1/*
2 * Copyright (C) 2010 David Xu <davidxu@freebsd.org>.
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice(s), this list of conditions and the following disclaimer as
10 *    the first lines of this file unmodified other than the possible
11 *    addition of one or more copyright notices.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 *    notice(s), this list of conditions and the following disclaimer in
14 *    the documentation and/or other materials provided with the
15 *    distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY
18 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
20 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE
21 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
22 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
23 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
24 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
25 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
26 * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
27 * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 *
29 * $FreeBSD$
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 <stdlib.h>
45#include <string.h>
46#include <time.h>
47#include <semaphore.h>
48#include <unistd.h>
49#include "un-namespace.h"
50#include "libc_private.h"
51
52__weak_reference(_sem_close, sem_close);
53__weak_reference(_sem_destroy, sem_destroy);
54__weak_reference(_sem_getvalue, sem_getvalue);
55__weak_reference(_sem_init, sem_init);
56__weak_reference(_sem_open, sem_open);
57__weak_reference(_sem_post, sem_post);
58__weak_reference(_sem_timedwait, sem_timedwait);
59__weak_reference(_sem_trywait, sem_trywait);
60__weak_reference(_sem_unlink, sem_unlink);
61__weak_reference(_sem_wait, sem_wait);
62
63#define SEM_PREFIX	"/tmp/SEMD"
64#define SEM_MAGIC	((u_int32_t)0x73656d31)
65
66struct sem_nameinfo {
67	int open_count;
68	char *name;
69	sem_t *sem;
70	LIST_ENTRY(sem_nameinfo) next;
71};
72
73static pthread_once_t once = PTHREAD_ONCE_INIT;
74static pthread_mutex_t sem_llock;
75static LIST_HEAD(,sem_nameinfo) sem_list = LIST_HEAD_INITIALIZER(sem_list);
76
77static void
78sem_prefork()
79{
80
81	_pthread_mutex_lock(&sem_llock);
82}
83
84static void
85sem_postfork()
86{
87	_pthread_mutex_unlock(&sem_llock);
88}
89
90static void
91sem_child_postfork()
92{
93	_pthread_mutex_unlock(&sem_llock);
94}
95
96static void
97sem_module_init(void)
98{
99	pthread_mutexattr_t ma;
100
101	_pthread_mutexattr_init(&ma);
102	_pthread_mutexattr_settype(&ma,  PTHREAD_MUTEX_RECURSIVE);
103	_pthread_mutex_init(&sem_llock, &ma);
104	_pthread_mutexattr_destroy(&ma);
105	_pthread_atfork(sem_prefork, sem_postfork, sem_child_postfork);
106}
107
108static inline int
109sem_check_validity(sem_t *sem)
110{
111
112	if (sem->_magic == SEM_MAGIC)
113		return (0);
114	else {
115		errno = EINVAL;
116		return (-1);
117	}
118}
119
120int
121_sem_init(sem_t *sem, int pshared, unsigned int value)
122{
123
124	if (value > SEM_VALUE_MAX) {
125		errno = EINVAL;
126		return (-1);
127	}
128
129	bzero(sem, sizeof(sem_t));
130	sem->_magic = SEM_MAGIC;
131	sem->_kern._count = (u_int32_t)value;
132	sem->_kern._has_waiters = 0;
133	sem->_kern._flags = pshared ? USYNC_PROCESS_SHARED : 0;
134	return (0);
135}
136
137sem_t *
138_sem_open(const char *name, int flags, ...)
139{
140	char path[PATH_MAX];
141
142	struct stat sb;
143	va_list ap;
144	struct sem_nameinfo *ni = NULL;
145	sem_t *sem = NULL;
146	int fd = -1, mode, len, errsave;
147	int value = 0;
148
149	if (name[0] != '/') {
150		errno = EINVAL;
151		return (SEM_FAILED);
152	}
153	name++;
154
155	if (flags & ~(O_CREAT|O_EXCL)) {
156		errno = EINVAL;
157		return (SEM_FAILED);
158	}
159
160	_pthread_once(&once, sem_module_init);
161
162	_pthread_mutex_lock(&sem_llock);
163	LIST_FOREACH(ni, &sem_list, next) {
164		if (strcmp(name, ni->name) == 0) {
165			if ((flags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL)) {
166				_pthread_mutex_unlock(&sem_llock);
167				errno = EEXIST;
168				return (SEM_FAILED);
169			} else {
170				ni->open_count++;
171				sem = ni->sem;
172				_pthread_mutex_unlock(&sem_llock);
173				return (sem);
174			}
175		}
176	}
177
178	if (flags & O_CREAT) {
179		va_start(ap, flags);
180		mode = va_arg(ap, int);
181		value = va_arg(ap, int);
182		va_end(ap);
183	}
184
185	len = sizeof(*ni) + strlen(name) + 1;
186	ni = (struct sem_nameinfo *)malloc(len);
187	if (ni == NULL) {
188		errno = ENOSPC;
189		goto error;
190	}
191
192	ni->name = (char *)(ni+1);
193	strcpy(ni->name, name);
194
195	strcpy(path, SEM_PREFIX);
196	if (strlcat(path, name, sizeof(path)) >= sizeof(path)) {
197		errno = ENAMETOOLONG;
198		goto error;
199	}
200
201	fd = _open(path, flags|O_RDWR|O_CLOEXEC|O_EXLOCK, mode);
202	if (fd == -1)
203		goto error;
204	if (_fstat(fd, &sb))
205		goto error;
206	if (sb.st_size < sizeof(sem_t)) {
207		sem_t tmp;
208
209		tmp._magic = SEM_MAGIC;
210		tmp._kern._has_waiters = 0;
211		tmp._kern._count = value;
212		tmp._kern._flags = USYNC_PROCESS_SHARED | SEM_NAMED;
213		if (_write(fd, &tmp, sizeof(tmp)) != sizeof(tmp))
214			goto error;
215	}
216	flock(fd, LOCK_UN);
217	sem = (sem_t *)mmap(NULL, sizeof(sem_t), PROT_READ|PROT_WRITE,
218		MAP_SHARED|MAP_NOSYNC, fd, 0);
219	if (sem == MAP_FAILED) {
220		sem = NULL;
221		if (errno == ENOMEM)
222			errno = ENOSPC;
223		goto error;
224	}
225	if (sem->_magic != SEM_MAGIC) {
226		errno = EINVAL;
227		goto error;
228	}
229	ni->open_count = 1;
230	ni->sem = sem;
231	LIST_INSERT_HEAD(&sem_list, ni, next);
232	_close(fd);
233	_pthread_mutex_unlock(&sem_llock);
234	return (sem);
235
236error:
237	errsave = errno;
238	if (fd != -1)
239		_close(fd);
240	if (sem != NULL)
241		munmap(sem, sizeof(sem_t));
242	free(ni);
243	_pthread_mutex_unlock(&sem_llock);
244	errno = errsave;
245	return (SEM_FAILED);
246}
247
248int
249_sem_close(sem_t *sem)
250{
251	struct sem_nameinfo *ni;
252
253	if (sem_check_validity(sem) != 0)
254		return (-1);
255
256	if (!(sem->_kern._flags & SEM_NAMED)) {
257		errno = EINVAL;
258		return (-1);
259	}
260
261	_pthread_once(&once, sem_module_init);
262
263	_pthread_mutex_lock(&sem_llock);
264	LIST_FOREACH(ni, &sem_list, next) {
265		if (sem == ni->sem) {
266			if (--ni->open_count > 0) {
267				_pthread_mutex_unlock(&sem_llock);
268				return (0);
269			}
270			else
271				break;
272		}
273	}
274
275	if (ni) {
276		LIST_REMOVE(ni, next);
277		_pthread_mutex_unlock(&sem_llock);
278		munmap(sem, sizeof(*sem));
279		free(ni);
280		return (0);
281	}
282	_pthread_mutex_unlock(&sem_llock);
283	errno = EINVAL;
284	return (-1);
285}
286
287int
288_sem_unlink(const char *name)
289{
290	char path[PATH_MAX];
291
292	if (name[0] != '/') {
293		errno = ENOENT;
294		return -1;
295	}
296	name++;
297
298	strcpy(path, SEM_PREFIX);
299	if (strlcat(path, name, sizeof(path)) >= sizeof(path)) {
300		errno = ENAMETOOLONG;
301		return (-1);
302	}
303	return unlink(path);
304}
305
306int
307_sem_destroy(sem_t *sem)
308{
309
310	if (sem_check_validity(sem) != 0)
311		return (-1);
312
313	if (sem->_kern._flags & SEM_NAMED) {
314		errno = EINVAL;
315		return (-1);
316	}
317	sem->_magic = 0;
318	return (0);
319}
320
321int
322_sem_getvalue(sem_t * __restrict sem, int * __restrict sval)
323{
324
325	if (sem_check_validity(sem) != 0)
326		return (-1);
327
328	*sval = (int)sem->_kern._count;
329	return (0);
330}
331
332static __inline int
333usem_wake(struct _usem *sem)
334{
335	return _umtx_op(sem, UMTX_OP_SEM_WAKE, 0, NULL, NULL);
336}
337
338static __inline int
339usem_wait(struct _usem *sem, const struct timespec *abstime)
340{
341	struct _umtx_time *tm_p, timeout;
342	size_t tm_size;
343
344	if (abstime == NULL) {
345		tm_p = NULL;
346		tm_size = 0;
347	} else {
348		timeout._clockid = CLOCK_REALTIME;
349		timeout._flags = UMTX_ABSTIME;
350		timeout._timeout = *abstime;
351		tm_p = &timeout;
352		tm_size = sizeof(timeout);
353	}
354	return _umtx_op(sem, UMTX_OP_SEM_WAIT, 0,
355		    (void *)tm_size, __DECONST(void*, tm_p));
356}
357
358int
359_sem_trywait(sem_t *sem)
360{
361	int val;
362
363	if (sem_check_validity(sem) != 0)
364		return (-1);
365
366	while ((val = sem->_kern._count) > 0) {
367		if (atomic_cmpset_acq_int(&sem->_kern._count, val, val - 1))
368			return (0);
369	}
370	errno = EAGAIN;
371	return (-1);
372}
373
374int
375_sem_timedwait(sem_t * __restrict sem,
376	const struct timespec * __restrict abstime)
377{
378	int val, retval;
379
380	if (sem_check_validity(sem) != 0)
381		return (-1);
382
383	retval = 0;
384	for (;;) {
385		while ((val = sem->_kern._count) > 0) {
386			if (atomic_cmpset_acq_int(&sem->_kern._count, val, val - 1))
387				return (0);
388		}
389
390		if (retval) {
391			_pthread_testcancel();
392			break;
393		}
394
395		/*
396		 * The timeout argument is only supposed to
397		 * be checked if the thread would have blocked.
398		 */
399		if (abstime != NULL) {
400			if (abstime->tv_nsec >= 1000000000 || abstime->tv_nsec < 0) {
401				errno = EINVAL;
402				return (-1);
403			}
404		}
405		_pthread_cancel_enter(1);
406		retval = usem_wait(&sem->_kern, abstime);
407		_pthread_cancel_leave(0);
408	}
409	return (retval);
410}
411
412int
413_sem_wait(sem_t *sem)
414{
415	return _sem_timedwait(sem, NULL);
416}
417
418/*
419 * POSIX:
420 * The sem_post() interface is reentrant with respect to signals and may be
421 * invoked from a signal-catching function.
422 * The implementation does not use lock, so it should be safe.
423 */
424int
425_sem_post(sem_t *sem)
426{
427	unsigned int count;
428
429	if (sem_check_validity(sem) != 0)
430		return (-1);
431
432	do {
433		count = sem->_kern._count;
434		if (count + 1 > SEM_VALUE_MAX)
435			return (EOVERFLOW);
436	} while(!atomic_cmpset_rel_int(&sem->_kern._count, count, count+1));
437	(void)usem_wake(&sem->_kern);
438	return (0);
439}
440