os0file.c 91.9 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

18
#undef HAVE_FDATASYNC
19 20 21 22 23 24 25

#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
26
/* This specifies the file permissions InnoDB uses when it creates files in
unknown's avatar
unknown committed
27 28 29 30 31 32 33 34 35
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

unknown's avatar
unknown committed
36
/* If the following is set to TRUE, we do not call os_file_flush in every
unknown's avatar
unknown committed
37
os_file_write. We can set this TRUE when the doublewrite buffer is used. */
unknown's avatar
unknown committed
38 39
ibool	os_do_not_call_flush_at_each_write	= FALSE;

40 41 42 43 44 45
/* 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
46
#define OS_AIO_MERGE_N_CONSECUTIVE	64
47 48 49 50 51 52 53

/* 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
54 55
ibool	os_aio_print_debug	= FALSE;

56 57 58 59 60 61 62 63
/* 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
64
	time_t		reservation_time;/* time when reserved */
65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84
	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 */
	char*		name;		/* file name or path */
	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 */
	void*		message1;	/* message which is given by the */
	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
85 86
        os_event_t	event;		/* event object we need in the
					OVERLAPPED struct */
87 88 89 90 91 92 93 94 95 96 97 98 99
	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 */
100
	os_event_t	not_full; /* The event which is set to the signaled
101 102
				  state when there is space in the aio
				  outside the ibuf segment */
103 104 105
	os_event_t	is_empty; /* The event which is set to the signaled
				  state when there are no pending i/os
				  in this array */
106 107 108 109 110 111 112 113
	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 */
	os_aio_slot_t* 	slots;	  /* Pointer to the slots in the array */
114 115 116 117 118 119
#ifdef __WIN__
	os_native_event_t* native_events;	 
				  /* 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
120
				  Windows */
121
#endif
122 123 124 125 126 127 128 129 130 131 132 133 134 135 136
};

/* 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. */
os_aio_array_t*	os_aio_read_array	= NULL;
os_aio_array_t*	os_aio_write_array	= NULL;
os_aio_array_t*	os_aio_ibuf_array	= NULL;
os_aio_array_t*	os_aio_log_array	= NULL;
os_aio_array_t*	os_aio_sync_array	= NULL;

ulint	os_aio_n_segments	= ULINT_UNDEFINED;

unknown's avatar
unknown committed
137 138 139 140
/* If the following is TRUE, read i/o handler threads try to
wait until a batch of new read requests have been posted */
ibool	os_aio_recommend_sleep_for_read_threads	= FALSE;

141
ulint	os_n_file_reads		= 0;
unknown's avatar
unknown committed
142
ulint	os_bytes_read_since_printout = 0;
143 144 145 146 147 148 149
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
150 151
ibool	os_has_said_disk_full	= FALSE;

unknown's avatar
unknown committed
152 153 154 155 156 157
/* The mutex protecting the following counts of pending pread and pwrite
operations */
os_mutex_t os_file_count_mutex;
ulint	os_file_n_pending_preads  = 0;
ulint	os_file_n_pending_pwrites = 0;

158 159 160 161 162 163
/***************************************************************************
Gets the operating system version. Currently works only on Windows. */

ulint
os_get_os_version(void)
/*===================*/
unknown's avatar
unknown committed
164
                  /* out: OS_WIN95, OS_WIN31, OS_WINNT, OS_WIN2000 */
165 166
{
#ifdef __WIN__
167 168 169 170 171 172 173 174 175 176 177
  	OSVERSIONINFO     os_info;

  	os_info.dwOSVersionInfoSize = sizeof(OSVERSIONINFO);

	ut_a(GetVersionEx(&os_info));

  	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
178 179 180 181 182
		if (os_info.dwMajorVersion <= 4) {
    			return(OS_WINNT);
    		} else {
			return(OS_WIN2000);
    		}
183 184 185 186
  	} else {
    		ut_error;
    		return(0);
  	}
187
#else
188
  	ut_error;
189

190
  	return(0);
191 192 193
#endif
}

194 195 196 197 198 199 200
/***************************************************************************
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
201 202 203 204 205 206
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 */
207 208 209 210 211 212 213
{
	ulint	err;

#ifdef __WIN__

	err = (ulint) GetLastError();

unknown's avatar
unknown committed
214 215
	if (report_all_errors
	    || (err != ERROR_DISK_FULL && err != ERROR_FILE_EXISTS)) {
unknown's avatar
unknown committed
216

unknown's avatar
unknown committed
217 218
		ut_print_timestamp(stderr);
	     	fprintf(stderr,
219
  "  InnoDB: Operating system error number %lu in a file operation.\n", (ulong) err);
unknown's avatar
unknown committed
220

unknown's avatar
unknown committed
221
		if (err == ERROR_PATH_NOT_FOUND) {
unknown's avatar
unknown committed
222 223 224 225 226 227 228 229
			fprintf(stderr,
  "InnoDB: The error means the system cannot find the path specified.\n");

			if (srv_is_being_started) {
				fprintf(stderr,
  "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
230
		} else if (err == ERROR_ACCESS_DENIED) {
unknown's avatar
unknown committed
231
			fprintf(stderr,
unknown's avatar
unknown committed
232 233 234
  "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"); 
unknown's avatar
unknown committed
235
		} else {
unknown's avatar
unknown committed
236
			fprintf(stderr,
237
  "InnoDB: See section 13.2 at http://www.innodb.com/ibman.php\n"
unknown's avatar
unknown committed
238
  "InnoDB: about operating system error numbers.\n");
unknown's avatar
unknown committed
239
		}
240
	}
241

unknown's avatar
unknown committed
242 243
	fflush(stderr);

244 245 246 247 248 249 250 251 252 253
	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);
	} else {
		return(100 + err);
	}
#else
254 255
	err = (ulint) errno;

unknown's avatar
unknown committed
256 257
	if (report_all_errors
	    || (err != ENOSPC && err != EEXIST)) {
unknown's avatar
unknown committed
258

unknown's avatar
unknown committed
259
		ut_print_timestamp(stderr);
unknown's avatar
unknown committed
260
	     	fprintf(stderr,
unknown's avatar
unknown committed
261
  "  InnoDB: Operating system error number %lu in a file operation.\n", (ulong) err);
unknown's avatar
unknown committed
262

unknown's avatar
unknown committed
263
		if (err == ENOENT) {
unknown's avatar
unknown committed
264 265 266 267 268 269 270 271
			fprintf(stderr,
  "InnoDB: The error means the system cannot find the path specified.\n");
			
			if (srv_is_being_started) {
				fprintf(stderr,
  "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
272
		} else if (err == EACCES) {
unknown's avatar
unknown committed
273
			fprintf(stderr,
unknown's avatar
unknown committed
274 275
  "InnoDB: The error means mysqld does not have the access rights to\n"
  "InnoDB: the directory.\n");
unknown's avatar
unknown committed
276
		} else {
unknown's avatar
unknown committed
277
			if (strerror((int)err) != NULL) {
unknown's avatar
unknown committed
278 279
				fprintf(stderr,
  "InnoDB: Error number %lu means '%s'.\n", err, strerror((int)err));
unknown's avatar
unknown committed
280
			}
unknown's avatar
unknown committed
281

unknown's avatar
unknown committed
282
			fprintf(stderr,
283
  "InnoDB: See also section 13.2 at http://www.innodb.com/ibman.php\n"
unknown's avatar
unknown committed
284
  "InnoDB: about operating system error numbers.\n");
unknown's avatar
unknown committed
285
		}
286
	}
287

unknown's avatar
unknown committed
288 289
	fflush(stderr);

290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306
	if (err == ENOSPC ) {
		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);
	} else {
		return(100 + err);
	}
#endif
}

/********************************************************************
unknown's avatar
unknown committed
307
Does error handling when a file operation fails. */
308 309 310 311
static
ibool
os_file_handle_error(
/*=================*/
unknown's avatar
unknown committed
312 313
				/* out: TRUE if we should retry the
				operation */
314
	os_file_t	file,	/* in: file pointer */
unknown's avatar
unknown committed
315
	char*		name,	/* in: name of a file or NULL */
unknown's avatar
unknown committed
316
	const char*	operation)/* in: operation */
317 318 319
{
	ulint	err;

320 321
	UT_NOT_USED(file);

unknown's avatar
unknown committed
322
	err = os_file_get_last_error(FALSE);
323 324
	
	if (err == OS_FILE_DISK_FULL) {
unknown's avatar
unknown committed
325 326 327 328 329 330 331
		/* We only print a warning about disk full once */

		if (os_has_said_disk_full) {

			return(FALSE);
		}
	
332
		if (name) {
unknown's avatar
unknown committed
333 334 335
			ut_print_timestamp(stderr);
			fprintf(stderr,
	"  InnoDB: Encountered a problem with file %s\n", name);
336
		}
unknown's avatar
unknown committed
337 338

		ut_print_timestamp(stderr);
unknown's avatar
unknown committed
339
	        fprintf(stderr,
unknown's avatar
unknown committed
340
	"  InnoDB: Disk is full. Try to clean the disk to free space.\n");
341

unknown's avatar
unknown committed
342 343
		os_has_said_disk_full = TRUE;

unknown's avatar
unknown committed
344 345
		fflush(stderr);

unknown's avatar
unknown committed
346
		return(FALSE);
347

348
	} else if (err == OS_FILE_AIO_RESOURCES_RESERVED) {
unknown's avatar
unknown committed
349

350
		return(TRUE);
351 352

	} else if (err == OS_FILE_ALREADY_EXISTS) {
unknown's avatar
unknown committed
353

354
		return(FALSE);
355
	} else {
unknown's avatar
unknown committed
356 357 358
	        if (name) {
	                fprintf(stderr, "InnoDB: File name %s\n", name);
	        }
359
	  
360 361
		fprintf(stderr, "InnoDB: File operation call: '%s'.\n",
							       operation);
362 363
		fprintf(stderr, "InnoDB: Cannot continue operation.\n");

unknown's avatar
unknown committed
364 365
		fflush(stderr);

366
		exit(1);
367 368 369 370 371
	}

	return(FALSE);	
}

unknown's avatar
unknown committed
372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430
/********************************************************************
Does error handling when a file operation fails. */
static
ibool
os_file_handle_error_no_exit(
/*=========================*/
				/* out: TRUE if we should retry the
				operation */
	os_file_t	file,	/* in: file pointer */
	char*		name,	/* in: name of a file or NULL */
	const char*	operation)/* in: operation */
{
	ulint	err;

	UT_NOT_USED(file);

	err = os_file_get_last_error(FALSE);
	
	if (err == OS_FILE_DISK_FULL) {
		/* We only print a warning about disk full once */

		if (os_has_said_disk_full) {

			return(FALSE);
		}
	
		if (name) {
			ut_print_timestamp(stderr);
			fprintf(stderr,
	"  InnoDB: Encountered a problem with file %s\n", name);
		}

		ut_print_timestamp(stderr);
	        fprintf(stderr,
	"  InnoDB: Disk is full. Try to clean the disk to free space.\n");

		os_has_said_disk_full = TRUE;

		fflush(stderr);

		return(FALSE);

	} else if (err == OS_FILE_AIO_RESOURCES_RESERVED) {

		return(TRUE);

	} else if (err == OS_FILE_ALREADY_EXISTS) {

		return(FALSE);
	} else {
	        if (name) {
	                fprintf(stderr, "InnoDB: File name %s\n", name);
	        }
	  
		fprintf(stderr, "InnoDB: File operation call: '%s'.\n",
							       operation);
		return (FALSE);
	}

431
	return(FALSE);		/* not reached */
unknown's avatar
unknown committed
432 433
}

unknown's avatar
unknown committed
434 435 436 437 438 439 440 441 442
/********************************************************************
Creates the seek mutexes used in positioned reads and writes. */

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

unknown's avatar
unknown committed
443 444
	os_file_count_mutex = os_mutex_create(NULL);

unknown's avatar
unknown committed
445 446 447 448 449
	for (i = 0; i < OS_FILE_N_SEEK_MUTEXES; i++) {
		os_file_seek_mutexes[i] = os_mutex_create(NULL);
	}
}

unknown's avatar
unknown committed
450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474
/***************************************************************************
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(
/*============*/
				/* out: directory stream, NULL if error */
	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 */
{
	os_file_dir_t		dir;
#ifdef __WIN__
        LPWIN32_FIND_DATA	lpFindFileData;
	char			path[OS_FILE_MAX_PATH + 3];

	ut_a(strlen(dirname) < OS_FILE_MAX_PATH);

	strcpy(path, dirname);
unknown's avatar
unknown committed
475
	strcpy(path + strlen(path), "\\*");
unknown's avatar
unknown committed
476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522

	/* 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));

	dir = FindFirstFile(path, lpFindFileData);

	ut_free(lpFindFileData);

	if (dir == INVALID_HANDLE_VALUE) {

		if (error_is_fatal) {
		        os_file_handle_error(NULL, dirname, "opendir");
		}

		return(NULL);
	}

	return(dir);	
#else
	dir = opendir(dirname);

	if (dir == NULL && error_is_fatal) {
	        os_file_handle_error(0, dirname, "opendir");
	}

	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) {
unknown's avatar
unknown committed
523
	        os_file_handle_error_no_exit(NULL, NULL, "closedir");
unknown's avatar
unknown committed
524 525 526 527 528 529 530 531 532 533 534
		
		return(-1);
	}
	
	return(0);
#else
	int	ret;
	
	ret = closedir(dir);

	if (ret) {
unknown's avatar
unknown committed
535
	        os_file_handle_error_no_exit(0, NULL, "closedir");
unknown's avatar
unknown committed
536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563
	}

	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 */
	char*		dirname,/* in: directory name or path */
	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
564
	        ut_a(strlen(lpFindFileData->cFileName) < OS_FILE_MAX_PATH);
unknown's avatar
unknown committed
565

unknown's avatar
unknown committed
566 567
		if (strcmp(lpFindFileData->cFileName, ".") == 0
		    || strcmp(lpFindFileData->cFileName, "..") == 0) {
unknown's avatar
unknown committed
568 569 570 571

		        goto next_file;
		}

unknown's avatar
unknown committed
572
		strcpy(info->name, lpFindFileData->cFileName);
unknown's avatar
unknown committed
573

unknown's avatar
unknown committed
574 575
		info->size = (ib_longlong)(lpFindFileData->nFileSizeLow)
		     + (((ib_longlong)(lpFindFileData->nFileSizeHigh)) << 32);
unknown's avatar
unknown committed
576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603

		if (lpFindFileData->dwFileAttributes
					& 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://www.mysql.com/doc/en/Windows_symbolic_links.html */
			info->type = OS_FILE_TYPE_LINK;
		} else if (lpFindFileData->dwFileAttributes
						& FILE_ATTRIBUTE_DIRECTORY) {
		        info->type = OS_FILE_TYPE_DIR;
		} else if (lpFindFileData->dwFileAttributes
						& FILE_ATTRIBUTE_NORMAL) {
/* TODO: are FILE_ATTRIBUTE_NORMAL files really all normal files? */	
			info->type = OS_FILE_TYPE_FILE;
		} else {
			info->type = OS_FILE_TYPE_UNKNOWN;
		}
	}

	ut_free(lpFindFileData);

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

		return(1);
	} else {
unknown's avatar
unknown committed
604 605
		os_file_handle_error_no_exit(NULL, dirname,
						"readdir_next_file");
unknown's avatar
unknown committed
606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635
		return(-1);
	}
#else
	struct dirent*	ent;
	char*		full_path;
	int		ret;
	struct stat	statinfo;
next_file:
	ent = readdir(dir);

	if (ent == NULL) {
		return(1);
	}

	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);
	
	sprintf(full_path, "%s/%s", dirname, ent->d_name);

	ret = stat(full_path, &statinfo);

	if (ret) {
unknown's avatar
unknown committed
636
		os_file_handle_error_no_exit(0, full_path, "stat");
unknown's avatar
unknown committed
637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660

		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)) {
	        info->type = OS_FILE_TYPE_LINK;
	} else if (S_ISREG(statinfo.st_mode)) {
	        info->type = OS_FILE_TYPE_FILE;
	} else {
	        info->type = OS_FILE_TYPE_UNKNOWN;
	}
			
	ut_free(full_path);

	return(0);
#endif
}

unknown's avatar
unknown committed
661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705
/*********************************************************************
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(
/*=====================*/
				/* out: TRUE if call succeeds, FALSE on
				error */
	char*	pathname,	/* in: directory name as null-terminated
				string */
	ibool	fail_if_exists)	/* in: if TRUE, pre-existing directory is
				treated as an error. */
{
#ifdef __WIN__
	BOOL	rcode;
    
	rcode = CreateDirectory(pathname, NULL);
	if (!(rcode != 0 ||
		   (GetLastError() == ERROR_FILE_EXISTS && !fail_if_exists))) {
		/* failure */
		os_file_handle_error(NULL, pathname, "CreateDirectory");

		return(FALSE);
	}
        
	return (TRUE);
#else
	int	rcode;

	rcode = mkdir(pathname, 0770);

	if (!(rcode == 0 || (errno == EEXIST && !fail_if_exists))) {
		/* failure */
		os_file_handle_error(0, pathname, "mkdir");

		return(FALSE);
	}
        
	return (TRUE);
#endif    
}

unknown's avatar
unknown committed
706 707 708 709 710 711 712
/********************************************************************
A simple function to open or create a file. */

os_file_t
os_file_create_simple(
/*==================*/
			/* out, own: handle to the file, not defined if error,
unknown's avatar
unknown committed
713 714
			error number can be retrieved with
			os_file_get_last_error */
unknown's avatar
unknown committed
715 716 717 718
	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
unknown's avatar
unknown committed
719 720 721
			file is created (if exists, error), or
                        OS_FILE_CREATE_PATH if new file (if exists, error) and
                        subdirectories along its path are created (if needed)*/
unknown's avatar
unknown committed
722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738
	ulint	access_type,/* in: OS_FILE_READ_ONLY or OS_FILE_READ_WRITE */
	ibool*	success)/* out: TRUE if succeed, FALSE if error */
{
#ifdef __WIN__
	os_file_t	file;
	DWORD		create_flag;
	DWORD		access;
	DWORD		attributes	= 0;
	ibool		retry;
	
try_again:	
	ut_a(name);

	if (create_mode == OS_FILE_OPEN) {
		create_flag = OPEN_EXISTING;
	} else if (create_mode == OS_FILE_CREATE) {
		create_flag = CREATE_NEW;
unknown's avatar
unknown committed
739 740 741 742 743 744 745 746
        } 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
747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762
	} 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;
	}

	file = CreateFile(name,
			access,
unknown's avatar
unknown committed
763 764
			FILE_SHARE_READ,/* file can be read also by other
					processes */
unknown's avatar
unknown committed
765 766 767 768 769 770 771 772
			NULL,	/* default security attributes */
			create_flag,
			attributes,
			NULL);	/* no template file */

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

unknown's avatar
unknown committed
773
		retry = os_file_handle_error(file, name,
unknown's avatar
unknown committed
774 775
				create_mode == OS_FILE_OPEN ?
				"open" : "create");
unknown's avatar
unknown committed
776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799
		if (retry) {
			goto try_again;
		}
	} else {
		*success = TRUE;
	}

	return(file);
#else
	os_file_t	file;
	int		create_flag;
	ibool		retry;
	
try_again:	
	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;
unknown's avatar
unknown committed
800 801 802 803 804 805 806 807
        } 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
808 809 810 811 812 813
	} else {
		create_flag = 0;
		ut_error;
	}

	if (create_mode == OS_FILE_CREATE) {
unknown's avatar
unknown committed
814 815
	        file = open(name, create_flag, S_IRUSR | S_IWUSR
						| S_IRGRP | S_IWGRP);
unknown's avatar
unknown committed
816 817 818 819 820 821 822
        } else {
                file = open(name, create_flag);
        }
	
	if (file == -1) {
		*success = FALSE;

unknown's avatar
unknown committed
823
		retry = os_file_handle_error(file, name,
unknown's avatar
unknown committed
824 825
				create_mode == OS_FILE_OPEN ?
				"open" : "create");
unknown's avatar
unknown committed
826 827 828 829 830 831 832 833 834 835
		if (retry) {
			goto try_again;
		}
	} else {
		*success = TRUE;
	}

	return(file);	
#endif
}
unknown's avatar
unknown committed
836 837 838 839 840 841 842 843

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

os_file_t
os_file_create_simple_no_error_handling(
/*====================================*/
			/* out, own: handle to the file, not defined if error,
unknown's avatar
unknown committed
844 845
			error number can be retrieved with
			os_file_get_last_error */
unknown's avatar
unknown committed
846 847 848 849 850
	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) */
unknown's avatar
unknown committed
851 852 853
	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 */
unknown's avatar
unknown committed
854 855 856 857 858 859 860
	ibool*	success)/* out: TRUE if succeed, FALSE if error */
{
#ifdef __WIN__
	os_file_t	file;
	DWORD		create_flag;
	DWORD		access;
	DWORD		attributes	= 0;
unknown's avatar
unknown committed
861
	DWORD		share_mode	= FILE_SHARE_READ;
unknown's avatar
unknown committed
862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877
	
	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
878 879 880 881 882 883 884
	} else if (access_type == OS_FILE_READ_ALLOW_DELETE) {
		access = GENERIC_READ;
		share_mode = FILE_SHARE_DELETE | FILE_SHARE_READ
			   | FILE_SHARE_WRITE;  /* A backup program has to give
						mysqld the maximum freedom to
						do what it likes with the
						file */
unknown's avatar
unknown committed
885 886 887 888 889 890 891
	} else {
		access = 0;
		ut_error;
	}

	file = CreateFile(name,
			access,
unknown's avatar
unknown committed
892
			share_mode,
unknown's avatar
unknown committed
893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940
			NULL,	/* default security attributes */
			create_flag,
			attributes,
			NULL);	/* no template file */

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

	return(file);
#else
	os_file_t	file;
	int		create_flag;
	
	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) {
	        file = open(name, create_flag, S_IRUSR | S_IWUSR
						| S_IRGRP | S_IWGRP);
        } else {
                file = open(name, create_flag);
        }
	
	if (file == -1) {
		*success = FALSE;
	} else {
		*success = TRUE;
	}

	return(file);	
#endif
}

941 942 943 944 945 946 947
/********************************************************************
Opens an existing file or creates a new. */

os_file_t
os_file_create(
/*===========*/
			/* out, own: handle to the file, not defined if error,
unknown's avatar
unknown committed
948 949
			error number can be retrieved with
			os_file_get_last_error */
950 951 952 953 954
	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
unknown's avatar
unknown committed
955 956 957
			if a new is created or an old overwritten,
			OS_FILE_OPEN_RAW, if a raw device or disk partition
			should be opened */
958
	ulint	purpose,/* in: OS_FILE_AIO, if asynchronous, non-buffered i/o
959 960 961 962 963
			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 */
964
	ulint	type,	/* in: OS_DATA_FILE or OS_LOG_FILE */
965 966 967 968
	ibool*	success)/* out: TRUE if succeed, FALSE if error */
{
#ifdef __WIN__
	os_file_t	file;
unknown's avatar
unknown committed
969
	DWORD		share_mode	= FILE_SHARE_READ;
970 971 972 973 974 975
	DWORD		create_flag;
	DWORD		attributes;
	ibool		retry;
try_again:	
	ut_a(name);

unknown's avatar
unknown committed
976 977 978
	if (create_mode == OS_FILE_OPEN_RAW) {
		create_flag = OPEN_EXISTING;
		share_mode = FILE_SHARE_WRITE;
unknown's avatar
unknown committed
979
	} else if (create_mode == OS_FILE_OPEN) {
980 981 982 983 984 985 986 987 988 989 990
		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) {
991 992
		/* If specified, use asynchronous (overlapped) io and no
		buffering of writes in the OS */
993 994 995 996 997 998 999
		attributes = 0;
#ifdef WIN_ASYNC_IO
		if (os_aio_use_native_aio) {
			attributes = attributes | FILE_FLAG_OVERLAPPED;
		}
#endif			
#ifdef UNIV_NON_BUFFERED_IO
1000
		if (type == OS_LOG_FILE && srv_flush_log_at_trx_commit == 2) {
unknown's avatar
unknown committed
1001
		        /* Do not use unbuffered i/o to log files because
1002 1003 1004 1005 1006
		        value 2 denotes that we do not flush the log at every
		        commit, but only once per second */
		} else if (srv_win_file_flush_method ==
						      SRV_WIN_IO_UNBUFFERED) {
		        attributes = attributes | FILE_FLAG_NO_BUFFERING;
unknown's avatar
unknown committed
1007
		}
1008 1009
#endif
	} else if (purpose == OS_FILE_NORMAL) {
unknown's avatar
unknown committed
1010
	        attributes = 0;
1011
#ifdef UNIV_NON_BUFFERED_IO
unknown's avatar
unknown committed
1012 1013 1014 1015
		if (type == OS_LOG_FILE && srv_flush_log_at_trx_commit == 2) {
		        /* 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 */
1016 1017 1018
		} else if (srv_win_file_flush_method ==
						      SRV_WIN_IO_UNBUFFERED) {
		        attributes = attributes | FILE_FLAG_NO_BUFFERING;
unknown's avatar
unknown committed
1019
		}
1020 1021 1022 1023 1024 1025 1026 1027 1028
#endif
	} else {
		attributes = 0;
		ut_error;
	}

	file = CreateFile(name,
			GENERIC_READ | GENERIC_WRITE, /* read and write
							access */
unknown's avatar
unknown committed
1029
			share_mode,     /* File can be read also by other
unknown's avatar
unknown committed
1030 1031 1032 1033 1034 1035
					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
1036
					database corruption! When opening
unknown's avatar
unknown committed
1037
					raw disk partitions, Microsoft manuals
unknown's avatar
unknown committed
1038 1039
					say that we must give also the write
					permission. */
1040 1041 1042 1043 1044 1045 1046 1047
			NULL,	/* default security attributes */
			create_flag,
			attributes,
			NULL);	/* no template file */

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

unknown's avatar
unknown committed
1048
		retry = os_file_handle_error(file, name,
unknown's avatar
unknown committed
1049 1050
				create_mode == OS_FILE_CREATE ?
				"create" : "open");
1051 1052
		if (retry) {
			goto try_again;
1053 1054 1055 1056 1057 1058 1059 1060 1061 1062
		}
	} else {
		*success = TRUE;
	}

	return(file);
#else
	os_file_t	file;
	int		create_flag;
	ibool		retry;
1063 1064 1065
	const char*	mode_str	= NULL;
	const char*	type_str	= NULL;
	const char*	purpose_str	= NULL;
1066 1067 1068 1069
	
try_again:	
	ut_a(name);

unknown's avatar
unknown committed
1070
	if (create_mode == OS_FILE_OPEN || create_mode == OS_FILE_OPEN_RAW) {
1071
		mode_str = "OPEN";
1072 1073
		create_flag = O_RDWR;
	} else if (create_mode == OS_FILE_CREATE) {
1074
		mode_str = "CREATE";
1075 1076
		create_flag = O_RDWR | O_CREAT | O_EXCL;
	} else if (create_mode == OS_FILE_OVERWRITE) {
1077
		mode_str = "OVERWRITE";
1078 1079 1080 1081 1082 1083
		create_flag = O_RDWR | O_CREAT | O_TRUNC;
	} else {
		create_flag = 0;
		ut_error;
	}

1084 1085 1086 1087 1088
	if (type == OS_LOG_FILE) {
		type_str = "LOG";
	} else if (type == OS_DATA_FILE) {
		type_str = "DATA";
	} else {
1089
	        ut_error;
1090 1091 1092 1093 1094 1095 1096
	}
	  
	if (purpose == OS_FILE_AIO) {
		purpose_str = "AIO";
	} else if (purpose == OS_FILE_NORMAL) {
		purpose_str = "NORMAL";
	} else {
1097
	        ut_error;
1098
	}
1099

1100
/*	fprintf(stderr, "Opening file %s, mode %s, type %s, purpose %s\n",
1101
			       name, mode_str, type_str, purpose_str); */
1102
#ifdef O_SYNC
1103 1104 1105 1106
        /* We let O_SYNC only affect log files; note that we map O_DSYNC to
	O_SYNC because the datasync options seemed to corrupt files in 2001
	in both Linux and Solaris */
	if (type == OS_LOG_FILE
1107 1108
	    && srv_unix_file_flush_method == SRV_UNIX_O_DSYNC) {

1109
/*		fprintf(stderr, "Using O_SYNC for file %s\n", name); */
1110

1111
	        create_flag = create_flag | O_SYNC;
1112
	}
1113 1114 1115 1116 1117 1118
#endif
#ifdef O_DIRECT
        /* We let O_DIRECT only affect data files */
	if (type != OS_LOG_FILE
	    && srv_unix_file_flush_method == SRV_UNIX_O_DIRECT) {

1119
/*		fprintf(stderr, "Using O_DIRECT for file %s\n", name); */
1120 1121 1122

	        create_flag = create_flag | O_DIRECT;
	}
1123
#endif
1124
	if (create_mode == OS_FILE_CREATE) {
unknown's avatar
unknown committed
1125
	        file = open(name, create_flag, os_innodb_umask);
1126 1127 1128 1129 1130 1131 1132
        } else {
                file = open(name, create_flag);
        }
	
	if (file == -1) {
		*success = FALSE;

unknown's avatar
unknown committed
1133
		retry = os_file_handle_error(file, name,
unknown's avatar
unknown committed
1134 1135
				create_mode == OS_FILE_CREATE ?
				"create" : "open");
1136 1137
		if (retry) {
			goto try_again;
1138 1139 1140 1141 1142 1143 1144 1145 1146
		}
	} else {
		*success = TRUE;
	}

	return(file);	
#endif
}

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 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207
/***************************************************************************
Deletes a file if it exists. The file has to be closed before calling this. */

ibool
os_file_delete_if_exists(
/*=====================*/
			/* out: TRUE if success */
	char*	name)	/* in: file path as a null-terminated string */
{
#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);
	}

	if (GetLastError() == ERROR_PATH_NOT_FOUND) {
		/* the file does not exist, this not an error */

		return(TRUE);
	}

	count++;

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

		return(FALSE);
	}

	return(TRUE);
#endif
}

unknown's avatar
unknown committed
1208 1209 1210 1211 1212 1213 1214 1215 1216 1217
/***************************************************************************
Deletes a file. The file has to be closed before calling this. */

ibool
os_file_delete(
/*===========*/
			/* out: TRUE if success */
	char*	name)	/* in: file path as a null-terminated string */
{
#ifdef __WIN__
unknown's avatar
unknown committed
1218
	BOOL	ret;
unknown's avatar
unknown committed
1219 1220 1221 1222
	ulint	count	= 0;
loop:
	/* In Windows, deleting an .ibd file may fail if ibbackup is copying
	it */
unknown's avatar
unknown committed
1223 1224 1225 1226 1227 1228 1229

	ret = DeleteFile((LPCTSTR)name);

	if (ret) {
		return(TRUE);
	}

unknown's avatar
unknown committed
1230 1231 1232
	if (GetLastError() == ERROR_PATH_NOT_FOUND) {
		/* If the file does not exist, we classify this as a 'mild'
		error and return */
unknown's avatar
unknown committed
1233

unknown's avatar
unknown committed
1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254
		return(FALSE);
	}

	count++;

	if (count > 100 && 0 == (count % 10)) {
		fprintf(stderr,
"InnoDB: Warning: cannot delete file %s\n"
"InnoDB: Are you running ibbackup to back up the file?\n", name);
		
		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
1255
#else
unknown's avatar
unknown committed
1256
	int	ret;
unknown's avatar
unknown committed
1257 1258 1259 1260

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

	if (ret != 0) {
unknown's avatar
unknown committed
1261
		os_file_handle_error(0, name, "delete");
unknown's avatar
unknown committed
1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282

		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 */
	char*	oldpath,	/* in: old file path as a null-terminated
				string */
	char*	newpath)	/* in: new file path */
{
#ifdef __WIN__
unknown's avatar
unknown committed
1283
	BOOL	ret;
unknown's avatar
unknown committed
1284 1285 1286 1287 1288 1289 1290

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

	if (ret) {
		return(TRUE);
	}

unknown's avatar
unknown committed
1291
	os_file_handle_error(NULL, oldpath, "delete");
unknown's avatar
unknown committed
1292 1293 1294

	return(FALSE);
#else
unknown's avatar
unknown committed
1295
	int	ret;
unknown's avatar
unknown committed
1296 1297 1298 1299

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

	if (ret != 0) {
unknown's avatar
unknown committed
1300
		os_file_handle_error(0, oldpath, "rename");
unknown's avatar
unknown committed
1301 1302 1303 1304 1305 1306 1307 1308

		return(FALSE);
	}

	return(TRUE);
#endif
}

1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329
/***************************************************************************
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);
	}

unknown's avatar
unknown committed
1330
	os_file_handle_error(file, NULL, "close");
unknown's avatar
unknown committed
1331

1332 1333 1334 1335 1336 1337 1338
	return(FALSE);
#else
	int	ret;

	ret = close(file);

	if (ret == -1) {
unknown's avatar
unknown committed
1339
		os_file_handle_error(file, NULL, "close");
unknown's avatar
unknown committed
1340

1341 1342 1343 1344 1345 1346 1347
		return(FALSE);
	}

	return(TRUE);
#endif
}

unknown's avatar
unknown committed
1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382
/***************************************************************************
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
}

1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409
/***************************************************************************
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
1410 1411 1412 1413
	off_t	offs;

	offs = lseek(file, 0, SEEK_END);

unknown's avatar
unknown committed
1414 1415 1416 1417 1418
	if (offs == ((off_t)-1)) {

		return(FALSE);
	}
	
unknown's avatar
Merge  
unknown committed
1419
	if (sizeof(off_t) > 4) {
unknown's avatar
unknown committed
1420
	        *size = (ulint)(offs & 0xFFFFFFFFUL);
1421
		*size_high = (ulint)(offs >> 32);
unknown's avatar
Merge  
unknown committed
1422 1423 1424 1425
	} else {
		*size = (ulint) offs;
		*size_high = 0;
	}
1426 1427 1428 1429 1430
	
	return(TRUE);	
#endif
}

1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453
/***************************************************************************
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);
}

1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467
/***************************************************************************
Sets a file size. This function can be used to extend or truncate a file. */

ibool
os_file_set_size(
/*=============*/
				/* out: TRUE if success */
	char*		name,	/* in: name of the file or path as a
				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 */
{
unknown's avatar
Merge  
unknown committed
1468 1469 1470 1471 1472
	ib_longlong	offset;
	ib_longlong	low;
	ulint   	n_bytes;
	ibool		ret;
	byte*   	buf;
unknown's avatar
unknown committed
1473
	byte*   	buf2;
unknown's avatar
Merge  
unknown committed
1474 1475
	ulint   	i;

1476
	ut_a(size == (size & 0xFFFFFFFF));
1477

1478
	/* We use a very big 8 MB buffer in writing because Linux may be
unknown's avatar
Merge  
unknown committed
1479
	extremely slow in fsync on 1 MB writes */
1480

unknown's avatar
unknown committed
1481 1482 1483 1484
	buf2 = ut_malloc(UNIV_PAGE_SIZE * 513);

	/* Align the buffer for possible raw i/o */
	buf = ut_align(buf2, UNIV_PAGE_SIZE);
1485 1486

	/* Write buffer full of zeros */
1487
	for (i = 0; i < UNIV_PAGE_SIZE * 512; i++) {
1488 1489 1490 1491
	        buf[i] = '\0';
	}

	offset = 0;
1492
	low = (ib_longlong)size + (((ib_longlong)size_high) << 32);
unknown's avatar
unknown committed
1493 1494

	if (low >= (ib_longlong)(100 * 1024 * 1024)) {
1495
				
unknown's avatar
unknown committed
1496 1497 1498
		fprintf(stderr, "InnoDB: Progress in MB:");
	}

1499
	while (offset < low) {
1500
	        if (low - offset < UNIV_PAGE_SIZE * 512) {
unknown's avatar
Merge  
unknown committed
1501
	        	n_bytes = (ulint)(low - offset);
1502
	        } else {
unknown's avatar
Merge  
unknown committed
1503
	        	n_bytes = UNIV_PAGE_SIZE * 512;
1504 1505
	        }
	  
unknown's avatar
Merge  
unknown committed
1506 1507
	        ret = os_file_write(name, file, buf,
				(ulint)(offset & 0xFFFFFFFF),
1508
				    (ulint)(offset >> 32),
unknown's avatar
Merge  
unknown committed
1509
				n_bytes);
1510
	        if (!ret) {
unknown's avatar
unknown committed
1511
			ut_free(buf2);
1512 1513
	         	goto error_handling;
	        }
unknown's avatar
unknown committed
1514 1515 1516 1517 1518 1519
				
		/* Print about progress for each 100 MB written */
		if ((offset + n_bytes) / (ib_longlong)(100 * 1024 * 1024)
		    != offset / (ib_longlong)(100 * 1024 * 1024)) {

		        fprintf(stderr, " %lu00",
1520
				(ulong) ((offset + n_bytes)
unknown's avatar
unknown committed
1521 1522 1523
					/ (ib_longlong)(100 * 1024 * 1024)));
		}
		
1524 1525 1526
	        offset += n_bytes;
	}

unknown's avatar
unknown committed
1527 1528 1529 1530 1531
	if (low >= (ib_longlong)(100 * 1024 * 1024)) {
				
		fprintf(stderr, "\n");
	}

unknown's avatar
unknown committed
1532
	ut_free(buf2);
1533 1534 1535 1536 1537 1538 1539 1540

	ret = os_file_flush(file);

	if (ret) {
	        return(TRUE);
	}

error_handling:
1541
	return(FALSE);
1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557
}

/***************************************************************************
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
1558 1559
	os_n_fsyncs++;

1560 1561 1562 1563 1564 1565
	ret = FlushFileBuffers(file);

	if (ret) {
		return(TRUE);
	}

unknown's avatar
unknown committed
1566 1567 1568 1569 1570 1571 1572 1573 1574
	/* 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()
						== ERROR_INVALID_FUNCTION) {
	        return(TRUE);
	}

unknown's avatar
unknown committed
1575
	os_file_handle_error(file, NULL, "flush");
1576

unknown's avatar
unknown committed
1577 1578
	/* It is a fatal error if a file flush does not succeed, because then
	the database can get corrupt on disk */
1579
	ut_error;
unknown's avatar
unknown committed
1580

1581 1582 1583
	return(FALSE);
#else
	int	ret;
1584

1585
#ifdef HAVE_FDATASYNC
1586 1587
	ret = fdatasync(file);
#else
1588
/*	fprintf(stderr, "Flushing to file %p\n", file); */
1589
	ret = fsync(file);
1590
#endif
1591 1592
	os_n_fsyncs++;

1593 1594 1595 1596
	if (ret == 0) {
		return(TRUE);
	}
	
unknown's avatar
unknown committed
1597
	/* Since Linux returns EINVAL if the 'file' is actually a raw device,
unknown's avatar
unknown committed
1598 1599 1600
	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
1601 1602 1603 1604

	        return(TRUE);
	}

unknown's avatar
unknown committed
1605 1606
	ut_print_timestamp(stderr);
	
1607
	fprintf(stderr,
unknown's avatar
unknown committed
1608
		"  InnoDB: Error: the OS said file flush did not succeed\n");
1609

unknown's avatar
unknown committed
1610
	os_file_handle_error(file, NULL, "flush");
1611

unknown's avatar
unknown committed
1612 1613
	/* It is a fatal error if a file flush does not succeed, because then
	the database can get corrupt on disk */
1614
	ut_error;
unknown's avatar
unknown committed
1615

1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630
	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 */
	ulint		n,	/* in: number of bytes to read */	
unknown's avatar
Merge  
unknown committed
1631 1632 1633 1634
	ulint		offset,	/* in: least significant 32 bits of file
				offset from where to read */
	ulint		offset_high) /* in: most significant 32 bits of
				offset */
1635
{
unknown's avatar
Merge  
unknown committed
1636
        off_t	offs;
unknown's avatar
unknown committed
1637
	ssize_t	n_bytes;
unknown's avatar
Merge  
unknown committed
1638

unknown's avatar
unknown committed
1639
	ut_a((offset & 0xFFFFFFFFUL) == offset);
unknown's avatar
Merge  
unknown committed
1640 1641 1642 1643 1644
        
        /* If off_t is > 4 bytes in size, then we assume we can pass a
	64-bit address */

        if (sizeof(off_t) > 4) {
1645 1646
	        offs = (off_t)offset + (((off_t)offset_high) << 32);
        				
unknown's avatar
Merge  
unknown committed
1647 1648 1649 1650 1651 1652 1653 1654
        } else {
        	offs = (off_t)offset;

        	if (offset_high > 0) {
        		fprintf(stderr,
			"InnoDB: Error: file read at offset > 4 GB\n");
		}
        }
1655

1656 1657
	os_n_file_reads++;

1658
#ifdef HAVE_PREAD
unknown's avatar
unknown committed
1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669
        os_mutex_enter(os_file_count_mutex);
	os_file_n_pending_preads++;
        os_mutex_exit(os_file_count_mutex);

        n_bytes = pread(file, buf, n, offs);

        os_mutex_enter(os_file_count_mutex);
	os_file_n_pending_preads--;
        os_mutex_exit(os_file_count_mutex);

	return(n_bytes);
1670
#else
unknown's avatar
unknown committed
1671
	{
1672 1673 1674 1675 1676 1677 1678 1679
	ssize_t	ret;
	ulint	i;

	/* Protect the seek / read operation with a mutex */
	i = ((ulint) file) % OS_FILE_N_SEEK_MUTEXES;
	
	os_mutex_enter(os_file_seek_mutexes[i]);

1680
	ret = lseek(file, offs, 0);
1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692

	if (ret < 0) {
		os_mutex_exit(os_file_seek_mutexes[i]);

		return(ret);
	}
	
	ret = read(file, buf, n);

	os_mutex_exit(os_file_seek_mutexes[i]);

	return(ret);
unknown's avatar
unknown committed
1693
	}
1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706
#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 */
	void*		buf,	/* in: buffer from where to write */
	ulint		n,	/* in: number of bytes to write */	
unknown's avatar
Merge  
unknown committed
1707 1708 1709 1710
	ulint		offset,	/* in: least significant 32 bits of file
				offset where to write */
	ulint		offset_high) /* in: most significant 32 bits of
				offset */
1711
{
1712
	ssize_t	ret;
unknown's avatar
Merge  
unknown committed
1713 1714
        off_t	offs;

unknown's avatar
unknown committed
1715
	ut_a((offset & 0xFFFFFFFFUL) == offset);
unknown's avatar
Merge  
unknown committed
1716 1717 1718 1719 1720

        /* If off_t is > 4 bytes in size, then we assume we can pass a
	64-bit address */

        if (sizeof(off_t) > 4) {
unknown's avatar
unknown committed
1721
	  	offs = (off_t)offset + (((off_t)offset_high) << 32);
unknown's avatar
Merge  
unknown committed
1722 1723 1724 1725 1726 1727 1728 1729
        } else {
        	offs = (off_t)offset;

        	if (offset_high > 0) {
        		fprintf(stderr,
			"InnoDB: Error: file write at offset > 4 GB\n");
		}
        }
1730

1731 1732
	os_n_file_writes++;

1733
#ifdef HAVE_PWRITE
unknown's avatar
unknown committed
1734 1735 1736 1737
        os_mutex_enter(os_file_count_mutex);
	os_file_n_pending_pwrites++;
        os_mutex_exit(os_file_count_mutex);

1738
	ret = pwrite(file, buf, n, offs);
1739

unknown's avatar
unknown committed
1740 1741 1742 1743
        os_mutex_enter(os_file_count_mutex);
	os_file_n_pending_pwrites--;
        os_mutex_exit(os_file_count_mutex);

1744
	if (srv_unix_file_flush_method != SRV_UNIX_LITTLESYNC
1745
	    && srv_unix_file_flush_method != SRV_UNIX_NOSYNC
unknown's avatar
unknown committed
1746 1747
	    && !os_do_not_call_flush_at_each_write) {
	    	
1748 1749 1750
	        /* Always do fsync to reduce the probability that when
                the OS crashes, a database page is only partially
                physically written to disk. */
1751

1752 1753
	        ut_a(TRUE == os_file_flush(file));
	}
1754 1755

        return(ret);
1756
#else
unknown's avatar
unknown committed
1757
	{
1758 1759 1760 1761 1762 1763 1764
	ulint	i;

	/* Protect the seek / write operation with a mutex */
	i = ((ulint) file) % OS_FILE_N_SEEK_MUTEXES;
	
	os_mutex_enter(os_file_seek_mutexes[i]);

1765
	ret = lseek(file, offs, 0);
1766 1767 1768 1769 1770 1771 1772 1773 1774

	if (ret < 0) {
		os_mutex_exit(os_file_seek_mutexes[i]);

		return(ret);
	}
	
	ret = write(file, buf, n);

1775
	if (srv_unix_file_flush_method != SRV_UNIX_LITTLESYNC
1776
	    && srv_unix_file_flush_method != SRV_UNIX_NOSYNC
unknown's avatar
unknown committed
1777
	    && !os_do_not_call_flush_at_each_write) {
1778

1779 1780 1781 1782 1783 1784
	        /* 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));
	}
1785

1786 1787 1788
	os_mutex_exit(os_file_seek_mutexes[i]);

	return(ret);
unknown's avatar
unknown committed
1789
	}
1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818
#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 */
	ulint		n)	/* in: number of bytes to read */	
{
#ifdef __WIN__
	BOOL		ret;
	DWORD		len;
	DWORD		ret2;
	DWORD		low;
	DWORD		high;
	ibool		retry;
	ulint		i;
	
unknown's avatar
unknown committed
1819
	ut_a((offset & 0xFFFFFFFFUL) == offset);
unknown's avatar
Merge  
unknown committed
1820

1821
	os_n_file_reads++;
unknown's avatar
unknown committed
1822
	os_bytes_read_since_printout += n;
1823

1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855
try_again:	
	ut_ad(file);
	ut_ad(buf);
	ut_ad(n > 0);

	low = offset;
	high = offset_high;

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

		goto error_handling;
	} 
	
	ret = ReadFile(file, buf, n, &len, NULL);

	os_mutex_exit(os_file_seek_mutexes[i]);
	
	if (ret && len == n) {
		return(TRUE);
	}		
#else
	ibool	retry;
	ssize_t	ret;
unknown's avatar
Merge  
unknown committed
1856

unknown's avatar
unknown committed
1857 1858
	os_bytes_read_since_printout += n;

1859
try_again:
unknown's avatar
Merge  
unknown committed
1860
	ret = os_file_pread(file, buf, n, offset, offset_high);
1861

1862
	if ((ulint)ret == n) {
1863 1864 1865 1866

		return(TRUE);
	}
#endif	
unknown's avatar
unknown committed
1867
#ifdef __WIN__
1868
error_handling:
unknown's avatar
unknown committed
1869
#endif
unknown's avatar
unknown committed
1870
	retry = os_file_handle_error(file, NULL, "read"); 
1871 1872 1873 1874

	if (retry) {
		goto try_again;
	}
unknown's avatar
unknown committed
1875 1876 1877 1878
       
	fprintf(stderr,
"InnoDB: Fatal error: cannot read from file. OS error number %lu.\n",
#ifdef __WIN__
1879
		(ulong) GetLastError()
unknown's avatar
unknown committed
1880
#else
1881
		(ulong) errno
unknown's avatar
unknown committed
1882 1883 1884 1885
#endif
		);
	fflush(stderr);

1886 1887 1888 1889 1890
	ut_error;

	return(FALSE);
}

unknown's avatar
unknown committed
1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976
/***********************************************************************
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 */
	ulint		n)	/* in: number of bytes to read */	
{
#ifdef __WIN__
	BOOL		ret;
	DWORD		len;
	DWORD		ret2;
	DWORD		low;
	DWORD		high;
	ibool		retry;
	ulint		i;
	
	ut_a((offset & 0xFFFFFFFFUL) == offset);

	os_n_file_reads++;
	os_bytes_read_since_printout += n;

try_again:	
	ut_ad(file);
	ut_ad(buf);
	ut_ad(n > 0);

	low = offset;
	high = offset_high;

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

		goto error_handling;
	} 
	
	ret = ReadFile(file, buf, n, &len, NULL);

	os_mutex_exit(os_file_seek_mutexes[i]);
	
	if (ret && len == n) {
		return(TRUE);
	}		
#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);
	}
#endif	
#ifdef __WIN__
error_handling:
#endif
	retry = os_file_handle_error_no_exit(file, NULL, "read"); 

	if (retry) {
		goto try_again;
	}
       
	return(FALSE);
}

1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001
/***********************************************************************
Requests a synchronous write operation. */

ibool
os_file_write(
/*==========*/
				/* out: TRUE if request was
				successful, FALSE if fail */
	char*		name,	/* in: name of the file or path as a
				null-terminated string */
	os_file_t	file,	/* in: handle to a file */
	void*		buf,	/* in: buffer from which to write */
	ulint		offset,	/* in: least significant 32 bits of file
				offset where to write */
	ulint		offset_high, /* in: most significant 32 bits of
				offset */
	ulint		n)	/* in: number of bytes to write */	
{
#ifdef __WIN__
	BOOL		ret;
	DWORD		len;
	DWORD		ret2;
	DWORD		low;
	DWORD		high;
	ulint		i;
unknown's avatar
unknown committed
2002
	ulint		n_retries	= 0;
2003
	ulint		err;
2004

2005
	ut_a((offset & 0xFFFFFFFF) == offset);
unknown's avatar
Merge  
unknown committed
2006

2007
	os_n_file_writes++;
unknown's avatar
unknown committed
2008

2009 2010 2011
	ut_ad(file);
	ut_ad(buf);
	ut_ad(n > 0);
unknown's avatar
unknown committed
2012
retry:
2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026
	low = offset;
	high = offset_high;
	
	/* Protect the seek / write operation with a mutex */
	i = ((ulint) file) % OS_FILE_N_SEEK_MUTEXES;
	
	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]);
		
unknown's avatar
unknown committed
2027 2028 2029 2030
		ut_print_timestamp(stderr);

		fprintf(stderr,
"  InnoDB: Error: File pointer positioning to file %s failed at\n"
unknown's avatar
unknown committed
2031
"InnoDB: offset %lu %lu. Operating system error number %lu.\n"
2032
"InnoDB: Look from section 13.2 at http://www.innodb.com/ibman.php\n"
unknown's avatar
unknown committed
2033
"InnoDB: what the error number means.\n",
2034 2035
			name, (ulong) offset_high, (ulong) offset,
			(ulong) GetLastError());
unknown's avatar
unknown committed
2036 2037

		return(FALSE);
2038 2039 2040
	} 

	ret = WriteFile(file, buf, n, &len, NULL);
2041 2042 2043 2044
	
	/* Always do fsync to reduce the probability that when the OS crashes,
	a database page is only partially physically written to disk. */

unknown's avatar
unknown committed
2045
	if (!os_do_not_call_flush_at_each_write) {
2046 2047
		ut_a(TRUE == os_file_flush(file));
	}
2048 2049

	os_mutex_exit(os_file_seek_mutexes[i]);
2050

2051
	if (ret && len == n) {
unknown's avatar
unknown committed
2052

2053 2054
		return(TRUE);
	}
unknown's avatar
unknown committed
2055

unknown's avatar
unknown committed
2056 2057 2058
	/* 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. */
unknown's avatar
unknown committed
2059
	
unknown's avatar
unknown committed
2060
	if (GetLastError() == ERROR_LOCK_VIOLATION && n_retries < 100) {
unknown's avatar
unknown committed
2061

unknown's avatar
unknown committed
2062 2063
		os_thread_sleep(1000000);
	
unknown's avatar
unknown committed
2064 2065 2066 2067 2068
		n_retries++;

		goto retry;
	}	
	
unknown's avatar
unknown committed
2069 2070
	if (!os_has_said_disk_full) {
	
2071 2072
		err = (ulint)GetLastError();

unknown's avatar
unknown committed
2073 2074 2075 2076 2077 2078 2079
		ut_print_timestamp(stderr);

		fprintf(stderr,
"  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"
unknown's avatar
unknown committed
2080
"InnoDB: Check also that the disk is not full or a disk quota exceeded.\n",
2081 2082
			name, (ulong) offset_high, (ulong) offset,
			(ulong) n, (ulong) len, (ulong) err);
2083 2084 2085

		if (strerror((int)err) != NULL) {
			fprintf(stderr,
2086
"InnoDB: Error number %lu means '%s'.\n", (ulong) err, strerror((int)err));
2087 2088 2089
		}

		fprintf(stderr,
2090
"InnoDB: See also section 13.2 at http://www.innodb.com/ibman.php\n"
2091
"InnoDB: about operating system error numbers.\n");
unknown's avatar
unknown committed
2092 2093 2094 2095 2096

		os_has_said_disk_full = TRUE;
	}

	return(FALSE);
2097 2098 2099
#else
	ssize_t	ret;
	
unknown's avatar
Merge  
unknown committed
2100
	ret = os_file_pwrite(file, buf, n, offset, offset_high);
2101

2102
	if ((ulint)ret == n) {
unknown's avatar
unknown committed
2103

2104 2105 2106
		return(TRUE);
	}

unknown's avatar
unknown committed
2107
	if (!os_has_said_disk_full) {
2108
	
unknown's avatar
unknown committed
2109
		ut_print_timestamp(stderr);
2110

unknown's avatar
unknown committed
2111 2112
		fprintf(stderr,
"  InnoDB: Error: Write to file %s failed at offset %lu %lu.\n"
unknown's avatar
unknown committed
2113
"InnoDB: %lu bytes should have been written, only %ld were written.\n"
unknown's avatar
unknown committed
2114 2115
"InnoDB: Operating system error number %lu.\n"
"InnoDB: Check that your OS and file system support files of this size.\n"
unknown's avatar
unknown committed
2116
"InnoDB: Check also that the disk is not full or a disk quota exceeded.\n",
unknown's avatar
unknown committed
2117
			name, offset_high, offset, n, (long int)ret,
unknown's avatar
unknown committed
2118
							(ulint)errno);
2119 2120 2121 2122 2123 2124
		if (strerror(errno) != NULL) {
			fprintf(stderr,
"InnoDB: Error number %lu means '%s'.\n", (ulint)errno, strerror(errno));
		}

		fprintf(stderr,
2125
"InnoDB: See also section 13.2 at http://www.innodb.com/ibman.php\n"
2126 2127
"InnoDB: about operating system error numbers.\n");

unknown's avatar
unknown committed
2128 2129 2130 2131 2132
		os_has_said_disk_full = TRUE;
	}

	return(FALSE);	
#endif
2133 2134
}

unknown's avatar
unknown committed
2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313
/***********************************************************************
Check the existence and type of the given file. */

ibool
os_file_status(
/*===========*/
				/* out: TRUE if call succeeded */
	char*		path,	/* in:  pathname of the file */
	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;
	
	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 */
		
		os_file_handle_error_no_exit(0, path, "stat");
		
		return(FALSE);
	}
	    
	if (_S_IFDIR & statinfo.st_mode) {
		*type = OS_FILE_TYPE_DIR;
	} else if (_S_IFREG & statinfo.st_mode) {
	        *type = OS_FILE_TYPE_FILE;
	} else {
	        *type = OS_FILE_TYPE_UNKNOWN;
	}

	*exists = TRUE;
	
	return(TRUE);
#else
	int		ret;
	struct stat	statinfo;
	
	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 */
		
		os_file_handle_error_no_exit(0, path, "stat");
		
		return(FALSE);
	}
	    
	if (S_ISDIR(statinfo.st_mode)) {
		*type = OS_FILE_TYPE_DIR;
	} else if (S_ISLNK(statinfo.st_mode)) {
	        *type = OS_FILE_TYPE_LINK;
	} else if (S_ISREG(statinfo.st_mode)) {
	        *type = OS_FILE_TYPE_FILE;
	} else {
	        *type = OS_FILE_TYPE_UNKNOWN;
	}

	*exists = TRUE;
	
	return(TRUE);
#endif
}

/* 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
is the component following the final '/'.  Trailing '/' charac­
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
to free it after it is no longer needed.	

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

       path           dirname        basename
       "/usr/lib"     "/usr"         "lib"
       "/usr/"        "/"            "usr"
       "usr"          "."            "usr"
       "/"            "/"            "/"
       "."            "."            "."
       ".."           "."            ".."
*/

char*
os_file_dirname(
/*============*/
				/* out, own: directory component of the
				pathname */
	char*		path)	/* in: pathname */
{
	char*	dir;
	int 	i, length, last_slash;

	/* find the offset of the last slash */
	length = ut_strlen(path);
	for (i = length - 1; i >= 0 && path[i] != OS_FILE_PATH_SEPARATOR; i++);
	last_slash = i;

	if (last_slash < 0) {
		/* no slash in the path, return "." */
		return(ut_strdup((char*)"."));
	}

	/* ok, there is a slash */

	if (last_slash == 0) {
		/* last slash is the first char of the path */
		return(ut_strdup((char*)"/"));
	}

	/* non-trivial directory component */
	dir = ut_strdup(path);
	dir[last_slash] = 0;

	return(dir);
}
	
/********************************************************************
Creates all missing subdirectories along the given path. */
	
ibool
os_file_create_subdirs_if_needed(
/*=============================*/
				/* out: TRUE if call succeeded
				   FALSE otherwise */
	char*		path)	/* in: path name */
{
	char*		subdir;
	static char 	rootdir[2] = { OS_FILE_PATH_SEPARATOR, 0 };
	ibool 		success, subdir_exists;
	os_file_type_t	type;

	subdir = os_file_dirname(path);
	if (0 == strcmp(subdir, rootdir) || 0 == strcmp(subdir, ".")) {
		/* subdir is root or cwd, nothing to do */
		ut_free(subdir);
		return(TRUE);
	}

	/* test if subdir exists */
	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) {
			ut_free(subdir);
			return(FALSE);
		}
		success = os_file_create_directory(subdir, FALSE);
	}

	ut_free(subdir);
	return(success);
}

2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352
/********************************************************************
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 */
	ulint	n_segments) 	/* in: number of segments in the aio array */
{
	os_aio_array_t*	array;
	ulint		i;
	os_aio_slot_t*	slot;
#ifdef WIN_ASYNC_IO
	OVERLAPPED*	over;
#endif	
	ut_a(n > 0);
	ut_a(n_segments > 0);

	array = ut_malloc(sizeof(os_aio_array_t));

	array->mutex 		= os_mutex_create(NULL);
	array->not_full		= os_event_create(NULL);
2353 2354 2355 2356
	array->is_empty		= os_event_create(NULL);

	os_event_set(array->is_empty);
	
2357 2358 2359 2360
	array->n_slots  	= n;
	array->n_segments	= n_segments;
	array->n_reserved	= 0;
	array->slots		= ut_malloc(n * sizeof(os_aio_slot_t));
2361 2362 2363
#ifdef __WIN__
	array->native_events	= ut_malloc(n * sizeof(os_native_event_t));
#endif	
2364 2365 2366 2367 2368 2369
	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
2370 2371
		slot->event = os_event_create(NULL);

2372 2373
		over = &(slot->control);

2374
		over->hEvent = slot->event->handle;
2375

2376
		*((array->native_events) + i) = over->hEvent;
2377 2378 2379 2380 2381 2382 2383
#endif
	}
	
	return(array);
}

/****************************************************************************
unknown's avatar
unknown committed
2384 2385
Initializes the asynchronous io system. Calls also os_io_init_simple.
Creates a separate aio array for
2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411
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
2412 2413
	os_io_init_simple();

unknown's avatar
unknown committed
2414
	for (i = 0; i < n_segments; i++) {
2415
	        srv_set_io_thread_op_info(i, "not started yet");
unknown's avatar
unknown committed
2416 2417
	}

2418 2419 2420 2421
	n_per_seg = n / n_segments;
	n_write_segs = (n_segments - 2) / 2;
	n_read_segs = n_segments - 2 - n_write_segs;
	
2422
	/* fprintf(stderr, "Array n per seg %lu\n", n_per_seg); */
2423

unknown's avatar
unknown committed
2424 2425
	os_aio_ibuf_array = os_aio_array_create(n_per_seg, 1);

2426
	srv_io_thread_function[0] = "insert buffer thread";
unknown's avatar
unknown committed
2427 2428 2429

	os_aio_log_array = os_aio_array_create(n_per_seg, 1);

2430
	srv_io_thread_function[1] = "log thread";
unknown's avatar
unknown committed
2431

2432 2433
	os_aio_read_array = os_aio_array_create(n_read_segs * n_per_seg,
							n_read_segs);
unknown's avatar
unknown committed
2434
	for (i = 2; i < 2 + n_read_segs; i++) {
unknown's avatar
unknown committed
2435
		ut_a(i < SRV_MAX_N_IO_THREADS);
2436
	        srv_io_thread_function[i] = "read thread";
unknown's avatar
unknown committed
2437 2438
	}

2439 2440
	os_aio_write_array = os_aio_array_create(n_write_segs * n_per_seg,
							n_write_segs);
unknown's avatar
unknown committed
2441
	for (i = 2 + n_read_segs; i < n_segments; i++) {
unknown's avatar
unknown committed
2442
		ut_a(i < SRV_MAX_N_IO_THREADS);
2443
	        srv_io_thread_function[i] = "write thread";
unknown's avatar
unknown committed
2444
	}
2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457

	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);
	}

2458 2459
	os_last_printout = time(NULL);

2460
#ifdef POSIX_ASYNC_IO
2461 2462 2463 2464 2465
	/* 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 */
	
2466 2467
	/* TODO: to work MySQL needs the SIGALARM signal; the following
	will not work yet! */
2468 2469 2470 2471 2472 2473
        sigemptyset(&sigset);
	sigaddset(&sigset, SIGRTMIN + 1 + 0);
	sigaddset(&sigset, SIGRTMIN + 1 + 1);
	sigaddset(&sigset, SIGRTMIN + 1 + 2);
	sigaddset(&sigset, SIGRTMIN + 1 + 3);

2474
	pthread_sigmask(SIG_BLOCK, &sigset, NULL); */
2475 2476
#endif
}
unknown's avatar
unknown committed
2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491

#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++) {

2492
	        os_event_set((array->slots + i)->event);
unknown's avatar
unknown committed
2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520
	}
}
#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
        /* This code wakes up all ai/o threads in Windows native aio */
	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++) {
	    	
		os_event_set(os_aio_segment_wait_events[i]);
	}	
}
2521
				
2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532
/****************************************************************************
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);
}

2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608
/**************************************************************************
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;
		
	} else if (array == os_aio_read_array) {
		seg_len = os_aio_read_array->n_slots /
				os_aio_read_array->n_segments;

		segment = 2 + slot->pos / seg_len;
	} else {
		ut_a(array == os_aio_write_array);
		seg_len = os_aio_write_array->n_slots /
				os_aio_write_array->n_segments;

		segment = os_aio_read_array->n_segments + 2
				+ slot->pos / seg_len;
	}

	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;

	ut_a(global_segment < os_aio_n_segments);	

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

	} else if (global_segment == 1) {
		*array = os_aio_log_array;
		segment = 0;
		
	} 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
2609 2610

#if !defined(WIN_ASYNC_IO) && defined(POSIX_ASYNC_IO)
2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631
static
ulint
os_aio_get_array_no(
/*================*/
	os_aio_array_t*	array)	/* in: aio array */
{	
	if (array == os_aio_ibuf_array) {
	
		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 {
2632
		ut_error;
2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658

		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 */
{	
	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 {
2659
		ut_error;
2660 2661 2662 2663

		return(NULL);
	}
}
unknown's avatar
unknown committed
2664
#endif /* if !defined(WIN_ASYNC_IO) && defined(POSIX_ASYNC_IO) */
2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727

/***********************************************************************
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 */
	void*		message1,/* in: message to be passed along with
				the aio operation */
	void*		message2,/* in: message to be passed along with
				the aio operation */
	os_file_t	file,	/* in: file handle */
	char*		name,	/* in: name of the file or path as a
				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();
		}
		
		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++;

2728 2729 2730 2731
	if (array->n_reserved == 1) {
		os_event_reset(array->is_empty);
	}

2732 2733 2734 2735 2736
	if (array->n_reserved == array->n_slots) {
		os_event_reset(array->not_full);
	}
	
	slot->reserved = TRUE;
unknown's avatar
unknown committed
2737
	slot->reservation_time = time(NULL);
2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752
	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;
	
#ifdef WIN_ASYNC_IO		
	control = &(slot->control);
	control->Offset = (DWORD)offset;
	control->OffsetHigh = (DWORD)offset_high;
2753
	os_event_reset(slot->event);
2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772

#elif defined(POSIX_ASYNC_IO)

#if (UNIV_WORD_SIZE == 8)
	offset = offset + (offset_high << 32);
#else
	ut_a(offset_high == 0);
#endif 
	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;
	control->aio_sigevent.sigev_signo =
			SIGRTMIN + 1 + os_aio_get_array_no(array);
			/* TODO: How to choose the signal numbers? */
/*
2773 2774
	fprintf(stderr, "AIO signal number %lu\n",
		(ulint) control->aio_sigevent.sigev_signo);
2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806
*/
	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);
	
	slot->reserved = FALSE;

	array->n_reserved--;

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

2807 2808 2809 2810
	if (array->n_reserved == 0) {
		os_event_set(array->is_empty);
	}

2811
#ifdef WIN_ASYNC_IO		
2812
	os_event_reset(slot->event);
2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827
#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
2828
	ulint		segment;
2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871
	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 */
			
			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;
unknown's avatar
unknown committed
2872 2873 2874
	}	

	os_aio_recommend_sleep_for_read_threads	= FALSE;
2875 2876 2877 2878 2879 2880

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

unknown's avatar
unknown committed
2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905
/**************************************************************************
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) {
		
			os_event_reset(os_aio_segment_wait_events[g]);
		}
	}
}

2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936
/***********************************************************************
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! */
	char*		name,	/* in: name of the file or path as a
				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
2937
	ulint		n,	/* in: number of bytes to read or write */
2938 2939 2940 2941 2942 2943 2944 2945 2946
	void*		message1,/* in: messages for the aio handler (these
				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
2947
	ibool		retval;
2948 2949 2950 2951
	BOOL		ret		= TRUE;
	DWORD		len		= n;
	void*		dummy_mess1;
	void*		dummy_mess2;
2952
	ulint		dummy_type;
2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980
#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);
	
	if (mode == OS_AIO_SYNC
#ifdef WIN_ASYNC_IO
				&& !os_aio_use_native_aio
#endif
	) {
		/* 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) {
2981 2982
			return(os_file_read(file, buf, offset,
							offset_high, n));
2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998
		}

		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);
2999 3000 3001 3002
		/* Reduce probability of deadlock bugs in connection with ibuf:
		do not let the ibuf i/o handler sleep */

		wake_later = FALSE;
3003 3004 3005 3006 3007 3008 3009 3010

		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 {
3011
		array = NULL; /* Eliminate compiler warning */
3012 3013 3014 3015 3016 3017 3018 3019
		ut_error;
	}
	
	slot = os_aio_array_reserve_slot(type, array, message1, message2, file,
					name, buf, offset, offset_high, n);
	if (type == OS_FILE_READ) {
		if (os_aio_use_native_aio) {
#ifdef WIN_ASYNC_IO
3020
			os_n_file_reads++;
unknown's avatar
unknown committed
3021 3022
			os_bytes_read_since_printout += len;
			
3023 3024 3025 3026 3027
			ret = ReadFile(file, buf, (DWORD)n, &len,
							&(slot->control));
#elif defined(POSIX_ASYNC_IO)
			slot->control.aio_lio_opcode = LIO_READ;
			err = (ulint) aio_read(&(slot->control));
3028
			fprintf(stderr, "Starting POSIX aio read %lu\n", err);
3029 3030 3031 3032 3033 3034 3035 3036 3037 3038
#endif
		} else {
			if (!wake_later) {
				os_aio_simulated_wake_handler_thread(
				 os_aio_get_segment_no_from_slot(array, slot));
			}
		}
	} else if (type == OS_FILE_WRITE) {
		if (os_aio_use_native_aio) {
#ifdef WIN_ASYNC_IO
3039
			os_n_file_writes++;
3040 3041 3042 3043 3044
			ret = WriteFile(file, buf, (DWORD)n, &len,
							&(slot->control));
#elif defined(POSIX_ASYNC_IO)
			slot->control.aio_lio_opcode = LIO_WRITE;
			err = (ulint) aio_write(&(slot->control));
3045
			fprintf(stderr, "Starting POSIX aio write %lu\n", err);
3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059
#endif
		} else {
			if (!wake_later) {
				os_aio_simulated_wake_handler_thread(
				 os_aio_get_segment_no_from_slot(array, slot));
			}
		}
	} else {
		ut_error;
	}

#ifdef WIN_ASYNC_IO
	if (os_aio_use_native_aio) {
		if ((ret && len == n)
3060
		    || (!ret && GetLastError() == ERROR_IO_PENDING)) {
3061 3062 3063 3064 3065 3066 3067
			/* aio was queued successfully! */
		
	    		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 */
	    		
unknown's avatar
unknown committed
3068
	    		    retval = os_aio_windows_handle(ULINT_UNDEFINED,
3069 3070
					slot->pos,
		    			&dummy_mess1, &dummy_mess2,
unknown's avatar
unknown committed
3071 3072 3073
					&dummy_type);

			    return(retval);
3074 3075 3076 3077 3078
	    		}

			return(TRUE);
		}

3079
		err = 1; /* Fall through the next if */
3080 3081 3082 3083 3084 3085 3086 3087 3088 3089
	}
#endif
	if (err == 0) {
		/* aio was queued successfully! */

		return(TRUE);
	}

	os_aio_array_free_slot(array, slot);

unknown's avatar
unknown committed
3090 3091
	retry = os_file_handle_error(file, name,
			type == OS_FILE_READ ? "aio read" : "aio write");
3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127
	if (retry) {

		goto try_again;
	}	

	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:
				wait for the aio slot at this position */  
	void**	message1,	/* out: the messages passed with the aio
				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 */
3128 3129
	void**	message2,
	ulint*	type)		/* out: OS_FILE_WRITE or ..._READ */
3130
{
3131
	ulint		orig_seg	= segment;
3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155
	os_aio_array_t*	array;
	os_aio_slot_t*	slot;
	ulint		n;
	ulint		i;
	ibool		ret_val;
	BOOL		ret;
	DWORD		len;

	if (segment == ULINT_UNDEFINED) {
		array = os_aio_sync_array;
		segment = 0;
	} else {
		segment = os_aio_get_array_and_local_segment(&array, segment);
	}
	
	/* 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) {
3156
		os_event_wait(os_aio_array_get_nth_slot(array, pos)->event);
3157 3158
		i = pos;
	} else {
3159
		srv_set_io_thread_op_info(orig_seg, "wait Windows aio");
3160 3161
		i = os_event_wait_multiple(n,
				(array->native_events) + segment * n);
3162 3163 3164 3165 3166 3167 3168 3169
	}

	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
3170
	if (orig_seg != ULINT_UNDEFINED) {
3171 3172
		srv_set_io_thread_op_info(orig_seg,
					"get windows aio return value");
unknown's avatar
unknown committed
3173 3174
	}

3175 3176 3177 3178 3179
	ret = GetOverlappedResult(slot->file, &(slot->control), &len, TRUE);

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

3180 3181
	*type = slot->type;

3182 3183
	if (ret && len == slot->len) {
		ret_val = TRUE;
3184

unknown's avatar
unknown committed
3185 3186
		if (slot->type == OS_FILE_WRITE
				&& !os_do_not_call_flush_at_each_write) {
3187 3188
		         ut_a(TRUE == os_file_flush(slot->file));
		}
3189
	} else {
unknown's avatar
unknown committed
3190
		os_file_handle_error(slot->file, slot->name, "Windows aio");
unknown's avatar
unknown committed
3191
		
3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240
		ret_val = FALSE;
	}		  

	os_mutex_exit(array->mutex);

	os_aio_array_free_slot(array, slot);
	
	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 */
	void**	message1,	/* out: the messages passed with the aio
				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;
	sigset_t        proc_sigset;
	sigset_t        thr_sigset;
	int		ret;
	int             i;
	int             sig;
	
        sigemptyset(&sigset);
	sigaddset(&sigset, SIGRTMIN + 1 + array_no);

	pthread_sigmask(SIG_UNBLOCK, &sigset, NULL);

	/*
	sigprocmask(0, NULL, &proc_sigset);
	pthread_sigmask(0, NULL, &thr_sigset);

	for (i = 32 ; i < 40; i++) {
3241
		fprintf(stderr, "%lu : %lu %lu\n", (ulint)i,
3242 3243 3244 3245 3246 3247 3248 3249 3250
		 (ulint)sigismember(&proc_sigset, i),
		 (ulint)sigismember(&thr_sigset, i));
	}
	*/

	ret = sigwaitinfo(&sigset, &info);

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

3251
		ut_error;
3252 3253 3254 3255
	
		return(FALSE);
	}
	
3256
	fputs("Handling POSIX aio\n", stderr);
3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268

	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;

unknown's avatar
unknown committed
3269 3270
	if (slot->type == OS_FILE_WRITE
				&& !os_do_not_call_flush_at_each_write) {
3271 3272 3273
		ut_a(TRUE == os_file_flush(slot->file));
	}

3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299
	os_mutex_exit(array->mutex);

	os_aio_array_free_slot(array, slot);
	
	return(TRUE);
}
#endif

/**************************************************************************
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 */
	void**	message1,	/* out: the messages passed with the aio
				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 */
3300 3301
	void**	message2,
	ulint*	type)		/* out: OS_FILE_WRITE or ..._READ */
3302 3303 3304 3305 3306 3307 3308 3309 3310 3311
{
	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
3312 3313
	ulint		biggest_age;
	ulint		age;
3314
	byte*		combined_buf;
unknown's avatar
unknown committed
3315
	byte*		combined_buf2;
3316 3317 3318
	ibool		ret;
	ulint		n;
	ulint		i;
unknown's avatar
unknown committed
3319
	ulint		len2;
unknown's avatar
unknown committed
3320
	
3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333
	segment = os_aio_get_array_and_local_segment(&array, global_segment);
	
restart:
	/* 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;

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

unknown's avatar
unknown committed
3334 3335 3336 3337 3338 3339 3340 3341 3342
	if (array == os_aio_read_array
	    && os_aio_recommend_sleep_for_read_threads) {

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

		goto recommended_sleep;
	}
	
3343 3344 3345 3346 3347 3348 3349 3350 3351 3352
	os_mutex_enter(array->mutex);

	/* Check if there is a slot for which the i/o has already been
	done */
	
	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
3353 3354
			if (os_aio_print_debug) {
				fprintf(stderr,
3355
"InnoDB: i/o for slot %lu already done, returning\n", (ulong) i);
unknown's avatar
unknown committed
3356 3357
			}

3358 3359 3360 3361 3362 3363 3364 3365
			ret = TRUE;
			
			goto slot_io_done;
		}
	}

	n_consecutive = 0;

unknown's avatar
unknown committed
3366 3367 3368
	/* 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. */
3369

unknown's avatar
unknown committed
3370
	biggest_age = 0;
3371
	lowest_offset = ULINT_MAX;
unknown's avatar
unknown committed
3372

3373 3374 3375
	for (i = 0; i < n; i++) {
		slot = os_aio_array_get_nth_slot(array, i + segment * n);

unknown's avatar
unknown committed
3376 3377 3378
		if (slot->reserved) {
		        age = (ulint)difftime(time(NULL),
						slot->reservation_time);
3379

unknown's avatar
unknown committed
3380 3381 3382
			if ((age >= 2 && age > biggest_age)
			    || (age >= 2 && age == biggest_age
			        && slot->offset < lowest_offset)) {
3383

unknown's avatar
unknown committed
3384 3385
			        /* Found an i/o request */
				consecutive_ios[0] = slot;
3386

unknown's avatar
unknown committed
3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404
				n_consecutive = 1;

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

	if (n_consecutive == 0) {
	        /* There were no old requests. Look for an i/o request at the
		lowest offset in the array (we ignore the high 32 bits of the
		offset in these heuristics) */

		lowest_offset = ULINT_MAX;
	
		for (i = 0; i < n; i++) {
		        slot = os_aio_array_get_nth_slot(array,
							i + segment * n);
3405

unknown's avatar
unknown committed
3406
			if (slot->reserved && slot->offset < lowest_offset) {
3407

unknown's avatar
unknown committed
3408 3409 3410 3411 3412 3413 3414
			        /* Found an i/o request */
				consecutive_ios[0] = slot;

				n_consecutive = 1;

				lowest_offset = slot->offset;
			}
3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471
		}
	}

	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 */

consecutive_loop:	
	for (i = 0; i < n; i++) {
		slot2 = os_aio_array_get_nth_slot(array, i + segment * n);

		if (slot2->reserved && slot2 != slot
		    && slot2->offset == slot->offset + slot->len
		    && slot->offset + slot->len > slot->offset /* check that
						sum does not wrap over */
		    && slot2->offset_high == slot->offset_high
		    && slot2->type == slot->type
		    && slot2->file == slot->file) {

			/* Found a consecutive i/o request */

			consecutive_ios[n_consecutive] = slot2;
			n_consecutive++;

			slot = slot2;

			if (n_consecutive < OS_AIO_MERGE_N_CONSECUTIVE) {
			
				goto consecutive_loop;
			} else {
				break;
			}
		}
	}

	/* 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];
	
	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;
	} else {
unknown's avatar
unknown committed
3472
		combined_buf2 = ut_malloc(total_len + UNIV_PAGE_SIZE);
3473

unknown's avatar
unknown committed
3474 3475 3476
		ut_a(combined_buf2);

		combined_buf = ut_align(combined_buf2, UNIV_PAGE_SIZE);
3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495
	}
	
	/* 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;
		
		for (i = 0; i < n_consecutive; i++) {

			ut_memcpy(combined_buf + offs, consecutive_ios[i]->buf,
						consecutive_ios[i]->len);
			offs += consecutive_ios[i]->len;
		}
	}
unknown's avatar
unknown committed
3496
	
3497
	srv_set_io_thread_op_info(global_segment, "doing file i/o");
3498

unknown's avatar
unknown committed
3499 3500 3501
	if (os_aio_print_debug) {
		fprintf(stderr,
"InnoDB: doing i/o of type %lu at offset %lu %lu, length %lu\n",
3502 3503
			(ulong) slot->type, (ulong) slot->offset_high,
			(ulong) slot->offset, (ulong) total_len);
unknown's avatar
unknown committed
3504 3505
	}

3506 3507
	/* Do the i/o with ordinary, synchronous i/o functions: */
	if (slot->type == OS_FILE_WRITE) {
unknown's avatar
unknown committed
3508
		if (array == os_aio_write_array) {
unknown's avatar
unknown committed
3509 3510
			if ((total_len % UNIV_PAGE_SIZE != 0)
			    || (slot->offset % UNIV_PAGE_SIZE != 0)) {
unknown's avatar
unknown committed
3511 3512
				fprintf(stderr,
"InnoDB: Error: trying a displaced write to %s %lu %lu, len %lu\n",
3513 3514 3515
					slot->name, (ulong) slot->offset_high,
					(ulong) slot->offset,
					(ulong) total_len);
3516
				ut_error;
unknown's avatar
unknown committed
3517 3518
			}
			  
unknown's avatar
unknown committed
3519 3520 3521 3522 3523 3524 3525 3526 3527 3528
			/* Do a 'last millisecond' check that the page end
			is sensible; reported page checksum errors from
			Linux seem to wipe over the page end */

			for (len2 = 0; len2 + UNIV_PAGE_SIZE <= total_len;
						len2 += UNIV_PAGE_SIZE) {
				if (mach_read_from_4(combined_buf + len2
						+ FIL_PAGE_LSN + 4)
				    != mach_read_from_4(combined_buf + len2
				    		+ UNIV_PAGE_SIZE
unknown's avatar
unknown committed
3529
				    	- FIL_PAGE_END_LSN_OLD_CHKSUM + 4)) {
unknown's avatar
unknown committed
3530 3531 3532
				    	ut_print_timestamp(stderr);
				    	fprintf(stderr,
"  InnoDB: ERROR: The page to be written seems corrupt!\n");
unknown's avatar
unknown committed
3533
					buf_page_print(combined_buf + len2);
unknown's avatar
unknown committed
3534 3535 3536 3537 3538 3539
				    	fprintf(stderr,
"InnoDB: ERROR: The page to be written seems corrupt!\n");
				}
			}
		}
	
3540 3541 3542 3543 3544 3545 3546 3547
		ret = os_file_write(slot->name, slot->file, combined_buf,
				slot->offset, slot->offset_high, total_len);
	} else {
		ret = os_file_read(slot->file, combined_buf,
				slot->offset, slot->offset_high, total_len);
	}

	ut_a(ret);
3548
	srv_set_io_thread_op_info(global_segment, "file i/o done");
3549

3550 3551 3552
/* fprintf(stderr,
	"aio: %lu consecutive %lu:th segment, first offs %lu blocks\n",
	n_consecutive, global_segment, slot->offset / UNIV_PAGE_SIZE); */
3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566

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

			ut_memcpy(consecutive_ios[i]->buf, combined_buf + offs, 
						consecutive_ios[i]->len);
			offs += consecutive_ios[i]->len;
		}
	}

	if (n_consecutive > 1) {
unknown's avatar
unknown committed
3567
		ut_free(combined_buf2);
3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588
	}

	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 */
	
slot_io_done:

	ut_a(slot->reserved);

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

3589 3590
	*type = slot->type;

3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604
	os_mutex_exit(array->mutex);

	os_aio_array_free_slot(array, slot);
	
	return(ret);

wait_for_io:
	/* We wait here until there again can be i/os in the segment
	of this thread */
	
	os_event_reset(os_aio_segment_wait_events[global_segment]);

	os_mutex_exit(array->mutex);

unknown's avatar
unknown committed
3605
recommended_sleep:
3606
	srv_set_io_thread_op_info(global_segment, "waiting for i/o request");
3607

3608 3609
	os_event_wait(os_aio_segment_wait_events[global_segment]);

unknown's avatar
unknown committed
3610 3611 3612
	if (os_aio_print_debug) {
		fprintf(stderr,
"InnoDB: i/o handler thread for i/o segment %lu wakes up\n",
3613
			(ulong) global_segment);
unknown's avatar
unknown committed
3614 3615
	}
	
3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675
	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;
	
	ut_a(array);

	os_mutex_enter(array->mutex);

	ut_a(array->n_slots > 0);
	ut_a(array->n_segments > 0);
	
	for (i = 0; i < array->n_slots; i++) {
		slot = os_aio_array_get_nth_slot(array, i);
	
		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
3676 3677
os_aio_print(
/*=========*/
3678
	FILE*	file)	/* in: file where to print */
3679 3680 3681 3682
{
	os_aio_array_t*	array;
	os_aio_slot_t*	slot;
	ulint		n_reserved;
3683 3684
	time_t		current_time;
	double		time_elapsed;
unknown's avatar
unknown committed
3685
	double		avg_bytes_read;
3686
	ulint		i;
3687

3688
	for (i = 0; i < srv_n_file_io_threads; i++) {
unknown's avatar
unknown committed
3689
		fprintf(file, "I/O thread %lu state: %s (%s)\n", (ulong) i,
unknown's avatar
unknown committed
3690 3691
					srv_io_thread_op_info[i],
					srv_io_thread_function[i]);
3692 3693
	}

3694
	fputs("Pending normal aio reads:", file);
3695

3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711
	array = os_aio_read_array;
loop:
	ut_a(array);
	
	os_mutex_enter(array->mutex);

	ut_a(array->n_slots > 0);
	ut_a(array->n_segments > 0);
	
	n_reserved = 0;

	for (i = 0; i < array->n_slots; i++) {
		slot = os_aio_array_get_nth_slot(array, i);
	
		if (slot->reserved) {
			n_reserved++;
3712 3713 3714
			/* fprintf(stderr, "Reserved slot, messages %p %p\n",
				slot->message1, slot->message2); */
			ut_a(slot->len > 0);
3715 3716 3717 3718 3719
		}
	}

	ut_a(array->n_reserved == n_reserved);

unknown's avatar
unknown committed
3720
	fprintf(file, " %lu", (ulong) n_reserved);
3721 3722 3723 3724
	
	os_mutex_exit(array->mutex);

	if (array == os_aio_read_array) {
3725
		fputs(", aio writes:", file);
3726
	
3727 3728 3729 3730 3731 3732
		array = os_aio_write_array;

		goto loop;
	}

	if (array == os_aio_write_array) {
3733
		fputs(",\n ibuf aio reads:", file);
3734 3735 3736 3737 3738 3739
		array = os_aio_ibuf_array;

		goto loop;
	}

	if (array == os_aio_ibuf_array) {
3740
		fputs(", log i/o's:", file);
3741 3742 3743 3744 3745 3746
		array = os_aio_log_array;

		goto loop;
	}

	if (array == os_aio_log_array) {
3747
		fputs(", sync i/o's:", file);
3748 3749 3750 3751
		array = os_aio_sync_array;

		goto loop;
	}
3752

3753
	putc('\n', file);
3754
	current_time = time(NULL);
unknown's avatar
unknown committed
3755
	time_elapsed = 0.001 + difftime(current_time, os_last_printout);
3756

3757 3758
	fprintf(file,
		"Pending flushes (fsync) log: %lu; buffer pool: %lu\n"
unknown's avatar
unknown committed
3759
		"%lu OS file reads, %lu OS file writes, %lu OS fsyncs\n",
unknown's avatar
unknown committed
3760 3761
		(ulong) fil_n_pending_log_flushes,
		(ulong) fil_n_pending_tablespace_flushes,
3762 3763
		(ulong) os_n_file_reads, (ulong) os_n_file_writes,
		(ulong) os_n_fsyncs);
unknown's avatar
unknown committed
3764

unknown's avatar
unknown committed
3765
	if (os_file_n_pending_preads != 0 || os_file_n_pending_pwrites != 0) {
3766
		fprintf(file,
unknown's avatar
unknown committed
3767
		    "%lu pending preads, %lu pending pwrites\n",
3768 3769
		    (ulong) os_file_n_pending_preads,
		    (ulong) os_file_n_pending_pwrites);
unknown's avatar
unknown committed
3770 3771
	}

unknown's avatar
unknown committed
3772 3773 3774 3775 3776 3777 3778
	if (os_n_file_reads == os_n_file_reads_old) {
		avg_bytes_read = 0.0;
	} else {
		avg_bytes_read = os_bytes_read_since_printout /
				(os_n_file_reads - os_n_file_reads_old);
	}

3779
	fprintf(file,
unknown's avatar
unknown committed
3780
"%.2f reads/s, %lu avg bytes/read, %.2f writes/s, %.2f fsyncs/s\n",
3781 3782
		(os_n_file_reads - os_n_file_reads_old)
		/ time_elapsed,
3783
		(ulong)avg_bytes_read,
3784 3785 3786 3787 3788 3789 3790 3791
		(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
3792
	os_bytes_read_since_printout = 0;
3793 3794
	
	os_last_printout = current_time;
3795 3796
}

unknown's avatar
unknown committed
3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811
/**************************************************************************
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;
	
	os_last_printout = time(NULL);
}

3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860 3861 3862 3863 3864 3865 3866 3867 3868 3869 3870
/**************************************************************************
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;
	
	array = os_aio_read_array;

	os_mutex_enter(array->mutex);

	n_res += array->n_reserved; 
	
	os_mutex_exit(array->mutex);

	array = os_aio_write_array;

	os_mutex_enter(array->mutex);

	n_res += array->n_reserved; 
	
	os_mutex_exit(array->mutex);

	array = os_aio_ibuf_array;

	os_mutex_enter(array->mutex);

	n_res += array->n_reserved; 
	
	os_mutex_exit(array->mutex);

	array = os_aio_log_array;

	os_mutex_enter(array->mutex);

	n_res += array->n_reserved; 
	
	os_mutex_exit(array->mutex);

	array = os_aio_sync_array;

	os_mutex_enter(array->mutex);

	n_res += array->n_reserved; 
	
	os_mutex_exit(array->mutex);

	if (n_res == 0) {

		return(TRUE);
	}

	return(FALSE);
}