xref: /freebsd/sys/kern/sysv_sem.c (revision 4cb6ea7e8e3a86f3162baeff92504d0c0ecc6986)
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 =	SEMMNI,	/* # of semaphore identifiers */
202 	.semmns =	SEMMNS,	/* # of semaphores in system */
203 	.semmnu =	SEMMNU,	/* # of undo structures in system */
204 	.semmsl =	SEMMSL,	/* max # of semaphores per id */
205 	.semopm =	SEMOPM,	/* max # of operations per semop call */
206 	.semume =	SEMUME,	/* max # of undo entries per process */
207 	.semusz =	SEMUSZ,	/* size in bytes of undo structure */
208 	.semvmx =	SEMVMX,	/* semaphore maximum value */
209 	.semaem =	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 	mtx_assert(&sem_mtx, MA_OWNED);
563 	semakptr = &sema[semidx];
564 	KASSERT(semakptr->u.__sem_base - sem + semakptr->u.sem_nsems <= semtot,
565 	    ("sem_remove: sema %d corrupted sem pointer %p %p %d %d",
566 	    semidx, semakptr->u.__sem_base, sem, semakptr->u.sem_nsems,
567 	    semtot));
568 
569 	semakptr->u.sem_perm.cuid = cred ? cred->cr_uid : 0;
570 	semakptr->u.sem_perm.uid = cred ? cred->cr_uid : 0;
571 	semakptr->u.sem_perm.mode = 0;
572 	racct_sub_cred(semakptr->cred, RACCT_NSEM, semakptr->u.sem_nsems);
573 	crfree(semakptr->cred);
574 	semakptr->cred = NULL;
575 	SEMUNDO_LOCK();
576 	semundo_clear(semidx, -1);
577 	SEMUNDO_UNLOCK();
578 #ifdef MAC
579 	mac_sysvsem_cleanup(semakptr);
580 #endif
581 	wakeup(semakptr);
582 	for (i = 0; i < seminfo.semmni; i++) {
583 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
584 		    sema[i].u.__sem_base > semakptr->u.__sem_base)
585 			mtx_lock_flags(&sema_mtx[i], LOP_DUPOK);
586 	}
587 	for (i = semakptr->u.__sem_base - sem; i < semtot; i++)
588 		sem[i] = sem[i + semakptr->u.sem_nsems];
589 	for (i = 0; i < seminfo.semmni; i++) {
590 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
591 		    sema[i].u.__sem_base > semakptr->u.__sem_base) {
592 			sema[i].u.__sem_base -= semakptr->u.sem_nsems;
593 			mtx_unlock(&sema_mtx[i]);
594 		}
595 	}
596 	semtot -= semakptr->u.sem_nsems;
597 }
598 
599 static struct prison *
600 sem_find_prison(struct ucred *cred)
601 {
602 	struct prison *pr, *rpr;
603 
604 	pr = cred->cr_prison;
605 	prison_lock(pr);
606 	rpr = osd_jail_get(pr, sem_prison_slot);
607 	prison_unlock(pr);
608 	return rpr;
609 }
610 
611 static int
612 sem_prison_cansee(struct prison *rpr, struct semid_kernel *semakptr)
613 {
614 
615 	if (semakptr->cred == NULL ||
616 	    !(rpr == semakptr->cred->cr_prison ||
617 	      prison_ischild(rpr, semakptr->cred->cr_prison)))
618 		return (EINVAL);
619 	return (0);
620 }
621 
622 /*
623  * Note that the user-mode half of this passes a union, not a pointer.
624  */
625 #ifndef _SYS_SYSPROTO_H_
626 struct __semctl_args {
627 	int	semid;
628 	int	semnum;
629 	int	cmd;
630 	union	semun *arg;
631 };
632 #endif
633 int
634 sys___semctl(struct thread *td, struct __semctl_args *uap)
635 {
636 	struct semid_ds dsbuf;
637 	union semun arg, semun;
638 	register_t rval;
639 	int error;
640 
641 	switch (uap->cmd) {
642 	case SEM_STAT:
643 	case IPC_SET:
644 	case IPC_STAT:
645 	case GETALL:
646 	case SETVAL:
647 	case SETALL:
648 		error = copyin(uap->arg, &arg, sizeof(arg));
649 		if (error)
650 			return (error);
651 		break;
652 	}
653 
654 	switch (uap->cmd) {
655 	case SEM_STAT:
656 	case IPC_STAT:
657 		semun.buf = &dsbuf;
658 		break;
659 	case IPC_SET:
660 		error = copyin(arg.buf, &dsbuf, sizeof(dsbuf));
661 		if (error)
662 			return (error);
663 		semun.buf = &dsbuf;
664 		break;
665 	case GETALL:
666 	case SETALL:
667 		semun.array = arg.array;
668 		break;
669 	case SETVAL:
670 		semun.val = arg.val;
671 		break;
672 	}
673 
674 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
675 	    &rval);
676 	if (error)
677 		return (error);
678 
679 	switch (uap->cmd) {
680 	case SEM_STAT:
681 	case IPC_STAT:
682 		error = copyout(&dsbuf, arg.buf, sizeof(dsbuf));
683 		break;
684 	}
685 
686 	if (error == 0)
687 		td->td_retval[0] = rval;
688 	return (error);
689 }
690 
691 int
692 kern_semctl(struct thread *td, int semid, int semnum, int cmd,
693     union semun *arg, register_t *rval)
694 {
695 	u_short *array;
696 	struct ucred *cred = td->td_ucred;
697 	int i, error;
698 	struct prison *rpr;
699 	struct semid_ds *sbuf;
700 	struct semid_kernel *semakptr;
701 	struct mtx *sema_mtxp;
702 	u_short usval, count;
703 	int semidx;
704 
705 	DPRINTF(("call to semctl(%d, %d, %d, 0x%p)\n",
706 	    semid, semnum, cmd, arg));
707 
708 	AUDIT_ARG_SVIPC_CMD(cmd);
709 	AUDIT_ARG_SVIPC_ID(semid);
710 
711 	rpr = sem_find_prison(td->td_ucred);
712 	if (sem == NULL)
713 		return (ENOSYS);
714 
715 	array = NULL;
716 
717 	switch(cmd) {
718 	case SEM_STAT:
719 		/*
720 		 * For this command we assume semid is an array index
721 		 * rather than an IPC id.
722 		 */
723 		if (semid < 0 || semid >= seminfo.semmni)
724 			return (EINVAL);
725 		semakptr = &sema[semid];
726 		sema_mtxp = &sema_mtx[semid];
727 		mtx_lock(sema_mtxp);
728 		if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
729 			error = EINVAL;
730 			goto done2;
731 		}
732 		if ((error = sem_prison_cansee(rpr, semakptr)))
733 			goto done2;
734 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
735 			goto done2;
736 #ifdef MAC
737 		error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
738 		if (error != 0)
739 			goto done2;
740 #endif
741 		bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
742 		if (cred->cr_prison != semakptr->cred->cr_prison)
743 			arg->buf->sem_perm.key = IPC_PRIVATE;
744 		*rval = IXSEQ_TO_IPCID(semid, semakptr->u.sem_perm);
745 		mtx_unlock(sema_mtxp);
746 		return (0);
747 	}
748 
749 	semidx = IPCID_TO_IX(semid);
750 	if (semidx < 0 || semidx >= seminfo.semmni)
751 		return (EINVAL);
752 
753 	semakptr = &sema[semidx];
754 	sema_mtxp = &sema_mtx[semidx];
755 	if (cmd == IPC_RMID)
756 		mtx_lock(&sem_mtx);
757 	mtx_lock(sema_mtxp);
758 
759 #ifdef MAC
760 	error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
761 	if (error != 0)
762 		goto done2;
763 #endif
764 
765 	error = 0;
766 	*rval = 0;
767 
768 	switch (cmd) {
769 	case IPC_RMID:
770 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
771 			goto done2;
772 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
773 			goto done2;
774 		sem_remove(semidx, cred);
775 		break;
776 
777 	case IPC_SET:
778 		AUDIT_ARG_SVIPC_PERM(&arg->buf->sem_perm);
779 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
780 			goto done2;
781 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
782 			goto done2;
783 		sbuf = arg->buf;
784 		semakptr->u.sem_perm.uid = sbuf->sem_perm.uid;
785 		semakptr->u.sem_perm.gid = sbuf->sem_perm.gid;
786 		semakptr->u.sem_perm.mode = (semakptr->u.sem_perm.mode &
787 		    ~0777) | (sbuf->sem_perm.mode & 0777);
788 		semakptr->u.sem_ctime = time_second;
789 		break;
790 
791 	case IPC_STAT:
792 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
793 			goto done2;
794 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
795 			goto done2;
796 		bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
797 		if (cred->cr_prison != semakptr->cred->cr_prison)
798 			arg->buf->sem_perm.key = IPC_PRIVATE;
799 		break;
800 
801 	case GETNCNT:
802 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
803 			goto done2;
804 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
805 			goto done2;
806 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
807 			error = EINVAL;
808 			goto done2;
809 		}
810 		*rval = semakptr->u.__sem_base[semnum].semncnt;
811 		break;
812 
813 	case GETPID:
814 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
815 			goto done2;
816 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
817 			goto done2;
818 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
819 			error = EINVAL;
820 			goto done2;
821 		}
822 		*rval = semakptr->u.__sem_base[semnum].sempid;
823 		break;
824 
825 	case GETVAL:
826 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
827 			goto done2;
828 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
829 			goto done2;
830 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
831 			error = EINVAL;
832 			goto done2;
833 		}
834 		*rval = semakptr->u.__sem_base[semnum].semval;
835 		break;
836 
837 	case GETALL:
838 		/*
839 		 * Unfortunately, callers of this function don't know
840 		 * in advance how many semaphores are in this set.
841 		 * While we could just allocate the maximum size array
842 		 * and pass the actual size back to the caller, that
843 		 * won't work for SETALL since we can't copyin() more
844 		 * data than the user specified as we may return a
845 		 * spurious EFAULT.
846 		 *
847 		 * Note that the number of semaphores in a set is
848 		 * fixed for the life of that set.  The only way that
849 		 * the 'count' could change while are blocked in
850 		 * malloc() is if this semaphore set were destroyed
851 		 * and a new one created with the same index.
852 		 * However, semvalid() will catch that due to the
853 		 * sequence number unless exactly 0x8000 (or a
854 		 * multiple thereof) semaphore sets for the same index
855 		 * are created and destroyed while we are in malloc!
856 		 *
857 		 */
858 		count = semakptr->u.sem_nsems;
859 		mtx_unlock(sema_mtxp);
860 		array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
861 		mtx_lock(sema_mtxp);
862 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
863 			goto done2;
864 		KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
865 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
866 			goto done2;
867 		for (i = 0; i < semakptr->u.sem_nsems; i++)
868 			array[i] = semakptr->u.__sem_base[i].semval;
869 		mtx_unlock(sema_mtxp);
870 		error = copyout(array, arg->array, count * sizeof(*array));
871 		mtx_lock(sema_mtxp);
872 		break;
873 
874 	case GETZCNT:
875 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
876 			goto done2;
877 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
878 			goto done2;
879 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
880 			error = EINVAL;
881 			goto done2;
882 		}
883 		*rval = semakptr->u.__sem_base[semnum].semzcnt;
884 		break;
885 
886 	case SETVAL:
887 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
888 			goto done2;
889 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
890 			goto done2;
891 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
892 			error = EINVAL;
893 			goto done2;
894 		}
895 		if (arg->val < 0 || arg->val > seminfo.semvmx) {
896 			error = ERANGE;
897 			goto done2;
898 		}
899 		semakptr->u.__sem_base[semnum].semval = arg->val;
900 		SEMUNDO_LOCK();
901 		semundo_clear(semidx, semnum);
902 		SEMUNDO_UNLOCK();
903 		wakeup(semakptr);
904 		break;
905 
906 	case SETALL:
907 		/*
908 		 * See comment on GETALL for why 'count' shouldn't change
909 		 * and why we require a userland buffer.
910 		 */
911 		count = semakptr->u.sem_nsems;
912 		mtx_unlock(sema_mtxp);
913 		array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
914 		error = copyin(arg->array, array, count * sizeof(*array));
915 		mtx_lock(sema_mtxp);
916 		if (error)
917 			break;
918 		if ((error = semvalid(semid, rpr, semakptr)) != 0)
919 			goto done2;
920 		KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
921 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
922 			goto done2;
923 		for (i = 0; i < semakptr->u.sem_nsems; i++) {
924 			usval = array[i];
925 			if (usval > seminfo.semvmx) {
926 				error = ERANGE;
927 				break;
928 			}
929 			semakptr->u.__sem_base[i].semval = usval;
930 		}
931 		SEMUNDO_LOCK();
932 		semundo_clear(semidx, -1);
933 		SEMUNDO_UNLOCK();
934 		wakeup(semakptr);
935 		break;
936 
937 	default:
938 		error = EINVAL;
939 		break;
940 	}
941 
942 done2:
943 	mtx_unlock(sema_mtxp);
944 	if (cmd == IPC_RMID)
945 		mtx_unlock(&sem_mtx);
946 	if (array != NULL)
947 		free(array, M_TEMP);
948 	return(error);
949 }
950 
951 #ifndef _SYS_SYSPROTO_H_
952 struct semget_args {
953 	key_t	key;
954 	int	nsems;
955 	int	semflg;
956 };
957 #endif
958 int
959 sys_semget(struct thread *td, struct semget_args *uap)
960 {
961 	int semid, error = 0;
962 	int key = uap->key;
963 	int nsems = uap->nsems;
964 	int semflg = uap->semflg;
965 	struct ucred *cred = td->td_ucred;
966 
967 	DPRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg));
968 
969 	AUDIT_ARG_VALUE(semflg);
970 
971 	if (sem_find_prison(cred) == NULL)
972 		return (ENOSYS);
973 
974 	mtx_lock(&sem_mtx);
975 	if (key != IPC_PRIVATE) {
976 		for (semid = 0; semid < seminfo.semmni; semid++) {
977 			if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) &&
978 			    sema[semid].cred != NULL &&
979 			    sema[semid].cred->cr_prison == cred->cr_prison &&
980 			    sema[semid].u.sem_perm.key == key)
981 				break;
982 		}
983 		if (semid < seminfo.semmni) {
984 			AUDIT_ARG_SVIPC_ID(semid);
985 			DPRINTF(("found public key\n"));
986 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
987 				DPRINTF(("not exclusive\n"));
988 				error = EEXIST;
989 				goto done2;
990 			}
991 			if ((error = ipcperm(td, &sema[semid].u.sem_perm,
992 			    semflg & 0700))) {
993 				goto done2;
994 			}
995 			if (nsems > 0 && sema[semid].u.sem_nsems < nsems) {
996 				DPRINTF(("too small\n"));
997 				error = EINVAL;
998 				goto done2;
999 			}
1000 #ifdef MAC
1001 			error = mac_sysvsem_check_semget(cred, &sema[semid]);
1002 			if (error != 0)
1003 				goto done2;
1004 #endif
1005 			goto found;
1006 		}
1007 	}
1008 
1009 	DPRINTF(("need to allocate the semid_kernel\n"));
1010 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
1011 		if (nsems <= 0 || nsems > seminfo.semmsl) {
1012 			DPRINTF(("nsems out of range (0<%d<=%d)\n", nsems,
1013 			    seminfo.semmsl));
1014 			error = EINVAL;
1015 			goto done2;
1016 		}
1017 		if (nsems > seminfo.semmns - semtot) {
1018 			DPRINTF((
1019 			    "not enough semaphores left (need %d, got %d)\n",
1020 			    nsems, seminfo.semmns - semtot));
1021 			error = ENOSPC;
1022 			goto done2;
1023 		}
1024 		for (semid = 0; semid < seminfo.semmni; semid++) {
1025 			if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0)
1026 				break;
1027 		}
1028 		if (semid == seminfo.semmni) {
1029 			DPRINTF(("no more semid_kernel's available\n"));
1030 			error = ENOSPC;
1031 			goto done2;
1032 		}
1033 #ifdef RACCT
1034 		if (racct_enable) {
1035 			PROC_LOCK(td->td_proc);
1036 			error = racct_add(td->td_proc, RACCT_NSEM, nsems);
1037 			PROC_UNLOCK(td->td_proc);
1038 			if (error != 0) {
1039 				error = ENOSPC;
1040 				goto done2;
1041 			}
1042 		}
1043 #endif
1044 		DPRINTF(("semid %d is available\n", semid));
1045 		mtx_lock(&sema_mtx[semid]);
1046 		KASSERT((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0,
1047 		    ("Lost semaphore %d", semid));
1048 		sema[semid].u.sem_perm.key = key;
1049 		sema[semid].u.sem_perm.cuid = cred->cr_uid;
1050 		sema[semid].u.sem_perm.uid = cred->cr_uid;
1051 		sema[semid].u.sem_perm.cgid = cred->cr_gid;
1052 		sema[semid].u.sem_perm.gid = cred->cr_gid;
1053 		sema[semid].u.sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
1054 		sema[semid].cred = crhold(cred);
1055 		sema[semid].u.sem_perm.seq =
1056 		    (sema[semid].u.sem_perm.seq + 1) & 0x7fff;
1057 		sema[semid].u.sem_nsems = nsems;
1058 		sema[semid].u.sem_otime = 0;
1059 		sema[semid].u.sem_ctime = time_second;
1060 		sema[semid].u.__sem_base = &sem[semtot];
1061 		semtot += nsems;
1062 		bzero(sema[semid].u.__sem_base,
1063 		    sizeof(sema[semid].u.__sem_base[0])*nsems);
1064 #ifdef MAC
1065 		mac_sysvsem_create(cred, &sema[semid]);
1066 #endif
1067 		mtx_unlock(&sema_mtx[semid]);
1068 		DPRINTF(("sembase = %p, next = %p\n",
1069 		    sema[semid].u.__sem_base, &sem[semtot]));
1070 	} else {
1071 		DPRINTF(("didn't find it and wasn't asked to create it\n"));
1072 		error = ENOENT;
1073 		goto done2;
1074 	}
1075 
1076 found:
1077 	td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].u.sem_perm);
1078 done2:
1079 	mtx_unlock(&sem_mtx);
1080 	return (error);
1081 }
1082 
1083 #ifndef _SYS_SYSPROTO_H_
1084 struct semop_args {
1085 	int	semid;
1086 	struct	sembuf *sops;
1087 	size_t	nsops;
1088 };
1089 #endif
1090 int
1091 sys_semop(struct thread *td, struct semop_args *uap)
1092 {
1093 #define SMALL_SOPS	8
1094 	struct sembuf small_sops[SMALL_SOPS];
1095 	int semid = uap->semid;
1096 	size_t nsops = uap->nsops;
1097 	struct prison *rpr;
1098 	struct sembuf *sops;
1099 	struct semid_kernel *semakptr;
1100 	struct sembuf *sopptr = NULL;
1101 	struct sem *semptr = NULL;
1102 	struct sem_undo *suptr;
1103 	struct mtx *sema_mtxp;
1104 	size_t i, j, k;
1105 	int error;
1106 	int do_wakeup, do_undos;
1107 	unsigned short seq;
1108 
1109 #ifdef SEM_DEBUG
1110 	sops = NULL;
1111 #endif
1112 	DPRINTF(("call to semop(%d, %p, %u)\n", semid, sops, nsops));
1113 
1114 	AUDIT_ARG_SVIPC_ID(semid);
1115 
1116 	rpr = sem_find_prison(td->td_ucred);
1117 	if (sem == NULL)
1118 		return (ENOSYS);
1119 
1120 	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
1121 
1122 	if (semid < 0 || semid >= seminfo.semmni)
1123 		return (EINVAL);
1124 
1125 	/* Allocate memory for sem_ops */
1126 	if (nsops <= SMALL_SOPS)
1127 		sops = small_sops;
1128 	else if (nsops > seminfo.semopm) {
1129 		DPRINTF(("too many sops (max=%d, nsops=%d)\n", seminfo.semopm,
1130 		    nsops));
1131 		return (E2BIG);
1132 	} else {
1133 #ifdef RACCT
1134 		if (racct_enable) {
1135 			PROC_LOCK(td->td_proc);
1136 			if (nsops >
1137 			    racct_get_available(td->td_proc, RACCT_NSEMOP)) {
1138 				PROC_UNLOCK(td->td_proc);
1139 				return (E2BIG);
1140 			}
1141 			PROC_UNLOCK(td->td_proc);
1142 		}
1143 #endif
1144 
1145 		sops = malloc(nsops * sizeof(*sops), M_TEMP, M_WAITOK);
1146 	}
1147 	if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) {
1148 		DPRINTF(("error = %d from copyin(%p, %p, %d)\n", error,
1149 		    uap->sops, sops, nsops * sizeof(sops[0])));
1150 		if (sops != small_sops)
1151 			free(sops, M_TEMP);
1152 		return (error);
1153 	}
1154 
1155 	semakptr = &sema[semid];
1156 	sema_mtxp = &sema_mtx[semid];
1157 	mtx_lock(sema_mtxp);
1158 	if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
1159 		error = EINVAL;
1160 		goto done2;
1161 	}
1162 	seq = semakptr->u.sem_perm.seq;
1163 	if (seq != IPCID_TO_SEQ(uap->semid)) {
1164 		error = EINVAL;
1165 		goto done2;
1166 	}
1167 	if ((error = sem_prison_cansee(rpr, semakptr)) != 0)
1168 		goto done2;
1169 	/*
1170 	 * Initial pass through sops to see what permissions are needed.
1171 	 * Also perform any checks that don't need repeating on each
1172 	 * attempt to satisfy the request vector.
1173 	 */
1174 	j = 0;		/* permission needed */
1175 	do_undos = 0;
1176 	for (i = 0; i < nsops; i++) {
1177 		sopptr = &sops[i];
1178 		if (sopptr->sem_num >= semakptr->u.sem_nsems) {
1179 			error = EFBIG;
1180 			goto done2;
1181 		}
1182 		if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0)
1183 			do_undos = 1;
1184 		j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A;
1185 	}
1186 
1187 	if ((error = ipcperm(td, &semakptr->u.sem_perm, j))) {
1188 		DPRINTF(("error = %d from ipaccess\n", error));
1189 		goto done2;
1190 	}
1191 #ifdef MAC
1192 	error = mac_sysvsem_check_semop(td->td_ucred, semakptr, j);
1193 	if (error != 0)
1194 		goto done2;
1195 #endif
1196 
1197 	/*
1198 	 * Loop trying to satisfy the vector of requests.
1199 	 * If we reach a point where we must wait, any requests already
1200 	 * performed are rolled back and we go to sleep until some other
1201 	 * process wakes us up.  At this point, we start all over again.
1202 	 *
1203 	 * This ensures that from the perspective of other tasks, a set
1204 	 * of requests is atomic (never partially satisfied).
1205 	 */
1206 	for (;;) {
1207 		do_wakeup = 0;
1208 		error = 0;	/* error return if necessary */
1209 
1210 		for (i = 0; i < nsops; i++) {
1211 			sopptr = &sops[i];
1212 			semptr = &semakptr->u.__sem_base[sopptr->sem_num];
1213 
1214 			DPRINTF((
1215 			    "semop:  semakptr=%p, __sem_base=%p, "
1216 			    "semptr=%p, sem[%d]=%d : op=%d, flag=%s\n",
1217 			    semakptr, semakptr->u.__sem_base, semptr,
1218 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
1219 			    (sopptr->sem_flg & IPC_NOWAIT) ?
1220 			    "nowait" : "wait"));
1221 
1222 			if (sopptr->sem_op < 0) {
1223 				if (semptr->semval + sopptr->sem_op < 0) {
1224 					DPRINTF(("semop:  can't do it now\n"));
1225 					break;
1226 				} else {
1227 					semptr->semval += sopptr->sem_op;
1228 					if (semptr->semval == 0 &&
1229 					    semptr->semzcnt > 0)
1230 						do_wakeup = 1;
1231 				}
1232 			} else if (sopptr->sem_op == 0) {
1233 				if (semptr->semval != 0) {
1234 					DPRINTF(("semop:  not zero now\n"));
1235 					break;
1236 				}
1237 			} else if (semptr->semval + sopptr->sem_op >
1238 			    seminfo.semvmx) {
1239 				error = ERANGE;
1240 				break;
1241 			} else {
1242 				if (semptr->semncnt > 0)
1243 					do_wakeup = 1;
1244 				semptr->semval += sopptr->sem_op;
1245 			}
1246 		}
1247 
1248 		/*
1249 		 * Did we get through the entire vector?
1250 		 */
1251 		if (i >= nsops)
1252 			goto done;
1253 
1254 		/*
1255 		 * No ... rollback anything that we've already done
1256 		 */
1257 		DPRINTF(("semop:  rollback 0 through %d\n", i-1));
1258 		for (j = 0; j < i; j++)
1259 			semakptr->u.__sem_base[sops[j].sem_num].semval -=
1260 			    sops[j].sem_op;
1261 
1262 		/* If we detected an error, return it */
1263 		if (error != 0)
1264 			goto done2;
1265 
1266 		/*
1267 		 * If the request that we couldn't satisfy has the
1268 		 * NOWAIT flag set then return with EAGAIN.
1269 		 */
1270 		if (sopptr->sem_flg & IPC_NOWAIT) {
1271 			error = EAGAIN;
1272 			goto done2;
1273 		}
1274 
1275 		if (sopptr->sem_op == 0)
1276 			semptr->semzcnt++;
1277 		else
1278 			semptr->semncnt++;
1279 
1280 		DPRINTF(("semop:  good night!\n"));
1281 		error = msleep(semakptr, sema_mtxp, (PZERO - 4) | PCATCH,
1282 		    "semwait", 0);
1283 		DPRINTF(("semop:  good morning (error=%d)!\n", error));
1284 		/* return code is checked below, after sem[nz]cnt-- */
1285 
1286 		/*
1287 		 * Make sure that the semaphore still exists
1288 		 */
1289 		seq = semakptr->u.sem_perm.seq;
1290 		if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1291 		    seq != IPCID_TO_SEQ(uap->semid)) {
1292 			error = EIDRM;
1293 			goto done2;
1294 		}
1295 
1296 		/*
1297 		 * Renew the semaphore's pointer after wakeup since
1298 		 * during msleep __sem_base may have been modified and semptr
1299 		 * is not valid any more
1300 		 */
1301 		semptr = &semakptr->u.__sem_base[sopptr->sem_num];
1302 
1303 		/*
1304 		 * The semaphore is still alive.  Readjust the count of
1305 		 * waiting processes.
1306 		 */
1307 		if (sopptr->sem_op == 0)
1308 			semptr->semzcnt--;
1309 		else
1310 			semptr->semncnt--;
1311 
1312 		/*
1313 		 * Is it really morning, or was our sleep interrupted?
1314 		 * (Delayed check of msleep() return code because we
1315 		 * need to decrement sem[nz]cnt either way.)
1316 		 */
1317 		if (error != 0) {
1318 			error = EINTR;
1319 			goto done2;
1320 		}
1321 		DPRINTF(("semop:  good morning!\n"));
1322 	}
1323 
1324 done:
1325 	/*
1326 	 * Process any SEM_UNDO requests.
1327 	 */
1328 	if (do_undos) {
1329 		SEMUNDO_LOCK();
1330 		suptr = NULL;
1331 		for (i = 0; i < nsops; i++) {
1332 			/*
1333 			 * We only need to deal with SEM_UNDO's for non-zero
1334 			 * op's.
1335 			 */
1336 			int adjval;
1337 
1338 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
1339 				continue;
1340 			adjval = sops[i].sem_op;
1341 			if (adjval == 0)
1342 				continue;
1343 			error = semundo_adjust(td, &suptr, semid, seq,
1344 			    sops[i].sem_num, -adjval);
1345 			if (error == 0)
1346 				continue;
1347 
1348 			/*
1349 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
1350 			 * Rollback the adjustments to this point and then
1351 			 * rollback the semaphore ups and down so we can return
1352 			 * with an error with all structures restored.  We
1353 			 * rollback the undo's in the exact reverse order that
1354 			 * we applied them.  This guarantees that we won't run
1355 			 * out of space as we roll things back out.
1356 			 */
1357 			for (j = 0; j < i; j++) {
1358 				k = i - j - 1;
1359 				if ((sops[k].sem_flg & SEM_UNDO) == 0)
1360 					continue;
1361 				adjval = sops[k].sem_op;
1362 				if (adjval == 0)
1363 					continue;
1364 				if (semundo_adjust(td, &suptr, semid, seq,
1365 				    sops[k].sem_num, adjval) != 0)
1366 					panic("semop - can't undo undos");
1367 			}
1368 
1369 			for (j = 0; j < nsops; j++)
1370 				semakptr->u.__sem_base[sops[j].sem_num].semval -=
1371 				    sops[j].sem_op;
1372 
1373 			DPRINTF(("error = %d from semundo_adjust\n", error));
1374 			SEMUNDO_UNLOCK();
1375 			goto done2;
1376 		} /* loop through the sops */
1377 		SEMUNDO_UNLOCK();
1378 	} /* if (do_undos) */
1379 
1380 	/* We're definitely done - set the sempid's and time */
1381 	for (i = 0; i < nsops; i++) {
1382 		sopptr = &sops[i];
1383 		semptr = &semakptr->u.__sem_base[sopptr->sem_num];
1384 		semptr->sempid = td->td_proc->p_pid;
1385 	}
1386 	semakptr->u.sem_otime = time_second;
1387 
1388 	/*
1389 	 * Do a wakeup if any semaphore was up'd whilst something was
1390 	 * sleeping on it.
1391 	 */
1392 	if (do_wakeup) {
1393 		DPRINTF(("semop:  doing wakeup\n"));
1394 		wakeup(semakptr);
1395 		DPRINTF(("semop:  back from wakeup\n"));
1396 	}
1397 	DPRINTF(("semop:  done\n"));
1398 	td->td_retval[0] = 0;
1399 done2:
1400 	mtx_unlock(sema_mtxp);
1401 	if (sops != small_sops)
1402 		free(sops, M_TEMP);
1403 	return (error);
1404 }
1405 
1406 /*
1407  * Go through the undo structures for this process and apply the adjustments to
1408  * semaphores.
1409  */
1410 static void
1411 semexit_myhook(void *arg, struct proc *p)
1412 {
1413 	struct sem_undo *suptr;
1414 	struct semid_kernel *semakptr;
1415 	struct mtx *sema_mtxp;
1416 	int semid, semnum, adjval, ix;
1417 	unsigned short seq;
1418 
1419 	/*
1420 	 * Go through the chain of undo vectors looking for one
1421 	 * associated with this process.
1422 	 */
1423 	if (LIST_EMPTY(&semu_list))
1424 		return;
1425 	SEMUNDO_LOCK();
1426 	LIST_FOREACH(suptr, &semu_list, un_next) {
1427 		if (suptr->un_proc == p)
1428 			break;
1429 	}
1430 	if (suptr == NULL) {
1431 		SEMUNDO_UNLOCK();
1432 		return;
1433 	}
1434 	LIST_REMOVE(suptr, un_next);
1435 
1436 	DPRINTF(("proc @%p has undo structure with %d entries\n", p,
1437 	    suptr->un_cnt));
1438 
1439 	/*
1440 	 * If there are any active undo elements then process them.
1441 	 */
1442 	if (suptr->un_cnt > 0) {
1443 		SEMUNDO_UNLOCK();
1444 		for (ix = 0; ix < suptr->un_cnt; ix++) {
1445 			semid = suptr->un_ent[ix].un_id;
1446 			semnum = suptr->un_ent[ix].un_num;
1447 			adjval = suptr->un_ent[ix].un_adjval;
1448 			seq = suptr->un_ent[ix].un_seq;
1449 			semakptr = &sema[semid];
1450 			sema_mtxp = &sema_mtx[semid];
1451 
1452 			mtx_lock(sema_mtxp);
1453 			if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1454 			    (semakptr->u.sem_perm.seq != seq)) {
1455 				mtx_unlock(sema_mtxp);
1456 				continue;
1457 			}
1458 			if (semnum >= semakptr->u.sem_nsems)
1459 				panic("semexit - semnum out of range");
1460 
1461 			DPRINTF((
1462 			    "semexit:  %p id=%d num=%d(adj=%d) ; sem=%d\n",
1463 			    suptr->un_proc, suptr->un_ent[ix].un_id,
1464 			    suptr->un_ent[ix].un_num,
1465 			    suptr->un_ent[ix].un_adjval,
1466 			    semakptr->u.__sem_base[semnum].semval));
1467 
1468 			if (adjval < 0 && semakptr->u.__sem_base[semnum].semval <
1469 			    -adjval)
1470 				semakptr->u.__sem_base[semnum].semval = 0;
1471 			else
1472 				semakptr->u.__sem_base[semnum].semval += adjval;
1473 
1474 			wakeup(semakptr);
1475 			DPRINTF(("semexit:  back from wakeup\n"));
1476 			mtx_unlock(sema_mtxp);
1477 		}
1478 		SEMUNDO_LOCK();
1479 	}
1480 
1481 	/*
1482 	 * Deallocate the undo vector.
1483 	 */
1484 	DPRINTF(("removing vector\n"));
1485 	suptr->un_proc = NULL;
1486 	suptr->un_cnt = 0;
1487 	LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
1488 	SEMUNDO_UNLOCK();
1489 }
1490 
1491 static int
1492 sysctl_sema(SYSCTL_HANDLER_ARGS)
1493 {
1494 	struct prison *pr, *rpr;
1495 	struct semid_kernel tsemak;
1496 #ifdef COMPAT_FREEBSD32
1497 	struct semid_kernel32 tsemak32;
1498 #endif
1499 	void *outaddr;
1500 	size_t outsize;
1501 	int error, i;
1502 
1503 	pr = req->td->td_ucred->cr_prison;
1504 	rpr = sem_find_prison(req->td->td_ucred);
1505 	error = 0;
1506 	for (i = 0; i < seminfo.semmni; i++) {
1507 		mtx_lock(&sema_mtx[i]);
1508 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) == 0 ||
1509 		    rpr == NULL || sem_prison_cansee(rpr, &sema[i]) != 0)
1510 			bzero(&tsemak, sizeof(tsemak));
1511 		else {
1512 			tsemak = sema[i];
1513 			if (tsemak.cred->cr_prison != pr)
1514 				tsemak.u.sem_perm.key = IPC_PRIVATE;
1515 		}
1516 		mtx_unlock(&sema_mtx[i]);
1517 #ifdef COMPAT_FREEBSD32
1518 		if (SV_CURPROC_FLAG(SV_ILP32)) {
1519 			bzero(&tsemak32, sizeof(tsemak32));
1520 			freebsd32_ipcperm_out(&tsemak.u.sem_perm,
1521 			    &tsemak32.u.sem_perm);
1522 			/* Don't copy u.__sem_base */
1523 			CP(tsemak, tsemak32, u.sem_nsems);
1524 			CP(tsemak, tsemak32, u.sem_otime);
1525 			CP(tsemak, tsemak32, u.sem_ctime);
1526 			/* Don't copy label or cred */
1527 			outaddr = &tsemak32;
1528 			outsize = sizeof(tsemak32);
1529 		} else
1530 #endif
1531 		{
1532 			tsemak.u.__sem_base = NULL;
1533 			tsemak.label = NULL;
1534 			tsemak.cred = NULL;
1535 			outaddr = &tsemak;
1536 			outsize = sizeof(tsemak);
1537 		}
1538 		error = SYSCTL_OUT(req, outaddr, outsize);
1539 		if (error != 0)
1540 			break;
1541 	}
1542 	return (error);
1543 }
1544 
1545 static int
1546 sem_prison_check(void *obj, void *data)
1547 {
1548 	struct prison *pr = obj;
1549 	struct prison *prpr;
1550 	struct vfsoptlist *opts = data;
1551 	int error, jsys;
1552 
1553 	/*
1554 	 * sysvsem is a jailsys integer.
1555 	 * It must be "disable" if the parent jail is disabled.
1556 	 */
1557 	error = vfs_copyopt(opts, "sysvsem", &jsys, sizeof(jsys));
1558 	if (error != ENOENT) {
1559 		if (error != 0)
1560 			return (error);
1561 		switch (jsys) {
1562 		case JAIL_SYS_DISABLE:
1563 			break;
1564 		case JAIL_SYS_NEW:
1565 		case JAIL_SYS_INHERIT:
1566 			prison_lock(pr->pr_parent);
1567 			prpr = osd_jail_get(pr->pr_parent, sem_prison_slot);
1568 			prison_unlock(pr->pr_parent);
1569 			if (prpr == NULL)
1570 				return (EPERM);
1571 			break;
1572 		default:
1573 			return (EINVAL);
1574 		}
1575 	}
1576 
1577 	return (0);
1578 }
1579 
1580 static int
1581 sem_prison_set(void *obj, void *data)
1582 {
1583 	struct prison *pr = obj;
1584 	struct prison *tpr, *orpr, *nrpr, *trpr;
1585 	struct vfsoptlist *opts = data;
1586 	void *rsv;
1587 	int jsys, descend;
1588 
1589 	/*
1590 	 * sysvsem controls which jail is the root of the associated sems (this
1591 	 * jail or same as the parent), or if the feature is available at all.
1592 	 */
1593 	if (vfs_copyopt(opts, "sysvsem", &jsys, sizeof(jsys)) == ENOENT)
1594 		jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0)
1595 		    ? JAIL_SYS_INHERIT
1596 		    : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0)
1597 		    ? JAIL_SYS_DISABLE
1598 		    : -1;
1599 	if (jsys == JAIL_SYS_DISABLE) {
1600 		prison_lock(pr);
1601 		orpr = osd_jail_get(pr, sem_prison_slot);
1602 		if (orpr != NULL)
1603 			osd_jail_del(pr, sem_prison_slot);
1604 		prison_unlock(pr);
1605 		if (orpr != NULL) {
1606 			if (orpr == pr)
1607 				sem_prison_cleanup(pr);
1608 			/* Disable all child jails as well. */
1609 			FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1610 				prison_lock(tpr);
1611 				trpr = osd_jail_get(tpr, sem_prison_slot);
1612 				if (trpr != NULL) {
1613 					osd_jail_del(tpr, sem_prison_slot);
1614 					prison_unlock(tpr);
1615 					if (trpr == tpr)
1616 						sem_prison_cleanup(tpr);
1617 				} else {
1618 					prison_unlock(tpr);
1619 					descend = 0;
1620 				}
1621 			}
1622 		}
1623 	} else if (jsys != -1) {
1624 		if (jsys == JAIL_SYS_NEW)
1625 			nrpr = pr;
1626 		else {
1627 			prison_lock(pr->pr_parent);
1628 			nrpr = osd_jail_get(pr->pr_parent, sem_prison_slot);
1629 			prison_unlock(pr->pr_parent);
1630 		}
1631 		rsv = osd_reserve(sem_prison_slot);
1632 		prison_lock(pr);
1633 		orpr = osd_jail_get(pr, sem_prison_slot);
1634 		if (orpr != nrpr)
1635 			(void)osd_jail_set_reserved(pr, sem_prison_slot, rsv,
1636 			    nrpr);
1637 		else
1638 			osd_free_reserved(rsv);
1639 		prison_unlock(pr);
1640 		if (orpr != nrpr) {
1641 			if (orpr == pr)
1642 				sem_prison_cleanup(pr);
1643 			if (orpr != NULL) {
1644 				/* Change child jails matching the old root, */
1645 				FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1646 					prison_lock(tpr);
1647 					trpr = osd_jail_get(tpr,
1648 					    sem_prison_slot);
1649 					if (trpr == orpr) {
1650 						(void)osd_jail_set(tpr,
1651 						    sem_prison_slot, nrpr);
1652 						prison_unlock(tpr);
1653 						if (trpr == tpr)
1654 							sem_prison_cleanup(tpr);
1655 					} else {
1656 						prison_unlock(tpr);
1657 						descend = 0;
1658 					}
1659 				}
1660 			}
1661 		}
1662 	}
1663 
1664 	return (0);
1665 }
1666 
1667 static int
1668 sem_prison_get(void *obj, void *data)
1669 {
1670 	struct prison *pr = obj;
1671 	struct prison *rpr;
1672 	struct vfsoptlist *opts = data;
1673 	int error, jsys;
1674 
1675 	/* Set sysvsem based on the jail's root prison. */
1676 	prison_lock(pr);
1677 	rpr = osd_jail_get(pr, sem_prison_slot);
1678 	prison_unlock(pr);
1679 	jsys = rpr == NULL ? JAIL_SYS_DISABLE
1680 	    : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT;
1681 	error = vfs_setopt(opts, "sysvsem", &jsys, sizeof(jsys));
1682 	if (error == ENOENT)
1683 		error = 0;
1684 	return (error);
1685 }
1686 
1687 static int
1688 sem_prison_remove(void *obj, void *data __unused)
1689 {
1690 	struct prison *pr = obj;
1691 	struct prison *rpr;
1692 
1693 	prison_lock(pr);
1694 	rpr = osd_jail_get(pr, sem_prison_slot);
1695 	prison_unlock(pr);
1696 	if (rpr == pr)
1697 		sem_prison_cleanup(pr);
1698 	return (0);
1699 }
1700 
1701 static void
1702 sem_prison_cleanup(struct prison *pr)
1703 {
1704 	int i;
1705 
1706 	/* Remove any sems that belong to this jail. */
1707 	mtx_lock(&sem_mtx);
1708 	for (i = 0; i < seminfo.semmni; i++) {
1709 		if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
1710 		    sema[i].cred != NULL && sema[i].cred->cr_prison == pr) {
1711 			mtx_lock(&sema_mtx[i]);
1712 			sem_remove(i, NULL);
1713 			mtx_unlock(&sema_mtx[i]);
1714 		}
1715 	}
1716 	mtx_unlock(&sem_mtx);
1717 }
1718 
1719 SYSCTL_JAIL_PARAM_SYS_NODE(sysvsem, CTLFLAG_RW, "SYSV semaphores");
1720 
1721 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1722     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1723 
1724 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1725 static sy_call_t *semcalls[] = {
1726 	(sy_call_t *)freebsd7___semctl, (sy_call_t *)sys_semget,
1727 	(sy_call_t *)sys_semop
1728 };
1729 
1730 /*
1731  * Entry point for all SEM calls.
1732  */
1733 int
1734 sys_semsys(td, uap)
1735 	struct thread *td;
1736 	/* XXX actually varargs. */
1737 	struct semsys_args /* {
1738 		int	which;
1739 		int	a2;
1740 		int	a3;
1741 		int	a4;
1742 		int	a5;
1743 	} */ *uap;
1744 {
1745 	int error;
1746 
1747 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1748 	if (uap->which < 0 || uap->which >= nitems(semcalls))
1749 		return (EINVAL);
1750 	error = (*semcalls[uap->which])(td, &uap->a2);
1751 	return (error);
1752 }
1753 
1754 #ifndef CP
1755 #define CP(src, dst, fld)	do { (dst).fld = (src).fld; } while (0)
1756 #endif
1757 
1758 #ifndef _SYS_SYSPROTO_H_
1759 struct freebsd7___semctl_args {
1760 	int	semid;
1761 	int	semnum;
1762 	int	cmd;
1763 	union	semun_old *arg;
1764 };
1765 #endif
1766 int
1767 freebsd7___semctl(struct thread *td, struct freebsd7___semctl_args *uap)
1768 {
1769 	struct semid_ds_old dsold;
1770 	struct semid_ds dsbuf;
1771 	union semun_old arg;
1772 	union semun semun;
1773 	register_t rval;
1774 	int error;
1775 
1776 	switch (uap->cmd) {
1777 	case SEM_STAT:
1778 	case IPC_SET:
1779 	case IPC_STAT:
1780 	case GETALL:
1781 	case SETVAL:
1782 	case SETALL:
1783 		error = copyin(uap->arg, &arg, sizeof(arg));
1784 		if (error)
1785 			return (error);
1786 		break;
1787 	}
1788 
1789 	switch (uap->cmd) {
1790 	case SEM_STAT:
1791 	case IPC_STAT:
1792 		semun.buf = &dsbuf;
1793 		break;
1794 	case IPC_SET:
1795 		error = copyin(arg.buf, &dsold, sizeof(dsold));
1796 		if (error)
1797 			return (error);
1798 		ipcperm_old2new(&dsold.sem_perm, &dsbuf.sem_perm);
1799 		CP(dsold, dsbuf, __sem_base);
1800 		CP(dsold, dsbuf, sem_nsems);
1801 		CP(dsold, dsbuf, sem_otime);
1802 		CP(dsold, dsbuf, sem_ctime);
1803 		semun.buf = &dsbuf;
1804 		break;
1805 	case GETALL:
1806 	case SETALL:
1807 		semun.array = arg.array;
1808 		break;
1809 	case SETVAL:
1810 		semun.val = arg.val;
1811 		break;
1812 	}
1813 
1814 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1815 	    &rval);
1816 	if (error)
1817 		return (error);
1818 
1819 	switch (uap->cmd) {
1820 	case SEM_STAT:
1821 	case IPC_STAT:
1822 		bzero(&dsold, sizeof(dsold));
1823 		ipcperm_new2old(&dsbuf.sem_perm, &dsold.sem_perm);
1824 		CP(dsbuf, dsold, __sem_base);
1825 		CP(dsbuf, dsold, sem_nsems);
1826 		CP(dsbuf, dsold, sem_otime);
1827 		CP(dsbuf, dsold, sem_ctime);
1828 		error = copyout(&dsold, arg.buf, sizeof(dsold));
1829 		break;
1830 	}
1831 
1832 	if (error == 0)
1833 		td->td_retval[0] = rval;
1834 	return (error);
1835 }
1836 
1837 #endif /* COMPAT_FREEBSD{4,5,6,7} */
1838 
1839 #ifdef COMPAT_FREEBSD32
1840 
1841 int
1842 freebsd32_semsys(struct thread *td, struct freebsd32_semsys_args *uap)
1843 {
1844 
1845 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1846     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1847 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1848 	switch (uap->which) {
1849 	case 0:
1850 		return (freebsd7_freebsd32_semctl(td,
1851 		    (struct freebsd7_freebsd32_semctl_args *)&uap->a2));
1852 	default:
1853 		return (sys_semsys(td, (struct semsys_args *)uap));
1854 	}
1855 #else
1856 	return (nosys(td, NULL));
1857 #endif
1858 }
1859 
1860 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1861     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1862 int
1863 freebsd7_freebsd32_semctl(struct thread *td,
1864     struct freebsd7_freebsd32_semctl_args *uap)
1865 {
1866 	struct semid_ds32_old dsbuf32;
1867 	struct semid_ds dsbuf;
1868 	union semun semun;
1869 	union semun32 arg;
1870 	register_t rval;
1871 	int error;
1872 
1873 	switch (uap->cmd) {
1874 	case SEM_STAT:
1875 	case IPC_SET:
1876 	case IPC_STAT:
1877 	case GETALL:
1878 	case SETVAL:
1879 	case SETALL:
1880 		error = copyin(uap->arg, &arg, sizeof(arg));
1881 		if (error)
1882 			return (error);
1883 		break;
1884 	}
1885 
1886 	switch (uap->cmd) {
1887 	case SEM_STAT:
1888 	case IPC_STAT:
1889 		semun.buf = &dsbuf;
1890 		break;
1891 	case IPC_SET:
1892 		error = copyin(PTRIN(arg.buf), &dsbuf32, sizeof(dsbuf32));
1893 		if (error)
1894 			return (error);
1895 		freebsd32_ipcperm_old_in(&dsbuf32.sem_perm, &dsbuf.sem_perm);
1896 		PTRIN_CP(dsbuf32, dsbuf, __sem_base);
1897 		CP(dsbuf32, dsbuf, sem_nsems);
1898 		CP(dsbuf32, dsbuf, sem_otime);
1899 		CP(dsbuf32, dsbuf, sem_ctime);
1900 		semun.buf = &dsbuf;
1901 		break;
1902 	case GETALL:
1903 	case SETALL:
1904 		semun.array = PTRIN(arg.array);
1905 		break;
1906 	case SETVAL:
1907 		semun.val = arg.val;
1908 		break;
1909 	}
1910 
1911 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1912 	    &rval);
1913 	if (error)
1914 		return (error);
1915 
1916 	switch (uap->cmd) {
1917 	case SEM_STAT:
1918 	case IPC_STAT:
1919 		bzero(&dsbuf32, sizeof(dsbuf32));
1920 		freebsd32_ipcperm_old_out(&dsbuf.sem_perm, &dsbuf32.sem_perm);
1921 		PTROUT_CP(dsbuf, dsbuf32, __sem_base);
1922 		CP(dsbuf, dsbuf32, sem_nsems);
1923 		CP(dsbuf, dsbuf32, sem_otime);
1924 		CP(dsbuf, dsbuf32, sem_ctime);
1925 		error = copyout(&dsbuf32, PTRIN(arg.buf), sizeof(dsbuf32));
1926 		break;
1927 	}
1928 
1929 	if (error == 0)
1930 		td->td_retval[0] = rval;
1931 	return (error);
1932 }
1933 #endif
1934 
1935 int
1936 freebsd32_semctl(struct thread *td, struct freebsd32_semctl_args *uap)
1937 {
1938 	struct semid_ds32 dsbuf32;
1939 	struct semid_ds dsbuf;
1940 	union semun semun;
1941 	union semun32 arg;
1942 	register_t rval;
1943 	int error;
1944 
1945 	switch (uap->cmd) {
1946 	case SEM_STAT:
1947 	case IPC_SET:
1948 	case IPC_STAT:
1949 	case GETALL:
1950 	case SETVAL:
1951 	case SETALL:
1952 		error = copyin(uap->arg, &arg, sizeof(arg));
1953 		if (error)
1954 			return (error);
1955 		break;
1956 	}
1957 
1958 	switch (uap->cmd) {
1959 	case SEM_STAT:
1960 	case IPC_STAT:
1961 		semun.buf = &dsbuf;
1962 		break;
1963 	case IPC_SET:
1964 		error = copyin(PTRIN(arg.buf), &dsbuf32, sizeof(dsbuf32));
1965 		if (error)
1966 			return (error);
1967 		freebsd32_ipcperm_in(&dsbuf32.sem_perm, &dsbuf.sem_perm);
1968 		PTRIN_CP(dsbuf32, dsbuf, __sem_base);
1969 		CP(dsbuf32, dsbuf, sem_nsems);
1970 		CP(dsbuf32, dsbuf, sem_otime);
1971 		CP(dsbuf32, dsbuf, sem_ctime);
1972 		semun.buf = &dsbuf;
1973 		break;
1974 	case GETALL:
1975 	case SETALL:
1976 		semun.array = PTRIN(arg.array);
1977 		break;
1978 	case SETVAL:
1979 		semun.val = arg.val;
1980 		break;
1981 	}
1982 
1983 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1984 	    &rval);
1985 	if (error)
1986 		return (error);
1987 
1988 	switch (uap->cmd) {
1989 	case SEM_STAT:
1990 	case IPC_STAT:
1991 		bzero(&dsbuf32, sizeof(dsbuf32));
1992 		freebsd32_ipcperm_out(&dsbuf.sem_perm, &dsbuf32.sem_perm);
1993 		PTROUT_CP(dsbuf, dsbuf32, __sem_base);
1994 		CP(dsbuf, dsbuf32, sem_nsems);
1995 		CP(dsbuf, dsbuf32, sem_otime);
1996 		CP(dsbuf, dsbuf32, sem_ctime);
1997 		error = copyout(&dsbuf32, PTRIN(arg.buf), sizeof(dsbuf32));
1998 		break;
1999 	}
2000 
2001 	if (error == 0)
2002 		td->td_retval[0] = rval;
2003 	return (error);
2004 }
2005 
2006 #endif /* COMPAT_FREEBSD32 */
2007