xref: /freebsd/sys/kern/sysv_sem.c (revision f5147e312f43a9050468de539aeafa072caa1a60)
1 /*-
2  * Implementation of SVID semaphores
3  *
4  * Author:  Daniel Boulet
5  *
6  * This software is provided ``AS IS'' without any warranties of any kind.
7  */
8 /*-
9  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
10  *
11  * Copyright (c) 2003-2005 McAfee, Inc.
12  * Copyright (c) 2016-2017 Robert N. M. Watson
13  * All rights reserved.
14  *
15  * This software was developed for the FreeBSD Project in part by McAfee
16  * Research, the Security Research Division of McAfee, Inc under DARPA/SPAWAR
17  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS research
18  * program.
19  *
20  * Portions of this software were developed by BAE Systems, the University of
21  * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL
22  * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent
23  * Computing (TC) research program.
24  *
25  * Redistribution and use in source and binary forms, with or without
26  * modification, are permitted provided that the following conditions
27  * are met:
28  * 1. Redistributions of source code must retain the above copyright
29  *    notice, this list of conditions and the following disclaimer.
30  * 2. Redistributions in binary form must reproduce the above copyright
31  *    notice, this list of conditions and the following disclaimer in the
32  *    documentation and/or other materials provided with the distribution.
33  *
34  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
35  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
36  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
37  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
38  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
39  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
40  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
41  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
42  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
43  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
44  * SUCH DAMAGE.
45  */
46 
47 #include <sys/cdefs.h>
48 __FBSDID("$FreeBSD$");
49 
50 #include "opt_compat.h"
51 #include "opt_sysvipc.h"
52 
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/sysproto.h>
56 #include <sys/eventhandler.h>
57 #include <sys/kernel.h>
58 #include <sys/proc.h>
59 #include <sys/lock.h>
60 #include <sys/module.h>
61 #include <sys/mutex.h>
62 #include <sys/racct.h>
63 #include <sys/sem.h>
64 #include <sys/sx.h>
65 #include <sys/syscall.h>
66 #include <sys/syscallsubr.h>
67 #include <sys/sysent.h>
68 #include <sys/sysctl.h>
69 #include <sys/uio.h>
70 #include <sys/malloc.h>
71 #include <sys/jail.h>
72 
73 #include <security/audit/audit.h>
74 #include <security/mac/mac_framework.h>
75 
76 FEATURE(sysv_sem, "System V semaphores support");
77 
78 static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores");
79 
80 #ifdef SEM_DEBUG
81 #define DPRINTF(a)	printf a
82 #else
83 #define DPRINTF(a)
84 #endif
85 
86 static int seminit(void);
87 static int sysvsem_modload(struct module *, int, void *);
88 static int semunload(void);
89 static void semexit_myhook(void *arg, struct proc *p);
90 static int sysctl_sema(SYSCTL_HANDLER_ARGS);
91 static int semvalid(int semid, struct prison *rpr,
92     struct semid_kernel *semakptr);
93 static void sem_remove(int semidx, struct ucred *cred);
94 static struct prison *sem_find_prison(struct ucred *);
95 static int sem_prison_cansee(struct prison *, struct semid_kernel *);
96 static int sem_prison_check(void *, void *);
97 static int sem_prison_set(void *, void *);
98 static int sem_prison_get(void *, void *);
99 static int sem_prison_remove(void *, void *);
100 static void sem_prison_cleanup(struct prison *);
101 
102 #ifndef _SYS_SYSPROTO_H_
103 struct __semctl_args;
104 int __semctl(struct thread *td, struct __semctl_args *uap);
105 struct semget_args;
106 int semget(struct thread *td, struct semget_args *uap);
107 struct semop_args;
108 int semop(struct thread *td, struct semop_args *uap);
109 #endif
110 
111 static struct sem_undo *semu_alloc(struct thread *td);
112 static int semundo_adjust(struct thread *td, struct sem_undo **supptr,
113     int semid, int semseq, int semnum, int adjval);
114 static void semundo_clear(int semid, int semnum);
115 
116 static struct mtx	sem_mtx;	/* semaphore global lock */
117 static struct mtx sem_undo_mtx;
118 static int	semtot = 0;
119 static struct semid_kernel *sema;	/* semaphore id pool */
120 static struct mtx *sema_mtx;	/* semaphore id pool mutexes*/
121 static struct sem *sem;		/* semaphore pool */
122 LIST_HEAD(, sem_undo) semu_list;	/* list of active undo structures */
123 LIST_HEAD(, sem_undo) semu_free_list;	/* list of free undo structures */
124 static int	*semu;		/* undo structure pool */
125 static eventhandler_tag semexit_tag;
126 static unsigned sem_prison_slot;	/* prison OSD slot */
127 
128 #define SEMUNDO_MTX		sem_undo_mtx
129 #define SEMUNDO_LOCK()		mtx_lock(&SEMUNDO_MTX);
130 #define SEMUNDO_UNLOCK()	mtx_unlock(&SEMUNDO_MTX);
131 #define SEMUNDO_LOCKASSERT(how)	mtx_assert(&SEMUNDO_MTX, (how));
132 
133 struct sem {
134 	u_short	semval;		/* semaphore value */
135 	pid_t	sempid;		/* pid of last operation */
136 	u_short	semncnt;	/* # awaiting semval > cval */
137 	u_short	semzcnt;	/* # awaiting semval = 0 */
138 };
139 
140 /*
141  * Undo structure (one per process)
142  */
143 struct sem_undo {
144 	LIST_ENTRY(sem_undo) un_next;	/* ptr to next active undo structure */
145 	struct	proc *un_proc;		/* owner of this structure */
146 	short	un_cnt;			/* # of active entries */
147 	struct undo {
148 		short	un_adjval;	/* adjust on exit values */
149 		short	un_num;		/* semaphore # */
150 		int	un_id;		/* semid */
151 		unsigned short un_seq;
152 	} un_ent[1];			/* undo entries */
153 };
154 
155 /*
156  * Configuration parameters
157  */
158 #ifndef SEMMNI
159 #define SEMMNI	50		/* # of semaphore identifiers */
160 #endif
161 #ifndef SEMMNS
162 #define SEMMNS	340		/* # of semaphores in system */
163 #endif
164 #ifndef SEMUME
165 #define SEMUME	50		/* max # of undo entries per process */
166 #endif
167 #ifndef SEMMNU
168 #define SEMMNU	150		/* # of undo structures in system */
169 #endif
170 
171 /* shouldn't need tuning */
172 #ifndef SEMMSL
173 #define SEMMSL	SEMMNS		/* max # of semaphores per id */
174 #endif
175 #ifndef SEMOPM
176 #define SEMOPM	100		/* max # of operations per semop call */
177 #endif
178 
179 #define SEMVMX	32767		/* semaphore maximum value */
180 #define SEMAEM	16384		/* adjust on exit max value */
181 
182 /*
183  * Due to the way semaphore memory is allocated, we have to ensure that
184  * SEMUSZ is properly aligned.
185  */
186 
187 #define	SEM_ALIGN(bytes) roundup2(bytes, sizeof(long))
188 
189 /* actual size of an undo structure */
190 #define SEMUSZ	SEM_ALIGN(offsetof(struct sem_undo, un_ent[SEMUME]))
191 
192 /*
193  * Macro to find a particular sem_undo vector
194  */
195 #define SEMU(ix) \
196 	((struct sem_undo *)(((intptr_t)semu)+ix * seminfo.semusz))
197 
198 /*
199  * semaphore info struct
200  */
201 struct seminfo seminfo = {
202                 SEMMNI,         /* # of semaphore identifiers */
203                 SEMMNS,         /* # of semaphores in system */
204                 SEMMNU,         /* # of undo structures in system */
205                 SEMMSL,         /* max # of semaphores per id */
206                 SEMOPM,         /* max # of operations per semop call */
207                 SEMUME,         /* max # of undo entries per process */
208                 SEMUSZ,         /* size in bytes of undo structure */
209                 SEMVMX,         /* semaphore maximum value */
210                 SEMAEM          /* adjust on exit max value */
211 };
212 
213 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RDTUN, &seminfo.semmni, 0,
214     "Number of semaphore identifiers");
215 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RDTUN, &seminfo.semmns, 0,
216     "Maximum number of semaphores in the system");
217 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RDTUN, &seminfo.semmnu, 0,
218     "Maximum number of undo structures in the system");
219 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RWTUN, &seminfo.semmsl, 0,
220     "Max semaphores per id");
221 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RDTUN, &seminfo.semopm, 0,
222     "Max operations per semop call");
223 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RDTUN, &seminfo.semume, 0,
224     "Max undo entries per process");
225 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RDTUN, &seminfo.semusz, 0,
226     "Size in bytes of undo structure");
227 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RWTUN, &seminfo.semvmx, 0,
228     "Semaphore maximum value");
229 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RWTUN, &seminfo.semaem, 0,
230     "Adjust on exit max value");
231 SYSCTL_PROC(_kern_ipc, OID_AUTO, sema,
232     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
233     NULL, 0, sysctl_sema, "",
234     "Array of struct semid_kernel for each potential semaphore");
235 
236 static struct syscall_helper_data sem_syscalls[] = {
237 	SYSCALL_INIT_HELPER(__semctl),
238 	SYSCALL_INIT_HELPER(semget),
239 	SYSCALL_INIT_HELPER(semop),
240 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
241     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
242 	SYSCALL_INIT_HELPER(semsys),
243 	SYSCALL_INIT_HELPER_COMPAT(freebsd7___semctl),
244 #endif
245 	SYSCALL_INIT_LAST
246 };
247 
248 #ifdef COMPAT_FREEBSD32
249 #include <compat/freebsd32/freebsd32.h>
250 #include <compat/freebsd32/freebsd32_ipc.h>
251 #include <compat/freebsd32/freebsd32_proto.h>
252 #include <compat/freebsd32/freebsd32_signal.h>
253 #include <compat/freebsd32/freebsd32_syscall.h>
254 #include <compat/freebsd32/freebsd32_util.h>
255 
256 static struct syscall_helper_data sem32_syscalls[] = {
257 	SYSCALL32_INIT_HELPER(freebsd32_semctl),
258 	SYSCALL32_INIT_HELPER_COMPAT(semget),
259 	SYSCALL32_INIT_HELPER_COMPAT(semop),
260 	SYSCALL32_INIT_HELPER(freebsd32_semsys),
261 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
262     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
263 	SYSCALL32_INIT_HELPER(freebsd7_freebsd32_semctl),
264 #endif
265 	SYSCALL_INIT_LAST
266 };
267 #endif
268 
269 static int
270 seminit(void)
271 {
272 	struct prison *pr;
273 	void **rsv;
274 	int i, error;
275 	osd_method_t methods[PR_MAXMETHOD] = {
276 	    [PR_METHOD_CHECK] =		sem_prison_check,
277 	    [PR_METHOD_SET] =		sem_prison_set,
278 	    [PR_METHOD_GET] =		sem_prison_get,
279 	    [PR_METHOD_REMOVE] =	sem_prison_remove,
280 	};
281 
282 	sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
283 	sema = malloc(sizeof(struct semid_kernel) * seminfo.semmni, M_SEM,
284 	    M_WAITOK | M_ZERO);
285 	sema_mtx = malloc(sizeof(struct mtx) * seminfo.semmni, M_SEM,
286 	    M_WAITOK | M_ZERO);
287 	semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
288 
289 	for (i = 0; i < seminfo.semmni; i++) {
290 		sema[i].u.__sem_base = 0;
291 		sema[i].u.sem_perm.mode = 0;
292 		sema[i].u.sem_perm.seq = 0;
293 #ifdef MAC
294 		mac_sysvsem_init(&sema[i]);
295 #endif
296 	}
297 	for (i = 0; i < seminfo.semmni; i++)
298 		mtx_init(&sema_mtx[i], "semid", NULL, MTX_DEF);
299 	LIST_INIT(&semu_free_list);
300 	for (i = 0; i < seminfo.semmnu; i++) {
301 		struct sem_undo *suptr = SEMU(i);
302 		suptr->un_proc = NULL;
303 		LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
304 	}
305 	LIST_INIT(&semu_list);
306 	mtx_init(&sem_mtx, "sem", NULL, MTX_DEF);
307 	mtx_init(&sem_undo_mtx, "semu", NULL, MTX_DEF);
308 	semexit_tag = EVENTHANDLER_REGISTER(process_exit, semexit_myhook, NULL,
309 	    EVENTHANDLER_PRI_ANY);
310 
311 	/* Set current prisons according to their allow.sysvipc. */
312 	sem_prison_slot = osd_jail_register(NULL, methods);
313 	rsv = osd_reserve(sem_prison_slot);
314 	prison_lock(&prison0);
315 	(void)osd_jail_set_reserved(&prison0, sem_prison_slot, rsv, &prison0);
316 	prison_unlock(&prison0);
317 	rsv = NULL;
318 	sx_slock(&allprison_lock);
319 	TAILQ_FOREACH(pr, &allprison, pr_list) {
320 		if (rsv == NULL)
321 			rsv = osd_reserve(sem_prison_slot);
322 		prison_lock(pr);
323 		if ((pr->pr_allow & PR_ALLOW_SYSVIPC) && pr->pr_ref > 0) {
324 			(void)osd_jail_set_reserved(pr, sem_prison_slot, rsv,
325 			    &prison0);
326 			rsv = NULL;
327 		}
328 		prison_unlock(pr);
329 	}
330 	if (rsv != NULL)
331 		osd_free_reserved(rsv);
332 	sx_sunlock(&allprison_lock);
333 
334 	error = syscall_helper_register(sem_syscalls, SY_THR_STATIC_KLD);
335 	if (error != 0)
336 		return (error);
337 #ifdef COMPAT_FREEBSD32
338 	error = syscall32_helper_register(sem32_syscalls, SY_THR_STATIC_KLD);
339 	if (error != 0)
340 		return (error);
341 #endif
342 	return (0);
343 }
344 
345 static int
346 semunload(void)
347 {
348 	int i;
349 
350 	/* XXXKIB */
351 	if (semtot != 0)
352 		return (EBUSY);
353 
354 #ifdef COMPAT_FREEBSD32
355 	syscall32_helper_unregister(sem32_syscalls);
356 #endif
357 	syscall_helper_unregister(sem_syscalls);
358 	EVENTHANDLER_DEREGISTER(process_exit, semexit_tag);
359 	if (sem_prison_slot != 0)
360 		osd_jail_deregister(sem_prison_slot);
361 #ifdef MAC
362 	for (i = 0; i < seminfo.semmni; i++)
363 		mac_sysvsem_destroy(&sema[i]);
364 #endif
365 	free(sem, M_SEM);
366 	free(sema, M_SEM);
367 	free(semu, M_SEM);
368 	for (i = 0; i < seminfo.semmni; i++)
369 		mtx_destroy(&sema_mtx[i]);
370 	free(sema_mtx, M_SEM);
371 	mtx_destroy(&sem_mtx);
372 	mtx_destroy(&sem_undo_mtx);
373 	return (0);
374 }
375 
376 static int
377 sysvsem_modload(struct module *module, int cmd, void *arg)
378 {
379 	int error = 0;
380 
381 	switch (cmd) {
382 	case MOD_LOAD:
383 		error = seminit();
384 		if (error != 0)
385 			semunload();
386 		break;
387 	case MOD_UNLOAD:
388 		error = semunload();
389 		break;
390 	case MOD_SHUTDOWN:
391 		break;
392 	default:
393 		error = EINVAL;
394 		break;
395 	}
396 	return (error);
397 }
398 
399 static moduledata_t sysvsem_mod = {
400 	"sysvsem",
401 	&sysvsem_modload,
402 	NULL
403 };
404 
405 DECLARE_MODULE(sysvsem, sysvsem_mod, SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
406 MODULE_VERSION(sysvsem, 1);
407 
408 /*
409  * Allocate a new sem_undo structure for a process
410  * (returns ptr to structure or NULL if no more room)
411  */
412 
413 static struct sem_undo *
414 semu_alloc(struct thread *td)
415 {
416 	struct sem_undo *suptr;
417 
418 	SEMUNDO_LOCKASSERT(MA_OWNED);
419 	if ((suptr = LIST_FIRST(&semu_free_list)) == NULL)
420 		return (NULL);
421 	LIST_REMOVE(suptr, un_next);
422 	LIST_INSERT_HEAD(&semu_list, suptr, un_next);
423 	suptr->un_cnt = 0;
424 	suptr->un_proc = td->td_proc;
425 	return (suptr);
426 }
427 
428 static int
429 semu_try_free(struct sem_undo *suptr)
430 {
431 
432 	SEMUNDO_LOCKASSERT(MA_OWNED);
433 
434 	if (suptr->un_cnt != 0)
435 		return (0);
436 	LIST_REMOVE(suptr, un_next);
437 	LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
438 	return (1);
439 }
440 
441 /*
442  * Adjust a particular entry for a particular proc
443  */
444 
445 static int
446 semundo_adjust(struct thread *td, struct sem_undo **supptr, int semid,
447     int semseq, int semnum, int adjval)
448 {
449 	struct proc *p = td->td_proc;
450 	struct sem_undo *suptr;
451 	struct undo *sunptr;
452 	int i;
453 
454 	SEMUNDO_LOCKASSERT(MA_OWNED);
455 	/* Look for and remember the sem_undo if the caller doesn't provide
456 	   it */
457 
458 	suptr = *supptr;
459 	if (suptr == NULL) {
460 		LIST_FOREACH(suptr, &semu_list, un_next) {
461 			if (suptr->un_proc == p) {
462 				*supptr = suptr;
463 				break;
464 			}
465 		}
466 		if (suptr == NULL) {
467 			if (adjval == 0)
468 				return(0);
469 			suptr = semu_alloc(td);
470 			if (suptr == NULL)
471 				return (ENOSPC);
472 			*supptr = suptr;
473 		}
474 	}
475 
476 	/*
477 	 * Look for the requested entry and adjust it (delete if adjval becomes
478 	 * 0).
479 	 */
480 	sunptr = &suptr->un_ent[0];
481 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
482 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
483 			continue;
484 		if (adjval != 0) {
485 			adjval += sunptr->un_adjval;
486 			if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
487 				return (ERANGE);
488 		}
489 		sunptr->un_adjval = adjval;
490 		if (sunptr->un_adjval == 0) {
491 			suptr->un_cnt--;
492 			if (i < suptr->un_cnt)
493 				suptr->un_ent[i] =
494 				    suptr->un_ent[suptr->un_cnt];
495 			if (suptr->un_cnt == 0)
496 				semu_try_free(suptr);
497 		}
498 		return (0);
499 	}
500 
501 	/* Didn't find the right entry - create it */
502 	if (adjval == 0)
503 		return (0);
504 	if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
505 		return (ERANGE);
506 	if (suptr->un_cnt != seminfo.semume) {
507 		sunptr = &suptr->un_ent[suptr->un_cnt];
508 		suptr->un_cnt++;
509 		sunptr->un_adjval = adjval;
510 		sunptr->un_id = semid;
511 		sunptr->un_num = semnum;
512 		sunptr->un_seq = semseq;
513 	} else
514 		return (EINVAL);
515 	return (0);
516 }
517 
518 static void
519 semundo_clear(int semid, int semnum)
520 {
521 	struct sem_undo *suptr, *suptr1;
522 	struct undo *sunptr;
523 	int i;
524 
525 	SEMUNDO_LOCKASSERT(MA_OWNED);
526 	LIST_FOREACH_SAFE(suptr, &semu_list, un_next, suptr1) {
527 		sunptr = &suptr->un_ent[0];
528 		for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
529 			if (sunptr->un_id != semid)
530 				continue;
531 			if (semnum == -1 || sunptr->un_num == semnum) {
532 				suptr->un_cnt--;
533 				if (i < suptr->un_cnt) {
534 					suptr->un_ent[i] =
535 					    suptr->un_ent[suptr->un_cnt];
536 					continue;
537 				}
538 				semu_try_free(suptr);
539 			}
540 			if (semnum != -1)
541 				break;
542 		}
543 	}
544 }
545 
546 static int
547 semvalid(int semid, struct prison *rpr, struct semid_kernel *semakptr)
548 {
549 
550 	return ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
551 	    semakptr->u.sem_perm.seq != IPCID_TO_SEQ(semid) ||
552 	    sem_prison_cansee(rpr, semakptr) ? EINVAL : 0);
553 }
554 
555 static void
556 sem_remove(int semidx, struct ucred *cred)
557 {
558 	struct semid_kernel *semakptr;
559 	int i;
560 
561 	KASSERT(semidx >= 0 && semidx < seminfo.semmni,
562 		("semidx out of bounds"));
563 	semakptr = &sema[semidx];
564 	semakptr->u.sem_perm.cuid = cred ? cred->cr_uid : 0;
565 	semakptr->u.sem_perm.uid = cred ? cred->cr_uid : 0;
566 	semakptr->u.sem_perm.mode = 0;
567 	racct_sub_cred(semakptr->cred, RACCT_NSEM, semakptr->u.sem_nsems);
568 	crfree(semakptr->cred);
569 	semakptr->cred = NULL;
570 	SEMUNDO_LOCK();
571 	semundo_clear(semidx, -1);
572 	SEMUNDO_UNLOCK();
573 #ifdef MAC
574 	mac_sysvsem_cleanup(semakptr);
575 #endif
576 	wakeup(semakptr);
577 	for (i = 0; i < seminfo.semmni; i++) {
578 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
579 		    sema[i].u.__sem_base > semakptr->u.__sem_base)
580 			mtx_lock_flags(&sema_mtx[i], LOP_DUPOK);
581 	}
582 	for (i = semakptr->u.__sem_base - sem; i < semtot; i++)
583 		sem[i] = sem[i + semakptr->u.sem_nsems];
584 	for (i = 0; i < seminfo.semmni; i++) {
585 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
586 		    sema[i].u.__sem_base > semakptr->u.__sem_base) {
587 			sema[i].u.__sem_base -= semakptr->u.sem_nsems;
588 			mtx_unlock(&sema_mtx[i]);
589 		}
590 	}
591 	semtot -= semakptr->u.sem_nsems;
592 }
593 
594 static struct prison *
595 sem_find_prison(struct ucred *cred)
596 {
597 	struct prison *pr, *rpr;
598 
599 	pr = cred->cr_prison;
600 	prison_lock(pr);
601 	rpr = osd_jail_get(pr, sem_prison_slot);
602 	prison_unlock(pr);
603 	return rpr;
604 }
605 
606 static int
607 sem_prison_cansee(struct prison *rpr, struct semid_kernel *semakptr)
608 {
609 
610 	if (semakptr->cred == NULL ||
611 	    !(rpr == semakptr->cred->cr_prison ||
612 	      prison_ischild(rpr, semakptr->cred->cr_prison)))
613 		return (EINVAL);
614 	return (0);
615 }
616 
617 /*
618  * Note that the user-mode half of this passes a union, not a pointer.
619  */
620 #ifndef _SYS_SYSPROTO_H_
621 struct __semctl_args {
622 	int	semid;
623 	int	semnum;
624 	int	cmd;
625 	union	semun *arg;
626 };
627 #endif
628 int
629 sys___semctl(struct thread *td, struct __semctl_args *uap)
630 {
631 	struct semid_ds dsbuf;
632 	union semun arg, semun;
633 	register_t rval;
634 	int error;
635 
636 	switch (uap->cmd) {
637 	case SEM_STAT:
638 	case IPC_SET:
639 	case IPC_STAT:
640 	case GETALL:
641 	case SETVAL:
642 	case SETALL:
643 		error = copyin(uap->arg, &arg, sizeof(arg));
644 		if (error)
645 			return (error);
646 		break;
647 	}
648 
649 	switch (uap->cmd) {
650 	case SEM_STAT:
651 	case IPC_STAT:
652 		semun.buf = &dsbuf;
653 		break;
654 	case IPC_SET:
655 		error = copyin(arg.buf, &dsbuf, sizeof(dsbuf));
656 		if (error)
657 			return (error);
658 		semun.buf = &dsbuf;
659 		break;
660 	case GETALL:
661 	case SETALL:
662 		semun.array = arg.array;
663 		break;
664 	case SETVAL:
665 		semun.val = arg.val;
666 		break;
667 	}
668 
669 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
670 	    &rval);
671 	if (error)
672 		return (error);
673 
674 	switch (uap->cmd) {
675 	case SEM_STAT:
676 	case IPC_STAT:
677 		error = copyout(&dsbuf, arg.buf, sizeof(dsbuf));
678 		break;
679 	}
680 
681 	if (error == 0)
682 		td->td_retval[0] = rval;
683 	return (error);
684 }
685 
686 int
687 kern_semctl(struct thread *td, int semid, int semnum, int cmd,
688     union semun *arg, register_t *rval)
689 {
690 	u_short *array;
691 	struct ucred *cred = td->td_ucred;
692 	int i, error;
693 	struct prison *rpr;
694 	struct semid_ds *sbuf;
695 	struct semid_kernel *semakptr;
696 	struct mtx *sema_mtxp;
697 	u_short usval, count;
698 	int semidx;
699 
700 	DPRINTF(("call to semctl(%d, %d, %d, 0x%p)\n",
701 	    semid, semnum, cmd, arg));
702 
703 	AUDIT_ARG_SVIPC_CMD(cmd);
704 	AUDIT_ARG_SVIPC_ID(semid);
705 
706 	rpr = sem_find_prison(td->td_ucred);
707 	if (sem == NULL)
708 		return (ENOSYS);
709 
710 	array = NULL;
711 
712 	switch(cmd) {
713 	case SEM_STAT:
714 		/*
715 		 * For this command we assume semid is an array index
716 		 * rather than an IPC id.
717 		 */
718 		if (semid < 0 || semid >= seminfo.semmni)
719 			return (EINVAL);
720 		semakptr = &sema[semid];
721 		sema_mtxp = &sema_mtx[semid];
722 		mtx_lock(sema_mtxp);
723 		if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
724 			error = EINVAL;
725 			goto done2;
726 		}
727 		if ((error = sem_prison_cansee(rpr, semakptr)))
728 			goto done2;
729 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
730 			goto done2;
731 #ifdef MAC
732 		error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
733 		if (error != 0)
734 			goto done2;
735 #endif
736 		bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
737 		if (cred->cr_prison != semakptr->cred->cr_prison)
738 			arg->buf->sem_perm.key = IPC_PRIVATE;
739 		*rval = IXSEQ_TO_IPCID(semid, semakptr->u.sem_perm);
740 		mtx_unlock(sema_mtxp);
741 		return (0);
742 	}
743 
744 	semidx = IPCID_TO_IX(semid);
745 	if (semidx < 0 || semidx >= seminfo.semmni)
746 		return (EINVAL);
747 
748 	semakptr = &sema[semidx];
749 	sema_mtxp = &sema_mtx[semidx];
750 	if (cmd == IPC_RMID)
751 		mtx_lock(&sem_mtx);
752 	mtx_lock(sema_mtxp);
753 
754 #ifdef MAC
755 	error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
756 	if (error != 0)
757 		goto done2;
758 #endif
759 
760 	error = 0;
761 	*rval = 0;
762 
763 	switch (cmd) {
764 	case IPC_RMID:
765 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
766 			goto done2;
767 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
768 			goto done2;
769 		sem_remove(semidx, cred);
770 		break;
771 
772 	case IPC_SET:
773 		AUDIT_ARG_SVIPC_PERM(&arg->buf->sem_perm);
774 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
775 			goto done2;
776 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
777 			goto done2;
778 		sbuf = arg->buf;
779 		semakptr->u.sem_perm.uid = sbuf->sem_perm.uid;
780 		semakptr->u.sem_perm.gid = sbuf->sem_perm.gid;
781 		semakptr->u.sem_perm.mode = (semakptr->u.sem_perm.mode &
782 		    ~0777) | (sbuf->sem_perm.mode & 0777);
783 		semakptr->u.sem_ctime = time_second;
784 		break;
785 
786 	case IPC_STAT:
787 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
788 			goto done2;
789 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
790 			goto done2;
791 		bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
792 		if (cred->cr_prison != semakptr->cred->cr_prison)
793 			arg->buf->sem_perm.key = IPC_PRIVATE;
794 		break;
795 
796 	case GETNCNT:
797 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
798 			goto done2;
799 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
800 			goto done2;
801 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
802 			error = EINVAL;
803 			goto done2;
804 		}
805 		*rval = semakptr->u.__sem_base[semnum].semncnt;
806 		break;
807 
808 	case GETPID:
809 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
810 			goto done2;
811 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
812 			goto done2;
813 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
814 			error = EINVAL;
815 			goto done2;
816 		}
817 		*rval = semakptr->u.__sem_base[semnum].sempid;
818 		break;
819 
820 	case GETVAL:
821 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
822 			goto done2;
823 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
824 			goto done2;
825 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
826 			error = EINVAL;
827 			goto done2;
828 		}
829 		*rval = semakptr->u.__sem_base[semnum].semval;
830 		break;
831 
832 	case GETALL:
833 		/*
834 		 * Unfortunately, callers of this function don't know
835 		 * in advance how many semaphores are in this set.
836 		 * While we could just allocate the maximum size array
837 		 * and pass the actual size back to the caller, that
838 		 * won't work for SETALL since we can't copyin() more
839 		 * data than the user specified as we may return a
840 		 * spurious EFAULT.
841 		 *
842 		 * Note that the number of semaphores in a set is
843 		 * fixed for the life of that set.  The only way that
844 		 * the 'count' could change while are blocked in
845 		 * malloc() is if this semaphore set were destroyed
846 		 * and a new one created with the same index.
847 		 * However, semvalid() will catch that due to the
848 		 * sequence number unless exactly 0x8000 (or a
849 		 * multiple thereof) semaphore sets for the same index
850 		 * are created and destroyed while we are in malloc!
851 		 *
852 		 */
853 		count = semakptr->u.sem_nsems;
854 		mtx_unlock(sema_mtxp);
855 		array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
856 		mtx_lock(sema_mtxp);
857 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
858 			goto done2;
859 		KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
860 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
861 			goto done2;
862 		for (i = 0; i < semakptr->u.sem_nsems; i++)
863 			array[i] = semakptr->u.__sem_base[i].semval;
864 		mtx_unlock(sema_mtxp);
865 		error = copyout(array, arg->array, count * sizeof(*array));
866 		mtx_lock(sema_mtxp);
867 		break;
868 
869 	case GETZCNT:
870 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
871 			goto done2;
872 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
873 			goto done2;
874 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
875 			error = EINVAL;
876 			goto done2;
877 		}
878 		*rval = semakptr->u.__sem_base[semnum].semzcnt;
879 		break;
880 
881 	case SETVAL:
882 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
883 			goto done2;
884 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
885 			goto done2;
886 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
887 			error = EINVAL;
888 			goto done2;
889 		}
890 		if (arg->val < 0 || arg->val > seminfo.semvmx) {
891 			error = ERANGE;
892 			goto done2;
893 		}
894 		semakptr->u.__sem_base[semnum].semval = arg->val;
895 		SEMUNDO_LOCK();
896 		semundo_clear(semidx, semnum);
897 		SEMUNDO_UNLOCK();
898 		wakeup(semakptr);
899 		break;
900 
901 	case SETALL:
902 		/*
903 		 * See comment on GETALL for why 'count' shouldn't change
904 		 * and why we require a userland buffer.
905 		 */
906 		count = semakptr->u.sem_nsems;
907 		mtx_unlock(sema_mtxp);
908 		array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
909 		error = copyin(arg->array, array, count * sizeof(*array));
910 		mtx_lock(sema_mtxp);
911 		if (error)
912 			break;
913 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
914 			goto done2;
915 		KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
916 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
917 			goto done2;
918 		for (i = 0; i < semakptr->u.sem_nsems; i++) {
919 			usval = array[i];
920 			if (usval > seminfo.semvmx) {
921 				error = ERANGE;
922 				break;
923 			}
924 			semakptr->u.__sem_base[i].semval = usval;
925 		}
926 		SEMUNDO_LOCK();
927 		semundo_clear(semidx, -1);
928 		SEMUNDO_UNLOCK();
929 		wakeup(semakptr);
930 		break;
931 
932 	default:
933 		error = EINVAL;
934 		break;
935 	}
936 
937 done2:
938 	mtx_unlock(sema_mtxp);
939 	if (cmd == IPC_RMID)
940 		mtx_unlock(&sem_mtx);
941 	if (array != NULL)
942 		free(array, M_TEMP);
943 	return(error);
944 }
945 
946 #ifndef _SYS_SYSPROTO_H_
947 struct semget_args {
948 	key_t	key;
949 	int	nsems;
950 	int	semflg;
951 };
952 #endif
953 int
954 sys_semget(struct thread *td, struct semget_args *uap)
955 {
956 	int semid, error = 0;
957 	int key = uap->key;
958 	int nsems = uap->nsems;
959 	int semflg = uap->semflg;
960 	struct ucred *cred = td->td_ucred;
961 
962 	DPRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg));
963 
964 	AUDIT_ARG_VALUE(semflg);
965 
966 	if (sem_find_prison(cred) == NULL)
967 		return (ENOSYS);
968 
969 	mtx_lock(&sem_mtx);
970 	if (key != IPC_PRIVATE) {
971 		for (semid = 0; semid < seminfo.semmni; semid++) {
972 			if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) &&
973 			    sema[semid].cred != NULL &&
974 			    sema[semid].cred->cr_prison == cred->cr_prison &&
975 			    sema[semid].u.sem_perm.key == key)
976 				break;
977 		}
978 		if (semid < seminfo.semmni) {
979 			AUDIT_ARG_SVIPC_ID(semid);
980 			DPRINTF(("found public key\n"));
981 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
982 				DPRINTF(("not exclusive\n"));
983 				error = EEXIST;
984 				goto done2;
985 			}
986 			if ((error = ipcperm(td, &sema[semid].u.sem_perm,
987 			    semflg & 0700))) {
988 				goto done2;
989 			}
990 			if (nsems > 0 && sema[semid].u.sem_nsems < nsems) {
991 				DPRINTF(("too small\n"));
992 				error = EINVAL;
993 				goto done2;
994 			}
995 #ifdef MAC
996 			error = mac_sysvsem_check_semget(cred, &sema[semid]);
997 			if (error != 0)
998 				goto done2;
999 #endif
1000 			goto found;
1001 		}
1002 	}
1003 
1004 	DPRINTF(("need to allocate the semid_kernel\n"));
1005 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
1006 		if (nsems <= 0 || nsems > seminfo.semmsl) {
1007 			DPRINTF(("nsems out of range (0<%d<=%d)\n", nsems,
1008 			    seminfo.semmsl));
1009 			error = EINVAL;
1010 			goto done2;
1011 		}
1012 		if (nsems > seminfo.semmns - semtot) {
1013 			DPRINTF((
1014 			    "not enough semaphores left (need %d, got %d)\n",
1015 			    nsems, seminfo.semmns - semtot));
1016 			error = ENOSPC;
1017 			goto done2;
1018 		}
1019 		for (semid = 0; semid < seminfo.semmni; semid++) {
1020 			if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0)
1021 				break;
1022 		}
1023 		if (semid == seminfo.semmni) {
1024 			DPRINTF(("no more semid_kernel's available\n"));
1025 			error = ENOSPC;
1026 			goto done2;
1027 		}
1028 #ifdef RACCT
1029 		if (racct_enable) {
1030 			PROC_LOCK(td->td_proc);
1031 			error = racct_add(td->td_proc, RACCT_NSEM, nsems);
1032 			PROC_UNLOCK(td->td_proc);
1033 			if (error != 0) {
1034 				error = ENOSPC;
1035 				goto done2;
1036 			}
1037 		}
1038 #endif
1039 		DPRINTF(("semid %d is available\n", semid));
1040 		mtx_lock(&sema_mtx[semid]);
1041 		KASSERT((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0,
1042 		    ("Lost semaphore %d", semid));
1043 		sema[semid].u.sem_perm.key = key;
1044 		sema[semid].u.sem_perm.cuid = cred->cr_uid;
1045 		sema[semid].u.sem_perm.uid = cred->cr_uid;
1046 		sema[semid].u.sem_perm.cgid = cred->cr_gid;
1047 		sema[semid].u.sem_perm.gid = cred->cr_gid;
1048 		sema[semid].u.sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
1049 		sema[semid].cred = crhold(cred);
1050 		sema[semid].u.sem_perm.seq =
1051 		    (sema[semid].u.sem_perm.seq + 1) & 0x7fff;
1052 		sema[semid].u.sem_nsems = nsems;
1053 		sema[semid].u.sem_otime = 0;
1054 		sema[semid].u.sem_ctime = time_second;
1055 		sema[semid].u.__sem_base = &sem[semtot];
1056 		semtot += nsems;
1057 		bzero(sema[semid].u.__sem_base,
1058 		    sizeof(sema[semid].u.__sem_base[0])*nsems);
1059 #ifdef MAC
1060 		mac_sysvsem_create(cred, &sema[semid]);
1061 #endif
1062 		mtx_unlock(&sema_mtx[semid]);
1063 		DPRINTF(("sembase = %p, next = %p\n",
1064 		    sema[semid].u.__sem_base, &sem[semtot]));
1065 	} else {
1066 		DPRINTF(("didn't find it and wasn't asked to create it\n"));
1067 		error = ENOENT;
1068 		goto done2;
1069 	}
1070 
1071 found:
1072 	td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].u.sem_perm);
1073 done2:
1074 	mtx_unlock(&sem_mtx);
1075 	return (error);
1076 }
1077 
1078 #ifndef _SYS_SYSPROTO_H_
1079 struct semop_args {
1080 	int	semid;
1081 	struct	sembuf *sops;
1082 	size_t	nsops;
1083 };
1084 #endif
1085 int
1086 sys_semop(struct thread *td, struct semop_args *uap)
1087 {
1088 #define SMALL_SOPS	8
1089 	struct sembuf small_sops[SMALL_SOPS];
1090 	int semid = uap->semid;
1091 	size_t nsops = uap->nsops;
1092 	struct prison *rpr;
1093 	struct sembuf *sops;
1094 	struct semid_kernel *semakptr;
1095 	struct sembuf *sopptr = NULL;
1096 	struct sem *semptr = NULL;
1097 	struct sem_undo *suptr;
1098 	struct mtx *sema_mtxp;
1099 	size_t i, j, k;
1100 	int error;
1101 	int do_wakeup, do_undos;
1102 	unsigned short seq;
1103 
1104 #ifdef SEM_DEBUG
1105 	sops = NULL;
1106 #endif
1107 	DPRINTF(("call to semop(%d, %p, %u)\n", semid, sops, nsops));
1108 
1109 	AUDIT_ARG_SVIPC_ID(semid);
1110 
1111 	rpr = sem_find_prison(td->td_ucred);
1112 	if (sem == NULL)
1113 		return (ENOSYS);
1114 
1115 	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
1116 
1117 	if (semid < 0 || semid >= seminfo.semmni)
1118 		return (EINVAL);
1119 
1120 	/* Allocate memory for sem_ops */
1121 	if (nsops <= SMALL_SOPS)
1122 		sops = small_sops;
1123 	else if (nsops > seminfo.semopm) {
1124 		DPRINTF(("too many sops (max=%d, nsops=%d)\n", seminfo.semopm,
1125 		    nsops));
1126 		return (E2BIG);
1127 	} else {
1128 #ifdef RACCT
1129 		if (racct_enable) {
1130 			PROC_LOCK(td->td_proc);
1131 			if (nsops >
1132 			    racct_get_available(td->td_proc, RACCT_NSEMOP)) {
1133 				PROC_UNLOCK(td->td_proc);
1134 				return (E2BIG);
1135 			}
1136 			PROC_UNLOCK(td->td_proc);
1137 		}
1138 #endif
1139 
1140 		sops = malloc(nsops * sizeof(*sops), M_TEMP, M_WAITOK);
1141 	}
1142 	if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) {
1143 		DPRINTF(("error = %d from copyin(%p, %p, %d)\n", error,
1144 		    uap->sops, sops, nsops * sizeof(sops[0])));
1145 		if (sops != small_sops)
1146 			free(sops, M_SEM);
1147 		return (error);
1148 	}
1149 
1150 	semakptr = &sema[semid];
1151 	sema_mtxp = &sema_mtx[semid];
1152 	mtx_lock(sema_mtxp);
1153 	if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
1154 		error = EINVAL;
1155 		goto done2;
1156 	}
1157 	seq = semakptr->u.sem_perm.seq;
1158 	if (seq != IPCID_TO_SEQ(uap->semid)) {
1159 		error = EINVAL;
1160 		goto done2;
1161 	}
1162 	if ((error = sem_prison_cansee(rpr, semakptr)) != 0)
1163 		goto done2;
1164 	/*
1165 	 * Initial pass through sops to see what permissions are needed.
1166 	 * Also perform any checks that don't need repeating on each
1167 	 * attempt to satisfy the request vector.
1168 	 */
1169 	j = 0;		/* permission needed */
1170 	do_undos = 0;
1171 	for (i = 0; i < nsops; i++) {
1172 		sopptr = &sops[i];
1173 		if (sopptr->sem_num >= semakptr->u.sem_nsems) {
1174 			error = EFBIG;
1175 			goto done2;
1176 		}
1177 		if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0)
1178 			do_undos = 1;
1179 		j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A;
1180 	}
1181 
1182 	if ((error = ipcperm(td, &semakptr->u.sem_perm, j))) {
1183 		DPRINTF(("error = %d from ipaccess\n", error));
1184 		goto done2;
1185 	}
1186 #ifdef MAC
1187 	error = mac_sysvsem_check_semop(td->td_ucred, semakptr, j);
1188 	if (error != 0)
1189 		goto done2;
1190 #endif
1191 
1192 	/*
1193 	 * Loop trying to satisfy the vector of requests.
1194 	 * If we reach a point where we must wait, any requests already
1195 	 * performed are rolled back and we go to sleep until some other
1196 	 * process wakes us up.  At this point, we start all over again.
1197 	 *
1198 	 * This ensures that from the perspective of other tasks, a set
1199 	 * of requests is atomic (never partially satisfied).
1200 	 */
1201 	for (;;) {
1202 		do_wakeup = 0;
1203 		error = 0;	/* error return if necessary */
1204 
1205 		for (i = 0; i < nsops; i++) {
1206 			sopptr = &sops[i];
1207 			semptr = &semakptr->u.__sem_base[sopptr->sem_num];
1208 
1209 			DPRINTF((
1210 			    "semop:  semakptr=%p, __sem_base=%p, "
1211 			    "semptr=%p, sem[%d]=%d : op=%d, flag=%s\n",
1212 			    semakptr, semakptr->u.__sem_base, semptr,
1213 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
1214 			    (sopptr->sem_flg & IPC_NOWAIT) ?
1215 			    "nowait" : "wait"));
1216 
1217 			if (sopptr->sem_op < 0) {
1218 				if (semptr->semval + sopptr->sem_op < 0) {
1219 					DPRINTF(("semop:  can't do it now\n"));
1220 					break;
1221 				} else {
1222 					semptr->semval += sopptr->sem_op;
1223 					if (semptr->semval == 0 &&
1224 					    semptr->semzcnt > 0)
1225 						do_wakeup = 1;
1226 				}
1227 			} else if (sopptr->sem_op == 0) {
1228 				if (semptr->semval != 0) {
1229 					DPRINTF(("semop:  not zero now\n"));
1230 					break;
1231 				}
1232 			} else if (semptr->semval + sopptr->sem_op >
1233 			    seminfo.semvmx) {
1234 				error = ERANGE;
1235 				break;
1236 			} else {
1237 				if (semptr->semncnt > 0)
1238 					do_wakeup = 1;
1239 				semptr->semval += sopptr->sem_op;
1240 			}
1241 		}
1242 
1243 		/*
1244 		 * Did we get through the entire vector?
1245 		 */
1246 		if (i >= nsops)
1247 			goto done;
1248 
1249 		/*
1250 		 * No ... rollback anything that we've already done
1251 		 */
1252 		DPRINTF(("semop:  rollback 0 through %d\n", i-1));
1253 		for (j = 0; j < i; j++)
1254 			semakptr->u.__sem_base[sops[j].sem_num].semval -=
1255 			    sops[j].sem_op;
1256 
1257 		/* If we detected an error, return it */
1258 		if (error != 0)
1259 			goto done2;
1260 
1261 		/*
1262 		 * If the request that we couldn't satisfy has the
1263 		 * NOWAIT flag set then return with EAGAIN.
1264 		 */
1265 		if (sopptr->sem_flg & IPC_NOWAIT) {
1266 			error = EAGAIN;
1267 			goto done2;
1268 		}
1269 
1270 		if (sopptr->sem_op == 0)
1271 			semptr->semzcnt++;
1272 		else
1273 			semptr->semncnt++;
1274 
1275 		DPRINTF(("semop:  good night!\n"));
1276 		error = msleep(semakptr, sema_mtxp, (PZERO - 4) | PCATCH,
1277 		    "semwait", 0);
1278 		DPRINTF(("semop:  good morning (error=%d)!\n", error));
1279 		/* return code is checked below, after sem[nz]cnt-- */
1280 
1281 		/*
1282 		 * Make sure that the semaphore still exists
1283 		 */
1284 		seq = semakptr->u.sem_perm.seq;
1285 		if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1286 		    seq != IPCID_TO_SEQ(uap->semid)) {
1287 			error = EIDRM;
1288 			goto done2;
1289 		}
1290 
1291 		/*
1292 		 * Renew the semaphore's pointer after wakeup since
1293 		 * during msleep __sem_base may have been modified and semptr
1294 		 * is not valid any more
1295 		 */
1296 		semptr = &semakptr->u.__sem_base[sopptr->sem_num];
1297 
1298 		/*
1299 		 * The semaphore is still alive.  Readjust the count of
1300 		 * waiting processes.
1301 		 */
1302 		if (sopptr->sem_op == 0)
1303 			semptr->semzcnt--;
1304 		else
1305 			semptr->semncnt--;
1306 
1307 		/*
1308 		 * Is it really morning, or was our sleep interrupted?
1309 		 * (Delayed check of msleep() return code because we
1310 		 * need to decrement sem[nz]cnt either way.)
1311 		 */
1312 		if (error != 0) {
1313 			error = EINTR;
1314 			goto done2;
1315 		}
1316 		DPRINTF(("semop:  good morning!\n"));
1317 	}
1318 
1319 done:
1320 	/*
1321 	 * Process any SEM_UNDO requests.
1322 	 */
1323 	if (do_undos) {
1324 		SEMUNDO_LOCK();
1325 		suptr = NULL;
1326 		for (i = 0; i < nsops; i++) {
1327 			/*
1328 			 * We only need to deal with SEM_UNDO's for non-zero
1329 			 * op's.
1330 			 */
1331 			int adjval;
1332 
1333 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
1334 				continue;
1335 			adjval = sops[i].sem_op;
1336 			if (adjval == 0)
1337 				continue;
1338 			error = semundo_adjust(td, &suptr, semid, seq,
1339 			    sops[i].sem_num, -adjval);
1340 			if (error == 0)
1341 				continue;
1342 
1343 			/*
1344 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
1345 			 * Rollback the adjustments to this point and then
1346 			 * rollback the semaphore ups and down so we can return
1347 			 * with an error with all structures restored.  We
1348 			 * rollback the undo's in the exact reverse order that
1349 			 * we applied them.  This guarantees that we won't run
1350 			 * out of space as we roll things back out.
1351 			 */
1352 			for (j = 0; j < i; j++) {
1353 				k = i - j - 1;
1354 				if ((sops[k].sem_flg & SEM_UNDO) == 0)
1355 					continue;
1356 				adjval = sops[k].sem_op;
1357 				if (adjval == 0)
1358 					continue;
1359 				if (semundo_adjust(td, &suptr, semid, seq,
1360 				    sops[k].sem_num, adjval) != 0)
1361 					panic("semop - can't undo undos");
1362 			}
1363 
1364 			for (j = 0; j < nsops; j++)
1365 				semakptr->u.__sem_base[sops[j].sem_num].semval -=
1366 				    sops[j].sem_op;
1367 
1368 			DPRINTF(("error = %d from semundo_adjust\n", error));
1369 			SEMUNDO_UNLOCK();
1370 			goto done2;
1371 		} /* loop through the sops */
1372 		SEMUNDO_UNLOCK();
1373 	} /* if (do_undos) */
1374 
1375 	/* We're definitely done - set the sempid's and time */
1376 	for (i = 0; i < nsops; i++) {
1377 		sopptr = &sops[i];
1378 		semptr = &semakptr->u.__sem_base[sopptr->sem_num];
1379 		semptr->sempid = td->td_proc->p_pid;
1380 	}
1381 	semakptr->u.sem_otime = time_second;
1382 
1383 	/*
1384 	 * Do a wakeup if any semaphore was up'd whilst something was
1385 	 * sleeping on it.
1386 	 */
1387 	if (do_wakeup) {
1388 		DPRINTF(("semop:  doing wakeup\n"));
1389 		wakeup(semakptr);
1390 		DPRINTF(("semop:  back from wakeup\n"));
1391 	}
1392 	DPRINTF(("semop:  done\n"));
1393 	td->td_retval[0] = 0;
1394 done2:
1395 	mtx_unlock(sema_mtxp);
1396 	if (sops != small_sops)
1397 		free(sops, M_SEM);
1398 	return (error);
1399 }
1400 
1401 /*
1402  * Go through the undo structures for this process and apply the adjustments to
1403  * semaphores.
1404  */
1405 static void
1406 semexit_myhook(void *arg, struct proc *p)
1407 {
1408 	struct sem_undo *suptr;
1409 	struct semid_kernel *semakptr;
1410 	struct mtx *sema_mtxp;
1411 	int semid, semnum, adjval, ix;
1412 	unsigned short seq;
1413 
1414 	/*
1415 	 * Go through the chain of undo vectors looking for one
1416 	 * associated with this process.
1417 	 */
1418 	if (LIST_EMPTY(&semu_list))
1419 		return;
1420 	SEMUNDO_LOCK();
1421 	LIST_FOREACH(suptr, &semu_list, un_next) {
1422 		if (suptr->un_proc == p)
1423 			break;
1424 	}
1425 	if (suptr == NULL) {
1426 		SEMUNDO_UNLOCK();
1427 		return;
1428 	}
1429 	LIST_REMOVE(suptr, un_next);
1430 
1431 	DPRINTF(("proc @%p has undo structure with %d entries\n", p,
1432 	    suptr->un_cnt));
1433 
1434 	/*
1435 	 * If there are any active undo elements then process them.
1436 	 */
1437 	if (suptr->un_cnt > 0) {
1438 		SEMUNDO_UNLOCK();
1439 		for (ix = 0; ix < suptr->un_cnt; ix++) {
1440 			semid = suptr->un_ent[ix].un_id;
1441 			semnum = suptr->un_ent[ix].un_num;
1442 			adjval = suptr->un_ent[ix].un_adjval;
1443 			seq = suptr->un_ent[ix].un_seq;
1444 			semakptr = &sema[semid];
1445 			sema_mtxp = &sema_mtx[semid];
1446 
1447 			mtx_lock(sema_mtxp);
1448 			if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1449 			    (semakptr->u.sem_perm.seq != seq)) {
1450 				mtx_unlock(sema_mtxp);
1451 				continue;
1452 			}
1453 			if (semnum >= semakptr->u.sem_nsems)
1454 				panic("semexit - semnum out of range");
1455 
1456 			DPRINTF((
1457 			    "semexit:  %p id=%d num=%d(adj=%d) ; sem=%d\n",
1458 			    suptr->un_proc, suptr->un_ent[ix].un_id,
1459 			    suptr->un_ent[ix].un_num,
1460 			    suptr->un_ent[ix].un_adjval,
1461 			    semakptr->u.__sem_base[semnum].semval));
1462 
1463 			if (adjval < 0 && semakptr->u.__sem_base[semnum].semval <
1464 			    -adjval)
1465 				semakptr->u.__sem_base[semnum].semval = 0;
1466 			else
1467 				semakptr->u.__sem_base[semnum].semval += adjval;
1468 
1469 			wakeup(semakptr);
1470 			DPRINTF(("semexit:  back from wakeup\n"));
1471 			mtx_unlock(sema_mtxp);
1472 		}
1473 		SEMUNDO_LOCK();
1474 	}
1475 
1476 	/*
1477 	 * Deallocate the undo vector.
1478 	 */
1479 	DPRINTF(("removing vector\n"));
1480 	suptr->un_proc = NULL;
1481 	suptr->un_cnt = 0;
1482 	LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
1483 	SEMUNDO_UNLOCK();
1484 }
1485 
1486 static int
1487 sysctl_sema(SYSCTL_HANDLER_ARGS)
1488 {
1489 	struct prison *pr, *rpr;
1490 	struct semid_kernel tsemak;
1491 #ifdef COMPAT_FREEBSD32
1492 	struct semid_kernel32 tsemak32;
1493 #endif
1494 	void *outaddr;
1495 	size_t outsize;
1496 	int error, i;
1497 
1498 	pr = req->td->td_ucred->cr_prison;
1499 	rpr = sem_find_prison(req->td->td_ucred);
1500 	error = 0;
1501 	for (i = 0; i < seminfo.semmni; i++) {
1502 		mtx_lock(&sema_mtx[i]);
1503 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) == 0 ||
1504 		    rpr == NULL || sem_prison_cansee(rpr, &sema[i]) != 0)
1505 			bzero(&tsemak, sizeof(tsemak));
1506 		else {
1507 			tsemak = sema[i];
1508 			if (tsemak.cred->cr_prison != pr)
1509 				tsemak.u.sem_perm.key = IPC_PRIVATE;
1510 		}
1511 		mtx_unlock(&sema_mtx[i]);
1512 #ifdef COMPAT_FREEBSD32
1513 		if (SV_CURPROC_FLAG(SV_ILP32)) {
1514 			bzero(&tsemak32, sizeof(tsemak32));
1515 			freebsd32_ipcperm_out(&tsemak.u.sem_perm,
1516 			    &tsemak32.u.sem_perm);
1517 			/* Don't copy u.__sem_base */
1518 			CP(tsemak, tsemak32, u.sem_nsems);
1519 			CP(tsemak, tsemak32, u.sem_otime);
1520 			CP(tsemak, tsemak32, u.sem_ctime);
1521 			/* Don't copy label or cred */
1522 			outaddr = &tsemak32;
1523 			outsize = sizeof(tsemak32);
1524 		} else
1525 #endif
1526 		{
1527 			tsemak.u.__sem_base = NULL;
1528 			tsemak.label = NULL;
1529 			tsemak.cred = NULL;
1530 			outaddr = &tsemak;
1531 			outsize = sizeof(tsemak);
1532 		}
1533 		error = SYSCTL_OUT(req, outaddr, outsize);
1534 		if (error != 0)
1535 			break;
1536 	}
1537 	return (error);
1538 }
1539 
1540 static int
1541 sem_prison_check(void *obj, void *data)
1542 {
1543 	struct prison *pr = obj;
1544 	struct prison *prpr;
1545 	struct vfsoptlist *opts = data;
1546 	int error, jsys;
1547 
1548 	/*
1549 	 * sysvsem is a jailsys integer.
1550 	 * It must be "disable" if the parent jail is disabled.
1551 	 */
1552 	error = vfs_copyopt(opts, "sysvsem", &jsys, sizeof(jsys));
1553 	if (error != ENOENT) {
1554 		if (error != 0)
1555 			return (error);
1556 		switch (jsys) {
1557 		case JAIL_SYS_DISABLE:
1558 			break;
1559 		case JAIL_SYS_NEW:
1560 		case JAIL_SYS_INHERIT:
1561 			prison_lock(pr->pr_parent);
1562 			prpr = osd_jail_get(pr->pr_parent, sem_prison_slot);
1563 			prison_unlock(pr->pr_parent);
1564 			if (prpr == NULL)
1565 				return (EPERM);
1566 			break;
1567 		default:
1568 			return (EINVAL);
1569 		}
1570 	}
1571 
1572 	return (0);
1573 }
1574 
1575 static int
1576 sem_prison_set(void *obj, void *data)
1577 {
1578 	struct prison *pr = obj;
1579 	struct prison *tpr, *orpr, *nrpr, *trpr;
1580 	struct vfsoptlist *opts = data;
1581 	void *rsv;
1582 	int jsys, descend;
1583 
1584 	/*
1585 	 * sysvsem controls which jail is the root of the associated sems (this
1586 	 * jail or same as the parent), or if the feature is available at all.
1587 	 */
1588 	if (vfs_copyopt(opts, "sysvsem", &jsys, sizeof(jsys)) == ENOENT)
1589 		jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0)
1590 		    ? JAIL_SYS_INHERIT
1591 		    : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0)
1592 		    ? JAIL_SYS_DISABLE
1593 		    : -1;
1594 	if (jsys == JAIL_SYS_DISABLE) {
1595 		prison_lock(pr);
1596 		orpr = osd_jail_get(pr, sem_prison_slot);
1597 		if (orpr != NULL)
1598 			osd_jail_del(pr, sem_prison_slot);
1599 		prison_unlock(pr);
1600 		if (orpr != NULL) {
1601 			if (orpr == pr)
1602 				sem_prison_cleanup(pr);
1603 			/* Disable all child jails as well. */
1604 			FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1605 				prison_lock(tpr);
1606 				trpr = osd_jail_get(tpr, sem_prison_slot);
1607 				if (trpr != NULL) {
1608 					osd_jail_del(tpr, sem_prison_slot);
1609 					prison_unlock(tpr);
1610 					if (trpr == tpr)
1611 						sem_prison_cleanup(tpr);
1612 				} else {
1613 					prison_unlock(tpr);
1614 					descend = 0;
1615 				}
1616 			}
1617 		}
1618 	} else if (jsys != -1) {
1619 		if (jsys == JAIL_SYS_NEW)
1620 			nrpr = pr;
1621 		else {
1622 			prison_lock(pr->pr_parent);
1623 			nrpr = osd_jail_get(pr->pr_parent, sem_prison_slot);
1624 			prison_unlock(pr->pr_parent);
1625 		}
1626 		rsv = osd_reserve(sem_prison_slot);
1627 		prison_lock(pr);
1628 		orpr = osd_jail_get(pr, sem_prison_slot);
1629 		if (orpr != nrpr)
1630 			(void)osd_jail_set_reserved(pr, sem_prison_slot, rsv,
1631 			    nrpr);
1632 		else
1633 			osd_free_reserved(rsv);
1634 		prison_unlock(pr);
1635 		if (orpr != nrpr) {
1636 			if (orpr == pr)
1637 				sem_prison_cleanup(pr);
1638 			if (orpr != NULL) {
1639 				/* Change child jails matching the old root, */
1640 				FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1641 					prison_lock(tpr);
1642 					trpr = osd_jail_get(tpr,
1643 					    sem_prison_slot);
1644 					if (trpr == orpr) {
1645 						(void)osd_jail_set(tpr,
1646 						    sem_prison_slot, nrpr);
1647 						prison_unlock(tpr);
1648 						if (trpr == tpr)
1649 							sem_prison_cleanup(tpr);
1650 					} else {
1651 						prison_unlock(tpr);
1652 						descend = 0;
1653 					}
1654 				}
1655 			}
1656 		}
1657 	}
1658 
1659 	return (0);
1660 }
1661 
1662 static int
1663 sem_prison_get(void *obj, void *data)
1664 {
1665 	struct prison *pr = obj;
1666 	struct prison *rpr;
1667 	struct vfsoptlist *opts = data;
1668 	int error, jsys;
1669 
1670 	/* Set sysvsem based on the jail's root prison. */
1671 	prison_lock(pr);
1672 	rpr = osd_jail_get(pr, sem_prison_slot);
1673 	prison_unlock(pr);
1674 	jsys = rpr == NULL ? JAIL_SYS_DISABLE
1675 	    : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT;
1676 	error = vfs_setopt(opts, "sysvsem", &jsys, sizeof(jsys));
1677 	if (error == ENOENT)
1678 		error = 0;
1679 	return (error);
1680 }
1681 
1682 static int
1683 sem_prison_remove(void *obj, void *data __unused)
1684 {
1685 	struct prison *pr = obj;
1686 	struct prison *rpr;
1687 
1688 	prison_lock(pr);
1689 	rpr = osd_jail_get(pr, sem_prison_slot);
1690 	prison_unlock(pr);
1691 	if (rpr == pr)
1692 		sem_prison_cleanup(pr);
1693 	return (0);
1694 }
1695 
1696 static void
1697 sem_prison_cleanup(struct prison *pr)
1698 {
1699 	int i;
1700 
1701 	/* Remove any sems that belong to this jail. */
1702 	mtx_lock(&sem_mtx);
1703 	for (i = 0; i < seminfo.semmni; i++) {
1704 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
1705 		    sema[i].cred != NULL && sema[i].cred->cr_prison == pr) {
1706 			mtx_lock(&sema_mtx[i]);
1707 			sem_remove(i, NULL);
1708 			mtx_unlock(&sema_mtx[i]);
1709 		}
1710 	}
1711 	mtx_unlock(&sem_mtx);
1712 }
1713 
1714 SYSCTL_JAIL_PARAM_SYS_NODE(sysvsem, CTLFLAG_RW, "SYSV semaphores");
1715 
1716 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1717     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1718 
1719 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1720 static sy_call_t *semcalls[] = {
1721 	(sy_call_t *)freebsd7___semctl, (sy_call_t *)sys_semget,
1722 	(sy_call_t *)sys_semop
1723 };
1724 
1725 /*
1726  * Entry point for all SEM calls.
1727  */
1728 int
1729 sys_semsys(td, uap)
1730 	struct thread *td;
1731 	/* XXX actually varargs. */
1732 	struct semsys_args /* {
1733 		int	which;
1734 		int	a2;
1735 		int	a3;
1736 		int	a4;
1737 		int	a5;
1738 	} */ *uap;
1739 {
1740 	int error;
1741 
1742 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1743 	if (uap->which < 0 || uap->which >= nitems(semcalls))
1744 		return (EINVAL);
1745 	error = (*semcalls[uap->which])(td, &uap->a2);
1746 	return (error);
1747 }
1748 
1749 #ifndef CP
1750 #define CP(src, dst, fld)	do { (dst).fld = (src).fld; } while (0)
1751 #endif
1752 
1753 #ifndef _SYS_SYSPROTO_H_
1754 struct freebsd7___semctl_args {
1755 	int	semid;
1756 	int	semnum;
1757 	int	cmd;
1758 	union	semun_old *arg;
1759 };
1760 #endif
1761 int
1762 freebsd7___semctl(struct thread *td, struct freebsd7___semctl_args *uap)
1763 {
1764 	struct semid_ds_old dsold;
1765 	struct semid_ds dsbuf;
1766 	union semun_old arg;
1767 	union semun semun;
1768 	register_t rval;
1769 	int error;
1770 
1771 	switch (uap->cmd) {
1772 	case SEM_STAT:
1773 	case IPC_SET:
1774 	case IPC_STAT:
1775 	case GETALL:
1776 	case SETVAL:
1777 	case SETALL:
1778 		error = copyin(uap->arg, &arg, sizeof(arg));
1779 		if (error)
1780 			return (error);
1781 		break;
1782 	}
1783 
1784 	switch (uap->cmd) {
1785 	case SEM_STAT:
1786 	case IPC_STAT:
1787 		semun.buf = &dsbuf;
1788 		break;
1789 	case IPC_SET:
1790 		error = copyin(arg.buf, &dsold, sizeof(dsold));
1791 		if (error)
1792 			return (error);
1793 		ipcperm_old2new(&dsold.sem_perm, &dsbuf.sem_perm);
1794 		CP(dsold, dsbuf, __sem_base);
1795 		CP(dsold, dsbuf, sem_nsems);
1796 		CP(dsold, dsbuf, sem_otime);
1797 		CP(dsold, dsbuf, sem_ctime);
1798 		semun.buf = &dsbuf;
1799 		break;
1800 	case GETALL:
1801 	case SETALL:
1802 		semun.array = arg.array;
1803 		break;
1804 	case SETVAL:
1805 		semun.val = arg.val;
1806 		break;
1807 	}
1808 
1809 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1810 	    &rval);
1811 	if (error)
1812 		return (error);
1813 
1814 	switch (uap->cmd) {
1815 	case SEM_STAT:
1816 	case IPC_STAT:
1817 		bzero(&dsold, sizeof(dsold));
1818 		ipcperm_new2old(&dsbuf.sem_perm, &dsold.sem_perm);
1819 		CP(dsbuf, dsold, __sem_base);
1820 		CP(dsbuf, dsold, sem_nsems);
1821 		CP(dsbuf, dsold, sem_otime);
1822 		CP(dsbuf, dsold, sem_ctime);
1823 		error = copyout(&dsold, arg.buf, sizeof(dsold));
1824 		break;
1825 	}
1826 
1827 	if (error == 0)
1828 		td->td_retval[0] = rval;
1829 	return (error);
1830 }
1831 
1832 #endif /* COMPAT_FREEBSD{4,5,6,7} */
1833 
1834 #ifdef COMPAT_FREEBSD32
1835 
1836 int
1837 freebsd32_semsys(struct thread *td, struct freebsd32_semsys_args *uap)
1838 {
1839 
1840 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1841     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1842 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1843 	switch (uap->which) {
1844 	case 0:
1845 		return (freebsd7_freebsd32_semctl(td,
1846 		    (struct freebsd7_freebsd32_semctl_args *)&uap->a2));
1847 	default:
1848 		return (sys_semsys(td, (struct semsys_args *)uap));
1849 	}
1850 #else
1851 	return (nosys(td, NULL));
1852 #endif
1853 }
1854 
1855 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1856     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1857 int
1858 freebsd7_freebsd32_semctl(struct thread *td,
1859     struct freebsd7_freebsd32_semctl_args *uap)
1860 {
1861 	struct semid_ds32_old dsbuf32;
1862 	struct semid_ds dsbuf;
1863 	union semun semun;
1864 	union semun32 arg;
1865 	register_t rval;
1866 	int error;
1867 
1868 	switch (uap->cmd) {
1869 	case SEM_STAT:
1870 	case IPC_SET:
1871 	case IPC_STAT:
1872 	case GETALL:
1873 	case SETVAL:
1874 	case SETALL:
1875 		error = copyin(uap->arg, &arg, sizeof(arg));
1876 		if (error)
1877 			return (error);
1878 		break;
1879 	}
1880 
1881 	switch (uap->cmd) {
1882 	case SEM_STAT:
1883 	case IPC_STAT:
1884 		semun.buf = &dsbuf;
1885 		break;
1886 	case IPC_SET:
1887 		error = copyin(PTRIN(arg.buf), &dsbuf32, sizeof(dsbuf32));
1888 		if (error)
1889 			return (error);
1890 		freebsd32_ipcperm_old_in(&dsbuf32.sem_perm, &dsbuf.sem_perm);
1891 		PTRIN_CP(dsbuf32, dsbuf, __sem_base);
1892 		CP(dsbuf32, dsbuf, sem_nsems);
1893 		CP(dsbuf32, dsbuf, sem_otime);
1894 		CP(dsbuf32, dsbuf, sem_ctime);
1895 		semun.buf = &dsbuf;
1896 		break;
1897 	case GETALL:
1898 	case SETALL:
1899 		semun.array = PTRIN(arg.array);
1900 		break;
1901 	case SETVAL:
1902 		semun.val = arg.val;
1903 		break;
1904 	}
1905 
1906 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1907 	    &rval);
1908 	if (error)
1909 		return (error);
1910 
1911 	switch (uap->cmd) {
1912 	case SEM_STAT:
1913 	case IPC_STAT:
1914 		bzero(&dsbuf32, sizeof(dsbuf32));
1915 		freebsd32_ipcperm_old_out(&dsbuf.sem_perm, &dsbuf32.sem_perm);
1916 		PTROUT_CP(dsbuf, dsbuf32, __sem_base);
1917 		CP(dsbuf, dsbuf32, sem_nsems);
1918 		CP(dsbuf, dsbuf32, sem_otime);
1919 		CP(dsbuf, dsbuf32, sem_ctime);
1920 		error = copyout(&dsbuf32, PTRIN(arg.buf), sizeof(dsbuf32));
1921 		break;
1922 	}
1923 
1924 	if (error == 0)
1925 		td->td_retval[0] = rval;
1926 	return (error);
1927 }
1928 #endif
1929 
1930 int
1931 freebsd32_semctl(struct thread *td, struct freebsd32_semctl_args *uap)
1932 {
1933 	struct semid_ds32 dsbuf32;
1934 	struct semid_ds dsbuf;
1935 	union semun semun;
1936 	union semun32 arg;
1937 	register_t rval;
1938 	int error;
1939 
1940 	switch (uap->cmd) {
1941 	case SEM_STAT:
1942 	case IPC_SET:
1943 	case IPC_STAT:
1944 	case GETALL:
1945 	case SETVAL:
1946 	case SETALL:
1947 		error = copyin(uap->arg, &arg, sizeof(arg));
1948 		if (error)
1949 			return (error);
1950 		break;
1951 	}
1952 
1953 	switch (uap->cmd) {
1954 	case SEM_STAT:
1955 	case IPC_STAT:
1956 		semun.buf = &dsbuf;
1957 		break;
1958 	case IPC_SET:
1959 		error = copyin(PTRIN(arg.buf), &dsbuf32, sizeof(dsbuf32));
1960 		if (error)
1961 			return (error);
1962 		freebsd32_ipcperm_in(&dsbuf32.sem_perm, &dsbuf.sem_perm);
1963 		PTRIN_CP(dsbuf32, dsbuf, __sem_base);
1964 		CP(dsbuf32, dsbuf, sem_nsems);
1965 		CP(dsbuf32, dsbuf, sem_otime);
1966 		CP(dsbuf32, dsbuf, sem_ctime);
1967 		semun.buf = &dsbuf;
1968 		break;
1969 	case GETALL:
1970 	case SETALL:
1971 		semun.array = PTRIN(arg.array);
1972 		break;
1973 	case SETVAL:
1974 		semun.val = arg.val;
1975 		break;
1976 	}
1977 
1978 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1979 	    &rval);
1980 	if (error)
1981 		return (error);
1982 
1983 	switch (uap->cmd) {
1984 	case SEM_STAT:
1985 	case IPC_STAT:
1986 		bzero(&dsbuf32, sizeof(dsbuf32));
1987 		freebsd32_ipcperm_out(&dsbuf.sem_perm, &dsbuf32.sem_perm);
1988 		PTROUT_CP(dsbuf, dsbuf32, __sem_base);
1989 		CP(dsbuf, dsbuf32, sem_nsems);
1990 		CP(dsbuf, dsbuf32, sem_otime);
1991 		CP(dsbuf, dsbuf32, sem_ctime);
1992 		error = copyout(&dsbuf32, PTRIN(arg.buf), sizeof(dsbuf32));
1993 		break;
1994 	}
1995 
1996 	if (error == 0)
1997 		td->td_retval[0] = rval;
1998 	return (error);
1999 }
2000 
2001 #endif /* COMPAT_FREEBSD32 */
2002