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/refcount.h>
56 #include <sys/semaphore.h>
57 #include <sys/_semaphore.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>
66 #include <sys/vnode.h>
68 #include <security/mac/mac_framework.h>
75 * - Replace global sem_lock with mtx_pool locks?
76 * - Add a MAC check_create() hook for creating new named semaphores.
84 #define DP(x) printf x
93 LIST_ENTRY(ksem_mapping) km_link;
96 static MALLOC_DEFINE(M_KSEM, "ksem", "semaphore file descriptor");
97 static LIST_HEAD(, ksem_mapping) *ksem_dictionary;
98 static struct sx ksem_dict_lock;
99 static struct mtx ksem_count_lock;
100 static struct mtx sem_lock;
101 static u_long ksem_hash;
102 static int ksem_dead;
104 #define KSEM_HASH(fnv) (&ksem_dictionary[(fnv) & ksem_hash])
106 static int nsems = 0;
107 SYSCTL_DECL(_p1003_1b);
108 SYSCTL_INT(_p1003_1b, OID_AUTO, nsems, CTLFLAG_RD, &nsems, 0,
109 "Number of active kernel POSIX semaphores");
111 static int kern_sem_wait(struct thread *td, semid_t id, int tryflag,
112 struct timespec *abstime);
113 static int ksem_access(struct ksem *ks, struct ucred *ucred);
114 static struct ksem *ksem_alloc(struct ucred *ucred, mode_t mode,
116 static int ksem_create(struct thread *td, const char *path,
117 semid_t *semidp, mode_t mode, unsigned int value,
119 static void ksem_drop(struct ksem *ks);
120 static int ksem_get(struct thread *td, semid_t id, struct file **fpp);
121 static struct ksem *ksem_hold(struct ksem *ks);
122 static void ksem_insert(char *path, Fnv32_t fnv, struct ksem *ks);
123 static struct ksem *ksem_lookup(char *path, Fnv32_t fnv);
124 static void ksem_module_destroy(void);
125 static int ksem_module_init(void);
126 static int ksem_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
127 static int sem_modload(struct module *module, int cmd, void *arg);
129 static fo_rdwr_t ksem_read;
130 static fo_rdwr_t ksem_write;
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_ioctl = ksem_ioctl,
142 .fo_poll = ksem_poll,
143 .fo_kqfilter = ksem_kqfilter,
144 .fo_stat = ksem_stat,
145 .fo_close = ksem_closef,
146 .fo_flags = DFLAG_PASSABLE
149 FEATURE(posix_sem, "POSIX semaphores");
152 ksem_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
153 int flags, struct thread *td)
160 ksem_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
161 int flags, struct thread *td)
168 ksem_ioctl(struct file *fp, u_long com, void *data,
169 struct ucred *active_cred, struct thread *td)
176 ksem_poll(struct file *fp, int events, struct ucred *active_cred,
184 ksem_kqfilter(struct file *fp, struct knote *kn)
191 ksem_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
203 * XXX: This isn't quite right, but adding a new stat() hook
204 * would break the ABI. This seems to be the closest match of
205 * the existing MAC hooks for POSIX semaphores in 7.
207 error = mac_check_posix_sem_getvalue(active_cred, ks);
213 * Attempt to return sanish values for fstat() on a semaphore
216 bzero(sb, sizeof(*sb));
217 sb->st_mode = S_IFREG | ks->ks_mode; /* XXX */
219 sb->st_atimespec = ks->ks_atime;
220 sb->st_ctimespec = ks->ks_ctime;
221 sb->st_mtimespec = ks->ks_mtime;
222 sb->st_birthtimespec = ks->ks_birthtime;
223 sb->st_uid = ks->ks_uid;
224 sb->st_gid = ks->ks_gid;
230 ksem_closef(struct file *fp, struct thread *td)
242 * ksem object management including creation and reference counting
246 ksem_alloc(struct ucred *ucred, mode_t mode, unsigned int value)
250 mtx_lock(&ksem_count_lock);
251 if (nsems == p31b_getcfg(CTL_P1003_1B_SEM_NSEMS_MAX) || ksem_dead) {
252 mtx_unlock(&ksem_count_lock);
256 mtx_unlock(&ksem_count_lock);
257 ks = malloc(sizeof(*ks), M_KSEM, M_WAITOK | M_ZERO);
258 ks->ks_uid = ucred->cr_uid;
259 ks->ks_gid = ucred->cr_gid;
261 ks->ks_value = value;
262 cv_init(&ks->ks_cv, "ksem");
263 vfs_timestamp(&ks->ks_birthtime);
264 ks->ks_atime = ks->ks_mtime = ks->ks_ctime = ks->ks_birthtime;
265 refcount_init(&ks->ks_ref, 1);
267 mac_init_posix_sem(ks);
268 mac_create_posix_sem(ucred, ks);
275 ksem_hold(struct ksem *ks)
278 refcount_acquire(&ks->ks_ref);
283 ksem_drop(struct ksem *ks)
286 if (refcount_release(&ks->ks_ref)) {
288 mac_destroy_posix_sem(ks);
290 cv_destroy(&ks->ks_cv);
292 mtx_lock(&ksem_count_lock);
294 mtx_unlock(&ksem_count_lock);
299 * Determine if the credentials have sufficient permissions for read
303 ksem_access(struct ksem *ks, struct ucred *ucred)
307 error = vaccess(VREG, ks->ks_mode, ks->ks_uid, ks->ks_gid,
308 VREAD | VWRITE, ucred, NULL);
310 error = priv_check_cred(ucred, PRIV_SEM_WRITE, 0);
315 * Dictionary management. We maintain an in-kernel dictionary to map
316 * paths to semaphore objects. We use the FNV hash on the path to
317 * store the mappings in a hash table.
320 ksem_lookup(char *path, Fnv32_t fnv)
322 struct ksem_mapping *map;
324 LIST_FOREACH(map, KSEM_HASH(fnv), km_link) {
325 if (map->km_fnv != fnv)
327 if (strcmp(map->km_path, path) == 0)
328 return (map->km_ksem);
335 ksem_insert(char *path, Fnv32_t fnv, struct ksem *ks)
337 struct ksem_mapping *map;
339 map = malloc(sizeof(struct ksem_mapping), M_KSEM, M_WAITOK);
342 map->km_ksem = ksem_hold(ks);
343 LIST_INSERT_HEAD(KSEM_HASH(fnv), map, km_link);
347 ksem_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
349 struct ksem_mapping *map;
352 LIST_FOREACH(map, KSEM_HASH(fnv), km_link) {
353 if (map->km_fnv != fnv)
355 if (strcmp(map->km_path, path) == 0) {
357 error = mac_check_posix_sem_unlink(ucred, map->km_ksem);
361 error = ksem_access(map->km_ksem, ucred);
364 LIST_REMOVE(map, km_link);
365 ksem_drop(map->km_ksem);
366 free(map->km_path, M_KSEM);
375 /* Other helper routines. */
377 ksem_create(struct thread *td, const char *name, semid_t *semidp, mode_t mode,
378 unsigned int value, int flags)
380 struct filedesc *fdp;
388 if (value > SEM_VALUE_MAX)
391 fdp = td->td_proc->p_fd;
392 mode = (mode & ~fdp->fd_cmask) & ACCESSPERMS;
393 error = falloc(td, &fp, &fd);
401 * Go ahead and copyout the file descriptor now. This is a bit
402 * premature, but it is a lot easier to handle errors as opposed
403 * to later when we've possibly created a new semaphore, etc.
406 error = copyout(&semid, semidp, sizeof(semid));
408 fdclose(fdp, fp, fd, td);
414 /* Create an anonymous semaphore. */
415 ks = ksem_alloc(td->td_ucred, mode, value);
419 ks->ks_flags |= KS_ANONYMOUS;
421 path = malloc(MAXPATHLEN, M_KSEM, M_WAITOK);
422 error = copyinstr(name, path, MAXPATHLEN, NULL);
424 /* Require paths to start with a '/' character. */
425 if (error == 0 && path[0] != '/')
428 fdclose(fdp, fp, fd, td);
434 fnv = fnv_32_str(path, FNV1_32_INIT);
435 sx_xlock(&ksem_dict_lock);
436 ks = ksem_lookup(path, fnv);
438 /* Object does not exist, create it if requested. */
439 if (flags & O_CREAT) {
440 ks = ksem_alloc(td->td_ucred, mode, value);
444 ksem_insert(path, fnv, ks);
451 * Object already exists, obtain a new
452 * reference if requested and permitted.
454 if ((flags & (O_CREAT | O_EXCL)) ==
459 error = mac_check_posix_sem_open(td->td_ucred,
463 error = ksem_access(ks, td->td_ucred);
472 sx_xunlock(&ksem_dict_lock);
478 KASSERT(ks == NULL, ("ksem_create error with a ksem"));
479 fdclose(fdp, fp, fd, td);
483 KASSERT(ks != NULL, ("ksem_create w/o a ksem"));
486 fp->f_flag = FREAD | FWRITE;
487 fp->f_type = DTYPE_SEM;
488 fp->f_ops = &ksem_ops;
491 if (fdp->fd_ofiles[fd] == fp)
492 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
493 FILEDESC_XUNLOCK(fdp);
500 ksem_get(struct thread *td, semid_t id, struct file **fpp)
506 error = fget(td, id, &fp);
509 if (fp->f_type != DTYPE_SEM) {
514 if (ks->ks_flags & KS_DEAD) {
523 #ifndef _SYS_SYSPROTO_H_
524 struct ksem_init_args {
530 ksem_init(struct thread *td, struct ksem_init_args *uap)
533 return (ksem_create(td, NULL, uap->idp, S_IRWXU | S_IRWXG, uap->value,
537 #ifndef _SYS_SYSPROTO_H_
538 struct ksem_open_args {
547 ksem_open(struct thread *td, struct ksem_open_args *uap)
550 if ((uap->oflag & ~(O_CREAT | O_EXCL)) != 0)
552 return (ksem_create(td, uap->name, uap->idp, uap->mode, uap->value,
556 #ifndef _SYS_SYSPROTO_H_
557 struct ksem_unlink_args {
562 ksem_unlink(struct thread *td, struct ksem_unlink_args *uap)
568 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
569 error = copyinstr(uap->name, path, MAXPATHLEN, NULL);
575 fnv = fnv_32_str(path, FNV1_32_INIT);
576 sx_xlock(&ksem_dict_lock);
577 error = ksem_remove(path, fnv, td->td_ucred);
578 sx_xunlock(&ksem_dict_lock);
584 #ifndef _SYS_SYSPROTO_H_
585 struct ksem_close_args {
590 ksem_close(struct thread *td, struct ksem_close_args *uap)
596 error = ksem_get(td, uap->id, &fp);
600 if (ks->ks_flags & KS_ANONYMOUS) {
604 error = kern_close(td, uap->id);
609 #ifndef _SYS_SYSPROTO_H_
610 struct ksem_post_args {
615 ksem_post(struct thread *td, struct ksem_post_args *uap)
621 error = ksem_get(td, uap->id, &fp);
628 error = mac_check_posix_sem_post(td->td_ucred, ks);
632 if (ks->ks_value == SEM_VALUE_MAX) {
637 if (ks->ks_waiters > 0)
638 cv_signal(&ks->ks_cv);
640 vfs_timestamp(&ks->ks_ctime);
642 mtx_unlock(&sem_lock);
647 #ifndef _SYS_SYSPROTO_H_
648 struct ksem_wait_args {
653 ksem_wait(struct thread *td, struct ksem_wait_args *uap)
656 return (kern_sem_wait(td, uap->id, 0, NULL));
659 #ifndef _SYS_SYSPROTO_H_
660 struct ksem_timedwait_args {
662 const struct timespec *abstime;
666 ksem_timedwait(struct thread *td, struct ksem_timedwait_args *uap)
668 struct timespec abstime;
673 * We allow a null timespec (wait forever).
675 if (uap->abstime == NULL)
678 error = copyin(uap->abstime, &abstime, sizeof(abstime));
681 if (abstime.tv_nsec >= 1000000000 || abstime.tv_nsec < 0)
685 return (kern_sem_wait(td, uap->id, 0, ts));
688 #ifndef _SYS_SYSPROTO_H_
689 struct ksem_trywait_args {
694 ksem_trywait(struct thread *td, struct ksem_trywait_args *uap)
697 return (kern_sem_wait(td, uap->id, 1, NULL));
701 kern_sem_wait(struct thread *td, semid_t id, int tryflag,
702 struct timespec *abstime)
704 struct timespec ts1, ts2;
710 DP((">>> kern_sem_wait entered!\n"));
711 error = ksem_get(td, id, &fp);
717 error = mac_check_posix_sem_wait(td->td_ucred, ks);
719 DP(("kern_sem_wait mac failed\n"));
723 DP(("kern_sem_wait value = %d, tryflag %d\n", ks->ks_value, tryflag));
724 vfs_timestamp(&ks->ks_atime);
725 while (ks->ks_value == 0) {
729 else if (abstime == NULL)
730 error = cv_wait_sig(&ks->ks_cv, &sem_lock);
735 timespecsub(&ts1, &ts2);
736 TIMESPEC_TO_TIMEVAL(&tv, &ts1);
741 error = cv_timedwait_sig(&ks->ks_cv,
742 &sem_lock, tvtohz(&tv));
743 if (error != EWOULDBLOCK)
754 mtx_unlock(&sem_lock);
756 DP(("<<< kern_sem_wait leaving, error = %d\n", error));
760 #ifndef _SYS_SYSPROTO_H_
761 struct ksem_getvalue_args {
767 ksem_getvalue(struct thread *td, struct ksem_getvalue_args *uap)
773 error = ksem_get(td, uap->id, &fp);
780 error = mac_check_posix_sem_getvalue(td->td_ucred, ks);
782 mtx_unlock(&sem_lock);
788 vfs_timestamp(&ks->ks_atime);
789 mtx_unlock(&sem_lock);
791 error = copyout(&val, uap->val, sizeof(val));
795 #ifndef _SYS_SYSPROTO_H_
796 struct ksem_destroy_args {
801 ksem_destroy(struct thread *td, struct ksem_destroy_args *uap)
807 error = ksem_get(td, uap->id, &fp);
811 if (!(ks->ks_flags & KS_ANONYMOUS)) {
816 if (ks->ks_waiters != 0) {
817 mtx_unlock(&sem_lock);
821 ks->ks_flags |= KS_DEAD;
822 mtx_unlock(&sem_lock);
824 error = kern_close(td, uap->id);
830 #define SYSCALL_DATA(syscallname) \
831 static int syscallname##_syscall = SYS_##syscallname; \
832 static int syscallname##_registered; \
833 static struct sysent syscallname##_old_sysent; \
834 MAKE_SYSENT(syscallname);
836 #define SYSCALL_REGISTER(syscallname) do { \
837 error = syscall_register(& syscallname##_syscall, \
838 & syscallname##_sysent, & syscallname##_old_sysent); \
841 syscallname##_registered = 1; \
844 #define SYSCALL_DEREGISTER(syscallname) do { \
845 if (syscallname##_registered) { \
846 syscallname##_registered = 0; \
847 syscall_deregister(& syscallname##_syscall, \
848 & syscallname##_old_sysent); \
852 SYSCALL_DATA(ksem_init);
853 SYSCALL_DATA(ksem_open);
854 SYSCALL_DATA(ksem_unlink);
855 SYSCALL_DATA(ksem_close);
856 SYSCALL_DATA(ksem_post);
857 SYSCALL_DATA(ksem_wait);
858 SYSCALL_DATA(ksem_timedwait);
859 SYSCALL_DATA(ksem_trywait);
860 SYSCALL_DATA(ksem_getvalue);
861 SYSCALL_DATA(ksem_destroy);
864 ksem_module_init(void)
868 mtx_init(&sem_lock, "sem", NULL, MTX_DEF);
869 mtx_init(&ksem_count_lock, "ksem count", NULL, MTX_DEF);
870 sx_init(&ksem_dict_lock, "ksem dictionary");
871 ksem_dictionary = hashinit(1024, M_KSEM, &ksem_hash);
872 p31b_setcfg(CTL_P1003_1B_SEM_NSEMS_MAX, SEM_MAX);
873 p31b_setcfg(CTL_P1003_1B_SEM_VALUE_MAX, SEM_VALUE_MAX);
875 SYSCALL_REGISTER(ksem_init);
876 SYSCALL_REGISTER(ksem_open);
877 SYSCALL_REGISTER(ksem_unlink);
878 SYSCALL_REGISTER(ksem_close);
879 SYSCALL_REGISTER(ksem_post);
880 SYSCALL_REGISTER(ksem_wait);
881 SYSCALL_REGISTER(ksem_timedwait);
882 SYSCALL_REGISTER(ksem_trywait);
883 SYSCALL_REGISTER(ksem_getvalue);
884 SYSCALL_REGISTER(ksem_destroy);
889 ksem_module_destroy(void)
892 SYSCALL_DEREGISTER(ksem_init);
893 SYSCALL_DEREGISTER(ksem_open);
894 SYSCALL_DEREGISTER(ksem_unlink);
895 SYSCALL_DEREGISTER(ksem_close);
896 SYSCALL_DEREGISTER(ksem_post);
897 SYSCALL_DEREGISTER(ksem_wait);
898 SYSCALL_DEREGISTER(ksem_timedwait);
899 SYSCALL_DEREGISTER(ksem_trywait);
900 SYSCALL_DEREGISTER(ksem_getvalue);
901 SYSCALL_DEREGISTER(ksem_destroy);
903 hashdestroy(ksem_dictionary, M_KSEM, ksem_hash);
904 sx_destroy(&ksem_dict_lock);
905 mtx_destroy(&ksem_count_lock);
906 mtx_destroy(&sem_lock);
910 sem_modload(struct module *module, int cmd, void *arg)
916 error = ksem_module_init();
918 ksem_module_destroy();
922 mtx_lock(&ksem_count_lock);
925 mtx_unlock(&ksem_count_lock);
929 mtx_unlock(&ksem_count_lock);
930 ksem_module_destroy();
942 static moduledata_t sem_mod = {
948 DECLARE_MODULE(sem, sem_mod, SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
949 MODULE_VERSION(sem, 1);