1/*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3 *
4 * Copyright (C) 2010 David Xu <davidxu@freebsd.org>.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice(s), this list of conditions and the following disclaimer as
12 *    the first lines of this file unmodified other than the possible
13 *    addition of one or more copyright notices.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 *    notice(s), this list of conditions and the following disclaimer in
16 *    the documentation and/or other materials provided with the
17 *    distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY
20 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE
23 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
26 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
27 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
28 * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
29 * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30 *
31 * $FreeBSD$
32 */
33
34#include "namespace.h"
35#include <sys/types.h>
36#include <sys/queue.h>
37#include <sys/mman.h>
38#include <sys/stat.h>
39#include <errno.h>
40#include <machine/atomic.h>
41#include <sys/umtx.h>
42#include <limits.h>
43#include <fcntl.h>
44#include <pthread.h>
45#include <stdarg.h>
46#include <stdbool.h>
47#include <stdlib.h>
48#include <string.h>
49#include <time.h>
50#include <semaphore.h>
51#include <unistd.h>
52#include "un-namespace.h"
53#include "libc_private.h"
54
55__weak_reference(_sem_close, sem_close);
56__weak_reference(_sem_destroy, sem_destroy);
57__weak_reference(_sem_getvalue, sem_getvalue);
58__weak_reference(_sem_init, sem_init);
59__weak_reference(_sem_open, sem_open);
60__weak_reference(_sem_post, sem_post);
61__weak_reference(_sem_timedwait, sem_timedwait);
62__weak_reference(_sem_clockwait_np, sem_clockwait_np);
63__weak_reference(_sem_trywait, sem_trywait);
64__weak_reference(_sem_unlink, sem_unlink);
65__weak_reference(_sem_wait, sem_wait);
66
67#define SEM_PREFIX	"/tmp/SEMD"
68#define SEM_MAGIC	((u_int32_t)0x73656d32)
69
70_Static_assert(SEM_VALUE_MAX <= USEM_MAX_COUNT, "SEM_VALUE_MAX too large");
71
72struct sem_nameinfo {
73	int open_count;
74	char *name;
75	dev_t dev;
76	ino_t ino;
77	sem_t *sem;
78	LIST_ENTRY(sem_nameinfo) next;
79};
80
81static pthread_once_t once = PTHREAD_ONCE_INIT;
82static pthread_mutex_t sem_llock;
83static LIST_HEAD(, sem_nameinfo) sem_list = LIST_HEAD_INITIALIZER(sem_list);
84
85static void
86sem_prefork(void)
87{
88
89	_pthread_mutex_lock(&sem_llock);
90}
91
92static void
93sem_postfork(void)
94{
95
96	_pthread_mutex_unlock(&sem_llock);
97}
98
99static void
100sem_child_postfork(void)
101{
102
103	_pthread_mutex_unlock(&sem_llock);
104}
105
106static void
107sem_module_init(void)
108{
109
110	_pthread_mutex_init(&sem_llock, NULL);
111	_pthread_atfork(sem_prefork, sem_postfork, sem_child_postfork);
112}
113
114static inline int
115sem_check_validity(sem_t *sem)
116{
117
118	if (sem->_magic == SEM_MAGIC)
119		return (0);
120	errno = EINVAL;
121	return (-1);
122}
123
124int
125_sem_init(sem_t *sem, int pshared, unsigned int value)
126{
127
128	if (value > SEM_VALUE_MAX) {
129		errno = EINVAL;
130		return (-1);
131	}
132
133	bzero(sem, sizeof(sem_t));
134	sem->_magic = SEM_MAGIC;
135	sem->_kern._count = (u_int32_t)value;
136	sem->_kern._flags = pshared ? USYNC_PROCESS_SHARED : 0;
137	return (0);
138}
139
140sem_t *
141_sem_open(const char *name, int flags, ...)
142{
143	char path[PATH_MAX];
144	struct stat sb;
145	va_list ap;
146	struct sem_nameinfo *ni;
147	sem_t *sem, tmp;
148	int errsave, fd, len, mode, value;
149
150	ni = NULL;
151	sem = NULL;
152	fd = -1;
153	value = 0;
154
155	if (name[0] != '/') {
156		errno = EINVAL;
157		return (SEM_FAILED);
158	}
159	name++;
160	strcpy(path, SEM_PREFIX);
161	if (strlcat(path, name, sizeof(path)) >= sizeof(path)) {
162		errno = ENAMETOOLONG;
163		return (SEM_FAILED);
164	}
165	if (flags & ~(O_CREAT|O_EXCL)) {
166		errno = EINVAL;
167		return (SEM_FAILED);
168	}
169	if ((flags & O_CREAT) != 0) {
170		va_start(ap, flags);
171		mode = va_arg(ap, int);
172		value = va_arg(ap, int);
173		va_end(ap);
174	}
175	fd = -1;
176	_pthread_once(&once, sem_module_init);
177
178	_pthread_mutex_lock(&sem_llock);
179	LIST_FOREACH(ni, &sem_list, next) {
180		if (ni->name != NULL && strcmp(name, ni->name) == 0) {
181			fd = _open(path, flags | O_RDWR | O_CLOEXEC |
182			    O_EXLOCK, mode);
183			if (fd == -1 || _fstat(fd, &sb) == -1) {
184				ni = NULL;
185				goto error;
186			}
187			if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT |
188			    O_EXCL) || ni->dev != sb.st_dev ||
189			    ni->ino != sb.st_ino) {
190				ni->name = NULL;
191				ni = NULL;
192				break;
193			}
194			ni->open_count++;
195			sem = ni->sem;
196			_pthread_mutex_unlock(&sem_llock);
197			_close(fd);
198			return (sem);
199		}
200	}
201
202	len = sizeof(*ni) + strlen(name) + 1;
203	ni = (struct sem_nameinfo *)malloc(len);
204	if (ni == NULL) {
205		errno = ENOSPC;
206		goto error;
207	}
208
209	ni->name = (char *)(ni+1);
210	strcpy(ni->name, name);
211
212	if (fd == -1) {
213		fd = _open(path, flags | O_RDWR | O_CLOEXEC | O_EXLOCK, mode);
214		if (fd == -1 || _fstat(fd, &sb) == -1)
215			goto error;
216	}
217	if (sb.st_size < sizeof(sem_t)) {
218		tmp._magic = SEM_MAGIC;
219		tmp._kern._count = value;
220		tmp._kern._flags = USYNC_PROCESS_SHARED | SEM_NAMED;
221		if (_write(fd, &tmp, sizeof(tmp)) != sizeof(tmp))
222			goto error;
223	}
224	flock(fd, LOCK_UN);
225	sem = mmap(NULL, sizeof(sem_t), PROT_READ | PROT_WRITE,
226	    MAP_SHARED | MAP_NOSYNC, fd, 0);
227	if (sem == MAP_FAILED) {
228		sem = NULL;
229		if (errno == ENOMEM)
230			errno = ENOSPC;
231		goto error;
232	}
233	if (sem->_magic != SEM_MAGIC) {
234		errno = EINVAL;
235		goto error;
236	}
237	ni->open_count = 1;
238	ni->sem = sem;
239	ni->dev = sb.st_dev;
240	ni->ino = sb.st_ino;
241	LIST_INSERT_HEAD(&sem_list, ni, next);
242	_close(fd);
243	_pthread_mutex_unlock(&sem_llock);
244	return (sem);
245
246error:
247	errsave = errno;
248	if (fd != -1)
249		_close(fd);
250	if (sem != NULL)
251		munmap(sem, sizeof(sem_t));
252	free(ni);
253	_pthread_mutex_unlock(&sem_llock);
254	errno = errsave;
255	return (SEM_FAILED);
256}
257
258int
259_sem_close(sem_t *sem)
260{
261	struct sem_nameinfo *ni;
262	bool last;
263
264	if (sem_check_validity(sem) != 0)
265		return (-1);
266
267	if (!(sem->_kern._flags & SEM_NAMED)) {
268		errno = EINVAL;
269		return (-1);
270	}
271
272	_pthread_once(&once, sem_module_init);
273
274	_pthread_mutex_lock(&sem_llock);
275	LIST_FOREACH(ni, &sem_list, next) {
276		if (sem == ni->sem) {
277			last = --ni->open_count == 0;
278			if (last)
279				LIST_REMOVE(ni, next);
280			_pthread_mutex_unlock(&sem_llock);
281			if (last) {
282				munmap(sem, sizeof(*sem));
283				free(ni);
284			}
285			return (0);
286		}
287	}
288	_pthread_mutex_unlock(&sem_llock);
289	errno = EINVAL;
290	return (-1);
291}
292
293int
294_sem_unlink(const char *name)
295{
296	char path[PATH_MAX];
297
298	if (name[0] != '/') {
299		errno = ENOENT;
300		return -1;
301	}
302	name++;
303	strcpy(path, SEM_PREFIX);
304	if (strlcat(path, name, sizeof(path)) >= sizeof(path)) {
305		errno = ENAMETOOLONG;
306		return (-1);
307	}
308
309	return (unlink(path));
310}
311
312int
313_sem_destroy(sem_t *sem)
314{
315
316	if (sem_check_validity(sem) != 0)
317		return (-1);
318
319	if (sem->_kern._flags & SEM_NAMED) {
320		errno = EINVAL;
321		return (-1);
322	}
323	sem->_magic = 0;
324	return (0);
325}
326
327int
328_sem_getvalue(sem_t * __restrict sem, int * __restrict sval)
329{
330
331	if (sem_check_validity(sem) != 0)
332		return (-1);
333
334	*sval = (int)USEM_COUNT(sem->_kern._count);
335	return (0);
336}
337
338static __inline int
339usem_wake(struct _usem2 *sem)
340{
341
342	return (_umtx_op(sem, UMTX_OP_SEM2_WAKE, 0, NULL, NULL));
343}
344
345static __inline int
346usem_wait(struct _usem2 *sem, clockid_t clock_id, int flags,
347    const struct timespec *rqtp, struct timespec *rmtp)
348{
349	struct {
350		struct _umtx_time timeout;
351		struct timespec remain;
352	} tms;
353	void *tm_p;
354	size_t tm_size;
355	int retval;
356
357	if (rqtp == NULL) {
358		tm_p = NULL;
359		tm_size = 0;
360	} else {
361		tms.timeout._clockid = clock_id;
362		tms.timeout._flags = (flags & TIMER_ABSTIME) ? UMTX_ABSTIME : 0;
363		tms.timeout._timeout = *rqtp;
364		tm_p = &tms;
365		tm_size = sizeof(tms);
366	}
367	retval = _umtx_op(sem, UMTX_OP_SEM2_WAIT, 0, (void *)tm_size, tm_p);
368	if (retval == -1 && errno == EINTR && (flags & TIMER_ABSTIME) == 0 &&
369	    rqtp != NULL && rmtp != NULL) {
370		*rmtp = tms.remain;
371	}
372
373	return (retval);
374}
375
376int
377_sem_trywait(sem_t *sem)
378{
379	int val;
380
381	if (sem_check_validity(sem) != 0)
382		return (-1);
383
384	while (USEM_COUNT(val = sem->_kern._count) > 0) {
385		if (atomic_cmpset_acq_int(&sem->_kern._count, val, val - 1))
386			return (0);
387	}
388	errno = EAGAIN;
389	return (-1);
390}
391
392int
393_sem_clockwait_np(sem_t * __restrict sem, clockid_t clock_id, int flags,
394	const struct timespec *rqtp, struct timespec *rmtp)
395{
396	int val, retval;
397
398	if (sem_check_validity(sem) != 0)
399		return (-1);
400
401	retval = 0;
402	_pthread_testcancel();
403	for (;;) {
404		while (USEM_COUNT(val = sem->_kern._count) > 0) {
405			if (atomic_cmpset_acq_int(&sem->_kern._count, val,
406			    val - 1))
407				return (0);
408		}
409
410		if (retval) {
411			_pthread_testcancel();
412			break;
413		}
414
415		/*
416		 * The timeout argument is only supposed to
417		 * be checked if the thread would have blocked.
418		 */
419		if (rqtp != NULL) {
420			if (rqtp->tv_nsec >= 1000000000 || rqtp->tv_nsec < 0) {
421				errno = EINVAL;
422				return (-1);
423			}
424		}
425		_pthread_cancel_enter(1);
426		retval = usem_wait(&sem->_kern, clock_id, flags, rqtp, rmtp);
427		_pthread_cancel_leave(0);
428	}
429	return (retval);
430}
431
432int
433_sem_timedwait(sem_t * __restrict sem,
434	const struct timespec * __restrict abstime)
435{
436
437	return (_sem_clockwait_np(sem, CLOCK_REALTIME, TIMER_ABSTIME, abstime,
438	    NULL));
439};
440
441int
442_sem_wait(sem_t *sem)
443{
444
445	return (_sem_timedwait(sem, NULL));
446}
447
448/*
449 * POSIX:
450 * The sem_post() interface is reentrant with respect to signals and may be
451 * invoked from a signal-catching function.
452 * The implementation does not use lock, so it should be safe.
453 */
454int
455_sem_post(sem_t *sem)
456{
457	unsigned int count;
458
459	if (sem_check_validity(sem) != 0)
460		return (-1);
461
462	do {
463		count = sem->_kern._count;
464		if (USEM_COUNT(count) + 1 > SEM_VALUE_MAX) {
465			errno = EOVERFLOW;
466			return (-1);
467		}
468	} while (!atomic_cmpset_rel_int(&sem->_kern._count, count, count + 1));
469	if (count & USEM_HAS_WAITERS)
470		usem_wake(&sem->_kern);
471	return (0);
472}
473