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/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,
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);
347 LIST_INSERT_HEAD(KSEM_HASH(fnv), map, km_link);
351 ksem_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
353 struct ksem_mapping *map;
356 LIST_FOREACH(map, KSEM_HASH(fnv), km_link) {
357 if (map->km_fnv != fnv)
359 if (strcmp(map->km_path, path) == 0) {
361 error = mac_posixsem_check_unlink(ucred, map->km_ksem);
365 error = ksem_access(map->km_ksem, ucred);
368 LIST_REMOVE(map, km_link);
369 ksem_drop(map->km_ksem);
370 free(map->km_path, M_KSEM);
379 /* Other helper routines. */
381 ksem_create(struct thread *td, const char *name, semid_t *semidp, mode_t mode,
382 unsigned int value, int flags)
384 struct filedesc *fdp;
392 if (value > SEM_VALUE_MAX)
395 fdp = td->td_proc->p_fd;
396 mode = (mode & ~fdp->fd_cmask) & ACCESSPERMS;
397 error = falloc(td, &fp, &fd);
405 * Go ahead and copyout the file descriptor now. This is a bit
406 * premature, but it is a lot easier to handle errors as opposed
407 * to later when we've possibly created a new semaphore, etc.
410 error = copyout(&semid, semidp, sizeof(semid));
412 fdclose(fdp, fp, fd, td);
418 /* Create an anonymous semaphore. */
419 ks = ksem_alloc(td->td_ucred, mode, value);
423 ks->ks_flags |= KS_ANONYMOUS;
425 path = malloc(MAXPATHLEN, M_KSEM, M_WAITOK);
426 error = copyinstr(name, path, MAXPATHLEN, NULL);
428 /* Require paths to start with a '/' character. */
429 if (error == 0 && path[0] != '/')
432 fdclose(fdp, fp, fd, td);
438 fnv = fnv_32_str(path, FNV1_32_INIT);
439 sx_xlock(&ksem_dict_lock);
440 ks = ksem_lookup(path, fnv);
442 /* Object does not exist, create it if requested. */
443 if (flags & O_CREAT) {
444 ks = ksem_alloc(td->td_ucred, mode, value);
448 ksem_insert(path, fnv, ks);
455 * Object already exists, obtain a new
456 * reference if requested and permitted.
458 if ((flags & (O_CREAT | O_EXCL)) ==
463 error = mac_posixsem_check_open(td->td_ucred,
467 error = ksem_access(ks, td->td_ucred);
476 sx_xunlock(&ksem_dict_lock);
482 KASSERT(ks == NULL, ("ksem_create error with a ksem"));
483 fdclose(fdp, fp, fd, td);
487 KASSERT(ks != NULL, ("ksem_create w/o a ksem"));
489 finit(fp, FREAD | FWRITE, DTYPE_SEM, ks, &ksem_ops);
492 if (fdp->fd_ofiles[fd] == fp)
493 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
494 FILEDESC_XUNLOCK(fdp);
501 ksem_get(struct thread *td, semid_t id, struct file **fpp)
507 error = fget(td, id, &fp);
510 if (fp->f_type != DTYPE_SEM) {
515 if (ks->ks_flags & KS_DEAD) {
524 #ifndef _SYS_SYSPROTO_H_
525 struct ksem_init_args {
531 ksem_init(struct thread *td, struct ksem_init_args *uap)
534 return (ksem_create(td, NULL, uap->idp, S_IRWXU | S_IRWXG, uap->value,
538 #ifndef _SYS_SYSPROTO_H_
539 struct ksem_open_args {
548 ksem_open(struct thread *td, struct ksem_open_args *uap)
551 if ((uap->oflag & ~(O_CREAT | O_EXCL)) != 0)
553 return (ksem_create(td, uap->name, uap->idp, uap->mode, uap->value,
557 #ifndef _SYS_SYSPROTO_H_
558 struct ksem_unlink_args {
563 ksem_unlink(struct thread *td, struct ksem_unlink_args *uap)
569 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
570 error = copyinstr(uap->name, path, MAXPATHLEN, NULL);
576 fnv = fnv_32_str(path, FNV1_32_INIT);
577 sx_xlock(&ksem_dict_lock);
578 error = ksem_remove(path, fnv, td->td_ucred);
579 sx_xunlock(&ksem_dict_lock);
585 #ifndef _SYS_SYSPROTO_H_
586 struct ksem_close_args {
591 ksem_close(struct thread *td, struct ksem_close_args *uap)
597 error = ksem_get(td, uap->id, &fp);
601 if (ks->ks_flags & KS_ANONYMOUS) {
605 error = kern_close(td, uap->id);
610 #ifndef _SYS_SYSPROTO_H_
611 struct ksem_post_args {
616 ksem_post(struct thread *td, struct ksem_post_args *uap)
622 error = ksem_get(td, uap->id, &fp);
629 error = mac_posixsem_check_post(td->td_ucred, fp->f_cred, ks);
633 if (ks->ks_value == SEM_VALUE_MAX) {
638 if (ks->ks_waiters > 0)
639 cv_signal(&ks->ks_cv);
641 vfs_timestamp(&ks->ks_ctime);
643 mtx_unlock(&sem_lock);
648 #ifndef _SYS_SYSPROTO_H_
649 struct ksem_wait_args {
654 ksem_wait(struct thread *td, struct ksem_wait_args *uap)
657 return (kern_sem_wait(td, uap->id, 0, NULL));
660 #ifndef _SYS_SYSPROTO_H_
661 struct ksem_timedwait_args {
663 const struct timespec *abstime;
667 ksem_timedwait(struct thread *td, struct ksem_timedwait_args *uap)
669 struct timespec abstime;
674 * We allow a null timespec (wait forever).
676 if (uap->abstime == NULL)
679 error = copyin(uap->abstime, &abstime, sizeof(abstime));
682 if (abstime.tv_nsec >= 1000000000 || abstime.tv_nsec < 0)
686 return (kern_sem_wait(td, uap->id, 0, ts));
689 #ifndef _SYS_SYSPROTO_H_
690 struct ksem_trywait_args {
695 ksem_trywait(struct thread *td, struct ksem_trywait_args *uap)
698 return (kern_sem_wait(td, uap->id, 1, NULL));
702 kern_sem_wait(struct thread *td, semid_t id, int tryflag,
703 struct timespec *abstime)
705 struct timespec ts1, ts2;
711 DP((">>> kern_sem_wait entered!\n"));
712 error = ksem_get(td, id, &fp);
718 error = mac_posixsem_check_wait(td->td_ucred, fp->f_cred, ks);
720 DP(("kern_sem_wait mac failed\n"));
724 DP(("kern_sem_wait value = %d, tryflag %d\n", ks->ks_value, tryflag));
725 vfs_timestamp(&ks->ks_atime);
726 if (ks->ks_value == 0) {
730 else if (abstime == NULL)
731 error = cv_wait_sig(&ks->ks_cv, &sem_lock);
736 timespecsub(&ts1, &ts2);
737 TIMESPEC_TO_TIMEVAL(&tv, &ts1);
742 error = cv_timedwait_sig(&ks->ks_cv,
743 &sem_lock, tvtohz(&tv));
744 if (error != EWOULDBLOCK)
755 mtx_unlock(&sem_lock);
757 DP(("<<< kern_sem_wait leaving, error = %d\n", error));
761 #ifndef _SYS_SYSPROTO_H_
762 struct ksem_getvalue_args {
768 ksem_getvalue(struct thread *td, struct ksem_getvalue_args *uap)
774 error = ksem_get(td, uap->id, &fp);
781 error = mac_posixsem_check_getvalue(td->td_ucred, fp->f_cred, ks);
783 mtx_unlock(&sem_lock);
789 vfs_timestamp(&ks->ks_atime);
790 mtx_unlock(&sem_lock);
792 error = copyout(&val, uap->val, sizeof(val));
796 #ifndef _SYS_SYSPROTO_H_
797 struct ksem_destroy_args {
802 ksem_destroy(struct thread *td, struct ksem_destroy_args *uap)
808 error = ksem_get(td, uap->id, &fp);
812 if (!(ks->ks_flags & KS_ANONYMOUS)) {
817 if (ks->ks_waiters != 0) {
818 mtx_unlock(&sem_lock);
822 ks->ks_flags |= KS_DEAD;
823 mtx_unlock(&sem_lock);
825 error = kern_close(td, uap->id);
831 #define SYSCALL_DATA(syscallname) \
832 static int syscallname##_syscall = SYS_##syscallname; \
833 static int syscallname##_registered; \
834 static struct sysent syscallname##_old_sysent; \
835 MAKE_SYSENT(syscallname);
837 #define SYSCALL_REGISTER(syscallname) do { \
838 error = syscall_register(& syscallname##_syscall, \
839 & syscallname##_sysent, & syscallname##_old_sysent); \
842 syscallname##_registered = 1; \
845 #define SYSCALL_DEREGISTER(syscallname) do { \
846 if (syscallname##_registered) { \
847 syscallname##_registered = 0; \
848 syscall_deregister(& syscallname##_syscall, \
849 & syscallname##_old_sysent); \
853 SYSCALL_DATA(ksem_init);
854 SYSCALL_DATA(ksem_open);
855 SYSCALL_DATA(ksem_unlink);
856 SYSCALL_DATA(ksem_close);
857 SYSCALL_DATA(ksem_post);
858 SYSCALL_DATA(ksem_wait);
859 SYSCALL_DATA(ksem_timedwait);
860 SYSCALL_DATA(ksem_trywait);
861 SYSCALL_DATA(ksem_getvalue);
862 SYSCALL_DATA(ksem_destroy);
865 ksem_module_init(void)
869 mtx_init(&sem_lock, "sem", NULL, MTX_DEF);
870 mtx_init(&ksem_count_lock, "ksem count", NULL, MTX_DEF);
871 sx_init(&ksem_dict_lock, "ksem dictionary");
872 ksem_dictionary = hashinit(1024, M_KSEM, &ksem_hash);
873 p31b_setcfg(CTL_P1003_1B_SEM_NSEMS_MAX, SEM_MAX);
874 p31b_setcfg(CTL_P1003_1B_SEM_VALUE_MAX, SEM_VALUE_MAX);
876 SYSCALL_REGISTER(ksem_init);
877 SYSCALL_REGISTER(ksem_open);
878 SYSCALL_REGISTER(ksem_unlink);
879 SYSCALL_REGISTER(ksem_close);
880 SYSCALL_REGISTER(ksem_post);
881 SYSCALL_REGISTER(ksem_wait);
882 SYSCALL_REGISTER(ksem_timedwait);
883 SYSCALL_REGISTER(ksem_trywait);
884 SYSCALL_REGISTER(ksem_getvalue);
885 SYSCALL_REGISTER(ksem_destroy);
890 ksem_module_destroy(void)
893 SYSCALL_DEREGISTER(ksem_init);
894 SYSCALL_DEREGISTER(ksem_open);
895 SYSCALL_DEREGISTER(ksem_unlink);
896 SYSCALL_DEREGISTER(ksem_close);
897 SYSCALL_DEREGISTER(ksem_post);
898 SYSCALL_DEREGISTER(ksem_wait);
899 SYSCALL_DEREGISTER(ksem_timedwait);
900 SYSCALL_DEREGISTER(ksem_trywait);
901 SYSCALL_DEREGISTER(ksem_getvalue);
902 SYSCALL_DEREGISTER(ksem_destroy);
904 hashdestroy(ksem_dictionary, M_KSEM, ksem_hash);
905 sx_destroy(&ksem_dict_lock);
906 mtx_destroy(&ksem_count_lock);
907 mtx_destroy(&sem_lock);
911 sem_modload(struct module *module, int cmd, void *arg)
917 error = ksem_module_init();
919 ksem_module_destroy();
923 mtx_lock(&ksem_count_lock);
926 mtx_unlock(&ksem_count_lock);
930 mtx_unlock(&ksem_count_lock);
931 ksem_module_destroy();
943 static moduledata_t sem_mod = {
949 DECLARE_MODULE(sem, sem_mod, SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
950 MODULE_VERSION(sem, 1);