xref: /freebsd/sys/kern/sysv_sem.c (revision a4bcd20486f8c20cc875b39bc75aa0d5a047373f)
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_sysvipc.h"
51 
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/sysproto.h>
55 #include <sys/abi_compat.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 =	SEMMNI,	/* # of semaphore identifiers */
203 	.semmns =	SEMMNS,	/* # of semaphores in system */
204 	.semmnu =	SEMMNU,	/* # of undo structures in system */
205 	.semmsl =	SEMMSL,	/* max # of semaphores per id */
206 	.semopm =	SEMOPM,	/* max # of operations per semop call */
207 	.semume =	SEMUME,	/* max # of undo entries per process */
208 	.semusz =	SEMUSZ,	/* size in bytes of undo structure */
209 	.semvmx =	SEMVMX,	/* semaphore maximum value */
210 	.semaem =	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 		break;
385 	case MOD_UNLOAD:
386 		error = semunload();
387 		break;
388 	case MOD_SHUTDOWN:
389 		break;
390 	default:
391 		error = EINVAL;
392 		break;
393 	}
394 	return (error);
395 }
396 
397 static moduledata_t sysvsem_mod = {
398 	"sysvsem",
399 	&sysvsem_modload,
400 	NULL
401 };
402 
403 DECLARE_MODULE(sysvsem, sysvsem_mod, SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
404 MODULE_VERSION(sysvsem, 1);
405 
406 /*
407  * Allocate a new sem_undo structure for a process
408  * (returns ptr to structure or NULL if no more room)
409  */
410 
411 static struct sem_undo *
412 semu_alloc(struct thread *td)
413 {
414 	struct sem_undo *suptr;
415 
416 	SEMUNDO_LOCKASSERT(MA_OWNED);
417 	if ((suptr = LIST_FIRST(&semu_free_list)) == NULL)
418 		return (NULL);
419 	LIST_REMOVE(suptr, un_next);
420 	LIST_INSERT_HEAD(&semu_list, suptr, un_next);
421 	suptr->un_cnt = 0;
422 	suptr->un_proc = td->td_proc;
423 	return (suptr);
424 }
425 
426 static int
427 semu_try_free(struct sem_undo *suptr)
428 {
429 
430 	SEMUNDO_LOCKASSERT(MA_OWNED);
431 
432 	if (suptr->un_cnt != 0)
433 		return (0);
434 	LIST_REMOVE(suptr, un_next);
435 	LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
436 	return (1);
437 }
438 
439 /*
440  * Adjust a particular entry for a particular proc
441  */
442 
443 static int
444 semundo_adjust(struct thread *td, struct sem_undo **supptr, int semid,
445     int semseq, int semnum, int adjval)
446 {
447 	struct proc *p = td->td_proc;
448 	struct sem_undo *suptr;
449 	struct undo *sunptr;
450 	int i;
451 
452 	SEMUNDO_LOCKASSERT(MA_OWNED);
453 	/* Look for and remember the sem_undo if the caller doesn't provide
454 	   it */
455 
456 	suptr = *supptr;
457 	if (suptr == NULL) {
458 		LIST_FOREACH(suptr, &semu_list, un_next) {
459 			if (suptr->un_proc == p) {
460 				*supptr = suptr;
461 				break;
462 			}
463 		}
464 		if (suptr == NULL) {
465 			if (adjval == 0)
466 				return(0);
467 			suptr = semu_alloc(td);
468 			if (suptr == NULL)
469 				return (ENOSPC);
470 			*supptr = suptr;
471 		}
472 	}
473 
474 	/*
475 	 * Look for the requested entry and adjust it (delete if adjval becomes
476 	 * 0).
477 	 */
478 	sunptr = &suptr->un_ent[0];
479 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
480 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
481 			continue;
482 		if (adjval != 0) {
483 			adjval += sunptr->un_adjval;
484 			if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
485 				return (ERANGE);
486 		}
487 		sunptr->un_adjval = adjval;
488 		if (sunptr->un_adjval == 0) {
489 			suptr->un_cnt--;
490 			if (i < suptr->un_cnt)
491 				suptr->un_ent[i] =
492 				    suptr->un_ent[suptr->un_cnt];
493 			if (suptr->un_cnt == 0)
494 				semu_try_free(suptr);
495 		}
496 		return (0);
497 	}
498 
499 	/* Didn't find the right entry - create it */
500 	if (adjval == 0)
501 		return (0);
502 	if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
503 		return (ERANGE);
504 	if (suptr->un_cnt != seminfo.semume) {
505 		sunptr = &suptr->un_ent[suptr->un_cnt];
506 		suptr->un_cnt++;
507 		sunptr->un_adjval = adjval;
508 		sunptr->un_id = semid;
509 		sunptr->un_num = semnum;
510 		sunptr->un_seq = semseq;
511 	} else
512 		return (EINVAL);
513 	return (0);
514 }
515 
516 static void
517 semundo_clear(int semid, int semnum)
518 {
519 	struct sem_undo *suptr, *suptr1;
520 	struct undo *sunptr;
521 	int i;
522 
523 	SEMUNDO_LOCKASSERT(MA_OWNED);
524 	LIST_FOREACH_SAFE(suptr, &semu_list, un_next, suptr1) {
525 		sunptr = &suptr->un_ent[0];
526 		for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
527 			if (sunptr->un_id != semid)
528 				continue;
529 			if (semnum == -1 || sunptr->un_num == semnum) {
530 				suptr->un_cnt--;
531 				if (i < suptr->un_cnt) {
532 					suptr->un_ent[i] =
533 					    suptr->un_ent[suptr->un_cnt];
534 					continue;
535 				}
536 				semu_try_free(suptr);
537 			}
538 			if (semnum != -1)
539 				break;
540 		}
541 	}
542 }
543 
544 static int
545 semvalid(int semid, struct prison *rpr, struct semid_kernel *semakptr)
546 {
547 
548 	return ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
549 	    semakptr->u.sem_perm.seq != IPCID_TO_SEQ(semid) ||
550 	    sem_prison_cansee(rpr, semakptr) ? EINVAL : 0);
551 }
552 
553 static void
554 sem_remove(int semidx, struct ucred *cred)
555 {
556 	struct semid_kernel *semakptr;
557 	int i;
558 
559 	KASSERT(semidx >= 0 && semidx < seminfo.semmni,
560 	    ("semidx out of bounds"));
561 	mtx_assert(&sem_mtx, MA_OWNED);
562 	semakptr = &sema[semidx];
563 	KASSERT(semakptr->u.__sem_base - sem + semakptr->u.sem_nsems <= semtot,
564 	    ("sem_remove: sema %d corrupted sem pointer %p %p %d %d",
565 	    semidx, semakptr->u.__sem_base, sem, semakptr->u.sem_nsems,
566 	    semtot));
567 
568 	semakptr->u.sem_perm.cuid = cred ? cred->cr_uid : 0;
569 	semakptr->u.sem_perm.uid = cred ? cred->cr_uid : 0;
570 	semakptr->u.sem_perm.mode = 0;
571 	racct_sub_cred(semakptr->cred, RACCT_NSEM, semakptr->u.sem_nsems);
572 	crfree(semakptr->cred);
573 	semakptr->cred = NULL;
574 	SEMUNDO_LOCK();
575 	semundo_clear(semidx, -1);
576 	SEMUNDO_UNLOCK();
577 #ifdef MAC
578 	mac_sysvsem_cleanup(semakptr);
579 #endif
580 	wakeup(semakptr);
581 	for (i = 0; i < seminfo.semmni; i++) {
582 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
583 		    sema[i].u.__sem_base > semakptr->u.__sem_base)
584 			mtx_lock_flags(&sema_mtx[i], LOP_DUPOK);
585 	}
586 	for (i = semakptr->u.__sem_base - sem + semakptr->u.sem_nsems;
587 	    i < semtot; i++)
588 		sem[i - semakptr->u.sem_nsems] = sem[i];
589 	for (i = 0; i < seminfo.semmni; i++) {
590 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
591 		    sema[i].u.__sem_base > semakptr->u.__sem_base) {
592 			sema[i].u.__sem_base -= semakptr->u.sem_nsems;
593 			mtx_unlock(&sema_mtx[i]);
594 		}
595 	}
596 	semtot -= semakptr->u.sem_nsems;
597 }
598 
599 static struct prison *
600 sem_find_prison(struct ucred *cred)
601 {
602 	struct prison *pr, *rpr;
603 
604 	pr = cred->cr_prison;
605 	prison_lock(pr);
606 	rpr = osd_jail_get(pr, sem_prison_slot);
607 	prison_unlock(pr);
608 	return rpr;
609 }
610 
611 static int
612 sem_prison_cansee(struct prison *rpr, struct semid_kernel *semakptr)
613 {
614 
615 	if (semakptr->cred == NULL ||
616 	    !(rpr == semakptr->cred->cr_prison ||
617 	      prison_ischild(rpr, semakptr->cred->cr_prison)))
618 		return (EINVAL);
619 	return (0);
620 }
621 
622 /*
623  * Note that the user-mode half of this passes a union, not a pointer.
624  */
625 #ifndef _SYS_SYSPROTO_H_
626 struct __semctl_args {
627 	int	semid;
628 	int	semnum;
629 	int	cmd;
630 	union	semun *arg;
631 };
632 #endif
633 int
634 sys___semctl(struct thread *td, struct __semctl_args *uap)
635 {
636 	struct semid_ds dsbuf;
637 	union semun arg, semun;
638 	register_t rval;
639 	int error;
640 
641 	switch (uap->cmd) {
642 	case SEM_STAT:
643 	case IPC_SET:
644 	case IPC_STAT:
645 	case GETALL:
646 	case SETVAL:
647 	case SETALL:
648 		error = copyin(uap->arg, &arg, sizeof(arg));
649 		if (error)
650 			return (error);
651 		break;
652 	}
653 
654 	switch (uap->cmd) {
655 	case SEM_STAT:
656 	case IPC_STAT:
657 		semun.buf = &dsbuf;
658 		break;
659 	case IPC_SET:
660 		error = copyin(arg.buf, &dsbuf, sizeof(dsbuf));
661 		if (error)
662 			return (error);
663 		semun.buf = &dsbuf;
664 		break;
665 	case GETALL:
666 	case SETALL:
667 		semun.array = arg.array;
668 		break;
669 	case SETVAL:
670 		semun.val = arg.val;
671 		break;
672 	}
673 
674 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
675 	    &rval);
676 	if (error)
677 		return (error);
678 
679 	switch (uap->cmd) {
680 	case SEM_STAT:
681 	case IPC_STAT:
682 		error = copyout(&dsbuf, arg.buf, sizeof(dsbuf));
683 		break;
684 	}
685 
686 	if (error == 0)
687 		td->td_retval[0] = rval;
688 	return (error);
689 }
690 
691 int
692 kern_semctl(struct thread *td, int semid, int semnum, int cmd,
693     union semun *arg, register_t *rval)
694 {
695 	u_short *array;
696 	struct ucred *cred = td->td_ucred;
697 	int i, error;
698 	struct prison *rpr;
699 	struct semid_ds *sbuf;
700 	struct semid_kernel *semakptr;
701 	struct mtx *sema_mtxp;
702 	u_short usval, count;
703 	int semidx;
704 
705 	DPRINTF(("call to semctl(%d, %d, %d, 0x%p)\n",
706 	    semid, semnum, cmd, arg));
707 
708 	AUDIT_ARG_SVIPC_CMD(cmd);
709 	AUDIT_ARG_SVIPC_ID(semid);
710 
711 	rpr = sem_find_prison(td->td_ucred);
712 	if (sem == NULL)
713 		return (ENOSYS);
714 
715 	array = NULL;
716 
717 	switch(cmd) {
718 	case SEM_STAT:
719 		/*
720 		 * For this command we assume semid is an array index
721 		 * rather than an IPC id.
722 		 */
723 		if (semid < 0 || semid >= seminfo.semmni)
724 			return (EINVAL);
725 		semakptr = &sema[semid];
726 		sema_mtxp = &sema_mtx[semid];
727 		mtx_lock(sema_mtxp);
728 		if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
729 			error = EINVAL;
730 			goto done2;
731 		}
732 		if ((error = sem_prison_cansee(rpr, semakptr)))
733 			goto done2;
734 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
735 			goto done2;
736 #ifdef MAC
737 		error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
738 		if (error != 0)
739 			goto done2;
740 #endif
741 		bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
742 		if (cred->cr_prison != semakptr->cred->cr_prison)
743 			arg->buf->sem_perm.key = IPC_PRIVATE;
744 		*rval = IXSEQ_TO_IPCID(semid, semakptr->u.sem_perm);
745 		mtx_unlock(sema_mtxp);
746 		return (0);
747 	}
748 
749 	semidx = IPCID_TO_IX(semid);
750 	if (semidx < 0 || semidx >= seminfo.semmni)
751 		return (EINVAL);
752 
753 	semakptr = &sema[semidx];
754 	sema_mtxp = &sema_mtx[semidx];
755 	if (cmd == IPC_RMID)
756 		mtx_lock(&sem_mtx);
757 	mtx_lock(sema_mtxp);
758 
759 #ifdef MAC
760 	error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
761 	if (error != 0)
762 		goto done2;
763 #endif
764 
765 	error = 0;
766 	*rval = 0;
767 
768 	switch (cmd) {
769 	case IPC_RMID:
770 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
771 			goto done2;
772 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
773 			goto done2;
774 		sem_remove(semidx, cred);
775 		break;
776 
777 	case IPC_SET:
778 		AUDIT_ARG_SVIPC_PERM(&arg->buf->sem_perm);
779 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
780 			goto done2;
781 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
782 			goto done2;
783 		sbuf = arg->buf;
784 		semakptr->u.sem_perm.uid = sbuf->sem_perm.uid;
785 		semakptr->u.sem_perm.gid = sbuf->sem_perm.gid;
786 		semakptr->u.sem_perm.mode = (semakptr->u.sem_perm.mode &
787 		    ~0777) | (sbuf->sem_perm.mode & 0777);
788 		semakptr->u.sem_ctime = time_second;
789 		break;
790 
791 	case IPC_STAT:
792 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
793 			goto done2;
794 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
795 			goto done2;
796 		bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
797 		if (cred->cr_prison != semakptr->cred->cr_prison)
798 			arg->buf->sem_perm.key = IPC_PRIVATE;
799 
800 		/*
801 		 * Try to hide the fact that the structure layout is shared by
802 		 * both the kernel and userland.  This pointer is not useful to
803 		 * userspace.
804 		 */
805 		arg->buf->__sem_base = NULL;
806 		break;
807 
808 	case GETNCNT:
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].semncnt;
818 		break;
819 
820 	case GETPID:
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].sempid;
830 		break;
831 
832 	case GETVAL:
833 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
834 			goto done2;
835 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
836 			goto done2;
837 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
838 			error = EINVAL;
839 			goto done2;
840 		}
841 		*rval = semakptr->u.__sem_base[semnum].semval;
842 		break;
843 
844 	case GETALL:
845 		/*
846 		 * Unfortunately, callers of this function don't know
847 		 * in advance how many semaphores are in this set.
848 		 * While we could just allocate the maximum size array
849 		 * and pass the actual size back to the caller, that
850 		 * won't work for SETALL since we can't copyin() more
851 		 * data than the user specified as we may return a
852 		 * spurious EFAULT.
853 		 *
854 		 * Note that the number of semaphores in a set is
855 		 * fixed for the life of that set.  The only way that
856 		 * the 'count' could change while are blocked in
857 		 * malloc() is if this semaphore set were destroyed
858 		 * and a new one created with the same index.
859 		 * However, semvalid() will catch that due to the
860 		 * sequence number unless exactly 0x8000 (or a
861 		 * multiple thereof) semaphore sets for the same index
862 		 * are created and destroyed while we are in malloc!
863 		 *
864 		 */
865 		count = semakptr->u.sem_nsems;
866 		mtx_unlock(sema_mtxp);
867 		array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
868 		mtx_lock(sema_mtxp);
869 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
870 			goto done2;
871 		KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
872 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
873 			goto done2;
874 		for (i = 0; i < semakptr->u.sem_nsems; i++)
875 			array[i] = semakptr->u.__sem_base[i].semval;
876 		mtx_unlock(sema_mtxp);
877 		error = copyout(array, arg->array, count * sizeof(*array));
878 		mtx_lock(sema_mtxp);
879 		break;
880 
881 	case GETZCNT:
882 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
883 			goto done2;
884 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
885 			goto done2;
886 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
887 			error = EINVAL;
888 			goto done2;
889 		}
890 		*rval = semakptr->u.__sem_base[semnum].semzcnt;
891 		break;
892 
893 	case SETVAL:
894 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
895 			goto done2;
896 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
897 			goto done2;
898 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
899 			error = EINVAL;
900 			goto done2;
901 		}
902 		if (arg->val < 0 || arg->val > seminfo.semvmx) {
903 			error = ERANGE;
904 			goto done2;
905 		}
906 		semakptr->u.__sem_base[semnum].semval = arg->val;
907 		SEMUNDO_LOCK();
908 		semundo_clear(semidx, semnum);
909 		SEMUNDO_UNLOCK();
910 		wakeup(semakptr);
911 		break;
912 
913 	case SETALL:
914 		/*
915 		 * See comment on GETALL for why 'count' shouldn't change
916 		 * and why we require a userland buffer.
917 		 */
918 		count = semakptr->u.sem_nsems;
919 		mtx_unlock(sema_mtxp);
920 		array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
921 		error = copyin(arg->array, array, count * sizeof(*array));
922 		mtx_lock(sema_mtxp);
923 		if (error)
924 			break;
925 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
926 			goto done2;
927 		KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
928 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
929 			goto done2;
930 		for (i = 0; i < semakptr->u.sem_nsems; i++) {
931 			usval = array[i];
932 			if (usval > seminfo.semvmx) {
933 				error = ERANGE;
934 				break;
935 			}
936 			semakptr->u.__sem_base[i].semval = usval;
937 		}
938 		SEMUNDO_LOCK();
939 		semundo_clear(semidx, -1);
940 		SEMUNDO_UNLOCK();
941 		wakeup(semakptr);
942 		break;
943 
944 	default:
945 		error = EINVAL;
946 		break;
947 	}
948 
949 done2:
950 	mtx_unlock(sema_mtxp);
951 	if (cmd == IPC_RMID)
952 		mtx_unlock(&sem_mtx);
953 	if (array != NULL)
954 		free(array, M_TEMP);
955 	return(error);
956 }
957 
958 #ifndef _SYS_SYSPROTO_H_
959 struct semget_args {
960 	key_t	key;
961 	int	nsems;
962 	int	semflg;
963 };
964 #endif
965 int
966 sys_semget(struct thread *td, struct semget_args *uap)
967 {
968 	int semid, error = 0;
969 	int key = uap->key;
970 	int nsems = uap->nsems;
971 	int semflg = uap->semflg;
972 	struct ucred *cred = td->td_ucred;
973 
974 	DPRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg));
975 
976 	AUDIT_ARG_VALUE(semflg);
977 
978 	if (sem_find_prison(cred) == NULL)
979 		return (ENOSYS);
980 
981 	mtx_lock(&sem_mtx);
982 	if (key != IPC_PRIVATE) {
983 		for (semid = 0; semid < seminfo.semmni; semid++) {
984 			if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) &&
985 			    sema[semid].cred != NULL &&
986 			    sema[semid].cred->cr_prison == cred->cr_prison &&
987 			    sema[semid].u.sem_perm.key == key)
988 				break;
989 		}
990 		if (semid < seminfo.semmni) {
991 			AUDIT_ARG_SVIPC_ID(semid);
992 			DPRINTF(("found public key\n"));
993 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
994 				DPRINTF(("not exclusive\n"));
995 				error = EEXIST;
996 				goto done2;
997 			}
998 			if ((error = ipcperm(td, &sema[semid].u.sem_perm,
999 			    semflg & 0700))) {
1000 				goto done2;
1001 			}
1002 			if (nsems > 0 && sema[semid].u.sem_nsems < nsems) {
1003 				DPRINTF(("too small\n"));
1004 				error = EINVAL;
1005 				goto done2;
1006 			}
1007 #ifdef MAC
1008 			error = mac_sysvsem_check_semget(cred, &sema[semid]);
1009 			if (error != 0)
1010 				goto done2;
1011 #endif
1012 			goto found;
1013 		}
1014 	}
1015 
1016 	DPRINTF(("need to allocate the semid_kernel\n"));
1017 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
1018 		if (nsems <= 0 || nsems > seminfo.semmsl) {
1019 			DPRINTF(("nsems out of range (0<%d<=%d)\n", nsems,
1020 			    seminfo.semmsl));
1021 			error = EINVAL;
1022 			goto done2;
1023 		}
1024 		if (nsems > seminfo.semmns - semtot) {
1025 			DPRINTF((
1026 			    "not enough semaphores left (need %d, got %d)\n",
1027 			    nsems, seminfo.semmns - semtot));
1028 			error = ENOSPC;
1029 			goto done2;
1030 		}
1031 		for (semid = 0; semid < seminfo.semmni; semid++) {
1032 			if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0)
1033 				break;
1034 		}
1035 		if (semid == seminfo.semmni) {
1036 			DPRINTF(("no more semid_kernel's available\n"));
1037 			error = ENOSPC;
1038 			goto done2;
1039 		}
1040 #ifdef RACCT
1041 		if (racct_enable) {
1042 			PROC_LOCK(td->td_proc);
1043 			error = racct_add(td->td_proc, RACCT_NSEM, nsems);
1044 			PROC_UNLOCK(td->td_proc);
1045 			if (error != 0) {
1046 				error = ENOSPC;
1047 				goto done2;
1048 			}
1049 		}
1050 #endif
1051 		DPRINTF(("semid %d is available\n", semid));
1052 		mtx_lock(&sema_mtx[semid]);
1053 		KASSERT((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0,
1054 		    ("Lost semaphore %d", semid));
1055 		sema[semid].u.sem_perm.key = key;
1056 		sema[semid].u.sem_perm.cuid = cred->cr_uid;
1057 		sema[semid].u.sem_perm.uid = cred->cr_uid;
1058 		sema[semid].u.sem_perm.cgid = cred->cr_gid;
1059 		sema[semid].u.sem_perm.gid = cred->cr_gid;
1060 		sema[semid].u.sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
1061 		sema[semid].cred = crhold(cred);
1062 		sema[semid].u.sem_perm.seq =
1063 		    (sema[semid].u.sem_perm.seq + 1) & 0x7fff;
1064 		sema[semid].u.sem_nsems = nsems;
1065 		sema[semid].u.sem_otime = 0;
1066 		sema[semid].u.sem_ctime = time_second;
1067 		sema[semid].u.__sem_base = &sem[semtot];
1068 		semtot += nsems;
1069 		bzero(sema[semid].u.__sem_base,
1070 		    sizeof(sema[semid].u.__sem_base[0])*nsems);
1071 #ifdef MAC
1072 		mac_sysvsem_create(cred, &sema[semid]);
1073 #endif
1074 		mtx_unlock(&sema_mtx[semid]);
1075 		DPRINTF(("sembase = %p, next = %p\n",
1076 		    sema[semid].u.__sem_base, &sem[semtot]));
1077 	} else {
1078 		DPRINTF(("didn't find it and wasn't asked to create it\n"));
1079 		error = ENOENT;
1080 		goto done2;
1081 	}
1082 
1083 found:
1084 	td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].u.sem_perm);
1085 done2:
1086 	mtx_unlock(&sem_mtx);
1087 	return (error);
1088 }
1089 
1090 #ifndef _SYS_SYSPROTO_H_
1091 struct semop_args {
1092 	int	semid;
1093 	struct	sembuf *sops;
1094 	size_t	nsops;
1095 };
1096 #endif
1097 int
1098 sys_semop(struct thread *td, struct semop_args *uap)
1099 {
1100 #define SMALL_SOPS	8
1101 	struct sembuf small_sops[SMALL_SOPS];
1102 	int semid = uap->semid;
1103 	size_t nsops = uap->nsops;
1104 	struct prison *rpr;
1105 	struct sembuf *sops;
1106 	struct semid_kernel *semakptr;
1107 	struct sembuf *sopptr = NULL;
1108 	struct sem *semptr = NULL;
1109 	struct sem_undo *suptr;
1110 	struct mtx *sema_mtxp;
1111 	size_t i, j, k;
1112 	int error;
1113 	int do_wakeup, do_undos;
1114 	unsigned short seq;
1115 
1116 #ifdef SEM_DEBUG
1117 	sops = NULL;
1118 #endif
1119 	DPRINTF(("call to semop(%d, %p, %u)\n", semid, sops, nsops));
1120 
1121 	AUDIT_ARG_SVIPC_ID(semid);
1122 
1123 	rpr = sem_find_prison(td->td_ucred);
1124 	if (sem == NULL)
1125 		return (ENOSYS);
1126 
1127 	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
1128 
1129 	if (semid < 0 || semid >= seminfo.semmni)
1130 		return (EINVAL);
1131 
1132 	/* Allocate memory for sem_ops */
1133 	if (nsops <= SMALL_SOPS)
1134 		sops = small_sops;
1135 	else if (nsops > seminfo.semopm) {
1136 		DPRINTF(("too many sops (max=%d, nsops=%d)\n", seminfo.semopm,
1137 		    nsops));
1138 		return (E2BIG);
1139 	} else {
1140 #ifdef RACCT
1141 		if (racct_enable) {
1142 			PROC_LOCK(td->td_proc);
1143 			if (nsops >
1144 			    racct_get_available(td->td_proc, RACCT_NSEMOP)) {
1145 				PROC_UNLOCK(td->td_proc);
1146 				return (E2BIG);
1147 			}
1148 			PROC_UNLOCK(td->td_proc);
1149 		}
1150 #endif
1151 
1152 		sops = malloc(nsops * sizeof(*sops), M_TEMP, M_WAITOK);
1153 	}
1154 	if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) {
1155 		DPRINTF(("error = %d from copyin(%p, %p, %d)\n", error,
1156 		    uap->sops, sops, nsops * sizeof(sops[0])));
1157 		if (sops != small_sops)
1158 			free(sops, M_TEMP);
1159 		return (error);
1160 	}
1161 
1162 	semakptr = &sema[semid];
1163 	sema_mtxp = &sema_mtx[semid];
1164 	mtx_lock(sema_mtxp);
1165 	if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
1166 		error = EINVAL;
1167 		goto done2;
1168 	}
1169 	seq = semakptr->u.sem_perm.seq;
1170 	if (seq != IPCID_TO_SEQ(uap->semid)) {
1171 		error = EINVAL;
1172 		goto done2;
1173 	}
1174 	if ((error = sem_prison_cansee(rpr, semakptr)) != 0)
1175 		goto done2;
1176 	/*
1177 	 * Initial pass through sops to see what permissions are needed.
1178 	 * Also perform any checks that don't need repeating on each
1179 	 * attempt to satisfy the request vector.
1180 	 */
1181 	j = 0;		/* permission needed */
1182 	do_undos = 0;
1183 	for (i = 0; i < nsops; i++) {
1184 		sopptr = &sops[i];
1185 		if (sopptr->sem_num >= semakptr->u.sem_nsems) {
1186 			error = EFBIG;
1187 			goto done2;
1188 		}
1189 		if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0)
1190 			do_undos = 1;
1191 		j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A;
1192 	}
1193 
1194 	if ((error = ipcperm(td, &semakptr->u.sem_perm, j))) {
1195 		DPRINTF(("error = %d from ipaccess\n", error));
1196 		goto done2;
1197 	}
1198 #ifdef MAC
1199 	error = mac_sysvsem_check_semop(td->td_ucred, semakptr, j);
1200 	if (error != 0)
1201 		goto done2;
1202 #endif
1203 
1204 	/*
1205 	 * Loop trying to satisfy the vector of requests.
1206 	 * If we reach a point where we must wait, any requests already
1207 	 * performed are rolled back and we go to sleep until some other
1208 	 * process wakes us up.  At this point, we start all over again.
1209 	 *
1210 	 * This ensures that from the perspective of other tasks, a set
1211 	 * of requests is atomic (never partially satisfied).
1212 	 */
1213 	for (;;) {
1214 		do_wakeup = 0;
1215 		error = 0;	/* error return if necessary */
1216 
1217 		for (i = 0; i < nsops; i++) {
1218 			sopptr = &sops[i];
1219 			semptr = &semakptr->u.__sem_base[sopptr->sem_num];
1220 
1221 			DPRINTF((
1222 			    "semop:  semakptr=%p, __sem_base=%p, "
1223 			    "semptr=%p, sem[%d]=%d : op=%d, flag=%s\n",
1224 			    semakptr, semakptr->u.__sem_base, semptr,
1225 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
1226 			    (sopptr->sem_flg & IPC_NOWAIT) ?
1227 			    "nowait" : "wait"));
1228 
1229 			if (sopptr->sem_op < 0) {
1230 				if (semptr->semval + sopptr->sem_op < 0) {
1231 					DPRINTF(("semop:  can't do it now\n"));
1232 					break;
1233 				} else {
1234 					semptr->semval += sopptr->sem_op;
1235 					if (semptr->semval == 0 &&
1236 					    semptr->semzcnt > 0)
1237 						do_wakeup = 1;
1238 				}
1239 			} else if (sopptr->sem_op == 0) {
1240 				if (semptr->semval != 0) {
1241 					DPRINTF(("semop:  not zero now\n"));
1242 					break;
1243 				}
1244 			} else if (semptr->semval + sopptr->sem_op >
1245 			    seminfo.semvmx) {
1246 				error = ERANGE;
1247 				break;
1248 			} else {
1249 				if (semptr->semncnt > 0)
1250 					do_wakeup = 1;
1251 				semptr->semval += sopptr->sem_op;
1252 			}
1253 		}
1254 
1255 		/*
1256 		 * Did we get through the entire vector?
1257 		 */
1258 		if (i >= nsops)
1259 			goto done;
1260 
1261 		/*
1262 		 * No ... rollback anything that we've already done
1263 		 */
1264 		DPRINTF(("semop:  rollback 0 through %d\n", i-1));
1265 		for (j = 0; j < i; j++)
1266 			semakptr->u.__sem_base[sops[j].sem_num].semval -=
1267 			    sops[j].sem_op;
1268 
1269 		/* If we detected an error, return it */
1270 		if (error != 0)
1271 			goto done2;
1272 
1273 		/*
1274 		 * If the request that we couldn't satisfy has the
1275 		 * NOWAIT flag set then return with EAGAIN.
1276 		 */
1277 		if (sopptr->sem_flg & IPC_NOWAIT) {
1278 			error = EAGAIN;
1279 			goto done2;
1280 		}
1281 
1282 		if (sopptr->sem_op == 0)
1283 			semptr->semzcnt++;
1284 		else
1285 			semptr->semncnt++;
1286 
1287 		DPRINTF(("semop:  good night!\n"));
1288 		error = msleep(semakptr, sema_mtxp, (PZERO - 4) | PCATCH,
1289 		    "semwait", 0);
1290 		DPRINTF(("semop:  good morning (error=%d)!\n", error));
1291 		/* return code is checked below, after sem[nz]cnt-- */
1292 
1293 		/*
1294 		 * Make sure that the semaphore still exists
1295 		 */
1296 		seq = semakptr->u.sem_perm.seq;
1297 		if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1298 		    seq != IPCID_TO_SEQ(uap->semid)) {
1299 			error = EIDRM;
1300 			goto done2;
1301 		}
1302 
1303 		/*
1304 		 * Renew the semaphore's pointer after wakeup since
1305 		 * during msleep __sem_base may have been modified and semptr
1306 		 * is not valid any more
1307 		 */
1308 		semptr = &semakptr->u.__sem_base[sopptr->sem_num];
1309 
1310 		/*
1311 		 * The semaphore is still alive.  Readjust the count of
1312 		 * waiting processes.
1313 		 */
1314 		if (sopptr->sem_op == 0)
1315 			semptr->semzcnt--;
1316 		else
1317 			semptr->semncnt--;
1318 
1319 		/*
1320 		 * Is it really morning, or was our sleep interrupted?
1321 		 * (Delayed check of msleep() return code because we
1322 		 * need to decrement sem[nz]cnt either way.)
1323 		 */
1324 		if (error != 0) {
1325 			error = EINTR;
1326 			goto done2;
1327 		}
1328 		DPRINTF(("semop:  good morning!\n"));
1329 	}
1330 
1331 done:
1332 	/*
1333 	 * Process any SEM_UNDO requests.
1334 	 */
1335 	if (do_undos) {
1336 		SEMUNDO_LOCK();
1337 		suptr = NULL;
1338 		for (i = 0; i < nsops; i++) {
1339 			/*
1340 			 * We only need to deal with SEM_UNDO's for non-zero
1341 			 * op's.
1342 			 */
1343 			int adjval;
1344 
1345 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
1346 				continue;
1347 			adjval = sops[i].sem_op;
1348 			if (adjval == 0)
1349 				continue;
1350 			error = semundo_adjust(td, &suptr, semid, seq,
1351 			    sops[i].sem_num, -adjval);
1352 			if (error == 0)
1353 				continue;
1354 
1355 			/*
1356 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
1357 			 * Rollback the adjustments to this point and then
1358 			 * rollback the semaphore ups and down so we can return
1359 			 * with an error with all structures restored.  We
1360 			 * rollback the undo's in the exact reverse order that
1361 			 * we applied them.  This guarantees that we won't run
1362 			 * out of space as we roll things back out.
1363 			 */
1364 			for (j = 0; j < i; j++) {
1365 				k = i - j - 1;
1366 				if ((sops[k].sem_flg & SEM_UNDO) == 0)
1367 					continue;
1368 				adjval = sops[k].sem_op;
1369 				if (adjval == 0)
1370 					continue;
1371 				if (semundo_adjust(td, &suptr, semid, seq,
1372 				    sops[k].sem_num, adjval) != 0)
1373 					panic("semop - can't undo undos");
1374 			}
1375 
1376 			for (j = 0; j < nsops; j++)
1377 				semakptr->u.__sem_base[sops[j].sem_num].semval -=
1378 				    sops[j].sem_op;
1379 
1380 			DPRINTF(("error = %d from semundo_adjust\n", error));
1381 			SEMUNDO_UNLOCK();
1382 			goto done2;
1383 		} /* loop through the sops */
1384 		SEMUNDO_UNLOCK();
1385 	} /* if (do_undos) */
1386 
1387 	/* We're definitely done - set the sempid's and time */
1388 	for (i = 0; i < nsops; i++) {
1389 		sopptr = &sops[i];
1390 		semptr = &semakptr->u.__sem_base[sopptr->sem_num];
1391 		semptr->sempid = td->td_proc->p_pid;
1392 	}
1393 	semakptr->u.sem_otime = time_second;
1394 
1395 	/*
1396 	 * Do a wakeup if any semaphore was up'd whilst something was
1397 	 * sleeping on it.
1398 	 */
1399 	if (do_wakeup) {
1400 		DPRINTF(("semop:  doing wakeup\n"));
1401 		wakeup(semakptr);
1402 		DPRINTF(("semop:  back from wakeup\n"));
1403 	}
1404 	DPRINTF(("semop:  done\n"));
1405 	td->td_retval[0] = 0;
1406 done2:
1407 	mtx_unlock(sema_mtxp);
1408 	if (sops != small_sops)
1409 		free(sops, M_TEMP);
1410 	return (error);
1411 }
1412 
1413 /*
1414  * Go through the undo structures for this process and apply the adjustments to
1415  * semaphores.
1416  */
1417 static void
1418 semexit_myhook(void *arg, struct proc *p)
1419 {
1420 	struct sem_undo *suptr;
1421 	struct semid_kernel *semakptr;
1422 	struct mtx *sema_mtxp;
1423 	int semid, semnum, adjval, ix;
1424 	unsigned short seq;
1425 
1426 	/*
1427 	 * Go through the chain of undo vectors looking for one
1428 	 * associated with this process.
1429 	 */
1430 	if (LIST_EMPTY(&semu_list))
1431 		return;
1432 	SEMUNDO_LOCK();
1433 	LIST_FOREACH(suptr, &semu_list, un_next) {
1434 		if (suptr->un_proc == p)
1435 			break;
1436 	}
1437 	if (suptr == NULL) {
1438 		SEMUNDO_UNLOCK();
1439 		return;
1440 	}
1441 	LIST_REMOVE(suptr, un_next);
1442 
1443 	DPRINTF(("proc @%p has undo structure with %d entries\n", p,
1444 	    suptr->un_cnt));
1445 
1446 	/*
1447 	 * If there are any active undo elements then process them.
1448 	 */
1449 	if (suptr->un_cnt > 0) {
1450 		SEMUNDO_UNLOCK();
1451 		for (ix = 0; ix < suptr->un_cnt; ix++) {
1452 			semid = suptr->un_ent[ix].un_id;
1453 			semnum = suptr->un_ent[ix].un_num;
1454 			adjval = suptr->un_ent[ix].un_adjval;
1455 			seq = suptr->un_ent[ix].un_seq;
1456 			semakptr = &sema[semid];
1457 			sema_mtxp = &sema_mtx[semid];
1458 
1459 			mtx_lock(sema_mtxp);
1460 			if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1461 			    (semakptr->u.sem_perm.seq != seq)) {
1462 				mtx_unlock(sema_mtxp);
1463 				continue;
1464 			}
1465 			if (semnum >= semakptr->u.sem_nsems)
1466 				panic("semexit - semnum out of range");
1467 
1468 			DPRINTF((
1469 			    "semexit:  %p id=%d num=%d(adj=%d) ; sem=%d\n",
1470 			    suptr->un_proc, suptr->un_ent[ix].un_id,
1471 			    suptr->un_ent[ix].un_num,
1472 			    suptr->un_ent[ix].un_adjval,
1473 			    semakptr->u.__sem_base[semnum].semval));
1474 
1475 			if (adjval < 0 && semakptr->u.__sem_base[semnum].semval <
1476 			    -adjval)
1477 				semakptr->u.__sem_base[semnum].semval = 0;
1478 			else
1479 				semakptr->u.__sem_base[semnum].semval += adjval;
1480 
1481 			wakeup(semakptr);
1482 			DPRINTF(("semexit:  back from wakeup\n"));
1483 			mtx_unlock(sema_mtxp);
1484 		}
1485 		SEMUNDO_LOCK();
1486 	}
1487 
1488 	/*
1489 	 * Deallocate the undo vector.
1490 	 */
1491 	DPRINTF(("removing vector\n"));
1492 	suptr->un_proc = NULL;
1493 	suptr->un_cnt = 0;
1494 	LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
1495 	SEMUNDO_UNLOCK();
1496 }
1497 
1498 static int
1499 sysctl_sema(SYSCTL_HANDLER_ARGS)
1500 {
1501 	struct prison *pr, *rpr;
1502 	struct semid_kernel tsemak;
1503 #ifdef COMPAT_FREEBSD32
1504 	struct semid_kernel32 tsemak32;
1505 #endif
1506 	void *outaddr;
1507 	size_t outsize;
1508 	int error, i;
1509 
1510 	pr = req->td->td_ucred->cr_prison;
1511 	rpr = sem_find_prison(req->td->td_ucred);
1512 	error = 0;
1513 	for (i = 0; i < seminfo.semmni; i++) {
1514 		mtx_lock(&sema_mtx[i]);
1515 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) == 0 ||
1516 		    rpr == NULL || sem_prison_cansee(rpr, &sema[i]) != 0)
1517 			bzero(&tsemak, sizeof(tsemak));
1518 		else {
1519 			tsemak = sema[i];
1520 			if (tsemak.cred->cr_prison != pr)
1521 				tsemak.u.sem_perm.key = IPC_PRIVATE;
1522 		}
1523 		mtx_unlock(&sema_mtx[i]);
1524 #ifdef COMPAT_FREEBSD32
1525 		if (SV_CURPROC_FLAG(SV_ILP32)) {
1526 			bzero(&tsemak32, sizeof(tsemak32));
1527 			freebsd32_ipcperm_out(&tsemak.u.sem_perm,
1528 			    &tsemak32.u.sem_perm);
1529 			/* Don't copy u.__sem_base */
1530 			CP(tsemak, tsemak32, u.sem_nsems);
1531 			CP(tsemak, tsemak32, u.sem_otime);
1532 			CP(tsemak, tsemak32, u.sem_ctime);
1533 			/* Don't copy label or cred */
1534 			outaddr = &tsemak32;
1535 			outsize = sizeof(tsemak32);
1536 		} else
1537 #endif
1538 		{
1539 			tsemak.u.__sem_base = NULL;
1540 			tsemak.label = NULL;
1541 			tsemak.cred = NULL;
1542 			outaddr = &tsemak;
1543 			outsize = sizeof(tsemak);
1544 		}
1545 		error = SYSCTL_OUT(req, outaddr, outsize);
1546 		if (error != 0)
1547 			break;
1548 	}
1549 	return (error);
1550 }
1551 
1552 static int
1553 sem_prison_check(void *obj, void *data)
1554 {
1555 	struct prison *pr = obj;
1556 	struct prison *prpr;
1557 	struct vfsoptlist *opts = data;
1558 	int error, jsys;
1559 
1560 	/*
1561 	 * sysvsem is a jailsys integer.
1562 	 * It must be "disable" if the parent jail is disabled.
1563 	 */
1564 	error = vfs_copyopt(opts, "sysvsem", &jsys, sizeof(jsys));
1565 	if (error != ENOENT) {
1566 		if (error != 0)
1567 			return (error);
1568 		switch (jsys) {
1569 		case JAIL_SYS_DISABLE:
1570 			break;
1571 		case JAIL_SYS_NEW:
1572 		case JAIL_SYS_INHERIT:
1573 			prison_lock(pr->pr_parent);
1574 			prpr = osd_jail_get(pr->pr_parent, sem_prison_slot);
1575 			prison_unlock(pr->pr_parent);
1576 			if (prpr == NULL)
1577 				return (EPERM);
1578 			break;
1579 		default:
1580 			return (EINVAL);
1581 		}
1582 	}
1583 
1584 	return (0);
1585 }
1586 
1587 static int
1588 sem_prison_set(void *obj, void *data)
1589 {
1590 	struct prison *pr = obj;
1591 	struct prison *tpr, *orpr, *nrpr, *trpr;
1592 	struct vfsoptlist *opts = data;
1593 	void *rsv;
1594 	int jsys, descend;
1595 
1596 	/*
1597 	 * sysvsem controls which jail is the root of the associated sems (this
1598 	 * jail or same as the parent), or if the feature is available at all.
1599 	 */
1600 	if (vfs_copyopt(opts, "sysvsem", &jsys, sizeof(jsys)) == ENOENT)
1601 		jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0)
1602 		    ? JAIL_SYS_INHERIT
1603 		    : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0)
1604 		    ? JAIL_SYS_DISABLE
1605 		    : -1;
1606 	if (jsys == JAIL_SYS_DISABLE) {
1607 		prison_lock(pr);
1608 		orpr = osd_jail_get(pr, sem_prison_slot);
1609 		if (orpr != NULL)
1610 			osd_jail_del(pr, sem_prison_slot);
1611 		prison_unlock(pr);
1612 		if (orpr != NULL) {
1613 			if (orpr == pr)
1614 				sem_prison_cleanup(pr);
1615 			/* Disable all child jails as well. */
1616 			FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1617 				prison_lock(tpr);
1618 				trpr = osd_jail_get(tpr, sem_prison_slot);
1619 				if (trpr != NULL) {
1620 					osd_jail_del(tpr, sem_prison_slot);
1621 					prison_unlock(tpr);
1622 					if (trpr == tpr)
1623 						sem_prison_cleanup(tpr);
1624 				} else {
1625 					prison_unlock(tpr);
1626 					descend = 0;
1627 				}
1628 			}
1629 		}
1630 	} else if (jsys != -1) {
1631 		if (jsys == JAIL_SYS_NEW)
1632 			nrpr = pr;
1633 		else {
1634 			prison_lock(pr->pr_parent);
1635 			nrpr = osd_jail_get(pr->pr_parent, sem_prison_slot);
1636 			prison_unlock(pr->pr_parent);
1637 		}
1638 		rsv = osd_reserve(sem_prison_slot);
1639 		prison_lock(pr);
1640 		orpr = osd_jail_get(pr, sem_prison_slot);
1641 		if (orpr != nrpr)
1642 			(void)osd_jail_set_reserved(pr, sem_prison_slot, rsv,
1643 			    nrpr);
1644 		else
1645 			osd_free_reserved(rsv);
1646 		prison_unlock(pr);
1647 		if (orpr != nrpr) {
1648 			if (orpr == pr)
1649 				sem_prison_cleanup(pr);
1650 			if (orpr != NULL) {
1651 				/* Change child jails matching the old root, */
1652 				FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1653 					prison_lock(tpr);
1654 					trpr = osd_jail_get(tpr,
1655 					    sem_prison_slot);
1656 					if (trpr == orpr) {
1657 						(void)osd_jail_set(tpr,
1658 						    sem_prison_slot, nrpr);
1659 						prison_unlock(tpr);
1660 						if (trpr == tpr)
1661 							sem_prison_cleanup(tpr);
1662 					} else {
1663 						prison_unlock(tpr);
1664 						descend = 0;
1665 					}
1666 				}
1667 			}
1668 		}
1669 	}
1670 
1671 	return (0);
1672 }
1673 
1674 static int
1675 sem_prison_get(void *obj, void *data)
1676 {
1677 	struct prison *pr = obj;
1678 	struct prison *rpr;
1679 	struct vfsoptlist *opts = data;
1680 	int error, jsys;
1681 
1682 	/* Set sysvsem based on the jail's root prison. */
1683 	prison_lock(pr);
1684 	rpr = osd_jail_get(pr, sem_prison_slot);
1685 	prison_unlock(pr);
1686 	jsys = rpr == NULL ? JAIL_SYS_DISABLE
1687 	    : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT;
1688 	error = vfs_setopt(opts, "sysvsem", &jsys, sizeof(jsys));
1689 	if (error == ENOENT)
1690 		error = 0;
1691 	return (error);
1692 }
1693 
1694 static int
1695 sem_prison_remove(void *obj, void *data __unused)
1696 {
1697 	struct prison *pr = obj;
1698 	struct prison *rpr;
1699 
1700 	prison_lock(pr);
1701 	rpr = osd_jail_get(pr, sem_prison_slot);
1702 	prison_unlock(pr);
1703 	if (rpr == pr)
1704 		sem_prison_cleanup(pr);
1705 	return (0);
1706 }
1707 
1708 static void
1709 sem_prison_cleanup(struct prison *pr)
1710 {
1711 	int i;
1712 
1713 	/* Remove any sems that belong to this jail. */
1714 	mtx_lock(&sem_mtx);
1715 	for (i = 0; i < seminfo.semmni; i++) {
1716 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
1717 		    sema[i].cred != NULL && sema[i].cred->cr_prison == pr) {
1718 			mtx_lock(&sema_mtx[i]);
1719 			sem_remove(i, NULL);
1720 			mtx_unlock(&sema_mtx[i]);
1721 		}
1722 	}
1723 	mtx_unlock(&sem_mtx);
1724 }
1725 
1726 SYSCTL_JAIL_PARAM_SYS_NODE(sysvsem, CTLFLAG_RW, "SYSV semaphores");
1727 
1728 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1729     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1730 
1731 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1732 static sy_call_t *semcalls[] = {
1733 	(sy_call_t *)freebsd7___semctl, (sy_call_t *)sys_semget,
1734 	(sy_call_t *)sys_semop
1735 };
1736 
1737 /*
1738  * Entry point for all SEM calls.
1739  */
1740 int
1741 sys_semsys(td, uap)
1742 	struct thread *td;
1743 	/* XXX actually varargs. */
1744 	struct semsys_args /* {
1745 		int	which;
1746 		int	a2;
1747 		int	a3;
1748 		int	a4;
1749 		int	a5;
1750 	} */ *uap;
1751 {
1752 	int error;
1753 
1754 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1755 	if (uap->which < 0 || uap->which >= nitems(semcalls))
1756 		return (EINVAL);
1757 	error = (*semcalls[uap->which])(td, &uap->a2);
1758 	return (error);
1759 }
1760 
1761 #ifndef _SYS_SYSPROTO_H_
1762 struct freebsd7___semctl_args {
1763 	int	semid;
1764 	int	semnum;
1765 	int	cmd;
1766 	union	semun_old *arg;
1767 };
1768 #endif
1769 int
1770 freebsd7___semctl(struct thread *td, struct freebsd7___semctl_args *uap)
1771 {
1772 	struct semid_ds_old dsold;
1773 	struct semid_ds dsbuf;
1774 	union semun_old arg;
1775 	union semun semun;
1776 	register_t rval;
1777 	int error;
1778 
1779 	switch (uap->cmd) {
1780 	case SEM_STAT:
1781 	case IPC_SET:
1782 	case IPC_STAT:
1783 	case GETALL:
1784 	case SETVAL:
1785 	case SETALL:
1786 		error = copyin(uap->arg, &arg, sizeof(arg));
1787 		if (error)
1788 			return (error);
1789 		break;
1790 	}
1791 
1792 	switch (uap->cmd) {
1793 	case SEM_STAT:
1794 	case IPC_STAT:
1795 		semun.buf = &dsbuf;
1796 		break;
1797 	case IPC_SET:
1798 		error = copyin(arg.buf, &dsold, sizeof(dsold));
1799 		if (error)
1800 			return (error);
1801 		ipcperm_old2new(&dsold.sem_perm, &dsbuf.sem_perm);
1802 		CP(dsold, dsbuf, __sem_base);
1803 		CP(dsold, dsbuf, sem_nsems);
1804 		CP(dsold, dsbuf, sem_otime);
1805 		CP(dsold, dsbuf, sem_ctime);
1806 		semun.buf = &dsbuf;
1807 		break;
1808 	case GETALL:
1809 	case SETALL:
1810 		semun.array = arg.array;
1811 		break;
1812 	case SETVAL:
1813 		semun.val = arg.val;
1814 		break;
1815 	}
1816 
1817 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1818 	    &rval);
1819 	if (error)
1820 		return (error);
1821 
1822 	switch (uap->cmd) {
1823 	case SEM_STAT:
1824 	case IPC_STAT:
1825 		bzero(&dsold, sizeof(dsold));
1826 		ipcperm_new2old(&dsbuf.sem_perm, &dsold.sem_perm);
1827 		CP(dsbuf, dsold, __sem_base);
1828 		CP(dsbuf, dsold, sem_nsems);
1829 		CP(dsbuf, dsold, sem_otime);
1830 		CP(dsbuf, dsold, sem_ctime);
1831 		error = copyout(&dsold, arg.buf, sizeof(dsold));
1832 		break;
1833 	}
1834 
1835 	if (error == 0)
1836 		td->td_retval[0] = rval;
1837 	return (error);
1838 }
1839 
1840 #endif /* COMPAT_FREEBSD{4,5,6,7} */
1841 
1842 #ifdef COMPAT_FREEBSD32
1843 
1844 int
1845 freebsd32_semsys(struct thread *td, struct freebsd32_semsys_args *uap)
1846 {
1847 
1848 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1849     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1850 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1851 	switch (uap->which) {
1852 	case 0:
1853 		return (freebsd7_freebsd32_semctl(td,
1854 		    (struct freebsd7_freebsd32_semctl_args *)&uap->a2));
1855 	default:
1856 		return (sys_semsys(td, (struct semsys_args *)uap));
1857 	}
1858 #else
1859 	return (nosys(td, NULL));
1860 #endif
1861 }
1862 
1863 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1864     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1865 int
1866 freebsd7_freebsd32_semctl(struct thread *td,
1867     struct freebsd7_freebsd32_semctl_args *uap)
1868 {
1869 	struct semid_ds32_old dsbuf32;
1870 	struct semid_ds dsbuf;
1871 	union semun semun;
1872 	union semun32 arg;
1873 	register_t rval;
1874 	int error;
1875 
1876 	switch (uap->cmd) {
1877 	case SEM_STAT:
1878 	case IPC_SET:
1879 	case IPC_STAT:
1880 	case GETALL:
1881 	case SETVAL:
1882 	case SETALL:
1883 		error = copyin(uap->arg, &arg, sizeof(arg));
1884 		if (error)
1885 			return (error);
1886 		break;
1887 	}
1888 
1889 	switch (uap->cmd) {
1890 	case SEM_STAT:
1891 	case IPC_STAT:
1892 		semun.buf = &dsbuf;
1893 		break;
1894 	case IPC_SET:
1895 		error = copyin(PTRIN(arg.buf), &dsbuf32, sizeof(dsbuf32));
1896 		if (error)
1897 			return (error);
1898 		freebsd32_ipcperm_old_in(&dsbuf32.sem_perm, &dsbuf.sem_perm);
1899 		PTRIN_CP(dsbuf32, dsbuf, __sem_base);
1900 		CP(dsbuf32, dsbuf, sem_nsems);
1901 		CP(dsbuf32, dsbuf, sem_otime);
1902 		CP(dsbuf32, dsbuf, sem_ctime);
1903 		semun.buf = &dsbuf;
1904 		break;
1905 	case GETALL:
1906 	case SETALL:
1907 		semun.array = PTRIN(arg.array);
1908 		break;
1909 	case SETVAL:
1910 		semun.val = arg.val;
1911 		break;
1912 	}
1913 
1914 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1915 	    &rval);
1916 	if (error)
1917 		return (error);
1918 
1919 	switch (uap->cmd) {
1920 	case SEM_STAT:
1921 	case IPC_STAT:
1922 		bzero(&dsbuf32, sizeof(dsbuf32));
1923 		freebsd32_ipcperm_old_out(&dsbuf.sem_perm, &dsbuf32.sem_perm);
1924 		PTROUT_CP(dsbuf, dsbuf32, __sem_base);
1925 		CP(dsbuf, dsbuf32, sem_nsems);
1926 		CP(dsbuf, dsbuf32, sem_otime);
1927 		CP(dsbuf, dsbuf32, sem_ctime);
1928 		error = copyout(&dsbuf32, PTRIN(arg.buf), sizeof(dsbuf32));
1929 		break;
1930 	}
1931 
1932 	if (error == 0)
1933 		td->td_retval[0] = rval;
1934 	return (error);
1935 }
1936 #endif
1937 
1938 int
1939 freebsd32_semctl(struct thread *td, struct freebsd32_semctl_args *uap)
1940 {
1941 	struct semid_ds32 dsbuf32;
1942 	struct semid_ds dsbuf;
1943 	union semun semun;
1944 	union semun32 arg;
1945 	register_t rval;
1946 	int error;
1947 
1948 	switch (uap->cmd) {
1949 	case SEM_STAT:
1950 	case IPC_SET:
1951 	case IPC_STAT:
1952 	case GETALL:
1953 	case SETVAL:
1954 	case SETALL:
1955 		error = copyin(uap->arg, &arg, sizeof(arg));
1956 		if (error)
1957 			return (error);
1958 		break;
1959 	}
1960 
1961 	switch (uap->cmd) {
1962 	case SEM_STAT:
1963 	case IPC_STAT:
1964 		semun.buf = &dsbuf;
1965 		break;
1966 	case IPC_SET:
1967 		error = copyin(PTRIN(arg.buf), &dsbuf32, sizeof(dsbuf32));
1968 		if (error)
1969 			return (error);
1970 		freebsd32_ipcperm_in(&dsbuf32.sem_perm, &dsbuf.sem_perm);
1971 		PTRIN_CP(dsbuf32, dsbuf, __sem_base);
1972 		CP(dsbuf32, dsbuf, sem_nsems);
1973 		CP(dsbuf32, dsbuf, sem_otime);
1974 		CP(dsbuf32, dsbuf, sem_ctime);
1975 		semun.buf = &dsbuf;
1976 		break;
1977 	case GETALL:
1978 	case SETALL:
1979 		semun.array = PTRIN(arg.array);
1980 		break;
1981 	case SETVAL:
1982 		semun.val = arg.val;
1983 		break;
1984 	}
1985 
1986 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1987 	    &rval);
1988 	if (error)
1989 		return (error);
1990 
1991 	switch (uap->cmd) {
1992 	case SEM_STAT:
1993 	case IPC_STAT:
1994 		bzero(&dsbuf32, sizeof(dsbuf32));
1995 		freebsd32_ipcperm_out(&dsbuf.sem_perm, &dsbuf32.sem_perm);
1996 		PTROUT_CP(dsbuf, dsbuf32, __sem_base);
1997 		CP(dsbuf, dsbuf32, sem_nsems);
1998 		CP(dsbuf, dsbuf32, sem_otime);
1999 		CP(dsbuf, dsbuf32, sem_ctime);
2000 		error = copyout(&dsbuf32, PTRIN(arg.buf), sizeof(dsbuf32));
2001 		break;
2002 	}
2003 
2004 	if (error == 0)
2005 		td->td_retval[0] = rval;
2006 	return (error);
2007 }
2008 
2009 #endif /* COMPAT_FREEBSD32 */
2010