xref: /freebsd/sys/kern/sysv_sem.c (revision 2f513db72b034fd5ef7f080b11be5c711c15186a)
1 /*-
2  * Implementation of SVID semaphores
3  *
4  * Author:  Daniel Boulet
5  *
6  * This software is provided ``AS IS'' without any warranties of any kind.
7  */
8 /*-
9  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
10  *
11  * Copyright (c) 2003-2005 McAfee, Inc.
12  * Copyright (c) 2016-2017 Robert N. M. Watson
13  * All rights reserved.
14  *
15  * This software was developed for the FreeBSD Project in part by McAfee
16  * Research, the Security Research Division of McAfee, Inc under DARPA/SPAWAR
17  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS research
18  * program.
19  *
20  * Portions of this software were developed by BAE Systems, the University of
21  * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL
22  * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent
23  * Computing (TC) research program.
24  *
25  * Redistribution and use in source and binary forms, with or without
26  * modification, are permitted provided that the following conditions
27  * are met:
28  * 1. Redistributions of source code must retain the above copyright
29  *    notice, this list of conditions and the following disclaimer.
30  * 2. Redistributions in binary form must reproduce the above copyright
31  *    notice, this list of conditions and the following disclaimer in the
32  *    documentation and/or other materials provided with the distribution.
33  *
34  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
35  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
36  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
37  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
38  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
39  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
40  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
41  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
42  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
43  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
44  * SUCH DAMAGE.
45  */
46 
47 #include <sys/cdefs.h>
48 __FBSDID("$FreeBSD$");
49 
50 #include "opt_sysvipc.h"
51 
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/sysproto.h>
55 #include <sys/eventhandler.h>
56 #include <sys/kernel.h>
57 #include <sys/proc.h>
58 #include <sys/lock.h>
59 #include <sys/module.h>
60 #include <sys/mutex.h>
61 #include <sys/racct.h>
62 #include <sys/sem.h>
63 #include <sys/sx.h>
64 #include <sys/syscall.h>
65 #include <sys/syscallsubr.h>
66 #include <sys/sysent.h>
67 #include <sys/sysctl.h>
68 #include <sys/uio.h>
69 #include <sys/malloc.h>
70 #include <sys/jail.h>
71 
72 #include <security/audit/audit.h>
73 #include <security/mac/mac_framework.h>
74 
75 FEATURE(sysv_sem, "System V semaphores support");
76 
77 static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores");
78 
79 #ifdef SEM_DEBUG
80 #define DPRINTF(a)	printf a
81 #else
82 #define DPRINTF(a)
83 #endif
84 
85 static int seminit(void);
86 static int sysvsem_modload(struct module *, int, void *);
87 static int semunload(void);
88 static void semexit_myhook(void *arg, struct proc *p);
89 static int sysctl_sema(SYSCTL_HANDLER_ARGS);
90 static int semvalid(int semid, struct prison *rpr,
91     struct semid_kernel *semakptr);
92 static void sem_remove(int semidx, struct ucred *cred);
93 static struct prison *sem_find_prison(struct ucred *);
94 static int sem_prison_cansee(struct prison *, struct semid_kernel *);
95 static int sem_prison_check(void *, void *);
96 static int sem_prison_set(void *, void *);
97 static int sem_prison_get(void *, void *);
98 static int sem_prison_remove(void *, void *);
99 static void sem_prison_cleanup(struct prison *);
100 
101 #ifndef _SYS_SYSPROTO_H_
102 struct __semctl_args;
103 int __semctl(struct thread *td, struct __semctl_args *uap);
104 struct semget_args;
105 int semget(struct thread *td, struct semget_args *uap);
106 struct semop_args;
107 int semop(struct thread *td, struct semop_args *uap);
108 #endif
109 
110 static struct sem_undo *semu_alloc(struct thread *td);
111 static int semundo_adjust(struct thread *td, struct sem_undo **supptr,
112     int semid, int semseq, int semnum, int adjval);
113 static void semundo_clear(int semid, int semnum);
114 
115 static struct mtx	sem_mtx;	/* semaphore global lock */
116 static struct mtx sem_undo_mtx;
117 static int	semtot = 0;
118 static struct semid_kernel *sema;	/* semaphore id pool */
119 static struct mtx *sema_mtx;	/* semaphore id pool mutexes*/
120 static struct sem *sem;		/* semaphore pool */
121 LIST_HEAD(, sem_undo) semu_list;	/* list of active undo structures */
122 LIST_HEAD(, sem_undo) semu_free_list;	/* list of free undo structures */
123 static int	*semu;		/* undo structure pool */
124 static eventhandler_tag semexit_tag;
125 static unsigned sem_prison_slot;	/* prison OSD slot */
126 
127 #define SEMUNDO_MTX		sem_undo_mtx
128 #define SEMUNDO_LOCK()		mtx_lock(&SEMUNDO_MTX);
129 #define SEMUNDO_UNLOCK()	mtx_unlock(&SEMUNDO_MTX);
130 #define SEMUNDO_LOCKASSERT(how)	mtx_assert(&SEMUNDO_MTX, (how));
131 
132 struct sem {
133 	u_short	semval;		/* semaphore value */
134 	pid_t	sempid;		/* pid of last operation */
135 	u_short	semncnt;	/* # awaiting semval > cval */
136 	u_short	semzcnt;	/* # awaiting semval = 0 */
137 };
138 
139 /*
140  * Undo structure (one per process)
141  */
142 struct sem_undo {
143 	LIST_ENTRY(sem_undo) un_next;	/* ptr to next active undo structure */
144 	struct	proc *un_proc;		/* owner of this structure */
145 	short	un_cnt;			/* # of active entries */
146 	struct undo {
147 		short	un_adjval;	/* adjust on exit values */
148 		short	un_num;		/* semaphore # */
149 		int	un_id;		/* semid */
150 		unsigned short un_seq;
151 	} un_ent[1];			/* undo entries */
152 };
153 
154 /*
155  * Configuration parameters
156  */
157 #ifndef SEMMNI
158 #define SEMMNI	50		/* # of semaphore identifiers */
159 #endif
160 #ifndef SEMMNS
161 #define SEMMNS	340		/* # of semaphores in system */
162 #endif
163 #ifndef SEMUME
164 #define SEMUME	50		/* max # of undo entries per process */
165 #endif
166 #ifndef SEMMNU
167 #define SEMMNU	150		/* # of undo structures in system */
168 #endif
169 
170 /* shouldn't need tuning */
171 #ifndef SEMMSL
172 #define SEMMSL	SEMMNS		/* max # of semaphores per id */
173 #endif
174 #ifndef SEMOPM
175 #define SEMOPM	100		/* max # of operations per semop call */
176 #endif
177 
178 #define SEMVMX	32767		/* semaphore maximum value */
179 #define SEMAEM	16384		/* adjust on exit max value */
180 
181 /*
182  * Due to the way semaphore memory is allocated, we have to ensure that
183  * SEMUSZ is properly aligned.
184  */
185 
186 #define	SEM_ALIGN(bytes) roundup2(bytes, sizeof(long))
187 
188 /* actual size of an undo structure */
189 #define SEMUSZ	SEM_ALIGN(offsetof(struct sem_undo, un_ent[SEMUME]))
190 
191 /*
192  * Macro to find a particular sem_undo vector
193  */
194 #define SEMU(ix) \
195 	((struct sem_undo *)(((intptr_t)semu)+ix * seminfo.semusz))
196 
197 /*
198  * semaphore info struct
199  */
200 struct seminfo seminfo = {
201                 SEMMNI,         /* # of semaphore identifiers */
202                 SEMMNS,         /* # of semaphores in system */
203                 SEMMNU,         /* # of undo structures in system */
204                 SEMMSL,         /* max # of semaphores per id */
205                 SEMOPM,         /* max # of operations per semop call */
206                 SEMUME,         /* max # of undo entries per process */
207                 SEMUSZ,         /* size in bytes of undo structure */
208                 SEMVMX,         /* semaphore maximum value */
209                 SEMAEM          /* adjust on exit max value */
210 };
211 
212 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RDTUN, &seminfo.semmni, 0,
213     "Number of semaphore identifiers");
214 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RDTUN, &seminfo.semmns, 0,
215     "Maximum number of semaphores in the system");
216 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RDTUN, &seminfo.semmnu, 0,
217     "Maximum number of undo structures in the system");
218 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RWTUN, &seminfo.semmsl, 0,
219     "Max semaphores per id");
220 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RDTUN, &seminfo.semopm, 0,
221     "Max operations per semop call");
222 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RDTUN, &seminfo.semume, 0,
223     "Max undo entries per process");
224 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RDTUN, &seminfo.semusz, 0,
225     "Size in bytes of undo structure");
226 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RWTUN, &seminfo.semvmx, 0,
227     "Semaphore maximum value");
228 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RWTUN, &seminfo.semaem, 0,
229     "Adjust on exit max value");
230 SYSCTL_PROC(_kern_ipc, OID_AUTO, sema,
231     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
232     NULL, 0, sysctl_sema, "",
233     "Array of struct semid_kernel for each potential semaphore");
234 
235 static struct syscall_helper_data sem_syscalls[] = {
236 	SYSCALL_INIT_HELPER(__semctl),
237 	SYSCALL_INIT_HELPER(semget),
238 	SYSCALL_INIT_HELPER(semop),
239 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
240     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
241 	SYSCALL_INIT_HELPER(semsys),
242 	SYSCALL_INIT_HELPER_COMPAT(freebsd7___semctl),
243 #endif
244 	SYSCALL_INIT_LAST
245 };
246 
247 #ifdef COMPAT_FREEBSD32
248 #include <compat/freebsd32/freebsd32.h>
249 #include <compat/freebsd32/freebsd32_ipc.h>
250 #include <compat/freebsd32/freebsd32_proto.h>
251 #include <compat/freebsd32/freebsd32_signal.h>
252 #include <compat/freebsd32/freebsd32_syscall.h>
253 #include <compat/freebsd32/freebsd32_util.h>
254 
255 static struct syscall_helper_data sem32_syscalls[] = {
256 	SYSCALL32_INIT_HELPER(freebsd32_semctl),
257 	SYSCALL32_INIT_HELPER_COMPAT(semget),
258 	SYSCALL32_INIT_HELPER_COMPAT(semop),
259 	SYSCALL32_INIT_HELPER(freebsd32_semsys),
260 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
261     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
262 	SYSCALL32_INIT_HELPER(freebsd7_freebsd32_semctl),
263 #endif
264 	SYSCALL_INIT_LAST
265 };
266 #endif
267 
268 static int
269 seminit(void)
270 {
271 	struct prison *pr;
272 	void **rsv;
273 	int i, error;
274 	osd_method_t methods[PR_MAXMETHOD] = {
275 	    [PR_METHOD_CHECK] =		sem_prison_check,
276 	    [PR_METHOD_SET] =		sem_prison_set,
277 	    [PR_METHOD_GET] =		sem_prison_get,
278 	    [PR_METHOD_REMOVE] =	sem_prison_remove,
279 	};
280 
281 	sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
282 	sema = malloc(sizeof(struct semid_kernel) * seminfo.semmni, M_SEM,
283 	    M_WAITOK | M_ZERO);
284 	sema_mtx = malloc(sizeof(struct mtx) * seminfo.semmni, M_SEM,
285 	    M_WAITOK | M_ZERO);
286 	semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
287 
288 	for (i = 0; i < seminfo.semmni; i++) {
289 		sema[i].u.__sem_base = 0;
290 		sema[i].u.sem_perm.mode = 0;
291 		sema[i].u.sem_perm.seq = 0;
292 #ifdef MAC
293 		mac_sysvsem_init(&sema[i]);
294 #endif
295 	}
296 	for (i = 0; i < seminfo.semmni; i++)
297 		mtx_init(&sema_mtx[i], "semid", NULL, MTX_DEF);
298 	LIST_INIT(&semu_free_list);
299 	for (i = 0; i < seminfo.semmnu; i++) {
300 		struct sem_undo *suptr = SEMU(i);
301 		suptr->un_proc = NULL;
302 		LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
303 	}
304 	LIST_INIT(&semu_list);
305 	mtx_init(&sem_mtx, "sem", NULL, MTX_DEF);
306 	mtx_init(&sem_undo_mtx, "semu", NULL, MTX_DEF);
307 	semexit_tag = EVENTHANDLER_REGISTER(process_exit, semexit_myhook, NULL,
308 	    EVENTHANDLER_PRI_ANY);
309 
310 	/* Set current prisons according to their allow.sysvipc. */
311 	sem_prison_slot = osd_jail_register(NULL, methods);
312 	rsv = osd_reserve(sem_prison_slot);
313 	prison_lock(&prison0);
314 	(void)osd_jail_set_reserved(&prison0, sem_prison_slot, rsv, &prison0);
315 	prison_unlock(&prison0);
316 	rsv = NULL;
317 	sx_slock(&allprison_lock);
318 	TAILQ_FOREACH(pr, &allprison, pr_list) {
319 		if (rsv == NULL)
320 			rsv = osd_reserve(sem_prison_slot);
321 		prison_lock(pr);
322 		if ((pr->pr_allow & PR_ALLOW_SYSVIPC) && pr->pr_ref > 0) {
323 			(void)osd_jail_set_reserved(pr, sem_prison_slot, rsv,
324 			    &prison0);
325 			rsv = NULL;
326 		}
327 		prison_unlock(pr);
328 	}
329 	if (rsv != NULL)
330 		osd_free_reserved(rsv);
331 	sx_sunlock(&allprison_lock);
332 
333 	error = syscall_helper_register(sem_syscalls, SY_THR_STATIC_KLD);
334 	if (error != 0)
335 		return (error);
336 #ifdef COMPAT_FREEBSD32
337 	error = syscall32_helper_register(sem32_syscalls, SY_THR_STATIC_KLD);
338 	if (error != 0)
339 		return (error);
340 #endif
341 	return (0);
342 }
343 
344 static int
345 semunload(void)
346 {
347 	int i;
348 
349 	/* XXXKIB */
350 	if (semtot != 0)
351 		return (EBUSY);
352 
353 #ifdef COMPAT_FREEBSD32
354 	syscall32_helper_unregister(sem32_syscalls);
355 #endif
356 	syscall_helper_unregister(sem_syscalls);
357 	EVENTHANDLER_DEREGISTER(process_exit, semexit_tag);
358 	if (sem_prison_slot != 0)
359 		osd_jail_deregister(sem_prison_slot);
360 #ifdef MAC
361 	for (i = 0; i < seminfo.semmni; i++)
362 		mac_sysvsem_destroy(&sema[i]);
363 #endif
364 	free(sem, M_SEM);
365 	free(sema, M_SEM);
366 	free(semu, M_SEM);
367 	for (i = 0; i < seminfo.semmni; i++)
368 		mtx_destroy(&sema_mtx[i]);
369 	free(sema_mtx, M_SEM);
370 	mtx_destroy(&sem_mtx);
371 	mtx_destroy(&sem_undo_mtx);
372 	return (0);
373 }
374 
375 static int
376 sysvsem_modload(struct module *module, int cmd, void *arg)
377 {
378 	int error = 0;
379 
380 	switch (cmd) {
381 	case MOD_LOAD:
382 		error = seminit();
383 		if (error != 0)
384 			semunload();
385 		break;
386 	case MOD_UNLOAD:
387 		error = semunload();
388 		break;
389 	case MOD_SHUTDOWN:
390 		break;
391 	default:
392 		error = EINVAL;
393 		break;
394 	}
395 	return (error);
396 }
397 
398 static moduledata_t sysvsem_mod = {
399 	"sysvsem",
400 	&sysvsem_modload,
401 	NULL
402 };
403 
404 DECLARE_MODULE(sysvsem, sysvsem_mod, SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
405 MODULE_VERSION(sysvsem, 1);
406 
407 /*
408  * Allocate a new sem_undo structure for a process
409  * (returns ptr to structure or NULL if no more room)
410  */
411 
412 static struct sem_undo *
413 semu_alloc(struct thread *td)
414 {
415 	struct sem_undo *suptr;
416 
417 	SEMUNDO_LOCKASSERT(MA_OWNED);
418 	if ((suptr = LIST_FIRST(&semu_free_list)) == NULL)
419 		return (NULL);
420 	LIST_REMOVE(suptr, un_next);
421 	LIST_INSERT_HEAD(&semu_list, suptr, un_next);
422 	suptr->un_cnt = 0;
423 	suptr->un_proc = td->td_proc;
424 	return (suptr);
425 }
426 
427 static int
428 semu_try_free(struct sem_undo *suptr)
429 {
430 
431 	SEMUNDO_LOCKASSERT(MA_OWNED);
432 
433 	if (suptr->un_cnt != 0)
434 		return (0);
435 	LIST_REMOVE(suptr, un_next);
436 	LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
437 	return (1);
438 }
439 
440 /*
441  * Adjust a particular entry for a particular proc
442  */
443 
444 static int
445 semundo_adjust(struct thread *td, struct sem_undo **supptr, int semid,
446     int semseq, int semnum, int adjval)
447 {
448 	struct proc *p = td->td_proc;
449 	struct sem_undo *suptr;
450 	struct undo *sunptr;
451 	int i;
452 
453 	SEMUNDO_LOCKASSERT(MA_OWNED);
454 	/* Look for and remember the sem_undo if the caller doesn't provide
455 	   it */
456 
457 	suptr = *supptr;
458 	if (suptr == NULL) {
459 		LIST_FOREACH(suptr, &semu_list, un_next) {
460 			if (suptr->un_proc == p) {
461 				*supptr = suptr;
462 				break;
463 			}
464 		}
465 		if (suptr == NULL) {
466 			if (adjval == 0)
467 				return(0);
468 			suptr = semu_alloc(td);
469 			if (suptr == NULL)
470 				return (ENOSPC);
471 			*supptr = suptr;
472 		}
473 	}
474 
475 	/*
476 	 * Look for the requested entry and adjust it (delete if adjval becomes
477 	 * 0).
478 	 */
479 	sunptr = &suptr->un_ent[0];
480 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
481 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
482 			continue;
483 		if (adjval != 0) {
484 			adjval += sunptr->un_adjval;
485 			if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
486 				return (ERANGE);
487 		}
488 		sunptr->un_adjval = adjval;
489 		if (sunptr->un_adjval == 0) {
490 			suptr->un_cnt--;
491 			if (i < suptr->un_cnt)
492 				suptr->un_ent[i] =
493 				    suptr->un_ent[suptr->un_cnt];
494 			if (suptr->un_cnt == 0)
495 				semu_try_free(suptr);
496 		}
497 		return (0);
498 	}
499 
500 	/* Didn't find the right entry - create it */
501 	if (adjval == 0)
502 		return (0);
503 	if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
504 		return (ERANGE);
505 	if (suptr->un_cnt != seminfo.semume) {
506 		sunptr = &suptr->un_ent[suptr->un_cnt];
507 		suptr->un_cnt++;
508 		sunptr->un_adjval = adjval;
509 		sunptr->un_id = semid;
510 		sunptr->un_num = semnum;
511 		sunptr->un_seq = semseq;
512 	} else
513 		return (EINVAL);
514 	return (0);
515 }
516 
517 static void
518 semundo_clear(int semid, int semnum)
519 {
520 	struct sem_undo *suptr, *suptr1;
521 	struct undo *sunptr;
522 	int i;
523 
524 	SEMUNDO_LOCKASSERT(MA_OWNED);
525 	LIST_FOREACH_SAFE(suptr, &semu_list, un_next, suptr1) {
526 		sunptr = &suptr->un_ent[0];
527 		for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
528 			if (sunptr->un_id != semid)
529 				continue;
530 			if (semnum == -1 || sunptr->un_num == semnum) {
531 				suptr->un_cnt--;
532 				if (i < suptr->un_cnt) {
533 					suptr->un_ent[i] =
534 					    suptr->un_ent[suptr->un_cnt];
535 					continue;
536 				}
537 				semu_try_free(suptr);
538 			}
539 			if (semnum != -1)
540 				break;
541 		}
542 	}
543 }
544 
545 static int
546 semvalid(int semid, struct prison *rpr, struct semid_kernel *semakptr)
547 {
548 
549 	return ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
550 	    semakptr->u.sem_perm.seq != IPCID_TO_SEQ(semid) ||
551 	    sem_prison_cansee(rpr, semakptr) ? EINVAL : 0);
552 }
553 
554 static void
555 sem_remove(int semidx, struct ucred *cred)
556 {
557 	struct semid_kernel *semakptr;
558 	int i;
559 
560 	KASSERT(semidx >= 0 && semidx < seminfo.semmni,
561 		("semidx out of bounds"));
562 	semakptr = &sema[semidx];
563 	semakptr->u.sem_perm.cuid = cred ? cred->cr_uid : 0;
564 	semakptr->u.sem_perm.uid = cred ? cred->cr_uid : 0;
565 	semakptr->u.sem_perm.mode = 0;
566 	racct_sub_cred(semakptr->cred, RACCT_NSEM, semakptr->u.sem_nsems);
567 	crfree(semakptr->cred);
568 	semakptr->cred = NULL;
569 	SEMUNDO_LOCK();
570 	semundo_clear(semidx, -1);
571 	SEMUNDO_UNLOCK();
572 #ifdef MAC
573 	mac_sysvsem_cleanup(semakptr);
574 #endif
575 	wakeup(semakptr);
576 	for (i = 0; i < seminfo.semmni; i++) {
577 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
578 		    sema[i].u.__sem_base > semakptr->u.__sem_base)
579 			mtx_lock_flags(&sema_mtx[i], LOP_DUPOK);
580 	}
581 	for (i = semakptr->u.__sem_base - sem; i < semtot; i++)
582 		sem[i] = sem[i + semakptr->u.sem_nsems];
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 			sema[i].u.__sem_base -= semakptr->u.sem_nsems;
587 			mtx_unlock(&sema_mtx[i]);
588 		}
589 	}
590 	semtot -= semakptr->u.sem_nsems;
591 }
592 
593 static struct prison *
594 sem_find_prison(struct ucred *cred)
595 {
596 	struct prison *pr, *rpr;
597 
598 	pr = cred->cr_prison;
599 	prison_lock(pr);
600 	rpr = osd_jail_get(pr, sem_prison_slot);
601 	prison_unlock(pr);
602 	return rpr;
603 }
604 
605 static int
606 sem_prison_cansee(struct prison *rpr, struct semid_kernel *semakptr)
607 {
608 
609 	if (semakptr->cred == NULL ||
610 	    !(rpr == semakptr->cred->cr_prison ||
611 	      prison_ischild(rpr, semakptr->cred->cr_prison)))
612 		return (EINVAL);
613 	return (0);
614 }
615 
616 /*
617  * Note that the user-mode half of this passes a union, not a pointer.
618  */
619 #ifndef _SYS_SYSPROTO_H_
620 struct __semctl_args {
621 	int	semid;
622 	int	semnum;
623 	int	cmd;
624 	union	semun *arg;
625 };
626 #endif
627 int
628 sys___semctl(struct thread *td, struct __semctl_args *uap)
629 {
630 	struct semid_ds dsbuf;
631 	union semun arg, semun;
632 	register_t rval;
633 	int error;
634 
635 	switch (uap->cmd) {
636 	case SEM_STAT:
637 	case IPC_SET:
638 	case IPC_STAT:
639 	case GETALL:
640 	case SETVAL:
641 	case SETALL:
642 		error = copyin(uap->arg, &arg, sizeof(arg));
643 		if (error)
644 			return (error);
645 		break;
646 	}
647 
648 	switch (uap->cmd) {
649 	case SEM_STAT:
650 	case IPC_STAT:
651 		semun.buf = &dsbuf;
652 		break;
653 	case IPC_SET:
654 		error = copyin(arg.buf, &dsbuf, sizeof(dsbuf));
655 		if (error)
656 			return (error);
657 		semun.buf = &dsbuf;
658 		break;
659 	case GETALL:
660 	case SETALL:
661 		semun.array = arg.array;
662 		break;
663 	case SETVAL:
664 		semun.val = arg.val;
665 		break;
666 	}
667 
668 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
669 	    &rval);
670 	if (error)
671 		return (error);
672 
673 	switch (uap->cmd) {
674 	case SEM_STAT:
675 	case IPC_STAT:
676 		error = copyout(&dsbuf, arg.buf, sizeof(dsbuf));
677 		break;
678 	}
679 
680 	if (error == 0)
681 		td->td_retval[0] = rval;
682 	return (error);
683 }
684 
685 int
686 kern_semctl(struct thread *td, int semid, int semnum, int cmd,
687     union semun *arg, register_t *rval)
688 {
689 	u_short *array;
690 	struct ucred *cred = td->td_ucred;
691 	int i, error;
692 	struct prison *rpr;
693 	struct semid_ds *sbuf;
694 	struct semid_kernel *semakptr;
695 	struct mtx *sema_mtxp;
696 	u_short usval, count;
697 	int semidx;
698 
699 	DPRINTF(("call to semctl(%d, %d, %d, 0x%p)\n",
700 	    semid, semnum, cmd, arg));
701 
702 	AUDIT_ARG_SVIPC_CMD(cmd);
703 	AUDIT_ARG_SVIPC_ID(semid);
704 
705 	rpr = sem_find_prison(td->td_ucred);
706 	if (sem == NULL)
707 		return (ENOSYS);
708 
709 	array = NULL;
710 
711 	switch(cmd) {
712 	case SEM_STAT:
713 		/*
714 		 * For this command we assume semid is an array index
715 		 * rather than an IPC id.
716 		 */
717 		if (semid < 0 || semid >= seminfo.semmni)
718 			return (EINVAL);
719 		semakptr = &sema[semid];
720 		sema_mtxp = &sema_mtx[semid];
721 		mtx_lock(sema_mtxp);
722 		if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
723 			error = EINVAL;
724 			goto done2;
725 		}
726 		if ((error = sem_prison_cansee(rpr, semakptr)))
727 			goto done2;
728 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
729 			goto done2;
730 #ifdef MAC
731 		error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
732 		if (error != 0)
733 			goto done2;
734 #endif
735 		bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
736 		if (cred->cr_prison != semakptr->cred->cr_prison)
737 			arg->buf->sem_perm.key = IPC_PRIVATE;
738 		*rval = IXSEQ_TO_IPCID(semid, semakptr->u.sem_perm);
739 		mtx_unlock(sema_mtxp);
740 		return (0);
741 	}
742 
743 	semidx = IPCID_TO_IX(semid);
744 	if (semidx < 0 || semidx >= seminfo.semmni)
745 		return (EINVAL);
746 
747 	semakptr = &sema[semidx];
748 	sema_mtxp = &sema_mtx[semidx];
749 	if (cmd == IPC_RMID)
750 		mtx_lock(&sem_mtx);
751 	mtx_lock(sema_mtxp);
752 
753 #ifdef MAC
754 	error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
755 	if (error != 0)
756 		goto done2;
757 #endif
758 
759 	error = 0;
760 	*rval = 0;
761 
762 	switch (cmd) {
763 	case IPC_RMID:
764 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
765 			goto done2;
766 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
767 			goto done2;
768 		sem_remove(semidx, cred);
769 		break;
770 
771 	case IPC_SET:
772 		AUDIT_ARG_SVIPC_PERM(&arg->buf->sem_perm);
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 		sbuf = arg->buf;
778 		semakptr->u.sem_perm.uid = sbuf->sem_perm.uid;
779 		semakptr->u.sem_perm.gid = sbuf->sem_perm.gid;
780 		semakptr->u.sem_perm.mode = (semakptr->u.sem_perm.mode &
781 		    ~0777) | (sbuf->sem_perm.mode & 0777);
782 		semakptr->u.sem_ctime = time_second;
783 		break;
784 
785 	case IPC_STAT:
786 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
787 			goto done2;
788 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
789 			goto done2;
790 		bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
791 		if (cred->cr_prison != semakptr->cred->cr_prison)
792 			arg->buf->sem_perm.key = IPC_PRIVATE;
793 		break;
794 
795 	case GETNCNT:
796 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
797 			goto done2;
798 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
799 			goto done2;
800 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
801 			error = EINVAL;
802 			goto done2;
803 		}
804 		*rval = semakptr->u.__sem_base[semnum].semncnt;
805 		break;
806 
807 	case GETPID:
808 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
809 			goto done2;
810 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
811 			goto done2;
812 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
813 			error = EINVAL;
814 			goto done2;
815 		}
816 		*rval = semakptr->u.__sem_base[semnum].sempid;
817 		break;
818 
819 	case GETVAL:
820 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
821 			goto done2;
822 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
823 			goto done2;
824 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
825 			error = EINVAL;
826 			goto done2;
827 		}
828 		*rval = semakptr->u.__sem_base[semnum].semval;
829 		break;
830 
831 	case GETALL:
832 		/*
833 		 * Unfortunately, callers of this function don't know
834 		 * in advance how many semaphores are in this set.
835 		 * While we could just allocate the maximum size array
836 		 * and pass the actual size back to the caller, that
837 		 * won't work for SETALL since we can't copyin() more
838 		 * data than the user specified as we may return a
839 		 * spurious EFAULT.
840 		 *
841 		 * Note that the number of semaphores in a set is
842 		 * fixed for the life of that set.  The only way that
843 		 * the 'count' could change while are blocked in
844 		 * malloc() is if this semaphore set were destroyed
845 		 * and a new one created with the same index.
846 		 * However, semvalid() will catch that due to the
847 		 * sequence number unless exactly 0x8000 (or a
848 		 * multiple thereof) semaphore sets for the same index
849 		 * are created and destroyed while we are in malloc!
850 		 *
851 		 */
852 		count = semakptr->u.sem_nsems;
853 		mtx_unlock(sema_mtxp);
854 		array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
855 		mtx_lock(sema_mtxp);
856 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
857 			goto done2;
858 		KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
859 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
860 			goto done2;
861 		for (i = 0; i < semakptr->u.sem_nsems; i++)
862 			array[i] = semakptr->u.__sem_base[i].semval;
863 		mtx_unlock(sema_mtxp);
864 		error = copyout(array, arg->array, count * sizeof(*array));
865 		mtx_lock(sema_mtxp);
866 		break;
867 
868 	case GETZCNT:
869 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
870 			goto done2;
871 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
872 			goto done2;
873 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
874 			error = EINVAL;
875 			goto done2;
876 		}
877 		*rval = semakptr->u.__sem_base[semnum].semzcnt;
878 		break;
879 
880 	case SETVAL:
881 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
882 			goto done2;
883 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
884 			goto done2;
885 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
886 			error = EINVAL;
887 			goto done2;
888 		}
889 		if (arg->val < 0 || arg->val > seminfo.semvmx) {
890 			error = ERANGE;
891 			goto done2;
892 		}
893 		semakptr->u.__sem_base[semnum].semval = arg->val;
894 		SEMUNDO_LOCK();
895 		semundo_clear(semidx, semnum);
896 		SEMUNDO_UNLOCK();
897 		wakeup(semakptr);
898 		break;
899 
900 	case SETALL:
901 		/*
902 		 * See comment on GETALL for why 'count' shouldn't change
903 		 * and why we require a userland buffer.
904 		 */
905 		count = semakptr->u.sem_nsems;
906 		mtx_unlock(sema_mtxp);
907 		array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
908 		error = copyin(arg->array, array, count * sizeof(*array));
909 		mtx_lock(sema_mtxp);
910 		if (error)
911 			break;
912 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
913 			goto done2;
914 		KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
915 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
916 			goto done2;
917 		for (i = 0; i < semakptr->u.sem_nsems; i++) {
918 			usval = array[i];
919 			if (usval > seminfo.semvmx) {
920 				error = ERANGE;
921 				break;
922 			}
923 			semakptr->u.__sem_base[i].semval = usval;
924 		}
925 		SEMUNDO_LOCK();
926 		semundo_clear(semidx, -1);
927 		SEMUNDO_UNLOCK();
928 		wakeup(semakptr);
929 		break;
930 
931 	default:
932 		error = EINVAL;
933 		break;
934 	}
935 
936 done2:
937 	mtx_unlock(sema_mtxp);
938 	if (cmd == IPC_RMID)
939 		mtx_unlock(&sem_mtx);
940 	if (array != NULL)
941 		free(array, M_TEMP);
942 	return(error);
943 }
944 
945 #ifndef _SYS_SYSPROTO_H_
946 struct semget_args {
947 	key_t	key;
948 	int	nsems;
949 	int	semflg;
950 };
951 #endif
952 int
953 sys_semget(struct thread *td, struct semget_args *uap)
954 {
955 	int semid, error = 0;
956 	int key = uap->key;
957 	int nsems = uap->nsems;
958 	int semflg = uap->semflg;
959 	struct ucred *cred = td->td_ucred;
960 
961 	DPRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg));
962 
963 	AUDIT_ARG_VALUE(semflg);
964 
965 	if (sem_find_prison(cred) == NULL)
966 		return (ENOSYS);
967 
968 	mtx_lock(&sem_mtx);
969 	if (key != IPC_PRIVATE) {
970 		for (semid = 0; semid < seminfo.semmni; semid++) {
971 			if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) &&
972 			    sema[semid].cred != NULL &&
973 			    sema[semid].cred->cr_prison == cred->cr_prison &&
974 			    sema[semid].u.sem_perm.key == key)
975 				break;
976 		}
977 		if (semid < seminfo.semmni) {
978 			AUDIT_ARG_SVIPC_ID(semid);
979 			DPRINTF(("found public key\n"));
980 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
981 				DPRINTF(("not exclusive\n"));
982 				error = EEXIST;
983 				goto done2;
984 			}
985 			if ((error = ipcperm(td, &sema[semid].u.sem_perm,
986 			    semflg & 0700))) {
987 				goto done2;
988 			}
989 			if (nsems > 0 && sema[semid].u.sem_nsems < nsems) {
990 				DPRINTF(("too small\n"));
991 				error = EINVAL;
992 				goto done2;
993 			}
994 #ifdef MAC
995 			error = mac_sysvsem_check_semget(cred, &sema[semid]);
996 			if (error != 0)
997 				goto done2;
998 #endif
999 			goto found;
1000 		}
1001 	}
1002 
1003 	DPRINTF(("need to allocate the semid_kernel\n"));
1004 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
1005 		if (nsems <= 0 || nsems > seminfo.semmsl) {
1006 			DPRINTF(("nsems out of range (0<%d<=%d)\n", nsems,
1007 			    seminfo.semmsl));
1008 			error = EINVAL;
1009 			goto done2;
1010 		}
1011 		if (nsems > seminfo.semmns - semtot) {
1012 			DPRINTF((
1013 			    "not enough semaphores left (need %d, got %d)\n",
1014 			    nsems, seminfo.semmns - semtot));
1015 			error = ENOSPC;
1016 			goto done2;
1017 		}
1018 		for (semid = 0; semid < seminfo.semmni; semid++) {
1019 			if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0)
1020 				break;
1021 		}
1022 		if (semid == seminfo.semmni) {
1023 			DPRINTF(("no more semid_kernel's available\n"));
1024 			error = ENOSPC;
1025 			goto done2;
1026 		}
1027 #ifdef RACCT
1028 		if (racct_enable) {
1029 			PROC_LOCK(td->td_proc);
1030 			error = racct_add(td->td_proc, RACCT_NSEM, nsems);
1031 			PROC_UNLOCK(td->td_proc);
1032 			if (error != 0) {
1033 				error = ENOSPC;
1034 				goto done2;
1035 			}
1036 		}
1037 #endif
1038 		DPRINTF(("semid %d is available\n", semid));
1039 		mtx_lock(&sema_mtx[semid]);
1040 		KASSERT((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0,
1041 		    ("Lost semaphore %d", semid));
1042 		sema[semid].u.sem_perm.key = key;
1043 		sema[semid].u.sem_perm.cuid = cred->cr_uid;
1044 		sema[semid].u.sem_perm.uid = cred->cr_uid;
1045 		sema[semid].u.sem_perm.cgid = cred->cr_gid;
1046 		sema[semid].u.sem_perm.gid = cred->cr_gid;
1047 		sema[semid].u.sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
1048 		sema[semid].cred = crhold(cred);
1049 		sema[semid].u.sem_perm.seq =
1050 		    (sema[semid].u.sem_perm.seq + 1) & 0x7fff;
1051 		sema[semid].u.sem_nsems = nsems;
1052 		sema[semid].u.sem_otime = 0;
1053 		sema[semid].u.sem_ctime = time_second;
1054 		sema[semid].u.__sem_base = &sem[semtot];
1055 		semtot += nsems;
1056 		bzero(sema[semid].u.__sem_base,
1057 		    sizeof(sema[semid].u.__sem_base[0])*nsems);
1058 #ifdef MAC
1059 		mac_sysvsem_create(cred, &sema[semid]);
1060 #endif
1061 		mtx_unlock(&sema_mtx[semid]);
1062 		DPRINTF(("sembase = %p, next = %p\n",
1063 		    sema[semid].u.__sem_base, &sem[semtot]));
1064 	} else {
1065 		DPRINTF(("didn't find it and wasn't asked to create it\n"));
1066 		error = ENOENT;
1067 		goto done2;
1068 	}
1069 
1070 found:
1071 	td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].u.sem_perm);
1072 done2:
1073 	mtx_unlock(&sem_mtx);
1074 	return (error);
1075 }
1076 
1077 #ifndef _SYS_SYSPROTO_H_
1078 struct semop_args {
1079 	int	semid;
1080 	struct	sembuf *sops;
1081 	size_t	nsops;
1082 };
1083 #endif
1084 int
1085 sys_semop(struct thread *td, struct semop_args *uap)
1086 {
1087 #define SMALL_SOPS	8
1088 	struct sembuf small_sops[SMALL_SOPS];
1089 	int semid = uap->semid;
1090 	size_t nsops = uap->nsops;
1091 	struct prison *rpr;
1092 	struct sembuf *sops;
1093 	struct semid_kernel *semakptr;
1094 	struct sembuf *sopptr = NULL;
1095 	struct sem *semptr = NULL;
1096 	struct sem_undo *suptr;
1097 	struct mtx *sema_mtxp;
1098 	size_t i, j, k;
1099 	int error;
1100 	int do_wakeup, do_undos;
1101 	unsigned short seq;
1102 
1103 #ifdef SEM_DEBUG
1104 	sops = NULL;
1105 #endif
1106 	DPRINTF(("call to semop(%d, %p, %u)\n", semid, sops, nsops));
1107 
1108 	AUDIT_ARG_SVIPC_ID(semid);
1109 
1110 	rpr = sem_find_prison(td->td_ucred);
1111 	if (sem == NULL)
1112 		return (ENOSYS);
1113 
1114 	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
1115 
1116 	if (semid < 0 || semid >= seminfo.semmni)
1117 		return (EINVAL);
1118 
1119 	/* Allocate memory for sem_ops */
1120 	if (nsops <= SMALL_SOPS)
1121 		sops = small_sops;
1122 	else if (nsops > seminfo.semopm) {
1123 		DPRINTF(("too many sops (max=%d, nsops=%d)\n", seminfo.semopm,
1124 		    nsops));
1125 		return (E2BIG);
1126 	} else {
1127 #ifdef RACCT
1128 		if (racct_enable) {
1129 			PROC_LOCK(td->td_proc);
1130 			if (nsops >
1131 			    racct_get_available(td->td_proc, RACCT_NSEMOP)) {
1132 				PROC_UNLOCK(td->td_proc);
1133 				return (E2BIG);
1134 			}
1135 			PROC_UNLOCK(td->td_proc);
1136 		}
1137 #endif
1138 
1139 		sops = malloc(nsops * sizeof(*sops), M_TEMP, M_WAITOK);
1140 	}
1141 	if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) {
1142 		DPRINTF(("error = %d from copyin(%p, %p, %d)\n", error,
1143 		    uap->sops, sops, nsops * sizeof(sops[0])));
1144 		if (sops != small_sops)
1145 			free(sops, M_TEMP);
1146 		return (error);
1147 	}
1148 
1149 	semakptr = &sema[semid];
1150 	sema_mtxp = &sema_mtx[semid];
1151 	mtx_lock(sema_mtxp);
1152 	if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
1153 		error = EINVAL;
1154 		goto done2;
1155 	}
1156 	seq = semakptr->u.sem_perm.seq;
1157 	if (seq != IPCID_TO_SEQ(uap->semid)) {
1158 		error = EINVAL;
1159 		goto done2;
1160 	}
1161 	if ((error = sem_prison_cansee(rpr, semakptr)) != 0)
1162 		goto done2;
1163 	/*
1164 	 * Initial pass through sops to see what permissions are needed.
1165 	 * Also perform any checks that don't need repeating on each
1166 	 * attempt to satisfy the request vector.
1167 	 */
1168 	j = 0;		/* permission needed */
1169 	do_undos = 0;
1170 	for (i = 0; i < nsops; i++) {
1171 		sopptr = &sops[i];
1172 		if (sopptr->sem_num >= semakptr->u.sem_nsems) {
1173 			error = EFBIG;
1174 			goto done2;
1175 		}
1176 		if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0)
1177 			do_undos = 1;
1178 		j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A;
1179 	}
1180 
1181 	if ((error = ipcperm(td, &semakptr->u.sem_perm, j))) {
1182 		DPRINTF(("error = %d from ipaccess\n", error));
1183 		goto done2;
1184 	}
1185 #ifdef MAC
1186 	error = mac_sysvsem_check_semop(td->td_ucred, semakptr, j);
1187 	if (error != 0)
1188 		goto done2;
1189 #endif
1190 
1191 	/*
1192 	 * Loop trying to satisfy the vector of requests.
1193 	 * If we reach a point where we must wait, any requests already
1194 	 * performed are rolled back and we go to sleep until some other
1195 	 * process wakes us up.  At this point, we start all over again.
1196 	 *
1197 	 * This ensures that from the perspective of other tasks, a set
1198 	 * of requests is atomic (never partially satisfied).
1199 	 */
1200 	for (;;) {
1201 		do_wakeup = 0;
1202 		error = 0;	/* error return if necessary */
1203 
1204 		for (i = 0; i < nsops; i++) {
1205 			sopptr = &sops[i];
1206 			semptr = &semakptr->u.__sem_base[sopptr->sem_num];
1207 
1208 			DPRINTF((
1209 			    "semop:  semakptr=%p, __sem_base=%p, "
1210 			    "semptr=%p, sem[%d]=%d : op=%d, flag=%s\n",
1211 			    semakptr, semakptr->u.__sem_base, semptr,
1212 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
1213 			    (sopptr->sem_flg & IPC_NOWAIT) ?
1214 			    "nowait" : "wait"));
1215 
1216 			if (sopptr->sem_op < 0) {
1217 				if (semptr->semval + sopptr->sem_op < 0) {
1218 					DPRINTF(("semop:  can't do it now\n"));
1219 					break;
1220 				} else {
1221 					semptr->semval += sopptr->sem_op;
1222 					if (semptr->semval == 0 &&
1223 					    semptr->semzcnt > 0)
1224 						do_wakeup = 1;
1225 				}
1226 			} else if (sopptr->sem_op == 0) {
1227 				if (semptr->semval != 0) {
1228 					DPRINTF(("semop:  not zero now\n"));
1229 					break;
1230 				}
1231 			} else if (semptr->semval + sopptr->sem_op >
1232 			    seminfo.semvmx) {
1233 				error = ERANGE;
1234 				break;
1235 			} else {
1236 				if (semptr->semncnt > 0)
1237 					do_wakeup = 1;
1238 				semptr->semval += sopptr->sem_op;
1239 			}
1240 		}
1241 
1242 		/*
1243 		 * Did we get through the entire vector?
1244 		 */
1245 		if (i >= nsops)
1246 			goto done;
1247 
1248 		/*
1249 		 * No ... rollback anything that we've already done
1250 		 */
1251 		DPRINTF(("semop:  rollback 0 through %d\n", i-1));
1252 		for (j = 0; j < i; j++)
1253 			semakptr->u.__sem_base[sops[j].sem_num].semval -=
1254 			    sops[j].sem_op;
1255 
1256 		/* If we detected an error, return it */
1257 		if (error != 0)
1258 			goto done2;
1259 
1260 		/*
1261 		 * If the request that we couldn't satisfy has the
1262 		 * NOWAIT flag set then return with EAGAIN.
1263 		 */
1264 		if (sopptr->sem_flg & IPC_NOWAIT) {
1265 			error = EAGAIN;
1266 			goto done2;
1267 		}
1268 
1269 		if (sopptr->sem_op == 0)
1270 			semptr->semzcnt++;
1271 		else
1272 			semptr->semncnt++;
1273 
1274 		DPRINTF(("semop:  good night!\n"));
1275 		error = msleep(semakptr, sema_mtxp, (PZERO - 4) | PCATCH,
1276 		    "semwait", 0);
1277 		DPRINTF(("semop:  good morning (error=%d)!\n", error));
1278 		/* return code is checked below, after sem[nz]cnt-- */
1279 
1280 		/*
1281 		 * Make sure that the semaphore still exists
1282 		 */
1283 		seq = semakptr->u.sem_perm.seq;
1284 		if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1285 		    seq != IPCID_TO_SEQ(uap->semid)) {
1286 			error = EIDRM;
1287 			goto done2;
1288 		}
1289 
1290 		/*
1291 		 * Renew the semaphore's pointer after wakeup since
1292 		 * during msleep __sem_base may have been modified and semptr
1293 		 * is not valid any more
1294 		 */
1295 		semptr = &semakptr->u.__sem_base[sopptr->sem_num];
1296 
1297 		/*
1298 		 * The semaphore is still alive.  Readjust the count of
1299 		 * waiting processes.
1300 		 */
1301 		if (sopptr->sem_op == 0)
1302 			semptr->semzcnt--;
1303 		else
1304 			semptr->semncnt--;
1305 
1306 		/*
1307 		 * Is it really morning, or was our sleep interrupted?
1308 		 * (Delayed check of msleep() return code because we
1309 		 * need to decrement sem[nz]cnt either way.)
1310 		 */
1311 		if (error != 0) {
1312 			error = EINTR;
1313 			goto done2;
1314 		}
1315 		DPRINTF(("semop:  good morning!\n"));
1316 	}
1317 
1318 done:
1319 	/*
1320 	 * Process any SEM_UNDO requests.
1321 	 */
1322 	if (do_undos) {
1323 		SEMUNDO_LOCK();
1324 		suptr = NULL;
1325 		for (i = 0; i < nsops; i++) {
1326 			/*
1327 			 * We only need to deal with SEM_UNDO's for non-zero
1328 			 * op's.
1329 			 */
1330 			int adjval;
1331 
1332 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
1333 				continue;
1334 			adjval = sops[i].sem_op;
1335 			if (adjval == 0)
1336 				continue;
1337 			error = semundo_adjust(td, &suptr, semid, seq,
1338 			    sops[i].sem_num, -adjval);
1339 			if (error == 0)
1340 				continue;
1341 
1342 			/*
1343 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
1344 			 * Rollback the adjustments to this point and then
1345 			 * rollback the semaphore ups and down so we can return
1346 			 * with an error with all structures restored.  We
1347 			 * rollback the undo's in the exact reverse order that
1348 			 * we applied them.  This guarantees that we won't run
1349 			 * out of space as we roll things back out.
1350 			 */
1351 			for (j = 0; j < i; j++) {
1352 				k = i - j - 1;
1353 				if ((sops[k].sem_flg & SEM_UNDO) == 0)
1354 					continue;
1355 				adjval = sops[k].sem_op;
1356 				if (adjval == 0)
1357 					continue;
1358 				if (semundo_adjust(td, &suptr, semid, seq,
1359 				    sops[k].sem_num, adjval) != 0)
1360 					panic("semop - can't undo undos");
1361 			}
1362 
1363 			for (j = 0; j < nsops; j++)
1364 				semakptr->u.__sem_base[sops[j].sem_num].semval -=
1365 				    sops[j].sem_op;
1366 
1367 			DPRINTF(("error = %d from semundo_adjust\n", error));
1368 			SEMUNDO_UNLOCK();
1369 			goto done2;
1370 		} /* loop through the sops */
1371 		SEMUNDO_UNLOCK();
1372 	} /* if (do_undos) */
1373 
1374 	/* We're definitely done - set the sempid's and time */
1375 	for (i = 0; i < nsops; i++) {
1376 		sopptr = &sops[i];
1377 		semptr = &semakptr->u.__sem_base[sopptr->sem_num];
1378 		semptr->sempid = td->td_proc->p_pid;
1379 	}
1380 	semakptr->u.sem_otime = time_second;
1381 
1382 	/*
1383 	 * Do a wakeup if any semaphore was up'd whilst something was
1384 	 * sleeping on it.
1385 	 */
1386 	if (do_wakeup) {
1387 		DPRINTF(("semop:  doing wakeup\n"));
1388 		wakeup(semakptr);
1389 		DPRINTF(("semop:  back from wakeup\n"));
1390 	}
1391 	DPRINTF(("semop:  done\n"));
1392 	td->td_retval[0] = 0;
1393 done2:
1394 	mtx_unlock(sema_mtxp);
1395 	if (sops != small_sops)
1396 		free(sops, M_TEMP);
1397 	return (error);
1398 }
1399 
1400 /*
1401  * Go through the undo structures for this process and apply the adjustments to
1402  * semaphores.
1403  */
1404 static void
1405 semexit_myhook(void *arg, struct proc *p)
1406 {
1407 	struct sem_undo *suptr;
1408 	struct semid_kernel *semakptr;
1409 	struct mtx *sema_mtxp;
1410 	int semid, semnum, adjval, ix;
1411 	unsigned short seq;
1412 
1413 	/*
1414 	 * Go through the chain of undo vectors looking for one
1415 	 * associated with this process.
1416 	 */
1417 	if (LIST_EMPTY(&semu_list))
1418 		return;
1419 	SEMUNDO_LOCK();
1420 	LIST_FOREACH(suptr, &semu_list, un_next) {
1421 		if (suptr->un_proc == p)
1422 			break;
1423 	}
1424 	if (suptr == NULL) {
1425 		SEMUNDO_UNLOCK();
1426 		return;
1427 	}
1428 	LIST_REMOVE(suptr, un_next);
1429 
1430 	DPRINTF(("proc @%p has undo structure with %d entries\n", p,
1431 	    suptr->un_cnt));
1432 
1433 	/*
1434 	 * If there are any active undo elements then process them.
1435 	 */
1436 	if (suptr->un_cnt > 0) {
1437 		SEMUNDO_UNLOCK();
1438 		for (ix = 0; ix < suptr->un_cnt; ix++) {
1439 			semid = suptr->un_ent[ix].un_id;
1440 			semnum = suptr->un_ent[ix].un_num;
1441 			adjval = suptr->un_ent[ix].un_adjval;
1442 			seq = suptr->un_ent[ix].un_seq;
1443 			semakptr = &sema[semid];
1444 			sema_mtxp = &sema_mtx[semid];
1445 
1446 			mtx_lock(sema_mtxp);
1447 			if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1448 			    (semakptr->u.sem_perm.seq != seq)) {
1449 				mtx_unlock(sema_mtxp);
1450 				continue;
1451 			}
1452 			if (semnum >= semakptr->u.sem_nsems)
1453 				panic("semexit - semnum out of range");
1454 
1455 			DPRINTF((
1456 			    "semexit:  %p id=%d num=%d(adj=%d) ; sem=%d\n",
1457 			    suptr->un_proc, suptr->un_ent[ix].un_id,
1458 			    suptr->un_ent[ix].un_num,
1459 			    suptr->un_ent[ix].un_adjval,
1460 			    semakptr->u.__sem_base[semnum].semval));
1461 
1462 			if (adjval < 0 && semakptr->u.__sem_base[semnum].semval <
1463 			    -adjval)
1464 				semakptr->u.__sem_base[semnum].semval = 0;
1465 			else
1466 				semakptr->u.__sem_base[semnum].semval += adjval;
1467 
1468 			wakeup(semakptr);
1469 			DPRINTF(("semexit:  back from wakeup\n"));
1470 			mtx_unlock(sema_mtxp);
1471 		}
1472 		SEMUNDO_LOCK();
1473 	}
1474 
1475 	/*
1476 	 * Deallocate the undo vector.
1477 	 */
1478 	DPRINTF(("removing vector\n"));
1479 	suptr->un_proc = NULL;
1480 	suptr->un_cnt = 0;
1481 	LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
1482 	SEMUNDO_UNLOCK();
1483 }
1484 
1485 static int
1486 sysctl_sema(SYSCTL_HANDLER_ARGS)
1487 {
1488 	struct prison *pr, *rpr;
1489 	struct semid_kernel tsemak;
1490 #ifdef COMPAT_FREEBSD32
1491 	struct semid_kernel32 tsemak32;
1492 #endif
1493 	void *outaddr;
1494 	size_t outsize;
1495 	int error, i;
1496 
1497 	pr = req->td->td_ucred->cr_prison;
1498 	rpr = sem_find_prison(req->td->td_ucred);
1499 	error = 0;
1500 	for (i = 0; i < seminfo.semmni; i++) {
1501 		mtx_lock(&sema_mtx[i]);
1502 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) == 0 ||
1503 		    rpr == NULL || sem_prison_cansee(rpr, &sema[i]) != 0)
1504 			bzero(&tsemak, sizeof(tsemak));
1505 		else {
1506 			tsemak = sema[i];
1507 			if (tsemak.cred->cr_prison != pr)
1508 				tsemak.u.sem_perm.key = IPC_PRIVATE;
1509 		}
1510 		mtx_unlock(&sema_mtx[i]);
1511 #ifdef COMPAT_FREEBSD32
1512 		if (SV_CURPROC_FLAG(SV_ILP32)) {
1513 			bzero(&tsemak32, sizeof(tsemak32));
1514 			freebsd32_ipcperm_out(&tsemak.u.sem_perm,
1515 			    &tsemak32.u.sem_perm);
1516 			/* Don't copy u.__sem_base */
1517 			CP(tsemak, tsemak32, u.sem_nsems);
1518 			CP(tsemak, tsemak32, u.sem_otime);
1519 			CP(tsemak, tsemak32, u.sem_ctime);
1520 			/* Don't copy label or cred */
1521 			outaddr = &tsemak32;
1522 			outsize = sizeof(tsemak32);
1523 		} else
1524 #endif
1525 		{
1526 			tsemak.u.__sem_base = NULL;
1527 			tsemak.label = NULL;
1528 			tsemak.cred = NULL;
1529 			outaddr = &tsemak;
1530 			outsize = sizeof(tsemak);
1531 		}
1532 		error = SYSCTL_OUT(req, outaddr, outsize);
1533 		if (error != 0)
1534 			break;
1535 	}
1536 	return (error);
1537 }
1538 
1539 static int
1540 sem_prison_check(void *obj, void *data)
1541 {
1542 	struct prison *pr = obj;
1543 	struct prison *prpr;
1544 	struct vfsoptlist *opts = data;
1545 	int error, jsys;
1546 
1547 	/*
1548 	 * sysvsem is a jailsys integer.
1549 	 * It must be "disable" if the parent jail is disabled.
1550 	 */
1551 	error = vfs_copyopt(opts, "sysvsem", &jsys, sizeof(jsys));
1552 	if (error != ENOENT) {
1553 		if (error != 0)
1554 			return (error);
1555 		switch (jsys) {
1556 		case JAIL_SYS_DISABLE:
1557 			break;
1558 		case JAIL_SYS_NEW:
1559 		case JAIL_SYS_INHERIT:
1560 			prison_lock(pr->pr_parent);
1561 			prpr = osd_jail_get(pr->pr_parent, sem_prison_slot);
1562 			prison_unlock(pr->pr_parent);
1563 			if (prpr == NULL)
1564 				return (EPERM);
1565 			break;
1566 		default:
1567 			return (EINVAL);
1568 		}
1569 	}
1570 
1571 	return (0);
1572 }
1573 
1574 static int
1575 sem_prison_set(void *obj, void *data)
1576 {
1577 	struct prison *pr = obj;
1578 	struct prison *tpr, *orpr, *nrpr, *trpr;
1579 	struct vfsoptlist *opts = data;
1580 	void *rsv;
1581 	int jsys, descend;
1582 
1583 	/*
1584 	 * sysvsem controls which jail is the root of the associated sems (this
1585 	 * jail or same as the parent), or if the feature is available at all.
1586 	 */
1587 	if (vfs_copyopt(opts, "sysvsem", &jsys, sizeof(jsys)) == ENOENT)
1588 		jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0)
1589 		    ? JAIL_SYS_INHERIT
1590 		    : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0)
1591 		    ? JAIL_SYS_DISABLE
1592 		    : -1;
1593 	if (jsys == JAIL_SYS_DISABLE) {
1594 		prison_lock(pr);
1595 		orpr = osd_jail_get(pr, sem_prison_slot);
1596 		if (orpr != NULL)
1597 			osd_jail_del(pr, sem_prison_slot);
1598 		prison_unlock(pr);
1599 		if (orpr != NULL) {
1600 			if (orpr == pr)
1601 				sem_prison_cleanup(pr);
1602 			/* Disable all child jails as well. */
1603 			FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1604 				prison_lock(tpr);
1605 				trpr = osd_jail_get(tpr, sem_prison_slot);
1606 				if (trpr != NULL) {
1607 					osd_jail_del(tpr, sem_prison_slot);
1608 					prison_unlock(tpr);
1609 					if (trpr == tpr)
1610 						sem_prison_cleanup(tpr);
1611 				} else {
1612 					prison_unlock(tpr);
1613 					descend = 0;
1614 				}
1615 			}
1616 		}
1617 	} else if (jsys != -1) {
1618 		if (jsys == JAIL_SYS_NEW)
1619 			nrpr = pr;
1620 		else {
1621 			prison_lock(pr->pr_parent);
1622 			nrpr = osd_jail_get(pr->pr_parent, sem_prison_slot);
1623 			prison_unlock(pr->pr_parent);
1624 		}
1625 		rsv = osd_reserve(sem_prison_slot);
1626 		prison_lock(pr);
1627 		orpr = osd_jail_get(pr, sem_prison_slot);
1628 		if (orpr != nrpr)
1629 			(void)osd_jail_set_reserved(pr, sem_prison_slot, rsv,
1630 			    nrpr);
1631 		else
1632 			osd_free_reserved(rsv);
1633 		prison_unlock(pr);
1634 		if (orpr != nrpr) {
1635 			if (orpr == pr)
1636 				sem_prison_cleanup(pr);
1637 			if (orpr != NULL) {
1638 				/* Change child jails matching the old root, */
1639 				FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1640 					prison_lock(tpr);
1641 					trpr = osd_jail_get(tpr,
1642 					    sem_prison_slot);
1643 					if (trpr == orpr) {
1644 						(void)osd_jail_set(tpr,
1645 						    sem_prison_slot, nrpr);
1646 						prison_unlock(tpr);
1647 						if (trpr == tpr)
1648 							sem_prison_cleanup(tpr);
1649 					} else {
1650 						prison_unlock(tpr);
1651 						descend = 0;
1652 					}
1653 				}
1654 			}
1655 		}
1656 	}
1657 
1658 	return (0);
1659 }
1660 
1661 static int
1662 sem_prison_get(void *obj, void *data)
1663 {
1664 	struct prison *pr = obj;
1665 	struct prison *rpr;
1666 	struct vfsoptlist *opts = data;
1667 	int error, jsys;
1668 
1669 	/* Set sysvsem based on the jail's root prison. */
1670 	prison_lock(pr);
1671 	rpr = osd_jail_get(pr, sem_prison_slot);
1672 	prison_unlock(pr);
1673 	jsys = rpr == NULL ? JAIL_SYS_DISABLE
1674 	    : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT;
1675 	error = vfs_setopt(opts, "sysvsem", &jsys, sizeof(jsys));
1676 	if (error == ENOENT)
1677 		error = 0;
1678 	return (error);
1679 }
1680 
1681 static int
1682 sem_prison_remove(void *obj, void *data __unused)
1683 {
1684 	struct prison *pr = obj;
1685 	struct prison *rpr;
1686 
1687 	prison_lock(pr);
1688 	rpr = osd_jail_get(pr, sem_prison_slot);
1689 	prison_unlock(pr);
1690 	if (rpr == pr)
1691 		sem_prison_cleanup(pr);
1692 	return (0);
1693 }
1694 
1695 static void
1696 sem_prison_cleanup(struct prison *pr)
1697 {
1698 	int i;
1699 
1700 	/* Remove any sems that belong to this jail. */
1701 	mtx_lock(&sem_mtx);
1702 	for (i = 0; i < seminfo.semmni; i++) {
1703 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
1704 		    sema[i].cred != NULL && sema[i].cred->cr_prison == pr) {
1705 			mtx_lock(&sema_mtx[i]);
1706 			sem_remove(i, NULL);
1707 			mtx_unlock(&sema_mtx[i]);
1708 		}
1709 	}
1710 	mtx_unlock(&sem_mtx);
1711 }
1712 
1713 SYSCTL_JAIL_PARAM_SYS_NODE(sysvsem, CTLFLAG_RW, "SYSV semaphores");
1714 
1715 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1716     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1717 
1718 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1719 static sy_call_t *semcalls[] = {
1720 	(sy_call_t *)freebsd7___semctl, (sy_call_t *)sys_semget,
1721 	(sy_call_t *)sys_semop
1722 };
1723 
1724 /*
1725  * Entry point for all SEM calls.
1726  */
1727 int
1728 sys_semsys(td, uap)
1729 	struct thread *td;
1730 	/* XXX actually varargs. */
1731 	struct semsys_args /* {
1732 		int	which;
1733 		int	a2;
1734 		int	a3;
1735 		int	a4;
1736 		int	a5;
1737 	} */ *uap;
1738 {
1739 	int error;
1740 
1741 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1742 	if (uap->which < 0 || uap->which >= nitems(semcalls))
1743 		return (EINVAL);
1744 	error = (*semcalls[uap->which])(td, &uap->a2);
1745 	return (error);
1746 }
1747 
1748 #ifndef CP
1749 #define CP(src, dst, fld)	do { (dst).fld = (src).fld; } while (0)
1750 #endif
1751 
1752 #ifndef _SYS_SYSPROTO_H_
1753 struct freebsd7___semctl_args {
1754 	int	semid;
1755 	int	semnum;
1756 	int	cmd;
1757 	union	semun_old *arg;
1758 };
1759 #endif
1760 int
1761 freebsd7___semctl(struct thread *td, struct freebsd7___semctl_args *uap)
1762 {
1763 	struct semid_ds_old dsold;
1764 	struct semid_ds dsbuf;
1765 	union semun_old arg;
1766 	union semun semun;
1767 	register_t rval;
1768 	int error;
1769 
1770 	switch (uap->cmd) {
1771 	case SEM_STAT:
1772 	case IPC_SET:
1773 	case IPC_STAT:
1774 	case GETALL:
1775 	case SETVAL:
1776 	case SETALL:
1777 		error = copyin(uap->arg, &arg, sizeof(arg));
1778 		if (error)
1779 			return (error);
1780 		break;
1781 	}
1782 
1783 	switch (uap->cmd) {
1784 	case SEM_STAT:
1785 	case IPC_STAT:
1786 		semun.buf = &dsbuf;
1787 		break;
1788 	case IPC_SET:
1789 		error = copyin(arg.buf, &dsold, sizeof(dsold));
1790 		if (error)
1791 			return (error);
1792 		ipcperm_old2new(&dsold.sem_perm, &dsbuf.sem_perm);
1793 		CP(dsold, dsbuf, __sem_base);
1794 		CP(dsold, dsbuf, sem_nsems);
1795 		CP(dsold, dsbuf, sem_otime);
1796 		CP(dsold, dsbuf, sem_ctime);
1797 		semun.buf = &dsbuf;
1798 		break;
1799 	case GETALL:
1800 	case SETALL:
1801 		semun.array = arg.array;
1802 		break;
1803 	case SETVAL:
1804 		semun.val = arg.val;
1805 		break;
1806 	}
1807 
1808 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1809 	    &rval);
1810 	if (error)
1811 		return (error);
1812 
1813 	switch (uap->cmd) {
1814 	case SEM_STAT:
1815 	case IPC_STAT:
1816 		bzero(&dsold, sizeof(dsold));
1817 		ipcperm_new2old(&dsbuf.sem_perm, &dsold.sem_perm);
1818 		CP(dsbuf, dsold, __sem_base);
1819 		CP(dsbuf, dsold, sem_nsems);
1820 		CP(dsbuf, dsold, sem_otime);
1821 		CP(dsbuf, dsold, sem_ctime);
1822 		error = copyout(&dsold, arg.buf, sizeof(dsold));
1823 		break;
1824 	}
1825 
1826 	if (error == 0)
1827 		td->td_retval[0] = rval;
1828 	return (error);
1829 }
1830 
1831 #endif /* COMPAT_FREEBSD{4,5,6,7} */
1832 
1833 #ifdef COMPAT_FREEBSD32
1834 
1835 int
1836 freebsd32_semsys(struct thread *td, struct freebsd32_semsys_args *uap)
1837 {
1838 
1839 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1840     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1841 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1842 	switch (uap->which) {
1843 	case 0:
1844 		return (freebsd7_freebsd32_semctl(td,
1845 		    (struct freebsd7_freebsd32_semctl_args *)&uap->a2));
1846 	default:
1847 		return (sys_semsys(td, (struct semsys_args *)uap));
1848 	}
1849 #else
1850 	return (nosys(td, NULL));
1851 #endif
1852 }
1853 
1854 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1855     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1856 int
1857 freebsd7_freebsd32_semctl(struct thread *td,
1858     struct freebsd7_freebsd32_semctl_args *uap)
1859 {
1860 	struct semid_ds32_old dsbuf32;
1861 	struct semid_ds dsbuf;
1862 	union semun semun;
1863 	union semun32 arg;
1864 	register_t rval;
1865 	int error;
1866 
1867 	switch (uap->cmd) {
1868 	case SEM_STAT:
1869 	case IPC_SET:
1870 	case IPC_STAT:
1871 	case GETALL:
1872 	case SETVAL:
1873 	case SETALL:
1874 		error = copyin(uap->arg, &arg, sizeof(arg));
1875 		if (error)
1876 			return (error);
1877 		break;
1878 	}
1879 
1880 	switch (uap->cmd) {
1881 	case SEM_STAT:
1882 	case IPC_STAT:
1883 		semun.buf = &dsbuf;
1884 		break;
1885 	case IPC_SET:
1886 		error = copyin(PTRIN(arg.buf), &dsbuf32, sizeof(dsbuf32));
1887 		if (error)
1888 			return (error);
1889 		freebsd32_ipcperm_old_in(&dsbuf32.sem_perm, &dsbuf.sem_perm);
1890 		PTRIN_CP(dsbuf32, dsbuf, __sem_base);
1891 		CP(dsbuf32, dsbuf, sem_nsems);
1892 		CP(dsbuf32, dsbuf, sem_otime);
1893 		CP(dsbuf32, dsbuf, sem_ctime);
1894 		semun.buf = &dsbuf;
1895 		break;
1896 	case GETALL:
1897 	case SETALL:
1898 		semun.array = PTRIN(arg.array);
1899 		break;
1900 	case SETVAL:
1901 		semun.val = arg.val;
1902 		break;
1903 	}
1904 
1905 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1906 	    &rval);
1907 	if (error)
1908 		return (error);
1909 
1910 	switch (uap->cmd) {
1911 	case SEM_STAT:
1912 	case IPC_STAT:
1913 		bzero(&dsbuf32, sizeof(dsbuf32));
1914 		freebsd32_ipcperm_old_out(&dsbuf.sem_perm, &dsbuf32.sem_perm);
1915 		PTROUT_CP(dsbuf, dsbuf32, __sem_base);
1916 		CP(dsbuf, dsbuf32, sem_nsems);
1917 		CP(dsbuf, dsbuf32, sem_otime);
1918 		CP(dsbuf, dsbuf32, sem_ctime);
1919 		error = copyout(&dsbuf32, PTRIN(arg.buf), sizeof(dsbuf32));
1920 		break;
1921 	}
1922 
1923 	if (error == 0)
1924 		td->td_retval[0] = rval;
1925 	return (error);
1926 }
1927 #endif
1928 
1929 int
1930 freebsd32_semctl(struct thread *td, struct freebsd32_semctl_args *uap)
1931 {
1932 	struct semid_ds32 dsbuf32;
1933 	struct semid_ds dsbuf;
1934 	union semun semun;
1935 	union semun32 arg;
1936 	register_t rval;
1937 	int error;
1938 
1939 	switch (uap->cmd) {
1940 	case SEM_STAT:
1941 	case IPC_SET:
1942 	case IPC_STAT:
1943 	case GETALL:
1944 	case SETVAL:
1945 	case SETALL:
1946 		error = copyin(uap->arg, &arg, sizeof(arg));
1947 		if (error)
1948 			return (error);
1949 		break;
1950 	}
1951 
1952 	switch (uap->cmd) {
1953 	case SEM_STAT:
1954 	case IPC_STAT:
1955 		semun.buf = &dsbuf;
1956 		break;
1957 	case IPC_SET:
1958 		error = copyin(PTRIN(arg.buf), &dsbuf32, sizeof(dsbuf32));
1959 		if (error)
1960 			return (error);
1961 		freebsd32_ipcperm_in(&dsbuf32.sem_perm, &dsbuf.sem_perm);
1962 		PTRIN_CP(dsbuf32, dsbuf, __sem_base);
1963 		CP(dsbuf32, dsbuf, sem_nsems);
1964 		CP(dsbuf32, dsbuf, sem_otime);
1965 		CP(dsbuf32, dsbuf, sem_ctime);
1966 		semun.buf = &dsbuf;
1967 		break;
1968 	case GETALL:
1969 	case SETALL:
1970 		semun.array = PTRIN(arg.array);
1971 		break;
1972 	case SETVAL:
1973 		semun.val = arg.val;
1974 		break;
1975 	}
1976 
1977 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1978 	    &rval);
1979 	if (error)
1980 		return (error);
1981 
1982 	switch (uap->cmd) {
1983 	case SEM_STAT:
1984 	case IPC_STAT:
1985 		bzero(&dsbuf32, sizeof(dsbuf32));
1986 		freebsd32_ipcperm_out(&dsbuf.sem_perm, &dsbuf32.sem_perm);
1987 		PTROUT_CP(dsbuf, dsbuf32, __sem_base);
1988 		CP(dsbuf, dsbuf32, sem_nsems);
1989 		CP(dsbuf, dsbuf32, sem_otime);
1990 		CP(dsbuf, dsbuf32, sem_ctime);
1991 		error = copyout(&dsbuf32, PTRIN(arg.buf), sizeof(dsbuf32));
1992 		break;
1993 	}
1994 
1995 	if (error == 0)
1996 		td->td_retval[0] = rval;
1997 	return (error);
1998 }
1999 
2000 #endif /* COMPAT_FREEBSD32 */
2001