buf0flu.c 42.7 KB
Newer Older
Vadim Tkachenko's avatar
Vadim Tkachenko committed
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
/*****************************************************************************

Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.

This program is free software; you can redistribute it and/or modify it under
the terms of the GNU General Public License as published by the Free Software
Foundation; version 2 of the License.

This program is distributed in the hope that it will be useful, but WITHOUT
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with
this program; if not, write to the Free Software Foundation, Inc., 59 Temple
Place, Suite 330, Boston, MA 02111-1307 USA

*****************************************************************************/

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
19 20
/**************************************************//**
@file buf/buf0flu.c
21 22 23 24 25 26 27 28 29 30 31
The database buffer buf_pool flush algorithm

Created 11/11/1995 Heikki Tuuri
*******************************************************/

#include "buf0flu.h"

#ifdef UNIV_NONINL
#include "buf0flu.ic"
#endif

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
32 33 34 35
#include "buf0buf.h"
#include "srv0srv.h"
#include "page0zip.h"
#ifndef UNIV_HOTBACKUP
36 37 38 39 40 41 42 43 44 45 46 47
#include "ut0byte.h"
#include "ut0lst.h"
#include "page0page.h"
#include "fil0fil.h"
#include "buf0lru.h"
#include "buf0rea.h"
#include "ibuf0ibuf.h"
#include "log0log.h"
#include "os0file.h"
#include "trx0sys.h"

/**********************************************************************
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83
These statistics are generated for heuristics used in estimating the
rate at which we should flush the dirty blocks to avoid bursty IO
activity. Note that the rate of flushing not only depends on how many
dirty pages we have in the buffer pool but it is also a fucntion of
how much redo the workload is generating and at what rate. */
/* @{ */

/** Number of intervals for which we keep the history of these stats.
Each interval is 1 second, defined by the rate at which
srv_error_monitor_thread() calls buf_flush_stat_update(). */
#define BUF_FLUSH_STAT_N_INTERVAL 20

/** Sampled values buf_flush_stat_cur.
Not protected by any mutex.  Updated by buf_flush_stat_update(). */
static buf_flush_stat_t	buf_flush_stat_arr[BUF_FLUSH_STAT_N_INTERVAL];

/** Cursor to buf_flush_stat_arr[]. Updated in a round-robin fashion. */
static ulint		buf_flush_stat_arr_ind;

/** Values at start of the current interval. Reset by
buf_flush_stat_update(). */
static buf_flush_stat_t	buf_flush_stat_cur;

/** Running sum of past values of buf_flush_stat_cur.
Updated by buf_flush_stat_update(). Not protected by any mutex. */
static buf_flush_stat_t	buf_flush_stat_sum;

/** Number of pages flushed through non flush_list flushes. */
static ulint buf_lru_flush_page_count = 0;

/* @} */

#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
/******************************************************************//**
Validates the flush list.
@return	TRUE if ok */
84 85 86 87 88 89
static
ibool
buf_flush_validate_low(void);
/*========================*/
#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
90
/********************************************************************//**
91 92 93 94 95
Inserts a modified block into the flush list. */
UNIV_INTERN
void
buf_flush_insert_into_flush_list(
/*=============================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
96
	buf_block_t*	block)	/*!< in/out: block which is modified */
97
{
Vadim Tkachenko's avatar
Vadim Tkachenko committed
98 99 100
	//ut_ad(buf_pool_mutex_own());
	ut_ad(mutex_own(&block->mutex));
	ut_ad(mutex_own(&flush_list_mutex));
101 102
	ut_ad((UT_LIST_GET_FIRST(buf_pool->flush_list) == NULL)
	      || (UT_LIST_GET_FIRST(buf_pool->flush_list)->oldest_modification
Vadim Tkachenko's avatar
Vadim Tkachenko committed
103
		  <= block->page.oldest_modification));
104

Vadim Tkachenko's avatar
Vadim Tkachenko committed
105 106 107 108 109 110
	ut_ad(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
	ut_ad(block->page.in_LRU_list);
	ut_ad(block->page.in_page_hash);
	ut_ad(!block->page.in_zip_hash);
	ut_ad(!block->page.in_flush_list);
	ut_d(block->page.in_flush_list = TRUE);
Vadim Tkachenko's avatar
Vadim Tkachenko committed
111
	UT_LIST_ADD_FIRST(flush_list, buf_pool->flush_list, &block->page);
112 113 114 115 116 117

#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
	ut_a(buf_flush_validate_low());
#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
118
/********************************************************************//**
119 120 121 122 123 124 125
Inserts a modified block into the flush list in the right sorted position.
This function is used by recovery, because there the modifications do not
necessarily come in the order of lsn's. */
UNIV_INTERN
void
buf_flush_insert_sorted_into_flush_list(
/*====================================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
126
	buf_block_t*	block)	/*!< in/out: block which is modified */
127 128 129 130
{
	buf_page_t*	prev_b;
	buf_page_t*	b;

Vadim Tkachenko's avatar
Vadim Tkachenko committed
131 132 133
	//ut_ad(buf_pool_mutex_own());
	ut_ad(mutex_own(&block->mutex));
	ut_ad(mutex_own(&flush_list_mutex));
Vadim Tkachenko's avatar
Vadim Tkachenko committed
134
	ut_ad(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
135

Vadim Tkachenko's avatar
Vadim Tkachenko committed
136 137 138 139 140
	ut_ad(block->page.in_LRU_list);
	ut_ad(block->page.in_page_hash);
	ut_ad(!block->page.in_zip_hash);
	ut_ad(!block->page.in_flush_list);
	ut_d(block->page.in_flush_list = TRUE);
141 142 143 144

	prev_b = NULL;
	b = UT_LIST_GET_FIRST(buf_pool->flush_list);

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
145 146 147 148 149 150 151 152 153 154 155 156 157 158
	if (srv_fast_recovery) {
	/* speed hack */
	if (b == NULL || b->oldest_modification < block->page.oldest_modification) {
		UT_LIST_ADD_FIRST(flush_list, buf_pool->flush_list, &block->page);
	} else {
		b = UT_LIST_GET_LAST(buf_pool->flush_list);
		if (b->oldest_modification < block->page.oldest_modification) {
			/* align oldest_modification not to sort */
			block->page.oldest_modification = b->oldest_modification;
		}
		UT_LIST_ADD_LAST(flush_list, buf_pool->flush_list, &block->page);
	}
	} else {
	/* normal */
Vadim Tkachenko's avatar
Vadim Tkachenko committed
159
	while (b && b->oldest_modification > block->page.oldest_modification) {
160 161
		ut_ad(b->in_flush_list);
		prev_b = b;
Vadim Tkachenko's avatar
Vadim Tkachenko committed
162
		b = UT_LIST_GET_NEXT(flush_list, b);
163 164 165
	}

	if (prev_b == NULL) {
Vadim Tkachenko's avatar
Vadim Tkachenko committed
166
		UT_LIST_ADD_FIRST(flush_list, buf_pool->flush_list, &block->page);
167
	} else {
Vadim Tkachenko's avatar
Vadim Tkachenko committed
168
		UT_LIST_INSERT_AFTER(flush_list, buf_pool->flush_list,
Vadim Tkachenko's avatar
Vadim Tkachenko committed
169
				     prev_b, &block->page);
170
	}
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
171
	}
172 173 174 175 176 177

#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
	ut_a(buf_flush_validate_low());
#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
178
/********************************************************************//**
179
Returns TRUE if the file page block is immediately suitable for replacement,
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
180 181
i.e., the transition FILE_PAGE => NOT_USED allowed.
@return	TRUE if can replace immediately */
182 183 184 185
UNIV_INTERN
ibool
buf_flush_ready_for_replace(
/*========================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
186
	buf_page_t*	bpage)	/*!< in: buffer control block, must be
187 188
				buf_page_in_file(bpage) and in the LRU list */
{
189
	//ut_ad(buf_pool_mutex_own());
Vadim Tkachenko's avatar
Vadim Tkachenko committed
190
	ut_ad(mutex_own(buf_page_get_mutex(bpage)));
191
	//ut_ad(bpage->in_LRU_list); /* optimistic use */
192

Vadim Tkachenko's avatar
Vadim Tkachenko committed
193
	if (UNIV_LIKELY(bpage->in_LRU_list && buf_page_in_file(bpage))) {
194 195 196 197 198 199

		return(bpage->oldest_modification == 0
		       && buf_page_get_io_fix(bpage) == BUF_IO_NONE
		       && bpage->buf_fix_count == 0);
	}

200 201
	/* permited not to own LRU_mutex..  */
/*
202 203 204 205 206 207
	ut_print_timestamp(stderr);
	fprintf(stderr,
		"  InnoDB: Error: buffer block state %lu"
		" in the LRU list!\n",
		(ulong) buf_page_get_state(bpage));
	ut_print_buf(stderr, bpage, sizeof(buf_page_t));
Vadim Tkachenko's avatar
Vadim Tkachenko committed
208
	putc('\n', stderr);
209
*/
210 211 212 213

	return(FALSE);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
214 215 216
/********************************************************************//**
Returns TRUE if the block is modified and ready for flushing.
@return	TRUE if can flush immediately */
217 218 219 220
UNIV_INLINE
ibool
buf_flush_ready_for_flush(
/*======================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
221
	buf_page_t*	bpage,	/*!< in: buffer control block, must be
222
				buf_page_in_file(bpage) */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
223
	enum buf_flush	flush_type)/*!< in: BUF_FLUSH_LRU or BUF_FLUSH_LIST */
224
{
Vadim Tkachenko's avatar
Vadim Tkachenko committed
225 226
	//ut_a(buf_page_in_file(bpage));
	//ut_ad(buf_pool_mutex_own()); /*optimistic...*/
227
	ut_ad(mutex_own(buf_page_get_mutex(bpage)));
Vadim Tkachenko's avatar
Vadim Tkachenko committed
228
	ut_ad(flush_type == BUF_FLUSH_LRU || BUF_FLUSH_LIST);
229

Vadim Tkachenko's avatar
Vadim Tkachenko committed
230
	if (buf_page_in_file(bpage) && bpage->oldest_modification != 0
231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250
	    && buf_page_get_io_fix(bpage) == BUF_IO_NONE) {
		ut_ad(bpage->in_flush_list);

		if (flush_type != BUF_FLUSH_LRU) {

			return(TRUE);

		} else if (bpage->buf_fix_count == 0) {

			/* If we are flushing the LRU list, to avoid deadlocks
			we require the block not to be bufferfixed, and hence
			not latched. */

			return(TRUE);
		}
	}

	return(FALSE);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
251
/********************************************************************//**
252 253 254 255 256
Remove a block from the flush list of modified blocks. */
UNIV_INTERN
void
buf_flush_remove(
/*=============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
257
	buf_page_t*	bpage)	/*!< in: pointer to the block in question */
258
{
Vadim Tkachenko's avatar
Vadim Tkachenko committed
259
	//ut_ad(buf_pool_mutex_own());
260
	ut_ad(mutex_own(buf_page_get_mutex(bpage)));
Vadim Tkachenko's avatar
Vadim Tkachenko committed
261 262 263

	mutex_enter(&flush_list_mutex);

264 265 266 267 268 269 270 271 272 273 274
	ut_ad(bpage->in_flush_list);
	ut_d(bpage->in_flush_list = FALSE);

	switch (buf_page_get_state(bpage)) {
	case BUF_BLOCK_ZIP_PAGE:
		/* clean compressed pages should not be on the flush list */
	case BUF_BLOCK_ZIP_FREE:
	case BUF_BLOCK_NOT_USED:
	case BUF_BLOCK_READY_FOR_USE:
	case BUF_BLOCK_MEMORY:
	case BUF_BLOCK_REMOVE_HASH:
Vadim Tkachenko's avatar
Vadim Tkachenko committed
275
		mutex_exit(&flush_list_mutex);
276 277 278 279
		ut_error;
		return;
	case BUF_BLOCK_ZIP_DIRTY:
		buf_page_set_state(bpage, BUF_BLOCK_ZIP_PAGE);
Vadim Tkachenko's avatar
Vadim Tkachenko committed
280
		UT_LIST_REMOVE(flush_list, buf_pool->flush_list, bpage);
281 282 283
		buf_LRU_insert_zip_clean(bpage);
		break;
	case BUF_BLOCK_FILE_PAGE:
Vadim Tkachenko's avatar
Vadim Tkachenko committed
284
		UT_LIST_REMOVE(flush_list, buf_pool->flush_list, bpage);
285 286 287 288 289
		break;
	}

	bpage->oldest_modification = 0;

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
290 291
	ut_d(UT_LIST_VALIDATE(flush_list, buf_page_t, buf_pool->flush_list,
			      ut_ad(ut_list_node_313->in_flush_list)));
Vadim Tkachenko's avatar
Vadim Tkachenko committed
292
	mutex_exit(&flush_list_mutex);
293 294
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
295
/********************************************************************//**
296 297 298 299 300
Updates the flush system data structures when a write is completed. */
UNIV_INTERN
void
buf_flush_write_complete(
/*=====================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
301
	buf_page_t*	bpage)	/*!< in: pointer to the block in question */
302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332
{
	enum buf_flush	flush_type;

	ut_ad(bpage);

	buf_flush_remove(bpage);

	flush_type = buf_page_get_flush_type(bpage);
	buf_pool->n_flush[flush_type]--;

	if (flush_type == BUF_FLUSH_LRU) {
		/* Put the block to the end of the LRU list to wait to be
		moved to the free list */

		buf_LRU_make_block_old(bpage);

		buf_pool->LRU_flush_ended++;
	}

	/* fprintf(stderr, "n pending flush %lu\n",
	buf_pool->n_flush[flush_type]); */

	if ((buf_pool->n_flush[flush_type] == 0)
	    && (buf_pool->init_flush[flush_type] == FALSE)) {

		/* The running flush batch has ended */

		os_event_set(buf_pool->no_flush[flush_type]);
	}
}

333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354
/********************************************************************//**
Flush a batch of writes to the datafiles that have already been
written by the OS. */
static
void
buf_flush_sync_datafiles(void)
/*==========================*/
{
	/* Wake possible simulated aio thread to actually post the
	writes to the operating system */
	os_aio_simulated_wake_handler_threads();

	/* Wait that all async writes to tablespaces have been posted to
	the OS */
	os_aio_wait_until_no_pending_writes();

	/* Now we flush the data to disk (for example, with fsync) */
	fil_flush_file_spaces(FIL_TABLESPACE);

	return;
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
355
/********************************************************************//**
356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371
Flushes possible buffered writes from the doublewrite memory buffer to disk,
and also wakes up the aio thread if simulated aio is used. It is very
important to call this function after a batch of writes has been posted,
and also when we may have to wait for a page latch! Otherwise a deadlock
of threads can occur. */
static
void
buf_flush_buffered_writes(void)
/*===========================*/
{
	byte*		write_buf;
	ulint		len;
	ulint		len2;
	ulint		i;

	if (!srv_use_doublewrite_buf || trx_doublewrite == NULL) {
372 373
		/* Sync the writes to the disk. */
		buf_flush_sync_datafiles();
374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580
		return;
	}

	mutex_enter(&(trx_doublewrite->mutex));

	/* Write first to doublewrite buffer blocks. We use synchronous
	aio and thus know that file write has been completed when the
	control returns. */

	if (trx_doublewrite->first_free == 0) {

		mutex_exit(&(trx_doublewrite->mutex));

		return;
	}

	for (i = 0; i < trx_doublewrite->first_free; i++) {

		const buf_block_t*	block;

		block = (buf_block_t*) trx_doublewrite->buf_block_arr[i];

		if (buf_block_get_state(block) != BUF_BLOCK_FILE_PAGE
		    || block->page.zip.data) {
			/* No simple validate for compressed pages exists. */
			continue;
		}

		if (UNIV_UNLIKELY
		    (memcmp(block->frame + (FIL_PAGE_LSN + 4),
			    block->frame + (UNIV_PAGE_SIZE
					    - FIL_PAGE_END_LSN_OLD_CHKSUM + 4),
			    4))) {
			ut_print_timestamp(stderr);
			fprintf(stderr,
				"  InnoDB: ERROR: The page to be written"
				" seems corrupt!\n"
				"InnoDB: The lsn fields do not match!"
				" Noticed in the buffer pool\n"
				"InnoDB: before posting to the"
				" doublewrite buffer.\n");
		}

		if (!block->check_index_page_at_flush) {
		} else if (page_is_comp(block->frame)) {
			if (UNIV_UNLIKELY
			    (!page_simple_validate_new(block->frame))) {
corrupted_page:
				buf_page_print(block->frame, 0);

				ut_print_timestamp(stderr);
				fprintf(stderr,
					"  InnoDB: Apparent corruption of an"
					" index page n:o %lu in space %lu\n"
					"InnoDB: to be written to data file."
					" We intentionally crash server\n"
					"InnoDB: to prevent corrupt data"
					" from ending up in data\n"
					"InnoDB: files.\n",
					(ulong) buf_block_get_page_no(block),
					(ulong) buf_block_get_space(block));

				ut_error;
			}
		} else if (UNIV_UNLIKELY
			   (!page_simple_validate_old(block->frame))) {

			goto corrupted_page;
		}
	}

	/* increment the doublewrite flushed pages counter */
	srv_dblwr_pages_written+= trx_doublewrite->first_free;
	srv_dblwr_writes++;

	len = ut_min(TRX_SYS_DOUBLEWRITE_BLOCK_SIZE,
		     trx_doublewrite->first_free) * UNIV_PAGE_SIZE;

	write_buf = trx_doublewrite->write_buf;
	i = 0;

	fil_io(OS_FILE_WRITE, TRUE, TRX_SYS_SPACE, 0,
	       trx_doublewrite->block1, 0, len,
	       (void*) write_buf, NULL);

	for (len2 = 0; len2 + UNIV_PAGE_SIZE <= len;
	     len2 += UNIV_PAGE_SIZE, i++) {
		const buf_block_t* block = (buf_block_t*)
			trx_doublewrite->buf_block_arr[i];

		if (UNIV_LIKELY(!block->page.zip.data)
		    && UNIV_LIKELY(buf_block_get_state(block)
				   == BUF_BLOCK_FILE_PAGE)
		    && UNIV_UNLIKELY
		    (memcmp(write_buf + len2 + (FIL_PAGE_LSN + 4),
			    write_buf + len2
			    + (UNIV_PAGE_SIZE
			       - FIL_PAGE_END_LSN_OLD_CHKSUM + 4), 4))) {
			ut_print_timestamp(stderr);
			fprintf(stderr,
				"  InnoDB: ERROR: The page to be written"
				" seems corrupt!\n"
				"InnoDB: The lsn fields do not match!"
				" Noticed in the doublewrite block1.\n");
		}
	}

	if (trx_doublewrite->first_free <= TRX_SYS_DOUBLEWRITE_BLOCK_SIZE) {
		goto flush;
	}

	len = (trx_doublewrite->first_free - TRX_SYS_DOUBLEWRITE_BLOCK_SIZE)
		* UNIV_PAGE_SIZE;

	write_buf = trx_doublewrite->write_buf
		+ TRX_SYS_DOUBLEWRITE_BLOCK_SIZE * UNIV_PAGE_SIZE;
	ut_ad(i == TRX_SYS_DOUBLEWRITE_BLOCK_SIZE);

	fil_io(OS_FILE_WRITE, TRUE, TRX_SYS_SPACE, 0,
	       trx_doublewrite->block2, 0, len,
	       (void*) write_buf, NULL);

	for (len2 = 0; len2 + UNIV_PAGE_SIZE <= len;
	     len2 += UNIV_PAGE_SIZE, i++) {
		const buf_block_t* block = (buf_block_t*)
			trx_doublewrite->buf_block_arr[i];

		if (UNIV_LIKELY(!block->page.zip.data)
		    && UNIV_LIKELY(buf_block_get_state(block)
				   == BUF_BLOCK_FILE_PAGE)
		    && UNIV_UNLIKELY
		    (memcmp(write_buf + len2 + (FIL_PAGE_LSN + 4),
			    write_buf + len2
			    + (UNIV_PAGE_SIZE
			       - FIL_PAGE_END_LSN_OLD_CHKSUM + 4), 4))) {
			ut_print_timestamp(stderr);
			fprintf(stderr,
				"  InnoDB: ERROR: The page to be"
				" written seems corrupt!\n"
				"InnoDB: The lsn fields do not match!"
				" Noticed in"
				" the doublewrite block2.\n");
		}
	}

flush:
	/* Now flush the doublewrite buffer data to disk */

	fil_flush(TRX_SYS_SPACE);

	/* We know that the writes have been flushed to disk now
	and in recovery we will find them in the doublewrite buffer
	blocks. Next do the writes to the intended positions. */

	for (i = 0; i < trx_doublewrite->first_free; i++) {
		const buf_block_t* block = (buf_block_t*)
			trx_doublewrite->buf_block_arr[i];

		ut_a(buf_page_in_file(&block->page));
		if (UNIV_LIKELY_NULL(block->page.zip.data)) {
			fil_io(OS_FILE_WRITE | OS_AIO_SIMULATED_WAKE_LATER,
			       FALSE, buf_page_get_space(&block->page),
			       buf_page_get_zip_size(&block->page),
			       buf_page_get_page_no(&block->page), 0,
			       buf_page_get_zip_size(&block->page),
			       (void*)block->page.zip.data,
			       (void*)block);

			/* Increment the counter of I/O operations used
			for selecting LRU policy. */
			buf_LRU_stat_inc_io();

			continue;
		}

		ut_a(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);

		if (UNIV_UNLIKELY(memcmp(block->frame + (FIL_PAGE_LSN + 4),
					 block->frame
					 + (UNIV_PAGE_SIZE
					    - FIL_PAGE_END_LSN_OLD_CHKSUM + 4),
					 4))) {
			ut_print_timestamp(stderr);
			fprintf(stderr,
				"  InnoDB: ERROR: The page to be written"
				" seems corrupt!\n"
				"InnoDB: The lsn fields do not match!"
				" Noticed in the buffer pool\n"
				"InnoDB: after posting and flushing"
				" the doublewrite buffer.\n"
				"InnoDB: Page buf fix count %lu,"
				" io fix %lu, state %lu\n",
				(ulong)block->page.buf_fix_count,
				(ulong)buf_block_get_io_fix(block),
				(ulong)buf_block_get_state(block));
		}

		fil_io(OS_FILE_WRITE | OS_AIO_SIMULATED_WAKE_LATER,
		       FALSE, buf_block_get_space(block), 0,
		       buf_block_get_page_no(block), 0, UNIV_PAGE_SIZE,
		       (void*)block->frame, (void*)block);

		/* Increment the counter of I/O operations used
		for selecting LRU policy. */
		buf_LRU_stat_inc_io();
	}

581 582
	/* Sync the writes to the disk. */
	buf_flush_sync_datafiles();
583 584 585 586 587 588 589

	/* We can now reuse the doublewrite memory buffer: */
	trx_doublewrite->first_free = 0;

	mutex_exit(&(trx_doublewrite->mutex));
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
590
/********************************************************************//**
591 592 593 594 595 596 597
Posts a buffer page for writing. If the doublewrite memory buffer is
full, calls buf_flush_buffered_writes and waits for for free space to
appear. */
static
void
buf_flush_post_to_doublewrite_buf(
/*==============================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
598
	buf_page_t*	bpage)	/*!< in: buffer block to write */
599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647
{
	ulint	zip_size;
try_again:
	mutex_enter(&(trx_doublewrite->mutex));

	ut_a(buf_page_in_file(bpage));

	if (trx_doublewrite->first_free
	    >= 2 * TRX_SYS_DOUBLEWRITE_BLOCK_SIZE) {
		mutex_exit(&(trx_doublewrite->mutex));

		buf_flush_buffered_writes();

		goto try_again;
	}

	zip_size = buf_page_get_zip_size(bpage);

	if (UNIV_UNLIKELY(zip_size)) {
		/* Copy the compressed page and clear the rest. */
		memcpy(trx_doublewrite->write_buf
		       + UNIV_PAGE_SIZE * trx_doublewrite->first_free,
		       bpage->zip.data, zip_size);
		memset(trx_doublewrite->write_buf
		       + UNIV_PAGE_SIZE * trx_doublewrite->first_free
		       + zip_size, 0, UNIV_PAGE_SIZE - zip_size);
	} else {
		ut_a(buf_page_get_state(bpage) == BUF_BLOCK_FILE_PAGE);

		memcpy(trx_doublewrite->write_buf
		       + UNIV_PAGE_SIZE * trx_doublewrite->first_free,
		       ((buf_block_t*) bpage)->frame, UNIV_PAGE_SIZE);
	}

	trx_doublewrite->buf_block_arr[trx_doublewrite->first_free] = bpage;

	trx_doublewrite->first_free++;

	if (trx_doublewrite->first_free
	    >= 2 * TRX_SYS_DOUBLEWRITE_BLOCK_SIZE) {
		mutex_exit(&(trx_doublewrite->mutex));

		buf_flush_buffered_writes();

		return;
	}

	mutex_exit(&(trx_doublewrite->mutex));
}
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
648
#endif /* !UNIV_HOTBACKUP */
649

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
650
/********************************************************************//**
651 652 653 654 655
Initializes a page for writing to the tablespace. */
UNIV_INTERN
void
buf_flush_init_for_writing(
/*=======================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
656 657 658
	byte*		page,		/*!< in/out: page */
	void*		page_zip_,	/*!< in/out: compressed page, or NULL */
	ib_uint64_t	newest_lsn)	/*!< in: newest modification lsn
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
					to the page */
{
	ut_ad(page);

	if (page_zip_) {
		page_zip_des_t*	page_zip = page_zip_;
		ulint		zip_size = page_zip_get_size(page_zip);
		ut_ad(zip_size);
		ut_ad(ut_is_2pow(zip_size));
		ut_ad(zip_size <= UNIV_PAGE_SIZE);

		switch (UNIV_EXPECT(fil_page_get_type(page), FIL_PAGE_INDEX)) {
		case FIL_PAGE_TYPE_ALLOCATED:
		case FIL_PAGE_INODE:
		case FIL_PAGE_IBUF_BITMAP:
		case FIL_PAGE_TYPE_FSP_HDR:
		case FIL_PAGE_TYPE_XDES:
			/* These are essentially uncompressed pages. */
			memcpy(page_zip->data, page, zip_size);
			/* fall through */
		case FIL_PAGE_TYPE_ZBLOB:
		case FIL_PAGE_TYPE_ZBLOB2:
		case FIL_PAGE_INDEX:
			mach_write_ull(page_zip->data
				       + FIL_PAGE_LSN, newest_lsn);
			memset(page_zip->data + FIL_PAGE_FILE_FLUSH_LSN, 0, 8);
			mach_write_to_4(page_zip->data
					+ FIL_PAGE_SPACE_OR_CHKSUM,
					srv_use_checksums
					? page_zip_calc_checksum(
						page_zip->data, zip_size)
					: BUF_NO_CHECKSUM_MAGIC);
			return;
		}

Vadim Tkachenko's avatar
Vadim Tkachenko committed
694 695 696 697 698 699 700
		ut_print_timestamp(stderr);
		fputs("  InnoDB: ERROR: The compressed page to be written"
		      " seems corrupt:", stderr);
		ut_print_buf(stderr, page, zip_size);
		fputs("\nInnoDB: Possibly older version of the page:", stderr);
		ut_print_buf(stderr, page_zip->data, zip_size);
		putc('\n', stderr);
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
		ut_error;
	}

	/* Write the newest modification lsn to the page header and trailer */
	mach_write_ull(page + FIL_PAGE_LSN, newest_lsn);

	mach_write_ull(page + UNIV_PAGE_SIZE - FIL_PAGE_END_LSN_OLD_CHKSUM,
		       newest_lsn);

	/* Store the new formula checksum */

	mach_write_to_4(page + FIL_PAGE_SPACE_OR_CHKSUM,
			srv_use_checksums
			? buf_calc_page_new_checksum(page)
			: BUF_NO_CHECKSUM_MAGIC);

	/* We overwrite the first 4 bytes of the end lsn field to store
	the old formula checksum. Since it depends also on the field
	FIL_PAGE_SPACE_OR_CHKSUM, it has to be calculated after storing the
	new formula checksum. */

	mach_write_to_4(page + UNIV_PAGE_SIZE - FIL_PAGE_END_LSN_OLD_CHKSUM,
			srv_use_checksums
			? buf_calc_page_old_checksum(page)
			: BUF_NO_CHECKSUM_MAGIC);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
728 729
#ifndef UNIV_HOTBACKUP
/********************************************************************//**
730 731 732 733 734 735 736
Does an asynchronous write of a buffer page. NOTE: in simulated aio and
also when the doublewrite buffer is used, we must call
buf_flush_buffered_writes after we have posted a batch of writes! */
static
void
buf_flush_write_block_low(
/*======================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
737
	buf_page_t*	bpage)	/*!< in: buffer block to write */
738 739 740 741 742 743 744 745 746
{
	ulint	zip_size	= buf_page_get_zip_size(bpage);
	page_t*	frame		= NULL;
#ifdef UNIV_LOG_DEBUG
	static ibool univ_log_debug_warned;
#endif /* UNIV_LOG_DEBUG */

	ut_ad(buf_page_in_file(bpage));

Vadim Tkachenko's avatar
Vadim Tkachenko committed
747 748 749 750 751
	/* We are not holding buf_pool_mutex or block_mutex here.
	Nevertheless, it is safe to access bpage, because it is
	io_fixed and oldest_modification != 0.  Thus, it cannot be
	relocated in the buffer pool or removed from flush_list or
	LRU_list. */
Vadim Tkachenko's avatar
Vadim Tkachenko committed
752 753 754
	//ut_ad(!buf_pool_mutex_own());
	ut_ad(!mutex_own(&LRU_list_mutex));
	ut_ad(!mutex_own(&flush_list_mutex));
Vadim Tkachenko's avatar
Vadim Tkachenko committed
755 756 757 758
	ut_ad(!mutex_own(buf_page_get_mutex(bpage)));
	ut_ad(buf_page_get_io_fix(bpage) == BUF_IO_WRITE);
	ut_ad(bpage->oldest_modification != 0);

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
#ifdef UNIV_IBUF_COUNT_DEBUG
	ut_a(ibuf_count_get(bpage->space, bpage->offset) == 0);
#endif
	ut_ad(bpage->newest_modification != 0);

#ifdef UNIV_LOG_DEBUG
	if (!univ_log_debug_warned) {
		univ_log_debug_warned = TRUE;
		fputs("Warning: cannot force log to disk if"
		      " UNIV_LOG_DEBUG is defined!\n"
		      "Crash recovery will not work!\n",
		      stderr);
	}
#else
	/* Force the log to the disk before writing the modified block */
	log_write_up_to(bpage->newest_modification, LOG_WAIT_ALL_GROUPS, TRUE);
#endif
	switch (buf_page_get_state(bpage)) {
	case BUF_BLOCK_ZIP_FREE:
	case BUF_BLOCK_ZIP_PAGE: /* The page should be dirty. */
	case BUF_BLOCK_NOT_USED:
	case BUF_BLOCK_READY_FOR_USE:
	case BUF_BLOCK_MEMORY:
	case BUF_BLOCK_REMOVE_HASH:
		ut_error;
		break;
	case BUF_BLOCK_ZIP_DIRTY:
		frame = bpage->zip.data;
		if (UNIV_LIKELY(srv_use_checksums)) {
			ut_a(mach_read_from_4(frame + FIL_PAGE_SPACE_OR_CHKSUM)
			     == page_zip_calc_checksum(frame, zip_size));
		}
		mach_write_ull(frame + FIL_PAGE_LSN,
			       bpage->newest_modification);
		memset(frame + FIL_PAGE_FILE_FLUSH_LSN, 0, 8);
		break;
	case BUF_BLOCK_FILE_PAGE:
		frame = bpage->zip.data;
		if (!frame) {
			frame = ((buf_block_t*) bpage)->frame;
		}

		buf_flush_init_for_writing(((buf_block_t*) bpage)->frame,
					   bpage->zip.data
					   ? &bpage->zip : NULL,
					   bpage->newest_modification);
		break;
	}

	if (!srv_use_doublewrite_buf || !trx_doublewrite) {
		fil_io(OS_FILE_WRITE | OS_AIO_SIMULATED_WAKE_LATER,
		       FALSE, buf_page_get_space(bpage), zip_size,
		       buf_page_get_page_no(bpage), 0,
		       zip_size ? zip_size : UNIV_PAGE_SIZE,
		       frame, bpage);
	} else {
		buf_flush_post_to_doublewrite_buf(bpage);
	}
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
819
/********************************************************************//**
Vadim Tkachenko's avatar
Vadim Tkachenko committed
820 821 822 823 824 825
Writes a flushable page asynchronously from the buffer pool to a file.
NOTE: in simulated aio we must call
os_aio_simulated_wake_handler_threads after we have posted a batch of
writes! NOTE: buf_pool_mutex and buf_page_get_mutex(bpage) must be
held upon entering this function, and they will be released by this
function. */
826
static
Vadim Tkachenko's avatar
Vadim Tkachenko committed
827 828 829
void
buf_flush_page(
/*===========*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
830 831
	buf_page_t*	bpage,		/*!< in: buffer control block */
	enum buf_flush	flush_type)	/*!< in: BUF_FLUSH_LRU
Vadim Tkachenko's avatar
Vadim Tkachenko committed
832
					or BUF_FLUSH_LIST */
833 834
{
	mutex_t*	block_mutex;
Vadim Tkachenko's avatar
Vadim Tkachenko committed
835
	ibool		is_uncompressed;
836

Vadim Tkachenko's avatar
Vadim Tkachenko committed
837
	ut_ad(flush_type == BUF_FLUSH_LRU || flush_type == BUF_FLUSH_LIST);
Vadim Tkachenko's avatar
Vadim Tkachenko committed
838 839 840 841 842
	//ut_ad(buf_pool_mutex_own());
#ifdef UNIV_SYNC_DEBUG
	ut_ad(rw_lock_own(&page_hash_latch, RW_LOCK_EX)
	      || rw_lock_own(&page_hash_latch, RW_LOCK_SHARED));
#endif
Vadim Tkachenko's avatar
Vadim Tkachenko committed
843
	ut_ad(buf_page_in_file(bpage));
844 845

	block_mutex = buf_page_get_mutex(bpage);
Vadim Tkachenko's avatar
Vadim Tkachenko committed
846
	ut_ad(mutex_own(block_mutex));
847

Vadim Tkachenko's avatar
Vadim Tkachenko committed
848 849 850
	mutex_enter(&buf_pool_mutex);
	rw_lock_s_unlock(&page_hash_latch);

Vadim Tkachenko's avatar
Vadim Tkachenko committed
851
	ut_ad(buf_flush_ready_for_flush(bpage, flush_type));
852

Vadim Tkachenko's avatar
Vadim Tkachenko committed
853
	buf_page_set_io_fix(bpage, BUF_IO_WRITE);
854

Vadim Tkachenko's avatar
Vadim Tkachenko committed
855
	buf_page_set_flush_type(bpage, flush_type);
856

Vadim Tkachenko's avatar
Vadim Tkachenko committed
857
	if (buf_pool->n_flush[flush_type] == 0) {
858

Vadim Tkachenko's avatar
Vadim Tkachenko committed
859 860
		os_event_reset(buf_pool->no_flush[flush_type]);
	}
861

Vadim Tkachenko's avatar
Vadim Tkachenko committed
862
	buf_pool->n_flush[flush_type]++;
863

Vadim Tkachenko's avatar
Vadim Tkachenko committed
864 865 866 867 868 869
	is_uncompressed = (buf_page_get_state(bpage) == BUF_BLOCK_FILE_PAGE);
	ut_ad(is_uncompressed == (block_mutex != &buf_pool_zip_mutex));

	switch (flush_type) {
		ibool	is_s_latched;
	case BUF_FLUSH_LIST:
870 871 872 873
		/* If the simulated aio thread is not running, we must
		not wait for any latch, as we may end up in a deadlock:
		if buf_fix_count == 0, then we know we need not wait */

Vadim Tkachenko's avatar
Vadim Tkachenko committed
874 875
		is_s_latched = (bpage->buf_fix_count == 0);
		if (is_s_latched && is_uncompressed) {
876 877 878 879 880
			rw_lock_s_lock_gen(&((buf_block_t*) bpage)->lock,
					   BUF_IO_WRITE);
		}

		mutex_exit(block_mutex);
Vadim Tkachenko's avatar
Vadim Tkachenko committed
881 882
		//buf_pool_mutex_exit();
		mutex_exit(&buf_pool_mutex);
883

Vadim Tkachenko's avatar
Vadim Tkachenko committed
884 885 886 887 888 889 890
		/* Even though bpage is not protected by any mutex at
		this point, it is safe to access bpage, because it is
		io_fixed and oldest_modification != 0.  Thus, it
		cannot be relocated in the buffer pool or removed from
		flush_list or LRU_list. */

		if (!is_s_latched) {
891 892
			buf_flush_buffered_writes();

Vadim Tkachenko's avatar
Vadim Tkachenko committed
893
			if (is_uncompressed) {
894 895 896 897 898 899 900 901 902 903 904 905
				rw_lock_s_lock_gen(&((buf_block_t*) bpage)
						   ->lock, BUF_IO_WRITE);
			}
		}

		break;

	case BUF_FLUSH_LRU:
		/* VERY IMPORTANT:
		Because any thread may call the LRU flush, even when owning
		locks on pages, to avoid deadlocks, we must make sure that the
		s-lock is acquired on the page without waiting: this is
Vadim Tkachenko's avatar
Vadim Tkachenko committed
906 907
		accomplished because buf_flush_ready_for_flush() must hold,
		and that requires the page not to be bufferfixed. */
908

Vadim Tkachenko's avatar
Vadim Tkachenko committed
909
		if (is_uncompressed) {
910 911 912 913 914 915 916 917 918
			rw_lock_s_lock_gen(&((buf_block_t*) bpage)->lock,
					   BUF_IO_WRITE);
		}

		/* Note that the s-latch is acquired before releasing the
		buf_pool mutex: this ensures that the latch is acquired
		immediately. */

		mutex_exit(block_mutex);
Vadim Tkachenko's avatar
Vadim Tkachenko committed
919 920
		//buf_pool_mutex_exit();
		mutex_exit(&buf_pool_mutex);
921 922 923 924 925 926
		break;

	default:
		ut_error;
	}

Vadim Tkachenko's avatar
Vadim Tkachenko committed
927 928 929 930 931
	/* Even though bpage is not protected by any mutex at this
	point, it is safe to access bpage, because it is io_fixed and
	oldest_modification != 0.  Thus, it cannot be relocated in the
	buffer pool or removed from flush_list or LRU_list. */

932 933 934 935 936 937 938 939 940 941
#ifdef UNIV_DEBUG
	if (buf_debug_prints) {
		fprintf(stderr,
			"Flushing %u space %u page %u\n",
			flush_type, bpage->space, bpage->offset);
	}
#endif /* UNIV_DEBUG */
	buf_flush_write_block_low(bpage);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
942 943 944
/***********************************************************//**
Flushes to disk all flushable pages within the flush area.
@return	number of pages flushed */
945 946 947 948
static
ulint
buf_flush_try_neighbors(
/*====================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
949 950 951
	ulint		space,		/*!< in: space id */
	ulint		offset,		/*!< in: page offset */
	enum buf_flush	flush_type,	/*!< in: BUF_FLUSH_LRU or
952
					BUF_FLUSH_LIST */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
953
	ulint		flush_neighbors)
954 955 956 957 958 959 960 961
{
	buf_page_t*	bpage;
	ulint		low, high;
	ulint		count		= 0;
	ulint		i;

	ut_ad(flush_type == BUF_FLUSH_LRU || flush_type == BUF_FLUSH_LIST);

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
962
	if (UT_LIST_GET_LEN(buf_pool->LRU) < BUF_LRU_OLD_MIN_LEN || !flush_neighbors) {
963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984
		/* If there is little space, it is better not to flush any
		block except from the end of the LRU list */

		low = offset;
		high = offset + 1;
	} else {
		/* When flushed, dirty blocks are searched in neighborhoods of
		this size, and flushed along with the original page. */

		ulint	buf_flush_area	= ut_min(BUF_READ_AHEAD_AREA,
						 buf_pool->curr_size / 16);

		low = (offset / buf_flush_area) * buf_flush_area;
		high = (offset / buf_flush_area + 1) * buf_flush_area;
	}

	/* fprintf(stderr, "Flush area: low %lu high %lu\n", low, high); */

	if (high > fil_space_get_size(space)) {
		high = fil_space_get_size(space);
	}

Vadim Tkachenko's avatar
Vadim Tkachenko committed
985 986
	//buf_pool_mutex_enter();
	rw_lock_s_lock(&page_hash_latch);
987 988 989 990 991 992 993 994

	for (i = low; i < high; i++) {

		bpage = buf_page_hash_get(space, i);

		if (!bpage) {

			continue;
Vadim Tkachenko's avatar
Vadim Tkachenko committed
995
		}
996

Vadim Tkachenko's avatar
Vadim Tkachenko committed
997
		ut_a(buf_page_in_file(bpage));
998

Vadim Tkachenko's avatar
Vadim Tkachenko committed
999 1000
		/* We avoid flushing 'non-old' blocks in an LRU flush,
		because the flushed blocks are soon freed */
1001

Vadim Tkachenko's avatar
Vadim Tkachenko committed
1002 1003 1004
		if (flush_type != BUF_FLUSH_LRU
		    || i == offset
		    || buf_page_is_old(bpage)) {
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1005
			mutex_t* block_mutex = buf_page_get_mutex_enter(bpage);
1006

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1007
			if (block_mutex && buf_flush_ready_for_flush(bpage, flush_type)
1008 1009 1010 1011 1012 1013 1014 1015 1016
			    && (i == offset || !bpage->buf_fix_count)) {
				/* We only try to flush those
				neighbors != offset where the buf fix count is
				zero, as we then know that we probably can
				latch the page without a semaphore wait.
				Semaphore waits are expensive because we must
				flush the doublewrite buffer before we start
				waiting. */

Vadim Tkachenko's avatar
Vadim Tkachenko committed
1017 1018 1019
				buf_flush_page(bpage, flush_type);
				ut_ad(!mutex_own(block_mutex));
				count++;
1020

Vadim Tkachenko's avatar
Vadim Tkachenko committed
1021 1022
				//buf_pool_mutex_enter();
				rw_lock_s_lock(&page_hash_latch);
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1023
			} else if (block_mutex) {
1024 1025 1026 1027 1028
				mutex_exit(block_mutex);
			}
		}
	}

Vadim Tkachenko's avatar
Vadim Tkachenko committed
1029 1030
	//buf_pool_mutex_exit();
	rw_lock_s_unlock(&page_hash_latch);
1031 1032 1033 1034

	return(count);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1035
/*******************************************************************//**
1036 1037 1038 1039
This utility flushes dirty blocks from the end of the LRU list or flush_list.
NOTE 1: in the case of an LRU flush the calling thread may own latches to
pages: to avoid deadlocks, this function must be written so that it cannot
end up waiting for these latches! NOTE 2: in the case of a flush list flush,
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1040 1041 1042
the calling thread is not allowed to own any latches on pages!
@return number of blocks for which the write request was queued;
ULINT_UNDEFINED if there was a flush of the same type already running */
1043 1044 1045 1046
UNIV_INTERN
ulint
buf_flush_batch(
/*============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1047
	enum buf_flush	flush_type,	/*!< in: BUF_FLUSH_LRU or
1048 1049 1050
					BUF_FLUSH_LIST; if BUF_FLUSH_LIST,
					then the caller must not own any
					latches on pages */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1051
	ulint		min_n,		/*!< in: wished minimum mumber of blocks
1052 1053
					flushed (it is not guaranteed that the
					actual number is that big, though) */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1054
	ib_uint64_t	lsn_limit)	/*!< in the case BUF_FLUSH_LIST all
1055 1056 1057 1058 1059 1060
					blocks whose oldest_modification is
					smaller than this should be flushed
					(if their number does not exceed
					min_n), otherwise ignored */
{
	buf_page_t*	bpage;
1061
	buf_page_t*	prev_bpage	= NULL;
1062 1063 1064 1065
	ulint		page_count	= 0;
	ulint		old_page_count;
	ulint		space;
	ulint		offset;
Vadim Tkachenko's avatar
Vadim Tkachenko committed
1066
	ulint		remaining	= 0;
1067 1068 1069 1070 1071 1072 1073

	ut_ad((flush_type == BUF_FLUSH_LRU)
	      || (flush_type == BUF_FLUSH_LIST));
#ifdef UNIV_SYNC_DEBUG
	ut_ad((flush_type != BUF_FLUSH_LIST)
	      || sync_thread_levels_empty_gen(TRUE));
#endif /* UNIV_SYNC_DEBUG */
Vadim Tkachenko's avatar
Vadim Tkachenko committed
1074 1075
	//buf_pool_mutex_enter();
	mutex_enter(&buf_pool_mutex);
1076 1077 1078 1079 1080 1081

	if ((buf_pool->n_flush[flush_type] > 0)
	    || (buf_pool->init_flush[flush_type] == TRUE)) {

		/* There is already a flush batch of the same type running */

Vadim Tkachenko's avatar
Vadim Tkachenko committed
1082 1083
		//buf_pool_mutex_exit();
		mutex_exit(&buf_pool_mutex);
1084 1085 1086 1087 1088 1089

		return(ULINT_UNDEFINED);
	}

	buf_pool->init_flush[flush_type] = TRUE;

Vadim Tkachenko's avatar
Vadim Tkachenko committed
1090 1091 1092 1093 1094 1095
	mutex_exit(&buf_pool_mutex);

	if (flush_type == BUF_FLUSH_LRU) {
		mutex_enter(&LRU_list_mutex);
	}

1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111
	for (;;) {
flush_next:
		/* If we have flushed enough, leave the loop */
		if (page_count >= min_n) {

			break;
		}

		/* Start from the end of the list looking for a suitable
		block to be flushed. */

		if (flush_type == BUF_FLUSH_LRU) {
			bpage = UT_LIST_GET_LAST(buf_pool->LRU);
		} else {
			ut_ad(flush_type == BUF_FLUSH_LIST);

Vadim Tkachenko's avatar
Vadim Tkachenko committed
1112 1113
			mutex_enter(&flush_list_mutex);
			remaining = UT_LIST_GET_LEN(buf_pool->flush_list);
1114
			bpage = UT_LIST_GET_LAST(buf_pool->flush_list);
1115 1116 1117
			if (bpage) {
				prev_bpage = UT_LIST_GET_PREV(flush_list, bpage);
			}
Vadim Tkachenko's avatar
Vadim Tkachenko committed
1118
			mutex_exit(&flush_list_mutex);
1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134
			if (!bpage
			    || bpage->oldest_modification >= lsn_limit) {
				/* We have flushed enough */

				break;
			}
			ut_ad(bpage->in_flush_list);
		}

		/* Note that after finding a single flushable page, we try to
		flush also all its neighbors, and after that start from the
		END of the LRU list or flush list again: the list may change
		during the flushing and we cannot safely preserve within this
		function a pointer to a block in the list! */

		do {
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1135
			mutex_t*block_mutex = buf_page_get_mutex_enter(bpage);
Vadim Tkachenko's avatar
Vadim Tkachenko committed
1136
			ibool	ready;
1137

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1138
			//ut_a(buf_page_in_file(bpage));
1139

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1140 1141 1142 1143 1144 1145
			if (block_mutex) {
				ready = buf_flush_ready_for_flush(bpage, flush_type);
				mutex_exit(block_mutex);
			} else {
				ready = FALSE;
			}
1146

Vadim Tkachenko's avatar
Vadim Tkachenko committed
1147
			if (ready) {
1148 1149 1150
				space = buf_page_get_space(bpage);
				offset = buf_page_get_page_no(bpage);

Vadim Tkachenko's avatar
Vadim Tkachenko committed
1151 1152 1153 1154
				//buf_pool_mutex_exit();
				if (flush_type == BUF_FLUSH_LRU) {
					mutex_exit(&LRU_list_mutex);
				}
1155 1156 1157 1158 1159

				old_page_count = page_count;

				/* Try to flush also all the neighbors */
				page_count += buf_flush_try_neighbors(
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1160
					space, offset, flush_type, srv_flush_neighbor_pages);
1161 1162 1163 1164 1165
				/* fprintf(stderr,
				"Flush type %lu, page no %lu, neighb %lu\n",
				flush_type, offset,
				page_count - old_page_count); */

Vadim Tkachenko's avatar
Vadim Tkachenko committed
1166 1167 1168 1169
				//buf_pool_mutex_enter();
				if (flush_type == BUF_FLUSH_LRU) {
					mutex_enter(&LRU_list_mutex);
				}
1170 1171 1172 1173 1174 1175 1176
				goto flush_next;

			} else if (flush_type == BUF_FLUSH_LRU) {
				bpage = UT_LIST_GET_PREV(LRU, bpage);
			} else {
				ut_ad(flush_type == BUF_FLUSH_LIST);

Vadim Tkachenko's avatar
Vadim Tkachenko committed
1177 1178 1179
				mutex_enter(&flush_list_mutex);
				bpage = UT_LIST_GET_PREV(flush_list, bpage);
				//ut_ad(!bpage || bpage->in_flush_list); /* optimistic */
1180 1181 1182 1183 1184 1185 1186
				if (bpage != prev_bpage) {
					/* the search may warp.. retrying */
					bpage = NULL;
				}
				if (bpage) {
					prev_bpage = UT_LIST_GET_PREV(flush_list, bpage);
				}
Vadim Tkachenko's avatar
Vadim Tkachenko committed
1187 1188
				mutex_exit(&flush_list_mutex);
				remaining--;
1189 1190 1191
			}
		} while (bpage != NULL);

Vadim Tkachenko's avatar
Vadim Tkachenko committed
1192 1193 1194
		if (remaining)
			goto flush_next;

1195 1196 1197 1198 1199
		/* If we could not find anything to flush, leave the loop */

		break;
	}

Vadim Tkachenko's avatar
Vadim Tkachenko committed
1200 1201 1202 1203 1204 1205
	if (flush_type == BUF_FLUSH_LRU) {
		mutex_exit(&LRU_list_mutex);
	}

	mutex_enter(&buf_pool_mutex);

1206 1207
	buf_pool->init_flush[flush_type] = FALSE;

Vadim Tkachenko's avatar
Vadim Tkachenko committed
1208
	if (buf_pool->n_flush[flush_type] == 0) {
1209 1210 1211 1212 1213 1214

		/* The running flush batch has ended */

		os_event_set(buf_pool->no_flush[flush_type]);
	}

Vadim Tkachenko's avatar
Vadim Tkachenko committed
1215 1216
	//buf_pool_mutex_exit();
	mutex_exit(&buf_pool_mutex);
1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232

	buf_flush_buffered_writes();

#ifdef UNIV_DEBUG
	if (buf_debug_prints && page_count > 0) {
		ut_a(flush_type == BUF_FLUSH_LRU
		     || flush_type == BUF_FLUSH_LIST);
		fprintf(stderr, flush_type == BUF_FLUSH_LRU
			? "Flushed %lu pages in LRU flush\n"
			: "Flushed %lu pages in flush list flush\n",
			(ulong) page_count);
	}
#endif /* UNIV_DEBUG */

	srv_buf_pool_flushed += page_count;

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1233 1234 1235 1236 1237 1238 1239
	/* We keep track of all flushes happening as part of LRU
	flush. When estimating the desired rate at which flush_list
	should be flushed we factor in this value. */
	if (flush_type == BUF_FLUSH_LRU) {
		buf_lru_flush_page_count += page_count;
	}

1240 1241 1242
	return(page_count);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1243
/******************************************************************//**
1244 1245 1246 1247 1248
Waits until a flush batch of the given type ends */
UNIV_INTERN
void
buf_flush_wait_batch_end(
/*=====================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1249
	enum buf_flush	type)	/*!< in: BUF_FLUSH_LRU or BUF_FLUSH_LIST */
1250 1251 1252 1253 1254 1255
{
	ut_ad((type == BUF_FLUSH_LRU) || (type == BUF_FLUSH_LIST));

	os_event_wait(buf_pool->no_flush[type]);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1256
/******************************************************************//**
1257 1258
Gives a recommendation of how many blocks should be flushed to establish
a big enough margin of replaceable blocks near the end of the LRU list
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1259 1260 1261
and in the free list.
@return number of blocks which should be flushed from the end of the
LRU list */
1262 1263 1264 1265 1266 1267 1268 1269
static
ulint
buf_flush_LRU_recommendation(void)
/*==============================*/
{
	buf_page_t*	bpage;
	ulint		n_replaceable;
	ulint		distance	= 0;
Vadim Tkachenko's avatar
Vadim Tkachenko committed
1270 1271 1272 1273
	ibool		have_LRU_mutex = FALSE;

	if(UT_LIST_GET_LEN(buf_pool->unzip_LRU))
		have_LRU_mutex = TRUE;
1274

1275
	//buf_pool_mutex_enter();
Vadim Tkachenko's avatar
Vadim Tkachenko committed
1276
	if (have_LRU_mutex)
Vadim Tkachenko's avatar
Vadim Tkachenko committed
1277
		mutex_enter(&LRU_list_mutex);
1278 1279 1280 1281 1282 1283 1284 1285 1286 1287

	n_replaceable = UT_LIST_GET_LEN(buf_pool->free);

	bpage = UT_LIST_GET_LAST(buf_pool->LRU);

	while ((bpage != NULL)
	       && (n_replaceable < BUF_FLUSH_FREE_BLOCK_MARGIN
		   + BUF_FLUSH_EXTRA_MARGIN)
	       && (distance < BUF_LRU_FREE_SEARCH_LEN)) {

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1288
		mutex_t* block_mutex;
Vadim Tkachenko's avatar
Vadim Tkachenko committed
1289 1290 1291 1292 1293
		if (!bpage->in_LRU_list) {
			/* reatart. but it is very optimistic */
			bpage = UT_LIST_GET_LAST(buf_pool->LRU);
			continue;
		}
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1294
		block_mutex = buf_page_get_mutex_enter(bpage);
1295

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1296
		if (block_mutex && buf_flush_ready_for_replace(bpage)) {
1297 1298 1299
			n_replaceable++;
		}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1300 1301 1302
		if (block_mutex) {
			mutex_exit(block_mutex);
		}
1303 1304 1305 1306 1307 1308

		distance++;

		bpage = UT_LIST_GET_PREV(LRU, bpage);
	}

1309
	//buf_pool_mutex_exit();
Vadim Tkachenko's avatar
Vadim Tkachenko committed
1310
	if (have_LRU_mutex)
Vadim Tkachenko's avatar
Vadim Tkachenko committed
1311
		mutex_exit(&LRU_list_mutex);
1312 1313 1314 1315 1316 1317 1318 1319 1320 1321

	if (n_replaceable >= BUF_FLUSH_FREE_BLOCK_MARGIN) {

		return(0);
	}

	return(BUF_FLUSH_FREE_BLOCK_MARGIN + BUF_FLUSH_EXTRA_MARGIN
	       - n_replaceable);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1322
/*********************************************************************//**
1323 1324 1325 1326 1327 1328 1329
Flushes pages from the end of the LRU list if there is too small a margin
of replaceable pages there or in the free list. VERY IMPORTANT: this function
is called also by threads which have locks on pages. To avoid deadlocks, we
flush only pages such that the s-lock required for flushing can be acquired
immediately, without waiting. */
UNIV_INTERN
void
1330
buf_flush_free_margin(
1331
/*=======================*/
1332
	ibool	wait)
1333 1334 1335 1336 1337 1338 1339 1340
{
	ulint	n_to_flush;
	ulint	n_flushed;

	n_to_flush = buf_flush_LRU_recommendation();

	if (n_to_flush > 0) {
		n_flushed = buf_flush_batch(BUF_FLUSH_LRU, n_to_flush, 0);
1341
		if (wait && n_flushed == ULINT_UNDEFINED) {
1342 1343 1344 1345 1346 1347 1348 1349
			/* There was an LRU type flush batch already running;
			let us wait for it to end */

			buf_flush_wait_batch_end(BUF_FLUSH_LRU);
		}
	}
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 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 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460
/*********************************************************************
Update the historical stats that we are collecting for flush rate
heuristics at the end of each interval.
Flush rate heuristic depends on (a) rate of redo log generation and
(b) the rate at which LRU flush is happening. */
UNIV_INTERN
void
buf_flush_stat_update(void)
/*=======================*/
{
	buf_flush_stat_t*	item;
	ib_uint64_t		lsn_diff;
	ib_uint64_t		lsn;
	ulint			n_flushed;

	lsn = log_get_lsn();
	if (buf_flush_stat_cur.redo == 0) {
		/* First time around. Just update the current LSN
		and return. */
		buf_flush_stat_cur.redo = lsn;
		return;
	}

	item = &buf_flush_stat_arr[buf_flush_stat_arr_ind];

	/* values for this interval */
	lsn_diff = lsn - buf_flush_stat_cur.redo;
	n_flushed = buf_lru_flush_page_count
		    - buf_flush_stat_cur.n_flushed;

	/* add the current value and subtract the obsolete entry. */
	buf_flush_stat_sum.redo += lsn_diff - item->redo;
	buf_flush_stat_sum.n_flushed += n_flushed - item->n_flushed;

	/* put current entry in the array. */
	item->redo = lsn_diff;
	item->n_flushed = n_flushed;

	/* update the index */
	buf_flush_stat_arr_ind++;
	buf_flush_stat_arr_ind %= BUF_FLUSH_STAT_N_INTERVAL;

	/* reset the current entry. */
	buf_flush_stat_cur.redo = lsn;
	buf_flush_stat_cur.n_flushed = buf_lru_flush_page_count;
}

/*********************************************************************
Determines the fraction of dirty pages that need to be flushed based
on the speed at which we generate redo log. Note that if redo log
is generated at a significant rate without corresponding increase
in the number of dirty pages (for example, an in-memory workload)
it can cause IO bursts of flushing. This function implements heuristics
to avoid this burstiness.
@return	number of dirty pages to be flushed / second */
UNIV_INTERN
ulint
buf_flush_get_desired_flush_rate(void)
/*==================================*/
{
	ulint			redo_avg;
	ulint			lru_flush_avg;
	ulint			n_dirty;
	ulint			n_flush_req;
	lint			rate;
	ib_uint64_t		lsn = log_get_lsn();
	ulint			log_capacity = log_get_capacity();

	/* log_capacity should never be zero after the initialization
	of log subsystem. */
	ut_ad(log_capacity != 0);

	/* Get total number of dirty pages. It is OK to access
	flush_list without holding any mtex as we are using this
	only for heuristics. */
	n_dirty = UT_LIST_GET_LEN(buf_pool->flush_list);

	/* An overflow can happen if we generate more than 2^32 bytes
	of redo in this interval i.e.: 4G of redo in 1 second. We can
	safely consider this as infinity because if we ever come close
	to 4G we'll start a synchronous flush of dirty pages. */
	/* redo_avg below is average at which redo is generated in
	past BUF_FLUSH_STAT_N_INTERVAL + redo generated in the current
	interval. */
	redo_avg = (ulint) (buf_flush_stat_sum.redo
			    / BUF_FLUSH_STAT_N_INTERVAL
			    + (lsn - buf_flush_stat_cur.redo));

	/* An overflow can happen possibly if we flush more than 2^32
	pages in BUF_FLUSH_STAT_N_INTERVAL. This is a very very
	unlikely scenario. Even when this happens it means that our
	flush rate will be off the mark. It won't affect correctness
	of any subsystem. */
	/* lru_flush_avg below is rate at which pages are flushed as
	part of LRU flush in past BUF_FLUSH_STAT_N_INTERVAL + the
	number of pages flushed in the current interval. */
	lru_flush_avg = buf_flush_stat_sum.n_flushed
			/ BUF_FLUSH_STAT_N_INTERVAL
			+ (buf_lru_flush_page_count
			   - buf_flush_stat_cur.n_flushed);

	n_flush_req = (n_dirty * redo_avg) / log_capacity;

	/* The number of pages that we want to flush from the flush
	list is the difference between the required rate and the
	number of pages that we are historically flushing from the
	LRU list */
	rate = n_flush_req - lru_flush_avg;
	return(rate > 0 ? (ulint) rate : 0);
}

1461
#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1462 1463 1464
/******************************************************************//**
Validates the flush list.
@return	TRUE if ok */
1465 1466 1467 1468 1469 1470 1471
static
ibool
buf_flush_validate_low(void)
/*========================*/
{
	buf_page_t*	bpage;

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1472 1473
	UT_LIST_VALIDATE(flush_list, buf_page_t, buf_pool->flush_list,
			 ut_ad(ut_list_node_313->in_flush_list));
1474 1475 1476 1477 1478 1479

	bpage = UT_LIST_GET_FIRST(buf_pool->flush_list);

	while (bpage != NULL) {
		const ib_uint64_t om = bpage->oldest_modification;
		ut_ad(bpage->in_flush_list);
Vadim Tkachenko's avatar
Vadim Tkachenko committed
1480
		//ut_a(buf_page_in_file(bpage)); /* optimistic */
1481 1482
		ut_a(om > 0);

Vadim Tkachenko's avatar
Vadim Tkachenko committed
1483
		bpage = UT_LIST_GET_NEXT(flush_list, bpage);
1484 1485 1486 1487 1488 1489 1490

		ut_a(!bpage || om >= bpage->oldest_modification);
	}

	return(TRUE);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1491 1492 1493
/******************************************************************//**
Validates the flush list.
@return	TRUE if ok */
1494 1495 1496 1497 1498 1499 1500
UNIV_INTERN
ibool
buf_flush_validate(void)
/*====================*/
{
	ibool	ret;

Vadim Tkachenko's avatar
Vadim Tkachenko committed
1501 1502
	//buf_pool_mutex_enter();
	mutex_enter(&flush_list_mutex);
1503 1504 1505

	ret = buf_flush_validate_low();

Vadim Tkachenko's avatar
Vadim Tkachenko committed
1506 1507
	//buf_pool_mutex_exit();
	mutex_exit(&flush_list_mutex);
1508 1509 1510 1511

	return(ret);
}
#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1512
#endif /* !UNIV_HOTBACKUP */