os0file.c 107 KB
Newer Older
1 2 3 4 5 6 7 8 9 10
/******************************************************
The interface to the operating system file i/o primitives

(c) 1995 Innobase Oy

Created 10/21/1995 Heikki Tuuri
*******************************************************/

#include "os0file.h"
#include "os0sync.h"
unknown's avatar
unknown committed
11
#include "os0thread.h"
12
#include "ut0mem.h"
13
#include "srv0srv.h"
unknown's avatar
unknown committed
14
#include "srv0start.h"
15
#include "fil0fil.h"
unknown's avatar
unknown committed
16
#include "buf0buf.h"
17

unknown's avatar
unknown committed
18 19 20 21 22 23 24
#if defined(UNIV_HOTBACKUP) && defined(__WIN__)
/* Add includes for the _stat() call to compile on Windows */
#include <sys/types.h>
#include <sys/stat.h>
#include <errno.h>
#endif /* UNIV_HOTBACKUP */

25 26 27 28 29 30
#ifdef POSIX_ASYNC_IO
/* We assume in this case that the OS has standard Posix aio (at least SunOS
2.6, HP-UX 11i and AIX 4.3 have) */

#endif

unknown's avatar
unknown committed
31
/* This specifies the file permissions InnoDB uses when it creates files in
unknown's avatar
unknown committed
32 33 34 35 36 37 38 39 40
Unix; the value of os_innodb_umask is initialized in ha_innodb.cc to
my_umask */

#ifndef __WIN__
ulint	os_innodb_umask		= S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP;
#else
ulint	os_innodb_umask		= 0;
#endif

41
#ifdef UNIV_DO_FLUSH
unknown's avatar
unknown committed
42
/* If the following is set to TRUE, we do not call os_file_flush in every
unknown's avatar
unknown committed
43
os_file_write. We can set this TRUE when the doublewrite buffer is used. */
unknown's avatar
unknown committed
44
ibool	os_do_not_call_flush_at_each_write	= FALSE;
45 46 47
#else
/* We do not call os_file_flush in every os_file_write. */
#endif /* UNIV_DO_FLUSH */
unknown's avatar
unknown committed
48

49 50 51 52 53 54
/* We use these mutexes to protect lseek + file i/o operation, if the
OS does not provide an atomic pread or pwrite, or similar */
#define OS_FILE_N_SEEK_MUTEXES	16
os_mutex_t	os_file_seek_mutexes[OS_FILE_N_SEEK_MUTEXES];

/* In simulated aio, merge at most this many consecutive i/os */
unknown's avatar
unknown committed
55
#define OS_AIO_MERGE_N_CONSECUTIVE	64
56 57 58 59 60 61 62

/* If this flag is TRUE, then we will use the native aio of the
OS (provided we compiled Innobase with it in), otherwise we will
use simulated aio we build below with threads */

ibool	os_aio_use_native_aio	= FALSE;

unknown's avatar
unknown committed
63 64
ibool	os_aio_print_debug	= FALSE;

65 66 67 68 69 70 71 72
/* The aio array slot structure */
typedef struct os_aio_slot_struct	os_aio_slot_t;

struct os_aio_slot_struct{
	ibool		is_read;	/* TRUE if a read operation */
	ulint		pos;		/* index of the slot in the aio
					array */
	ibool		reserved;	/* TRUE if this slot is reserved */
unknown's avatar
unknown committed
73
	time_t		reservation_time;/* time when reserved */
74 75 76 77 78 79 80 81
	ulint		len;		/* length of the block to read or
					write */
	byte*		buf;		/* buffer used in i/o */
	ulint		type;		/* OS_FILE_READ or OS_FILE_WRITE */
	ulint		offset;		/* 32 low bits of file offset in
					bytes */
	ulint		offset_high;	/* 32 high bits of file offset */
	os_file_t	file;		/* file where to read or write */
82
	const char*	name;		/* file name or path */
83 84 85 86 87
	ibool		io_already_done;/* used only in simulated aio:
					TRUE if the physical i/o already
					made and only the slot message
					needs to be passed to the caller
					of os_aio_simulated_handle */
unknown's avatar
unknown committed
88
	fil_node_t*	message1;	/* message which is given by the */
89 90 91 92 93
	void*		message2;	/* the requester of an aio operation
					and which can be used to identify
					which pending aio operation was
					completed */
#ifdef WIN_ASYNC_IO
94
	os_event_t	event;		/* event object we need in the
95
					OVERLAPPED struct */
96 97 98 99 100 101 102 103 104 105 106 107 108
	OVERLAPPED	control;	/* Windows control block for the
					aio request */
#elif defined(POSIX_ASYNC_IO)
	struct aiocb	control;	/* Posix control block for aio
					request */
#endif
};

/* The aio array structure */
typedef struct os_aio_array_struct	os_aio_array_t;

struct os_aio_array_struct{
	os_mutex_t	mutex;	  /* the mutex protecting the aio array */
109
	os_event_t	not_full; /* The event which is set to the signaled
110 111
				  state when there is space in the aio
				  outside the ibuf segment */
112 113 114
	os_event_t	is_empty; /* The event which is set to the signaled
				  state when there are no pending i/os
				  in this array */
115 116 117 118 119 120 121
	ulint		n_slots;  /* Total number of slots in the aio array.
				  This must be divisible by n_threads. */
	ulint		n_segments;/* Number of segments in the aio array of
				  pending aio requests. A thread can wait
				  separately for any one of the segments. */
	ulint		n_reserved;/* Number of reserved slots in the
				  aio array outside the ibuf segment */
122
	os_aio_slot_t*	slots;	  /* Pointer to the slots in the array */
123
#ifdef __WIN__
124
	os_native_event_t* native_events;
125 126 127 128
				  /* Pointer to an array of OS native event
				  handles where we copied the handles from
				  slots, in the same order. This can be used
				  in WaitForMultipleObjects; used only in
129
				  Windows */
130
#endif
131 132 133 134 135 136 137
};

/* Array of events used in simulated aio */
os_event_t*	os_aio_segment_wait_events	= NULL;

/* The aio arrays for non-ibuf i/o and ibuf i/o, as well as sync aio. These
are NULL when the module has not yet been initialized. */
unknown's avatar
unknown committed
138 139 140 141 142
static os_aio_array_t*	os_aio_read_array	= NULL;
static os_aio_array_t*	os_aio_write_array	= NULL;
static os_aio_array_t*	os_aio_ibuf_array	= NULL;
static os_aio_array_t*	os_aio_log_array	= NULL;
static os_aio_array_t*	os_aio_sync_array	= NULL;
143

unknown's avatar
unknown committed
144
static ulint	os_aio_n_segments	= ULINT_UNDEFINED;
145

unknown's avatar
unknown committed
146 147
/* If the following is TRUE, read i/o handler threads try to
wait until a batch of new read requests have been posted */
unknown's avatar
unknown committed
148
static ibool	os_aio_recommend_sleep_for_read_threads	= FALSE;
unknown's avatar
unknown committed
149

150
ulint	os_n_file_reads		= 0;
unknown's avatar
unknown committed
151
ulint	os_bytes_read_since_printout = 0;
152 153 154 155 156 157 158
ulint	os_n_file_writes	= 0;
ulint	os_n_fsyncs		= 0;
ulint	os_n_file_reads_old	= 0;
ulint	os_n_file_writes_old	= 0;
ulint	os_n_fsyncs_old		= 0;
time_t	os_last_printout;

unknown's avatar
unknown committed
159 160
ibool	os_has_said_disk_full	= FALSE;

161
/* The mutex protecting the following counts of pending I/O operations */
unknown's avatar
unknown committed
162
static os_mutex_t os_file_count_mutex;
unknown's avatar
unknown committed
163 164
ulint	os_file_n_pending_preads  = 0;
ulint	os_file_n_pending_pwrites = 0;
165 166
ulint	os_n_pending_writes = 0;
ulint	os_n_pending_reads = 0;
167

168 169 170 171 172 173
/***************************************************************************
Gets the operating system version. Currently works only on Windows. */

ulint
os_get_os_version(void)
/*===================*/
174
		  /* out: OS_WIN95, OS_WIN31, OS_WINNT, OS_WIN2000 */
175 176
{
#ifdef __WIN__
177
	OSVERSIONINFO	  os_info;
178

179
	os_info.dwOSVersionInfoSize = sizeof(OSVERSIONINFO);
180 181 182

	ut_a(GetVersionEx(&os_info));

183 184 185 186 187
	if (os_info.dwPlatformId == VER_PLATFORM_WIN32s) {
		return(OS_WIN31);
	} else if (os_info.dwPlatformId == VER_PLATFORM_WIN32_WINDOWS) {
		return(OS_WIN95);
	} else if (os_info.dwPlatformId == VER_PLATFORM_WIN32_NT) {
unknown's avatar
unknown committed
188
		if (os_info.dwMajorVersion <= 4) {
189 190
			return(OS_WINNT);
		} else {
unknown's avatar
unknown committed
191
			return(OS_WIN2000);
192 193 194 195 196
		}
	} else {
		ut_error;
		return(0);
	}
197
#else
198
	ut_error;
199

200
	return(0);
201 202 203
#endif
}

204 205 206 207 208 209 210
/***************************************************************************
Retrieves the last error number if an error occurs in a file io function.
The number should be retrieved before any other OS calls (because they may
overwrite the error number). If the number is not known to this program,
the OS error number + 100 is returned. */

ulint
unknown's avatar
unknown committed
211 212 213 214 215 216
os_file_get_last_error(
/*===================*/
					/* out: error number, or OS error
					number + 100 */
	ibool	report_all_errors)	/* in: TRUE if we want an error message
					printed of all errors */
217 218 219 220 221 222 223
{
	ulint	err;

#ifdef __WIN__

	err = (ulint) GetLastError();

unknown's avatar
unknown committed
224
	if (report_all_errors
unknown's avatar
unknown committed
225
	    || (err != ERROR_DISK_FULL && err != ERROR_FILE_EXISTS)) {
unknown's avatar
unknown committed
226

unknown's avatar
unknown committed
227
		ut_print_timestamp(stderr);
228
		fprintf(stderr,
unknown's avatar
unknown committed
229 230
			"  InnoDB: Operating system error number %lu"
			" in a file operation.\n", (ulong) err);
unknown's avatar
unknown committed
231

unknown's avatar
unknown committed
232
		if (err == ERROR_PATH_NOT_FOUND) {
unknown's avatar
unknown committed
233
			fprintf(stderr,
unknown's avatar
unknown committed
234 235
				"InnoDB: The error means the system"
				" cannot find the path specified.\n");
unknown's avatar
unknown committed
236 237 238

			if (srv_is_being_started) {
				fprintf(stderr,
unknown's avatar
unknown committed
239 240 241 242
					"InnoDB: If you are installing InnoDB,"
					" remember that you must create\n"
					"InnoDB: directories yourself, InnoDB"
					" does not create them.\n");
unknown's avatar
unknown committed
243
			}
unknown's avatar
unknown committed
244
		} else if (err == ERROR_ACCESS_DENIED) {
unknown's avatar
unknown committed
245
			fprintf(stderr,
unknown's avatar
unknown committed
246 247 248 249 250
				"InnoDB: The error means mysqld does not have"
				" the access rights to\n"
				"InnoDB: the directory. It may also be"
				" you have created a subdirectory\n"
				"InnoDB: of the same name as a data file.\n");
251 252 253 254 255 256 257 258 259
		} else if (err == ERROR_SHARING_VIOLATION
			   || err == ERROR_LOCK_VIOLATION) {
			fprintf(stderr,
				"InnoDB: The error means that another program"
				" is using InnoDB's files.\n"
				"InnoDB: This might be a backup or antivirus"
				" software or another instance\n"
				"InnoDB: of MySQL."
				" Please close it to get rid of this error.\n");
260 261 262 263 264 265 266
		} else if (err == ERROR_OPERATION_ABORTED) {
			fprintf(stderr,
				"InnoDB: The error means that the I/O"
				" operation has been aborted\n"
				"InnoDB: because of either a thread exit"
				" or an application request.\n"
				"InnoDB: Retry attempt is made.\n");
unknown's avatar
unknown committed
267
		} else {
unknown's avatar
unknown committed
268
			fprintf(stderr,
unknown's avatar
unknown committed
269 270 271 272 273
				"InnoDB: Some operating system error numbers"
				" are described at\n"
				"InnoDB: "
				"http://dev.mysql.com/doc/refman/5.1/en/"
				"operating-system-error-codes.html\n");
unknown's avatar
unknown committed
274
		}
275
	}
276

unknown's avatar
unknown committed
277 278
	fflush(stderr);

279 280 281 282 283 284
	if (err == ERROR_FILE_NOT_FOUND) {
		return(OS_FILE_NOT_FOUND);
	} else if (err == ERROR_DISK_FULL) {
		return(OS_FILE_DISK_FULL);
	} else if (err == ERROR_FILE_EXISTS) {
		return(OS_FILE_ALREADY_EXISTS);
285 286 287
	} else if (err == ERROR_SHARING_VIOLATION
		   || err == ERROR_LOCK_VIOLATION) {
		return(OS_FILE_SHARING_VIOLATION);
288 289
	} else if (err == ERROR_OPERATION_ABORTED) {
		return(OS_FILE_OPERATION_ABORTED);
290 291 292 293
	} else {
		return(100 + err);
	}
#else
294 295
	err = (ulint) errno;

unknown's avatar
unknown committed
296
	if (report_all_errors
unknown's avatar
unknown committed
297
	    || (err != ENOSPC && err != EEXIST)) {
unknown's avatar
unknown committed
298

unknown's avatar
unknown committed
299
		ut_print_timestamp(stderr);
300
		fprintf(stderr,
unknown's avatar
unknown committed
301 302
			"  InnoDB: Operating system error number %lu"
			" in a file operation.\n", (ulong) err);
unknown's avatar
unknown committed
303

unknown's avatar
unknown committed
304
		if (err == ENOENT) {
unknown's avatar
unknown committed
305
			fprintf(stderr,
unknown's avatar
unknown committed
306 307
				"InnoDB: The error means the system"
				" cannot find the path specified.\n");
308

unknown's avatar
unknown committed
309 310
			if (srv_is_being_started) {
				fprintf(stderr,
unknown's avatar
unknown committed
311 312 313 314
					"InnoDB: If you are installing InnoDB,"
					" remember that you must create\n"
					"InnoDB: directories yourself, InnoDB"
					" does not create them.\n");
unknown's avatar
unknown committed
315
			}
unknown's avatar
unknown committed
316
		} else if (err == EACCES) {
unknown's avatar
unknown committed
317
			fprintf(stderr,
unknown's avatar
unknown committed
318 319 320
				"InnoDB: The error means mysqld does not have"
				" the access rights to\n"
				"InnoDB: the directory.\n");
unknown's avatar
unknown committed
321
		} else {
unknown's avatar
unknown committed
322
			if (strerror((int)err) != NULL) {
unknown's avatar
unknown committed
323
				fprintf(stderr,
unknown's avatar
unknown committed
324 325 326
					"InnoDB: Error number %lu"
					" means '%s'.\n",
					err, strerror((int)err));
unknown's avatar
unknown committed
327
			}
unknown's avatar
unknown committed
328

unknown's avatar
unknown committed
329
			fprintf(stderr,
unknown's avatar
unknown committed
330 331 332 333 334
				"InnoDB: Some operating system"
				" error numbers are described at\n"
				"InnoDB: "
				"http://dev.mysql.com/doc/refman/5.1/en/"
				"operating-system-error-codes.html\n");
unknown's avatar
unknown committed
335
		}
336
	}
337

unknown's avatar
unknown committed
338 339
	fflush(stderr);

unknown's avatar
unknown committed
340
	if (err == ENOSPC) {
341 342 343 344 345 346 347 348 349
		return(OS_FILE_DISK_FULL);
#ifdef POSIX_ASYNC_IO
	} else if (err == EAGAIN) {
		return(OS_FILE_AIO_RESOURCES_RESERVED);
#endif
	} else if (err == ENOENT) {
		return(OS_FILE_NOT_FOUND);
	} else if (err == EEXIST) {
		return(OS_FILE_ALREADY_EXISTS);
350 351
	} else if (err == EXDEV || err == ENOTDIR || err == EISDIR) {
		return(OS_FILE_PATH_ERROR);
352 353 354 355 356 357 358
	} else {
		return(100 + err);
	}
#endif
}

/********************************************************************
unknown's avatar
unknown committed
359 360 361 362
Does error handling when a file operation fails.
Conditionally exits (calling exit(3)) based on should_exit value and the
error type */

363 364
static
ibool
unknown's avatar
unknown committed
365 366 367 368 369 370 371 372
os_file_handle_error_cond_exit(
/*===========================*/
					/* out: TRUE if we should retry the
					operation */
	const char*	name,		/* in: name of a file or NULL */
	const char*	operation,	/* in: operation */
	ibool		should_exit)	/* in: call exit(3) if unknown error
					and this parameter is TRUE */
373 374 375
{
	ulint	err;

unknown's avatar
unknown committed
376
	err = os_file_get_last_error(FALSE);
377

378
	if (err == OS_FILE_DISK_FULL) {
unknown's avatar
unknown committed
379 380 381 382 383 384
		/* We only print a warning about disk full once */

		if (os_has_said_disk_full) {

			return(FALSE);
		}
385

386
		if (name) {
unknown's avatar
unknown committed
387 388
			ut_print_timestamp(stderr);
			fprintf(stderr,
unknown's avatar
unknown committed
389 390
				"  InnoDB: Encountered a problem with"
				" file %s\n", name);
391
		}
unknown's avatar
unknown committed
392 393

		ut_print_timestamp(stderr);
394
		fprintf(stderr,
unknown's avatar
unknown committed
395 396
			"  InnoDB: Disk is full. Try to clean the disk"
			" to free space.\n");
397

unknown's avatar
unknown committed
398 399
		os_has_said_disk_full = TRUE;

unknown's avatar
unknown committed
400 401
		fflush(stderr);

unknown's avatar
unknown committed
402
		return(FALSE);
403
	} else if (err == OS_FILE_AIO_RESOURCES_RESERVED) {
unknown's avatar
unknown committed
404

405
		return(TRUE);
406
	} else if (err == OS_FILE_ALREADY_EXISTS
unknown's avatar
unknown committed
407
		   || err == OS_FILE_PATH_ERROR) {
unknown's avatar
unknown committed
408

409
		return(FALSE);
410 411 412 413
	} else if (err == OS_FILE_SHARING_VIOLATION) {

		os_thread_sleep(10000000);  /* 10 sec */
		return(TRUE);
414 415 416 417
	} else if (err == OS_FILE_OPERATION_ABORTED) {

		os_thread_sleep(100000);	/* 100 ms */
		return(TRUE);
418
	} else {
419 420 421 422
		if (name) {
			fprintf(stderr, "InnoDB: File name %s\n", name);
		}

423
		fprintf(stderr, "InnoDB: File operation call: '%s'.\n",
424
			operation);
425

unknown's avatar
unknown committed
426 427 428 429
		if (should_exit) {
			fprintf(stderr, "InnoDB: Cannot continue operation.\n");

			fflush(stderr);
unknown's avatar
unknown committed
430

unknown's avatar
unknown committed
431 432
			exit(1);
		}
433 434
	}

435
	return(FALSE);
436 437
}

unknown's avatar
unknown committed
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
/********************************************************************
Does error handling when a file operation fails. */
static
ibool
os_file_handle_error(
/*=================*/
				/* out: TRUE if we should retry the
				operation */
	const char*	name,	/* in: name of a file or NULL */
	const char*	operation)/* in: operation */
{
	/* exit in case of unknown error */
	return(os_file_handle_error_cond_exit(name, operation, TRUE));
}

/********************************************************************
Does error handling when a file operation fails. */
static
ibool
os_file_handle_error_no_exit(
/*=========================*/
				/* out: TRUE if we should retry the
				operation */
	const char*	name,	/* in: name of a file or NULL */
	const char*	operation)/* in: operation */
{
	/* don't exit in case of unknown error */
	return(os_file_handle_error_cond_exit(name, operation, FALSE));
}

468 469
#undef USE_FILE_LOCK
#define USE_FILE_LOCK
unknown's avatar
unknown committed
470
#if defined(UNIV_HOTBACKUP) || defined(__WIN__) || defined(__NETWARE__)
471 472 473 474 475 476
/* InnoDB Hot Backup does not lock the data files.
 * On Windows, mandatory locking is used.
 */
# undef USE_FILE_LOCK
#endif
#ifdef USE_FILE_LOCK
477 478 479 480 481 482 483 484
/********************************************************************
Obtain an exclusive lock on a file. */
static
int
os_file_lock(
/*=========*/
				/* out: 0 on success */
	int		fd,	/* in: file descriptor */
unknown's avatar
unknown committed
485
	const char*	name)	/* in: file name */
486 487
{
	struct flock lk;
unknown's avatar
unknown committed
488
	lk.l_type = F_WRLCK;
489 490 491 492
	lk.l_whence = SEEK_SET;
	lk.l_start = lk.l_len = 0;
	if (fcntl(fd, F_SETLK, &lk) == -1) {
		fprintf(stderr,
unknown's avatar
unknown committed
493 494 495 496
			"InnoDB: Unable to lock %s, error: %d\n", name, errno);

		if (errno == EAGAIN || errno == EACCES) {
			fprintf(stderr,
unknown's avatar
unknown committed
497 498 499 500
				"InnoDB: Check that you do not already have"
				" another mysqld process\n"
				"InnoDB: using the same InnoDB data"
				" or log files.\n");
unknown's avatar
unknown committed
501 502
		}

503 504
		return(-1);
	}
unknown's avatar
unknown committed
505

unknown's avatar
unknown committed
506
	return(0);
507
}
508
#endif /* USE_FILE_LOCK */
509

unknown's avatar
unknown committed
510 511 512 513 514 515 516 517 518
/********************************************************************
Creates the seek mutexes used in positioned reads and writes. */

void
os_io_init_simple(void)
/*===================*/
{
	ulint	i;

unknown's avatar
unknown committed
519 520
	os_file_count_mutex = os_mutex_create(NULL);

unknown's avatar
unknown committed
521 522 523 524 525
	for (i = 0; i < OS_FILE_N_SEEK_MUTEXES; i++) {
		os_file_seek_mutexes[i] = os_mutex_create(NULL);
	}
}

526
#if !defined(UNIV_HOTBACKUP) && !defined(__NETWARE__)
527
/*************************************************************************
528 529
Creates a temporary file that will be deleted on close.
This function is defined in ha_innodb.cc. */
530 531 532 533 534

int
innobase_mysql_tmpfile(void);
/*========================*/
			/* out: temporary file descriptor, or < 0 on error */
535
#endif /* !UNIV_HOTBACKUP && !__NETWARE__ */
536

537
/***************************************************************************
538 539 540 541
Creates a temporary file.  This function is like tmpfile(3), but
the temporary file is created in the MySQL temporary directory.
On Netware, this function is like tmpfile(3), because the C run-time
library of Netware does not expose the delete-on-close flag. */
542 543 544 545

FILE*
os_file_create_tmpfile(void)
/*========================*/
546
			/* out: temporary file handle, or NULL on error */
547
{
548 549 550 551 552 553
#ifdef UNIV_HOTBACKUP
	ut_error;

	return(NULL);
#else
# ifdef __NETWARE__
554
	FILE*	file	= tmpfile();
555
# else /* __NETWARE__ */
556
	FILE*	file	= NULL;
557
	int	fd	= innobase_mysql_tmpfile();
558 559 560

	if (fd >= 0) {
		file = fdopen(fd, "w+b");
561
	}
562
# endif /* __NETWARE__ */
563 564

	if (!file) {
565
		ut_print_timestamp(stderr);
unknown's avatar
unknown committed
566 567 568
		fprintf(stderr,
			"  InnoDB: Error: unable to create temporary file;"
			" errno: %d\n", errno);
569
# ifndef __NETWARE__
570 571 572
		if (fd >= 0) {
			close(fd);
		}
573
# endif /* !__NETWARE__ */
574
	}
575

576
	return(file);
577
#endif /* UNIV_HOTBACKUP */
578 579
}

unknown's avatar
unknown committed
580 581 582 583 584 585 586 587 588
/***************************************************************************
The os_file_opendir() function opens a directory stream corresponding to the
directory named by the dirname argument. The directory stream is positioned
at the first entry. In both Unix and Windows we automatically skip the '.'
and '..' items at the start of the directory listing. */

os_file_dir_t
os_file_opendir(
/*============*/
589 590 591 592 593 594 595 596 597
					/* out: directory stream, NULL if
					error */
	const char*	dirname,	/* in: directory name; it must not
					contain a trailing '\' or '/' */
	ibool		error_is_fatal)	/* in: TRUE if we should treat an
					error as a fatal error; if we try to
					open symlinks then we do not wish a
					fatal error if it happens not to be
					a directory */
unknown's avatar
unknown committed
598 599 600
{
	os_file_dir_t		dir;
#ifdef __WIN__
601
	LPWIN32_FIND_DATA	lpFindFileData;
unknown's avatar
unknown committed
602 603 604 605 606
	char			path[OS_FILE_MAX_PATH + 3];

	ut_a(strlen(dirname) < OS_FILE_MAX_PATH);

	strcpy(path, dirname);
unknown's avatar
unknown committed
607
	strcpy(path + strlen(path), "\\*");
unknown's avatar
unknown committed
608 609 610 611 612 613 614

	/* Note that in Windows opening the 'directory stream' also retrieves
	the first entry in the directory. Since it is '.', that is no problem,
	as we will skip over the '.' and '..' entries anyway. */

	lpFindFileData = ut_malloc(sizeof(WIN32_FIND_DATA));

unknown's avatar
unknown committed
615
	dir = FindFirstFile((LPCTSTR) path, lpFindFileData);
unknown's avatar
unknown committed
616 617 618 619 620 621

	ut_free(lpFindFileData);

	if (dir == INVALID_HANDLE_VALUE) {

		if (error_is_fatal) {
622
			os_file_handle_error(dirname, "opendir");
unknown's avatar
unknown committed
623 624 625 626 627
		}

		return(NULL);
	}

628
	return(dir);
unknown's avatar
unknown committed
629 630 631 632
#else
	dir = opendir(dirname);

	if (dir == NULL && error_is_fatal) {
633
		os_file_handle_error(dirname, "opendir");
unknown's avatar
unknown committed
634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654
	}

	return(dir);
#endif
}

/***************************************************************************
Closes a directory stream. */

int
os_file_closedir(
/*=============*/
				/* out: 0 if success, -1 if failure */
	os_file_dir_t	dir)	/* in: directory stream */
{
#ifdef __WIN__
	BOOL		ret;

	ret = FindClose(dir);

	if (!ret) {
655 656
		os_file_handle_error_no_exit(NULL, "closedir");

unknown's avatar
unknown committed
657 658
		return(-1);
	}
659

unknown's avatar
unknown committed
660 661 662
	return(0);
#else
	int	ret;
663

unknown's avatar
unknown committed
664 665 666
	ret = closedir(dir);

	if (ret) {
667
		os_file_handle_error_no_exit(NULL, "closedir");
unknown's avatar
unknown committed
668 669 670 671 672 673 674 675 676 677 678 679 680 681 682
	}

	return(ret);
#endif
}

/***************************************************************************
This function returns information of the next file in the directory. We jump
over the '.' and '..' entries in the directory. */

int
os_file_readdir_next_file(
/*======================*/
				/* out: 0 if ok, -1 if error, 1 if at the end
				of the directory */
683
	const char*	dirname,/* in: directory name or path */
unknown's avatar
unknown committed
684 685 686 687 688 689 690 691 692 693 694 695
	os_file_dir_t	dir,	/* in: directory stream */
	os_file_stat_t*	info)	/* in/out: buffer where the info is returned */
{
#ifdef __WIN__
	LPWIN32_FIND_DATA	lpFindFileData;
	BOOL			ret;

	lpFindFileData = ut_malloc(sizeof(WIN32_FIND_DATA));
next_file:
	ret = FindNextFile(dir, lpFindFileData);

	if (ret) {
unknown's avatar
unknown committed
696 697
		ut_a(strlen((char *) lpFindFileData->cFileName)
		     < OS_FILE_MAX_PATH);
unknown's avatar
unknown committed
698

unknown's avatar
unknown committed
699
		if (strcmp((char *) lpFindFileData->cFileName, ".") == 0
unknown's avatar
unknown committed
700
		    || strcmp((char *) lpFindFileData->cFileName, "..") == 0) {
unknown's avatar
unknown committed
701

702
			goto next_file;
unknown's avatar
unknown committed
703 704
		}

unknown's avatar
unknown committed
705
		strcpy(info->name, (char *) lpFindFileData->cFileName);
unknown's avatar
unknown committed
706

unknown's avatar
unknown committed
707
		info->size = (ib_longlong)(lpFindFileData->nFileSizeLow)
unknown's avatar
unknown committed
708 709
			+ (((ib_longlong)(lpFindFileData->nFileSizeHigh))
			   << 32);
unknown's avatar
unknown committed
710 711

		if (lpFindFileData->dwFileAttributes
unknown's avatar
unknown committed
712 713 714 715 716 717 718
		    & FILE_ATTRIBUTE_REPARSE_POINT) {
			/* TODO: test Windows symlinks */
			/* TODO: MySQL has apparently its own symlink
			implementation in Windows, dbname.sym can
			redirect a database directory:
			http://dev.mysql.com/doc/refman/5.1/en/
			windows-symbolic-links.html */
unknown's avatar
unknown committed
719 720
			info->type = OS_FILE_TYPE_LINK;
		} else if (lpFindFileData->dwFileAttributes
unknown's avatar
unknown committed
721
			   & FILE_ATTRIBUTE_DIRECTORY) {
722
			info->type = OS_FILE_TYPE_DIR;
unknown's avatar
unknown committed
723
		} else {
unknown's avatar
unknown committed
724 725 726 727 728
			/* It is probably safest to assume that all other
			file types are normal. Better to check them rather
			than blindly skip them. */

			info->type = OS_FILE_TYPE_FILE;
unknown's avatar
unknown committed
729 730 731 732 733 734 735 736 737 738 739
		}
	}

	ut_free(lpFindFileData);

	if (ret) {
		return(0);
	} else if (GetLastError() == ERROR_NO_MORE_FILES) {

		return(1);
	} else {
unknown's avatar
unknown committed
740
		os_file_handle_error_no_exit(dirname,
unknown's avatar
unknown committed
741
					     "readdir_next_file");
unknown's avatar
unknown committed
742 743 744 745 746 747 748
		return(-1);
	}
#else
	struct dirent*	ent;
	char*		full_path;
	int		ret;
	struct stat	statinfo;
unknown's avatar
unknown committed
749
#ifdef HAVE_READDIR_R
unknown's avatar
unknown committed
750 751 752 753 754
	char		dirent_buf[sizeof(struct dirent)
				   + _POSIX_PATH_MAX + 100];
	/* In /mysys/my_lib.c, _POSIX_PATH_MAX + 1 is used as
	the max file name len; but in most standards, the
	length is NAME_MAX; we add 100 to be even safer */
unknown's avatar
unknown committed
755 756
#endif

unknown's avatar
unknown committed
757
next_file:
unknown's avatar
unknown committed
758 759 760 761

#ifdef HAVE_READDIR_R
	ret = readdir_r(dir, (struct dirent*)dirent_buf, &ent);

762 763 764 765 766 767 768 769 770
	if (ret != 0
#ifdef UNIV_AIX
	    /* On AIX, only if we got non-NULL 'ent' (result) value and
	    a non-zero 'ret' (return) value, it indicates a failed
	    readdir_r() call. An NULL 'ent' with an non-zero 'ret'
	    would indicate the "end of the directory" is reached. */
	    && ent != NULL
#endif
	   ) {
unknown's avatar
unknown committed
771
		fprintf(stderr,
unknown's avatar
unknown committed
772 773
			"InnoDB: cannot read directory %s, error %lu\n",
			dirname, (ulong)ret);
unknown's avatar
unknown committed
774 775 776

		return(-1);
	}
unknown's avatar
unknown committed
777 778

	if (ent == NULL) {
unknown's avatar
unknown committed
779
		/* End of directory */
780

unknown's avatar
unknown committed
781 782 783
		return(1);
	}

unknown's avatar
unknown committed
784 785 786 787 788 789 790 791 792
	ut_a(strlen(ent->d_name) < _POSIX_PATH_MAX + 100 - 1);
#else
	ent = readdir(dir);

	if (ent == NULL) {

		return(1);
	}
#endif
unknown's avatar
unknown committed
793 794 795 796 797 798 799 800 801 802
	ut_a(strlen(ent->d_name) < OS_FILE_MAX_PATH);

	if (strcmp(ent->d_name, ".") == 0 || strcmp(ent->d_name, "..") == 0) {

		goto next_file;
	}

	strcpy(info->name, ent->d_name);

	full_path = ut_malloc(strlen(dirname) + strlen(ent->d_name) + 10);
803

unknown's avatar
unknown committed
804 805 806 807 808
	sprintf(full_path, "%s/%s", dirname, ent->d_name);

	ret = stat(full_path, &statinfo);

	if (ret) {
unknown's avatar
unknown committed
809
		os_file_handle_error_no_exit(full_path, "stat");
unknown's avatar
unknown committed
810 811 812 813 814 815 816 817 818 819 820

		ut_free(full_path);

		return(-1);
	}

	info->size = (ib_longlong)statinfo.st_size;

	if (S_ISDIR(statinfo.st_mode)) {
		info->type = OS_FILE_TYPE_DIR;
	} else if (S_ISLNK(statinfo.st_mode)) {
821
		info->type = OS_FILE_TYPE_LINK;
unknown's avatar
unknown committed
822
	} else if (S_ISREG(statinfo.st_mode)) {
823
		info->type = OS_FILE_TYPE_FILE;
unknown's avatar
unknown committed
824
	} else {
825
		info->type = OS_FILE_TYPE_UNKNOWN;
unknown's avatar
unknown committed
826
	}
827

unknown's avatar
unknown committed
828 829 830 831 832 833
	ut_free(full_path);

	return(0);
#endif
}

unknown's avatar
unknown committed
834 835 836 837 838 839 840 841 842
/*********************************************************************
This function attempts to create a directory named pathname. The new directory
gets default permissions. On Unix the permissions are (0770 & ~umask). If the
directory exists already, nothing is done and the call succeeds, unless the
fail_if_exists arguments is true. */

ibool
os_file_create_directory(
/*=====================*/
843 844 845 846 847 848
					/* out: TRUE if call succeeds,
					FALSE on error */
	const char*	pathname,	/* in: directory name as
					null-terminated string */
	ibool		fail_if_exists)	/* in: if TRUE, pre-existing directory
					is treated as an error. */
unknown's avatar
unknown committed
849 850 851
{
#ifdef __WIN__
	BOOL	rcode;
852

unknown's avatar
unknown committed
853
	rcode = CreateDirectory((LPCTSTR) pathname, NULL);
unknown's avatar
unknown committed
854 855 856
	if (!(rcode != 0
	      || (GetLastError() == ERROR_ALREADY_EXISTS
		  && !fail_if_exists))) {
unknown's avatar
unknown committed
857
		/* failure */
858
		os_file_handle_error(pathname, "CreateDirectory");
unknown's avatar
unknown committed
859 860 861

		return(FALSE);
	}
862

unknown's avatar
unknown committed
863 864 865 866 867 868 869 870
	return (TRUE);
#else
	int	rcode;

	rcode = mkdir(pathname, 0770);

	if (!(rcode == 0 || (errno == EEXIST && !fail_if_exists))) {
		/* failure */
unknown's avatar
unknown committed
871
		os_file_handle_error(pathname, "mkdir");
unknown's avatar
unknown committed
872 873 874

		return(FALSE);
	}
875

unknown's avatar
unknown committed
876
	return (TRUE);
877
#endif
unknown's avatar
unknown committed
878 879
}

unknown's avatar
unknown committed
880 881 882 883 884 885
/********************************************************************
A simple function to open or create a file. */

os_file_t
os_file_create_simple(
/*==================*/
886 887 888 889 890 891 892 893 894
				/* out, own: handle to the file, not defined
				if error, error number can be retrieved with
				os_file_get_last_error */
	const char*	name,	/* in: name of the file or path as a
				null-terminated string */
	ulint		create_mode,/* in: OS_FILE_OPEN if an existing file is
				opened (if does not exist, error), or
				OS_FILE_CREATE if a new file is created
				(if exists, error), or
895
				OS_FILE_CREATE_PATH if new file
896 897 898 899 900
				(if exists, error) and subdirectories along
				its path are created (if needed)*/
	ulint		access_type,/* in: OS_FILE_READ_ONLY or
				OS_FILE_READ_WRITE */
	ibool*		success)/* out: TRUE if succeed, FALSE if error */
unknown's avatar
unknown committed
901 902 903 904 905 906 907
{
#ifdef __WIN__
	os_file_t	file;
	DWORD		create_flag;
	DWORD		access;
	DWORD		attributes	= 0;
	ibool		retry;
908 909

try_again:
unknown's avatar
unknown committed
910 911 912 913 914 915
	ut_a(name);

	if (create_mode == OS_FILE_OPEN) {
		create_flag = OPEN_EXISTING;
	} else if (create_mode == OS_FILE_CREATE) {
		create_flag = CREATE_NEW;
916 917 918 919 920 921 922 923
	} else if (create_mode == OS_FILE_CREATE_PATH) {
		/* create subdirs along the path if needed  */
		*success = os_file_create_subdirs_if_needed(name);
		if (!*success) {
			ut_error;
		}
		create_flag = CREATE_NEW;
		create_mode = OS_FILE_CREATE;
unknown's avatar
unknown committed
924 925 926 927 928 929 930 931 932 933 934 935 936 937
	} else {
		create_flag = 0;
		ut_error;
	}

	if (access_type == OS_FILE_READ_ONLY) {
		access = GENERIC_READ;
	} else if (access_type == OS_FILE_READ_WRITE) {
		access = GENERIC_READ | GENERIC_WRITE;
	} else {
		access = 0;
		ut_error;
	}

unknown's avatar
unknown committed
938
	file = CreateFile((LPCTSTR) name,
unknown's avatar
unknown committed
939 940
			  access,
			  FILE_SHARE_READ | FILE_SHARE_WRITE,
unknown's avatar
unknown committed
941
			  /* file can be read and written also
unknown's avatar
unknown committed
942 943 944 945 946
			  by other processes */
			  NULL,	/* default security attributes */
			  create_flag,
			  attributes,
			  NULL);	/* no template file */
unknown's avatar
unknown committed
947 948 949 950

	if (file == INVALID_HANDLE_VALUE) {
		*success = FALSE;

951
		retry = os_file_handle_error(name,
unknown's avatar
unknown committed
952 953
					     create_mode == OS_FILE_OPEN ?
					     "open" : "create");
unknown's avatar
unknown committed
954 955 956 957 958 959 960 961
		if (retry) {
			goto try_again;
		}
	} else {
		*success = TRUE;
	}

	return(file);
962
#else /* __WIN__ */
unknown's avatar
unknown committed
963 964 965
	os_file_t	file;
	int		create_flag;
	ibool		retry;
966 967

try_again:
unknown's avatar
unknown committed
968 969 970 971 972 973 974 975 976 977
	ut_a(name);

	if (create_mode == OS_FILE_OPEN) {
		if (access_type == OS_FILE_READ_ONLY) {
			create_flag = O_RDONLY;
		} else {
			create_flag = O_RDWR;
		}
	} else if (create_mode == OS_FILE_CREATE) {
		create_flag = O_RDWR | O_CREAT | O_EXCL;
978 979 980 981 982 983 984 985
	} else if (create_mode == OS_FILE_CREATE_PATH) {
		/* create subdirs along the path if needed  */
		*success = os_file_create_subdirs_if_needed(name);
		if (!*success) {
			return (-1);
		}
		create_flag = O_RDWR | O_CREAT | O_EXCL;
		create_mode = OS_FILE_CREATE;
unknown's avatar
unknown committed
986 987 988 989 990 991
	} else {
		create_flag = 0;
		ut_error;
	}

	if (create_mode == OS_FILE_CREATE) {
992
		file = open(name, create_flag, S_IRUSR | S_IWUSR
unknown's avatar
unknown committed
993
			    | S_IRGRP | S_IWGRP);
994 995 996 997
	} else {
		file = open(name, create_flag);
	}

unknown's avatar
unknown committed
998 999 1000
	if (file == -1) {
		*success = FALSE;

1001
		retry = os_file_handle_error(name,
unknown's avatar
unknown committed
1002 1003
					     create_mode == OS_FILE_OPEN ?
					     "open" : "create");
unknown's avatar
unknown committed
1004 1005 1006
		if (retry) {
			goto try_again;
		}
1007
#ifdef USE_FILE_LOCK
unknown's avatar
unknown committed
1008
	} else if (access_type == OS_FILE_READ_WRITE
unknown's avatar
unknown committed
1009
		   && os_file_lock(file, name)) {
1010
		*success = FALSE;
1011
		close(file);
1012 1013
		file = -1;
#endif
unknown's avatar
unknown committed
1014 1015 1016 1017
	} else {
		*success = TRUE;
	}

1018
	return(file);
1019
#endif /* __WIN__ */
unknown's avatar
unknown committed
1020
}
unknown's avatar
unknown committed
1021 1022 1023 1024 1025 1026 1027

/********************************************************************
A simple function to open or create a file. */

os_file_t
os_file_create_simple_no_error_handling(
/*====================================*/
1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041
				/* out, own: handle to the file, not defined
				if error, error number can be retrieved with
				os_file_get_last_error */
	const char*	name,	/* in: name of the file or path as a
				null-terminated string */
	ulint		create_mode,/* in: OS_FILE_OPEN if an existing file
				is opened (if does not exist, error), or
				OS_FILE_CREATE if a new file is created
				(if exists, error) */
	ulint		access_type,/* in: OS_FILE_READ_ONLY,
				OS_FILE_READ_WRITE, or
				OS_FILE_READ_ALLOW_DELETE; the last option is
				used by a backup program reading the file */
	ibool*		success)/* out: TRUE if succeed, FALSE if error */
unknown's avatar
unknown committed
1042 1043 1044 1045 1046 1047
{
#ifdef __WIN__
	os_file_t	file;
	DWORD		create_flag;
	DWORD		access;
	DWORD		attributes	= 0;
unknown's avatar
unknown committed
1048
	DWORD		share_mode	= FILE_SHARE_READ | FILE_SHARE_WRITE;
1049

unknown's avatar
unknown committed
1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064
	ut_a(name);

	if (create_mode == OS_FILE_OPEN) {
		create_flag = OPEN_EXISTING;
	} else if (create_mode == OS_FILE_CREATE) {
		create_flag = CREATE_NEW;
	} else {
		create_flag = 0;
		ut_error;
	}

	if (access_type == OS_FILE_READ_ONLY) {
		access = GENERIC_READ;
	} else if (access_type == OS_FILE_READ_WRITE) {
		access = GENERIC_READ | GENERIC_WRITE;
unknown's avatar
unknown committed
1065 1066 1067
	} else if (access_type == OS_FILE_READ_ALLOW_DELETE) {
		access = GENERIC_READ;
		share_mode = FILE_SHARE_DELETE | FILE_SHARE_READ
unknown's avatar
unknown committed
1068
			| FILE_SHARE_WRITE;	/* A backup program has to give
unknown's avatar
unknown committed
1069 1070 1071
						mysqld the maximum freedom to
						do what it likes with the
						file */
unknown's avatar
unknown committed
1072 1073 1074 1075 1076
	} else {
		access = 0;
		ut_error;
	}

unknown's avatar
unknown committed
1077
	file = CreateFile((LPCTSTR) name,
unknown's avatar
unknown committed
1078 1079 1080 1081 1082 1083
			  access,
			  share_mode,
			  NULL,	/* default security attributes */
			  create_flag,
			  attributes,
			  NULL);	/* no template file */
unknown's avatar
unknown committed
1084 1085 1086 1087 1088 1089 1090 1091

	if (file == INVALID_HANDLE_VALUE) {
		*success = FALSE;
	} else {
		*success = TRUE;
	}

	return(file);
1092
#else /* __WIN__ */
unknown's avatar
unknown committed
1093 1094
	os_file_t	file;
	int		create_flag;
1095

unknown's avatar
unknown committed
1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111
	ut_a(name);

	if (create_mode == OS_FILE_OPEN) {
		if (access_type == OS_FILE_READ_ONLY) {
			create_flag = O_RDONLY;
		} else {
			create_flag = O_RDWR;
		}
	} else if (create_mode == OS_FILE_CREATE) {
		create_flag = O_RDWR | O_CREAT | O_EXCL;
	} else {
		create_flag = 0;
		ut_error;
	}

	if (create_mode == OS_FILE_CREATE) {
1112
		file = open(name, create_flag, S_IRUSR | S_IWUSR
unknown's avatar
unknown committed
1113
			    | S_IRGRP | S_IWGRP);
1114 1115 1116 1117
	} else {
		file = open(name, create_flag);
	}

unknown's avatar
unknown committed
1118 1119
	if (file == -1) {
		*success = FALSE;
1120
#ifdef USE_FILE_LOCK
unknown's avatar
unknown committed
1121
	} else if (access_type == OS_FILE_READ_WRITE
unknown's avatar
unknown committed
1122
		   && os_file_lock(file, name)) {
1123
		*success = FALSE;
1124
		close(file);
1125 1126
		file = -1;
#endif
unknown's avatar
unknown committed
1127 1128 1129 1130
	} else {
		*success = TRUE;
	}

1131
	return(file);
1132
#endif /* __WIN__ */
unknown's avatar
unknown committed
1133 1134
}

unknown's avatar
unknown committed
1135 1136 1137 1138 1139 1140
/********************************************************************
Tries to disable OS caching on an opened file descriptor. */

void
os_file_set_nocache(
/*================*/
1141 1142 1143 1144 1145 1146
	int		fd		/* in: file descriptor to alter */
	__attribute__((unused)),
	const char*	file_name	/* in: used in the diagnostic message */
	__attribute__((unused)),
	const char*	operation_name __attribute__((unused)))
					/* in: used in the diagnostic message,
unknown's avatar
unknown committed
1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182
					we call os_file_set_nocache()
					immediately after opening or creating
					a file, so this is either "open" or
					"create" */
{
	/* some versions of Solaris may not have DIRECTIO_ON */
#if defined(UNIV_SOLARIS) && defined(DIRECTIO_ON)
	if (directio(fd, DIRECTIO_ON) == -1) {
		int	errno_save;
		errno_save = (int)errno;
		ut_print_timestamp(stderr);
		fprintf(stderr,
			"  InnoDB: Failed to set DIRECTIO_ON "
			"on file %s: %s: %s, continuing anyway\n",
			file_name, operation_name, strerror(errno_save));
	}
#elif defined(O_DIRECT)
	if (fcntl(fd, F_SETFL, O_DIRECT) == -1) {
		int	errno_save;
		errno_save = (int)errno;
		ut_print_timestamp(stderr);
		fprintf(stderr,
			"  InnoDB: Failed to set O_DIRECT "
			"on file %s: %s: %s, continuing anyway\n",
			file_name, operation_name, strerror(errno_save));
		if (errno_save == EINVAL) {
			ut_print_timestamp(stderr);
			fprintf(stderr,
				"  InnoDB: O_DIRECT is known to result in "
				"'Invalid argument' on Linux on tmpfs, "
				"see MySQL Bug#26662\n");
		}
	}
#endif
}

1183 1184 1185 1186 1187 1188
/********************************************************************
Opens an existing file or creates a new. */

os_file_t
os_file_create(
/*===========*/
1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210
				/* out, own: handle to the file, not defined
				if error, error number can be retrieved with
				os_file_get_last_error */
	const char*	name,	/* in: name of the file or path as a
				null-terminated string */
	ulint		create_mode,/* in: OS_FILE_OPEN if an existing file
				is opened (if does not exist, error), or
				OS_FILE_CREATE if a new file is created
				(if exists, error),
				OS_FILE_OVERWRITE if a new file is created
				or an old overwritten;
				OS_FILE_OPEN_RAW, if a raw device or disk
				partition should be opened */
	ulint		purpose,/* in: OS_FILE_AIO, if asynchronous,
				non-buffered i/o is desired,
				OS_FILE_NORMAL, if any normal file;
				NOTE that it also depends on type, os_aio_..
				and srv_.. variables whether we really use
				async i/o or unbuffered i/o: look in the
				function source code for the exact rules */
	ulint		type,	/* in: OS_DATA_FILE or OS_LOG_FILE */
	ibool*		success)/* out: TRUE if succeed, FALSE if error */
1211 1212 1213
{
#ifdef __WIN__
	os_file_t	file;
unknown's avatar
unknown committed
1214
	DWORD		share_mode	= FILE_SHARE_READ;
1215 1216 1217
	DWORD		create_flag;
	DWORD		attributes;
	ibool		retry;
1218
try_again:
1219 1220
	ut_a(name);

unknown's avatar
unknown committed
1221 1222 1223
	if (create_mode == OS_FILE_OPEN_RAW) {
		create_flag = OPEN_EXISTING;
		share_mode = FILE_SHARE_WRITE;
1224
	} else if (create_mode == OS_FILE_OPEN
unknown's avatar
unknown committed
1225
		   || create_mode == OS_FILE_OPEN_RETRY) {
1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236
		create_flag = OPEN_EXISTING;
	} else if (create_mode == OS_FILE_CREATE) {
		create_flag = CREATE_NEW;
	} else if (create_mode == OS_FILE_OVERWRITE) {
		create_flag = CREATE_ALWAYS;
	} else {
		create_flag = 0;
		ut_error;
	}

	if (purpose == OS_FILE_AIO) {
1237 1238
		/* If specified, use asynchronous (overlapped) io and no
		buffering of writes in the OS */
1239 1240 1241 1242 1243
		attributes = 0;
#ifdef WIN_ASYNC_IO
		if (os_aio_use_native_aio) {
			attributes = attributes | FILE_FLAG_OVERLAPPED;
		}
1244
#endif
1245
#ifdef UNIV_NON_BUFFERED_IO
1246
		if (type == OS_LOG_FILE && srv_flush_log_at_trx_commit == 2) {
1247 1248 1249
			/* Do not use unbuffered i/o to log files because
			value 2 denotes that we do not flush the log at every
			commit, but only once per second */
unknown's avatar
unknown committed
1250 1251
		} else if (srv_win_file_flush_method
			   == SRV_WIN_IO_UNBUFFERED) {
1252
			attributes = attributes | FILE_FLAG_NO_BUFFERING;
unknown's avatar
unknown committed
1253
		}
1254 1255
#endif
	} else if (purpose == OS_FILE_NORMAL) {
1256
		attributes = 0;
1257
#ifdef UNIV_NON_BUFFERED_IO
unknown's avatar
unknown committed
1258
		if (type == OS_LOG_FILE && srv_flush_log_at_trx_commit == 2) {
1259 1260 1261
			/* Do not use unbuffered i/o to log files because
			value 2 denotes that we do not flush the log at every
			commit, but only once per second */
unknown's avatar
unknown committed
1262 1263
		} else if (srv_win_file_flush_method
			   == SRV_WIN_IO_UNBUFFERED) {
1264
			attributes = attributes | FILE_FLAG_NO_BUFFERING;
unknown's avatar
unknown committed
1265
		}
1266 1267 1268 1269 1270 1271
#endif
	} else {
		attributes = 0;
		ut_error;
	}

unknown's avatar
unknown committed
1272
	file = CreateFile((LPCTSTR) name,
unknown's avatar
unknown committed
1273
			  GENERIC_READ | GENERIC_WRITE, /* read and write
1274
							access */
unknown's avatar
unknown committed
1275
			  share_mode,	/* File can be read also by other
unknown's avatar
unknown committed
1276 1277 1278 1279 1280 1281
					processes; we must give the read
					permission because of ibbackup. We do
					not give the write permission to
					others because if one would succeed to
					start 2 instances of mysqld on the
					SAME files, that could cause severe
unknown's avatar
unknown committed
1282
					database corruption! When opening
unknown's avatar
unknown committed
1283
					raw disk partitions, Microsoft manuals
unknown's avatar
unknown committed
1284 1285
					say that we must give also the write
					permission. */
unknown's avatar
unknown committed
1286 1287 1288 1289
			  NULL,	/* default security attributes */
			  create_flag,
			  attributes,
			  NULL);	/* no template file */
1290 1291 1292 1293

	if (file == INVALID_HANDLE_VALUE) {
		*success = FALSE;

1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306
		/* When srv_file_per_table is on, file creation failure may not
		be critical to the whole instance. Do not crash the server in
		case of unknown errors. */
		if (srv_file_per_table) {
			retry = os_file_handle_error_no_exit(name,
						create_mode == OS_FILE_CREATE ?
						"create" : "open");
		} else {
			retry = os_file_handle_error(name,
						create_mode == OS_FILE_CREATE ?
						"create" : "open");
		}

1307 1308
		if (retry) {
			goto try_again;
1309 1310 1311 1312 1313 1314
		}
	} else {
		*success = TRUE;
	}

	return(file);
1315
#else /* __WIN__ */
1316 1317 1318
	os_file_t	file;
	int		create_flag;
	ibool		retry;
1319
	const char*	mode_str	= NULL;
1320 1321

try_again:
1322 1323
	ut_a(name);

1324
	if (create_mode == OS_FILE_OPEN || create_mode == OS_FILE_OPEN_RAW
unknown's avatar
unknown committed
1325
	    || create_mode == OS_FILE_OPEN_RETRY) {
1326
		mode_str = "OPEN";
1327 1328
		create_flag = O_RDWR;
	} else if (create_mode == OS_FILE_CREATE) {
1329
		mode_str = "CREATE";
1330 1331
		create_flag = O_RDWR | O_CREAT | O_EXCL;
	} else if (create_mode == OS_FILE_OVERWRITE) {
1332
		mode_str = "OVERWRITE";
1333 1334 1335 1336 1337 1338
		create_flag = O_RDWR | O_CREAT | O_TRUNC;
	} else {
		create_flag = 0;
		ut_error;
	}

1339 1340
	ut_a(type == OS_LOG_FILE || type == OS_DATA_FILE);
	ut_a(purpose == OS_FILE_AIO || purpose == OS_FILE_NORMAL);
1341

1342
#ifdef O_SYNC
1343
	/* We let O_SYNC only affect log files; note that we map O_DSYNC to
1344 1345 1346
	O_SYNC because the datasync options seemed to corrupt files in 2001
	in both Linux and Solaris */
	if (type == OS_LOG_FILE
unknown's avatar
unknown committed
1347
	    && srv_unix_file_flush_method == SRV_UNIX_O_DSYNC) {
1348

unknown's avatar
unknown committed
1349 1350 1351
# if 0
		fprintf(stderr, "Using O_SYNC for file %s\n", name);
# endif
1352

1353
		create_flag = create_flag | O_SYNC;
1354
	}
unknown's avatar
unknown committed
1355
#endif /* O_SYNC */
unknown's avatar
unknown committed
1356 1357

	file = open(name, create_flag, os_innodb_umask);
1358

1359 1360 1361
	if (file == -1) {
		*success = FALSE;

1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374
		/* When srv_file_per_table is on, file creation failure may not
		be critical to the whole instance. Do not crash the server in
		case of unknown errors. */
		if (srv_file_per_table) {
			retry = os_file_handle_error_no_exit(name,
						create_mode == OS_FILE_CREATE ?
						"create" : "open");
		} else {
			retry = os_file_handle_error(name,
						create_mode == OS_FILE_CREATE ?
						"create" : "open");
		}

1375 1376
		if (retry) {
			goto try_again;
unknown's avatar
unknown committed
1377 1378
		} else {
			return(file /* -1 */);
1379
		}
unknown's avatar
unknown committed
1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391
	}
	/* else */

	*success = TRUE;

	/* We disable OS caching (O_DIRECT) only on data files */
	if (type != OS_LOG_FILE
	    && srv_unix_file_flush_method == SRV_UNIX_O_DIRECT) {
		
		os_file_set_nocache(file, name, mode_str);
	}

1392
#ifdef USE_FILE_LOCK
unknown's avatar
unknown committed
1393 1394
	if (create_mode != OS_FILE_OPEN_RAW && os_file_lock(file, name)) {

1395 1396 1397
		if (create_mode == OS_FILE_OPEN_RETRY) {
			int i;
			ut_print_timestamp(stderr);
unknown's avatar
unknown committed
1398 1399 1400
			fputs("  InnoDB: Retrying to lock"
			      " the first data file\n",
			      stderr);
1401 1402 1403 1404 1405 1406 1407 1408 1409
			for (i = 0; i < 100; i++) {
				os_thread_sleep(1000000);
				if (!os_file_lock(file, name)) {
					*success = TRUE;
					return(file);
				}
			}
			ut_print_timestamp(stderr);
			fputs("  InnoDB: Unable to open the first data file\n",
unknown's avatar
unknown committed
1410
			      stderr);
1411
		}
unknown's avatar
unknown committed
1412 1413

		*success = FALSE;
1414
		close(file);
1415
		file = -1;
1416
	}
unknown's avatar
unknown committed
1417
#endif /* USE_FILE_LOCK */
1418

1419
	return(file);
1420
#endif /* __WIN__ */
1421 1422
}

unknown's avatar
unknown committed
1423 1424 1425 1426 1427 1428
/***************************************************************************
Deletes a file if it exists. The file has to be closed before calling this. */

ibool
os_file_delete_if_exists(
/*=====================*/
1429 1430
				/* out: TRUE if success */
	const char*	name)	/* in: file path as a null-terminated string */
unknown's avatar
unknown committed
1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444
{
#ifdef __WIN__
	BOOL	ret;
	ulint	count	= 0;
loop:
	/* In Windows, deleting an .ibd file may fail if ibbackup is copying
	it */

	ret = DeleteFile((LPCTSTR)name);

	if (ret) {
		return(TRUE);
	}

unknown's avatar
unknown committed
1445
	if (GetLastError() == ERROR_FILE_NOT_FOUND) {
unknown's avatar
unknown committed
1446 1447 1448 1449 1450 1451 1452 1453 1454
		/* the file does not exist, this not an error */

		return(TRUE);
	}

	count++;

	if (count > 100 && 0 == (count % 10)) {
		fprintf(stderr,
unknown's avatar
unknown committed
1455 1456 1457
			"InnoDB: Warning: cannot delete file %s\n"
			"InnoDB: Are you running ibbackup"
			" to back up the file?\n", name);
1458

unknown's avatar
unknown committed
1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475
		os_file_get_last_error(TRUE); /* print error information */
	}

	os_thread_sleep(1000000);	/* sleep for a second */

	if (count > 2000) {

		return(FALSE);
	}

	goto loop;
#else
	int	ret;

	ret = unlink((const char*)name);

	if (ret != 0 && errno != ENOENT) {
unknown's avatar
unknown committed
1476
		os_file_handle_error_no_exit(name, "delete");
unknown's avatar
unknown committed
1477 1478 1479 1480 1481 1482 1483 1484

		return(FALSE);
	}

	return(TRUE);
#endif
}

unknown's avatar
unknown committed
1485 1486 1487 1488 1489 1490
/***************************************************************************
Deletes a file. The file has to be closed before calling this. */

ibool
os_file_delete(
/*===========*/
1491 1492
				/* out: TRUE if success */
	const char*	name)	/* in: file path as a null-terminated string */
unknown's avatar
unknown committed
1493 1494
{
#ifdef __WIN__
unknown's avatar
unknown committed
1495
	BOOL	ret;
unknown's avatar
unknown committed
1496 1497 1498 1499
	ulint	count	= 0;
loop:
	/* In Windows, deleting an .ibd file may fail if ibbackup is copying
	it */
unknown's avatar
unknown committed
1500 1501 1502 1503 1504 1505 1506

	ret = DeleteFile((LPCTSTR)name);

	if (ret) {
		return(TRUE);
	}

unknown's avatar
unknown committed
1507
	if (GetLastError() == ERROR_FILE_NOT_FOUND) {
unknown's avatar
unknown committed
1508 1509
		/* If the file does not exist, we classify this as a 'mild'
		error and return */
unknown's avatar
unknown committed
1510

unknown's avatar
unknown committed
1511 1512 1513 1514 1515 1516 1517
		return(FALSE);
	}

	count++;

	if (count > 100 && 0 == (count % 10)) {
		fprintf(stderr,
unknown's avatar
unknown committed
1518 1519 1520
			"InnoDB: Warning: cannot delete file %s\n"
			"InnoDB: Are you running ibbackup"
			" to back up the file?\n", name);
1521

unknown's avatar
unknown committed
1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532
		os_file_get_last_error(TRUE); /* print error information */
	}

	os_thread_sleep(1000000);	/* sleep for a second */

	if (count > 2000) {

		return(FALSE);
	}

	goto loop;
unknown's avatar
unknown committed
1533
#else
unknown's avatar
unknown committed
1534
	int	ret;
unknown's avatar
unknown committed
1535 1536 1537 1538

	ret = unlink((const char*)name);

	if (ret != 0) {
unknown's avatar
unknown committed
1539
		os_file_handle_error_no_exit(name, "delete");
unknown's avatar
unknown committed
1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555

		return(FALSE);
	}

	return(TRUE);
#endif
}

/***************************************************************************
Renames a file (can also move it to another directory). It is safest that the
file is closed before calling this function. */

ibool
os_file_rename(
/*===========*/
				/* out: TRUE if success */
1556
	const char*	oldpath,/* in: old file path as a null-terminated
unknown's avatar
unknown committed
1557
				string */
1558
	const char*	newpath)/* in: new file path */
unknown's avatar
unknown committed
1559 1560
{
#ifdef __WIN__
unknown's avatar
unknown committed
1561
	BOOL	ret;
unknown's avatar
unknown committed
1562 1563 1564 1565 1566 1567 1568

	ret = MoveFile((LPCTSTR)oldpath, (LPCTSTR)newpath);

	if (ret) {
		return(TRUE);
	}

unknown's avatar
unknown committed
1569
	os_file_handle_error_no_exit(oldpath, "rename");
unknown's avatar
unknown committed
1570 1571 1572

	return(FALSE);
#else
unknown's avatar
unknown committed
1573
	int	ret;
unknown's avatar
unknown committed
1574 1575 1576 1577

	ret = rename((const char*)oldpath, (const char*)newpath);

	if (ret != 0) {
unknown's avatar
unknown committed
1578
		os_file_handle_error_no_exit(oldpath, "rename");
unknown's avatar
unknown committed
1579 1580 1581 1582 1583 1584 1585 1586

		return(FALSE);
	}

	return(TRUE);
#endif
}

1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607
/***************************************************************************
Closes a file handle. In case of error, error number can be retrieved with
os_file_get_last_error. */

ibool
os_file_close(
/*==========*/
				/* out: TRUE if success */
	os_file_t	file)	/* in, own: handle to a file */
{
#ifdef __WIN__
	BOOL	ret;

	ut_a(file);

	ret = CloseHandle(file);

	if (ret) {
		return(TRUE);
	}

1608
	os_file_handle_error(NULL, "close");
unknown's avatar
unknown committed
1609

1610 1611 1612 1613 1614 1615 1616
	return(FALSE);
#else
	int	ret;

	ret = close(file);

	if (ret == -1) {
1617
		os_file_handle_error(NULL, "close");
unknown's avatar
unknown committed
1618

1619 1620 1621 1622 1623 1624 1625
		return(FALSE);
	}

	return(TRUE);
#endif
}

unknown's avatar
unknown committed
1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660
/***************************************************************************
Closes a file handle. */

ibool
os_file_close_no_error_handling(
/*============================*/
				/* out: TRUE if success */
	os_file_t	file)	/* in, own: handle to a file */
{
#ifdef __WIN__
	BOOL	ret;

	ut_a(file);

	ret = CloseHandle(file);

	if (ret) {
		return(TRUE);
	}

	return(FALSE);
#else
	int	ret;

	ret = close(file);

	if (ret == -1) {

		return(FALSE);
	}

	return(TRUE);
#endif
}

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
/***************************************************************************
Gets a file size. */

ibool
os_file_get_size(
/*=============*/
				/* out: TRUE if success */
	os_file_t	file,	/* in: handle to a file */
	ulint*		size,	/* out: least significant 32 bits of file
				size */
	ulint*		size_high)/* out: most significant 32 bits of size */
{
#ifdef __WIN__
	DWORD	high;
	DWORD	low;

	low = GetFileSize(file, &high);

	if ((low == 0xFFFFFFFF) && (GetLastError() != NO_ERROR)) {
		return(FALSE);
	}

	*size = low;
	*size_high = high;

	return(TRUE);
#else
unknown's avatar
Merge  
unknown committed
1688 1689 1690 1691
	off_t	offs;

	offs = lseek(file, 0, SEEK_END);

unknown's avatar
unknown committed
1692 1693 1694 1695
	if (offs == ((off_t)-1)) {

		return(FALSE);
	}
1696

unknown's avatar
Merge  
unknown committed
1697
	if (sizeof(off_t) > 4) {
1698
		*size = (ulint)(offs & 0xFFFFFFFFUL);
1699
		*size_high = (ulint)(offs >> 32);
unknown's avatar
Merge  
unknown committed
1700 1701 1702 1703
	} else {
		*size = (ulint) offs;
		*size_high = 0;
	}
1704 1705

	return(TRUE);
1706 1707 1708
#endif
}

1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731
/***************************************************************************
Gets file size as a 64-bit integer ib_longlong. */

ib_longlong
os_file_get_size_as_iblonglong(
/*===========================*/
				/* out: size in bytes, -1 if error */
	os_file_t	file)	/* in: handle to a file */
{
	ulint	size;
	ulint	size_high;
	ibool	success;

	success = os_file_get_size(file, &size, &size_high);

	if (!success) {

		return(-1);
	}

	return((((ib_longlong)size_high) << 32) + (ib_longlong)size);
}

1732
/***************************************************************************
1733
Write the specified number of zeros to a newly created file. */
1734 1735 1736 1737 1738

ibool
os_file_set_size(
/*=============*/
				/* out: TRUE if success */
1739
	const char*	name,	/* in: name of the file or path as a
1740 1741 1742 1743 1744 1745
				null-terminated string */
	os_file_t	file,	/* in: handle to a file */
	ulint		size,	/* in: least significant 32 bits of file
				size */
	ulint		size_high)/* in: most significant 32 bits of size */
{
1746 1747
	ib_longlong	current_size;
	ib_longlong	desired_size;
unknown's avatar
Merge  
unknown committed
1748
	ibool		ret;
1749 1750
	byte*		buf;
	byte*		buf2;
1751
	ulint		buf_size;
unknown's avatar
Merge  
unknown committed
1752

1753
	ut_a(size == (size & 0xFFFFFFFF));
1754

1755 1756
	current_size = 0;
	desired_size = (ib_longlong)size + (((ib_longlong)size_high) << 32);
1757

1758
	/* Write up to 1 megabyte at a time. */
unknown's avatar
unknown committed
1759
	buf_size = ut_min(64, (ulint) (desired_size / UNIV_PAGE_SIZE))
unknown's avatar
unknown committed
1760
		* UNIV_PAGE_SIZE;
1761
	buf2 = ut_malloc(buf_size + UNIV_PAGE_SIZE);
unknown's avatar
unknown committed
1762 1763 1764

	/* Align the buffer for possible raw i/o */
	buf = ut_align(buf2, UNIV_PAGE_SIZE);
1765 1766

	/* Write buffer full of zeros */
1767
	memset(buf, 0, buf_size);
1768

1769
	if (desired_size >= (ib_longlong)(100 * 1024 * 1024)) {
1770

unknown's avatar
unknown committed
1771 1772 1773
		fprintf(stderr, "InnoDB: Progress in MB:");
	}

1774
	while (current_size < desired_size) {
unknown's avatar
unknown committed
1775 1776 1777 1778 1779 1780 1781 1782
		ulint	n_bytes;

		if (desired_size - current_size < (ib_longlong) buf_size) {
			n_bytes = (ulint) (desired_size - current_size);
		} else {
			n_bytes = buf_size;
		}

1783
		ret = os_file_write(name, file, buf,
unknown's avatar
unknown committed
1784 1785 1786
				    (ulint)(current_size & 0xFFFFFFFF),
				    (ulint)(current_size >> 32),
				    n_bytes);
1787
		if (!ret) {
unknown's avatar
unknown committed
1788
			ut_free(buf2);
1789 1790 1791
			goto error_handling;
		}

unknown's avatar
unknown committed
1792
		/* Print about progress for each 100 MB written */
unknown's avatar
unknown committed
1793
		if ((ib_longlong) (current_size + n_bytes) / (ib_longlong)(100 * 1024 * 1024)
unknown's avatar
unknown committed
1794
		    != current_size / (ib_longlong)(100 * 1024 * 1024)) {
unknown's avatar
unknown committed
1795

1796
			fprintf(stderr, " %lu00",
1797
				(ulong) ((current_size + n_bytes)
unknown's avatar
unknown committed
1798
					 / (ib_longlong)(100 * 1024 * 1024)));
unknown's avatar
unknown committed
1799
		}
1800 1801

		current_size += n_bytes;
1802 1803
	}

1804
	if (desired_size >= (ib_longlong)(100 * 1024 * 1024)) {
1805

unknown's avatar
unknown committed
1806 1807 1808
		fprintf(stderr, "\n");
	}

unknown's avatar
unknown committed
1809
	ut_free(buf2);
1810 1811 1812 1813

	ret = os_file_flush(file);

	if (ret) {
1814
		return(TRUE);
1815 1816 1817
	}

error_handling:
1818
	return(FALSE);
1819 1820 1821
}

/***************************************************************************
1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837
Truncates a file at its current position. */

ibool
os_file_set_eof(
/*============*/
				/* out: TRUE if success */
	FILE*		file)	/* in: file to be truncated */
{
#ifdef __WIN__
	HANDLE h = (HANDLE) _get_osfhandle(fileno(file));
	return(SetEndOfFile(h));
#else /* __WIN__ */
	return(!ftruncate(fileno(file), ftell(file)));
#endif /* __WIN__ */
}

unknown's avatar
unknown committed
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 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886
#ifndef __WIN__
/***************************************************************************
Wrapper to fsync(2) that retries the call on some errors.
Returns the value 0 if successful; otherwise the value -1 is returned and
the global variable errno is set to indicate the error. */

static
int
os_file_fsync(
/*==========*/
				/* out: 0 if success, -1 otherwise */
	os_file_t	file)	/* in: handle to a file */
{
	int	ret;
	int	failures;
	ibool	retry;

	failures = 0;

	do {
		ret = fsync(file);

		os_n_fsyncs++;

		if (ret == -1 && errno == ENOLCK) {

			if (failures % 100 == 0) {

				ut_print_timestamp(stderr);
				fprintf(stderr,
					"  InnoDB: fsync(): "
					"No locks available; retrying\n");
			}

			os_thread_sleep(200000 /* 0.2 sec */);

			failures++;

			retry = TRUE;
		} else {

			retry = FALSE;
		}
	} while (retry);

	return(ret);
}
#endif /* !__WIN__ */

1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900
/***************************************************************************
Flushes the write buffers of a given file to the disk. */

ibool
os_file_flush(
/*==========*/
				/* out: TRUE if success */
	os_file_t	file)	/* in, own: handle to a file */
{
#ifdef __WIN__
	BOOL	ret;

	ut_a(file);

unknown's avatar
unknown committed
1901 1902
	os_n_fsyncs++;

1903 1904 1905 1906 1907 1908
	ret = FlushFileBuffers(file);

	if (ret) {
		return(TRUE);
	}

unknown's avatar
unknown committed
1909 1910 1911 1912 1913
	/* Since Windows returns ERROR_INVALID_FUNCTION if the 'file' is
	actually a raw device, we choose to ignore that error if we are using
	raw disks */

	if (srv_start_raw_disk_in_use && GetLastError()
unknown's avatar
unknown committed
1914
	    == ERROR_INVALID_FUNCTION) {
1915
		return(TRUE);
unknown's avatar
unknown committed
1916 1917
	}

1918
	os_file_handle_error(NULL, "flush");
1919

unknown's avatar
unknown committed
1920 1921
	/* It is a fatal error if a file flush does not succeed, because then
	the database can get corrupt on disk */
1922
	ut_error;
unknown's avatar
unknown committed
1923

1924 1925 1926
	return(FALSE);
#else
	int	ret;
1927

1928 1929 1930 1931 1932 1933 1934
#if defined(HAVE_DARWIN_THREADS)
# ifndef F_FULLFSYNC
	/* The following definition is from the Mac OS X 10.3 <sys/fcntl.h> */
#  define F_FULLFSYNC 51 /* fsync + ask the drive to flush to the media */
# elif F_FULLFSYNC != 51
#  error "F_FULLFSYNC != 51: ABI incompatibility with Mac OS X 10.3"
# endif
unknown's avatar
unknown committed
1935 1936 1937 1938 1939
	/* Apple has disabled fsync() for internal disk drives in OS X. That
	caused corruption for a user when he tested a power outage. Let us in
	OS X use a nonstandard flush method recommended by an Apple
	engineer. */

1940 1941
	if (!srv_have_fullfsync) {
		/* If we are not on an operating system that supports this,
1942
		then fall back to a plain fsync. */
unknown's avatar
unknown committed
1943

unknown's avatar
unknown committed
1944
		ret = os_file_fsync(file);
1945 1946 1947 1948 1949
	} else {
		ret = fcntl(file, F_FULLFSYNC, NULL);

		if (ret) {
			/* If we are not on a file system that supports this,
1950
			then fall back to a plain fsync. */
unknown's avatar
unknown committed
1951
			ret = os_file_fsync(file);
1952
		}
unknown's avatar
unknown committed
1953
	}
1954
#else
unknown's avatar
unknown committed
1955
	ret = os_file_fsync(file);
1956
#endif
1957

1958 1959 1960
	if (ret == 0) {
		return(TRUE);
	}
1961

unknown's avatar
unknown committed
1962
	/* Since Linux returns EINVAL if the 'file' is actually a raw device,
unknown's avatar
unknown committed
1963 1964 1965
	we choose to ignore that error if we are using raw disks */

	if (srv_start_raw_disk_in_use && errno == EINVAL) {
unknown's avatar
unknown committed
1966

1967
		return(TRUE);
unknown's avatar
unknown committed
1968 1969
	}

unknown's avatar
unknown committed
1970
	ut_print_timestamp(stderr);
1971

1972
	fprintf(stderr,
unknown's avatar
unknown committed
1973
		"  InnoDB: Error: the OS said file flush did not succeed\n");
1974

1975
	os_file_handle_error(NULL, "flush");
1976

unknown's avatar
unknown committed
1977 1978
	/* It is a fatal error if a file flush does not succeed, because then
	the database can get corrupt on disk */
1979
	ut_error;
unknown's avatar
unknown committed
1980

1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994
	return(FALSE);
#endif
}

#ifndef __WIN__
/***********************************************************************
Does a synchronous read operation in Posix. */
static
ssize_t
os_file_pread(
/*==========*/
				/* out: number of bytes read, -1 if error */
	os_file_t	file,	/* in: handle to a file */
	void*		buf,	/* in: buffer where to read */
1995
	ulint		n,	/* in: number of bytes to read */
unknown's avatar
Merge  
unknown committed
1996 1997 1998 1999
	ulint		offset,	/* in: least significant 32 bits of file
				offset from where to read */
	ulint		offset_high) /* in: most significant 32 bits of
				offset */
2000
{
2001
	off_t	offs;
unknown's avatar
unknown committed
2002
	ssize_t	n_bytes;
unknown's avatar
Merge  
unknown committed
2003

unknown's avatar
unknown committed
2004
	ut_a((offset & 0xFFFFFFFFUL) == offset);
2005 2006

	/* If off_t is > 4 bytes in size, then we assume we can pass a
unknown's avatar
Merge  
unknown committed
2007 2008
	64-bit address */

2009 2010 2011 2012 2013
	if (sizeof(off_t) > 4) {
		offs = (off_t)offset + (((off_t)offset_high) << 32);

	} else {
		offs = (off_t)offset;
unknown's avatar
Merge  
unknown committed
2014

2015 2016
		if (offset_high > 0) {
			fprintf(stderr,
unknown's avatar
unknown committed
2017
				"InnoDB: Error: file read at offset > 4 GB\n");
unknown's avatar
Merge  
unknown committed
2018
		}
2019
	}
2020

2021 2022
	os_n_file_reads++;

unknown's avatar
unknown committed
2023
#if defined(HAVE_PREAD) && !defined(HAVE_BROKEN_PREAD)
2024
	os_mutex_enter(os_file_count_mutex);
unknown's avatar
unknown committed
2025
	os_file_n_pending_preads++;
2026 2027
	os_n_pending_reads++;
	os_mutex_exit(os_file_count_mutex);
unknown's avatar
unknown committed
2028

2029
	n_bytes = pread(file, buf, (ssize_t)n, offs);
unknown's avatar
unknown committed
2030

2031
	os_mutex_enter(os_file_count_mutex);
unknown's avatar
unknown committed
2032
	os_file_n_pending_preads--;
2033 2034
	os_n_pending_reads--;
	os_mutex_exit(os_file_count_mutex);
unknown's avatar
unknown committed
2035 2036

	return(n_bytes);
2037
#else
unknown's avatar
unknown committed
2038
	{
unknown's avatar
unknown committed
2039 2040 2041
		off_t	ret_offset;
		ssize_t	ret;
		ulint	i;
2042

unknown's avatar
unknown committed
2043 2044 2045
		os_mutex_enter(os_file_count_mutex);
		os_n_pending_reads++;
		os_mutex_exit(os_file_count_mutex);
2046

unknown's avatar
unknown committed
2047 2048
		/* Protect the seek / read operation with a mutex */
		i = ((ulint) file) % OS_FILE_N_SEEK_MUTEXES;
2049

unknown's avatar
unknown committed
2050
		os_mutex_enter(os_file_seek_mutexes[i]);
2051

unknown's avatar
unknown committed
2052
		ret_offset = lseek(file, offs, SEEK_SET);
2053

unknown's avatar
unknown committed
2054 2055 2056 2057 2058
		if (ret_offset < 0) {
			ret = -1;
		} else {
			ret = read(file, buf, (ssize_t)n);
		}
2059

unknown's avatar
unknown committed
2060
		os_mutex_exit(os_file_seek_mutexes[i]);
2061

unknown's avatar
unknown committed
2062 2063 2064
		os_mutex_enter(os_file_count_mutex);
		os_n_pending_reads--;
		os_mutex_exit(os_file_count_mutex);
2065

unknown's avatar
unknown committed
2066
		return(ret);
unknown's avatar
unknown committed
2067
	}
2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078
#endif
}

/***********************************************************************
Does a synchronous write operation in Posix. */
static
ssize_t
os_file_pwrite(
/*===========*/
				/* out: number of bytes written, -1 if error */
	os_file_t	file,	/* in: handle to a file */
2079
	const void*	buf,	/* in: buffer from where to write */
2080
	ulint		n,	/* in: number of bytes to write */
unknown's avatar
Merge  
unknown committed
2081 2082 2083 2084
	ulint		offset,	/* in: least significant 32 bits of file
				offset where to write */
	ulint		offset_high) /* in: most significant 32 bits of
				offset */
2085
{
2086
	ssize_t	ret;
2087
	off_t	offs;
unknown's avatar
Merge  
unknown committed
2088

unknown's avatar
unknown committed
2089
	ut_a((offset & 0xFFFFFFFFUL) == offset);
unknown's avatar
Merge  
unknown committed
2090

2091
	/* If off_t is > 4 bytes in size, then we assume we can pass a
unknown's avatar
Merge  
unknown committed
2092 2093
	64-bit address */

2094 2095 2096 2097
	if (sizeof(off_t) > 4) {
		offs = (off_t)offset + (((off_t)offset_high) << 32);
	} else {
		offs = (off_t)offset;
unknown's avatar
Merge  
unknown committed
2098

2099 2100
		if (offset_high > 0) {
			fprintf(stderr,
unknown's avatar
unknown committed
2101 2102
				"InnoDB: Error: file write"
				" at offset > 4 GB\n");
unknown's avatar
Merge  
unknown committed
2103
		}
2104
	}
2105

2106 2107
	os_n_file_writes++;

unknown's avatar
unknown committed
2108
#if defined(HAVE_PWRITE) && !defined(HAVE_BROKEN_PREAD)
2109
	os_mutex_enter(os_file_count_mutex);
unknown's avatar
unknown committed
2110
	os_file_n_pending_pwrites++;
2111 2112
	os_n_pending_writes++;
	os_mutex_exit(os_file_count_mutex);
unknown's avatar
unknown committed
2113

unknown's avatar
unknown committed
2114
	ret = pwrite(file, buf, (ssize_t)n, offs);
2115

2116
	os_mutex_enter(os_file_count_mutex);
unknown's avatar
unknown committed
2117
	os_file_n_pending_pwrites--;
2118 2119
	os_n_pending_writes--;
	os_mutex_exit(os_file_count_mutex);
unknown's avatar
unknown committed
2120

2121
# ifdef UNIV_DO_FLUSH
2122
	if (srv_unix_file_flush_method != SRV_UNIX_LITTLESYNC
unknown's avatar
unknown committed
2123 2124
	    && srv_unix_file_flush_method != SRV_UNIX_NOSYNC
	    && !os_do_not_call_flush_at_each_write) {
2125

2126 2127 2128 2129 2130
		/* Always do fsync to reduce the probability that when
		the OS crashes, a database page is only partially
		physically written to disk. */

		ut_a(TRUE == os_file_flush(file));
2131
	}
2132
# endif /* UNIV_DO_FLUSH */
2133

2134
	return(ret);
2135
#else
unknown's avatar
unknown committed
2136
	{
unknown's avatar
unknown committed
2137 2138
		off_t	ret_offset;
		ulint	i;
2139

unknown's avatar
unknown committed
2140 2141 2142
		os_mutex_enter(os_file_count_mutex);
		os_n_pending_writes++;
		os_mutex_exit(os_file_count_mutex);
2143

unknown's avatar
unknown committed
2144 2145
		/* Protect the seek / write operation with a mutex */
		i = ((ulint) file) % OS_FILE_N_SEEK_MUTEXES;
2146

unknown's avatar
unknown committed
2147
		os_mutex_enter(os_file_seek_mutexes[i]);
2148

unknown's avatar
unknown committed
2149
		ret_offset = lseek(file, offs, SEEK_SET);
2150

unknown's avatar
unknown committed
2151 2152
		if (ret_offset < 0) {
			ret = -1;
2153

unknown's avatar
unknown committed
2154 2155
			goto func_exit;
		}
2156

unknown's avatar
unknown committed
2157
		ret = write(file, buf, (ssize_t)n);
2158

2159
# ifdef UNIV_DO_FLUSH
unknown's avatar
unknown committed
2160 2161 2162
		if (srv_unix_file_flush_method != SRV_UNIX_LITTLESYNC
		    && srv_unix_file_flush_method != SRV_UNIX_NOSYNC
		    && !os_do_not_call_flush_at_each_write) {
2163

unknown's avatar
unknown committed
2164 2165 2166
			/* Always do fsync to reduce the probability that when
			the OS crashes, a database page is only partially
			physically written to disk. */
2167

unknown's avatar
unknown committed
2168 2169
			ut_a(TRUE == os_file_flush(file));
		}
2170
# endif /* UNIV_DO_FLUSH */
2171

2172
func_exit:
unknown's avatar
unknown committed
2173
		os_mutex_exit(os_file_seek_mutexes[i]);
2174

unknown's avatar
unknown committed
2175 2176 2177
		os_mutex_enter(os_file_count_mutex);
		os_n_pending_writes--;
		os_mutex_exit(os_file_count_mutex);
2178

unknown's avatar
unknown committed
2179
		return(ret);
unknown's avatar
unknown committed
2180
	}
2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198
#endif
}
#endif

/***********************************************************************
Requests a synchronous positioned read operation. */

ibool
os_file_read(
/*=========*/
				/* out: TRUE if request was
				successful, FALSE if fail */
	os_file_t	file,	/* in: handle to a file */
	void*		buf,	/* in: buffer where to read */
	ulint		offset,	/* in: least significant 32 bits of file
				offset where to read */
	ulint		offset_high, /* in: most significant 32 bits of
				offset */
2199
	ulint		n)	/* in: number of bytes to read */
2200 2201 2202 2203 2204 2205 2206 2207 2208
{
#ifdef __WIN__
	BOOL		ret;
	DWORD		len;
	DWORD		ret2;
	DWORD		low;
	DWORD		high;
	ibool		retry;
	ulint		i;
2209

unknown's avatar
unknown committed
2210
	ut_a((offset & 0xFFFFFFFFUL) == offset);
unknown's avatar
Merge  
unknown committed
2211

2212
	os_n_file_reads++;
unknown's avatar
unknown committed
2213
	os_bytes_read_since_printout += n;
2214

2215
try_again:
2216 2217 2218 2219
	ut_ad(file);
	ut_ad(buf);
	ut_ad(n > 0);

2220 2221
	low = (DWORD) offset;
	high = (DWORD) offset_high;
2222

2223 2224 2225 2226
	os_mutex_enter(os_file_count_mutex);
	os_n_pending_reads++;
	os_mutex_exit(os_file_count_mutex);

2227 2228
	/* Protect the seek / read operation with a mutex */
	i = ((ulint) file) % OS_FILE_N_SEEK_MUTEXES;
2229

2230 2231 2232 2233 2234 2235 2236 2237
	os_mutex_enter(os_file_seek_mutexes[i]);

	ret2 = SetFilePointer(file, low, &high, FILE_BEGIN);

	if (ret2 == 0xFFFFFFFF && GetLastError() != NO_ERROR) {

		os_mutex_exit(os_file_seek_mutexes[i]);

2238 2239 2240 2241
		os_mutex_enter(os_file_count_mutex);
		os_n_pending_reads--;
		os_mutex_exit(os_file_count_mutex);

2242
		goto error_handling;
2243
	}
2244

2245
	ret = ReadFile(file, buf, (DWORD) n, &len, NULL);
2246

2247
	os_mutex_exit(os_file_seek_mutexes[i]);
2248 2249 2250 2251 2252

	os_mutex_enter(os_file_count_mutex);
	os_n_pending_reads--;
	os_mutex_exit(os_file_count_mutex);

2253 2254
	if (ret && len == n) {
		return(TRUE);
2255
	}
2256 2257 2258
#else
	ibool	retry;
	ssize_t	ret;
unknown's avatar
Merge  
unknown committed
2259

unknown's avatar
unknown committed
2260 2261
	os_bytes_read_since_printout += n;

2262
try_again:
unknown's avatar
Merge  
unknown committed
2263
	ret = os_file_pread(file, buf, n, offset, offset_high);
2264

2265
	if ((ulint)ret == n) {
2266 2267 2268

		return(TRUE);
	}
unknown's avatar
unknown committed
2269 2270

	fprintf(stderr,
unknown's avatar
unknown committed
2271 2272 2273 2274
		"InnoDB: Error: tried to read %lu bytes at offset %lu %lu.\n"
		"InnoDB: Was only able to read %ld.\n",
		(ulong)n, (ulong)offset_high,
		(ulong)offset, (long)ret);
2275
#endif
unknown's avatar
unknown committed
2276
#ifdef __WIN__
2277
error_handling:
unknown's avatar
unknown committed
2278
#endif
2279
	retry = os_file_handle_error(NULL, "read");
2280 2281 2282 2283

	if (retry) {
		goto try_again;
	}
2284

unknown's avatar
unknown committed
2285
	fprintf(stderr,
unknown's avatar
unknown committed
2286 2287
		"InnoDB: Fatal error: cannot read from file."
		" OS error number %lu.\n",
unknown's avatar
unknown committed
2288
#ifdef __WIN__
2289
		(ulong) GetLastError()
unknown's avatar
unknown committed
2290
#else
2291
		(ulong) errno
unknown's avatar
unknown committed
2292 2293 2294 2295
#endif
		);
	fflush(stderr);

2296 2297 2298 2299 2300
	ut_error;

	return(FALSE);
}

unknown's avatar
unknown committed
2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315
/***********************************************************************
Requests a synchronous positioned read operation. This function does not do
any error handling. In case of error it returns FALSE. */

ibool
os_file_read_no_error_handling(
/*===========================*/
				/* out: TRUE if request was
				successful, FALSE if fail */
	os_file_t	file,	/* in: handle to a file */
	void*		buf,	/* in: buffer where to read */
	ulint		offset,	/* in: least significant 32 bits of file
				offset where to read */
	ulint		offset_high, /* in: most significant 32 bits of
				offset */
2316
	ulint		n)	/* in: number of bytes to read */
unknown's avatar
unknown committed
2317 2318 2319 2320 2321 2322 2323 2324 2325
{
#ifdef __WIN__
	BOOL		ret;
	DWORD		len;
	DWORD		ret2;
	DWORD		low;
	DWORD		high;
	ibool		retry;
	ulint		i;
2326

unknown's avatar
unknown committed
2327 2328 2329 2330 2331
	ut_a((offset & 0xFFFFFFFFUL) == offset);

	os_n_file_reads++;
	os_bytes_read_since_printout += n;

2332
try_again:
unknown's avatar
unknown committed
2333 2334 2335 2336
	ut_ad(file);
	ut_ad(buf);
	ut_ad(n > 0);

2337 2338
	low = (DWORD) offset;
	high = (DWORD) offset_high;
unknown's avatar
unknown committed
2339

2340 2341 2342 2343
	os_mutex_enter(os_file_count_mutex);
	os_n_pending_reads++;
	os_mutex_exit(os_file_count_mutex);

unknown's avatar
unknown committed
2344 2345
	/* Protect the seek / read operation with a mutex */
	i = ((ulint) file) % OS_FILE_N_SEEK_MUTEXES;
2346

unknown's avatar
unknown committed
2347 2348 2349 2350 2351 2352 2353 2354
	os_mutex_enter(os_file_seek_mutexes[i]);

	ret2 = SetFilePointer(file, low, &high, FILE_BEGIN);

	if (ret2 == 0xFFFFFFFF && GetLastError() != NO_ERROR) {

		os_mutex_exit(os_file_seek_mutexes[i]);

2355 2356 2357 2358
		os_mutex_enter(os_file_count_mutex);
		os_n_pending_reads--;
		os_mutex_exit(os_file_count_mutex);

unknown's avatar
unknown committed
2359
		goto error_handling;
2360 2361
	}

2362
	ret = ReadFile(file, buf, (DWORD) n, &len, NULL);
unknown's avatar
unknown committed
2363 2364

	os_mutex_exit(os_file_seek_mutexes[i]);
2365 2366 2367 2368 2369

	os_mutex_enter(os_file_count_mutex);
	os_n_pending_reads--;
	os_mutex_exit(os_file_count_mutex);

unknown's avatar
unknown committed
2370 2371
	if (ret && len == n) {
		return(TRUE);
2372
	}
unknown's avatar
unknown committed
2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385
#else
	ibool	retry;
	ssize_t	ret;

	os_bytes_read_since_printout += n;

try_again:
	ret = os_file_pread(file, buf, n, offset, offset_high);

	if ((ulint)ret == n) {

		return(TRUE);
	}
2386
#endif
unknown's avatar
unknown committed
2387 2388 2389
#ifdef __WIN__
error_handling:
#endif
2390
	retry = os_file_handle_error_no_exit(NULL, "read");
unknown's avatar
unknown committed
2391 2392 2393 2394

	if (retry) {
		goto try_again;
	}
2395

unknown's avatar
unknown committed
2396 2397 2398
	return(FALSE);
}

2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415
/***********************************************************************
Rewind file to its start, read at most size - 1 bytes from it to str, and
NUL-terminate str. All errors are silently ignored. This function is
mostly meant to be used with temporary files. */

void
os_file_read_string(
/*================*/
	FILE*	file,	/* in: file to read from */
	char*	str,	/* in: buffer where to read */
	ulint	size)	/* in: size of buffer */
{
	size_t	flen;

	if (size == 0) {
		return;
	}
2416

2417 2418 2419 2420 2421
	rewind(file);
	flen = fread(str, 1, size - 1, file);
	str[flen] = '\0';
}

2422 2423 2424 2425 2426 2427 2428 2429
/***********************************************************************
Requests a synchronous write operation. */

ibool
os_file_write(
/*==========*/
				/* out: TRUE if request was
				successful, FALSE if fail */
2430
	const char*	name,	/* in: name of the file or path as a
2431 2432
				null-terminated string */
	os_file_t	file,	/* in: handle to a file */
2433
	const void*	buf,	/* in: buffer from which to write */
2434 2435 2436 2437
	ulint		offset,	/* in: least significant 32 bits of file
				offset where to write */
	ulint		offset_high, /* in: most significant 32 bits of
				offset */
2438
	ulint		n)	/* in: number of bytes to write */
2439 2440 2441 2442 2443 2444 2445 2446
{
#ifdef __WIN__
	BOOL		ret;
	DWORD		len;
	DWORD		ret2;
	DWORD		low;
	DWORD		high;
	ulint		i;
unknown's avatar
unknown committed
2447
	ulint		n_retries	= 0;
2448
	ulint		err;
2449

2450
	ut_a((offset & 0xFFFFFFFF) == offset);
unknown's avatar
Merge  
unknown committed
2451

2452
	os_n_file_writes++;
unknown's avatar
unknown committed
2453

2454 2455 2456
	ut_ad(file);
	ut_ad(buf);
	ut_ad(n > 0);
unknown's avatar
unknown committed
2457
retry:
2458 2459
	low = (DWORD) offset;
	high = (DWORD) offset_high;
2460 2461 2462 2463 2464

	os_mutex_enter(os_file_count_mutex);
	os_n_pending_writes++;
	os_mutex_exit(os_file_count_mutex);

2465 2466
	/* Protect the seek / write operation with a mutex */
	i = ((ulint) file) % OS_FILE_N_SEEK_MUTEXES;
2467

2468 2469 2470 2471 2472 2473 2474
	os_mutex_enter(os_file_seek_mutexes[i]);

	ret2 = SetFilePointer(file, low, &high, FILE_BEGIN);

	if (ret2 == 0xFFFFFFFF && GetLastError() != NO_ERROR) {

		os_mutex_exit(os_file_seek_mutexes[i]);
2475 2476 2477 2478 2479

		os_mutex_enter(os_file_count_mutex);
		os_n_pending_writes--;
		os_mutex_exit(os_file_count_mutex);

unknown's avatar
unknown committed
2480 2481 2482
		ut_print_timestamp(stderr);

		fprintf(stderr,
unknown's avatar
unknown committed
2483 2484 2485 2486 2487 2488 2489 2490 2491
			"  InnoDB: Error: File pointer positioning to"
			" file %s failed at\n"
			"InnoDB: offset %lu %lu. Operating system"
			" error number %lu.\n"
			"InnoDB: Some operating system error numbers"
			" are described at\n"
			"InnoDB: "
			"http://dev.mysql.com/doc/refman/5.1/en/"
			"operating-system-error-codes.html\n",
2492 2493
			name, (ulong) offset_high, (ulong) offset,
			(ulong) GetLastError());
unknown's avatar
unknown committed
2494 2495

		return(FALSE);
2496
	}
2497

2498
	ret = WriteFile(file, buf, (DWORD) n, &len, NULL);
2499

2500 2501 2502
	/* Always do fsync to reduce the probability that when the OS crashes,
	a database page is only partially physically written to disk. */

2503
# ifdef UNIV_DO_FLUSH
unknown's avatar
unknown committed
2504
	if (!os_do_not_call_flush_at_each_write) {
2505 2506
		ut_a(TRUE == os_file_flush(file));
	}
2507
# endif /* UNIV_DO_FLUSH */
2508 2509

	os_mutex_exit(os_file_seek_mutexes[i]);
2510

2511 2512 2513 2514
	os_mutex_enter(os_file_count_mutex);
	os_n_pending_writes--;
	os_mutex_exit(os_file_count_mutex);

2515
	if (ret && len == n) {
unknown's avatar
unknown committed
2516

2517 2518
		return(TRUE);
	}
unknown's avatar
unknown committed
2519

unknown's avatar
unknown committed
2520 2521 2522
	/* If some background file system backup tool is running, then, at
	least in Windows 2000, we may get here a specific error. Let us
	retry the operation 100 times, with 1 second waits. */
2523

unknown's avatar
unknown committed
2524
	if (GetLastError() == ERROR_LOCK_VIOLATION && n_retries < 100) {
unknown's avatar
unknown committed
2525

unknown's avatar
unknown committed
2526
		os_thread_sleep(1000000);
2527

unknown's avatar
unknown committed
2528 2529 2530
		n_retries++;

		goto retry;
2531 2532
	}

unknown's avatar
unknown committed
2533
	if (!os_has_said_disk_full) {
2534

2535 2536
		err = (ulint)GetLastError();

unknown's avatar
unknown committed
2537 2538 2539
		ut_print_timestamp(stderr);

		fprintf(stderr,
unknown's avatar
unknown committed
2540 2541 2542 2543 2544 2545 2546 2547 2548
			"  InnoDB: Error: Write to file %s failed"
			" at offset %lu %lu.\n"
			"InnoDB: %lu bytes should have been written,"
			" only %lu were written.\n"
			"InnoDB: Operating system error number %lu.\n"
			"InnoDB: Check that your OS and file system"
			" support files of this size.\n"
			"InnoDB: Check also that the disk is not full"
			" or a disk quota exceeded.\n",
2549 2550
			name, (ulong) offset_high, (ulong) offset,
			(ulong) n, (ulong) len, (ulong) err);
2551 2552 2553

		if (strerror((int)err) != NULL) {
			fprintf(stderr,
unknown's avatar
unknown committed
2554 2555
				"InnoDB: Error number %lu means '%s'.\n",
				(ulong) err, strerror((int)err));
2556 2557 2558
		}

		fprintf(stderr,
unknown's avatar
unknown committed
2559 2560 2561 2562 2563
			"InnoDB: Some operating system error numbers"
			" are described at\n"
			"InnoDB: "
			"http://dev.mysql.com/doc/refman/5.1/en/"
			"operating-system-error-codes.html\n");
unknown's avatar
unknown committed
2564 2565 2566 2567 2568

		os_has_said_disk_full = TRUE;
	}

	return(FALSE);
2569 2570
#else
	ssize_t	ret;
2571

unknown's avatar
Merge  
unknown committed
2572
	ret = os_file_pwrite(file, buf, n, offset, offset_high);
2573

2574
	if ((ulint)ret == n) {
unknown's avatar
unknown committed
2575

2576 2577 2578
		return(TRUE);
	}

unknown's avatar
unknown committed
2579
	if (!os_has_said_disk_full) {
2580

unknown's avatar
unknown committed
2581
		ut_print_timestamp(stderr);
2582

unknown's avatar
unknown committed
2583
		fprintf(stderr,
unknown's avatar
unknown committed
2584 2585 2586 2587 2588 2589 2590 2591 2592
			"  InnoDB: Error: Write to file %s failed"
			" at offset %lu %lu.\n"
			"InnoDB: %lu bytes should have been written,"
			" only %ld were written.\n"
			"InnoDB: Operating system error number %lu.\n"
			"InnoDB: Check that your OS and file system"
			" support files of this size.\n"
			"InnoDB: Check also that the disk is not full"
			" or a disk quota exceeded.\n",
unknown's avatar
unknown committed
2593
			name, offset_high, offset, n, (long int)ret,
unknown's avatar
unknown committed
2594
			(ulint)errno);
2595 2596
		if (strerror(errno) != NULL) {
			fprintf(stderr,
unknown's avatar
unknown committed
2597 2598
				"InnoDB: Error number %lu means '%s'.\n",
				(ulint)errno, strerror(errno));
2599 2600 2601
		}

		fprintf(stderr,
unknown's avatar
unknown committed
2602 2603 2604 2605 2606
			"InnoDB: Some operating system error numbers"
			" are described at\n"
			"InnoDB: "
			"http://dev.mysql.com/doc/refman/5.1/en/"
			"operating-system-error-codes.html\n");
2607

unknown's avatar
unknown committed
2608 2609 2610
		os_has_said_disk_full = TRUE;
	}

2611
	return(FALSE);
unknown's avatar
unknown committed
2612
#endif
2613 2614
}

unknown's avatar
unknown committed
2615 2616 2617 2618 2619 2620 2621
/***********************************************************************
Check the existence and type of the given file. */

ibool
os_file_status(
/*===========*/
				/* out: TRUE if call succeeded */
2622
	const char*	path,	/* in:	pathname of the file */
unknown's avatar
unknown committed
2623 2624 2625 2626 2627 2628
	ibool*		exists,	/* out: TRUE if file exists */
	os_file_type_t* type)	/* out: type of the file (if it exists) */
{
#ifdef __WIN__
	int		ret;
	struct _stat	statinfo;
2629

unknown's avatar
unknown committed
2630 2631 2632 2633 2634 2635 2636
	ret = _stat(path, &statinfo);
	if (ret && (errno == ENOENT || errno == ENOTDIR)) {
		/* file does not exist */
		*exists = FALSE;
		return(TRUE);
	} else if (ret) {
		/* file exists, but stat call failed */
2637

unknown's avatar
unknown committed
2638
		os_file_handle_error_no_exit(path, "stat");
2639

unknown's avatar
unknown committed
2640 2641
		return(FALSE);
	}
2642

unknown's avatar
unknown committed
2643 2644 2645
	if (_S_IFDIR & statinfo.st_mode) {
		*type = OS_FILE_TYPE_DIR;
	} else if (_S_IFREG & statinfo.st_mode) {
2646
		*type = OS_FILE_TYPE_FILE;
unknown's avatar
unknown committed
2647
	} else {
2648
		*type = OS_FILE_TYPE_UNKNOWN;
unknown's avatar
unknown committed
2649 2650 2651
	}

	*exists = TRUE;
2652

unknown's avatar
unknown committed
2653 2654 2655 2656
	return(TRUE);
#else
	int		ret;
	struct stat	statinfo;
2657

unknown's avatar
unknown committed
2658 2659 2660 2661 2662 2663 2664
	ret = stat(path, &statinfo);
	if (ret && (errno == ENOENT || errno == ENOTDIR)) {
		/* file does not exist */
		*exists = FALSE;
		return(TRUE);
	} else if (ret) {
		/* file exists, but stat call failed */
2665

unknown's avatar
unknown committed
2666
		os_file_handle_error_no_exit(path, "stat");
2667

unknown's avatar
unknown committed
2668 2669
		return(FALSE);
	}
2670

unknown's avatar
unknown committed
2671 2672 2673
	if (S_ISDIR(statinfo.st_mode)) {
		*type = OS_FILE_TYPE_DIR;
	} else if (S_ISLNK(statinfo.st_mode)) {
2674
		*type = OS_FILE_TYPE_LINK;
unknown's avatar
unknown committed
2675
	} else if (S_ISREG(statinfo.st_mode)) {
2676
		*type = OS_FILE_TYPE_FILE;
unknown's avatar
unknown committed
2677
	} else {
2678
		*type = OS_FILE_TYPE_UNKNOWN;
unknown's avatar
unknown committed
2679 2680 2681
	}

	*exists = TRUE;
2682

unknown's avatar
unknown committed
2683 2684 2685 2686
	return(TRUE);
#endif
}

2687 2688 2689 2690 2691
/***********************************************************************
This function returns information about the specified file */

ibool
os_file_get_status(
2692
/*===============*/
unknown's avatar
unknown committed
2693 2694 2695 2696 2697
					/* out: TRUE if stat
					information found */
	const char*	path,		/* in:	pathname of the file */
	os_file_stat_t* stat_info)	/* information of a file in a
					directory */
2698 2699 2700 2701
{
#ifdef __WIN__
	int		ret;
	struct _stat	statinfo;
2702

2703 2704 2705 2706 2707 2708 2709
	ret = _stat(path, &statinfo);
	if (ret && (errno == ENOENT || errno == ENOTDIR)) {
		/* file does not exist */

		return(FALSE);
	} else if (ret) {
		/* file exists, but stat call failed */
2710

unknown's avatar
unknown committed
2711
		os_file_handle_error_no_exit(path, "stat");
2712

2713 2714 2715 2716 2717
		return(FALSE);
	}
	if (_S_IFDIR & statinfo.st_mode) {
		stat_info->type = OS_FILE_TYPE_DIR;
	} else if (_S_IFREG & statinfo.st_mode) {
2718
		stat_info->type = OS_FILE_TYPE_FILE;
2719
	} else {
2720
		stat_info->type = OS_FILE_TYPE_UNKNOWN;
2721 2722 2723 2724 2725
	}

	stat_info->ctime = statinfo.st_ctime;
	stat_info->atime = statinfo.st_atime;
	stat_info->mtime = statinfo.st_mtime;
2726 2727
	stat_info->size	 = statinfo.st_size;

2728 2729 2730 2731
	return(TRUE);
#else
	int		ret;
	struct stat	statinfo;
2732

2733 2734 2735 2736 2737 2738 2739 2740
	ret = stat(path, &statinfo);

	if (ret && (errno == ENOENT || errno == ENOTDIR)) {
		/* file does not exist */

		return(FALSE);
	} else if (ret) {
		/* file exists, but stat call failed */
2741

unknown's avatar
unknown committed
2742
		os_file_handle_error_no_exit(path, "stat");
2743

2744 2745
		return(FALSE);
	}
2746

2747 2748 2749
	if (S_ISDIR(statinfo.st_mode)) {
		stat_info->type = OS_FILE_TYPE_DIR;
	} else if (S_ISLNK(statinfo.st_mode)) {
2750
		stat_info->type = OS_FILE_TYPE_LINK;
2751
	} else if (S_ISREG(statinfo.st_mode)) {
2752
		stat_info->type = OS_FILE_TYPE_FILE;
2753
	} else {
2754
		stat_info->type = OS_FILE_TYPE_UNKNOWN;
2755 2756 2757 2758 2759
	}

	stat_info->ctime = statinfo.st_ctime;
	stat_info->atime = statinfo.st_atime;
	stat_info->mtime = statinfo.st_mtime;
2760 2761
	stat_info->size	 = statinfo.st_size;

2762 2763 2764 2765
	return(TRUE);
#endif
}

unknown's avatar
unknown committed
2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776
/* path name separator character */
#ifdef __WIN__
#  define OS_FILE_PATH_SEPARATOR	'\\'
#else
#  define OS_FILE_PATH_SEPARATOR	'/'
#endif

/********************************************************************
The function os_file_dirname returns a directory component of a
null-terminated pathname string.  In the usual case, dirname returns
the string up to, but not including, the final '/', and basename
unknown's avatar
unknown committed
2777
is the component following the final '/'.  Trailing '/' charac­
unknown's avatar
unknown committed
2778 2779 2780 2781 2782 2783 2784 2785 2786
ters are not counted as part of the pathname.

If path does not contain a slash, dirname returns the string ".".

Concatenating the string returned by dirname, a "/", and the basename
yields a complete pathname.

The return value is  a copy of the directory component of the pathname.
The copy is allocated from heap. It is the caller responsibility
2787
to free it after it is no longer needed.
unknown's avatar
unknown committed
2788 2789 2790 2791

The following list of examples (taken from SUSv2) shows the strings
returned by dirname and basename for different paths:

2792 2793 2794 2795 2796 2797 2798
       path	      dirname	     basename
       "/usr/lib"     "/usr"	     "lib"
       "/usr/"	      "/"	     "usr"
       "usr"	      "."	     "usr"
       "/"	      "/"	     "/"
       "."	      "."	     "."
       ".."	      "."	     ".."
unknown's avatar
unknown committed
2799 2800 2801 2802 2803 2804 2805
*/

char*
os_file_dirname(
/*============*/
				/* out, own: directory component of the
				pathname */
2806
	const char*	path)	/* in: pathname */
unknown's avatar
unknown committed
2807
{
unknown's avatar
unknown committed
2808
	/* Find the offset of the last slash */
2809 2810
	const char* last_slash = strrchr(path, OS_FILE_PATH_SEPARATOR);
	if (!last_slash) {
unknown's avatar
unknown committed
2811 2812
		/* No slash in the path, return "." */

2813
		return(mem_strdup("."));
unknown's avatar
unknown committed
2814 2815
	}

unknown's avatar
unknown committed
2816
	/* Ok, there is a slash */
unknown's avatar
unknown committed
2817

2818
	if (last_slash == path) {
unknown's avatar
unknown committed
2819
		/* last slash is the first char of the path */
unknown's avatar
unknown committed
2820

2821
		return(mem_strdup("/"));
unknown's avatar
unknown committed
2822 2823
	}

unknown's avatar
unknown committed
2824 2825
	/* Non-trivial directory component */

2826
	return(mem_strdupl(path, last_slash - path));
unknown's avatar
unknown committed
2827
}
2828

unknown's avatar
unknown committed
2829 2830
/********************************************************************
Creates all missing subdirectories along the given path. */
2831

unknown's avatar
unknown committed
2832 2833 2834 2835 2836
ibool
os_file_create_subdirs_if_needed(
/*=============================*/
				/* out: TRUE if call succeeded
				   FALSE otherwise */
2837
	const char*	path)	/* in: path name */
unknown's avatar
unknown committed
2838 2839
{
	char*		subdir;
2840
	ibool		success, subdir_exists;
unknown's avatar
unknown committed
2841 2842 2843
	os_file_type_t	type;

	subdir = os_file_dirname(path);
2844
	if (strlen(subdir) == 1
unknown's avatar
unknown committed
2845
	    && (*subdir == OS_FILE_PATH_SEPARATOR || *subdir == '.')) {
unknown's avatar
unknown committed
2846
		/* subdir is root or cwd, nothing to do */
unknown's avatar
unknown committed
2847 2848
		mem_free(subdir);

unknown's avatar
unknown committed
2849 2850 2851
		return(TRUE);
	}

unknown's avatar
unknown committed
2852
	/* Test if subdir exists */
unknown's avatar
unknown committed
2853 2854 2855 2856 2857
	success = os_file_status(subdir, &subdir_exists, &type);
	if (success && !subdir_exists) {
		/* subdir does not exist, create it */
		success = os_file_create_subdirs_if_needed(subdir);
		if (!success) {
unknown's avatar
unknown committed
2858 2859
			mem_free(subdir);

unknown's avatar
unknown committed
2860 2861 2862 2863 2864
			return(FALSE);
		}
		success = os_file_create_directory(subdir, FALSE);
	}

unknown's avatar
unknown committed
2865 2866
	mem_free(subdir);

unknown's avatar
unknown committed
2867 2868 2869
	return(success);
}

2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893
/********************************************************************
Returns a pointer to the nth slot in the aio array. */
static
os_aio_slot_t*
os_aio_array_get_nth_slot(
/*======================*/
					/* out: pointer to slot */
	os_aio_array_t*		array,	/* in: aio array */
	ulint			index)	/* in: index of the slot */
{
	ut_a(index < array->n_slots);

	return((array->slots) + index);
}

/****************************************************************************
Creates an aio wait array. */
static
os_aio_array_t*
os_aio_array_create(
/*================*/
				/* out, own: aio array */
	ulint	n,		/* in: maximum number of pending aio operations
				allowed; n must be divisible by n_segments */
2894
	ulint	n_segments)	/* in: number of segments in the aio array */
2895 2896 2897 2898 2899 2900
{
	os_aio_array_t*	array;
	ulint		i;
	os_aio_slot_t*	slot;
#ifdef WIN_ASYNC_IO
	OVERLAPPED*	over;
2901
#endif
2902 2903 2904 2905 2906
	ut_a(n > 0);
	ut_a(n_segments > 0);

	array = ut_malloc(sizeof(os_aio_array_t));

2907
	array->mutex		= os_mutex_create(NULL);
2908
	array->not_full		= os_event_create(NULL);
2909 2910 2911
	array->is_empty		= os_event_create(NULL);

	os_event_set(array->is_empty);
2912 2913

	array->n_slots		= n;
2914 2915 2916
	array->n_segments	= n_segments;
	array->n_reserved	= 0;
	array->slots		= ut_malloc(n * sizeof(os_aio_slot_t));
2917 2918
#ifdef __WIN__
	array->native_events	= ut_malloc(n * sizeof(os_native_event_t));
2919
#endif
2920 2921 2922 2923 2924 2925
	for (i = 0; i < n; i++) {
		slot = os_aio_array_get_nth_slot(array, i);

		slot->pos = i;
		slot->reserved = FALSE;
#ifdef WIN_ASYNC_IO
2926 2927
		slot->event = os_event_create(NULL);

2928 2929
		over = &(slot->control);

2930
		over->hEvent = slot->event->handle;
2931

2932
		*((array->native_events) + i) = over->hEvent;
2933 2934
#endif
	}
2935

2936 2937 2938 2939
	return(array);
}

/****************************************************************************
unknown's avatar
unknown committed
2940 2941
Initializes the asynchronous io system. Calls also os_io_init_simple.
Creates a separate aio array for
2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967
non-ibuf read and write, a third aio array for the ibuf i/o, with just one
segment, two aio arrays for log reads and writes with one segment, and a
synchronous aio array of the specified size. The combined number of segments
in the three first aio arrays is the parameter n_segments given to the
function. The caller must create an i/o handler thread for each segment in
the four first arrays, but not for the sync aio array. */

void
os_aio_init(
/*========*/
	ulint	n,		/* in: maximum number of pending aio operations
				allowed; n must be divisible by n_segments */
	ulint	n_segments,	/* in: combined number of segments in the four
				first aio arrays; must be >= 4 */
	ulint	n_slots_sync)	/* in: number of slots in the sync aio array */
{
	ulint	n_read_segs;
	ulint	n_write_segs;
	ulint	n_per_seg;
	ulint	i;
#ifdef POSIX_ASYNC_IO
	sigset_t   sigset;
#endif
	ut_ad(n % n_segments == 0);
	ut_ad(n_segments >= 4);

unknown's avatar
unknown committed
2968 2969
	os_io_init_simple();

unknown's avatar
unknown committed
2970
	for (i = 0; i < n_segments; i++) {
2971
		srv_set_io_thread_op_info(i, "not started yet");
unknown's avatar
unknown committed
2972 2973
	}

2974 2975 2976
	n_per_seg = n / n_segments;
	n_write_segs = (n_segments - 2) / 2;
	n_read_segs = n_segments - 2 - n_write_segs;
2977

2978
	/* fprintf(stderr, "Array n per seg %lu\n", n_per_seg); */
2979

unknown's avatar
unknown committed
2980 2981
	os_aio_ibuf_array = os_aio_array_create(n_per_seg, 1);

2982
	srv_io_thread_function[0] = "insert buffer thread";
unknown's avatar
unknown committed
2983 2984 2985

	os_aio_log_array = os_aio_array_create(n_per_seg, 1);

2986
	srv_io_thread_function[1] = "log thread";
unknown's avatar
unknown committed
2987

2988
	os_aio_read_array = os_aio_array_create(n_read_segs * n_per_seg,
unknown's avatar
unknown committed
2989
						n_read_segs);
unknown's avatar
unknown committed
2990
	for (i = 2; i < 2 + n_read_segs; i++) {
unknown's avatar
unknown committed
2991
		ut_a(i < SRV_MAX_N_IO_THREADS);
2992
		srv_io_thread_function[i] = "read thread";
unknown's avatar
unknown committed
2993 2994
	}

2995
	os_aio_write_array = os_aio_array_create(n_write_segs * n_per_seg,
unknown's avatar
unknown committed
2996
						 n_write_segs);
unknown's avatar
unknown committed
2997
	for (i = 2 + n_read_segs; i < n_segments; i++) {
unknown's avatar
unknown committed
2998
		ut_a(i < SRV_MAX_N_IO_THREADS);
2999
		srv_io_thread_function[i] = "write thread";
unknown's avatar
unknown committed
3000
	}
3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013

	os_aio_sync_array = os_aio_array_create(n_slots_sync, 1);

	os_aio_n_segments = n_segments;

	os_aio_validate();

	os_aio_segment_wait_events = ut_malloc(n_segments * sizeof(void*));

	for (i = 0; i < n_segments; i++) {
		os_aio_segment_wait_events[i] = os_event_create(NULL);
	}

3014 3015
	os_last_printout = time(NULL);

3016
#ifdef POSIX_ASYNC_IO
3017 3018 3019 3020
	/* Block aio signals from the current thread and its children:
	for this to work, the current thread must be the first created
	in the database, so that all its children will inherit its
	signal mask */
3021

3022 3023
	/* TODO: to work MySQL needs the SIGALARM signal; the following
	will not work yet! */
3024
	sigemptyset(&sigset);
3025 3026 3027 3028 3029
	sigaddset(&sigset, SIGRTMIN + 1 + 0);
	sigaddset(&sigset, SIGRTMIN + 1 + 1);
	sigaddset(&sigset, SIGRTMIN + 1 + 2);
	sigaddset(&sigset, SIGRTMIN + 1 + 3);

3030
	pthread_sigmask(SIG_BLOCK, &sigset, NULL); */
3031
#endif
unknown's avatar
unknown committed
3032
		}
unknown's avatar
unknown committed
3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047

#ifdef WIN_ASYNC_IO
/****************************************************************************
Wakes up all async i/o threads in the array in Windows async i/o at
shutdown. */
static
void
os_aio_array_wake_win_aio_at_shutdown(
/*==================================*/
	os_aio_array_t*	array)	/* in: aio array */
{
	ulint	i;

	for (i = 0; i < array->n_slots; i++) {

3048
		os_event_set((array->slots + i)->event);
unknown's avatar
unknown committed
3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063
	}
}
#endif

/****************************************************************************
Wakes up all async i/o threads so that they know to exit themselves in
shutdown. */

void
os_aio_wake_all_threads_at_shutdown(void)
/*=====================================*/
{
	ulint	i;

#ifdef WIN_ASYNC_IO
3064
	/* This code wakes up all ai/o threads in Windows native aio */
unknown's avatar
unknown committed
3065 3066 3067 3068 3069 3070 3071 3072
	os_aio_array_wake_win_aio_at_shutdown(os_aio_read_array);
	os_aio_array_wake_win_aio_at_shutdown(os_aio_write_array);
	os_aio_array_wake_win_aio_at_shutdown(os_aio_ibuf_array);
	os_aio_array_wake_win_aio_at_shutdown(os_aio_log_array);
#endif
	/* This loop wakes up all simulated ai/o threads */

	for (i = 0; i < os_aio_n_segments; i++) {
3073

unknown's avatar
unknown committed
3074
		os_event_set(os_aio_segment_wait_events[i]);
3075
	}
unknown's avatar
unknown committed
3076
}
3077

3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088
/****************************************************************************
Waits until there are no pending writes in os_aio_write_array. There can
be other, synchronous, pending writes. */

void
os_aio_wait_until_no_pending_writes(void)
/*=====================================*/
{
	os_event_wait(os_aio_write_array->is_empty);
}

3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107
/**************************************************************************
Calculates segment number for a slot. */
static
ulint
os_aio_get_segment_no_from_slot(
/*============================*/
				/* out: segment number (which is the number
				used by, for example, i/o-handler threads) */
	os_aio_array_t*	array,	/* in: aio wait array */
	os_aio_slot_t*	slot)	/* in: slot in this array */
{
	ulint	segment;
	ulint	seg_len;

	if (array == os_aio_ibuf_array) {
		segment = 0;

	} else if (array == os_aio_log_array) {
		segment = 1;
3108

3109
	} else if (array == os_aio_read_array) {
unknown's avatar
unknown committed
3110 3111
		seg_len = os_aio_read_array->n_slots
			/ os_aio_read_array->n_segments;
3112 3113 3114 3115

		segment = 2 + slot->pos / seg_len;
	} else {
		ut_a(array == os_aio_write_array);
unknown's avatar
unknown committed
3116 3117
		seg_len = os_aio_write_array->n_slots
			/ os_aio_write_array->n_segments;
3118 3119

		segment = os_aio_read_array->n_segments + 2
unknown's avatar
unknown committed
3120
			+ slot->pos / seg_len;
3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138
	}

	return(segment);
}

/**************************************************************************
Calculates local segment number and aio array from global segment number. */
static
ulint
os_aio_get_array_and_local_segment(
/*===============================*/
					/* out: local segment number within
					the aio array */
	os_aio_array_t** array,		/* out: aio wait array */
	ulint		 global_segment)/* in: global segment number */
{
	ulint	segment;

3139
	ut_a(global_segment < os_aio_n_segments);
3140 3141 3142 3143 3144 3145 3146 3147

	if (global_segment == 0) {
		*array = os_aio_ibuf_array;
		segment = 0;

	} else if (global_segment == 1) {
		*array = os_aio_log_array;
		segment = 0;
3148

3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164
	} else if (global_segment < os_aio_read_array->n_segments + 2) {
		*array = os_aio_read_array;

		segment = global_segment - 2;
	} else {
		*array = os_aio_write_array;

		segment = global_segment - (os_aio_read_array->n_segments + 2);
	}

	return(segment);
}

/***********************************************************************
Gets an integer value designating a specified aio array. This is used
to give numbers to signals in Posix aio. */
unknown's avatar
unknown committed
3165 3166

#if !defined(WIN_ASYNC_IO) && defined(POSIX_ASYNC_IO)
3167 3168 3169 3170 3171
static
ulint
os_aio_get_array_no(
/*================*/
	os_aio_array_t*	array)	/* in: aio array */
3172
{
3173
	if (array == os_aio_ibuf_array) {
3174

3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187
		return(0);

	} else if (array == os_aio_log_array) {

		return(1);

	} else if (array == os_aio_read_array) {

		return(2);
	} else if (array == os_aio_write_array) {

		return(3);
	} else {
3188
		ut_error;
3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201

		return(0);
	}
}

/***********************************************************************
Gets the aio array for its number. */
static
os_aio_array_t*
os_aio_get_array_from_no(
/*=====================*/
			/* out: aio array */
	ulint	n)	/* in: array number */
3202
{
3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214
	if (n == 0) {
		return(os_aio_ibuf_array);
	} else if (n == 1) {

		return(os_aio_log_array);
	} else if (n == 2) {

		return(os_aio_read_array);
	} else if (n == 3) {

		return(os_aio_write_array);
	} else {
3215
		ut_error;
3216 3217 3218 3219

		return(NULL);
	}
}
unknown's avatar
unknown committed
3220
#endif /* if !defined(WIN_ASYNC_IO) && defined(POSIX_ASYNC_IO) */
3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231

/***********************************************************************
Requests for a slot in the aio array. If no slot is available, waits until
not_full-event becomes signaled. */
static
os_aio_slot_t*
os_aio_array_reserve_slot(
/*======================*/
				/* out: pointer to slot */
	ulint		type,	/* in: OS_FILE_READ or OS_FILE_WRITE */
	os_aio_array_t*	array,	/* in: aio array */
unknown's avatar
unknown committed
3232
	fil_node_t*	message1,/* in: message to be passed along with
3233 3234 3235 3236
				the aio operation */
	void*		message2,/* in: message to be passed along with
				the aio operation */
	os_file_t	file,	/* in: file handle */
3237
	const char*	name,	/* in: name of the file or path as a
3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267
				null-terminated string */
	void*		buf,	/* in: buffer where to read or from which
				to write */
	ulint		offset,	/* in: least significant 32 bits of file
				offset */
	ulint		offset_high, /* in: most significant 32 bits of
				offset */
	ulint		len)	/* in: length of the block to read or write */
{
	os_aio_slot_t*	slot;
#ifdef WIN_ASYNC_IO
	OVERLAPPED*	control;

#elif defined(POSIX_ASYNC_IO)

	struct aiocb*	control;
#endif
	ulint		i;
loop:
	os_mutex_enter(array->mutex);

	if (array->n_reserved == array->n_slots) {
		os_mutex_exit(array->mutex);

		if (!os_aio_use_native_aio) {
			/* If the handler threads are suspended, wake them
			so that we get more slots */

			os_aio_simulated_wake_handler_threads();
		}
3268

3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283
		os_event_wait(array->not_full);

		goto loop;
	}

	for (i = 0;; i++) {
		slot = os_aio_array_get_nth_slot(array, i);

		if (slot->reserved == FALSE) {
			break;
		}
	}

	array->n_reserved++;

3284 3285 3286 3287
	if (array->n_reserved == 1) {
		os_event_reset(array->is_empty);
	}

3288 3289 3290
	if (array->n_reserved == array->n_slots) {
		os_event_reset(array->not_full);
	}
3291

3292
	slot->reserved = TRUE;
unknown's avatar
unknown committed
3293
	slot->reservation_time = time(NULL);
3294 3295 3296 3297 3298 3299 3300 3301 3302 3303
	slot->message1 = message1;
	slot->message2 = message2;
	slot->file     = file;
	slot->name     = name;
	slot->len      = len;
	slot->type     = type;
	slot->buf      = buf;
	slot->offset   = offset;
	slot->offset_high = offset_high;
	slot->io_already_done = FALSE;
3304 3305

#ifdef WIN_ASYNC_IO
3306 3307 3308
	control = &(slot->control);
	control->Offset = (DWORD)offset;
	control->OffsetHigh = (DWORD)offset_high;
3309
	os_event_reset(slot->event);
3310 3311 3312 3313 3314 3315 3316

#elif defined(POSIX_ASYNC_IO)

#if (UNIV_WORD_SIZE == 8)
	offset = offset + (offset_high << 32);
#else
	ut_a(offset_high == 0);
3317
#endif
3318 3319 3320 3321 3322 3323 3324
	control = &(slot->control);
	control->aio_fildes = file;
	control->aio_buf = buf;
	control->aio_nbytes = len;
	control->aio_offset = offset;
	control->aio_reqprio = 0;
	control->aio_sigevent.sigev_notify = SIGEV_SIGNAL;
unknown's avatar
unknown committed
3325 3326 3327 3328
	control->aio_sigevent.sigev_signo
		= SIGRTMIN + 1 + os_aio_get_array_no(array);
	/* TODO: How to choose the signal numbers? */
	/*
3329
	fprintf(stderr, "AIO signal number %lu\n",
unknown's avatar
unknown committed
3330 3331
	(ulint) control->aio_sigevent.sigev_signo);
	*/
3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353
	control->aio_sigevent.sigev_value.sival_ptr = slot;
#endif
	os_mutex_exit(array->mutex);

	return(slot);
}

/***********************************************************************
Frees a slot in the aio array. */
static
void
os_aio_array_free_slot(
/*===================*/
	os_aio_array_t*	array,	/* in: aio array */
	os_aio_slot_t*	slot)	/* in: pointer to slot */
{
	ut_ad(array);
	ut_ad(slot);

	os_mutex_enter(array->mutex);

	ut_ad(slot->reserved);
3354

3355 3356 3357 3358 3359 3360 3361 3362
	slot->reserved = FALSE;

	array->n_reserved--;

	if (array->n_reserved == array->n_slots - 1) {
		os_event_set(array->not_full);
	}

3363 3364 3365 3366
	if (array->n_reserved == 0) {
		os_event_set(array->is_empty);
	}

3367
#ifdef WIN_ASYNC_IO
3368
	os_event_reset(slot->event);
3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383
#endif
	os_mutex_exit(array->mutex);
}

/**************************************************************************
Wakes up a simulated aio i/o-handler thread if it has something to do. */
static
void
os_aio_simulated_wake_handler_thread(
/*=================================*/
	ulint	global_segment)	/* in: the number of the segment in the aio
				arrays */
{
	os_aio_array_t*	array;
	os_aio_slot_t*	slot;
unknown's avatar
Merge  
unknown committed
3384
	ulint		segment;
3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402
	ulint		n;
	ulint		i;

	ut_ad(!os_aio_use_native_aio);

	segment = os_aio_get_array_and_local_segment(&array, global_segment);

	n = array->n_slots / array->n_segments;

	/* Look through n slots after the segment * n'th slot */

	os_mutex_enter(array->mutex);

	for (i = 0; i < n; i++) {
		slot = os_aio_array_get_nth_slot(array, i + segment * n);

		if (slot->reserved) {
			/* Found an i/o request */
3403

3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427
			break;
		}
	}

	os_mutex_exit(array->mutex);

	if (i < n) {
		os_event_set(os_aio_segment_wait_events[global_segment]);
	}
}

/**************************************************************************
Wakes up simulated aio i/o-handler threads if they have something to do. */

void
os_aio_simulated_wake_handler_threads(void)
/*=======================================*/
{
	ulint	i;

	if (os_aio_use_native_aio) {
		/* We do not use simulated aio: do nothing */

		return;
3428
	}
unknown's avatar
unknown committed
3429 3430

	os_aio_recommend_sleep_for_read_threads	= FALSE;
3431 3432 3433 3434 3435 3436

	for (i = 0; i < os_aio_n_segments; i++) {
		os_aio_simulated_wake_handler_thread(i);
	}
}

unknown's avatar
unknown committed
3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455
/**************************************************************************
This function can be called if one wants to post a batch of reads and
prefers an i/o-handler thread to handle them all at once later. You must
call os_aio_simulated_wake_handler_threads later to ensure the threads
are not left sleeping! */

void
os_aio_simulated_put_read_threads_to_sleep(void)
/*============================================*/
{
	os_aio_array_t*	array;
	ulint		g;

	os_aio_recommend_sleep_for_read_threads	= TRUE;

	for (g = 0; g < os_aio_n_segments; g++) {
		os_aio_get_array_and_local_segment(&array, g);

		if (array == os_aio_read_array) {
3456

unknown's avatar
unknown committed
3457 3458 3459 3460 3461
			os_event_reset(os_aio_segment_wait_events[g]);
		}
	}
}

3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483
/***********************************************************************
Requests an asynchronous i/o operation. */

ibool
os_aio(
/*===*/
				/* out: TRUE if request was queued
				successfully, FALSE if fail */
	ulint		type,	/* in: OS_FILE_READ or OS_FILE_WRITE */
	ulint		mode,	/* in: OS_AIO_NORMAL, ..., possibly ORed
				to OS_AIO_SIMULATED_WAKE_LATER: the
				last flag advises this function not to wake
				i/o-handler threads, but the caller will
				do the waking explicitly later, in this
				way the caller can post several requests in
				a batch; NOTE that the batch must not be
				so big that it exhausts the slots in aio
				arrays! NOTE that a simulated batch
				may introduce hidden chances of deadlocks,
				because i/os are not actually handled until
				all have been posted: use with great
				caution! */
3484
	const char*	name,	/* in: name of the file or path as a
3485 3486 3487 3488 3489 3490 3491 3492
				null-terminated string */
	os_file_t	file,	/* in: handle to a file */
	void*		buf,	/* in: buffer where to read or from which
				to write */
	ulint		offset,	/* in: least significant 32 bits of file
				offset where to read or write */
	ulint		offset_high, /* in: most significant 32 bits of
				offset */
unknown's avatar
unknown committed
3493
	ulint		n,	/* in: number of bytes to read or write */
unknown's avatar
unknown committed
3494
	fil_node_t*	message1,/* in: messages for the aio handler (these
3495 3496 3497 3498 3499 3500 3501 3502
				can be used to identify a completed aio
				operation); if mode is OS_AIO_SYNC, these
				are ignored */
	void*		message2)
{
	os_aio_array_t*	array;
	os_aio_slot_t*	slot;
#ifdef WIN_ASYNC_IO
unknown's avatar
unknown committed
3503
	ibool		retval;
3504
	BOOL		ret		= TRUE;
3505
	DWORD		len		= (DWORD) n;
unknown's avatar
unknown committed
3506
	struct fil_node_struct * dummy_mess1;
3507
	void*		dummy_mess2;
3508
	ulint		dummy_type;
3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522
#endif
	ulint		err		= 0;
	ibool		retry;
	ulint		wake_later;

	ut_ad(file);
	ut_ad(buf);
	ut_ad(n > 0);
	ut_ad(n % OS_FILE_LOG_BLOCK_SIZE == 0);
	ut_ad(offset % OS_FILE_LOG_BLOCK_SIZE == 0);
	ut_ad(os_aio_validate());

	wake_later = mode & OS_AIO_SIMULATED_WAKE_LATER;
	mode = mode & (~OS_AIO_SIMULATED_WAKE_LATER);
3523

3524 3525
	if (mode == OS_AIO_SYNC
#ifdef WIN_ASYNC_IO
unknown's avatar
unknown committed
3526
	    && !os_aio_use_native_aio
3527
#endif
unknown's avatar
unknown committed
3528
	    ) {
3529 3530 3531 3532 3533 3534 3535 3536
		/* This is actually an ordinary synchronous read or write:
		no need to use an i/o-handler thread. NOTE that if we use
		Windows async i/o, Windows does not allow us to use
		ordinary synchronous os_file_read etc. on the same file,
		therefore we have built a special mechanism for synchronous
		wait in the Windows case. */

		if (type == OS_FILE_READ) {
3537
			return(os_file_read(file, buf, offset,
unknown's avatar
unknown committed
3538
					    offset_high, n));
3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554
		}

		ut_a(type == OS_FILE_WRITE);

		return(os_file_write(name, file, buf, offset, offset_high, n));
	}

try_again:
	if (mode == OS_AIO_NORMAL) {
		if (type == OS_FILE_READ) {
			array = os_aio_read_array;
		} else {
			array = os_aio_write_array;
		}
	} else if (mode == OS_AIO_IBUF) {
		ut_ad(type == OS_FILE_READ);
3555 3556 3557 3558
		/* Reduce probability of deadlock bugs in connection with ibuf:
		do not let the ibuf i/o handler sleep */

		wake_later = FALSE;
3559 3560 3561 3562 3563 3564 3565 3566

		array = os_aio_ibuf_array;
	} else if (mode == OS_AIO_LOG) {

		array = os_aio_log_array;
	} else if (mode == OS_AIO_SYNC) {
		array = os_aio_sync_array;
	} else {
3567
		array = NULL; /* Eliminate compiler warning */
3568 3569
		ut_error;
	}
3570

3571
	slot = os_aio_array_reserve_slot(type, array, message1, message2, file,
unknown's avatar
unknown committed
3572
					 name, buf, offset, offset_high, n);
3573 3574 3575
	if (type == OS_FILE_READ) {
		if (os_aio_use_native_aio) {
#ifdef WIN_ASYNC_IO
3576
			os_n_file_reads++;
unknown's avatar
unknown committed
3577
			os_bytes_read_since_printout += len;
3578

3579
			ret = ReadFile(file, buf, (DWORD)n, &len,
unknown's avatar
unknown committed
3580
				       &(slot->control));
3581 3582 3583
#elif defined(POSIX_ASYNC_IO)
			slot->control.aio_lio_opcode = LIO_READ;
			err = (ulint) aio_read(&(slot->control));
3584
			fprintf(stderr, "Starting POSIX aio read %lu\n", err);
3585 3586 3587
#endif
		} else {
			if (!wake_later) {
3588 3589 3590
				os_aio_simulated_wake_handler_thread(
					os_aio_get_segment_no_from_slot(
						array, slot));
3591 3592 3593 3594 3595
			}
		}
	} else if (type == OS_FILE_WRITE) {
		if (os_aio_use_native_aio) {
#ifdef WIN_ASYNC_IO
3596
			os_n_file_writes++;
3597
			ret = WriteFile(file, buf, (DWORD)n, &len,
unknown's avatar
unknown committed
3598
					&(slot->control));
3599 3600 3601
#elif defined(POSIX_ASYNC_IO)
			slot->control.aio_lio_opcode = LIO_WRITE;
			err = (ulint) aio_write(&(slot->control));
3602
			fprintf(stderr, "Starting POSIX aio write %lu\n", err);
3603 3604 3605
#endif
		} else {
			if (!wake_later) {
3606 3607 3608
				os_aio_simulated_wake_handler_thread(
					os_aio_get_segment_no_from_slot(
						array, slot));
3609 3610 3611 3612 3613 3614 3615 3616 3617
			}
		}
	} else {
		ut_error;
	}

#ifdef WIN_ASYNC_IO
	if (os_aio_use_native_aio) {
		if ((ret && len == n)
unknown's avatar
unknown committed
3618
		    || (!ret && GetLastError() == ERROR_IO_PENDING)) {
3619
			/* aio was queued successfully! */
3620 3621 3622 3623 3624 3625 3626 3627

			if (mode == OS_AIO_SYNC) {
				/* We want a synchronous i/o operation on a
				file where we also use async i/o: in Windows
				we must use the same wait mechanism as for
				async i/o */

				retval = os_aio_windows_handle(ULINT_UNDEFINED,
unknown's avatar
unknown committed
3628 3629 3630 3631
							       slot->pos,
							       &dummy_mess1,
							       &dummy_mess2,
							       &dummy_type);
unknown's avatar
unknown committed
3632

3633 3634
				return(retval);
			}
3635 3636 3637 3638

			return(TRUE);
		}

3639
		err = 1; /* Fall through the next if */
3640 3641 3642 3643 3644 3645 3646 3647 3648 3649
	}
#endif
	if (err == 0) {
		/* aio was queued successfully! */

		return(TRUE);
	}

	os_aio_array_free_slot(array, slot);

3650
	retry = os_file_handle_error(name,
unknown's avatar
unknown committed
3651 3652
				     type == OS_FILE_READ
				     ? "aio read" : "aio write");
3653 3654 3655
	if (retry) {

		goto try_again;
3656
	}
3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682

	return(FALSE);
}

#ifdef WIN_ASYNC_IO
/**************************************************************************
This function is only used in Windows asynchronous i/o.
Waits for an aio operation to complete. This function is used to wait the
for completed requests. The aio array of pending requests is divided
into segments. The thread specifies which segment or slot it wants to wait
for. NOTE: this function will also take care of freeing the aio slot,
therefore no other thread is allowed to do the freeing! */

ibool
os_aio_windows_handle(
/*==================*/
				/* out: TRUE if the aio operation succeeded */
	ulint	segment,	/* in: the number of the segment in the aio
				arrays to wait for; segment 0 is the ibuf
				i/o thread, segment 1 the log i/o thread,
				then follow the non-ibuf read threads, and as
				the last are the non-ibuf write threads; if
				this is ULINT_UNDEFINED, then it means that
				sync aio is used, and this parameter is
				ignored */
	ulint	pos,		/* this parameter is used only in sync aio:
3683
				wait for the aio slot at this position */
unknown's avatar
unknown committed
3684
	fil_node_t**message1,	/* out: the messages passed with the aio
3685 3686 3687 3688
				request; note that also in the case where
				the aio operation failed, these output
				parameters are valid and can be used to
				restart the operation, for example */
3689 3690
	void**	message2,
	ulint*	type)		/* out: OS_FILE_WRITE or ..._READ */
3691
{
3692
	ulint		orig_seg	= segment;
3693 3694 3695 3696 3697 3698 3699
	os_aio_array_t*	array;
	os_aio_slot_t*	slot;
	ulint		n;
	ulint		i;
	ibool		ret_val;
	BOOL		ret;
	DWORD		len;
3700
	BOOL		retry		= FALSE;
3701 3702 3703 3704 3705 3706 3707

	if (segment == ULINT_UNDEFINED) {
		array = os_aio_sync_array;
		segment = 0;
	} else {
		segment = os_aio_get_array_and_local_segment(&array, segment);
	}
3708

3709 3710 3711 3712 3713 3714 3715 3716 3717
	/* NOTE! We only access constant fields in os_aio_array. Therefore
	we do not have to acquire the protecting mutex yet */

	ut_ad(os_aio_validate());
	ut_ad(segment < array->n_segments);

	n = array->n_slots / array->n_segments;

	if (array == os_aio_sync_array) {
3718
		os_event_wait(os_aio_array_get_nth_slot(array, pos)->event);
3719 3720
		i = pos;
	} else {
3721
		srv_set_io_thread_op_info(orig_seg, "wait Windows aio");
3722
		i = os_event_wait_multiple(n,
unknown's avatar
unknown committed
3723 3724
					   (array->native_events)
					   + segment * n);
3725 3726 3727 3728 3729 3730 3731 3732
	}

	os_mutex_enter(array->mutex);

	slot = os_aio_array_get_nth_slot(array, i + segment * n);

	ut_a(slot->reserved);

unknown's avatar
unknown committed
3733
	if (orig_seg != ULINT_UNDEFINED) {
3734
		srv_set_io_thread_op_info(orig_seg,
unknown's avatar
unknown committed
3735
					  "get windows aio return value");
unknown's avatar
unknown committed
3736 3737
	}

3738 3739 3740 3741 3742
	ret = GetOverlappedResult(slot->file, &(slot->control), &len, TRUE);

	*message1 = slot->message1;
	*message2 = slot->message2;

3743 3744
	*type = slot->type;

3745 3746
	if (ret && len == slot->len) {
		ret_val = TRUE;
3747

3748
# ifdef UNIV_DO_FLUSH
unknown's avatar
unknown committed
3749
		if (slot->type == OS_FILE_WRITE
unknown's avatar
unknown committed
3750 3751
		    && !os_do_not_call_flush_at_each_write) {
			ut_a(TRUE == os_file_flush(slot->file));
3752
		}
3753
# endif /* UNIV_DO_FLUSH */
3754 3755 3756
	} else if (os_file_handle_error(slot->name, "Windows aio")) {

		retry = TRUE;
3757
	} else {
3758

3759
		ret_val = FALSE;
3760
	}
3761 3762 3763

	os_mutex_exit(array->mutex);

3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799
	if (retry) {
		/* retry failed read/write operation synchronously.
		No need to hold array->mutex. */

		switch (slot->type) {
		case OS_FILE_WRITE:
			ret = WriteFile(slot->file, slot->buf,
					slot->len, &len,
					&(slot->control));

			break;
		case OS_FILE_READ:
			ret = ReadFile(slot->file, slot->buf,
				       slot->len, &len,
				       &(slot->control));

			break;
		default:
			ut_error;
		}

		if (!ret && GetLastError() == ERROR_IO_PENDING) {
			/* aio was queued successfully!
			We want a synchronous i/o operation on a
			file where we also use async i/o: in Windows
			we must use the same wait mechanism as for
			async i/o */

			ret = GetOverlappedResult(slot->file,
						  &(slot->control),
						  &len, TRUE);
		}

		ret_val = ret && len == slot->len;
	}

3800
	os_aio_array_free_slot(array, slot);
3801

3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816
	return(ret_val);
}
#endif

#ifdef POSIX_ASYNC_IO

/**************************************************************************
This function is only used in Posix asynchronous i/o. Waits for an aio
operation to complete. */

ibool
os_aio_posix_handle(
/*================*/
				/* out: TRUE if the aio operation succeeded */
	ulint	array_no,	/* in: array number 0 - 3 */
unknown's avatar
unknown committed
3817
	fil_node_t**message1,	/* out: the messages passed with the aio
3818 3819 3820 3821 3822 3823 3824 3825 3826 3827
				request; note that also in the case where
				the aio operation failed, these output
				parameters are valid and can be used to
				restart the operation, for example */
	void**	message2)
{
	os_aio_array_t*	array;
	os_aio_slot_t*	slot;
	siginfo_t	info;
	sigset_t	sigset;
3828 3829
	sigset_t	proc_sigset;
	sigset_t	thr_sigset;
3830
	int		ret;
3831 3832 3833 3834
	int		i;
	int		sig;

	sigemptyset(&sigset);
3835 3836 3837 3838
	sigaddset(&sigset, SIGRTMIN + 1 + array_no);

	pthread_sigmask(SIG_UNBLOCK, &sigset, NULL);

unknown's avatar
unknown committed
3839
#if 0
3840 3841 3842 3843
	sigprocmask(0, NULL, &proc_sigset);
	pthread_sigmask(0, NULL, &thr_sigset);

	for (i = 32 ; i < 40; i++) {
3844
		fprintf(stderr, "%lu : %lu %lu\n", (ulint)i,
unknown's avatar
unknown committed
3845 3846
			(ulint) sigismember(&proc_sigset, i),
			(ulint) sigismember(&thr_sigset, i));
3847
	}
unknown's avatar
unknown committed
3848
#endif
3849 3850 3851 3852 3853

	ret = sigwaitinfo(&sigset, &info);

	if (sig != SIGRTMIN + 1 + array_no) {

3854
		ut_error;
3855

3856 3857
		return(FALSE);
	}
3858

3859
	fputs("Handling POSIX aio\n", stderr);
3860 3861 3862 3863 3864 3865 3866 3867 3868 3869 3870 3871

	array = os_aio_get_array_from_no(array_no);

	os_mutex_enter(array->mutex);

	slot = info.si_value.sival_ptr;

	ut_a(slot->reserved);

	*message1 = slot->message1;
	*message2 = slot->message2;

3872
# ifdef UNIV_DO_FLUSH
unknown's avatar
unknown committed
3873
	if (slot->type == OS_FILE_WRITE
unknown's avatar
unknown committed
3874
	    && !os_do_not_call_flush_at_each_write) {
3875 3876
		ut_a(TRUE == os_file_flush(slot->file));
	}
3877
# endif /* UNIV_DO_FLUSH */
3878

3879 3880 3881
	os_mutex_exit(array->mutex);

	os_aio_array_free_slot(array, slot);
3882

3883 3884 3885 3886
	return(TRUE);
}
#endif

unknown's avatar
unknown committed
3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898 3899 3900
/**************************************************************************
Do a 'last millisecond' check that the page end is sensible;
reported page checksum errors from Linux seem to wipe over the page end. */
static
void
os_file_check_page_trailers(
/*========================*/
	byte*	combined_buf,	/* in: combined write buffer */
	ulint	total_len)	/* in: size of combined_buf, in bytes
				(a multiple of UNIV_PAGE_SIZE) */
{
	ulint	len;

	for (len = 0; len + UNIV_PAGE_SIZE <= total_len;
unknown's avatar
unknown committed
3901
	     len += UNIV_PAGE_SIZE) {
unknown's avatar
unknown committed
3902 3903
		byte*	buf = combined_buf + len;

unknown's avatar
unknown committed
3904 3905 3906 3907
		if (UNIV_UNLIKELY
		    (memcmp(buf + (FIL_PAGE_LSN + 4),
			    buf + (UNIV_PAGE_SIZE
				   - FIL_PAGE_END_LSN_OLD_CHKSUM + 4), 4))) {
unknown's avatar
unknown committed
3908 3909
		    	ut_print_timestamp(stderr);
		    	fprintf(stderr,
unknown's avatar
unknown committed
3910 3911 3912 3913 3914
				"  InnoDB: ERROR: The page to be written"
				" seems corrupt!\n"
				"InnoDB: Writing a block of %lu bytes,"
				" currently at offset %lu\n",
				(ulong)total_len, (ulong)len);
unknown's avatar
unknown committed
3915 3916
			buf_page_print(buf);
		    	fprintf(stderr,
unknown's avatar
unknown committed
3917 3918
				"InnoDB: ERROR: The page to be written"
				" seems corrupt!\n");
unknown's avatar
unknown committed
3919 3920 3921 3922
		}
	}
}

3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935
/**************************************************************************
Does simulated aio. This function should be called by an i/o-handler
thread. */

ibool
os_aio_simulated_handle(
/*====================*/
				/* out: TRUE if the aio operation succeeded */
	ulint	global_segment,	/* in: the number of the segment in the aio
				arrays to wait for; segment 0 is the ibuf
				i/o thread, segment 1 the log i/o thread,
				then follow the non-ibuf read threads, and as
				the last are the non-ibuf write threads */
unknown's avatar
unknown committed
3936
	fil_node_t**message1,	/* out: the messages passed with the aio
3937 3938 3939 3940
				request; note that also in the case where
				the aio operation failed, these output
				parameters are valid and can be used to
				restart the operation, for example */
3941 3942
	void**	message2,
	ulint*	type)		/* out: OS_FILE_WRITE or ..._READ */
3943 3944 3945 3946 3947 3948 3949 3950 3951 3952
{
	os_aio_array_t*	array;
	ulint		segment;
	os_aio_slot_t*	slot;
	os_aio_slot_t*	slot2;
	os_aio_slot_t*	consecutive_ios[OS_AIO_MERGE_N_CONSECUTIVE];
	ulint		n_consecutive;
	ulint		total_len;
	ulint		offs;
	ulint		lowest_offset;
unknown's avatar
unknown committed
3953 3954
	ulint		biggest_age;
	ulint		age;
3955
	byte*		combined_buf;
unknown's avatar
unknown committed
3956
	byte*		combined_buf2;
3957 3958 3959
	ibool		ret;
	ulint		n;
	ulint		i;
3960

3961 3962 3963
	/* Fix compiler warning */
	*consecutive_ios = NULL;

3964
	segment = os_aio_get_array_and_local_segment(&array, global_segment);
3965

3966 3967 3968 3969
restart:
	/* NOTE! We only access constant fields in os_aio_array. Therefore
	we do not have to acquire the protecting mutex yet */

unknown's avatar
unknown committed
3970
	srv_set_io_thread_op_info(global_segment,
unknown's avatar
unknown committed
3971
				  "looking for i/o requests (a)");
3972 3973 3974 3975 3976 3977 3978
	ut_ad(os_aio_validate());
	ut_ad(segment < array->n_segments);

	n = array->n_slots / array->n_segments;

	/* Look through n slots after the segment * n'th slot */

unknown's avatar
unknown committed
3979
	if (array == os_aio_read_array
unknown's avatar
unknown committed
3980
	    && os_aio_recommend_sleep_for_read_threads) {
unknown's avatar
unknown committed
3981 3982 3983 3984 3985 3986

		/* Give other threads chance to add several i/os to the array
		at once. */

		goto recommended_sleep;
	}
3987

3988 3989
	os_mutex_enter(array->mutex);

unknown's avatar
unknown committed
3990
	srv_set_io_thread_op_info(global_segment,
unknown's avatar
unknown committed
3991
				  "looking for i/o requests (b)");
unknown's avatar
unknown committed
3992

3993 3994
	/* Check if there is a slot for which the i/o has already been
	done */
3995

3996 3997 3998 3999 4000
	for (i = 0; i < n; i++) {
		slot = os_aio_array_get_nth_slot(array, i + segment * n);

		if (slot->reserved && slot->io_already_done) {

unknown's avatar
unknown committed
4001 4002
			if (os_aio_print_debug) {
				fprintf(stderr,
unknown's avatar
unknown committed
4003 4004 4005
					"InnoDB: i/o for slot %lu"
					" already done, returning\n",
					(ulong) i);
unknown's avatar
unknown committed
4006 4007
			}

4008
			ret = TRUE;
4009

4010 4011 4012 4013 4014 4015
			goto slot_io_done;
		}
	}

	n_consecutive = 0;

unknown's avatar
unknown committed
4016 4017 4018
	/* If there are at least 2 seconds old requests, then pick the oldest
	one to prevent starvation. If several requests have the same age,
	then pick the one at the lowest offset. */
4019

unknown's avatar
unknown committed
4020
	biggest_age = 0;
4021
	lowest_offset = ULINT_MAX;
unknown's avatar
unknown committed
4022

4023 4024 4025
	for (i = 0; i < n; i++) {
		slot = os_aio_array_get_nth_slot(array, i + segment * n);

unknown's avatar
unknown committed
4026
		if (slot->reserved) {
4027
			age = (ulint)difftime(time(NULL),
unknown's avatar
unknown committed
4028
					      slot->reservation_time);
4029

unknown's avatar
unknown committed
4030
			if ((age >= 2 && age > biggest_age)
unknown's avatar
unknown committed
4031 4032
			    || (age >= 2 && age == biggest_age
				&& slot->offset < lowest_offset)) {
4033

4034
				/* Found an i/o request */
unknown's avatar
unknown committed
4035
				consecutive_ios[0] = slot;
4036

unknown's avatar
unknown committed
4037 4038 4039 4040 4041 4042 4043 4044 4045
				n_consecutive = 1;

				biggest_age = age;
				lowest_offset = slot->offset;
			}
		}
	}

	if (n_consecutive == 0) {
4046
		/* There were no old requests. Look for an i/o request at the
unknown's avatar
unknown committed
4047 4048 4049 4050
		lowest offset in the array (we ignore the high 32 bits of the
		offset in these heuristics) */

		lowest_offset = ULINT_MAX;
4051

unknown's avatar
unknown committed
4052
		for (i = 0; i < n; i++) {
4053
			slot = os_aio_array_get_nth_slot(array,
unknown's avatar
unknown committed
4054
							 i + segment * n);
4055

unknown's avatar
unknown committed
4056
			if (slot->reserved && slot->offset < lowest_offset) {
4057

4058
				/* Found an i/o request */
unknown's avatar
unknown committed
4059 4060 4061 4062 4063 4064
				consecutive_ios[0] = slot;

				n_consecutive = 1;

				lowest_offset = slot->offset;
			}
4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078
		}
	}

	if (n_consecutive == 0) {

		/* No i/o requested at the moment */

		goto wait_for_io;
	}

	slot = consecutive_ios[0];

	/* Check if there are several consecutive blocks to read or write */

4079
consecutive_loop:
4080 4081 4082 4083
	for (i = 0; i < n; i++) {
		slot2 = os_aio_array_get_nth_slot(array, i + segment * n);

		if (slot2->reserved && slot2 != slot
unknown's avatar
unknown committed
4084 4085 4086 4087 4088 4089
		    && slot2->offset == slot->offset + slot->len
		    /* check that sum does not wrap over */
		    && slot->offset + slot->len > slot->offset
		    && slot2->offset_high == slot->offset_high
		    && slot2->type == slot->type
		    && slot2->file == slot->file) {
4090 4091 4092 4093 4094 4095 4096 4097 4098

			/* Found a consecutive i/o request */

			consecutive_ios[n_consecutive] = slot2;
			n_consecutive++;

			slot = slot2;

			if (n_consecutive < OS_AIO_MERGE_N_CONSECUTIVE) {
4099

4100 4101 4102 4103 4104 4105 4106
				goto consecutive_loop;
			} else {
				break;
			}
		}
	}

unknown's avatar
unknown committed
4107 4108
	srv_set_io_thread_op_info(global_segment, "consecutive i/o requests");

4109 4110 4111 4112 4113 4114
	/* We have now collected n_consecutive i/o requests in the array;
	allocate a single buffer which can hold all data, and perform the
	i/o */

	total_len = 0;
	slot = consecutive_ios[0];
4115

4116 4117 4118 4119 4120 4121 4122
	for (i = 0; i < n_consecutive; i++) {
		total_len += consecutive_ios[i]->len;
	}

	if (n_consecutive == 1) {
		/* We can use the buffer of the i/o request */
		combined_buf = slot->buf;
4123
		combined_buf2 = NULL;
4124
	} else {
unknown's avatar
unknown committed
4125
		combined_buf2 = ut_malloc(total_len + UNIV_PAGE_SIZE);
4126

unknown's avatar
unknown committed
4127 4128 4129
		ut_a(combined_buf2);

		combined_buf = ut_align(combined_buf2, UNIV_PAGE_SIZE);
4130
	}
4131

4132 4133 4134 4135 4136 4137 4138 4139 4140
	/* We release the array mutex for the time of the i/o: NOTE that
	this assumes that there is just one i/o-handler thread serving
	a single segment of slots! */

	os_mutex_exit(array->mutex);

	if (slot->type == OS_FILE_WRITE && n_consecutive > 1) {
		/* Copy the buffers to the combined buffer */
		offs = 0;
4141

4142 4143 4144
		for (i = 0; i < n_consecutive; i++) {

			ut_memcpy(combined_buf + offs, consecutive_ios[i]->buf,
unknown's avatar
unknown committed
4145
				  consecutive_ios[i]->len);
4146 4147 4148
			offs += consecutive_ios[i]->len;
		}
	}
4149

4150
	srv_set_io_thread_op_info(global_segment, "doing file i/o");
4151

unknown's avatar
unknown committed
4152 4153
	if (os_aio_print_debug) {
		fprintf(stderr,
unknown's avatar
unknown committed
4154 4155
			"InnoDB: doing i/o of type %lu at offset %lu %lu,"
			" length %lu\n",
4156 4157
			(ulong) slot->type, (ulong) slot->offset_high,
			(ulong) slot->offset, (ulong) total_len);
unknown's avatar
unknown committed
4158 4159
	}

4160 4161
	/* Do the i/o with ordinary, synchronous i/o functions: */
	if (slot->type == OS_FILE_WRITE) {
unknown's avatar
unknown committed
4162
		if (array == os_aio_write_array) {
unknown's avatar
unknown committed
4163
			if ((total_len % UNIV_PAGE_SIZE != 0)
unknown's avatar
unknown committed
4164
			    || (slot->offset % UNIV_PAGE_SIZE != 0)) {
unknown's avatar
unknown committed
4165
				fprintf(stderr,
unknown's avatar
unknown committed
4166 4167
					"InnoDB: Error: trying a displaced"
					" write to %s %lu %lu, len %lu\n",
4168 4169 4170
					slot->name, (ulong) slot->offset_high,
					(ulong) slot->offset,
					(ulong) total_len);
4171
				ut_error;
unknown's avatar
unknown committed
4172
			}
4173

unknown's avatar
unknown committed
4174
			os_file_check_page_trailers(combined_buf, total_len);
unknown's avatar
unknown committed
4175
		}
4176

4177
		ret = os_file_write(slot->name, slot->file, combined_buf,
unknown's avatar
unknown committed
4178 4179
				    slot->offset, slot->offset_high,
				    total_len);
unknown's avatar
unknown committed
4180 4181 4182 4183

		if (array == os_aio_write_array) {
			os_file_check_page_trailers(combined_buf, total_len);
		}
4184 4185
	} else {
		ret = os_file_read(slot->file, combined_buf,
unknown's avatar
unknown committed
4186
				   slot->offset, slot->offset_high, total_len);
4187 4188 4189
	}

	ut_a(ret);
4190
	srv_set_io_thread_op_info(global_segment, "file i/o done");
4191

unknown's avatar
unknown committed
4192 4193 4194 4195 4196
#if 0
	fprintf(stderr,
		"aio: %lu consecutive %lu:th segment, first offs %lu blocks\n",
		n_consecutive, global_segment, slot->offset / UNIV_PAGE_SIZE);
#endif
4197 4198 4199 4200

	if (slot->type == OS_FILE_READ && n_consecutive > 1) {
		/* Copy the combined buffer to individual buffers */
		offs = 0;
4201

4202 4203
		for (i = 0; i < n_consecutive; i++) {

4204
			ut_memcpy(consecutive_ios[i]->buf, combined_buf + offs,
unknown's avatar
unknown committed
4205
				  consecutive_ios[i]->len);
4206 4207 4208 4209
			offs += consecutive_ios[i]->len;
		}
	}

4210
	if (combined_buf2) {
unknown's avatar
unknown committed
4211
		ut_free(combined_buf2);
4212 4213 4214 4215 4216 4217 4218 4219 4220 4221 4222 4223 4224
	}

	os_mutex_enter(array->mutex);

	/* Mark the i/os done in slots */

	for (i = 0; i < n_consecutive; i++) {
		consecutive_ios[i]->io_already_done = TRUE;
	}

	/* We return the messages for the first slot now, and if there were
	several slots, the messages will be returned with subsequent calls
	of this function */
4225

4226 4227 4228 4229 4230 4231 4232
slot_io_done:

	ut_a(slot->reserved);

	*message1 = slot->message1;
	*message2 = slot->message2;

4233 4234
	*type = slot->type;

4235 4236 4237
	os_mutex_exit(array->mutex);

	os_aio_array_free_slot(array, slot);
4238

4239 4240 4241
	return(ret);

wait_for_io:
unknown's avatar
unknown committed
4242 4243
	srv_set_io_thread_op_info(global_segment, "resetting wait event");

4244 4245
	/* We wait here until there again can be i/os in the segment
	of this thread */
4246

4247 4248 4249 4250
	os_event_reset(os_aio_segment_wait_events[global_segment]);

	os_mutex_exit(array->mutex);

unknown's avatar
unknown committed
4251
recommended_sleep:
4252
	srv_set_io_thread_op_info(global_segment, "waiting for i/o request");
4253

4254 4255
	os_event_wait(os_aio_segment_wait_events[global_segment]);

unknown's avatar
unknown committed
4256 4257
	if (os_aio_print_debug) {
		fprintf(stderr,
unknown's avatar
unknown committed
4258 4259
			"InnoDB: i/o handler thread for i/o"
			" segment %lu wakes up\n",
4260
			(ulong) global_segment);
unknown's avatar
unknown committed
4261
	}
4262

4263 4264 4265 4266 4267 4268 4269 4270 4271 4272 4273 4274 4275 4276 4277
	goto restart;
}

/**************************************************************************
Validates the consistency of an aio array. */
static
ibool
os_aio_array_validate(
/*==================*/
				/* out: TRUE if ok */
	os_aio_array_t*	array)	/* in: aio wait array */
{
	os_aio_slot_t*	slot;
	ulint		n_reserved	= 0;
	ulint		i;
4278

4279 4280 4281 4282 4283 4284
	ut_a(array);

	os_mutex_enter(array->mutex);

	ut_a(array->n_slots > 0);
	ut_a(array->n_segments > 0);
4285

4286 4287
	for (i = 0; i < array->n_slots; i++) {
		slot = os_aio_array_get_nth_slot(array, i);
4288

4289 4290 4291 4292 4293 4294 4295 4296 4297 4298 4299 4300 4301 4302 4303 4304 4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322
		if (slot->reserved) {
			n_reserved++;
			ut_a(slot->len > 0);
		}
	}

	ut_a(array->n_reserved == n_reserved);

	os_mutex_exit(array->mutex);

	return(TRUE);
}

/**************************************************************************
Validates the consistency the aio system. */

ibool
os_aio_validate(void)
/*=================*/
				/* out: TRUE if ok */
{
	os_aio_array_validate(os_aio_read_array);
	os_aio_array_validate(os_aio_write_array);
	os_aio_array_validate(os_aio_ibuf_array);
	os_aio_array_validate(os_aio_log_array);
	os_aio_array_validate(os_aio_sync_array);

	return(TRUE);
}

/**************************************************************************
Prints info of the aio arrays. */

void
unknown's avatar
unknown committed
4323 4324
os_aio_print(
/*=========*/
4325
	FILE*	file)	/* in: file where to print */
4326 4327 4328 4329
{
	os_aio_array_t*	array;
	os_aio_slot_t*	slot;
	ulint		n_reserved;
4330 4331
	time_t		current_time;
	double		time_elapsed;
unknown's avatar
unknown committed
4332
	double		avg_bytes_read;
4333
	ulint		i;
4334

4335
	for (i = 0; i < srv_n_file_io_threads; i++) {
unknown's avatar
unknown committed
4336
		fprintf(file, "I/O thread %lu state: %s (%s)", (ulong) i,
unknown's avatar
unknown committed
4337 4338
			srv_io_thread_op_info[i],
			srv_io_thread_function[i]);
unknown's avatar
unknown committed
4339

unknown's avatar
unknown committed
4340
#ifndef __WIN__
4341
		if (os_aio_segment_wait_events[i]->is_set) {
unknown's avatar
unknown committed
4342 4343
			fprintf(file, " ev set");
		}
unknown's avatar
unknown committed
4344
#endif
4345

unknown's avatar
unknown committed
4346
		fprintf(file, "\n");
4347 4348
	}

4349
	fputs("Pending normal aio reads:", file);
4350

4351 4352 4353
	array = os_aio_read_array;
loop:
	ut_a(array);
4354

4355 4356 4357 4358
	os_mutex_enter(array->mutex);

	ut_a(array->n_slots > 0);
	ut_a(array->n_segments > 0);
4359

4360 4361 4362 4363
	n_reserved = 0;

	for (i = 0; i < array->n_slots; i++) {
		slot = os_aio_array_get_nth_slot(array, i);
4364

4365 4366
		if (slot->reserved) {
			n_reserved++;
unknown's avatar
unknown committed
4367 4368 4369 4370 4371
#if 0
			fprintf(stderr, "Reserved slot, messages %p %p\n",
				(void*) slot->message1,
				(void*) slot->message2);
#endif
4372
			ut_a(slot->len > 0);
4373 4374 4375 4376 4377
		}
	}

	ut_a(array->n_reserved == n_reserved);

unknown's avatar
unknown committed
4378
	fprintf(file, " %lu", (ulong) n_reserved);
4379

4380 4381 4382
	os_mutex_exit(array->mutex);

	if (array == os_aio_read_array) {
4383
		fputs(", aio writes:", file);
4384

4385 4386 4387 4388 4389 4390
		array = os_aio_write_array;

		goto loop;
	}

	if (array == os_aio_write_array) {
4391
		fputs(",\n ibuf aio reads:", file);
4392 4393 4394 4395 4396 4397
		array = os_aio_ibuf_array;

		goto loop;
	}

	if (array == os_aio_ibuf_array) {
4398
		fputs(", log i/o's:", file);
4399 4400 4401 4402 4403 4404
		array = os_aio_log_array;

		goto loop;
	}

	if (array == os_aio_log_array) {
4405
		fputs(", sync i/o's:", file);
4406 4407 4408 4409
		array = os_aio_sync_array;

		goto loop;
	}
4410

4411
	putc('\n', file);
4412
	current_time = time(NULL);
unknown's avatar
unknown committed
4413
	time_elapsed = 0.001 + difftime(current_time, os_last_printout);
4414

4415 4416
	fprintf(file,
		"Pending flushes (fsync) log: %lu; buffer pool: %lu\n"
unknown's avatar
unknown committed
4417
		"%lu OS file reads, %lu OS file writes, %lu OS fsyncs\n",
unknown's avatar
unknown committed
4418 4419
		(ulong) fil_n_pending_log_flushes,
		(ulong) fil_n_pending_tablespace_flushes,
4420 4421
		(ulong) os_n_file_reads, (ulong) os_n_file_writes,
		(ulong) os_n_fsyncs);
unknown's avatar
unknown committed
4422

unknown's avatar
unknown committed
4423
	if (os_file_n_pending_preads != 0 || os_file_n_pending_pwrites != 0) {
4424
		fprintf(file,
4425 4426 4427
			"%lu pending preads, %lu pending pwrites\n",
			(ulong) os_file_n_pending_preads,
			(ulong) os_file_n_pending_pwrites);
unknown's avatar
unknown committed
4428 4429
	}

unknown's avatar
unknown committed
4430 4431 4432
	if (os_n_file_reads == os_n_file_reads_old) {
		avg_bytes_read = 0.0;
	} else {
unknown's avatar
unknown committed
4433 4434
		avg_bytes_read = (double) os_bytes_read_since_printout
			/ (os_n_file_reads - os_n_file_reads_old);
unknown's avatar
unknown committed
4435 4436
	}

4437
	fprintf(file,
unknown's avatar
unknown committed
4438 4439
		"%.2f reads/s, %lu avg bytes/read,"
		" %.2f writes/s, %.2f fsyncs/s\n",
4440 4441
		(os_n_file_reads - os_n_file_reads_old)
		/ time_elapsed,
4442
		(ulong)avg_bytes_read,
4443 4444 4445 4446 4447 4448 4449 4450
		(os_n_file_writes - os_n_file_writes_old)
		/ time_elapsed,
		(os_n_fsyncs - os_n_fsyncs_old)
		/ time_elapsed);

	os_n_file_reads_old = os_n_file_reads;
	os_n_file_writes_old = os_n_file_writes;
	os_n_fsyncs_old = os_n_fsyncs;
unknown's avatar
unknown committed
4451
	os_bytes_read_since_printout = 0;
4452

4453
	os_last_printout = current_time;
4454 4455
}

unknown's avatar
unknown committed
4456 4457 4458 4459 4460 4461 4462 4463 4464 4465 4466
/**************************************************************************
Refreshes the statistics used to print per-second averages. */

void
os_aio_refresh_stats(void)
/*======================*/
{
	os_n_file_reads_old = os_n_file_reads;
	os_n_file_writes_old = os_n_file_writes;
	os_n_fsyncs_old = os_n_fsyncs;
	os_bytes_read_since_printout = 0;
4467

unknown's avatar
unknown committed
4468 4469 4470
	os_last_printout = time(NULL);
}

unknown's avatar
unknown committed
4471
#ifdef UNIV_DEBUG
4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482
/**************************************************************************
Checks that all slots in the system have been freed, that is, there are
no pending io operations. */

ibool
os_aio_all_slots_free(void)
/*=======================*/
				/* out: TRUE if all free */
{
	os_aio_array_t*	array;
	ulint		n_res	= 0;
4483

4484 4485 4486 4487
	array = os_aio_read_array;

	os_mutex_enter(array->mutex);

4488 4489
	n_res += array->n_reserved;

4490 4491 4492 4493 4494 4495
	os_mutex_exit(array->mutex);

	array = os_aio_write_array;

	os_mutex_enter(array->mutex);

4496 4497
	n_res += array->n_reserved;

4498 4499 4500 4501 4502 4503
	os_mutex_exit(array->mutex);

	array = os_aio_ibuf_array;

	os_mutex_enter(array->mutex);

4504 4505
	n_res += array->n_reserved;

4506 4507 4508 4509 4510 4511
	os_mutex_exit(array->mutex);

	array = os_aio_log_array;

	os_mutex_enter(array->mutex);

4512 4513
	n_res += array->n_reserved;

4514 4515 4516 4517 4518 4519
	os_mutex_exit(array->mutex);

	array = os_aio_sync_array;

	os_mutex_enter(array->mutex);

4520 4521
	n_res += array->n_reserved;

4522 4523 4524 4525 4526 4527 4528 4529 4530
	os_mutex_exit(array->mutex);

	if (n_res == 0) {

		return(TRUE);
	}

	return(FALSE);
}
unknown's avatar
unknown committed
4531
#endif /* UNIV_DEBUG */