| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* posix_sema.c |
| 4 |
* Implement PGSemaphores using POSIX semaphore facilities |
| 5 |
* |
| 6 |
* We prefer the unnamed style of POSIX semaphore (the kind made with |
| 7 |
* sem_init). We can cope with the kind made with sem_open, however. |
| 8 |
* |
| 9 |
* In either implementation, typedef PGSemaphore is equivalent to "sem_t *". |
| 10 |
* With unnamed semaphores, the sem_t structs live in an array in shared |
| 11 |
* memory. With named semaphores, that's not true because we cannot persuade |
| 12 |
* sem_open to do its allocation there. Therefore, the named-semaphore code |
| 13 |
* *does not cope with EXEC_BACKEND*. The sem_t structs will just be in the |
| 14 |
* postmaster's private memory, where they are successfully inherited by |
| 15 |
* forked backends, but they could not be accessed by exec'd backends. |
| 16 |
* |
| 17 |
* |
| 18 |
* Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group |
| 19 |
* Portions Copyright (c) 1994, Regents of the University of California |
| 20 |
* |
| 21 |
* IDENTIFICATION |
| 22 |
* src/backend/port/posix_sema.c |
| 23 |
* |
| 24 |
*------------------------------------------------------------------------- |
| 25 |
*/ |
| 26 |
#include "postgres.h" |
| 27 |
|
| 28 |
#include <fcntl.h> |
| 29 |
#include <semaphore.h> |
| 30 |
#include <signal.h> |
| 31 |
#include <unistd.h> |
| 32 |
#include <sys/stat.h> |
| 33 |
|
| 34 |
#include "miscadmin.h" |
| 35 |
#include "storage/ipc.h" |
| 36 |
#include "storage/pg_sema.h" |
| 37 |
#include "storage/shmem.h" |
| 38 |
|
| 39 |
|
| 40 |
/* see file header comment */ |
| 41 |
#if defined(USE_NAMED_POSIX_SEMAPHORES) && defined(EXEC_BACKEND) |
| 42 |
#error cannot use named POSIX semaphores with EXEC_BACKEND |
| 43 |
#endif |
| 44 |
|
| 45 |
typedef union SemTPadded |
| 46 |
{ |
| 47 |
sem_t pgsem; |
| 48 |
char pad[PG_CACHE_LINE_SIZE]; |
| 49 |
} SemTPadded; |
| 50 |
|
| 51 |
/* typedef PGSemaphore is equivalent to pointer to sem_t */ |
| 52 |
typedef struct PGSemaphoreData |
| 53 |
{ |
| 54 |
SemTPadded sem_padded; |
| 55 |
} PGSemaphoreData; |
| 56 |
|
| 57 |
#define PG_SEM_REF(x) (&(x)->sem_padded.pgsem) |
| 58 |
|
| 59 |
#define IPCProtection (0600) /* access/modify by user only */ |
| 60 |
|
| 61 |
#ifdef USE_NAMED_POSIX_SEMAPHORES |
| 62 |
static sem_t **mySemPointers; /* keep track of created semaphores */ |
| 63 |
#else |
| 64 |
static PGSemaphore sharedSemas; /* array of PGSemaphoreData in shared memory */ |
| 65 |
#endif |
| 66 |
static int numSems; /* number of semas acquired so far */ |
| 67 |
static int maxSems; /* allocated size of above arrays */ |
| 68 |
static int nextSemKey; /* next name to try */ |
| 69 |
|
| 70 |
|
| 71 |
static void ReleaseSemaphores(int status, Datum arg); |
| 72 |
|
| 73 |
|
| 74 |
#ifdef USE_NAMED_POSIX_SEMAPHORES |
| 75 |
|
| 76 |
/* |
| 77 |
* PosixSemaphoreCreate |
| 78 |
* |
| 79 |
* Attempt to create a new named semaphore. |
| 80 |
* |
| 81 |
* If we fail with a failure code other than collision-with-existing-sema, |
| 82 |
* print out an error and abort. Other types of errors suggest nonrecoverable |
| 83 |
* problems. |
| 84 |
*/ |
| 85 |
static sem_t * |
| 86 |
PosixSemaphoreCreate(void) |
| 87 |
{ |
| 88 |
int semKey; |
| 89 |
char semname[64]; |
| 90 |
sem_t *mySem; |
| 91 |
|
| 92 |
for (;;) |
| 93 |
{ |
| 94 |
semKey = nextSemKey++; |
| 95 |
|
| 96 |
snprintf(semname, sizeof(semname), "/pgsql-%d", semKey); |
| 97 |
|
| 98 |
mySem = sem_open(semname, O_CREAT | O_EXCL, |
| 99 |
(mode_t) IPCProtection, (unsigned) 1); |
| 100 |
|
| 101 |
#ifdef SEM_FAILED |
| 102 |
if (mySem != (sem_t *) SEM_FAILED) |
| 103 |
break; |
| 104 |
#else |
| 105 |
if (mySem != (sem_t *) (-1)) |
| 106 |
break; |
| 107 |
#endif |
| 108 |
|
| 109 |
/* Loop if error indicates a collision */ |
| 110 |
if (errno == EEXIST || errno == EACCES || errno == EINTR) |
| 111 |
continue; |
| 112 |
|
| 113 |
/* |
| 114 |
* Else complain and abort |
| 115 |
*/ |
| 116 |
elog(FATAL, "sem_open(\"%s\") failed: %m", semname); |
| 117 |
} |
| 118 |
|
| 119 |
/* |
| 120 |
* Unlink the semaphore immediately, so it can't be accessed externally. |
| 121 |
* This also ensures that it will go away if we crash. |
| 122 |
*/ |
| 123 |
sem_unlink(semname); |
| 124 |
|
| 125 |
return mySem; |
| 126 |
} |
| 127 |
#else /* !USE_NAMED_POSIX_SEMAPHORES */ |
| 128 |
|
| 129 |
/* |
| 130 |
* PosixSemaphoreCreate |
| 131 |
* |
| 132 |
* Attempt to create a new unnamed semaphore. |
| 133 |
*/ |
| 134 |
static void |
| 135 |
PosixSemaphoreCreate(sem_t *sem) |
| 136 |
{ |
| 137 |
if (sem_init(sem, 1, 1) < 0) |
| 138 |
elog(FATAL, "sem_init failed: %m"); |
| 139 |
} |
| 140 |
#endif /* USE_NAMED_POSIX_SEMAPHORES */ |
| 141 |
|
| 142 |
|
| 143 |
/* |
| 144 |
* PosixSemaphoreKill - removes a semaphore |
| 145 |
*/ |
| 146 |
static void |
| 147 |
PosixSemaphoreKill(sem_t *sem) |
| 148 |
{ |
| 149 |
#ifdef USE_NAMED_POSIX_SEMAPHORES |
| 150 |
/* Got to use sem_close for named semaphores */ |
| 151 |
if (sem_close(sem) < 0) |
| 152 |
elog(LOG, "sem_close failed: %m"); |
| 153 |
#else |
| 154 |
/* Got to use sem_destroy for unnamed semaphores */ |
| 155 |
if (sem_destroy(sem) < 0) |
| 156 |
elog(LOG, "sem_destroy failed: %m"); |
| 157 |
#endif |
| 158 |
} |
| 159 |
|
| 160 |
|
| 161 |
/* |
| 162 |
* Report amount of shared memory needed for semaphores |
| 163 |
*/ |
| 164 |
Size |
| 165 |
PGSemaphoreShmemSize(int maxSemas) |
| 166 |
{ |
| 167 |
#ifdef USE_NAMED_POSIX_SEMAPHORES |
| 168 |
/* No shared memory needed in this case */ |
| 169 |
return 0; |
| 170 |
#else |
| 171 |
/* Need a PGSemaphoreData per semaphore */ |
| 172 |
return mul_size(maxSemas, sizeof(PGSemaphoreData)); |
| 173 |
#endif |
| 174 |
} |
| 175 |
|
| 176 |
/* |
| 177 |
* PGReserveSemaphores --- initialize semaphore support |
| 178 |
* |
| 179 |
* This is called during postmaster start or shared memory reinitialization. |
| 180 |
* It should do whatever is needed to be able to support up to maxSemas |
| 181 |
* subsequent PGSemaphoreCreate calls. Also, if any system resources |
| 182 |
* are acquired here or in PGSemaphoreCreate, register an on_shmem_exit |
| 183 |
* callback to release them. |
| 184 |
* |
| 185 |
* In the Posix implementation, we acquire semaphores on-demand; the |
| 186 |
* maxSemas parameter is just used to size the arrays. For unnamed |
| 187 |
* semaphores, there is an array of PGSemaphoreData structs in shared memory. |
| 188 |
* For named semaphores, we keep a postmaster-local array of sem_t pointers, |
| 189 |
* which we use for releasing the semaphores when done. |
| 190 |
* (This design minimizes the dependency of postmaster shutdown on the |
| 191 |
* contents of shared memory, which a failed backend might have clobbered. |
| 192 |
* We can't do much about the possibility of sem_destroy() crashing, but |
| 193 |
* we don't have to expose the counters to other processes.) |
| 194 |
*/ |
| 195 |
void |
| 196 |
PGReserveSemaphores(int maxSemas) |
| 197 |
{ |
| 198 |
struct stat statbuf; |
| 199 |
|
| 200 |
/* |
| 201 |
* We use the data directory's inode number to seed the search for free |
| 202 |
* semaphore keys. This minimizes the odds of collision with other |
| 203 |
* postmasters, while maximizing the odds that we will detect and clean up |
| 204 |
* semaphores left over from a crashed postmaster in our own directory. |
| 205 |
*/ |
| 206 |
if (stat(DataDir, &statbuf) < 0) |
| 207 |
ereport(FATAL, |
| 208 |
(errcode_for_file_access(), |
| 209 |
errmsg("could not stat data directory \"%s\": %m", |
| 210 |
DataDir))); |
| 211 |
|
| 212 |
#ifdef USE_NAMED_POSIX_SEMAPHORES |
| 213 |
mySemPointers = (sem_t **) malloc(maxSemas * sizeof(sem_t *)); |
| 214 |
if (mySemPointers == NULL) |
| 215 |
elog(PANIC, "out of memory"); |
| 216 |
#else |
| 217 |
|
| 218 |
/* |
| 219 |
* We must use ShmemAllocUnlocked(), since the spinlock protecting |
| 220 |
* ShmemAlloc() won't be ready yet. |
| 221 |
*/ |
| 222 |
sharedSemas = (PGSemaphore) |
| 223 |
ShmemAllocUnlocked(PGSemaphoreShmemSize(maxSemas)); |
| 224 |
#endif |
| 225 |
|
| 226 |
numSems = 0; |
| 227 |
maxSems = maxSemas; |
| 228 |
nextSemKey = statbuf.st_ino; |
| 229 |
|
| 230 |
on_shmem_exit(ReleaseSemaphores, 0); |
| 231 |
} |
| 232 |
|
| 233 |
/* |
| 234 |
* Release semaphores at shutdown or shmem reinitialization |
| 235 |
* |
| 236 |
* (called as an on_shmem_exit callback, hence funny argument list) |
| 237 |
*/ |
| 238 |
static void |
| 239 |
ReleaseSemaphores(int status, Datum arg) |
| 240 |
{ |
| 241 |
int i; |
| 242 |
|
| 243 |
#ifdef USE_NAMED_POSIX_SEMAPHORES |
| 244 |
for (i = 0; i < numSems; i++) |
| 245 |
PosixSemaphoreKill(mySemPointers[i]); |
| 246 |
free(mySemPointers); |
| 247 |
#endif |
| 248 |
|
| 249 |
#ifdef USE_UNNAMED_POSIX_SEMAPHORES |
| 250 |
for (i = 0; i < numSems; i++) |
| 251 |
PosixSemaphoreKill(PG_SEM_REF(sharedSemas + i)); |
| 252 |
#endif |
| 253 |
} |
| 254 |
|
| 255 |
/* |
| 256 |
* PGSemaphoreCreate |
| 257 |
* |
| 258 |
* Allocate a PGSemaphore structure with initial count 1 |
| 259 |
*/ |
| 260 |
PGSemaphore |
| 261 |
PGSemaphoreCreate(void) |
| 262 |
{ |
| 263 |
PGSemaphore sema; |
| 264 |
sem_t *newsem; |
| 265 |
|
| 266 |
/* Can't do this in a backend, because static state is postmaster's */ |
| 267 |
Assert(!IsUnderPostmaster); |
| 268 |
|
| 269 |
if (numSems >= maxSems) |
| 270 |
elog(PANIC, "too many semaphores created"); |
| 271 |
|
| 272 |
#ifdef USE_NAMED_POSIX_SEMAPHORES |
| 273 |
newsem = PosixSemaphoreCreate(); |
| 274 |
/* Remember new sema for ReleaseSemaphores */ |
| 275 |
mySemPointers[numSems] = newsem; |
| 276 |
sema = (PGSemaphore) newsem; |
| 277 |
#else |
| 278 |
sema = &sharedSemas[numSems]; |
| 279 |
newsem = PG_SEM_REF(sema); |
| 280 |
PosixSemaphoreCreate(newsem); |
| 281 |
#endif |
| 282 |
|
| 283 |
numSems++; |
| 284 |
|
| 285 |
return sema; |
| 286 |
} |
| 287 |
|
| 288 |
/* |
| 289 |
* PGSemaphoreReset |
| 290 |
* |
| 291 |
* Reset a previously-initialized PGSemaphore to have count 0 |
| 292 |
*/ |
| 293 |
void |
| 294 |
PGSemaphoreReset(PGSemaphore sema) |
| 295 |
{ |
| 296 |
/* |
| 297 |
* There's no direct API for this in POSIX, so we have to ratchet the |
| 298 |
* semaphore down to 0 with repeated trywait's. |
| 299 |
*/ |
| 300 |
for (;;) |
| 301 |
{ |
| 302 |
if (sem_trywait(PG_SEM_REF(sema)) < 0) |
| 303 |
{ |
| 304 |
if (errno == EAGAIN || errno == EDEADLK) |
| 305 |
break; /* got it down to 0 */ |
| 306 |
if (errno == EINTR) |
| 307 |
continue; /* can this happen? */ |
| 308 |
elog(FATAL, "sem_trywait failed: %m"); |
| 309 |
} |
| 310 |
} |
| 311 |
} |
| 312 |
|
| 313 |
/* |
| 314 |
* PGSemaphoreLock |
| 315 |
* |
| 316 |
* Lock a semaphore (decrement count), blocking if count would be < 0 |
| 317 |
*/ |
| 318 |
void |
| 319 |
PGSemaphoreLock(PGSemaphore sema) |
| 320 |
{ |
| 321 |
int errStatus; |
| 322 |
|
| 323 |
/* See notes in sysv_sema.c's implementation of PGSemaphoreLock. */ |
| 324 |
do |
| 325 |
{ |
| 326 |
errStatus = sem_wait(PG_SEM_REF(sema)); |
| 327 |
} while (errStatus < 0 && errno == EINTR); |
| 328 |
|
| 329 |
if (errStatus < 0) |
| 330 |
elog(FATAL, "sem_wait failed: %m"); |
| 331 |
} |
| 332 |
|
| 333 |
/* |
| 334 |
* PGSemaphoreUnlock |
| 335 |
* |
| 336 |
* Unlock a semaphore (increment count) |
| 337 |
*/ |
| 338 |
void |
| 339 |
PGSemaphoreUnlock(PGSemaphore sema) |
| 340 |
{ |
| 341 |
int errStatus; |
| 342 |
|
| 343 |
/* |
| 344 |
* Note: if errStatus is -1 and errno == EINTR then it means we returned |
| 345 |
* from the operation prematurely because we were sent a signal. So we |
| 346 |
* try and unlock the semaphore again. Not clear this can really happen, |
| 347 |
* but might as well cope. |
| 348 |
*/ |
| 349 |
do |
| 350 |
{ |
| 351 |
errStatus = sem_post(PG_SEM_REF(sema)); |
| 352 |
} while (errStatus < 0 && errno == EINTR); |
| 353 |
|
| 354 |
if (errStatus < 0) |
| 355 |
elog(FATAL, "sem_post failed: %m"); |
| 356 |
} |
| 357 |
|
| 358 |
/* |
| 359 |
* PGSemaphoreTryLock |
| 360 |
* |
| 361 |
* Lock a semaphore only if able to do so without blocking |
| 362 |
*/ |
| 363 |
bool |
| 364 |
PGSemaphoreTryLock(PGSemaphore sema) |
| 365 |
{ |
| 366 |
int errStatus; |
| 367 |
|
| 368 |
/* |
| 369 |
* Note: if errStatus is -1 and errno == EINTR then it means we returned |
| 370 |
* from the operation prematurely because we were sent a signal. So we |
| 371 |
* try and lock the semaphore again. |
| 372 |
*/ |
| 373 |
do |
| 374 |
{ |
| 375 |
errStatus = sem_trywait(PG_SEM_REF(sema)); |
| 376 |
} while (errStatus < 0 && errno == EINTR); |
| 377 |
|
| 378 |
if (errStatus < 0) |
| 379 |
{ |
| 380 |
if (errno == EAGAIN || errno == EDEADLK) |
| 381 |
return false; /* failed to lock it */ |
| 382 |
/* Otherwise we got trouble */ |
| 383 |
elog(FATAL, "sem_trywait failed: %m"); |
| 384 |
} |
| 385 |
|
| 386 |
return true; |
| 387 |
} |