xref: /freebsd/sys/kern/sysv_msg.c (revision eb69d1f144a6fcc765d1b9d44a5ae8082353e70b)
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-FreeBSD
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 __FBSDID("$FreeBSD$");
60 
61 #include "opt_compat.h"
62 #include "opt_sysvipc.h"
63 
64 #include <sys/param.h>
65 #include <sys/systm.h>
66 #include <sys/sysproto.h>
67 #include <sys/kernel.h>
68 #include <sys/priv.h>
69 #include <sys/proc.h>
70 #include <sys/lock.h>
71 #include <sys/mutex.h>
72 #include <sys/module.h>
73 #include <sys/mount.h>
74 #include <sys/msg.h>
75 #include <sys/racct.h>
76 #include <sys/sx.h>
77 #include <sys/syscall.h>
78 #include <sys/syscallsubr.h>
79 #include <sys/sysent.h>
80 #include <sys/sysctl.h>
81 #include <sys/malloc.h>
82 #include <sys/jail.h>
83 
84 #include <security/audit/audit.h>
85 #include <security/mac/mac_framework.h>
86 
87 FEATURE(sysv_msg, "System V message queues support");
88 
89 static MALLOC_DEFINE(M_MSG, "msg", "SVID compatible message queues");
90 
91 static int msginit(void);
92 static int msgunload(void);
93 static int sysvmsg_modload(struct module *, int, void *);
94 static void msq_remove(struct msqid_kernel *);
95 static struct prison *msg_find_prison(struct ucred *);
96 static int msq_prison_cansee(struct prison *, struct msqid_kernel *);
97 static int msg_prison_check(void *, void *);
98 static int msg_prison_set(void *, void *);
99 static int msg_prison_get(void *, void *);
100 static int msg_prison_remove(void *, void *);
101 static void msg_prison_cleanup(struct prison *);
102 
103 
104 #ifdef MSG_DEBUG
105 #define DPRINTF(a)	printf a
106 #else
107 #define DPRINTF(a)	(void)0
108 #endif
109 
110 static void msg_freehdr(struct msg *msghdr);
111 
112 #ifndef MSGSSZ
113 #define MSGSSZ	8		/* Each segment must be 2^N long */
114 #endif
115 #ifndef MSGSEG
116 #define MSGSEG	2048		/* must be less than 32767 */
117 #endif
118 #define MSGMAX	(MSGSSZ*MSGSEG)
119 #ifndef MSGMNB
120 #define MSGMNB	2048		/* max # of bytes in a queue */
121 #endif
122 #ifndef MSGMNI
123 #define MSGMNI	40
124 #endif
125 #ifndef MSGTQL
126 #define MSGTQL	40
127 #endif
128 
129 /*
130  * Based on the configuration parameters described in an SVR2 (yes, two)
131  * config(1m) man page.
132  *
133  * Each message is broken up and stored in segments that are msgssz bytes
134  * long.  For efficiency reasons, this should be a power of two.  Also,
135  * it doesn't make sense if it is less than 8 or greater than about 256.
136  * Consequently, msginit in kern/sysv_msg.c checks that msgssz is a power of
137  * two between 8 and 1024 inclusive (and panic's if it isn't).
138  */
139 struct msginfo msginfo = {
140                 MSGMAX,         /* max chars in a message */
141                 MSGMNI,         /* # of message queue identifiers */
142                 MSGMNB,         /* max chars in a queue */
143                 MSGTQL,         /* max messages in system */
144                 MSGSSZ,         /* size of a message segment */
145                 		/* (must be small power of 2 greater than 4) */
146                 MSGSEG          /* number of message segments */
147 };
148 
149 /*
150  * macros to convert between msqid_ds's and msqid's.
151  * (specific to this implementation)
152  */
153 #define MSQID(ix,ds)	((ix) & 0xffff | (((ds).msg_perm.seq << 16) & 0xffff0000))
154 #define MSQID_IX(id)	((id) & 0xffff)
155 #define MSQID_SEQ(id)	(((id) >> 16) & 0xffff)
156 
157 /*
158  * The rest of this file is specific to this particular implementation.
159  */
160 
161 struct msgmap {
162 	short	next;		/* next segment in buffer */
163     				/* -1 -> available */
164     				/* 0..(MSGSEG-1) -> index of next segment */
165 };
166 
167 #define MSG_LOCKED	01000	/* Is this msqid_ds locked? */
168 
169 static int nfree_msgmaps;	/* # of free map entries */
170 static short free_msgmaps;	/* head of linked list of free map entries */
171 static struct msg *free_msghdrs;/* list of free msg headers */
172 static char *msgpool;		/* MSGMAX byte long msg buffer pool */
173 static struct msgmap *msgmaps;	/* MSGSEG msgmap structures */
174 static struct msg *msghdrs;	/* MSGTQL msg headers */
175 static struct msqid_kernel *msqids;	/* MSGMNI msqid_kernel struct's */
176 static struct mtx msq_mtx;	/* global mutex for message queues. */
177 static unsigned msg_prison_slot;/* prison OSD slot */
178 
179 static struct syscall_helper_data msg_syscalls[] = {
180 	SYSCALL_INIT_HELPER(msgctl),
181 	SYSCALL_INIT_HELPER(msgget),
182 	SYSCALL_INIT_HELPER(msgsnd),
183 	SYSCALL_INIT_HELPER(msgrcv),
184 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
185     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
186 	SYSCALL_INIT_HELPER(msgsys),
187 	SYSCALL_INIT_HELPER_COMPAT(freebsd7_msgctl),
188 #endif
189 	SYSCALL_INIT_LAST
190 };
191 
192 #ifdef COMPAT_FREEBSD32
193 #include <compat/freebsd32/freebsd32.h>
194 #include <compat/freebsd32/freebsd32_ipc.h>
195 #include <compat/freebsd32/freebsd32_proto.h>
196 #include <compat/freebsd32/freebsd32_signal.h>
197 #include <compat/freebsd32/freebsd32_syscall.h>
198 #include <compat/freebsd32/freebsd32_util.h>
199 
200 static struct syscall_helper_data msg32_syscalls[] = {
201 	SYSCALL32_INIT_HELPER(freebsd32_msgctl),
202 	SYSCALL32_INIT_HELPER(freebsd32_msgsnd),
203 	SYSCALL32_INIT_HELPER(freebsd32_msgrcv),
204 	SYSCALL32_INIT_HELPER_COMPAT(msgget),
205 	SYSCALL32_INIT_HELPER(freebsd32_msgsys),
206 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
207     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
208 	SYSCALL32_INIT_HELPER(freebsd7_freebsd32_msgctl),
209 #endif
210 	SYSCALL_INIT_LAST
211 };
212 #endif
213 
214 static int
215 msginit()
216 {
217 	struct prison *pr;
218 	void **rsv;
219 	int i, error;
220 	osd_method_t methods[PR_MAXMETHOD] = {
221 	    [PR_METHOD_CHECK] =		msg_prison_check,
222 	    [PR_METHOD_SET] =		msg_prison_set,
223 	    [PR_METHOD_GET] =		msg_prison_get,
224 	    [PR_METHOD_REMOVE] =	msg_prison_remove,
225 	};
226 
227 	msginfo.msgmax = msginfo.msgseg * msginfo.msgssz;
228 	msgpool = malloc(msginfo.msgmax, M_MSG, M_WAITOK);
229 	msgmaps = malloc(sizeof(struct msgmap) * msginfo.msgseg, M_MSG, M_WAITOK);
230 	msghdrs = malloc(sizeof(struct msg) * msginfo.msgtql, M_MSG, M_WAITOK);
231 	msqids = malloc(sizeof(struct msqid_kernel) * msginfo.msgmni, M_MSG,
232 	    M_WAITOK | M_ZERO);
233 
234 	/*
235 	 * msginfo.msgssz should be a power of two for efficiency reasons.
236 	 * It is also pretty silly if msginfo.msgssz is less than 8
237 	 * or greater than about 256 so ...
238 	 */
239 
240 	i = 8;
241 	while (i < 1024 && i != msginfo.msgssz)
242 		i <<= 1;
243     	if (i != msginfo.msgssz) {
244 		DPRINTF(("msginfo.msgssz=%d (0x%x)\n", msginfo.msgssz,
245 		    msginfo.msgssz));
246 		panic("msginfo.msgssz not a small power of 2");
247 	}
248 
249 	if (msginfo.msgseg > 32767) {
250 		DPRINTF(("msginfo.msgseg=%d\n", msginfo.msgseg));
251 		panic("msginfo.msgseg > 32767");
252 	}
253 
254 	for (i = 0; i < msginfo.msgseg; i++) {
255 		if (i > 0)
256 			msgmaps[i-1].next = i;
257 		msgmaps[i].next = -1;	/* implies entry is available */
258 	}
259 	free_msgmaps = 0;
260 	nfree_msgmaps = msginfo.msgseg;
261 
262 	for (i = 0; i < msginfo.msgtql; i++) {
263 		msghdrs[i].msg_type = 0;
264 		if (i > 0)
265 			msghdrs[i-1].msg_next = &msghdrs[i];
266 		msghdrs[i].msg_next = NULL;
267 #ifdef MAC
268 		mac_sysvmsg_init(&msghdrs[i]);
269 #endif
270     	}
271 	free_msghdrs = &msghdrs[0];
272 
273 	for (i = 0; i < msginfo.msgmni; i++) {
274 		msqids[i].u.msg_qbytes = 0;	/* implies entry is available */
275 		msqids[i].u.msg_perm.seq = 0;	/* reset to a known value */
276 		msqids[i].u.msg_perm.mode = 0;
277 #ifdef MAC
278 		mac_sysvmsq_init(&msqids[i]);
279 #endif
280 	}
281 	mtx_init(&msq_mtx, "msq", NULL, MTX_DEF);
282 
283 	/* Set current prisons according to their allow.sysvipc. */
284 	msg_prison_slot = osd_jail_register(NULL, methods);
285 	rsv = osd_reserve(msg_prison_slot);
286 	prison_lock(&prison0);
287 	(void)osd_jail_set_reserved(&prison0, msg_prison_slot, rsv, &prison0);
288 	prison_unlock(&prison0);
289 	rsv = NULL;
290 	sx_slock(&allprison_lock);
291 	TAILQ_FOREACH(pr, &allprison, pr_list) {
292 		if (rsv == NULL)
293 			rsv = osd_reserve(msg_prison_slot);
294 		prison_lock(pr);
295 		if ((pr->pr_allow & PR_ALLOW_SYSVIPC) && pr->pr_ref > 0) {
296 			(void)osd_jail_set_reserved(pr, msg_prison_slot, rsv,
297 			    &prison0);
298 			rsv = NULL;
299 		}
300 		prison_unlock(pr);
301 	}
302 	if (rsv != NULL)
303 		osd_free_reserved(rsv);
304 	sx_sunlock(&allprison_lock);
305 
306 	error = syscall_helper_register(msg_syscalls, SY_THR_STATIC_KLD);
307 	if (error != 0)
308 		return (error);
309 #ifdef COMPAT_FREEBSD32
310 	error = syscall32_helper_register(msg32_syscalls, SY_THR_STATIC_KLD);
311 	if (error != 0)
312 		return (error);
313 #endif
314 	return (0);
315 }
316 
317 static int
318 msgunload()
319 {
320 	struct msqid_kernel *msqkptr;
321 	int msqid;
322 #ifdef MAC
323 	int i;
324 #endif
325 
326 	syscall_helper_unregister(msg_syscalls);
327 #ifdef COMPAT_FREEBSD32
328 	syscall32_helper_unregister(msg32_syscalls);
329 #endif
330 
331 	for (msqid = 0; msqid < msginfo.msgmni; msqid++) {
332 		msqkptr = &msqids[msqid];
333 		if (msqkptr->u.msg_qbytes != 0 ||
334 		    (msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0)
335 			break;
336 	}
337 	if (msqid != msginfo.msgmni)
338 		return (EBUSY);
339 
340 	if (msg_prison_slot != 0)
341 		osd_jail_deregister(msg_prison_slot);
342 #ifdef MAC
343 	for (i = 0; i < msginfo.msgtql; i++)
344 		mac_sysvmsg_destroy(&msghdrs[i]);
345 	for (msqid = 0; msqid < msginfo.msgmni; msqid++)
346 		mac_sysvmsq_destroy(&msqids[msqid]);
347 #endif
348 	free(msgpool, M_MSG);
349 	free(msgmaps, M_MSG);
350 	free(msghdrs, M_MSG);
351 	free(msqids, M_MSG);
352 	mtx_destroy(&msq_mtx);
353 	return (0);
354 }
355 
356 
357 static int
358 sysvmsg_modload(struct module *module, int cmd, void *arg)
359 {
360 	int error = 0;
361 
362 	switch (cmd) {
363 	case MOD_LOAD:
364 		error = msginit();
365 		if (error != 0)
366 			msgunload();
367 		break;
368 	case MOD_UNLOAD:
369 		error = msgunload();
370 		break;
371 	case MOD_SHUTDOWN:
372 		break;
373 	default:
374 		error = EINVAL;
375 		break;
376 	}
377 	return (error);
378 }
379 
380 static moduledata_t sysvmsg_mod = {
381 	"sysvmsg",
382 	&sysvmsg_modload,
383 	NULL
384 };
385 
386 DECLARE_MODULE(sysvmsg, sysvmsg_mod, SI_SUB_SYSV_MSG, SI_ORDER_FIRST);
387 MODULE_VERSION(sysvmsg, 1);
388 
389 static void
390 msg_freehdr(msghdr)
391 	struct msg *msghdr;
392 {
393 	while (msghdr->msg_ts > 0) {
394 		short next;
395 		if (msghdr->msg_spot < 0 || msghdr->msg_spot >= msginfo.msgseg)
396 			panic("msghdr->msg_spot out of range");
397 		next = msgmaps[msghdr->msg_spot].next;
398 		msgmaps[msghdr->msg_spot].next = free_msgmaps;
399 		free_msgmaps = msghdr->msg_spot;
400 		nfree_msgmaps++;
401 		msghdr->msg_spot = next;
402 		if (msghdr->msg_ts >= msginfo.msgssz)
403 			msghdr->msg_ts -= msginfo.msgssz;
404 		else
405 			msghdr->msg_ts = 0;
406 	}
407 	if (msghdr->msg_spot != -1)
408 		panic("msghdr->msg_spot != -1");
409 	msghdr->msg_next = free_msghdrs;
410 	free_msghdrs = msghdr;
411 #ifdef MAC
412 	mac_sysvmsg_cleanup(msghdr);
413 #endif
414 }
415 
416 static void
417 msq_remove(struct msqid_kernel *msqkptr)
418 {
419 	struct msg *msghdr;
420 
421 	racct_sub_cred(msqkptr->cred, RACCT_NMSGQ, 1);
422 	racct_sub_cred(msqkptr->cred, RACCT_MSGQQUEUED, msqkptr->u.msg_qnum);
423 	racct_sub_cred(msqkptr->cred, RACCT_MSGQSIZE, msqkptr->u.msg_cbytes);
424 	crfree(msqkptr->cred);
425 	msqkptr->cred = NULL;
426 
427 	/* Free the message headers */
428 	msghdr = msqkptr->u.msg_first;
429 	while (msghdr != NULL) {
430 		struct msg *msghdr_tmp;
431 
432 		/* Free the segments of each message */
433 		msqkptr->u.msg_cbytes -= msghdr->msg_ts;
434 		msqkptr->u.msg_qnum--;
435 		msghdr_tmp = msghdr;
436 		msghdr = msghdr->msg_next;
437 		msg_freehdr(msghdr_tmp);
438 	}
439 
440 	if (msqkptr->u.msg_cbytes != 0)
441 		panic("msg_cbytes is screwed up");
442 	if (msqkptr->u.msg_qnum != 0)
443 		panic("msg_qnum is screwed up");
444 
445 	msqkptr->u.msg_qbytes = 0;	/* Mark it as free */
446 
447 #ifdef MAC
448 	mac_sysvmsq_cleanup(msqkptr);
449 #endif
450 
451 	wakeup(msqkptr);
452 }
453 
454 static struct prison *
455 msg_find_prison(struct ucred *cred)
456 {
457 	struct prison *pr, *rpr;
458 
459 	pr = cred->cr_prison;
460 	prison_lock(pr);
461 	rpr = osd_jail_get(pr, msg_prison_slot);
462 	prison_unlock(pr);
463 	return rpr;
464 }
465 
466 static int
467 msq_prison_cansee(struct prison *rpr, struct msqid_kernel *msqkptr)
468 {
469 
470 	if (msqkptr->cred == NULL ||
471 	    !(rpr == msqkptr->cred->cr_prison ||
472 	      prison_ischild(rpr, msqkptr->cred->cr_prison)))
473 		return (EINVAL);
474 	return (0);
475 }
476 
477 #ifndef _SYS_SYSPROTO_H_
478 struct msgctl_args {
479 	int	msqid;
480 	int	cmd;
481 	struct	msqid_ds *buf;
482 };
483 #endif
484 int
485 sys_msgctl(struct thread *td, struct msgctl_args *uap)
486 {
487 	int msqid = uap->msqid;
488 	int cmd = uap->cmd;
489 	struct msqid_ds msqbuf;
490 	int error;
491 
492 	DPRINTF(("call to msgctl(%d, %d, %p)\n", msqid, cmd, uap->buf));
493 	if (cmd == IPC_SET &&
494 	    (error = copyin(uap->buf, &msqbuf, sizeof(msqbuf))) != 0)
495 		return (error);
496 	error = kern_msgctl(td, msqid, cmd, &msqbuf);
497 	if (cmd == IPC_STAT && error == 0)
498 		error = copyout(&msqbuf, uap->buf, sizeof(struct msqid_ds));
499 	return (error);
500 }
501 
502 int
503 kern_msgctl(td, msqid, cmd, msqbuf)
504 	struct thread *td;
505 	int msqid;
506 	int cmd;
507 	struct msqid_ds *msqbuf;
508 {
509 	int rval, error, msqix;
510 	struct msqid_kernel *msqkptr;
511 	struct prison *rpr;
512 
513 	rpr = msg_find_prison(td->td_ucred);
514 	if (rpr == NULL)
515 		return (ENOSYS);
516 
517 	AUDIT_ARG_SVIPC_CMD(cmd);
518 	AUDIT_ARG_SVIPC_ID(msqid);
519 	msqix = IPCID_TO_IX(msqid);
520 
521 	if (msqix < 0 || msqix >= msginfo.msgmni) {
522 		DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix,
523 		    msginfo.msgmni));
524 		return (EINVAL);
525 	}
526 
527 	msqkptr = &msqids[msqix];
528 
529 	mtx_lock(&msq_mtx);
530 	if (msqkptr->u.msg_qbytes == 0) {
531 		DPRINTF(("no such msqid\n"));
532 		error = EINVAL;
533 		goto done2;
534 	}
535 	if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) {
536 		DPRINTF(("wrong sequence number\n"));
537 		error = EINVAL;
538 		goto done2;
539 	}
540 
541 	error = msq_prison_cansee(rpr, msqkptr);
542 	if (error != 0) {
543 		DPRINTF(("requester can't see prison\n"));
544 		goto done2;
545 	}
546 
547 #ifdef MAC
548 	error = mac_sysvmsq_check_msqctl(td->td_ucred, msqkptr, cmd);
549 	if (error != 0)
550 		goto done2;
551 #endif
552 
553 	error = 0;
554 	rval = 0;
555 
556 	switch (cmd) {
557 
558 	case IPC_RMID:
559 	{
560 #ifdef MAC
561 		struct msg *msghdr;
562 #endif
563 		if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M)))
564 			goto done2;
565 
566 #ifdef MAC
567 		/*
568 		 * Check that the thread has MAC access permissions to
569 		 * individual msghdrs.  Note: We need to do this in a
570 		 * separate loop because the actual loop alters the
571 		 * msq/msghdr info as it progresses, and there is no going
572 		 * back if half the way through we discover that the
573 		 * thread cannot free a certain msghdr.  The msq will get
574 		 * into an inconsistent state.
575 		 */
576 		for (msghdr = msqkptr->u.msg_first; msghdr != NULL;
577 		    msghdr = msghdr->msg_next) {
578 			error = mac_sysvmsq_check_msgrmid(td->td_ucred, msghdr);
579 			if (error != 0)
580 				goto done2;
581 		}
582 #endif
583 
584 		msq_remove(msqkptr);
585 	}
586 
587 		break;
588 
589 	case IPC_SET:
590 		AUDIT_ARG_SVIPC_PERM(&msqbuf->msg_perm);
591 		if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M)))
592 			goto done2;
593 		if (msqbuf->msg_qbytes > msqkptr->u.msg_qbytes) {
594 			error = priv_check(td, PRIV_IPC_MSGSIZE);
595 			if (error)
596 				goto done2;
597 		}
598 		if (msqbuf->msg_qbytes > msginfo.msgmnb) {
599 			DPRINTF(("can't increase msg_qbytes beyond %d"
600 			    "(truncating)\n", msginfo.msgmnb));
601 			msqbuf->msg_qbytes = msginfo.msgmnb;	/* silently restrict qbytes to system limit */
602 		}
603 		if (msqbuf->msg_qbytes == 0) {
604 			DPRINTF(("can't reduce msg_qbytes to 0\n"));
605 			error = EINVAL;		/* non-standard errno! */
606 			goto done2;
607 		}
608 		msqkptr->u.msg_perm.uid = msqbuf->msg_perm.uid;	/* change the owner */
609 		msqkptr->u.msg_perm.gid = msqbuf->msg_perm.gid;	/* change the owner */
610 		msqkptr->u.msg_perm.mode = (msqkptr->u.msg_perm.mode & ~0777) |
611 		    (msqbuf->msg_perm.mode & 0777);
612 		msqkptr->u.msg_qbytes = msqbuf->msg_qbytes;
613 		msqkptr->u.msg_ctime = time_second;
614 		break;
615 
616 	case IPC_STAT:
617 		if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) {
618 			DPRINTF(("requester doesn't have read access\n"));
619 			goto done2;
620 		}
621 		*msqbuf = msqkptr->u;
622 		if (td->td_ucred->cr_prison != msqkptr->cred->cr_prison)
623 			msqbuf->msg_perm.key = IPC_PRIVATE;
624 		break;
625 
626 	default:
627 		DPRINTF(("invalid command %d\n", cmd));
628 		error = EINVAL;
629 		goto done2;
630 	}
631 
632 	if (error == 0)
633 		td->td_retval[0] = rval;
634 done2:
635 	mtx_unlock(&msq_mtx);
636 	return (error);
637 }
638 
639 #ifndef _SYS_SYSPROTO_H_
640 struct msgget_args {
641 	key_t	key;
642 	int	msgflg;
643 };
644 #endif
645 
646 int
647 sys_msgget(struct thread *td, struct msgget_args *uap)
648 {
649 	int msqid, error = 0;
650 	int key = uap->key;
651 	int msgflg = uap->msgflg;
652 	struct ucred *cred = td->td_ucred;
653 	struct msqid_kernel *msqkptr = NULL;
654 
655 	DPRINTF(("msgget(0x%x, 0%o)\n", key, msgflg));
656 
657 	if (msg_find_prison(cred) == NULL)
658 		return (ENOSYS);
659 
660 	mtx_lock(&msq_mtx);
661 	if (key != IPC_PRIVATE) {
662 		for (msqid = 0; msqid < msginfo.msgmni; msqid++) {
663 			msqkptr = &msqids[msqid];
664 			if (msqkptr->u.msg_qbytes != 0 &&
665 			    msqkptr->cred != NULL &&
666 			    msqkptr->cred->cr_prison == cred->cr_prison &&
667 			    msqkptr->u.msg_perm.key == key)
668 				break;
669 		}
670 		if (msqid < msginfo.msgmni) {
671 			DPRINTF(("found public key\n"));
672 			if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) {
673 				DPRINTF(("not exclusive\n"));
674 				error = EEXIST;
675 				goto done2;
676 			}
677 			AUDIT_ARG_SVIPC_ID(IXSEQ_TO_IPCID(msqid,
678 			    msqkptr->u.msg_perm));
679 			if ((error = ipcperm(td, &msqkptr->u.msg_perm,
680 			    msgflg & 0700))) {
681 				DPRINTF(("requester doesn't have 0%o access\n",
682 				    msgflg & 0700));
683 				goto done2;
684 			}
685 #ifdef MAC
686 			error = mac_sysvmsq_check_msqget(cred, msqkptr);
687 			if (error != 0)
688 				goto done2;
689 #endif
690 			goto found;
691 		}
692 	}
693 
694 	DPRINTF(("need to allocate the msqid_ds\n"));
695 	if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) {
696 		for (msqid = 0; msqid < msginfo.msgmni; msqid++) {
697 			/*
698 			 * Look for an unallocated and unlocked msqid_ds.
699 			 * msqid_ds's can be locked by msgsnd or msgrcv while
700 			 * they are copying the message in/out.  We can't
701 			 * re-use the entry until they release it.
702 			 */
703 			msqkptr = &msqids[msqid];
704 			if (msqkptr->u.msg_qbytes == 0 &&
705 			    (msqkptr->u.msg_perm.mode & MSG_LOCKED) == 0)
706 				break;
707 		}
708 		if (msqid == msginfo.msgmni) {
709 			DPRINTF(("no more msqid_ds's available\n"));
710 			error = ENOSPC;
711 			goto done2;
712 		}
713 #ifdef RACCT
714 		if (racct_enable) {
715 			PROC_LOCK(td->td_proc);
716 			error = racct_add(td->td_proc, RACCT_NMSGQ, 1);
717 			PROC_UNLOCK(td->td_proc);
718 			if (error != 0) {
719 				error = ENOSPC;
720 				goto done2;
721 			}
722 		}
723 #endif
724 		DPRINTF(("msqid %d is available\n", msqid));
725 		msqkptr->u.msg_perm.key = key;
726 		msqkptr->u.msg_perm.cuid = cred->cr_uid;
727 		msqkptr->u.msg_perm.uid = cred->cr_uid;
728 		msqkptr->u.msg_perm.cgid = cred->cr_gid;
729 		msqkptr->u.msg_perm.gid = cred->cr_gid;
730 		msqkptr->u.msg_perm.mode = (msgflg & 0777);
731 		msqkptr->cred = crhold(cred);
732 		/* Make sure that the returned msqid is unique */
733 		msqkptr->u.msg_perm.seq = (msqkptr->u.msg_perm.seq + 1) & 0x7fff;
734 		msqkptr->u.msg_first = NULL;
735 		msqkptr->u.msg_last = NULL;
736 		msqkptr->u.msg_cbytes = 0;
737 		msqkptr->u.msg_qnum = 0;
738 		msqkptr->u.msg_qbytes = msginfo.msgmnb;
739 		msqkptr->u.msg_lspid = 0;
740 		msqkptr->u.msg_lrpid = 0;
741 		msqkptr->u.msg_stime = 0;
742 		msqkptr->u.msg_rtime = 0;
743 		msqkptr->u.msg_ctime = time_second;
744 #ifdef MAC
745 		mac_sysvmsq_create(cred, msqkptr);
746 #endif
747 		AUDIT_ARG_SVIPC_PERM(&msqkptr->u.msg_perm);
748 	} else {
749 		DPRINTF(("didn't find it and wasn't asked to create it\n"));
750 		error = ENOENT;
751 		goto done2;
752 	}
753 
754 found:
755 	/* Construct the unique msqid */
756 	td->td_retval[0] = IXSEQ_TO_IPCID(msqid, msqkptr->u.msg_perm);
757 done2:
758 	mtx_unlock(&msq_mtx);
759 	return (error);
760 }
761 
762 #ifndef _SYS_SYSPROTO_H_
763 struct msgsnd_args {
764 	int	msqid;
765 	const void	*msgp;	/* XXX msgp is actually mtext. */
766 	size_t	msgsz;
767 	int	msgflg;
768 };
769 #endif
770 int
771 kern_msgsnd(struct thread *td, int msqid, const void *msgp,
772     size_t msgsz, int msgflg, long mtype)
773 {
774 	int msqix, segs_needed, error = 0;
775 	struct msqid_kernel *msqkptr;
776 	struct msg *msghdr;
777 	struct prison *rpr;
778 	short next;
779 #ifdef RACCT
780 	size_t saved_msgsz;
781 #endif
782 
783 	rpr = msg_find_prison(td->td_ucred);
784 	if (rpr == NULL)
785 		return (ENOSYS);
786 
787 	mtx_lock(&msq_mtx);
788 	AUDIT_ARG_SVIPC_ID(msqid);
789 	msqix = IPCID_TO_IX(msqid);
790 
791 	if (msqix < 0 || msqix >= msginfo.msgmni) {
792 		DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix,
793 		    msginfo.msgmni));
794 		error = EINVAL;
795 		goto done2;
796 	}
797 
798 	msqkptr = &msqids[msqix];
799 	AUDIT_ARG_SVIPC_PERM(&msqkptr->u.msg_perm);
800 	if (msqkptr->u.msg_qbytes == 0) {
801 		DPRINTF(("no such message queue id\n"));
802 		error = EINVAL;
803 		goto done2;
804 	}
805 	if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) {
806 		DPRINTF(("wrong sequence number\n"));
807 		error = EINVAL;
808 		goto done2;
809 	}
810 
811 	if ((error = msq_prison_cansee(rpr, msqkptr))) {
812 		DPRINTF(("requester can't see prison\n"));
813 		goto done2;
814 	}
815 
816 	if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_W))) {
817 		DPRINTF(("requester doesn't have write access\n"));
818 		goto done2;
819 	}
820 
821 #ifdef MAC
822 	error = mac_sysvmsq_check_msqsnd(td->td_ucred, msqkptr);
823 	if (error != 0)
824 		goto done2;
825 #endif
826 
827 #ifdef RACCT
828 	if (racct_enable) {
829 		PROC_LOCK(td->td_proc);
830 		if (racct_add(td->td_proc, RACCT_MSGQQUEUED, 1)) {
831 			PROC_UNLOCK(td->td_proc);
832 			error = EAGAIN;
833 			goto done2;
834 		}
835 		saved_msgsz = msgsz;
836 		if (racct_add(td->td_proc, RACCT_MSGQSIZE, msgsz)) {
837 			racct_sub(td->td_proc, RACCT_MSGQQUEUED, 1);
838 			PROC_UNLOCK(td->td_proc);
839 			error = EAGAIN;
840 			goto done2;
841 		}
842 		PROC_UNLOCK(td->td_proc);
843 	}
844 #endif
845 
846 	segs_needed = howmany(msgsz, msginfo.msgssz);
847 	DPRINTF(("msgsz=%zu, msgssz=%d, segs_needed=%d\n", msgsz,
848 	    msginfo.msgssz, segs_needed));
849 	for (;;) {
850 		int need_more_resources = 0;
851 
852 		/*
853 		 * check msgsz
854 		 * (inside this loop in case msg_qbytes changes while we sleep)
855 		 */
856 
857 		if (msgsz > msqkptr->u.msg_qbytes) {
858 			DPRINTF(("msgsz > msqkptr->u.msg_qbytes\n"));
859 			error = EINVAL;
860 			goto done3;
861 		}
862 
863 		if (msqkptr->u.msg_perm.mode & MSG_LOCKED) {
864 			DPRINTF(("msqid is locked\n"));
865 			need_more_resources = 1;
866 		}
867 		if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) {
868 			DPRINTF(("msgsz + msg_cbytes > msg_qbytes\n"));
869 			need_more_resources = 1;
870 		}
871 		if (segs_needed > nfree_msgmaps) {
872 			DPRINTF(("segs_needed > nfree_msgmaps\n"));
873 			need_more_resources = 1;
874 		}
875 		if (free_msghdrs == NULL) {
876 			DPRINTF(("no more msghdrs\n"));
877 			need_more_resources = 1;
878 		}
879 
880 		if (need_more_resources) {
881 			int we_own_it;
882 
883 			if ((msgflg & IPC_NOWAIT) != 0) {
884 				DPRINTF(("need more resources but caller "
885 				    "doesn't want to wait\n"));
886 				error = EAGAIN;
887 				goto done3;
888 			}
889 
890 			if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) {
891 				DPRINTF(("we don't own the msqid_ds\n"));
892 				we_own_it = 0;
893 			} else {
894 				/* Force later arrivals to wait for our
895 				   request */
896 				DPRINTF(("we own the msqid_ds\n"));
897 				msqkptr->u.msg_perm.mode |= MSG_LOCKED;
898 				we_own_it = 1;
899 			}
900 			DPRINTF(("msgsnd:  goodnight\n"));
901 			error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH,
902 			    "msgsnd", hz);
903 			DPRINTF(("msgsnd:  good morning, error=%d\n", error));
904 			if (we_own_it)
905 				msqkptr->u.msg_perm.mode &= ~MSG_LOCKED;
906 			if (error == EWOULDBLOCK) {
907 				DPRINTF(("msgsnd:  timed out\n"));
908 				continue;
909 			}
910 			if (error != 0) {
911 				DPRINTF(("msgsnd:  interrupted system call\n"));
912 				error = EINTR;
913 				goto done3;
914 			}
915 
916 			/*
917 			 * Make sure that the msq queue still exists
918 			 */
919 
920 			if (msqkptr->u.msg_qbytes == 0) {
921 				DPRINTF(("msqid deleted\n"));
922 				error = EIDRM;
923 				goto done3;
924 			}
925 
926 		} else {
927 			DPRINTF(("got all the resources that we need\n"));
928 			break;
929 		}
930 	}
931 
932 	/*
933 	 * We have the resources that we need.
934 	 * Make sure!
935 	 */
936 
937 	if (msqkptr->u.msg_perm.mode & MSG_LOCKED)
938 		panic("msg_perm.mode & MSG_LOCKED");
939 	if (segs_needed > nfree_msgmaps)
940 		panic("segs_needed > nfree_msgmaps");
941 	if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes)
942 		panic("msgsz + msg_cbytes > msg_qbytes");
943 	if (free_msghdrs == NULL)
944 		panic("no more msghdrs");
945 
946 	/*
947 	 * Re-lock the msqid_ds in case we page-fault when copying in the
948 	 * message
949 	 */
950 
951 	if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0)
952 		panic("msqid_ds is already locked");
953 	msqkptr->u.msg_perm.mode |= MSG_LOCKED;
954 
955 	/*
956 	 * Allocate a message header
957 	 */
958 
959 	msghdr = free_msghdrs;
960 	free_msghdrs = msghdr->msg_next;
961 	msghdr->msg_spot = -1;
962 	msghdr->msg_ts = msgsz;
963 	msghdr->msg_type = mtype;
964 #ifdef MAC
965 	/*
966 	 * XXXMAC: Should the mac_sysvmsq_check_msgmsq check follow here
967 	 * immediately?  Or, should it be checked just before the msg is
968 	 * enqueued in the msgq (as it is done now)?
969 	 */
970 	mac_sysvmsg_create(td->td_ucred, msqkptr, msghdr);
971 #endif
972 
973 	/*
974 	 * Allocate space for the message
975 	 */
976 
977 	while (segs_needed > 0) {
978 		if (nfree_msgmaps <= 0)
979 			panic("not enough msgmaps");
980 		if (free_msgmaps == -1)
981 			panic("nil free_msgmaps");
982 		next = free_msgmaps;
983 		if (next <= -1)
984 			panic("next too low #1");
985 		if (next >= msginfo.msgseg)
986 			panic("next out of range #1");
987 		DPRINTF(("allocating segment %d to message\n", next));
988 		free_msgmaps = msgmaps[next].next;
989 		nfree_msgmaps--;
990 		msgmaps[next].next = msghdr->msg_spot;
991 		msghdr->msg_spot = next;
992 		segs_needed--;
993 	}
994 
995 	/*
996 	 * Validate the message type
997 	 */
998 
999 	if (msghdr->msg_type < 1) {
1000 		msg_freehdr(msghdr);
1001 		msqkptr->u.msg_perm.mode &= ~MSG_LOCKED;
1002 		wakeup(msqkptr);
1003 		DPRINTF(("mtype (%ld) < 1\n", msghdr->msg_type));
1004 		error = EINVAL;
1005 		goto done3;
1006 	}
1007 
1008 	/*
1009 	 * Copy in the message body
1010 	 */
1011 
1012 	next = msghdr->msg_spot;
1013 	while (msgsz > 0) {
1014 		size_t tlen;
1015 		if (msgsz > msginfo.msgssz)
1016 			tlen = msginfo.msgssz;
1017 		else
1018 			tlen = msgsz;
1019 		if (next <= -1)
1020 			panic("next too low #2");
1021 		if (next >= msginfo.msgseg)
1022 			panic("next out of range #2");
1023 		mtx_unlock(&msq_mtx);
1024 		if ((error = copyin(msgp, &msgpool[next * msginfo.msgssz],
1025 		    tlen)) != 0) {
1026 			mtx_lock(&msq_mtx);
1027 			DPRINTF(("error %d copying in message segment\n",
1028 			    error));
1029 			msg_freehdr(msghdr);
1030 			msqkptr->u.msg_perm.mode &= ~MSG_LOCKED;
1031 			wakeup(msqkptr);
1032 			goto done3;
1033 		}
1034 		mtx_lock(&msq_mtx);
1035 		msgsz -= tlen;
1036 		msgp = (const char *)msgp + tlen;
1037 		next = msgmaps[next].next;
1038 	}
1039 	if (next != -1)
1040 		panic("didn't use all the msg segments");
1041 
1042 	/*
1043 	 * We've got the message.  Unlock the msqid_ds.
1044 	 */
1045 
1046 	msqkptr->u.msg_perm.mode &= ~MSG_LOCKED;
1047 
1048 	/*
1049 	 * Make sure that the msqid_ds is still allocated.
1050 	 */
1051 
1052 	if (msqkptr->u.msg_qbytes == 0) {
1053 		msg_freehdr(msghdr);
1054 		wakeup(msqkptr);
1055 		error = EIDRM;
1056 		goto done3;
1057 	}
1058 
1059 #ifdef MAC
1060 	/*
1061 	 * Note: Since the task/thread allocates the msghdr and usually
1062 	 * primes it with its own MAC label, for a majority of policies, it
1063 	 * won't be necessary to check whether the msghdr has access
1064 	 * permissions to the msgq.  The mac_sysvmsq_check_msqsnd check would
1065 	 * suffice in that case.  However, this hook may be required where
1066 	 * individual policies derive a non-identical label for the msghdr
1067 	 * from the current thread label and may want to check the msghdr
1068 	 * enqueue permissions, along with read/write permissions to the
1069 	 * msgq.
1070 	 */
1071 	error = mac_sysvmsq_check_msgmsq(td->td_ucred, msghdr, msqkptr);
1072 	if (error != 0) {
1073 		msg_freehdr(msghdr);
1074 		wakeup(msqkptr);
1075 		goto done3;
1076 	}
1077 #endif
1078 
1079 	/*
1080 	 * Put the message into the queue
1081 	 */
1082 	if (msqkptr->u.msg_first == NULL) {
1083 		msqkptr->u.msg_first = msghdr;
1084 		msqkptr->u.msg_last = msghdr;
1085 	} else {
1086 		msqkptr->u.msg_last->msg_next = msghdr;
1087 		msqkptr->u.msg_last = msghdr;
1088 	}
1089 	msqkptr->u.msg_last->msg_next = NULL;
1090 
1091 	msqkptr->u.msg_cbytes += msghdr->msg_ts;
1092 	msqkptr->u.msg_qnum++;
1093 	msqkptr->u.msg_lspid = td->td_proc->p_pid;
1094 	msqkptr->u.msg_stime = time_second;
1095 
1096 	wakeup(msqkptr);
1097 	td->td_retval[0] = 0;
1098 done3:
1099 #ifdef RACCT
1100 	if (racct_enable && error != 0) {
1101 		PROC_LOCK(td->td_proc);
1102 		racct_sub(td->td_proc, RACCT_MSGQQUEUED, 1);
1103 		racct_sub(td->td_proc, RACCT_MSGQSIZE, saved_msgsz);
1104 		PROC_UNLOCK(td->td_proc);
1105 	}
1106 #endif
1107 done2:
1108 	mtx_unlock(&msq_mtx);
1109 	return (error);
1110 }
1111 
1112 int
1113 sys_msgsnd(struct thread *td, struct msgsnd_args *uap)
1114 {
1115 	int error;
1116 	long mtype;
1117 
1118 	DPRINTF(("call to msgsnd(%d, %p, %zu, %d)\n", uap->msqid, uap->msgp,
1119 	    uap->msgsz, uap->msgflg));
1120 
1121 	if ((error = copyin(uap->msgp, &mtype, sizeof(mtype))) != 0) {
1122 		DPRINTF(("error %d copying the message type\n", error));
1123 		return (error);
1124 	}
1125 	return (kern_msgsnd(td, uap->msqid,
1126 	    (const char *)uap->msgp + sizeof(mtype),
1127 	    uap->msgsz, uap->msgflg, mtype));
1128 }
1129 
1130 #ifndef _SYS_SYSPROTO_H_
1131 struct msgrcv_args {
1132 	int	msqid;
1133 	void	*msgp;
1134 	size_t	msgsz;
1135 	long	msgtyp;
1136 	int	msgflg;
1137 };
1138 #endif
1139 /* XXX msgp is actually mtext. */
1140 int
1141 kern_msgrcv(struct thread *td, int msqid, void *msgp, size_t msgsz, long msgtyp,
1142     int msgflg, long *mtype)
1143 {
1144 	size_t len;
1145 	struct msqid_kernel *msqkptr;
1146 	struct msg *msghdr;
1147 	struct prison *rpr;
1148 	int msqix, error = 0;
1149 	short next;
1150 
1151 	rpr = msg_find_prison(td->td_ucred);
1152 	if (rpr == NULL)
1153 		return (ENOSYS);
1154 
1155 	AUDIT_ARG_SVIPC_ID(msqid);
1156 	msqix = IPCID_TO_IX(msqid);
1157 
1158 	if (msqix < 0 || msqix >= msginfo.msgmni) {
1159 		DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix,
1160 		    msginfo.msgmni));
1161 		return (EINVAL);
1162 	}
1163 
1164 	msqkptr = &msqids[msqix];
1165 	mtx_lock(&msq_mtx);
1166 	AUDIT_ARG_SVIPC_PERM(&msqkptr->u.msg_perm);
1167 	if (msqkptr->u.msg_qbytes == 0) {
1168 		DPRINTF(("no such message queue id\n"));
1169 		error = EINVAL;
1170 		goto done2;
1171 	}
1172 	if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) {
1173 		DPRINTF(("wrong sequence number\n"));
1174 		error = EINVAL;
1175 		goto done2;
1176 	}
1177 
1178 	if ((error = msq_prison_cansee(rpr, msqkptr))) {
1179 		DPRINTF(("requester can't see prison\n"));
1180 		goto done2;
1181 	}
1182 
1183 	if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) {
1184 		DPRINTF(("requester doesn't have read access\n"));
1185 		goto done2;
1186 	}
1187 
1188 #ifdef MAC
1189 	error = mac_sysvmsq_check_msqrcv(td->td_ucred, msqkptr);
1190 	if (error != 0)
1191 		goto done2;
1192 #endif
1193 
1194 	msghdr = NULL;
1195 	while (msghdr == NULL) {
1196 		if (msgtyp == 0) {
1197 			msghdr = msqkptr->u.msg_first;
1198 			if (msghdr != NULL) {
1199 				if (msgsz < msghdr->msg_ts &&
1200 				    (msgflg & MSG_NOERROR) == 0) {
1201 					DPRINTF(("first message on the queue "
1202 					    "is too big (want %zu, got %d)\n",
1203 					    msgsz, msghdr->msg_ts));
1204 					error = E2BIG;
1205 					goto done2;
1206 				}
1207 #ifdef MAC
1208 				error = mac_sysvmsq_check_msgrcv(td->td_ucred,
1209 				    msghdr);
1210 				if (error != 0)
1211 					goto done2;
1212 #endif
1213 				if (msqkptr->u.msg_first == msqkptr->u.msg_last) {
1214 					msqkptr->u.msg_first = NULL;
1215 					msqkptr->u.msg_last = NULL;
1216 				} else {
1217 					msqkptr->u.msg_first = msghdr->msg_next;
1218 					if (msqkptr->u.msg_first == NULL)
1219 						panic("msg_first/last screwed up #1");
1220 				}
1221 			}
1222 		} else {
1223 			struct msg *previous;
1224 			struct msg **prev;
1225 
1226 			previous = NULL;
1227 			prev = &(msqkptr->u.msg_first);
1228 			while ((msghdr = *prev) != NULL) {
1229 				/*
1230 				 * Is this message's type an exact match or is
1231 				 * this message's type less than or equal to
1232 				 * the absolute value of a negative msgtyp?
1233 				 * Note that the second half of this test can
1234 				 * NEVER be true if msgtyp is positive since
1235 				 * msg_type is always positive!
1236 				 */
1237 
1238 				if (msgtyp == msghdr->msg_type ||
1239 				    msghdr->msg_type <= -msgtyp) {
1240 					DPRINTF(("found message type %ld, "
1241 					    "requested %ld\n",
1242 					    msghdr->msg_type, msgtyp));
1243 					if (msgsz < msghdr->msg_ts &&
1244 					    (msgflg & MSG_NOERROR) == 0) {
1245 						DPRINTF(("requested message "
1246 						    "on the queue is too big "
1247 						    "(want %zu, got %hu)\n",
1248 						    msgsz, msghdr->msg_ts));
1249 						error = E2BIG;
1250 						goto done2;
1251 					}
1252 #ifdef MAC
1253 					error = mac_sysvmsq_check_msgrcv(
1254 					    td->td_ucred, msghdr);
1255 					if (error != 0)
1256 						goto done2;
1257 #endif
1258 					*prev = msghdr->msg_next;
1259 					if (msghdr == msqkptr->u.msg_last) {
1260 						if (previous == NULL) {
1261 							if (prev !=
1262 							    &msqkptr->u.msg_first)
1263 								panic("msg_first/last screwed up #2");
1264 							msqkptr->u.msg_first =
1265 							    NULL;
1266 							msqkptr->u.msg_last =
1267 							    NULL;
1268 						} else {
1269 							if (prev ==
1270 							    &msqkptr->u.msg_first)
1271 								panic("msg_first/last screwed up #3");
1272 							msqkptr->u.msg_last =
1273 							    previous;
1274 						}
1275 					}
1276 					break;
1277 				}
1278 				previous = msghdr;
1279 				prev = &(msghdr->msg_next);
1280 			}
1281 		}
1282 
1283 		/*
1284 		 * We've either extracted the msghdr for the appropriate
1285 		 * message or there isn't one.
1286 		 * If there is one then bail out of this loop.
1287 		 */
1288 
1289 		if (msghdr != NULL)
1290 			break;
1291 
1292 		/*
1293 		 * Hmph!  No message found.  Does the user want to wait?
1294 		 */
1295 
1296 		if ((msgflg & IPC_NOWAIT) != 0) {
1297 			DPRINTF(("no appropriate message found (msgtyp=%ld)\n",
1298 			    msgtyp));
1299 			/* The SVID says to return ENOMSG. */
1300 			error = ENOMSG;
1301 			goto done2;
1302 		}
1303 
1304 		/*
1305 		 * Wait for something to happen
1306 		 */
1307 
1308 		DPRINTF(("msgrcv:  goodnight\n"));
1309 		error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH,
1310 		    "msgrcv", 0);
1311 		DPRINTF(("msgrcv:  good morning (error=%d)\n", error));
1312 
1313 		if (error != 0) {
1314 			DPRINTF(("msgrcv:  interrupted system call\n"));
1315 			error = EINTR;
1316 			goto done2;
1317 		}
1318 
1319 		/*
1320 		 * Make sure that the msq queue still exists
1321 		 */
1322 
1323 		if (msqkptr->u.msg_qbytes == 0 ||
1324 		    msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) {
1325 			DPRINTF(("msqid deleted\n"));
1326 			error = EIDRM;
1327 			goto done2;
1328 		}
1329 	}
1330 
1331 	/*
1332 	 * Return the message to the user.
1333 	 *
1334 	 * First, do the bookkeeping (before we risk being interrupted).
1335 	 */
1336 
1337 	msqkptr->u.msg_cbytes -= msghdr->msg_ts;
1338 	msqkptr->u.msg_qnum--;
1339 	msqkptr->u.msg_lrpid = td->td_proc->p_pid;
1340 	msqkptr->u.msg_rtime = time_second;
1341 
1342 	racct_sub_cred(msqkptr->cred, RACCT_MSGQQUEUED, 1);
1343 	racct_sub_cred(msqkptr->cred, RACCT_MSGQSIZE, msghdr->msg_ts);
1344 
1345 	/*
1346 	 * Make msgsz the actual amount that we'll be returning.
1347 	 * Note that this effectively truncates the message if it is too long
1348 	 * (since msgsz is never increased).
1349 	 */
1350 
1351 	DPRINTF(("found a message, msgsz=%zu, msg_ts=%hu\n", msgsz,
1352 	    msghdr->msg_ts));
1353 	if (msgsz > msghdr->msg_ts)
1354 		msgsz = msghdr->msg_ts;
1355 	*mtype = msghdr->msg_type;
1356 
1357 	/*
1358 	 * Return the segments to the user
1359 	 */
1360 
1361 	next = msghdr->msg_spot;
1362 	for (len = 0; len < msgsz; len += msginfo.msgssz) {
1363 		size_t tlen;
1364 
1365 		if (msgsz - len > msginfo.msgssz)
1366 			tlen = msginfo.msgssz;
1367 		else
1368 			tlen = msgsz - len;
1369 		if (next <= -1)
1370 			panic("next too low #3");
1371 		if (next >= msginfo.msgseg)
1372 			panic("next out of range #3");
1373 		mtx_unlock(&msq_mtx);
1374 		error = copyout(&msgpool[next * msginfo.msgssz], msgp, tlen);
1375 		mtx_lock(&msq_mtx);
1376 		if (error != 0) {
1377 			DPRINTF(("error (%d) copying out message segment\n",
1378 			    error));
1379 			msg_freehdr(msghdr);
1380 			wakeup(msqkptr);
1381 			goto done2;
1382 		}
1383 		msgp = (char *)msgp + tlen;
1384 		next = msgmaps[next].next;
1385 	}
1386 
1387 	/*
1388 	 * Done, return the actual number of bytes copied out.
1389 	 */
1390 
1391 	msg_freehdr(msghdr);
1392 	wakeup(msqkptr);
1393 	td->td_retval[0] = msgsz;
1394 done2:
1395 	mtx_unlock(&msq_mtx);
1396 	return (error);
1397 }
1398 
1399 int
1400 sys_msgrcv(struct thread *td, struct msgrcv_args *uap)
1401 {
1402 	int error;
1403 	long mtype;
1404 
1405 	DPRINTF(("call to msgrcv(%d, %p, %zu, %ld, %d)\n", uap->msqid,
1406 	    uap->msgp, uap->msgsz, uap->msgtyp, uap->msgflg));
1407 
1408 	if ((error = kern_msgrcv(td, uap->msqid,
1409 	    (char *)uap->msgp + sizeof(mtype), uap->msgsz,
1410 	    uap->msgtyp, uap->msgflg, &mtype)) != 0)
1411 		return (error);
1412 	if ((error = copyout(&mtype, uap->msgp, sizeof(mtype))) != 0)
1413 		DPRINTF(("error %d copying the message type\n", error));
1414 	return (error);
1415 }
1416 
1417 static int
1418 sysctl_msqids(SYSCTL_HANDLER_ARGS)
1419 {
1420 	struct msqid_kernel tmsqk;
1421 #ifdef COMPAT_FREEBSD32
1422 	struct msqid_kernel32 tmsqk32;
1423 #endif
1424 	struct prison *pr, *rpr;
1425 	void *outaddr;
1426 	size_t outsize;
1427 	int error, i;
1428 
1429 	pr = req->td->td_ucred->cr_prison;
1430 	rpr = msg_find_prison(req->td->td_ucred);
1431 	error = 0;
1432 	for (i = 0; i < msginfo.msgmni; i++) {
1433 		mtx_lock(&msq_mtx);
1434 		if (msqids[i].u.msg_qbytes == 0 || rpr == NULL ||
1435 		    msq_prison_cansee(rpr, &msqids[i]) != 0)
1436 			bzero(&tmsqk, sizeof(tmsqk));
1437 		else {
1438 			tmsqk = msqids[i];
1439 			if (tmsqk.cred->cr_prison != pr)
1440 				tmsqk.u.msg_perm.key = IPC_PRIVATE;
1441 		}
1442 		mtx_unlock(&msq_mtx);
1443 #ifdef COMPAT_FREEBSD32
1444 		if (SV_CURPROC_FLAG(SV_ILP32)) {
1445 			bzero(&tmsqk32, sizeof(tmsqk32));
1446 			freebsd32_ipcperm_out(&tmsqk.u.msg_perm,
1447 			    &tmsqk32.u.msg_perm);
1448 			/* Don't copy u.msg_first or u.msg_last */
1449 			CP(tmsqk, tmsqk32, u.msg_cbytes);
1450 			CP(tmsqk, tmsqk32, u.msg_qnum);
1451 			CP(tmsqk, tmsqk32, u.msg_qbytes);
1452 			CP(tmsqk, tmsqk32, u.msg_lspid);
1453 			CP(tmsqk, tmsqk32, u.msg_lrpid);
1454 			CP(tmsqk, tmsqk32, u.msg_stime);
1455 			CP(tmsqk, tmsqk32, u.msg_rtime);
1456 			CP(tmsqk, tmsqk32, u.msg_ctime);
1457 			/* Don't copy label or cred */
1458 			outaddr = &tmsqk32;
1459 			outsize = sizeof(tmsqk32);
1460 		} else
1461 #endif
1462 		{
1463 			/* Don't leak kernel pointers */
1464 			tmsqk.u.msg_first = NULL;
1465 			tmsqk.u.msg_last = NULL;
1466 			tmsqk.label = NULL;
1467 			tmsqk.cred = NULL;
1468 			/*
1469 			 * XXX: some padding also exists, but we take care to
1470 			 * allocate our pool of msqid_kernel structs with
1471 			 * zeroed memory so this should be OK.
1472 			 */
1473 			outaddr = &tmsqk;
1474 			outsize = sizeof(tmsqk);
1475 		}
1476 		error = SYSCTL_OUT(req, outaddr, outsize);
1477 		if (error != 0)
1478 			break;
1479 	}
1480 	return (error);
1481 }
1482 
1483 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmax, CTLFLAG_RD, &msginfo.msgmax, 0,
1484     "Maximum message size");
1485 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmni, CTLFLAG_RDTUN, &msginfo.msgmni, 0,
1486     "Number of message queue identifiers");
1487 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmnb, CTLFLAG_RDTUN, &msginfo.msgmnb, 0,
1488     "Maximum number of bytes in a queue");
1489 SYSCTL_INT(_kern_ipc, OID_AUTO, msgtql, CTLFLAG_RDTUN, &msginfo.msgtql, 0,
1490     "Maximum number of messages in the system");
1491 SYSCTL_INT(_kern_ipc, OID_AUTO, msgssz, CTLFLAG_RDTUN, &msginfo.msgssz, 0,
1492     "Size of a message segment");
1493 SYSCTL_INT(_kern_ipc, OID_AUTO, msgseg, CTLFLAG_RDTUN, &msginfo.msgseg, 0,
1494     "Number of message segments");
1495 SYSCTL_PROC(_kern_ipc, OID_AUTO, msqids,
1496     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1497     NULL, 0, sysctl_msqids, "",
1498     "Array of struct msqid_kernel for each potential message queue");
1499 
1500 static int
1501 msg_prison_check(void *obj, void *data)
1502 {
1503 	struct prison *pr = obj;
1504 	struct prison *prpr;
1505 	struct vfsoptlist *opts = data;
1506 	int error, jsys;
1507 
1508 	/*
1509 	 * sysvmsg is a jailsys integer.
1510 	 * It must be "disable" if the parent jail is disabled.
1511 	 */
1512 	error = vfs_copyopt(opts, "sysvmsg", &jsys, sizeof(jsys));
1513 	if (error != ENOENT) {
1514 		if (error != 0)
1515 			return (error);
1516 		switch (jsys) {
1517 		case JAIL_SYS_DISABLE:
1518 			break;
1519 		case JAIL_SYS_NEW:
1520 		case JAIL_SYS_INHERIT:
1521 			prison_lock(pr->pr_parent);
1522 			prpr = osd_jail_get(pr->pr_parent, msg_prison_slot);
1523 			prison_unlock(pr->pr_parent);
1524 			if (prpr == NULL)
1525 				return (EPERM);
1526 			break;
1527 		default:
1528 			return (EINVAL);
1529 		}
1530 	}
1531 
1532 	return (0);
1533 }
1534 
1535 static int
1536 msg_prison_set(void *obj, void *data)
1537 {
1538 	struct prison *pr = obj;
1539 	struct prison *tpr, *orpr, *nrpr, *trpr;
1540 	struct vfsoptlist *opts = data;
1541 	void *rsv;
1542 	int jsys, descend;
1543 
1544 	/*
1545 	 * sysvmsg controls which jail is the root of the associated msgs (this
1546 	 * jail or same as the parent), or if the feature is available at all.
1547 	 */
1548 	if (vfs_copyopt(opts, "sysvmsg", &jsys, sizeof(jsys)) == ENOENT)
1549 		jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0)
1550 		    ? JAIL_SYS_INHERIT
1551 		    : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0)
1552 		    ? JAIL_SYS_DISABLE
1553 		    : -1;
1554 	if (jsys == JAIL_SYS_DISABLE) {
1555 		prison_lock(pr);
1556 		orpr = osd_jail_get(pr, msg_prison_slot);
1557 		if (orpr != NULL)
1558 			osd_jail_del(pr, msg_prison_slot);
1559 		prison_unlock(pr);
1560 		if (orpr != NULL) {
1561 			if (orpr == pr)
1562 				msg_prison_cleanup(pr);
1563 			/* Disable all child jails as well. */
1564 			FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1565 				prison_lock(tpr);
1566 				trpr = osd_jail_get(tpr, msg_prison_slot);
1567 				if (trpr != NULL) {
1568 					osd_jail_del(tpr, msg_prison_slot);
1569 					prison_unlock(tpr);
1570 					if (trpr == tpr)
1571 						msg_prison_cleanup(tpr);
1572 				} else {
1573 					prison_unlock(tpr);
1574 					descend = 0;
1575 				}
1576 			}
1577 		}
1578 	} else if (jsys != -1) {
1579 		if (jsys == JAIL_SYS_NEW)
1580 			nrpr = pr;
1581 		else {
1582 			prison_lock(pr->pr_parent);
1583 			nrpr = osd_jail_get(pr->pr_parent, msg_prison_slot);
1584 			prison_unlock(pr->pr_parent);
1585 		}
1586 		rsv = osd_reserve(msg_prison_slot);
1587 		prison_lock(pr);
1588 		orpr = osd_jail_get(pr, msg_prison_slot);
1589 		if (orpr != nrpr)
1590 			(void)osd_jail_set_reserved(pr, msg_prison_slot, rsv,
1591 			    nrpr);
1592 		else
1593 			osd_free_reserved(rsv);
1594 		prison_unlock(pr);
1595 		if (orpr != nrpr) {
1596 			if (orpr == pr)
1597 				msg_prison_cleanup(pr);
1598 			if (orpr != NULL) {
1599 				/* Change child jails matching the old root, */
1600 				FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1601 					prison_lock(tpr);
1602 					trpr = osd_jail_get(tpr,
1603 					    msg_prison_slot);
1604 					if (trpr == orpr) {
1605 						(void)osd_jail_set(tpr,
1606 						    msg_prison_slot, nrpr);
1607 						prison_unlock(tpr);
1608 						if (trpr == tpr)
1609 							msg_prison_cleanup(tpr);
1610 					} else {
1611 						prison_unlock(tpr);
1612 						descend = 0;
1613 					}
1614 				}
1615 			}
1616 		}
1617 	}
1618 
1619 	return (0);
1620 }
1621 
1622 static int
1623 msg_prison_get(void *obj, void *data)
1624 {
1625 	struct prison *pr = obj;
1626 	struct prison *rpr;
1627 	struct vfsoptlist *opts = data;
1628 	int error, jsys;
1629 
1630 	/* Set sysvmsg based on the jail's root prison. */
1631 	prison_lock(pr);
1632 	rpr = osd_jail_get(pr, msg_prison_slot);
1633 	prison_unlock(pr);
1634 	jsys = rpr == NULL ? JAIL_SYS_DISABLE
1635 	    : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT;
1636 	error = vfs_setopt(opts, "sysvmsg", &jsys, sizeof(jsys));
1637 	if (error == ENOENT)
1638 		error = 0;
1639 	return (error);
1640 }
1641 
1642 static int
1643 msg_prison_remove(void *obj, void *data __unused)
1644 {
1645 	struct prison *pr = obj;
1646 	struct prison *rpr;
1647 
1648 	prison_lock(pr);
1649 	rpr = osd_jail_get(pr, msg_prison_slot);
1650 	prison_unlock(pr);
1651 	if (rpr == pr)
1652 		msg_prison_cleanup(pr);
1653 	return (0);
1654 }
1655 
1656 static void
1657 msg_prison_cleanup(struct prison *pr)
1658 {
1659 	struct msqid_kernel *msqkptr;
1660 	int i;
1661 
1662 	/* Remove any msqs that belong to this jail. */
1663 	mtx_lock(&msq_mtx);
1664 	for (i = 0; i < msginfo.msgmni; i++) {
1665 		msqkptr = &msqids[i];
1666 		if (msqkptr->u.msg_qbytes != 0 &&
1667 		    msqkptr->cred != NULL && msqkptr->cred->cr_prison == pr)
1668 			msq_remove(msqkptr);
1669 	}
1670 	mtx_unlock(&msq_mtx);
1671 }
1672 
1673 SYSCTL_JAIL_PARAM_SYS_NODE(sysvmsg, CTLFLAG_RW, "SYSV message queues");
1674 
1675 #ifdef COMPAT_FREEBSD32
1676 int
1677 freebsd32_msgsys(struct thread *td, struct freebsd32_msgsys_args *uap)
1678 {
1679 
1680 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1681     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1682 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1683 	switch (uap->which) {
1684 	case 0:
1685 		return (freebsd7_freebsd32_msgctl(td,
1686 		    (struct freebsd7_freebsd32_msgctl_args *)&uap->a2));
1687 	case 2:
1688 		return (freebsd32_msgsnd(td,
1689 		    (struct freebsd32_msgsnd_args *)&uap->a2));
1690 	case 3:
1691 		return (freebsd32_msgrcv(td,
1692 		    (struct freebsd32_msgrcv_args *)&uap->a2));
1693 	default:
1694 		return (sys_msgsys(td, (struct msgsys_args *)uap));
1695 	}
1696 #else
1697 	return (nosys(td, NULL));
1698 #endif
1699 }
1700 
1701 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1702     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1703 int
1704 freebsd7_freebsd32_msgctl(struct thread *td,
1705     struct freebsd7_freebsd32_msgctl_args *uap)
1706 {
1707 	struct msqid_ds msqbuf;
1708 	struct msqid_ds32_old msqbuf32;
1709 	int error;
1710 
1711 	if (uap->cmd == IPC_SET) {
1712 		error = copyin(uap->buf, &msqbuf32, sizeof(msqbuf32));
1713 		if (error)
1714 			return (error);
1715 		freebsd32_ipcperm_old_in(&msqbuf32.msg_perm, &msqbuf.msg_perm);
1716 		PTRIN_CP(msqbuf32, msqbuf, msg_first);
1717 		PTRIN_CP(msqbuf32, msqbuf, msg_last);
1718 		CP(msqbuf32, msqbuf, msg_cbytes);
1719 		CP(msqbuf32, msqbuf, msg_qnum);
1720 		CP(msqbuf32, msqbuf, msg_qbytes);
1721 		CP(msqbuf32, msqbuf, msg_lspid);
1722 		CP(msqbuf32, msqbuf, msg_lrpid);
1723 		CP(msqbuf32, msqbuf, msg_stime);
1724 		CP(msqbuf32, msqbuf, msg_rtime);
1725 		CP(msqbuf32, msqbuf, msg_ctime);
1726 	}
1727 	error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf);
1728 	if (error)
1729 		return (error);
1730 	if (uap->cmd == IPC_STAT) {
1731 		bzero(&msqbuf32, sizeof(msqbuf32));
1732 		freebsd32_ipcperm_old_out(&msqbuf.msg_perm, &msqbuf32.msg_perm);
1733 		PTROUT_CP(msqbuf, msqbuf32, msg_first);
1734 		PTROUT_CP(msqbuf, msqbuf32, msg_last);
1735 		CP(msqbuf, msqbuf32, msg_cbytes);
1736 		CP(msqbuf, msqbuf32, msg_qnum);
1737 		CP(msqbuf, msqbuf32, msg_qbytes);
1738 		CP(msqbuf, msqbuf32, msg_lspid);
1739 		CP(msqbuf, msqbuf32, msg_lrpid);
1740 		CP(msqbuf, msqbuf32, msg_stime);
1741 		CP(msqbuf, msqbuf32, msg_rtime);
1742 		CP(msqbuf, msqbuf32, msg_ctime);
1743 		error = copyout(&msqbuf32, uap->buf, sizeof(struct msqid_ds32));
1744 	}
1745 	return (error);
1746 }
1747 #endif
1748 
1749 int
1750 freebsd32_msgctl(struct thread *td, struct freebsd32_msgctl_args *uap)
1751 {
1752 	struct msqid_ds msqbuf;
1753 	struct msqid_ds32 msqbuf32;
1754 	int error;
1755 
1756 	if (uap->cmd == IPC_SET) {
1757 		error = copyin(uap->buf, &msqbuf32, sizeof(msqbuf32));
1758 		if (error)
1759 			return (error);
1760 		freebsd32_ipcperm_in(&msqbuf32.msg_perm, &msqbuf.msg_perm);
1761 		PTRIN_CP(msqbuf32, msqbuf, msg_first);
1762 		PTRIN_CP(msqbuf32, msqbuf, msg_last);
1763 		CP(msqbuf32, msqbuf, msg_cbytes);
1764 		CP(msqbuf32, msqbuf, msg_qnum);
1765 		CP(msqbuf32, msqbuf, msg_qbytes);
1766 		CP(msqbuf32, msqbuf, msg_lspid);
1767 		CP(msqbuf32, msqbuf, msg_lrpid);
1768 		CP(msqbuf32, msqbuf, msg_stime);
1769 		CP(msqbuf32, msqbuf, msg_rtime);
1770 		CP(msqbuf32, msqbuf, msg_ctime);
1771 	}
1772 	error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf);
1773 	if (error)
1774 		return (error);
1775 	if (uap->cmd == IPC_STAT) {
1776 		freebsd32_ipcperm_out(&msqbuf.msg_perm, &msqbuf32.msg_perm);
1777 		PTROUT_CP(msqbuf, msqbuf32, msg_first);
1778 		PTROUT_CP(msqbuf, msqbuf32, msg_last);
1779 		CP(msqbuf, msqbuf32, msg_cbytes);
1780 		CP(msqbuf, msqbuf32, msg_qnum);
1781 		CP(msqbuf, msqbuf32, msg_qbytes);
1782 		CP(msqbuf, msqbuf32, msg_lspid);
1783 		CP(msqbuf, msqbuf32, msg_lrpid);
1784 		CP(msqbuf, msqbuf32, msg_stime);
1785 		CP(msqbuf, msqbuf32, msg_rtime);
1786 		CP(msqbuf, msqbuf32, msg_ctime);
1787 		error = copyout(&msqbuf32, uap->buf, sizeof(struct msqid_ds32));
1788 	}
1789 	return (error);
1790 }
1791 
1792 int
1793 freebsd32_msgsnd(struct thread *td, struct freebsd32_msgsnd_args *uap)
1794 {
1795 	const void *msgp;
1796 	long mtype;
1797 	int32_t mtype32;
1798 	int error;
1799 
1800 	msgp = PTRIN(uap->msgp);
1801 	if ((error = copyin(msgp, &mtype32, sizeof(mtype32))) != 0)
1802 		return (error);
1803 	mtype = mtype32;
1804 	return (kern_msgsnd(td, uap->msqid,
1805 	    (const char *)msgp + sizeof(mtype32),
1806 	    uap->msgsz, uap->msgflg, mtype));
1807 }
1808 
1809 int
1810 freebsd32_msgrcv(struct thread *td, struct freebsd32_msgrcv_args *uap)
1811 {
1812 	void *msgp;
1813 	long mtype;
1814 	int32_t mtype32;
1815 	int error;
1816 
1817 	msgp = PTRIN(uap->msgp);
1818 	if ((error = kern_msgrcv(td, uap->msqid,
1819 	    (char *)msgp + sizeof(mtype32), uap->msgsz,
1820 	    uap->msgtyp, uap->msgflg, &mtype)) != 0)
1821 		return (error);
1822 	mtype32 = (int32_t)mtype;
1823 	return (copyout(&mtype32, msgp, sizeof(mtype32)));
1824 }
1825 #endif
1826 
1827 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1828     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1829 
1830 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1831 static sy_call_t *msgcalls[] = {
1832 	(sy_call_t *)freebsd7_msgctl, (sy_call_t *)sys_msgget,
1833 	(sy_call_t *)sys_msgsnd, (sy_call_t *)sys_msgrcv
1834 };
1835 
1836 /*
1837  * Entry point for all MSG calls.
1838  *
1839  * XXX actually varargs.
1840  * struct msgsys_args {
1841  *		int	which;
1842  *		int	a2;
1843  *		int	a3;
1844  *		int	a4;
1845  *		int	a5;
1846  *		int	a6;
1847  *	} *uap;
1848  */
1849 int
1850 sys_msgsys(struct thread *td, struct msgsys_args *uap)
1851 {
1852 	int error;
1853 
1854 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1855 	if (uap->which < 0 || uap->which >= nitems(msgcalls))
1856 		return (EINVAL);
1857 	error = (*msgcalls[uap->which])(td, &uap->a2);
1858 	return (error);
1859 }
1860 
1861 #ifndef CP
1862 #define CP(src, dst, fld)	do { (dst).fld = (src).fld; } while (0)
1863 #endif
1864 
1865 #ifndef _SYS_SYSPROTO_H_
1866 struct freebsd7_msgctl_args {
1867 	int	msqid;
1868 	int	cmd;
1869 	struct	msqid_ds_old *buf;
1870 };
1871 #endif
1872 int
1873 freebsd7_msgctl(struct thread *td, struct freebsd7_msgctl_args *uap)
1874 {
1875 	struct msqid_ds_old msqold;
1876 	struct msqid_ds msqbuf;
1877 	int error;
1878 
1879 	DPRINTF(("call to freebsd7_msgctl(%d, %d, %p)\n", uap->msqid, uap->cmd,
1880 	    uap->buf));
1881 	if (uap->cmd == IPC_SET) {
1882 		error = copyin(uap->buf, &msqold, sizeof(msqold));
1883 		if (error)
1884 			return (error);
1885 		ipcperm_old2new(&msqold.msg_perm, &msqbuf.msg_perm);
1886 		CP(msqold, msqbuf, msg_first);
1887 		CP(msqold, msqbuf, msg_last);
1888 		CP(msqold, msqbuf, msg_cbytes);
1889 		CP(msqold, msqbuf, msg_qnum);
1890 		CP(msqold, msqbuf, msg_qbytes);
1891 		CP(msqold, msqbuf, msg_lspid);
1892 		CP(msqold, msqbuf, msg_lrpid);
1893 		CP(msqold, msqbuf, msg_stime);
1894 		CP(msqold, msqbuf, msg_rtime);
1895 		CP(msqold, msqbuf, msg_ctime);
1896 	}
1897 	error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf);
1898 	if (error)
1899 		return (error);
1900 	if (uap->cmd == IPC_STAT) {
1901 		bzero(&msqold, sizeof(msqold));
1902 		ipcperm_new2old(&msqbuf.msg_perm, &msqold.msg_perm);
1903 		CP(msqbuf, msqold, msg_first);
1904 		CP(msqbuf, msqold, msg_last);
1905 		CP(msqbuf, msqold, msg_cbytes);
1906 		CP(msqbuf, msqold, msg_qnum);
1907 		CP(msqbuf, msqold, msg_qbytes);
1908 		CP(msqbuf, msqold, msg_lspid);
1909 		CP(msqbuf, msqold, msg_lrpid);
1910 		CP(msqbuf, msqold, msg_stime);
1911 		CP(msqbuf, msqold, msg_rtime);
1912 		CP(msqbuf, msqold, msg_ctime);
1913 		error = copyout(&msqold, uap->buf, sizeof(struct msqid_ds_old));
1914 	}
1915 	return (error);
1916 }
1917 
1918 #undef CP
1919 
1920 #endif	/* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 ||
1921 	   COMPAT_FREEBSD7 */
1922