xref: /freebsd/sys/kern/sysv_sem.c (revision ca987d4641cdcd7f27e153db17c5bf064934faf5)
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  * Copyright (c) 2003-2005 McAfee, Inc.
10  * Copyright (c) 2016-2017 Robert N. M. Watson
11  * All rights reserved.
12  *
13  * This software was developed for the FreeBSD Project in part by McAfee
14  * Research, the Security Research Division of McAfee, Inc under DARPA/SPAWAR
15  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS research
16  * program.
17  *
18  * Portions of this software were developed by BAE Systems, the University of
19  * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL
20  * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent
21  * Computing (TC) research program.
22  *
23  * Redistribution and use in source and binary forms, with or without
24  * modification, are permitted provided that the following conditions
25  * are met:
26  * 1. Redistributions of source code must retain the above copyright
27  *    notice, this list of conditions and the following disclaimer.
28  * 2. Redistributions in binary form must reproduce the above copyright
29  *    notice, this list of conditions and the following disclaimer in the
30  *    documentation and/or other materials provided with the distribution.
31  *
32  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
33  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
34  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
35  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
36  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
37  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
38  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
39  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
40  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
41  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
42  * SUCH DAMAGE.
43  */
44 
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47 
48 #include "opt_compat.h"
49 #include "opt_sysvipc.h"
50 
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/sysproto.h>
54 #include <sys/eventhandler.h>
55 #include <sys/kernel.h>
56 #include <sys/proc.h>
57 #include <sys/lock.h>
58 #include <sys/module.h>
59 #include <sys/mutex.h>
60 #include <sys/racct.h>
61 #include <sys/sem.h>
62 #include <sys/sx.h>
63 #include <sys/syscall.h>
64 #include <sys/syscallsubr.h>
65 #include <sys/sysent.h>
66 #include <sys/sysctl.h>
67 #include <sys/uio.h>
68 #include <sys/malloc.h>
69 #include <sys/jail.h>
70 
71 #include <security/audit/audit.h>
72 #include <security/mac/mac_framework.h>
73 
74 FEATURE(sysv_sem, "System V semaphores support");
75 
76 static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores");
77 
78 #ifdef SEM_DEBUG
79 #define DPRINTF(a)	printf a
80 #else
81 #define DPRINTF(a)
82 #endif
83 
84 static int seminit(void);
85 static int sysvsem_modload(struct module *, int, void *);
86 static int semunload(void);
87 static void semexit_myhook(void *arg, struct proc *p);
88 static int sysctl_sema(SYSCTL_HANDLER_ARGS);
89 static int semvalid(int semid, struct prison *rpr,
90     struct semid_kernel *semakptr);
91 static void sem_remove(int semidx, struct ucred *cred);
92 static struct prison *sem_find_prison(struct ucred *);
93 static int sem_prison_cansee(struct prison *, struct semid_kernel *);
94 static int sem_prison_check(void *, void *);
95 static int sem_prison_set(void *, void *);
96 static int sem_prison_get(void *, void *);
97 static int sem_prison_remove(void *, void *);
98 static void sem_prison_cleanup(struct prison *);
99 
100 #ifndef _SYS_SYSPROTO_H_
101 struct __semctl_args;
102 int __semctl(struct thread *td, struct __semctl_args *uap);
103 struct semget_args;
104 int semget(struct thread *td, struct semget_args *uap);
105 struct semop_args;
106 int semop(struct thread *td, struct semop_args *uap);
107 #endif
108 
109 static struct sem_undo *semu_alloc(struct thread *td);
110 static int semundo_adjust(struct thread *td, struct sem_undo **supptr,
111     int semid, int semseq, int semnum, int adjval);
112 static void semundo_clear(int semid, int semnum);
113 
114 static struct mtx	sem_mtx;	/* semaphore global lock */
115 static struct mtx sem_undo_mtx;
116 static int	semtot = 0;
117 static struct semid_kernel *sema;	/* semaphore id pool */
118 static struct mtx *sema_mtx;	/* semaphore id pool mutexes*/
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 		unsigned short 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	SEM_ALIGN(offsetof(struct sem_undo, un_ent[SEMUME]))
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,         /* # of semaphore identifiers */
201                 SEMMNS,         /* # of semaphores in system */
202                 SEMMNU,         /* # of undo structures in system */
203                 SEMMSL,         /* max # of semaphores per id */
204                 SEMOPM,         /* max # of operations per semop call */
205                 SEMUME,         /* max # of undo entries per process */
206                 SEMUSZ,         /* size in bytes of undo structure */
207                 SEMVMX,         /* semaphore maximum value */
208                 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_RDTUN, &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, "", "Semaphore id pool");
232 
233 static struct syscall_helper_data sem_syscalls[] = {
234 	SYSCALL_INIT_HELPER(__semctl),
235 	SYSCALL_INIT_HELPER(semget),
236 	SYSCALL_INIT_HELPER(semop),
237 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
238     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
239 	SYSCALL_INIT_HELPER(semsys),
240 	SYSCALL_INIT_HELPER_COMPAT(freebsd7___semctl),
241 #endif
242 	SYSCALL_INIT_LAST
243 };
244 
245 #ifdef COMPAT_FREEBSD32
246 #include <compat/freebsd32/freebsd32.h>
247 #include <compat/freebsd32/freebsd32_ipc.h>
248 #include <compat/freebsd32/freebsd32_proto.h>
249 #include <compat/freebsd32/freebsd32_signal.h>
250 #include <compat/freebsd32/freebsd32_syscall.h>
251 #include <compat/freebsd32/freebsd32_util.h>
252 
253 static struct syscall_helper_data sem32_syscalls[] = {
254 	SYSCALL32_INIT_HELPER(freebsd32_semctl),
255 	SYSCALL32_INIT_HELPER_COMPAT(semget),
256 	SYSCALL32_INIT_HELPER_COMPAT(semop),
257 	SYSCALL32_INIT_HELPER(freebsd32_semsys),
258 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
259     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
260 	SYSCALL32_INIT_HELPER(freebsd7_freebsd32_semctl),
261 #endif
262 	SYSCALL_INIT_LAST
263 };
264 #endif
265 
266 static int
267 seminit(void)
268 {
269 	struct prison *pr;
270 	void **rsv;
271 	int i, error;
272 	osd_method_t methods[PR_MAXMETHOD] = {
273 	    [PR_METHOD_CHECK] =		sem_prison_check,
274 	    [PR_METHOD_SET] =		sem_prison_set,
275 	    [PR_METHOD_GET] =		sem_prison_get,
276 	    [PR_METHOD_REMOVE] =	sem_prison_remove,
277 	};
278 
279 	sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
280 	sema = malloc(sizeof(struct semid_kernel) * seminfo.semmni, M_SEM,
281 	    M_WAITOK);
282 	sema_mtx = malloc(sizeof(struct mtx) * seminfo.semmni, M_SEM,
283 	    M_WAITOK | M_ZERO);
284 	semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
285 
286 	for (i = 0; i < seminfo.semmni; i++) {
287 		sema[i].u.sem_base = 0;
288 		sema[i].u.sem_perm.mode = 0;
289 		sema[i].u.sem_perm.seq = 0;
290 #ifdef MAC
291 		mac_sysvsem_init(&sema[i]);
292 #endif
293 	}
294 	for (i = 0; i < seminfo.semmni; i++)
295 		mtx_init(&sema_mtx[i], "semid", NULL, MTX_DEF);
296 	LIST_INIT(&semu_free_list);
297 	for (i = 0; i < seminfo.semmnu; i++) {
298 		struct sem_undo *suptr = SEMU(i);
299 		suptr->un_proc = NULL;
300 		LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
301 	}
302 	LIST_INIT(&semu_list);
303 	mtx_init(&sem_mtx, "sem", NULL, MTX_DEF);
304 	mtx_init(&sem_undo_mtx, "semu", NULL, MTX_DEF);
305 	semexit_tag = EVENTHANDLER_REGISTER(process_exit, semexit_myhook, NULL,
306 	    EVENTHANDLER_PRI_ANY);
307 
308 	/* Set current prisons according to their allow.sysvipc. */
309 	sem_prison_slot = osd_jail_register(NULL, methods);
310 	rsv = osd_reserve(sem_prison_slot);
311 	prison_lock(&prison0);
312 	(void)osd_jail_set_reserved(&prison0, sem_prison_slot, rsv, &prison0);
313 	prison_unlock(&prison0);
314 	rsv = NULL;
315 	sx_slock(&allprison_lock);
316 	TAILQ_FOREACH(pr, &allprison, pr_list) {
317 		if (rsv == NULL)
318 			rsv = osd_reserve(sem_prison_slot);
319 		prison_lock(pr);
320 		if ((pr->pr_allow & PR_ALLOW_SYSVIPC) && pr->pr_ref > 0) {
321 			(void)osd_jail_set_reserved(pr, sem_prison_slot, rsv,
322 			    &prison0);
323 			rsv = NULL;
324 		}
325 		prison_unlock(pr);
326 	}
327 	if (rsv != NULL)
328 		osd_free_reserved(rsv);
329 	sx_sunlock(&allprison_lock);
330 
331 	error = syscall_helper_register(sem_syscalls, SY_THR_STATIC_KLD);
332 	if (error != 0)
333 		return (error);
334 #ifdef COMPAT_FREEBSD32
335 	error = syscall32_helper_register(sem32_syscalls, SY_THR_STATIC_KLD);
336 	if (error != 0)
337 		return (error);
338 #endif
339 	return (0);
340 }
341 
342 static int
343 semunload(void)
344 {
345 	int i;
346 
347 	/* XXXKIB */
348 	if (semtot != 0)
349 		return (EBUSY);
350 
351 #ifdef COMPAT_FREEBSD32
352 	syscall32_helper_unregister(sem32_syscalls);
353 #endif
354 	syscall_helper_unregister(sem_syscalls);
355 	EVENTHANDLER_DEREGISTER(process_exit, semexit_tag);
356 	if (sem_prison_slot != 0)
357 		osd_jail_deregister(sem_prison_slot);
358 #ifdef MAC
359 	for (i = 0; i < seminfo.semmni; i++)
360 		mac_sysvsem_destroy(&sema[i]);
361 #endif
362 	free(sem, M_SEM);
363 	free(sema, M_SEM);
364 	free(semu, M_SEM);
365 	for (i = 0; i < seminfo.semmni; i++)
366 		mtx_destroy(&sema_mtx[i]);
367 	free(sema_mtx, M_SEM);
368 	mtx_destroy(&sem_mtx);
369 	mtx_destroy(&sem_undo_mtx);
370 	return (0);
371 }
372 
373 static int
374 sysvsem_modload(struct module *module, int cmd, void *arg)
375 {
376 	int error = 0;
377 
378 	switch (cmd) {
379 	case MOD_LOAD:
380 		error = seminit();
381 		if (error != 0)
382 			semunload();
383 		break;
384 	case MOD_UNLOAD:
385 		error = semunload();
386 		break;
387 	case MOD_SHUTDOWN:
388 		break;
389 	default:
390 		error = EINVAL;
391 		break;
392 	}
393 	return (error);
394 }
395 
396 static moduledata_t sysvsem_mod = {
397 	"sysvsem",
398 	&sysvsem_modload,
399 	NULL
400 };
401 
402 DECLARE_MODULE(sysvsem, sysvsem_mod, SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
403 MODULE_VERSION(sysvsem, 1);
404 
405 /*
406  * Allocate a new sem_undo structure for a process
407  * (returns ptr to structure or NULL if no more room)
408  */
409 
410 static struct sem_undo *
411 semu_alloc(struct thread *td)
412 {
413 	struct sem_undo *suptr;
414 
415 	SEMUNDO_LOCKASSERT(MA_OWNED);
416 	if ((suptr = LIST_FIRST(&semu_free_list)) == NULL)
417 		return (NULL);
418 	LIST_REMOVE(suptr, un_next);
419 	LIST_INSERT_HEAD(&semu_list, suptr, un_next);
420 	suptr->un_cnt = 0;
421 	suptr->un_proc = td->td_proc;
422 	return (suptr);
423 }
424 
425 static int
426 semu_try_free(struct sem_undo *suptr)
427 {
428 
429 	SEMUNDO_LOCKASSERT(MA_OWNED);
430 
431 	if (suptr->un_cnt != 0)
432 		return (0);
433 	LIST_REMOVE(suptr, un_next);
434 	LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
435 	return (1);
436 }
437 
438 /*
439  * Adjust a particular entry for a particular proc
440  */
441 
442 static int
443 semundo_adjust(struct thread *td, struct sem_undo **supptr, int semid,
444     int semseq, int semnum, int adjval)
445 {
446 	struct proc *p = td->td_proc;
447 	struct sem_undo *suptr;
448 	struct undo *sunptr;
449 	int i;
450 
451 	SEMUNDO_LOCKASSERT(MA_OWNED);
452 	/* Look for and remember the sem_undo if the caller doesn't provide
453 	   it */
454 
455 	suptr = *supptr;
456 	if (suptr == NULL) {
457 		LIST_FOREACH(suptr, &semu_list, un_next) {
458 			if (suptr->un_proc == p) {
459 				*supptr = suptr;
460 				break;
461 			}
462 		}
463 		if (suptr == NULL) {
464 			if (adjval == 0)
465 				return(0);
466 			suptr = semu_alloc(td);
467 			if (suptr == NULL)
468 				return (ENOSPC);
469 			*supptr = suptr;
470 		}
471 	}
472 
473 	/*
474 	 * Look for the requested entry and adjust it (delete if adjval becomes
475 	 * 0).
476 	 */
477 	sunptr = &suptr->un_ent[0];
478 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
479 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
480 			continue;
481 		if (adjval != 0) {
482 			adjval += sunptr->un_adjval;
483 			if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
484 				return (ERANGE);
485 		}
486 		sunptr->un_adjval = adjval;
487 		if (sunptr->un_adjval == 0) {
488 			suptr->un_cnt--;
489 			if (i < suptr->un_cnt)
490 				suptr->un_ent[i] =
491 				    suptr->un_ent[suptr->un_cnt];
492 			if (suptr->un_cnt == 0)
493 				semu_try_free(suptr);
494 		}
495 		return (0);
496 	}
497 
498 	/* Didn't find the right entry - create it */
499 	if (adjval == 0)
500 		return (0);
501 	if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
502 		return (ERANGE);
503 	if (suptr->un_cnt != seminfo.semume) {
504 		sunptr = &suptr->un_ent[suptr->un_cnt];
505 		suptr->un_cnt++;
506 		sunptr->un_adjval = adjval;
507 		sunptr->un_id = semid;
508 		sunptr->un_num = semnum;
509 		sunptr->un_seq = semseq;
510 	} else
511 		return (EINVAL);
512 	return (0);
513 }
514 
515 static void
516 semundo_clear(int semid, int semnum)
517 {
518 	struct sem_undo *suptr, *suptr1;
519 	struct undo *sunptr;
520 	int i;
521 
522 	SEMUNDO_LOCKASSERT(MA_OWNED);
523 	LIST_FOREACH_SAFE(suptr, &semu_list, un_next, suptr1) {
524 		sunptr = &suptr->un_ent[0];
525 		for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
526 			if (sunptr->un_id != semid)
527 				continue;
528 			if (semnum == -1 || sunptr->un_num == semnum) {
529 				suptr->un_cnt--;
530 				if (i < suptr->un_cnt) {
531 					suptr->un_ent[i] =
532 					    suptr->un_ent[suptr->un_cnt];
533 					continue;
534 				}
535 				semu_try_free(suptr);
536 			}
537 			if (semnum != -1)
538 				break;
539 		}
540 	}
541 }
542 
543 static int
544 semvalid(int semid, struct prison *rpr, struct semid_kernel *semakptr)
545 {
546 
547 	return ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
548 	    semakptr->u.sem_perm.seq != IPCID_TO_SEQ(semid) ||
549 	    sem_prison_cansee(rpr, semakptr) ? EINVAL : 0);
550 }
551 
552 static void
553 sem_remove(int semidx, struct ucred *cred)
554 {
555 	struct semid_kernel *semakptr;
556 	int i;
557 
558 	KASSERT(semidx >= 0 && semidx < seminfo.semmni,
559 		("semidx out of bounds"));
560 	semakptr = &sema[semidx];
561 	semakptr->u.sem_perm.cuid = cred ? cred->cr_uid : 0;
562 	semakptr->u.sem_perm.uid = cred ? cred->cr_uid : 0;
563 	semakptr->u.sem_perm.mode = 0;
564 	racct_sub_cred(semakptr->cred, RACCT_NSEM, semakptr->u.sem_nsems);
565 	crfree(semakptr->cred);
566 	semakptr->cred = NULL;
567 	SEMUNDO_LOCK();
568 	semundo_clear(semidx, -1);
569 	SEMUNDO_UNLOCK();
570 #ifdef MAC
571 	mac_sysvsem_cleanup(semakptr);
572 #endif
573 	wakeup(semakptr);
574 	for (i = 0; i < seminfo.semmni; i++) {
575 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
576 		    sema[i].u.sem_base > semakptr->u.sem_base)
577 			mtx_lock_flags(&sema_mtx[i], LOP_DUPOK);
578 	}
579 	for (i = semakptr->u.sem_base - sem; i < semtot; i++)
580 		sem[i] = sem[i + semakptr->u.sem_nsems];
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 			sema[i].u.sem_base -= semakptr->u.sem_nsems;
585 			mtx_unlock(&sema_mtx[i]);
586 		}
587 	}
588 	semtot -= semakptr->u.sem_nsems;
589 }
590 
591 static struct prison *
592 sem_find_prison(struct ucred *cred)
593 {
594 	struct prison *pr, *rpr;
595 
596 	pr = cred->cr_prison;
597 	prison_lock(pr);
598 	rpr = osd_jail_get(pr, sem_prison_slot);
599 	prison_unlock(pr);
600 	return rpr;
601 }
602 
603 static int
604 sem_prison_cansee(struct prison *rpr, struct semid_kernel *semakptr)
605 {
606 
607 	if (semakptr->cred == NULL ||
608 	    !(rpr == semakptr->cred->cr_prison ||
609 	      prison_ischild(rpr, semakptr->cred->cr_prison)))
610 		return (EINVAL);
611 	return (0);
612 }
613 
614 /*
615  * Note that the user-mode half of this passes a union, not a pointer.
616  */
617 #ifndef _SYS_SYSPROTO_H_
618 struct __semctl_args {
619 	int	semid;
620 	int	semnum;
621 	int	cmd;
622 	union	semun *arg;
623 };
624 #endif
625 int
626 sys___semctl(struct thread *td, struct __semctl_args *uap)
627 {
628 	struct semid_ds dsbuf;
629 	union semun arg, semun;
630 	register_t rval;
631 	int error;
632 
633 	switch (uap->cmd) {
634 	case SEM_STAT:
635 	case IPC_SET:
636 	case IPC_STAT:
637 	case GETALL:
638 	case SETVAL:
639 	case SETALL:
640 		error = copyin(uap->arg, &arg, sizeof(arg));
641 		if (error)
642 			return (error);
643 		break;
644 	}
645 
646 	switch (uap->cmd) {
647 	case SEM_STAT:
648 	case IPC_STAT:
649 		semun.buf = &dsbuf;
650 		break;
651 	case IPC_SET:
652 		error = copyin(arg.buf, &dsbuf, sizeof(dsbuf));
653 		if (error)
654 			return (error);
655 		semun.buf = &dsbuf;
656 		break;
657 	case GETALL:
658 	case SETALL:
659 		semun.array = arg.array;
660 		break;
661 	case SETVAL:
662 		semun.val = arg.val;
663 		break;
664 	}
665 
666 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
667 	    &rval);
668 	if (error)
669 		return (error);
670 
671 	switch (uap->cmd) {
672 	case SEM_STAT:
673 	case IPC_STAT:
674 		error = copyout(&dsbuf, arg.buf, sizeof(dsbuf));
675 		break;
676 	}
677 
678 	if (error == 0)
679 		td->td_retval[0] = rval;
680 	return (error);
681 }
682 
683 int
684 kern_semctl(struct thread *td, int semid, int semnum, int cmd,
685     union semun *arg, register_t *rval)
686 {
687 	u_short *array;
688 	struct ucred *cred = td->td_ucred;
689 	int i, error;
690 	struct prison *rpr;
691 	struct semid_ds *sbuf;
692 	struct semid_kernel *semakptr;
693 	struct mtx *sema_mtxp;
694 	u_short usval, count;
695 	int semidx;
696 
697 	DPRINTF(("call to semctl(%d, %d, %d, 0x%p)\n",
698 	    semid, semnum, cmd, arg));
699 
700 	AUDIT_ARG_SVIPC_CMD(cmd);
701 	AUDIT_ARG_SVIPC_ID(semid);
702 
703 	rpr = sem_find_prison(td->td_ucred);
704 	if (sem == NULL)
705 		return (ENOSYS);
706 
707 	array = NULL;
708 
709 	switch(cmd) {
710 	case SEM_STAT:
711 		/*
712 		 * For this command we assume semid is an array index
713 		 * rather than an IPC id.
714 		 */
715 		if (semid < 0 || semid >= seminfo.semmni)
716 			return (EINVAL);
717 		semakptr = &sema[semid];
718 		sema_mtxp = &sema_mtx[semid];
719 		mtx_lock(sema_mtxp);
720 		if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
721 			error = EINVAL;
722 			goto done2;
723 		}
724 		if ((error = sem_prison_cansee(rpr, semakptr)))
725 			goto done2;
726 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
727 			goto done2;
728 #ifdef MAC
729 		error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
730 		if (error != 0)
731 			goto done2;
732 #endif
733 		bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
734 		if (cred->cr_prison != semakptr->cred->cr_prison)
735 			arg->buf->sem_perm.key = IPC_PRIVATE;
736 		*rval = IXSEQ_TO_IPCID(semid, semakptr->u.sem_perm);
737 		mtx_unlock(sema_mtxp);
738 		return (0);
739 	}
740 
741 	semidx = IPCID_TO_IX(semid);
742 	if (semidx < 0 || semidx >= seminfo.semmni)
743 		return (EINVAL);
744 
745 	semakptr = &sema[semidx];
746 	sema_mtxp = &sema_mtx[semidx];
747 	if (cmd == IPC_RMID)
748 		mtx_lock(&sem_mtx);
749 	mtx_lock(sema_mtxp);
750 
751 #ifdef MAC
752 	error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
753 	if (error != 0)
754 		goto done2;
755 #endif
756 
757 	error = 0;
758 	*rval = 0;
759 
760 	switch (cmd) {
761 	case IPC_RMID:
762 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
763 			goto done2;
764 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
765 			goto done2;
766 		sem_remove(semidx, cred);
767 		break;
768 
769 	case IPC_SET:
770 		AUDIT_ARG_SVIPC_PERM(&arg->buf->sem_perm);
771 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
772 			goto done2;
773 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
774 			goto done2;
775 		sbuf = arg->buf;
776 		semakptr->u.sem_perm.uid = sbuf->sem_perm.uid;
777 		semakptr->u.sem_perm.gid = sbuf->sem_perm.gid;
778 		semakptr->u.sem_perm.mode = (semakptr->u.sem_perm.mode &
779 		    ~0777) | (sbuf->sem_perm.mode & 0777);
780 		semakptr->u.sem_ctime = time_second;
781 		break;
782 
783 	case IPC_STAT:
784 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
785 			goto done2;
786 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
787 			goto done2;
788 		bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
789 		if (cred->cr_prison != semakptr->cred->cr_prison)
790 			arg->buf->sem_perm.key = IPC_PRIVATE;
791 		break;
792 
793 	case GETNCNT:
794 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
795 			goto done2;
796 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
797 			goto done2;
798 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
799 			error = EINVAL;
800 			goto done2;
801 		}
802 		*rval = semakptr->u.sem_base[semnum].semncnt;
803 		break;
804 
805 	case GETPID:
806 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
807 			goto done2;
808 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
809 			goto done2;
810 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
811 			error = EINVAL;
812 			goto done2;
813 		}
814 		*rval = semakptr->u.sem_base[semnum].sempid;
815 		break;
816 
817 	case GETVAL:
818 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
819 			goto done2;
820 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
821 			goto done2;
822 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
823 			error = EINVAL;
824 			goto done2;
825 		}
826 		*rval = semakptr->u.sem_base[semnum].semval;
827 		break;
828 
829 	case GETALL:
830 		/*
831 		 * Unfortunately, callers of this function don't know
832 		 * in advance how many semaphores are in this set.
833 		 * While we could just allocate the maximum size array
834 		 * and pass the actual size back to the caller, that
835 		 * won't work for SETALL since we can't copyin() more
836 		 * data than the user specified as we may return a
837 		 * spurious EFAULT.
838 		 *
839 		 * Note that the number of semaphores in a set is
840 		 * fixed for the life of that set.  The only way that
841 		 * the 'count' could change while are blocked in
842 		 * malloc() is if this semaphore set were destroyed
843 		 * and a new one created with the same index.
844 		 * However, semvalid() will catch that due to the
845 		 * sequence number unless exactly 0x8000 (or a
846 		 * multiple thereof) semaphore sets for the same index
847 		 * are created and destroyed while we are in malloc!
848 		 *
849 		 */
850 		count = semakptr->u.sem_nsems;
851 		mtx_unlock(sema_mtxp);
852 		array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
853 		mtx_lock(sema_mtxp);
854 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
855 			goto done2;
856 		KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
857 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
858 			goto done2;
859 		for (i = 0; i < semakptr->u.sem_nsems; i++)
860 			array[i] = semakptr->u.sem_base[i].semval;
861 		mtx_unlock(sema_mtxp);
862 		error = copyout(array, arg->array, count * sizeof(*array));
863 		mtx_lock(sema_mtxp);
864 		break;
865 
866 	case GETZCNT:
867 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
868 			goto done2;
869 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
870 			goto done2;
871 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
872 			error = EINVAL;
873 			goto done2;
874 		}
875 		*rval = semakptr->u.sem_base[semnum].semzcnt;
876 		break;
877 
878 	case SETVAL:
879 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
880 			goto done2;
881 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
882 			goto done2;
883 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
884 			error = EINVAL;
885 			goto done2;
886 		}
887 		if (arg->val < 0 || arg->val > seminfo.semvmx) {
888 			error = ERANGE;
889 			goto done2;
890 		}
891 		semakptr->u.sem_base[semnum].semval = arg->val;
892 		SEMUNDO_LOCK();
893 		semundo_clear(semidx, semnum);
894 		SEMUNDO_UNLOCK();
895 		wakeup(semakptr);
896 		break;
897 
898 	case SETALL:
899 		/*
900 		 * See comment on GETALL for why 'count' shouldn't change
901 		 * and why we require a userland buffer.
902 		 */
903 		count = semakptr->u.sem_nsems;
904 		mtx_unlock(sema_mtxp);
905 		array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
906 		error = copyin(arg->array, array, count * sizeof(*array));
907 		mtx_lock(sema_mtxp);
908 		if (error)
909 			break;
910 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
911 			goto done2;
912 		KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
913 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
914 			goto done2;
915 		for (i = 0; i < semakptr->u.sem_nsems; i++) {
916 			usval = array[i];
917 			if (usval > seminfo.semvmx) {
918 				error = ERANGE;
919 				break;
920 			}
921 			semakptr->u.sem_base[i].semval = usval;
922 		}
923 		SEMUNDO_LOCK();
924 		semundo_clear(semidx, -1);
925 		SEMUNDO_UNLOCK();
926 		wakeup(semakptr);
927 		break;
928 
929 	default:
930 		error = EINVAL;
931 		break;
932 	}
933 
934 done2:
935 	mtx_unlock(sema_mtxp);
936 	if (cmd == IPC_RMID)
937 		mtx_unlock(&sem_mtx);
938 	if (array != NULL)
939 		free(array, M_TEMP);
940 	return(error);
941 }
942 
943 #ifndef _SYS_SYSPROTO_H_
944 struct semget_args {
945 	key_t	key;
946 	int	nsems;
947 	int	semflg;
948 };
949 #endif
950 int
951 sys_semget(struct thread *td, struct semget_args *uap)
952 {
953 	int semid, error = 0;
954 	int key = uap->key;
955 	int nsems = uap->nsems;
956 	int semflg = uap->semflg;
957 	struct ucred *cred = td->td_ucred;
958 
959 	DPRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg));
960 
961 	AUDIT_ARG_VALUE(semflg);
962 
963 	if (sem_find_prison(cred) == NULL)
964 		return (ENOSYS);
965 
966 	mtx_lock(&sem_mtx);
967 	if (key != IPC_PRIVATE) {
968 		for (semid = 0; semid < seminfo.semmni; semid++) {
969 			if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) &&
970 			    sema[semid].cred != NULL &&
971 			    sema[semid].cred->cr_prison == cred->cr_prison &&
972 			    sema[semid].u.sem_perm.key == key)
973 				break;
974 		}
975 		if (semid < seminfo.semmni) {
976 			AUDIT_ARG_SVIPC_ID(semid);
977 			DPRINTF(("found public key\n"));
978 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
979 				DPRINTF(("not exclusive\n"));
980 				error = EEXIST;
981 				goto done2;
982 			}
983 			if ((error = ipcperm(td, &sema[semid].u.sem_perm,
984 			    semflg & 0700))) {
985 				goto done2;
986 			}
987 			if (nsems > 0 && sema[semid].u.sem_nsems < nsems) {
988 				DPRINTF(("too small\n"));
989 				error = EINVAL;
990 				goto done2;
991 			}
992 #ifdef MAC
993 			error = mac_sysvsem_check_semget(cred, &sema[semid]);
994 			if (error != 0)
995 				goto done2;
996 #endif
997 			goto found;
998 		}
999 	}
1000 
1001 	DPRINTF(("need to allocate the semid_kernel\n"));
1002 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
1003 		if (nsems <= 0 || nsems > seminfo.semmsl) {
1004 			DPRINTF(("nsems out of range (0<%d<=%d)\n", nsems,
1005 			    seminfo.semmsl));
1006 			error = EINVAL;
1007 			goto done2;
1008 		}
1009 		if (nsems > seminfo.semmns - semtot) {
1010 			DPRINTF((
1011 			    "not enough semaphores left (need %d, got %d)\n",
1012 			    nsems, seminfo.semmns - semtot));
1013 			error = ENOSPC;
1014 			goto done2;
1015 		}
1016 		for (semid = 0; semid < seminfo.semmni; semid++) {
1017 			if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0)
1018 				break;
1019 		}
1020 		if (semid == seminfo.semmni) {
1021 			DPRINTF(("no more semid_kernel's available\n"));
1022 			error = ENOSPC;
1023 			goto done2;
1024 		}
1025 #ifdef RACCT
1026 		if (racct_enable) {
1027 			PROC_LOCK(td->td_proc);
1028 			error = racct_add(td->td_proc, RACCT_NSEM, nsems);
1029 			PROC_UNLOCK(td->td_proc);
1030 			if (error != 0) {
1031 				error = ENOSPC;
1032 				goto done2;
1033 			}
1034 		}
1035 #endif
1036 		DPRINTF(("semid %d is available\n", semid));
1037 		mtx_lock(&sema_mtx[semid]);
1038 		KASSERT((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0,
1039 		    ("Lost semaphore %d", semid));
1040 		sema[semid].u.sem_perm.key = key;
1041 		sema[semid].u.sem_perm.cuid = cred->cr_uid;
1042 		sema[semid].u.sem_perm.uid = cred->cr_uid;
1043 		sema[semid].u.sem_perm.cgid = cred->cr_gid;
1044 		sema[semid].u.sem_perm.gid = cred->cr_gid;
1045 		sema[semid].u.sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
1046 		sema[semid].cred = crhold(cred);
1047 		sema[semid].u.sem_perm.seq =
1048 		    (sema[semid].u.sem_perm.seq + 1) & 0x7fff;
1049 		sema[semid].u.sem_nsems = nsems;
1050 		sema[semid].u.sem_otime = 0;
1051 		sema[semid].u.sem_ctime = time_second;
1052 		sema[semid].u.sem_base = &sem[semtot];
1053 		semtot += nsems;
1054 		bzero(sema[semid].u.sem_base,
1055 		    sizeof(sema[semid].u.sem_base[0])*nsems);
1056 #ifdef MAC
1057 		mac_sysvsem_create(cred, &sema[semid]);
1058 #endif
1059 		mtx_unlock(&sema_mtx[semid]);
1060 		DPRINTF(("sembase = %p, next = %p\n",
1061 		    sema[semid].u.sem_base, &sem[semtot]));
1062 	} else {
1063 		DPRINTF(("didn't find it and wasn't asked to create it\n"));
1064 		error = ENOENT;
1065 		goto done2;
1066 	}
1067 
1068 found:
1069 	td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].u.sem_perm);
1070 done2:
1071 	mtx_unlock(&sem_mtx);
1072 	return (error);
1073 }
1074 
1075 #ifndef _SYS_SYSPROTO_H_
1076 struct semop_args {
1077 	int	semid;
1078 	struct	sembuf *sops;
1079 	size_t	nsops;
1080 };
1081 #endif
1082 int
1083 sys_semop(struct thread *td, struct semop_args *uap)
1084 {
1085 #define SMALL_SOPS	8
1086 	struct sembuf small_sops[SMALL_SOPS];
1087 	int semid = uap->semid;
1088 	size_t nsops = uap->nsops;
1089 	struct prison *rpr;
1090 	struct sembuf *sops;
1091 	struct semid_kernel *semakptr;
1092 	struct sembuf *sopptr = NULL;
1093 	struct sem *semptr = NULL;
1094 	struct sem_undo *suptr;
1095 	struct mtx *sema_mtxp;
1096 	size_t i, j, k;
1097 	int error;
1098 	int do_wakeup, do_undos;
1099 	unsigned short seq;
1100 
1101 #ifdef SEM_DEBUG
1102 	sops = NULL;
1103 #endif
1104 	DPRINTF(("call to semop(%d, %p, %u)\n", semid, sops, nsops));
1105 
1106 	AUDIT_ARG_SVIPC_ID(semid);
1107 
1108 	rpr = sem_find_prison(td->td_ucred);
1109 	if (sem == NULL)
1110 		return (ENOSYS);
1111 
1112 	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
1113 
1114 	if (semid < 0 || semid >= seminfo.semmni)
1115 		return (EINVAL);
1116 
1117 	/* Allocate memory for sem_ops */
1118 	if (nsops <= SMALL_SOPS)
1119 		sops = small_sops;
1120 	else if (nsops > seminfo.semopm) {
1121 		DPRINTF(("too many sops (max=%d, nsops=%d)\n", seminfo.semopm,
1122 		    nsops));
1123 		return (E2BIG);
1124 	} else {
1125 #ifdef RACCT
1126 		if (racct_enable) {
1127 			PROC_LOCK(td->td_proc);
1128 			if (nsops >
1129 			    racct_get_available(td->td_proc, RACCT_NSEMOP)) {
1130 				PROC_UNLOCK(td->td_proc);
1131 				return (E2BIG);
1132 			}
1133 			PROC_UNLOCK(td->td_proc);
1134 		}
1135 #endif
1136 
1137 		sops = malloc(nsops * sizeof(*sops), M_TEMP, M_WAITOK);
1138 	}
1139 	if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) {
1140 		DPRINTF(("error = %d from copyin(%p, %p, %d)\n", error,
1141 		    uap->sops, sops, nsops * sizeof(sops[0])));
1142 		if (sops != small_sops)
1143 			free(sops, M_SEM);
1144 		return (error);
1145 	}
1146 
1147 	semakptr = &sema[semid];
1148 	sema_mtxp = &sema_mtx[semid];
1149 	mtx_lock(sema_mtxp);
1150 	if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
1151 		error = EINVAL;
1152 		goto done2;
1153 	}
1154 	seq = semakptr->u.sem_perm.seq;
1155 	if (seq != IPCID_TO_SEQ(uap->semid)) {
1156 		error = EINVAL;
1157 		goto done2;
1158 	}
1159 	if ((error = sem_prison_cansee(rpr, semakptr)) != 0)
1160 		goto done2;
1161 	/*
1162 	 * Initial pass through sops to see what permissions are needed.
1163 	 * Also perform any checks that don't need repeating on each
1164 	 * attempt to satisfy the request vector.
1165 	 */
1166 	j = 0;		/* permission needed */
1167 	do_undos = 0;
1168 	for (i = 0; i < nsops; i++) {
1169 		sopptr = &sops[i];
1170 		if (sopptr->sem_num >= semakptr->u.sem_nsems) {
1171 			error = EFBIG;
1172 			goto done2;
1173 		}
1174 		if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0)
1175 			do_undos = 1;
1176 		j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A;
1177 	}
1178 
1179 	if ((error = ipcperm(td, &semakptr->u.sem_perm, j))) {
1180 		DPRINTF(("error = %d from ipaccess\n", error));
1181 		goto done2;
1182 	}
1183 #ifdef MAC
1184 	error = mac_sysvsem_check_semop(td->td_ucred, semakptr, j);
1185 	if (error != 0)
1186 		goto done2;
1187 #endif
1188 
1189 	/*
1190 	 * Loop trying to satisfy the vector of requests.
1191 	 * If we reach a point where we must wait, any requests already
1192 	 * performed are rolled back and we go to sleep until some other
1193 	 * process wakes us up.  At this point, we start all over again.
1194 	 *
1195 	 * This ensures that from the perspective of other tasks, a set
1196 	 * of requests is atomic (never partially satisfied).
1197 	 */
1198 	for (;;) {
1199 		do_wakeup = 0;
1200 		error = 0;	/* error return if necessary */
1201 
1202 		for (i = 0; i < nsops; i++) {
1203 			sopptr = &sops[i];
1204 			semptr = &semakptr->u.sem_base[sopptr->sem_num];
1205 
1206 			DPRINTF((
1207 			    "semop:  semakptr=%p, sem_base=%p, "
1208 			    "semptr=%p, sem[%d]=%d : op=%d, flag=%s\n",
1209 			    semakptr, semakptr->u.sem_base, semptr,
1210 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
1211 			    (sopptr->sem_flg & IPC_NOWAIT) ?
1212 			    "nowait" : "wait"));
1213 
1214 			if (sopptr->sem_op < 0) {
1215 				if (semptr->semval + sopptr->sem_op < 0) {
1216 					DPRINTF(("semop:  can't do it now\n"));
1217 					break;
1218 				} else {
1219 					semptr->semval += sopptr->sem_op;
1220 					if (semptr->semval == 0 &&
1221 					    semptr->semzcnt > 0)
1222 						do_wakeup = 1;
1223 				}
1224 			} else if (sopptr->sem_op == 0) {
1225 				if (semptr->semval != 0) {
1226 					DPRINTF(("semop:  not zero now\n"));
1227 					break;
1228 				}
1229 			} else if (semptr->semval + sopptr->sem_op >
1230 			    seminfo.semvmx) {
1231 				error = ERANGE;
1232 				break;
1233 			} else {
1234 				if (semptr->semncnt > 0)
1235 					do_wakeup = 1;
1236 				semptr->semval += sopptr->sem_op;
1237 			}
1238 		}
1239 
1240 		/*
1241 		 * Did we get through the entire vector?
1242 		 */
1243 		if (i >= nsops)
1244 			goto done;
1245 
1246 		/*
1247 		 * No ... rollback anything that we've already done
1248 		 */
1249 		DPRINTF(("semop:  rollback 0 through %d\n", i-1));
1250 		for (j = 0; j < i; j++)
1251 			semakptr->u.sem_base[sops[j].sem_num].semval -=
1252 			    sops[j].sem_op;
1253 
1254 		/* If we detected an error, return it */
1255 		if (error != 0)
1256 			goto done2;
1257 
1258 		/*
1259 		 * If the request that we couldn't satisfy has the
1260 		 * NOWAIT flag set then return with EAGAIN.
1261 		 */
1262 		if (sopptr->sem_flg & IPC_NOWAIT) {
1263 			error = EAGAIN;
1264 			goto done2;
1265 		}
1266 
1267 		if (sopptr->sem_op == 0)
1268 			semptr->semzcnt++;
1269 		else
1270 			semptr->semncnt++;
1271 
1272 		DPRINTF(("semop:  good night!\n"));
1273 		error = msleep(semakptr, sema_mtxp, (PZERO - 4) | PCATCH,
1274 		    "semwait", 0);
1275 		DPRINTF(("semop:  good morning (error=%d)!\n", error));
1276 		/* return code is checked below, after sem[nz]cnt-- */
1277 
1278 		/*
1279 		 * Make sure that the semaphore still exists
1280 		 */
1281 		seq = semakptr->u.sem_perm.seq;
1282 		if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1283 		    seq != IPCID_TO_SEQ(uap->semid)) {
1284 			error = EIDRM;
1285 			goto done2;
1286 		}
1287 
1288 		/*
1289 		 * Renew the semaphore's pointer after wakeup since
1290 		 * during msleep sem_base may have been modified and semptr
1291 		 * is not valid any more
1292 		 */
1293 		semptr = &semakptr->u.sem_base[sopptr->sem_num];
1294 
1295 		/*
1296 		 * The semaphore is still alive.  Readjust the count of
1297 		 * waiting processes.
1298 		 */
1299 		if (sopptr->sem_op == 0)
1300 			semptr->semzcnt--;
1301 		else
1302 			semptr->semncnt--;
1303 
1304 		/*
1305 		 * Is it really morning, or was our sleep interrupted?
1306 		 * (Delayed check of msleep() return code because we
1307 		 * need to decrement sem[nz]cnt either way.)
1308 		 */
1309 		if (error != 0) {
1310 			error = EINTR;
1311 			goto done2;
1312 		}
1313 		DPRINTF(("semop:  good morning!\n"));
1314 	}
1315 
1316 done:
1317 	/*
1318 	 * Process any SEM_UNDO requests.
1319 	 */
1320 	if (do_undos) {
1321 		SEMUNDO_LOCK();
1322 		suptr = NULL;
1323 		for (i = 0; i < nsops; i++) {
1324 			/*
1325 			 * We only need to deal with SEM_UNDO's for non-zero
1326 			 * op's.
1327 			 */
1328 			int adjval;
1329 
1330 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
1331 				continue;
1332 			adjval = sops[i].sem_op;
1333 			if (adjval == 0)
1334 				continue;
1335 			error = semundo_adjust(td, &suptr, semid, seq,
1336 			    sops[i].sem_num, -adjval);
1337 			if (error == 0)
1338 				continue;
1339 
1340 			/*
1341 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
1342 			 * Rollback the adjustments to this point and then
1343 			 * rollback the semaphore ups and down so we can return
1344 			 * with an error with all structures restored.  We
1345 			 * rollback the undo's in the exact reverse order that
1346 			 * we applied them.  This guarantees that we won't run
1347 			 * out of space as we roll things back out.
1348 			 */
1349 			for (j = 0; j < i; j++) {
1350 				k = i - j - 1;
1351 				if ((sops[k].sem_flg & SEM_UNDO) == 0)
1352 					continue;
1353 				adjval = sops[k].sem_op;
1354 				if (adjval == 0)
1355 					continue;
1356 				if (semundo_adjust(td, &suptr, semid, seq,
1357 				    sops[k].sem_num, adjval) != 0)
1358 					panic("semop - can't undo undos");
1359 			}
1360 
1361 			for (j = 0; j < nsops; j++)
1362 				semakptr->u.sem_base[sops[j].sem_num].semval -=
1363 				    sops[j].sem_op;
1364 
1365 			DPRINTF(("error = %d from semundo_adjust\n", error));
1366 			SEMUNDO_UNLOCK();
1367 			goto done2;
1368 		} /* loop through the sops */
1369 		SEMUNDO_UNLOCK();
1370 	} /* if (do_undos) */
1371 
1372 	/* We're definitely done - set the sempid's and time */
1373 	for (i = 0; i < nsops; i++) {
1374 		sopptr = &sops[i];
1375 		semptr = &semakptr->u.sem_base[sopptr->sem_num];
1376 		semptr->sempid = td->td_proc->p_pid;
1377 	}
1378 	semakptr->u.sem_otime = time_second;
1379 
1380 	/*
1381 	 * Do a wakeup if any semaphore was up'd whilst something was
1382 	 * sleeping on it.
1383 	 */
1384 	if (do_wakeup) {
1385 		DPRINTF(("semop:  doing wakeup\n"));
1386 		wakeup(semakptr);
1387 		DPRINTF(("semop:  back from wakeup\n"));
1388 	}
1389 	DPRINTF(("semop:  done\n"));
1390 	td->td_retval[0] = 0;
1391 done2:
1392 	mtx_unlock(sema_mtxp);
1393 	if (sops != small_sops)
1394 		free(sops, M_SEM);
1395 	return (error);
1396 }
1397 
1398 /*
1399  * Go through the undo structures for this process and apply the adjustments to
1400  * semaphores.
1401  */
1402 static void
1403 semexit_myhook(void *arg, struct proc *p)
1404 {
1405 	struct sem_undo *suptr;
1406 	struct semid_kernel *semakptr;
1407 	struct mtx *sema_mtxp;
1408 	int semid, semnum, adjval, ix;
1409 	unsigned short seq;
1410 
1411 	/*
1412 	 * Go through the chain of undo vectors looking for one
1413 	 * associated with this process.
1414 	 */
1415 	if (LIST_EMPTY(&semu_list))
1416 		return;
1417 	SEMUNDO_LOCK();
1418 	LIST_FOREACH(suptr, &semu_list, un_next) {
1419 		if (suptr->un_proc == p)
1420 			break;
1421 	}
1422 	if (suptr == NULL) {
1423 		SEMUNDO_UNLOCK();
1424 		return;
1425 	}
1426 	LIST_REMOVE(suptr, un_next);
1427 
1428 	DPRINTF(("proc @%p has undo structure with %d entries\n", p,
1429 	    suptr->un_cnt));
1430 
1431 	/*
1432 	 * If there are any active undo elements then process them.
1433 	 */
1434 	if (suptr->un_cnt > 0) {
1435 		SEMUNDO_UNLOCK();
1436 		for (ix = 0; ix < suptr->un_cnt; ix++) {
1437 			semid = suptr->un_ent[ix].un_id;
1438 			semnum = suptr->un_ent[ix].un_num;
1439 			adjval = suptr->un_ent[ix].un_adjval;
1440 			seq = suptr->un_ent[ix].un_seq;
1441 			semakptr = &sema[semid];
1442 			sema_mtxp = &sema_mtx[semid];
1443 
1444 			mtx_lock(sema_mtxp);
1445 			if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1446 			    (semakptr->u.sem_perm.seq != seq)) {
1447 				mtx_unlock(sema_mtxp);
1448 				continue;
1449 			}
1450 			if (semnum >= semakptr->u.sem_nsems)
1451 				panic("semexit - semnum out of range");
1452 
1453 			DPRINTF((
1454 			    "semexit:  %p id=%d num=%d(adj=%d) ; sem=%d\n",
1455 			    suptr->un_proc, suptr->un_ent[ix].un_id,
1456 			    suptr->un_ent[ix].un_num,
1457 			    suptr->un_ent[ix].un_adjval,
1458 			    semakptr->u.sem_base[semnum].semval));
1459 
1460 			if (adjval < 0 && semakptr->u.sem_base[semnum].semval <
1461 			    -adjval)
1462 				semakptr->u.sem_base[semnum].semval = 0;
1463 			else
1464 				semakptr->u.sem_base[semnum].semval += adjval;
1465 
1466 			wakeup(semakptr);
1467 			DPRINTF(("semexit:  back from wakeup\n"));
1468 			mtx_unlock(sema_mtxp);
1469 		}
1470 		SEMUNDO_LOCK();
1471 	}
1472 
1473 	/*
1474 	 * Deallocate the undo vector.
1475 	 */
1476 	DPRINTF(("removing vector\n"));
1477 	suptr->un_proc = NULL;
1478 	suptr->un_cnt = 0;
1479 	LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
1480 	SEMUNDO_UNLOCK();
1481 }
1482 
1483 static int
1484 sysctl_sema(SYSCTL_HANDLER_ARGS)
1485 {
1486 	struct prison *pr, *rpr;
1487 	struct semid_kernel tsemak;
1488 	int error, i;
1489 
1490 	pr = req->td->td_ucred->cr_prison;
1491 	rpr = sem_find_prison(req->td->td_ucred);
1492 	error = 0;
1493 	for (i = 0; i < seminfo.semmni; i++) {
1494 		mtx_lock(&sema_mtx[i]);
1495 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) == 0 ||
1496 		    rpr == NULL || sem_prison_cansee(rpr, &sema[i]) != 0)
1497 			bzero(&tsemak, sizeof(tsemak));
1498 		else {
1499 			tsemak = sema[i];
1500 			if (tsemak.cred->cr_prison != pr)
1501 				tsemak.u.sem_perm.key = IPC_PRIVATE;
1502 		}
1503 		mtx_unlock(&sema_mtx[i]);
1504 		error = SYSCTL_OUT(req, &tsemak, sizeof(tsemak));
1505 		if (error != 0)
1506 			break;
1507 	}
1508 	return (error);
1509 }
1510 
1511 static int
1512 sem_prison_check(void *obj, void *data)
1513 {
1514 	struct prison *pr = obj;
1515 	struct prison *prpr;
1516 	struct vfsoptlist *opts = data;
1517 	int error, jsys;
1518 
1519 	/*
1520 	 * sysvsem is a jailsys integer.
1521 	 * It must be "disable" if the parent jail is disabled.
1522 	 */
1523 	error = vfs_copyopt(opts, "sysvsem", &jsys, sizeof(jsys));
1524 	if (error != ENOENT) {
1525 		if (error != 0)
1526 			return (error);
1527 		switch (jsys) {
1528 		case JAIL_SYS_DISABLE:
1529 			break;
1530 		case JAIL_SYS_NEW:
1531 		case JAIL_SYS_INHERIT:
1532 			prison_lock(pr->pr_parent);
1533 			prpr = osd_jail_get(pr->pr_parent, sem_prison_slot);
1534 			prison_unlock(pr->pr_parent);
1535 			if (prpr == NULL)
1536 				return (EPERM);
1537 			break;
1538 		default:
1539 			return (EINVAL);
1540 		}
1541 	}
1542 
1543 	return (0);
1544 }
1545 
1546 static int
1547 sem_prison_set(void *obj, void *data)
1548 {
1549 	struct prison *pr = obj;
1550 	struct prison *tpr, *orpr, *nrpr, *trpr;
1551 	struct vfsoptlist *opts = data;
1552 	void *rsv;
1553 	int jsys, descend;
1554 
1555 	/*
1556 	 * sysvsem controls which jail is the root of the associated sems (this
1557 	 * jail or same as the parent), or if the feature is available at all.
1558 	 */
1559 	if (vfs_copyopt(opts, "sysvsem", &jsys, sizeof(jsys)) == ENOENT)
1560 		jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0)
1561 		    ? JAIL_SYS_INHERIT
1562 		    : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0)
1563 		    ? JAIL_SYS_DISABLE
1564 		    : -1;
1565 	if (jsys == JAIL_SYS_DISABLE) {
1566 		prison_lock(pr);
1567 		orpr = osd_jail_get(pr, sem_prison_slot);
1568 		if (orpr != NULL)
1569 			osd_jail_del(pr, sem_prison_slot);
1570 		prison_unlock(pr);
1571 		if (orpr != NULL) {
1572 			if (orpr == pr)
1573 				sem_prison_cleanup(pr);
1574 			/* Disable all child jails as well. */
1575 			FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1576 				prison_lock(tpr);
1577 				trpr = osd_jail_get(tpr, sem_prison_slot);
1578 				if (trpr != NULL) {
1579 					osd_jail_del(tpr, sem_prison_slot);
1580 					prison_unlock(tpr);
1581 					if (trpr == tpr)
1582 						sem_prison_cleanup(tpr);
1583 				} else {
1584 					prison_unlock(tpr);
1585 					descend = 0;
1586 				}
1587 			}
1588 		}
1589 	} else if (jsys != -1) {
1590 		if (jsys == JAIL_SYS_NEW)
1591 			nrpr = pr;
1592 		else {
1593 			prison_lock(pr->pr_parent);
1594 			nrpr = osd_jail_get(pr->pr_parent, sem_prison_slot);
1595 			prison_unlock(pr->pr_parent);
1596 		}
1597 		rsv = osd_reserve(sem_prison_slot);
1598 		prison_lock(pr);
1599 		orpr = osd_jail_get(pr, sem_prison_slot);
1600 		if (orpr != nrpr)
1601 			(void)osd_jail_set_reserved(pr, sem_prison_slot, rsv,
1602 			    nrpr);
1603 		else
1604 			osd_free_reserved(rsv);
1605 		prison_unlock(pr);
1606 		if (orpr != nrpr) {
1607 			if (orpr == pr)
1608 				sem_prison_cleanup(pr);
1609 			if (orpr != NULL) {
1610 				/* Change child jails matching the old root, */
1611 				FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1612 					prison_lock(tpr);
1613 					trpr = osd_jail_get(tpr,
1614 					    sem_prison_slot);
1615 					if (trpr == orpr) {
1616 						(void)osd_jail_set(tpr,
1617 						    sem_prison_slot, nrpr);
1618 						prison_unlock(tpr);
1619 						if (trpr == tpr)
1620 							sem_prison_cleanup(tpr);
1621 					} else {
1622 						prison_unlock(tpr);
1623 						descend = 0;
1624 					}
1625 				}
1626 			}
1627 		}
1628 	}
1629 
1630 	return (0);
1631 }
1632 
1633 static int
1634 sem_prison_get(void *obj, void *data)
1635 {
1636 	struct prison *pr = obj;
1637 	struct prison *rpr;
1638 	struct vfsoptlist *opts = data;
1639 	int error, jsys;
1640 
1641 	/* Set sysvsem based on the jail's root prison. */
1642 	prison_lock(pr);
1643 	rpr = osd_jail_get(pr, sem_prison_slot);
1644 	prison_unlock(pr);
1645 	jsys = rpr == NULL ? JAIL_SYS_DISABLE
1646 	    : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT;
1647 	error = vfs_setopt(opts, "sysvsem", &jsys, sizeof(jsys));
1648 	if (error == ENOENT)
1649 		error = 0;
1650 	return (error);
1651 }
1652 
1653 static int
1654 sem_prison_remove(void *obj, void *data __unused)
1655 {
1656 	struct prison *pr = obj;
1657 	struct prison *rpr;
1658 
1659 	prison_lock(pr);
1660 	rpr = osd_jail_get(pr, sem_prison_slot);
1661 	prison_unlock(pr);
1662 	if (rpr == pr)
1663 		sem_prison_cleanup(pr);
1664 	return (0);
1665 }
1666 
1667 static void
1668 sem_prison_cleanup(struct prison *pr)
1669 {
1670 	int i;
1671 
1672 	/* Remove any sems that belong to this jail. */
1673 	mtx_lock(&sem_mtx);
1674 	for (i = 0; i < seminfo.semmni; i++) {
1675 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
1676 		    sema[i].cred != NULL && sema[i].cred->cr_prison == pr) {
1677 			mtx_lock(&sema_mtx[i]);
1678 			sem_remove(i, NULL);
1679 			mtx_unlock(&sema_mtx[i]);
1680 		}
1681 	}
1682 	mtx_unlock(&sem_mtx);
1683 }
1684 
1685 SYSCTL_JAIL_PARAM_SYS_NODE(sysvsem, CTLFLAG_RW, "SYSV semaphores");
1686 
1687 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1688     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1689 
1690 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1691 static sy_call_t *semcalls[] = {
1692 	(sy_call_t *)freebsd7___semctl, (sy_call_t *)sys_semget,
1693 	(sy_call_t *)sys_semop
1694 };
1695 
1696 /*
1697  * Entry point for all SEM calls.
1698  */
1699 int
1700 sys_semsys(td, uap)
1701 	struct thread *td;
1702 	/* XXX actually varargs. */
1703 	struct semsys_args /* {
1704 		int	which;
1705 		int	a2;
1706 		int	a3;
1707 		int	a4;
1708 		int	a5;
1709 	} */ *uap;
1710 {
1711 	int error;
1712 
1713 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1714 	if (uap->which < 0 || uap->which >= nitems(semcalls))
1715 		return (EINVAL);
1716 	error = (*semcalls[uap->which])(td, &uap->a2);
1717 	return (error);
1718 }
1719 
1720 #ifndef CP
1721 #define CP(src, dst, fld)	do { (dst).fld = (src).fld; } while (0)
1722 #endif
1723 
1724 #ifndef _SYS_SYSPROTO_H_
1725 struct freebsd7___semctl_args {
1726 	int	semid;
1727 	int	semnum;
1728 	int	cmd;
1729 	union	semun_old *arg;
1730 };
1731 #endif
1732 int
1733 freebsd7___semctl(struct thread *td, struct freebsd7___semctl_args *uap)
1734 {
1735 	struct semid_ds_old dsold;
1736 	struct semid_ds dsbuf;
1737 	union semun_old arg;
1738 	union semun semun;
1739 	register_t rval;
1740 	int error;
1741 
1742 	switch (uap->cmd) {
1743 	case SEM_STAT:
1744 	case IPC_SET:
1745 	case IPC_STAT:
1746 	case GETALL:
1747 	case SETVAL:
1748 	case SETALL:
1749 		error = copyin(uap->arg, &arg, sizeof(arg));
1750 		if (error)
1751 			return (error);
1752 		break;
1753 	}
1754 
1755 	switch (uap->cmd) {
1756 	case SEM_STAT:
1757 	case IPC_STAT:
1758 		semun.buf = &dsbuf;
1759 		break;
1760 	case IPC_SET:
1761 		error = copyin(arg.buf, &dsold, sizeof(dsold));
1762 		if (error)
1763 			return (error);
1764 		ipcperm_old2new(&dsold.sem_perm, &dsbuf.sem_perm);
1765 		CP(dsold, dsbuf, sem_base);
1766 		CP(dsold, dsbuf, sem_nsems);
1767 		CP(dsold, dsbuf, sem_otime);
1768 		CP(dsold, dsbuf, sem_ctime);
1769 		semun.buf = &dsbuf;
1770 		break;
1771 	case GETALL:
1772 	case SETALL:
1773 		semun.array = arg.array;
1774 		break;
1775 	case SETVAL:
1776 		semun.val = arg.val;
1777 		break;
1778 	}
1779 
1780 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1781 	    &rval);
1782 	if (error)
1783 		return (error);
1784 
1785 	switch (uap->cmd) {
1786 	case SEM_STAT:
1787 	case IPC_STAT:
1788 		bzero(&dsold, sizeof(dsold));
1789 		ipcperm_new2old(&dsbuf.sem_perm, &dsold.sem_perm);
1790 		CP(dsbuf, dsold, sem_base);
1791 		CP(dsbuf, dsold, sem_nsems);
1792 		CP(dsbuf, dsold, sem_otime);
1793 		CP(dsbuf, dsold, sem_ctime);
1794 		error = copyout(&dsold, arg.buf, sizeof(dsold));
1795 		break;
1796 	}
1797 
1798 	if (error == 0)
1799 		td->td_retval[0] = rval;
1800 	return (error);
1801 }
1802 
1803 #endif /* COMPAT_FREEBSD{4,5,6,7} */
1804 
1805 #ifdef COMPAT_FREEBSD32
1806 
1807 int
1808 freebsd32_semsys(struct thread *td, struct freebsd32_semsys_args *uap)
1809 {
1810 
1811 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1812     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1813 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1814 	switch (uap->which) {
1815 	case 0:
1816 		return (freebsd7_freebsd32_semctl(td,
1817 		    (struct freebsd7_freebsd32_semctl_args *)&uap->a2));
1818 	default:
1819 		return (sys_semsys(td, (struct semsys_args *)uap));
1820 	}
1821 #else
1822 	return (nosys(td, NULL));
1823 #endif
1824 }
1825 
1826 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1827     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1828 int
1829 freebsd7_freebsd32_semctl(struct thread *td,
1830     struct freebsd7_freebsd32_semctl_args *uap)
1831 {
1832 	struct semid_ds32_old dsbuf32;
1833 	struct semid_ds dsbuf;
1834 	union semun semun;
1835 	union semun32 arg;
1836 	register_t rval;
1837 	int error;
1838 
1839 	switch (uap->cmd) {
1840 	case SEM_STAT:
1841 	case IPC_SET:
1842 	case IPC_STAT:
1843 	case GETALL:
1844 	case SETVAL:
1845 	case SETALL:
1846 		error = copyin(uap->arg, &arg, sizeof(arg));
1847 		if (error)
1848 			return (error);
1849 		break;
1850 	}
1851 
1852 	switch (uap->cmd) {
1853 	case SEM_STAT:
1854 	case IPC_STAT:
1855 		semun.buf = &dsbuf;
1856 		break;
1857 	case IPC_SET:
1858 		error = copyin(PTRIN(arg.buf), &dsbuf32, sizeof(dsbuf32));
1859 		if (error)
1860 			return (error);
1861 		freebsd32_ipcperm_old_in(&dsbuf32.sem_perm, &dsbuf.sem_perm);
1862 		PTRIN_CP(dsbuf32, dsbuf, sem_base);
1863 		CP(dsbuf32, dsbuf, sem_nsems);
1864 		CP(dsbuf32, dsbuf, sem_otime);
1865 		CP(dsbuf32, dsbuf, sem_ctime);
1866 		semun.buf = &dsbuf;
1867 		break;
1868 	case GETALL:
1869 	case SETALL:
1870 		semun.array = PTRIN(arg.array);
1871 		break;
1872 	case SETVAL:
1873 		semun.val = arg.val;
1874 		break;
1875 	}
1876 
1877 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1878 	    &rval);
1879 	if (error)
1880 		return (error);
1881 
1882 	switch (uap->cmd) {
1883 	case SEM_STAT:
1884 	case IPC_STAT:
1885 		bzero(&dsbuf32, sizeof(dsbuf32));
1886 		freebsd32_ipcperm_old_out(&dsbuf.sem_perm, &dsbuf32.sem_perm);
1887 		PTROUT_CP(dsbuf, dsbuf32, sem_base);
1888 		CP(dsbuf, dsbuf32, sem_nsems);
1889 		CP(dsbuf, dsbuf32, sem_otime);
1890 		CP(dsbuf, dsbuf32, sem_ctime);
1891 		error = copyout(&dsbuf32, PTRIN(arg.buf), sizeof(dsbuf32));
1892 		break;
1893 	}
1894 
1895 	if (error == 0)
1896 		td->td_retval[0] = rval;
1897 	return (error);
1898 }
1899 #endif
1900 
1901 int
1902 freebsd32_semctl(struct thread *td, struct freebsd32_semctl_args *uap)
1903 {
1904 	struct semid_ds32 dsbuf32;
1905 	struct semid_ds dsbuf;
1906 	union semun semun;
1907 	union semun32 arg;
1908 	register_t rval;
1909 	int error;
1910 
1911 	switch (uap->cmd) {
1912 	case SEM_STAT:
1913 	case IPC_SET:
1914 	case IPC_STAT:
1915 	case GETALL:
1916 	case SETVAL:
1917 	case SETALL:
1918 		error = copyin(uap->arg, &arg, sizeof(arg));
1919 		if (error)
1920 			return (error);
1921 		break;
1922 	}
1923 
1924 	switch (uap->cmd) {
1925 	case SEM_STAT:
1926 	case IPC_STAT:
1927 		semun.buf = &dsbuf;
1928 		break;
1929 	case IPC_SET:
1930 		error = copyin(PTRIN(arg.buf), &dsbuf32, sizeof(dsbuf32));
1931 		if (error)
1932 			return (error);
1933 		freebsd32_ipcperm_in(&dsbuf32.sem_perm, &dsbuf.sem_perm);
1934 		PTRIN_CP(dsbuf32, dsbuf, sem_base);
1935 		CP(dsbuf32, dsbuf, sem_nsems);
1936 		CP(dsbuf32, dsbuf, sem_otime);
1937 		CP(dsbuf32, dsbuf, sem_ctime);
1938 		semun.buf = &dsbuf;
1939 		break;
1940 	case GETALL:
1941 	case SETALL:
1942 		semun.array = PTRIN(arg.array);
1943 		break;
1944 	case SETVAL:
1945 		semun.val = arg.val;
1946 		break;
1947 	}
1948 
1949 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1950 	    &rval);
1951 	if (error)
1952 		return (error);
1953 
1954 	switch (uap->cmd) {
1955 	case SEM_STAT:
1956 	case IPC_STAT:
1957 		bzero(&dsbuf32, sizeof(dsbuf32));
1958 		freebsd32_ipcperm_out(&dsbuf.sem_perm, &dsbuf32.sem_perm);
1959 		PTROUT_CP(dsbuf, dsbuf32, sem_base);
1960 		CP(dsbuf, dsbuf32, sem_nsems);
1961 		CP(dsbuf, dsbuf32, sem_otime);
1962 		CP(dsbuf, dsbuf32, sem_ctime);
1963 		error = copyout(&dsbuf32, PTRIN(arg.buf), sizeof(dsbuf32));
1964 		break;
1965 	}
1966 
1967 	if (error == 0)
1968 		td->td_retval[0] = rval;
1969 	return (error);
1970 }
1971 
1972 #endif /* COMPAT_FREEBSD32 */
1973