2 * Implementation of SVID semaphores
4 * Author: Daniel Boulet
6 * This software is provided ``AS IS'' without any warranties of any kind.
9 * Copyright (c) 2003-2005 McAfee, Inc.
10 * All rights reserved.
12 * This software was developed for the FreeBSD Project in part by McAfee
13 * Research, the Security Research Division of McAfee, Inc under DARPA/SPAWAR
14 * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS research
17 * Redistribution and use in source and binary forms, with or without
18 * modification, are permitted provided that the following conditions
20 * 1. Redistributions of source code must retain the above copyright
21 * notice, this list of conditions and the following disclaimer.
22 * 2. Redistributions in binary form must reproduce the above copyright
23 * notice, this list of conditions and the following disclaimer in the
24 * documentation and/or other materials provided with the distribution.
26 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
42 #include "opt_compat.h"
43 #include "opt_sysvipc.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/eventhandler.h>
49 #include <sys/kernel.h>
52 #include <sys/module.h>
53 #include <sys/mutex.h>
55 #include <sys/syscall.h>
56 #include <sys/syscallsubr.h>
57 #include <sys/sysent.h>
58 #include <sys/sysctl.h>
60 #include <sys/malloc.h>
63 #include <security/mac/mac_framework.h>
65 static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores");
68 #define DPRINTF(a) printf a
73 static int seminit(void);
74 static int sysvsem_modload(struct module *, int, void *);
75 static int semunload(void);
76 static void semexit_myhook(void *arg, struct proc *p);
77 static int sysctl_sema(SYSCTL_HANDLER_ARGS);
78 static int semvalid(int semid, struct semid_kernel *semakptr);
80 #ifndef _SYS_SYSPROTO_H_
82 int __semctl(struct thread *td, struct __semctl_args *uap);
84 int semget(struct thread *td, struct semget_args *uap);
86 int semop(struct thread *td, struct semop_args *uap);
89 static struct sem_undo *semu_alloc(struct thread *td);
90 static int semundo_adjust(struct thread *td, struct sem_undo **supptr,
91 int semid, int semseq, int semnum, int adjval);
92 static void semundo_clear(int semid, int semnum);
94 static struct mtx sem_mtx; /* semaphore global lock */
95 static struct mtx sem_undo_mtx;
96 static int semtot = 0;
97 static struct semid_kernel *sema; /* semaphore id pool */
98 static struct mtx *sema_mtx; /* semaphore id pool mutexes*/
99 static struct sem *sem; /* semaphore pool */
100 LIST_HEAD(, sem_undo) semu_list; /* list of active undo structures */
101 LIST_HEAD(, sem_undo) semu_free_list; /* list of free undo structures */
102 static int *semu; /* undo structure pool */
103 static eventhandler_tag semexit_tag;
105 #define SEMUNDO_MTX sem_undo_mtx
106 #define SEMUNDO_LOCK() mtx_lock(&SEMUNDO_MTX);
107 #define SEMUNDO_UNLOCK() mtx_unlock(&SEMUNDO_MTX);
108 #define SEMUNDO_LOCKASSERT(how) mtx_assert(&SEMUNDO_MTX, (how));
111 u_short semval; /* semaphore value */
112 pid_t sempid; /* pid of last operation */
113 u_short semncnt; /* # awaiting semval > cval */
114 u_short semzcnt; /* # awaiting semval = 0 */
118 * Undo structure (one per process)
121 LIST_ENTRY(sem_undo) un_next; /* ptr to next active undo structure */
122 struct proc *un_proc; /* owner of this structure */
123 short un_cnt; /* # of active entries */
125 short un_adjval; /* adjust on exit values */
126 short un_num; /* semaphore # */
127 int un_id; /* semid */
128 unsigned short un_seq;
129 } un_ent[1]; /* undo entries */
133 * Configuration parameters
136 #define SEMMNI 10 /* # of semaphore identifiers */
139 #define SEMMNS 60 /* # of semaphores in system */
142 #define SEMUME 10 /* max # of undo entries per process */
145 #define SEMMNU 30 /* # of undo structures in system */
148 /* shouldn't need tuning */
150 #define SEMMAP 30 /* # of entries in semaphore map */
153 #define SEMMSL SEMMNS /* max # of semaphores per id */
156 #define SEMOPM 100 /* max # of operations per semop call */
159 #define SEMVMX 32767 /* semaphore maximum value */
160 #define SEMAEM 16384 /* adjust on exit max value */
163 * Due to the way semaphore memory is allocated, we have to ensure that
164 * SEMUSZ is properly aligned.
167 #define SEM_ALIGN(bytes) (((bytes) + (sizeof(long) - 1)) & ~(sizeof(long) - 1))
169 /* actual size of an undo structure */
170 #define SEMUSZ SEM_ALIGN(offsetof(struct sem_undo, un_ent[SEMUME]))
173 * Macro to find a particular sem_undo vector
176 ((struct sem_undo *)(((intptr_t)semu)+ix * seminfo.semusz))
179 * semaphore info struct
181 struct seminfo seminfo = {
182 SEMMAP, /* # of entries in semaphore map */
183 SEMMNI, /* # of semaphore identifiers */
184 SEMMNS, /* # of semaphores in system */
185 SEMMNU, /* # of undo structures in system */
186 SEMMSL, /* max # of semaphores per id */
187 SEMOPM, /* max # of operations per semop call */
188 SEMUME, /* max # of undo entries per process */
189 SEMUSZ, /* size in bytes of undo structure */
190 SEMVMX, /* semaphore maximum value */
191 SEMAEM /* adjust on exit max value */
194 SYSCTL_INT(_kern_ipc, OID_AUTO, semmap, CTLFLAG_RW, &seminfo.semmap, 0,
195 "Number of entries in the semaphore map");
196 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RDTUN, &seminfo.semmni, 0,
197 "Number of semaphore identifiers");
198 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RDTUN, &seminfo.semmns, 0,
199 "Maximum number of semaphores in the system");
200 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RDTUN, &seminfo.semmnu, 0,
201 "Maximum number of undo structures in the system");
202 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0,
203 "Max semaphores per id");
204 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RDTUN, &seminfo.semopm, 0,
205 "Max operations per semop call");
206 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RDTUN, &seminfo.semume, 0,
207 "Max undo entries per process");
208 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RDTUN, &seminfo.semusz, 0,
209 "Size in bytes of undo structure");
210 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RW, &seminfo.semvmx, 0,
211 "Semaphore maximum value");
212 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &seminfo.semaem, 0,
213 "Adjust on exit max value");
214 SYSCTL_PROC(_kern_ipc, OID_AUTO, sema, CTLFLAG_RD,
215 NULL, 0, sysctl_sema, "", "Semaphore id pool");
217 static struct syscall_helper_data sem_syscalls[] = {
218 SYSCALL_INIT_HELPER(__semctl),
219 SYSCALL_INIT_HELPER(semget),
220 SYSCALL_INIT_HELPER(semop),
221 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
222 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
223 SYSCALL_INIT_HELPER(semsys),
224 SYSCALL_INIT_HELPER(freebsd7___semctl),
229 #ifdef COMPAT_FREEBSD32
230 #include <compat/freebsd32/freebsd32.h>
231 #include <compat/freebsd32/freebsd32_ipc.h>
232 #include <compat/freebsd32/freebsd32_proto.h>
233 #include <compat/freebsd32/freebsd32_signal.h>
234 #include <compat/freebsd32/freebsd32_syscall.h>
235 #include <compat/freebsd32/freebsd32_util.h>
237 static struct syscall_helper_data sem32_syscalls[] = {
238 SYSCALL32_INIT_HELPER(freebsd32_semctl),
239 SYSCALL32_INIT_HELPER(semget),
240 SYSCALL32_INIT_HELPER(semop),
241 SYSCALL32_INIT_HELPER(freebsd32_semsys),
242 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
243 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
244 SYSCALL32_INIT_HELPER(freebsd7_freebsd32_semctl),
255 TUNABLE_INT_FETCH("kern.ipc.semmap", &seminfo.semmap);
256 TUNABLE_INT_FETCH("kern.ipc.semmni", &seminfo.semmni);
257 TUNABLE_INT_FETCH("kern.ipc.semmns", &seminfo.semmns);
258 TUNABLE_INT_FETCH("kern.ipc.semmnu", &seminfo.semmnu);
259 TUNABLE_INT_FETCH("kern.ipc.semmsl", &seminfo.semmsl);
260 TUNABLE_INT_FETCH("kern.ipc.semopm", &seminfo.semopm);
261 TUNABLE_INT_FETCH("kern.ipc.semume", &seminfo.semume);
262 TUNABLE_INT_FETCH("kern.ipc.semusz", &seminfo.semusz);
263 TUNABLE_INT_FETCH("kern.ipc.semvmx", &seminfo.semvmx);
264 TUNABLE_INT_FETCH("kern.ipc.semaem", &seminfo.semaem);
266 sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
267 sema = malloc(sizeof(struct semid_kernel) * seminfo.semmni, M_SEM,
269 sema_mtx = malloc(sizeof(struct mtx) * seminfo.semmni, M_SEM,
271 semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
273 for (i = 0; i < seminfo.semmni; i++) {
274 sema[i].u.sem_base = 0;
275 sema[i].u.sem_perm.mode = 0;
276 sema[i].u.sem_perm.seq = 0;
278 mac_sysvsem_init(&sema[i]);
281 for (i = 0; i < seminfo.semmni; i++)
282 mtx_init(&sema_mtx[i], "semid", NULL, MTX_DEF);
283 LIST_INIT(&semu_free_list);
284 for (i = 0; i < seminfo.semmnu; i++) {
285 struct sem_undo *suptr = SEMU(i);
286 suptr->un_proc = NULL;
287 LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
289 LIST_INIT(&semu_list);
290 mtx_init(&sem_mtx, "sem", NULL, MTX_DEF);
291 mtx_init(&sem_undo_mtx, "semu", NULL, MTX_DEF);
292 semexit_tag = EVENTHANDLER_REGISTER(process_exit, semexit_myhook, NULL,
293 EVENTHANDLER_PRI_ANY);
295 error = syscall_helper_register(sem_syscalls);
298 #ifdef COMPAT_FREEBSD32
299 error = syscall32_helper_register(sem32_syscalls);
315 #ifdef COMPAT_FREEBSD32
316 syscall32_helper_unregister(sem32_syscalls);
318 syscall_helper_unregister(sem_syscalls);
319 EVENTHANDLER_DEREGISTER(process_exit, semexit_tag);
321 for (i = 0; i < seminfo.semmni; i++)
322 mac_sysvsem_destroy(&sema[i]);
327 for (i = 0; i < seminfo.semmni; i++)
328 mtx_destroy(&sema_mtx[i]);
329 free(sema_mtx, M_SEM);
330 mtx_destroy(&sem_mtx);
331 mtx_destroy(&sem_undo_mtx);
336 sysvsem_modload(struct module *module, int cmd, void *arg)
358 static moduledata_t sysvsem_mod = {
364 DECLARE_MODULE(sysvsem, sysvsem_mod, SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
365 MODULE_VERSION(sysvsem, 1);
368 * Allocate a new sem_undo structure for a process
369 * (returns ptr to structure or NULL if no more room)
372 static struct sem_undo *
373 semu_alloc(struct thread *td)
375 struct sem_undo *suptr;
377 SEMUNDO_LOCKASSERT(MA_OWNED);
378 if ((suptr = LIST_FIRST(&semu_free_list)) == NULL)
380 LIST_REMOVE(suptr, un_next);
381 LIST_INSERT_HEAD(&semu_list, suptr, un_next);
383 suptr->un_proc = td->td_proc;
388 semu_try_free(struct sem_undo *suptr)
391 SEMUNDO_LOCKASSERT(MA_OWNED);
393 if (suptr->un_cnt != 0)
395 LIST_REMOVE(suptr, un_next);
396 LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
401 * Adjust a particular entry for a particular proc
405 semundo_adjust(struct thread *td, struct sem_undo **supptr, int semid,
406 int semseq, int semnum, int adjval)
408 struct proc *p = td->td_proc;
409 struct sem_undo *suptr;
413 SEMUNDO_LOCKASSERT(MA_OWNED);
414 /* Look for and remember the sem_undo if the caller doesn't provide
419 LIST_FOREACH(suptr, &semu_list, un_next) {
420 if (suptr->un_proc == p) {
428 suptr = semu_alloc(td);
436 * Look for the requested entry and adjust it (delete if adjval becomes
439 sunptr = &suptr->un_ent[0];
440 for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
441 if (sunptr->un_id != semid || sunptr->un_num != semnum)
444 adjval += sunptr->un_adjval;
445 if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
448 sunptr->un_adjval = adjval;
449 if (sunptr->un_adjval == 0) {
451 if (i < suptr->un_cnt)
453 suptr->un_ent[suptr->un_cnt];
454 if (suptr->un_cnt == 0)
455 semu_try_free(suptr);
460 /* Didn't find the right entry - create it */
463 if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
465 if (suptr->un_cnt != seminfo.semume) {
466 sunptr = &suptr->un_ent[suptr->un_cnt];
468 sunptr->un_adjval = adjval;
469 sunptr->un_id = semid;
470 sunptr->un_num = semnum;
471 sunptr->un_seq = semseq;
478 semundo_clear(int semid, int semnum)
480 struct sem_undo *suptr, *suptr1;
484 SEMUNDO_LOCKASSERT(MA_OWNED);
485 LIST_FOREACH_SAFE(suptr, &semu_list, un_next, suptr1) {
486 sunptr = &suptr->un_ent[0];
487 for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
488 if (sunptr->un_id != semid)
490 if (semnum == -1 || sunptr->un_num == semnum) {
492 if (i < suptr->un_cnt) {
494 suptr->un_ent[suptr->un_cnt];
497 semu_try_free(suptr);
506 semvalid(int semid, struct semid_kernel *semakptr)
509 return ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
510 semakptr->u.sem_perm.seq != IPCID_TO_SEQ(semid) ? EINVAL : 0);
514 * Note that the user-mode half of this passes a union, not a pointer.
516 #ifndef _SYS_SYSPROTO_H_
517 struct __semctl_args {
525 __semctl(struct thread *td, struct __semctl_args *uap)
527 struct semid_ds dsbuf;
528 union semun arg, semun;
539 error = copyin(uap->arg, &arg, sizeof(arg));
551 error = copyin(arg.buf, &dsbuf, sizeof(dsbuf));
558 semun.array = arg.array;
565 error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
573 error = copyout(&dsbuf, arg.buf, sizeof(dsbuf));
578 td->td_retval[0] = rval;
583 kern_semctl(struct thread *td, int semid, int semnum, int cmd,
584 union semun *arg, register_t *rval)
587 struct ucred *cred = td->td_ucred;
589 struct semid_ds *sbuf;
590 struct semid_kernel *semakptr;
591 struct mtx *sema_mtxp;
592 u_short usval, count;
595 DPRINTF(("call to semctl(%d, %d, %d, 0x%p)\n",
596 semid, semnum, cmd, arg));
597 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
605 * For this command we assume semid is an array index
606 * rather than an IPC id.
608 if (semid < 0 || semid >= seminfo.semmni)
610 semakptr = &sema[semid];
611 sema_mtxp = &sema_mtx[semid];
613 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
617 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
620 error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
624 bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
625 *rval = IXSEQ_TO_IPCID(semid, semakptr->u.sem_perm);
626 mtx_unlock(sema_mtxp);
630 semidx = IPCID_TO_IX(semid);
631 if (semidx < 0 || semidx >= seminfo.semmni)
634 semakptr = &sema[semidx];
635 sema_mtxp = &sema_mtx[semidx];
640 error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
650 if ((error = semvalid(semid, semakptr)) != 0)
652 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
654 semakptr->u.sem_perm.cuid = cred->cr_uid;
655 semakptr->u.sem_perm.uid = cred->cr_uid;
656 semakptr->u.sem_perm.mode = 0;
658 semundo_clear(semidx, -1);
661 mac_sysvsem_cleanup(semakptr);
664 for (i = 0; i < seminfo.semmni; i++) {
665 if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
666 sema[i].u.sem_base > semakptr->u.sem_base)
667 mtx_lock_flags(&sema_mtx[i], LOP_DUPOK);
669 for (i = semakptr->u.sem_base - sem; i < semtot; i++)
670 sem[i] = sem[i + semakptr->u.sem_nsems];
671 for (i = 0; i < seminfo.semmni; i++) {
672 if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
673 sema[i].u.sem_base > semakptr->u.sem_base) {
674 sema[i].u.sem_base -= semakptr->u.sem_nsems;
675 mtx_unlock(&sema_mtx[i]);
678 semtot -= semakptr->u.sem_nsems;
682 if ((error = semvalid(semid, semakptr)) != 0)
684 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
687 semakptr->u.sem_perm.uid = sbuf->sem_perm.uid;
688 semakptr->u.sem_perm.gid = sbuf->sem_perm.gid;
689 semakptr->u.sem_perm.mode = (semakptr->u.sem_perm.mode &
690 ~0777) | (sbuf->sem_perm.mode & 0777);
691 semakptr->u.sem_ctime = time_second;
695 if ((error = semvalid(semid, semakptr)) != 0)
697 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
699 bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
703 if ((error = semvalid(semid, semakptr)) != 0)
705 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
707 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
711 *rval = semakptr->u.sem_base[semnum].semncnt;
715 if ((error = semvalid(semid, semakptr)) != 0)
717 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
719 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
723 *rval = semakptr->u.sem_base[semnum].sempid;
727 if ((error = semvalid(semid, semakptr)) != 0)
729 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
731 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
735 *rval = semakptr->u.sem_base[semnum].semval;
740 * Unfortunately, callers of this function don't know
741 * in advance how many semaphores are in this set.
742 * While we could just allocate the maximum size array
743 * and pass the actual size back to the caller, that
744 * won't work for SETALL since we can't copyin() more
745 * data than the user specified as we may return a
748 * Note that the number of semaphores in a set is
749 * fixed for the life of that set. The only way that
750 * the 'count' could change while are blocked in
751 * malloc() is if this semaphore set were destroyed
752 * and a new one created with the same index.
753 * However, semvalid() will catch that due to the
754 * sequence number unless exactly 0x8000 (or a
755 * multiple thereof) semaphore sets for the same index
756 * are created and destroyed while we are in malloc!
759 count = semakptr->u.sem_nsems;
760 mtx_unlock(sema_mtxp);
761 array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
763 if ((error = semvalid(semid, semakptr)) != 0)
765 KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
766 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
768 for (i = 0; i < semakptr->u.sem_nsems; i++)
769 array[i] = semakptr->u.sem_base[i].semval;
770 mtx_unlock(sema_mtxp);
771 error = copyout(array, arg->array, count * sizeof(*array));
776 if ((error = semvalid(semid, semakptr)) != 0)
778 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
780 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
784 *rval = semakptr->u.sem_base[semnum].semzcnt;
788 if ((error = semvalid(semid, semakptr)) != 0)
790 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
792 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
796 if (arg->val < 0 || arg->val > seminfo.semvmx) {
800 semakptr->u.sem_base[semnum].semval = arg->val;
802 semundo_clear(semidx, semnum);
809 * See comment on GETALL for why 'count' shouldn't change
810 * and why we require a userland buffer.
812 count = semakptr->u.sem_nsems;
813 mtx_unlock(sema_mtxp);
814 array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
815 error = copyin(arg->array, array, count * sizeof(*array));
819 if ((error = semvalid(semid, semakptr)) != 0)
821 KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
822 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
824 for (i = 0; i < semakptr->u.sem_nsems; i++) {
826 if (usval > seminfo.semvmx) {
830 semakptr->u.sem_base[i].semval = usval;
833 semundo_clear(semidx, -1);
844 mtx_unlock(sema_mtxp);
846 mtx_unlock(&sem_mtx);
852 #ifndef _SYS_SYSPROTO_H_
860 semget(struct thread *td, struct semget_args *uap)
862 int semid, error = 0;
864 int nsems = uap->nsems;
865 int semflg = uap->semflg;
866 struct ucred *cred = td->td_ucred;
868 DPRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg));
869 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
873 if (key != IPC_PRIVATE) {
874 for (semid = 0; semid < seminfo.semmni; semid++) {
875 if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) &&
876 sema[semid].u.sem_perm.key == key)
879 if (semid < seminfo.semmni) {
880 DPRINTF(("found public key\n"));
881 if ((error = ipcperm(td, &sema[semid].u.sem_perm,
885 if (nsems > 0 && sema[semid].u.sem_nsems < nsems) {
886 DPRINTF(("too small\n"));
890 if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
891 DPRINTF(("not exclusive\n"));
896 error = mac_sysvsem_check_semget(cred, &sema[semid]);
904 DPRINTF(("need to allocate the semid_kernel\n"));
905 if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
906 if (nsems <= 0 || nsems > seminfo.semmsl) {
907 DPRINTF(("nsems out of range (0<%d<=%d)\n", nsems,
912 if (nsems > seminfo.semmns - semtot) {
914 "not enough semaphores left (need %d, got %d)\n",
915 nsems, seminfo.semmns - semtot));
919 for (semid = 0; semid < seminfo.semmni; semid++) {
920 if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0)
923 if (semid == seminfo.semmni) {
924 DPRINTF(("no more semid_kernel's available\n"));
928 DPRINTF(("semid %d is available\n", semid));
929 mtx_lock(&sema_mtx[semid]);
930 KASSERT((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0,
931 ("Lost semaphore %d", semid));
932 sema[semid].u.sem_perm.key = key;
933 sema[semid].u.sem_perm.cuid = cred->cr_uid;
934 sema[semid].u.sem_perm.uid = cred->cr_uid;
935 sema[semid].u.sem_perm.cgid = cred->cr_gid;
936 sema[semid].u.sem_perm.gid = cred->cr_gid;
937 sema[semid].u.sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
938 sema[semid].u.sem_perm.seq =
939 (sema[semid].u.sem_perm.seq + 1) & 0x7fff;
940 sema[semid].u.sem_nsems = nsems;
941 sema[semid].u.sem_otime = 0;
942 sema[semid].u.sem_ctime = time_second;
943 sema[semid].u.sem_base = &sem[semtot];
945 bzero(sema[semid].u.sem_base,
946 sizeof(sema[semid].u.sem_base[0])*nsems);
948 mac_sysvsem_create(cred, &sema[semid]);
950 mtx_unlock(&sema_mtx[semid]);
951 DPRINTF(("sembase = %p, next = %p\n",
952 sema[semid].u.sem_base, &sem[semtot]));
954 DPRINTF(("didn't find it and wasn't asked to create it\n"));
960 td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].u.sem_perm);
962 mtx_unlock(&sem_mtx);
966 #ifndef _SYS_SYSPROTO_H_
974 semop(struct thread *td, struct semop_args *uap)
977 struct sembuf small_sops[SMALL_SOPS];
978 int semid = uap->semid;
979 size_t nsops = uap->nsops;
981 struct semid_kernel *semakptr;
982 struct sembuf *sopptr = 0;
983 struct sem *semptr = 0;
984 struct sem_undo *suptr;
985 struct mtx *sema_mtxp;
988 int do_wakeup, do_undos;
994 DPRINTF(("call to semop(%d, %p, %u)\n", semid, sops, nsops));
996 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
999 semid = IPCID_TO_IX(semid); /* Convert back to zero origin */
1001 if (semid < 0 || semid >= seminfo.semmni)
1004 /* Allocate memory for sem_ops */
1005 if (nsops <= SMALL_SOPS)
1007 else if (nsops <= seminfo.semopm)
1008 sops = malloc(nsops * sizeof(*sops), M_TEMP, M_WAITOK);
1010 DPRINTF(("too many sops (max=%d, nsops=%d)\n", seminfo.semopm,
1014 if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) {
1015 DPRINTF(("error = %d from copyin(%p, %p, %d)\n", error,
1016 uap->sops, sops, nsops * sizeof(sops[0])));
1017 if (sops != small_sops)
1022 semakptr = &sema[semid];
1023 sema_mtxp = &sema_mtx[semid];
1024 mtx_lock(sema_mtxp);
1025 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
1029 seq = semakptr->u.sem_perm.seq;
1030 if (seq != IPCID_TO_SEQ(uap->semid)) {
1035 * Initial pass thru sops to see what permissions are needed.
1036 * Also perform any checks that don't need repeating on each
1037 * attempt to satisfy the request vector.
1039 j = 0; /* permission needed */
1041 for (i = 0; i < nsops; i++) {
1043 if (sopptr->sem_num >= semakptr->u.sem_nsems) {
1047 if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0)
1049 j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A;
1052 if ((error = ipcperm(td, &semakptr->u.sem_perm, j))) {
1053 DPRINTF(("error = %d from ipaccess\n", error));
1057 error = mac_sysvsem_check_semop(td->td_ucred, semakptr, j);
1063 * Loop trying to satisfy the vector of requests.
1064 * If we reach a point where we must wait, any requests already
1065 * performed are rolled back and we go to sleep until some other
1066 * process wakes us up. At this point, we start all over again.
1068 * This ensures that from the perspective of other tasks, a set
1069 * of requests is atomic (never partially satisfied).
1073 error = 0; /* error return if necessary */
1075 for (i = 0; i < nsops; i++) {
1077 semptr = &semakptr->u.sem_base[sopptr->sem_num];
1080 "semop: semakptr=%p, sem_base=%p, "
1081 "semptr=%p, sem[%d]=%d : op=%d, flag=%s\n",
1082 semakptr, semakptr->u.sem_base, semptr,
1083 sopptr->sem_num, semptr->semval, sopptr->sem_op,
1084 (sopptr->sem_flg & IPC_NOWAIT) ?
1085 "nowait" : "wait"));
1087 if (sopptr->sem_op < 0) {
1088 if (semptr->semval + sopptr->sem_op < 0) {
1089 DPRINTF(("semop: can't do it now\n"));
1092 semptr->semval += sopptr->sem_op;
1093 if (semptr->semval == 0 &&
1094 semptr->semzcnt > 0)
1097 } else if (sopptr->sem_op == 0) {
1098 if (semptr->semval != 0) {
1099 DPRINTF(("semop: not zero now\n"));
1102 } else if (semptr->semval + sopptr->sem_op >
1107 if (semptr->semncnt > 0)
1109 semptr->semval += sopptr->sem_op;
1114 * Did we get through the entire vector?
1120 * No ... rollback anything that we've already done
1122 DPRINTF(("semop: rollback 0 through %d\n", i-1));
1123 for (j = 0; j < i; j++)
1124 semakptr->u.sem_base[sops[j].sem_num].semval -=
1127 /* If we detected an error, return it */
1132 * If the request that we couldn't satisfy has the
1133 * NOWAIT flag set then return with EAGAIN.
1135 if (sopptr->sem_flg & IPC_NOWAIT) {
1140 if (sopptr->sem_op == 0)
1145 DPRINTF(("semop: good night!\n"));
1146 error = msleep(semakptr, sema_mtxp, (PZERO - 4) | PCATCH,
1148 DPRINTF(("semop: good morning (error=%d)!\n", error));
1149 /* return code is checked below, after sem[nz]cnt-- */
1152 * Make sure that the semaphore still exists
1154 seq = semakptr->u.sem_perm.seq;
1155 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1156 seq != IPCID_TO_SEQ(uap->semid)) {
1162 * Renew the semaphore's pointer after wakeup since
1163 * during msleep sem_base may have been modified and semptr
1164 * is not valid any more
1166 semptr = &semakptr->u.sem_base[sopptr->sem_num];
1169 * The semaphore is still alive. Readjust the count of
1170 * waiting processes.
1172 if (sopptr->sem_op == 0)
1178 * Is it really morning, or was our sleep interrupted?
1179 * (Delayed check of msleep() return code because we
1180 * need to decrement sem[nz]cnt either way.)
1186 DPRINTF(("semop: good morning!\n"));
1191 * Process any SEM_UNDO requests.
1196 for (i = 0; i < nsops; i++) {
1198 * We only need to deal with SEM_UNDO's for non-zero
1203 if ((sops[i].sem_flg & SEM_UNDO) == 0)
1205 adjval = sops[i].sem_op;
1208 error = semundo_adjust(td, &suptr, semid, seq,
1209 sops[i].sem_num, -adjval);
1214 * Oh-Oh! We ran out of either sem_undo's or undo's.
1215 * Rollback the adjustments to this point and then
1216 * rollback the semaphore ups and down so we can return
1217 * with an error with all structures restored. We
1218 * rollback the undo's in the exact reverse order that
1219 * we applied them. This guarantees that we won't run
1220 * out of space as we roll things back out.
1222 for (j = 0; j < i; j++) {
1224 if ((sops[k].sem_flg & SEM_UNDO) == 0)
1226 adjval = sops[k].sem_op;
1229 if (semundo_adjust(td, &suptr, semid, seq,
1230 sops[k].sem_num, adjval) != 0)
1231 panic("semop - can't undo undos");
1234 for (j = 0; j < nsops; j++)
1235 semakptr->u.sem_base[sops[j].sem_num].semval -=
1238 DPRINTF(("error = %d from semundo_adjust\n", error));
1241 } /* loop through the sops */
1243 } /* if (do_undos) */
1245 /* We're definitely done - set the sempid's and time */
1246 for (i = 0; i < nsops; i++) {
1248 semptr = &semakptr->u.sem_base[sopptr->sem_num];
1249 semptr->sempid = td->td_proc->p_pid;
1251 semakptr->u.sem_otime = time_second;
1254 * Do a wakeup if any semaphore was up'd whilst something was
1258 DPRINTF(("semop: doing wakeup\n"));
1260 DPRINTF(("semop: back from wakeup\n"));
1262 DPRINTF(("semop: done\n"));
1263 td->td_retval[0] = 0;
1265 mtx_unlock(sema_mtxp);
1266 if (sops != small_sops)
1272 * Go through the undo structures for this process and apply the adjustments to
1276 semexit_myhook(void *arg, struct proc *p)
1278 struct sem_undo *suptr;
1279 struct semid_kernel *semakptr;
1280 struct mtx *sema_mtxp;
1281 int semid, semnum, adjval, ix;
1285 * Go through the chain of undo vectors looking for one
1286 * associated with this process.
1289 LIST_FOREACH(suptr, &semu_list, un_next) {
1290 if (suptr->un_proc == p)
1293 if (suptr == NULL) {
1297 LIST_REMOVE(suptr, un_next);
1299 DPRINTF(("proc @%p has undo structure with %d entries\n", p,
1303 * If there are any active undo elements then process them.
1305 if (suptr->un_cnt > 0) {
1307 for (ix = 0; ix < suptr->un_cnt; ix++) {
1308 semid = suptr->un_ent[ix].un_id;
1309 semnum = suptr->un_ent[ix].un_num;
1310 adjval = suptr->un_ent[ix].un_adjval;
1311 seq = suptr->un_ent[ix].un_seq;
1312 semakptr = &sema[semid];
1313 sema_mtxp = &sema_mtx[semid];
1315 mtx_lock(sema_mtxp);
1316 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1317 (semakptr->u.sem_perm.seq != seq)) {
1318 mtx_unlock(sema_mtxp);
1321 if (semnum >= semakptr->u.sem_nsems)
1322 panic("semexit - semnum out of range");
1325 "semexit: %p id=%d num=%d(adj=%d) ; sem=%d\n",
1326 suptr->un_proc, suptr->un_ent[ix].un_id,
1327 suptr->un_ent[ix].un_num,
1328 suptr->un_ent[ix].un_adjval,
1329 semakptr->u.sem_base[semnum].semval));
1331 if (adjval < 0 && semakptr->u.sem_base[semnum].semval <
1333 semakptr->u.sem_base[semnum].semval = 0;
1335 semakptr->u.sem_base[semnum].semval += adjval;
1338 DPRINTF(("semexit: back from wakeup\n"));
1339 mtx_unlock(sema_mtxp);
1345 * Deallocate the undo vector.
1347 DPRINTF(("removing vector\n"));
1348 suptr->un_proc = NULL;
1350 LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
1355 sysctl_sema(SYSCTL_HANDLER_ARGS)
1358 return (SYSCTL_OUT(req, sema,
1359 sizeof(struct semid_kernel) * seminfo.semmni));
1362 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1363 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1365 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1366 static sy_call_t *semcalls[] = {
1367 (sy_call_t *)freebsd7___semctl, (sy_call_t *)semget,
1372 * Entry point for all SEM calls.
1377 /* XXX actually varargs. */
1378 struct semsys_args /* {
1388 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
1390 if (uap->which < 0 ||
1391 uap->which >= sizeof(semcalls)/sizeof(semcalls[0]))
1393 error = (*semcalls[uap->which])(td, &uap->a2);
1398 #define CP(src, dst, fld) do { (dst).fld = (src).fld; } while (0)
1401 #ifndef _SYS_SYSPROTO_H_
1402 struct freebsd7___semctl_args {
1406 union semun_old *arg;
1410 freebsd7___semctl(struct thread *td, struct freebsd7___semctl_args *uap)
1412 struct semid_ds_old dsold;
1413 struct semid_ds dsbuf;
1414 union semun_old arg;
1426 error = copyin(uap->arg, &arg, sizeof(arg));
1438 error = copyin(arg.buf, &dsold, sizeof(dsold));
1441 ipcperm_old2new(&dsold.sem_perm, &dsbuf.sem_perm);
1442 CP(dsold, dsbuf, sem_base);
1443 CP(dsold, dsbuf, sem_nsems);
1444 CP(dsold, dsbuf, sem_otime);
1445 CP(dsold, dsbuf, sem_ctime);
1450 semun.array = arg.array;
1453 semun.val = arg.val;
1457 error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1465 bzero(&dsold, sizeof(dsold));
1466 ipcperm_new2old(&dsbuf.sem_perm, &dsold.sem_perm);
1467 CP(dsbuf, dsold, sem_base);
1468 CP(dsbuf, dsold, sem_nsems);
1469 CP(dsbuf, dsold, sem_otime);
1470 CP(dsbuf, dsold, sem_ctime);
1471 error = copyout(&dsold, arg.buf, sizeof(dsold));
1476 td->td_retval[0] = rval;
1480 #endif /* COMPAT_FREEBSD{4,5,6,7} */
1482 #ifdef COMPAT_FREEBSD32
1485 freebsd32_semsys(struct thread *td, struct freebsd32_semsys_args *uap)
1488 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1489 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1490 switch (uap->which) {
1492 return (freebsd7_freebsd32_semctl(td,
1493 (struct freebsd7_freebsd32_semctl_args *)&uap->a2));
1495 return (semsys(td, (struct semsys_args *)uap));
1498 return (nosys(td, NULL));
1502 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1503 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1505 freebsd7_freebsd32_semctl(struct thread *td,
1506 struct freebsd7_freebsd32_semctl_args *uap)
1508 struct semid_ds32_old dsbuf32;
1509 struct semid_ds dsbuf;
1522 error = copyin(uap->arg, &arg, sizeof(arg));
1534 error = copyin(PTRIN(arg.buf), &dsbuf32, sizeof(dsbuf32));
1537 freebsd32_ipcperm_old_in(&dsbuf32.sem_perm, &dsbuf.sem_perm);
1538 PTRIN_CP(dsbuf32, dsbuf, sem_base);
1539 CP(dsbuf32, dsbuf, sem_nsems);
1540 CP(dsbuf32, dsbuf, sem_otime);
1541 CP(dsbuf32, dsbuf, sem_ctime);
1546 semun.array = PTRIN(arg.array);
1549 semun.val = arg.val;
1553 error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1561 bzero(&dsbuf32, sizeof(dsbuf32));
1562 freebsd32_ipcperm_old_out(&dsbuf.sem_perm, &dsbuf32.sem_perm);
1563 PTROUT_CP(dsbuf, dsbuf32, sem_base);
1564 CP(dsbuf, dsbuf32, sem_nsems);
1565 CP(dsbuf, dsbuf32, sem_otime);
1566 CP(dsbuf, dsbuf32, sem_ctime);
1567 error = copyout(&dsbuf32, PTRIN(arg.buf), sizeof(dsbuf32));
1572 td->td_retval[0] = rval;
1578 freebsd32_semctl(struct thread *td, struct freebsd32_semctl_args *uap)
1580 struct semid_ds32 dsbuf32;
1581 struct semid_ds dsbuf;
1594 error = copyin(uap->arg, &arg, sizeof(arg));
1606 error = copyin(PTRIN(arg.buf), &dsbuf32, sizeof(dsbuf32));
1609 freebsd32_ipcperm_in(&dsbuf32.sem_perm, &dsbuf.sem_perm);
1610 PTRIN_CP(dsbuf32, dsbuf, sem_base);
1611 CP(dsbuf32, dsbuf, sem_nsems);
1612 CP(dsbuf32, dsbuf, sem_otime);
1613 CP(dsbuf32, dsbuf, sem_ctime);
1618 semun.array = PTRIN(arg.array);
1621 semun.val = arg.val;
1625 error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1633 bzero(&dsbuf32, sizeof(dsbuf32));
1634 freebsd32_ipcperm_out(&dsbuf.sem_perm, &dsbuf32.sem_perm);
1635 PTROUT_CP(dsbuf, dsbuf32, sem_base);
1636 CP(dsbuf, dsbuf32, sem_nsems);
1637 CP(dsbuf, dsbuf32, sem_otime);
1638 CP(dsbuf, dsbuf32, sem_ctime);
1639 error = copyout(&dsbuf32, PTRIN(arg.buf), sizeof(dsbuf32));
1644 td->td_retval[0] = rval;
1648 #endif /* COMPAT_FREEBSD32 */