page0zip.c 101 KB
Newer Older
marko's avatar
marko committed
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
/******************************************************
Compressed page interface

(c) 2005 Innobase Oy

Created June 2005 by Marko Makela
*******************************************************/

#define THIS_MODULE
#include "page0zip.h"
#ifdef UNIV_NONINL
# include "page0zip.ic"
#endif
#undef THIS_MODULE
#include "page0page.h"
#include "mtr0log.h"
17
#include "ut0sort.h"
18
#include "dict0boot.h"
19
#include "dict0dict.h"
20
#include "btr0sea.h"
21 22
#include "btr0cur.h"
#include "page0types.h"
23
#include "lock0lock.h"
24
#include "log0recv.h"
marko's avatar
marko committed
25 26
#include "zlib.h"

27 28 29
/* Please refer to ../include/page0zip.ic for a description of the
compressed page format. */

30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
/* The infimum and supremum records are omitted from the compressed page.
On compress, we compare that the records are there, and on uncompress we
restore the records. */
static const byte infimum_extra[] = {
	0x01,			/* info_bits=0, n_owned=1 */
	0x00, 0x02		/* heap_no=0, status=2 */
	/* ?, ?	*/		/* next=(first user rec, or supremum) */
};
static const byte infimum_data[] = {
	0x69, 0x6e, 0x66, 0x69,
	0x6d, 0x75, 0x6d, 0x00	/* "infimum\0" */
};
static const byte supremum_extra_data[] = {
	/* 0x0?, */		/* info_bits=0, n_owned=1..8 */
	0x00, 0x0b,		/* heap_no=1, status=3 */
	0x00, 0x00,		/* next=0 */
	0x73, 0x75, 0x70, 0x72,
	0x65, 0x6d, 0x75, 0x6d	/* "supremum" */
};

marko's avatar
marko committed
50 51 52 53 54
#ifdef UNIV_DEBUG
/* Array of zeros, used for debug assertions */
static const byte zero[BTR_EXTERN_FIELD_REF_SIZE] = { 0, };
#endif

55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78
/**************************************************************************
Determine the guaranteed free space on an empty page. */

ulint
page_zip_empty_size(
/*================*/
				/* out: minimum payload size on the page */
	ulint	n_fields,	/* in: number of columns in the index */
	ulint	zip_size)	/* in: compressed page size in bytes */
{
	lint	size = zip_size
		/* subtract the page header and the longest
		uncompressed data needed for one record */
		- (PAGE_DATA
		   + PAGE_ZIP_DIR_SLOT_SIZE
		   + DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN
		   + 1/* encoded heap_no==2 in page_zip_write_rec() */
		   + 1/* end of modification log */
		   - REC_N_NEW_EXTRA_BYTES/* omitted bytes */)
		/* subtract the space for page_zip_fields_encode() */
		- compressBound(2 * (n_fields + 1));
	return(size > 0 ? (ulint) size : 0);
}

79 80 81 82 83 84 85 86 87 88 89 90 91
/*****************************************************************
Gets the size of the compressed page trailer (the dense page directory),
including deleted records (the free list). */
UNIV_INLINE
ulint
page_zip_dir_size(
/*==============*/
						/* out: length of dense page
						directory, in bytes */
	const page_zip_des_t*	page_zip)	/* in: compressed page */
{
	/* Exclude the page infimum and supremum from the record count. */
	ulint	size = PAGE_ZIP_DIR_SLOT_SIZE
92
		* (page_dir_get_n_heap((page_t*) page_zip->data) - 2);
93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108
	return(size);
}

/*****************************************************************
Gets the size of the compressed page trailer (the dense page directory),
only including user records (excluding the free list). */
UNIV_INLINE
ulint
page_zip_dir_user_size(
/*===================*/
						/* out: length of dense page
						directory comprising existing
						records, in bytes */
	const page_zip_des_t*	page_zip)	/* in: compressed page */
{
	ulint	size = PAGE_ZIP_DIR_SLOT_SIZE
109
		* page_get_n_recs((page_t*) page_zip->data);
110
	ut_ad(size <= page_zip_dir_size(page_zip));
111 112 113 114
	return(size);
}

/*****************************************************************
115
Find the slot of the given record in the dense page directory. */
116 117
UNIV_INLINE
byte*
118 119 120 121 122 123 124
page_zip_dir_find_low(
/*==================*/
					/* out: dense directory slot,
					or NULL if record not found */
	byte*	slot,			/* in: start of records */
	byte*	end,			/* in: end of records */
	ulint	offset)			/* in: offset of user record */
125
{
126
	ut_ad(slot <= end);
127 128 129

	for (; slot < end; slot += PAGE_ZIP_DIR_SLOT_SIZE) {
		if ((mach_read_from_2(slot) & PAGE_ZIP_DIR_SLOT_MASK)
130
		    == offset) {
131 132 133 134 135 136 137 138
			return(slot);
		}
	}

	return(NULL);
}

/*****************************************************************
139
Find the slot of the given non-free record in the dense page directory. */
140 141
UNIV_INLINE
byte*
142 143 144 145 146 147
page_zip_dir_find(
/*==============*/
						/* out: dense directory slot,
						or NULL if record not found */
	page_zip_des_t*	page_zip,		/* in: compressed page */
	ulint		offset)			/* in: offset of user record */
148
{
149
	byte*	end	= page_zip->data + page_zip->size;
150

151
	ut_ad(page_zip_simple_validate(page_zip));
152

153 154 155
	return(page_zip_dir_find_low(end - page_zip_dir_user_size(page_zip),
				     end,
				     offset));
156 157
}

158 159 160 161 162 163 164 165 166 167 168 169 170 171 172
/*****************************************************************
Find the slot of the given free record in the dense page directory. */
UNIV_INLINE
byte*
page_zip_dir_find_free(
/*===================*/
						/* out: dense directory slot,
						or NULL if record not found */
	page_zip_des_t*	page_zip,		/* in: compressed page */
	ulint		offset)			/* in: offset of user record */
{
	byte*	end	= page_zip->data + page_zip->size;

	ut_ad(page_zip_simple_validate(page_zip));

173 174 175
	return(page_zip_dir_find_low(end - page_zip_dir_size(page_zip),
				     end - page_zip_dir_user_size(page_zip),
				     offset));
176 177
}

178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195
/*****************************************************************
Read a given slot in the dense page directory. */
UNIV_INLINE
ulint
page_zip_dir_get(
/*=============*/
						/* out: record offset
						on the uncompressed page,
						possibly ORed with
						PAGE_ZIP_DIR_SLOT_DEL or
						PAGE_ZIP_DIR_SLOT_OWNED */
	const page_zip_des_t*	page_zip,	/* in: compressed page */
	ulint			slot)		/* in: slot
						(0=first user record) */
{
	ut_ad(page_zip_simple_validate(page_zip));
	ut_ad(slot < page_zip_dir_size(page_zip) / PAGE_ZIP_DIR_SLOT_SIZE);
	return(mach_read_from_2(page_zip->data + page_zip->size
196
				- PAGE_ZIP_DIR_SLOT_SIZE * (slot + 1)));
197 198
}

199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223
/**************************************************************************
Write a log record of compressing an index page. */
static
void
page_zip_compress_write_log(
/*========================*/
	const page_zip_des_t*	page_zip,/* in: compressed page */
	const page_t*		page,	/* in: uncompressed page */
	dict_index_t*		index,	/* in: index of the B-tree node */
	mtr_t*			mtr)	/* in: mini-transaction */
{
	byte*	log_ptr;
	ulint	trailer_size;

	log_ptr = mlog_open(mtr, 11 + 2 + 2);

	if (!log_ptr) {

		return;
	}

	/* Read the number of user records.
	Subtract 2 for the infimum and supremum records. */
	trailer_size = page_dir_get_n_heap(page_zip->data) - 2;
	/* Multiply by uncompressed of size stored per record */
224
	if (!page_is_leaf(page)) {
225
		trailer_size *= PAGE_ZIP_DIR_SLOT_SIZE + REC_NODE_PTR_SIZE;
226 227
	} else if (dict_index_is_clust(index)) {
		trailer_size *= PAGE_ZIP_DIR_SLOT_SIZE
228
			+ DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN;
229 230
	} else {
		trailer_size *= PAGE_ZIP_DIR_SLOT_SIZE;
231 232 233 234 235 236 237 238 239
	}
	/* Add the space occupied by BLOB pointers. */
	trailer_size += page_zip->n_blobs * BTR_EXTERN_FIELD_REF_SIZE;
	ut_a(page_zip->m_end > PAGE_DATA);
#if FIL_PAGE_DATA > PAGE_DATA
# error "FIL_PAGE_DATA > PAGE_DATA"
#endif
	ut_a(page_zip->m_end + trailer_size <= page_zip->size);

240 241 242
	log_ptr = mlog_write_initial_log_record_fast((page_t*) page,
						     MLOG_ZIP_PAGE_COMPRESS,
						     log_ptr, mtr);
243 244 245 246 247 248 249 250 251 252 253 254
	mach_write_to_2(log_ptr, page_zip->m_end - FIL_PAGE_TYPE);
	log_ptr += 2;
	mach_write_to_2(log_ptr, trailer_size);
	log_ptr += 2;
	mlog_close(mtr, log_ptr);

	/* Write FIL_PAGE_PREV and FIL_PAGE_NEXT */
	mlog_catenate_string(mtr, page_zip->data + FIL_PAGE_PREV, 4);
	mlog_catenate_string(mtr, page_zip->data + FIL_PAGE_NEXT, 4);
	/* Write most of the page header, the compressed stream and
	the modification log. */
	mlog_catenate_string(mtr, page_zip->data + FIL_PAGE_TYPE,
255
			     page_zip->m_end - FIL_PAGE_TYPE);
256 257
	/* Write the uncompressed trailer of the compressed page. */
	mlog_catenate_string(mtr, page_zip->data + page_zip->size
258
			     - trailer_size, trailer_size);
259 260
}

261 262
/**********************************************************
Determine how many externally stored columns are contained
263
in existing records with smaller heap_no than rec. */
264
static
265 266 267
ulint
page_zip_get_n_prev_extern(
/*=======================*/
268 269 270 271 272
	const page_zip_des_t*	page_zip,/* in: dense page directory on
					compressed page */
	const rec_t*		rec,	/* in: compact physical record
					on a B-tree leaf page */
	dict_index_t*		index)	/* in: record descriptor */
273
{
274
	page_t*	page	= page_align((rec_t*) rec);
275 276 277 278
	ulint	n_ext	= 0;
	ulint	i;
	ulint	left;
	ulint	heap_no;
279
	ulint	n_recs	= page_get_n_recs((page_t*) page_zip->data);
280 281 282 283

	ut_ad(page_is_leaf(page));
	ut_ad(page_is_comp(page));
	ut_ad(dict_table_is_comp(index->table));
284
	ut_ad(dict_index_is_clust(index));
285

286 287 288 289 290 291
	heap_no = rec_get_heap_no_new((rec_t*) rec);
	ut_ad(heap_no >= 2);	/* exclude infimum and supremum */
	left = heap_no - 2;
	if (UNIV_UNLIKELY(!left)) {
		return(0);
	}
292

293 294
	for (i = 0; i < n_recs; i++) {
		rec_t*	r	= page + (page_zip_dir_get(page_zip, i)
295
					  & PAGE_ZIP_DIR_SLOT_MASK);
296

297
		if (rec_get_heap_no_new(r) < heap_no) {
298 299
			n_ext += rec_get_n_extern_new(r, index,
						      ULINT_UNDEFINED);
300 301 302
			if (!--left) {
				break;
			}
303
		}
304
	}
305

306
	return(n_ext);
307 308
}

309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329
/**************************************************************************
Encode the length of a fixed-length column. */
static
byte*
page_zip_fixed_field_encode(
/*========================*/
			/* out: buf + length of encoded val */
	byte*	buf,	/* in: pointer to buffer where to write */
	ulint	val)	/* in: value to write */
{
	ut_ad(val >= 2);

	if (UNIV_LIKELY(val < 126)) {
		/*
		0 = nullable variable field of at most 255 bytes length;
		1 = not null variable field of at most 255 bytes length;
		126 = nullable variable field with maximum length >255;
		127 = not null variable field with maximum length >255
		*/
		*buf++ = val;
	} else {
330
		*buf++ = 0x80 | val >> 8;
331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363
		*buf++ = 0xff & val;
	}

	return(buf);
}

/**************************************************************************
Write the index information for the compressed page. */
static
ulint
page_zip_fields_encode(
/*===================*/
				/* out: used size of buf */
	ulint		n,	/* in: number of fields to compress */
	dict_index_t*	index,	/* in: index comprising at least n fields */
	ulint		trx_id_pos,/* in: position of the trx_id column
				in the index, or ULINT_UNDEFINED if
				this is a non-leaf page */
	byte*		buf)	/* out: buffer of (n + 1) * 2 bytes */
{
	const byte*	buf_start	= buf;
	ulint		i;
	ulint		col;
	ulint		trx_id_col	= 0;
	/* sum of lengths of preceding non-nullable fixed fields, or 0 */
	ulint		fixed_sum	= 0;

	ut_ad(trx_id_pos == ULINT_UNDEFINED || trx_id_pos < n);

	for (i = col = 0; i < n; i++) {
		dict_field_t*	field = dict_index_get_nth_field(index, i);
		ulint		val;

364
		if (dict_field_get_col(field)->prtype & DATA_NOT_NULL) {
365 366 367 368 369 370 371
			val = 1; /* set the "not nullable" flag */
		} else {
			val = 0; /* nullable field */
		}

		if (!field->fixed_len) {
			/* variable-length field */
372 373
			const dict_col_t*	column
				= dict_field_get_col(field);
374

375 376
			if (UNIV_UNLIKELY(column->len > 255)
			    || UNIV_UNLIKELY(column->mtype == DATA_BLOB)) {
377 378 379 380 381 382
				val |= 0x7e; /* max > 255 bytes */
			}

			if (fixed_sum) {
				/* write out the length of any
				preceding non-nullable fields */
383 384
				buf = page_zip_fixed_field_encode(
					buf, fixed_sum << 1 | 1);
385 386 387 388 389 390 391 392
				fixed_sum = 0;
				col++;
			}

			*buf++ = val;
			col++;
		} else if (val) {
			/* fixed-length non-nullable field */
393

394 395 396
			if (fixed_sum && UNIV_UNLIKELY
			    (fixed_sum + field->fixed_len
			     > DICT_MAX_INDEX_COL_LEN)) {
397 398 399 400
				/* Write out the length of the
				preceding non-nullable fields,
				to avoid exceeding the maximum
				length of a fixed-length column. */
401 402
				buf = page_zip_fixed_field_encode(
					buf, fixed_sum << 1 | 1);
403 404 405 406
				fixed_sum = 0;
				col++;
			}

407 408 409 410 411
			if (i && UNIV_UNLIKELY(i == trx_id_pos)) {
				if (fixed_sum) {
					/* Write out the length of any
					preceding non-nullable fields,
					and start a new trx_id column. */
412 413
					buf = page_zip_fixed_field_encode(
						buf, fixed_sum << 1 | 1);
414
					col++;
415 416
				}

417
				trx_id_col = col;
418 419 420 421 422 423 424 425 426 427 428
				fixed_sum = field->fixed_len;
			} else {
				/* add to the sum */
				fixed_sum += field->fixed_len;
			}
		} else {
			/* fixed-length nullable field */

			if (fixed_sum) {
				/* write out the length of any
				preceding non-nullable fields */
429 430
				buf = page_zip_fixed_field_encode(
					buf, fixed_sum << 1 | 1);
431 432 433 434
				fixed_sum = 0;
				col++;
			}

435 436
			buf = page_zip_fixed_field_encode(
				buf, field->fixed_len << 1);
437 438 439 440 441 442 443 444 445 446 447
			col++;
		}
	}

	if (fixed_sum) {
		/* Write out the lengths of last fixed-length columns. */
		buf = page_zip_fixed_field_encode(buf, fixed_sum << 1 | 1);
	}

	if (trx_id_pos != ULINT_UNDEFINED) {
		/* Write out the position of the trx_id column */
marko's avatar
marko committed
448 449 450 451
		i = trx_id_col;
	} else {
		/* Write out the number of nullable fields */
		i = index->n_nullable;
452 453
	}

marko's avatar
marko committed
454 455 456 457 458 459 460 461
	if (i < 128) {
		*buf++ = i;
	} else {
		*buf++ = 0x80 | i >> 8;
		*buf++ = 0xff & i;
	}

	ut_ad((ulint) (buf - buf_start) <= (n + 2) * 2);
462 463 464
	return((ulint) (buf - buf_start));
}

465 466 467 468 469 470 471
/**************************************************************************
Populate the dense page directory from the sparse directory. */
static
void
page_zip_dir_encode(
/*================*/
	const page_t*	page,	/* in: compact page */
472 473 474
	byte*		buf,	/* in: pointer to dense page directory[-1];
				out: dense directory on compressed page */
	const rec_t**	recs)	/* in: pointer to an array of 0, or NULL;
475 476
				out: dense page directory sorted by ascending
				address (and heap_no) */
477 478 479 480 481 482 483 484 485 486 487
{
	byte*	rec;
	ulint	status;
	ulint	min_mark;
	ulint	heap_no;
	ulint	i;
	ulint	n_heap;
	ulint	offs;

	min_mark = 0;

488 489 490
	if (page_is_leaf(page)) {
		status = REC_STATUS_ORDINARY;
	} else {
491
		status = REC_STATUS_NODE_PTR;
492
		if (UNIV_UNLIKELY
493
		    (mach_read_from_4(page + FIL_PAGE_PREV) == FIL_NULL)) {
494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514
			min_mark = REC_INFO_MIN_REC_FLAG;
		}
	}

	n_heap = page_dir_get_n_heap((page_t*) page);

	/* Traverse the list of stored records in the collation order,
	starting from the first user record. */

	rec = (page_t*) page + PAGE_NEW_INFIMUM, TRUE;

	i = 0;

	for (;;) {
		ulint	info_bits;
		offs = rec_get_next_offs(rec, TRUE);
		if (UNIV_UNLIKELY(offs == PAGE_NEW_SUPREMUM)) {
			break;
		}
		rec = (page_t*) page + offs;
		heap_no = rec_get_heap_no_new(rec);
515
		ut_a(heap_no >= 2); /* not infimum or supremum */
516
		ut_a(heap_no < n_heap);
517 518 519 520 521
		ut_a(offs < UNIV_PAGE_SIZE - PAGE_DIR);
		ut_a(offs >= PAGE_ZIP_START);
#if PAGE_ZIP_DIR_SLOT_MASK & UNIV_PAGE_SIZE
# error "PAGE_ZIP_DIR_SLOT_MASK & UNIV_PAGE_SIZE"
#endif
522 523 524 525 526 527 528 529 530 531 532 533 534 535
		if (UNIV_UNLIKELY(rec_get_n_owned_new(rec))) {
			offs |= PAGE_ZIP_DIR_SLOT_OWNED;
		}

		info_bits = rec_get_info_bits(rec, TRUE);
		if (UNIV_UNLIKELY(info_bits & REC_INFO_DELETED_FLAG)) {
			info_bits &= ~REC_INFO_DELETED_FLAG;
			offs |= PAGE_ZIP_DIR_SLOT_DEL;
		}
		ut_a(info_bits == min_mark);
		/* Only the smallest user record can have
		REC_INFO_MIN_REC_FLAG set. */
		min_mark = 0;

536
		mach_write_to_2(buf - PAGE_ZIP_DIR_SLOT_SIZE * ++i, offs);
537

538 539 540 541 542 543
		if (UNIV_LIKELY_NULL(recs)) {
			/* Ensure that each heap_no occurs at most once. */
			ut_a(!recs[heap_no - 2]);
			/* exclude infimum and supremum */
			recs[heap_no - 2] = rec;
		}
544 545 546 547 548 549 550 551 552 553 554 555

		ut_a(rec_get_status(rec) == status);
	}

	offs = page_header_get_field((page_t*) page, PAGE_FREE);

	/* Traverse the free list (of deleted records). */
	while (offs) {
		ut_ad(!(offs & ~PAGE_ZIP_DIR_SLOT_MASK));
		rec = (page_t*) page + offs;

		heap_no = rec_get_heap_no_new(rec);
556
		ut_a(heap_no >= 2); /* not infimum or supremum */
557 558 559 560 561
		ut_a(heap_no < n_heap);

		ut_a(!rec[-REC_N_NEW_EXTRA_BYTES]); /* info_bits and n_owned */
		ut_a(rec_get_status(rec) == status);

562
		mach_write_to_2(buf - PAGE_ZIP_DIR_SLOT_SIZE * ++i, offs);
563

564 565 566 567 568 569
		if (UNIV_LIKELY_NULL(recs)) {
			/* Ensure that each heap_no occurs at most once. */
			ut_a(!recs[heap_no - 2]);
			/* exclude infimum and supremum */
			recs[heap_no - 2] = rec;
		}
570 571 572 573 574

		offs = rec_get_next_offs(rec, TRUE);
	}

	/* Ensure that each heap no occurs at least once. */
575
	ut_a(i + 2/* infimum and supremum */ == n_heap);
576 577
}

578 579 580 581
/**************************************************************************
Allocate memory for zlib. */
static
void*
582 583
page_zip_malloc(
/*============*/
584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599
	void*	opaque __attribute__((unused)),
	uInt	items,
	uInt	size)
{
	return(ut_malloc(items * size));
}

/**************************************************************************
Deallocate memory for zlib. */
static
void
page_zip_free(
/*==========*/
	 void*	opaque __attribute__((unused)),
	 void*	address)
{
600
	ut_free(address);
601 602
}

603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632
#if defined UNIV_DEBUG || defined UNIV_ZIP_DEBUG
/* Set this variable in a debugger to enable
excessive logging in page_zip_compress(). */
ibool	page_zip_compress_dbg;

/**************************************************************************
Wrapper for deflate().  Log the operation if page_zip_compress_dbg is set. */
static
ibool
page_zip_compress_deflate(
/*======================*/
	z_streamp	strm,	/* in/out: compressed stream for deflate() */
	int		flush)	/* in: deflate() flushing method */
{
	int	status;
	if (UNIV_UNLIKELY(page_zip_compress_dbg)) {
		ut_print_buf(stderr, strm->next_in, strm->avail_in);
	}
	status = deflate(strm, flush);
	if (UNIV_UNLIKELY(page_zip_compress_dbg)) {
		fprintf(stderr, " -> %d\n", status);
	}
	return(status);
}

/* Redefine deflate(). */
# undef deflate
# define deflate page_zip_compress_deflate
#endif /* UNIV_DEBUG || UNIV_ZIP_DEBUG */

633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834
/**************************************************************************
Compress the records of a node pointer page. */
static
int
page_zip_compress_node_ptrs(
/*========================*/
					/* out: Z_OK, or a zlib error code */
	z_stream*	c_stream,	/* in/out: compressed page stream */
	const rec_t**	recs,		/* in: dense page directory
					sorted by address */
	ulint		n_dense,	/* in: size of recs[] */
	dict_index_t*	index,		/* in: the index of the page */
	byte*		storage,	/* in: end of dense page directory */
	mem_heap_t*	heap)		/* in: temporary memory heap */
{
	int	err	= Z_OK;
	ulint*	offsets = NULL;

	do {
		rec_t*	rec = (rec_t*) *recs++;

		offsets = rec_get_offsets(rec, index, offsets,
					  ULINT_UNDEFINED, &heap);
		/* Only leaf nodes may contain externally stored columns. */
		ut_ad(!rec_offs_any_extern(offsets));

		/* Compress the extra bytes. */
		c_stream->avail_in = rec - REC_N_NEW_EXTRA_BYTES
			- c_stream->next_in;

		if (c_stream->avail_in) {
			err = deflate(c_stream, Z_NO_FLUSH);
			if (UNIV_UNLIKELY(err != Z_OK)) {
				break;
			}
		}
		ut_ad(!c_stream->avail_in);

		/* Compress the data bytes, except node_ptr. */
		c_stream->next_in = rec;
		c_stream->avail_in = rec_offs_data_size(offsets)
			- REC_NODE_PTR_SIZE;
		ut_ad(c_stream->avail_in);

		err = deflate(c_stream, Z_NO_FLUSH);
		if (UNIV_UNLIKELY(err != Z_OK)) {
			break;
		}

		ut_ad(!c_stream->avail_in);

		memcpy(storage - REC_NODE_PTR_SIZE
		       * (rec_get_heap_no_new(rec) - 1),
		       c_stream->next_in, REC_NODE_PTR_SIZE);
		c_stream->next_in += REC_NODE_PTR_SIZE;
	} while (--n_dense);

	return(err);
}

/**************************************************************************
Compress the records of a leaf node of a secondary index. */
static
int
page_zip_compress_sec(
/*==================*/
					/* out: Z_OK, or a zlib error code */
	z_stream*	c_stream,	/* in/out: compressed page stream */
	const rec_t**	recs,		/* in: dense page directory
					sorted by address */
	ulint		n_dense)	/* in: size of recs[] */
{
	int		err	= Z_OK;

	ut_ad(n_dense > 0);

	do {
		const rec_t*	rec = *recs++;

		/* Compress everything up to this record. */
		c_stream->avail_in = rec - REC_N_NEW_EXTRA_BYTES
			- c_stream->next_in;

		if (UNIV_LIKELY(c_stream->avail_in)) {
			err = deflate(c_stream, Z_NO_FLUSH);
			if (UNIV_UNLIKELY(err != Z_OK)) {
				break;
			}
		}

		ut_ad(!c_stream->avail_in);
		ut_ad(c_stream->next_in == rec - REC_N_NEW_EXTRA_BYTES);

		/* Skip the REC_N_NEW_EXTRA_BYTES. */

		c_stream->next_in = (byte*) rec;
	} while (--n_dense);

	return(err);
}

/**************************************************************************
Compress the records of a leaf node of a clustered index. */
static
int
page_zip_compress_clust(
/*====================*/
					/* out: Z_OK, or a zlib error code */
	z_stream*	c_stream,	/* in/out: compressed page stream */
	const rec_t**	recs,		/* in: dense page directory
					sorted by address */
	ulint		n_dense,	/* in: size of recs[] */
	dict_index_t*	index,		/* in: the index of the page */
	ulint*		n_blobs_out,	/* out: number of
					externally stored columns */
	ulint		trx_id_col,	/* index of the trx_id column */
	byte*		deleted,	/* in: dense directory entry pointing
					to the head of the free list */
	byte*		storage,	/* in: end of dense page directory */
	mem_heap_t*	heap)		/* in: temporary memory heap */
{
	int	err		= Z_OK;
	ulint*	offsets		= NULL;
	/* BTR_EXTERN_FIELD_REF storage */
	byte*	externs		= storage - n_dense
		* (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
	ulint	n_blobs		= 0;
	ulint	n_fields	= dict_index_get_n_fields(index);

	do {
		ulint	i;
		rec_t*	rec = (rec_t*) *recs++;

		offsets = rec_get_offsets(rec, index, offsets,
					  ULINT_UNDEFINED, &heap);
		ut_ad(rec_offs_n_fields(offsets) == n_fields);

		/* Compress the extra bytes. */
		c_stream->avail_in = rec - REC_N_NEW_EXTRA_BYTES
			- c_stream->next_in;

		if (c_stream->avail_in) {
			err = deflate(c_stream, Z_NO_FLUSH);
			if (UNIV_UNLIKELY(err != Z_OK)) {

				goto func_exit;
			}
		}
		ut_ad(!c_stream->avail_in);
		ut_ad(c_stream->next_in == rec - REC_N_NEW_EXTRA_BYTES);

		/* Compress the data bytes. */

		c_stream->next_in = rec;

		/* Check if there are any externally stored columns.
		For each externally stored column, store the
		BTR_EXTERN_FIELD_REF separately. */

		for (i = 0; i < n_fields; i++) {
			ulint	len;
			byte*	src;

			if (UNIV_UNLIKELY(i == trx_id_col)) {
				ut_ad(!rec_offs_nth_extern(offsets, i));
				/* Store trx_id and roll_ptr
				in uncompressed form. */
				src = rec_get_nth_field(rec, offsets, i, &len);
				ut_ad(src + DATA_TRX_ID_LEN
				      == rec_get_nth_field(rec, offsets,
							   i + 1, &len));
				ut_ad(len == DATA_ROLL_PTR_LEN);

				/* Compress any preceding bytes. */
				c_stream->avail_in
					= src - c_stream->next_in;

				if (c_stream->avail_in) {
					err = deflate(c_stream, Z_NO_FLUSH);
					if (UNIV_UNLIKELY(err != Z_OK)) {

						goto func_exit;
					}
				}

				ut_ad(!c_stream->avail_in);
				ut_ad(c_stream->next_in == src);

				memcpy(storage
				       - (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN)
				       * (rec_get_heap_no_new(rec) - 1),
				       c_stream->next_in,
				       DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);

				c_stream->next_in
					+= DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN;

				/* Skip also roll_ptr */
				i++;
			} else if (rec_offs_nth_extern(offsets, i)) {
				src = rec_get_nth_field(rec, offsets, i, &len);
				ut_ad(dict_index_is_clust(index));
835
				ut_ad(len >= BTR_EXTERN_FIELD_REF_SIZE);
836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908
				src += len - BTR_EXTERN_FIELD_REF_SIZE;

				c_stream->avail_in = src
					- c_stream->next_in;
				if (UNIV_LIKELY(c_stream->avail_in)) {
					err = deflate(c_stream, Z_NO_FLUSH);
					if (UNIV_UNLIKELY(err != Z_OK)) {

						goto func_exit;
					}
				}

				ut_ad(!c_stream->avail_in);
				ut_ad(c_stream->next_in == src);

				/* Reserve space for the data at
				the end of the space reserved for
				the compressed data and the page
				modification log. */

				if (UNIV_UNLIKELY
				    (c_stream->avail_out
				     <= BTR_EXTERN_FIELD_REF_SIZE)) {
					/* out of space */
					goto func_exit;
				}

				ut_ad(externs == c_stream->next_out
				      + c_stream->avail_out
				      + 1/* end of modif. log */);

				c_stream->next_in
					+= BTR_EXTERN_FIELD_REF_SIZE;

				/* Skip deleted records. */
				if (UNIV_LIKELY_NULL
				    (page_zip_dir_find_low(
					    storage, deleted,
					    page_offset(rec)))) {
					continue;
				}

				n_blobs++;
				c_stream->avail_out
					-= BTR_EXTERN_FIELD_REF_SIZE;
				externs -= BTR_EXTERN_FIELD_REF_SIZE;

				/* Copy the BLOB pointer */
				memcpy(externs, c_stream->next_in
				       - BTR_EXTERN_FIELD_REF_SIZE,
				       BTR_EXTERN_FIELD_REF_SIZE);
			}
		}

		/* Compress the last bytes of the record. */
		c_stream->avail_in = rec_get_end(rec, offsets)
			- c_stream->next_in;

		if (c_stream->avail_in) {
			err = deflate(c_stream, Z_NO_FLUSH);
			if (UNIV_UNLIKELY(err != Z_OK)) {

				goto func_exit;
			}
		}
		ut_ad(!c_stream->avail_in);
	} while (--n_dense);

func_exit:
	*n_blobs_out = n_blobs;
	return(err);
}

marko's avatar
marko committed
909 910 911 912 913 914 915 916
/**************************************************************************
Compress a page. */

ibool
page_zip_compress(
/*==============*/
				/* out: TRUE on success, FALSE on failure;
				page_zip will be left intact on failure. */
917 918 919
	page_zip_des_t*	page_zip,/* in: size; out: data, n_blobs,
				m_start, m_end */
	const page_t*	page,	/* in: uncompressed page */
920 921
	dict_index_t*	index,	/* in: index of the B-tree node */
	mtr_t*		mtr)	/* in: mini-transaction, or NULL */
marko's avatar
marko committed
922 923 924
{
	z_stream	c_stream;
	int		err;
925 926 927
	ulint		n_fields;/* number of index fields needed */
	byte*		fields;	/* index field information */
	byte*		buf;	/* compressed payload of the page */
928
	byte*		buf_end;/* end of buf */
929
	ulint		n_dense;
930
	ulint		slot_size;/* amount of uncompressed bytes per record */
931
	const rec_t**	recs;	/* dense page directory, sorted by address */
932
	mem_heap_t*	heap;
933 934 935 936
	ulint		trx_id_col;
	ulint*		offsets	= NULL;
	ulint		n_blobs	= 0;
	byte*		storage;/* storage of uncompressed columns */
937 938

	ut_a(page_is_comp((page_t*) page));
939
	ut_a(fil_page_get_type((page_t*) page) == FIL_PAGE_INDEX);
940
	ut_ad(page_simple_validate_new((page_t*) page));
941
	ut_ad(page_zip_simple_validate(page_zip));
942 943 944 945 946 947 948

	/* Check the data that will be omitted. */
	ut_a(!memcmp(page + (PAGE_NEW_INFIMUM - REC_N_NEW_EXTRA_BYTES),
		     infimum_extra, sizeof infimum_extra));
	ut_a(!memcmp(page + PAGE_NEW_INFIMUM,
		     infimum_data, sizeof infimum_data));
	ut_a(page[PAGE_NEW_SUPREMUM - REC_N_NEW_EXTRA_BYTES]
949 950
	     /* info_bits == 0, n_owned <= max */
	     <= PAGE_DIR_SLOT_MAX_N_OWNED);
951 952
	ut_a(!memcmp(page + (PAGE_NEW_SUPREMUM - REC_N_NEW_EXTRA_BYTES + 1),
		     supremum_extra_data, sizeof supremum_extra_data));
953

954 955
	if (UNIV_UNLIKELY(!page_get_n_recs((page_t*) page))) {
		ut_a(rec_get_next_offs((page_t*) page + PAGE_NEW_INFIMUM, TRUE)
956
		     == PAGE_NEW_SUPREMUM);
957
	}
958

959 960 961 962 963 964
	if (page_is_leaf(page)) {
		n_fields = dict_index_get_n_fields(index);
	} else {
		n_fields = dict_index_get_n_unique_in_tree(index);
	}

965 966
	/* The dense directory excludes the infimum and supremum records. */
	n_dense = page_dir_get_n_heap((page_t*) page) - 2;
967 968 969 970 971 972 973 974
#if defined UNIV_DEBUG || defined UNIV_ZIP_DEBUG
	if (UNIV_UNLIKELY(page_zip_compress_dbg)) {
		fprintf(stderr, "compress %p %p %lu %lu %lu\n",
			(void*) page_zip, (void*) page,
			page_is_leaf(page),
			n_fields, n_dense);
	}
#endif /* UNIV_DEBUG || UNIV_ZIP_DEBUG */
975
	if (UNIV_UNLIKELY(n_dense * PAGE_ZIP_DIR_SLOT_SIZE
976
			  >= page_zip->size)) {
977 978
		return(FALSE);
	}
979 980

	heap = mem_heap_create(page_zip->size
981 982 983
			       + n_fields * (2 + sizeof *offsets)
			       + n_dense * ((sizeof *recs)
					    - PAGE_ZIP_DIR_SLOT_SIZE));
984

985 986
	recs = mem_heap_alloc(heap, n_dense * sizeof *recs);
	memset(recs, 0, n_dense * sizeof *recs);
987

988 989
	fields = mem_heap_alloc(heap, (n_fields + 1) * 2);

990 991
	buf = mem_heap_alloc(heap, page_zip->size - PAGE_DATA);
	buf_end = buf + page_zip->size - PAGE_DATA;
992

marko's avatar
marko committed
993
	/* Compress the data payload. */
994
	c_stream.zalloc = page_zip_malloc;
995
	c_stream.zfree = page_zip_free;
marko's avatar
marko committed
996 997 998 999 1000 1001
	c_stream.opaque = (voidpf) 0;

	err = deflateInit(&c_stream, Z_DEFAULT_COMPRESSION);
	ut_a(err == Z_OK);

	c_stream.next_out = buf;
1002
	/* Subtract the space reserved for uncompressed data. */
1003 1004
	/* Page header and the end marker of the modification log */
	c_stream.avail_out = buf_end - buf - 1;
1005 1006
	/* Dense page directory and uncompressed columns, if any */
	if (page_is_leaf(page)) {
1007
		if (dict_index_is_clust(index)) {
1008 1009
			trx_id_col = dict_index_get_sys_col_pos(
				index, DATA_TRX_ID);
1010 1011 1012
			ut_ad(trx_id_col > 0);
			ut_ad(trx_id_col != ULINT_UNDEFINED);

1013
			slot_size = PAGE_ZIP_DIR_SLOT_SIZE
1014
				+ DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN;
1015
		} else {
1016 1017
			/* Signal the absence of trx_id
			in page_zip_fields_encode() */
1018 1019
			ut_ad(dict_index_get_sys_col_pos(index, DATA_TRX_ID)
			      == ULINT_UNDEFINED);
1020
			trx_id_col = 0;
1021
			slot_size = PAGE_ZIP_DIR_SLOT_SIZE;
1022
		}
1023
	} else {
1024
		slot_size = PAGE_ZIP_DIR_SLOT_SIZE + REC_NODE_PTR_SIZE;
1025
		trx_id_col = ULINT_UNDEFINED;
1026 1027
	}

1028 1029 1030 1031 1032
	if (UNIV_UNLIKELY(c_stream.avail_out < n_dense * slot_size)) {
		goto zlib_error;
	}

	c_stream.avail_out -= n_dense * slot_size;
1033 1034
	c_stream.avail_in = page_zip_fields_encode(n_fields, index,
						   trx_id_col, fields);
1035
	c_stream.next_in = fields;
1036
	if (UNIV_LIKELY(!trx_id_col)) {
1037 1038
		trx_id_col = ULINT_UNDEFINED;
	}
1039

1040 1041 1042 1043 1044
	err = deflate(&c_stream, Z_FULL_FLUSH);
	if (err != Z_OK) {
		goto zlib_error;
	}

1045 1046
	ut_ad(!c_stream.avail_in);

1047
	page_zip_dir_encode(page, buf_end, recs);
1048 1049 1050

	c_stream.next_in = (byte*) page + PAGE_ZIP_START;

1051
	storage = buf_end - n_dense * PAGE_ZIP_DIR_SLOT_SIZE;
1052

1053
	/* Compress the records in heap_no order. */
1054
	if (UNIV_UNLIKELY(!n_dense)) {
1055 1056
	} else if (!page_is_leaf(page)) {
		/* This is a node pointer page. */
1057 1058 1059 1060 1061
		err = page_zip_compress_node_ptrs(&c_stream, recs, n_dense,
						  index, storage, heap);
		if (UNIV_UNLIKELY(err != Z_OK)) {
			goto zlib_error;
		}
1062
	} else if (UNIV_LIKELY(trx_id_col == ULINT_UNDEFINED)) {
1063 1064 1065 1066 1067
		/* This is a leaf page in a secondary index. */
		err = page_zip_compress_sec(&c_stream, recs, n_dense);
		if (UNIV_UNLIKELY(err != Z_OK)) {
			goto zlib_error;
		}
1068 1069
	} else {
		/* This is a leaf page in a clustered index. */
1070 1071
		err = page_zip_compress_clust(&c_stream, recs, n_dense,
					      index, &n_blobs, trx_id_col,
1072 1073
					      buf_end - PAGE_ZIP_DIR_SLOT_SIZE
					      * page_get_n_recs((page_t*)page),
1074 1075 1076 1077
					      storage, heap);
		if (UNIV_UNLIKELY(err != Z_OK)) {
			goto zlib_error;
		}
1078 1079
	}

1080 1081
	/* Finish the compression. */
	ut_ad(!c_stream.avail_in);
1082
	/* Compress any trailing garbage, in case the last record was
1083 1084
	allocated from an originally longer space on the free list,
	or the data of the last record from page_zip_compress_sec(). */
1085 1086 1087
	c_stream.avail_in
		= page_header_get_field((page_t*) page, PAGE_HEAP_TOP)
		- (c_stream.next_in - page);
1088
	ut_a(c_stream.avail_in <= UNIV_PAGE_SIZE - PAGE_ZIP_START - PAGE_DIR);
marko's avatar
marko committed
1089 1090

	err = deflate(&c_stream, Z_FINISH);
1091

1092
	if (UNIV_UNLIKELY(err != Z_STREAM_END)) {
1093
zlib_error:
marko's avatar
marko committed
1094
		deflateEnd(&c_stream);
1095
		mem_heap_free(heap);
marko's avatar
marko committed
1096 1097 1098 1099 1100 1101
		return(FALSE);
	}

	err = deflateEnd(&c_stream);
	ut_a(err == Z_OK);

1102
	ut_ad(buf + c_stream.total_out == c_stream.next_out);
1103 1104
	ut_ad((ulint) (storage - c_stream.next_out) >= c_stream.avail_out);

1105 1106 1107 1108
	/* Zero out the area reserved for the modification log.
	Space for the end marker of the modification log is not
	included in avail_out. */
	memset(c_stream.next_out, 0, c_stream.avail_out + 1/* end marker */);
1109

marko's avatar
marko committed
1110
	page_zip->m_end = page_zip->m_start = PAGE_DATA + c_stream.total_out;
1111
	page_zip->n_blobs = n_blobs;
1112 1113 1114
	/* Copy those header fields that will not be written
	in buf_flush_init_for_writing() */
	memcpy(page_zip->data + FIL_PAGE_PREV, page + FIL_PAGE_PREV,
1115
	       FIL_PAGE_LSN - FIL_PAGE_PREV);
1116 1117
	memcpy(page_zip->data + FIL_PAGE_TYPE, page + FIL_PAGE_TYPE, 2);
	memcpy(page_zip->data + FIL_PAGE_DATA, page + FIL_PAGE_DATA,
1118
	       PAGE_DATA - FIL_PAGE_DATA);
1119 1120
	/* Copy the rest of the compressed page */
	memcpy(page_zip->data + PAGE_DATA, buf, page_zip->size - PAGE_DATA);
1121
	mem_heap_free(heap);
1122
#ifdef UNIV_ZIP_DEBUG
1123
	ut_a(page_zip_validate(page_zip, page));
1124
#endif /* UNIV_ZIP_DEBUG */
marko's avatar
marko committed
1125

1126 1127 1128 1129
	if (mtr) {
		page_zip_compress_write_log(page_zip, page, index, mtr);
	}

1130
	return(TRUE);
marko's avatar
marko committed
1131 1132
}

1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153
/**************************************************************************
Compare two page directory entries. */
UNIV_INLINE
ibool
page_zip_dir_cmp(
/*=============*/
				/* out: positive if rec1 > rec2 */
	const rec_t*	rec1,	/* in: rec1 */
	const rec_t*	rec2)	/* in: rec2 */
{
	return(rec1 > rec2);
}

/**************************************************************************
Sort the dense page directory by address (heap_no). */
static
void
page_zip_dir_sort(
/*==============*/
	rec_t**	arr,	/* in/out: dense page directory */
	rec_t**	aux_arr,/* in/out: work area */
1154 1155
	ulint	low,	/* in: lower bound of the sorting area, inclusive */
	ulint	high)	/* in: upper bound of the sorting area, exclusive */
1156 1157
{
	UT_SORT_FUNCTION_BODY(page_zip_dir_sort, arr, aux_arr, low, high,
1158
			      page_zip_dir_cmp);
1159 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 1190 1191 1192 1193
}

/**************************************************************************
Deallocate the index information initialized by page_zip_fields_decode(). */
static
void
page_zip_fields_free(
/*=================*/
	dict_index_t*	index)	/* in: dummy index to be freed */
{
	if (index) {
		dict_table_t*	table = index->table;
		mem_heap_free(index->heap);
		mutex_free(&(table->autoinc_mutex));
		mem_heap_free(table->heap);
	}
}

/**************************************************************************
Read the index information for the compressed page. */
static
dict_index_t*
page_zip_fields_decode(
/*===================*/
				/* out,own: dummy index describing the page,
				or NULL on error */
	const byte*	buf,	/* in: index information */
	const byte*	end,	/* in: end of buf */
	ulint*		trx_id_col)/* in: NULL for non-leaf pages;
				for leaf pages, pointer to where to store
				the position of the trx_id column */
{
	const byte*	b;
	ulint		n;
	ulint		i;
marko's avatar
marko committed
1194
	ulint		val;
1195 1196 1197 1198 1199 1200 1201 1202 1203 1204
	dict_table_t*	table;
	dict_index_t*	index;

	/* Determine the number of fields. */
	for (b = buf, n = 0; b < end; n++) {
		if (*b++ & 0x80) {
			b++; /* skip the second byte */
		}
	}

marko's avatar
marko committed
1205 1206
	n--; /* n_nullable or trx_id */

1207
	if (UNIV_UNLIKELY(n > REC_MAX_N_FIELDS)
1208
	    || UNIV_UNLIKELY(b > end)) {
1209 1210 1211 1212

		return(NULL);
	}

1213
	table = dict_mem_table_create("ZIP_DUMMY", DICT_HDR_SPACE, n,
1214
				      DICT_TF_COMPACT);
1215
	index = dict_mem_index_create("ZIP_DUMMY", "ZIP_DUMMY",
1216
				      DICT_HDR_SPACE, 0, n);
1217 1218 1219 1220 1221 1222 1223 1224 1225 1226
	index->table = table;
	index->n_uniq = n;
	/* avoid ut_ad(index->cached) in dict_index_get_n_unique_in_tree */
	index->cached = TRUE;

	/* Initialize the fields. */
	for (b = buf, i = 0; i < n; i++) {
		ulint	mtype;
		ulint	len;

marko's avatar
marko committed
1227 1228
		val = *b++;

1229
		if (UNIV_UNLIKELY(val & 0x80)) {
1230
			/* fixed length > 62 bytes */
1231
			val = (val & 0x7f) << 8 | *b++;
1232 1233 1234 1235
			len = val >> 1;
			mtype = DATA_FIXBINARY;
		} else if (UNIV_UNLIKELY(val >= 126)) {
			/* variable length with max > 255 bytes */
1236 1237
			len = 0x7fff;
			mtype = DATA_BINARY;
1238 1239 1240 1241 1242 1243 1244
		} else if (val <= 1) {
			/* variable length with max <= 255 bytes */
			len = 0;
			mtype = DATA_BINARY;
		} else {
			/* fixed length < 62 bytes */
			len = val >> 1;
1245 1246
			mtype = DATA_FIXBINARY;
		}
1247

1248
		dict_mem_table_add_col(table, "DUMMY", mtype,
1249 1250
				       val & 1 ? DATA_NOT_NULL : 0, len);
		dict_index_add_col(index, table, (dict_col_t*)
1251
				   dict_table_get_nth_col(table, i), 0);
1252 1253
	}

marko's avatar
marko committed
1254 1255 1256 1257 1258
	val = *b++;
	if (UNIV_UNLIKELY(val & 0x80)) {
		val = (val & 0x7f) << 8 | *b++;
	}

1259 1260 1261 1262 1263 1264 1265 1266 1267
	/* Decode the position of the trx_id column. */
	if (trx_id_col) {
		if (UNIV_UNLIKELY(val >= n)) {
			page_zip_fields_free(index);
			index = NULL;
		}

		if (!val) {
			val = ULINT_UNDEFINED;
1268 1269
		} else {
			index->type = DICT_CLUSTERED;
1270 1271 1272
		}

		*trx_id_col = val;
marko's avatar
marko committed
1273 1274 1275 1276 1277 1278 1279 1280
	} else {
		/* Decode the number of nullable fields. */
		if (UNIV_UNLIKELY(index->n_nullable > val)) {
			page_zip_fields_free(index);
			index = NULL;
		} else {
			index->n_nullable = val;
		}
1281 1282 1283 1284 1285
	}

	ut_ad(b == end);

	return(index);
1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303
}

/**************************************************************************
Populate the sparse page directory from the dense directory. */
static
ibool
page_zip_dir_decode(
/*================*/
					/* out: TRUE on success,
					FALSE on failure */
	const page_zip_des_t*	page_zip,/* in: dense page directory on
					compressed page */
	page_t*			page,	/* in: compact page with valid header;
					out: trailer and sparse page directory
					filled in */
	rec_t**			recs,	/* out: dense page directory sorted by
					ascending address (and heap_no) */
	rec_t**			recs_aux,/* in/out: scratch area */
1304
	ulint			n_dense)/* in: number of user records, and
1305 1306 1307 1308 1309 1310 1311 1312
					size of recs[] and recs_aux[] */
{
	ulint	i;
	ulint	n_recs;
	byte*	slot;

	n_recs = page_get_n_recs(page);

1313 1314 1315 1316
	if (UNIV_UNLIKELY(n_recs > n_dense)) {
		return(FALSE);
	}

1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339
	/* Traverse the list of stored records in the sorting order,
	starting from the first user record. */

	slot = page + (UNIV_PAGE_SIZE - PAGE_DIR - PAGE_DIR_SLOT_SIZE);
	UNIV_PREFETCH_RW(slot);

	/* Zero out the page trailer. */
	memset(slot + PAGE_DIR_SLOT_SIZE, 0, PAGE_DIR);

	mach_write_to_2(slot, PAGE_NEW_INFIMUM);
	slot -= PAGE_DIR_SLOT_SIZE;
	UNIV_PREFETCH_RW(slot);

	/* Initialize the sparse directory and copy the dense directory. */
	for (i = 0; i < n_recs; i++) {
		ulint	offs = page_zip_dir_get(page_zip, i);

		if (offs & PAGE_ZIP_DIR_SLOT_OWNED) {
			mach_write_to_2(slot, offs & PAGE_ZIP_DIR_SLOT_MASK);
			slot -= PAGE_DIR_SLOT_SIZE;
			UNIV_PREFETCH_RW(slot);
		}

1340
		if (UNIV_UNLIKELY((offs & PAGE_ZIP_DIR_SLOT_MASK)
1341
				  < PAGE_ZIP_START + REC_N_NEW_EXTRA_BYTES)) {
1342 1343
			return(FALSE);
		}
1344

1345 1346 1347 1348
		recs[i] = page + (offs & PAGE_ZIP_DIR_SLOT_MASK);
	}

	mach_write_to_2(slot, PAGE_NEW_SUPREMUM);
1349 1350 1351
	if (UNIV_UNLIKELY
	    (slot != page_dir_get_nth_slot(page,
					   page_dir_get_n_slots(page) - 1))) {
1352 1353 1354 1355
		return(FALSE);
	}

	/* Copy the rest of the dense directory. */
1356
	for (; i < n_dense; i++) {
1357 1358 1359 1360 1361 1362 1363 1364 1365
		ulint	offs = page_zip_dir_get(page_zip, i);

		if (UNIV_UNLIKELY(offs & ~PAGE_ZIP_DIR_SLOT_MASK)) {
			return(FALSE);
		}

		recs[i] = page + offs;
	}

1366
	if (UNIV_LIKELY(n_dense > 1)) {
1367
		page_zip_dir_sort(recs, recs_aux, 0, n_dense);
1368 1369 1370 1371
	}
	return(TRUE);
}

1372 1373
/**************************************************************************
Initialize the REC_N_NEW_EXTRA_BYTES of each record. */
1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405
static
ibool
page_zip_set_extra_bytes(
/*=====================*/
					/* out: TRUE on success,
					FALSE on failure */
	const page_zip_des_t*	page_zip,/* in: compressed page */
	page_t*			page,	/* in/out: uncompressed page */
	ulint			info_bits)/* in: REC_INFO_MIN_REC_FLAG or 0 */
{
	ulint	n;
	ulint	i;
	ulint	n_owned = 1;
	ulint	offs;
	rec_t*	rec;

	n = page_get_n_recs(page);
	rec = page + PAGE_NEW_INFIMUM;

	for (i = 0; i < n; i++) {
		offs = page_zip_dir_get(page_zip, i);

		if (UNIV_UNLIKELY(offs & PAGE_ZIP_DIR_SLOT_DEL)) {
			info_bits |= REC_INFO_DELETED_FLAG;
		}
		if (UNIV_UNLIKELY(offs & PAGE_ZIP_DIR_SLOT_OWNED)) {
			info_bits |= n_owned;
			n_owned = 1;
		} else {
			n_owned++;
		}
		offs &= PAGE_ZIP_DIR_SLOT_MASK;
1406
		if (UNIV_UNLIKELY(offs < PAGE_ZIP_START
1407
				  + REC_N_NEW_EXTRA_BYTES)) {
1408 1409
			return(FALSE);
		}
1410

1411
		rec_set_next_offs_new(rec, offs);
1412
		rec = page + offs;
1413 1414
		rec[-REC_N_NEW_EXTRA_BYTES] = info_bits;
		info_bits = 0;
1415 1416 1417
	}

	/* Set the next pointer of the last user record. */
1418
	rec_set_next_offs_new(rec, PAGE_NEW_SUPREMUM);
1419 1420 1421 1422

	/* Set n_owned of the supremum record. */
	page[PAGE_NEW_SUPREMUM - REC_N_NEW_EXTRA_BYTES] = n_owned;

1423 1424 1425 1426
	/* The dense directory excludes the infimum and supremum records. */
	n = page_dir_get_n_heap(page) - 2;

	if (i >= n) {
1427

1428
		return(UNIV_LIKELY(i == n));
1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442
	}

	offs = page_zip_dir_get(page_zip, i);

	/* Set the extra bytes of deleted records on the free list. */
	for (;;) {
		if (UNIV_UNLIKELY(!offs)
		    || UNIV_UNLIKELY(offs & ~PAGE_ZIP_DIR_SLOT_MASK)) {
			return(FALSE);
		}

		rec = page + offs;
		rec[-REC_N_NEW_EXTRA_BYTES] = 0; /* info_bits and n_owned */

1443 1444 1445 1446 1447
		if (++i == n) {
			break;
		}

		offs = page_zip_dir_get(page_zip, i);
1448
		rec_set_next_offs_new(rec, offs);
1449 1450 1451 1452
	}

	/* Terminate the free list. */
	rec[-REC_N_NEW_EXTRA_BYTES] = 0; /* info_bits and n_owned */
1453
	rec_set_next_offs_new(rec, 0);
1454

1455
	return(TRUE);
1456 1457
}

1458
/**************************************************************************
1459 1460
Apply the modification log to an uncompressed page.
Do not copy the fields that are stored separately. */
1461 1462 1463 1464 1465 1466 1467
static
const byte*
page_zip_apply_log(
/*===============*/
				/* out: pointer to end of modification log,
				or NULL on failure */
	const byte*	data,	/* in: modification log */
1468
	ulint		size,	/* in: maximum length of the log, in bytes */
1469 1470 1471
	rec_t**		recs,	/* in: dense page directory,
				sorted by address (indexed by heap_no - 2) */
	ulint		n_dense,/* in: size of recs[] */
1472 1473
	ulint		trx_id_col,/* in: column number of trx_id in the index,
				or ULINT_UNDEFINED if none */
1474 1475 1476 1477 1478 1479
	ulint		heap_status,
				/* in: heap_no and status bits for
				the next record to uncompress */
	dict_index_t*	index,	/* in: index of the page */
	ulint*		offsets)/* in/out: work area for
				rec_get_offsets_reverse() */
1480
{
1481 1482
	const byte* const end = data + size;

1483
	for (;;) {
1484
		ulint	val;
1485 1486 1487 1488
		rec_t*	rec;
		ulint	len;
		ulint	hs;

1489 1490
		val = *data++;
		if (UNIV_UNLIKELY(!val)) {
1491
			return(data - 1);
1492
		}
1493
		if (val & 0x80) {
1494
			val = (val & 0x7f) << 8 | *data++;
1495 1496 1497
			if (UNIV_UNLIKELY(!val)) {
				return(NULL);
			}
1498 1499
		}
		if (UNIV_UNLIKELY(data >= end)) {
1500 1501
			return(NULL);
		}
1502
		if (UNIV_UNLIKELY((val >> 1) > n_dense)) {
1503 1504
			return(NULL);
		}
1505 1506

		/* Determine the heap number and status bits of the record. */
1507 1508 1509 1510 1511 1512 1513
		rec = recs[(val >> 1) - 1];

		if (val & 1) {
			/* Clear the data bytes of the record. */
			mem_heap_t*	heap	= NULL;
			ulint*		offs;
			offs = rec_get_offsets(rec, index, offsets,
1514
					       ULINT_UNDEFINED, &heap);
1515 1516 1517 1518 1519 1520 1521 1522 1523
			memset(rec, 0, rec_offs_data_size(offs));

			if (UNIV_LIKELY_NULL(heap)) {
				mem_heap_free(heap);
			}
			continue;
		}

		hs = ((val >> 1) + 1) << REC_HEAP_NO_SHIFT;
1524 1525 1526 1527 1528 1529 1530 1531 1532 1533
		hs |= heap_status & ((1 << REC_HEAP_NO_SHIFT) - 1);

		/* This may either be an old record that is being
		overwritten (updated in place, or allocated from
		the free list), or a new record, with the next
		available_heap_no. */
		if (UNIV_UNLIKELY(hs > heap_status)) {
			return(NULL);
		} else if (hs == heap_status) {
			/* A new record was allocated from the heap. */
marko's avatar
marko committed
1534
			heap_status += 1 << REC_HEAP_NO_SHIFT;
1535 1536
		}

marko's avatar
marko committed
1537 1538 1539 1540
		mach_write_to_2(rec - REC_NEW_HEAP_NO, hs);
#if REC_STATUS_NODE_PTR != TRUE
# error "REC_STATUS_NODE_PTR != TRUE"
#endif
1541
		rec_get_offsets_reverse(data, index,
1542 1543
					hs & REC_STATUS_NODE_PTR,
					offsets);
marko's avatar
marko committed
1544
		rec_offs_make_valid(rec, index, offsets);
1545 1546 1547

		/* Copy the extra bytes (backwards). */
		{
1548 1549 1550 1551
			byte*	start	= rec_get_start(rec, offsets);
			byte*	b	= rec - REC_N_NEW_EXTRA_BYTES;
			while (b != start) {
				*--b = *data++;
1552 1553 1554 1555
			}
		}

		/* Copy the data bytes. */
marko's avatar
marko committed
1556
		if (UNIV_UNLIKELY(hs & REC_STATUS_NODE_PTR)) {
1557 1558 1559 1560 1561
			/* Non-leaf nodes should not contain any
			externally stored columns. */
			if (UNIV_UNLIKELY(rec_offs_any_extern(offsets))) {
				return(NULL);
			}
1562

1563
			len = rec_offs_data_size(offsets)
1564
				- REC_NODE_PTR_SIZE;
1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576
			/* Copy the data bytes, except node_ptr. */
			if (UNIV_UNLIKELY(data + len >= end)) {
				return(NULL);
			}
			memcpy(rec, data, len);
			data += len;
		} else {
			ulint	i;
			byte*	next_out = rec;

			/* Check if there are any externally stored columns.
			For each externally stored column, skip the
1577
			BTR_EXTERN_FIELD_REF. */
1578 1579

			for (i = 0; i < rec_offs_n_fields(offsets); i++) {
1580 1581 1582 1583
				byte*	dst;

				if (UNIV_UNLIKELY(i == trx_id_col)) {
					/* Skip trx_id and roll_ptr */
1584 1585
					dst = rec_get_nth_field(rec, offsets,
								i, &len);
1586
					if (UNIV_UNLIKELY(dst - next_out
1587 1588 1589 1590
							  >= end - data)
					    || UNIV_UNLIKELY
					    (len < (DATA_TRX_ID_LEN
						    + DATA_ROLL_PTR_LEN))
1591 1592
					    || rec_offs_nth_extern(offsets,
								   i)) {
1593 1594 1595 1596 1597 1598
						return(NULL);
					}

					memcpy(next_out, data, dst - next_out);
					data += dst - next_out;
					next_out = dst + (DATA_TRX_ID_LEN
1599
							  + DATA_ROLL_PTR_LEN);
1600
				} else if (rec_offs_nth_extern(offsets, i)) {
1601 1602
					dst = rec_get_nth_field(rec, offsets,
								i, &len);
1603 1604
					ut_ad(len
					      >= BTR_EXTERN_FIELD_REF_SIZE);
1605 1606 1607 1608 1609 1610 1611

					len += dst - next_out
						- BTR_EXTERN_FIELD_REF_SIZE;

					if (UNIV_UNLIKELY(data + len >= end)) {
						return(NULL);
					}
1612

1613 1614 1615 1616 1617 1618 1619
					memcpy(next_out, data, len);
					data += len;
					next_out += len
						+ BTR_EXTERN_FIELD_REF_SIZE;
				}
			}

1620 1621
			/* Copy the last bytes of the record. */
			len = rec_get_end(rec, offsets) - next_out;
1622 1623 1624 1625 1626 1627
			if (UNIV_UNLIKELY(data + len >= end)) {
				return(NULL);
			}
			memcpy(next_out, data, len);
			data += len;
		}
1628 1629 1630
	}
}

marko's avatar
marko committed
1631
/**************************************************************************
1632 1633
Decompress the records of a node pointer page. */
static
marko's avatar
marko committed
1634
ibool
1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646
page_zip_decompress_node_ptrs(
/*==========================*/
					/* out: TRUE on success,
					FALSE on failure */
	page_zip_des_t*	page_zip,	/* in/out: compressed page */
	z_stream*	d_stream,	/* in/out: compressed page stream */
	rec_t**		recs,		/* in: dense page directory
					sorted by address */
	ulint		n_dense,	/* in: size of recs[] */
	dict_index_t*	index,		/* in: the index of the page */
	ulint*		offsets,	/* in/out: temporary offsets */
	mem_heap_t*	heap)		/* in: temporary memory heap */
marko's avatar
marko committed
1647
{
1648 1649
	ulint		heap_status = REC_STATUS_NODE_PTR
		| 2 << REC_HEAP_NO_SHIFT;
1650
	ulint		slot;
1651
	const byte*	storage;
marko's avatar
marko committed
1652

1653 1654 1655
	/* Subtract the space reserved for uncompressed data. */
	d_stream->avail_in -= n_dense
		* (PAGE_ZIP_DIR_SLOT_SIZE + REC_NODE_PTR_SIZE);
marko's avatar
marko committed
1656

1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 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
	/* Decompress the records in heap_no order. */
	for (slot = 0; slot < n_dense; slot++) {
		rec_t*	rec = recs[slot];

		d_stream->avail_out = rec - REC_N_NEW_EXTRA_BYTES
			- d_stream->next_out;

		ut_ad(d_stream->avail_out < UNIV_PAGE_SIZE
		      - PAGE_ZIP_START - PAGE_DIR);
		switch (inflate(d_stream, Z_SYNC_FLUSH)) {
		case Z_STREAM_END:
			/* Apparently, n_dense has grown
			since the time the page was last compressed. */
			goto zlib_done;
		case Z_OK:
		case Z_BUF_ERROR:
			if (!d_stream->avail_out) {
				break;
			}
			/* fall through */
		default:
			goto zlib_error;
		}

		ut_ad(d_stream->next_out == rec - REC_N_NEW_EXTRA_BYTES);
		/* Prepare to decompress the data bytes. */
		d_stream->next_out = rec;
		/* Set heap_no and the status bits. */
		mach_write_to_2(rec - REC_NEW_HEAP_NO, heap_status);
		heap_status += 1 << REC_HEAP_NO_SHIFT;

		/* Read the offsets. The status bits are needed here. */
		offsets = rec_get_offsets(rec, index, offsets,
					  ULINT_UNDEFINED, &heap);

		/* Non-leaf nodes should not have any externally
		stored columns. */
		ut_ad(!rec_offs_any_extern(offsets));

		/* Decompress the data bytes, except node_ptr. */
		d_stream->avail_out = rec_offs_data_size(offsets)
			- REC_NODE_PTR_SIZE;

		switch (inflate(d_stream, Z_SYNC_FLUSH)) {
		case Z_STREAM_END:
			goto zlib_done;
		case Z_OK:
		case Z_BUF_ERROR:
			if (!d_stream->avail_out) {
				break;
			}
			/* fall through */
		default:
			goto zlib_error;
		}

		/* Clear the node pointer in case the record
		will be deleted and the space will be reallocated
		to a smaller record. */
		memset(d_stream->next_out, 0, REC_NODE_PTR_SIZE);
		d_stream->next_out += REC_NODE_PTR_SIZE;

		ut_ad(d_stream->next_out == rec_get_end(rec, offsets));
1720
	}
1721

1722 1723 1724 1725 1726 1727 1728
	/* Decompress any trailing garbage, in case the last record was
	allocated from an originally longer space on the free list. */
	d_stream->avail_out = page_header_get_field(page_zip->data,
						    PAGE_HEAP_TOP)
		- page_offset(d_stream->next_out);
	if (UNIV_UNLIKELY(d_stream->avail_out > UNIV_PAGE_SIZE
			  - PAGE_ZIP_START - PAGE_DIR)) {
1729

1730 1731
		goto zlib_error;
	}
1732

1733 1734 1735
	if (UNIV_UNLIKELY(inflate(d_stream, Z_FINISH) != Z_STREAM_END)) {
zlib_error:
		inflateEnd(d_stream);
1736 1737 1738
		return(FALSE);
	}

1739 1740 1741 1742 1743 1744
	/* Note that d_stream->avail_out > 0 may hold here
	if the modification log is nonempty. */

zlib_done:
	if (UNIV_UNLIKELY(inflateEnd(d_stream) != Z_OK)) {
		ut_error;
1745 1746
	}

1747 1748
	{
		page_t*	page = page_align(d_stream->next_out);
marko's avatar
marko committed
1749

1750 1751 1752 1753 1754
		/* Clear the unused heap space on the uncompressed page. */
		memset(d_stream->next_out, 0,
		       page_dir_get_nth_slot(page,
					     page_dir_get_n_slots(page) - 1)
		       - d_stream->next_out);
marko's avatar
marko committed
1755
	}
marko's avatar
marko committed
1756

1757
	page_zip->m_start = PAGE_DATA + d_stream->total_in;
1758

1759 1760 1761 1762 1763 1764 1765 1766 1767
	/* Apply the modification log. */
	{
		const byte*	mod_log_ptr;
		mod_log_ptr = page_zip_apply_log(page_zip->data
						 + page_zip->m_start,
						 d_stream->avail_in + 1,
						 recs, n_dense,
						 ULINT_UNDEFINED, heap_status,
						 index, offsets);
1768

1769 1770 1771 1772 1773 1774 1775
		if (UNIV_UNLIKELY(!mod_log_ptr)) {
			return(FALSE);
		}
		page_zip->m_end = mod_log_ptr - page_zip->data;
		ut_a(page_zip_get_trailer_len(page_zip, index, NULL)
		     + page_zip->m_end < page_zip->size);
	}
1776

1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792
	/* Restore the uncompressed columns in heap_no order. */
	storage	= page_zip->data + page_zip->size
		- n_dense * PAGE_ZIP_DIR_SLOT_SIZE;

	for (slot = 0; slot < n_dense; slot++) {
		rec_t*		rec	= recs[slot];

		offsets = rec_get_offsets(rec, index, offsets,
					  ULINT_UNDEFINED, &heap);
		/* Non-leaf nodes should not have any externally
		stored columns. */
		ut_ad(!rec_offs_any_extern(offsets));
		storage -= REC_NODE_PTR_SIZE;

		memcpy(rec_get_end(rec, offsets) - REC_NODE_PTR_SIZE,
		       storage, REC_NODE_PTR_SIZE);
1793 1794
	}

1795 1796
	return(TRUE);
}
1797

1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862
/**************************************************************************
Decompress the records of a leaf node of a secondary index. */
static
ibool
page_zip_decompress_sec(
/*====================*/
					/* out: TRUE on success,
					FALSE on failure */
	page_zip_des_t*	page_zip,	/* in/out: compressed page */
	z_stream*	d_stream,	/* in/out: compressed page stream */
	rec_t**		recs,		/* in: dense page directory
					sorted by address */
	ulint		n_dense,	/* in: size of recs[] */
	dict_index_t*	index,		/* in: the index of the page */
	ulint*		offsets)	/* in/out: temporary offsets */
{
	ulint	heap_status	= REC_STATUS_ORDINARY | 2 << REC_HEAP_NO_SHIFT;
	ulint	slot;

	/* Subtract the space reserved for uncompressed data. */
	d_stream->avail_in -= n_dense * PAGE_ZIP_DIR_SLOT_SIZE;

	for (slot = 0; slot < n_dense; slot++) {
		rec_t*	rec = recs[slot];

		/* Decompress everything up to this record. */
		d_stream->avail_out = rec - REC_N_NEW_EXTRA_BYTES
			- d_stream->next_out;

		if (UNIV_LIKELY(d_stream->avail_out)) {
			switch (inflate(d_stream, Z_SYNC_FLUSH)) {
			case Z_STREAM_END:
				/* Apparently, n_dense has grown
				since the time the page was last compressed. */
				goto zlib_done;
			case Z_OK:
			case Z_BUF_ERROR:
				if (!d_stream->avail_out) {
					break;
				}
				/* fall through */
			default:
				goto zlib_error;
			}
		}

		ut_ad(d_stream->next_out == rec - REC_N_NEW_EXTRA_BYTES);

		/* Skip the REC_N_NEW_EXTRA_BYTES. */

		d_stream->next_out = rec;

		/* Set heap_no and the status bits. */
		mach_write_to_2(rec - REC_NEW_HEAP_NO, heap_status);
		heap_status += 1 << REC_HEAP_NO_SHIFT;
	}

	/* Decompress the data of the last record and any trailing garbage,
	in case the last record was allocated from an originally longer space
	on the free list. */
	d_stream->avail_out = page_header_get_field(page_zip->data,
						    PAGE_HEAP_TOP)
		- page_offset(d_stream->next_out);
	if (UNIV_UNLIKELY(d_stream->avail_out > UNIV_PAGE_SIZE
			  - PAGE_ZIP_START - PAGE_DIR)) {
1863 1864 1865 1866

		goto zlib_error;
	}

1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879
	if (UNIV_UNLIKELY(inflate(d_stream, Z_FINISH) != Z_STREAM_END)) {
zlib_error:
		inflateEnd(d_stream);
		return(FALSE);
	}

	/* Note that d_stream->avail_out > 0 may hold here
	if the modification log is nonempty. */

zlib_done:
	if (UNIV_UNLIKELY(inflateEnd(d_stream) != Z_OK)) {
		ut_error;
	}
1880 1881

	{
1882
		page_t*	page = page_align(d_stream->next_out);
1883

1884 1885 1886 1887 1888 1889
		/* Clear the unused heap space on the uncompressed page. */
		memset(d_stream->next_out, 0,
		       page_dir_get_nth_slot(page,
					     page_dir_get_n_slots(page) - 1)
		       - d_stream->next_out);
	}
1890

1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908
	page_zip->m_start = PAGE_DATA + d_stream->total_in;

	/* Apply the modification log. */
	{
		const byte*	mod_log_ptr;
		mod_log_ptr = page_zip_apply_log(page_zip->data
						 + page_zip->m_start,
						 d_stream->avail_in + 1,
						 recs, n_dense,
						 ULINT_UNDEFINED, heap_status,
						 index, offsets);

		if (UNIV_UNLIKELY(!mod_log_ptr)) {
			return(FALSE);
		}
		page_zip->m_end = mod_log_ptr - page_zip->data;
		ut_a(page_zip_get_trailer_len(page_zip, index, NULL)
		     + page_zip->m_end < page_zip->size);
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
	/* There are no uncompressed columns on leaf pages of
	secondary indexes. */

	return(TRUE);
}

/**************************************************************************
Compress the records of a leaf node of a clustered index. */
static
ibool
page_zip_decompress_clust(
/*======================*/
					/* out: TRUE on success,
					FALSE on failure */
	page_zip_des_t*	page_zip,	/* in/out: compressed page */
	z_stream*	d_stream,	/* in/out: compressed page stream */
	rec_t**		recs,		/* in: dense page directory
					sorted by address */
	ulint		n_dense,	/* in: size of recs[] */
	dict_index_t*	index,		/* in: the index of the page */
	ulint		trx_id_col,	/* index of the trx_id column */
	ulint*		offsets,	/* in/out: temporary offsets */
	mem_heap_t*	heap)		/* in: temporary memory heap */
{
	int		err;
	ulint		slot;
	ulint		heap_status	= REC_STATUS_ORDINARY
		| 2 << REC_HEAP_NO_SHIFT;
	const byte*	storage;
	const byte*	externs;

	/* Subtract the space reserved for uncompressed data. */
	d_stream->avail_in -= n_dense * (PAGE_ZIP_DIR_SLOT_SIZE
					 + DATA_TRX_ID_LEN
					 + DATA_ROLL_PTR_LEN);

1947
	/* Decompress the records in heap_no order. */
1948
	for (slot = 0; slot < n_dense; slot++) {
1949
		rec_t*	rec	= recs[slot];
1950

1951 1952
		d_stream->avail_out = rec - REC_N_NEW_EXTRA_BYTES
			- d_stream->next_out;
1953

1954
		ut_ad(d_stream->avail_out < UNIV_PAGE_SIZE
1955
		      - PAGE_ZIP_START - PAGE_DIR);
1956 1957
		err = inflate(d_stream, Z_SYNC_FLUSH);
		switch (err) {
1958 1959
		case Z_STREAM_END:
			/* Apparently, n_dense has grown
1960
			since the time the page was last compressed. */
1961
			goto zlib_done;
1962
		case Z_OK:
1963
		case Z_BUF_ERROR:
1964
			if (UNIV_LIKELY(!d_stream->avail_out)) {
1965 1966
				break;
			}
1967
			/* fall through */
1968 1969 1970 1971
		default:
			goto zlib_error;
		}

1972
		ut_ad(d_stream->next_out == rec - REC_N_NEW_EXTRA_BYTES);
1973
		/* Prepare to decompress the data bytes. */
1974
		d_stream->next_out = rec;
1975 1976 1977
		/* Set heap_no and the status bits. */
		mach_write_to_2(rec - REC_NEW_HEAP_NO, heap_status);
		heap_status += 1 << REC_HEAP_NO_SHIFT;
1978

1979 1980
		/* Read the offsets. The status bits are needed here. */
		offsets = rec_get_offsets(rec, index, offsets,
1981
					  ULINT_UNDEFINED, &heap);
1982

1983 1984
		/* This is a leaf page in a clustered index. */
		ulint	i;
1985

1986 1987 1988
		/* Check if there are any externally stored columns.
		For each externally stored column, restore the
		BTR_EXTERN_FIELD_REF separately. */
1989

1990 1991 1992
		for (i = 0; i < rec_offs_n_fields(offsets); i++) {
			ulint	len;
			byte*	dst;
1993

1994 1995 1996 1997 1998 1999
			if (UNIV_UNLIKELY(i == trx_id_col)) {
				/* Skip trx_id and roll_ptr */
				dst = rec_get_nth_field(rec, offsets, i, &len);
				if (UNIV_UNLIKELY(len < DATA_TRX_ID_LEN
						  + DATA_ROLL_PTR_LEN)
				    || rec_offs_nth_extern(offsets, i)) {
2000

2001 2002
					goto zlib_error;
				}
2003

2004
				d_stream->avail_out = dst - d_stream->next_out;
2005

2006 2007 2008 2009 2010 2011
				switch (inflate(d_stream, Z_SYNC_FLUSH)) {
				case Z_STREAM_END:
				case Z_OK:
				case Z_BUF_ERROR:
					if (!d_stream->avail_out) {
						break;
2012
					}
2013 2014 2015 2016
					/* fall through */
				default:
					goto zlib_error;
				}
2017

2018
				ut_ad(d_stream->next_out == dst);
2019

2020 2021 2022 2023
				d_stream->next_out += DATA_TRX_ID_LEN
					+ DATA_ROLL_PTR_LEN;
			} else if (rec_offs_nth_extern(offsets, i)) {
				dst = rec_get_nth_field(rec, offsets, i, &len);
2024
				ut_ad(len >= BTR_EXTERN_FIELD_REF_SIZE);
2025
				dst += len - BTR_EXTERN_FIELD_REF_SIZE;
2026

2027 2028 2029 2030 2031 2032 2033 2034
				d_stream->avail_out = dst - d_stream->next_out;
				switch (inflate(d_stream,
						Z_SYNC_FLUSH)) {
				case Z_STREAM_END:
				case Z_OK:
				case Z_BUF_ERROR:
					if (!d_stream->avail_out) {
						break;
2035
					}
2036 2037 2038 2039
					/* fall through */
				default:
					goto zlib_error;
				}
2040

2041
				ut_ad(d_stream->next_out == dst);
2042

2043 2044 2045 2046
				/* Reserve space for the data at
				the end of the space reserved for
				the compressed data and the
				page modification log. */
2047

2048 2049 2050 2051 2052
				if (UNIV_UNLIKELY
				    (d_stream->avail_in
				     <= BTR_EXTERN_FIELD_REF_SIZE)) {
					/* out of space */
					goto zlib_error;
2053
				}
2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069

				/* Clear the BLOB pointer in case
				the record will be deleted and the
				space will not be reused.  Note that
				the final initialization of the BLOB
				pointers (copying from "externs"
				or clearing) will have to take place
				only after the page modification log
				has been applied.  Otherwise, we
				could end up with an uninitialized
				BLOB pointer when a record is deleted,
				reallocated and deleted. */
				memset(d_stream->next_out, 0,
				       BTR_EXTERN_FIELD_REF_SIZE);
				d_stream->next_out
					+= BTR_EXTERN_FIELD_REF_SIZE;
2070
			}
2071
		}
marko's avatar
marko committed
2072

2073 2074 2075
		/* Decompress the last bytes of the record. */
		d_stream->avail_out = rec_get_end(rec, offsets)
			- d_stream->next_out;
2076

2077 2078 2079 2080 2081 2082
		switch (inflate(d_stream, Z_SYNC_FLUSH)) {
		case Z_STREAM_END:
		case Z_OK:
		case Z_BUF_ERROR:
			if (!d_stream->avail_out) {
				break;
2083
			}
2084 2085 2086
			/* fall through */
		default:
			goto zlib_error;
2087
		}
2088 2089
	}

2090 2091
	/* Decompress any trailing garbage, in case the last record was
	allocated from an originally longer space on the free list. */
2092 2093 2094 2095
	d_stream->avail_out = page_header_get_field(page_zip->data,
						    PAGE_HEAP_TOP)
		- page_offset(d_stream->next_out);
	if (UNIV_UNLIKELY(d_stream->avail_out > UNIV_PAGE_SIZE
2096
			  - PAGE_ZIP_START - PAGE_DIR)) {
2097 2098 2099 2100

		goto zlib_error;
	}

2101
	if (UNIV_UNLIKELY(inflate(d_stream, Z_FINISH) != Z_STREAM_END)) {
2102
zlib_error:
2103 2104
		inflateEnd(d_stream);
		return(FALSE);
2105 2106
	}

2107
	/* Note that d_stream->avail_out > 0 may hold here
2108
	if the modification log is nonempty. */
2109 2110

zlib_done:
2111
	if (UNIV_UNLIKELY(inflateEnd(d_stream) != Z_OK)) {
marko's avatar
marko committed
2112 2113
		ut_error;
	}
2114

2115 2116
	{
		page_t*	page = page_align(d_stream->next_out);
2117

2118 2119 2120 2121 2122 2123
		/* Clear the unused heap space on the uncompressed page. */
		memset(d_stream->next_out, 0,
		       page_dir_get_nth_slot(page,
					     page_dir_get_n_slots(page) - 1)
		       - d_stream->next_out);
	}
2124

2125
	page_zip->m_start = PAGE_DATA + d_stream->total_in;
marko's avatar
marko committed
2126 2127

	/* Apply the modification log. */
2128 2129
	{
		const byte*	mod_log_ptr;
2130 2131
		mod_log_ptr = page_zip_apply_log(page_zip->data
						 + page_zip->m_start,
2132
						 d_stream->avail_in + 1,
2133 2134 2135
						 recs, n_dense,
						 trx_id_col, heap_status,
						 index, offsets);
2136

2137
		if (UNIV_UNLIKELY(!mod_log_ptr)) {
2138
			return(FALSE);
marko's avatar
marko committed
2139
		}
2140
		page_zip->m_end = mod_log_ptr - page_zip->data;
2141
		ut_a(page_zip_get_trailer_len(page_zip, index, NULL)
2142
		     + page_zip->m_end < page_zip->size);
marko's avatar
marko committed
2143 2144
	}

2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 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
	storage = page_zip->data + page_zip->size
		- n_dense * PAGE_ZIP_DIR_SLOT_SIZE;

	externs = storage - n_dense
		* (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);

	/* Restore the uncompressed columns in heap_no order. */

	for (slot = 0; slot < n_dense; slot++) {
		ulint	i;
		ulint	len;
		byte*	dst;
		rec_t*	rec	= recs[slot];
		ibool	exists	= !page_zip_dir_find_free(
			page_zip, page_offset(rec));
		offsets = rec_get_offsets(rec, index, offsets,
					  ULINT_UNDEFINED, &heap);

		dst = rec_get_nth_field(rec, offsets,
					trx_id_col, &len);
		ut_ad(len >= DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
		storage -= DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN;
		memcpy(dst, storage,
		       DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);

		/* Check if there are any externally stored
		columns in this record.  For each externally
		stored column, restore or clear the
		BTR_EXTERN_FIELD_REF. */

		for (i = 0; i < rec_offs_n_fields(offsets); i++) {
			if (!rec_offs_nth_extern(offsets, i)) {
				continue;
			}
			dst = rec_get_nth_field(rec, offsets, i, &len);
			ut_ad(len >= BTR_EXTERN_FIELD_REF_SIZE);
			dst += len - BTR_EXTERN_FIELD_REF_SIZE;

			if (UNIV_LIKELY(exists)) {
				/* Existing record:
				restore the BLOB pointer */
				externs -= BTR_EXTERN_FIELD_REF_SIZE;

				memcpy(dst, externs,
				       BTR_EXTERN_FIELD_REF_SIZE);

				page_zip->n_blobs++;
			} else {
				/* Deleted record:
				clear the BLOB pointer */
				memset(dst, 0,
				       BTR_EXTERN_FIELD_REF_SIZE);
			}
		}
	}

	return(TRUE);
}

/**************************************************************************
Decompress a page.  This function should tolerate errors on the compressed
page.  Instead of letting assertions fail, it will return FALSE if an
inconsistency is detected. */

ibool
page_zip_decompress(
/*================*/
				/* out: TRUE on success, FALSE on failure */
	page_zip_des_t*	page_zip,/* in: data, size;
				out: m_start, m_end, n_blobs */
	page_t*		page)	/* out: uncompressed page, may be trashed */
{
	z_stream	d_stream;
	dict_index_t*	index	= NULL;
	rec_t**		recs;	/* dense page directory, sorted by address */
	ulint		n_dense;/* number of user records on the page */
	ulint		trx_id_col = ULINT_UNDEFINED;
	mem_heap_t*	heap;
	ulint*		offsets;

	ut_ad(page_zip_simple_validate(page_zip));

	/* The dense directory excludes the infimum and supremum records. */
	n_dense = page_dir_get_n_heap(page_zip->data) - 2;
	if (UNIV_UNLIKELY(n_dense * PAGE_ZIP_DIR_SLOT_SIZE
			  >= page_zip->size)) {
		return(FALSE);
	}

	heap = mem_heap_create(n_dense * (3 * sizeof *recs));
	recs = mem_heap_alloc(heap, n_dense * (2 * sizeof *recs));

#ifdef UNIV_ZIP_DEBUG
	/* Clear the page. */
	memset(page, 0x55, UNIV_PAGE_SIZE);
#endif /* UNIV_ZIP_DEBUG */
	/* Copy the page header. */
	memcpy(page, page_zip->data, PAGE_DATA);

	/* Copy the page directory. */
	if (UNIV_UNLIKELY(!page_zip_dir_decode(page_zip, page, recs,
					       recs + n_dense, n_dense))) {
zlib_error:
2248 2249 2250 2251
		mem_heap_free(heap);
		return(FALSE);
	}

2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265
	/* Copy the infimum and supremum records. */
	memcpy(page + (PAGE_NEW_INFIMUM - REC_N_NEW_EXTRA_BYTES),
	       infimum_extra, sizeof infimum_extra);
	if (UNIV_UNLIKELY(!page_get_n_recs(page))) {
		rec_set_next_offs_new(page + PAGE_NEW_INFIMUM,
				      PAGE_NEW_SUPREMUM);
	} else {
		rec_set_next_offs_new(page + PAGE_NEW_INFIMUM,
				      page_zip_dir_get(page_zip, 0)
				      & PAGE_ZIP_DIR_SLOT_MASK);
	}
	memcpy(page + PAGE_NEW_INFIMUM, infimum_data, sizeof infimum_data);
	memcpy(page + (PAGE_NEW_SUPREMUM - REC_N_NEW_EXTRA_BYTES + 1),
	       supremum_extra_data, sizeof supremum_extra_data);
2266

2267 2268 2269
	d_stream.zalloc = page_zip_malloc;
	d_stream.zfree = page_zip_free;
	d_stream.opaque = (voidpf) 0;
2270

2271 2272
	if (UNIV_UNLIKELY(inflateInit(&d_stream) != Z_OK)) {
		ut_error;
2273
	}
2274

2275 2276 2277 2278
	d_stream.next_in = page_zip->data + PAGE_DATA;
	/* Subtract the space reserved for
	the page header and the end marker of the modification log. */
	d_stream.avail_in = page_zip->size - (PAGE_DATA + 1);
2279

2280 2281
	d_stream.next_out = page + PAGE_ZIP_START;
	d_stream.avail_out = UNIV_PAGE_SIZE - PAGE_ZIP_START;
2282

2283 2284 2285
	/* Decode the zlib header and the index information. */
	if (UNIV_UNLIKELY(inflate(&d_stream, Z_BLOCK) != Z_OK)
	    || UNIV_UNLIKELY(inflate(&d_stream, Z_BLOCK) != Z_OK)) {
2286

2287 2288
		goto zlib_error;
	}
2289

2290 2291 2292
	index = page_zip_fields_decode(
		page + PAGE_ZIP_START, d_stream.next_out,
		page_is_leaf(page) ? &trx_id_col : NULL);
2293

2294
	if (UNIV_UNLIKELY(!index)) {
2295

2296 2297
		goto zlib_error;
	}
2298

2299 2300 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
	/* Decompress the user records. */
	page_zip->n_blobs = 0;
	d_stream.next_out = page + PAGE_ZIP_START;

	{
		/* Pre-allocate the offsets for rec_get_offsets_reverse(). */
		ulint	n = 1 + 1/* node ptr */ + REC_OFFS_HEADER_SIZE
			+ dict_index_get_n_fields(index);
		offsets = mem_heap_alloc(heap, n * sizeof(ulint));
		*offsets = n;
	}

	/* Decompress the records in heap_no order. */
	if (!page_is_leaf(page)) {
		/* This is a node pointer page. */
		ulint	info_bits;

		if (UNIV_UNLIKELY
		    (!page_zip_decompress_node_ptrs(page_zip, &d_stream,
						    recs, n_dense, index,
						    offsets, heap))) {
			goto err_exit;
		}

		info_bits = mach_read_from_4(page + FIL_PAGE_PREV) == FIL_NULL
			? REC_INFO_MIN_REC_FLAG : 0;

		if (UNIV_UNLIKELY(!page_zip_set_extra_bytes(page_zip, page,
							    info_bits))) {
			goto err_exit;
		}
	} else if (UNIV_LIKELY(trx_id_col == ULINT_UNDEFINED)) {
		/* This is a leaf page in a secondary index. */
		if (UNIV_UNLIKELY(!page_zip_decompress_sec(page_zip, &d_stream,
							   recs, n_dense,
							   index, offsets))) {
			goto err_exit;
		}
2337

2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353
		if (UNIV_UNLIKELY(!page_zip_set_extra_bytes(page_zip,
							    page, 0))) {
err_exit:
			page_zip_fields_free(index);
			mem_heap_free(heap);
			return(FALSE);
		}
	} else {
		/* This is a leaf page in a clustered index. */
		if (UNIV_UNLIKELY(!page_zip_decompress_clust(page_zip,
							     &d_stream, recs,
							     n_dense, index,
							     trx_id_col,
							     offsets, heap))) {
			goto err_exit;
		}
2354

2355 2356 2357 2358
		if (UNIV_UNLIKELY(!page_zip_set_extra_bytes(page_zip,
							    page, 0))) {
			goto err_exit;
		}
2359 2360
	}

marko's avatar
marko committed
2361 2362
	ut_a(page_is_comp(page));

2363 2364 2365
	page_zip_fields_free(index);
	mem_heap_free(heap);

marko's avatar
marko committed
2366 2367 2368
	return(TRUE);
}

2369
#ifdef UNIV_ZIP_DEBUG
2370 2371 2372
/* Flag: make page_zip_validate() compare page headers only */
ibool	page_zip_validate_header_only = FALSE;

marko's avatar
marko committed
2373 2374 2375 2376 2377 2378
/**************************************************************************
Check that the compressed and decompressed pages match. */

ibool
page_zip_validate(
/*==============*/
2379
					/* out: TRUE if valid, FALSE if not */
2380 2381
	const page_zip_des_t*	page_zip,/* in: compressed page */
	const page_t*		page)	/* in: uncompressed page */
marko's avatar
marko committed
2382 2383
{
	page_zip_des_t	temp_page_zip = *page_zip;
2384
	byte*		temp_page_buf;
2385
	page_t*		temp_page;
2386
	ibool		valid;
marko's avatar
marko committed
2387

2388
	ut_a(buf_frame_get_page_zip((byte*) page) == page_zip);
2389
	ut_a(page_is_comp((page_t*) page));
marko's avatar
marko committed
2390

2391
	if (memcmp(page_zip->data + FIL_PAGE_PREV, page + FIL_PAGE_PREV,
2392
		   FIL_PAGE_LSN - FIL_PAGE_PREV)
2393 2394
	    || memcmp(page_zip->data + FIL_PAGE_TYPE, page + FIL_PAGE_TYPE, 2)
	    || memcmp(page_zip->data + FIL_PAGE_DATA, page + FIL_PAGE_DATA,
2395
		      PAGE_DATA - FIL_PAGE_DATA)) {
2396 2397 2398 2399 2400 2401 2402 2403
		fputs("page_zip_validate(): page header mismatch\n", stderr);
		return(FALSE);
	}

	if (page_zip_validate_header_only) {
		return(TRUE);
	}

2404 2405 2406 2407
	/* page_zip_decompress() expects the uncompressed page to be
	UNIV_PAGE_SIZE aligned. */
	temp_page_buf = ut_malloc(2 * UNIV_PAGE_SIZE);
	temp_page = ut_align(temp_page_buf, UNIV_PAGE_SIZE);
2408

2409
	valid = page_zip_decompress(&temp_page_zip, temp_page);
2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432
	if (!valid) {
		fputs("page_zip_validate(): failed to decompress\n", stderr);
		goto func_exit;
	}
	if (page_zip->n_blobs != temp_page_zip.n_blobs) {
		fprintf(stderr,
			"page_zip_validate(): n_blobs mismatch: %lu!=%lu\n",
			page_zip->n_blobs, temp_page_zip.n_blobs);
		valid = FALSE;
	}
	if (page_zip->m_start != temp_page_zip.m_start) {
		fprintf(stderr,
			"page_zip_validate(): m_start mismatch: %lu!=%lu\n",
			page_zip->m_start, temp_page_zip.m_start);
		valid = FALSE;
	}
	if (page_zip->m_end != temp_page_zip.m_end) {
		fprintf(stderr,
			"page_zip_validate(): m_end mismatch: %lu!=%lu\n",
			page_zip->m_end, temp_page_zip.m_end);
		valid = FALSE;
	}
	if (memcmp(page + PAGE_HEADER, temp_page + PAGE_HEADER,
2433
		   UNIV_PAGE_SIZE - PAGE_HEADER - FIL_PAGE_DATA_END)) {
2434
		fputs("page_zip_validate(): content mismatch\n", stderr);
2435 2436 2437 2438
		valid = FALSE;
	}

func_exit:
2439
	ut_free(temp_page_buf);
2440
	return(valid);
marko's avatar
marko committed
2441
}
2442
#endif /* UNIV_ZIP_DEBUG */
marko's avatar
marko committed
2443

2444 2445 2446 2447 2448 2449 2450 2451 2452 2453
#ifdef UNIV_DEBUG
static
ibool
page_zip_header_cmp(
/*================*/
					/* out: TRUE */
	const page_zip_des_t*	page_zip,/* in: compressed page */
	const byte*		page)	/* in: uncompressed page */
{
	ut_ad(!memcmp(page_zip->data + FIL_PAGE_PREV, page + FIL_PAGE_PREV,
2454
		      FIL_PAGE_LSN - FIL_PAGE_PREV));
2455
	ut_ad(!memcmp(page_zip->data + FIL_PAGE_TYPE, page + FIL_PAGE_TYPE,
2456
		      2));
2457
	ut_ad(!memcmp(page_zip->data + FIL_PAGE_DATA, page + FIL_PAGE_DATA,
2458
		      PAGE_DATA - FIL_PAGE_DATA));
2459 2460 2461 2462 2463

	return(TRUE);
}
#endif /* UNIV_DEBUG */

marko's avatar
marko committed
2464
/**************************************************************************
2465
Write an entire record on the compressed page.  The data must already
marko's avatar
marko committed
2466 2467 2468
have been written to the uncompressed page. */

void
2469 2470
page_zip_write_rec(
/*===============*/
marko's avatar
marko committed
2471
	page_zip_des_t*	page_zip,/* in/out: compressed page */
2472
	const byte*	rec,	/* in: record being written */
2473
	dict_index_t*	index,	/* in: the index the record belongs to */
2474 2475
	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
	ulint		create)	/* in: nonzero=insert, zero=update */
marko's avatar
marko committed
2476
{
2477 2478 2479
	page_t*	page;
	byte*	data;
	byte*	storage;
2480
	ulint	heap_no;
marko's avatar
marko committed
2481
	byte*	slot;
marko's avatar
marko committed
2482

2483
	ut_ad(buf_frame_get_page_zip((byte*) rec) == page_zip);
marko's avatar
marko committed
2484
	ut_ad(page_zip_simple_validate(page_zip));
2485 2486
	ut_ad(page_zip->size > PAGE_DATA + page_zip_dir_size(page_zip));
	ut_ad(rec_offs_comp(offsets));
2487
	ut_ad(rec_offs_validate((rec_t*) rec, index, offsets));
marko's avatar
marko committed
2488

2489
	ut_ad(page_zip->m_start >= PAGE_DATA);
marko's avatar
marko committed
2490

2491
	page = page_align((rec_t*) rec);
2492 2493

	ut_ad(page_zip_header_cmp(page_zip, page));
2494
	ut_ad(page_simple_validate_new(page));
2495

2496
	slot = page_zip_dir_find(page_zip, page_offset(rec));
marko's avatar
marko committed
2497
	ut_a(slot);
2498 2499 2500 2501 2502 2503
	/* Copy the delete mark. */
	if (rec_get_deleted_flag((rec_t*) rec, TRUE)) {
		*slot |= PAGE_ZIP_DIR_SLOT_DEL >> 8;
	} else {
		*slot &= ~(PAGE_ZIP_DIR_SLOT_DEL >> 8);
	}
marko's avatar
marko committed
2504

2505 2506
	ut_ad(rec_get_start((rec_t*) rec, offsets) >= page + PAGE_ZIP_START);
	ut_ad(rec_get_end((rec_t*) rec, offsets) <= page + UNIV_PAGE_SIZE
2507 2508
	      - PAGE_DIR - PAGE_DIR_SLOT_SIZE
	      * page_dir_get_n_slots(page));
marko's avatar
marko committed
2509

2510 2511 2512 2513
	heap_no = rec_get_heap_no_new((rec_t*) rec);
	ut_ad(heap_no >= 2); /* not infimum or supremum */
	ut_ad(heap_no < page_dir_get_n_heap(page));

marko's avatar
marko committed
2514
	/* Append to the modification log. */
2515
	data = page_zip->data + page_zip->m_end;
2516 2517 2518 2519 2520
	ut_ad(!*data);

	/* Identify the record by writing its heap number - 1.
	0 is reserved to indicate the end of the modification log. */

2521 2522
	if (UNIV_UNLIKELY(heap_no - 1 >= 64)) {
		*data++ = 0x80 | (heap_no - 1) >> 7;
marko's avatar
marko committed
2523
		ut_ad(!*data);
2524
	}
2525
	*data++ = (heap_no - 1) << 1;
marko's avatar
marko committed
2526
	ut_ad(!*data);
2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538

	{
		const byte*	start	= rec_get_start((rec_t*) rec, offsets);
		const byte*	b	= rec - REC_N_NEW_EXTRA_BYTES;

		/* Write the extra bytes backwards, so that
		rec_offs_extra_size() can be easily computed in
		page_zip_apply_log() by invoking
		rec_get_offsets_reverse(). */

		while (b != start) {
			*data++ = *--b;
marko's avatar
marko committed
2539
			ut_ad(!*data);
2540 2541 2542 2543 2544
		}
	}

	/* Write the data bytes.  Store the uncompressed bytes separately. */
	storage = page_zip->data + page_zip->size
2545 2546
		- (page_dir_get_n_heap(page) - 2)
		* PAGE_ZIP_DIR_SLOT_SIZE;
2547 2548 2549

	if (page_is_leaf(page)) {
		ulint		len;
2550
		const byte*	start	= rec;
2551 2552

		if (dict_index_is_clust(index)) {
2553 2554 2555 2556 2557
			ulint		i;
			byte*		externs	= storage;
			ulint		trx_id_col;
			ulint		n_ext	= rec_offs_n_extern(offsets);

2558 2559
			trx_id_col = dict_index_get_sys_col_pos(index,
								DATA_TRX_ID);
2560
			ut_ad(trx_id_col != ULINT_UNDEFINED);
2561

2562
			externs -= (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN)
2563
				* (page_dir_get_n_heap(page) - 2);
2564 2565 2566 2567

			/* Note that this will not take into account
			the BLOB columns of rec if create==TRUE */
			ut_ad(data + rec_offs_data_size(offsets)
2568 2569 2570 2571 2572
			      - (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN)
			      - n_ext * BTR_EXTERN_FIELD_REF_SIZE
			      < externs
			      - BTR_EXTERN_FIELD_REF_SIZE
			      * page_zip->n_blobs);
2573

2574
			if (UNIV_UNLIKELY(n_ext)) {
2575 2576
				ulint	blob_no = page_zip_get_n_prev_extern(
					page_zip, rec, index);
2577
				byte*	ext_end = externs - page_zip->n_blobs
2578
					* BTR_EXTERN_FIELD_REF_SIZE;
2579 2580 2581 2582 2583
				ut_ad(blob_no <= page_zip->n_blobs);
				externs -= blob_no * BTR_EXTERN_FIELD_REF_SIZE;

				if (create) {
					page_zip->n_blobs += n_ext;
2584 2585 2586 2587 2588
					ut_ad(!memcmp
					      (ext_end - n_ext
					       * BTR_EXTERN_FIELD_REF_SIZE,
					       zero,
					       BTR_EXTERN_FIELD_REF_SIZE));
2589
					memmove(ext_end - n_ext
2590 2591 2592
						* BTR_EXTERN_FIELD_REF_SIZE,
						ext_end,
						externs - ext_end);
2593 2594 2595
				}

				ut_a(blob_no + n_ext <= page_zip->n_blobs);
2596 2597
			}

2598 2599
			/* Store separately trx_id, roll_ptr and
			the BTR_EXTERN_FIELD_REF of each BLOB column. */
2600

2601 2602 2603 2604
			for (i = 0; i < rec_offs_n_fields(offsets); i++) {
				const byte*	src;

				if (UNIV_UNLIKELY(i == trx_id_col)) {
2605 2606 2607 2608
					ut_ad(!rec_offs_nth_extern(offsets,
								   i));
					ut_ad(!rec_offs_nth_extern(offsets,
								   i + 1));
2609
					/* Locate trx_id and roll_ptr. */
2610 2611 2612
					src = rec_get_nth_field((rec_t*) rec,
								offsets,
								i, &len);
2613 2614
					ut_ad(len == DATA_TRX_ID_LEN);
					ut_ad(src + DATA_TRX_ID_LEN
2615 2616 2617
					      == rec_get_nth_field(
						      (rec_t*) rec, offsets,
						      i + 1, &len));
2618 2619 2620
					ut_ad(len == DATA_ROLL_PTR_LEN);

					/* Log the preceding fields. */
2621 2622 2623
					ut_ad(!memcmp(data, zero,
						      ut_min(src - start,
							     sizeof zero)));
2624 2625 2626
					memcpy(data, start, src - start);
					data += src - start;
					start = src + (DATA_TRX_ID_LEN
2627
						       + DATA_ROLL_PTR_LEN);
2628 2629 2630

					/* Store trx_id and roll_ptr. */
					memcpy(storage - (DATA_TRX_ID_LEN
2631 2632 2633 2634
							  + DATA_ROLL_PTR_LEN)
					       * (heap_no - 1),
					       src, DATA_TRX_ID_LEN
					       + DATA_ROLL_PTR_LEN);
2635 2636
					i++; /* skip also roll_ptr */
				} else if (rec_offs_nth_extern(offsets, i)) {
2637 2638 2639
					src = rec_get_nth_field((rec_t*) rec,
								offsets,
								i, &len);
2640 2641 2642

					ut_ad(dict_index_is_clust(index));
					ut_ad(len
2643
					      >= BTR_EXTERN_FIELD_REF_SIZE);
2644 2645
					src += len - BTR_EXTERN_FIELD_REF_SIZE;

2646 2647 2648
					ut_ad(!memcmp(data, zero,
						      ut_min(src - start,
							     sizeof zero)));
2649 2650 2651 2652
					memcpy(data, start, src - start);
					data += src - start;
					start = src
						+ BTR_EXTERN_FIELD_REF_SIZE;
2653

2654 2655 2656 2657
					/* Store the BLOB pointer. */
					externs -= BTR_EXTERN_FIELD_REF_SIZE;
					ut_ad(data < externs);
					memcpy(externs, src,
2658
					       BTR_EXTERN_FIELD_REF_SIZE);
2659
				}
2660
			}
2661 2662 2663 2664
		} else {
			/* Leaf page of a secondary index:
			no externally stored columns */
			ut_ad(dict_index_get_sys_col_pos(index, DATA_TRX_ID)
2665
			      == ULINT_UNDEFINED);
2666
			ut_ad(!rec_offs_any_extern(offsets));
2667 2668
		}

2669 2670
		/* Log the last bytes of the record. */
		len = rec_get_end((rec_t*) rec, offsets) - start;
2671

marko's avatar
marko committed
2672
		ut_ad(!memcmp(data, zero, ut_min(len, sizeof zero)));
2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684
		memcpy(data, start, len);
		data += len;
	} else {
		/* This is a node pointer page. */
		ulint	len;

		/* Non-leaf nodes should not have any externally
		stored columns. */
		ut_ad(!rec_offs_any_extern(offsets));

		/* Copy the data bytes, except node_ptr. */
		len = rec_offs_data_size(offsets) - REC_NODE_PTR_SIZE;
2685
		ut_ad(data + len < storage - REC_NODE_PTR_SIZE
2686
		      * (page_dir_get_n_heap(page) - 2));
marko's avatar
marko committed
2687
		ut_ad(!memcmp(data, zero, ut_min(len, sizeof zero)));
2688 2689 2690 2691 2692
		memcpy(data, rec, len);
		data += len;

		/* Copy the node pointer to the uncompressed area. */
		memcpy(storage - REC_NODE_PTR_SIZE
2693 2694 2695
		       * (heap_no - 1),
		       rec + len,
		       REC_NODE_PTR_SIZE);
2696 2697
	}

2698
	ut_a(!*data);
2699
	ut_ad((ulint) (data - page_zip->data) < page_zip->size);
2700
	page_zip->m_end = data - page_zip->data;
2701

2702
#ifdef UNIV_ZIP_DEBUG
2703
	ut_a(page_zip_validate(page_zip, page_align((byte*) rec)));
2704
#endif /* UNIV_ZIP_DEBUG */
2705 2706
}

2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723
/***************************************************************
Parses a log record of writing a BLOB pointer of a record. */

byte*
page_zip_parse_write_blob_ptr(
/*==========================*/
				/* out: end of log record or NULL */
	byte*		ptr,	/* in: redo log buffer */
	byte*		end_ptr,/* in: redo log buffer end */
	page_t*		page,	/* in/out: uncompressed page */
	page_zip_des_t*	page_zip)/* in/out: compressed page */
{
	ulint	offset;
	ulint	z_offset;

	ut_ad(!page == !page_zip);

2724 2725
	if (UNIV_UNLIKELY
	    (end_ptr < ptr + (2 + 2 + BTR_EXTERN_FIELD_REF_SIZE))) {
2726 2727 2728 2729 2730 2731 2732 2733

		return(NULL);
	}

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

	if (UNIV_UNLIKELY(offset < PAGE_ZIP_START)
2734 2735
	    || UNIV_UNLIKELY(offset >= UNIV_PAGE_SIZE)
	    || UNIV_UNLIKELY(z_offset >= UNIV_PAGE_SIZE)) {
2736 2737 2738 2739 2740 2741 2742
corrupt:
		recv_sys->found_corrupt_log = TRUE;

		return(NULL);
	}

	if (page) {
2743
		if (UNIV_UNLIKELY(!page_zip)
2744
		    || UNIV_UNLIKELY(!page_is_leaf(page))) {
2745 2746 2747 2748

			goto corrupt;
		}

2749
#ifdef UNIV_ZIP_DEBUG
2750
		ut_a(page_zip_validate(page_zip, page));
2751
#endif /* UNIV_ZIP_DEBUG */
2752

2753
		memcpy(page + offset,
2754
		       ptr + 4, BTR_EXTERN_FIELD_REF_SIZE);
2755
		memcpy(page_zip->data + z_offset,
2756
		       ptr + 4, BTR_EXTERN_FIELD_REF_SIZE);
2757

2758
#ifdef UNIV_ZIP_DEBUG
2759
		ut_a(page_zip_validate(page_zip, page));
2760
#endif /* UNIV_ZIP_DEBUG */
2761 2762 2763 2764 2765
	}

	return(ptr + (2 + 2 + BTR_EXTERN_FIELD_REF_SIZE));
}

2766
/**************************************************************************
2767
Write a BLOB pointer of a record on the leaf page of a clustered index.
2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782
The information must already have been updated on the uncompressed page. */

void
page_zip_write_blob_ptr(
/*====================*/
	page_zip_des_t*	page_zip,/* in/out: compressed page */
	const byte*	rec,	/* in/out: record whose data is being
				written */
	dict_index_t*	index,	/* in: index of the page */
	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
	ulint		n,	/* in: column index */
	mtr_t*		mtr)	/* in: mini-transaction handle,
				or NULL if no logging is needed */
{
	byte*	field;
2783
	byte*	externs;
2784
	page_t*	page	= page_align((byte*) rec);
2785 2786 2787
	ulint	blob_no;
	ulint	len;

2788
	ut_ad(buf_frame_get_page_zip((byte*) rec) == page_zip);
2789
	ut_ad(page_simple_validate_new(page));
2790 2791 2792 2793 2794 2795 2796
	ut_ad(page_zip_simple_validate(page_zip));
	ut_ad(page_zip->size > PAGE_DATA + page_zip_dir_size(page_zip));
	ut_ad(rec_offs_comp(offsets));
	ut_ad(rec_offs_validate((rec_t*) rec, NULL, offsets));
	ut_ad(rec_offs_nth_extern(offsets, n));

	ut_ad(page_zip->m_start >= PAGE_DATA);
2797
	ut_ad(page_zip_header_cmp(page_zip, page));
2798 2799

	ut_ad(page_is_leaf(page));
2800
	ut_ad(dict_index_is_clust(index));
2801

2802
	blob_no = page_zip_get_n_prev_extern(page_zip, rec, index)
2803
		+ rec_get_n_extern_new(rec, index, n);
2804
	ut_a(blob_no < page_zip->n_blobs);
2805 2806

	/* The heap number of the first user record is 2. */
2807
	externs = page_zip->data + page_zip->size
2808 2809 2810
		- (page_dir_get_n_heap(page) - 2)
		* (PAGE_ZIP_DIR_SLOT_SIZE
		   + DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
2811

2812 2813
	field = rec_get_nth_field((rec_t*) rec, offsets, n, &len);

2814 2815 2816 2817
	externs -= (blob_no + 1) * BTR_EXTERN_FIELD_REF_SIZE;
	field += len - BTR_EXTERN_FIELD_REF_SIZE;

	memcpy(externs, field, BTR_EXTERN_FIELD_REF_SIZE);
2818

2819
#ifdef UNIV_ZIP_DEBUG
2820
	ut_a(page_zip_validate(page_zip, page_align((rec_t*) rec)));
2821
#endif /* UNIV_ZIP_DEBUG */
2822

2823
	if (mtr) {
2824 2825
		byte*	log_ptr	= mlog_open(
			mtr, 11 + 2 + 2 + BTR_EXTERN_FIELD_REF_SIZE);
2826 2827 2828 2829
		if (UNIV_UNLIKELY(!log_ptr)) {
			return;
		}

2830 2831 2832
		log_ptr = mlog_write_initial_log_record_fast(
			(byte*) field, MLOG_ZIP_WRITE_BLOB_PTR, log_ptr, mtr);
		mach_write_to_2(log_ptr, page_offset(field));
2833
		log_ptr += 2;
2834 2835 2836 2837 2838
		mach_write_to_2(log_ptr, externs - page_zip->data);
		log_ptr += 2;
		memcpy(log_ptr, externs, BTR_EXTERN_FIELD_REF_SIZE);
		log_ptr += BTR_EXTERN_FIELD_REF_SIZE;
		mlog_close(mtr, log_ptr);
2839 2840 2841
	}
}

2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858
/***************************************************************
Parses a log record of writing the node pointer of a record. */

byte*
page_zip_parse_write_node_ptr(
/*==========================*/
				/* out: end of log record or NULL */
	byte*		ptr,	/* in: redo log buffer */
	byte*		end_ptr,/* in: redo log buffer end */
	page_t*		page,	/* in/out: uncompressed page */
	page_zip_des_t*	page_zip)/* in/out: compressed page */
{
	ulint	offset;
	ulint	z_offset;

	ut_ad(!page == !page_zip);

2859
	if (UNIV_UNLIKELY(end_ptr < ptr + (2 + 2 + REC_NODE_PTR_SIZE))) {
2860 2861 2862 2863 2864 2865 2866 2867

		return(NULL);
	}

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

	if (UNIV_UNLIKELY(offset < PAGE_ZIP_START)
2868 2869
	    || UNIV_UNLIKELY(offset >= UNIV_PAGE_SIZE)
	    || UNIV_UNLIKELY(z_offset >= UNIV_PAGE_SIZE)) {
2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881
corrupt:
		recv_sys->found_corrupt_log = TRUE;

		return(NULL);
	}

	if (page) {
		byte*	storage_end;
		byte*	field;
		byte*	storage;
		ulint	heap_no;

2882
		if (UNIV_UNLIKELY(!page_zip)
2883
		    || UNIV_UNLIKELY(page_is_leaf(page))) {
2884 2885 2886 2887

			goto corrupt;
		}

2888
#ifdef UNIV_ZIP_DEBUG
2889
		ut_a(page_zip_validate(page_zip, page));
2890
#endif /* UNIV_ZIP_DEBUG */
2891

2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910
		field = page + offset;
		storage = page_zip->data + z_offset;

		storage_end = page_zip->data + page_zip->size
			- (page_dir_get_n_heap(page) - 2)
			* PAGE_ZIP_DIR_SLOT_SIZE;

		heap_no = 1 + (storage_end - storage) / REC_NODE_PTR_SIZE;

		if (UNIV_UNLIKELY((storage_end - storage) % REC_NODE_PTR_SIZE)
		    || UNIV_UNLIKELY(heap_no < 2)
		    || UNIV_UNLIKELY(heap_no >= page_dir_get_n_heap(page))) {

			goto corrupt;
		}

		memcpy(field, ptr + 4, REC_NODE_PTR_SIZE);
		memcpy(storage, ptr + 4, REC_NODE_PTR_SIZE);

2911
#ifdef UNIV_ZIP_DEBUG
2912
		ut_a(page_zip_validate(page_zip, page));
2913
#endif /* UNIV_ZIP_DEBUG */
2914 2915
	}

2916
	return(ptr + (2 + 2 + REC_NODE_PTR_SIZE));
2917 2918
}

2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932
/**************************************************************************
Write the node pointer of a record on a non-leaf compressed page. */

void
page_zip_write_node_ptr(
/*====================*/
	page_zip_des_t*	page_zip,/* in/out: compressed page */
	byte*		rec,	/* in/out: record */
	ulint		size,	/* in: data size of rec */
	ulint		ptr,	/* in: node pointer */
	mtr_t*		mtr)	/* in: mini-transaction, or NULL */
{
	byte*	field;
	byte*	storage;
2933
	page_t*	page	= page_align(rec);
2934

2935
	ut_ad(buf_frame_get_page_zip(rec) == page_zip);
2936
	ut_ad(page_simple_validate_new(page));
2937 2938 2939 2940 2941
	ut_ad(page_zip_simple_validate(page_zip));
	ut_ad(page_zip->size > PAGE_DATA + page_zip_dir_size(page_zip));
	ut_ad(page_rec_is_comp(rec));

	ut_ad(page_zip->m_start >= PAGE_DATA);
2942
	ut_ad(page_zip_header_cmp(page_zip, page));
2943 2944 2945 2946 2947

	ut_ad(!page_is_leaf(page));

	/* The heap number of the first user record is 2. */
	storage = page_zip->data + page_zip->size
2948 2949 2950
		- (page_dir_get_n_heap(page) - 2)
		* PAGE_ZIP_DIR_SLOT_SIZE
		- (rec_get_heap_no_new(rec) - 1) * REC_NODE_PTR_SIZE;
2951 2952
	field = rec + size - REC_NODE_PTR_SIZE;

2953
#if defined UNIV_DEBUG || defined UNIV_ZIP_DEBUG
2954
	ut_a(!memcmp(storage, field, REC_NODE_PTR_SIZE));
2955
#endif /* UNIV_DEBUG || UNIV_ZIP_DEBUG */
2956 2957 2958 2959 2960 2961 2962
#if REC_NODE_PTR_SIZE != 4
# error "REC_NODE_PTR_SIZE != 4"
#endif
	mach_write_to_4(field, ptr);
	memcpy(storage, field, REC_NODE_PTR_SIZE);

	if (mtr) {
2963
		byte*	log_ptr	= mlog_open(mtr,
2964
					    11 + 2 + 2 + REC_NODE_PTR_SIZE);
2965 2966 2967 2968
		if (UNIV_UNLIKELY(!log_ptr)) {
			return;
		}

2969 2970 2971
		log_ptr = mlog_write_initial_log_record_fast(
			field, MLOG_ZIP_WRITE_NODE_PTR, log_ptr, mtr);
		mach_write_to_2(log_ptr, page_offset(field));
2972
		log_ptr += 2;
2973 2974
		mach_write_to_2(log_ptr, storage - page_zip->data);
		log_ptr += 2;
2975
		memcpy(log_ptr, field, REC_NODE_PTR_SIZE);
2976 2977
		log_ptr += REC_NODE_PTR_SIZE;
		mlog_close(mtr, log_ptr);
2978
	}
marko's avatar
marko committed
2979 2980 2981
}

/**************************************************************************
2982
Write the trx_id and roll_ptr of a record on a B-tree leaf node page. */
marko's avatar
marko committed
2983

2984
void
2985 2986
page_zip_write_trx_id_and_roll_ptr(
/*===============================*/
2987 2988
	page_zip_des_t*	page_zip,/* in/out: compressed page */
	byte*		rec,	/* in/out: record */
2989 2990
	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
	ulint		trx_id_col,/* in: column number of TRX_ID in rec */
2991
	dulint		trx_id,	/* in: transaction identifier */
2992
	dulint		roll_ptr)/* in: roll_ptr */
marko's avatar
marko committed
2993
{
2994 2995
	byte*	field;
	byte*	storage;
2996
	page_t*	page	= page_align(rec);
2997
	ulint	len;
2998

2999
	ut_ad(buf_frame_get_page_zip(rec) == page_zip);
3000
	ut_ad(page_simple_validate_new(page));
3001 3002
	ut_ad(page_zip_simple_validate(page_zip));
	ut_ad(page_zip->size > PAGE_DATA + page_zip_dir_size(page_zip));
3003 3004
	ut_ad(rec_offs_validate(rec, NULL, offsets));
	ut_ad(rec_offs_comp(offsets));
3005 3006

	ut_ad(page_zip->m_start >= PAGE_DATA);
3007
	ut_ad(page_zip_header_cmp(page_zip, page));
3008 3009 3010 3011 3012

	ut_ad(page_is_leaf(page));

	/* The heap number of the first user record is 2. */
	storage = page_zip->data + page_zip->size
3013 3014 3015 3016
		- (page_dir_get_n_heap(page) - 2)
		* PAGE_ZIP_DIR_SLOT_SIZE
		- (rec_get_heap_no_new(rec) - 1)
		* (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
3017 3018 3019 3020 3021 3022 3023

#if DATA_TRX_ID + 1 != DATA_ROLL_PTR
# error "DATA_TRX_ID + 1 != DATA_ROLL_PTR"
#endif
	field = rec_get_nth_field(rec, offsets, trx_id_col, &len);
	ut_ad(len == DATA_TRX_ID_LEN);
	ut_ad(field + DATA_TRX_ID_LEN
3024
	      == rec_get_nth_field(rec, offsets, trx_id_col + 1, &len));
3025
	ut_ad(len == DATA_ROLL_PTR_LEN);
3026 3027 3028 3029 3030 3031 3032 3033 3034 3035
#if defined UNIV_DEBUG || defined UNIV_ZIP_DEBUG
	ut_a(!memcmp(storage, field, DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN));
#endif /* UNIV_DEBUG || UNIV_ZIP_DEBUG */
#if DATA_TRX_ID_LEN != 6
# error "DATA_TRX_ID_LEN != 6"
#endif
	mach_write_to_6(field, trx_id);
#if DATA_ROLL_PTR_LEN != 7
# error "DATA_ROLL_PTR_LEN != 7"
#endif
marko's avatar
marko committed
3036
	mach_write_to_7(field + DATA_TRX_ID_LEN, roll_ptr);
3037
	memcpy(storage, field, DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
3038 3039
}

3040 3041 3042
#ifdef UNIV_ZIP_DEBUG
/* Set this variable in a debugger to disable page_zip_clear_rec().
The only observable effect should be the compression ratio due to
3043 3044 3045
deleted records not being zeroed out.  In rare cases, there can be
page_zip_validate() failures on the node_ptr, trx_id and roll_ptr
columns if the space is reallocated for a smaller record. */
3046 3047 3048
ibool	page_zip_clear_rec_disable;
#endif /* UNIV_ZIP_DEBUG */

3049 3050
/**************************************************************************
Clear an area on the uncompressed and compressed page, if possible. */
3051
static
3052 3053 3054 3055 3056 3057
void
page_zip_clear_rec(
/*===============*/
	page_zip_des_t*	page_zip,/* in/out: compressed page */
	byte*		rec,	/* in: record to clear */
	dict_index_t*	index,	/* in: index of rec */
3058
	const ulint*	offsets)/* in: rec_get_offsets(rec, index) */
3059
{
3060
	ulint	heap_no;
3061
	page_t*	page	= page_align(rec);
3062 3063
	/* page_zip_validate() would fail here if a record
	containing externally stored columns is being deleted. */
3064
	ut_ad(rec_offs_validate(rec, index, offsets));
3065 3066
	ut_ad(!page_zip_dir_find(page_zip, page_offset(rec)));
	ut_ad(page_zip_dir_find_free(page_zip, page_offset(rec)));
3067
	ut_ad(page_zip_header_cmp(page_zip, page));
3068

3069
	heap_no = rec_get_heap_no_new(rec);
3070
	ut_ad(heap_no >= 2); /* exclude infimum and supremum */
3071

3072 3073
	if (
#ifdef UNIV_ZIP_DEBUG
3074
	    !page_zip_clear_rec_disable &&
3075
#endif /* UNIV_ZIP_DEBUG */
3076 3077 3078 3079
	    page_zip->m_end
	    + 1 + ((heap_no - 1) >= 64)/* size of the log entry */
	    + page_zip_get_trailer_len(page_zip, index, NULL)
	    < page_zip->size) {
3080
		byte*	data;
3081 3082 3083

		/* Clear only the data bytes, because the allocator and
		the decompressor depend on the extra bytes. */
3084 3085
		memset(rec, 0, rec_offs_data_size(offsets));

marko's avatar
marko committed
3086 3087 3088
		if (!page_is_leaf(page)) {
			/* Clear node_ptr on the compressed page. */
			byte*	storage	= page_zip->data + page_zip->size
3089 3090
				- (page_dir_get_n_heap(page) - 2)
				* PAGE_ZIP_DIR_SLOT_SIZE;
marko's avatar
marko committed
3091 3092

			memset(storage - (heap_no - 1) * REC_NODE_PTR_SIZE,
3093
			       0, REC_NODE_PTR_SIZE);
3094
		} else if (dict_index_is_clust(index)) {
3095 3096
			/* Clear trx_id and roll_ptr on the compressed page. */
			byte*	storage	= page_zip->data + page_zip->size
3097 3098
				- (page_dir_get_n_heap(page) - 2)
				* PAGE_ZIP_DIR_SLOT_SIZE;
3099 3100

			memset(storage - (heap_no - 1)
3101 3102
			       * (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN),
			       0, DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
3103 3104
		}

3105
		/* Log that the data was zeroed out. */
3106 3107 3108 3109
		data = page_zip->data + page_zip->m_end;
		ut_ad(!*data);
		if (UNIV_UNLIKELY(heap_no - 1 >= 64)) {
			*data++ = 0x80 | (heap_no - 1) >> 7;
marko's avatar
marko committed
3110
			ut_ad(!*data);
3111 3112 3113
		}
		*data++ = (heap_no - 1) << 1 | 1;
		ut_ad(!*data);
3114
		ut_ad((ulint) (data - page_zip->data) < page_zip->size);
3115
		page_zip->m_end = data - page_zip->data;
3116
	} else if (page_is_leaf(page) && dict_index_is_clust(index)) {
marko's avatar
marko committed
3117 3118 3119 3120 3121 3122 3123 3124 3125
		/* Do not clear the record, because there is not enough space
		to log the operation. */

		ulint	i;
		for (i = rec_offs_n_fields(offsets); i--; ) {
			/* Clear all BLOB pointers in order to make
			page_zip_validate() pass. */
			if (rec_offs_nth_extern(offsets, i)) {
				ulint	len;
3126 3127
				byte*	field = rec_get_nth_field(rec, offsets,
								  i, &len);
marko's avatar
marko committed
3128
				memset(field + len - BTR_EXTERN_FIELD_REF_SIZE,
3129
				       0, BTR_EXTERN_FIELD_REF_SIZE);
marko's avatar
marko committed
3130
			}
3131 3132
		}
	}
marko's avatar
marko committed
3133 3134 3135 3136

#ifdef UNIV_ZIP_DEBUG
	ut_a(page_zip_validate(page_zip, page));
#endif /* UNIV_ZIP_DEBUG */
3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149
}

/**************************************************************************
Write the "deleted" flag of a record on a compressed page.  The flag must
already have been written on the uncompressed page. */

void
page_zip_rec_set_deleted(
/*=====================*/
	page_zip_des_t*	page_zip,/* in/out: compressed page */
	const byte*	rec,	/* in: record on the uncompressed page */
	ulint		flag)	/* in: the deleted flag (nonzero=TRUE) */
{
3150
	byte*	slot = page_zip_dir_find(page_zip, page_offset(rec));
3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169
	ut_a(slot);
	if (flag) {
		*slot |= (PAGE_ZIP_DIR_SLOT_DEL >> 8);
	} else {
		*slot &= ~(PAGE_ZIP_DIR_SLOT_DEL >> 8);
	}
}

/**************************************************************************
Write the "owned" flag of a record on a compressed page.  The n_owned field
must already have been written on the uncompressed page. */

void
page_zip_rec_set_owned(
/*===================*/
	page_zip_des_t*	page_zip,/* in/out: compressed page */
	const byte*	rec,	/* in: record on the uncompressed page */
	ulint		flag)	/* in: the owned flag (nonzero=TRUE) */
{
3170
	byte*	slot = page_zip_dir_find(page_zip, page_offset(rec));
3171 3172 3173 3174 3175 3176 3177 3178
	ut_a(slot);
	if (flag) {
		*slot |= (PAGE_ZIP_DIR_SLOT_OWNED >> 8);
	} else {
		*slot &= ~(PAGE_ZIP_DIR_SLOT_OWNED >> 8);
	}
}

3179 3180 3181 3182 3183 3184 3185 3186 3187 3188
/**************************************************************************
Insert a record to the dense page directory. */

void
page_zip_dir_insert(
/*================*/
	page_zip_des_t*	page_zip,/* in/out: compressed page */
	const byte*	prev_rec,/* in: record after which to insert */
	const byte*	free_rec,/* in: record from which rec was
				allocated, or NULL */
3189
	byte*		rec)	/* in: record to insert */
3190 3191 3192 3193 3194 3195 3196
{
	ulint	n_dense;
	byte*	slot_rec;
	byte*	slot_free;

	ut_ad(prev_rec != rec);
	ut_ad(page_rec_get_next((rec_t*) prev_rec) == rec);
3197
	ut_ad(page_zip_simple_validate(page_zip));
3198 3199 3200 3201 3202

	if (page_rec_is_infimum(prev_rec)) {
		/* Use the first slot. */
		slot_rec = page_zip->data + page_zip->size;
	} else {
3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213
		byte*	end	= page_zip->data + page_zip->size;
		byte*	start	= end - page_zip_dir_user_size(page_zip);

		if (UNIV_LIKELY(!free_rec)) {
			/* PAGE_N_RECS was already incremented
			in page_cur_insert_rec_low(), but the
			dense directory slot at that position
			contains garbage.  Skip it. */
			start += PAGE_ZIP_DIR_SLOT_SIZE;
		}

3214 3215
		slot_rec = page_zip_dir_find_low(start, end,
						 page_offset(prev_rec));
3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229
		ut_a(slot_rec);
	}

	/* Read the old n_dense (n_heap may have been incremented).
	Subtract 2 for the infimum and supremum records. */
	n_dense = page_dir_get_n_heap(page_zip->data) - 3;

	if (UNIV_LIKELY_NULL(free_rec)) {
		/* The record was allocated from the free list.
		Shift the dense directory only up to that slot.
		Note that in this case, n_dense is actually
		off by one, because page_cur_insert_rec_low()
		did not increment n_heap. */
		ut_ad(rec_get_heap_no_new(rec) < n_dense + 1
3230
		      + 2/* infimum and supremum */);
3231
		ut_ad(rec >= free_rec);
3232
		slot_free = page_zip_dir_find(page_zip, page_offset(free_rec));
3233 3234 3235 3236 3237 3238
		ut_ad(slot_free);
		slot_free += PAGE_ZIP_DIR_SLOT_SIZE;
	} else {
		/* The record was allocated from the heap.
		Shift the entire dense directory. */
		ut_ad(rec_get_heap_no_new(rec) == n_dense
3239
		      + 2/* infimum and supremum */);
3240 3241 3242

		/* Shift to the end of the dense page directory. */
		slot_free = page_zip->data + page_zip->size
3243
			- PAGE_ZIP_DIR_SLOT_SIZE * n_dense;
3244 3245 3246 3247
	}

	/* Shift the dense directory to allocate place for rec. */
	memmove(slot_free - PAGE_ZIP_DIR_SLOT_SIZE, slot_free,
3248
		slot_rec - slot_free);
3249 3250 3251

	/* Write the entry for the inserted record.
	The "owned" and "deleted" flags must be zero. */
3252
	mach_write_to_2(slot_rec - PAGE_ZIP_DIR_SLOT_SIZE, page_offset(rec));
3253 3254
}

3255
/**************************************************************************
3256 3257
Shift the dense page directory and the array of BLOB pointers
when a record is deleted. */
3258 3259 3260 3261 3262

void
page_zip_dir_delete(
/*================*/
	page_zip_des_t*	page_zip,/* in/out: compressed page */
3263 3264 3265
	byte*		rec,	/* in: record to delete */
	dict_index_t*	index,	/* in: index of rec */
	const ulint*	offsets,/* in: rec_get_offsets(rec) */
3266 3267 3268 3269
	const byte*	free)	/* in: previous start of the free list */
{
	byte*	slot_rec;
	byte*	slot_free;
3270
	ulint	n_ext;
3271
	page_t*	page	= page_align(rec);
3272

3273 3274
	ut_ad(rec_offs_validate(rec, index, offsets));
	ut_ad(rec_offs_comp(offsets));
marko's avatar
marko committed
3275

3276
	slot_rec = page_zip_dir_find(page_zip, page_offset(rec));
3277 3278 3279

	ut_a(slot_rec);

3280 3281
	/* This could not be done before page_zip_dir_find(). */
	page_header_set_field(page, page_zip, PAGE_N_RECS,
3282
			      (ulint)(page_get_n_recs(page) - 1));
3283

marko's avatar
marko committed
3284
	if (UNIV_UNLIKELY(!free)) {
3285 3286
		/* Make the last slot the start of the free list. */
		slot_free = page_zip->data + page_zip->size
3287 3288
			- PAGE_ZIP_DIR_SLOT_SIZE
			* (page_dir_get_n_heap(page_zip->data) - 2);
3289
	} else {
3290 3291
		slot_free = page_zip_dir_find_free(page_zip,
						   page_offset(free));
3292 3293 3294 3295 3296
		ut_a(slot_free < slot_rec);
		/* Grow the free list by one slot by moving the start. */
		slot_free += PAGE_ZIP_DIR_SLOT_SIZE;
	}

marko's avatar
marko committed
3297
	if (UNIV_LIKELY(slot_rec > slot_free)) {
3298 3299
		memmove(slot_free + PAGE_ZIP_DIR_SLOT_SIZE,
			slot_free,
marko's avatar
marko committed
3300
			slot_rec - slot_free);
3301 3302 3303 3304
	}

	/* Write the entry for the deleted record.
	The "owned" and "deleted" flags will be cleared. */
3305
	mach_write_to_2(slot_free, page_offset(rec));
3306

3307 3308 3309 3310 3311
	if (!page_is_leaf(page) || !dict_index_is_clust(index)) {
		ut_ad(!rec_offs_any_extern(offsets));
		goto skip_blobs;
	}

3312 3313 3314 3315 3316 3317 3318 3319 3320 3321
	n_ext = rec_offs_n_extern(offsets);
	if (UNIV_UNLIKELY(n_ext)) {
		/* Shift and zero fill the array of BLOB pointers. */
		ulint	blob_no;
		byte*	externs;
		byte*	ext_end;

		blob_no = page_zip_get_n_prev_extern(page_zip, rec, index);
		ut_a(blob_no + n_ext <= page_zip->n_blobs);

3322
		externs = page_zip->data + page_zip->size
3323 3324 3325
			- (page_dir_get_n_heap(page) - 2)
			* (PAGE_ZIP_DIR_SLOT_SIZE
			   + DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
3326 3327

		ext_end = externs - page_zip->n_blobs
3328
			* BTR_EXTERN_FIELD_REF_SIZE;
3329 3330 3331 3332 3333
		externs -= blob_no * BTR_EXTERN_FIELD_REF_SIZE;

		page_zip->n_blobs -= n_ext;
		/* Shift and zero fill the array. */
		memmove(ext_end + n_ext * BTR_EXTERN_FIELD_REF_SIZE, ext_end,
3334 3335
			(page_zip->n_blobs - blob_no)
			* BTR_EXTERN_FIELD_REF_SIZE);
3336 3337 3338
		memset(ext_end, 0, n_ext * BTR_EXTERN_FIELD_REF_SIZE);
	}

3339
skip_blobs:
3340 3341 3342 3343
	/* The compression algorithm expects info_bits and n_owned
	to be 0 for deleted records. */
	rec[-REC_N_NEW_EXTRA_BYTES] = 0; /* info_bits and n_owned */

3344
	page_zip_clear_rec(page_zip, rec, index, offsets);
marko's avatar
marko committed
3345
}
3346

3347 3348 3349 3350 3351 3352
/**************************************************************************
Add a slot to the dense page directory. */

void
page_zip_dir_add_slot(
/*==================*/
3353 3354 3355
	page_zip_des_t*	page_zip,	/* in/out: compressed page */
	ulint		is_clustered)	/* in: nonzero for clustered index,
					zero for others */
3356 3357 3358 3359 3360
{
	ulint	n_dense;
	byte*	dir;
	byte*	stored;

3361 3362 3363 3364 3365
	ut_ad(page_is_comp(page_zip->data));

	/* Read the old n_dense (n_heap has already been incremented).
	Subtract 2 for the infimum and supremum records. */
	n_dense = page_dir_get_n_heap(page_zip->data) - 3;
3366 3367

	dir = page_zip->data + page_zip->size
3368
		- PAGE_ZIP_DIR_SLOT_SIZE * n_dense;
3369

3370 3371 3372 3373
	if (!page_is_leaf(page_zip->data)) {
		ut_ad(!page_zip->n_blobs);
		stored = dir - n_dense * REC_NODE_PTR_SIZE;
	} else if (UNIV_UNLIKELY(is_clustered)) {
3374 3375 3376 3377
		/* Move the BLOB pointer array backwards to make space for the
		roll_ptr and trx_id columns and the dense directory slot. */
		byte*	externs;

3378
		stored = dir - n_dense
3379 3380
			* (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
		externs = stored
3381
			- page_zip->n_blobs * BTR_EXTERN_FIELD_REF_SIZE;
3382 3383 3384 3385 3386
		ut_ad(!memcmp(zero, externs
			      - (PAGE_ZIP_DIR_SLOT_SIZE
				 + DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN),
			      PAGE_ZIP_DIR_SLOT_SIZE
			      + DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN));
3387
		memmove(externs - (PAGE_ZIP_DIR_SLOT_SIZE
3388
				   + DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN),
3389
			externs, stored - externs);
3390
	} else {
3391 3392
		stored = dir
			- page_zip->n_blobs * BTR_EXTERN_FIELD_REF_SIZE;
marko's avatar
marko committed
3393
		ut_ad(!memcmp(zero, stored - PAGE_ZIP_DIR_SLOT_SIZE,
3394
			      PAGE_ZIP_DIR_SLOT_SIZE));
3395 3396
	}

3397 3398 3399
	/* Move the uncompressed area backwards to make space
	for one directory slot. */
	memmove(stored - PAGE_ZIP_DIR_SLOT_SIZE, stored, dir - stored);
3400 3401
}

3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416
/***************************************************************
Parses a log record of writing to the header of a page. */

byte*
page_zip_parse_write_header(
/*========================*/
				/* out: end of log record or NULL */
	byte*		ptr,	/* in: redo log buffer */
	byte*		end_ptr,/* in: redo log buffer end */
	page_t*		page,	/* in/out: uncompressed page */
	page_zip_des_t*	page_zip)/* in/out: compressed page */
{
	ulint	offset;
	ulint	len;

3417
	ut_ad(ptr && end_ptr);
3418 3419
	ut_ad(!page == !page_zip);

3420
	if (UNIV_UNLIKELY(end_ptr < ptr + (1 + 1))) {
3421 3422 3423 3424

		return(NULL);
	}

3425
	offset = (ulint) *ptr++;
3426 3427
	len = (ulint) *ptr++;

3428 3429
	if (UNIV_UNLIKELY(!len) || UNIV_UNLIKELY(offset + len >= PAGE_DATA)) {
corrupt:
3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440
		recv_sys->found_corrupt_log = TRUE;

		return(NULL);
	}

	if (UNIV_UNLIKELY(end_ptr < ptr + len)) {

		return(NULL);
	}

	if (page) {
3441 3442 3443 3444
		if (UNIV_UNLIKELY(!page_zip)) {

			goto corrupt;
		}
3445
#ifdef UNIV_ZIP_DEBUG
3446
		ut_a(page_zip_validate(page_zip, page));
3447
#endif /* UNIV_ZIP_DEBUG */
3448 3449 3450 3451

		memcpy(page + offset, ptr, len);
		memcpy(page_zip->data + offset, ptr, len);

3452
#ifdef UNIV_ZIP_DEBUG
3453
		ut_a(page_zip_validate(page_zip, page));
3454
#endif /* UNIV_ZIP_DEBUG */
3455 3456 3457 3458 3459
	}

	return(ptr + len);
}

3460 3461 3462 3463 3464 3465
/**************************************************************************
Write a log record of writing to the uncompressed header portion of a page. */

void
page_zip_write_header_log(
/*======================*/
3466 3467 3468
	const byte*	data,	/* in: data on the uncompressed page */
	ulint		length,	/* in: length of the data */
	mtr_t*		mtr)	/* in: mini-transaction */
3469
{
3470
	byte*	log_ptr	= mlog_open(mtr, 11 + 1 + 1);
3471
	ulint	offset	= page_offset(data);
3472

3473 3474 3475 3476 3477 3478 3479 3480
	ut_ad(offset < PAGE_DATA);
	ut_ad(offset + length < PAGE_DATA);
#if PAGE_DATA > 255
# error "PAGE_DATA > 255"
#endif
	ut_ad(length < 256);

	/* If no logging is requested, we may return now */
3481
	if (UNIV_UNLIKELY(!log_ptr)) {
3482 3483 3484 3485

		return;
	}

3486 3487
	log_ptr = mlog_write_initial_log_record_fast(
		(byte*) data, MLOG_ZIP_WRITE_HEADER, log_ptr, mtr);
3488
	*log_ptr++ = (byte) offset;
3489
	*log_ptr++ = (byte) length;
3490 3491
	mlog_close(mtr, log_ptr);

3492
	mlog_catenate_string(mtr, data, length);
3493
}
3494

3495 3496 3497
/**************************************************************************
Reorganize and compress a page.  This is a low-level operation for
compressed pages, to be used when page_zip_compress() fails.
3498
On success, a redo log entry MLOG_ZIP_PAGE_COMPRESS will be written.
3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516
The function btr_page_reorganize() should be preferred whenever possible. */

ibool
page_zip_reorganize(
/*================*/
				/* out: TRUE on success, FALSE on failure;
				page and page_zip will be left intact
				on failure. */
	page_zip_des_t*	page_zip,/* in: size; out: data, n_blobs,
				m_start, m_end */
	page_t*		page,	/* in/out: uncompressed page */
	dict_index_t*	index,	/* in: index of the B-tree node */
	mtr_t*		mtr)	/* in: mini-transaction */
{
	page_t*	temp_page;
	ulint	log_mode;

	ut_ad(mtr_memo_contains(mtr, buf_block_align(page),
3517
				MTR_MEMO_PAGE_X_FIX));
3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537
	ut_ad(page_is_comp(page));
	/* Note that page_zip_validate(page_zip, page) may fail here. */

	/* Disable logging */
	log_mode = mtr_set_log_mode(mtr, MTR_LOG_NONE);

	temp_page = buf_frame_alloc();

	/* Copy the old page to temporary space */
	buf_frame_copy(temp_page, page);

	/* Recreate the page: note that global data on page (possible
	segment headers, next page-field, etc.) is preserved intact */

	page_create(page, mtr, dict_table_is_comp(index->table));
	buf_block_align(page)->check_index_page_at_flush = TRUE;

	/* Copy the records from the temporary space to the recreated page;
	do not copy the lock bits yet */

3538 3539
	page_copy_rec_list_end_no_locks(page, page_get_infimum_rec(temp_page),
					index, mtr);
3540 3541 3542
	/* Copy max trx id to recreated page */
	page_set_max_trx_id(page, NULL, page_get_max_trx_id(temp_page));

3543 3544 3545
	/* Restore logging. */
	mtr_set_log_mode(mtr, log_mode);

3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561
	if (UNIV_UNLIKELY(!page_zip_compress(page_zip, page, index, mtr))) {

		/* Restore the old page and exit. */
		buf_frame_copy(page, temp_page);

		buf_frame_free(temp_page);
		return(FALSE);
	}

	lock_move_reorganize_page(page, temp_page);
	btr_search_drop_page_hash_index(page);

	buf_frame_free(temp_page);
	return(TRUE);
}

3562 3563 3564 3565 3566 3567 3568 3569 3570 3571
/**************************************************************************
Copy a page byte for byte, except for the file page header and trailer. */

void
page_zip_copy(
/*==========*/
	page_zip_des_t*		page_zip,	/* out: copy of src_zip */
	page_t*			page,		/* out: copy of src */
	const page_zip_des_t*	src_zip,	/* in: compressed page */
	const page_t*		src,		/* in: page */
3572
	dict_index_t*		index,		/* in: index of the B-tree */
3573 3574
	mtr_t*			mtr)		/* in: mini-transaction */
{
3575
	ut_ad(mtr_memo_contains(mtr, buf_block_align(page),
3576
				MTR_MEMO_PAGE_X_FIX));
3577
	ut_ad(mtr_memo_contains(mtr, buf_block_align((page_t*) src),
3578
				MTR_MEMO_PAGE_X_FIX));
3579
#ifdef UNIV_ZIP_DEBUG
3580
	ut_a(page_zip_validate(src_zip, src));
3581
#endif /* UNIV_ZIP_DEBUG */
3582
	ut_a(page_zip->size == src_zip->size);
3583 3584 3585 3586
	if (UNIV_UNLIKELY(src_zip->n_blobs)) {
		ut_a(page_is_leaf((page_t*) src));
		ut_a(dict_index_is_clust(index));
	}
3587 3588 3589

	/* Skip the file page header and trailer. */
	memcpy(page + FIL_PAGE_DATA, src + FIL_PAGE_DATA,
3590 3591
	       UNIV_PAGE_SIZE - FIL_PAGE_DATA
	       - FIL_PAGE_DATA_END);
3592
	memcpy(page_zip->data + FIL_PAGE_DATA,
3593 3594
	       src_zip->data + FIL_PAGE_DATA,
	       page_zip->size - FIL_PAGE_DATA);
3595 3596 3597 3598

	page_zip->n_blobs = src_zip->n_blobs;
	page_zip->m_start = src_zip->m_start;
	page_zip->m_end = src_zip->m_end;
3599
	ut_ad(page_zip_get_trailer_len(page_zip, index, NULL)
3600
	      + page_zip->m_end < page_zip->size);
3601

3602
	if (!page_is_leaf(src)
3603
	    && UNIV_UNLIKELY(mach_read_from_4(src + FIL_PAGE_PREV) == FIL_NULL)
3604 3605
	    && UNIV_LIKELY(mach_read_from_4(page
					    + FIL_PAGE_PREV) != FIL_NULL)) {
3606
		/* Clear the REC_INFO_MIN_REC_FLAG of the first user record. */
3607 3608
		ulint	offs = rec_get_next_offs(page + PAGE_NEW_INFIMUM,
						 TRUE);
3609 3610 3611
		if (UNIV_LIKELY(offs != PAGE_NEW_SUPREMUM)) {
			rec_t*	rec = page + offs;
			ut_a(rec[-REC_N_NEW_EXTRA_BYTES]
3612
			     & REC_INFO_MIN_REC_FLAG);
3613 3614 3615 3616
			rec[-REC_N_NEW_EXTRA_BYTES] &= ~ REC_INFO_MIN_REC_FLAG;
		}
	}

3617
#ifdef UNIV_ZIP_DEBUG
3618
	ut_a(page_zip_validate(page_zip, page));
3619
#endif /* UNIV_ZIP_DEBUG */
3620

3621
	page_zip_compress_write_log(page_zip, page, index, mtr);
3622 3623
}

3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636
/**************************************************************************
Parses a log record of compressing an index page. */

byte*
page_zip_parse_compress(
/*====================*/
				/* out: end of log record or NULL */
	byte*		ptr,	/* in: buffer */
	byte*		end_ptr,/* in: buffer end */
	page_t*		page,	/* out: uncompressed page */
	page_zip_des_t*	page_zip)/* out: compressed page */
{
	ulint	size;
3637
	ulint	trailer_size;
3638 3639

	ut_ad(ptr && end_ptr);
3640
	ut_ad(!page == !page_zip);
3641

3642
	if (UNIV_UNLIKELY(ptr + (2 + 2) > end_ptr)) {
3643 3644 3645 3646 3647 3648

		return(NULL);
	}

	size = mach_read_from_2(ptr);
	ptr += 2;
3649 3650
	trailer_size = mach_read_from_2(ptr);
	ptr += 2;
3651

3652
	if (UNIV_UNLIKELY(ptr + 8 + size + trailer_size > end_ptr)) {
3653 3654 3655 3656 3657 3658

		return(NULL);
	}

	if (page) {
		if (UNIV_UNLIKELY(!page_zip)
3659
		    || UNIV_UNLIKELY(page_zip->size < size)) {
3660 3661 3662 3663 3664 3665
corrupt:
			recv_sys->found_corrupt_log = TRUE;

			return(NULL);
		}

3666 3667 3668 3669
		memcpy(page_zip->data + FIL_PAGE_PREV, ptr, 4);
		memcpy(page_zip->data + FIL_PAGE_NEXT, ptr + 4, 4);
		memcpy(page_zip->data + FIL_PAGE_TYPE, ptr + 8, size);
		memset(page_zip->data + FIL_PAGE_TYPE + size, 0,
3670 3671
		       page_zip->size - trailer_size
		       - (FIL_PAGE_TYPE + size));
3672
		memcpy(page_zip->data + page_zip->size - trailer_size,
3673
		       ptr + 8 + size, trailer_size);
3674

3675 3676 3677 3678 3679 3680
		if (UNIV_UNLIKELY(!page_zip_decompress(page_zip, page))) {

			goto corrupt;
		}
	}

3681
	return(ptr + 8 + size + trailer_size);
3682
}
3683 3684 3685 3686 3687 3688 3689 3690 3691 3692 3693 3694 3695

/**************************************************************************
Calculate the compressed page checksum. */

ulint
page_zip_calc_checksum(
/*===================*/
				/* out: page checksum */
	const void*	data,	/* in: compressed page */
	ulint		size)	/* in: size of compressed page */
{
	/* Exclude the 32-bit checksum field from the checksum. */
	return((ulint) adler32(0,
3696 3697
			       ((const Bytef*) data) + FIL_PAGE_OFFSET,
			       size - FIL_PAGE_OFFSET));
3698
}