old explicit lock state on the record */

	lock_rec_restore_from_page_infimum(rec, page);

        page_cur_move_to_next(page_cursor);

	mem_heap_free(heap);		

	return(DB_SUCCESS);
}

/*****************************************************************
If, in a split, a new supremum record was created as the predecessor of the
updated record, the supremum record must inherit exactly the locks on the
updated record. In the split it may have inherited locks from the successor
of the updated record, which is not correct. This function restores the
right locks for the new supremum. */
static
void
btr_cur_pess_upd_restore_supremum(
/*==============================*/
	rec_t*	rec,	/* in: updated record */
	mtr_t*	mtr)	/* in: mtr */
{
	page_t*	page;
	page_t*	prev_page;
	ulint	space;
	ulint	prev_page_no;
	
	page = buf_frame_align(rec);

	if (page_rec_get_next(page_get_infimum_rec(page)) != rec) {
		/* Updated record is not the first user record on its page */ 
	
		return;
	}

	space = buf_frame_get_space_id(page);
	prev_page_no = btr_page_get_prev(page, mtr);
	
	ut_ad(prev_page_no != FIL_NULL);
	prev_page = buf_page_get_with_no_latch(space, prev_page_no, mtr);

	/* We must already have an x-latch to prev_page! */
	ut_ad(mtr_memo_contains(mtr, buf_block_align(prev_page),
		      				MTR_MEMO_PAGE_X_FIX));

	lock_rec_reset_and_inherit_gap_locks(page_get_supremum_rec(prev_page),
									rec);
}
1643

1644 1645 1646 1647
/*****************************************************************
Performs an update of a record on a page of a tree. It is assumed
that mtr holds an x-latch on the tree and on the cursor page. If the
update is made on the leaf level, to avoid deadlocks, mtr must also
unknown's avatar
unknown committed
1648 1649
own x-latches to brothers of page, if those brothers exist. We assume
here that the ordering fields of the record do not change. */
1650 1651 1652 1653 1654 1655 1656

ulint
btr_cur_pessimistic_update(
/*=======================*/
				/* out: DB_SUCCESS or error code */
	ulint		flags,	/* in: undo logging, locking, and rollback
				flags */
1657 1658 1659
	btr_cur_t*	cursor,	/* in: cursor on the record to update */
	big_rec_t**	big_rec,/* out: big rec vector whose fields have to
				be stored externally by the caller, or NULL */
1660 1661 1662 1663 1664 1665 1666 1667
	upd_t*		update,	/* in: update vector; this is allowed also
				contain trx id and roll ptr fields, but
				the values in update vector have no effect */
	ulint		cmpl_info,/* in: compiler info on secondary index
				updates */
	que_thr_t*	thr,	/* in: query thread */
	mtr_t*		mtr)	/* in: mtr */
{
1668 1669
	big_rec_t*	big_rec_vec	= NULL;
	big_rec_t*	dummy_big_rec;
1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684
	dict_index_t*	index;
	page_t*		page;
	dict_tree_t*	tree;
	rec_t*		rec;
	page_cur_t*	page_cursor;
	dtuple_t*	new_entry;
	mem_heap_t*	heap;
	ulint		err;
	ulint		optim_err;
	ibool		dummy_reorganized;
	dulint		roll_ptr;
	trx_t*		trx;
	ibool		was_first;
	ibool		success;
	ulint		n_extents	= 0;
unknown's avatar
unknown committed
1685
	ulint		n_reserved;
1686 1687 1688 1689 1690
	ulint*		ext_vect;
	ulint		n_ext_vect;
	ulint		reserve_flag;
	
	*big_rec = NULL;
1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724
	
	page = btr_cur_get_page(cursor);
	rec = btr_cur_get_rec(cursor);
	index = cursor->index;
	tree = index->tree;

	ut_ad(mtr_memo_contains(mtr, dict_tree_get_lock(tree),
							MTR_MEMO_X_LOCK));
	ut_ad(mtr_memo_contains(mtr, buf_block_align(page),
							MTR_MEMO_PAGE_X_FIX));

	optim_err = btr_cur_optimistic_update(flags, cursor, update,
							cmpl_info, thr, mtr);

	if (optim_err != DB_UNDERFLOW && optim_err != DB_OVERFLOW) {

		return(optim_err);
	}

	/* Do lock checking and undo logging */
	err = btr_cur_upd_lock_and_undo(flags, cursor, update, cmpl_info,
							thr, &roll_ptr);
	if (err != DB_SUCCESS) {

		return(err);
	}

	if (optim_err == DB_OVERFLOW) {
		/* First reserve enough free space for the file segments
		of the index tree, so that the update will not fail because
		of lack of space */

		n_extents = cursor->tree_height / 16 + 3;

1725 1726 1727 1728 1729 1730
		if (flags & BTR_NO_UNDO_LOG_FLAG) {
			reserve_flag = FSP_CLEANING;
		} else {
			reserve_flag = FSP_NORMAL;
		}
		
unknown's avatar
unknown committed
1731 1732
		success = fsp_reserve_free_extents(&n_reserved,
						cursor->index->space,
1733
						n_extents, reserve_flag, mtr);
1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746
		if (!success) {
			err = DB_OUT_OF_FILE_SPACE;

			return(err);
		}
	}
	
	heap = mem_heap_create(1024);

	trx = thr_get_trx(thr);
	
	new_entry = row_rec_to_index_entry(ROW_COPY_DATA, index, rec, heap);

unknown's avatar
unknown committed
1747 1748
	row_upd_index_replace_new_col_vals_index_pos(new_entry, index, update,
									heap);
1749 1750 1751 1752 1753 1754 1755
	if (!(flags & BTR_KEEP_SYS_FLAG)) {
		row_upd_index_entry_sys_field(new_entry, index, DATA_ROLL_PTR,
								roll_ptr);
		row_upd_index_entry_sys_field(new_entry, index, DATA_TRX_ID,
								trx->id);
	}

1756 1757 1758
	if (flags & BTR_NO_UNDO_LOG_FLAG) {
		/* We are in a transaction rollback undoing a row
		update: we must free possible externally stored fields
1759 1760 1761 1762
		which got new values in the update, if they are not
		inherited values. They can be inherited if we have
		updated the primary key to another value, and then
		update it back again. */
1763 1764 1765

		ut_a(big_rec_vec == NULL);
		
1766 1767
		btr_rec_free_updated_extern_fields(index, rec, update,
						 		TRUE, mtr);
1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779
	}

	/* We have to set appropriate extern storage bits in the new
	record to be inserted: we have to remember which fields were such */

	ext_vect = mem_heap_alloc(heap, sizeof(ulint) * rec_get_n_fields(rec));
	n_ext_vect = btr_push_update_extern_fields(ext_vect, rec, update);
	
	if ((rec_get_converted_size(new_entry) >=
				page_get_free_space_of_empty() / 2)
	    || (rec_get_converted_size(new_entry) >= REC_MAX_DATA_SIZE)) {

1780 1781
                big_rec_vec = dtuple_convert_big_rec(index, new_entry,
                					ext_vect, n_ext_vect);
1782 1783 1784 1785
		if (big_rec_vec == NULL) {

			mem_heap_free(heap);
		
unknown's avatar
unknown committed
1786 1787
			err = DB_TOO_BIG_RECORD;

1788 1789 1790 1791
			goto return_after_reservations;
		}
	}

1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805
	page_cursor = btr_cur_get_page_cur(cursor);

	/* Store state of explicit locks on rec on the page infimum record,
	before deleting rec. The page infimum acts as a dummy carrier of the
	locks, taking care also of lock releases, before we can move the locks
	back on the actual record. There is a special case: if we are
	inserting on the root page and the insert causes a call of
	btr_root_raise_and_insert. Therefore we cannot in the lock system
	delete the lock structs set on the root page even if the root
	page carries just node pointers. */

	lock_rec_store_on_page_infimum(rec);

	btr_search_update_hash_on_delete(cursor);
1806

1807 1808 1809 1810
	page_cur_delete_rec(page_cursor, mtr);

	page_cur_move_to_prev(page_cursor);

1811
	rec = btr_cur_insert_if_possible(cursor, new_entry,
1812
						&dummy_reorganized, mtr);
1813
	ut_a(rec || optim_err != DB_UNDERFLOW);
1814

1815
	if (rec) {
1816
		lock_rec_restore_from_page_infimum(rec, page);
1817 1818
		rec_set_field_extern_bits(rec, ext_vect, n_ext_vect, mtr);
		
1819 1820 1821 1822 1823 1824 1825
		if (!rec_get_deleted_flag(rec)) {
			/* The new inserted record owns its possible externally
			stored fields */

			btr_cur_unmark_extern_fields(rec, mtr);
		}

1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848
		btr_cur_compress_if_useful(cursor, mtr);

		err = DB_SUCCESS;
		mem_heap_free(heap);

		goto return_after_reservations;
	}

	if (page_cur_is_before_first(page_cursor)) {
		/* The record to be updated was positioned as the first user
		record on its page */

		was_first = TRUE;
	} else {
		was_first = FALSE;
	}

	/* The first parameter means that no lock checking and undo logging
	is made in the insert */

	err = btr_cur_pessimistic_insert(BTR_NO_UNDO_LOG_FLAG
					| BTR_NO_LOCKING_FLAG
					| BTR_KEEP_SYS_FLAG,
1849 1850
					cursor, new_entry, &rec,
					&dummy_big_rec, NULL, mtr);
1851 1852
	ut_a(rec);
	ut_a(err == DB_SUCCESS);
1853 1854 1855
	ut_a(dummy_big_rec == NULL);

	rec_set_field_extern_bits(rec, ext_vect, n_ext_vect, mtr);
1856

1857 1858 1859 1860 1861 1862 1863
	if (!rec_get_deleted_flag(rec)) {
		/* The new inserted record owns its possible externally
		stored fields */

		btr_cur_unmark_extern_fields(rec, mtr);
	}

1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879
	lock_rec_restore_from_page_infimum(rec, page);

	/* If necessary, restore also the correct lock state for a new,
	preceding supremum record created in a page split. While the old
	record was nonexistent, the supremum might have inherited its locks
	from a wrong record. */

	if (!was_first) {
		btr_cur_pess_upd_restore_supremum(rec, mtr);
	}

	mem_heap_free(heap);

return_after_reservations:

	if (n_extents > 0) {
1880
		fil_space_release_free_extents(cursor->index->space,
unknown's avatar
unknown committed
1881
							n_reserved);
1882 1883
	}

1884 1885
	*big_rec = big_rec_vec;

1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970
	return(err);
}

/*==================== B-TREE DELETE MARK AND UNMARK ===============*/

/********************************************************************
Writes the redo log record for delete marking or unmarking of an index
record. */
UNIV_INLINE
void
btr_cur_del_mark_set_clust_rec_log(
/*===============================*/
	ulint		flags,	/* in: flags */
	rec_t*		rec,	/* in: record */
	dict_index_t*	index,	/* in: index of the record */
	ibool		val,	/* in: value to set */
	trx_t*		trx,	/* in: deleting transaction */
	dulint		roll_ptr,/* in: roll ptr to the undo log record */
	mtr_t*		mtr)	/* in: mtr */
{
	byte*	log_ptr;

	log_ptr = mlog_open(mtr, 30);

	log_ptr = mlog_write_initial_log_record_fast(rec,
				MLOG_REC_CLUST_DELETE_MARK, log_ptr, mtr);

	mach_write_to_1(log_ptr, flags);
	log_ptr++;
	mach_write_to_1(log_ptr, val);
	log_ptr++;

	log_ptr = row_upd_write_sys_vals_to_log(index, trx, roll_ptr, log_ptr,
									mtr);
	mach_write_to_2(log_ptr, rec - buf_frame_align(rec));
	log_ptr += 2;

	mlog_close(mtr, log_ptr);
}

/********************************************************************
Parses the redo log record for delete marking or unmarking of a clustered
index record. */

byte*
btr_cur_parse_del_mark_set_clust_rec(
/*=================================*/
			/* out: end of log record or NULL */
	byte*	ptr,	/* in: buffer */
	byte*	end_ptr,/* in: buffer end */
	page_t*	page)	/* in: page or NULL */	
{
	ulint	flags;
	ibool	val;
	ulint	pos;
	dulint	trx_id;
	dulint	roll_ptr;
	ulint	offset;
	rec_t*	rec;

	if (end_ptr < ptr + 2) {

		return(NULL);
	}
	
	flags = mach_read_from_1(ptr);
	ptr++;
	val = mach_read_from_1(ptr);
	ptr++;

	ptr = row_upd_parse_sys_vals(ptr, end_ptr, &pos, &trx_id, &roll_ptr);

	if (ptr == NULL) {

		return(NULL);
	}

	if (end_ptr < ptr + 2) {

		return(NULL);
	}

	offset = mach_read_from_2(ptr);
	ptr += 2;

unknown's avatar
unknown committed
1971 1972
	ut_a(offset <= UNIV_PAGE_SIZE);

1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
	if (page) {
		rec = page + offset;
	
		if (!(flags & BTR_KEEP_SYS_FLAG)) {
			row_upd_rec_sys_fields_in_recovery(rec, pos, trx_id,
								roll_ptr);
		}

		/* We do not need to reserve btr_search_latch, as the page
		is only being recovered, and there cannot be a hash index to
		it. */

		rec_set_deleted_flag(rec, val);
	}
	
	return(ptr);
}

/***************************************************************
Marks a clustered index record deleted. Writes an undo log record to
undo log on this delete marking. Writes in the trx id field the id
of the deleting transaction, and in the roll ptr field pointer to the
undo log record created. */

ulint
btr_cur_del_mark_set_clust_rec(
/*===========================*/
				/* out: DB_SUCCESS, DB_LOCK_WAIT, or error
				number */
	ulint		flags,	/* in: undo logging and locking flags */
	btr_cur_t*	cursor,	/* in: cursor */
	ibool		val,	/* in: value to set */
	que_thr_t*	thr,	/* in: query thread */
	mtr_t*		mtr)	/* in: mtr */
{
	dict_index_t*	index;
	buf_block_t*	block;
	dulint		roll_ptr;
	ulint		err;
	rec_t*		rec;
	trx_t*		trx;
	
	rec = btr_cur_get_rec(cursor);
	index = cursor->index;
	
unknown's avatar
unknown committed
2018 2019 2020
	if (btr_cur_print_record_ops && thr) {
		printf(
	"Trx with id %lu %lu going to del mark table %s index %s\n",
2021 2022
		(unsigned long) ut_dulint_get_high(thr_get_trx(thr)->id),
		(unsigned long) ut_dulint_get_low(thr_get_trx(thr)->id),
unknown's avatar
unknown committed
2023 2024 2025 2026
		index->table_name, index->name);
		rec_print(rec);
	}

2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122
	ut_ad(index->type & DICT_CLUSTERED);
	ut_ad(rec_get_deleted_flag(rec) == FALSE);

	err = lock_clust_rec_modify_check_and_lock(flags, rec, index, thr);

	if (err != DB_SUCCESS) {

		return(err);
	}

	err = trx_undo_report_row_operation(flags, TRX_UNDO_MODIFY_OP, thr,
						index, NULL, NULL, 0, rec,
						&roll_ptr);
	if (err != DB_SUCCESS) {

		return(err);
	}

	block = buf_block_align(rec);

	if (block->is_hashed) {
		rw_lock_x_lock(&btr_search_latch);
	}

	rec_set_deleted_flag(rec, val);

	trx = thr_get_trx(thr);
	
	if (!(flags & BTR_KEEP_SYS_FLAG)) {

		row_upd_rec_sys_fields(rec, index, trx, roll_ptr);
	}
	
	if (block->is_hashed) {
		rw_lock_x_unlock(&btr_search_latch);
	}

	btr_cur_del_mark_set_clust_rec_log(flags, rec, index, val, trx,
							roll_ptr, mtr);
	return(DB_SUCCESS);
}

/********************************************************************
Writes the redo log record for a delete mark setting of a secondary
index record. */
UNIV_INLINE
void
btr_cur_del_mark_set_sec_rec_log(
/*=============================*/
	rec_t*	rec,	/* in: record */
	ibool	val,	/* in: value to set */
	mtr_t*	mtr)	/* in: mtr */
{
	byte*	log_ptr;

	log_ptr = mlog_open(mtr, 30);

	log_ptr = mlog_write_initial_log_record_fast(rec,
				MLOG_REC_SEC_DELETE_MARK, log_ptr, mtr);

	mach_write_to_1(log_ptr, val);
	log_ptr++;

	mach_write_to_2(log_ptr, rec - buf_frame_align(rec));
	log_ptr += 2;

	mlog_close(mtr, log_ptr);
}

/********************************************************************
Parses the redo log record for delete marking or unmarking of a secondary
index record. */

byte*
btr_cur_parse_del_mark_set_sec_rec(
/*===============================*/
			/* out: end of log record or NULL */
	byte*	ptr,	/* in: buffer */
	byte*	end_ptr,/* in: buffer end */
	page_t*	page)	/* in: page or NULL */	
{
	ibool	val;
	ulint	offset;
	rec_t*	rec;

	if (end_ptr < ptr + 3) {

		return(NULL);
	}
	
	val = mach_read_from_1(ptr);
	ptr++;

	offset = mach_read_from_2(ptr);
	ptr += 2;

unknown's avatar
unknown committed
2123 2124
	ut_a(offset <= UNIV_PAGE_SIZE);

2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157
	if (page) {
		rec = page + offset;
	
		/* We do not need to reserve btr_search_latch, as the page
		is only being recovered, and there cannot be a hash index to
		it. */

		rec_set_deleted_flag(rec, val);
	}
	
	return(ptr);
}
	
/***************************************************************
Sets a secondary index record delete mark to TRUE or FALSE. */

ulint
btr_cur_del_mark_set_sec_rec(
/*=========================*/
				/* out: DB_SUCCESS, DB_LOCK_WAIT, or error
				number */
	ulint		flags,	/* in: locking flag */
	btr_cur_t*	cursor,	/* in: cursor */
	ibool		val,	/* in: value to set */
	que_thr_t*	thr,	/* in: query thread */
	mtr_t*		mtr)	/* in: mtr */
{
	buf_block_t*	block;
	rec_t*		rec;
	ulint		err;

	rec = btr_cur_get_rec(cursor);

unknown's avatar
unknown committed
2158 2159 2160
	if (btr_cur_print_record_ops && thr) {
		printf(
	"Trx with id %lu %lu going to del mark table %s index %s\n",
2161 2162
		(unsigned long) ut_dulint_get_high(thr_get_trx(thr)->id),
		(unsigned long) ut_dulint_get_low(thr_get_trx(thr)->id),
unknown's avatar
unknown committed
2163 2164 2165 2166
		cursor->index->table_name, cursor->index->name);
		rec_print(rec);
	}

2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295
	err = lock_sec_rec_modify_check_and_lock(flags, rec, cursor->index,
									thr);
	if (err != DB_SUCCESS) {

		return(err);
	}

	block = buf_block_align(rec);
	
	if (block->is_hashed) {
		rw_lock_x_lock(&btr_search_latch);
	}

	rec_set_deleted_flag(rec, val);

	if (block->is_hashed) {
		rw_lock_x_unlock(&btr_search_latch);
	}

	btr_cur_del_mark_set_sec_rec_log(rec, val, mtr);

	return(DB_SUCCESS);
}

/***************************************************************
Sets a secondary index record delete mark to FALSE. This function is only
used by the insert buffer insert merge mechanism. */

void
btr_cur_del_unmark_for_ibuf(
/*========================*/
	rec_t*	rec,	/* in: record to delete unmark */
	mtr_t*	mtr)	/* in: mtr */
{
	/* We do not need to reserve btr_search_latch, as the page has just
	been read to the buffer pool and there cannot be a hash index to it. */

	rec_set_deleted_flag(rec, FALSE);

	btr_cur_del_mark_set_sec_rec_log(rec, FALSE, mtr);
}

/*==================== B-TREE RECORD REMOVE =========================*/

/*****************************************************************
Tries to compress a page of the tree on the leaf level. It is assumed
that mtr holds an x-latch on the tree and on the cursor page. To avoid
deadlocks, mtr must also own x-latches to brothers of page, if those
brothers exist. NOTE: it is assumed that the caller has reserved enough
free extents so that the compression will always succeed if done! */

void
btr_cur_compress(
/*=============*/
	btr_cur_t*	cursor,	/* in: cursor on the page to compress;
				cursor does not stay valid */
	mtr_t*		mtr)	/* in: mtr */
{
	ut_ad(mtr_memo_contains(mtr,
				dict_tree_get_lock(btr_cur_get_tree(cursor)),
							MTR_MEMO_X_LOCK));
	ut_ad(mtr_memo_contains(mtr, buf_block_align(
						btr_cur_get_page(cursor)),
				MTR_MEMO_PAGE_X_FIX));
	ut_ad(btr_page_get_level(btr_cur_get_page(cursor), mtr) == 0);

	btr_compress(cursor, mtr);	
}

/*****************************************************************
Tries to compress a page of the tree if it seems useful. It is assumed
that mtr holds an x-latch on the tree and on the cursor page. To avoid
deadlocks, mtr must also own x-latches to brothers of page, if those
brothers exist. NOTE: it is assumed that the caller has reserved enough
free extents so that the compression will always succeed if done! */

ibool
btr_cur_compress_if_useful(
/*=======================*/
				/* out: TRUE if compression occurred */
	btr_cur_t*	cursor,	/* in: cursor on the page to compress;
				cursor does not stay valid if compression
				occurs */
	mtr_t*		mtr)	/* in: mtr */
{
	ut_ad(mtr_memo_contains(mtr,
				dict_tree_get_lock(btr_cur_get_tree(cursor)),
							MTR_MEMO_X_LOCK));
	ut_ad(mtr_memo_contains(mtr, buf_block_align(
						btr_cur_get_page(cursor)),
				MTR_MEMO_PAGE_X_FIX));

	if (btr_cur_compress_recommendation(cursor, mtr)) {

		btr_compress(cursor, mtr);

		return(TRUE);
	}

	return(FALSE);
}

/***********************************************************
Removes the record on which the tree cursor is positioned on a leaf page.
It is assumed that the mtr has an x-latch on the page where the cursor is
positioned, but no latch on the whole tree. */

ibool
btr_cur_optimistic_delete(
/*======================*/
				/* out: TRUE if success, i.e., the page
				did not become too empty */
	btr_cur_t*	cursor,	/* in: cursor on leaf page, on the record to
				delete; cursor stays valid: if deletion
				succeeds, on function exit it points to the
				successor of the deleted record */
	mtr_t*		mtr)	/* in: mtr */
{
	page_t*	page;
	ulint	max_ins_size;

	ut_ad(mtr_memo_contains(mtr, buf_block_align(btr_cur_get_page(cursor)),
							MTR_MEMO_PAGE_X_FIX));
	/* This is intended only for leaf page deletions */

	page = btr_cur_get_page(cursor);
	
	ut_ad(btr_page_get_level(page, mtr) == 0);

2296 2297 2298 2299 2300
	if (rec_contains_externally_stored_field(btr_cur_get_rec(cursor))) {

		return(FALSE);
	}

2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343
	if (btr_cur_can_delete_without_compress(cursor, mtr)) {

		lock_update_delete(btr_cur_get_rec(cursor));

		btr_search_update_hash_on_delete(cursor);

		max_ins_size = page_get_max_insert_size_after_reorganize(page,
									1);
		page_cur_delete_rec(btr_cur_get_page_cur(cursor), mtr);

		ibuf_update_free_bits_low(cursor->index, page, max_ins_size,
									mtr);
		return(TRUE);
	}

	return(FALSE);
}

/*****************************************************************
Removes the record on which the tree cursor is positioned. Tries
to compress the page if its fillfactor drops below a threshold
or if it is the only page on the level. It is assumed that mtr holds
an x-latch on the tree and on the cursor page. To avoid deadlocks,
mtr must also own x-latches to brothers of page, if those brothers
exist. */

ibool
btr_cur_pessimistic_delete(
/*=======================*/
				/* out: TRUE if compression occurred */
	ulint*		err,	/* out: DB_SUCCESS or DB_OUT_OF_FILE_SPACE;
				the latter may occur because we may have
				to update node pointers on upper levels,
				and in the case of variable length keys
				these may actually grow in size */
	ibool		has_reserved_extents, /* in: TRUE if the
				caller has already reserved enough free
				extents so that he knows that the operation
				will succeed */
	btr_cur_t*	cursor,	/* in: cursor on the record to delete;
				if compression does not occur, the cursor
				stays valid: it points to successor of
				deleted record on function exit */
2344
	ibool		in_rollback,/* in: TRUE if called in rollback */
2345 2346 2347 2348 2349 2350 2351
	mtr_t*		mtr)	/* in: mtr */
{
	page_t*		page;
	dict_tree_t*	tree;
	rec_t*		rec;
	dtuple_t*	node_ptr;
	ulint		n_extents	= 0;
unknown's avatar
unknown committed
2352
	ulint		n_reserved;
2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370
	ibool		success;
	ibool		ret		= FALSE;
	mem_heap_t*	heap;
	
	page = btr_cur_get_page(cursor);
	tree = btr_cur_get_tree(cursor);

	ut_ad(mtr_memo_contains(mtr, dict_tree_get_lock(tree),
							MTR_MEMO_X_LOCK));
	ut_ad(mtr_memo_contains(mtr, buf_block_align(page),
							MTR_MEMO_PAGE_X_FIX));
	if (!has_reserved_extents) {
		/* First reserve enough free space for the file segments
		of the index tree, so that the node pointer updates will
		not fail because of lack of space */

		n_extents = cursor->tree_height / 32 + 1;

unknown's avatar
unknown committed
2371 2372
		success = fsp_reserve_free_extents(&n_reserved,
						cursor->index->space,
2373 2374 2375 2376 2377 2378 2379 2380
						n_extents, FSP_CLEANING, mtr);
		if (!success) {
			*err = DB_OUT_OF_FILE_SPACE;

			return(FALSE);
		}
	}

2381
	btr_rec_free_externally_stored_fields(cursor->index,
2382 2383
			btr_cur_get_rec(cursor), in_rollback, mtr);

2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423
	if ((page_get_n_recs(page) < 2)
	    && (dict_tree_get_page(btr_cur_get_tree(cursor))
					!= buf_frame_get_page_no(page))) {

		/* If there is only one record, drop the whole page in
		btr_discard_page, if this is not the root page */
	
		btr_discard_page(cursor, mtr);

		*err = DB_SUCCESS;
		ret = TRUE;

		goto return_after_reservations;	
	}

	rec = btr_cur_get_rec(cursor);
	
	lock_update_delete(rec);

	if ((btr_page_get_level(page, mtr) > 0)
	    && (page_rec_get_next(page_get_infimum_rec(page)) == rec)) {

		if (btr_page_get_prev(page, mtr) == FIL_NULL) {

			/* If we delete the leftmost node pointer on a
			non-leaf level, we must mark the new leftmost node
			pointer as the predefined minimum record */

	    		btr_set_min_rec_mark(page_rec_get_next(rec), mtr);
		} else {
			/* Otherwise, if we delete the leftmost node pointer
			on a page, we have to change the father node pointer
			so that it is equal to the new leftmost node pointer
			on the page */

			btr_node_ptr_delete(tree, page, mtr);

			heap = mem_heap_create(256);

			node_ptr = dict_tree_build_node_ptr(
2424 2425 2426
					tree, page_rec_get_next(rec),
					buf_frame_get_page_no(page),
       					heap, btr_page_get_level(page, mtr));
2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450

			btr_insert_on_non_leaf_level(tree,
					btr_page_get_level(page, mtr) + 1,
					node_ptr, mtr);

			mem_heap_free(heap);
		}
	} 

	btr_search_update_hash_on_delete(cursor);

	page_cur_delete_rec(btr_cur_get_page_cur(cursor), mtr);

	ut_ad(btr_check_node_ptr(tree, page, mtr));

	*err = DB_SUCCESS;
	
return_after_reservations:

	if (ret == FALSE) {
		ret = btr_cur_compress_if_useful(cursor, mtr);
	}

	if (n_extents > 0) {
unknown's avatar
unknown committed
2451 2452
		fil_space_release_free_extents(cursor->index->space,
								n_reserved);
2453 2454
	}

2455
	return(ret);
2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500
}

/***********************************************************************
Adds path information to the cursor for the current page, for which
the binary search has been performed. */
static
void
btr_cur_add_path_info(
/*==================*/
	btr_cur_t*	cursor,		/* in: cursor positioned on a page */
	ulint		height,		/* in: height of the page in tree;
					0 means leaf node */
	ulint		root_height)	/* in: root node height in tree */
{
	btr_path_t*	slot;
	rec_t*		rec;

	ut_a(cursor->path_arr);

	if (root_height >= BTR_PATH_ARRAY_N_SLOTS - 1) {
		/* Do nothing; return empty path */

		slot = cursor->path_arr;
		slot->nth_rec = ULINT_UNDEFINED;

		return;
	}

	if (height == 0) {
		/* Mark end of slots for path */
		slot = cursor->path_arr + root_height + 1;
		slot->nth_rec = ULINT_UNDEFINED;
	}

	rec = btr_cur_get_rec(cursor);
	
	slot = cursor->path_arr + (root_height - height);

	slot->nth_rec = page_rec_get_n_recs_before(rec);
	slot->n_recs = page_get_n_recs(buf_frame_align(rec));
}

/***********************************************************************
Estimates the number of rows in a given index range. */

unknown's avatar
unknown committed
2501
ib_longlong
2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516
btr_estimate_n_rows_in_range(
/*=========================*/
				/* out: estimated number of rows */
	dict_index_t*	index,	/* in: index */
	dtuple_t*	tuple1,	/* in: range start, may also be empty tuple */
	ulint		mode1,	/* in: search mode for range start */
	dtuple_t*	tuple2,	/* in: range end, may also be empty tuple */
	ulint		mode2)	/* in: search mode for range end */
{
	btr_path_t	path1[BTR_PATH_ARRAY_N_SLOTS];
	btr_path_t	path2[BTR_PATH_ARRAY_N_SLOTS];
	btr_cur_t	cursor;
	btr_path_t*	slot1;
	btr_path_t*	slot2;
	ibool		diverged;
unknown's avatar
unknown committed
2517
	ibool           diverged_lot;
2518
	ulint           divergence_level;           
unknown's avatar
unknown committed
2519
	ib_longlong	n_rows;
2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559
	ulint		i;
	mtr_t		mtr;

	mtr_start(&mtr);

	cursor.path_arr = path1;

	if (dtuple_get_n_fields(tuple1) > 0) {
	
		btr_cur_search_to_nth_level(index, 0, tuple1, mode1,
					BTR_SEARCH_LEAF	| BTR_ESTIMATE,
					&cursor, 0, &mtr);
	} else {
		btr_cur_open_at_index_side(TRUE, index,
					BTR_SEARCH_LEAF	| BTR_ESTIMATE,
					&cursor, &mtr);
	}
	
	mtr_commit(&mtr);

	mtr_start(&mtr);

	cursor.path_arr = path2;

	if (dtuple_get_n_fields(tuple2) > 0) {
	
		btr_cur_search_to_nth_level(index, 0, tuple2, mode2,
					BTR_SEARCH_LEAF	| BTR_ESTIMATE,
					&cursor, 0, &mtr);
	} else {
		btr_cur_open_at_index_side(FALSE, index,
					BTR_SEARCH_LEAF	| BTR_ESTIMATE,
					&cursor, &mtr);
	}
		
	mtr_commit(&mtr);

	/* We have the path information for the range in path1 and path2 */

	n_rows = 1;
unknown's avatar
unknown committed
2560 2561 2562 2563 2564 2565 2566
	diverged = FALSE;           /* This becomes true when the path is not
				    the same any more */
	diverged_lot = FALSE;       /* This becomes true when the paths are
				    not the same or adjacent any more */
	divergence_level = 1000000; /* This is the level where paths diverged
				    a lot */ 
       	for (i = 0; ; i++) {
2567 2568 2569 2570 2571 2572 2573 2574
		ut_ad(i < BTR_PATH_ARRAY_N_SLOTS);
	
		slot1 = path1 + i;
		slot2 = path2 + i;

		if (slot1->nth_rec == ULINT_UNDEFINED
				|| slot2->nth_rec == ULINT_UNDEFINED) {

2575 2576 2577 2578 2579 2580 2581
		        if (i > divergence_level + 1) {
		                /* In trees whose height is > 1 our algorithm
		                tends to underestimate: multiply the estimate
		                by 2: */

		                n_rows = n_rows * 2;
		        }
unknown's avatar
unknown committed
2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593

			/* Do not estimate the number of rows in the range
		        to over 1 / 2 of the estimated rows in the whole
			table */

			if (n_rows > index->table->stat_n_rows / 2) {
			        n_rows = index->table->stat_n_rows / 2;

				/* If there are just 0 or 1 rows in the table,
				then we estimate all rows are in the range */
			  
			        if (n_rows == 0) {
unknown's avatar
unknown committed
2594
				        n_rows = index->table->stat_n_rows;
unknown's avatar
unknown committed
2595 2596 2597
			        }
			}

2598 2599 2600 2601 2602
			return(n_rows);
		}

		if (!diverged && slot1->nth_rec != slot2->nth_rec) {

unknown's avatar
unknown committed
2603 2604
			diverged = TRUE;

2605 2606
			if (slot1->nth_rec < slot2->nth_rec) {
				n_rows = slot2->nth_rec - slot1->nth_rec;
unknown's avatar
unknown committed
2607 2608 2609 2610 2611

				if (n_rows > 1) {
				          diverged_lot = TRUE;
					  divergence_level = i;
				}
2612 2613 2614 2615 2616 2617 2618
			} else {
				/* Maybe the tree has changed between
				searches */

				return(10);
			}

unknown's avatar
unknown committed
2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638
		} else if (diverged && !diverged_lot) {

		        if (slot1->nth_rec < slot1->n_recs
		            || slot2->nth_rec > 1) {

		                diverged_lot = TRUE;
				divergence_level = i;

				n_rows = 0;

		                if (slot1->nth_rec < slot1->n_recs) {
				        n_rows += slot1->n_recs
					             - slot1->nth_rec;
				}

				if (slot2->nth_rec > 1) {
				        n_rows += slot2->nth_rec - 1;
				}
      			}
		} else if (diverged_lot) {
2639

2640 2641 2642 2643 2644 2645 2646
			n_rows = (n_rows * (slot1->n_recs + slot2->n_recs))
									/ 2;
		}	
	}
}

/***********************************************************************
2647 2648 2649
Estimates the number of different key values in a given index, for
each n-column prefix of the index where n <= dict_index_get_n_unique(index).
The estimates are stored in the array index->stat_n_diff_key_vals. */
2650

2651
void
2652 2653 2654 2655 2656 2657 2658
btr_estimate_number_of_different_key_vals(
/*======================================*/
	dict_index_t*	index)	/* in: index */
{
	btr_cur_t	cursor;
	page_t*		page;
	rec_t*		rec;
2659
	ulint		n_cols;
2660 2661
	ulint		matched_fields;
	ulint		matched_bytes;
unknown's avatar
unknown committed
2662
	ib_longlong*	n_diff;
2663
	ulint		not_empty_flag	= 0;
unknown's avatar
unknown committed
2664
	ulint		total_external_size = 0;
2665
	ulint		i;
2666
	ulint		j;
unknown's avatar
unknown committed
2667
	ulint		add_on;
2668 2669
	mtr_t		mtr;

2670 2671 2672 2673 2674 2675
	n_cols = dict_index_get_n_unique(index);

	n_diff = mem_alloc((n_cols + 1) * sizeof(ib_longlong));

	for (j = 0; j <= n_cols; j++) {
		n_diff[j] = 0;
2676 2677 2678 2679 2680 2681 2682 2683 2684
	}

	/* We sample some pages in the index to get an estimate */
	
	for (i = 0; i < BTR_KEY_VAL_ESTIMATE_N_PAGES; i++) {
		mtr_start(&mtr);

		btr_cur_open_at_rnd_pos(index, BTR_SEARCH_LEAF, &cursor, &mtr);
		
unknown's avatar
unknown committed
2685 2686 2687 2688 2689
		/* Count the number of different key values for each prefix of
		the key on this index page. If the prefix does not determine
		the index record uniquely in te B-tree, then we subtract one
		because otherwise our algorithm would give a wrong estimate
		for an index where there is just one key value. */
2690 2691 2692 2693 2694 2695

		page = btr_cur_get_page(&cursor);

		rec = page_get_infimum_rec(page);
		rec = page_rec_get_next(rec);

2696 2697 2698
		if (rec != page_get_supremum_rec(page)) {
			not_empty_flag = 1;
		}
2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709
		
		while (rec != page_get_supremum_rec(page)
		       && page_rec_get_next(rec)
					!= page_get_supremum_rec(page)) {
			matched_fields = 0;
			matched_bytes = 0;

			cmp_rec_rec_with_match(rec, page_rec_get_next(rec),
						index, &matched_fields,
						&matched_bytes);

2710
			for (j = matched_fields + 1; j <= n_cols; j++) {
unknown's avatar
unknown committed
2711 2712 2713
				/* We add one if this index record has
				a different prefix from the previous */

2714 2715
				n_diff[j]++;
			}
unknown's avatar
unknown committed
2716 2717 2718 2719

			total_external_size +=
				btr_rec_get_externally_stored_len(rec);
			
2720 2721 2722
			rec = page_rec_get_next(rec);
		}
		
unknown's avatar
unknown committed
2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734
		if (n_cols == dict_index_get_n_unique_in_tree(index)) {
			/* We add one because we know that the first record
			on the page certainly had a different prefix than the
			last record on the previous index page in the
			alphabetical order. Before this fix, if there was
			just one big record on each clustered index page, the
			algorithm grossly underestimated the number of rows
			in the table. */

			n_diff[n_cols]++;
		}

unknown's avatar
unknown committed
2735 2736
		total_external_size +=
				btr_rec_get_externally_stored_len(rec);
2737 2738 2739
		mtr_commit(&mtr);
	}

2740 2741 2742 2743
	/* If we saw k borders between different key values on
	BTR_KEY_VAL_ESTIMATE_N_PAGES leaf pages, we can estimate how many
	there will be in index->stat_n_leaf_pages */
	
unknown's avatar
unknown committed
2744 2745 2746 2747
	/* We must take into account that our sample actually represents
	also the pages used for external storage of fields (those pages are
	included in index->stat_n_leaf_pages) */ 

2748 2749
	for (j = 0; j <= n_cols; j++) {
		index->stat_n_diff_key_vals[j] =
unknown's avatar
unknown committed
2750 2751
				(n_diff[j]
				 * (ib_longlong)index->stat_n_leaf_pages
2752
				 + BTR_KEY_VAL_ESTIMATE_N_PAGES - 1
unknown's avatar
unknown committed
2753
				 + total_external_size
2754
				 + not_empty_flag)
unknown's avatar
unknown committed
2755 2756
		                	/ (BTR_KEY_VAL_ESTIMATE_N_PAGES
		                	   + total_external_size);
2757
	
unknown's avatar
unknown committed
2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775
		/* If the tree is small, smaller than <
		10 * BTR_KEY_VAL_ESTIMATE_N_PAGES + total_external_size, then
		the above estimate is ok. For bigger trees it is common that we
		do not see any borders between key values in the few pages
		we pick. But still there may be BTR_KEY_VAL_ESTIMATE_N_PAGES
		different key values, or even more. Let us try to approximate
		that: */

		add_on = index->stat_n_leaf_pages /
		   (10 * (BTR_KEY_VAL_ESTIMATE_N_PAGES + total_external_size));

		if (add_on > BTR_KEY_VAL_ESTIMATE_N_PAGES) {
			add_on = BTR_KEY_VAL_ESTIMATE_N_PAGES;
		}
		
		index->stat_n_diff_key_vals[j] += add_on;
	}
		
2776
	mem_free(n_diff);
2777
}
2778 2779 2780

/*================== EXTERNAL STORAGE OF BIG FIELDS ===================*/

unknown's avatar
unknown committed
2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822
/***************************************************************
Gets the externally stored size of a record, in units of a database page. */
static
ulint
btr_rec_get_externally_stored_len(
/*==============================*/
			/* out: externally stored part, in units of a
			database page */
	rec_t*	rec)	/* in: record */
{
	ulint	n_fields;
	byte*	data;
	ulint	local_len;
	ulint	extern_len;
	ulint	total_extern_len = 0;
	ulint	i;

	if (rec_get_data_size(rec) <= REC_1BYTE_OFFS_LIMIT) {

		return(0);
	}
	
	n_fields = rec_get_n_fields(rec);

	for (i = 0; i < n_fields; i++) {
		if (rec_get_nth_field_extern_bit(rec, i)) {

			data = rec_get_nth_field(rec, i, &local_len);

			local_len -= BTR_EXTERN_FIELD_REF_SIZE;
	
			extern_len = mach_read_from_4(data + local_len
						+ BTR_EXTERN_LEN + 4);

			total_extern_len += ut_calc_align(extern_len,
							UNIV_PAGE_SIZE);
		}
	}

	return(total_extern_len / UNIV_PAGE_SIZE);
}

2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961
/***********************************************************************
Sets the ownership bit of an externally stored field in a record. */
static
void
btr_cur_set_ownership_of_extern_field(
/*==================================*/
	rec_t*	rec,	/* in: clustered index record */
	ulint	i,	/* in: field number */
	ibool	val,	/* in: value to set */
	mtr_t*	mtr)	/* in: mtr */
{
	byte*	data;
	ulint	local_len;
	ulint	byte_val;

	data = rec_get_nth_field(rec, i, &local_len);
	
	ut_a(local_len >= BTR_EXTERN_FIELD_REF_SIZE);

	local_len -= BTR_EXTERN_FIELD_REF_SIZE;

	byte_val = mach_read_from_1(data + local_len + BTR_EXTERN_LEN);

	if (val) {
		byte_val = byte_val & (~BTR_EXTERN_OWNER_FLAG);
	} else {
		byte_val = byte_val | BTR_EXTERN_OWNER_FLAG;
	}
	
	mlog_write_ulint(data + local_len + BTR_EXTERN_LEN, byte_val,
							MLOG_1BYTE, mtr);
}

/***********************************************************************
Marks not updated extern fields as not-owned by this record. The ownership
is transferred to the updated record which is inserted elsewhere in the
index tree. In purge only the owner of externally stored field is allowed
to free the field. */

void
btr_cur_mark_extern_inherited_fields(
/*=================================*/
	rec_t*	rec,	/* in: record in a clustered index */
	upd_t*	update,	/* in: update vector */
	mtr_t*	mtr)	/* in: mtr */
{
	ibool	is_updated;
	ulint	n;
	ulint	j;
	ulint	i;
	
	n = rec_get_n_fields(rec);

	for (i = 0; i < n; i++) {
		if (rec_get_nth_field_extern_bit(rec, i)) {
			
			/* Check it is not in updated fields */
			is_updated = FALSE;

			if (update) {
				for (j = 0; j < upd_get_n_fields(update);
								j++) {
					if (upd_get_nth_field(update, j)
							->field_no == i) {
						is_updated = TRUE;
					}
				}
			}

			if (!is_updated) {
				btr_cur_set_ownership_of_extern_field(rec, i,
								FALSE, mtr);
			}
		}
	}
}

/***********************************************************************
The complement of the previous function: in an update entry may inherit
some externally stored fields from a record. We must mark them as inherited
in entry, so that they are not freed in a rollback. */

void
btr_cur_mark_dtuple_inherited_extern(
/*=================================*/
	dtuple_t*	entry,		/* in: updated entry to be inserted to
					clustered index */
	ulint*		ext_vec,	/* in: array of extern fields in the
					original record */
	ulint		n_ext_vec,	/* in: number of elements in ext_vec */
	upd_t*		update)		/* in: update vector */
{
	dfield_t* dfield;
	ulint	byte_val;
	byte*	data;
	ulint	len;
	ibool	is_updated;
	ulint	j;
	ulint	i;

	if (ext_vec == NULL) {

		return;
	}
	
	for (i = 0; i < n_ext_vec; i++) {

		/* Check ext_vec[i] is in updated fields */
		is_updated = FALSE;

		for (j = 0; j < upd_get_n_fields(update); j++) {
			if (upd_get_nth_field(update, j)->field_no
							== ext_vec[i]) {
				is_updated = TRUE;
			}
		}

		if (!is_updated) {
			dfield = dtuple_get_nth_field(entry, ext_vec[i]);

			data = dfield_get_data(dfield);
			len = dfield_get_len(dfield);
		
			len -= BTR_EXTERN_FIELD_REF_SIZE;

			byte_val = mach_read_from_1(data + len
							+ BTR_EXTERN_LEN);

			byte_val = byte_val | BTR_EXTERN_INHERITED_FLAG;
		
			mach_write_to_1(data + len + BTR_EXTERN_LEN, byte_val);
		}
	}
}

/***********************************************************************
Marks all extern fields in a record as owned by the record. This function
should be called if the delete mark of a record is removed: a not delete
marked record always owns all its extern fields. */
2962
static
2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015
void
btr_cur_unmark_extern_fields(
/*=========================*/
	rec_t*	rec,	/* in: record in a clustered index */
	mtr_t*	mtr)	/* in: mtr */
{
	ulint	n;
	ulint	i;

	n = rec_get_n_fields(rec);

	for (i = 0; i < n; i++) {
		if (rec_get_nth_field_extern_bit(rec, i)) {
			
			btr_cur_set_ownership_of_extern_field(rec, i,
								TRUE, mtr);
		}
	}	
}

/***********************************************************************
Marks all extern fields in a dtuple as owned by the record. */

void
btr_cur_unmark_dtuple_extern_fields(
/*================================*/
	dtuple_t*	entry,		/* in: clustered index entry */
	ulint*		ext_vec,	/* in: array of numbers of fields
					which have been stored externally */
	ulint		n_ext_vec)	/* in: number of elements in ext_vec */
{
	dfield_t* dfield;
	ulint	byte_val;
	byte*	data;
	ulint	len;
	ulint	i;

	for (i = 0; i < n_ext_vec; i++) {
		dfield = dtuple_get_nth_field(entry, ext_vec[i]);

		data = dfield_get_data(dfield);
		len = dfield_get_len(dfield);
		
		len -= BTR_EXTERN_FIELD_REF_SIZE;

		byte_val = mach_read_from_1(data + len + BTR_EXTERN_LEN);

		byte_val = byte_val & (~BTR_EXTERN_OWNER_FLAG);
		
		mach_write_to_1(data + len + BTR_EXTERN_LEN, byte_val);
	}	
}

3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119
/***********************************************************************
Stores the positions of the fields marked as extern storage in the update
vector, and also those fields who are marked as extern storage in rec
and not mentioned in updated fields. We use this function to remember
which fields we must mark as extern storage in a record inserted for an
update. */

ulint
btr_push_update_extern_fields(
/*==========================*/
				/* out: number of values stored in ext_vect */
	ulint*	ext_vect,	/* in: array of ulints, must be preallocated
				to have space for all fields in rec */
	rec_t*	rec,		/* in: record */
	upd_t*	update)		/* in: update vector or NULL */
{
	ulint	n_pushed	= 0;
	ibool	is_updated;
	ulint	n;
	ulint	j;
	ulint	i;

	if (update) {
		n = upd_get_n_fields(update);
	
		for (i = 0; i < n; i++) {

			if (upd_get_nth_field(update, i)->extern_storage) {

				ext_vect[n_pushed] =
					upd_get_nth_field(update, i)->field_no;

				n_pushed++;
			}
		}
	}

	n = rec_get_n_fields(rec);

	for (i = 0; i < n; i++) {
		if (rec_get_nth_field_extern_bit(rec, i)) {
			
			/* Check it is not in updated fields */
			is_updated = FALSE;

			if (update) {
				for (j = 0; j < upd_get_n_fields(update);
								j++) {
					if (upd_get_nth_field(update, j)
							->field_no == i) {
						is_updated = TRUE;
					}
				}
			}

			if (!is_updated) {
				ext_vect[n_pushed] = i;
				n_pushed++;
			}
		}
	}		

	return(n_pushed);
}

/***********************************************************************
Returns the length of a BLOB part stored on the header page. */
static
ulint
btr_blob_get_part_len(
/*==================*/
				/* out: part length */
	byte*	blob_header)	/* in: blob header */
{
	return(mach_read_from_4(blob_header + BTR_BLOB_HDR_PART_LEN));
}

/***********************************************************************
Returns the page number where the next BLOB part is stored. */
static
ulint
btr_blob_get_next_page_no(
/*======================*/
				/* out: page number or FIL_NULL if
				no more pages */
	byte*	blob_header)	/* in: blob header */
{
	return(mach_read_from_4(blob_header + BTR_BLOB_HDR_NEXT_PAGE_NO));
}

/***********************************************************************
Stores the fields in big_rec_vec to the tablespace and puts pointers to
them in rec. The fields are stored on pages allocated from leaf node
file segment of the index tree. */

ulint
btr_store_big_rec_extern_fields(
/*============================*/
					/* out: DB_SUCCESS or error */
	dict_index_t*	index,		/* in: index of rec; the index tree
					MUST be X-latched */
	rec_t*		rec,		/* in: record */
	big_rec_t*	big_rec_vec,	/* in: vector containing fields
					to be stored externally */
unknown's avatar
unknown committed
3120 3121 3122
	mtr_t*		local_mtr __attribute__((unused))) /* in: mtr
                                        containing the latch to rec and to the
                                        tree */
3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139
{
	byte*	data;
	ulint	local_len;
	ulint	extern_len;
	ulint	store_len;
	ulint	page_no;
	page_t*	page;
	ulint	space_id;
	page_t*	prev_page;
	page_t*	rec_page;
	ulint	prev_page_no;
	ulint	hint_page_no;
	ulint	i;
	mtr_t	mtr;

	ut_ad(mtr_memo_contains(local_mtr, dict_tree_get_lock(index->tree),
							MTR_MEMO_X_LOCK));
3140
	ut_ad(mtr_memo_contains(local_mtr, buf_block_align(rec),
unknown's avatar
unknown committed
3141
							MTR_MEMO_PAGE_X_FIX));
3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185
	ut_a(index->type & DICT_CLUSTERED);
							
	space_id = buf_frame_get_space_id(rec);
	
	/* We have to create a file segment to the tablespace
	for each field and put the pointer to the field in rec */

	for (i = 0; i < big_rec_vec->n_fields; i++) {

		data = rec_get_nth_field(rec, big_rec_vec->fields[i].field_no,
								&local_len);
		ut_a(local_len >= BTR_EXTERN_FIELD_REF_SIZE);
		local_len -= BTR_EXTERN_FIELD_REF_SIZE;
		extern_len = big_rec_vec->fields[i].len;

		ut_a(extern_len > 0);

		prev_page_no = FIL_NULL;

		while (extern_len > 0) {
			mtr_start(&mtr);

			if (prev_page_no == FIL_NULL) {
				hint_page_no = buf_frame_get_page_no(rec) + 1;
			} else {
				hint_page_no = prev_page_no + 1;
			}
			
			page = btr_page_alloc(index->tree, hint_page_no,
						FSP_NO_DIR, 0, &mtr);
			if (page == NULL) {

				mtr_commit(&mtr);

				return(DB_OUT_OF_FILE_SPACE);
			}

			page_no = buf_frame_get_page_no(page);

			if (prev_page_no != FIL_NULL) {
				prev_page = buf_page_get(space_id,
						prev_page_no,
						RW_X_LATCH, &mtr);

3186
#ifdef UNIV_SYNC_DEBUG
3187 3188
				buf_page_dbg_add_level(prev_page,
							SYNC_EXTERN_STORAGE);
3189
#endif /* UNIV_SYNC_DEBUG */
3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223
							
				mlog_write_ulint(prev_page + FIL_PAGE_DATA
						+ BTR_BLOB_HDR_NEXT_PAGE_NO,
						page_no, MLOG_4BYTES, &mtr);
			}

			if (extern_len > (UNIV_PAGE_SIZE - FIL_PAGE_DATA
						- BTR_BLOB_HDR_SIZE
						- FIL_PAGE_DATA_END)) {
				store_len = UNIV_PAGE_SIZE - FIL_PAGE_DATA
						- BTR_BLOB_HDR_SIZE
						- FIL_PAGE_DATA_END;
			} else {
				store_len = extern_len;
			}

			mlog_write_string(page + FIL_PAGE_DATA
						+ BTR_BLOB_HDR_SIZE,
					big_rec_vec->fields[i].data
						+ big_rec_vec->fields[i].len
						- extern_len,
					store_len, &mtr);
			mlog_write_ulint(page + FIL_PAGE_DATA
						+ BTR_BLOB_HDR_PART_LEN,
					store_len, MLOG_4BYTES, &mtr);
			mlog_write_ulint(page + FIL_PAGE_DATA
						+ BTR_BLOB_HDR_NEXT_PAGE_NO,
					FIL_NULL, MLOG_4BYTES, &mtr);
					
			extern_len -= store_len;

			rec_page = buf_page_get(space_id,
						buf_frame_get_page_no(data),
							RW_X_LATCH, &mtr);
3224
#ifdef UNIV_SYNC_DEBUG
3225
			buf_page_dbg_add_level(rec_page, SYNC_NO_ORDER_CHECK);
3226
#endif /* UNIV_SYNC_DEBUG */
3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268
			mlog_write_ulint(data + local_len + BTR_EXTERN_LEN, 0,
						MLOG_4BYTES, &mtr);
			mlog_write_ulint(data + local_len + BTR_EXTERN_LEN + 4,
					big_rec_vec->fields[i].len
								- extern_len,
					MLOG_4BYTES, &mtr);

			if (prev_page_no == FIL_NULL) {
				mlog_write_ulint(data + local_len
							+ BTR_EXTERN_SPACE_ID,
						space_id,
						MLOG_4BYTES, &mtr);

				mlog_write_ulint(data + local_len
							+ BTR_EXTERN_PAGE_NO,
						page_no,
						MLOG_4BYTES, &mtr);
				
				mlog_write_ulint(data + local_len
							+ BTR_EXTERN_OFFSET,
						FIL_PAGE_DATA,
						MLOG_4BYTES, &mtr);

				/* Set the bit denoting that this field
				in rec is stored externally */

				rec_set_nth_field_extern_bit(rec,
					big_rec_vec->fields[i].field_no,
					TRUE, &mtr);
			}

			prev_page_no = page_no;

			mtr_commit(&mtr);
		}
	}

	return(DB_SUCCESS);
}

/***********************************************************************
Frees the space in an externally stored field to the file space
3269 3270 3271
management if the field in data is owned the externally stored field,
in a rollback we may have the additional condition that the field must
not be inherited. */
3272 3273 3274 3275 3276

void
btr_free_externally_stored_field(
/*=============================*/
	dict_index_t*	index,		/* in: index of the data, the index
3277 3278 3279 3280 3281 3282 3283
					tree MUST be X-latched; if the tree
					height is 1, then also the root page
					must be X-latched! (this is relevant
					in the case this function is called
					from purge where 'data' is located on
					an undo log page, not an index
					page) */
3284 3285 3286 3287
	byte*		data,		/* in: internally stored data
					+ reference to the externally
					stored part */
	ulint		local_len,	/* in: length of data */
3288 3289 3290
	ibool		do_not_free_inherited,/* in: TRUE if called in a
					rollback and we do not want to free
					inherited fields */
unknown's avatar
unknown committed
3291 3292 3293
	mtr_t*		local_mtr __attribute__((unused))) /* in: mtr 
                                        containing the latch to data an an 
                                        X-latch to the index tree */
3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308
{
	page_t*	page;
	page_t*	rec_page;
	ulint	space_id;
	ulint	page_no;
	ulint	offset;
	ulint	extern_len;
	ulint	next_page_no;
	ulint	part_len;
	mtr_t	mtr;

	ut_a(local_len >= BTR_EXTERN_FIELD_REF_SIZE);
	ut_ad(mtr_memo_contains(local_mtr, dict_tree_get_lock(index->tree),
							MTR_MEMO_X_LOCK));
	ut_ad(mtr_memo_contains(local_mtr, buf_block_align(data),
unknown's avatar
unknown committed
3309
							MTR_MEMO_PAGE_X_FIX));
3310 3311 3312 3313 3314 3315 3316 3317
	ut_a(local_len >= BTR_EXTERN_FIELD_REF_SIZE);
	local_len -= BTR_EXTERN_FIELD_REF_SIZE;
	
	for (;;) {
		mtr_start(&mtr);

		rec_page = buf_page_get(buf_frame_get_space_id(data),
				buf_frame_get_page_no(data), RW_X_LATCH, &mtr);
3318
#ifdef UNIV_SYNC_DEBUG
3319
		buf_page_dbg_add_level(rec_page, SYNC_NO_ORDER_CHECK);
3320
#endif /* UNIV_SYNC_DEBUG */
3321 3322 3323 3324 3325 3326
		space_id = mach_read_from_4(data + local_len
						+ BTR_EXTERN_SPACE_ID);

		page_no = mach_read_from_4(data + local_len
						+ BTR_EXTERN_PAGE_NO);

unknown's avatar
unknown committed
3327 3328
		offset = mach_read_from_4(data + local_len
							+ BTR_EXTERN_OFFSET);
3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341
		extern_len = mach_read_from_4(data + local_len
						+ BTR_EXTERN_LEN + 4);

		/* If extern len is 0, then there is no external storage data
		at all */

		if (extern_len == 0) {

			mtr_commit(&mtr);

			return;
		}

3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361
		if (mach_read_from_1(data + local_len + BTR_EXTERN_LEN)
						& BTR_EXTERN_OWNER_FLAG) {
			/* This field does not own the externally
			stored field: do not free! */

			mtr_commit(&mtr);

			return;
		}

		if (do_not_free_inherited
			&& mach_read_from_1(data + local_len + BTR_EXTERN_LEN)
						& BTR_EXTERN_INHERITED_FLAG) {
			/* Rollback and inherited field: do not free! */

			mtr_commit(&mtr);

			return;
		}
		
3362
		page = buf_page_get(space_id, page_no, RW_X_LATCH, &mtr);
3363
#ifdef UNIV_SYNC_DEBUG
3364
		buf_page_dbg_add_level(page, SYNC_EXTERN_STORAGE);
3365
#endif /* UNIV_SYNC_DEBUG */
3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405
		next_page_no = mach_read_from_4(page + FIL_PAGE_DATA
						+ BTR_BLOB_HDR_NEXT_PAGE_NO);

		part_len = btr_blob_get_part_len(page + FIL_PAGE_DATA);

		ut_a(extern_len >= part_len);

		/* We must supply the page level (= 0) as an argument
		because we did not store it on the page (we save the space
		overhead from an index page header. */

		btr_page_free_low(index->tree, page, 0, &mtr);

		mlog_write_ulint(data + local_len + BTR_EXTERN_PAGE_NO,
						next_page_no,
						MLOG_4BYTES, &mtr);
		mlog_write_ulint(data + local_len + BTR_EXTERN_LEN + 4,
						extern_len - part_len,
						MLOG_4BYTES, &mtr);
		if (next_page_no == FIL_NULL) {
			ut_a(extern_len - part_len == 0);
		}

		if (extern_len - part_len == 0) {
			ut_a(next_page_no == FIL_NULL);
		}

		mtr_commit(&mtr);
	}
}

/***************************************************************
Frees the externally stored fields for a record. */

void
btr_rec_free_externally_stored_fields(
/*==================================*/
	dict_index_t*	index,	/* in: index of the data, the index
				tree MUST be X-latched */
	rec_t*		rec,	/* in: record */
3406 3407 3408
	ibool		do_not_free_inherited,/* in: TRUE if called in a
				rollback and we do not want to free
				inherited fields */
3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432
	mtr_t*		mtr)	/* in: mini-transaction handle which contains
				an X-latch to record page and to the index
				tree */
{
	ulint	n_fields;
	byte*	data;
	ulint	len;
	ulint	i;

	ut_ad(mtr_memo_contains(mtr, buf_block_align(rec),
							MTR_MEMO_PAGE_X_FIX));
	if (rec_get_data_size(rec) <= REC_1BYTE_OFFS_LIMIT) {

		return;
	}
	
	/* Free possible externally stored fields in the record */

	n_fields = rec_get_n_fields(rec);

	for (i = 0; i < n_fields; i++) {
		if (rec_get_nth_field_extern_bit(rec, i)) {

			data = rec_get_nth_field(rec, i, &len);
3433 3434
			btr_free_externally_stored_field(index, data, len,
						do_not_free_inherited, mtr);
3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449
		}
	}
}

/***************************************************************
Frees the externally stored fields for a record, if the field is mentioned
in the update vector. */
static
void
btr_rec_free_updated_extern_fields(
/*===============================*/
	dict_index_t*	index,	/* in: index of rec; the index tree MUST be
				X-latched */
	rec_t*		rec,	/* in: record */
	upd_t*		update,	/* in: update vector */
3450 3451 3452
	ibool		do_not_free_inherited,/* in: TRUE if called in a
				rollback and we do not want to free
				inherited fields */
3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478
	mtr_t*		mtr)	/* in: mini-transaction handle which contains
				an X-latch to record page and to the tree */
{
	upd_field_t*	ufield;
	ulint		n_fields;
	byte*		data;
	ulint		len;
	ulint		i;

	ut_ad(mtr_memo_contains(mtr, buf_block_align(rec),
							MTR_MEMO_PAGE_X_FIX));
	if (rec_get_data_size(rec) <= REC_1BYTE_OFFS_LIMIT) {

		return;
	}
	
	/* Free possible externally stored fields in the record */

	n_fields = upd_get_n_fields(update);

	for (i = 0; i < n_fields; i++) {
		ufield = upd_get_nth_field(update, i);
	
		if (rec_get_nth_field_extern_bit(rec, ufield->field_no)) {

			data = rec_get_nth_field(rec, ufield->field_no, &len);
3479 3480
			btr_free_externally_stored_field(index, data, len,
						do_not_free_inherited, mtr);
3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542
		}
	}
}

/***********************************************************************
Copies an externally stored field of a record to mem heap. Parameter
data contains a pointer to 'internally' stored part of the field:
possibly some data, and the reference to the externally stored part in
the last 20 bytes of data. */

byte*
btr_copy_externally_stored_field(
/*=============================*/
				/* out: the whole field copied to heap */
	ulint*		len,	/* out: length of the whole field */
	byte*		data,	/* in: 'internally' stored part of the
				field containing also the reference to
				the external part */
	ulint		local_len,/* in: length of data */
	mem_heap_t*	heap)	/* in: mem heap */
{
	page_t*	page;
	ulint	space_id;
	ulint	page_no;
	ulint	offset;
	ulint	extern_len;
	byte*	blob_header;
	ulint	part_len;
	byte*	buf;
	ulint	copied_len;
	mtr_t	mtr;

	ut_a(local_len >= BTR_EXTERN_FIELD_REF_SIZE);

	local_len -= BTR_EXTERN_FIELD_REF_SIZE;

	space_id = mach_read_from_4(data + local_len + BTR_EXTERN_SPACE_ID);

	page_no = mach_read_from_4(data + local_len + BTR_EXTERN_PAGE_NO);

	offset = mach_read_from_4(data + local_len + BTR_EXTERN_OFFSET);

	/* Currently a BLOB cannot be bigger that 4 GB; we
	leave the 4 upper bytes in the length field unused */
	
	extern_len = mach_read_from_4(data + local_len + BTR_EXTERN_LEN + 4);

	buf = mem_heap_alloc(heap, local_len + extern_len);

	ut_memcpy(buf, data, local_len);
	copied_len = local_len;

	if (extern_len == 0) {
		*len = copied_len;
		
		return(buf);
	}

	for (;;) {	
		mtr_start(&mtr);

		page = buf_page_get(space_id, page_no, RW_S_LATCH, &mtr);
3543
#ifdef UNIV_SYNC_DEBUG
3544
		buf_page_dbg_add_level(page, SYNC_EXTERN_STORAGE);
3545
#endif /* UNIV_SYNC_DEBUG */
3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604
		blob_header = page + offset;

		part_len = btr_blob_get_part_len(blob_header);

		ut_memcpy(buf + copied_len, blob_header + BTR_BLOB_HDR_SIZE,
							part_len);
		copied_len += part_len;

		page_no = btr_blob_get_next_page_no(blob_header);

		/* On other BLOB pages except the first the BLOB header
		always is at the page data start: */

		offset = FIL_PAGE_DATA;

		mtr_commit(&mtr);

		if (page_no == FIL_NULL) {
			ut_a(copied_len == local_len + extern_len);

			*len = copied_len;
		
			return(buf);
		}

		ut_a(copied_len < local_len + extern_len);
	}
}

/***********************************************************************
Copies an externally stored field of a record to mem heap. */

byte*
btr_rec_copy_externally_stored_field(
/*=================================*/
				/* out: the field copied to heap */
	rec_t*		rec,	/* in: record */	
	ulint		no,	/* in: field number */
	ulint*		len,	/* out: length of the field */
	mem_heap_t*	heap)	/* in: mem heap */
{
	ulint	local_len;
	byte*	data;

	ut_a(rec_get_nth_field_extern_bit(rec, no));

	/* An externally stored field can contain some initial
	data from the field, and in the last 20 bytes it has the
	space id, page number, and offset where the rest of the
	field data is stored, and the data length in addition to
	the data stored locally. We may need to store some data
	locally to get the local record length above the 128 byte
	limit so that field offsets are stored in two bytes, and
	the extern bit is available in those two bytes. */

	data = rec_get_nth_field(rec, no, &local_len);

	return(btr_copy_externally_stored_field(len, data, local_len, heap));
}