xref: /freebsd/sys/kern/sysv_msg.c (revision a970610a3af63b3f4df5b69d91c6b4093a00ed8f)
1 /*-
2  * Implementation of SVID messages
3  *
4  * Author:  Daniel Boulet
5  *
6  * Copyright 1993 Daniel Boulet and RTMX Inc.
7  *
8  * This system call was implemented by Daniel Boulet under contract from RTMX.
9  *
10  * Redistribution and use in source forms, with and without modification,
11  * are permitted provided that this entire comment appears intact.
12  *
13  * Redistribution in binary form may occur without any restrictions.
14  * Obviously, it would be nice if you gave credit where credit is due
15  * but requiring it would be too onerous.
16  *
17  * This software is provided ``AS IS'' without any warranties of any kind.
18  */
19 /*-
20  * SPDX-License-Identifier: BSD-2-Clause
21  *
22  * Copyright (c) 2003-2005 McAfee, Inc.
23  * Copyright (c) 2016-2017 Robert N. M. Watson
24  * All rights reserved.
25  *
26  * This software was developed for the FreeBSD Project in part by McAfee
27  * Research, the Security Research Division of McAfee, Inc under DARPA/SPAWAR
28  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS research
29  * program.
30  *
31  * Portions of this software were developed by BAE Systems, the University of
32  * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL
33  * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent
34  * Computing (TC) research program.
35  *
36  * Redistribution and use in source and binary forms, with or without
37  * modification, are permitted provided that the following conditions
38  * are met:
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  *
45  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
46  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
47  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
48  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
49  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
50  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
51  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
52  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
53  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
54  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55  * SUCH DAMAGE.
56  */
57 
58 #include <sys/cdefs.h>
59 #include "opt_sysvipc.h"
60 
61 #include <sys/param.h>
62 #include <sys/systm.h>
63 #include <sys/sysproto.h>
64 #include <sys/kernel.h>
65 #include <sys/priv.h>
66 #include <sys/proc.h>
67 #include <sys/lock.h>
68 #include <sys/mutex.h>
69 #include <sys/module.h>
70 #include <sys/mount.h>
71 #include <sys/msg.h>
72 #include <sys/racct.h>
73 #include <sys/sx.h>
74 #include <sys/syscall.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysent.h>
77 #include <sys/sysctl.h>
78 #include <sys/malloc.h>
79 #include <sys/jail.h>
80 
81 #include <security/audit/audit.h>
82 #include <security/mac/mac_framework.h>
83 
84 FEATURE(sysv_msg, "System V message queues support");
85 
86 static MALLOC_DEFINE(M_MSG, "msg", "SVID compatible message queues");
87 
88 static int msginit(void);
89 static int msgunload(void);
90 static int sysvmsg_modload(struct module *, int, void *);
91 static void msq_remove(struct msqid_kernel *);
92 static struct prison *msg_find_prison(struct ucred *);
93 static int msq_prison_cansee(struct prison *, struct msqid_kernel *);
94 static int msg_prison_check(void *, void *);
95 static int msg_prison_set(void *, void *);
96 static int msg_prison_get(void *, void *);
97 static int msg_prison_remove(void *, void *);
98 static void msg_prison_cleanup(struct prison *);
99 
100 #ifdef MSG_DEBUG
101 #define DPRINTF(a)	printf a
102 #else
103 #define DPRINTF(a)	(void)0
104 #endif
105 
106 static void msg_freehdr(struct msg *msghdr);
107 
108 #ifndef MSGSSZ
109 #define MSGSSZ	8		/* Each segment must be 2^N long */
110 #endif
111 #ifndef MSGSEG
112 #define MSGSEG	2048		/* must be less than 32767 */
113 #endif
114 #define MSGMAX	(MSGSSZ*MSGSEG)
115 #ifndef MSGMNB
116 #define MSGMNB	2048		/* max # of bytes in a queue */
117 #endif
118 #ifndef MSGMNI
119 #define MSGMNI	40
120 #endif
121 #ifndef MSGTQL
122 #define MSGTQL	40
123 #endif
124 
125 /*
126  * Based on the configuration parameters described in an SVR2 (yes, two)
127  * config(1m) man page.
128  *
129  * Each message is broken up and stored in segments that are msgssz bytes
130  * long.  For efficiency reasons, this should be a power of two.  Also,
131  * it doesn't make sense if it is less than 8 or greater than about 256.
132  * Consequently, msginit in kern/sysv_msg.c checks that msgssz is a power of
133  * two between 8 and 1024 inclusive (and panic's if it isn't).
134  */
135 struct msginfo msginfo = {
136                 MSGMAX,         /* max chars in a message */
137                 MSGMNI,         /* # of message queue identifiers */
138                 MSGMNB,         /* max chars in a queue */
139                 MSGTQL,         /* max messages in system */
140                 MSGSSZ,         /* size of a message segment */
141                 		/* (must be small power of 2 greater than 4) */
142                 MSGSEG          /* number of message segments */
143 };
144 
145 /*
146  * macros to convert between msqid_ds's and msqid's.
147  * (specific to this implementation)
148  */
149 #define MSQID(ix,ds)	((ix) & 0xffff | (((ds).msg_perm.seq << 16) & 0xffff0000))
150 #define MSQID_IX(id)	((id) & 0xffff)
151 #define MSQID_SEQ(id)	(((id) >> 16) & 0xffff)
152 
153 /*
154  * The rest of this file is specific to this particular implementation.
155  */
156 
157 struct msgmap {
158 	short	next;		/* next segment in buffer */
159     				/* -1 -> available */
160     				/* 0..(MSGSEG-1) -> index of next segment */
161 };
162 
163 #define MSG_LOCKED	01000	/* Is this msqid_ds locked? */
164 
165 static int nfree_msgmaps;	/* # of free map entries */
166 static short free_msgmaps;	/* head of linked list of free map entries */
167 static struct msg *free_msghdrs;/* list of free msg headers */
168 static char *msgpool;		/* MSGMAX byte long msg buffer pool */
169 static struct msgmap *msgmaps;	/* MSGSEG msgmap structures */
170 static struct msg *msghdrs;	/* MSGTQL msg headers */
171 static struct msqid_kernel *msqids;	/* MSGMNI msqid_kernel struct's */
172 static struct mtx msq_mtx;	/* global mutex for message queues. */
173 static unsigned msg_prison_slot;/* prison OSD slot */
174 
175 static struct syscall_helper_data msg_syscalls[] = {
176 	SYSCALL_INIT_HELPER(msgctl),
177 	SYSCALL_INIT_HELPER(msgget),
178 	SYSCALL_INIT_HELPER(msgsnd),
179 	SYSCALL_INIT_HELPER(msgrcv),
180 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
181     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
182 	SYSCALL_INIT_HELPER(msgsys),
183 	SYSCALL_INIT_HELPER_COMPAT(freebsd7_msgctl),
184 #endif
185 	SYSCALL_INIT_LAST
186 };
187 
188 #ifdef COMPAT_FREEBSD32
189 #include <compat/freebsd32/freebsd32.h>
190 #include <compat/freebsd32/freebsd32_ipc.h>
191 #include <compat/freebsd32/freebsd32_proto.h>
192 #include <compat/freebsd32/freebsd32_signal.h>
193 #include <compat/freebsd32/freebsd32_syscall.h>
194 #include <compat/freebsd32/freebsd32_util.h>
195 
196 static struct syscall_helper_data msg32_syscalls[] = {
197 	SYSCALL32_INIT_HELPER(freebsd32_msgctl),
198 	SYSCALL32_INIT_HELPER(freebsd32_msgsnd),
199 	SYSCALL32_INIT_HELPER(freebsd32_msgrcv),
200 	SYSCALL32_INIT_HELPER_COMPAT(msgget),
201 	SYSCALL32_INIT_HELPER(freebsd32_msgsys),
202 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
203     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
204 	SYSCALL32_INIT_HELPER(freebsd7_freebsd32_msgctl),
205 #endif
206 	SYSCALL_INIT_LAST
207 };
208 #endif
209 
210 static int
211 msginit(void)
212 {
213 	struct prison *pr;
214 	void **rsv;
215 	int i, error;
216 	osd_method_t methods[PR_MAXMETHOD] = {
217 	    [PR_METHOD_CHECK] =		msg_prison_check,
218 	    [PR_METHOD_SET] =		msg_prison_set,
219 	    [PR_METHOD_GET] =		msg_prison_get,
220 	    [PR_METHOD_REMOVE] =	msg_prison_remove,
221 	};
222 
223 	msginfo.msgmax = msginfo.msgseg * msginfo.msgssz;
224 	msgpool = malloc(msginfo.msgmax, M_MSG, M_WAITOK);
225 	msgmaps = malloc(sizeof(struct msgmap) * msginfo.msgseg, M_MSG, M_WAITOK);
226 	msghdrs = malloc(sizeof(struct msg) * msginfo.msgtql, M_MSG, M_WAITOK);
227 	msqids = malloc(sizeof(struct msqid_kernel) * msginfo.msgmni, M_MSG,
228 	    M_WAITOK | M_ZERO);
229 
230 	/*
231 	 * msginfo.msgssz should be a power of two for efficiency reasons.
232 	 * It is also pretty silly if msginfo.msgssz is less than 8
233 	 * or greater than about 256 so ...
234 	 */
235 
236 	i = 8;
237 	while (i < 1024 && i != msginfo.msgssz)
238 		i <<= 1;
239     	if (i != msginfo.msgssz) {
240 		DPRINTF(("msginfo.msgssz=%d (0x%x)\n", msginfo.msgssz,
241 		    msginfo.msgssz));
242 		panic("msginfo.msgssz not a small power of 2");
243 	}
244 
245 	if (msginfo.msgseg > 32767) {
246 		DPRINTF(("msginfo.msgseg=%d\n", msginfo.msgseg));
247 		panic("msginfo.msgseg > 32767");
248 	}
249 
250 	for (i = 0; i < msginfo.msgseg; i++) {
251 		if (i > 0)
252 			msgmaps[i-1].next = i;
253 		msgmaps[i].next = -1;	/* implies entry is available */
254 	}
255 	free_msgmaps = 0;
256 	nfree_msgmaps = msginfo.msgseg;
257 
258 	for (i = 0; i < msginfo.msgtql; i++) {
259 		msghdrs[i].msg_type = 0;
260 		if (i > 0)
261 			msghdrs[i-1].msg_next = &msghdrs[i];
262 		msghdrs[i].msg_next = NULL;
263 #ifdef MAC
264 		mac_sysvmsg_init(&msghdrs[i]);
265 #endif
266     	}
267 	free_msghdrs = &msghdrs[0];
268 
269 	for (i = 0; i < msginfo.msgmni; i++) {
270 		msqids[i].u.msg_qbytes = 0;	/* implies entry is available */
271 		msqids[i].u.msg_perm.seq = 0;	/* reset to a known value */
272 		msqids[i].u.msg_perm.mode = 0;
273 #ifdef MAC
274 		mac_sysvmsq_init(&msqids[i]);
275 #endif
276 	}
277 	mtx_init(&msq_mtx, "msq", NULL, MTX_DEF);
278 
279 	/* Set current prisons according to their allow.sysvipc. */
280 	msg_prison_slot = osd_jail_register(NULL, methods);
281 	rsv = osd_reserve(msg_prison_slot);
282 	prison_lock(&prison0);
283 	(void)osd_jail_set_reserved(&prison0, msg_prison_slot, rsv, &prison0);
284 	prison_unlock(&prison0);
285 	rsv = NULL;
286 	sx_slock(&allprison_lock);
287 	TAILQ_FOREACH(pr, &allprison, pr_list) {
288 		if (rsv == NULL)
289 			rsv = osd_reserve(msg_prison_slot);
290 		prison_lock(pr);
291 		if (pr->pr_allow & PR_ALLOW_SYSVIPC) {
292 			(void)osd_jail_set_reserved(pr, msg_prison_slot, rsv,
293 			    &prison0);
294 			rsv = NULL;
295 		}
296 		prison_unlock(pr);
297 	}
298 	if (rsv != NULL)
299 		osd_free_reserved(rsv);
300 	sx_sunlock(&allprison_lock);
301 
302 	error = syscall_helper_register(msg_syscalls, SY_THR_STATIC_KLD);
303 	if (error != 0)
304 		return (error);
305 #ifdef COMPAT_FREEBSD32
306 	error = syscall32_helper_register(msg32_syscalls, SY_THR_STATIC_KLD);
307 	if (error != 0)
308 		return (error);
309 #endif
310 	return (0);
311 }
312 
313 static int
314 msgunload(void)
315 {
316 	struct msqid_kernel *msqkptr;
317 	int msqid;
318 #ifdef MAC
319 	int i;
320 #endif
321 
322 	syscall_helper_unregister(msg_syscalls);
323 #ifdef COMPAT_FREEBSD32
324 	syscall32_helper_unregister(msg32_syscalls);
325 #endif
326 
327 	for (msqid = 0; msqid < msginfo.msgmni; msqid++) {
328 		msqkptr = &msqids[msqid];
329 		if (msqkptr->u.msg_qbytes != 0 ||
330 		    (msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0)
331 			break;
332 	}
333 	if (msqid != msginfo.msgmni)
334 		return (EBUSY);
335 
336 	if (msg_prison_slot != 0)
337 		osd_jail_deregister(msg_prison_slot);
338 #ifdef MAC
339 	for (i = 0; i < msginfo.msgtql; i++)
340 		mac_sysvmsg_destroy(&msghdrs[i]);
341 	for (msqid = 0; msqid < msginfo.msgmni; msqid++)
342 		mac_sysvmsq_destroy(&msqids[msqid]);
343 #endif
344 	free(msgpool, M_MSG);
345 	free(msgmaps, M_MSG);
346 	free(msghdrs, M_MSG);
347 	free(msqids, M_MSG);
348 	mtx_destroy(&msq_mtx);
349 	return (0);
350 }
351 
352 static int
353 sysvmsg_modload(struct module *module, int cmd, void *arg)
354 {
355 	int error = 0;
356 
357 	switch (cmd) {
358 	case MOD_LOAD:
359 		error = msginit();
360 		if (error != 0)
361 			msgunload();
362 		break;
363 	case MOD_UNLOAD:
364 		error = msgunload();
365 		break;
366 	case MOD_SHUTDOWN:
367 		break;
368 	default:
369 		error = EINVAL;
370 		break;
371 	}
372 	return (error);
373 }
374 
375 static moduledata_t sysvmsg_mod = {
376 	"sysvmsg",
377 	&sysvmsg_modload,
378 	NULL
379 };
380 
381 DECLARE_MODULE(sysvmsg, sysvmsg_mod, SI_SUB_SYSV_MSG, SI_ORDER_FIRST);
382 MODULE_VERSION(sysvmsg, 1);
383 
384 static void
385 msg_freehdr(struct msg *msghdr)
386 {
387 	while (msghdr->msg_ts > 0) {
388 		short next;
389 		if (msghdr->msg_spot < 0 || msghdr->msg_spot >= msginfo.msgseg)
390 			panic("msghdr->msg_spot out of range");
391 		next = msgmaps[msghdr->msg_spot].next;
392 		msgmaps[msghdr->msg_spot].next = free_msgmaps;
393 		free_msgmaps = msghdr->msg_spot;
394 		nfree_msgmaps++;
395 		msghdr->msg_spot = next;
396 		if (msghdr->msg_ts >= msginfo.msgssz)
397 			msghdr->msg_ts -= msginfo.msgssz;
398 		else
399 			msghdr->msg_ts = 0;
400 	}
401 	if (msghdr->msg_spot != -1)
402 		panic("msghdr->msg_spot != -1");
403 	msghdr->msg_next = free_msghdrs;
404 	free_msghdrs = msghdr;
405 #ifdef MAC
406 	mac_sysvmsg_cleanup(msghdr);
407 #endif
408 }
409 
410 static void
411 msq_remove(struct msqid_kernel *msqkptr)
412 {
413 	struct msg *msghdr;
414 
415 	racct_sub_cred(msqkptr->cred, RACCT_NMSGQ, 1);
416 	racct_sub_cred(msqkptr->cred, RACCT_MSGQQUEUED, msqkptr->u.msg_qnum);
417 	racct_sub_cred(msqkptr->cred, RACCT_MSGQSIZE, msqkptr->u.msg_cbytes);
418 	crfree(msqkptr->cred);
419 	msqkptr->cred = NULL;
420 
421 	/* Free the message headers */
422 	msghdr = msqkptr->u.__msg_first;
423 	while (msghdr != NULL) {
424 		struct msg *msghdr_tmp;
425 
426 		/* Free the segments of each message */
427 		msqkptr->u.msg_cbytes -= msghdr->msg_ts;
428 		msqkptr->u.msg_qnum--;
429 		msghdr_tmp = msghdr;
430 		msghdr = msghdr->msg_next;
431 		msg_freehdr(msghdr_tmp);
432 	}
433 
434 	if (msqkptr->u.msg_cbytes != 0)
435 		panic("msg_cbytes is screwed up");
436 	if (msqkptr->u.msg_qnum != 0)
437 		panic("msg_qnum is screwed up");
438 
439 	msqkptr->u.msg_qbytes = 0;	/* Mark it as free */
440 
441 #ifdef MAC
442 	mac_sysvmsq_cleanup(msqkptr);
443 #endif
444 
445 	wakeup(msqkptr);
446 }
447 
448 static struct prison *
449 msg_find_prison(struct ucred *cred)
450 {
451 	struct prison *pr, *rpr;
452 
453 	pr = cred->cr_prison;
454 	prison_lock(pr);
455 	rpr = osd_jail_get(pr, msg_prison_slot);
456 	prison_unlock(pr);
457 	return rpr;
458 }
459 
460 static int
461 msq_prison_cansee(struct prison *rpr, struct msqid_kernel *msqkptr)
462 {
463 
464 	if (msqkptr->cred == NULL ||
465 	    !(rpr == msqkptr->cred->cr_prison ||
466 	      prison_ischild(rpr, msqkptr->cred->cr_prison)))
467 		return (EINVAL);
468 	return (0);
469 }
470 
471 #ifndef _SYS_SYSPROTO_H_
472 struct msgctl_args {
473 	int	msqid;
474 	int	cmd;
475 	struct	msqid_ds *buf;
476 };
477 #endif
478 int
479 sys_msgctl(struct thread *td, struct msgctl_args *uap)
480 {
481 	int msqid = uap->msqid;
482 	int cmd = uap->cmd;
483 	struct msqid_ds msqbuf;
484 	int error;
485 
486 	DPRINTF(("call to msgctl(%d, %d, %p)\n", msqid, cmd, uap->buf));
487 	if (cmd == IPC_SET &&
488 	    (error = copyin(uap->buf, &msqbuf, sizeof(msqbuf))) != 0)
489 		return (error);
490 	error = kern_msgctl(td, msqid, cmd, &msqbuf);
491 	if (cmd == IPC_STAT && error == 0)
492 		error = copyout(&msqbuf, uap->buf, sizeof(struct msqid_ds));
493 	return (error);
494 }
495 
496 int
497 kern_msgctl(struct thread *td, int msqid, int cmd, struct msqid_ds *msqbuf)
498 {
499 	int rval, error, msqix;
500 	struct msqid_kernel *msqkptr;
501 	struct prison *rpr;
502 
503 	rpr = msg_find_prison(td->td_ucred);
504 	if (rpr == NULL)
505 		return (ENOSYS);
506 
507 	AUDIT_ARG_SVIPC_CMD(cmd);
508 	AUDIT_ARG_SVIPC_ID(msqid);
509 	msqix = IPCID_TO_IX(msqid);
510 
511 	if (msqix < 0 || msqix >= msginfo.msgmni) {
512 		DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix,
513 		    msginfo.msgmni));
514 		return (EINVAL);
515 	}
516 
517 	msqkptr = &msqids[msqix];
518 
519 	mtx_lock(&msq_mtx);
520 	if (msqkptr->u.msg_qbytes == 0) {
521 		DPRINTF(("no such msqid\n"));
522 		error = EINVAL;
523 		goto done2;
524 	}
525 	if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) {
526 		DPRINTF(("wrong sequence number\n"));
527 		error = EINVAL;
528 		goto done2;
529 	}
530 
531 	error = msq_prison_cansee(rpr, msqkptr);
532 	if (error != 0) {
533 		DPRINTF(("requester can't see prison\n"));
534 		goto done2;
535 	}
536 
537 #ifdef MAC
538 	error = mac_sysvmsq_check_msqctl(td->td_ucred, msqkptr, cmd);
539 	if (error != 0)
540 		goto done2;
541 #endif
542 
543 	error = 0;
544 	rval = 0;
545 
546 	switch (cmd) {
547 	case IPC_RMID:
548 	{
549 #ifdef MAC
550 		struct msg *msghdr;
551 #endif
552 		if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M)))
553 			goto done2;
554 
555 #ifdef MAC
556 		/*
557 		 * Check that the thread has MAC access permissions to
558 		 * individual msghdrs.  Note: We need to do this in a
559 		 * separate loop because the actual loop alters the
560 		 * msq/msghdr info as it progresses, and there is no going
561 		 * back if half the way through we discover that the
562 		 * thread cannot free a certain msghdr.  The msq will get
563 		 * into an inconsistent state.
564 		 */
565 		for (msghdr = msqkptr->u.__msg_first; msghdr != NULL;
566 		    msghdr = msghdr->msg_next) {
567 			error = mac_sysvmsq_check_msgrmid(td->td_ucred, msghdr);
568 			if (error != 0)
569 				goto done2;
570 		}
571 #endif
572 
573 		msq_remove(msqkptr);
574 	}
575 
576 		break;
577 
578 	case IPC_SET:
579 		AUDIT_ARG_SVIPC_PERM(&msqbuf->msg_perm);
580 		if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M)))
581 			goto done2;
582 		if (msqbuf->msg_qbytes > msqkptr->u.msg_qbytes) {
583 			error = priv_check(td, PRIV_IPC_MSGSIZE);
584 			if (error)
585 				goto done2;
586 		}
587 		if (msqbuf->msg_qbytes > msginfo.msgmnb) {
588 			DPRINTF(("can't increase msg_qbytes beyond %d"
589 			    "(truncating)\n", msginfo.msgmnb));
590 			msqbuf->msg_qbytes = msginfo.msgmnb;	/* silently restrict qbytes to system limit */
591 		}
592 		if (msqbuf->msg_qbytes == 0) {
593 			DPRINTF(("can't reduce msg_qbytes to 0\n"));
594 			error = EINVAL;		/* non-standard errno! */
595 			goto done2;
596 		}
597 		msqkptr->u.msg_perm.uid = msqbuf->msg_perm.uid;	/* change the owner */
598 		msqkptr->u.msg_perm.gid = msqbuf->msg_perm.gid;	/* change the owner */
599 		msqkptr->u.msg_perm.mode = (msqkptr->u.msg_perm.mode & ~0777) |
600 		    (msqbuf->msg_perm.mode & 0777);
601 		msqkptr->u.msg_qbytes = msqbuf->msg_qbytes;
602 		msqkptr->u.msg_ctime = time_second;
603 		break;
604 
605 	case IPC_STAT:
606 		if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) {
607 			DPRINTF(("requester doesn't have read access\n"));
608 			goto done2;
609 		}
610 		*msqbuf = msqkptr->u;
611 		if (td->td_ucred->cr_prison != msqkptr->cred->cr_prison)
612 			msqbuf->msg_perm.key = IPC_PRIVATE;
613 
614 		/*
615 		 * Try to hide the fact that the structure layout is shared by
616 		 * both the kernel and userland.  These pointers are not useful
617 		 * to userspace.
618 		 */
619 		msqbuf->__msg_first = msqbuf->__msg_last = NULL;
620 		break;
621 
622 	default:
623 		DPRINTF(("invalid command %d\n", cmd));
624 		error = EINVAL;
625 		goto done2;
626 	}
627 
628 	if (error == 0)
629 		td->td_retval[0] = rval;
630 done2:
631 	mtx_unlock(&msq_mtx);
632 	return (error);
633 }
634 
635 #ifndef _SYS_SYSPROTO_H_
636 struct msgget_args {
637 	key_t	key;
638 	int	msgflg;
639 };
640 #endif
641 
642 int
643 sys_msgget(struct thread *td, struct msgget_args *uap)
644 {
645 	int msqid, error = 0;
646 	int key = uap->key;
647 	int msgflg = uap->msgflg;
648 	struct ucred *cred = td->td_ucred;
649 	struct msqid_kernel *msqkptr = NULL;
650 
651 	DPRINTF(("msgget(0x%x, 0%o)\n", key, msgflg));
652 
653 	if (msg_find_prison(cred) == NULL)
654 		return (ENOSYS);
655 
656 	mtx_lock(&msq_mtx);
657 	if (key != IPC_PRIVATE) {
658 		for (msqid = 0; msqid < msginfo.msgmni; msqid++) {
659 			msqkptr = &msqids[msqid];
660 			if (msqkptr->u.msg_qbytes != 0 &&
661 			    msqkptr->cred != NULL &&
662 			    msqkptr->cred->cr_prison == cred->cr_prison &&
663 			    msqkptr->u.msg_perm.key == key)
664 				break;
665 		}
666 		if (msqid < msginfo.msgmni) {
667 			DPRINTF(("found public key\n"));
668 			if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) {
669 				DPRINTF(("not exclusive\n"));
670 				error = EEXIST;
671 				goto done2;
672 			}
673 			AUDIT_ARG_SVIPC_ID(IXSEQ_TO_IPCID(msqid,
674 			    msqkptr->u.msg_perm));
675 			if ((error = ipcperm(td, &msqkptr->u.msg_perm,
676 			    msgflg & 0700))) {
677 				DPRINTF(("requester doesn't have 0%o access\n",
678 				    msgflg & 0700));
679 				goto done2;
680 			}
681 #ifdef MAC
682 			error = mac_sysvmsq_check_msqget(cred, msqkptr);
683 			if (error != 0)
684 				goto done2;
685 #endif
686 			goto found;
687 		}
688 	}
689 
690 	DPRINTF(("need to allocate the msqid_ds\n"));
691 	if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) {
692 		for (msqid = 0; msqid < msginfo.msgmni; msqid++) {
693 			/*
694 			 * Look for an unallocated and unlocked msqid_ds.
695 			 * msqid_ds's can be locked by msgsnd or msgrcv while
696 			 * they are copying the message in/out.  We can't
697 			 * re-use the entry until they release it.
698 			 */
699 			msqkptr = &msqids[msqid];
700 			if (msqkptr->u.msg_qbytes == 0 &&
701 			    (msqkptr->u.msg_perm.mode & MSG_LOCKED) == 0)
702 				break;
703 		}
704 		if (msqid == msginfo.msgmni) {
705 			DPRINTF(("no more msqid_ds's available\n"));
706 			error = ENOSPC;
707 			goto done2;
708 		}
709 #ifdef RACCT
710 		if (racct_enable) {
711 			PROC_LOCK(td->td_proc);
712 			error = racct_add(td->td_proc, RACCT_NMSGQ, 1);
713 			PROC_UNLOCK(td->td_proc);
714 			if (error != 0) {
715 				error = ENOSPC;
716 				goto done2;
717 			}
718 		}
719 #endif
720 		DPRINTF(("msqid %d is available\n", msqid));
721 		msqkptr->u.msg_perm.key = key;
722 		msqkptr->u.msg_perm.cuid = cred->cr_uid;
723 		msqkptr->u.msg_perm.uid = cred->cr_uid;
724 		msqkptr->u.msg_perm.cgid = cred->cr_gid;
725 		msqkptr->u.msg_perm.gid = cred->cr_gid;
726 		msqkptr->u.msg_perm.mode = (msgflg & 0777);
727 		msqkptr->cred = crhold(cred);
728 		/* Make sure that the returned msqid is unique */
729 		msqkptr->u.msg_perm.seq = (msqkptr->u.msg_perm.seq + 1) & 0x7fff;
730 		msqkptr->u.__msg_first = NULL;
731 		msqkptr->u.__msg_last = NULL;
732 		msqkptr->u.msg_cbytes = 0;
733 		msqkptr->u.msg_qnum = 0;
734 		msqkptr->u.msg_qbytes = msginfo.msgmnb;
735 		msqkptr->u.msg_lspid = 0;
736 		msqkptr->u.msg_lrpid = 0;
737 		msqkptr->u.msg_stime = 0;
738 		msqkptr->u.msg_rtime = 0;
739 		msqkptr->u.msg_ctime = time_second;
740 #ifdef MAC
741 		mac_sysvmsq_create(cred, msqkptr);
742 #endif
743 		AUDIT_ARG_SVIPC_PERM(&msqkptr->u.msg_perm);
744 	} else {
745 		DPRINTF(("didn't find it and wasn't asked to create it\n"));
746 		error = ENOENT;
747 		goto done2;
748 	}
749 
750 found:
751 	/* Construct the unique msqid */
752 	td->td_retval[0] = IXSEQ_TO_IPCID(msqid, msqkptr->u.msg_perm);
753 done2:
754 	mtx_unlock(&msq_mtx);
755 	return (error);
756 }
757 
758 #ifndef _SYS_SYSPROTO_H_
759 struct msgsnd_args {
760 	int	msqid;
761 	const void	*msgp;	/* XXX msgp is actually mtext. */
762 	size_t	msgsz;
763 	int	msgflg;
764 };
765 #endif
766 int
767 kern_msgsnd(struct thread *td, int msqid, const void *msgp,
768     size_t msgsz, int msgflg, long mtype)
769 {
770 	int msqix, segs_needed, error = 0;
771 	struct msqid_kernel *msqkptr;
772 	struct msg *msghdr;
773 	struct prison *rpr;
774 	short next;
775 #ifdef RACCT
776 	size_t saved_msgsz = 0;
777 #endif
778 
779 	rpr = msg_find_prison(td->td_ucred);
780 	if (rpr == NULL)
781 		return (ENOSYS);
782 
783 	mtx_lock(&msq_mtx);
784 	AUDIT_ARG_SVIPC_ID(msqid);
785 	msqix = IPCID_TO_IX(msqid);
786 
787 	if (msqix < 0 || msqix >= msginfo.msgmni) {
788 		DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix,
789 		    msginfo.msgmni));
790 		error = EINVAL;
791 		goto done2;
792 	}
793 
794 	msqkptr = &msqids[msqix];
795 	AUDIT_ARG_SVIPC_PERM(&msqkptr->u.msg_perm);
796 	if (msqkptr->u.msg_qbytes == 0) {
797 		DPRINTF(("no such message queue id\n"));
798 		error = EINVAL;
799 		goto done2;
800 	}
801 	if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) {
802 		DPRINTF(("wrong sequence number\n"));
803 		error = EINVAL;
804 		goto done2;
805 	}
806 
807 	if ((error = msq_prison_cansee(rpr, msqkptr))) {
808 		DPRINTF(("requester can't see prison\n"));
809 		goto done2;
810 	}
811 
812 	if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_W))) {
813 		DPRINTF(("requester doesn't have write access\n"));
814 		goto done2;
815 	}
816 
817 #ifdef MAC
818 	error = mac_sysvmsq_check_msqsnd(td->td_ucred, msqkptr);
819 	if (error != 0)
820 		goto done2;
821 #endif
822 
823 #ifdef RACCT
824 	if (racct_enable) {
825 		PROC_LOCK(td->td_proc);
826 		if (racct_add(td->td_proc, RACCT_MSGQQUEUED, 1)) {
827 			PROC_UNLOCK(td->td_proc);
828 			error = EAGAIN;
829 			goto done2;
830 		}
831 		saved_msgsz = msgsz;
832 		if (racct_add(td->td_proc, RACCT_MSGQSIZE, msgsz)) {
833 			racct_sub(td->td_proc, RACCT_MSGQQUEUED, 1);
834 			PROC_UNLOCK(td->td_proc);
835 			error = EAGAIN;
836 			goto done2;
837 		}
838 		PROC_UNLOCK(td->td_proc);
839 	}
840 #endif
841 
842 	segs_needed = howmany(msgsz, msginfo.msgssz);
843 	DPRINTF(("msgsz=%zu, msgssz=%d, segs_needed=%d\n", msgsz,
844 	    msginfo.msgssz, segs_needed));
845 	for (;;) {
846 		int need_more_resources = 0;
847 
848 		/*
849 		 * check msgsz
850 		 * (inside this loop in case msg_qbytes changes while we sleep)
851 		 */
852 
853 		if (msgsz > msqkptr->u.msg_qbytes) {
854 			DPRINTF(("msgsz > msqkptr->u.msg_qbytes\n"));
855 			error = EINVAL;
856 			goto done3;
857 		}
858 
859 		if (msqkptr->u.msg_perm.mode & MSG_LOCKED) {
860 			DPRINTF(("msqid is locked\n"));
861 			need_more_resources = 1;
862 		}
863 		if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) {
864 			DPRINTF(("msgsz + msg_cbytes > msg_qbytes\n"));
865 			need_more_resources = 1;
866 		}
867 		if (segs_needed > nfree_msgmaps) {
868 			DPRINTF(("segs_needed > nfree_msgmaps\n"));
869 			need_more_resources = 1;
870 		}
871 		if (free_msghdrs == NULL) {
872 			DPRINTF(("no more msghdrs\n"));
873 			need_more_resources = 1;
874 		}
875 
876 		if (need_more_resources) {
877 			int we_own_it;
878 
879 			if ((msgflg & IPC_NOWAIT) != 0) {
880 				DPRINTF(("need more resources but caller "
881 				    "doesn't want to wait\n"));
882 				error = EAGAIN;
883 				goto done3;
884 			}
885 
886 			if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) {
887 				DPRINTF(("we don't own the msqid_ds\n"));
888 				we_own_it = 0;
889 			} else {
890 				/* Force later arrivals to wait for our
891 				   request */
892 				DPRINTF(("we own the msqid_ds\n"));
893 				msqkptr->u.msg_perm.mode |= MSG_LOCKED;
894 				we_own_it = 1;
895 			}
896 			DPRINTF(("msgsnd:  goodnight\n"));
897 			error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH,
898 			    "msgsnd", hz);
899 			DPRINTF(("msgsnd:  good morning, error=%d\n", error));
900 			if (we_own_it)
901 				msqkptr->u.msg_perm.mode &= ~MSG_LOCKED;
902 			if (error == EWOULDBLOCK) {
903 				DPRINTF(("msgsnd:  timed out\n"));
904 				continue;
905 			}
906 			if (error != 0) {
907 				DPRINTF(("msgsnd:  interrupted system call\n"));
908 				error = EINTR;
909 				goto done3;
910 			}
911 
912 			/*
913 			 * Make sure that the msq queue still exists
914 			 */
915 
916 			if (msqkptr->u.msg_qbytes == 0) {
917 				DPRINTF(("msqid deleted\n"));
918 				error = EIDRM;
919 				goto done3;
920 			}
921 
922 		} else {
923 			DPRINTF(("got all the resources that we need\n"));
924 			break;
925 		}
926 	}
927 
928 	/*
929 	 * We have the resources that we need.
930 	 * Make sure!
931 	 */
932 
933 	if (msqkptr->u.msg_perm.mode & MSG_LOCKED)
934 		panic("msg_perm.mode & MSG_LOCKED");
935 	if (segs_needed > nfree_msgmaps)
936 		panic("segs_needed > nfree_msgmaps");
937 	if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes)
938 		panic("msgsz + msg_cbytes > msg_qbytes");
939 	if (free_msghdrs == NULL)
940 		panic("no more msghdrs");
941 
942 	/*
943 	 * Re-lock the msqid_ds in case we page-fault when copying in the
944 	 * message
945 	 */
946 
947 	if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0)
948 		panic("msqid_ds is already locked");
949 	msqkptr->u.msg_perm.mode |= MSG_LOCKED;
950 
951 	/*
952 	 * Allocate a message header
953 	 */
954 
955 	msghdr = free_msghdrs;
956 	free_msghdrs = msghdr->msg_next;
957 	msghdr->msg_spot = -1;
958 	msghdr->msg_ts = msgsz;
959 	msghdr->msg_type = mtype;
960 #ifdef MAC
961 	/*
962 	 * XXXMAC: Should the mac_sysvmsq_check_msgmsq check follow here
963 	 * immediately?  Or, should it be checked just before the msg is
964 	 * enqueued in the msgq (as it is done now)?
965 	 */
966 	mac_sysvmsg_create(td->td_ucred, msqkptr, msghdr);
967 #endif
968 
969 	/*
970 	 * Allocate space for the message
971 	 */
972 
973 	while (segs_needed > 0) {
974 		if (nfree_msgmaps <= 0)
975 			panic("not enough msgmaps");
976 		if (free_msgmaps == -1)
977 			panic("nil free_msgmaps");
978 		next = free_msgmaps;
979 		if (next <= -1)
980 			panic("next too low #1");
981 		if (next >= msginfo.msgseg)
982 			panic("next out of range #1");
983 		DPRINTF(("allocating segment %d to message\n", next));
984 		free_msgmaps = msgmaps[next].next;
985 		nfree_msgmaps--;
986 		msgmaps[next].next = msghdr->msg_spot;
987 		msghdr->msg_spot = next;
988 		segs_needed--;
989 	}
990 
991 	/*
992 	 * Validate the message type
993 	 */
994 
995 	if (msghdr->msg_type < 1) {
996 		msg_freehdr(msghdr);
997 		msqkptr->u.msg_perm.mode &= ~MSG_LOCKED;
998 		wakeup(msqkptr);
999 		DPRINTF(("mtype (%ld) < 1\n", msghdr->msg_type));
1000 		error = EINVAL;
1001 		goto done3;
1002 	}
1003 
1004 	/*
1005 	 * Copy in the message body
1006 	 */
1007 
1008 	next = msghdr->msg_spot;
1009 	while (msgsz > 0) {
1010 		size_t tlen;
1011 		if (msgsz > msginfo.msgssz)
1012 			tlen = msginfo.msgssz;
1013 		else
1014 			tlen = msgsz;
1015 		if (next <= -1)
1016 			panic("next too low #2");
1017 		if (next >= msginfo.msgseg)
1018 			panic("next out of range #2");
1019 		mtx_unlock(&msq_mtx);
1020 		if ((error = copyin(msgp, &msgpool[next * msginfo.msgssz],
1021 		    tlen)) != 0) {
1022 			mtx_lock(&msq_mtx);
1023 			DPRINTF(("error %d copying in message segment\n",
1024 			    error));
1025 			msg_freehdr(msghdr);
1026 			msqkptr->u.msg_perm.mode &= ~MSG_LOCKED;
1027 			wakeup(msqkptr);
1028 			goto done3;
1029 		}
1030 		mtx_lock(&msq_mtx);
1031 		msgsz -= tlen;
1032 		msgp = (const char *)msgp + tlen;
1033 		next = msgmaps[next].next;
1034 	}
1035 	if (next != -1)
1036 		panic("didn't use all the msg segments");
1037 
1038 	/*
1039 	 * We've got the message.  Unlock the msqid_ds.
1040 	 */
1041 
1042 	msqkptr->u.msg_perm.mode &= ~MSG_LOCKED;
1043 
1044 	/*
1045 	 * Make sure that the msqid_ds is still allocated.
1046 	 */
1047 
1048 	if (msqkptr->u.msg_qbytes == 0) {
1049 		msg_freehdr(msghdr);
1050 		wakeup(msqkptr);
1051 		error = EIDRM;
1052 		goto done3;
1053 	}
1054 
1055 #ifdef MAC
1056 	/*
1057 	 * Note: Since the task/thread allocates the msghdr and usually
1058 	 * primes it with its own MAC label, for a majority of policies, it
1059 	 * won't be necessary to check whether the msghdr has access
1060 	 * permissions to the msgq.  The mac_sysvmsq_check_msqsnd check would
1061 	 * suffice in that case.  However, this hook may be required where
1062 	 * individual policies derive a non-identical label for the msghdr
1063 	 * from the current thread label and may want to check the msghdr
1064 	 * enqueue permissions, along with read/write permissions to the
1065 	 * msgq.
1066 	 */
1067 	error = mac_sysvmsq_check_msgmsq(td->td_ucred, msghdr, msqkptr);
1068 	if (error != 0) {
1069 		msg_freehdr(msghdr);
1070 		wakeup(msqkptr);
1071 		goto done3;
1072 	}
1073 #endif
1074 
1075 	/*
1076 	 * Put the message into the queue
1077 	 */
1078 	if (msqkptr->u.__msg_first == NULL) {
1079 		msqkptr->u.__msg_first = msghdr;
1080 		msqkptr->u.__msg_last = msghdr;
1081 	} else {
1082 		msqkptr->u.__msg_last->msg_next = msghdr;
1083 		msqkptr->u.__msg_last = msghdr;
1084 	}
1085 	msqkptr->u.__msg_last->msg_next = NULL;
1086 
1087 	msqkptr->u.msg_cbytes += msghdr->msg_ts;
1088 	msqkptr->u.msg_qnum++;
1089 	msqkptr->u.msg_lspid = td->td_proc->p_pid;
1090 	msqkptr->u.msg_stime = time_second;
1091 
1092 	wakeup(msqkptr);
1093 	td->td_retval[0] = 0;
1094 done3:
1095 #ifdef RACCT
1096 	if (racct_enable && error != 0) {
1097 		PROC_LOCK(td->td_proc);
1098 		racct_sub(td->td_proc, RACCT_MSGQQUEUED, 1);
1099 		racct_sub(td->td_proc, RACCT_MSGQSIZE, saved_msgsz);
1100 		PROC_UNLOCK(td->td_proc);
1101 	}
1102 #endif
1103 done2:
1104 	mtx_unlock(&msq_mtx);
1105 	return (error);
1106 }
1107 
1108 int
1109 sys_msgsnd(struct thread *td, struct msgsnd_args *uap)
1110 {
1111 	int error;
1112 	long mtype;
1113 
1114 	DPRINTF(("call to msgsnd(%d, %p, %zu, %d)\n", uap->msqid, uap->msgp,
1115 	    uap->msgsz, uap->msgflg));
1116 
1117 	if ((error = copyin(uap->msgp, &mtype, sizeof(mtype))) != 0) {
1118 		DPRINTF(("error %d copying the message type\n", error));
1119 		return (error);
1120 	}
1121 	return (kern_msgsnd(td, uap->msqid,
1122 	    (const char *)uap->msgp + sizeof(mtype),
1123 	    uap->msgsz, uap->msgflg, mtype));
1124 }
1125 
1126 #ifndef _SYS_SYSPROTO_H_
1127 struct msgrcv_args {
1128 	int	msqid;
1129 	void	*msgp;
1130 	size_t	msgsz;
1131 	long	msgtyp;
1132 	int	msgflg;
1133 };
1134 #endif
1135 /* XXX msgp is actually mtext. */
1136 int
1137 kern_msgrcv(struct thread *td, int msqid, void *msgp, size_t msgsz, long msgtyp,
1138     int msgflg, long *mtype)
1139 {
1140 	size_t len;
1141 	struct msqid_kernel *msqkptr;
1142 	struct msg *msghdr;
1143 	struct prison *rpr;
1144 	int msqix, error = 0;
1145 	short next;
1146 
1147 	rpr = msg_find_prison(td->td_ucred);
1148 	if (rpr == NULL)
1149 		return (ENOSYS);
1150 
1151 	AUDIT_ARG_SVIPC_ID(msqid);
1152 	msqix = IPCID_TO_IX(msqid);
1153 
1154 	if (msqix < 0 || msqix >= msginfo.msgmni) {
1155 		DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix,
1156 		    msginfo.msgmni));
1157 		return (EINVAL);
1158 	}
1159 
1160 	msqkptr = &msqids[msqix];
1161 	mtx_lock(&msq_mtx);
1162 	AUDIT_ARG_SVIPC_PERM(&msqkptr->u.msg_perm);
1163 	if (msqkptr->u.msg_qbytes == 0) {
1164 		DPRINTF(("no such message queue id\n"));
1165 		error = EINVAL;
1166 		goto done2;
1167 	}
1168 	if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) {
1169 		DPRINTF(("wrong sequence number\n"));
1170 		error = EINVAL;
1171 		goto done2;
1172 	}
1173 
1174 	if ((error = msq_prison_cansee(rpr, msqkptr))) {
1175 		DPRINTF(("requester can't see prison\n"));
1176 		goto done2;
1177 	}
1178 
1179 	if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) {
1180 		DPRINTF(("requester doesn't have read access\n"));
1181 		goto done2;
1182 	}
1183 
1184 #ifdef MAC
1185 	error = mac_sysvmsq_check_msqrcv(td->td_ucred, msqkptr);
1186 	if (error != 0)
1187 		goto done2;
1188 #endif
1189 
1190 	msghdr = NULL;
1191 	while (msghdr == NULL) {
1192 		if (msgtyp == 0) {
1193 			msghdr = msqkptr->u.__msg_first;
1194 			if (msghdr != NULL) {
1195 				if (msgsz < msghdr->msg_ts &&
1196 				    (msgflg & MSG_NOERROR) == 0) {
1197 					DPRINTF(("first message on the queue "
1198 					    "is too big (want %zu, got %d)\n",
1199 					    msgsz, msghdr->msg_ts));
1200 					error = E2BIG;
1201 					goto done2;
1202 				}
1203 #ifdef MAC
1204 				error = mac_sysvmsq_check_msgrcv(td->td_ucred,
1205 				    msghdr);
1206 				if (error != 0)
1207 					goto done2;
1208 #endif
1209 				if (msqkptr->u.__msg_first ==
1210 				    msqkptr->u.__msg_last) {
1211 					msqkptr->u.__msg_first = NULL;
1212 					msqkptr->u.__msg_last = NULL;
1213 				} else {
1214 					msqkptr->u.__msg_first = msghdr->msg_next;
1215 					if (msqkptr->u.__msg_first == NULL)
1216 						panic("msg_first/last screwed up #1");
1217 				}
1218 			}
1219 		} else {
1220 			struct msg *previous;
1221 			struct msg **prev;
1222 
1223 			previous = NULL;
1224 			prev = &(msqkptr->u.__msg_first);
1225 			while ((msghdr = *prev) != NULL) {
1226 				/*
1227 				 * Is this message's type an exact match or is
1228 				 * this message's type less than or equal to
1229 				 * the absolute value of a negative msgtyp?
1230 				 * Note that the second half of this test can
1231 				 * NEVER be true if msgtyp is positive since
1232 				 * msg_type is always positive!
1233 				 */
1234 
1235 				if (msgtyp == msghdr->msg_type ||
1236 				    msghdr->msg_type <= -msgtyp) {
1237 					DPRINTF(("found message type %ld, "
1238 					    "requested %ld\n",
1239 					    msghdr->msg_type, msgtyp));
1240 					if (msgsz < msghdr->msg_ts &&
1241 					    (msgflg & MSG_NOERROR) == 0) {
1242 						DPRINTF(("requested message "
1243 						    "on the queue is too big "
1244 						    "(want %zu, got %hu)\n",
1245 						    msgsz, msghdr->msg_ts));
1246 						error = E2BIG;
1247 						goto done2;
1248 					}
1249 #ifdef MAC
1250 					error = mac_sysvmsq_check_msgrcv(
1251 					    td->td_ucred, msghdr);
1252 					if (error != 0)
1253 						goto done2;
1254 #endif
1255 					*prev = msghdr->msg_next;
1256 					if (msghdr == msqkptr->u.__msg_last) {
1257 						if (previous == NULL) {
1258 							if (prev !=
1259 							    &msqkptr->u.__msg_first)
1260 								panic("__msg_first/last screwed up #2");
1261 							msqkptr->u.__msg_first =
1262 							    NULL;
1263 							msqkptr->u.__msg_last =
1264 							    NULL;
1265 						} else {
1266 							if (prev ==
1267 							    &msqkptr->u.__msg_first)
1268 								panic("__msg_first/last screwed up #3");
1269 							msqkptr->u.__msg_last =
1270 							    previous;
1271 						}
1272 					}
1273 					break;
1274 				}
1275 				previous = msghdr;
1276 				prev = &(msghdr->msg_next);
1277 			}
1278 		}
1279 
1280 		/*
1281 		 * We've either extracted the msghdr for the appropriate
1282 		 * message or there isn't one.
1283 		 * If there is one then bail out of this loop.
1284 		 */
1285 
1286 		if (msghdr != NULL)
1287 			break;
1288 
1289 		/*
1290 		 * Hmph!  No message found.  Does the user want to wait?
1291 		 */
1292 
1293 		if ((msgflg & IPC_NOWAIT) != 0) {
1294 			DPRINTF(("no appropriate message found (msgtyp=%ld)\n",
1295 			    msgtyp));
1296 			/* The SVID says to return ENOMSG. */
1297 			error = ENOMSG;
1298 			goto done2;
1299 		}
1300 
1301 		/*
1302 		 * Wait for something to happen
1303 		 */
1304 
1305 		DPRINTF(("msgrcv:  goodnight\n"));
1306 		error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH,
1307 		    "msgrcv", 0);
1308 		DPRINTF(("msgrcv:  good morning (error=%d)\n", error));
1309 
1310 		if (error != 0) {
1311 			DPRINTF(("msgrcv:  interrupted system call\n"));
1312 			error = EINTR;
1313 			goto done2;
1314 		}
1315 
1316 		/*
1317 		 * Make sure that the msq queue still exists
1318 		 */
1319 
1320 		if (msqkptr->u.msg_qbytes == 0 ||
1321 		    msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) {
1322 			DPRINTF(("msqid deleted\n"));
1323 			error = EIDRM;
1324 			goto done2;
1325 		}
1326 	}
1327 
1328 	/*
1329 	 * Return the message to the user.
1330 	 *
1331 	 * First, do the bookkeeping (before we risk being interrupted).
1332 	 */
1333 
1334 	msqkptr->u.msg_cbytes -= msghdr->msg_ts;
1335 	msqkptr->u.msg_qnum--;
1336 	msqkptr->u.msg_lrpid = td->td_proc->p_pid;
1337 	msqkptr->u.msg_rtime = time_second;
1338 
1339 	racct_sub_cred(msqkptr->cred, RACCT_MSGQQUEUED, 1);
1340 	racct_sub_cred(msqkptr->cred, RACCT_MSGQSIZE, msghdr->msg_ts);
1341 
1342 	/*
1343 	 * Make msgsz the actual amount that we'll be returning.
1344 	 * Note that this effectively truncates the message if it is too long
1345 	 * (since msgsz is never increased).
1346 	 */
1347 
1348 	DPRINTF(("found a message, msgsz=%zu, msg_ts=%hu\n", msgsz,
1349 	    msghdr->msg_ts));
1350 	if (msgsz > msghdr->msg_ts)
1351 		msgsz = msghdr->msg_ts;
1352 	*mtype = msghdr->msg_type;
1353 
1354 	/*
1355 	 * Return the segments to the user
1356 	 */
1357 
1358 	next = msghdr->msg_spot;
1359 	for (len = 0; len < msgsz; len += msginfo.msgssz) {
1360 		size_t tlen;
1361 
1362 		if (msgsz - len > msginfo.msgssz)
1363 			tlen = msginfo.msgssz;
1364 		else
1365 			tlen = msgsz - len;
1366 		if (next <= -1)
1367 			panic("next too low #3");
1368 		if (next >= msginfo.msgseg)
1369 			panic("next out of range #3");
1370 		mtx_unlock(&msq_mtx);
1371 		error = copyout(&msgpool[next * msginfo.msgssz], msgp, tlen);
1372 		mtx_lock(&msq_mtx);
1373 		if (error != 0) {
1374 			DPRINTF(("error (%d) copying out message segment\n",
1375 			    error));
1376 			msg_freehdr(msghdr);
1377 			wakeup(msqkptr);
1378 			goto done2;
1379 		}
1380 		msgp = (char *)msgp + tlen;
1381 		next = msgmaps[next].next;
1382 	}
1383 
1384 	/*
1385 	 * Done, return the actual number of bytes copied out.
1386 	 */
1387 
1388 	msg_freehdr(msghdr);
1389 	wakeup(msqkptr);
1390 	td->td_retval[0] = msgsz;
1391 done2:
1392 	mtx_unlock(&msq_mtx);
1393 	return (error);
1394 }
1395 
1396 int
1397 sys_msgrcv(struct thread *td, struct msgrcv_args *uap)
1398 {
1399 	int error;
1400 	long mtype;
1401 
1402 	DPRINTF(("call to msgrcv(%d, %p, %zu, %ld, %d)\n", uap->msqid,
1403 	    uap->msgp, uap->msgsz, uap->msgtyp, uap->msgflg));
1404 
1405 	if ((error = kern_msgrcv(td, uap->msqid,
1406 	    (char *)uap->msgp + sizeof(mtype), uap->msgsz,
1407 	    uap->msgtyp, uap->msgflg, &mtype)) != 0)
1408 		return (error);
1409 	if ((error = copyout(&mtype, uap->msgp, sizeof(mtype))) != 0)
1410 		DPRINTF(("error %d copying the message type\n", error));
1411 	return (error);
1412 }
1413 
1414 static int
1415 sysctl_msqids(SYSCTL_HANDLER_ARGS)
1416 {
1417 	struct msqid_kernel tmsqk;
1418 #ifdef COMPAT_FREEBSD32
1419 	struct msqid_kernel32 tmsqk32;
1420 #endif
1421 	struct prison *pr, *rpr;
1422 	void *outaddr;
1423 	size_t outsize;
1424 	int error, i;
1425 
1426 	pr = req->td->td_ucred->cr_prison;
1427 	rpr = msg_find_prison(req->td->td_ucred);
1428 	error = 0;
1429 	for (i = 0; i < msginfo.msgmni; i++) {
1430 		mtx_lock(&msq_mtx);
1431 		if (msqids[i].u.msg_qbytes == 0 || rpr == NULL ||
1432 		    msq_prison_cansee(rpr, &msqids[i]) != 0)
1433 			bzero(&tmsqk, sizeof(tmsqk));
1434 		else {
1435 			tmsqk = msqids[i];
1436 			if (tmsqk.cred->cr_prison != pr)
1437 				tmsqk.u.msg_perm.key = IPC_PRIVATE;
1438 		}
1439 		mtx_unlock(&msq_mtx);
1440 #ifdef COMPAT_FREEBSD32
1441 		if (SV_CURPROC_FLAG(SV_ILP32)) {
1442 			bzero(&tmsqk32, sizeof(tmsqk32));
1443 			freebsd32_ipcperm_out(&tmsqk.u.msg_perm,
1444 			    &tmsqk32.u.msg_perm);
1445 			/* Don't copy u.msg_first or u.msg_last */
1446 			CP(tmsqk, tmsqk32, u.msg_cbytes);
1447 			CP(tmsqk, tmsqk32, u.msg_qnum);
1448 			CP(tmsqk, tmsqk32, u.msg_qbytes);
1449 			CP(tmsqk, tmsqk32, u.msg_lspid);
1450 			CP(tmsqk, tmsqk32, u.msg_lrpid);
1451 			CP(tmsqk, tmsqk32, u.msg_stime);
1452 			CP(tmsqk, tmsqk32, u.msg_rtime);
1453 			CP(tmsqk, tmsqk32, u.msg_ctime);
1454 			/* Don't copy label or cred */
1455 			outaddr = &tmsqk32;
1456 			outsize = sizeof(tmsqk32);
1457 		} else
1458 #endif
1459 		{
1460 			/* Don't leak kernel pointers */
1461 			tmsqk.u.__msg_first = NULL;
1462 			tmsqk.u.__msg_last = NULL;
1463 			tmsqk.label = NULL;
1464 			tmsqk.cred = NULL;
1465 			/*
1466 			 * XXX: some padding also exists, but we take care to
1467 			 * allocate our pool of msqid_kernel structs with
1468 			 * zeroed memory so this should be OK.
1469 			 */
1470 			outaddr = &tmsqk;
1471 			outsize = sizeof(tmsqk);
1472 		}
1473 		error = SYSCTL_OUT(req, outaddr, outsize);
1474 		if (error != 0)
1475 			break;
1476 	}
1477 	return (error);
1478 }
1479 
1480 int
1481 kern_get_msqids(struct thread *td, struct msqid_kernel **res, size_t *sz)
1482 {
1483 	struct msqid_kernel *pmsqk;
1484 	struct prison *pr, *rpr;
1485 	int i, mi;
1486 
1487 	*sz = mi = msginfo.msgmni;
1488 	if (res == NULL)
1489 		return (0);
1490 
1491 	pr = td->td_ucred->cr_prison;
1492 	rpr = msg_find_prison(td->td_ucred);
1493 	*res = malloc(sizeof(struct msqid_kernel) * mi, M_TEMP, M_WAITOK);
1494 	for (i = 0; i < mi; i++) {
1495 		pmsqk = &(*res)[i];
1496 		mtx_lock(&msq_mtx);
1497 		if (msqids[i].u.msg_qbytes == 0 || rpr == NULL ||
1498 		    msq_prison_cansee(rpr, &msqids[i]) != 0)
1499 			bzero(pmsqk, sizeof(*pmsqk));
1500 		else {
1501 			*pmsqk = msqids[i];
1502 			if (pmsqk->cred->cr_prison != pr)
1503 				pmsqk->u.msg_perm.key = IPC_PRIVATE;
1504 		}
1505 		mtx_unlock(&msq_mtx);
1506 		pmsqk->u.__msg_first = NULL;
1507 		pmsqk->u.__msg_last = NULL;
1508 		pmsqk->label = NULL;
1509 		pmsqk->cred = NULL;
1510 	}
1511 	return (0);
1512 }
1513 
1514 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmax, CTLFLAG_RD, &msginfo.msgmax, 0,
1515     "Maximum message size");
1516 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmni, CTLFLAG_RDTUN, &msginfo.msgmni, 0,
1517     "Number of message queue identifiers");
1518 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmnb, CTLFLAG_RDTUN, &msginfo.msgmnb, 0,
1519     "Maximum number of bytes in a queue");
1520 SYSCTL_INT(_kern_ipc, OID_AUTO, msgtql, CTLFLAG_RDTUN, &msginfo.msgtql, 0,
1521     "Maximum number of messages in the system");
1522 SYSCTL_INT(_kern_ipc, OID_AUTO, msgssz, CTLFLAG_RDTUN, &msginfo.msgssz, 0,
1523     "Size of a message segment");
1524 SYSCTL_INT(_kern_ipc, OID_AUTO, msgseg, CTLFLAG_RDTUN, &msginfo.msgseg, 0,
1525     "Number of message segments");
1526 SYSCTL_PROC(_kern_ipc, OID_AUTO, msqids,
1527     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1528     NULL, 0, sysctl_msqids, "",
1529     "Array of struct msqid_kernel for each potential message queue");
1530 
1531 static int
1532 msg_prison_check(void *obj, void *data)
1533 {
1534 	struct prison *pr = obj;
1535 	struct prison *prpr;
1536 	struct vfsoptlist *opts = data;
1537 	int error, jsys;
1538 
1539 	/*
1540 	 * sysvmsg is a jailsys integer.
1541 	 * It must be "disable" if the parent jail is disabled.
1542 	 */
1543 	error = vfs_copyopt(opts, "sysvmsg", &jsys, sizeof(jsys));
1544 	if (error != ENOENT) {
1545 		if (error != 0)
1546 			return (error);
1547 		switch (jsys) {
1548 		case JAIL_SYS_DISABLE:
1549 			break;
1550 		case JAIL_SYS_NEW:
1551 		case JAIL_SYS_INHERIT:
1552 			prison_lock(pr->pr_parent);
1553 			prpr = osd_jail_get(pr->pr_parent, msg_prison_slot);
1554 			prison_unlock(pr->pr_parent);
1555 			if (prpr == NULL)
1556 				return (EPERM);
1557 			break;
1558 		default:
1559 			return (EINVAL);
1560 		}
1561 	}
1562 
1563 	return (0);
1564 }
1565 
1566 static int
1567 msg_prison_set(void *obj, void *data)
1568 {
1569 	struct prison *pr = obj;
1570 	struct prison *tpr, *orpr, *nrpr, *trpr;
1571 	struct vfsoptlist *opts = data;
1572 	void *rsv;
1573 	int jsys, descend;
1574 
1575 	/*
1576 	 * sysvmsg controls which jail is the root of the associated msgs (this
1577 	 * jail or same as the parent), or if the feature is available at all.
1578 	 */
1579 	if (vfs_copyopt(opts, "sysvmsg", &jsys, sizeof(jsys)) == ENOENT)
1580 		jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0)
1581 		    ? JAIL_SYS_INHERIT
1582 		    : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0)
1583 		    ? JAIL_SYS_DISABLE
1584 		    : -1;
1585 	if (jsys == JAIL_SYS_DISABLE) {
1586 		prison_lock(pr);
1587 		orpr = osd_jail_get(pr, msg_prison_slot);
1588 		if (orpr != NULL)
1589 			osd_jail_del(pr, msg_prison_slot);
1590 		prison_unlock(pr);
1591 		if (orpr != NULL) {
1592 			if (orpr == pr)
1593 				msg_prison_cleanup(pr);
1594 			/* Disable all child jails as well. */
1595 			FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1596 				prison_lock(tpr);
1597 				trpr = osd_jail_get(tpr, msg_prison_slot);
1598 				if (trpr != NULL) {
1599 					osd_jail_del(tpr, msg_prison_slot);
1600 					prison_unlock(tpr);
1601 					if (trpr == tpr)
1602 						msg_prison_cleanup(tpr);
1603 				} else {
1604 					prison_unlock(tpr);
1605 					descend = 0;
1606 				}
1607 			}
1608 		}
1609 	} else if (jsys != -1) {
1610 		if (jsys == JAIL_SYS_NEW)
1611 			nrpr = pr;
1612 		else {
1613 			prison_lock(pr->pr_parent);
1614 			nrpr = osd_jail_get(pr->pr_parent, msg_prison_slot);
1615 			prison_unlock(pr->pr_parent);
1616 		}
1617 		rsv = osd_reserve(msg_prison_slot);
1618 		prison_lock(pr);
1619 		orpr = osd_jail_get(pr, msg_prison_slot);
1620 		if (orpr != nrpr)
1621 			(void)osd_jail_set_reserved(pr, msg_prison_slot, rsv,
1622 			    nrpr);
1623 		else
1624 			osd_free_reserved(rsv);
1625 		prison_unlock(pr);
1626 		if (orpr != nrpr) {
1627 			if (orpr == pr)
1628 				msg_prison_cleanup(pr);
1629 			if (orpr != NULL) {
1630 				/* Change child jails matching the old root, */
1631 				FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1632 					prison_lock(tpr);
1633 					trpr = osd_jail_get(tpr,
1634 					    msg_prison_slot);
1635 					if (trpr == orpr) {
1636 						(void)osd_jail_set(tpr,
1637 						    msg_prison_slot, nrpr);
1638 						prison_unlock(tpr);
1639 						if (trpr == tpr)
1640 							msg_prison_cleanup(tpr);
1641 					} else {
1642 						prison_unlock(tpr);
1643 						descend = 0;
1644 					}
1645 				}
1646 			}
1647 		}
1648 	}
1649 
1650 	return (0);
1651 }
1652 
1653 static int
1654 msg_prison_get(void *obj, void *data)
1655 {
1656 	struct prison *pr = obj;
1657 	struct prison *rpr;
1658 	struct vfsoptlist *opts = data;
1659 	int error, jsys;
1660 
1661 	/* Set sysvmsg based on the jail's root prison. */
1662 	prison_lock(pr);
1663 	rpr = osd_jail_get(pr, msg_prison_slot);
1664 	prison_unlock(pr);
1665 	jsys = rpr == NULL ? JAIL_SYS_DISABLE
1666 	    : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT;
1667 	error = vfs_setopt(opts, "sysvmsg", &jsys, sizeof(jsys));
1668 	if (error == ENOENT)
1669 		error = 0;
1670 	return (error);
1671 }
1672 
1673 static int
1674 msg_prison_remove(void *obj, void *data __unused)
1675 {
1676 	struct prison *pr = obj;
1677 	struct prison *rpr;
1678 
1679 	prison_lock(pr);
1680 	rpr = osd_jail_get(pr, msg_prison_slot);
1681 	prison_unlock(pr);
1682 	if (rpr == pr)
1683 		msg_prison_cleanup(pr);
1684 	return (0);
1685 }
1686 
1687 static void
1688 msg_prison_cleanup(struct prison *pr)
1689 {
1690 	struct msqid_kernel *msqkptr;
1691 	int i;
1692 
1693 	/* Remove any msqs that belong to this jail. */
1694 	mtx_lock(&msq_mtx);
1695 	for (i = 0; i < msginfo.msgmni; i++) {
1696 		msqkptr = &msqids[i];
1697 		if (msqkptr->u.msg_qbytes != 0 &&
1698 		    msqkptr->cred != NULL && msqkptr->cred->cr_prison == pr)
1699 			msq_remove(msqkptr);
1700 	}
1701 	mtx_unlock(&msq_mtx);
1702 }
1703 
1704 SYSCTL_JAIL_PARAM_SYS_NODE(sysvmsg, CTLFLAG_RW, "SYSV message queues");
1705 
1706 #ifdef COMPAT_FREEBSD32
1707 int
1708 freebsd32_msgsys(struct thread *td, struct freebsd32_msgsys_args *uap)
1709 {
1710 
1711 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1712     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1713 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1714 	switch (uap->which) {
1715 	case 0:
1716 		return (freebsd7_freebsd32_msgctl(td,
1717 		    (struct freebsd7_freebsd32_msgctl_args *)&uap->a2));
1718 	case 2:
1719 		return (freebsd32_msgsnd(td,
1720 		    (struct freebsd32_msgsnd_args *)&uap->a2));
1721 	case 3:
1722 		return (freebsd32_msgrcv(td,
1723 		    (struct freebsd32_msgrcv_args *)&uap->a2));
1724 	default:
1725 		return (sys_msgsys(td, (struct msgsys_args *)uap));
1726 	}
1727 #else
1728 	return (nosys(td, NULL));
1729 #endif
1730 }
1731 
1732 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1733     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1734 int
1735 freebsd7_freebsd32_msgctl(struct thread *td,
1736     struct freebsd7_freebsd32_msgctl_args *uap)
1737 {
1738 	struct msqid_ds msqbuf;
1739 	struct msqid_ds_old32 msqbuf32;
1740 	int error;
1741 
1742 	if (uap->cmd == IPC_SET) {
1743 		error = copyin(uap->buf, &msqbuf32, sizeof(msqbuf32));
1744 		if (error)
1745 			return (error);
1746 		freebsd32_ipcperm_old_in(&msqbuf32.msg_perm, &msqbuf.msg_perm);
1747 		PTRIN_CP(msqbuf32, msqbuf, __msg_first);
1748 		PTRIN_CP(msqbuf32, msqbuf, __msg_last);
1749 		CP(msqbuf32, msqbuf, msg_cbytes);
1750 		CP(msqbuf32, msqbuf, msg_qnum);
1751 		CP(msqbuf32, msqbuf, msg_qbytes);
1752 		CP(msqbuf32, msqbuf, msg_lspid);
1753 		CP(msqbuf32, msqbuf, msg_lrpid);
1754 		CP(msqbuf32, msqbuf, msg_stime);
1755 		CP(msqbuf32, msqbuf, msg_rtime);
1756 		CP(msqbuf32, msqbuf, msg_ctime);
1757 	}
1758 	error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf);
1759 	if (error)
1760 		return (error);
1761 	if (uap->cmd == IPC_STAT) {
1762 		bzero(&msqbuf32, sizeof(msqbuf32));
1763 		freebsd32_ipcperm_old_out(&msqbuf.msg_perm, &msqbuf32.msg_perm);
1764 		PTROUT_CP(msqbuf, msqbuf32, __msg_first);
1765 		PTROUT_CP(msqbuf, msqbuf32, __msg_last);
1766 		CP(msqbuf, msqbuf32, msg_cbytes);
1767 		CP(msqbuf, msqbuf32, msg_qnum);
1768 		CP(msqbuf, msqbuf32, msg_qbytes);
1769 		CP(msqbuf, msqbuf32, msg_lspid);
1770 		CP(msqbuf, msqbuf32, msg_lrpid);
1771 		CP(msqbuf, msqbuf32, msg_stime);
1772 		CP(msqbuf, msqbuf32, msg_rtime);
1773 		CP(msqbuf, msqbuf32, msg_ctime);
1774 		error = copyout(&msqbuf32, uap->buf, sizeof(struct msqid_ds32));
1775 	}
1776 	return (error);
1777 }
1778 #endif
1779 
1780 int
1781 freebsd32_msgctl(struct thread *td, struct freebsd32_msgctl_args *uap)
1782 {
1783 	struct msqid_ds msqbuf;
1784 	struct msqid_ds32 msqbuf32;
1785 	int error;
1786 
1787 	if (uap->cmd == IPC_SET) {
1788 		error = copyin(uap->buf, &msqbuf32, sizeof(msqbuf32));
1789 		if (error)
1790 			return (error);
1791 		freebsd32_ipcperm_in(&msqbuf32.msg_perm, &msqbuf.msg_perm);
1792 		PTRIN_CP(msqbuf32, msqbuf, __msg_first);
1793 		PTRIN_CP(msqbuf32, msqbuf, __msg_last);
1794 		CP(msqbuf32, msqbuf, msg_cbytes);
1795 		CP(msqbuf32, msqbuf, msg_qnum);
1796 		CP(msqbuf32, msqbuf, msg_qbytes);
1797 		CP(msqbuf32, msqbuf, msg_lspid);
1798 		CP(msqbuf32, msqbuf, msg_lrpid);
1799 		CP(msqbuf32, msqbuf, msg_stime);
1800 		CP(msqbuf32, msqbuf, msg_rtime);
1801 		CP(msqbuf32, msqbuf, msg_ctime);
1802 	}
1803 	error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf);
1804 	if (error)
1805 		return (error);
1806 	if (uap->cmd == IPC_STAT) {
1807 		freebsd32_ipcperm_out(&msqbuf.msg_perm, &msqbuf32.msg_perm);
1808 		PTROUT_CP(msqbuf, msqbuf32, __msg_first);
1809 		PTROUT_CP(msqbuf, msqbuf32, __msg_last);
1810 		CP(msqbuf, msqbuf32, msg_cbytes);
1811 		CP(msqbuf, msqbuf32, msg_qnum);
1812 		CP(msqbuf, msqbuf32, msg_qbytes);
1813 		CP(msqbuf, msqbuf32, msg_lspid);
1814 		CP(msqbuf, msqbuf32, msg_lrpid);
1815 		CP(msqbuf, msqbuf32, msg_stime);
1816 		CP(msqbuf, msqbuf32, msg_rtime);
1817 		CP(msqbuf, msqbuf32, msg_ctime);
1818 		error = copyout(&msqbuf32, uap->buf, sizeof(struct msqid_ds32));
1819 	}
1820 	return (error);
1821 }
1822 
1823 int
1824 freebsd32_msgsnd(struct thread *td, struct freebsd32_msgsnd_args *uap)
1825 {
1826 	const void *msgp;
1827 	long mtype;
1828 	int32_t mtype32;
1829 	int error;
1830 
1831 	msgp = PTRIN(uap->msgp);
1832 	if ((error = copyin(msgp, &mtype32, sizeof(mtype32))) != 0)
1833 		return (error);
1834 	mtype = mtype32;
1835 	return (kern_msgsnd(td, uap->msqid,
1836 	    (const char *)msgp + sizeof(mtype32),
1837 	    uap->msgsz, uap->msgflg, mtype));
1838 }
1839 
1840 int
1841 freebsd32_msgrcv(struct thread *td, struct freebsd32_msgrcv_args *uap)
1842 {
1843 	void *msgp;
1844 	long mtype;
1845 	int32_t mtype32;
1846 	int error;
1847 
1848 	msgp = PTRIN(uap->msgp);
1849 	if ((error = kern_msgrcv(td, uap->msqid,
1850 	    (char *)msgp + sizeof(mtype32), uap->msgsz,
1851 	    uap->msgtyp, uap->msgflg, &mtype)) != 0)
1852 		return (error);
1853 	mtype32 = (int32_t)mtype;
1854 	return (copyout(&mtype32, msgp, sizeof(mtype32)));
1855 }
1856 #endif
1857 
1858 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1859     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1860 
1861 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1862 static sy_call_t *msgcalls[] = {
1863 	(sy_call_t *)freebsd7_msgctl, (sy_call_t *)sys_msgget,
1864 	(sy_call_t *)sys_msgsnd, (sy_call_t *)sys_msgrcv
1865 };
1866 
1867 /*
1868  * Entry point for all MSG calls.
1869  *
1870  * XXX actually varargs.
1871  * struct msgsys_args {
1872  *		int	which;
1873  *		int	a2;
1874  *		int	a3;
1875  *		int	a4;
1876  *		int	a5;
1877  *		int	a6;
1878  *	} *uap;
1879  */
1880 int
1881 sys_msgsys(struct thread *td, struct msgsys_args *uap)
1882 {
1883 	int error;
1884 
1885 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1886 	if (uap->which < 0 || uap->which >= nitems(msgcalls))
1887 		return (EINVAL);
1888 	error = (*msgcalls[uap->which])(td, &uap->a2);
1889 	return (error);
1890 }
1891 
1892 #ifndef CP
1893 #define CP(src, dst, fld)	do { (dst).fld = (src).fld; } while (0)
1894 #endif
1895 
1896 #ifndef _SYS_SYSPROTO_H_
1897 struct freebsd7_msgctl_args {
1898 	int	msqid;
1899 	int	cmd;
1900 	struct	msqid_ds_old *buf;
1901 };
1902 #endif
1903 int
1904 freebsd7_msgctl(struct thread *td, struct freebsd7_msgctl_args *uap)
1905 {
1906 	struct msqid_ds_old msqold;
1907 	struct msqid_ds msqbuf;
1908 	int error;
1909 
1910 	DPRINTF(("call to freebsd7_msgctl(%d, %d, %p)\n", uap->msqid, uap->cmd,
1911 	    uap->buf));
1912 	if (uap->cmd == IPC_SET) {
1913 		error = copyin(uap->buf, &msqold, sizeof(msqold));
1914 		if (error)
1915 			return (error);
1916 		ipcperm_old2new(&msqold.msg_perm, &msqbuf.msg_perm);
1917 		CP(msqold, msqbuf, __msg_first);
1918 		CP(msqold, msqbuf, __msg_last);
1919 		CP(msqold, msqbuf, msg_cbytes);
1920 		CP(msqold, msqbuf, msg_qnum);
1921 		CP(msqold, msqbuf, msg_qbytes);
1922 		CP(msqold, msqbuf, msg_lspid);
1923 		CP(msqold, msqbuf, msg_lrpid);
1924 		CP(msqold, msqbuf, msg_stime);
1925 		CP(msqold, msqbuf, msg_rtime);
1926 		CP(msqold, msqbuf, msg_ctime);
1927 	}
1928 	error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf);
1929 	if (error)
1930 		return (error);
1931 	if (uap->cmd == IPC_STAT) {
1932 		bzero(&msqold, sizeof(msqold));
1933 		ipcperm_new2old(&msqbuf.msg_perm, &msqold.msg_perm);
1934 		CP(msqbuf, msqold, __msg_first);
1935 		CP(msqbuf, msqold, __msg_last);
1936 		CP(msqbuf, msqold, msg_cbytes);
1937 		CP(msqbuf, msqold, msg_qnum);
1938 		CP(msqbuf, msqold, msg_qbytes);
1939 		CP(msqbuf, msqold, msg_lspid);
1940 		CP(msqbuf, msqold, msg_lrpid);
1941 		CP(msqbuf, msqold, msg_stime);
1942 		CP(msqbuf, msqold, msg_rtime);
1943 		CP(msqbuf, msqold, msg_ctime);
1944 		error = copyout(&msqold, uap->buf, sizeof(struct msqid_ds_old));
1945 	}
1946 	return (error);
1947 }
1948 
1949 #undef CP
1950 
1951 #endif	/* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 ||
1952 	   COMPAT_FREEBSD7 */
1953