1/*-
2 * SPDX-License-Identifier: BSD-2-Clause
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
32#include "namespace.h"
33#include <sys/types.h>
34#include <sys/queue.h>
35#include <sys/mman.h>
36#include <sys/stat.h>
37#include <errno.h>
38#include <machine/atomic.h>
39#include <sys/umtx.h>
40#include <limits.h>
41#include <fcntl.h>
42#include <pthread.h>
43#include <stdarg.h>
44#include <stdbool.h>
45#include <stdlib.h>
46#include <string.h>
47#include <time.h>
48#include <semaphore.h>
49#include <unistd.h>
50#include "un-namespace.h"
51#include "libc_private.h"
52
53__weak_reference(_sem_close, sem_close);
54__weak_reference(_sem_destroy, sem_destroy);
55__weak_reference(_sem_getvalue, sem_getvalue);
56__weak_reference(_sem_init, sem_init);
57__weak_reference(_sem_open, sem_open);
58__weak_reference(_sem_post, sem_post);
59__weak_reference(_sem_timedwait, sem_timedwait);
60__weak_reference(_sem_clockwait_np, sem_clockwait_np);
61__weak_reference(_sem_trywait, sem_trywait);
62__weak_reference(_sem_unlink, sem_unlink);
63__weak_reference(_sem_wait, sem_wait);
64
65#define SEM_PREFIX	"/tmp/SEMD"
66#define SEM_MAGIC	((u_int32_t)0x73656d32)
67
68_Static_assert(SEM_VALUE_MAX <= USEM_MAX_COUNT, "SEM_VALUE_MAX too large");
69
70struct sem_nameinfo {
71	int open_count;
72	char *name;
73	dev_t dev;
74	ino_t ino;
75	sem_t *sem;
76	LIST_ENTRY(sem_nameinfo) next;
77};
78
79static pthread_once_t once = PTHREAD_ONCE_INIT;
80static pthread_mutex_t sem_llock;
81static LIST_HEAD(, sem_nameinfo) sem_list = LIST_HEAD_INITIALIZER(sem_list);
82
83static void
84sem_prefork(void)
85{
86
87	_pthread_mutex_lock(&sem_llock);
88}
89
90static void
91sem_postfork(void)
92{
93
94	_pthread_mutex_unlock(&sem_llock);
95}
96
97static void
98sem_child_postfork(void)
99{
100
101	_pthread_mutex_unlock(&sem_llock);
102}
103
104static void
105sem_module_init(void)
106{
107
108	_pthread_mutex_init(&sem_llock, NULL);
109	_pthread_atfork(sem_prefork, sem_postfork, sem_child_postfork);
110}
111
112static inline int
113sem_check_validity(sem_t *sem)
114{
115
116	if (sem->_magic == SEM_MAGIC)
117		return (0);
118	errno = EINVAL;
119	return (-1);
120}
121
122int
123_sem_init(sem_t *sem, int pshared, unsigned int value)
124{
125
126	if (value > SEM_VALUE_MAX) {
127		errno = EINVAL;
128		return (-1);
129	}
130
131	bzero(sem, sizeof(sem_t));
132	sem->_magic = SEM_MAGIC;
133	sem->_kern._count = (u_int32_t)value;
134	sem->_kern._flags = pshared ? USYNC_PROCESS_SHARED : 0;
135	return (0);
136}
137
138sem_t *
139_sem_open(const char *name, int flags, ...)
140{
141	char path[PATH_MAX];
142	struct stat sb;
143	va_list ap;
144	struct sem_nameinfo *ni;
145	sem_t *sem, tmp;
146	int errsave, fd, len, mode, value;
147
148	ni = NULL;
149	sem = NULL;
150	fd = -1;
151	value = 0;
152
153	if (name[0] != '/') {
154		errno = EINVAL;
155		return (SEM_FAILED);
156	}
157	name++;
158	strcpy(path, SEM_PREFIX);
159	if (strlcat(path, name, sizeof(path)) >= sizeof(path)) {
160		errno = ENAMETOOLONG;
161		return (SEM_FAILED);
162	}
163	if (flags & ~(O_CREAT|O_EXCL)) {
164		errno = EINVAL;
165		return (SEM_FAILED);
166	}
167	if ((flags & O_CREAT) != 0) {
168		va_start(ap, flags);
169		mode = va_arg(ap, int);
170		value = va_arg(ap, int);
171		va_end(ap);
172	}
173	fd = -1;
174	_pthread_once(&once, sem_module_init);
175
176	_pthread_mutex_lock(&sem_llock);
177	LIST_FOREACH(ni, &sem_list, next) {
178		if (ni->name != NULL && strcmp(name, ni->name) == 0) {
179			fd = _open(path, flags | O_RDWR | O_CLOEXEC |
180			    O_EXLOCK, mode);
181			if (fd == -1 || _fstat(fd, &sb) == -1) {
182				ni = NULL;
183				goto error;
184			}
185			if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT |
186			    O_EXCL) || ni->dev != sb.st_dev ||
187			    ni->ino != sb.st_ino) {
188				ni->name = NULL;
189				ni = NULL;
190				break;
191			}
192			ni->open_count++;
193			sem = ni->sem;
194			_pthread_mutex_unlock(&sem_llock);
195			_close(fd);
196			return (sem);
197		}
198	}
199
200	len = sizeof(*ni) + strlen(name) + 1;
201	ni = (struct sem_nameinfo *)malloc(len);
202	if (ni == NULL) {
203		errno = ENOSPC;
204		goto error;
205	}
206
207	ni->name = (char *)(ni+1);
208	strcpy(ni->name, name);
209
210	if (fd == -1) {
211		fd = _open(path, flags | O_RDWR | O_CLOEXEC | O_EXLOCK, mode);
212		if (fd == -1 || _fstat(fd, &sb) == -1)
213			goto error;
214	}
215	if (sb.st_size < sizeof(sem_t)) {
216		tmp._magic = SEM_MAGIC;
217		tmp._kern._count = value;
218		tmp._kern._flags = USYNC_PROCESS_SHARED | SEM_NAMED;
219		if (_write(fd, &tmp, sizeof(tmp)) != sizeof(tmp))
220			goto error;
221	}
222	flock(fd, LOCK_UN);
223	sem = mmap(NULL, sizeof(sem_t), PROT_READ | PROT_WRITE,
224	    MAP_SHARED | MAP_NOSYNC, fd, 0);
225	if (sem == MAP_FAILED) {
226		sem = NULL;
227		if (errno == ENOMEM)
228			errno = ENOSPC;
229		goto error;
230	}
231	if (sem->_magic != SEM_MAGIC) {
232		errno = EINVAL;
233		goto error;
234	}
235	ni->open_count = 1;
236	ni->sem = sem;
237	ni->dev = sb.st_dev;
238	ni->ino = sb.st_ino;
239	LIST_INSERT_HEAD(&sem_list, ni, next);
240	_close(fd);
241	_pthread_mutex_unlock(&sem_llock);
242	return (sem);
243
244error:
245	errsave = errno;
246	if (fd != -1)
247		_close(fd);
248	if (sem != NULL)
249		munmap(sem, sizeof(sem_t));
250	free(ni);
251	_pthread_mutex_unlock(&sem_llock);
252	errno = errsave;
253	return (SEM_FAILED);
254}
255
256int
257_sem_close(sem_t *sem)
258{
259	struct sem_nameinfo *ni;
260	bool last;
261
262	if (sem_check_validity(sem) != 0)
263		return (-1);
264
265	if (!(sem->_kern._flags & SEM_NAMED)) {
266		errno = EINVAL;
267		return (-1);
268	}
269
270	_pthread_once(&once, sem_module_init);
271
272	_pthread_mutex_lock(&sem_llock);
273	LIST_FOREACH(ni, &sem_list, next) {
274		if (sem == ni->sem) {
275			last = --ni->open_count == 0;
276			if (last)
277				LIST_REMOVE(ni, next);
278			_pthread_mutex_unlock(&sem_llock);
279			if (last) {
280				munmap(sem, sizeof(*sem));
281				free(ni);
282			}
283			return (0);
284		}
285	}
286	_pthread_mutex_unlock(&sem_llock);
287	errno = EINVAL;
288	return (-1);
289}
290
291int
292_sem_unlink(const char *name)
293{
294	char path[PATH_MAX];
295
296	if (name[0] != '/') {
297		errno = ENOENT;
298		return -1;
299	}
300	name++;
301	strcpy(path, SEM_PREFIX);
302	if (strlcat(path, name, sizeof(path)) >= sizeof(path)) {
303		errno = ENAMETOOLONG;
304		return (-1);
305	}
306
307	return (unlink(path));
308}
309
310int
311_sem_destroy(sem_t *sem)
312{
313
314	if (sem_check_validity(sem) != 0)
315		return (-1);
316
317	if (sem->_kern._flags & SEM_NAMED) {
318		errno = EINVAL;
319		return (-1);
320	}
321	sem->_magic = 0;
322	return (0);
323}
324
325int
326_sem_getvalue(sem_t * __restrict sem, int * __restrict sval)
327{
328
329	if (sem_check_validity(sem) != 0)
330		return (-1);
331
332	*sval = (int)USEM_COUNT(sem->_kern._count);
333	return (0);
334}
335
336static __inline int
337usem_wake(struct _usem2 *sem)
338{
339
340	return (_umtx_op(sem, UMTX_OP_SEM2_WAKE, 0, NULL, NULL));
341}
342
343static __inline int
344usem_wait(struct _usem2 *sem, clockid_t clock_id, int flags,
345    const struct timespec *rqtp, struct timespec *rmtp)
346{
347	struct {
348		struct _umtx_time timeout;
349		struct timespec remain;
350	} tms;
351	void *tm_p;
352	size_t tm_size;
353	int retval;
354
355	if (rqtp == NULL) {
356		tm_p = NULL;
357		tm_size = 0;
358	} else {
359		tms.timeout._clockid = clock_id;
360		tms.timeout._flags = (flags & TIMER_ABSTIME) ? UMTX_ABSTIME : 0;
361		tms.timeout._timeout = *rqtp;
362		tm_p = &tms;
363		tm_size = sizeof(tms);
364	}
365	retval = _umtx_op(sem, UMTX_OP_SEM2_WAIT, 0, (void *)tm_size, tm_p);
366	if (retval == -1 && errno == EINTR && (flags & TIMER_ABSTIME) == 0 &&
367	    rqtp != NULL && rmtp != NULL) {
368		*rmtp = tms.remain;
369	}
370
371	return (retval);
372}
373
374int
375_sem_trywait(sem_t *sem)
376{
377	int val;
378
379	if (sem_check_validity(sem) != 0)
380		return (-1);
381
382	while (USEM_COUNT(val = sem->_kern._count) > 0) {
383		if (atomic_cmpset_acq_int(&sem->_kern._count, val, val - 1))
384			return (0);
385	}
386	errno = EAGAIN;
387	return (-1);
388}
389
390int
391_sem_clockwait_np(sem_t * __restrict sem, clockid_t clock_id, int flags,
392	const struct timespec *rqtp, struct timespec *rmtp)
393{
394	int val, retval;
395
396	if (sem_check_validity(sem) != 0)
397		return (-1);
398
399	retval = 0;
400	_pthread_testcancel();
401	for (;;) {
402		while (USEM_COUNT(val = sem->_kern._count) > 0) {
403			if (atomic_cmpset_acq_int(&sem->_kern._count, val,
404			    val - 1))
405				return (0);
406		}
407
408		if (retval) {
409			_pthread_testcancel();
410			break;
411		}
412
413		/*
414		 * The timeout argument is only supposed to
415		 * be checked if the thread would have blocked.
416		 */
417		if (rqtp != NULL) {
418			if (rqtp->tv_nsec >= 1000000000 || rqtp->tv_nsec < 0) {
419				errno = EINVAL;
420				return (-1);
421			}
422		}
423		_pthread_cancel_enter(1);
424		retval = usem_wait(&sem->_kern, clock_id, flags, rqtp, rmtp);
425		_pthread_cancel_leave(0);
426	}
427	return (retval);
428}
429
430int
431_sem_timedwait(sem_t * __restrict sem,
432	const struct timespec * __restrict abstime)
433{
434
435	return (_sem_clockwait_np(sem, CLOCK_REALTIME, TIMER_ABSTIME, abstime,
436	    NULL));
437};
438
439int
440_sem_wait(sem_t *sem)
441{
442
443	return (_sem_timedwait(sem, NULL));
444}
445
446/*
447 * POSIX:
448 * The sem_post() interface is reentrant with respect to signals and may be
449 * invoked from a signal-catching function.
450 * The implementation does not use lock, so it should be safe.
451 */
452int
453_sem_post(sem_t *sem)
454{
455	unsigned int count;
456
457	if (sem_check_validity(sem) != 0)
458		return (-1);
459
460	do {
461		count = sem->_kern._count;
462		if (USEM_COUNT(count) + 1 > SEM_VALUE_MAX) {
463			errno = EOVERFLOW;
464			return (-1);
465		}
466	} while (!atomic_cmpset_rel_int(&sem->_kern._count, count, count + 1));
467	if (count & USEM_HAS_WAITERS)
468		usem_wake(&sem->_kern);
469	return (0);
470}
471