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_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 void 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 /* XXX casting to (sy_call_t *) is bogus, as usual. */
95 static sy_call_t *semcalls[] = {
96 (sy_call_t *)__semctl, (sy_call_t *)semget,
100 static struct mtx sem_mtx; /* semaphore global lock */
101 static struct mtx sem_undo_mtx;
102 static int semtot = 0;
103 static struct semid_kernel *sema; /* semaphore id pool */
104 static struct mtx *sema_mtx; /* semaphore id pool mutexes*/
105 static struct sem *sem; /* semaphore pool */
106 LIST_HEAD(, sem_undo) semu_list; /* list of active undo structures */
107 LIST_HEAD(, sem_undo) semu_free_list; /* list of free undo structures */
108 static int *semu; /* undo structure pool */
109 static eventhandler_tag semexit_tag;
111 #define SEMUNDO_MTX sem_undo_mtx
112 #define SEMUNDO_LOCK() mtx_lock(&SEMUNDO_MTX);
113 #define SEMUNDO_UNLOCK() mtx_unlock(&SEMUNDO_MTX);
114 #define SEMUNDO_LOCKASSERT(how) mtx_assert(&SEMUNDO_MTX, (how));
117 u_short semval; /* semaphore value */
118 pid_t sempid; /* pid of last operation */
119 u_short semncnt; /* # awaiting semval > cval */
120 u_short semzcnt; /* # awaiting semval = 0 */
124 * Undo structure (one per process)
127 LIST_ENTRY(sem_undo) un_next; /* ptr to next active undo structure */
128 struct proc *un_proc; /* owner of this structure */
129 short un_cnt; /* # of active entries */
131 short un_adjval; /* adjust on exit values */
132 short un_num; /* semaphore # */
133 int un_id; /* semid */
134 unsigned short un_seq;
135 } un_ent[1]; /* undo entries */
139 * Configuration parameters
142 #define SEMMNI 10 /* # of semaphore identifiers */
145 #define SEMMNS 60 /* # of semaphores in system */
148 #define SEMUME 10 /* max # of undo entries per process */
151 #define SEMMNU 30 /* # of undo structures in system */
154 /* shouldn't need tuning */
156 #define SEMMAP 30 /* # of entries in semaphore map */
159 #define SEMMSL SEMMNS /* max # of semaphores per id */
162 #define SEMOPM 100 /* max # of operations per semop call */
165 #define SEMVMX 32767 /* semaphore maximum value */
166 #define SEMAEM 16384 /* adjust on exit max value */
169 * Due to the way semaphore memory is allocated, we have to ensure that
170 * SEMUSZ is properly aligned.
173 #define SEM_ALIGN(bytes) (((bytes) + (sizeof(long) - 1)) & ~(sizeof(long) - 1))
175 /* actual size of an undo structure */
176 #define SEMUSZ SEM_ALIGN(offsetof(struct sem_undo, un_ent[SEMUME]))
179 * Macro to find a particular sem_undo vector
182 ((struct sem_undo *)(((intptr_t)semu)+ix * seminfo.semusz))
185 * semaphore info struct
187 struct seminfo seminfo = {
188 SEMMAP, /* # of entries in semaphore map */
189 SEMMNI, /* # of semaphore identifiers */
190 SEMMNS, /* # of semaphores in system */
191 SEMMNU, /* # of undo structures in system */
192 SEMMSL, /* max # of semaphores per id */
193 SEMOPM, /* max # of operations per semop call */
194 SEMUME, /* max # of undo entries per process */
195 SEMUSZ, /* size in bytes of undo structure */
196 SEMVMX, /* semaphore maximum value */
197 SEMAEM /* adjust on exit max value */
200 SYSCTL_INT(_kern_ipc, OID_AUTO, semmap, CTLFLAG_RW, &seminfo.semmap, 0,
201 "Number of entries in the semaphore map");
202 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RDTUN, &seminfo.semmni, 0,
203 "Number of semaphore identifiers");
204 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RDTUN, &seminfo.semmns, 0,
205 "Maximum number of semaphores in the system");
206 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RDTUN, &seminfo.semmnu, 0,
207 "Maximum number of undo structures in the system");
208 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0,
209 "Max semaphores per id");
210 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RDTUN, &seminfo.semopm, 0,
211 "Max operations per semop call");
212 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RDTUN, &seminfo.semume, 0,
213 "Max undo entries per process");
214 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RDTUN, &seminfo.semusz, 0,
215 "Size in bytes of undo structure");
216 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RW, &seminfo.semvmx, 0,
217 "Semaphore maximum value");
218 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &seminfo.semaem, 0,
219 "Adjust on exit max value");
220 SYSCTL_PROC(_kern_ipc, OID_AUTO, sema, CTLFLAG_RD,
221 NULL, 0, sysctl_sema, "", "");
228 TUNABLE_INT_FETCH("kern.ipc.semmap", &seminfo.semmap);
229 TUNABLE_INT_FETCH("kern.ipc.semmni", &seminfo.semmni);
230 TUNABLE_INT_FETCH("kern.ipc.semmns", &seminfo.semmns);
231 TUNABLE_INT_FETCH("kern.ipc.semmnu", &seminfo.semmnu);
232 TUNABLE_INT_FETCH("kern.ipc.semmsl", &seminfo.semmsl);
233 TUNABLE_INT_FETCH("kern.ipc.semopm", &seminfo.semopm);
234 TUNABLE_INT_FETCH("kern.ipc.semume", &seminfo.semume);
235 TUNABLE_INT_FETCH("kern.ipc.semusz", &seminfo.semusz);
236 TUNABLE_INT_FETCH("kern.ipc.semvmx", &seminfo.semvmx);
237 TUNABLE_INT_FETCH("kern.ipc.semaem", &seminfo.semaem);
239 sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
240 sema = malloc(sizeof(struct semid_kernel) * seminfo.semmni, M_SEM,
242 sema_mtx = malloc(sizeof(struct mtx) * seminfo.semmni, M_SEM,
244 semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
246 for (i = 0; i < seminfo.semmni; i++) {
247 sema[i].u.sem_base = 0;
248 sema[i].u.sem_perm.mode = 0;
249 sema[i].u.sem_perm.seq = 0;
251 mac_init_sysv_sem(&sema[i]);
254 for (i = 0; i < seminfo.semmni; i++)
255 mtx_init(&sema_mtx[i], "semid", NULL, MTX_DEF);
256 LIST_INIT(&semu_free_list);
257 for (i = 0; i < seminfo.semmnu; i++) {
258 struct sem_undo *suptr = SEMU(i);
259 suptr->un_proc = NULL;
260 LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
262 LIST_INIT(&semu_list);
263 mtx_init(&sem_mtx, "sem", NULL, MTX_DEF);
264 mtx_init(&sem_undo_mtx, "semu", NULL, MTX_DEF);
265 semexit_tag = EVENTHANDLER_REGISTER(process_exit, semexit_myhook, NULL,
266 EVENTHANDLER_PRI_ANY);
278 EVENTHANDLER_DEREGISTER(process_exit, semexit_tag);
280 for (i = 0; i < seminfo.semmni; i++)
281 mac_destroy_sysv_sem(&sema[i]);
286 for (i = 0; i < seminfo.semmni; i++)
287 mtx_destroy(&sema_mtx[i]);
288 free(sema_mtx, M_SEM);
289 mtx_destroy(&sem_mtx);
290 mtx_destroy(&sem_undo_mtx);
295 sysvsem_modload(struct module *module, int cmd, void *arg)
315 static moduledata_t sysvsem_mod = {
321 SYSCALL_MODULE_HELPER(semsys);
322 SYSCALL_MODULE_HELPER(__semctl);
323 SYSCALL_MODULE_HELPER(semget);
324 SYSCALL_MODULE_HELPER(semop);
326 DECLARE_MODULE(sysvsem, sysvsem_mod,
327 SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
328 MODULE_VERSION(sysvsem, 1);
331 * Entry point for all SEM calls.
336 /* XXX actually varargs. */
337 struct semsys_args /* {
347 if (!jail_sysvipc_allowed && jailed(td->td_ucred))
349 if (uap->which < 0 ||
350 uap->which >= sizeof(semcalls)/sizeof(semcalls[0]))
352 error = (*semcalls[uap->which])(td, &uap->a2);
357 * Allocate a new sem_undo structure for a process
358 * (returns ptr to structure or NULL if no more room)
361 static struct sem_undo *
362 semu_alloc(struct thread *td)
364 struct sem_undo *suptr;
366 SEMUNDO_LOCKASSERT(MA_OWNED);
367 if ((suptr = LIST_FIRST(&semu_free_list)) == NULL)
369 LIST_REMOVE(suptr, un_next);
370 LIST_INSERT_HEAD(&semu_list, suptr, un_next);
372 suptr->un_proc = td->td_proc;
377 semu_try_free(struct sem_undo *suptr)
380 SEMUNDO_LOCKASSERT(MA_OWNED);
382 if (suptr->un_cnt != 0)
384 LIST_REMOVE(suptr, un_next);
385 LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
390 * Adjust a particular entry for a particular proc
394 semundo_adjust(struct thread *td, struct sem_undo **supptr, int semid,
395 int semseq, int semnum, int adjval)
397 struct proc *p = td->td_proc;
398 struct sem_undo *suptr;
402 SEMUNDO_LOCKASSERT(MA_OWNED);
403 /* Look for and remember the sem_undo if the caller doesn't provide
408 LIST_FOREACH(suptr, &semu_list, un_next) {
409 if (suptr->un_proc == p) {
417 suptr = semu_alloc(td);
425 * Look for the requested entry and adjust it (delete if adjval becomes
428 sunptr = &suptr->un_ent[0];
429 for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
430 if (sunptr->un_id != semid || sunptr->un_num != semnum)
433 adjval += sunptr->un_adjval;
434 if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
437 sunptr->un_adjval = adjval;
438 if (sunptr->un_adjval == 0) {
440 if (i < suptr->un_cnt)
442 suptr->un_ent[suptr->un_cnt];
443 if (suptr->un_cnt == 0)
444 semu_try_free(suptr);
449 /* Didn't find the right entry - create it */
452 if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
454 if (suptr->un_cnt != seminfo.semume) {
455 sunptr = &suptr->un_ent[suptr->un_cnt];
457 sunptr->un_adjval = adjval;
458 sunptr->un_id = semid;
459 sunptr->un_num = semnum;
460 sunptr->un_seq = semseq;
467 semundo_clear(int semid, int semnum)
469 struct sem_undo *suptr, *suptr1;
473 SEMUNDO_LOCKASSERT(MA_OWNED);
474 LIST_FOREACH_SAFE(suptr, &semu_list, un_next, suptr1) {
475 sunptr = &suptr->un_ent[0];
476 for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
477 if (sunptr->un_id != semid)
479 if (semnum == -1 || sunptr->un_num == semnum) {
481 if (i < suptr->un_cnt) {
483 suptr->un_ent[suptr->un_cnt];
486 semu_try_free(suptr);
495 semvalid(int semid, struct semid_kernel *semakptr)
498 return ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
499 semakptr->u.sem_perm.seq != IPCID_TO_SEQ(semid) ? EINVAL : 0);
503 * Note that the user-mode half of this passes a union, not a pointer.
505 #ifndef _SYS_SYSPROTO_H_
506 struct __semctl_args {
514 __semctl(struct thread *td, struct __semctl_args *uap)
516 struct semid_ds dsbuf;
517 union semun arg, semun;
528 error = copyin(uap->arg, &arg, sizeof(arg));
540 error = copyin(arg.buf, &dsbuf, sizeof(dsbuf));
547 semun.array = arg.array;
554 error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
562 error = copyout(&dsbuf, arg.buf, sizeof(dsbuf));
567 td->td_retval[0] = rval;
572 kern_semctl(struct thread *td, int semid, int semnum, int cmd,
573 union semun *arg, register_t *rval)
576 struct ucred *cred = td->td_ucred;
578 struct semid_ds *sbuf;
579 struct semid_kernel *semakptr;
580 struct mtx *sema_mtxp;
581 u_short usval, count;
584 DPRINTF(("call to semctl(%d, %d, %d, 0x%p)\n",
585 semid, semnum, cmd, arg));
586 if (!jail_sysvipc_allowed && jailed(td->td_ucred))
594 * For this command we assume semid is an array index
595 * rather than an IPC id.
597 if (semid < 0 || semid >= seminfo.semmni)
599 semakptr = &sema[semid];
600 sema_mtxp = &sema_mtx[semid];
602 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
606 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
609 error = mac_check_sysv_semctl(cred, semakptr, cmd);
613 bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
614 *rval = IXSEQ_TO_IPCID(semid, semakptr->u.sem_perm);
615 mtx_unlock(sema_mtxp);
619 semidx = IPCID_TO_IX(semid);
620 if (semidx < 0 || semidx >= seminfo.semmni)
623 semakptr = &sema[semidx];
624 sema_mtxp = &sema_mtx[semidx];
629 error = mac_check_sysv_semctl(cred, semakptr, cmd);
639 if ((error = semvalid(semid, semakptr)) != 0)
641 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
643 semakptr->u.sem_perm.cuid = cred->cr_uid;
644 semakptr->u.sem_perm.uid = cred->cr_uid;
645 semakptr->u.sem_perm.mode = 0;
647 semundo_clear(semidx, -1);
650 mac_cleanup_sysv_sem(semakptr);
653 for (i = 0; i < seminfo.semmni; i++) {
654 if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
655 sema[i].u.sem_base > semakptr->u.sem_base)
656 mtx_lock_flags(&sema_mtx[i], LOP_DUPOK);
658 for (i = semakptr->u.sem_base - sem; i < semtot; i++)
659 sem[i] = sem[i + semakptr->u.sem_nsems];
660 for (i = 0; i < seminfo.semmni; i++) {
661 if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
662 sema[i].u.sem_base > semakptr->u.sem_base) {
663 sema[i].u.sem_base -= semakptr->u.sem_nsems;
664 mtx_unlock(&sema_mtx[i]);
667 semtot -= semakptr->u.sem_nsems;
671 if ((error = semvalid(semid, semakptr)) != 0)
673 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
676 semakptr->u.sem_perm.uid = sbuf->sem_perm.uid;
677 semakptr->u.sem_perm.gid = sbuf->sem_perm.gid;
678 semakptr->u.sem_perm.mode = (semakptr->u.sem_perm.mode &
679 ~0777) | (sbuf->sem_perm.mode & 0777);
680 semakptr->u.sem_ctime = time_second;
684 if ((error = semvalid(semid, semakptr)) != 0)
686 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
688 bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
692 if ((error = semvalid(semid, semakptr)) != 0)
694 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
696 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
700 *rval = semakptr->u.sem_base[semnum].semncnt;
704 if ((error = semvalid(semid, semakptr)) != 0)
706 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
708 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
712 *rval = semakptr->u.sem_base[semnum].sempid;
716 if ((error = semvalid(semid, semakptr)) != 0)
718 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
720 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
724 *rval = semakptr->u.sem_base[semnum].semval;
729 * Unfortunately, callers of this function don't know
730 * in advance how many semaphores are in this set.
731 * While we could just allocate the maximum size array
732 * and pass the actual size back to the caller, that
733 * won't work for SETALL since we can't copyin() more
734 * data than the user specified as we may return a
737 * Note that the number of semaphores in a set is
738 * fixed for the life of that set. The only way that
739 * the 'count' could change while are blocked in
740 * malloc() is if this semaphore set were destroyed
741 * and a new one created with the same index.
742 * However, semvalid() will catch that due to the
743 * sequence number unless exactly 0x8000 (or a
744 * multiple thereof) semaphore sets for the same index
745 * are created and destroyed while we are in malloc!
748 count = semakptr->u.sem_nsems;
749 mtx_unlock(sema_mtxp);
750 array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
752 if ((error = semvalid(semid, semakptr)) != 0)
754 KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
755 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
757 for (i = 0; i < semakptr->u.sem_nsems; i++)
758 array[i] = semakptr->u.sem_base[i].semval;
759 mtx_unlock(sema_mtxp);
760 error = copyout(array, arg->array, count * sizeof(*array));
765 if ((error = semvalid(semid, semakptr)) != 0)
767 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
769 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
773 *rval = semakptr->u.sem_base[semnum].semzcnt;
777 if ((error = semvalid(semid, semakptr)) != 0)
779 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
781 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
785 if (arg->val < 0 || arg->val > seminfo.semvmx) {
789 semakptr->u.sem_base[semnum].semval = arg->val;
791 semundo_clear(semidx, semnum);
798 * See comment on GETALL for why 'count' shouldn't change
799 * and why we require a userland buffer.
801 count = semakptr->u.sem_nsems;
802 mtx_unlock(sema_mtxp);
803 array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
804 error = copyin(arg->array, array, count * sizeof(*array));
808 if ((error = semvalid(semid, semakptr)) != 0)
810 KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
811 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
813 for (i = 0; i < semakptr->u.sem_nsems; i++) {
815 if (usval > seminfo.semvmx) {
819 semakptr->u.sem_base[i].semval = usval;
822 semundo_clear(semidx, -1);
833 mtx_unlock(sema_mtxp);
835 mtx_unlock(&sem_mtx);
841 #ifndef _SYS_SYSPROTO_H_
849 semget(struct thread *td, struct semget_args *uap)
851 int semid, error = 0;
853 int nsems = uap->nsems;
854 int semflg = uap->semflg;
855 struct ucred *cred = td->td_ucred;
857 DPRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg));
858 if (!jail_sysvipc_allowed && jailed(td->td_ucred))
862 if (key != IPC_PRIVATE) {
863 for (semid = 0; semid < seminfo.semmni; semid++) {
864 if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) &&
865 sema[semid].u.sem_perm.key == key)
868 if (semid < seminfo.semmni) {
869 DPRINTF(("found public key\n"));
870 if ((error = ipcperm(td, &sema[semid].u.sem_perm,
874 if (nsems > 0 && sema[semid].u.sem_nsems < nsems) {
875 DPRINTF(("too small\n"));
879 if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
880 DPRINTF(("not exclusive\n"));
885 error = mac_check_sysv_semget(cred, &sema[semid]);
893 DPRINTF(("need to allocate the semid_kernel\n"));
894 if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
895 if (nsems <= 0 || nsems > seminfo.semmsl) {
896 DPRINTF(("nsems out of range (0<%d<=%d)\n", nsems,
901 if (nsems > seminfo.semmns - semtot) {
903 "not enough semaphores left (need %d, got %d)\n",
904 nsems, seminfo.semmns - semtot));
908 for (semid = 0; semid < seminfo.semmni; semid++) {
909 if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0)
912 if (semid == seminfo.semmni) {
913 DPRINTF(("no more semid_kernel's available\n"));
917 DPRINTF(("semid %d is available\n", semid));
918 mtx_lock(&sema_mtx[semid]);
919 KASSERT((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0,
920 ("Lost semaphore %d", semid));
921 sema[semid].u.sem_perm.key = key;
922 sema[semid].u.sem_perm.cuid = cred->cr_uid;
923 sema[semid].u.sem_perm.uid = cred->cr_uid;
924 sema[semid].u.sem_perm.cgid = cred->cr_gid;
925 sema[semid].u.sem_perm.gid = cred->cr_gid;
926 sema[semid].u.sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
927 sema[semid].u.sem_perm.seq =
928 (sema[semid].u.sem_perm.seq + 1) & 0x7fff;
929 sema[semid].u.sem_nsems = nsems;
930 sema[semid].u.sem_otime = 0;
931 sema[semid].u.sem_ctime = time_second;
932 sema[semid].u.sem_base = &sem[semtot];
934 bzero(sema[semid].u.sem_base,
935 sizeof(sema[semid].u.sem_base[0])*nsems);
937 mac_create_sysv_sem(cred, &sema[semid]);
939 mtx_unlock(&sema_mtx[semid]);
940 DPRINTF(("sembase = %p, next = %p\n",
941 sema[semid].u.sem_base, &sem[semtot]));
943 DPRINTF(("didn't find it and wasn't asked to create it\n"));
949 td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].u.sem_perm);
951 mtx_unlock(&sem_mtx);
955 #ifndef _SYS_SYSPROTO_H_
963 semop(struct thread *td, struct semop_args *uap)
966 struct sembuf small_sops[SMALL_SOPS];
967 int semid = uap->semid;
968 size_t nsops = uap->nsops;
970 struct semid_kernel *semakptr;
971 struct sembuf *sopptr = 0;
972 struct sem *semptr = 0;
973 struct sem_undo *suptr;
974 struct mtx *sema_mtxp;
977 int do_wakeup, do_undos;
983 DPRINTF(("call to semop(%d, %p, %u)\n", semid, sops, nsops));
985 if (!jail_sysvipc_allowed && jailed(td->td_ucred))
988 semid = IPCID_TO_IX(semid); /* Convert back to zero origin */
990 if (semid < 0 || semid >= seminfo.semmni)
993 /* Allocate memory for sem_ops */
994 if (nsops <= SMALL_SOPS)
996 else if (nsops <= seminfo.semopm)
997 sops = malloc(nsops * sizeof(*sops), M_TEMP, M_WAITOK);
999 DPRINTF(("too many sops (max=%d, nsops=%d)\n", seminfo.semopm,
1003 if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) {
1004 DPRINTF(("error = %d from copyin(%p, %p, %d)\n", error,
1005 uap->sops, sops, nsops * sizeof(sops[0])));
1006 if (sops != small_sops)
1011 semakptr = &sema[semid];
1012 sema_mtxp = &sema_mtx[semid];
1013 mtx_lock(sema_mtxp);
1014 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
1018 seq = semakptr->u.sem_perm.seq;
1019 if (seq != IPCID_TO_SEQ(uap->semid)) {
1024 * Initial pass thru sops to see what permissions are needed.
1025 * Also perform any checks that don't need repeating on each
1026 * attempt to satisfy the request vector.
1028 j = 0; /* permission needed */
1030 for (i = 0; i < nsops; i++) {
1032 if (sopptr->sem_num >= semakptr->u.sem_nsems) {
1036 if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0)
1038 j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A;
1041 if ((error = ipcperm(td, &semakptr->u.sem_perm, j))) {
1042 DPRINTF(("error = %d from ipaccess\n", error));
1046 error = mac_check_sysv_semop(td->td_ucred, semakptr, j);
1052 * Loop trying to satisfy the vector of requests.
1053 * If we reach a point where we must wait, any requests already
1054 * performed are rolled back and we go to sleep until some other
1055 * process wakes us up. At this point, we start all over again.
1057 * This ensures that from the perspective of other tasks, a set
1058 * of requests is atomic (never partially satisfied).
1062 error = 0; /* error return if necessary */
1064 for (i = 0; i < nsops; i++) {
1066 semptr = &semakptr->u.sem_base[sopptr->sem_num];
1069 "semop: semakptr=%p, sem_base=%p, "
1070 "semptr=%p, sem[%d]=%d : op=%d, flag=%s\n",
1071 semakptr, semakptr->u.sem_base, semptr,
1072 sopptr->sem_num, semptr->semval, sopptr->sem_op,
1073 (sopptr->sem_flg & IPC_NOWAIT) ?
1074 "nowait" : "wait"));
1076 if (sopptr->sem_op < 0) {
1077 if (semptr->semval + sopptr->sem_op < 0) {
1078 DPRINTF(("semop: can't do it now\n"));
1081 semptr->semval += sopptr->sem_op;
1082 if (semptr->semval == 0 &&
1083 semptr->semzcnt > 0)
1086 } else if (sopptr->sem_op == 0) {
1087 if (semptr->semval != 0) {
1088 DPRINTF(("semop: not zero now\n"));
1091 } else if (semptr->semval + sopptr->sem_op >
1096 if (semptr->semncnt > 0)
1098 semptr->semval += sopptr->sem_op;
1103 * Did we get through the entire vector?
1109 * No ... rollback anything that we've already done
1111 DPRINTF(("semop: rollback 0 through %d\n", i-1));
1112 for (j = 0; j < i; j++)
1113 semakptr->u.sem_base[sops[j].sem_num].semval -=
1116 /* If we detected an error, return it */
1121 * If the request that we couldn't satisfy has the
1122 * NOWAIT flag set then return with EAGAIN.
1124 if (sopptr->sem_flg & IPC_NOWAIT) {
1129 if (sopptr->sem_op == 0)
1134 DPRINTF(("semop: good night!\n"));
1135 error = msleep(semakptr, sema_mtxp, (PZERO - 4) | PCATCH,
1137 DPRINTF(("semop: good morning (error=%d)!\n", error));
1138 /* return code is checked below, after sem[nz]cnt-- */
1141 * Make sure that the semaphore still exists
1143 seq = semakptr->u.sem_perm.seq;
1144 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1145 seq != IPCID_TO_SEQ(uap->semid)) {
1151 * Renew the semaphore's pointer after wakeup since
1152 * during msleep sem_base may have been modified and semptr
1153 * is not valid any more
1155 semptr = &semakptr->u.sem_base[sopptr->sem_num];
1158 * The semaphore is still alive. Readjust the count of
1159 * waiting processes.
1161 if (sopptr->sem_op == 0)
1167 * Is it really morning, or was our sleep interrupted?
1168 * (Delayed check of msleep() return code because we
1169 * need to decrement sem[nz]cnt either way.)
1175 DPRINTF(("semop: good morning!\n"));
1180 * Process any SEM_UNDO requests.
1185 for (i = 0; i < nsops; i++) {
1187 * We only need to deal with SEM_UNDO's for non-zero
1192 if ((sops[i].sem_flg & SEM_UNDO) == 0)
1194 adjval = sops[i].sem_op;
1197 error = semundo_adjust(td, &suptr, semid, seq,
1198 sops[i].sem_num, -adjval);
1203 * Oh-Oh! We ran out of either sem_undo's or undo's.
1204 * Rollback the adjustments to this point and then
1205 * rollback the semaphore ups and down so we can return
1206 * with an error with all structures restored. We
1207 * rollback the undo's in the exact reverse order that
1208 * we applied them. This guarantees that we won't run
1209 * out of space as we roll things back out.
1211 for (j = 0; j < i; j++) {
1213 if ((sops[k].sem_flg & SEM_UNDO) == 0)
1215 adjval = sops[k].sem_op;
1218 if (semundo_adjust(td, &suptr, semid, seq,
1219 sops[k].sem_num, adjval) != 0)
1220 panic("semop - can't undo undos");
1223 for (j = 0; j < nsops; j++)
1224 semakptr->u.sem_base[sops[j].sem_num].semval -=
1227 DPRINTF(("error = %d from semundo_adjust\n", error));
1230 } /* loop through the sops */
1232 } /* if (do_undos) */
1234 /* We're definitely done - set the sempid's and time */
1235 for (i = 0; i < nsops; i++) {
1237 semptr = &semakptr->u.sem_base[sopptr->sem_num];
1238 semptr->sempid = td->td_proc->p_pid;
1240 semakptr->u.sem_otime = time_second;
1243 * Do a wakeup if any semaphore was up'd whilst something was
1247 DPRINTF(("semop: doing wakeup\n"));
1249 DPRINTF(("semop: back from wakeup\n"));
1251 DPRINTF(("semop: done\n"));
1252 td->td_retval[0] = 0;
1254 mtx_unlock(sema_mtxp);
1255 if (sops != small_sops)
1261 * Go through the undo structures for this process and apply the adjustments to
1265 semexit_myhook(void *arg, struct proc *p)
1267 struct sem_undo *suptr;
1268 struct semid_kernel *semakptr;
1269 struct mtx *sema_mtxp;
1270 int semid, semnum, adjval, ix;
1274 * Go through the chain of undo vectors looking for one
1275 * associated with this process.
1278 LIST_FOREACH(suptr, &semu_list, un_next) {
1279 if (suptr->un_proc == p)
1282 if (suptr == NULL) {
1286 LIST_REMOVE(suptr, un_next);
1288 DPRINTF(("proc @%p has undo structure with %d entries\n", p,
1292 * If there are any active undo elements then process them.
1294 if (suptr->un_cnt > 0) {
1296 for (ix = 0; ix < suptr->un_cnt; ix++) {
1297 semid = suptr->un_ent[ix].un_id;
1298 semnum = suptr->un_ent[ix].un_num;
1299 adjval = suptr->un_ent[ix].un_adjval;
1300 seq = suptr->un_ent[ix].un_seq;
1301 semakptr = &sema[semid];
1302 sema_mtxp = &sema_mtx[semid];
1304 mtx_lock(sema_mtxp);
1305 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1306 (semakptr->u.sem_perm.seq != seq)) {
1307 mtx_unlock(sema_mtxp);
1310 if (semnum >= semakptr->u.sem_nsems)
1311 panic("semexit - semnum out of range");
1314 "semexit: %p id=%d num=%d(adj=%d) ; sem=%d\n",
1315 suptr->un_proc, suptr->un_ent[ix].un_id,
1316 suptr->un_ent[ix].un_num,
1317 suptr->un_ent[ix].un_adjval,
1318 semakptr->u.sem_base[semnum].semval));
1320 if (adjval < 0 && semakptr->u.sem_base[semnum].semval <
1322 semakptr->u.sem_base[semnum].semval = 0;
1324 semakptr->u.sem_base[semnum].semval += adjval;
1327 DPRINTF(("semexit: back from wakeup\n"));
1328 mtx_unlock(sema_mtxp);
1334 * Deallocate the undo vector.
1336 DPRINTF(("removing vector\n"));
1337 suptr->un_proc = NULL;
1339 LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
1344 sysctl_sema(SYSCTL_HANDLER_ARGS)
1347 return (SYSCTL_OUT(req, sema,
1348 sizeof(struct semid_kernel) * seminfo.semmni));