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("$FreeBSD$");
37 #include "opt_compat.h"
38 #include "opt_posix.h"
40 #include <sys/param.h>
41 #include <sys/condvar.h>
42 #include <sys/fcntl.h>
44 #include <sys/filedesc.h>
45 #include <sys/fnv_hash.h>
46 #include <sys/kernel.h>
49 #include <sys/malloc.h>
50 #include <sys/module.h>
51 #include <sys/mutex.h>
54 #include <sys/posix4.h>
55 #include <sys/semaphore.h>
56 #include <sys/_semaphore.h>
58 #include <sys/syscall.h>
59 #include <sys/syscallsubr.h>
60 #include <sys/sysctl.h>
61 #include <sys/sysent.h>
62 #include <sys/sysproto.h>
63 #include <sys/systm.h>
65 #include <sys/vnode.h>
67 #include <security/mac/mac_framework.h>
74 * - Replace global sem_lock with mtx_pool locks?
75 * - Add a MAC check_create() hook for creating new named semaphores.
83 #define DP(x) printf x
92 LIST_ENTRY(ksem_mapping) km_link;
95 static MALLOC_DEFINE(M_KSEM, "ksem", "semaphore file descriptor");
96 static LIST_HEAD(, ksem_mapping) *ksem_dictionary;
97 static struct sx ksem_dict_lock;
98 static struct mtx ksem_count_lock;
99 static struct mtx sem_lock;
100 static u_long ksem_hash;
101 static int ksem_dead;
103 #define KSEM_HASH(fnv) (&ksem_dictionary[(fnv) & ksem_hash])
105 static int nsems = 0;
106 SYSCTL_DECL(_p1003_1b);
107 SYSCTL_INT(_p1003_1b, OID_AUTO, nsems, CTLFLAG_RD, &nsems, 0,
108 "Number of active kernel POSIX semaphores");
110 static int kern_sem_wait(struct thread *td, semid_t id, int tryflag,
111 struct timespec *abstime);
112 static int ksem_access(struct ksem *ks, struct ucred *ucred);
113 static struct ksem *ksem_alloc(struct ucred *ucred, mode_t mode,
115 static int ksem_create(struct thread *td, const char *path,
116 semid_t *semidp, mode_t mode, unsigned int value,
117 int flags, int compat32);
118 static void ksem_drop(struct ksem *ks);
119 static int ksem_get(struct thread *td, semid_t id, struct file **fpp);
120 static struct ksem *ksem_hold(struct ksem *ks);
121 static void ksem_insert(char *path, Fnv32_t fnv, struct ksem *ks);
122 static struct ksem *ksem_lookup(char *path, Fnv32_t fnv);
123 static void ksem_module_destroy(void);
124 static int ksem_module_init(void);
125 static int ksem_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
126 static int sem_modload(struct module *module, int cmd, void *arg);
128 static fo_rdwr_t ksem_read;
129 static fo_rdwr_t ksem_write;
130 static fo_truncate_t ksem_truncate;
131 static fo_ioctl_t ksem_ioctl;
132 static fo_poll_t ksem_poll;
133 static fo_kqfilter_t ksem_kqfilter;
134 static fo_stat_t ksem_stat;
135 static fo_close_t ksem_closef;
137 /* File descriptor operations. */
138 static struct fileops ksem_ops = {
139 .fo_read = ksem_read,
140 .fo_write = ksem_write,
141 .fo_truncate = ksem_truncate,
142 .fo_ioctl = ksem_ioctl,
143 .fo_poll = ksem_poll,
144 .fo_kqfilter = ksem_kqfilter,
145 .fo_stat = ksem_stat,
146 .fo_close = ksem_closef,
147 .fo_flags = DFLAG_PASSABLE
150 FEATURE(posix_sem, "POSIX semaphores");
153 ksem_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
154 int flags, struct thread *td)
161 ksem_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
162 int flags, struct thread *td)
169 ksem_truncate(struct file *fp, off_t length, struct ucred *active_cred,
177 ksem_ioctl(struct file *fp, u_long com, void *data,
178 struct ucred *active_cred, struct thread *td)
185 ksem_poll(struct file *fp, int events, struct ucred *active_cred,
193 ksem_kqfilter(struct file *fp, struct knote *kn)
200 ksem_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
211 error = mac_posixsem_check_stat(active_cred, fp->f_cred, ks);
217 * Attempt to return sanish values for fstat() on a semaphore
220 bzero(sb, sizeof(*sb));
221 sb->st_mode = S_IFREG | ks->ks_mode; /* XXX */
223 sb->st_atimespec = ks->ks_atime;
224 sb->st_ctimespec = ks->ks_ctime;
225 sb->st_mtimespec = ks->ks_mtime;
226 sb->st_birthtimespec = ks->ks_birthtime;
227 sb->st_uid = ks->ks_uid;
228 sb->st_gid = ks->ks_gid;
234 ksem_closef(struct file *fp, struct thread *td)
246 * ksem object management including creation and reference counting
250 ksem_alloc(struct ucred *ucred, mode_t mode, unsigned int value)
254 mtx_lock(&ksem_count_lock);
255 if (nsems == p31b_getcfg(CTL_P1003_1B_SEM_NSEMS_MAX) || ksem_dead) {
256 mtx_unlock(&ksem_count_lock);
260 mtx_unlock(&ksem_count_lock);
261 ks = malloc(sizeof(*ks), M_KSEM, M_WAITOK | M_ZERO);
262 ks->ks_uid = ucred->cr_uid;
263 ks->ks_gid = ucred->cr_gid;
265 ks->ks_value = value;
266 cv_init(&ks->ks_cv, "ksem");
267 vfs_timestamp(&ks->ks_birthtime);
268 ks->ks_atime = ks->ks_mtime = ks->ks_ctime = ks->ks_birthtime;
269 refcount_init(&ks->ks_ref, 1);
271 mac_posixsem_init(ks);
272 mac_posixsem_create(ucred, ks);
279 ksem_hold(struct ksem *ks)
282 refcount_acquire(&ks->ks_ref);
287 ksem_drop(struct ksem *ks)
290 if (refcount_release(&ks->ks_ref)) {
292 mac_posixsem_destroy(ks);
294 cv_destroy(&ks->ks_cv);
296 mtx_lock(&ksem_count_lock);
298 mtx_unlock(&ksem_count_lock);
303 * Determine if the credentials have sufficient permissions for read
307 ksem_access(struct ksem *ks, struct ucred *ucred)
311 error = vaccess(VREG, ks->ks_mode, ks->ks_uid, ks->ks_gid,
312 VREAD | VWRITE, ucred, NULL);
314 error = priv_check_cred(ucred, PRIV_SEM_WRITE, 0);
319 * Dictionary management. We maintain an in-kernel dictionary to map
320 * paths to semaphore objects. We use the FNV hash on the path to
321 * store the mappings in a hash table.
324 ksem_lookup(char *path, Fnv32_t fnv)
326 struct ksem_mapping *map;
328 LIST_FOREACH(map, KSEM_HASH(fnv), km_link) {
329 if (map->km_fnv != fnv)
331 if (strcmp(map->km_path, path) == 0)
332 return (map->km_ksem);
339 ksem_insert(char *path, Fnv32_t fnv, struct ksem *ks)
341 struct ksem_mapping *map;
343 map = malloc(sizeof(struct ksem_mapping), M_KSEM, M_WAITOK);
346 map->km_ksem = ksem_hold(ks);
348 LIST_INSERT_HEAD(KSEM_HASH(fnv), map, km_link);
352 ksem_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
354 struct ksem_mapping *map;
357 LIST_FOREACH(map, KSEM_HASH(fnv), km_link) {
358 if (map->km_fnv != fnv)
360 if (strcmp(map->km_path, path) == 0) {
362 error = mac_posixsem_check_unlink(ucred, map->km_ksem);
366 error = ksem_access(map->km_ksem, ucred);
369 map->km_ksem->ks_path = NULL;
370 LIST_REMOVE(map, km_link);
371 ksem_drop(map->km_ksem);
372 free(map->km_path, M_KSEM);
382 ksem_info_impl(struct ksem *ks, char *path, size_t size, uint32_t *value)
385 if (ks->ks_path == NULL)
387 sx_slock(&ksem_dict_lock);
388 if (ks->ks_path != NULL)
389 strlcpy(path, ks->ks_path, size);
391 *value = ks->ks_value;
392 sx_sunlock(&ksem_dict_lock);
396 ksem_create_copyout_semid(struct thread *td, semid_t *semidp, int fd,
400 #ifdef COMPAT_FREEBSD32
406 #ifdef COMPAT_FREEBSD32
410 ptrs = sizeof(semid32);
415 ptrs = sizeof(semid);
416 compat32 = 0; /* silence gcc */
417 #ifdef COMPAT_FREEBSD32
421 return (copyout(ptr, semidp, ptrs));
424 /* Other helper routines. */
426 ksem_create(struct thread *td, const char *name, semid_t *semidp, mode_t mode,
427 unsigned int value, int flags, int compat32)
429 struct filedesc *fdp;
436 if (value > SEM_VALUE_MAX)
439 fdp = td->td_proc->p_fd;
440 mode = (mode & ~fdp->fd_cmask) & ACCESSPERMS;
441 error = falloc(td, &fp, &fd);
449 * Go ahead and copyout the file descriptor now. This is a bit
450 * premature, but it is a lot easier to handle errors as opposed
451 * to later when we've possibly created a new semaphore, etc.
453 error = ksem_create_copyout_semid(td, semidp, fd, compat32);
455 fdclose(fdp, fp, fd, td);
461 /* Create an anonymous semaphore. */
462 ks = ksem_alloc(td->td_ucred, mode, value);
466 ks->ks_flags |= KS_ANONYMOUS;
468 path = malloc(MAXPATHLEN, M_KSEM, M_WAITOK);
469 error = copyinstr(name, path, MAXPATHLEN, NULL);
471 /* Require paths to start with a '/' character. */
472 if (error == 0 && path[0] != '/')
475 fdclose(fdp, fp, fd, td);
481 fnv = fnv_32_str(path, FNV1_32_INIT);
482 sx_xlock(&ksem_dict_lock);
483 ks = ksem_lookup(path, fnv);
485 /* Object does not exist, create it if requested. */
486 if (flags & O_CREAT) {
487 ks = ksem_alloc(td->td_ucred, mode, value);
491 ksem_insert(path, fnv, ks);
498 * Object already exists, obtain a new
499 * reference if requested and permitted.
501 if ((flags & (O_CREAT | O_EXCL)) ==
506 error = mac_posixsem_check_open(td->td_ucred,
510 error = ksem_access(ks, td->td_ucred);
519 sx_xunlock(&ksem_dict_lock);
525 KASSERT(ks == NULL, ("ksem_create error with a ksem"));
526 fdclose(fdp, fp, fd, td);
530 KASSERT(ks != NULL, ("ksem_create w/o a ksem"));
532 finit(fp, FREAD | FWRITE, DTYPE_SEM, ks, &ksem_ops);
535 if (fdp->fd_ofiles[fd] == fp)
536 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
537 FILEDESC_XUNLOCK(fdp);
544 ksem_get(struct thread *td, semid_t id, struct file **fpp)
550 error = fget(td, id, &fp);
553 if (fp->f_type != DTYPE_SEM) {
558 if (ks->ks_flags & KS_DEAD) {
567 #ifndef _SYS_SYSPROTO_H_
568 struct ksem_init_args {
574 ksem_init(struct thread *td, struct ksem_init_args *uap)
577 return (ksem_create(td, NULL, uap->idp, S_IRWXU | S_IRWXG, uap->value,
581 #ifndef _SYS_SYSPROTO_H_
582 struct ksem_open_args {
591 ksem_open(struct thread *td, struct ksem_open_args *uap)
594 DP((">>> ksem_open start, pid=%d\n", (int)td->td_proc->p_pid));
596 if ((uap->oflag & ~(O_CREAT | O_EXCL)) != 0)
598 return (ksem_create(td, uap->name, uap->idp, uap->mode, uap->value,
602 #ifndef _SYS_SYSPROTO_H_
603 struct ksem_unlink_args {
608 ksem_unlink(struct thread *td, struct ksem_unlink_args *uap)
614 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
615 error = copyinstr(uap->name, path, MAXPATHLEN, NULL);
621 fnv = fnv_32_str(path, FNV1_32_INIT);
622 sx_xlock(&ksem_dict_lock);
623 error = ksem_remove(path, fnv, td->td_ucred);
624 sx_xunlock(&ksem_dict_lock);
630 #ifndef _SYS_SYSPROTO_H_
631 struct ksem_close_args {
636 ksem_close(struct thread *td, struct ksem_close_args *uap)
642 error = ksem_get(td, uap->id, &fp);
646 if (ks->ks_flags & KS_ANONYMOUS) {
650 error = kern_close(td, uap->id);
655 #ifndef _SYS_SYSPROTO_H_
656 struct ksem_post_args {
661 ksem_post(struct thread *td, struct ksem_post_args *uap)
667 error = ksem_get(td, uap->id, &fp);
674 error = mac_posixsem_check_post(td->td_ucred, fp->f_cred, ks);
678 if (ks->ks_value == SEM_VALUE_MAX) {
683 if (ks->ks_waiters > 0)
684 cv_signal(&ks->ks_cv);
686 vfs_timestamp(&ks->ks_ctime);
688 mtx_unlock(&sem_lock);
693 #ifndef _SYS_SYSPROTO_H_
694 struct ksem_wait_args {
699 ksem_wait(struct thread *td, struct ksem_wait_args *uap)
702 return (kern_sem_wait(td, uap->id, 0, NULL));
705 #ifndef _SYS_SYSPROTO_H_
706 struct ksem_timedwait_args {
708 const struct timespec *abstime;
712 ksem_timedwait(struct thread *td, struct ksem_timedwait_args *uap)
714 struct timespec abstime;
719 * We allow a null timespec (wait forever).
721 if (uap->abstime == NULL)
724 error = copyin(uap->abstime, &abstime, sizeof(abstime));
727 if (abstime.tv_nsec >= 1000000000 || abstime.tv_nsec < 0)
731 return (kern_sem_wait(td, uap->id, 0, ts));
734 #ifndef _SYS_SYSPROTO_H_
735 struct ksem_trywait_args {
740 ksem_trywait(struct thread *td, struct ksem_trywait_args *uap)
743 return (kern_sem_wait(td, uap->id, 1, NULL));
747 kern_sem_wait(struct thread *td, semid_t id, int tryflag,
748 struct timespec *abstime)
750 struct timespec ts1, ts2;
756 DP((">>> kern_sem_wait entered! pid=%d\n", (int)td->td_proc->p_pid));
757 error = ksem_get(td, id, &fp);
762 DP((">>> kern_sem_wait critical section entered! pid=%d\n",
763 (int)td->td_proc->p_pid));
765 error = mac_posixsem_check_wait(td->td_ucred, fp->f_cred, ks);
767 DP(("kern_sem_wait mac failed\n"));
771 DP(("kern_sem_wait value = %d, tryflag %d\n", ks->ks_value, tryflag));
772 vfs_timestamp(&ks->ks_atime);
773 while (ks->ks_value == 0) {
777 else if (abstime == NULL)
778 error = cv_wait_sig(&ks->ks_cv, &sem_lock);
783 timespecsub(&ts1, &ts2);
784 TIMESPEC_TO_TIMEVAL(&tv, &ts1);
789 error = cv_timedwait_sig(&ks->ks_cv,
790 &sem_lock, tvtohz(&tv));
791 if (error != EWOULDBLOCK)
800 DP(("kern_sem_wait value post-decrement = %d\n", ks->ks_value));
803 mtx_unlock(&sem_lock);
805 DP(("<<< kern_sem_wait leaving, pid=%d, error = %d\n",
806 (int)td->td_proc->p_pid, error));
810 #ifndef _SYS_SYSPROTO_H_
811 struct ksem_getvalue_args {
817 ksem_getvalue(struct thread *td, struct ksem_getvalue_args *uap)
823 error = ksem_get(td, uap->id, &fp);
830 error = mac_posixsem_check_getvalue(td->td_ucred, fp->f_cred, ks);
832 mtx_unlock(&sem_lock);
838 vfs_timestamp(&ks->ks_atime);
839 mtx_unlock(&sem_lock);
841 error = copyout(&val, uap->val, sizeof(val));
845 #ifndef _SYS_SYSPROTO_H_
846 struct ksem_destroy_args {
851 ksem_destroy(struct thread *td, struct ksem_destroy_args *uap)
857 error = ksem_get(td, uap->id, &fp);
861 if (!(ks->ks_flags & KS_ANONYMOUS)) {
866 if (ks->ks_waiters != 0) {
867 mtx_unlock(&sem_lock);
871 ks->ks_flags |= KS_DEAD;
872 mtx_unlock(&sem_lock);
874 error = kern_close(td, uap->id);
880 static struct syscall_helper_data ksem_syscalls[] = {
881 SYSCALL_INIT_HELPER(ksem_init),
882 SYSCALL_INIT_HELPER(ksem_open),
883 SYSCALL_INIT_HELPER(ksem_unlink),
884 SYSCALL_INIT_HELPER(ksem_close),
885 SYSCALL_INIT_HELPER(ksem_post),
886 SYSCALL_INIT_HELPER(ksem_wait),
887 SYSCALL_INIT_HELPER(ksem_timedwait),
888 SYSCALL_INIT_HELPER(ksem_trywait),
889 SYSCALL_INIT_HELPER(ksem_getvalue),
890 SYSCALL_INIT_HELPER(ksem_destroy),
894 #ifdef COMPAT_FREEBSD32
895 #include <compat/freebsd32/freebsd32.h>
896 #include <compat/freebsd32/freebsd32_proto.h>
897 #include <compat/freebsd32/freebsd32_signal.h>
898 #include <compat/freebsd32/freebsd32_syscall.h>
899 #include <compat/freebsd32/freebsd32_util.h>
902 freebsd32_ksem_init(struct thread *td, struct freebsd32_ksem_init_args *uap)
905 return (ksem_create(td, NULL, uap->idp, S_IRWXU | S_IRWXG, uap->value,
910 freebsd32_ksem_open(struct thread *td, struct freebsd32_ksem_open_args *uap)
913 if ((uap->oflag & ~(O_CREAT | O_EXCL)) != 0)
915 return (ksem_create(td, uap->name, uap->idp, uap->mode, uap->value,
920 freebsd32_ksem_timedwait(struct thread *td,
921 struct freebsd32_ksem_timedwait_args *uap)
923 struct timespec32 abstime32;
924 struct timespec *ts, abstime;
928 * We allow a null timespec (wait forever).
930 if (uap->abstime == NULL)
933 error = copyin(uap->abstime, &abstime32, sizeof(abstime32));
936 CP(abstime32, abstime, tv_sec);
937 CP(abstime32, abstime, tv_nsec);
938 if (abstime.tv_nsec >= 1000000000 || abstime.tv_nsec < 0)
942 return (kern_sem_wait(td, uap->id, 0, ts));
945 static struct syscall_helper_data ksem32_syscalls[] = {
946 SYSCALL32_INIT_HELPER(freebsd32_ksem_init),
947 SYSCALL32_INIT_HELPER(freebsd32_ksem_open),
948 SYSCALL32_INIT_HELPER(ksem_unlink),
949 SYSCALL32_INIT_HELPER(ksem_close),
950 SYSCALL32_INIT_HELPER(ksem_post),
951 SYSCALL32_INIT_HELPER(ksem_wait),
952 SYSCALL32_INIT_HELPER(freebsd32_ksem_timedwait),
953 SYSCALL32_INIT_HELPER(ksem_trywait),
954 SYSCALL32_INIT_HELPER(ksem_getvalue),
955 SYSCALL32_INIT_HELPER(ksem_destroy),
961 ksem_module_init(void)
965 mtx_init(&sem_lock, "sem", NULL, MTX_DEF);
966 mtx_init(&ksem_count_lock, "ksem count", NULL, MTX_DEF);
967 sx_init(&ksem_dict_lock, "ksem dictionary");
968 ksem_dictionary = hashinit(1024, M_KSEM, &ksem_hash);
969 p31b_setcfg(CTL_P1003_1B_SEMAPHORES, 200112L);
970 p31b_setcfg(CTL_P1003_1B_SEM_NSEMS_MAX, SEM_MAX);
971 p31b_setcfg(CTL_P1003_1B_SEM_VALUE_MAX, SEM_VALUE_MAX);
972 ksem_info = ksem_info_impl;
974 error = syscall_helper_register(ksem_syscalls);
977 #ifdef COMPAT_FREEBSD32
978 error = syscall32_helper_register(ksem32_syscalls);
986 ksem_module_destroy(void)
989 #ifdef COMPAT_FREEBSD32
990 syscall32_helper_unregister(ksem32_syscalls);
992 syscall_helper_unregister(ksem_syscalls);
995 p31b_setcfg(CTL_P1003_1B_SEMAPHORES, 0);
996 hashdestroy(ksem_dictionary, M_KSEM, ksem_hash);
997 sx_destroy(&ksem_dict_lock);
998 mtx_destroy(&ksem_count_lock);
999 mtx_destroy(&sem_lock);
1000 p31b_unsetcfg(CTL_P1003_1B_SEM_VALUE_MAX);
1001 p31b_unsetcfg(CTL_P1003_1B_SEM_NSEMS_MAX);
1005 sem_modload(struct module *module, int cmd, void *arg)
1011 error = ksem_module_init();
1013 ksem_module_destroy();
1017 mtx_lock(&ksem_count_lock);
1020 mtx_unlock(&ksem_count_lock);
1024 mtx_unlock(&ksem_count_lock);
1025 ksem_module_destroy();
1037 static moduledata_t sem_mod = {
1043 DECLARE_MODULE(sem, sem_mod, SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
1044 MODULE_VERSION(sem, 1);