xref: /freebsd/sys/kern/sysv_sem.c (revision 7c5dd794f2cc6c5ba2675cfc451308dd493b078e)
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, uint64_t 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 uint64_t *sema_seq;	/* semaphore id sequence numbers */
119 static struct sem *sem;		/* semaphore pool */
120 LIST_HEAD(, sem_undo) semu_list;	/* list of active undo structures */
121 LIST_HEAD(, sem_undo) semu_free_list;	/* list of free undo structures */
122 static int	*semu;		/* undo structure pool */
123 static eventhandler_tag semexit_tag;
124 static unsigned sem_prison_slot;	/* prison OSD slot */
125 
126 #define SEMUNDO_MTX		sem_undo_mtx
127 #define SEMUNDO_LOCK()		mtx_lock(&SEMUNDO_MTX);
128 #define SEMUNDO_UNLOCK()	mtx_unlock(&SEMUNDO_MTX);
129 #define SEMUNDO_LOCKASSERT(how)	mtx_assert(&SEMUNDO_MTX, (how));
130 
131 struct sem {
132 	u_short	semval;		/* semaphore value */
133 	pid_t	sempid;		/* pid of last operation */
134 	u_short	semncnt;	/* # awaiting semval > cval */
135 	u_short	semzcnt;	/* # awaiting semval = 0 */
136 };
137 
138 /*
139  * Undo structure (one per process)
140  */
141 struct sem_undo {
142 	LIST_ENTRY(sem_undo) un_next;	/* ptr to next active undo structure */
143 	struct	proc *un_proc;		/* owner of this structure */
144 	short	un_cnt;			/* # of active entries */
145 	struct undo {
146 		short	un_adjval;	/* adjust on exit values */
147 		short	un_num;		/* semaphore # */
148 		int	un_id;		/* semid */
149 		uint64_t un_seq;
150 	} un_ent[1];			/* undo entries */
151 };
152 
153 /*
154  * Configuration parameters
155  */
156 #ifndef SEMMNI
157 #define SEMMNI	50		/* # of semaphore identifiers */
158 #endif
159 #ifndef SEMMNS
160 #define SEMMNS	340		/* # of semaphores in system */
161 #endif
162 #ifndef SEMUME
163 #define SEMUME	50		/* max # of undo entries per process */
164 #endif
165 #ifndef SEMMNU
166 #define SEMMNU	150		/* # of undo structures in system */
167 #endif
168 
169 /* shouldn't need tuning */
170 #ifndef SEMMSL
171 #define SEMMSL	SEMMNS		/* max # of semaphores per id */
172 #endif
173 #ifndef SEMOPM
174 #define SEMOPM	100		/* max # of operations per semop call */
175 #endif
176 
177 #define SEMVMX	32767		/* semaphore maximum value */
178 #define SEMAEM	16384		/* adjust on exit max value */
179 
180 /*
181  * Due to the way semaphore memory is allocated, we have to ensure that
182  * SEMUSZ is properly aligned.
183  */
184 
185 #define	SEM_ALIGN(bytes) roundup2(bytes, sizeof(long))
186 
187 /* actual size of an undo structure */
188 #define SEMUSZ(x)	SEM_ALIGN(offsetof(struct sem_undo, un_ent[(x)]))
189 
190 /*
191  * Macro to find a particular sem_undo vector
192  */
193 #define SEMU(ix) \
194 	((struct sem_undo *)(((intptr_t)semu) + (ix) * seminfo.semusz))
195 
196 /*
197  * semaphore info struct
198  */
199 struct seminfo seminfo = {
200 	.semmni =	SEMMNI,	/* # of semaphore identifiers */
201 	.semmns =	SEMMNS,	/* # of semaphores in system */
202 	.semmnu =	SEMMNU,	/* # of undo structures in system */
203 	.semmsl =	SEMMSL,	/* max # of semaphores per id */
204 	.semopm =	SEMOPM,	/* max # of operations per semop call */
205 	.semume =	SEMUME,	/* max # of undo entries per process */
206 	.semusz =	SEMUSZ(SEMUME),	/* size in bytes of undo structure */
207 	.semvmx =	SEMVMX,	/* semaphore maximum value */
208 	.semaem =	SEMAEM,	/* adjust on exit max value */
209 };
210 
211 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RDTUN, &seminfo.semmni, 0,
212     "Number of semaphore identifiers");
213 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RDTUN, &seminfo.semmns, 0,
214     "Maximum number of semaphores in the system");
215 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RDTUN, &seminfo.semmnu, 0,
216     "Maximum number of undo structures in the system");
217 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RWTUN, &seminfo.semmsl, 0,
218     "Max semaphores per id");
219 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RDTUN, &seminfo.semopm, 0,
220     "Max operations per semop call");
221 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RDTUN, &seminfo.semume, 0,
222     "Max undo entries per process");
223 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RD, &seminfo.semusz, 0,
224     "Size in bytes of undo structure");
225 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RWTUN, &seminfo.semvmx, 0,
226     "Semaphore maximum value");
227 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RWTUN, &seminfo.semaem, 0,
228     "Adjust on exit max value");
229 SYSCTL_PROC(_kern_ipc, OID_AUTO, sema,
230     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
231     NULL, 0, sysctl_sema, "",
232     "Array of struct semid_kernel for each potential semaphore");
233 
234 static struct syscall_helper_data sem_syscalls[] = {
235 	SYSCALL_INIT_HELPER(__semctl),
236 	SYSCALL_INIT_HELPER(semget),
237 	SYSCALL_INIT_HELPER(semop),
238 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
239     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
240 	SYSCALL_INIT_HELPER(semsys),
241 	SYSCALL_INIT_HELPER_COMPAT(freebsd7___semctl),
242 #endif
243 	SYSCALL_INIT_LAST
244 };
245 
246 #ifdef COMPAT_FREEBSD32
247 #include <compat/freebsd32/freebsd32.h>
248 #include <compat/freebsd32/freebsd32_ipc.h>
249 #include <compat/freebsd32/freebsd32_proto.h>
250 #include <compat/freebsd32/freebsd32_signal.h>
251 #include <compat/freebsd32/freebsd32_syscall.h>
252 #include <compat/freebsd32/freebsd32_util.h>
253 
254 static struct syscall_helper_data sem32_syscalls[] = {
255 	SYSCALL32_INIT_HELPER(freebsd32___semctl),
256 	SYSCALL32_INIT_HELPER_COMPAT(semget),
257 	SYSCALL32_INIT_HELPER_COMPAT(semop),
258 	SYSCALL32_INIT_HELPER(freebsd32_semsys),
259 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
260     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
261 	SYSCALL32_INIT_HELPER(freebsd7_freebsd32___semctl),
262 #endif
263 	SYSCALL_INIT_LAST
264 };
265 #endif
266 
267 static int
268 seminit(void)
269 {
270 	struct prison *pr;
271 	void **rsv;
272 	int i, error;
273 	osd_method_t methods[PR_MAXMETHOD] = {
274 	    [PR_METHOD_CHECK] =		sem_prison_check,
275 	    [PR_METHOD_SET] =		sem_prison_set,
276 	    [PR_METHOD_GET] =		sem_prison_get,
277 	    [PR_METHOD_REMOVE] =	sem_prison_remove,
278 	};
279 
280 	sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
281 	sema = malloc(sizeof(struct semid_kernel) * seminfo.semmni, M_SEM,
282 	    M_WAITOK | M_ZERO);
283 	sema_mtx = malloc(sizeof(struct mtx) * seminfo.semmni, M_SEM,
284 	    M_WAITOK | M_ZERO);
285 	sema_seq = malloc(sizeof(uint64_t) * seminfo.semmni, M_SEM,
286 	    M_WAITOK | M_ZERO);
287 	seminfo.semusz = SEMUSZ(seminfo.semume);
288 	semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
289 
290 	for (i = 0; i < seminfo.semmni; i++) {
291 		sema[i].u.__sem_base = 0;
292 		sema[i].u.sem_perm.mode = 0;
293 		sema[i].u.sem_perm.seq = 0;
294 #ifdef MAC
295 		mac_sysvsem_init(&sema[i]);
296 #endif
297 	}
298 	for (i = 0; i < seminfo.semmni; i++)
299 		mtx_init(&sema_mtx[i], "semid", NULL, MTX_DEF);
300 	LIST_INIT(&semu_free_list);
301 	for (i = 0; i < seminfo.semmnu; i++) {
302 		struct sem_undo *suptr = SEMU(i);
303 		suptr->un_proc = NULL;
304 		LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
305 	}
306 	LIST_INIT(&semu_list);
307 	mtx_init(&sem_mtx, "sem", NULL, MTX_DEF);
308 	mtx_init(&sem_undo_mtx, "semu", NULL, MTX_DEF);
309 	semexit_tag = EVENTHANDLER_REGISTER(process_exit, semexit_myhook, NULL,
310 	    EVENTHANDLER_PRI_ANY);
311 
312 	/* Set current prisons according to their allow.sysvipc. */
313 	sem_prison_slot = osd_jail_register(NULL, methods);
314 	rsv = osd_reserve(sem_prison_slot);
315 	prison_lock(&prison0);
316 	(void)osd_jail_set_reserved(&prison0, sem_prison_slot, rsv, &prison0);
317 	prison_unlock(&prison0);
318 	rsv = NULL;
319 	sx_slock(&allprison_lock);
320 	TAILQ_FOREACH(pr, &allprison, pr_list) {
321 		if (rsv == NULL)
322 			rsv = osd_reserve(sem_prison_slot);
323 		prison_lock(pr);
324 		if (pr->pr_allow & PR_ALLOW_SYSVIPC) {
325 			(void)osd_jail_set_reserved(pr, sem_prison_slot, rsv,
326 			    &prison0);
327 			rsv = NULL;
328 		}
329 		prison_unlock(pr);
330 	}
331 	if (rsv != NULL)
332 		osd_free_reserved(rsv);
333 	sx_sunlock(&allprison_lock);
334 
335 	error = syscall_helper_register(sem_syscalls, SY_THR_STATIC_KLD);
336 	if (error != 0)
337 		return (error);
338 #ifdef COMPAT_FREEBSD32
339 	error = syscall32_helper_register(sem32_syscalls, SY_THR_STATIC_KLD);
340 	if (error != 0)
341 		return (error);
342 #endif
343 	return (0);
344 }
345 
346 static int
347 semunload(void)
348 {
349 	int i;
350 
351 	/* XXXKIB */
352 	if (semtot != 0)
353 		return (EBUSY);
354 
355 #ifdef COMPAT_FREEBSD32
356 	syscall32_helper_unregister(sem32_syscalls);
357 #endif
358 	syscall_helper_unregister(sem_syscalls);
359 	EVENTHANDLER_DEREGISTER(process_exit, semexit_tag);
360 	if (sem_prison_slot != 0)
361 		osd_jail_deregister(sem_prison_slot);
362 #ifdef MAC
363 	for (i = 0; i < seminfo.semmni; i++)
364 		mac_sysvsem_destroy(&sema[i]);
365 #endif
366 	free(sem, M_SEM);
367 	free(sema, M_SEM);
368 	free(semu, M_SEM);
369 	for (i = 0; i < seminfo.semmni; i++)
370 		mtx_destroy(&sema_mtx[i]);
371 	free(sema_mtx, M_SEM);
372 	free(sema_seq, M_SEM);
373 	mtx_destroy(&sem_mtx);
374 	mtx_destroy(&sem_undo_mtx);
375 	return (0);
376 }
377 
378 static int
379 sysvsem_modload(struct module *module, int cmd, void *arg)
380 {
381 	int error = 0;
382 
383 	switch (cmd) {
384 	case MOD_LOAD:
385 		error = seminit();
386 		break;
387 	case MOD_UNLOAD:
388 		error = semunload();
389 		break;
390 	case MOD_SHUTDOWN:
391 		break;
392 	default:
393 		error = EINVAL;
394 		break;
395 	}
396 	return (error);
397 }
398 
399 static moduledata_t sysvsem_mod = {
400 	"sysvsem",
401 	&sysvsem_modload,
402 	NULL
403 };
404 
405 DECLARE_MODULE(sysvsem, sysvsem_mod, SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
406 MODULE_VERSION(sysvsem, 1);
407 
408 /*
409  * Allocate a new sem_undo structure for a process
410  * (returns ptr to structure or NULL if no more room)
411  */
412 
413 static struct sem_undo *
414 semu_alloc(struct thread *td)
415 {
416 	struct sem_undo *suptr;
417 
418 	SEMUNDO_LOCKASSERT(MA_OWNED);
419 	if ((suptr = LIST_FIRST(&semu_free_list)) == NULL)
420 		return (NULL);
421 	LIST_REMOVE(suptr, un_next);
422 	LIST_INSERT_HEAD(&semu_list, suptr, un_next);
423 	suptr->un_cnt = 0;
424 	suptr->un_proc = td->td_proc;
425 	return (suptr);
426 }
427 
428 static int
429 semu_try_free(struct sem_undo *suptr)
430 {
431 
432 	SEMUNDO_LOCKASSERT(MA_OWNED);
433 
434 	if (suptr->un_cnt != 0)
435 		return (0);
436 	LIST_REMOVE(suptr, un_next);
437 	LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
438 	return (1);
439 }
440 
441 /*
442  * Adjust a particular entry for a particular proc
443  */
444 
445 static int
446 semundo_adjust(struct thread *td, struct sem_undo **supptr, int semid,
447     uint64_t semseq, int semnum, int adjval)
448 {
449 	struct proc *p = td->td_proc;
450 	struct sem_undo *suptr;
451 	struct undo *sunptr;
452 	int i;
453 
454 	SEMUNDO_LOCKASSERT(MA_OWNED);
455 	/* Look for and remember the sem_undo if the caller doesn't provide
456 	   it */
457 
458 	suptr = *supptr;
459 	if (suptr == NULL) {
460 		LIST_FOREACH(suptr, &semu_list, un_next) {
461 			if (suptr->un_proc == p) {
462 				*supptr = suptr;
463 				break;
464 			}
465 		}
466 		if (suptr == NULL) {
467 			if (adjval == 0)
468 				return(0);
469 			suptr = semu_alloc(td);
470 			if (suptr == NULL)
471 				return (ENOSPC);
472 			*supptr = suptr;
473 		}
474 	}
475 
476 	/*
477 	 * Look for the requested entry and adjust it (delete if adjval becomes
478 	 * 0).
479 	 */
480 	sunptr = &suptr->un_ent[0];
481 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
482 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
483 			continue;
484 		if (adjval != 0) {
485 			adjval += sunptr->un_adjval;
486 			if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
487 				return (ERANGE);
488 		}
489 		sunptr->un_adjval = adjval;
490 		if (sunptr->un_adjval == 0) {
491 			suptr->un_cnt--;
492 			if (i < suptr->un_cnt)
493 				suptr->un_ent[i] =
494 				    suptr->un_ent[suptr->un_cnt];
495 			if (suptr->un_cnt == 0)
496 				semu_try_free(suptr);
497 		}
498 		return (0);
499 	}
500 
501 	/* Didn't find the right entry - create it */
502 	if (adjval == 0)
503 		return (0);
504 	if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
505 		return (ERANGE);
506 	if (suptr->un_cnt != seminfo.semume) {
507 		sunptr = &suptr->un_ent[suptr->un_cnt];
508 		suptr->un_cnt++;
509 		sunptr->un_adjval = adjval;
510 		sunptr->un_id = semid;
511 		sunptr->un_num = semnum;
512 		sunptr->un_seq = semseq;
513 	} else
514 		return (EINVAL);
515 	return (0);
516 }
517 
518 static void
519 semundo_clear(int semid, int semnum)
520 {
521 	struct sem_undo *suptr, *suptr1;
522 	struct undo *sunptr;
523 	int i;
524 
525 	SEMUNDO_LOCKASSERT(MA_OWNED);
526 	LIST_FOREACH_SAFE(suptr, &semu_list, un_next, suptr1) {
527 		sunptr = &suptr->un_ent[0];
528 		for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
529 			if (sunptr->un_id != semid)
530 				continue;
531 			if (semnum == -1 || sunptr->un_num == semnum) {
532 				suptr->un_cnt--;
533 				if (i < suptr->un_cnt) {
534 					suptr->un_ent[i] =
535 					    suptr->un_ent[suptr->un_cnt];
536 					continue;
537 				}
538 				semu_try_free(suptr);
539 			}
540 			if (semnum != -1)
541 				break;
542 		}
543 	}
544 }
545 
546 static int
547 semvalid(int semid, struct prison *rpr, struct semid_kernel *semakptr)
548 {
549 
550 	return ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
551 	    semakptr->u.sem_perm.seq != IPCID_TO_SEQ(semid) ||
552 	    sem_prison_cansee(rpr, semakptr) ? EINVAL : 0);
553 }
554 
555 static void
556 sem_remove(int semidx, struct ucred *cred)
557 {
558 	struct semid_kernel *semakptr;
559 	int i;
560 
561 	KASSERT(semidx >= 0 && semidx < seminfo.semmni,
562 	    ("semidx out of bounds"));
563 	mtx_assert(&sem_mtx, MA_OWNED);
564 	semakptr = &sema[semidx];
565 	KASSERT(semakptr->u.__sem_base - sem + semakptr->u.sem_nsems <= semtot,
566 	    ("sem_remove: sema %d corrupted sem pointer %p %p %d %d",
567 	    semidx, semakptr->u.__sem_base, sem, semakptr->u.sem_nsems,
568 	    semtot));
569 
570 	semakptr->u.sem_perm.cuid = cred ? cred->cr_uid : 0;
571 	semakptr->u.sem_perm.uid = cred ? cred->cr_uid : 0;
572 	semakptr->u.sem_perm.mode = 0;
573 	racct_sub_cred(semakptr->cred, RACCT_NSEM, semakptr->u.sem_nsems);
574 	crfree(semakptr->cred);
575 	semakptr->cred = NULL;
576 	SEMUNDO_LOCK();
577 	semundo_clear(semidx, -1);
578 	SEMUNDO_UNLOCK();
579 #ifdef MAC
580 	mac_sysvsem_cleanup(semakptr);
581 #endif
582 	wakeup(semakptr);
583 	for (i = 0; i < seminfo.semmni; i++) {
584 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
585 		    sema[i].u.__sem_base > semakptr->u.__sem_base)
586 			mtx_lock_flags(&sema_mtx[i], LOP_DUPOK);
587 	}
588 	for (i = semakptr->u.__sem_base - sem + semakptr->u.sem_nsems;
589 	    i < semtot; i++)
590 		sem[i - semakptr->u.sem_nsems] = sem[i];
591 	for (i = 0; i < seminfo.semmni; i++) {
592 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
593 		    sema[i].u.__sem_base > semakptr->u.__sem_base) {
594 			sema[i].u.__sem_base -= semakptr->u.sem_nsems;
595 			mtx_unlock(&sema_mtx[i]);
596 		}
597 	}
598 	semtot -= semakptr->u.sem_nsems;
599 }
600 
601 static struct prison *
602 sem_find_prison(struct ucred *cred)
603 {
604 	struct prison *pr, *rpr;
605 
606 	pr = cred->cr_prison;
607 	prison_lock(pr);
608 	rpr = osd_jail_get(pr, sem_prison_slot);
609 	prison_unlock(pr);
610 	return (rpr);
611 }
612 
613 static int
614 sem_prison_cansee(struct prison *rpr, struct semid_kernel *semakptr)
615 {
616 
617 	if (semakptr->cred == NULL ||
618 	    !(rpr == semakptr->cred->cr_prison ||
619 	      prison_ischild(rpr, semakptr->cred->cr_prison)))
620 		return (EINVAL);
621 	return (0);
622 }
623 
624 /*
625  * Note that the user-mode half of this passes a union, not a pointer.
626  */
627 #ifndef _SYS_SYSPROTO_H_
628 struct __semctl_args {
629 	int	semid;
630 	int	semnum;
631 	int	cmd;
632 	union	semun *arg;
633 };
634 #endif
635 int
636 sys___semctl(struct thread *td, struct __semctl_args *uap)
637 {
638 	struct semid_ds dsbuf;
639 	union semun arg, semun;
640 	register_t rval;
641 	int error;
642 
643 	switch (uap->cmd) {
644 	case SEM_STAT:
645 	case IPC_SET:
646 	case IPC_STAT:
647 	case GETALL:
648 	case SETVAL:
649 	case SETALL:
650 		error = copyin(uap->arg, &arg, sizeof(arg));
651 		if (error)
652 			return (error);
653 		break;
654 	}
655 
656 	switch (uap->cmd) {
657 	case SEM_STAT:
658 	case IPC_STAT:
659 		semun.buf = &dsbuf;
660 		break;
661 	case IPC_SET:
662 		error = copyin(arg.buf, &dsbuf, sizeof(dsbuf));
663 		if (error)
664 			return (error);
665 		semun.buf = &dsbuf;
666 		break;
667 	case GETALL:
668 	case SETALL:
669 		semun.array = arg.array;
670 		break;
671 	case SETVAL:
672 		semun.val = arg.val;
673 		break;
674 	}
675 
676 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
677 	    &rval);
678 	if (error)
679 		return (error);
680 
681 	switch (uap->cmd) {
682 	case SEM_STAT:
683 	case IPC_STAT:
684 		error = copyout(&dsbuf, arg.buf, sizeof(dsbuf));
685 		break;
686 	}
687 
688 	if (error == 0)
689 		td->td_retval[0] = rval;
690 	return (error);
691 }
692 
693 int
694 kern_semctl(struct thread *td, int semid, int semnum, int cmd,
695     union semun *arg, register_t *rval)
696 {
697 	u_short *array;
698 	struct ucred *cred = td->td_ucred;
699 	int i, error;
700 	struct prison *rpr;
701 	struct semid_ds *sbuf;
702 	struct semid_kernel *semakptr;
703 	struct mtx *sema_mtxp;
704 	uint64_t seq;
705 	u_short usval, count;
706 	int semidx;
707 
708 	DPRINTF(("call to semctl(%d, %d, %d, 0x%p)\n",
709 	    semid, semnum, cmd, arg));
710 
711 	AUDIT_ARG_SVIPC_CMD(cmd);
712 	AUDIT_ARG_SVIPC_ID(semid);
713 
714 	rpr = sem_find_prison(td->td_ucred);
715 	if (rpr == NULL)
716 		return (ENOSYS);
717 
718 	array = NULL;
719 
720 	switch(cmd) {
721 	case SEM_STAT:
722 		/*
723 		 * For this command we assume semid is an array index
724 		 * rather than an IPC id.
725 		 */
726 		if (semid < 0 || semid >= seminfo.semmni)
727 			return (EINVAL);
728 		semakptr = &sema[semid];
729 		sema_mtxp = &sema_mtx[semid];
730 		mtx_lock(sema_mtxp);
731 		if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
732 			error = EINVAL;
733 			goto done2;
734 		}
735 		if ((error = sem_prison_cansee(rpr, semakptr)))
736 			goto done2;
737 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
738 			goto done2;
739 #ifdef MAC
740 		error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
741 		if (error != 0)
742 			goto done2;
743 #endif
744 		bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
745 		if (cred->cr_prison != semakptr->cred->cr_prison)
746 			arg->buf->sem_perm.key = IPC_PRIVATE;
747 		*rval = IXSEQ_TO_IPCID(semid, semakptr->u.sem_perm);
748 		mtx_unlock(sema_mtxp);
749 		return (0);
750 	}
751 
752 	semidx = IPCID_TO_IX(semid);
753 	if (semidx < 0 || semidx >= seminfo.semmni)
754 		return (EINVAL);
755 
756 	semakptr = &sema[semidx];
757 	sema_mtxp = &sema_mtx[semidx];
758 	if (cmd == IPC_RMID)
759 		mtx_lock(&sem_mtx);
760 	mtx_lock(sema_mtxp);
761 
762 #ifdef MAC
763 	error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
764 	if (error != 0)
765 		goto done2;
766 #endif
767 
768 	error = 0;
769 	*rval = 0;
770 
771 	switch (cmd) {
772 	case IPC_RMID:
773 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
774 			goto done2;
775 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
776 			goto done2;
777 		sem_remove(semidx, cred);
778 		break;
779 
780 	case IPC_SET:
781 		AUDIT_ARG_SVIPC_PERM(&arg->buf->sem_perm);
782 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
783 			goto done2;
784 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
785 			goto done2;
786 		sbuf = arg->buf;
787 		semakptr->u.sem_perm.uid = sbuf->sem_perm.uid;
788 		semakptr->u.sem_perm.gid = sbuf->sem_perm.gid;
789 		semakptr->u.sem_perm.mode = (semakptr->u.sem_perm.mode &
790 		    ~0777) | (sbuf->sem_perm.mode & 0777);
791 		semakptr->u.sem_ctime = time_second;
792 		break;
793 
794 	case IPC_STAT:
795 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
796 			goto done2;
797 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
798 			goto done2;
799 		bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
800 		if (cred->cr_prison != semakptr->cred->cr_prison)
801 			arg->buf->sem_perm.key = IPC_PRIVATE;
802 
803 		/*
804 		 * Try to hide the fact that the structure layout is shared by
805 		 * both the kernel and userland.  This pointer is not useful to
806 		 * userspace.
807 		 */
808 		arg->buf->__sem_base = NULL;
809 		break;
810 
811 	case GETNCNT:
812 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
813 			goto done2;
814 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
815 			goto done2;
816 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
817 			error = EINVAL;
818 			goto done2;
819 		}
820 		*rval = semakptr->u.__sem_base[semnum].semncnt;
821 		break;
822 
823 	case GETPID:
824 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
825 			goto done2;
826 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
827 			goto done2;
828 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
829 			error = EINVAL;
830 			goto done2;
831 		}
832 		*rval = semakptr->u.__sem_base[semnum].sempid;
833 		break;
834 
835 	case GETVAL:
836 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
837 			goto done2;
838 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
839 			goto done2;
840 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
841 			error = EINVAL;
842 			goto done2;
843 		}
844 		*rval = semakptr->u.__sem_base[semnum].semval;
845 		break;
846 
847 	case GETALL:
848 		/*
849 		 * Unfortunately, callers of this function don't know
850 		 * in advance how many semaphores are in this set.
851 		 * While we could just allocate the maximum size array
852 		 * and pass the actual size back to the caller, that
853 		 * won't work for SETALL since we can't copyin() more
854 		 * data than the user specified as we may return a
855 		 * spurious EFAULT.
856 		 */
857 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
858 			goto done2;
859 		count = semakptr->u.sem_nsems;
860 		seq = sema_seq[semidx];
861 		mtx_unlock(sema_mtxp);
862 		array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
863 		mtx_lock(sema_mtxp);
864 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
865 			goto done2;
866 		if (seq != sema_seq[semidx]) {
867 			error = EAGAIN;
868 			goto done2;
869 		}
870 		KASSERT(count == semakptr->u.sem_nsems,
871 		    ("sem_nsems changed from %d to %d",
872 		    count, semakptr->u.sem_nsems));
873 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
874 			goto done2;
875 		for (i = 0; i < semakptr->u.sem_nsems; i++)
876 			array[i] = semakptr->u.__sem_base[i].semval;
877 		mtx_unlock(sema_mtxp);
878 		error = copyout(array, arg->array, count * sizeof(*array));
879 		mtx_lock(sema_mtxp);
880 		break;
881 
882 	case GETZCNT:
883 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
884 			goto done2;
885 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
886 			goto done2;
887 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
888 			error = EINVAL;
889 			goto done2;
890 		}
891 		*rval = semakptr->u.__sem_base[semnum].semzcnt;
892 		break;
893 
894 	case SETVAL:
895 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
896 			goto done2;
897 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
898 			goto done2;
899 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
900 			error = EINVAL;
901 			goto done2;
902 		}
903 		if (arg->val < 0 || arg->val > seminfo.semvmx) {
904 			error = ERANGE;
905 			goto done2;
906 		}
907 		semakptr->u.__sem_base[semnum].semval = arg->val;
908 		SEMUNDO_LOCK();
909 		semundo_clear(semidx, semnum);
910 		SEMUNDO_UNLOCK();
911 		wakeup(semakptr);
912 		break;
913 
914 	case SETALL:
915 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
916 			goto done2;
917 		count = semakptr->u.sem_nsems;
918 		seq = sema_seq[semidx];
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 		if (seq != sema_seq[semidx]) {
928 			error = EAGAIN;
929 			goto done2;
930 		}
931 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
932 			goto done2;
933 		for (i = 0; i < semakptr->u.sem_nsems; i++) {
934 			usval = array[i];
935 			if (usval > seminfo.semvmx) {
936 				error = ERANGE;
937 				break;
938 			}
939 			semakptr->u.__sem_base[i].semval = usval;
940 		}
941 		SEMUNDO_LOCK();
942 		semundo_clear(semidx, -1);
943 		SEMUNDO_UNLOCK();
944 		wakeup(semakptr);
945 		break;
946 
947 	default:
948 		error = EINVAL;
949 		break;
950 	}
951 
952 done2:
953 	mtx_unlock(sema_mtxp);
954 	if (cmd == IPC_RMID)
955 		mtx_unlock(&sem_mtx);
956 	if (array != NULL)
957 		free(array, M_TEMP);
958 	return(error);
959 }
960 
961 #ifndef _SYS_SYSPROTO_H_
962 struct semget_args {
963 	key_t	key;
964 	int	nsems;
965 	int	semflg;
966 };
967 #endif
968 int
969 sys_semget(struct thread *td, struct semget_args *uap)
970 {
971 	int semid, error = 0;
972 	int key = uap->key;
973 	int nsems = uap->nsems;
974 	int semflg = uap->semflg;
975 	struct ucred *cred = td->td_ucred;
976 
977 	DPRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg));
978 
979 	AUDIT_ARG_VALUE(semflg);
980 
981 	if (sem_find_prison(cred) == NULL)
982 		return (ENOSYS);
983 
984 	mtx_lock(&sem_mtx);
985 	if (key != IPC_PRIVATE) {
986 		for (semid = 0; semid < seminfo.semmni; semid++) {
987 			if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) &&
988 			    sema[semid].cred != NULL &&
989 			    sema[semid].cred->cr_prison == cred->cr_prison &&
990 			    sema[semid].u.sem_perm.key == key)
991 				break;
992 		}
993 		if (semid < seminfo.semmni) {
994 			AUDIT_ARG_SVIPC_ID(semid);
995 			DPRINTF(("found public key\n"));
996 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
997 				DPRINTF(("not exclusive\n"));
998 				error = EEXIST;
999 				goto done2;
1000 			}
1001 			if ((error = ipcperm(td, &sema[semid].u.sem_perm,
1002 			    semflg & 0700))) {
1003 				goto done2;
1004 			}
1005 			if (nsems > 0 && sema[semid].u.sem_nsems < nsems) {
1006 				DPRINTF(("too small\n"));
1007 				error = EINVAL;
1008 				goto done2;
1009 			}
1010 #ifdef MAC
1011 			error = mac_sysvsem_check_semget(cred, &sema[semid]);
1012 			if (error != 0)
1013 				goto done2;
1014 #endif
1015 			goto found;
1016 		}
1017 	}
1018 
1019 	DPRINTF(("need to allocate the semid_kernel\n"));
1020 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
1021 		if (nsems <= 0 || nsems > seminfo.semmsl) {
1022 			DPRINTF(("nsems out of range (0<%d<=%d)\n", nsems,
1023 			    seminfo.semmsl));
1024 			error = EINVAL;
1025 			goto done2;
1026 		}
1027 		if (nsems > seminfo.semmns - semtot) {
1028 			DPRINTF((
1029 			    "not enough semaphores left (need %d, got %d)\n",
1030 			    nsems, seminfo.semmns - semtot));
1031 			error = ENOSPC;
1032 			goto done2;
1033 		}
1034 		for (semid = 0; semid < seminfo.semmni; semid++) {
1035 			if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0)
1036 				break;
1037 		}
1038 		if (semid == seminfo.semmni) {
1039 			DPRINTF(("no more semid_kernel's available\n"));
1040 			error = ENOSPC;
1041 			goto done2;
1042 		}
1043 #ifdef RACCT
1044 		if (racct_enable) {
1045 			PROC_LOCK(td->td_proc);
1046 			error = racct_add(td->td_proc, RACCT_NSEM, nsems);
1047 			PROC_UNLOCK(td->td_proc);
1048 			if (error != 0) {
1049 				error = ENOSPC;
1050 				goto done2;
1051 			}
1052 		}
1053 #endif
1054 		DPRINTF(("semid %d is available\n", semid));
1055 		mtx_lock(&sema_mtx[semid]);
1056 		KASSERT((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0,
1057 		    ("Lost semaphore %d", semid));
1058 		sema[semid].u.sem_perm.key = key;
1059 		sema[semid].u.sem_perm.cuid = cred->cr_uid;
1060 		sema[semid].u.sem_perm.uid = cred->cr_uid;
1061 		sema[semid].u.sem_perm.cgid = cred->cr_gid;
1062 		sema[semid].u.sem_perm.gid = cred->cr_gid;
1063 		sema[semid].u.sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
1064 		sema[semid].cred = crhold(cred);
1065 		sema_seq[semid]++;
1066 		sema[semid].u.sem_perm.seq = sema_seq[semid] & 0x7fff;
1067 		sema[semid].u.sem_nsems = nsems;
1068 		sema[semid].u.sem_otime = 0;
1069 		sema[semid].u.sem_ctime = time_second;
1070 		sema[semid].u.__sem_base = &sem[semtot];
1071 		semtot += nsems;
1072 		bzero(sema[semid].u.__sem_base,
1073 		    sizeof(sema[semid].u.__sem_base[0])*nsems);
1074 #ifdef MAC
1075 		mac_sysvsem_create(cred, &sema[semid]);
1076 #endif
1077 		mtx_unlock(&sema_mtx[semid]);
1078 		DPRINTF(("sembase = %p, next = %p\n",
1079 		    sema[semid].u.__sem_base, &sem[semtot]));
1080 	} else {
1081 		DPRINTF(("didn't find it and wasn't asked to create it\n"));
1082 		error = ENOENT;
1083 		goto done2;
1084 	}
1085 
1086 found:
1087 	td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].u.sem_perm);
1088 done2:
1089 	mtx_unlock(&sem_mtx);
1090 	return (error);
1091 }
1092 
1093 #ifndef _SYS_SYSPROTO_H_
1094 struct semop_args {
1095 	int	semid;
1096 	struct	sembuf *sops;
1097 	size_t	nsops;
1098 };
1099 #endif
1100 int
1101 sys_semop(struct thread *td, struct semop_args *uap)
1102 {
1103 
1104 	return (kern_semop(td, uap->semid, uap->sops, uap->nsops, NULL));
1105 }
1106 
1107 int
1108 kern_semop(struct thread *td, int usemid, struct sembuf *usops,
1109     size_t nsops, struct timespec *timeout)
1110 {
1111 #define SMALL_SOPS	8
1112 	struct sembuf small_sops[SMALL_SOPS];
1113 	int semid;
1114 	struct prison *rpr;
1115 	struct sembuf *sops;
1116 	struct semid_kernel *semakptr;
1117 	struct sembuf *sopptr = NULL;
1118 	struct sem *semptr = NULL;
1119 	struct sem_undo *suptr;
1120 	struct mtx *sema_mtxp;
1121 	sbintime_t sbt, precision;
1122 	uint64_t seq;
1123 	size_t i, j, k, perms;
1124 	int error;
1125 	bool do_wakeup, do_undos;
1126 
1127 #ifdef SEM_DEBUG
1128 	sops = NULL;
1129 #endif
1130 	DPRINTF(("call to semop(%d, %p, %u)\n", usemid, usops, nsops));
1131 
1132 	AUDIT_ARG_SVIPC_ID(usemid);
1133 
1134 	rpr = sem_find_prison(td->td_ucred);
1135 	if (rpr == NULL)
1136 		return (ENOSYS);
1137 
1138 	semid = IPCID_TO_IX(usemid);	/* Convert back to zero origin */
1139 
1140 	if (semid < 0 || semid >= seminfo.semmni)
1141 		return (EINVAL);
1142 	if (timeout != NULL) {
1143 		if (!timespecvalid_interval(timeout))
1144 			return (EINVAL);
1145 		precision = 0;
1146 		if (timespecisset(timeout)) {
1147 			if (timeout->tv_sec < INT32_MAX / 2) {
1148 				precision = tstosbt(*timeout);
1149 				if (TIMESEL(&sbt, precision))
1150 					sbt += tc_tick_sbt;
1151 				sbt += precision;
1152 				precision >>= tc_precexp;
1153 			} else
1154 				sbt = 0;
1155 		} else
1156 			sbt = -1;
1157 	} else
1158 		precision = sbt = 0;
1159 
1160 	/* Allocate memory for sem_ops */
1161 	if (nsops <= SMALL_SOPS)
1162 		sops = small_sops;
1163 	else if (nsops > seminfo.semopm) {
1164 		DPRINTF(("too many sops (max=%d, nsops=%d)\n", seminfo.semopm,
1165 		    nsops));
1166 		return (E2BIG);
1167 	} else {
1168 #ifdef RACCT
1169 		if (racct_enable) {
1170 			PROC_LOCK(td->td_proc);
1171 			if (nsops >
1172 			    racct_get_available(td->td_proc, RACCT_NSEMOP)) {
1173 				PROC_UNLOCK(td->td_proc);
1174 				return (E2BIG);
1175 			}
1176 			PROC_UNLOCK(td->td_proc);
1177 		}
1178 #endif
1179 
1180 		sops = malloc(nsops * sizeof(*sops), M_TEMP, M_WAITOK);
1181 	}
1182 	if ((error = copyin(usops, sops, nsops * sizeof(sops[0]))) != 0) {
1183 		DPRINTF(("error = %d from copyin(%p, %p, %d)\n", error,
1184 		    usops, sops, nsops * sizeof(sops[0])));
1185 		if (sops != small_sops)
1186 			free(sops, M_TEMP);
1187 		return (error);
1188 	}
1189 
1190 	semakptr = &sema[semid];
1191 	sema_mtxp = &sema_mtx[semid];
1192 	mtx_lock(sema_mtxp);
1193 	if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
1194 		error = EINVAL;
1195 		goto done2;
1196 	}
1197 	if (semvalid(usemid, rpr, semakptr) != 0) {
1198 		error = EINVAL;
1199 		goto done2;
1200 	}
1201 
1202 	/*
1203 	 * Initial pass through sops to see what permissions are needed.
1204 	 * Also perform any checks that don't need repeating on each
1205 	 * attempt to satisfy the request vector.
1206 	 */
1207 	perms = 0;
1208 	do_undos = false;
1209 	for (i = 0; i < nsops; i++) {
1210 		sopptr = &sops[i];
1211 		if (sopptr->sem_num >= semakptr->u.sem_nsems) {
1212 			error = EFBIG;
1213 			goto done2;
1214 		}
1215 		if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0)
1216 			do_undos = true;
1217 		perms |= (sopptr->sem_op == 0) ? SEM_R : SEM_A;
1218 	}
1219 
1220 	if ((error = ipcperm(td, &semakptr->u.sem_perm, perms))) {
1221 		DPRINTF(("error = %d from ipaccess\n", error));
1222 		goto done2;
1223 	}
1224 #ifdef MAC
1225 	error = mac_sysvsem_check_semop(td->td_ucred, semakptr, perms);
1226 	if (error != 0)
1227 		goto done2;
1228 #endif
1229 
1230 	/*
1231 	 * Loop trying to satisfy the vector of requests.
1232 	 * If we reach a point where we must wait, any requests already
1233 	 * performed are rolled back and we go to sleep until some other
1234 	 * process wakes us up.  At this point, we start all over again.
1235 	 *
1236 	 * This ensures that from the perspective of other tasks, a set
1237 	 * of requests is atomic (never partially satisfied).
1238 	 */
1239 	for (;;) {
1240 		do_wakeup = false;
1241 		error = 0;	/* error return if necessary */
1242 		seq = sema_seq[semid];
1243 
1244 		for (i = 0; i < nsops; i++) {
1245 			sopptr = &sops[i];
1246 			semptr = &semakptr->u.__sem_base[sopptr->sem_num];
1247 
1248 			DPRINTF((
1249 			    "semop:  semakptr=%p, __sem_base=%p, "
1250 			    "semptr=%p, sem[%d]=%d : op=%d, flag=%s\n",
1251 			    semakptr, semakptr->u.__sem_base, semptr,
1252 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
1253 			    (sopptr->sem_flg & IPC_NOWAIT) ?
1254 			    "nowait" : "wait"));
1255 
1256 			if (sopptr->sem_op < 0) {
1257 				if (semptr->semval + sopptr->sem_op < 0) {
1258 					DPRINTF(("semop:  can't do it now\n"));
1259 					break;
1260 				} else {
1261 					semptr->semval += sopptr->sem_op;
1262 					if (semptr->semval == 0 &&
1263 					    semptr->semzcnt > 0)
1264 						do_wakeup = true;
1265 				}
1266 			} else if (sopptr->sem_op == 0) {
1267 				if (semptr->semval != 0) {
1268 					DPRINTF(("semop:  not zero now\n"));
1269 					break;
1270 				}
1271 			} else if (semptr->semval + sopptr->sem_op >
1272 			    seminfo.semvmx) {
1273 				error = ERANGE;
1274 				break;
1275 			} else {
1276 				if (semptr->semncnt > 0)
1277 					do_wakeup = true;
1278 				semptr->semval += sopptr->sem_op;
1279 			}
1280 		}
1281 
1282 		/*
1283 		 * Did we get through the entire vector?
1284 		 */
1285 		if (i >= nsops)
1286 			goto done;
1287 
1288 		/*
1289 		 * No ... rollback anything that we've already done
1290 		 */
1291 		DPRINTF(("semop:  rollback 0 through %d\n", i-1));
1292 		for (j = 0; j < i; j++)
1293 			semakptr->u.__sem_base[sops[j].sem_num].semval -=
1294 			    sops[j].sem_op;
1295 
1296 		/* If we detected an error, return it */
1297 		if (error != 0)
1298 			goto done2;
1299 
1300 		/*
1301 		 * If the request that we couldn't satisfy has the
1302 		 * NOWAIT flag set then return with EAGAIN.
1303 		 */
1304 		if (sopptr->sem_flg & IPC_NOWAIT) {
1305 			error = EAGAIN;
1306 			goto done2;
1307 		}
1308 
1309 		if (sopptr->sem_op == 0)
1310 			semptr->semzcnt++;
1311 		else
1312 			semptr->semncnt++;
1313 
1314 		DPRINTF(("semop:  good night!\n"));
1315 		error = msleep_sbt(semakptr, sema_mtxp, PVFS | PCATCH,
1316 		    "semwait", sbt, precision, C_ABSOLUTE);
1317 		DPRINTF(("semop:  good morning (error=%d)!\n", error));
1318 		/* return code is checked below, after sem[nz]cnt-- */
1319 
1320 		/*
1321 		 * Make sure that the semaphore still exists.  The embedded
1322 		 * sequence number check isn't sufficient to detect reallocation
1323 		 * since it's too narrow.
1324 		 */
1325 		if (semvalid(usemid, rpr, semakptr) != 0 ||
1326 		    seq != sema_seq[semid]) {
1327 			error = EIDRM;
1328 			goto done2;
1329 		}
1330 
1331 		/*
1332 		 * Renew the semaphore's pointer after wakeup since
1333 		 * during msleep __sem_base may have been modified and semptr
1334 		 * is not valid any more
1335 		 */
1336 		semptr = &semakptr->u.__sem_base[sopptr->sem_num];
1337 
1338 		/*
1339 		 * The semaphore is still alive.  Readjust the count of
1340 		 * waiting processes.
1341 		 */
1342 		if (sopptr->sem_op == 0)
1343 			semptr->semzcnt--;
1344 		else
1345 			semptr->semncnt--;
1346 
1347 		/*
1348 		 * Is it really morning, or was our sleep interrupted?
1349 		 * (Delayed check of msleep() return code because we
1350 		 * need to decrement sem[nz]cnt either way.)
1351 		 */
1352 		if (error != 0) {
1353 			if (error == ERESTART)
1354 				error = EINTR;
1355 			goto done2;
1356 		}
1357 		DPRINTF(("semop:  good morning!\n"));
1358 	}
1359 
1360 done:
1361 	/*
1362 	 * Process any SEM_UNDO requests.
1363 	 */
1364 	if (do_undos) {
1365 		SEMUNDO_LOCK();
1366 		suptr = NULL;
1367 		for (i = 0; i < nsops; i++) {
1368 			/*
1369 			 * We only need to deal with SEM_UNDO's for non-zero
1370 			 * op's.
1371 			 */
1372 			int adjval;
1373 
1374 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
1375 				continue;
1376 			adjval = sops[i].sem_op;
1377 			if (adjval == 0)
1378 				continue;
1379 			error = semundo_adjust(td, &suptr, semid, seq,
1380 			    sops[i].sem_num, -adjval);
1381 			if (error == 0)
1382 				continue;
1383 
1384 			/*
1385 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
1386 			 * Rollback the adjustments to this point and then
1387 			 * rollback the semaphore ups and down so we can return
1388 			 * with an error with all structures restored.  We
1389 			 * rollback the undo's in the exact reverse order that
1390 			 * we applied them.  This guarantees that we won't run
1391 			 * out of space as we roll things back out.
1392 			 */
1393 			for (j = 0; j < i; j++) {
1394 				k = i - j - 1;
1395 				if ((sops[k].sem_flg & SEM_UNDO) == 0)
1396 					continue;
1397 				adjval = sops[k].sem_op;
1398 				if (adjval == 0)
1399 					continue;
1400 				if (semundo_adjust(td, &suptr, semid, seq,
1401 				    sops[k].sem_num, adjval) != 0)
1402 					panic("semop - can't undo undos");
1403 			}
1404 
1405 			for (j = 0; j < nsops; j++)
1406 				semakptr->u.__sem_base[sops[j].sem_num].semval -=
1407 				    sops[j].sem_op;
1408 
1409 			DPRINTF(("error = %d from semundo_adjust\n", error));
1410 			SEMUNDO_UNLOCK();
1411 			goto done2;
1412 		} /* loop through the sops */
1413 		SEMUNDO_UNLOCK();
1414 	} /* if (do_undos) */
1415 
1416 	/* We're definitely done - set the sempid's and time */
1417 	for (i = 0; i < nsops; i++) {
1418 		sopptr = &sops[i];
1419 		semptr = &semakptr->u.__sem_base[sopptr->sem_num];
1420 		semptr->sempid = td->td_proc->p_pid;
1421 	}
1422 	semakptr->u.sem_otime = time_second;
1423 
1424 	/*
1425 	 * Do a wakeup if any semaphore was up'd whilst something was
1426 	 * sleeping on it.
1427 	 */
1428 	if (do_wakeup) {
1429 		DPRINTF(("semop:  doing wakeup\n"));
1430 		wakeup(semakptr);
1431 		DPRINTF(("semop:  back from wakeup\n"));
1432 	}
1433 	DPRINTF(("semop:  done\n"));
1434 	td->td_retval[0] = 0;
1435 done2:
1436 	mtx_unlock(sema_mtxp);
1437 	if (sops != small_sops)
1438 		free(sops, M_TEMP);
1439 	return (error);
1440 }
1441 
1442 /*
1443  * Go through the undo structures for this process and apply the adjustments to
1444  * semaphores.
1445  */
1446 static void
1447 semexit_myhook(void *arg, struct proc *p)
1448 {
1449 	struct sem_undo *suptr;
1450 	struct semid_kernel *semakptr;
1451 	struct mtx *sema_mtxp;
1452 	uint64_t seq;
1453 	int semid, semnum, adjval, ix;
1454 
1455 	/*
1456 	 * Go through the chain of undo vectors looking for one
1457 	 * associated with this process.
1458 	 */
1459 	if (LIST_EMPTY(&semu_list))
1460 		return;
1461 	SEMUNDO_LOCK();
1462 	LIST_FOREACH(suptr, &semu_list, un_next) {
1463 		if (suptr->un_proc == p)
1464 			break;
1465 	}
1466 	if (suptr == NULL) {
1467 		SEMUNDO_UNLOCK();
1468 		return;
1469 	}
1470 	LIST_REMOVE(suptr, un_next);
1471 
1472 	DPRINTF(("proc @%p has undo structure with %d entries\n", p,
1473 	    suptr->un_cnt));
1474 
1475 	/*
1476 	 * If there are any active undo elements then process them.
1477 	 */
1478 	if (suptr->un_cnt > 0) {
1479 		SEMUNDO_UNLOCK();
1480 		for (ix = 0; ix < suptr->un_cnt; ix++) {
1481 			semid = suptr->un_ent[ix].un_id;
1482 			semnum = suptr->un_ent[ix].un_num;
1483 			adjval = suptr->un_ent[ix].un_adjval;
1484 			seq = suptr->un_ent[ix].un_seq;
1485 			semakptr = &sema[semid];
1486 			sema_mtxp = &sema_mtx[semid];
1487 
1488 			mtx_lock(sema_mtxp);
1489 			if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1490 			    sema_seq[semid] != seq ||
1491 			    semakptr->u.sem_nsems <= semnum) {
1492 				mtx_unlock(sema_mtxp);
1493 				continue;
1494 			}
1495 
1496 			DPRINTF((
1497 			    "semexit:  %p id=%d num=%d(adj=%d) ; sem=%d\n",
1498 			    suptr->un_proc, suptr->un_ent[ix].un_id,
1499 			    suptr->un_ent[ix].un_num,
1500 			    suptr->un_ent[ix].un_adjval,
1501 			    semakptr->u.__sem_base[semnum].semval));
1502 
1503 			if (adjval < 0 && semakptr->u.__sem_base[semnum].semval <
1504 			    -adjval)
1505 				semakptr->u.__sem_base[semnum].semval = 0;
1506 			else
1507 				semakptr->u.__sem_base[semnum].semval += adjval;
1508 
1509 			wakeup(semakptr);
1510 			DPRINTF(("semexit:  back from wakeup\n"));
1511 			mtx_unlock(sema_mtxp);
1512 		}
1513 		SEMUNDO_LOCK();
1514 	}
1515 
1516 	/*
1517 	 * Deallocate the undo vector.
1518 	 */
1519 	DPRINTF(("removing vector\n"));
1520 	suptr->un_proc = NULL;
1521 	suptr->un_cnt = 0;
1522 	LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
1523 	SEMUNDO_UNLOCK();
1524 }
1525 
1526 static int
1527 sysctl_sema(SYSCTL_HANDLER_ARGS)
1528 {
1529 	struct prison *pr, *rpr;
1530 	struct semid_kernel tsemak;
1531 #ifdef COMPAT_FREEBSD32
1532 	struct semid_kernel32 tsemak32;
1533 #endif
1534 	void *outaddr;
1535 	size_t outsize;
1536 	int error, i;
1537 
1538 	pr = req->td->td_ucred->cr_prison;
1539 	rpr = sem_find_prison(req->td->td_ucred);
1540 	error = 0;
1541 	for (i = 0; i < seminfo.semmni; i++) {
1542 		mtx_lock(&sema_mtx[i]);
1543 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) == 0 ||
1544 		    rpr == NULL || sem_prison_cansee(rpr, &sema[i]) != 0)
1545 			bzero(&tsemak, sizeof(tsemak));
1546 		else {
1547 			tsemak = sema[i];
1548 			if (tsemak.cred->cr_prison != pr)
1549 				tsemak.u.sem_perm.key = IPC_PRIVATE;
1550 		}
1551 		mtx_unlock(&sema_mtx[i]);
1552 #ifdef COMPAT_FREEBSD32
1553 		if (SV_CURPROC_FLAG(SV_ILP32)) {
1554 			bzero(&tsemak32, sizeof(tsemak32));
1555 			freebsd32_ipcperm_out(&tsemak.u.sem_perm,
1556 			    &tsemak32.u.sem_perm);
1557 			/* Don't copy u.__sem_base */
1558 			CP(tsemak, tsemak32, u.sem_nsems);
1559 			CP(tsemak, tsemak32, u.sem_otime);
1560 			CP(tsemak, tsemak32, u.sem_ctime);
1561 			/* Don't copy label or cred */
1562 			outaddr = &tsemak32;
1563 			outsize = sizeof(tsemak32);
1564 		} else
1565 #endif
1566 		{
1567 			tsemak.u.__sem_base = NULL;
1568 			tsemak.label = NULL;
1569 			tsemak.cred = NULL;
1570 			outaddr = &tsemak;
1571 			outsize = sizeof(tsemak);
1572 		}
1573 		error = SYSCTL_OUT(req, outaddr, outsize);
1574 		if (error != 0)
1575 			break;
1576 	}
1577 	return (error);
1578 }
1579 
1580 int
1581 kern_get_sema(struct thread *td, struct semid_kernel **res, size_t *sz)
1582 {
1583 	struct prison *pr, *rpr;
1584 	struct semid_kernel *psemak;
1585 	int i, mi;
1586 
1587 	*sz = mi = seminfo.semmni;
1588 	if (res == NULL)
1589 		return (0);
1590 
1591 	pr = td->td_ucred->cr_prison;
1592 	rpr = sem_find_prison(td->td_ucred);
1593 	*res = malloc(sizeof(struct semid_kernel) * mi, M_TEMP, M_WAITOK);
1594 	for (i = 0; i < mi; i++) {
1595 		psemak = &(*res)[i];
1596 		mtx_lock(&sema_mtx[i]);
1597 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) == 0 ||
1598 		    rpr == NULL || sem_prison_cansee(rpr, &sema[i]) != 0)
1599 			bzero(psemak, sizeof(*psemak));
1600 		else {
1601 			*psemak = sema[i];
1602 			if (psemak->cred->cr_prison != pr)
1603 				psemak->u.sem_perm.key = IPC_PRIVATE;
1604 		}
1605 		mtx_unlock(&sema_mtx[i]);
1606 		psemak->u.__sem_base = NULL;
1607 		psemak->label = NULL;
1608 		psemak->cred = NULL;
1609 	}
1610 	return (0);
1611 }
1612 
1613 static int
1614 sem_prison_check(void *obj, void *data)
1615 {
1616 	struct prison *pr = obj;
1617 	struct prison *prpr;
1618 	struct vfsoptlist *opts = data;
1619 	int error, jsys;
1620 
1621 	/*
1622 	 * sysvsem is a jailsys integer.
1623 	 * It must be "disable" if the parent jail is disabled.
1624 	 */
1625 	error = vfs_copyopt(opts, "sysvsem", &jsys, sizeof(jsys));
1626 	if (error != ENOENT) {
1627 		if (error != 0)
1628 			return (error);
1629 		switch (jsys) {
1630 		case JAIL_SYS_DISABLE:
1631 			break;
1632 		case JAIL_SYS_NEW:
1633 		case JAIL_SYS_INHERIT:
1634 			prison_lock(pr->pr_parent);
1635 			prpr = osd_jail_get(pr->pr_parent, sem_prison_slot);
1636 			prison_unlock(pr->pr_parent);
1637 			if (prpr == NULL)
1638 				return (EPERM);
1639 			break;
1640 		default:
1641 			return (EINVAL);
1642 		}
1643 	}
1644 
1645 	return (0);
1646 }
1647 
1648 static int
1649 sem_prison_set(void *obj, void *data)
1650 {
1651 	struct prison *pr = obj;
1652 	struct prison *tpr, *orpr, *nrpr, *trpr;
1653 	struct vfsoptlist *opts = data;
1654 	void *rsv;
1655 	int jsys, descend;
1656 
1657 	/*
1658 	 * sysvsem controls which jail is the root of the associated sems (this
1659 	 * jail or same as the parent), or if the feature is available at all.
1660 	 */
1661 	if (vfs_copyopt(opts, "sysvsem", &jsys, sizeof(jsys)) == ENOENT)
1662 		jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0)
1663 		    ? JAIL_SYS_INHERIT
1664 		    : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0)
1665 		    ? JAIL_SYS_DISABLE
1666 		    : -1;
1667 	if (jsys == JAIL_SYS_DISABLE) {
1668 		prison_lock(pr);
1669 		orpr = osd_jail_get(pr, sem_prison_slot);
1670 		if (orpr != NULL)
1671 			osd_jail_del(pr, sem_prison_slot);
1672 		prison_unlock(pr);
1673 		if (orpr != NULL) {
1674 			if (orpr == pr)
1675 				sem_prison_cleanup(pr);
1676 			/* Disable all child jails as well. */
1677 			FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1678 				prison_lock(tpr);
1679 				trpr = osd_jail_get(tpr, sem_prison_slot);
1680 				if (trpr != NULL) {
1681 					osd_jail_del(tpr, sem_prison_slot);
1682 					prison_unlock(tpr);
1683 					if (trpr == tpr)
1684 						sem_prison_cleanup(tpr);
1685 				} else {
1686 					prison_unlock(tpr);
1687 					descend = 0;
1688 				}
1689 			}
1690 		}
1691 	} else if (jsys != -1) {
1692 		if (jsys == JAIL_SYS_NEW)
1693 			nrpr = pr;
1694 		else {
1695 			prison_lock(pr->pr_parent);
1696 			nrpr = osd_jail_get(pr->pr_parent, sem_prison_slot);
1697 			prison_unlock(pr->pr_parent);
1698 		}
1699 		rsv = osd_reserve(sem_prison_slot);
1700 		prison_lock(pr);
1701 		orpr = osd_jail_get(pr, sem_prison_slot);
1702 		if (orpr != nrpr)
1703 			(void)osd_jail_set_reserved(pr, sem_prison_slot, rsv,
1704 			    nrpr);
1705 		else
1706 			osd_free_reserved(rsv);
1707 		prison_unlock(pr);
1708 		if (orpr != nrpr) {
1709 			if (orpr == pr)
1710 				sem_prison_cleanup(pr);
1711 			if (orpr != NULL) {
1712 				/* Change child jails matching the old root, */
1713 				FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1714 					prison_lock(tpr);
1715 					trpr = osd_jail_get(tpr,
1716 					    sem_prison_slot);
1717 					if (trpr == orpr) {
1718 						(void)osd_jail_set(tpr,
1719 						    sem_prison_slot, nrpr);
1720 						prison_unlock(tpr);
1721 						if (trpr == tpr)
1722 							sem_prison_cleanup(tpr);
1723 					} else {
1724 						prison_unlock(tpr);
1725 						descend = 0;
1726 					}
1727 				}
1728 			}
1729 		}
1730 	}
1731 
1732 	return (0);
1733 }
1734 
1735 static int
1736 sem_prison_get(void *obj, void *data)
1737 {
1738 	struct prison *pr = obj;
1739 	struct prison *rpr;
1740 	struct vfsoptlist *opts = data;
1741 	int error, jsys;
1742 
1743 	/* Set sysvsem based on the jail's root prison. */
1744 	prison_lock(pr);
1745 	rpr = osd_jail_get(pr, sem_prison_slot);
1746 	prison_unlock(pr);
1747 	jsys = rpr == NULL ? JAIL_SYS_DISABLE
1748 	    : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT;
1749 	error = vfs_setopt(opts, "sysvsem", &jsys, sizeof(jsys));
1750 	if (error == ENOENT)
1751 		error = 0;
1752 	return (error);
1753 }
1754 
1755 static int
1756 sem_prison_remove(void *obj, void *data __unused)
1757 {
1758 	struct prison *pr = obj;
1759 	struct prison *rpr;
1760 
1761 	prison_lock(pr);
1762 	rpr = osd_jail_get(pr, sem_prison_slot);
1763 	prison_unlock(pr);
1764 	if (rpr == pr)
1765 		sem_prison_cleanup(pr);
1766 	return (0);
1767 }
1768 
1769 static void
1770 sem_prison_cleanup(struct prison *pr)
1771 {
1772 	int i;
1773 
1774 	/* Remove any sems that belong to this jail. */
1775 	mtx_lock(&sem_mtx);
1776 	for (i = 0; i < seminfo.semmni; i++) {
1777 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
1778 		    sema[i].cred != NULL && sema[i].cred->cr_prison == pr) {
1779 			mtx_lock(&sema_mtx[i]);
1780 			sem_remove(i, NULL);
1781 			mtx_unlock(&sema_mtx[i]);
1782 		}
1783 	}
1784 	mtx_unlock(&sem_mtx);
1785 }
1786 
1787 SYSCTL_JAIL_PARAM_SYS_NODE(sysvsem, CTLFLAG_RW, "SYSV semaphores");
1788 
1789 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1790     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1791 
1792 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1793 static sy_call_t *semcalls[] = {
1794 	(sy_call_t *)freebsd7___semctl, (sy_call_t *)sys_semget,
1795 	(sy_call_t *)sys_semop
1796 };
1797 
1798 /*
1799  * Entry point for all SEM calls.
1800  */
1801 int
1802 sys_semsys(struct thread *td, struct semsys_args *uap)
1803 {
1804 	int error;
1805 
1806 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1807 	if (uap->which < 0 || uap->which >= nitems(semcalls))
1808 		return (EINVAL);
1809 	error = (*semcalls[uap->which])(td, &uap->a2);
1810 	return (error);
1811 }
1812 
1813 #ifndef _SYS_SYSPROTO_H_
1814 struct freebsd7___semctl_args {
1815 	int	semid;
1816 	int	semnum;
1817 	int	cmd;
1818 	union	semun_old *arg;
1819 };
1820 #endif
1821 int
1822 freebsd7___semctl(struct thread *td, struct freebsd7___semctl_args *uap)
1823 {
1824 	struct semid_ds_old dsold;
1825 	struct semid_ds dsbuf;
1826 	union semun_old arg;
1827 	union semun semun;
1828 	register_t rval;
1829 	int error;
1830 
1831 	switch (uap->cmd) {
1832 	case SEM_STAT:
1833 	case IPC_SET:
1834 	case IPC_STAT:
1835 	case GETALL:
1836 	case SETVAL:
1837 	case SETALL:
1838 		error = copyin(uap->arg, &arg, sizeof(arg));
1839 		if (error)
1840 			return (error);
1841 		break;
1842 	}
1843 
1844 	switch (uap->cmd) {
1845 	case SEM_STAT:
1846 	case IPC_STAT:
1847 		semun.buf = &dsbuf;
1848 		break;
1849 	case IPC_SET:
1850 		error = copyin(arg.buf, &dsold, sizeof(dsold));
1851 		if (error)
1852 			return (error);
1853 		ipcperm_old2new(&dsold.sem_perm, &dsbuf.sem_perm);
1854 		CP(dsold, dsbuf, __sem_base);
1855 		CP(dsold, dsbuf, sem_nsems);
1856 		CP(dsold, dsbuf, sem_otime);
1857 		CP(dsold, dsbuf, sem_ctime);
1858 		semun.buf = &dsbuf;
1859 		break;
1860 	case GETALL:
1861 	case SETALL:
1862 		semun.array = arg.array;
1863 		break;
1864 	case SETVAL:
1865 		semun.val = arg.val;
1866 		break;
1867 	}
1868 
1869 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1870 	    &rval);
1871 	if (error)
1872 		return (error);
1873 
1874 	switch (uap->cmd) {
1875 	case SEM_STAT:
1876 	case IPC_STAT:
1877 		bzero(&dsold, sizeof(dsold));
1878 		ipcperm_new2old(&dsbuf.sem_perm, &dsold.sem_perm);
1879 		CP(dsbuf, dsold, __sem_base);
1880 		CP(dsbuf, dsold, sem_nsems);
1881 		CP(dsbuf, dsold, sem_otime);
1882 		CP(dsbuf, dsold, sem_ctime);
1883 		error = copyout(&dsold, arg.buf, sizeof(dsold));
1884 		break;
1885 	}
1886 
1887 	if (error == 0)
1888 		td->td_retval[0] = rval;
1889 	return (error);
1890 }
1891 
1892 #endif /* COMPAT_FREEBSD{4,5,6,7} */
1893 
1894 #ifdef COMPAT_FREEBSD32
1895 
1896 int
1897 freebsd32_semsys(struct thread *td, struct freebsd32_semsys_args *uap)
1898 {
1899 
1900 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1901     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1902 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1903 	switch (uap->which) {
1904 	case 0:
1905 		return (freebsd7_freebsd32___semctl(td,
1906 		    (struct freebsd7_freebsd32___semctl_args *)&uap->a2));
1907 	default:
1908 		return (sys_semsys(td, (struct semsys_args *)uap));
1909 	}
1910 #else
1911 	return (kern_nosys(td, 0));
1912 #endif
1913 }
1914 
1915 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1916     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1917 int
1918 freebsd7_freebsd32___semctl(struct thread *td,
1919     struct freebsd7_freebsd32___semctl_args *uap)
1920 {
1921 	struct semid_ds_old32 dsbuf32;
1922 	struct semid_ds dsbuf;
1923 	union semun semun;
1924 	union semun_old32 arg;
1925 	register_t rval;
1926 	int error;
1927 
1928 	switch (uap->cmd) {
1929 	case SEM_STAT:
1930 	case IPC_SET:
1931 	case IPC_STAT:
1932 	case GETALL:
1933 	case SETVAL:
1934 	case SETALL:
1935 		error = copyin(uap->arg, &arg, sizeof(arg));
1936 		if (error)
1937 			return (error);
1938 		break;
1939 	}
1940 
1941 	switch (uap->cmd) {
1942 	case SEM_STAT:
1943 	case IPC_STAT:
1944 		semun.buf = &dsbuf;
1945 		break;
1946 	case IPC_SET:
1947 		error = copyin(PTRIN(arg.buf), &dsbuf32, sizeof(dsbuf32));
1948 		if (error)
1949 			return (error);
1950 		freebsd32_ipcperm_old_in(&dsbuf32.sem_perm, &dsbuf.sem_perm);
1951 		PTRIN_CP(dsbuf32, dsbuf, __sem_base);
1952 		CP(dsbuf32, dsbuf, sem_nsems);
1953 		CP(dsbuf32, dsbuf, sem_otime);
1954 		CP(dsbuf32, dsbuf, sem_ctime);
1955 		semun.buf = &dsbuf;
1956 		break;
1957 	case GETALL:
1958 	case SETALL:
1959 		semun.array = PTRIN(arg.array);
1960 		break;
1961 	case SETVAL:
1962 		semun.val = arg.val;
1963 		break;
1964 	}
1965 
1966 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1967 	    &rval);
1968 	if (error)
1969 		return (error);
1970 
1971 	switch (uap->cmd) {
1972 	case SEM_STAT:
1973 	case IPC_STAT:
1974 		bzero(&dsbuf32, sizeof(dsbuf32));
1975 		freebsd32_ipcperm_old_out(&dsbuf.sem_perm, &dsbuf32.sem_perm);
1976 		PTROUT_CP(dsbuf, dsbuf32, __sem_base);
1977 		CP(dsbuf, dsbuf32, sem_nsems);
1978 		CP(dsbuf, dsbuf32, sem_otime);
1979 		CP(dsbuf, dsbuf32, sem_ctime);
1980 		error = copyout(&dsbuf32, PTRIN(arg.buf), sizeof(dsbuf32));
1981 		break;
1982 	}
1983 
1984 	if (error == 0)
1985 		td->td_retval[0] = rval;
1986 	return (error);
1987 }
1988 #endif
1989 
1990 int
1991 freebsd32___semctl(struct thread *td, struct freebsd32___semctl_args *uap)
1992 {
1993 	struct semid_ds32 dsbuf32;
1994 	struct semid_ds dsbuf;
1995 	union semun semun;
1996 	union semun32 arg;
1997 	register_t rval;
1998 	int error;
1999 
2000 	switch (uap->cmd) {
2001 	case SEM_STAT:
2002 	case IPC_SET:
2003 	case IPC_STAT:
2004 	case GETALL:
2005 	case SETVAL:
2006 	case SETALL:
2007 		error = copyin(uap->arg, &arg, sizeof(arg));
2008 		if (error)
2009 			return (error);
2010 		break;
2011 	}
2012 
2013 	switch (uap->cmd) {
2014 	case SEM_STAT:
2015 	case IPC_STAT:
2016 		semun.buf = &dsbuf;
2017 		break;
2018 	case IPC_SET:
2019 		error = copyin(PTRIN(arg.buf), &dsbuf32, sizeof(dsbuf32));
2020 		if (error)
2021 			return (error);
2022 		freebsd32_ipcperm_in(&dsbuf32.sem_perm, &dsbuf.sem_perm);
2023 		PTRIN_CP(dsbuf32, dsbuf, __sem_base);
2024 		CP(dsbuf32, dsbuf, sem_nsems);
2025 		CP(dsbuf32, dsbuf, sem_otime);
2026 		CP(dsbuf32, dsbuf, sem_ctime);
2027 		semun.buf = &dsbuf;
2028 		break;
2029 	case GETALL:
2030 	case SETALL:
2031 		semun.array = PTRIN(arg.array);
2032 		break;
2033 	case SETVAL:
2034 		semun.val = arg.val;
2035 		break;
2036 	}
2037 
2038 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
2039 	    &rval);
2040 	if (error)
2041 		return (error);
2042 
2043 	switch (uap->cmd) {
2044 	case SEM_STAT:
2045 	case IPC_STAT:
2046 		bzero(&dsbuf32, sizeof(dsbuf32));
2047 		freebsd32_ipcperm_out(&dsbuf.sem_perm, &dsbuf32.sem_perm);
2048 		PTROUT_CP(dsbuf, dsbuf32, __sem_base);
2049 		CP(dsbuf, dsbuf32, sem_nsems);
2050 		CP(dsbuf, dsbuf32, sem_otime);
2051 		CP(dsbuf, dsbuf32, sem_ctime);
2052 		error = copyout(&dsbuf32, PTRIN(arg.buf), sizeof(dsbuf32));
2053 		break;
2054 	}
2055 
2056 	if (error == 0)
2057 		td->td_retval[0] = rval;
2058 	return (error);
2059 }
2060 
2061 #endif /* COMPAT_FREEBSD32 */
2062