2 * Copyright (c) 2002 Alfred Perlstein <alfred@FreeBSD.org>
3 * Copyright (c) 2003-2005 SPARTA, Inc.
4 * Copyright (c) 2005 Robert N. M. Watson
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.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
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.
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
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
38 #include "opt_posix.h"
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/sysproto.h>
43 #include <sys/eventhandler.h>
44 #include <sys/kernel.h>
47 #include <sys/mutex.h>
48 #include <sys/module.h>
49 #include <sys/condvar.h>
52 #include <sys/syscall.h>
54 #include <sys/sysent.h>
55 #include <sys/sysctl.h>
58 #include <sys/malloc.h>
59 #include <sys/fcntl.h>
61 #include <posix4/ksem.h>
62 #include <posix4/posix4.h>
63 #include <posix4/semaphore.h>
64 #include <posix4/_semaphore.h>
66 static int sem_count_proc(struct proc *p);
67 static struct ksem *sem_lookup_byname(const char *name);
68 static int sem_create(struct thread *td, const char *name,
69 struct ksem **ksret, mode_t mode, unsigned int value);
70 static void sem_free(struct ksem *ksnew);
71 static int sem_perm(struct thread *td, struct ksem *ks);
72 static void sem_enter(struct proc *p, struct ksem *ks);
73 static int sem_leave(struct proc *p, struct ksem *ks);
74 static void sem_exithook(void *arg, struct proc *p);
75 static void sem_forkhook(void *arg, struct proc *p1, struct proc *p2,
77 static int sem_hasopen(struct thread *td, struct ksem *ks);
79 static int kern_sem_close(struct thread *td, semid_t id);
80 static int kern_sem_post(struct thread *td, semid_t id);
81 static int kern_sem_wait(struct thread *td, semid_t id, int tryflag,
82 struct timespec *abstime);
83 static int kern_sem_init(struct thread *td, int dir, unsigned int value,
85 static int kern_sem_open(struct thread *td, int dir, const char *name,
86 int oflag, mode_t mode, unsigned int value, semid_t *idp);
87 static int kern_sem_unlink(struct thread *td, const char *name);
93 #define SEM_MAX_NAMELEN 14
95 #define SEM_TO_ID(x) ((intptr_t)(x))
96 #define ID_TO_SEM(x) id_to_sem(x)
99 * available semaphores go here, this includes sem_init and any semaphores
100 * created via sem_open that have not yet been unlinked.
102 LIST_HEAD(, ksem) ksem_head = LIST_HEAD_INITIALIZER(&ksem_head);
104 * semaphores still in use but have been sem_unlink()'d go here.
106 LIST_HEAD(, ksem) ksem_deadhead = LIST_HEAD_INITIALIZER(&ksem_deadhead);
108 static struct mtx sem_lock;
109 static MALLOC_DEFINE(M_SEM, "sems", "semaphore data");
111 static int nsems = 0;
112 SYSCTL_DECL(_p1003_1b);
113 SYSCTL_INT(_p1003_1b, OID_AUTO, nsems, CTLFLAG_RD, &nsems, 0, "");
115 static eventhandler_tag sem_exit_tag, sem_exec_tag, sem_fork_tag;
118 #define DP(x) printf x
125 sem_ref(struct ksem *ks)
128 mtx_assert(&sem_lock, MA_OWNED);
130 DP(("sem_ref: ks = %p, ref = %d\n", ks, ks->ks_ref));
135 sem_rel(struct ksem *ks)
138 mtx_assert(&sem_lock, MA_OWNED);
139 DP(("sem_rel: ks = %p, ref = %d\n", ks, ks->ks_ref - 1));
140 if (--ks->ks_ref == 0)
144 static __inline struct ksem *id_to_sem(semid_t id);
148 id_to_sem(semid_t id)
152 mtx_assert(&sem_lock, MA_OWNED);
153 DP(("id_to_sem: id = %0x,%p\n", id, (struct ksem *)id));
154 LIST_FOREACH(ks, &ksem_head, ks_entry) {
155 DP(("id_to_sem: ks = %p\n", ks));
156 if (ks == (struct ksem *)id)
163 sem_lookup_byname(const char *name)
167 mtx_assert(&sem_lock, MA_OWNED);
168 LIST_FOREACH(ks, &ksem_head, ks_entry)
169 if (ks->ks_name != NULL && strcmp(ks->ks_name, name) == 0)
175 sem_create(struct thread *td, const char *name, struct ksem **ksret,
176 mode_t mode, unsigned int value)
184 DP(("sem_create\n"));
187 if (value > SEM_VALUE_MAX)
189 ret = malloc(sizeof(*ret), M_SEM, M_WAITOK | M_ZERO);
192 if (len > SEM_MAX_NAMELEN) {
194 return (ENAMETOOLONG);
196 /* name must start with a '/' but not contain one. */
197 if (*name != '/' || len < 2 || index(name + 1, '/') != NULL) {
201 ret->ks_name = malloc(len + 1, M_SEM, M_WAITOK);
202 strcpy(ret->ks_name, name);
207 ret->ks_value = value;
210 ret->ks_uid = uc->cr_uid;
211 ret->ks_gid = uc->cr_gid;
213 cv_init(&ret->ks_cv, "sem");
214 LIST_INIT(&ret->ks_users);
216 mac_init_posix_sem(ret);
217 mac_create_posix_sem(uc, ret);
220 sem_enter(td->td_proc, ret);
223 if (nsems >= p31b_getcfg(CTL_P1003_1B_SEM_NSEMS_MAX)) {
224 sem_leave(td->td_proc, ret);
231 mtx_unlock(&sem_lock);
235 #ifndef _SYS_SYSPROTO_H_
236 struct ksem_init_args {
240 int ksem_init(struct thread *td, struct ksem_init_args *uap);
243 ksem_init(struct thread *td, struct ksem_init_args *uap)
247 error = kern_sem_init(td, UIO_USERSPACE, uap->value, uap->idp);
252 kern_sem_init(struct thread *td, int dir, unsigned int value, semid_t *idp)
258 error = sem_create(td, NULL, &ks, S_IRWXU | S_IRWXG, value);
262 if (dir == UIO_USERSPACE) {
263 error = copyout(&id, idp, sizeof(id));
267 mtx_unlock(&sem_lock);
274 LIST_INSERT_HEAD(&ksem_head, ks, ks_entry);
276 mtx_unlock(&sem_lock);
280 #ifndef _SYS_SYSPROTO_H_
281 struct ksem_open_args {
288 int ksem_open(struct thread *td, struct ksem_open_args *uap);
291 ksem_open(struct thread *td, struct ksem_open_args *uap)
293 char name[SEM_MAX_NAMELEN + 1];
297 error = copyinstr(uap->name, name, SEM_MAX_NAMELEN + 1, &done);
300 DP((">>> sem_open start\n"));
301 error = kern_sem_open(td, UIO_USERSPACE,
302 name, uap->oflag, uap->mode, uap->value, uap->idp);
303 DP(("<<< sem_open end\n"));
308 kern_sem_open(struct thread *td, int dir, const char *name, int oflag,
309 mode_t mode, unsigned int value, semid_t *idp)
311 struct ksem *ksnew, *ks;
317 ks = sem_lookup_byname(name);
319 * If we found it but O_EXCL is set, error.
321 if (ks != NULL && (oflag & O_EXCL) != 0) {
322 mtx_unlock(&sem_lock);
326 * If we didn't find it...
330 * didn't ask for creation? error.
332 if ((oflag & O_CREAT) == 0) {
333 mtx_unlock(&sem_lock);
337 * We may block during creation, so drop the lock.
339 mtx_unlock(&sem_lock);
340 error = sem_create(td, name, &ksnew, mode, value);
343 id = SEM_TO_ID(ksnew);
344 if (dir == UIO_USERSPACE) {
345 DP(("about to copyout! %d to %p\n", id, idp));
346 error = copyout(&id, idp, sizeof(id));
349 sem_leave(td->td_proc, ksnew);
351 mtx_unlock(&sem_lock);
355 DP(("about to set! %d to %p\n", id, idp));
359 * We need to make sure we haven't lost a race while
360 * allocating during creation.
363 ks = sem_lookup_byname(name);
366 sem_leave(td->td_proc, ksnew);
368 /* we lost and we can't loose... */
369 if ((oflag & O_EXCL) != 0) {
370 mtx_unlock(&sem_lock);
374 DP(("sem_create: about to add to list...\n"));
375 LIST_INSERT_HEAD(&ksem_head, ksnew, ks_entry);
376 DP(("sem_create: setting list bit...\n"));
377 ksnew->ks_onlist = 1;
378 DP(("sem_create: done, about to unlock...\n"));
382 error = mac_check_posix_sem_open(td->td_ucred, ks);
387 * if we aren't the creator, then enforce permissions.
389 error = sem_perm(td, ks);
393 mtx_unlock(&sem_lock);
395 if (dir == UIO_USERSPACE) {
396 error = copyout(&id, idp, sizeof(id));
400 mtx_unlock(&sem_lock);
406 sem_enter(td->td_proc, ks);
411 mtx_unlock(&sem_lock);
416 sem_perm(struct thread *td, struct ksem *ks)
421 DP(("sem_perm: uc(%d,%d) ks(%d,%d,%o)\n",
422 uc->cr_uid, uc->cr_gid,
423 ks->ks_uid, ks->ks_gid, ks->ks_mode));
424 if ((uc->cr_uid == ks->ks_uid && (ks->ks_mode & S_IWUSR) != 0) ||
425 (uc->cr_gid == ks->ks_gid && (ks->ks_mode & S_IWGRP) != 0) ||
426 (ks->ks_mode & S_IWOTH) != 0 || suser(td) == 0)
432 sem_free(struct ksem *ks)
437 LIST_REMOVE(ks, ks_entry);
438 if (ks->ks_name != NULL)
439 free(ks->ks_name, M_SEM);
440 cv_destroy(&ks->ks_cv);
444 static __inline struct kuser *sem_getuser(struct proc *p, struct ksem *ks);
446 static __inline struct kuser *
447 sem_getuser(struct proc *p, struct ksem *ks)
451 LIST_FOREACH(k, &ks->ks_users, ku_next)
452 if (k->ku_pid == p->p_pid)
458 sem_hasopen(struct thread *td, struct ksem *ks)
461 return ((ks->ks_name == NULL && sem_perm(td, ks) == 0)
462 || sem_getuser(td->td_proc, ks) != NULL);
466 sem_leave(struct proc *p, struct ksem *ks)
470 DP(("sem_leave: ks = %p\n", ks));
471 k = sem_getuser(p, ks);
472 DP(("sem_leave: ks = %p, k = %p\n", ks, k));
474 LIST_REMOVE(k, ku_next);
476 DP(("sem_leave: about to free k\n"));
478 DP(("sem_leave: returning\n"));
489 struct kuser *ku, *k;
491 ku = malloc(sizeof(*ku), M_SEM, M_WAITOK);
492 ku->ku_pid = p->p_pid;
494 k = sem_getuser(p, ks);
496 mtx_unlock(&sem_lock);
500 LIST_INSERT_HEAD(&ks->ks_users, ku, ku_next);
502 mtx_unlock(&sem_lock);
505 #ifndef _SYS_SYSPROTO_H_
506 struct ksem_unlink_args {
509 int ksem_unlink(struct thread *td, struct ksem_unlink_args *uap);
513 ksem_unlink(struct thread *td, struct ksem_unlink_args *uap)
515 char name[SEM_MAX_NAMELEN + 1];
519 error = copyinstr(uap->name, name, SEM_MAX_NAMELEN + 1, &done);
520 return (error ? error :
521 kern_sem_unlink(td, name));
525 kern_sem_unlink(struct thread *td, const char *name)
531 ks = sem_lookup_byname(name);
534 error = mac_check_posix_sem_unlink(td->td_ucred, ks);
536 mtx_unlock(&sem_lock);
540 error = sem_perm(td, ks);
543 DP(("sem_unlink: '%s' ks = %p, error = %d\n", name, ks, error));
545 LIST_REMOVE(ks, ks_entry);
546 LIST_INSERT_HEAD(&ksem_deadhead, ks, ks_entry);
549 mtx_unlock(&sem_lock);
553 #ifndef _SYS_SYSPROTO_H_
554 struct ksem_close_args {
557 int ksem_close(struct thread *td, struct ksem_close_args *uap);
561 ksem_close(struct thread *td, struct ksem_close_args *uap)
564 return (kern_sem_close(td, uap->id));
568 kern_sem_close(struct thread *td, semid_t id)
576 /* this is not a valid operation for unnamed sems */
577 if (ks != NULL && ks->ks_name != NULL)
578 error = sem_leave(td->td_proc, ks);
579 mtx_unlock(&sem_lock);
583 #ifndef _SYS_SYSPROTO_H_
584 struct ksem_post_args {
587 int ksem_post(struct thread *td, struct ksem_post_args *uap);
590 ksem_post(struct thread *td, struct ksem_post_args *uap)
593 return (kern_sem_post(td, uap->id));
597 kern_sem_post(struct thread *td, semid_t id)
604 if (ks == NULL || !sem_hasopen(td, ks)) {
609 error = mac_check_posix_sem_post(td->td_ucred, ks);
613 if (ks->ks_value == SEM_VALUE_MAX) {
618 if (ks->ks_waiters > 0)
619 cv_signal(&ks->ks_cv);
622 mtx_unlock(&sem_lock);
626 #ifndef _SYS_SYSPROTO_H_
627 struct ksem_wait_args {
630 int ksem_wait(struct thread *td, struct ksem_wait_args *uap);
634 ksem_wait(struct thread *td, struct ksem_wait_args *uap)
637 return (kern_sem_wait(td, uap->id, 0, NULL));
640 #ifndef _SYS_SYSPROTO_H_
641 struct ksem_timedwait_args {
643 const struct timespec *abstime;
645 int ksem_timedwait(struct thread *td, struct ksem_timedwait_args *uap);
648 ksem_timedwait(struct thread *td, struct ksem_timedwait_args *uap)
650 struct timespec abstime;
654 /* We allow a null timespec (wait forever). */
655 if (uap->abstime == NULL)
658 error = copyin(uap->abstime, &abstime, sizeof(abstime));
661 if (abstime.tv_nsec >= 1000000000 || abstime.tv_nsec < 0)
665 return (kern_sem_wait(td, uap->id, 0, ts));
668 #ifndef _SYS_SYSPROTO_H_
669 struct ksem_trywait_args {
672 int ksem_trywait(struct thread *td, struct ksem_trywait_args *uap);
675 ksem_trywait(struct thread *td, struct ksem_trywait_args *uap)
678 return (kern_sem_wait(td, uap->id, 1, NULL));
682 kern_sem_wait(struct thread *td, semid_t id, int tryflag,
683 struct timespec *abstime)
685 struct timespec ts1, ts2;
690 DP((">>> kern_sem_wait entered!\n"));
694 DP(("kern_sem_wait ks == NULL\n"));
699 if (!sem_hasopen(td, ks)) {
700 DP(("kern_sem_wait hasopen failed\n"));
705 error = mac_check_posix_sem_wait(td->td_ucred, ks);
707 DP(("kern_sem_wait mac failed\n"));
711 DP(("kern_sem_wait value = %d, tryflag %d\n", ks->ks_value, tryflag));
712 if (ks->ks_value == 0) {
716 else if (abstime == NULL)
717 error = cv_wait_sig(&ks->ks_cv, &sem_lock);
722 timespecsub(&ts1, &ts2);
723 TIMESPEC_TO_TIMEVAL(&tv, &ts1);
728 error = cv_timedwait_sig(&ks->ks_cv,
729 &sem_lock, tvtohz(&tv));
730 if (error != EWOULDBLOCK)
743 mtx_unlock(&sem_lock);
744 DP(("<<< kern_sem_wait leaving, error = %d\n", error));
748 #ifndef _SYS_SYSPROTO_H_
749 struct ksem_getvalue_args {
753 int ksem_getvalue(struct thread *td, struct ksem_getvalue_args *uap);
756 ksem_getvalue(struct thread *td, struct ksem_getvalue_args *uap)
762 ks = ID_TO_SEM(uap->id);
763 if (ks == NULL || !sem_hasopen(td, ks)) {
764 mtx_unlock(&sem_lock);
768 error = mac_check_posix_sem_getvalue(td->td_ucred, ks);
770 mtx_unlock(&sem_lock);
775 mtx_unlock(&sem_lock);
776 error = copyout(&val, uap->val, sizeof(val));
780 #ifndef _SYS_SYSPROTO_H_
781 struct ksem_destroy_args {
784 int ksem_destroy(struct thread *td, struct ksem_destroy_args *uap);
787 ksem_destroy(struct thread *td, struct ksem_destroy_args *uap)
793 ks = ID_TO_SEM(uap->id);
794 if (ks == NULL || !sem_hasopen(td, ks) ||
795 ks->ks_name != NULL) {
800 error = mac_check_posix_sem_destroy(td->td_ucred, ks);
804 if (ks->ks_waiters != 0) {
811 mtx_unlock(&sem_lock);
816 * Count the number of kusers associated with a proc, so as to guess at how
817 * many to allocate when forking.
820 sem_count_proc(struct proc *p)
826 mtx_assert(&sem_lock, MA_OWNED);
829 LIST_FOREACH(ks, &ksem_head, ks_entry) {
830 LIST_FOREACH(ku, &ks->ks_users, ku_next) {
831 if (ku->ku_pid == p->p_pid)
835 LIST_FOREACH(ks, &ksem_deadhead, ks_entry) {
836 LIST_FOREACH(ku, &ks->ks_users, ku_next) {
837 if (ku->ku_pid == p->p_pid)
845 * When a process forks, the child process must gain a reference to each open
846 * semaphore in the parent process, whether it is unlinked or not. This
847 * requires allocating a kuser structure for each semaphore reference in the
848 * new process. Because the set of semaphores in the parent can change while
849 * the fork is in progress, we have to handle races -- first we attempt to
850 * allocate enough storage to acquire references to each of the semaphores,
851 * then we enter the semaphores and release the temporary references.
854 sem_forkhook(void *arg, struct proc *p1, struct proc *p2, int flags)
856 struct ksem *ks, **sem_array;
857 int count, i, new_count;
861 count = sem_count_proc(p1);
863 mtx_unlock(&sem_lock);
867 mtx_assert(&sem_lock, MA_OWNED);
868 mtx_unlock(&sem_lock);
869 sem_array = malloc(sizeof(struct ksem *) * count, M_TEMP, M_WAITOK);
871 new_count = sem_count_proc(p1);
872 if (count < new_count) {
873 /* Lost race, repeat and allocate more storage. */
874 free(sem_array, M_TEMP);
879 * Given an array capable of storing an adequate number of semaphore
880 * references, now walk the list of semaphores and acquire a new
881 * reference for any semaphore opened by p1.
885 LIST_FOREACH(ks, &ksem_head, ks_entry) {
886 LIST_FOREACH(ku, &ks->ks_users, ku_next) {
887 if (ku->ku_pid == p1->p_pid) {
895 LIST_FOREACH(ks, &ksem_deadhead, ks_entry) {
896 LIST_FOREACH(ku, &ks->ks_users, ku_next) {
897 if (ku->ku_pid == p1->p_pid) {
905 mtx_unlock(&sem_lock);
906 KASSERT(i == count, ("sem_forkhook: i != count (%d, %d)", i, count));
908 * Now cause p2 to enter each of the referenced semaphores, then
909 * release our temporary reference. This is pretty inefficient.
910 * Finally, free our temporary array.
912 for (i = 0; i < count; i++) {
913 sem_enter(p2, sem_array[i]);
915 sem_rel(sem_array[i]);
916 mtx_unlock(&sem_lock);
918 free(sem_array, M_TEMP);
922 sem_exithook(void *arg, struct proc *p)
924 struct ksem *ks, *ksnext;
927 ks = LIST_FIRST(&ksem_head);
929 ksnext = LIST_NEXT(ks, ks_entry);
933 ks = LIST_FIRST(&ksem_deadhead);
935 ksnext = LIST_NEXT(ks, ks_entry);
939 mtx_unlock(&sem_lock);
943 sem_modload(struct module *module, int cmd, void *arg)
949 mtx_init(&sem_lock, "sem", "semaphore", MTX_DEF);
950 p31b_setcfg(CTL_P1003_1B_SEM_NSEMS_MAX, SEM_MAX);
951 p31b_setcfg(CTL_P1003_1B_SEM_VALUE_MAX, SEM_VALUE_MAX);
952 sem_exit_tag = EVENTHANDLER_REGISTER(process_exit, sem_exithook,
953 NULL, EVENTHANDLER_PRI_ANY);
954 sem_exec_tag = EVENTHANDLER_REGISTER(process_exec, sem_exithook,
955 NULL, EVENTHANDLER_PRI_ANY);
956 sem_fork_tag = EVENTHANDLER_REGISTER(process_fork, sem_forkhook, NULL, EVENTHANDLER_PRI_ANY);
963 EVENTHANDLER_DEREGISTER(process_exit, sem_exit_tag);
964 EVENTHANDLER_DEREGISTER(process_exec, sem_exec_tag);
965 EVENTHANDLER_DEREGISTER(process_fork, sem_fork_tag);
966 mtx_destroy(&sem_lock);
977 static moduledata_t sem_mod = {
983 SYSCALL_MODULE_HELPER(ksem_init);
984 SYSCALL_MODULE_HELPER(ksem_open);
985 SYSCALL_MODULE_HELPER(ksem_unlink);
986 SYSCALL_MODULE_HELPER(ksem_close);
987 SYSCALL_MODULE_HELPER(ksem_post);
988 SYSCALL_MODULE_HELPER(ksem_wait);
989 SYSCALL_MODULE_HELPER(ksem_timedwait);
990 SYSCALL_MODULE_HELPER(ksem_trywait);
991 SYSCALL_MODULE_HELPER(ksem_getvalue);
992 SYSCALL_MODULE_HELPER(ksem_destroy);
994 DECLARE_MODULE(sem, sem_mod, SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
995 MODULE_VERSION(sem, 1);