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 149560 : PosixSemaphoreCreate(sem_t *sem)
136 : {
137 149560 : if (sem_init(sem, 1, 1) < 0)
138 0 : elog(FATAL, "sem_init failed: %m");
139 149560 : }
140 : #endif /* USE_NAMED_POSIX_SEMAPHORES */
141 :
142 :
143 : /*
144 : * PosixSemaphoreKill - removes a semaphore
145 : */
146 : static void
147 149560 : 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 149560 : if (sem_destroy(sem) < 0)
156 0 : elog(LOG, "sem_destroy failed: %m");
157 : #endif
158 149560 : }
159 :
160 :
161 : /*
162 : * Report amount of shared memory needed for semaphores
163 : */
164 : Size
165 3297 : 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 3297 : 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 1150 : 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 1150 : if (stat(DataDir, &statbuf) < 0)
207 0 : 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 1150 : sharedSemas = (PGSemaphore)
219 1150 : ShmemAlloc(PGSemaphoreShmemSize(maxSemas));
220 : #endif
221 :
222 1150 : numSems = 0;
223 1150 : maxSems = maxSemas;
224 1150 : nextSemKey = statbuf.st_ino;
225 :
226 1150 : on_shmem_exit(ReleaseSemaphores, 0);
227 1150 : }
228 :
229 : /*
230 : * Release semaphores at shutdown or shmem reinitialization
231 : *
232 : * (called as an on_shmem_exit callback, hence funny argument list)
233 : */
234 : static void
235 1150 : ReleaseSemaphores(int status, Datum arg)
236 : {
237 : int i;
238 :
239 : #ifdef USE_NAMED_POSIX_SEMAPHORES
240 : for (i = 0; i < numSems; i++)
241 : PosixSemaphoreKill(mySemPointers[i]);
242 : free(mySemPointers);
243 : #endif
244 :
245 : #ifdef USE_UNNAMED_POSIX_SEMAPHORES
246 150710 : for (i = 0; i < numSems; i++)
247 149560 : PosixSemaphoreKill(PG_SEM_REF(sharedSemas + i));
248 : #endif
249 1150 : }
250 :
251 : /*
252 : * PGSemaphoreCreate
253 : *
254 : * Allocate a PGSemaphore structure with initial count 1
255 : */
256 : PGSemaphore
257 149560 : PGSemaphoreCreate(void)
258 : {
259 : PGSemaphore sema;
260 : sem_t *newsem;
261 :
262 : /* Can't do this in a backend, because static state is postmaster's */
263 : Assert(!IsUnderPostmaster);
264 :
265 149560 : if (numSems >= maxSems)
266 0 : elog(PANIC, "too many semaphores created");
267 :
268 : #ifdef USE_NAMED_POSIX_SEMAPHORES
269 : newsem = PosixSemaphoreCreate();
270 : /* Remember new sema for ReleaseSemaphores */
271 : mySemPointers[numSems] = newsem;
272 : sema = (PGSemaphore) newsem;
273 : #else
274 149560 : sema = &sharedSemas[numSems];
275 149560 : newsem = PG_SEM_REF(sema);
276 149560 : PosixSemaphoreCreate(newsem);
277 : #endif
278 :
279 149560 : numSems++;
280 :
281 149560 : return sema;
282 : }
283 :
284 : /*
285 : * PGSemaphoreReset
286 : *
287 : * Reset a previously-initialized PGSemaphore to have count 0
288 : */
289 : void
290 23180 : PGSemaphoreReset(PGSemaphore sema)
291 : {
292 : /*
293 : * There's no direct API for this in POSIX, so we have to ratchet the
294 : * semaphore down to 0 with repeated trywait's.
295 : */
296 : for (;;)
297 : {
298 37508 : if (sem_trywait(PG_SEM_REF(sema)) < 0)
299 : {
300 23180 : if (errno == EAGAIN || errno == EDEADLK)
301 : break; /* got it down to 0 */
302 0 : if (errno == EINTR)
303 0 : continue; /* can this happen? */
304 0 : elog(FATAL, "sem_trywait failed: %m");
305 : }
306 : }
307 23180 : }
308 :
309 : /*
310 : * PGSemaphoreLock
311 : *
312 : * Lock a semaphore (decrement count), blocking if count would be < 0
313 : */
314 : void
315 4719019 : PGSemaphoreLock(PGSemaphore sema)
316 : {
317 : int errStatus;
318 :
319 : /* See notes in sysv_sema.c's implementation of PGSemaphoreLock. */
320 : do
321 : {
322 4719019 : errStatus = sem_wait(PG_SEM_REF(sema));
323 4719019 : } while (errStatus < 0 && errno == EINTR);
324 :
325 4719019 : if (errStatus < 0)
326 0 : elog(FATAL, "sem_wait failed: %m");
327 4719019 : }
328 :
329 : /*
330 : * PGSemaphoreUnlock
331 : *
332 : * Unlock a semaphore (increment count)
333 : */
334 : void
335 4719454 : PGSemaphoreUnlock(PGSemaphore sema)
336 : {
337 : int errStatus;
338 :
339 : /*
340 : * Note: if errStatus is -1 and errno == EINTR then it means we returned
341 : * from the operation prematurely because we were sent a signal. So we
342 : * try and unlock the semaphore again. Not clear this can really happen,
343 : * but might as well cope.
344 : */
345 : do
346 : {
347 4719454 : errStatus = sem_post(PG_SEM_REF(sema));
348 4719454 : } while (errStatus < 0 && errno == EINTR);
349 :
350 4719454 : if (errStatus < 0)
351 0 : elog(FATAL, "sem_post failed: %m");
352 4719454 : }
353 :
354 : /*
355 : * PGSemaphoreTryLock
356 : *
357 : * Lock a semaphore only if able to do so without blocking
358 : */
359 : bool
360 0 : PGSemaphoreTryLock(PGSemaphore sema)
361 : {
362 : int errStatus;
363 :
364 : /*
365 : * Note: if errStatus is -1 and errno == EINTR then it means we returned
366 : * from the operation prematurely because we were sent a signal. So we
367 : * try and lock the semaphore again.
368 : */
369 : do
370 : {
371 0 : errStatus = sem_trywait(PG_SEM_REF(sema));
372 0 : } while (errStatus < 0 && errno == EINTR);
373 :
374 0 : if (errStatus < 0)
375 : {
376 0 : if (errno == EAGAIN || errno == EDEADLK)
377 0 : return false; /* failed to lock it */
378 : /* Otherwise we got trouble */
379 0 : elog(FATAL, "sem_trywait failed: %m");
380 : }
381 :
382 0 : return true;
383 : }
|
