xref: /freebsd/sys/kern/sysv_sem.c (revision 7fdf597e96a02165cfe22ff357b857d5fa15ed8a)
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
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 "opt_sysvipc.h"
48 
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/sysproto.h>
52 #include <sys/abi_compat.h>
53 #include <sys/eventhandler.h>
54 #include <sys/kernel.h>
55 #include <sys/proc.h>
56 #include <sys/lock.h>
57 #include <sys/module.h>
58 #include <sys/mutex.h>
59 #include <sys/racct.h>
60 #include <sys/sem.h>
61 #include <sys/sx.h>
62 #include <sys/syscall.h>
63 #include <sys/syscallsubr.h>
64 #include <sys/sysent.h>
65 #include <sys/sysctl.h>
66 #include <sys/uio.h>
67 #include <sys/malloc.h>
68 #include <sys/jail.h>
69 
70 #include <security/audit/audit.h>
71 #include <security/mac/mac_framework.h>
72 
73 FEATURE(sysv_sem, "System V semaphores support");
74 
75 static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores");
76 
77 #ifdef SEM_DEBUG
78 #define DPRINTF(a)	printf a
79 #else
80 #define DPRINTF(a)
81 #endif
82 
83 static int seminit(void);
84 static int sysvsem_modload(struct module *, int, void *);
85 static int semunload(void);
86 static void semexit_myhook(void *arg, struct proc *p);
87 static int sysctl_sema(SYSCTL_HANDLER_ARGS);
88 static int semvalid(int semid, struct prison *rpr,
89     struct semid_kernel *semakptr);
90 static void sem_remove(int semidx, struct ucred *cred);
91 static struct prison *sem_find_prison(struct ucred *);
92 static int sem_prison_cansee(struct prison *, struct semid_kernel *);
93 static int sem_prison_check(void *, void *);
94 static int sem_prison_set(void *, void *);
95 static int sem_prison_get(void *, void *);
96 static int sem_prison_remove(void *, void *);
97 static void sem_prison_cleanup(struct prison *);
98 
99 #ifndef _SYS_SYSPROTO_H_
100 struct __semctl_args;
101 int __semctl(struct thread *td, struct __semctl_args *uap);
102 struct semget_args;
103 int semget(struct thread *td, struct semget_args *uap);
104 struct semop_args;
105 int semop(struct thread *td, struct semop_args *uap);
106 #endif
107 
108 static struct sem_undo *semu_alloc(struct thread *td);
109 static int semundo_adjust(struct thread *td, struct sem_undo **supptr,
110     int semid, int semseq, int semnum, int adjval);
111 static void semundo_clear(int semid, int semnum);
112 
113 static struct mtx	sem_mtx;	/* semaphore global lock */
114 static struct mtx sem_undo_mtx;
115 static int	semtot = 0;
116 static struct semid_kernel *sema;	/* semaphore id pool */
117 static struct mtx *sema_mtx;	/* semaphore id pool mutexes*/
118 static struct sem *sem;		/* semaphore pool */
119 LIST_HEAD(, sem_undo) semu_list;	/* list of active undo structures */
120 LIST_HEAD(, sem_undo) semu_free_list;	/* list of free undo structures */
121 static int	*semu;		/* undo structure pool */
122 static eventhandler_tag semexit_tag;
123 static unsigned sem_prison_slot;	/* prison OSD slot */
124 
125 #define SEMUNDO_MTX		sem_undo_mtx
126 #define SEMUNDO_LOCK()		mtx_lock(&SEMUNDO_MTX);
127 #define SEMUNDO_UNLOCK()	mtx_unlock(&SEMUNDO_MTX);
128 #define SEMUNDO_LOCKASSERT(how)	mtx_assert(&SEMUNDO_MTX, (how));
129 
130 struct sem {
131 	u_short	semval;		/* semaphore value */
132 	pid_t	sempid;		/* pid of last operation */
133 	u_short	semncnt;	/* # awaiting semval > cval */
134 	u_short	semzcnt;	/* # awaiting semval = 0 */
135 };
136 
137 /*
138  * Undo structure (one per process)
139  */
140 struct sem_undo {
141 	LIST_ENTRY(sem_undo) un_next;	/* ptr to next active undo structure */
142 	struct	proc *un_proc;		/* owner of this structure */
143 	short	un_cnt;			/* # of active entries */
144 	struct undo {
145 		short	un_adjval;	/* adjust on exit values */
146 		short	un_num;		/* semaphore # */
147 		int	un_id;		/* semid */
148 		unsigned short un_seq;
149 	} un_ent[1];			/* undo entries */
150 };
151 
152 /*
153  * Configuration parameters
154  */
155 #ifndef SEMMNI
156 #define SEMMNI	50		/* # of semaphore identifiers */
157 #endif
158 #ifndef SEMMNS
159 #define SEMMNS	340		/* # of semaphores in system */
160 #endif
161 #ifndef SEMUME
162 #define SEMUME	50		/* max # of undo entries per process */
163 #endif
164 #ifndef SEMMNU
165 #define SEMMNU	150		/* # of undo structures in system */
166 #endif
167 
168 /* shouldn't need tuning */
169 #ifndef SEMMSL
170 #define SEMMSL	SEMMNS		/* max # of semaphores per id */
171 #endif
172 #ifndef SEMOPM
173 #define SEMOPM	100		/* max # of operations per semop call */
174 #endif
175 
176 #define SEMVMX	32767		/* semaphore maximum value */
177 #define SEMAEM	16384		/* adjust on exit max value */
178 
179 /*
180  * Due to the way semaphore memory is allocated, we have to ensure that
181  * SEMUSZ is properly aligned.
182  */
183 
184 #define	SEM_ALIGN(bytes) roundup2(bytes, sizeof(long))
185 
186 /* actual size of an undo structure */
187 #define SEMUSZ(x)	SEM_ALIGN(offsetof(struct sem_undo, un_ent[(x)]))
188 
189 /*
190  * Macro to find a particular sem_undo vector
191  */
192 #define SEMU(ix) \
193 	((struct sem_undo *)(((intptr_t)semu) + (ix) * seminfo.semusz))
194 
195 /*
196  * semaphore info struct
197  */
198 struct seminfo seminfo = {
199 	.semmni =	SEMMNI,	/* # of semaphore identifiers */
200 	.semmns =	SEMMNS,	/* # of semaphores in system */
201 	.semmnu =	SEMMNU,	/* # of undo structures in system */
202 	.semmsl =	SEMMSL,	/* max # of semaphores per id */
203 	.semopm =	SEMOPM,	/* max # of operations per semop call */
204 	.semume =	SEMUME,	/* max # of undo entries per process */
205 	.semusz =	SEMUSZ(SEMUME),	/* size in bytes of undo structure */
206 	.semvmx =	SEMVMX,	/* semaphore maximum value */
207 	.semaem =	SEMAEM,	/* adjust on exit max value */
208 };
209 
210 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RDTUN, &seminfo.semmni, 0,
211     "Number of semaphore identifiers");
212 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RDTUN, &seminfo.semmns, 0,
213     "Maximum number of semaphores in the system");
214 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RDTUN, &seminfo.semmnu, 0,
215     "Maximum number of undo structures in the system");
216 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RWTUN, &seminfo.semmsl, 0,
217     "Max semaphores per id");
218 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RDTUN, &seminfo.semopm, 0,
219     "Max operations per semop call");
220 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RDTUN, &seminfo.semume, 0,
221     "Max undo entries per process");
222 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RD, &seminfo.semusz, 0,
223     "Size in bytes of undo structure");
224 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RWTUN, &seminfo.semvmx, 0,
225     "Semaphore maximum value");
226 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RWTUN, &seminfo.semaem, 0,
227     "Adjust on exit max value");
228 SYSCTL_PROC(_kern_ipc, OID_AUTO, sema,
229     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
230     NULL, 0, sysctl_sema, "",
231     "Array of struct semid_kernel for each potential semaphore");
232 
233 static struct syscall_helper_data sem_syscalls[] = {
234 	SYSCALL_INIT_HELPER(__semctl),
235 	SYSCALL_INIT_HELPER(semget),
236 	SYSCALL_INIT_HELPER(semop),
237 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
238     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
239 	SYSCALL_INIT_HELPER(semsys),
240 	SYSCALL_INIT_HELPER_COMPAT(freebsd7___semctl),
241 #endif
242 	SYSCALL_INIT_LAST
243 };
244 
245 #ifdef COMPAT_FREEBSD32
246 #include <compat/freebsd32/freebsd32.h>
247 #include <compat/freebsd32/freebsd32_ipc.h>
248 #include <compat/freebsd32/freebsd32_proto.h>
249 #include <compat/freebsd32/freebsd32_signal.h>
250 #include <compat/freebsd32/freebsd32_syscall.h>
251 #include <compat/freebsd32/freebsd32_util.h>
252 
253 static struct syscall_helper_data sem32_syscalls[] = {
254 	SYSCALL32_INIT_HELPER(freebsd32___semctl),
255 	SYSCALL32_INIT_HELPER_COMPAT(semget),
256 	SYSCALL32_INIT_HELPER_COMPAT(semop),
257 	SYSCALL32_INIT_HELPER(freebsd32_semsys),
258 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
259     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
260 	SYSCALL32_INIT_HELPER(freebsd7_freebsd32___semctl),
261 #endif
262 	SYSCALL_INIT_LAST
263 };
264 #endif
265 
266 static int
267 seminit(void)
268 {
269 	struct prison *pr;
270 	void **rsv;
271 	int i, error;
272 	osd_method_t methods[PR_MAXMETHOD] = {
273 	    [PR_METHOD_CHECK] =		sem_prison_check,
274 	    [PR_METHOD_SET] =		sem_prison_set,
275 	    [PR_METHOD_GET] =		sem_prison_get,
276 	    [PR_METHOD_REMOVE] =	sem_prison_remove,
277 	};
278 
279 	sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
280 	sema = malloc(sizeof(struct semid_kernel) * seminfo.semmni, M_SEM,
281 	    M_WAITOK | M_ZERO);
282 	sema_mtx = malloc(sizeof(struct mtx) * seminfo.semmni, M_SEM,
283 	    M_WAITOK | M_ZERO);
284 	seminfo.semusz = SEMUSZ(seminfo.semume);
285 	semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
286 
287 	for (i = 0; i < seminfo.semmni; i++) {
288 		sema[i].u.__sem_base = 0;
289 		sema[i].u.sem_perm.mode = 0;
290 		sema[i].u.sem_perm.seq = 0;
291 #ifdef MAC
292 		mac_sysvsem_init(&sema[i]);
293 #endif
294 	}
295 	for (i = 0; i < seminfo.semmni; i++)
296 		mtx_init(&sema_mtx[i], "semid", NULL, MTX_DEF);
297 	LIST_INIT(&semu_free_list);
298 	for (i = 0; i < seminfo.semmnu; i++) {
299 		struct sem_undo *suptr = SEMU(i);
300 		suptr->un_proc = NULL;
301 		LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
302 	}
303 	LIST_INIT(&semu_list);
304 	mtx_init(&sem_mtx, "sem", NULL, MTX_DEF);
305 	mtx_init(&sem_undo_mtx, "semu", NULL, MTX_DEF);
306 	semexit_tag = EVENTHANDLER_REGISTER(process_exit, semexit_myhook, NULL,
307 	    EVENTHANDLER_PRI_ANY);
308 
309 	/* Set current prisons according to their allow.sysvipc. */
310 	sem_prison_slot = osd_jail_register(NULL, methods);
311 	rsv = osd_reserve(sem_prison_slot);
312 	prison_lock(&prison0);
313 	(void)osd_jail_set_reserved(&prison0, sem_prison_slot, rsv, &prison0);
314 	prison_unlock(&prison0);
315 	rsv = NULL;
316 	sx_slock(&allprison_lock);
317 	TAILQ_FOREACH(pr, &allprison, pr_list) {
318 		if (rsv == NULL)
319 			rsv = osd_reserve(sem_prison_slot);
320 		prison_lock(pr);
321 		if (pr->pr_allow & PR_ALLOW_SYSVIPC) {
322 			(void)osd_jail_set_reserved(pr, sem_prison_slot, rsv,
323 			    &prison0);
324 			rsv = NULL;
325 		}
326 		prison_unlock(pr);
327 	}
328 	if (rsv != NULL)
329 		osd_free_reserved(rsv);
330 	sx_sunlock(&allprison_lock);
331 
332 	error = syscall_helper_register(sem_syscalls, SY_THR_STATIC_KLD);
333 	if (error != 0)
334 		return (error);
335 #ifdef COMPAT_FREEBSD32
336 	error = syscall32_helper_register(sem32_syscalls, SY_THR_STATIC_KLD);
337 	if (error != 0)
338 		return (error);
339 #endif
340 	return (0);
341 }
342 
343 static int
344 semunload(void)
345 {
346 	int i;
347 
348 	/* XXXKIB */
349 	if (semtot != 0)
350 		return (EBUSY);
351 
352 #ifdef COMPAT_FREEBSD32
353 	syscall32_helper_unregister(sem32_syscalls);
354 #endif
355 	syscall_helper_unregister(sem_syscalls);
356 	EVENTHANDLER_DEREGISTER(process_exit, semexit_tag);
357 	if (sem_prison_slot != 0)
358 		osd_jail_deregister(sem_prison_slot);
359 #ifdef MAC
360 	for (i = 0; i < seminfo.semmni; i++)
361 		mac_sysvsem_destroy(&sema[i]);
362 #endif
363 	free(sem, M_SEM);
364 	free(sema, M_SEM);
365 	free(semu, M_SEM);
366 	for (i = 0; i < seminfo.semmni; i++)
367 		mtx_destroy(&sema_mtx[i]);
368 	free(sema_mtx, M_SEM);
369 	mtx_destroy(&sem_mtx);
370 	mtx_destroy(&sem_undo_mtx);
371 	return (0);
372 }
373 
374 static int
375 sysvsem_modload(struct module *module, int cmd, void *arg)
376 {
377 	int error = 0;
378 
379 	switch (cmd) {
380 	case MOD_LOAD:
381 		error = seminit();
382 		break;
383 	case MOD_UNLOAD:
384 		error = semunload();
385 		break;
386 	case MOD_SHUTDOWN:
387 		break;
388 	default:
389 		error = EINVAL;
390 		break;
391 	}
392 	return (error);
393 }
394 
395 static moduledata_t sysvsem_mod = {
396 	"sysvsem",
397 	&sysvsem_modload,
398 	NULL
399 };
400 
401 DECLARE_MODULE(sysvsem, sysvsem_mod, SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
402 MODULE_VERSION(sysvsem, 1);
403 
404 /*
405  * Allocate a new sem_undo structure for a process
406  * (returns ptr to structure or NULL if no more room)
407  */
408 
409 static struct sem_undo *
410 semu_alloc(struct thread *td)
411 {
412 	struct sem_undo *suptr;
413 
414 	SEMUNDO_LOCKASSERT(MA_OWNED);
415 	if ((suptr = LIST_FIRST(&semu_free_list)) == NULL)
416 		return (NULL);
417 	LIST_REMOVE(suptr, un_next);
418 	LIST_INSERT_HEAD(&semu_list, suptr, un_next);
419 	suptr->un_cnt = 0;
420 	suptr->un_proc = td->td_proc;
421 	return (suptr);
422 }
423 
424 static int
425 semu_try_free(struct sem_undo *suptr)
426 {
427 
428 	SEMUNDO_LOCKASSERT(MA_OWNED);
429 
430 	if (suptr->un_cnt != 0)
431 		return (0);
432 	LIST_REMOVE(suptr, un_next);
433 	LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
434 	return (1);
435 }
436 
437 /*
438  * Adjust a particular entry for a particular proc
439  */
440 
441 static int
442 semundo_adjust(struct thread *td, struct sem_undo **supptr, int semid,
443     int semseq, int semnum, int adjval)
444 {
445 	struct proc *p = td->td_proc;
446 	struct sem_undo *suptr;
447 	struct undo *sunptr;
448 	int i;
449 
450 	SEMUNDO_LOCKASSERT(MA_OWNED);
451 	/* Look for and remember the sem_undo if the caller doesn't provide
452 	   it */
453 
454 	suptr = *supptr;
455 	if (suptr == NULL) {
456 		LIST_FOREACH(suptr, &semu_list, un_next) {
457 			if (suptr->un_proc == p) {
458 				*supptr = suptr;
459 				break;
460 			}
461 		}
462 		if (suptr == NULL) {
463 			if (adjval == 0)
464 				return(0);
465 			suptr = semu_alloc(td);
466 			if (suptr == NULL)
467 				return (ENOSPC);
468 			*supptr = suptr;
469 		}
470 	}
471 
472 	/*
473 	 * Look for the requested entry and adjust it (delete if adjval becomes
474 	 * 0).
475 	 */
476 	sunptr = &suptr->un_ent[0];
477 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
478 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
479 			continue;
480 		if (adjval != 0) {
481 			adjval += sunptr->un_adjval;
482 			if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
483 				return (ERANGE);
484 		}
485 		sunptr->un_adjval = adjval;
486 		if (sunptr->un_adjval == 0) {
487 			suptr->un_cnt--;
488 			if (i < suptr->un_cnt)
489 				suptr->un_ent[i] =
490 				    suptr->un_ent[suptr->un_cnt];
491 			if (suptr->un_cnt == 0)
492 				semu_try_free(suptr);
493 		}
494 		return (0);
495 	}
496 
497 	/* Didn't find the right entry - create it */
498 	if (adjval == 0)
499 		return (0);
500 	if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
501 		return (ERANGE);
502 	if (suptr->un_cnt != seminfo.semume) {
503 		sunptr = &suptr->un_ent[suptr->un_cnt];
504 		suptr->un_cnt++;
505 		sunptr->un_adjval = adjval;
506 		sunptr->un_id = semid;
507 		sunptr->un_num = semnum;
508 		sunptr->un_seq = semseq;
509 	} else
510 		return (EINVAL);
511 	return (0);
512 }
513 
514 static void
515 semundo_clear(int semid, int semnum)
516 {
517 	struct sem_undo *suptr, *suptr1;
518 	struct undo *sunptr;
519 	int i;
520 
521 	SEMUNDO_LOCKASSERT(MA_OWNED);
522 	LIST_FOREACH_SAFE(suptr, &semu_list, un_next, suptr1) {
523 		sunptr = &suptr->un_ent[0];
524 		for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
525 			if (sunptr->un_id != semid)
526 				continue;
527 			if (semnum == -1 || sunptr->un_num == semnum) {
528 				suptr->un_cnt--;
529 				if (i < suptr->un_cnt) {
530 					suptr->un_ent[i] =
531 					    suptr->un_ent[suptr->un_cnt];
532 					continue;
533 				}
534 				semu_try_free(suptr);
535 			}
536 			if (semnum != -1)
537 				break;
538 		}
539 	}
540 }
541 
542 static int
543 semvalid(int semid, struct prison *rpr, struct semid_kernel *semakptr)
544 {
545 
546 	return ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
547 	    semakptr->u.sem_perm.seq != IPCID_TO_SEQ(semid) ||
548 	    sem_prison_cansee(rpr, semakptr) ? EINVAL : 0);
549 }
550 
551 static void
552 sem_remove(int semidx, struct ucred *cred)
553 {
554 	struct semid_kernel *semakptr;
555 	int i;
556 
557 	KASSERT(semidx >= 0 && semidx < seminfo.semmni,
558 	    ("semidx out of bounds"));
559 	mtx_assert(&sem_mtx, MA_OWNED);
560 	semakptr = &sema[semidx];
561 	KASSERT(semakptr->u.__sem_base - sem + semakptr->u.sem_nsems <= semtot,
562 	    ("sem_remove: sema %d corrupted sem pointer %p %p %d %d",
563 	    semidx, semakptr->u.__sem_base, sem, semakptr->u.sem_nsems,
564 	    semtot));
565 
566 	semakptr->u.sem_perm.cuid = cred ? cred->cr_uid : 0;
567 	semakptr->u.sem_perm.uid = cred ? cred->cr_uid : 0;
568 	semakptr->u.sem_perm.mode = 0;
569 	racct_sub_cred(semakptr->cred, RACCT_NSEM, semakptr->u.sem_nsems);
570 	crfree(semakptr->cred);
571 	semakptr->cred = NULL;
572 	SEMUNDO_LOCK();
573 	semundo_clear(semidx, -1);
574 	SEMUNDO_UNLOCK();
575 #ifdef MAC
576 	mac_sysvsem_cleanup(semakptr);
577 #endif
578 	wakeup(semakptr);
579 	for (i = 0; i < seminfo.semmni; i++) {
580 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
581 		    sema[i].u.__sem_base > semakptr->u.__sem_base)
582 			mtx_lock_flags(&sema_mtx[i], LOP_DUPOK);
583 	}
584 	for (i = semakptr->u.__sem_base - sem + semakptr->u.sem_nsems;
585 	    i < semtot; i++)
586 		sem[i - semakptr->u.sem_nsems] = sem[i];
587 	for (i = 0; i < seminfo.semmni; i++) {
588 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
589 		    sema[i].u.__sem_base > semakptr->u.__sem_base) {
590 			sema[i].u.__sem_base -= semakptr->u.sem_nsems;
591 			mtx_unlock(&sema_mtx[i]);
592 		}
593 	}
594 	semtot -= semakptr->u.sem_nsems;
595 }
596 
597 static struct prison *
598 sem_find_prison(struct ucred *cred)
599 {
600 	struct prison *pr, *rpr;
601 
602 	pr = cred->cr_prison;
603 	prison_lock(pr);
604 	rpr = osd_jail_get(pr, sem_prison_slot);
605 	prison_unlock(pr);
606 	return (rpr);
607 }
608 
609 static int
610 sem_prison_cansee(struct prison *rpr, struct semid_kernel *semakptr)
611 {
612 
613 	if (semakptr->cred == NULL ||
614 	    !(rpr == semakptr->cred->cr_prison ||
615 	      prison_ischild(rpr, semakptr->cred->cr_prison)))
616 		return (EINVAL);
617 	return (0);
618 }
619 
620 /*
621  * Note that the user-mode half of this passes a union, not a pointer.
622  */
623 #ifndef _SYS_SYSPROTO_H_
624 struct __semctl_args {
625 	int	semid;
626 	int	semnum;
627 	int	cmd;
628 	union	semun *arg;
629 };
630 #endif
631 int
632 sys___semctl(struct thread *td, struct __semctl_args *uap)
633 {
634 	struct semid_ds dsbuf;
635 	union semun arg, semun;
636 	register_t rval;
637 	int error;
638 
639 	switch (uap->cmd) {
640 	case SEM_STAT:
641 	case IPC_SET:
642 	case IPC_STAT:
643 	case GETALL:
644 	case SETVAL:
645 	case SETALL:
646 		error = copyin(uap->arg, &arg, sizeof(arg));
647 		if (error)
648 			return (error);
649 		break;
650 	}
651 
652 	switch (uap->cmd) {
653 	case SEM_STAT:
654 	case IPC_STAT:
655 		semun.buf = &dsbuf;
656 		break;
657 	case IPC_SET:
658 		error = copyin(arg.buf, &dsbuf, sizeof(dsbuf));
659 		if (error)
660 			return (error);
661 		semun.buf = &dsbuf;
662 		break;
663 	case GETALL:
664 	case SETALL:
665 		semun.array = arg.array;
666 		break;
667 	case SETVAL:
668 		semun.val = arg.val;
669 		break;
670 	}
671 
672 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
673 	    &rval);
674 	if (error)
675 		return (error);
676 
677 	switch (uap->cmd) {
678 	case SEM_STAT:
679 	case IPC_STAT:
680 		error = copyout(&dsbuf, arg.buf, sizeof(dsbuf));
681 		break;
682 	}
683 
684 	if (error == 0)
685 		td->td_retval[0] = rval;
686 	return (error);
687 }
688 
689 int
690 kern_semctl(struct thread *td, int semid, int semnum, int cmd,
691     union semun *arg, register_t *rval)
692 {
693 	u_short *array;
694 	struct ucred *cred = td->td_ucred;
695 	int i, error;
696 	struct prison *rpr;
697 	struct semid_ds *sbuf;
698 	struct semid_kernel *semakptr;
699 	struct mtx *sema_mtxp;
700 	u_short usval, count;
701 	int semidx;
702 
703 	DPRINTF(("call to semctl(%d, %d, %d, 0x%p)\n",
704 	    semid, semnum, cmd, arg));
705 
706 	AUDIT_ARG_SVIPC_CMD(cmd);
707 	AUDIT_ARG_SVIPC_ID(semid);
708 
709 	rpr = sem_find_prison(td->td_ucred);
710 	if (rpr == NULL)
711 		return (ENOSYS);
712 
713 	array = NULL;
714 
715 	switch(cmd) {
716 	case SEM_STAT:
717 		/*
718 		 * For this command we assume semid is an array index
719 		 * rather than an IPC id.
720 		 */
721 		if (semid < 0 || semid >= seminfo.semmni)
722 			return (EINVAL);
723 		semakptr = &sema[semid];
724 		sema_mtxp = &sema_mtx[semid];
725 		mtx_lock(sema_mtxp);
726 		if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
727 			error = EINVAL;
728 			goto done2;
729 		}
730 		if ((error = sem_prison_cansee(rpr, semakptr)))
731 			goto done2;
732 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
733 			goto done2;
734 #ifdef MAC
735 		error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
736 		if (error != 0)
737 			goto done2;
738 #endif
739 		bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
740 		if (cred->cr_prison != semakptr->cred->cr_prison)
741 			arg->buf->sem_perm.key = IPC_PRIVATE;
742 		*rval = IXSEQ_TO_IPCID(semid, semakptr->u.sem_perm);
743 		mtx_unlock(sema_mtxp);
744 		return (0);
745 	}
746 
747 	semidx = IPCID_TO_IX(semid);
748 	if (semidx < 0 || semidx >= seminfo.semmni)
749 		return (EINVAL);
750 
751 	semakptr = &sema[semidx];
752 	sema_mtxp = &sema_mtx[semidx];
753 	if (cmd == IPC_RMID)
754 		mtx_lock(&sem_mtx);
755 	mtx_lock(sema_mtxp);
756 
757 #ifdef MAC
758 	error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
759 	if (error != 0)
760 		goto done2;
761 #endif
762 
763 	error = 0;
764 	*rval = 0;
765 
766 	switch (cmd) {
767 	case IPC_RMID:
768 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
769 			goto done2;
770 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
771 			goto done2;
772 		sem_remove(semidx, cred);
773 		break;
774 
775 	case IPC_SET:
776 		AUDIT_ARG_SVIPC_PERM(&arg->buf->sem_perm);
777 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
778 			goto done2;
779 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
780 			goto done2;
781 		sbuf = arg->buf;
782 		semakptr->u.sem_perm.uid = sbuf->sem_perm.uid;
783 		semakptr->u.sem_perm.gid = sbuf->sem_perm.gid;
784 		semakptr->u.sem_perm.mode = (semakptr->u.sem_perm.mode &
785 		    ~0777) | (sbuf->sem_perm.mode & 0777);
786 		semakptr->u.sem_ctime = time_second;
787 		break;
788 
789 	case IPC_STAT:
790 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
791 			goto done2;
792 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
793 			goto done2;
794 		bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
795 		if (cred->cr_prison != semakptr->cred->cr_prison)
796 			arg->buf->sem_perm.key = IPC_PRIVATE;
797 
798 		/*
799 		 * Try to hide the fact that the structure layout is shared by
800 		 * both the kernel and userland.  This pointer is not useful to
801 		 * userspace.
802 		 */
803 		arg->buf->__sem_base = NULL;
804 		break;
805 
806 	case GETNCNT:
807 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
808 			goto done2;
809 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
810 			goto done2;
811 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
812 			error = EINVAL;
813 			goto done2;
814 		}
815 		*rval = semakptr->u.__sem_base[semnum].semncnt;
816 		break;
817 
818 	case GETPID:
819 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
820 			goto done2;
821 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
822 			goto done2;
823 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
824 			error = EINVAL;
825 			goto done2;
826 		}
827 		*rval = semakptr->u.__sem_base[semnum].sempid;
828 		break;
829 
830 	case GETVAL:
831 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
832 			goto done2;
833 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
834 			goto done2;
835 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
836 			error = EINVAL;
837 			goto done2;
838 		}
839 		*rval = semakptr->u.__sem_base[semnum].semval;
840 		break;
841 
842 	case GETALL:
843 		/*
844 		 * Unfortunately, callers of this function don't know
845 		 * in advance how many semaphores are in this set.
846 		 * While we could just allocate the maximum size array
847 		 * and pass the actual size back to the caller, that
848 		 * won't work for SETALL since we can't copyin() more
849 		 * data than the user specified as we may return a
850 		 * spurious EFAULT.
851 		 *
852 		 * Note that the number of semaphores in a set is
853 		 * fixed for the life of that set.  The only way that
854 		 * the 'count' could change while are blocked in
855 		 * malloc() is if this semaphore set were destroyed
856 		 * and a new one created with the same index.
857 		 * However, semvalid() will catch that due to the
858 		 * sequence number unless exactly 0x8000 (or a
859 		 * multiple thereof) semaphore sets for the same index
860 		 * are created and destroyed while we are in malloc!
861 		 *
862 		 */
863 		count = semakptr->u.sem_nsems;
864 		mtx_unlock(sema_mtxp);
865 		array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
866 		mtx_lock(sema_mtxp);
867 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
868 			goto done2;
869 		KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
870 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
871 			goto done2;
872 		for (i = 0; i < semakptr->u.sem_nsems; i++)
873 			array[i] = semakptr->u.__sem_base[i].semval;
874 		mtx_unlock(sema_mtxp);
875 		error = copyout(array, arg->array, count * sizeof(*array));
876 		mtx_lock(sema_mtxp);
877 		break;
878 
879 	case GETZCNT:
880 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
881 			goto done2;
882 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
883 			goto done2;
884 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
885 			error = EINVAL;
886 			goto done2;
887 		}
888 		*rval = semakptr->u.__sem_base[semnum].semzcnt;
889 		break;
890 
891 	case SETVAL:
892 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
893 			goto done2;
894 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
895 			goto done2;
896 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
897 			error = EINVAL;
898 			goto done2;
899 		}
900 		if (arg->val < 0 || arg->val > seminfo.semvmx) {
901 			error = ERANGE;
902 			goto done2;
903 		}
904 		semakptr->u.__sem_base[semnum].semval = arg->val;
905 		SEMUNDO_LOCK();
906 		semundo_clear(semidx, semnum);
907 		SEMUNDO_UNLOCK();
908 		wakeup(semakptr);
909 		break;
910 
911 	case SETALL:
912 		/*
913 		 * See comment on GETALL for why 'count' shouldn't change
914 		 * and why we require a userland buffer.
915 		 */
916 		count = semakptr->u.sem_nsems;
917 		mtx_unlock(sema_mtxp);
918 		array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
919 		error = copyin(arg->array, array, count * sizeof(*array));
920 		mtx_lock(sema_mtxp);
921 		if (error)
922 			break;
923 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
924 			goto done2;
925 		KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
926 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
927 			goto done2;
928 		for (i = 0; i < semakptr->u.sem_nsems; i++) {
929 			usval = array[i];
930 			if (usval > seminfo.semvmx) {
931 				error = ERANGE;
932 				break;
933 			}
934 			semakptr->u.__sem_base[i].semval = usval;
935 		}
936 		SEMUNDO_LOCK();
937 		semundo_clear(semidx, -1);
938 		SEMUNDO_UNLOCK();
939 		wakeup(semakptr);
940 		break;
941 
942 	default:
943 		error = EINVAL;
944 		break;
945 	}
946 
947 done2:
948 	mtx_unlock(sema_mtxp);
949 	if (cmd == IPC_RMID)
950 		mtx_unlock(&sem_mtx);
951 	if (array != NULL)
952 		free(array, M_TEMP);
953 	return(error);
954 }
955 
956 #ifndef _SYS_SYSPROTO_H_
957 struct semget_args {
958 	key_t	key;
959 	int	nsems;
960 	int	semflg;
961 };
962 #endif
963 int
964 sys_semget(struct thread *td, struct semget_args *uap)
965 {
966 	int semid, error = 0;
967 	int key = uap->key;
968 	int nsems = uap->nsems;
969 	int semflg = uap->semflg;
970 	struct ucred *cred = td->td_ucred;
971 
972 	DPRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg));
973 
974 	AUDIT_ARG_VALUE(semflg);
975 
976 	if (sem_find_prison(cred) == NULL)
977 		return (ENOSYS);
978 
979 	mtx_lock(&sem_mtx);
980 	if (key != IPC_PRIVATE) {
981 		for (semid = 0; semid < seminfo.semmni; semid++) {
982 			if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) &&
983 			    sema[semid].cred != NULL &&
984 			    sema[semid].cred->cr_prison == cred->cr_prison &&
985 			    sema[semid].u.sem_perm.key == key)
986 				break;
987 		}
988 		if (semid < seminfo.semmni) {
989 			AUDIT_ARG_SVIPC_ID(semid);
990 			DPRINTF(("found public key\n"));
991 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
992 				DPRINTF(("not exclusive\n"));
993 				error = EEXIST;
994 				goto done2;
995 			}
996 			if ((error = ipcperm(td, &sema[semid].u.sem_perm,
997 			    semflg & 0700))) {
998 				goto done2;
999 			}
1000 			if (nsems > 0 && sema[semid].u.sem_nsems < nsems) {
1001 				DPRINTF(("too small\n"));
1002 				error = EINVAL;
1003 				goto done2;
1004 			}
1005 #ifdef MAC
1006 			error = mac_sysvsem_check_semget(cred, &sema[semid]);
1007 			if (error != 0)
1008 				goto done2;
1009 #endif
1010 			goto found;
1011 		}
1012 	}
1013 
1014 	DPRINTF(("need to allocate the semid_kernel\n"));
1015 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
1016 		if (nsems <= 0 || nsems > seminfo.semmsl) {
1017 			DPRINTF(("nsems out of range (0<%d<=%d)\n", nsems,
1018 			    seminfo.semmsl));
1019 			error = EINVAL;
1020 			goto done2;
1021 		}
1022 		if (nsems > seminfo.semmns - semtot) {
1023 			DPRINTF((
1024 			    "not enough semaphores left (need %d, got %d)\n",
1025 			    nsems, seminfo.semmns - semtot));
1026 			error = ENOSPC;
1027 			goto done2;
1028 		}
1029 		for (semid = 0; semid < seminfo.semmni; semid++) {
1030 			if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0)
1031 				break;
1032 		}
1033 		if (semid == seminfo.semmni) {
1034 			DPRINTF(("no more semid_kernel's available\n"));
1035 			error = ENOSPC;
1036 			goto done2;
1037 		}
1038 #ifdef RACCT
1039 		if (racct_enable) {
1040 			PROC_LOCK(td->td_proc);
1041 			error = racct_add(td->td_proc, RACCT_NSEM, nsems);
1042 			PROC_UNLOCK(td->td_proc);
1043 			if (error != 0) {
1044 				error = ENOSPC;
1045 				goto done2;
1046 			}
1047 		}
1048 #endif
1049 		DPRINTF(("semid %d is available\n", semid));
1050 		mtx_lock(&sema_mtx[semid]);
1051 		KASSERT((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0,
1052 		    ("Lost semaphore %d", semid));
1053 		sema[semid].u.sem_perm.key = key;
1054 		sema[semid].u.sem_perm.cuid = cred->cr_uid;
1055 		sema[semid].u.sem_perm.uid = cred->cr_uid;
1056 		sema[semid].u.sem_perm.cgid = cred->cr_gid;
1057 		sema[semid].u.sem_perm.gid = cred->cr_gid;
1058 		sema[semid].u.sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
1059 		sema[semid].cred = crhold(cred);
1060 		sema[semid].u.sem_perm.seq =
1061 		    (sema[semid].u.sem_perm.seq + 1) & 0x7fff;
1062 		sema[semid].u.sem_nsems = nsems;
1063 		sema[semid].u.sem_otime = 0;
1064 		sema[semid].u.sem_ctime = time_second;
1065 		sema[semid].u.__sem_base = &sem[semtot];
1066 		semtot += nsems;
1067 		bzero(sema[semid].u.__sem_base,
1068 		    sizeof(sema[semid].u.__sem_base[0])*nsems);
1069 #ifdef MAC
1070 		mac_sysvsem_create(cred, &sema[semid]);
1071 #endif
1072 		mtx_unlock(&sema_mtx[semid]);
1073 		DPRINTF(("sembase = %p, next = %p\n",
1074 		    sema[semid].u.__sem_base, &sem[semtot]));
1075 	} else {
1076 		DPRINTF(("didn't find it and wasn't asked to create it\n"));
1077 		error = ENOENT;
1078 		goto done2;
1079 	}
1080 
1081 found:
1082 	td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].u.sem_perm);
1083 done2:
1084 	mtx_unlock(&sem_mtx);
1085 	return (error);
1086 }
1087 
1088 #ifndef _SYS_SYSPROTO_H_
1089 struct semop_args {
1090 	int	semid;
1091 	struct	sembuf *sops;
1092 	size_t	nsops;
1093 };
1094 #endif
1095 int
1096 sys_semop(struct thread *td, struct semop_args *uap)
1097 {
1098 
1099 	return (kern_semop(td, uap->semid, uap->sops, uap->nsops, NULL));
1100 }
1101 
1102 int
1103 kern_semop(struct thread *td, int usemid, struct sembuf *usops,
1104     size_t nsops, struct timespec *timeout)
1105 {
1106 #define SMALL_SOPS	8
1107 	struct sembuf small_sops[SMALL_SOPS];
1108 	int semid;
1109 	struct prison *rpr;
1110 	struct sembuf *sops;
1111 	struct semid_kernel *semakptr;
1112 	struct sembuf *sopptr = NULL;
1113 	struct sem *semptr = NULL;
1114 	struct sem_undo *suptr;
1115 	struct mtx *sema_mtxp;
1116 	sbintime_t sbt, precision;
1117 	size_t i, j, k;
1118 	int error;
1119 	int do_wakeup, do_undos;
1120 	unsigned short seq;
1121 
1122 #ifdef SEM_DEBUG
1123 	sops = NULL;
1124 #endif
1125 	DPRINTF(("call to semop(%d, %p, %u)\n", usemid, usops, nsops));
1126 
1127 	AUDIT_ARG_SVIPC_ID(usemid);
1128 
1129 	rpr = sem_find_prison(td->td_ucred);
1130 	if (rpr == NULL)
1131 		return (ENOSYS);
1132 
1133 	semid = IPCID_TO_IX(usemid);	/* Convert back to zero origin */
1134 
1135 	if (semid < 0 || semid >= seminfo.semmni)
1136 		return (EINVAL);
1137 	if (timeout != NULL) {
1138 		if (!timespecvalid_interval(timeout))
1139 			return (EINVAL);
1140 		precision = 0;
1141 		if (timespecisset(timeout)) {
1142 			if (timeout->tv_sec < INT32_MAX / 2) {
1143 				precision = tstosbt(*timeout);
1144 				if (TIMESEL(&sbt, precision))
1145 					sbt += tc_tick_sbt;
1146 				sbt += precision;
1147 				precision >>= tc_precexp;
1148 			} else
1149 				sbt = 0;
1150 		} else
1151 			sbt = -1;
1152 	} else
1153 		precision = sbt = 0;
1154 
1155 	/* Allocate memory for sem_ops */
1156 	if (nsops <= SMALL_SOPS)
1157 		sops = small_sops;
1158 	else if (nsops > seminfo.semopm) {
1159 		DPRINTF(("too many sops (max=%d, nsops=%d)\n", seminfo.semopm,
1160 		    nsops));
1161 		return (E2BIG);
1162 	} else {
1163 #ifdef RACCT
1164 		if (racct_enable) {
1165 			PROC_LOCK(td->td_proc);
1166 			if (nsops >
1167 			    racct_get_available(td->td_proc, RACCT_NSEMOP)) {
1168 				PROC_UNLOCK(td->td_proc);
1169 				return (E2BIG);
1170 			}
1171 			PROC_UNLOCK(td->td_proc);
1172 		}
1173 #endif
1174 
1175 		sops = malloc(nsops * sizeof(*sops), M_TEMP, M_WAITOK);
1176 	}
1177 	if ((error = copyin(usops, sops, nsops * sizeof(sops[0]))) != 0) {
1178 		DPRINTF(("error = %d from copyin(%p, %p, %d)\n", error,
1179 		    usops, sops, nsops * sizeof(sops[0])));
1180 		if (sops != small_sops)
1181 			free(sops, M_TEMP);
1182 		return (error);
1183 	}
1184 
1185 	semakptr = &sema[semid];
1186 	sema_mtxp = &sema_mtx[semid];
1187 	mtx_lock(sema_mtxp);
1188 	if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
1189 		error = EINVAL;
1190 		goto done2;
1191 	}
1192 	seq = semakptr->u.sem_perm.seq;
1193 	if (seq != IPCID_TO_SEQ(usemid)) {
1194 		error = EINVAL;
1195 		goto done2;
1196 	}
1197 	if ((error = sem_prison_cansee(rpr, semakptr)) != 0)
1198 		goto done2;
1199 	/*
1200 	 * Initial pass through sops to see what permissions are needed.
1201 	 * Also perform any checks that don't need repeating on each
1202 	 * attempt to satisfy the request vector.
1203 	 */
1204 	j = 0;		/* permission needed */
1205 	do_undos = 0;
1206 	for (i = 0; i < nsops; i++) {
1207 		sopptr = &sops[i];
1208 		if (sopptr->sem_num >= semakptr->u.sem_nsems) {
1209 			error = EFBIG;
1210 			goto done2;
1211 		}
1212 		if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0)
1213 			do_undos = 1;
1214 		j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A;
1215 	}
1216 
1217 	if ((error = ipcperm(td, &semakptr->u.sem_perm, j))) {
1218 		DPRINTF(("error = %d from ipaccess\n", error));
1219 		goto done2;
1220 	}
1221 #ifdef MAC
1222 	error = mac_sysvsem_check_semop(td->td_ucred, semakptr, j);
1223 	if (error != 0)
1224 		goto done2;
1225 #endif
1226 
1227 	/*
1228 	 * Loop trying to satisfy the vector of requests.
1229 	 * If we reach a point where we must wait, any requests already
1230 	 * performed are rolled back and we go to sleep until some other
1231 	 * process wakes us up.  At this point, we start all over again.
1232 	 *
1233 	 * This ensures that from the perspective of other tasks, a set
1234 	 * of requests is atomic (never partially satisfied).
1235 	 */
1236 	for (;;) {
1237 		do_wakeup = 0;
1238 		error = 0;	/* error return if necessary */
1239 
1240 		for (i = 0; i < nsops; i++) {
1241 			sopptr = &sops[i];
1242 			semptr = &semakptr->u.__sem_base[sopptr->sem_num];
1243 
1244 			DPRINTF((
1245 			    "semop:  semakptr=%p, __sem_base=%p, "
1246 			    "semptr=%p, sem[%d]=%d : op=%d, flag=%s\n",
1247 			    semakptr, semakptr->u.__sem_base, semptr,
1248 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
1249 			    (sopptr->sem_flg & IPC_NOWAIT) ?
1250 			    "nowait" : "wait"));
1251 
1252 			if (sopptr->sem_op < 0) {
1253 				if (semptr->semval + sopptr->sem_op < 0) {
1254 					DPRINTF(("semop:  can't do it now\n"));
1255 					break;
1256 				} else {
1257 					semptr->semval += sopptr->sem_op;
1258 					if (semptr->semval == 0 &&
1259 					    semptr->semzcnt > 0)
1260 						do_wakeup = 1;
1261 				}
1262 			} else if (sopptr->sem_op == 0) {
1263 				if (semptr->semval != 0) {
1264 					DPRINTF(("semop:  not zero now\n"));
1265 					break;
1266 				}
1267 			} else if (semptr->semval + sopptr->sem_op >
1268 			    seminfo.semvmx) {
1269 				error = ERANGE;
1270 				break;
1271 			} else {
1272 				if (semptr->semncnt > 0)
1273 					do_wakeup = 1;
1274 				semptr->semval += sopptr->sem_op;
1275 			}
1276 		}
1277 
1278 		/*
1279 		 * Did we get through the entire vector?
1280 		 */
1281 		if (i >= nsops)
1282 			goto done;
1283 
1284 		/*
1285 		 * No ... rollback anything that we've already done
1286 		 */
1287 		DPRINTF(("semop:  rollback 0 through %d\n", i-1));
1288 		for (j = 0; j < i; j++)
1289 			semakptr->u.__sem_base[sops[j].sem_num].semval -=
1290 			    sops[j].sem_op;
1291 
1292 		/* If we detected an error, return it */
1293 		if (error != 0)
1294 			goto done2;
1295 
1296 		/*
1297 		 * If the request that we couldn't satisfy has the
1298 		 * NOWAIT flag set then return with EAGAIN.
1299 		 */
1300 		if (sopptr->sem_flg & IPC_NOWAIT) {
1301 			error = EAGAIN;
1302 			goto done2;
1303 		}
1304 
1305 		if (sopptr->sem_op == 0)
1306 			semptr->semzcnt++;
1307 		else
1308 			semptr->semncnt++;
1309 
1310 		DPRINTF(("semop:  good night!\n"));
1311 		error = msleep_sbt(semakptr, sema_mtxp, (PZERO - 4) | PCATCH,
1312 		    "semwait", sbt, precision, C_ABSOLUTE);
1313 		DPRINTF(("semop:  good morning (error=%d)!\n", error));
1314 		/* return code is checked below, after sem[nz]cnt-- */
1315 
1316 		/*
1317 		 * Make sure that the semaphore still exists
1318 		 */
1319 		seq = semakptr->u.sem_perm.seq;
1320 		if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1321 		    seq != IPCID_TO_SEQ(usemid)) {
1322 			error = EIDRM;
1323 			goto done2;
1324 		}
1325 
1326 		/*
1327 		 * Renew the semaphore's pointer after wakeup since
1328 		 * during msleep __sem_base may have been modified and semptr
1329 		 * is not valid any more
1330 		 */
1331 		semptr = &semakptr->u.__sem_base[sopptr->sem_num];
1332 
1333 		/*
1334 		 * The semaphore is still alive.  Readjust the count of
1335 		 * waiting processes.
1336 		 */
1337 		if (sopptr->sem_op == 0)
1338 			semptr->semzcnt--;
1339 		else
1340 			semptr->semncnt--;
1341 
1342 		/*
1343 		 * Is it really morning, or was our sleep interrupted?
1344 		 * (Delayed check of msleep() return code because we
1345 		 * need to decrement sem[nz]cnt either way.)
1346 		 */
1347 		if (error != 0) {
1348 			if (error == ERESTART)
1349 				error = EINTR;
1350 			goto done2;
1351 		}
1352 		DPRINTF(("semop:  good morning!\n"));
1353 	}
1354 
1355 done:
1356 	/*
1357 	 * Process any SEM_UNDO requests.
1358 	 */
1359 	if (do_undos) {
1360 		SEMUNDO_LOCK();
1361 		suptr = NULL;
1362 		for (i = 0; i < nsops; i++) {
1363 			/*
1364 			 * We only need to deal with SEM_UNDO's for non-zero
1365 			 * op's.
1366 			 */
1367 			int adjval;
1368 
1369 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
1370 				continue;
1371 			adjval = sops[i].sem_op;
1372 			if (adjval == 0)
1373 				continue;
1374 			error = semundo_adjust(td, &suptr, semid, seq,
1375 			    sops[i].sem_num, -adjval);
1376 			if (error == 0)
1377 				continue;
1378 
1379 			/*
1380 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
1381 			 * Rollback the adjustments to this point and then
1382 			 * rollback the semaphore ups and down so we can return
1383 			 * with an error with all structures restored.  We
1384 			 * rollback the undo's in the exact reverse order that
1385 			 * we applied them.  This guarantees that we won't run
1386 			 * out of space as we roll things back out.
1387 			 */
1388 			for (j = 0; j < i; j++) {
1389 				k = i - j - 1;
1390 				if ((sops[k].sem_flg & SEM_UNDO) == 0)
1391 					continue;
1392 				adjval = sops[k].sem_op;
1393 				if (adjval == 0)
1394 					continue;
1395 				if (semundo_adjust(td, &suptr, semid, seq,
1396 				    sops[k].sem_num, adjval) != 0)
1397 					panic("semop - can't undo undos");
1398 			}
1399 
1400 			for (j = 0; j < nsops; j++)
1401 				semakptr->u.__sem_base[sops[j].sem_num].semval -=
1402 				    sops[j].sem_op;
1403 
1404 			DPRINTF(("error = %d from semundo_adjust\n", error));
1405 			SEMUNDO_UNLOCK();
1406 			goto done2;
1407 		} /* loop through the sops */
1408 		SEMUNDO_UNLOCK();
1409 	} /* if (do_undos) */
1410 
1411 	/* We're definitely done - set the sempid's and time */
1412 	for (i = 0; i < nsops; i++) {
1413 		sopptr = &sops[i];
1414 		semptr = &semakptr->u.__sem_base[sopptr->sem_num];
1415 		semptr->sempid = td->td_proc->p_pid;
1416 	}
1417 	semakptr->u.sem_otime = time_second;
1418 
1419 	/*
1420 	 * Do a wakeup if any semaphore was up'd whilst something was
1421 	 * sleeping on it.
1422 	 */
1423 	if (do_wakeup) {
1424 		DPRINTF(("semop:  doing wakeup\n"));
1425 		wakeup(semakptr);
1426 		DPRINTF(("semop:  back from wakeup\n"));
1427 	}
1428 	DPRINTF(("semop:  done\n"));
1429 	td->td_retval[0] = 0;
1430 done2:
1431 	mtx_unlock(sema_mtxp);
1432 	if (sops != small_sops)
1433 		free(sops, M_TEMP);
1434 	return (error);
1435 }
1436 
1437 /*
1438  * Go through the undo structures for this process and apply the adjustments to
1439  * semaphores.
1440  */
1441 static void
1442 semexit_myhook(void *arg, struct proc *p)
1443 {
1444 	struct sem_undo *suptr;
1445 	struct semid_kernel *semakptr;
1446 	struct mtx *sema_mtxp;
1447 	int semid, semnum, adjval, ix;
1448 	unsigned short seq;
1449 
1450 	/*
1451 	 * Go through the chain of undo vectors looking for one
1452 	 * associated with this process.
1453 	 */
1454 	if (LIST_EMPTY(&semu_list))
1455 		return;
1456 	SEMUNDO_LOCK();
1457 	LIST_FOREACH(suptr, &semu_list, un_next) {
1458 		if (suptr->un_proc == p)
1459 			break;
1460 	}
1461 	if (suptr == NULL) {
1462 		SEMUNDO_UNLOCK();
1463 		return;
1464 	}
1465 	LIST_REMOVE(suptr, un_next);
1466 
1467 	DPRINTF(("proc @%p has undo structure with %d entries\n", p,
1468 	    suptr->un_cnt));
1469 
1470 	/*
1471 	 * If there are any active undo elements then process them.
1472 	 */
1473 	if (suptr->un_cnt > 0) {
1474 		SEMUNDO_UNLOCK();
1475 		for (ix = 0; ix < suptr->un_cnt; ix++) {
1476 			semid = suptr->un_ent[ix].un_id;
1477 			semnum = suptr->un_ent[ix].un_num;
1478 			adjval = suptr->un_ent[ix].un_adjval;
1479 			seq = suptr->un_ent[ix].un_seq;
1480 			semakptr = &sema[semid];
1481 			sema_mtxp = &sema_mtx[semid];
1482 
1483 			mtx_lock(sema_mtxp);
1484 			if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1485 			    (semakptr->u.sem_perm.seq != seq)) {
1486 				mtx_unlock(sema_mtxp);
1487 				continue;
1488 			}
1489 			if (semnum >= semakptr->u.sem_nsems)
1490 				panic("semexit - semnum out of range");
1491 
1492 			DPRINTF((
1493 			    "semexit:  %p id=%d num=%d(adj=%d) ; sem=%d\n",
1494 			    suptr->un_proc, suptr->un_ent[ix].un_id,
1495 			    suptr->un_ent[ix].un_num,
1496 			    suptr->un_ent[ix].un_adjval,
1497 			    semakptr->u.__sem_base[semnum].semval));
1498 
1499 			if (adjval < 0 && semakptr->u.__sem_base[semnum].semval <
1500 			    -adjval)
1501 				semakptr->u.__sem_base[semnum].semval = 0;
1502 			else
1503 				semakptr->u.__sem_base[semnum].semval += adjval;
1504 
1505 			wakeup(semakptr);
1506 			DPRINTF(("semexit:  back from wakeup\n"));
1507 			mtx_unlock(sema_mtxp);
1508 		}
1509 		SEMUNDO_LOCK();
1510 	}
1511 
1512 	/*
1513 	 * Deallocate the undo vector.
1514 	 */
1515 	DPRINTF(("removing vector\n"));
1516 	suptr->un_proc = NULL;
1517 	suptr->un_cnt = 0;
1518 	LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
1519 	SEMUNDO_UNLOCK();
1520 }
1521 
1522 static int
1523 sysctl_sema(SYSCTL_HANDLER_ARGS)
1524 {
1525 	struct prison *pr, *rpr;
1526 	struct semid_kernel tsemak;
1527 #ifdef COMPAT_FREEBSD32
1528 	struct semid_kernel32 tsemak32;
1529 #endif
1530 	void *outaddr;
1531 	size_t outsize;
1532 	int error, i;
1533 
1534 	pr = req->td->td_ucred->cr_prison;
1535 	rpr = sem_find_prison(req->td->td_ucred);
1536 	error = 0;
1537 	for (i = 0; i < seminfo.semmni; i++) {
1538 		mtx_lock(&sema_mtx[i]);
1539 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) == 0 ||
1540 		    rpr == NULL || sem_prison_cansee(rpr, &sema[i]) != 0)
1541 			bzero(&tsemak, sizeof(tsemak));
1542 		else {
1543 			tsemak = sema[i];
1544 			if (tsemak.cred->cr_prison != pr)
1545 				tsemak.u.sem_perm.key = IPC_PRIVATE;
1546 		}
1547 		mtx_unlock(&sema_mtx[i]);
1548 #ifdef COMPAT_FREEBSD32
1549 		if (SV_CURPROC_FLAG(SV_ILP32)) {
1550 			bzero(&tsemak32, sizeof(tsemak32));
1551 			freebsd32_ipcperm_out(&tsemak.u.sem_perm,
1552 			    &tsemak32.u.sem_perm);
1553 			/* Don't copy u.__sem_base */
1554 			CP(tsemak, tsemak32, u.sem_nsems);
1555 			CP(tsemak, tsemak32, u.sem_otime);
1556 			CP(tsemak, tsemak32, u.sem_ctime);
1557 			/* Don't copy label or cred */
1558 			outaddr = &tsemak32;
1559 			outsize = sizeof(tsemak32);
1560 		} else
1561 #endif
1562 		{
1563 			tsemak.u.__sem_base = NULL;
1564 			tsemak.label = NULL;
1565 			tsemak.cred = NULL;
1566 			outaddr = &tsemak;
1567 			outsize = sizeof(tsemak);
1568 		}
1569 		error = SYSCTL_OUT(req, outaddr, outsize);
1570 		if (error != 0)
1571 			break;
1572 	}
1573 	return (error);
1574 }
1575 
1576 int
1577 kern_get_sema(struct thread *td, struct semid_kernel **res, size_t *sz)
1578 {
1579 	struct prison *pr, *rpr;
1580 	struct semid_kernel *psemak;
1581 	int i, mi;
1582 
1583 	*sz = mi = seminfo.semmni;
1584 	if (res == NULL)
1585 		return (0);
1586 
1587 	pr = td->td_ucred->cr_prison;
1588 	rpr = sem_find_prison(td->td_ucred);
1589 	*res = malloc(sizeof(struct semid_kernel) * mi, M_TEMP, M_WAITOK);
1590 	for (i = 0; i < mi; i++) {
1591 		psemak = &(*res)[i];
1592 		mtx_lock(&sema_mtx[i]);
1593 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) == 0 ||
1594 		    rpr == NULL || sem_prison_cansee(rpr, &sema[i]) != 0)
1595 			bzero(psemak, sizeof(*psemak));
1596 		else {
1597 			*psemak = sema[i];
1598 			if (psemak->cred->cr_prison != pr)
1599 				psemak->u.sem_perm.key = IPC_PRIVATE;
1600 		}
1601 		mtx_unlock(&sema_mtx[i]);
1602 		psemak->u.__sem_base = NULL;
1603 		psemak->label = NULL;
1604 		psemak->cred = NULL;
1605 	}
1606 	return (0);
1607 }
1608 
1609 static int
1610 sem_prison_check(void *obj, void *data)
1611 {
1612 	struct prison *pr = obj;
1613 	struct prison *prpr;
1614 	struct vfsoptlist *opts = data;
1615 	int error, jsys;
1616 
1617 	/*
1618 	 * sysvsem is a jailsys integer.
1619 	 * It must be "disable" if the parent jail is disabled.
1620 	 */
1621 	error = vfs_copyopt(opts, "sysvsem", &jsys, sizeof(jsys));
1622 	if (error != ENOENT) {
1623 		if (error != 0)
1624 			return (error);
1625 		switch (jsys) {
1626 		case JAIL_SYS_DISABLE:
1627 			break;
1628 		case JAIL_SYS_NEW:
1629 		case JAIL_SYS_INHERIT:
1630 			prison_lock(pr->pr_parent);
1631 			prpr = osd_jail_get(pr->pr_parent, sem_prison_slot);
1632 			prison_unlock(pr->pr_parent);
1633 			if (prpr == NULL)
1634 				return (EPERM);
1635 			break;
1636 		default:
1637 			return (EINVAL);
1638 		}
1639 	}
1640 
1641 	return (0);
1642 }
1643 
1644 static int
1645 sem_prison_set(void *obj, void *data)
1646 {
1647 	struct prison *pr = obj;
1648 	struct prison *tpr, *orpr, *nrpr, *trpr;
1649 	struct vfsoptlist *opts = data;
1650 	void *rsv;
1651 	int jsys, descend;
1652 
1653 	/*
1654 	 * sysvsem controls which jail is the root of the associated sems (this
1655 	 * jail or same as the parent), or if the feature is available at all.
1656 	 */
1657 	if (vfs_copyopt(opts, "sysvsem", &jsys, sizeof(jsys)) == ENOENT)
1658 		jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0)
1659 		    ? JAIL_SYS_INHERIT
1660 		    : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0)
1661 		    ? JAIL_SYS_DISABLE
1662 		    : -1;
1663 	if (jsys == JAIL_SYS_DISABLE) {
1664 		prison_lock(pr);
1665 		orpr = osd_jail_get(pr, sem_prison_slot);
1666 		if (orpr != NULL)
1667 			osd_jail_del(pr, sem_prison_slot);
1668 		prison_unlock(pr);
1669 		if (orpr != NULL) {
1670 			if (orpr == pr)
1671 				sem_prison_cleanup(pr);
1672 			/* Disable all child jails as well. */
1673 			FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1674 				prison_lock(tpr);
1675 				trpr = osd_jail_get(tpr, sem_prison_slot);
1676 				if (trpr != NULL) {
1677 					osd_jail_del(tpr, sem_prison_slot);
1678 					prison_unlock(tpr);
1679 					if (trpr == tpr)
1680 						sem_prison_cleanup(tpr);
1681 				} else {
1682 					prison_unlock(tpr);
1683 					descend = 0;
1684 				}
1685 			}
1686 		}
1687 	} else if (jsys != -1) {
1688 		if (jsys == JAIL_SYS_NEW)
1689 			nrpr = pr;
1690 		else {
1691 			prison_lock(pr->pr_parent);
1692 			nrpr = osd_jail_get(pr->pr_parent, sem_prison_slot);
1693 			prison_unlock(pr->pr_parent);
1694 		}
1695 		rsv = osd_reserve(sem_prison_slot);
1696 		prison_lock(pr);
1697 		orpr = osd_jail_get(pr, sem_prison_slot);
1698 		if (orpr != nrpr)
1699 			(void)osd_jail_set_reserved(pr, sem_prison_slot, rsv,
1700 			    nrpr);
1701 		else
1702 			osd_free_reserved(rsv);
1703 		prison_unlock(pr);
1704 		if (orpr != nrpr) {
1705 			if (orpr == pr)
1706 				sem_prison_cleanup(pr);
1707 			if (orpr != NULL) {
1708 				/* Change child jails matching the old root, */
1709 				FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1710 					prison_lock(tpr);
1711 					trpr = osd_jail_get(tpr,
1712 					    sem_prison_slot);
1713 					if (trpr == orpr) {
1714 						(void)osd_jail_set(tpr,
1715 						    sem_prison_slot, nrpr);
1716 						prison_unlock(tpr);
1717 						if (trpr == tpr)
1718 							sem_prison_cleanup(tpr);
1719 					} else {
1720 						prison_unlock(tpr);
1721 						descend = 0;
1722 					}
1723 				}
1724 			}
1725 		}
1726 	}
1727 
1728 	return (0);
1729 }
1730 
1731 static int
1732 sem_prison_get(void *obj, void *data)
1733 {
1734 	struct prison *pr = obj;
1735 	struct prison *rpr;
1736 	struct vfsoptlist *opts = data;
1737 	int error, jsys;
1738 
1739 	/* Set sysvsem based on the jail's root prison. */
1740 	prison_lock(pr);
1741 	rpr = osd_jail_get(pr, sem_prison_slot);
1742 	prison_unlock(pr);
1743 	jsys = rpr == NULL ? JAIL_SYS_DISABLE
1744 	    : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT;
1745 	error = vfs_setopt(opts, "sysvsem", &jsys, sizeof(jsys));
1746 	if (error == ENOENT)
1747 		error = 0;
1748 	return (error);
1749 }
1750 
1751 static int
1752 sem_prison_remove(void *obj, void *data __unused)
1753 {
1754 	struct prison *pr = obj;
1755 	struct prison *rpr;
1756 
1757 	prison_lock(pr);
1758 	rpr = osd_jail_get(pr, sem_prison_slot);
1759 	prison_unlock(pr);
1760 	if (rpr == pr)
1761 		sem_prison_cleanup(pr);
1762 	return (0);
1763 }
1764 
1765 static void
1766 sem_prison_cleanup(struct prison *pr)
1767 {
1768 	int i;
1769 
1770 	/* Remove any sems that belong to this jail. */
1771 	mtx_lock(&sem_mtx);
1772 	for (i = 0; i < seminfo.semmni; i++) {
1773 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
1774 		    sema[i].cred != NULL && sema[i].cred->cr_prison == pr) {
1775 			mtx_lock(&sema_mtx[i]);
1776 			sem_remove(i, NULL);
1777 			mtx_unlock(&sema_mtx[i]);
1778 		}
1779 	}
1780 	mtx_unlock(&sem_mtx);
1781 }
1782 
1783 SYSCTL_JAIL_PARAM_SYS_NODE(sysvsem, CTLFLAG_RW, "SYSV semaphores");
1784 
1785 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1786     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1787 
1788 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1789 static sy_call_t *semcalls[] = {
1790 	(sy_call_t *)freebsd7___semctl, (sy_call_t *)sys_semget,
1791 	(sy_call_t *)sys_semop
1792 };
1793 
1794 /*
1795  * Entry point for all SEM calls.
1796  */
1797 int
1798 sys_semsys(struct thread *td, struct semsys_args *uap)
1799 {
1800 	int error;
1801 
1802 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1803 	if (uap->which < 0 || uap->which >= nitems(semcalls))
1804 		return (EINVAL);
1805 	error = (*semcalls[uap->which])(td, &uap->a2);
1806 	return (error);
1807 }
1808 
1809 #ifndef _SYS_SYSPROTO_H_
1810 struct freebsd7___semctl_args {
1811 	int	semid;
1812 	int	semnum;
1813 	int	cmd;
1814 	union	semun_old *arg;
1815 };
1816 #endif
1817 int
1818 freebsd7___semctl(struct thread *td, struct freebsd7___semctl_args *uap)
1819 {
1820 	struct semid_ds_old dsold;
1821 	struct semid_ds dsbuf;
1822 	union semun_old arg;
1823 	union semun semun;
1824 	register_t rval;
1825 	int error;
1826 
1827 	switch (uap->cmd) {
1828 	case SEM_STAT:
1829 	case IPC_SET:
1830 	case IPC_STAT:
1831 	case GETALL:
1832 	case SETVAL:
1833 	case SETALL:
1834 		error = copyin(uap->arg, &arg, sizeof(arg));
1835 		if (error)
1836 			return (error);
1837 		break;
1838 	}
1839 
1840 	switch (uap->cmd) {
1841 	case SEM_STAT:
1842 	case IPC_STAT:
1843 		semun.buf = &dsbuf;
1844 		break;
1845 	case IPC_SET:
1846 		error = copyin(arg.buf, &dsold, sizeof(dsold));
1847 		if (error)
1848 			return (error);
1849 		ipcperm_old2new(&dsold.sem_perm, &dsbuf.sem_perm);
1850 		CP(dsold, dsbuf, __sem_base);
1851 		CP(dsold, dsbuf, sem_nsems);
1852 		CP(dsold, dsbuf, sem_otime);
1853 		CP(dsold, dsbuf, sem_ctime);
1854 		semun.buf = &dsbuf;
1855 		break;
1856 	case GETALL:
1857 	case SETALL:
1858 		semun.array = arg.array;
1859 		break;
1860 	case SETVAL:
1861 		semun.val = arg.val;
1862 		break;
1863 	}
1864 
1865 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1866 	    &rval);
1867 	if (error)
1868 		return (error);
1869 
1870 	switch (uap->cmd) {
1871 	case SEM_STAT:
1872 	case IPC_STAT:
1873 		bzero(&dsold, sizeof(dsold));
1874 		ipcperm_new2old(&dsbuf.sem_perm, &dsold.sem_perm);
1875 		CP(dsbuf, dsold, __sem_base);
1876 		CP(dsbuf, dsold, sem_nsems);
1877 		CP(dsbuf, dsold, sem_otime);
1878 		CP(dsbuf, dsold, sem_ctime);
1879 		error = copyout(&dsold, arg.buf, sizeof(dsold));
1880 		break;
1881 	}
1882 
1883 	if (error == 0)
1884 		td->td_retval[0] = rval;
1885 	return (error);
1886 }
1887 
1888 #endif /* COMPAT_FREEBSD{4,5,6,7} */
1889 
1890 #ifdef COMPAT_FREEBSD32
1891 
1892 int
1893 freebsd32_semsys(struct thread *td, struct freebsd32_semsys_args *uap)
1894 {
1895 
1896 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1897     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1898 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1899 	switch (uap->which) {
1900 	case 0:
1901 		return (freebsd7_freebsd32___semctl(td,
1902 		    (struct freebsd7_freebsd32___semctl_args *)&uap->a2));
1903 	default:
1904 		return (sys_semsys(td, (struct semsys_args *)uap));
1905 	}
1906 #else
1907 	return (nosys(td, NULL));
1908 #endif
1909 }
1910 
1911 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1912     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1913 int
1914 freebsd7_freebsd32___semctl(struct thread *td,
1915     struct freebsd7_freebsd32___semctl_args *uap)
1916 {
1917 	struct semid_ds_old32 dsbuf32;
1918 	struct semid_ds dsbuf;
1919 	union semun semun;
1920 	union semun_old32 arg;
1921 	register_t rval;
1922 	int error;
1923 
1924 	switch (uap->cmd) {
1925 	case SEM_STAT:
1926 	case IPC_SET:
1927 	case IPC_STAT:
1928 	case GETALL:
1929 	case SETVAL:
1930 	case SETALL:
1931 		error = copyin(uap->arg, &arg, sizeof(arg));
1932 		if (error)
1933 			return (error);
1934 		break;
1935 	}
1936 
1937 	switch (uap->cmd) {
1938 	case SEM_STAT:
1939 	case IPC_STAT:
1940 		semun.buf = &dsbuf;
1941 		break;
1942 	case IPC_SET:
1943 		error = copyin(PTRIN(arg.buf), &dsbuf32, sizeof(dsbuf32));
1944 		if (error)
1945 			return (error);
1946 		freebsd32_ipcperm_old_in(&dsbuf32.sem_perm, &dsbuf.sem_perm);
1947 		PTRIN_CP(dsbuf32, dsbuf, __sem_base);
1948 		CP(dsbuf32, dsbuf, sem_nsems);
1949 		CP(dsbuf32, dsbuf, sem_otime);
1950 		CP(dsbuf32, dsbuf, sem_ctime);
1951 		semun.buf = &dsbuf;
1952 		break;
1953 	case GETALL:
1954 	case SETALL:
1955 		semun.array = PTRIN(arg.array);
1956 		break;
1957 	case SETVAL:
1958 		semun.val = arg.val;
1959 		break;
1960 	}
1961 
1962 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1963 	    &rval);
1964 	if (error)
1965 		return (error);
1966 
1967 	switch (uap->cmd) {
1968 	case SEM_STAT:
1969 	case IPC_STAT:
1970 		bzero(&dsbuf32, sizeof(dsbuf32));
1971 		freebsd32_ipcperm_old_out(&dsbuf.sem_perm, &dsbuf32.sem_perm);
1972 		PTROUT_CP(dsbuf, dsbuf32, __sem_base);
1973 		CP(dsbuf, dsbuf32, sem_nsems);
1974 		CP(dsbuf, dsbuf32, sem_otime);
1975 		CP(dsbuf, dsbuf32, sem_ctime);
1976 		error = copyout(&dsbuf32, PTRIN(arg.buf), sizeof(dsbuf32));
1977 		break;
1978 	}
1979 
1980 	if (error == 0)
1981 		td->td_retval[0] = rval;
1982 	return (error);
1983 }
1984 #endif
1985 
1986 int
1987 freebsd32___semctl(struct thread *td, struct freebsd32___semctl_args *uap)
1988 {
1989 	struct semid_ds32 dsbuf32;
1990 	struct semid_ds dsbuf;
1991 	union semun semun;
1992 	union semun32 arg;
1993 	register_t rval;
1994 	int error;
1995 
1996 	switch (uap->cmd) {
1997 	case SEM_STAT:
1998 	case IPC_SET:
1999 	case IPC_STAT:
2000 	case GETALL:
2001 	case SETVAL:
2002 	case SETALL:
2003 		error = copyin(uap->arg, &arg, sizeof(arg));
2004 		if (error)
2005 			return (error);
2006 		break;
2007 	}
2008 
2009 	switch (uap->cmd) {
2010 	case SEM_STAT:
2011 	case IPC_STAT:
2012 		semun.buf = &dsbuf;
2013 		break;
2014 	case IPC_SET:
2015 		error = copyin(PTRIN(arg.buf), &dsbuf32, sizeof(dsbuf32));
2016 		if (error)
2017 			return (error);
2018 		freebsd32_ipcperm_in(&dsbuf32.sem_perm, &dsbuf.sem_perm);
2019 		PTRIN_CP(dsbuf32, dsbuf, __sem_base);
2020 		CP(dsbuf32, dsbuf, sem_nsems);
2021 		CP(dsbuf32, dsbuf, sem_otime);
2022 		CP(dsbuf32, dsbuf, sem_ctime);
2023 		semun.buf = &dsbuf;
2024 		break;
2025 	case GETALL:
2026 	case SETALL:
2027 		semun.array = PTRIN(arg.array);
2028 		break;
2029 	case SETVAL:
2030 		semun.val = arg.val;
2031 		break;
2032 	}
2033 
2034 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
2035 	    &rval);
2036 	if (error)
2037 		return (error);
2038 
2039 	switch (uap->cmd) {
2040 	case SEM_STAT:
2041 	case IPC_STAT:
2042 		bzero(&dsbuf32, sizeof(dsbuf32));
2043 		freebsd32_ipcperm_out(&dsbuf.sem_perm, &dsbuf32.sem_perm);
2044 		PTROUT_CP(dsbuf, dsbuf32, __sem_base);
2045 		CP(dsbuf, dsbuf32, sem_nsems);
2046 		CP(dsbuf, dsbuf32, sem_otime);
2047 		CP(dsbuf, dsbuf32, sem_ctime);
2048 		error = copyout(&dsbuf32, PTRIN(arg.buf), sizeof(dsbuf32));
2049 		break;
2050 	}
2051 
2052 	if (error == 0)
2053 		td->td_retval[0] = rval;
2054 	return (error);
2055 }
2056 
2057 #endif /* COMPAT_FREEBSD32 */
2058