Line data Source code
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) 1996-2026, 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 162082 : PosixSemaphoreCreate(sem_t *sem)
136 : {
137 162082 : if (sem_init(sem, 1, 1) < 0)
138 0 : elog(FATAL, "sem_init failed: %m");
139 162082 : }
140 : #endif /* USE_NAMED_POSIX_SEMAPHORES */
141 :
142 :
143 : /*
144 : * PosixSemaphoreKill - removes a semaphore
145 : */
146 : static void
147 162082 : 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 162082 : if (sem_destroy(sem) < 0)
156 0 : elog(LOG, "sem_destroy failed: %m");
157 : #endif
158 162082 : }
159 :
160 :
161 : /*
162 : * Request shared memory needed for semaphores
163 : */
164 : void
165 1234 : PGSemaphoreShmemRequest(int maxSemas)
166 : {
167 : #ifdef USE_NAMED_POSIX_SEMAPHORES
168 : /* No shared memory needed in this case */
169 : #else
170 : /* Need a PGSemaphoreData per semaphore */
171 1234 : ShmemRequestStruct(.name = "Semaphores",
172 : .size = mul_size(maxSemas, sizeof(PGSemaphoreData)),
173 : .ptr = (void **) &sharedSemas,
174 : );
175 : #endif
176 1234 : }
177 :
178 : /*
179 : * PGSemaphoreInit --- initialize semaphore support
180 : *
181 : * This is called during postmaster start or shared memory reinitialization.
182 : * It should do whatever is needed to be able to support up to maxSemas
183 : * subsequent PGSemaphoreCreate calls. Also, if any system resources
184 : * are acquired here or in PGSemaphoreCreate, register an on_shmem_exit
185 : * callback to release them.
186 : *
187 : * In the Posix implementation, we acquire semaphores on-demand; the
188 : * maxSemas parameter is just used to size the arrays. For unnamed
189 : * semaphores, there is an array of PGSemaphoreData structs in shared memory.
190 : * For named semaphores, we keep a postmaster-local array of sem_t pointers,
191 : * which we use for releasing the semaphores when done.
192 : * (This design minimizes the dependency of postmaster shutdown on the
193 : * contents of shared memory, which a failed backend might have clobbered.
194 : * We can't do much about the possibility of sem_destroy() crashing, but
195 : * we don't have to expose the counters to other processes.)
196 : */
197 : void
198 1231 : PGSemaphoreInit(int maxSemas)
199 : {
200 : struct stat statbuf;
201 :
202 : /*
203 : * We use the data directory's inode number to seed the search for free
204 : * semaphore keys. This minimizes the odds of collision with other
205 : * postmasters, while maximizing the odds that we will detect and clean up
206 : * semaphores left over from a crashed postmaster in our own directory.
207 : */
208 1231 : if (stat(DataDir, &statbuf) < 0)
209 0 : ereport(FATAL,
210 : (errcode_for_file_access(),
211 : errmsg("could not stat data directory \"%s\": %m",
212 : DataDir)));
213 :
214 : #ifdef USE_NAMED_POSIX_SEMAPHORES
215 : mySemPointers = (sem_t **) malloc(maxSemas * sizeof(sem_t *));
216 : if (mySemPointers == NULL)
217 : elog(PANIC, "out of memory");
218 : #endif
219 :
220 1231 : numSems = 0;
221 1231 : maxSems = maxSemas;
222 1231 : nextSemKey = statbuf.st_ino;
223 :
224 1231 : on_shmem_exit(ReleaseSemaphores, 0);
225 1231 : }
226 :
227 : /*
228 : * Release semaphores at shutdown or shmem reinitialization
229 : *
230 : * (called as an on_shmem_exit callback, hence funny argument list)
231 : */
232 : static void
233 1231 : ReleaseSemaphores(int status, Datum arg)
234 : {
235 : int i;
236 :
237 : #ifdef USE_NAMED_POSIX_SEMAPHORES
238 : for (i = 0; i < numSems; i++)
239 : PosixSemaphoreKill(mySemPointers[i]);
240 : free(mySemPointers);
241 : #endif
242 :
243 : #ifdef USE_UNNAMED_POSIX_SEMAPHORES
244 163313 : for (i = 0; i < numSems; i++)
245 162082 : PosixSemaphoreKill(PG_SEM_REF(sharedSemas + i));
246 : #endif
247 1231 : }
248 :
249 : /*
250 : * PGSemaphoreCreate
251 : *
252 : * Allocate a PGSemaphore structure with initial count 1
253 : */
254 : PGSemaphore
255 162082 : PGSemaphoreCreate(void)
256 : {
257 : PGSemaphore sema;
258 : sem_t *newsem;
259 :
260 : /* Can't do this in a backend, because static state is postmaster's */
261 : Assert(!IsUnderPostmaster);
262 :
263 162082 : if (numSems >= maxSems)
264 0 : elog(PANIC, "too many semaphores created");
265 :
266 : #ifdef USE_NAMED_POSIX_SEMAPHORES
267 : newsem = PosixSemaphoreCreate();
268 : /* Remember new sema for ReleaseSemaphores */
269 : mySemPointers[numSems] = newsem;
270 : sema = (PGSemaphore) newsem;
271 : #else
272 162082 : sema = &sharedSemas[numSems];
273 162082 : newsem = PG_SEM_REF(sema);
274 162082 : PosixSemaphoreCreate(newsem);
275 : #endif
276 :
277 162082 : numSems++;
278 :
279 162082 : return sema;
280 : }
281 :
282 : /*
283 : * PGSemaphoreReset
284 : *
285 : * Reset a previously-initialized PGSemaphore to have count 0
286 : */
287 : void
288 24748 : PGSemaphoreReset(PGSemaphore sema)
289 : {
290 : /*
291 : * There's no direct API for this in POSIX, so we have to ratchet the
292 : * semaphore down to 0 with repeated trywait's.
293 : */
294 : for (;;)
295 : {
296 40163 : if (sem_trywait(PG_SEM_REF(sema)) < 0)
297 : {
298 24748 : if (errno == EAGAIN || errno == EDEADLK)
299 : break; /* got it down to 0 */
300 0 : if (errno == EINTR)
301 0 : continue; /* can this happen? */
302 0 : elog(FATAL, "sem_trywait failed: %m");
303 : }
304 : }
305 24748 : }
306 :
307 : /*
308 : * PGSemaphoreLock
309 : *
310 : * Lock a semaphore (decrement count), blocking if count would be < 0
311 : */
312 : void
313 5647775 : PGSemaphoreLock(PGSemaphore sema)
314 : {
315 : int errStatus;
316 :
317 : /* See notes in sysv_sema.c's implementation of PGSemaphoreLock. */
318 : do
319 : {
320 5647775 : errStatus = sem_wait(PG_SEM_REF(sema));
321 5647775 : } while (errStatus < 0 && errno == EINTR);
322 :
323 5647775 : if (errStatus < 0)
324 0 : elog(FATAL, "sem_wait failed: %m");
325 5647775 : }
326 :
327 : /*
328 : * PGSemaphoreUnlock
329 : *
330 : * Unlock a semaphore (increment count)
331 : */
332 : void
333 5647911 : PGSemaphoreUnlock(PGSemaphore sema)
334 : {
335 : int errStatus;
336 :
337 : /*
338 : * Note: if errStatus is -1 and errno == EINTR then it means we returned
339 : * from the operation prematurely because we were sent a signal. So we
340 : * try and unlock the semaphore again. Not clear this can really happen,
341 : * but might as well cope.
342 : */
343 : do
344 : {
345 5647911 : errStatus = sem_post(PG_SEM_REF(sema));
346 5647911 : } while (errStatus < 0 && errno == EINTR);
347 :
348 5647911 : if (errStatus < 0)
349 0 : elog(FATAL, "sem_post failed: %m");
350 5647911 : }
351 :
352 : /*
353 : * PGSemaphoreTryLock
354 : *
355 : * Lock a semaphore only if able to do so without blocking
356 : */
357 : bool
358 0 : PGSemaphoreTryLock(PGSemaphore sema)
359 : {
360 : int errStatus;
361 :
362 : /*
363 : * Note: if errStatus is -1 and errno == EINTR then it means we returned
364 : * from the operation prematurely because we were sent a signal. So we
365 : * try and lock the semaphore again.
366 : */
367 : do
368 : {
369 0 : errStatus = sem_trywait(PG_SEM_REF(sema));
370 0 : } while (errStatus < 0 && errno == EINTR);
371 :
372 0 : if (errStatus < 0)
373 : {
374 0 : if (errno == EAGAIN || errno == EDEADLK)
375 0 : return false; /* failed to lock it */
376 : /* Otherwise we got trouble */
377 0 : elog(FATAL, "sem_trywait failed: %m");
378 : }
379 :
380 0 : return true;
381 : }
|
