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"
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/eventhandler.h>
48 #include <sys/kernel.h>
51 #include <sys/module.h>
52 #include <sys/mutex.h>
54 #include <sys/syscall.h>
55 #include <sys/syscallsubr.h>
56 #include <sys/sysent.h>
57 #include <sys/sysctl.h>
59 #include <sys/malloc.h>
62 #include <security/mac/mac_framework.h>
64 static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores");
67 #define DPRINTF(a) printf a
72 static void seminit(void);
73 static int sysvsem_modload(struct module *, int, void *);
74 static int semunload(void);
75 static void semexit_myhook(void *arg, struct proc *p);
76 static int sysctl_sema(SYSCTL_HANDLER_ARGS);
77 static int semvalid(int semid, struct semid_kernel *semakptr);
79 #ifndef _SYS_SYSPROTO_H_
81 int __semctl(struct thread *td, struct __semctl_args *uap);
83 int semget(struct thread *td, struct semget_args *uap);
85 int semop(struct thread *td, struct semop_args *uap);
88 static struct sem_undo *semu_alloc(struct thread *td);
89 static int semundo_adjust(struct thread *td, struct sem_undo **supptr,
90 int semid, int semseq, int semnum, int adjval);
91 static void semundo_clear(int semid, int semnum);
93 /* XXX casting to (sy_call_t *) is bogus, as usual. */
94 static sy_call_t *semcalls[] = {
95 (sy_call_t *)__semctl, (sy_call_t *)semget,
99 static struct mtx sem_mtx; /* semaphore global lock */
100 static struct mtx sem_undo_mtx;
101 static int semtot = 0;
102 static struct semid_kernel *sema; /* semaphore id pool */
103 static struct mtx *sema_mtx; /* semaphore id pool mutexes*/
104 static struct sem *sem; /* semaphore pool */
105 LIST_HEAD(, sem_undo) semu_list; /* list of active undo structures */
106 LIST_HEAD(, sem_undo) semu_free_list; /* list of free undo structures */
107 static int *semu; /* undo structure pool */
108 static eventhandler_tag semexit_tag;
110 #define SEMUNDO_MTX sem_undo_mtx
111 #define SEMUNDO_LOCK() mtx_lock(&SEMUNDO_MTX);
112 #define SEMUNDO_UNLOCK() mtx_unlock(&SEMUNDO_MTX);
113 #define SEMUNDO_LOCKASSERT(how) mtx_assert(&SEMUNDO_MTX, (how));
116 u_short semval; /* semaphore value */
117 pid_t sempid; /* pid of last operation */
118 u_short semncnt; /* # awaiting semval > cval */
119 u_short semzcnt; /* # awaiting semval = 0 */
123 * Undo structure (one per process)
126 LIST_ENTRY(sem_undo) un_next; /* ptr to next active undo structure */
127 struct proc *un_proc; /* owner of this structure */
128 short un_cnt; /* # of active entries */
130 short un_adjval; /* adjust on exit values */
131 short un_num; /* semaphore # */
132 int un_id; /* semid */
133 unsigned short un_seq;
134 } un_ent[1]; /* undo entries */
138 * Configuration parameters
141 #define SEMMNI 10 /* # of semaphore identifiers */
144 #define SEMMNS 60 /* # of semaphores in system */
147 #define SEMUME 10 /* max # of undo entries per process */
150 #define SEMMNU 30 /* # of undo structures in system */
153 /* shouldn't need tuning */
155 #define SEMMAP 30 /* # of entries in semaphore map */
158 #define SEMMSL SEMMNS /* max # of semaphores per id */
161 #define SEMOPM 100 /* max # of operations per semop call */
164 #define SEMVMX 32767 /* semaphore maximum value */
165 #define SEMAEM 16384 /* adjust on exit max value */
168 * Due to the way semaphore memory is allocated, we have to ensure that
169 * SEMUSZ is properly aligned.
172 #define SEM_ALIGN(bytes) (((bytes) + (sizeof(long) - 1)) & ~(sizeof(long) - 1))
174 /* actual size of an undo structure */
175 #define SEMUSZ SEM_ALIGN(offsetof(struct sem_undo, un_ent[SEMUME]))
178 * Macro to find a particular sem_undo vector
181 ((struct sem_undo *)(((intptr_t)semu)+ix * seminfo.semusz))
184 * semaphore info struct
186 struct seminfo seminfo = {
187 SEMMAP, /* # of entries in semaphore map */
188 SEMMNI, /* # of semaphore identifiers */
189 SEMMNS, /* # of semaphores in system */
190 SEMMNU, /* # of undo structures in system */
191 SEMMSL, /* max # of semaphores per id */
192 SEMOPM, /* max # of operations per semop call */
193 SEMUME, /* max # of undo entries per process */
194 SEMUSZ, /* size in bytes of undo structure */
195 SEMVMX, /* semaphore maximum value */
196 SEMAEM /* adjust on exit max value */
199 SYSCTL_INT(_kern_ipc, OID_AUTO, semmap, CTLFLAG_RW, &seminfo.semmap, 0,
200 "Number of entries in the semaphore map");
201 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RDTUN, &seminfo.semmni, 0,
202 "Number of semaphore identifiers");
203 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RDTUN, &seminfo.semmns, 0,
204 "Maximum number of semaphores in the system");
205 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RDTUN, &seminfo.semmnu, 0,
206 "Maximum number of undo structures in the system");
207 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0,
208 "Max semaphores per id");
209 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RDTUN, &seminfo.semopm, 0,
210 "Max operations per semop call");
211 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RDTUN, &seminfo.semume, 0,
212 "Max undo entries per process");
213 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RDTUN, &seminfo.semusz, 0,
214 "Size in bytes of undo structure");
215 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RW, &seminfo.semvmx, 0,
216 "Semaphore maximum value");
217 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &seminfo.semaem, 0,
218 "Adjust on exit max value");
219 SYSCTL_PROC(_kern_ipc, OID_AUTO, sema, CTLFLAG_RD,
220 NULL, 0, sysctl_sema, "", "");
227 TUNABLE_INT_FETCH("kern.ipc.semmap", &seminfo.semmap);
228 TUNABLE_INT_FETCH("kern.ipc.semmni", &seminfo.semmni);
229 TUNABLE_INT_FETCH("kern.ipc.semmns", &seminfo.semmns);
230 TUNABLE_INT_FETCH("kern.ipc.semmnu", &seminfo.semmnu);
231 TUNABLE_INT_FETCH("kern.ipc.semmsl", &seminfo.semmsl);
232 TUNABLE_INT_FETCH("kern.ipc.semopm", &seminfo.semopm);
233 TUNABLE_INT_FETCH("kern.ipc.semume", &seminfo.semume);
234 TUNABLE_INT_FETCH("kern.ipc.semusz", &seminfo.semusz);
235 TUNABLE_INT_FETCH("kern.ipc.semvmx", &seminfo.semvmx);
236 TUNABLE_INT_FETCH("kern.ipc.semaem", &seminfo.semaem);
238 sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
239 sema = malloc(sizeof(struct semid_kernel) * seminfo.semmni, M_SEM,
241 sema_mtx = malloc(sizeof(struct mtx) * seminfo.semmni, M_SEM,
243 semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
245 for (i = 0; i < seminfo.semmni; i++) {
246 sema[i].u.sem_base = 0;
247 sema[i].u.sem_perm.mode = 0;
248 sema[i].u.sem_perm.seq = 0;
250 mac_sysvsem_init(&sema[i]);
253 for (i = 0; i < seminfo.semmni; i++)
254 mtx_init(&sema_mtx[i], "semid", NULL, MTX_DEF);
255 LIST_INIT(&semu_free_list);
256 for (i = 0; i < seminfo.semmnu; i++) {
257 struct sem_undo *suptr = SEMU(i);
258 suptr->un_proc = NULL;
259 LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
261 LIST_INIT(&semu_list);
262 mtx_init(&sem_mtx, "sem", NULL, MTX_DEF);
263 mtx_init(&sem_undo_mtx, "semu", NULL, MTX_DEF);
264 semexit_tag = EVENTHANDLER_REGISTER(process_exit, semexit_myhook, NULL,
265 EVENTHANDLER_PRI_ANY);
277 EVENTHANDLER_DEREGISTER(process_exit, semexit_tag);
279 for (i = 0; i < seminfo.semmni; i++)
280 mac_sysvsem_destroy(&sema[i]);
285 for (i = 0; i < seminfo.semmni; i++)
286 mtx_destroy(&sema_mtx[i]);
287 free(sema_mtx, M_SEM);
288 mtx_destroy(&sem_mtx);
289 mtx_destroy(&sem_undo_mtx);
294 sysvsem_modload(struct module *module, int cmd, void *arg)
314 static moduledata_t sysvsem_mod = {
320 SYSCALL_MODULE_HELPER(semsys);
321 SYSCALL_MODULE_HELPER(__semctl);
322 SYSCALL_MODULE_HELPER(semget);
323 SYSCALL_MODULE_HELPER(semop);
325 DECLARE_MODULE(sysvsem, sysvsem_mod, SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
326 MODULE_VERSION(sysvsem, 1);
329 * Entry point for all SEM calls.
334 /* XXX actually varargs. */
335 struct semsys_args /* {
345 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
347 if (uap->which < 0 ||
348 uap->which >= sizeof(semcalls)/sizeof(semcalls[0]))
350 error = (*semcalls[uap->which])(td, &uap->a2);
355 * Allocate a new sem_undo structure for a process
356 * (returns ptr to structure or NULL if no more room)
359 static struct sem_undo *
360 semu_alloc(struct thread *td)
362 struct sem_undo *suptr;
364 SEMUNDO_LOCKASSERT(MA_OWNED);
365 if ((suptr = LIST_FIRST(&semu_free_list)) == NULL)
367 LIST_REMOVE(suptr, un_next);
368 LIST_INSERT_HEAD(&semu_list, suptr, un_next);
370 suptr->un_proc = td->td_proc;
375 semu_try_free(struct sem_undo *suptr)
378 SEMUNDO_LOCKASSERT(MA_OWNED);
380 if (suptr->un_cnt != 0)
382 LIST_REMOVE(suptr, un_next);
383 LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
388 * Adjust a particular entry for a particular proc
392 semundo_adjust(struct thread *td, struct sem_undo **supptr, int semid,
393 int semseq, int semnum, int adjval)
395 struct proc *p = td->td_proc;
396 struct sem_undo *suptr;
400 SEMUNDO_LOCKASSERT(MA_OWNED);
401 /* Look for and remember the sem_undo if the caller doesn't provide
406 LIST_FOREACH(suptr, &semu_list, un_next) {
407 if (suptr->un_proc == p) {
415 suptr = semu_alloc(td);
423 * Look for the requested entry and adjust it (delete if adjval becomes
426 sunptr = &suptr->un_ent[0];
427 for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
428 if (sunptr->un_id != semid || sunptr->un_num != semnum)
431 adjval += sunptr->un_adjval;
432 if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
435 sunptr->un_adjval = adjval;
436 if (sunptr->un_adjval == 0) {
438 if (i < suptr->un_cnt)
440 suptr->un_ent[suptr->un_cnt];
441 if (suptr->un_cnt == 0)
442 semu_try_free(suptr);
447 /* Didn't find the right entry - create it */
450 if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
452 if (suptr->un_cnt != seminfo.semume) {
453 sunptr = &suptr->un_ent[suptr->un_cnt];
455 sunptr->un_adjval = adjval;
456 sunptr->un_id = semid;
457 sunptr->un_num = semnum;
458 sunptr->un_seq = semseq;
465 semundo_clear(int semid, int semnum)
467 struct sem_undo *suptr, *suptr1;
471 SEMUNDO_LOCKASSERT(MA_OWNED);
472 LIST_FOREACH_SAFE(suptr, &semu_list, un_next, suptr1) {
473 sunptr = &suptr->un_ent[0];
474 for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
475 if (sunptr->un_id != semid)
477 if (semnum == -1 || sunptr->un_num == semnum) {
479 if (i < suptr->un_cnt) {
481 suptr->un_ent[suptr->un_cnt];
484 semu_try_free(suptr);
493 semvalid(int semid, struct semid_kernel *semakptr)
496 return ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
497 semakptr->u.sem_perm.seq != IPCID_TO_SEQ(semid) ? EINVAL : 0);
501 * Note that the user-mode half of this passes a union, not a pointer.
503 #ifndef _SYS_SYSPROTO_H_
504 struct __semctl_args {
512 __semctl(struct thread *td, struct __semctl_args *uap)
514 struct semid_ds dsbuf;
515 union semun arg, semun;
526 error = copyin(uap->arg, &arg, sizeof(arg));
538 error = copyin(arg.buf, &dsbuf, sizeof(dsbuf));
545 semun.array = arg.array;
552 error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
560 error = copyout(&dsbuf, arg.buf, sizeof(dsbuf));
565 td->td_retval[0] = rval;
570 kern_semctl(struct thread *td, int semid, int semnum, int cmd,
571 union semun *arg, register_t *rval)
574 struct ucred *cred = td->td_ucred;
576 struct semid_ds *sbuf;
577 struct semid_kernel *semakptr;
578 struct mtx *sema_mtxp;
579 u_short usval, count;
582 DPRINTF(("call to semctl(%d, %d, %d, 0x%p)\n",
583 semid, semnum, cmd, arg));
584 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
592 * For this command we assume semid is an array index
593 * rather than an IPC id.
595 if (semid < 0 || semid >= seminfo.semmni)
597 semakptr = &sema[semid];
598 sema_mtxp = &sema_mtx[semid];
600 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
604 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
607 error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
611 bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
612 *rval = IXSEQ_TO_IPCID(semid, semakptr->u.sem_perm);
613 mtx_unlock(sema_mtxp);
617 semidx = IPCID_TO_IX(semid);
618 if (semidx < 0 || semidx >= seminfo.semmni)
621 semakptr = &sema[semidx];
622 sema_mtxp = &sema_mtx[semidx];
627 error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
637 if ((error = semvalid(semid, semakptr)) != 0)
639 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
641 semakptr->u.sem_perm.cuid = cred->cr_uid;
642 semakptr->u.sem_perm.uid = cred->cr_uid;
643 semakptr->u.sem_perm.mode = 0;
645 semundo_clear(semidx, -1);
648 mac_sysvsem_cleanup(semakptr);
651 for (i = 0; i < seminfo.semmni; i++) {
652 if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
653 sema[i].u.sem_base > semakptr->u.sem_base)
654 mtx_lock_flags(&sema_mtx[i], LOP_DUPOK);
656 for (i = semakptr->u.sem_base - sem; i < semtot; i++)
657 sem[i] = sem[i + semakptr->u.sem_nsems];
658 for (i = 0; i < seminfo.semmni; i++) {
659 if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
660 sema[i].u.sem_base > semakptr->u.sem_base) {
661 sema[i].u.sem_base -= semakptr->u.sem_nsems;
662 mtx_unlock(&sema_mtx[i]);
665 semtot -= semakptr->u.sem_nsems;
669 if ((error = semvalid(semid, semakptr)) != 0)
671 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
674 semakptr->u.sem_perm.uid = sbuf->sem_perm.uid;
675 semakptr->u.sem_perm.gid = sbuf->sem_perm.gid;
676 semakptr->u.sem_perm.mode = (semakptr->u.sem_perm.mode &
677 ~0777) | (sbuf->sem_perm.mode & 0777);
678 semakptr->u.sem_ctime = time_second;
682 if ((error = semvalid(semid, semakptr)) != 0)
684 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
686 bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
690 if ((error = semvalid(semid, semakptr)) != 0)
692 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
694 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
698 *rval = semakptr->u.sem_base[semnum].semncnt;
702 if ((error = semvalid(semid, semakptr)) != 0)
704 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
706 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
710 *rval = semakptr->u.sem_base[semnum].sempid;
714 if ((error = semvalid(semid, semakptr)) != 0)
716 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
718 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
722 *rval = semakptr->u.sem_base[semnum].semval;
727 * Unfortunately, callers of this function don't know
728 * in advance how many semaphores are in this set.
729 * While we could just allocate the maximum size array
730 * and pass the actual size back to the caller, that
731 * won't work for SETALL since we can't copyin() more
732 * data than the user specified as we may return a
735 * Note that the number of semaphores in a set is
736 * fixed for the life of that set. The only way that
737 * the 'count' could change while are blocked in
738 * malloc() is if this semaphore set were destroyed
739 * and a new one created with the same index.
740 * However, semvalid() will catch that due to the
741 * sequence number unless exactly 0x8000 (or a
742 * multiple thereof) semaphore sets for the same index
743 * are created and destroyed while we are in malloc!
746 count = semakptr->u.sem_nsems;
747 mtx_unlock(sema_mtxp);
748 array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
750 if ((error = semvalid(semid, semakptr)) != 0)
752 KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
753 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
755 for (i = 0; i < semakptr->u.sem_nsems; i++)
756 array[i] = semakptr->u.sem_base[i].semval;
757 mtx_unlock(sema_mtxp);
758 error = copyout(array, arg->array, count * sizeof(*array));
763 if ((error = semvalid(semid, semakptr)) != 0)
765 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
767 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
771 *rval = semakptr->u.sem_base[semnum].semzcnt;
775 if ((error = semvalid(semid, semakptr)) != 0)
777 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
779 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
783 if (arg->val < 0 || arg->val > seminfo.semvmx) {
787 semakptr->u.sem_base[semnum].semval = arg->val;
789 semundo_clear(semidx, semnum);
796 * See comment on GETALL for why 'count' shouldn't change
797 * and why we require a userland buffer.
799 count = semakptr->u.sem_nsems;
800 mtx_unlock(sema_mtxp);
801 array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
802 error = copyin(arg->array, array, count * sizeof(*array));
806 if ((error = semvalid(semid, semakptr)) != 0)
808 KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
809 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
811 for (i = 0; i < semakptr->u.sem_nsems; i++) {
813 if (usval > seminfo.semvmx) {
817 semakptr->u.sem_base[i].semval = usval;
820 semundo_clear(semidx, -1);
831 mtx_unlock(sema_mtxp);
833 mtx_unlock(&sem_mtx);
839 #ifndef _SYS_SYSPROTO_H_
847 semget(struct thread *td, struct semget_args *uap)
849 int semid, error = 0;
851 int nsems = uap->nsems;
852 int semflg = uap->semflg;
853 struct ucred *cred = td->td_ucred;
855 DPRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg));
856 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
860 if (key != IPC_PRIVATE) {
861 for (semid = 0; semid < seminfo.semmni; semid++) {
862 if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) &&
863 sema[semid].u.sem_perm.key == key)
866 if (semid < seminfo.semmni) {
867 DPRINTF(("found public key\n"));
868 if ((error = ipcperm(td, &sema[semid].u.sem_perm,
872 if (nsems > 0 && sema[semid].u.sem_nsems < nsems) {
873 DPRINTF(("too small\n"));
877 if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
878 DPRINTF(("not exclusive\n"));
883 error = mac_sysvsem_check_semget(cred, &sema[semid]);
891 DPRINTF(("need to allocate the semid_kernel\n"));
892 if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
893 if (nsems <= 0 || nsems > seminfo.semmsl) {
894 DPRINTF(("nsems out of range (0<%d<=%d)\n", nsems,
899 if (nsems > seminfo.semmns - semtot) {
901 "not enough semaphores left (need %d, got %d)\n",
902 nsems, seminfo.semmns - semtot));
906 for (semid = 0; semid < seminfo.semmni; semid++) {
907 if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0)
910 if (semid == seminfo.semmni) {
911 DPRINTF(("no more semid_kernel's available\n"));
915 DPRINTF(("semid %d is available\n", semid));
916 mtx_lock(&sema_mtx[semid]);
917 KASSERT((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0,
918 ("Lost semaphore %d", semid));
919 sema[semid].u.sem_perm.key = key;
920 sema[semid].u.sem_perm.cuid = cred->cr_uid;
921 sema[semid].u.sem_perm.uid = cred->cr_uid;
922 sema[semid].u.sem_perm.cgid = cred->cr_gid;
923 sema[semid].u.sem_perm.gid = cred->cr_gid;
924 sema[semid].u.sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
925 sema[semid].u.sem_perm.seq =
926 (sema[semid].u.sem_perm.seq + 1) & 0x7fff;
927 sema[semid].u.sem_nsems = nsems;
928 sema[semid].u.sem_otime = 0;
929 sema[semid].u.sem_ctime = time_second;
930 sema[semid].u.sem_base = &sem[semtot];
932 bzero(sema[semid].u.sem_base,
933 sizeof(sema[semid].u.sem_base[0])*nsems);
935 mac_sysvsem_create(cred, &sema[semid]);
937 mtx_unlock(&sema_mtx[semid]);
938 DPRINTF(("sembase = %p, next = %p\n",
939 sema[semid].u.sem_base, &sem[semtot]));
941 DPRINTF(("didn't find it and wasn't asked to create it\n"));
947 td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].u.sem_perm);
949 mtx_unlock(&sem_mtx);
953 #ifndef _SYS_SYSPROTO_H_
961 semop(struct thread *td, struct semop_args *uap)
964 struct sembuf small_sops[SMALL_SOPS];
965 int semid = uap->semid;
966 size_t nsops = uap->nsops;
968 struct semid_kernel *semakptr;
969 struct sembuf *sopptr = 0;
970 struct sem *semptr = 0;
971 struct sem_undo *suptr;
972 struct mtx *sema_mtxp;
975 int do_wakeup, do_undos;
981 DPRINTF(("call to semop(%d, %p, %u)\n", semid, sops, nsops));
983 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
986 semid = IPCID_TO_IX(semid); /* Convert back to zero origin */
988 if (semid < 0 || semid >= seminfo.semmni)
991 /* Allocate memory for sem_ops */
992 if (nsops <= SMALL_SOPS)
994 else if (nsops <= seminfo.semopm)
995 sops = malloc(nsops * sizeof(*sops), M_TEMP, M_WAITOK);
997 DPRINTF(("too many sops (max=%d, nsops=%d)\n", seminfo.semopm,
1001 if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) {
1002 DPRINTF(("error = %d from copyin(%p, %p, %d)\n", error,
1003 uap->sops, sops, nsops * sizeof(sops[0])));
1004 if (sops != small_sops)
1009 semakptr = &sema[semid];
1010 sema_mtxp = &sema_mtx[semid];
1011 mtx_lock(sema_mtxp);
1012 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
1016 seq = semakptr->u.sem_perm.seq;
1017 if (seq != IPCID_TO_SEQ(uap->semid)) {
1022 * Initial pass thru sops to see what permissions are needed.
1023 * Also perform any checks that don't need repeating on each
1024 * attempt to satisfy the request vector.
1026 j = 0; /* permission needed */
1028 for (i = 0; i < nsops; i++) {
1030 if (sopptr->sem_num >= semakptr->u.sem_nsems) {
1034 if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0)
1036 j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A;
1039 if ((error = ipcperm(td, &semakptr->u.sem_perm, j))) {
1040 DPRINTF(("error = %d from ipaccess\n", error));
1044 error = mac_sysvsem_check_semop(td->td_ucred, semakptr, j);
1050 * Loop trying to satisfy the vector of requests.
1051 * If we reach a point where we must wait, any requests already
1052 * performed are rolled back and we go to sleep until some other
1053 * process wakes us up. At this point, we start all over again.
1055 * This ensures that from the perspective of other tasks, a set
1056 * of requests is atomic (never partially satisfied).
1060 error = 0; /* error return if necessary */
1062 for (i = 0; i < nsops; i++) {
1064 semptr = &semakptr->u.sem_base[sopptr->sem_num];
1067 "semop: semakptr=%p, sem_base=%p, "
1068 "semptr=%p, sem[%d]=%d : op=%d, flag=%s\n",
1069 semakptr, semakptr->u.sem_base, semptr,
1070 sopptr->sem_num, semptr->semval, sopptr->sem_op,
1071 (sopptr->sem_flg & IPC_NOWAIT) ?
1072 "nowait" : "wait"));
1074 if (sopptr->sem_op < 0) {
1075 if (semptr->semval + sopptr->sem_op < 0) {
1076 DPRINTF(("semop: can't do it now\n"));
1079 semptr->semval += sopptr->sem_op;
1080 if (semptr->semval == 0 &&
1081 semptr->semzcnt > 0)
1084 } else if (sopptr->sem_op == 0) {
1085 if (semptr->semval != 0) {
1086 DPRINTF(("semop: not zero now\n"));
1089 } else if (semptr->semval + sopptr->sem_op >
1094 if (semptr->semncnt > 0)
1096 semptr->semval += sopptr->sem_op;
1101 * Did we get through the entire vector?
1107 * No ... rollback anything that we've already done
1109 DPRINTF(("semop: rollback 0 through %d\n", i-1));
1110 for (j = 0; j < i; j++)
1111 semakptr->u.sem_base[sops[j].sem_num].semval -=
1114 /* If we detected an error, return it */
1119 * If the request that we couldn't satisfy has the
1120 * NOWAIT flag set then return with EAGAIN.
1122 if (sopptr->sem_flg & IPC_NOWAIT) {
1127 if (sopptr->sem_op == 0)
1132 DPRINTF(("semop: good night!\n"));
1133 error = msleep(semakptr, sema_mtxp, (PZERO - 4) | PCATCH,
1135 DPRINTF(("semop: good morning (error=%d)!\n", error));
1136 /* return code is checked below, after sem[nz]cnt-- */
1139 * Make sure that the semaphore still exists
1141 seq = semakptr->u.sem_perm.seq;
1142 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1143 seq != IPCID_TO_SEQ(uap->semid)) {
1149 * Renew the semaphore's pointer after wakeup since
1150 * during msleep sem_base may have been modified and semptr
1151 * is not valid any more
1153 semptr = &semakptr->u.sem_base[sopptr->sem_num];
1156 * The semaphore is still alive. Readjust the count of
1157 * waiting processes.
1159 if (sopptr->sem_op == 0)
1165 * Is it really morning, or was our sleep interrupted?
1166 * (Delayed check of msleep() return code because we
1167 * need to decrement sem[nz]cnt either way.)
1173 DPRINTF(("semop: good morning!\n"));
1178 * Process any SEM_UNDO requests.
1183 for (i = 0; i < nsops; i++) {
1185 * We only need to deal with SEM_UNDO's for non-zero
1190 if ((sops[i].sem_flg & SEM_UNDO) == 0)
1192 adjval = sops[i].sem_op;
1195 error = semundo_adjust(td, &suptr, semid, seq,
1196 sops[i].sem_num, -adjval);
1201 * Oh-Oh! We ran out of either sem_undo's or undo's.
1202 * Rollback the adjustments to this point and then
1203 * rollback the semaphore ups and down so we can return
1204 * with an error with all structures restored. We
1205 * rollback the undo's in the exact reverse order that
1206 * we applied them. This guarantees that we won't run
1207 * out of space as we roll things back out.
1209 for (j = 0; j < i; j++) {
1211 if ((sops[k].sem_flg & SEM_UNDO) == 0)
1213 adjval = sops[k].sem_op;
1216 if (semundo_adjust(td, &suptr, semid, seq,
1217 sops[k].sem_num, adjval) != 0)
1218 panic("semop - can't undo undos");
1221 for (j = 0; j < nsops; j++)
1222 semakptr->u.sem_base[sops[j].sem_num].semval -=
1225 DPRINTF(("error = %d from semundo_adjust\n", error));
1228 } /* loop through the sops */
1230 } /* if (do_undos) */
1232 /* We're definitely done - set the sempid's and time */
1233 for (i = 0; i < nsops; i++) {
1235 semptr = &semakptr->u.sem_base[sopptr->sem_num];
1236 semptr->sempid = td->td_proc->p_pid;
1238 semakptr->u.sem_otime = time_second;
1241 * Do a wakeup if any semaphore was up'd whilst something was
1245 DPRINTF(("semop: doing wakeup\n"));
1247 DPRINTF(("semop: back from wakeup\n"));
1249 DPRINTF(("semop: done\n"));
1250 td->td_retval[0] = 0;
1252 mtx_unlock(sema_mtxp);
1253 if (sops != small_sops)
1259 * Go through the undo structures for this process and apply the adjustments to
1263 semexit_myhook(void *arg, struct proc *p)
1265 struct sem_undo *suptr;
1266 struct semid_kernel *semakptr;
1267 struct mtx *sema_mtxp;
1268 int semid, semnum, adjval, ix;
1272 * Go through the chain of undo vectors looking for one
1273 * associated with this process.
1276 LIST_FOREACH(suptr, &semu_list, un_next) {
1277 if (suptr->un_proc == p)
1280 if (suptr == NULL) {
1284 LIST_REMOVE(suptr, un_next);
1286 DPRINTF(("proc @%p has undo structure with %d entries\n", p,
1290 * If there are any active undo elements then process them.
1292 if (suptr->un_cnt > 0) {
1294 for (ix = 0; ix < suptr->un_cnt; ix++) {
1295 semid = suptr->un_ent[ix].un_id;
1296 semnum = suptr->un_ent[ix].un_num;
1297 adjval = suptr->un_ent[ix].un_adjval;
1298 seq = suptr->un_ent[ix].un_seq;
1299 semakptr = &sema[semid];
1300 sema_mtxp = &sema_mtx[semid];
1302 mtx_lock(sema_mtxp);
1303 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1304 (semakptr->u.sem_perm.seq != seq)) {
1305 mtx_unlock(sema_mtxp);
1308 if (semnum >= semakptr->u.sem_nsems)
1309 panic("semexit - semnum out of range");
1312 "semexit: %p id=%d num=%d(adj=%d) ; sem=%d\n",
1313 suptr->un_proc, suptr->un_ent[ix].un_id,
1314 suptr->un_ent[ix].un_num,
1315 suptr->un_ent[ix].un_adjval,
1316 semakptr->u.sem_base[semnum].semval));
1318 if (adjval < 0 && semakptr->u.sem_base[semnum].semval <
1320 semakptr->u.sem_base[semnum].semval = 0;
1322 semakptr->u.sem_base[semnum].semval += adjval;
1325 DPRINTF(("semexit: back from wakeup\n"));
1326 mtx_unlock(sema_mtxp);
1332 * Deallocate the undo vector.
1334 DPRINTF(("removing vector\n"));
1335 suptr->un_proc = NULL;
1337 LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
1342 sysctl_sema(SYSCTL_HANDLER_ARGS)
1345 return (SYSCTL_OUT(req, sema,
1346 sizeof(struct semid_kernel) * seminfo.semmni));