ipc.c 25.9 KB
Newer Older
1
/*-------------------------------------------------------------------------
2
 *
3
 * ipc.c
4
 *	  POSTGRES inter-process communication definitions.
5
 *
B
Add:  
Bruce Momjian 已提交
6 7
 * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
 * Portions Copyright (c) 1994, Regents of the University of California
8 9 10
 *
 *
 * IDENTIFICATION
11
 *	  $Header: /cvsroot/pgsql/src/backend/storage/ipc/ipc.c,v 1.54 2000/11/28 23:27:56 tgl Exp $
12 13 14
 *
 * NOTES
 *
15 16 17 18 19 20 21 22 23 24
 *	  Currently, semaphores are used (my understanding anyway) in two
 *	  different ways:
 *		1. as mutexes on machines that don't have test-and-set (eg.
 *		   mips R3000).
 *		2. for putting processes to sleep when waiting on a lock
 *		   and waking them up when the lock is free.
 *	  The number of semaphores in (1) is fixed and those are shared
 *	  among all backends. In (2), there is 1 semaphore per process and those
 *	  are not shared with anyone else.
 *														  -ay 4/95
25 26 27
 *
 *-------------------------------------------------------------------------
 */
28 29
#include "postgres.h"

30 31 32
#include <sys/types.h>
#include <sys/file.h>
#include <errno.h>
33
#include <unistd.h>
34 35

#include "storage/ipc.h"
36
#include "storage/s_lock.h"
37
/* In Ultrix, sem.h and shm.h must be included AFTER ipc.h */
38
#ifdef HAVE_SYS_SEM_H
39
#include <sys/sem.h>
40 41
#endif
#ifdef HAVE_SYS_SHM_H
42
#include <sys/shm.h>
43 44 45 46
#endif
#ifdef HAVE_KERNEL_OS_H
#include <kernel/OS.h>
#endif
47
#include "miscadmin.h"
48
#include "utils/memutils.h"
49
#include "libpq/libpq.h"
50

B
Bruce Momjian 已提交
51
#if defined(solaris_sparc)
52 53 54
#include <sys/ipc.h>
#endif

55

56 57 58 59 60
/*
 * This flag is set during proc_exit() to change elog()'s behavior,
 * so that an elog() from an on_proc_exit routine cannot get us out
 * of the exit procedure.  We do NOT want to go back to the idle loop...
 */
61
bool		proc_exit_inprogress = false;
62

63 64 65 66 67 68 69 70 71 72
static IpcSemaphoreId InternalIpcSemaphoreCreate(IpcSemaphoreKey semKey,
						   int numSems, int permission,
						   int semStartValue, bool removeOnExit);
static void CallbackSemaphoreKill(int status, Datum semId);
static void *InternalIpcMemoryCreate(IpcMemoryKey memKey, uint32 size,
									 int permission);
static void IpcMemoryDetach(int status, Datum shmaddr);
static void IpcMemoryDelete(int status, Datum shmId);
static void *PrivateMemoryCreate(uint32 size);
static void PrivateMemoryDelete(int status, Datum memaddr);
73

74

75
/* ----------------------------------------------------------------
76
 *						exit() handling stuff
77 78 79 80 81 82 83 84 85 86 87
 *
 * These functions are in generally the same spirit as atexit(2),
 * but provide some additional features we need --- in particular,
 * we want to register callbacks to invoke when we are disconnecting
 * from a broken shared-memory context but not exiting the postmaster.
 *
 * Callback functions can take zero, one, or two args: the first passed
 * arg is the integer exitcode, the second is the Datum supplied when
 * the callback was registered.
 *
 * XXX these functions probably ought to live in some other module.
88 89 90 91 92
 * ----------------------------------------------------------------
 */

#define MAX_ON_EXITS 20

93 94
static struct ONEXIT
{
95
	void		(*function) ();
96
	Datum		arg;
97 98
}			on_proc_exit_list[MAX_ON_EXITS],
			on_shmem_exit_list[MAX_ON_EXITS];
99

100 101
static int	on_proc_exit_index,
			on_shmem_exit_index;
102 103 104


/* ----------------------------------------------------------------
105
 *		proc_exit
106
 *
107 108 109 110
 *		this function calls all the callbacks registered
 *		for it (to free resources) and then calls exit.
 *		This should be the only function to call exit().
 *		-cim 2/6/90
111 112 113
 * ----------------------------------------------------------------
 */
void
114
proc_exit(int code)
115
{
116

117
	/*
118 119
	 * Once we set this flag, we are committed to exit.  Any elog() will
	 * NOT send control back to the main loop, but right back here.
M
 
Marc G. Fournier 已提交
120
	 */
121
	proc_exit_inprogress = true;
B
Bruce Momjian 已提交
122

123 124
	if (DebugLvl > 1)
		elog(DEBUG, "proc_exit(%d)", code);
125

126 127
	/* do our shared memory exits first */
	shmem_exit(code);
128

129 130
	/* ----------------
	 *	call all the callbacks registered before calling exit().
131 132 133 134 135 136
	 *
	 *	Note that since we decrement on_proc_exit_index each time,
	 *	if a callback calls elog(ERROR) or elog(FATAL) then it won't
	 *	be invoked again when control comes back here (nor will the
	 *	previously-completed callbacks).  So, an infinite loop
	 *	should not be possible.
137 138
	 * ----------------
	 */
139 140
	while (--on_proc_exit_index >= 0)
		(*on_proc_exit_list[on_proc_exit_index].function) (code,
141
							  on_proc_exit_list[on_proc_exit_index].arg);
142

143 144
	if (DebugLvl > 1)
		elog(DEBUG, "exit(%d)", code);
145
	exit(code);
146 147 148
}

/* ------------------
149
 * Run all of the on_shmem_exit routines --- but don't actually exit.
150
 * This is used by the postmaster to re-initialize shared memory and
151
 * semaphores after a backend dies horribly.
152 153 154
 * ------------------
 */
void
155
shmem_exit(int code)
156
{
157 158
	if (DebugLvl > 1)
		elog(DEBUG, "shmem_exit(%d)", code);
M
 
Marc G. Fournier 已提交
159

160
	/* ----------------
161 162 163 164
	 *	call all the registered callbacks.
	 *
	 *	As with proc_exit(), we remove each callback from the list
	 *	before calling it, to avoid infinite loop in case of error.
165 166
	 * ----------------
	 */
167 168
	while (--on_shmem_exit_index >= 0)
		(*on_shmem_exit_list[on_shmem_exit_index].function) (code,
169
							on_shmem_exit_list[on_shmem_exit_index].arg);
170 171 172

	on_shmem_exit_index = 0;
}
173

174 175 176 177 178 179 180
/* ----------------------------------------------------------------
 *		on_proc_exit
 *
 *		this function adds a callback function to the list of
 *		functions invoked by proc_exit().	-cim 2/6/90
 * ----------------------------------------------------------------
 */
181
void
182
on_proc_exit(void (*function) (), Datum arg)
183 184
{
	if (on_proc_exit_index >= MAX_ON_EXITS)
185
		elog(FATAL, "Out of on_proc_exit slots");
186 187 188 189 190

	on_proc_exit_list[on_proc_exit_index].function = function;
	on_proc_exit_list[on_proc_exit_index].arg = arg;

	++on_proc_exit_index;
191 192 193
}

/* ----------------------------------------------------------------
194
 *		on_shmem_exit
195
 *
196
 *		this function adds a callback function to the list of
197
 *		functions invoked by shmem_exit().	-cim 2/6/90
198 199
 * ----------------------------------------------------------------
 */
200
void
201
on_shmem_exit(void (*function) (), Datum arg)
202
{
203
	if (on_shmem_exit_index >= MAX_ON_EXITS)
204
		elog(FATAL, "Out of on_shmem_exit slots");
205

206 207
	on_shmem_exit_list[on_shmem_exit_index].function = function;
	on_shmem_exit_list[on_shmem_exit_index].arg = arg;
208

209
	++on_shmem_exit_index;
210 211
}

212
/* ----------------------------------------------------------------
213
 *		on_exit_reset
214
 *
215 216 217 218
 *		this function clears all on_proc_exit() and on_shmem_exit()
 *		registered functions.  This is used just after forking a backend,
 *		so that the backend doesn't believe it should call the postmaster's
 *		on-exit routines when it exits...
219 220 221
 * ----------------------------------------------------------------
 */
void
222
on_exit_reset(void)
223
{
224
	on_shmem_exit_index = 0;
225
	on_proc_exit_index = 0;
226 227
}

228

229 230
/* ----------------------------------------------------------------
 *						Semaphore support
231
 *
232 233 234
 * These routines represent a fairly thin layer on top of SysV semaphore
 * functionality.
 * ----------------------------------------------------------------
235 236
 */

237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253
/* ----------------------------------------------------------------
 *	InternalIpcSemaphoreCreate(semKey, numSems, permission,
 *							   semStartValue, removeOnExit)
 *
 * Attempt to create a new semaphore set with the specified key.
 * Will fail (return -1) if such a set already exists.
 * On success, a callback is optionally registered with on_shmem_exit
 * to delete the semaphore set when on_shmem_exit is called.
 *
 * If we fail with a failure code other than collision-with-existing-set,
 * print out an error and abort.  Other types of errors are not recoverable.
 * ----------------------------------------------------------------
 */
static IpcSemaphoreId
InternalIpcSemaphoreCreate(IpcSemaphoreKey semKey,
						   int numSems, int permission,
						   int semStartValue, bool removeOnExit)
254
{
255
	int			semId;
256 257 258
	int			i;
	u_short		array[IPC_NMAXSEM];
	union semun semun;
259

260
	Assert(numSems > 0 && numSems <= IPC_NMAXSEM);
261

262
	semId = semget(semKey, numSems, IPC_CREAT | IPC_EXCL | permission);
263

264
	if (semId < 0)
265
	{
266 267 268 269 270 271 272 273 274 275 276 277 278
		/*
		 * Fail quietly if error indicates a collision with existing set.
		 * One would expect EEXIST, given that we said IPC_EXCL, but perhaps
		 * we could get a permission violation instead?
		 */
		if (errno == EEXIST || errno == EACCES)
			return -1;
		/*
		 * Else complain and abort
		 */
		fprintf(stderr, "IpcSemaphoreCreate: semget(key=%d, num=%d, 0%o) failed: %s\n",
				(int) semKey, numSems, (IPC_CREAT|IPC_EXCL|permission),
				strerror(errno));
279

280 281 282 283 284 285 286 287
		if (errno == ENOSPC)
			fprintf(stderr,
					"\nThis error does *not* mean that you have run out of disk space.\n\n"
					"It occurs either because system limit for the maximum number of\n"
					"semaphore sets (SEMMNI), or the system wide maximum number of\n"
					"semaphores (SEMMNS), would be exceeded.  You need to raise the\n"
					"respective kernel parameter.  Look into the PostgreSQL documentation\n"
					"for details.\n\n");
288

289 290
		proc_exit(1);
	}
291

292 293 294 295 296 297 298 299
	/* Initialize new semas to specified start value */
	for (i = 0; i < numSems; i++)
		array[i] = semStartValue;
	semun.array = array;
	if (semctl(semId, 0, SETALL, semun) < 0)
	{
		fprintf(stderr, "IpcSemaphoreCreate: semctl(id=%d, 0, SETALL, ...) failed: %s\n",
				semId, strerror(errno));
300

301 302 303 304 305
		if (errno == ERANGE)
			fprintf(stderr,
					"You possibly need to raise your kernel's SEMVMX value to be at least\n"
					"%d.  Look into the PostgreSQL documentation for details.\n",
					semStartValue);
306

307 308
		IpcSemaphoreKill(semId);
		proc_exit(1);
309 310
	}

311 312 313
	/* Register on-exit routine to delete the new set */
	if (removeOnExit)
		on_shmem_exit(CallbackSemaphoreKill, Int32GetDatum(semId));
314

315
	return semId;
316 317 318
}

/****************************************************************************/
319
/*	 IpcSemaphoreKill(semId)	- removes a semaphore set					*/
320
/*																			*/
321 322
/****************************************************************************/
void
323
IpcSemaphoreKill(IpcSemaphoreId semId)
324
{
325
	union semun semun;
326

327
	semun.val = 0;		/* unused, but keep compiler quiet */
328

329 330
	if (semctl(semId, 0, IPC_RMID, semun) < 0)
		fprintf(stderr, "IpcSemaphoreKill: semctl(%d, 0, IPC_RMID, ...) failed: %s\n",
331
				semId, strerror(errno));
332 333 334
	/* We used to report a failure via elog(NOTICE), but that's pretty
	 * pointless considering any client has long since disconnected ...
	 */
335
}
336

337
/****************************************************************************/
338 339
/*	 CallbackSemaphoreKill(status, semId)									*/
/*	(called as an on_shmem_exit callback, hence funny argument list)		*/
340
/****************************************************************************/
341 342
static void
CallbackSemaphoreKill(int status, Datum semId)
343
{
344
	IpcSemaphoreKill(DatumGetInt32(semId));
345 346 347
}

/****************************************************************************/
348
/*	 IpcSemaphoreLock(semId, sem) - locks a semaphore						*/
349 350
/****************************************************************************/
void
351
IpcSemaphoreLock(IpcSemaphoreId semId, int sem)
352
{
353 354
	int			errStatus;
	struct sembuf sops;
355

356
	sops.sem_op = -1;			/* decrement */
357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372
	sops.sem_flg = 0;
	sops.sem_num = sem;

	/* ----------------
	 *	Note: if errStatus is -1 and errno == EINTR then it means we
	 *		  returned from the operation prematurely because we were
	 *		  sent a signal.  So we try and lock the semaphore again.
	 * ----------------
	 */
	do
	{
		errStatus = semop(semId, &sops, 1);
	} while (errStatus == -1 && errno == EINTR);

	if (errStatus == -1)
	{
373 374
        fprintf(stderr, "IpcSemaphoreLock: semop(id=%d) failed: %s\n",
				semId, strerror(errno));
375
		proc_exit(255);
376
	}
377 378 379
}

/****************************************************************************/
380
/*	 IpcSemaphoreUnlock(semId, sem)		- unlocks a semaphore				*/
381 382
/****************************************************************************/
void
383
IpcSemaphoreUnlock(IpcSemaphoreId semId, int sem)
384
{
385 386
	int			errStatus;
	struct sembuf sops;
387

388
	sops.sem_op = 1;			/* increment */
389 390 391 392 393 394 395
	sops.sem_flg = 0;
	sops.sem_num = sem;


	/* ----------------
	 *	Note: if errStatus is -1 and errno == EINTR then it means we
	 *		  returned from the operation prematurely because we were
396 397
	 *		  sent a signal.  So we try and unlock the semaphore again.
	 *		  Not clear this can really happen, but might as well cope.
398 399 400 401 402 403 404 405 406
	 * ----------------
	 */
	do
	{
		errStatus = semop(semId, &sops, 1);
	} while (errStatus == -1 && errno == EINTR);

	if (errStatus == -1)
	{
407 408
		fprintf(stderr, "IpcSemaphoreUnlock: semop(id=%d) failed: %s\n",
				semId, strerror(errno));
409
		proc_exit(255);
410
	}
411 412
}

413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458
/****************************************************************************/
/*	 IpcSemaphoreTryLock(semId, sem)	- conditionally locks a semaphore	*/
/* Lock the semaphore if it's free, but don't block.						*/
/****************************************************************************/
bool
IpcSemaphoreTryLock(IpcSemaphoreId semId, int sem)
{
	int			errStatus;
	struct sembuf sops;

	sops.sem_op = -1;			/* decrement */
	sops.sem_flg = IPC_NOWAIT;	/* but don't block */
	sops.sem_num = sem;

	/* ----------------
	 *	Note: if errStatus is -1 and errno == EINTR then it means we
	 *		  returned from the operation prematurely because we were
	 *		  sent a signal.  So we try and lock the semaphore again.
	 * ----------------
	 */
	do
	{
		errStatus = semop(semId, &sops, 1);
	} while (errStatus == -1 && errno == EINTR);

	if (errStatus == -1)
	{
		/* Expect EAGAIN or EWOULDBLOCK (platform-dependent) */
#ifdef EAGAIN
		if (errno == EAGAIN)
			return false;		/* failed to lock it */
#endif
#if defined(EWOULDBLOCK) && (!defined(EAGAIN) || (EWOULDBLOCK != EAGAIN))
		if (errno == EWOULDBLOCK)
			return false;		/* failed to lock it */
#endif
		/* Otherwise we got trouble */
        fprintf(stderr, "IpcSemaphoreTryLock: semop(id=%d) failed: %s\n",
				semId, strerror(errno));
		proc_exit(255);
	}

	return true;
}

/* Get the current value (semval) of the semaphore */
459
int
460
IpcSemaphoreGetValue(IpcSemaphoreId semId, int sem)
461
{
462
	union semun dummy;			/* for Solaris */
463
	dummy.val = 0;		/* unused */
464

465
	return semctl(semId, sem, GETVAL, dummy);
466 467
}

468 469 470
/* Get the PID of the last process to do semop() on the semaphore */
static pid_t
IpcSemaphoreGetLastPID(IpcSemaphoreId semId, int sem)
471
{
472
	union semun dummy;			/* for Solaris */
473
	dummy.val = 0;		/* unused */
474

475
	return semctl(semId, sem, GETPID, dummy);
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
/* ----------------------------------------------------------------
 *						Shared memory support
 *
 * These routines represent a fairly thin layer on top of SysV shared
 * memory functionality.
 * ----------------------------------------------------------------
 */

/* ----------------------------------------------------------------
 *	InternalIpcMemoryCreate(memKey, size, permission)
 *
 * Attempt to create a new shared memory segment with the specified key.
 * Will fail (return NULL) if such a segment already exists.  If successful,
 * attach the segment to the current process and return its attached address.
 * On success, callbacks are registered with on_shmem_exit to detach and
 * delete the segment when on_shmem_exit is called.
 *
 * If we fail with a failure code other than collision-with-existing-segment,
 * print out an error and abort.  Other types of errors are not recoverable.
 * ----------------------------------------------------------------
 */
static void *
InternalIpcMemoryCreate(IpcMemoryKey memKey, uint32 size, int permission)
502
{
503
	IpcMemoryId shmid;
504
	void	   *memAddress;
505

506
	shmid = shmget(memKey, size, IPC_CREAT | IPC_EXCL | permission);
507 508 509

	if (shmid < 0)
	{
510 511 512 513 514 515 516 517 518 519 520 521
		/*
		 * Fail quietly if error indicates a collision with existing segment.
		 * One would expect EEXIST, given that we said IPC_EXCL, but perhaps
		 * we could get a permission violation instead?
		 */
		if (errno == EEXIST || errno == EACCES)
			return NULL;
		/*
		 * Else complain and abort
		 */
		fprintf(stderr, "IpcMemoryCreate: shmget(key=%d, size=%u, 0%o) failed: %s\n",
				(int) memKey, size, (IPC_CREAT | IPC_EXCL | permission),
522 523 524 525 526 527 528
				strerror(errno));

		if (errno == EINVAL)
			fprintf(stderr,
					"\nThis error can be caused by one of three things:\n\n"
					"1. The maximum size for shared memory segments on your system was\n"
					"   exceeded.  You need to raise the SHMMAX parameter in your kernel\n"
529
					"   to be at least %u bytes.\n\n"
530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548
					"2. The requested shared memory segment was too small for your system.\n"
					"   You need to lower the SHMMIN parameter in your kernel.\n\n"
					"3. The requested shared memory segment already exists but is of the\n"
					"   wrong size.  This is most likely the case if an old version of\n"
					"   PostgreSQL crashed and didn't clean up.  The `ipcclean' utility\n"
					"   can be used to remedy this.\n\n"
					"The PostgreSQL Administrator's Guide contains more information about\n"
					"shared memory configuration.\n\n",
					size);

		else if (errno == ENOSPC)
			fprintf(stderr,
					"\nThis error does *not* mean that you have run out of disk space.\n\n"
					"It occurs either if all available shared memory ids have been taken,\n"
					"in which case you need to raise the SHMMNI parameter in your kernel,\n"
					"or because the system's overall limit for shared memory has been\n"
					"reached.  The PostgreSQL Administrator's Guide contains more\n"
					"information about shared memory configuration.\n\n");

549
		proc_exit(1);
550 551
	}

552 553
	/* Register on-exit routine to delete the new segment */
	on_shmem_exit(IpcMemoryDelete, Int32GetDatum(shmid));
554

555 556
	/* OK, should be able to attach to the segment */
	memAddress = shmat(shmid, 0, 0);
557

558
	if (memAddress == (void *) -1)
559
	{
560 561 562
        fprintf(stderr, "IpcMemoryCreate: shmat(id=%d) failed: %s\n",
				shmid, strerror(errno));
		proc_exit(1);
563 564
	}

565 566 567 568
	/* Register on-exit routine to detach new segment before deleting */
	on_shmem_exit(IpcMemoryDetach, PointerGetDatum(memAddress));

	return memAddress;
569 570 571
}

/****************************************************************************/
572
/*	IpcMemoryDetach(status, shmaddr)	removes a shared memory segment		*/
573 574
/*										from process' address spaceq		*/
/*	(called as an on_shmem_exit callback, hence funny argument list)		*/
575
/****************************************************************************/
576
static void
577
IpcMemoryDetach(int status, Datum shmaddr)
578
{
579 580 581 582 583 584
	if (shmdt(DatumGetPointer(shmaddr)) < 0)
		fprintf(stderr, "IpcMemoryDetach: shmdt(%p) failed: %s\n",
				DatumGetPointer(shmaddr), strerror(errno));
	/* We used to report a failure via elog(NOTICE), but that's pretty
	 * pointless considering any client has long since disconnected ...
	 */
585 586 587
}

/****************************************************************************/
588 589
/*	IpcMemoryDelete(status, shmId)		deletes a shared memory segment		*/
/*	(called as an on_shmem_exit callback, hence funny argument list)		*/
590
/****************************************************************************/
591 592
static void
IpcMemoryDelete(int status, Datum shmId)
593
{
594 595 596 597 598 599 600
	if (shmctl(DatumGetInt32(shmId), IPC_RMID, (struct shmid_ds *) NULL) < 0)
		fprintf(stderr, "IpcMemoryDelete: shmctl(%d, %d, 0) failed: %s\n",
				DatumGetInt32(shmId), IPC_RMID, strerror(errno));
	/* We used to report a failure via elog(NOTICE), but that's pretty
	 * pointless considering any client has long since disconnected ...
	 */
}
601

602 603 604 605 606 607 608 609
/* ----------------------------------------------------------------
 *						private memory support
 *
 * Rather than allocating shmem segments with IPC_PRIVATE key, we
 * just malloc() the requested amount of space.  This code emulates
 * the needed shmem functions.
 * ----------------------------------------------------------------
 */
610

611 612 613 614 615 616 617
static void *
PrivateMemoryCreate(uint32 size)
{
	void	   *memAddress;

	memAddress = malloc(size);
	if (!memAddress)
618
	{
619 620
		fprintf(stderr, "PrivateMemoryCreate: malloc(%u) failed\n", size);
		proc_exit(1);
621
	}
622
	MemSet(memAddress, 0, size);		/* keep Purify quiet */
623

624 625
	/* Register on-exit routine to release storage */
	on_shmem_exit(PrivateMemoryDelete, PointerGetDatum(memAddress));
626

627
	return memAddress;
628 629
}

630 631
static void
PrivateMemoryDelete(int status, Datum memaddr)
632
{
633
	free(DatumGetPointer(memaddr));
634
}
635

636

637
/* ------------------
638 639 640 641
 *				Routines to assign keys for new IPC objects
 *
 * The idea here is to detect and re-use keys that may have been assigned
 * by a crashed postmaster or backend.
642 643
 * ------------------
 */
644

645 646
static IpcMemoryKey NextShmemSegID = 0;
static IpcSemaphoreKey NextSemaID = 0;
647

648 649 650 651 652 653 654 655 656 657
/*
 * (Re) initialize key assignment at startup of postmaster or standalone
 * backend, also at postmaster reset.
 */
void
IpcInitKeyAssignment(int port)
{
	NextShmemSegID = port * 1000;
	NextSemaID = port * 1000;
}
658

659 660 661 662 663 664 665 666 667 668
/*
 * Create a shared memory segment of the given size and initialize its
 * standard header.  Dead Postgres segments are recycled if found,
 * but we do not fail upon collision with non-Postgres shmem segments.
 */
PGShmemHeader *
IpcMemoryCreate(uint32 size, bool private, int permission)
{
	void   *memAddress;
	PGShmemHeader *hdr;
669

670 671
	/* Room for a header? */
	Assert(size > MAXALIGN(sizeof(PGShmemHeader)));
672

673 674
	/* Loop till we find a free IPC key */
	for (NextShmemSegID++ ; ; NextShmemSegID++)
675
	{
676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736
		IpcMemoryId shmid;

		/* Special case if creating a private segment --- just malloc() it */
		if (private)
		{
			memAddress = PrivateMemoryCreate(size);
			break;
		}

		/* Try to create new segment */
		memAddress = InternalIpcMemoryCreate(NextShmemSegID, size, permission);
		if (memAddress)
			break;				/* successful create and attach */

		/* See if it looks to be leftover from a dead Postgres process */
		shmid = shmget(NextShmemSegID, sizeof(PGShmemHeader), 0);
		if (shmid < 0)
			continue;			/* failed: must be some other app's */
		memAddress = shmat(shmid, 0, 0);
		if (memAddress == (void *) -1)
			continue;			/* failed: must be some other app's */
		hdr = (PGShmemHeader *) memAddress;
		if (hdr->magic != PGShmemMagic)
		{
			shmdt(memAddress);
			continue;			/* segment belongs to a non-Postgres app */
		}
		/*
		 * If the creator PID is my own PID or does not belong to any
		 * extant process, it's safe to zap it.
		 */
		if (hdr->creatorPID != getpid())
		{
			if (kill(hdr->creatorPID, 0) == 0 ||
				errno != ESRCH)
			{
				shmdt(memAddress);
				continue;		/* segment belongs to a live process */
			}
		}
		/*
		 * The segment appears to be from a dead Postgres process, or
		 * from a previous cycle of life in this same process.  Zap it,
		 * if possible.  This probably shouldn't fail, but if it does,
		 * assume the segment belongs to someone else after all,
		 * and continue quietly.
		 */
		shmdt(memAddress);
		if (shmctl(shmid, IPC_RMID, (struct shmid_ds *) NULL) < 0)
			continue;
		/*
		 * Now try again to create the segment.
		 */
		memAddress = InternalIpcMemoryCreate(NextShmemSegID, size, permission);
		if (memAddress)
			break;				/* successful create and attach */
		/*
		 * Can only get here if some other process managed to create the
		 * same shmem key before we did.  Let him have that one,
		 * loop around to try next key.
		 */
737
	}
738 739 740 741 742 743 744 745 746 747 748 749 750 751
	/*
	 * OK, we created a new segment.  Mark it as created by this process.
	 * The order of assignments here is critical so that another Postgres
	 * process can't see the header as valid but belonging to an invalid
	 * PID!
	 */
	hdr = (PGShmemHeader *) memAddress;
	hdr->creatorPID = getpid();
	hdr->magic = PGShmemMagic;
	/*
	 * Initialize space allocation status for segment.
	 */
	hdr->totalsize = size;
	hdr->freeoffset = MAXALIGN(sizeof(PGShmemHeader));
752

753
	return hdr;
754 755
}

756 757 758 759 760 761 762 763 764
/*
 * Create a semaphore set with the given number of useful semaphores
 * (an additional sema is actually allocated to serve as identifier).
 * Dead Postgres sema sets are recycled if found, but we do not fail
 * upon collision with non-Postgres sema sets.
 */
IpcSemaphoreId
IpcSemaphoreCreate(int numSems, int permission,
				   int semStartValue, bool removeOnExit)
765
{
766 767
	IpcSemaphoreId	semId;
	union semun semun;
768

769 770 771 772 773 774 775 776 777 778 779
	/* Loop till we find a free IPC key */
	for (NextSemaID++ ; ; NextSemaID++)
	{
		pid_t	creatorPID;

		/* Try to create new semaphore set */
		semId = InternalIpcSemaphoreCreate(NextSemaID, numSems+1,
										   permission, semStartValue,
										   removeOnExit);
		if (semId >= 0)
			break;				/* successful create */
780

781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847
		/* See if it looks to be leftover from a dead Postgres process */
		semId = semget(NextSemaID, numSems+1, 0);
		if (semId < 0)
			continue;			/* failed: must be some other app's */
		if (IpcSemaphoreGetValue(semId, numSems) != PGSemaMagic)
			continue;			/* sema belongs to a non-Postgres app */
		/*
		 * If the creator PID is my own PID or does not belong to any
		 * extant process, it's safe to zap it.
		 */
		creatorPID = IpcSemaphoreGetLastPID(semId, numSems);
		if (creatorPID <= 0)
			continue;			/* oops, GETPID failed */
		if (creatorPID != getpid())
		{
			if (kill(creatorPID, 0) == 0 ||
				errno != ESRCH)
				continue;		/* sema belongs to a live process */
		}
		/*
		 * The sema set appears to be from a dead Postgres process, or
		 * from a previous cycle of life in this same process.  Zap it,
		 * if possible.  This probably shouldn't fail, but if it does,
		 * assume the sema set belongs to someone else after all,
		 * and continue quietly.
		 */
		semun.val = 0;			/* unused, but keep compiler quiet */
		if (semctl(semId, 0, IPC_RMID, semun) < 0)
			continue;
		/*
		 * Now try again to create the sema set.
		 */
		semId = InternalIpcSemaphoreCreate(NextSemaID, numSems+1,
										   permission, semStartValue,
										   removeOnExit);
		if (semId >= 0)
			break;				/* successful create */
		/*
		 * Can only get here if some other process managed to create the
		 * same sema key before we did.  Let him have that one,
		 * loop around to try next key.
		 */
	}
	/*
	 * OK, we created a new sema set.  Mark it as created by this process.
	 * We do this by setting the spare semaphore to PGSemaMagic-1 and then
	 * incrementing it with semop().  That leaves it with value PGSemaMagic
	 * and sempid referencing this process.
	 */
	semun.val = PGSemaMagic-1;
	if (semctl(semId, numSems, SETVAL, semun) < 0)
	{
		fprintf(stderr, "IpcSemaphoreCreate: semctl(id=%d, %d, SETVAL, %d) failed: %s\n",
				semId, numSems, PGSemaMagic-1, strerror(errno));

		if (errno == ERANGE)
			fprintf(stderr,
					"You possibly need to raise your kernel's SEMVMX value to be at least\n"
					"%d.  Look into the PostgreSQL documentation for details.\n",
					PGSemaMagic);

		proc_exit(1);
	}
	IpcSemaphoreUnlock(semId, numSems);

	return semId;
}