xref: /freebsd/sys/kern/uipc_sem.c (revision d0b2dbfa0ecf2bbc9709efc5e20baf8e4b44bbbf)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2002 Alfred Perlstein <alfred@FreeBSD.org>
5  * Copyright (c) 2003-2005 SPARTA, Inc.
6  * Copyright (c) 2005, 2016-2017 Robert N. M. Watson
7  * All rights reserved.
8  *
9  * This software was developed for the FreeBSD Project in part by Network
10  * Associates Laboratories, the Security Research Division of Network
11  * Associates, Inc. under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"),
12  * as part of the DARPA CHATS research program.
13  *
14  * Portions of this software were developed by BAE Systems, the University of
15  * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL
16  * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent
17  * Computing (TC) research program.
18  *
19  * Redistribution and use in source and binary forms, with or without
20  * modification, are permitted provided that the following conditions
21  * are met:
22  * 1. Redistributions of source code must retain the above copyright
23  *    notice, this list of conditions and the following disclaimer.
24  * 2. Redistributions in binary form must reproduce the above copyright
25  *    notice, this list of conditions and the following disclaimer in the
26  *    documentation and/or other materials provided with the distribution.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38  * SUCH DAMAGE.
39  */
40 
41 #include <sys/cdefs.h>
42 #include "opt_posix.h"
43 
44 #include <sys/param.h>
45 #include <sys/capsicum.h>
46 #include <sys/condvar.h>
47 #include <sys/fcntl.h>
48 #include <sys/file.h>
49 #include <sys/filedesc.h>
50 #include <sys/fnv_hash.h>
51 #include <sys/jail.h>
52 #include <sys/kernel.h>
53 #include <sys/ksem.h>
54 #include <sys/lock.h>
55 #include <sys/malloc.h>
56 #include <sys/module.h>
57 #include <sys/mutex.h>
58 #include <sys/priv.h>
59 #include <sys/proc.h>
60 #include <sys/posix4.h>
61 #include <sys/_semaphore.h>
62 #include <sys/stat.h>
63 #include <sys/syscall.h>
64 #include <sys/syscallsubr.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysent.h>
67 #include <sys/sysproto.h>
68 #include <sys/systm.h>
69 #include <sys/sx.h>
70 #include <sys/user.h>
71 #include <sys/vnode.h>
72 
73 #include <security/audit/audit.h>
74 #include <security/mac/mac_framework.h>
75 
76 FEATURE(p1003_1b_semaphores, "POSIX P1003.1B semaphores support");
77 /*
78  * TODO
79  *
80  * - Resource limits?
81  * - Replace global sem_lock with mtx_pool locks?
82  * - Add a MAC check_create() hook for creating new named semaphores.
83  */
84 
85 #ifndef SEM_MAX
86 #define	SEM_MAX	30
87 #endif
88 
89 #ifdef SEM_DEBUG
90 #define	DP(x)	printf x
91 #else
92 #define	DP(x)
93 #endif
94 
95 struct ksem_mapping {
96 	char		*km_path;
97 	Fnv32_t		km_fnv;
98 	struct ksem	*km_ksem;
99 	LIST_ENTRY(ksem_mapping) km_link;
100 };
101 
102 static MALLOC_DEFINE(M_KSEM, "ksem", "semaphore file descriptor");
103 static LIST_HEAD(, ksem_mapping) *ksem_dictionary;
104 static struct sx ksem_dict_lock;
105 static struct mtx ksem_count_lock;
106 static struct mtx sem_lock;
107 static u_long ksem_hash;
108 static int ksem_dead;
109 
110 #define	KSEM_HASH(fnv)	(&ksem_dictionary[(fnv) & ksem_hash])
111 
112 static int nsems = 0;
113 SYSCTL_DECL(_p1003_1b);
114 SYSCTL_INT(_p1003_1b, OID_AUTO, nsems, CTLFLAG_RD, &nsems, 0,
115     "Number of active kernel POSIX semaphores");
116 
117 static int	kern_sem_wait(struct thread *td, semid_t id, int tryflag,
118 		    struct timespec *abstime);
119 static int	ksem_access(struct ksem *ks, struct ucred *ucred);
120 static struct ksem *ksem_alloc(struct ucred *ucred, mode_t mode,
121 		    unsigned int value);
122 static int	ksem_create(struct thread *td, const char *path,
123 		    semid_t *semidp, mode_t mode, unsigned int value,
124 		    int flags, int compat32);
125 static void	ksem_drop(struct ksem *ks);
126 static int	ksem_get(struct thread *td, semid_t id, cap_rights_t *rightsp,
127     struct file **fpp);
128 static struct ksem *ksem_hold(struct ksem *ks);
129 static void	ksem_insert(char *path, Fnv32_t fnv, struct ksem *ks);
130 static struct ksem *ksem_lookup(char *path, Fnv32_t fnv);
131 static void	ksem_module_destroy(void);
132 static int	ksem_module_init(void);
133 static int	ksem_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
134 static int	sem_modload(struct module *module, int cmd, void *arg);
135 
136 static fo_stat_t	ksem_stat;
137 static fo_close_t	ksem_closef;
138 static fo_chmod_t	ksem_chmod;
139 static fo_chown_t	ksem_chown;
140 static fo_fill_kinfo_t	ksem_fill_kinfo;
141 
142 /* File descriptor operations. */
143 static struct fileops ksem_ops = {
144 	.fo_read = invfo_rdwr,
145 	.fo_write = invfo_rdwr,
146 	.fo_truncate = invfo_truncate,
147 	.fo_ioctl = invfo_ioctl,
148 	.fo_poll = invfo_poll,
149 	.fo_kqfilter = invfo_kqfilter,
150 	.fo_stat = ksem_stat,
151 	.fo_close = ksem_closef,
152 	.fo_chmod = ksem_chmod,
153 	.fo_chown = ksem_chown,
154 	.fo_sendfile = invfo_sendfile,
155 	.fo_fill_kinfo = ksem_fill_kinfo,
156 	.fo_flags = DFLAG_PASSABLE
157 };
158 
159 FEATURE(posix_sem, "POSIX semaphores");
160 
161 static int
162 ksem_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
163 {
164 	struct ksem *ks;
165 #ifdef MAC
166 	int error;
167 #endif
168 
169 	ks = fp->f_data;
170 
171 #ifdef MAC
172 	error = mac_posixsem_check_stat(active_cred, fp->f_cred, ks);
173 	if (error)
174 		return (error);
175 #endif
176 
177 	/*
178 	 * Attempt to return sanish values for fstat() on a semaphore
179 	 * file descriptor.
180 	 */
181 	bzero(sb, sizeof(*sb));
182 
183 	mtx_lock(&sem_lock);
184 	sb->st_atim = ks->ks_atime;
185 	sb->st_ctim = ks->ks_ctime;
186 	sb->st_mtim = ks->ks_mtime;
187 	sb->st_birthtim = ks->ks_birthtime;
188 	sb->st_uid = ks->ks_uid;
189 	sb->st_gid = ks->ks_gid;
190 	sb->st_mode = S_IFREG | ks->ks_mode;		/* XXX */
191 	mtx_unlock(&sem_lock);
192 
193 	return (0);
194 }
195 
196 static int
197 ksem_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
198     struct thread *td)
199 {
200 	struct ksem *ks;
201 	int error;
202 
203 	error = 0;
204 	ks = fp->f_data;
205 	mtx_lock(&sem_lock);
206 #ifdef MAC
207 	error = mac_posixsem_check_setmode(active_cred, ks, mode);
208 	if (error != 0)
209 		goto out;
210 #endif
211 	error = vaccess(VREG, ks->ks_mode, ks->ks_uid, ks->ks_gid, VADMIN,
212 	    active_cred);
213 	if (error != 0)
214 		goto out;
215 	ks->ks_mode = mode & ACCESSPERMS;
216 out:
217 	mtx_unlock(&sem_lock);
218 	return (error);
219 }
220 
221 static int
222 ksem_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
223     struct thread *td)
224 {
225 	struct ksem *ks;
226 	int error;
227 
228 	error = 0;
229 	ks = fp->f_data;
230 	mtx_lock(&sem_lock);
231 #ifdef MAC
232 	error = mac_posixsem_check_setowner(active_cred, ks, uid, gid);
233 	if (error != 0)
234 		goto out;
235 #endif
236 	if (uid == (uid_t)-1)
237 		uid = ks->ks_uid;
238 	if (gid == (gid_t)-1)
239                  gid = ks->ks_gid;
240 	if (((uid != ks->ks_uid && uid != active_cred->cr_uid) ||
241 	    (gid != ks->ks_gid && !groupmember(gid, active_cred))) &&
242 	    (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN)))
243 		goto out;
244 	ks->ks_uid = uid;
245 	ks->ks_gid = gid;
246 out:
247 	mtx_unlock(&sem_lock);
248 	return (error);
249 }
250 
251 static int
252 ksem_closef(struct file *fp, struct thread *td)
253 {
254 	struct ksem *ks;
255 
256 	ks = fp->f_data;
257 	fp->f_data = NULL;
258 	ksem_drop(ks);
259 
260 	return (0);
261 }
262 
263 static int
264 ksem_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
265 {
266 	const char *path, *pr_path;
267 	struct ksem *ks;
268 	size_t pr_pathlen;
269 
270 	kif->kf_type = KF_TYPE_SEM;
271 	ks = fp->f_data;
272 	mtx_lock(&sem_lock);
273 	kif->kf_un.kf_sem.kf_sem_value = ks->ks_value;
274 	kif->kf_un.kf_sem.kf_sem_mode = S_IFREG | ks->ks_mode;	/* XXX */
275 	mtx_unlock(&sem_lock);
276 	if (ks->ks_path != NULL) {
277 		sx_slock(&ksem_dict_lock);
278 		if (ks->ks_path != NULL) {
279 			path = ks->ks_path;
280 			pr_path = curthread->td_ucred->cr_prison->pr_path;
281 			if (strcmp(pr_path, "/") != 0) {
282 				/* Return the jail-rooted pathname. */
283 				pr_pathlen = strlen(pr_path);
284 				if (strncmp(path, pr_path, pr_pathlen) == 0 &&
285 				    path[pr_pathlen] == '/')
286 					path += pr_pathlen;
287 			}
288 			strlcpy(kif->kf_path, path, sizeof(kif->kf_path));
289 		}
290 		sx_sunlock(&ksem_dict_lock);
291 	}
292 	return (0);
293 }
294 
295 /*
296  * ksem object management including creation and reference counting
297  * routines.
298  */
299 static struct ksem *
300 ksem_alloc(struct ucred *ucred, mode_t mode, unsigned int value)
301 {
302 	struct ksem *ks;
303 
304 	mtx_lock(&ksem_count_lock);
305 	if (nsems == p31b_getcfg(CTL_P1003_1B_SEM_NSEMS_MAX) || ksem_dead) {
306 		mtx_unlock(&ksem_count_lock);
307 		return (NULL);
308 	}
309 	nsems++;
310 	mtx_unlock(&ksem_count_lock);
311 	ks = malloc(sizeof(*ks), M_KSEM, M_WAITOK | M_ZERO);
312 	ks->ks_uid = ucred->cr_uid;
313 	ks->ks_gid = ucred->cr_gid;
314 	ks->ks_mode = mode;
315 	ks->ks_value = value;
316 	cv_init(&ks->ks_cv, "ksem");
317 	vfs_timestamp(&ks->ks_birthtime);
318 	ks->ks_atime = ks->ks_mtime = ks->ks_ctime = ks->ks_birthtime;
319 	refcount_init(&ks->ks_ref, 1);
320 #ifdef MAC
321 	mac_posixsem_init(ks);
322 	mac_posixsem_create(ucred, ks);
323 #endif
324 
325 	return (ks);
326 }
327 
328 static struct ksem *
329 ksem_hold(struct ksem *ks)
330 {
331 
332 	refcount_acquire(&ks->ks_ref);
333 	return (ks);
334 }
335 
336 static void
337 ksem_drop(struct ksem *ks)
338 {
339 
340 	if (refcount_release(&ks->ks_ref)) {
341 #ifdef MAC
342 		mac_posixsem_destroy(ks);
343 #endif
344 		cv_destroy(&ks->ks_cv);
345 		free(ks, M_KSEM);
346 		mtx_lock(&ksem_count_lock);
347 		nsems--;
348 		mtx_unlock(&ksem_count_lock);
349 	}
350 }
351 
352 /*
353  * Determine if the credentials have sufficient permissions for read
354  * and write access.
355  */
356 static int
357 ksem_access(struct ksem *ks, struct ucred *ucred)
358 {
359 	int error;
360 
361 	error = vaccess(VREG, ks->ks_mode, ks->ks_uid, ks->ks_gid,
362 	    VREAD | VWRITE, ucred);
363 	if (error)
364 		error = priv_check_cred(ucred, PRIV_SEM_WRITE);
365 	return (error);
366 }
367 
368 /*
369  * Dictionary management.  We maintain an in-kernel dictionary to map
370  * paths to semaphore objects.  We use the FNV hash on the path to
371  * store the mappings in a hash table.
372  */
373 static struct ksem *
374 ksem_lookup(char *path, Fnv32_t fnv)
375 {
376 	struct ksem_mapping *map;
377 
378 	LIST_FOREACH(map, KSEM_HASH(fnv), km_link) {
379 		if (map->km_fnv != fnv)
380 			continue;
381 		if (strcmp(map->km_path, path) == 0)
382 			return (map->km_ksem);
383 	}
384 
385 	return (NULL);
386 }
387 
388 static void
389 ksem_insert(char *path, Fnv32_t fnv, struct ksem *ks)
390 {
391 	struct ksem_mapping *map;
392 
393 	map = malloc(sizeof(struct ksem_mapping), M_KSEM, M_WAITOK);
394 	map->km_path = path;
395 	map->km_fnv = fnv;
396 	map->km_ksem = ksem_hold(ks);
397 	ks->ks_path = path;
398 	LIST_INSERT_HEAD(KSEM_HASH(fnv), map, km_link);
399 }
400 
401 static int
402 ksem_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
403 {
404 	struct ksem_mapping *map;
405 	int error;
406 
407 	LIST_FOREACH(map, KSEM_HASH(fnv), km_link) {
408 		if (map->km_fnv != fnv)
409 			continue;
410 		if (strcmp(map->km_path, path) == 0) {
411 #ifdef MAC
412 			error = mac_posixsem_check_unlink(ucred, map->km_ksem);
413 			if (error)
414 				return (error);
415 #endif
416 			error = ksem_access(map->km_ksem, ucred);
417 			if (error)
418 				return (error);
419 			map->km_ksem->ks_path = NULL;
420 			LIST_REMOVE(map, km_link);
421 			ksem_drop(map->km_ksem);
422 			free(map->km_path, M_KSEM);
423 			free(map, M_KSEM);
424 			return (0);
425 		}
426 	}
427 
428 	return (ENOENT);
429 }
430 
431 static int
432 ksem_create_copyout_semid(struct thread *td, semid_t *semidp, int fd,
433     int compat32)
434 {
435 	semid_t semid;
436 #ifdef COMPAT_FREEBSD32
437 	int32_t semid32;
438 #endif
439 	void *ptr;
440 	size_t ptrs;
441 
442 #ifdef COMPAT_FREEBSD32
443 	if (compat32) {
444 		semid32 = fd;
445 		ptr = &semid32;
446 		ptrs = sizeof(semid32);
447 	} else {
448 #endif
449 		semid = fd;
450 		ptr = &semid;
451 		ptrs = sizeof(semid);
452 		compat32 = 0; /* silence gcc */
453 #ifdef COMPAT_FREEBSD32
454 	}
455 #endif
456 
457 	return (copyout(ptr, semidp, ptrs));
458 }
459 
460 /* Other helper routines. */
461 static int
462 ksem_create(struct thread *td, const char *name, semid_t *semidp, mode_t mode,
463     unsigned int value, int flags, int compat32)
464 {
465 	struct pwddesc *pdp;
466 	struct ksem *ks;
467 	struct file *fp;
468 	char *path;
469 	const char *pr_path;
470 	size_t pr_pathlen;
471 	Fnv32_t fnv;
472 	int error, fd;
473 
474 	AUDIT_ARG_FFLAGS(flags);
475 	AUDIT_ARG_MODE(mode);
476 	AUDIT_ARG_VALUE(value);
477 
478 	if (value > SEM_VALUE_MAX)
479 		return (EINVAL);
480 
481 	pdp = td->td_proc->p_pd;
482 	mode = (mode & ~pdp->pd_cmask) & ACCESSPERMS;
483 	error = falloc(td, &fp, &fd, O_CLOEXEC);
484 	if (error) {
485 		if (name == NULL)
486 			error = ENOSPC;
487 		return (error);
488 	}
489 
490 	/*
491 	 * Go ahead and copyout the file descriptor now.  This is a bit
492 	 * premature, but it is a lot easier to handle errors as opposed
493 	 * to later when we've possibly created a new semaphore, etc.
494 	 */
495 	error = ksem_create_copyout_semid(td, semidp, fd, compat32);
496 	if (error) {
497 		fdclose(td, fp, fd);
498 		fdrop(fp, td);
499 		return (error);
500 	}
501 
502 	if (name == NULL) {
503 		/* Create an anonymous semaphore. */
504 		ks = ksem_alloc(td->td_ucred, mode, value);
505 		if (ks == NULL)
506 			error = ENOSPC;
507 		else
508 			ks->ks_flags |= KS_ANONYMOUS;
509 	} else {
510 		path = malloc(MAXPATHLEN, M_KSEM, M_WAITOK);
511 		pr_path = td->td_ucred->cr_prison->pr_path;
512 
513 		/* Construct a full pathname for jailed callers. */
514 		pr_pathlen = strcmp(pr_path, "/") == 0 ? 0
515 		    : strlcpy(path, pr_path, MAXPATHLEN);
516 		error = copyinstr(name, path + pr_pathlen,
517 		    MAXPATHLEN - pr_pathlen, NULL);
518 
519 		/* Require paths to start with a '/' character. */
520 		if (error == 0 && path[pr_pathlen] != '/')
521 			error = EINVAL;
522 		if (error) {
523 			fdclose(td, fp, fd);
524 			fdrop(fp, td);
525 			free(path, M_KSEM);
526 			return (error);
527 		}
528 
529 		AUDIT_ARG_UPATH1_CANON(path);
530 		fnv = fnv_32_str(path, FNV1_32_INIT);
531 		sx_xlock(&ksem_dict_lock);
532 		ks = ksem_lookup(path, fnv);
533 		if (ks == NULL) {
534 			/* Object does not exist, create it if requested. */
535 			if (flags & O_CREAT) {
536 				ks = ksem_alloc(td->td_ucred, mode, value);
537 				if (ks == NULL)
538 					error = ENFILE;
539 				else {
540 					ksem_insert(path, fnv, ks);
541 					path = NULL;
542 				}
543 			} else
544 				error = ENOENT;
545 		} else {
546 			/*
547 			 * Object already exists, obtain a new
548 			 * reference if requested and permitted.
549 			 */
550 			if ((flags & (O_CREAT | O_EXCL)) ==
551 			    (O_CREAT | O_EXCL))
552 				error = EEXIST;
553 			else {
554 #ifdef MAC
555 				error = mac_posixsem_check_open(td->td_ucred,
556 				    ks);
557 				if (error == 0)
558 #endif
559 				error = ksem_access(ks, td->td_ucred);
560 			}
561 			if (error == 0)
562 				ksem_hold(ks);
563 #ifdef INVARIANTS
564 			else
565 				ks = NULL;
566 #endif
567 		}
568 		sx_xunlock(&ksem_dict_lock);
569 		if (path)
570 			free(path, M_KSEM);
571 	}
572 
573 	if (error) {
574 		KASSERT(ks == NULL, ("ksem_create error with a ksem"));
575 		fdclose(td, fp, fd);
576 		fdrop(fp, td);
577 		return (error);
578 	}
579 	KASSERT(ks != NULL, ("ksem_create w/o a ksem"));
580 
581 	finit(fp, FREAD | FWRITE, DTYPE_SEM, ks, &ksem_ops);
582 
583 	fdrop(fp, td);
584 
585 	return (0);
586 }
587 
588 static int
589 ksem_get(struct thread *td, semid_t id, cap_rights_t *rightsp,
590     struct file **fpp)
591 {
592 	struct ksem *ks;
593 	struct file *fp;
594 	int error;
595 
596 	error = fget(td, id, rightsp, &fp);
597 	if (error)
598 		return (EINVAL);
599 	if (fp->f_type != DTYPE_SEM) {
600 		fdrop(fp, td);
601 		return (EINVAL);
602 	}
603 	ks = fp->f_data;
604 	if (ks->ks_flags & KS_DEAD) {
605 		fdrop(fp, td);
606 		return (EINVAL);
607 	}
608 	*fpp = fp;
609 	return (0);
610 }
611 
612 /* System calls. */
613 #ifndef _SYS_SYSPROTO_H_
614 struct ksem_init_args {
615 	unsigned int	value;
616 	semid_t		*idp;
617 };
618 #endif
619 int
620 sys_ksem_init(struct thread *td, struct ksem_init_args *uap)
621 {
622 
623 	return (ksem_create(td, NULL, uap->idp, S_IRWXU | S_IRWXG, uap->value,
624 	    0, 0));
625 }
626 
627 #ifndef _SYS_SYSPROTO_H_
628 struct ksem_open_args {
629 	char		*name;
630 	int		oflag;
631 	mode_t		mode;
632 	unsigned int	value;
633 	semid_t		*idp;
634 };
635 #endif
636 int
637 sys_ksem_open(struct thread *td, struct ksem_open_args *uap)
638 {
639 
640 	DP((">>> ksem_open start, pid=%d\n", (int)td->td_proc->p_pid));
641 
642 	if ((uap->oflag & ~(O_CREAT | O_EXCL)) != 0)
643 		return (EINVAL);
644 	return (ksem_create(td, uap->name, uap->idp, uap->mode, uap->value,
645 	    uap->oflag, 0));
646 }
647 
648 #ifndef _SYS_SYSPROTO_H_
649 struct ksem_unlink_args {
650 	char		*name;
651 };
652 #endif
653 int
654 sys_ksem_unlink(struct thread *td, struct ksem_unlink_args *uap)
655 {
656 	char *path;
657 	const char *pr_path;
658 	size_t pr_pathlen;
659 	Fnv32_t fnv;
660 	int error;
661 
662 	path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
663 	pr_path = td->td_ucred->cr_prison->pr_path;
664 	pr_pathlen = strcmp(pr_path, "/") == 0 ? 0
665 	    : strlcpy(path, pr_path, MAXPATHLEN);
666 	error = copyinstr(uap->name, path + pr_pathlen, MAXPATHLEN - pr_pathlen,
667 	    NULL);
668 	if (error) {
669 		free(path, M_TEMP);
670 		return (error);
671 	}
672 
673 	AUDIT_ARG_UPATH1_CANON(path);
674 	fnv = fnv_32_str(path, FNV1_32_INIT);
675 	sx_xlock(&ksem_dict_lock);
676 	error = ksem_remove(path, fnv, td->td_ucred);
677 	sx_xunlock(&ksem_dict_lock);
678 	free(path, M_TEMP);
679 
680 	return (error);
681 }
682 
683 #ifndef _SYS_SYSPROTO_H_
684 struct ksem_close_args {
685 	semid_t		id;
686 };
687 #endif
688 int
689 sys_ksem_close(struct thread *td, struct ksem_close_args *uap)
690 {
691 	struct ksem *ks;
692 	struct file *fp;
693 	int error;
694 
695 	/* No capability rights required to close a semaphore. */
696 	AUDIT_ARG_FD(uap->id);
697 	error = ksem_get(td, uap->id, &cap_no_rights, &fp);
698 	if (error)
699 		return (error);
700 	ks = fp->f_data;
701 	if (ks->ks_flags & KS_ANONYMOUS) {
702 		fdrop(fp, td);
703 		return (EINVAL);
704 	}
705 	error = kern_close(td, uap->id);
706 	fdrop(fp, td);
707 	return (error);
708 }
709 
710 #ifndef _SYS_SYSPROTO_H_
711 struct ksem_post_args {
712 	semid_t	id;
713 };
714 #endif
715 int
716 sys_ksem_post(struct thread *td, struct ksem_post_args *uap)
717 {
718 	cap_rights_t rights;
719 	struct file *fp;
720 	struct ksem *ks;
721 	int error;
722 
723 	AUDIT_ARG_FD(uap->id);
724 	error = ksem_get(td, uap->id,
725 	    cap_rights_init_one(&rights, CAP_SEM_POST), &fp);
726 	if (error)
727 		return (error);
728 	ks = fp->f_data;
729 
730 	mtx_lock(&sem_lock);
731 #ifdef MAC
732 	error = mac_posixsem_check_post(td->td_ucred, fp->f_cred, ks);
733 	if (error)
734 		goto err;
735 #endif
736 	if (ks->ks_value == SEM_VALUE_MAX) {
737 		error = EOVERFLOW;
738 		goto err;
739 	}
740 	++ks->ks_value;
741 	if (ks->ks_waiters > 0)
742 		cv_signal(&ks->ks_cv);
743 	error = 0;
744 	vfs_timestamp(&ks->ks_ctime);
745 err:
746 	mtx_unlock(&sem_lock);
747 	fdrop(fp, td);
748 	return (error);
749 }
750 
751 #ifndef _SYS_SYSPROTO_H_
752 struct ksem_wait_args {
753 	semid_t		id;
754 };
755 #endif
756 int
757 sys_ksem_wait(struct thread *td, struct ksem_wait_args *uap)
758 {
759 
760 	return (kern_sem_wait(td, uap->id, 0, NULL));
761 }
762 
763 #ifndef _SYS_SYSPROTO_H_
764 struct ksem_timedwait_args {
765 	semid_t		id;
766 	const struct timespec *abstime;
767 };
768 #endif
769 int
770 sys_ksem_timedwait(struct thread *td, struct ksem_timedwait_args *uap)
771 {
772 	struct timespec abstime;
773 	struct timespec *ts;
774 	int error;
775 
776 	/*
777 	 * We allow a null timespec (wait forever).
778 	 */
779 	if (uap->abstime == NULL)
780 		ts = NULL;
781 	else {
782 		error = copyin(uap->abstime, &abstime, sizeof(abstime));
783 		if (error != 0)
784 			return (error);
785 		if (abstime.tv_nsec >= 1000000000 || abstime.tv_nsec < 0)
786 			return (EINVAL);
787 		ts = &abstime;
788 	}
789 	return (kern_sem_wait(td, uap->id, 0, ts));
790 }
791 
792 #ifndef _SYS_SYSPROTO_H_
793 struct ksem_trywait_args {
794 	semid_t		id;
795 };
796 #endif
797 int
798 sys_ksem_trywait(struct thread *td, struct ksem_trywait_args *uap)
799 {
800 
801 	return (kern_sem_wait(td, uap->id, 1, NULL));
802 }
803 
804 static int
805 kern_sem_wait(struct thread *td, semid_t id, int tryflag,
806     struct timespec *abstime)
807 {
808 	struct timespec ts1, ts2;
809 	struct timeval tv;
810 	cap_rights_t rights;
811 	struct file *fp;
812 	struct ksem *ks;
813 	int error;
814 
815 	DP((">>> kern_sem_wait entered! pid=%d\n", (int)td->td_proc->p_pid));
816 	AUDIT_ARG_FD(id);
817 	error = ksem_get(td, id, cap_rights_init_one(&rights, CAP_SEM_WAIT),
818 	    &fp);
819 	if (error)
820 		return (error);
821 	ks = fp->f_data;
822 	mtx_lock(&sem_lock);
823 	DP((">>> kern_sem_wait critical section entered! pid=%d\n",
824 	    (int)td->td_proc->p_pid));
825 #ifdef MAC
826 	error = mac_posixsem_check_wait(td->td_ucred, fp->f_cred, ks);
827 	if (error) {
828 		DP(("kern_sem_wait mac failed\n"));
829 		goto err;
830 	}
831 #endif
832 	DP(("kern_sem_wait value = %d, tryflag %d\n", ks->ks_value, tryflag));
833 	vfs_timestamp(&ks->ks_atime);
834 	while (ks->ks_value == 0) {
835 		ks->ks_waiters++;
836 		if (tryflag != 0)
837 			error = EAGAIN;
838 		else if (abstime == NULL)
839 			error = cv_wait_sig(&ks->ks_cv, &sem_lock);
840 		else {
841 			for (;;) {
842 				ts1 = *abstime;
843 				getnanotime(&ts2);
844 				timespecsub(&ts1, &ts2, &ts1);
845 				TIMESPEC_TO_TIMEVAL(&tv, &ts1);
846 				if (tv.tv_sec < 0) {
847 					error = ETIMEDOUT;
848 					break;
849 				}
850 				error = cv_timedwait_sig(&ks->ks_cv,
851 				    &sem_lock, tvtohz(&tv));
852 				if (error != EWOULDBLOCK)
853 					break;
854 			}
855 		}
856 		ks->ks_waiters--;
857 		if (error)
858 			goto err;
859 	}
860 	ks->ks_value--;
861 	DP(("kern_sem_wait value post-decrement = %d\n", ks->ks_value));
862 	error = 0;
863 err:
864 	mtx_unlock(&sem_lock);
865 	fdrop(fp, td);
866 	DP(("<<< kern_sem_wait leaving, pid=%d, error = %d\n",
867 	    (int)td->td_proc->p_pid, error));
868 	return (error);
869 }
870 
871 #ifndef _SYS_SYSPROTO_H_
872 struct ksem_getvalue_args {
873 	semid_t		id;
874 	int		*val;
875 };
876 #endif
877 int
878 sys_ksem_getvalue(struct thread *td, struct ksem_getvalue_args *uap)
879 {
880 	cap_rights_t rights;
881 	struct file *fp;
882 	struct ksem *ks;
883 	int error, val;
884 
885 	AUDIT_ARG_FD(uap->id);
886 	error = ksem_get(td, uap->id,
887 	    cap_rights_init_one(&rights, CAP_SEM_GETVALUE), &fp);
888 	if (error)
889 		return (error);
890 	ks = fp->f_data;
891 
892 	mtx_lock(&sem_lock);
893 #ifdef MAC
894 	error = mac_posixsem_check_getvalue(td->td_ucred, fp->f_cred, ks);
895 	if (error) {
896 		mtx_unlock(&sem_lock);
897 		fdrop(fp, td);
898 		return (error);
899 	}
900 #endif
901 	val = ks->ks_value;
902 	vfs_timestamp(&ks->ks_atime);
903 	mtx_unlock(&sem_lock);
904 	fdrop(fp, td);
905 	error = copyout(&val, uap->val, sizeof(val));
906 	return (error);
907 }
908 
909 #ifndef _SYS_SYSPROTO_H_
910 struct ksem_destroy_args {
911 	semid_t		id;
912 };
913 #endif
914 int
915 sys_ksem_destroy(struct thread *td, struct ksem_destroy_args *uap)
916 {
917 	struct file *fp;
918 	struct ksem *ks;
919 	int error;
920 
921 	/* No capability rights required to close a semaphore. */
922 	AUDIT_ARG_FD(uap->id);
923 	error = ksem_get(td, uap->id, &cap_no_rights, &fp);
924 	if (error)
925 		return (error);
926 	ks = fp->f_data;
927 	if (!(ks->ks_flags & KS_ANONYMOUS)) {
928 		fdrop(fp, td);
929 		return (EINVAL);
930 	}
931 	mtx_lock(&sem_lock);
932 	if (ks->ks_waiters != 0) {
933 		mtx_unlock(&sem_lock);
934 		error = EBUSY;
935 		goto err;
936 	}
937 	ks->ks_flags |= KS_DEAD;
938 	mtx_unlock(&sem_lock);
939 
940 	error = kern_close(td, uap->id);
941 err:
942 	fdrop(fp, td);
943 	return (error);
944 }
945 
946 static struct syscall_helper_data ksem_syscalls[] = {
947 	SYSCALL_INIT_HELPER(ksem_init),
948 	SYSCALL_INIT_HELPER(ksem_open),
949 	SYSCALL_INIT_HELPER(ksem_unlink),
950 	SYSCALL_INIT_HELPER(ksem_close),
951 	SYSCALL_INIT_HELPER(ksem_post),
952 	SYSCALL_INIT_HELPER(ksem_wait),
953 	SYSCALL_INIT_HELPER(ksem_timedwait),
954 	SYSCALL_INIT_HELPER(ksem_trywait),
955 	SYSCALL_INIT_HELPER(ksem_getvalue),
956 	SYSCALL_INIT_HELPER(ksem_destroy),
957 	SYSCALL_INIT_LAST
958 };
959 
960 #ifdef COMPAT_FREEBSD32
961 #include <compat/freebsd32/freebsd32.h>
962 #include <compat/freebsd32/freebsd32_proto.h>
963 #include <compat/freebsd32/freebsd32_signal.h>
964 #include <compat/freebsd32/freebsd32_syscall.h>
965 #include <compat/freebsd32/freebsd32_util.h>
966 
967 int
968 freebsd32_ksem_init(struct thread *td, struct freebsd32_ksem_init_args *uap)
969 {
970 
971 	return (ksem_create(td, NULL, (semid_t *)uap->idp, S_IRWXU | S_IRWXG, uap->value,
972 	    0, 1));
973 }
974 
975 int
976 freebsd32_ksem_open(struct thread *td, struct freebsd32_ksem_open_args *uap)
977 {
978 
979 	if ((uap->oflag & ~(O_CREAT | O_EXCL)) != 0)
980 		return (EINVAL);
981 	return (ksem_create(td, uap->name, (semid_t *)uap->idp, uap->mode, uap->value,
982 	    uap->oflag, 1));
983 }
984 
985 int
986 freebsd32_ksem_timedwait(struct thread *td,
987     struct freebsd32_ksem_timedwait_args *uap)
988 {
989 	struct timespec32 abstime32;
990 	struct timespec *ts, abstime;
991 	int error;
992 
993 	/*
994 	 * We allow a null timespec (wait forever).
995 	 */
996 	if (uap->abstime == NULL)
997 		ts = NULL;
998 	else {
999 		error = copyin(uap->abstime, &abstime32, sizeof(abstime32));
1000 		if (error != 0)
1001 			return (error);
1002 		CP(abstime32, abstime, tv_sec);
1003 		CP(abstime32, abstime, tv_nsec);
1004 		if (abstime.tv_nsec >= 1000000000 || abstime.tv_nsec < 0)
1005 			return (EINVAL);
1006 		ts = &abstime;
1007 	}
1008 	return (kern_sem_wait(td, uap->id, 0, ts));
1009 }
1010 
1011 static struct syscall_helper_data ksem32_syscalls[] = {
1012 	SYSCALL32_INIT_HELPER(freebsd32_ksem_init),
1013 	SYSCALL32_INIT_HELPER(freebsd32_ksem_open),
1014 	SYSCALL32_INIT_HELPER_COMPAT(ksem_unlink),
1015 	SYSCALL32_INIT_HELPER_COMPAT(ksem_close),
1016 	SYSCALL32_INIT_HELPER_COMPAT(ksem_post),
1017 	SYSCALL32_INIT_HELPER_COMPAT(ksem_wait),
1018 	SYSCALL32_INIT_HELPER(freebsd32_ksem_timedwait),
1019 	SYSCALL32_INIT_HELPER_COMPAT(ksem_trywait),
1020 	SYSCALL32_INIT_HELPER_COMPAT(ksem_getvalue),
1021 	SYSCALL32_INIT_HELPER_COMPAT(ksem_destroy),
1022 	SYSCALL_INIT_LAST
1023 };
1024 #endif
1025 
1026 static int
1027 ksem_module_init(void)
1028 {
1029 	int error;
1030 
1031 	mtx_init(&sem_lock, "sem", NULL, MTX_DEF);
1032 	mtx_init(&ksem_count_lock, "ksem count", NULL, MTX_DEF);
1033 	sx_init(&ksem_dict_lock, "ksem dictionary");
1034 	ksem_dictionary = hashinit(1024, M_KSEM, &ksem_hash);
1035 	p31b_setcfg(CTL_P1003_1B_SEMAPHORES, 200112L);
1036 	p31b_setcfg(CTL_P1003_1B_SEM_NSEMS_MAX, SEM_MAX);
1037 	p31b_setcfg(CTL_P1003_1B_SEM_VALUE_MAX, SEM_VALUE_MAX);
1038 
1039 	error = syscall_helper_register(ksem_syscalls, SY_THR_STATIC_KLD);
1040 	if (error)
1041 		return (error);
1042 #ifdef COMPAT_FREEBSD32
1043 	error = syscall32_helper_register(ksem32_syscalls, SY_THR_STATIC_KLD);
1044 	if (error)
1045 		return (error);
1046 #endif
1047 	return (0);
1048 }
1049 
1050 static void
1051 ksem_module_destroy(void)
1052 {
1053 
1054 #ifdef COMPAT_FREEBSD32
1055 	syscall32_helper_unregister(ksem32_syscalls);
1056 #endif
1057 	syscall_helper_unregister(ksem_syscalls);
1058 
1059 	p31b_setcfg(CTL_P1003_1B_SEMAPHORES, 0);
1060 	hashdestroy(ksem_dictionary, M_KSEM, ksem_hash);
1061 	sx_destroy(&ksem_dict_lock);
1062 	mtx_destroy(&ksem_count_lock);
1063 	mtx_destroy(&sem_lock);
1064 	p31b_unsetcfg(CTL_P1003_1B_SEM_VALUE_MAX);
1065 	p31b_unsetcfg(CTL_P1003_1B_SEM_NSEMS_MAX);
1066 }
1067 
1068 static int
1069 sem_modload(struct module *module, int cmd, void *arg)
1070 {
1071         int error = 0;
1072 
1073         switch (cmd) {
1074         case MOD_LOAD:
1075 		error = ksem_module_init();
1076 		if (error)
1077 			ksem_module_destroy();
1078                 break;
1079 
1080         case MOD_UNLOAD:
1081 		mtx_lock(&ksem_count_lock);
1082 		if (nsems != 0) {
1083 			error = EOPNOTSUPP;
1084 			mtx_unlock(&ksem_count_lock);
1085 			break;
1086 		}
1087 		ksem_dead = 1;
1088 		mtx_unlock(&ksem_count_lock);
1089 		ksem_module_destroy();
1090                 break;
1091 
1092         case MOD_SHUTDOWN:
1093                 break;
1094         default:
1095                 error = EINVAL;
1096                 break;
1097         }
1098         return (error);
1099 }
1100 
1101 static moduledata_t sem_mod = {
1102         "sem",
1103         &sem_modload,
1104         NULL
1105 };
1106 
1107 DECLARE_MODULE(sem, sem_mod, SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
1108 MODULE_VERSION(sem, 1);
1109