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