xref: /freebsd/sys/kern/sysv_msg.c (revision d7d962ead0b6e5e8a39202d0590022082bf5bfb6)
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_sysvipc.h"
62 
63 #include <sys/param.h>
64 #include <sys/systm.h>
65 #include <sys/sysproto.h>
66 #include <sys/kernel.h>
67 #include <sys/priv.h>
68 #include <sys/proc.h>
69 #include <sys/lock.h>
70 #include <sys/mutex.h>
71 #include <sys/module.h>
72 #include <sys/mount.h>
73 #include <sys/msg.h>
74 #include <sys/racct.h>
75 #include <sys/sx.h>
76 #include <sys/syscall.h>
77 #include <sys/syscallsubr.h>
78 #include <sys/sysent.h>
79 #include <sys/sysctl.h>
80 #include <sys/malloc.h>
81 #include <sys/jail.h>
82 
83 #include <security/audit/audit.h>
84 #include <security/mac/mac_framework.h>
85 
86 FEATURE(sysv_msg, "System V message queues support");
87 
88 static MALLOC_DEFINE(M_MSG, "msg", "SVID compatible message queues");
89 
90 static int msginit(void);
91 static int msgunload(void);
92 static int sysvmsg_modload(struct module *, int, void *);
93 static void msq_remove(struct msqid_kernel *);
94 static struct prison *msg_find_prison(struct ucred *);
95 static int msq_prison_cansee(struct prison *, struct msqid_kernel *);
96 static int msg_prison_check(void *, void *);
97 static int msg_prison_set(void *, void *);
98 static int msg_prison_get(void *, void *);
99 static int msg_prison_remove(void *, void *);
100 static void msg_prison_cleanup(struct prison *);
101 
102 #ifdef MSG_DEBUG
103 #define DPRINTF(a)	printf a
104 #else
105 #define DPRINTF(a)	(void)0
106 #endif
107 
108 static void msg_freehdr(struct msg *msghdr);
109 
110 #ifndef MSGSSZ
111 #define MSGSSZ	8		/* Each segment must be 2^N long */
112 #endif
113 #ifndef MSGSEG
114 #define MSGSEG	2048		/* must be less than 32767 */
115 #endif
116 #define MSGMAX	(MSGSSZ*MSGSEG)
117 #ifndef MSGMNB
118 #define MSGMNB	2048		/* max # of bytes in a queue */
119 #endif
120 #ifndef MSGMNI
121 #define MSGMNI	40
122 #endif
123 #ifndef MSGTQL
124 #define MSGTQL	40
125 #endif
126 
127 /*
128  * Based on the configuration parameters described in an SVR2 (yes, two)
129  * config(1m) man page.
130  *
131  * Each message is broken up and stored in segments that are msgssz bytes
132  * long.  For efficiency reasons, this should be a power of two.  Also,
133  * it doesn't make sense if it is less than 8 or greater than about 256.
134  * Consequently, msginit in kern/sysv_msg.c checks that msgssz is a power of
135  * two between 8 and 1024 inclusive (and panic's if it isn't).
136  */
137 struct msginfo msginfo = {
138                 MSGMAX,         /* max chars in a message */
139                 MSGMNI,         /* # of message queue identifiers */
140                 MSGMNB,         /* max chars in a queue */
141                 MSGTQL,         /* max messages in system */
142                 MSGSSZ,         /* size of a message segment */
143                 		/* (must be small power of 2 greater than 4) */
144                 MSGSEG          /* number of message segments */
145 };
146 
147 /*
148  * macros to convert between msqid_ds's and msqid's.
149  * (specific to this implementation)
150  */
151 #define MSQID(ix,ds)	((ix) & 0xffff | (((ds).msg_perm.seq << 16) & 0xffff0000))
152 #define MSQID_IX(id)	((id) & 0xffff)
153 #define MSQID_SEQ(id)	(((id) >> 16) & 0xffff)
154 
155 /*
156  * The rest of this file is specific to this particular implementation.
157  */
158 
159 struct msgmap {
160 	short	next;		/* next segment in buffer */
161     				/* -1 -> available */
162     				/* 0..(MSGSEG-1) -> index of next segment */
163 };
164 
165 #define MSG_LOCKED	01000	/* Is this msqid_ds locked? */
166 
167 static int nfree_msgmaps;	/* # of free map entries */
168 static short free_msgmaps;	/* head of linked list of free map entries */
169 static struct msg *free_msghdrs;/* list of free msg headers */
170 static char *msgpool;		/* MSGMAX byte long msg buffer pool */
171 static struct msgmap *msgmaps;	/* MSGSEG msgmap structures */
172 static struct msg *msghdrs;	/* MSGTQL msg headers */
173 static struct msqid_kernel *msqids;	/* MSGMNI msqid_kernel struct's */
174 static struct mtx msq_mtx;	/* global mutex for message queues. */
175 static unsigned msg_prison_slot;/* prison OSD slot */
176 
177 static struct syscall_helper_data msg_syscalls[] = {
178 	SYSCALL_INIT_HELPER(msgctl),
179 	SYSCALL_INIT_HELPER(msgget),
180 	SYSCALL_INIT_HELPER(msgsnd),
181 	SYSCALL_INIT_HELPER(msgrcv),
182 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
183     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
184 	SYSCALL_INIT_HELPER(msgsys),
185 	SYSCALL_INIT_HELPER_COMPAT(freebsd7_msgctl),
186 #endif
187 	SYSCALL_INIT_LAST
188 };
189 
190 #ifdef COMPAT_FREEBSD32
191 #include <compat/freebsd32/freebsd32.h>
192 #include <compat/freebsd32/freebsd32_ipc.h>
193 #include <compat/freebsd32/freebsd32_proto.h>
194 #include <compat/freebsd32/freebsd32_signal.h>
195 #include <compat/freebsd32/freebsd32_syscall.h>
196 #include <compat/freebsd32/freebsd32_util.h>
197 
198 static struct syscall_helper_data msg32_syscalls[] = {
199 	SYSCALL32_INIT_HELPER(freebsd32_msgctl),
200 	SYSCALL32_INIT_HELPER(freebsd32_msgsnd),
201 	SYSCALL32_INIT_HELPER(freebsd32_msgrcv),
202 	SYSCALL32_INIT_HELPER_COMPAT(msgget),
203 	SYSCALL32_INIT_HELPER(freebsd32_msgsys),
204 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
205     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
206 	SYSCALL32_INIT_HELPER(freebsd7_freebsd32_msgctl),
207 #endif
208 	SYSCALL_INIT_LAST
209 };
210 #endif
211 
212 static int
213 msginit()
214 {
215 	struct prison *pr;
216 	void **rsv;
217 	int i, error;
218 	osd_method_t methods[PR_MAXMETHOD] = {
219 	    [PR_METHOD_CHECK] =		msg_prison_check,
220 	    [PR_METHOD_SET] =		msg_prison_set,
221 	    [PR_METHOD_GET] =		msg_prison_get,
222 	    [PR_METHOD_REMOVE] =	msg_prison_remove,
223 	};
224 
225 	msginfo.msgmax = msginfo.msgseg * msginfo.msgssz;
226 	msgpool = malloc(msginfo.msgmax, M_MSG, M_WAITOK);
227 	msgmaps = malloc(sizeof(struct msgmap) * msginfo.msgseg, M_MSG, M_WAITOK);
228 	msghdrs = malloc(sizeof(struct msg) * msginfo.msgtql, M_MSG, M_WAITOK);
229 	msqids = malloc(sizeof(struct msqid_kernel) * msginfo.msgmni, M_MSG,
230 	    M_WAITOK | M_ZERO);
231 
232 	/*
233 	 * msginfo.msgssz should be a power of two for efficiency reasons.
234 	 * It is also pretty silly if msginfo.msgssz is less than 8
235 	 * or greater than about 256 so ...
236 	 */
237 
238 	i = 8;
239 	while (i < 1024 && i != msginfo.msgssz)
240 		i <<= 1;
241     	if (i != msginfo.msgssz) {
242 		DPRINTF(("msginfo.msgssz=%d (0x%x)\n", msginfo.msgssz,
243 		    msginfo.msgssz));
244 		panic("msginfo.msgssz not a small power of 2");
245 	}
246 
247 	if (msginfo.msgseg > 32767) {
248 		DPRINTF(("msginfo.msgseg=%d\n", msginfo.msgseg));
249 		panic("msginfo.msgseg > 32767");
250 	}
251 
252 	for (i = 0; i < msginfo.msgseg; i++) {
253 		if (i > 0)
254 			msgmaps[i-1].next = i;
255 		msgmaps[i].next = -1;	/* implies entry is available */
256 	}
257 	free_msgmaps = 0;
258 	nfree_msgmaps = msginfo.msgseg;
259 
260 	for (i = 0; i < msginfo.msgtql; i++) {
261 		msghdrs[i].msg_type = 0;
262 		if (i > 0)
263 			msghdrs[i-1].msg_next = &msghdrs[i];
264 		msghdrs[i].msg_next = NULL;
265 #ifdef MAC
266 		mac_sysvmsg_init(&msghdrs[i]);
267 #endif
268     	}
269 	free_msghdrs = &msghdrs[0];
270 
271 	for (i = 0; i < msginfo.msgmni; i++) {
272 		msqids[i].u.msg_qbytes = 0;	/* implies entry is available */
273 		msqids[i].u.msg_perm.seq = 0;	/* reset to a known value */
274 		msqids[i].u.msg_perm.mode = 0;
275 #ifdef MAC
276 		mac_sysvmsq_init(&msqids[i]);
277 #endif
278 	}
279 	mtx_init(&msq_mtx, "msq", NULL, MTX_DEF);
280 
281 	/* Set current prisons according to their allow.sysvipc. */
282 	msg_prison_slot = osd_jail_register(NULL, methods);
283 	rsv = osd_reserve(msg_prison_slot);
284 	prison_lock(&prison0);
285 	(void)osd_jail_set_reserved(&prison0, msg_prison_slot, rsv, &prison0);
286 	prison_unlock(&prison0);
287 	rsv = NULL;
288 	sx_slock(&allprison_lock);
289 	TAILQ_FOREACH(pr, &allprison, pr_list) {
290 		if (rsv == NULL)
291 			rsv = osd_reserve(msg_prison_slot);
292 		prison_lock(pr);
293 		if (pr->pr_allow & PR_ALLOW_SYSVIPC) {
294 			(void)osd_jail_set_reserved(pr, msg_prison_slot, rsv,
295 			    &prison0);
296 			rsv = NULL;
297 		}
298 		prison_unlock(pr);
299 	}
300 	if (rsv != NULL)
301 		osd_free_reserved(rsv);
302 	sx_sunlock(&allprison_lock);
303 
304 	error = syscall_helper_register(msg_syscalls, SY_THR_STATIC_KLD);
305 	if (error != 0)
306 		return (error);
307 #ifdef COMPAT_FREEBSD32
308 	error = syscall32_helper_register(msg32_syscalls, SY_THR_STATIC_KLD);
309 	if (error != 0)
310 		return (error);
311 #endif
312 	return (0);
313 }
314 
315 static int
316 msgunload()
317 {
318 	struct msqid_kernel *msqkptr;
319 	int msqid;
320 #ifdef MAC
321 	int i;
322 #endif
323 
324 	syscall_helper_unregister(msg_syscalls);
325 #ifdef COMPAT_FREEBSD32
326 	syscall32_helper_unregister(msg32_syscalls);
327 #endif
328 
329 	for (msqid = 0; msqid < msginfo.msgmni; msqid++) {
330 		msqkptr = &msqids[msqid];
331 		if (msqkptr->u.msg_qbytes != 0 ||
332 		    (msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0)
333 			break;
334 	}
335 	if (msqid != msginfo.msgmni)
336 		return (EBUSY);
337 
338 	if (msg_prison_slot != 0)
339 		osd_jail_deregister(msg_prison_slot);
340 #ifdef MAC
341 	for (i = 0; i < msginfo.msgtql; i++)
342 		mac_sysvmsg_destroy(&msghdrs[i]);
343 	for (msqid = 0; msqid < msginfo.msgmni; msqid++)
344 		mac_sysvmsq_destroy(&msqids[msqid]);
345 #endif
346 	free(msgpool, M_MSG);
347 	free(msgmaps, M_MSG);
348 	free(msghdrs, M_MSG);
349 	free(msqids, M_MSG);
350 	mtx_destroy(&msq_mtx);
351 	return (0);
352 }
353 
354 static int
355 sysvmsg_modload(struct module *module, int cmd, void *arg)
356 {
357 	int error = 0;
358 
359 	switch (cmd) {
360 	case MOD_LOAD:
361 		error = msginit();
362 		if (error != 0)
363 			msgunload();
364 		break;
365 	case MOD_UNLOAD:
366 		error = msgunload();
367 		break;
368 	case MOD_SHUTDOWN:
369 		break;
370 	default:
371 		error = EINVAL;
372 		break;
373 	}
374 	return (error);
375 }
376 
377 static moduledata_t sysvmsg_mod = {
378 	"sysvmsg",
379 	&sysvmsg_modload,
380 	NULL
381 };
382 
383 DECLARE_MODULE(sysvmsg, sysvmsg_mod, SI_SUB_SYSV_MSG, SI_ORDER_FIRST);
384 MODULE_VERSION(sysvmsg, 1);
385 
386 static void
387 msg_freehdr(struct msg *msghdr)
388 {
389 	while (msghdr->msg_ts > 0) {
390 		short next;
391 		if (msghdr->msg_spot < 0 || msghdr->msg_spot >= msginfo.msgseg)
392 			panic("msghdr->msg_spot out of range");
393 		next = msgmaps[msghdr->msg_spot].next;
394 		msgmaps[msghdr->msg_spot].next = free_msgmaps;
395 		free_msgmaps = msghdr->msg_spot;
396 		nfree_msgmaps++;
397 		msghdr->msg_spot = next;
398 		if (msghdr->msg_ts >= msginfo.msgssz)
399 			msghdr->msg_ts -= msginfo.msgssz;
400 		else
401 			msghdr->msg_ts = 0;
402 	}
403 	if (msghdr->msg_spot != -1)
404 		panic("msghdr->msg_spot != -1");
405 	msghdr->msg_next = free_msghdrs;
406 	free_msghdrs = msghdr;
407 #ifdef MAC
408 	mac_sysvmsg_cleanup(msghdr);
409 #endif
410 }
411 
412 static void
413 msq_remove(struct msqid_kernel *msqkptr)
414 {
415 	struct msg *msghdr;
416 
417 	racct_sub_cred(msqkptr->cred, RACCT_NMSGQ, 1);
418 	racct_sub_cred(msqkptr->cred, RACCT_MSGQQUEUED, msqkptr->u.msg_qnum);
419 	racct_sub_cred(msqkptr->cred, RACCT_MSGQSIZE, msqkptr->u.msg_cbytes);
420 	crfree(msqkptr->cred);
421 	msqkptr->cred = NULL;
422 
423 	/* Free the message headers */
424 	msghdr = msqkptr->u.__msg_first;
425 	while (msghdr != NULL) {
426 		struct msg *msghdr_tmp;
427 
428 		/* Free the segments of each message */
429 		msqkptr->u.msg_cbytes -= msghdr->msg_ts;
430 		msqkptr->u.msg_qnum--;
431 		msghdr_tmp = msghdr;
432 		msghdr = msghdr->msg_next;
433 		msg_freehdr(msghdr_tmp);
434 	}
435 
436 	if (msqkptr->u.msg_cbytes != 0)
437 		panic("msg_cbytes is screwed up");
438 	if (msqkptr->u.msg_qnum != 0)
439 		panic("msg_qnum is screwed up");
440 
441 	msqkptr->u.msg_qbytes = 0;	/* Mark it as free */
442 
443 #ifdef MAC
444 	mac_sysvmsq_cleanup(msqkptr);
445 #endif
446 
447 	wakeup(msqkptr);
448 }
449 
450 static struct prison *
451 msg_find_prison(struct ucred *cred)
452 {
453 	struct prison *pr, *rpr;
454 
455 	pr = cred->cr_prison;
456 	prison_lock(pr);
457 	rpr = osd_jail_get(pr, msg_prison_slot);
458 	prison_unlock(pr);
459 	return rpr;
460 }
461 
462 static int
463 msq_prison_cansee(struct prison *rpr, struct msqid_kernel *msqkptr)
464 {
465 
466 	if (msqkptr->cred == NULL ||
467 	    !(rpr == msqkptr->cred->cr_prison ||
468 	      prison_ischild(rpr, msqkptr->cred->cr_prison)))
469 		return (EINVAL);
470 	return (0);
471 }
472 
473 #ifndef _SYS_SYSPROTO_H_
474 struct msgctl_args {
475 	int	msqid;
476 	int	cmd;
477 	struct	msqid_ds *buf;
478 };
479 #endif
480 int
481 sys_msgctl(struct thread *td, struct msgctl_args *uap)
482 {
483 	int msqid = uap->msqid;
484 	int cmd = uap->cmd;
485 	struct msqid_ds msqbuf;
486 	int error;
487 
488 	DPRINTF(("call to msgctl(%d, %d, %p)\n", msqid, cmd, uap->buf));
489 	if (cmd == IPC_SET &&
490 	    (error = copyin(uap->buf, &msqbuf, sizeof(msqbuf))) != 0)
491 		return (error);
492 	error = kern_msgctl(td, msqid, cmd, &msqbuf);
493 	if (cmd == IPC_STAT && error == 0)
494 		error = copyout(&msqbuf, uap->buf, sizeof(struct msqid_ds));
495 	return (error);
496 }
497 
498 int
499 kern_msgctl(struct thread *td, int msqid, int cmd, struct msqid_ds *msqbuf)
500 {
501 	int rval, error, msqix;
502 	struct msqid_kernel *msqkptr;
503 	struct prison *rpr;
504 
505 	rpr = msg_find_prison(td->td_ucred);
506 	if (rpr == NULL)
507 		return (ENOSYS);
508 
509 	AUDIT_ARG_SVIPC_CMD(cmd);
510 	AUDIT_ARG_SVIPC_ID(msqid);
511 	msqix = IPCID_TO_IX(msqid);
512 
513 	if (msqix < 0 || msqix >= msginfo.msgmni) {
514 		DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix,
515 		    msginfo.msgmni));
516 		return (EINVAL);
517 	}
518 
519 	msqkptr = &msqids[msqix];
520 
521 	mtx_lock(&msq_mtx);
522 	if (msqkptr->u.msg_qbytes == 0) {
523 		DPRINTF(("no such msqid\n"));
524 		error = EINVAL;
525 		goto done2;
526 	}
527 	if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) {
528 		DPRINTF(("wrong sequence number\n"));
529 		error = EINVAL;
530 		goto done2;
531 	}
532 
533 	error = msq_prison_cansee(rpr, msqkptr);
534 	if (error != 0) {
535 		DPRINTF(("requester can't see prison\n"));
536 		goto done2;
537 	}
538 
539 #ifdef MAC
540 	error = mac_sysvmsq_check_msqctl(td->td_ucred, msqkptr, cmd);
541 	if (error != 0)
542 		goto done2;
543 #endif
544 
545 	error = 0;
546 	rval = 0;
547 
548 	switch (cmd) {
549 	case IPC_RMID:
550 	{
551 #ifdef MAC
552 		struct msg *msghdr;
553 #endif
554 		if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M)))
555 			goto done2;
556 
557 #ifdef MAC
558 		/*
559 		 * Check that the thread has MAC access permissions to
560 		 * individual msghdrs.  Note: We need to do this in a
561 		 * separate loop because the actual loop alters the
562 		 * msq/msghdr info as it progresses, and there is no going
563 		 * back if half the way through we discover that the
564 		 * thread cannot free a certain msghdr.  The msq will get
565 		 * into an inconsistent state.
566 		 */
567 		for (msghdr = msqkptr->u.__msg_first; msghdr != NULL;
568 		    msghdr = msghdr->msg_next) {
569 			error = mac_sysvmsq_check_msgrmid(td->td_ucred, msghdr);
570 			if (error != 0)
571 				goto done2;
572 		}
573 #endif
574 
575 		msq_remove(msqkptr);
576 	}
577 
578 		break;
579 
580 	case IPC_SET:
581 		AUDIT_ARG_SVIPC_PERM(&msqbuf->msg_perm);
582 		if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M)))
583 			goto done2;
584 		if (msqbuf->msg_qbytes > msqkptr->u.msg_qbytes) {
585 			error = priv_check(td, PRIV_IPC_MSGSIZE);
586 			if (error)
587 				goto done2;
588 		}
589 		if (msqbuf->msg_qbytes > msginfo.msgmnb) {
590 			DPRINTF(("can't increase msg_qbytes beyond %d"
591 			    "(truncating)\n", msginfo.msgmnb));
592 			msqbuf->msg_qbytes = msginfo.msgmnb;	/* silently restrict qbytes to system limit */
593 		}
594 		if (msqbuf->msg_qbytes == 0) {
595 			DPRINTF(("can't reduce msg_qbytes to 0\n"));
596 			error = EINVAL;		/* non-standard errno! */
597 			goto done2;
598 		}
599 		msqkptr->u.msg_perm.uid = msqbuf->msg_perm.uid;	/* change the owner */
600 		msqkptr->u.msg_perm.gid = msqbuf->msg_perm.gid;	/* change the owner */
601 		msqkptr->u.msg_perm.mode = (msqkptr->u.msg_perm.mode & ~0777) |
602 		    (msqbuf->msg_perm.mode & 0777);
603 		msqkptr->u.msg_qbytes = msqbuf->msg_qbytes;
604 		msqkptr->u.msg_ctime = time_second;
605 		break;
606 
607 	case IPC_STAT:
608 		if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) {
609 			DPRINTF(("requester doesn't have read access\n"));
610 			goto done2;
611 		}
612 		*msqbuf = msqkptr->u;
613 		if (td->td_ucred->cr_prison != msqkptr->cred->cr_prison)
614 			msqbuf->msg_perm.key = IPC_PRIVATE;
615 
616 		/*
617 		 * Try to hide the fact that the structure layout is shared by
618 		 * both the kernel and userland.  These pointers are not useful
619 		 * to userspace.
620 		 */
621 		msqbuf->__msg_first = msqbuf->__msg_last = NULL;
622 		break;
623 
624 	default:
625 		DPRINTF(("invalid command %d\n", cmd));
626 		error = EINVAL;
627 		goto done2;
628 	}
629 
630 	if (error == 0)
631 		td->td_retval[0] = rval;
632 done2:
633 	mtx_unlock(&msq_mtx);
634 	return (error);
635 }
636 
637 #ifndef _SYS_SYSPROTO_H_
638 struct msgget_args {
639 	key_t	key;
640 	int	msgflg;
641 };
642 #endif
643 
644 int
645 sys_msgget(struct thread *td, struct msgget_args *uap)
646 {
647 	int msqid, error = 0;
648 	int key = uap->key;
649 	int msgflg = uap->msgflg;
650 	struct ucred *cred = td->td_ucred;
651 	struct msqid_kernel *msqkptr = NULL;
652 
653 	DPRINTF(("msgget(0x%x, 0%o)\n", key, msgflg));
654 
655 	if (msg_find_prison(cred) == NULL)
656 		return (ENOSYS);
657 
658 	mtx_lock(&msq_mtx);
659 	if (key != IPC_PRIVATE) {
660 		for (msqid = 0; msqid < msginfo.msgmni; msqid++) {
661 			msqkptr = &msqids[msqid];
662 			if (msqkptr->u.msg_qbytes != 0 &&
663 			    msqkptr->cred != NULL &&
664 			    msqkptr->cred->cr_prison == cred->cr_prison &&
665 			    msqkptr->u.msg_perm.key == key)
666 				break;
667 		}
668 		if (msqid < msginfo.msgmni) {
669 			DPRINTF(("found public key\n"));
670 			if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) {
671 				DPRINTF(("not exclusive\n"));
672 				error = EEXIST;
673 				goto done2;
674 			}
675 			AUDIT_ARG_SVIPC_ID(IXSEQ_TO_IPCID(msqid,
676 			    msqkptr->u.msg_perm));
677 			if ((error = ipcperm(td, &msqkptr->u.msg_perm,
678 			    msgflg & 0700))) {
679 				DPRINTF(("requester doesn't have 0%o access\n",
680 				    msgflg & 0700));
681 				goto done2;
682 			}
683 #ifdef MAC
684 			error = mac_sysvmsq_check_msqget(cred, msqkptr);
685 			if (error != 0)
686 				goto done2;
687 #endif
688 			goto found;
689 		}
690 	}
691 
692 	DPRINTF(("need to allocate the msqid_ds\n"));
693 	if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) {
694 		for (msqid = 0; msqid < msginfo.msgmni; msqid++) {
695 			/*
696 			 * Look for an unallocated and unlocked msqid_ds.
697 			 * msqid_ds's can be locked by msgsnd or msgrcv while
698 			 * they are copying the message in/out.  We can't
699 			 * re-use the entry until they release it.
700 			 */
701 			msqkptr = &msqids[msqid];
702 			if (msqkptr->u.msg_qbytes == 0 &&
703 			    (msqkptr->u.msg_perm.mode & MSG_LOCKED) == 0)
704 				break;
705 		}
706 		if (msqid == msginfo.msgmni) {
707 			DPRINTF(("no more msqid_ds's available\n"));
708 			error = ENOSPC;
709 			goto done2;
710 		}
711 #ifdef RACCT
712 		if (racct_enable) {
713 			PROC_LOCK(td->td_proc);
714 			error = racct_add(td->td_proc, RACCT_NMSGQ, 1);
715 			PROC_UNLOCK(td->td_proc);
716 			if (error != 0) {
717 				error = ENOSPC;
718 				goto done2;
719 			}
720 		}
721 #endif
722 		DPRINTF(("msqid %d is available\n", msqid));
723 		msqkptr->u.msg_perm.key = key;
724 		msqkptr->u.msg_perm.cuid = cred->cr_uid;
725 		msqkptr->u.msg_perm.uid = cred->cr_uid;
726 		msqkptr->u.msg_perm.cgid = cred->cr_gid;
727 		msqkptr->u.msg_perm.gid = cred->cr_gid;
728 		msqkptr->u.msg_perm.mode = (msgflg & 0777);
729 		msqkptr->cred = crhold(cred);
730 		/* Make sure that the returned msqid is unique */
731 		msqkptr->u.msg_perm.seq = (msqkptr->u.msg_perm.seq + 1) & 0x7fff;
732 		msqkptr->u.__msg_first = NULL;
733 		msqkptr->u.__msg_last = NULL;
734 		msqkptr->u.msg_cbytes = 0;
735 		msqkptr->u.msg_qnum = 0;
736 		msqkptr->u.msg_qbytes = msginfo.msgmnb;
737 		msqkptr->u.msg_lspid = 0;
738 		msqkptr->u.msg_lrpid = 0;
739 		msqkptr->u.msg_stime = 0;
740 		msqkptr->u.msg_rtime = 0;
741 		msqkptr->u.msg_ctime = time_second;
742 #ifdef MAC
743 		mac_sysvmsq_create(cred, msqkptr);
744 #endif
745 		AUDIT_ARG_SVIPC_PERM(&msqkptr->u.msg_perm);
746 	} else {
747 		DPRINTF(("didn't find it and wasn't asked to create it\n"));
748 		error = ENOENT;
749 		goto done2;
750 	}
751 
752 found:
753 	/* Construct the unique msqid */
754 	td->td_retval[0] = IXSEQ_TO_IPCID(msqid, msqkptr->u.msg_perm);
755 done2:
756 	mtx_unlock(&msq_mtx);
757 	return (error);
758 }
759 
760 #ifndef _SYS_SYSPROTO_H_
761 struct msgsnd_args {
762 	int	msqid;
763 	const void	*msgp;	/* XXX msgp is actually mtext. */
764 	size_t	msgsz;
765 	int	msgflg;
766 };
767 #endif
768 int
769 kern_msgsnd(struct thread *td, int msqid, const void *msgp,
770     size_t msgsz, int msgflg, long mtype)
771 {
772 	int msqix, segs_needed, error = 0;
773 	struct msqid_kernel *msqkptr;
774 	struct msg *msghdr;
775 	struct prison *rpr;
776 	short next;
777 #ifdef RACCT
778 	size_t saved_msgsz = 0;
779 #endif
780 
781 	rpr = msg_find_prison(td->td_ucred);
782 	if (rpr == NULL)
783 		return (ENOSYS);
784 
785 	mtx_lock(&msq_mtx);
786 	AUDIT_ARG_SVIPC_ID(msqid);
787 	msqix = IPCID_TO_IX(msqid);
788 
789 	if (msqix < 0 || msqix >= msginfo.msgmni) {
790 		DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix,
791 		    msginfo.msgmni));
792 		error = EINVAL;
793 		goto done2;
794 	}
795 
796 	msqkptr = &msqids[msqix];
797 	AUDIT_ARG_SVIPC_PERM(&msqkptr->u.msg_perm);
798 	if (msqkptr->u.msg_qbytes == 0) {
799 		DPRINTF(("no such message queue id\n"));
800 		error = EINVAL;
801 		goto done2;
802 	}
803 	if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) {
804 		DPRINTF(("wrong sequence number\n"));
805 		error = EINVAL;
806 		goto done2;
807 	}
808 
809 	if ((error = msq_prison_cansee(rpr, msqkptr))) {
810 		DPRINTF(("requester can't see prison\n"));
811 		goto done2;
812 	}
813 
814 	if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_W))) {
815 		DPRINTF(("requester doesn't have write access\n"));
816 		goto done2;
817 	}
818 
819 #ifdef MAC
820 	error = mac_sysvmsq_check_msqsnd(td->td_ucred, msqkptr);
821 	if (error != 0)
822 		goto done2;
823 #endif
824 
825 #ifdef RACCT
826 	if (racct_enable) {
827 		PROC_LOCK(td->td_proc);
828 		if (racct_add(td->td_proc, RACCT_MSGQQUEUED, 1)) {
829 			PROC_UNLOCK(td->td_proc);
830 			error = EAGAIN;
831 			goto done2;
832 		}
833 		saved_msgsz = msgsz;
834 		if (racct_add(td->td_proc, RACCT_MSGQSIZE, msgsz)) {
835 			racct_sub(td->td_proc, RACCT_MSGQQUEUED, 1);
836 			PROC_UNLOCK(td->td_proc);
837 			error = EAGAIN;
838 			goto done2;
839 		}
840 		PROC_UNLOCK(td->td_proc);
841 	}
842 #endif
843 
844 	segs_needed = howmany(msgsz, msginfo.msgssz);
845 	DPRINTF(("msgsz=%zu, msgssz=%d, segs_needed=%d\n", msgsz,
846 	    msginfo.msgssz, segs_needed));
847 	for (;;) {
848 		int need_more_resources = 0;
849 
850 		/*
851 		 * check msgsz
852 		 * (inside this loop in case msg_qbytes changes while we sleep)
853 		 */
854 
855 		if (msgsz > msqkptr->u.msg_qbytes) {
856 			DPRINTF(("msgsz > msqkptr->u.msg_qbytes\n"));
857 			error = EINVAL;
858 			goto done3;
859 		}
860 
861 		if (msqkptr->u.msg_perm.mode & MSG_LOCKED) {
862 			DPRINTF(("msqid is locked\n"));
863 			need_more_resources = 1;
864 		}
865 		if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) {
866 			DPRINTF(("msgsz + msg_cbytes > msg_qbytes\n"));
867 			need_more_resources = 1;
868 		}
869 		if (segs_needed > nfree_msgmaps) {
870 			DPRINTF(("segs_needed > nfree_msgmaps\n"));
871 			need_more_resources = 1;
872 		}
873 		if (free_msghdrs == NULL) {
874 			DPRINTF(("no more msghdrs\n"));
875 			need_more_resources = 1;
876 		}
877 
878 		if (need_more_resources) {
879 			int we_own_it;
880 
881 			if ((msgflg & IPC_NOWAIT) != 0) {
882 				DPRINTF(("need more resources but caller "
883 				    "doesn't want to wait\n"));
884 				error = EAGAIN;
885 				goto done3;
886 			}
887 
888 			if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) {
889 				DPRINTF(("we don't own the msqid_ds\n"));
890 				we_own_it = 0;
891 			} else {
892 				/* Force later arrivals to wait for our
893 				   request */
894 				DPRINTF(("we own the msqid_ds\n"));
895 				msqkptr->u.msg_perm.mode |= MSG_LOCKED;
896 				we_own_it = 1;
897 			}
898 			DPRINTF(("msgsnd:  goodnight\n"));
899 			error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH,
900 			    "msgsnd", hz);
901 			DPRINTF(("msgsnd:  good morning, error=%d\n", error));
902 			if (we_own_it)
903 				msqkptr->u.msg_perm.mode &= ~MSG_LOCKED;
904 			if (error == EWOULDBLOCK) {
905 				DPRINTF(("msgsnd:  timed out\n"));
906 				continue;
907 			}
908 			if (error != 0) {
909 				DPRINTF(("msgsnd:  interrupted system call\n"));
910 				error = EINTR;
911 				goto done3;
912 			}
913 
914 			/*
915 			 * Make sure that the msq queue still exists
916 			 */
917 
918 			if (msqkptr->u.msg_qbytes == 0) {
919 				DPRINTF(("msqid deleted\n"));
920 				error = EIDRM;
921 				goto done3;
922 			}
923 
924 		} else {
925 			DPRINTF(("got all the resources that we need\n"));
926 			break;
927 		}
928 	}
929 
930 	/*
931 	 * We have the resources that we need.
932 	 * Make sure!
933 	 */
934 
935 	if (msqkptr->u.msg_perm.mode & MSG_LOCKED)
936 		panic("msg_perm.mode & MSG_LOCKED");
937 	if (segs_needed > nfree_msgmaps)
938 		panic("segs_needed > nfree_msgmaps");
939 	if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes)
940 		panic("msgsz + msg_cbytes > msg_qbytes");
941 	if (free_msghdrs == NULL)
942 		panic("no more msghdrs");
943 
944 	/*
945 	 * Re-lock the msqid_ds in case we page-fault when copying in the
946 	 * message
947 	 */
948 
949 	if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0)
950 		panic("msqid_ds is already locked");
951 	msqkptr->u.msg_perm.mode |= MSG_LOCKED;
952 
953 	/*
954 	 * Allocate a message header
955 	 */
956 
957 	msghdr = free_msghdrs;
958 	free_msghdrs = msghdr->msg_next;
959 	msghdr->msg_spot = -1;
960 	msghdr->msg_ts = msgsz;
961 	msghdr->msg_type = mtype;
962 #ifdef MAC
963 	/*
964 	 * XXXMAC: Should the mac_sysvmsq_check_msgmsq check follow here
965 	 * immediately?  Or, should it be checked just before the msg is
966 	 * enqueued in the msgq (as it is done now)?
967 	 */
968 	mac_sysvmsg_create(td->td_ucred, msqkptr, msghdr);
969 #endif
970 
971 	/*
972 	 * Allocate space for the message
973 	 */
974 
975 	while (segs_needed > 0) {
976 		if (nfree_msgmaps <= 0)
977 			panic("not enough msgmaps");
978 		if (free_msgmaps == -1)
979 			panic("nil free_msgmaps");
980 		next = free_msgmaps;
981 		if (next <= -1)
982 			panic("next too low #1");
983 		if (next >= msginfo.msgseg)
984 			panic("next out of range #1");
985 		DPRINTF(("allocating segment %d to message\n", next));
986 		free_msgmaps = msgmaps[next].next;
987 		nfree_msgmaps--;
988 		msgmaps[next].next = msghdr->msg_spot;
989 		msghdr->msg_spot = next;
990 		segs_needed--;
991 	}
992 
993 	/*
994 	 * Validate the message type
995 	 */
996 
997 	if (msghdr->msg_type < 1) {
998 		msg_freehdr(msghdr);
999 		msqkptr->u.msg_perm.mode &= ~MSG_LOCKED;
1000 		wakeup(msqkptr);
1001 		DPRINTF(("mtype (%ld) < 1\n", msghdr->msg_type));
1002 		error = EINVAL;
1003 		goto done3;
1004 	}
1005 
1006 	/*
1007 	 * Copy in the message body
1008 	 */
1009 
1010 	next = msghdr->msg_spot;
1011 	while (msgsz > 0) {
1012 		size_t tlen;
1013 		if (msgsz > msginfo.msgssz)
1014 			tlen = msginfo.msgssz;
1015 		else
1016 			tlen = msgsz;
1017 		if (next <= -1)
1018 			panic("next too low #2");
1019 		if (next >= msginfo.msgseg)
1020 			panic("next out of range #2");
1021 		mtx_unlock(&msq_mtx);
1022 		if ((error = copyin(msgp, &msgpool[next * msginfo.msgssz],
1023 		    tlen)) != 0) {
1024 			mtx_lock(&msq_mtx);
1025 			DPRINTF(("error %d copying in message segment\n",
1026 			    error));
1027 			msg_freehdr(msghdr);
1028 			msqkptr->u.msg_perm.mode &= ~MSG_LOCKED;
1029 			wakeup(msqkptr);
1030 			goto done3;
1031 		}
1032 		mtx_lock(&msq_mtx);
1033 		msgsz -= tlen;
1034 		msgp = (const char *)msgp + tlen;
1035 		next = msgmaps[next].next;
1036 	}
1037 	if (next != -1)
1038 		panic("didn't use all the msg segments");
1039 
1040 	/*
1041 	 * We've got the message.  Unlock the msqid_ds.
1042 	 */
1043 
1044 	msqkptr->u.msg_perm.mode &= ~MSG_LOCKED;
1045 
1046 	/*
1047 	 * Make sure that the msqid_ds is still allocated.
1048 	 */
1049 
1050 	if (msqkptr->u.msg_qbytes == 0) {
1051 		msg_freehdr(msghdr);
1052 		wakeup(msqkptr);
1053 		error = EIDRM;
1054 		goto done3;
1055 	}
1056 
1057 #ifdef MAC
1058 	/*
1059 	 * Note: Since the task/thread allocates the msghdr and usually
1060 	 * primes it with its own MAC label, for a majority of policies, it
1061 	 * won't be necessary to check whether the msghdr has access
1062 	 * permissions to the msgq.  The mac_sysvmsq_check_msqsnd check would
1063 	 * suffice in that case.  However, this hook may be required where
1064 	 * individual policies derive a non-identical label for the msghdr
1065 	 * from the current thread label and may want to check the msghdr
1066 	 * enqueue permissions, along with read/write permissions to the
1067 	 * msgq.
1068 	 */
1069 	error = mac_sysvmsq_check_msgmsq(td->td_ucred, msghdr, msqkptr);
1070 	if (error != 0) {
1071 		msg_freehdr(msghdr);
1072 		wakeup(msqkptr);
1073 		goto done3;
1074 	}
1075 #endif
1076 
1077 	/*
1078 	 * Put the message into the queue
1079 	 */
1080 	if (msqkptr->u.__msg_first == NULL) {
1081 		msqkptr->u.__msg_first = msghdr;
1082 		msqkptr->u.__msg_last = msghdr;
1083 	} else {
1084 		msqkptr->u.__msg_last->msg_next = msghdr;
1085 		msqkptr->u.__msg_last = msghdr;
1086 	}
1087 	msqkptr->u.__msg_last->msg_next = NULL;
1088 
1089 	msqkptr->u.msg_cbytes += msghdr->msg_ts;
1090 	msqkptr->u.msg_qnum++;
1091 	msqkptr->u.msg_lspid = td->td_proc->p_pid;
1092 	msqkptr->u.msg_stime = time_second;
1093 
1094 	wakeup(msqkptr);
1095 	td->td_retval[0] = 0;
1096 done3:
1097 #ifdef RACCT
1098 	if (racct_enable && error != 0) {
1099 		PROC_LOCK(td->td_proc);
1100 		racct_sub(td->td_proc, RACCT_MSGQQUEUED, 1);
1101 		racct_sub(td->td_proc, RACCT_MSGQSIZE, saved_msgsz);
1102 		PROC_UNLOCK(td->td_proc);
1103 	}
1104 #endif
1105 done2:
1106 	mtx_unlock(&msq_mtx);
1107 	return (error);
1108 }
1109 
1110 int
1111 sys_msgsnd(struct thread *td, struct msgsnd_args *uap)
1112 {
1113 	int error;
1114 	long mtype;
1115 
1116 	DPRINTF(("call to msgsnd(%d, %p, %zu, %d)\n", uap->msqid, uap->msgp,
1117 	    uap->msgsz, uap->msgflg));
1118 
1119 	if ((error = copyin(uap->msgp, &mtype, sizeof(mtype))) != 0) {
1120 		DPRINTF(("error %d copying the message type\n", error));
1121 		return (error);
1122 	}
1123 	return (kern_msgsnd(td, uap->msqid,
1124 	    (const char *)uap->msgp + sizeof(mtype),
1125 	    uap->msgsz, uap->msgflg, mtype));
1126 }
1127 
1128 #ifndef _SYS_SYSPROTO_H_
1129 struct msgrcv_args {
1130 	int	msqid;
1131 	void	*msgp;
1132 	size_t	msgsz;
1133 	long	msgtyp;
1134 	int	msgflg;
1135 };
1136 #endif
1137 /* XXX msgp is actually mtext. */
1138 int
1139 kern_msgrcv(struct thread *td, int msqid, void *msgp, size_t msgsz, long msgtyp,
1140     int msgflg, long *mtype)
1141 {
1142 	size_t len;
1143 	struct msqid_kernel *msqkptr;
1144 	struct msg *msghdr;
1145 	struct prison *rpr;
1146 	int msqix, error = 0;
1147 	short next;
1148 
1149 	rpr = msg_find_prison(td->td_ucred);
1150 	if (rpr == NULL)
1151 		return (ENOSYS);
1152 
1153 	AUDIT_ARG_SVIPC_ID(msqid);
1154 	msqix = IPCID_TO_IX(msqid);
1155 
1156 	if (msqix < 0 || msqix >= msginfo.msgmni) {
1157 		DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix,
1158 		    msginfo.msgmni));
1159 		return (EINVAL);
1160 	}
1161 
1162 	msqkptr = &msqids[msqix];
1163 	mtx_lock(&msq_mtx);
1164 	AUDIT_ARG_SVIPC_PERM(&msqkptr->u.msg_perm);
1165 	if (msqkptr->u.msg_qbytes == 0) {
1166 		DPRINTF(("no such message queue id\n"));
1167 		error = EINVAL;
1168 		goto done2;
1169 	}
1170 	if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) {
1171 		DPRINTF(("wrong sequence number\n"));
1172 		error = EINVAL;
1173 		goto done2;
1174 	}
1175 
1176 	if ((error = msq_prison_cansee(rpr, msqkptr))) {
1177 		DPRINTF(("requester can't see prison\n"));
1178 		goto done2;
1179 	}
1180 
1181 	if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) {
1182 		DPRINTF(("requester doesn't have read access\n"));
1183 		goto done2;
1184 	}
1185 
1186 #ifdef MAC
1187 	error = mac_sysvmsq_check_msqrcv(td->td_ucred, msqkptr);
1188 	if (error != 0)
1189 		goto done2;
1190 #endif
1191 
1192 	msghdr = NULL;
1193 	while (msghdr == NULL) {
1194 		if (msgtyp == 0) {
1195 			msghdr = msqkptr->u.__msg_first;
1196 			if (msghdr != NULL) {
1197 				if (msgsz < msghdr->msg_ts &&
1198 				    (msgflg & MSG_NOERROR) == 0) {
1199 					DPRINTF(("first message on the queue "
1200 					    "is too big (want %zu, got %d)\n",
1201 					    msgsz, msghdr->msg_ts));
1202 					error = E2BIG;
1203 					goto done2;
1204 				}
1205 #ifdef MAC
1206 				error = mac_sysvmsq_check_msgrcv(td->td_ucred,
1207 				    msghdr);
1208 				if (error != 0)
1209 					goto done2;
1210 #endif
1211 				if (msqkptr->u.__msg_first ==
1212 				    msqkptr->u.__msg_last) {
1213 					msqkptr->u.__msg_first = NULL;
1214 					msqkptr->u.__msg_last = NULL;
1215 				} else {
1216 					msqkptr->u.__msg_first = msghdr->msg_next;
1217 					if (msqkptr->u.__msg_first == NULL)
1218 						panic("msg_first/last screwed up #1");
1219 				}
1220 			}
1221 		} else {
1222 			struct msg *previous;
1223 			struct msg **prev;
1224 
1225 			previous = NULL;
1226 			prev = &(msqkptr->u.__msg_first);
1227 			while ((msghdr = *prev) != NULL) {
1228 				/*
1229 				 * Is this message's type an exact match or is
1230 				 * this message's type less than or equal to
1231 				 * the absolute value of a negative msgtyp?
1232 				 * Note that the second half of this test can
1233 				 * NEVER be true if msgtyp is positive since
1234 				 * msg_type is always positive!
1235 				 */
1236 
1237 				if (msgtyp == msghdr->msg_type ||
1238 				    msghdr->msg_type <= -msgtyp) {
1239 					DPRINTF(("found message type %ld, "
1240 					    "requested %ld\n",
1241 					    msghdr->msg_type, msgtyp));
1242 					if (msgsz < msghdr->msg_ts &&
1243 					    (msgflg & MSG_NOERROR) == 0) {
1244 						DPRINTF(("requested message "
1245 						    "on the queue is too big "
1246 						    "(want %zu, got %hu)\n",
1247 						    msgsz, msghdr->msg_ts));
1248 						error = E2BIG;
1249 						goto done2;
1250 					}
1251 #ifdef MAC
1252 					error = mac_sysvmsq_check_msgrcv(
1253 					    td->td_ucred, msghdr);
1254 					if (error != 0)
1255 						goto done2;
1256 #endif
1257 					*prev = msghdr->msg_next;
1258 					if (msghdr == msqkptr->u.__msg_last) {
1259 						if (previous == NULL) {
1260 							if (prev !=
1261 							    &msqkptr->u.__msg_first)
1262 								panic("__msg_first/last screwed up #2");
1263 							msqkptr->u.__msg_first =
1264 							    NULL;
1265 							msqkptr->u.__msg_last =
1266 							    NULL;
1267 						} else {
1268 							if (prev ==
1269 							    &msqkptr->u.__msg_first)
1270 								panic("__msg_first/last screwed up #3");
1271 							msqkptr->u.__msg_last =
1272 							    previous;
1273 						}
1274 					}
1275 					break;
1276 				}
1277 				previous = msghdr;
1278 				prev = &(msghdr->msg_next);
1279 			}
1280 		}
1281 
1282 		/*
1283 		 * We've either extracted the msghdr for the appropriate
1284 		 * message or there isn't one.
1285 		 * If there is one then bail out of this loop.
1286 		 */
1287 
1288 		if (msghdr != NULL)
1289 			break;
1290 
1291 		/*
1292 		 * Hmph!  No message found.  Does the user want to wait?
1293 		 */
1294 
1295 		if ((msgflg & IPC_NOWAIT) != 0) {
1296 			DPRINTF(("no appropriate message found (msgtyp=%ld)\n",
1297 			    msgtyp));
1298 			/* The SVID says to return ENOMSG. */
1299 			error = ENOMSG;
1300 			goto done2;
1301 		}
1302 
1303 		/*
1304 		 * Wait for something to happen
1305 		 */
1306 
1307 		DPRINTF(("msgrcv:  goodnight\n"));
1308 		error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH,
1309 		    "msgrcv", 0);
1310 		DPRINTF(("msgrcv:  good morning (error=%d)\n", error));
1311 
1312 		if (error != 0) {
1313 			DPRINTF(("msgrcv:  interrupted system call\n"));
1314 			error = EINTR;
1315 			goto done2;
1316 		}
1317 
1318 		/*
1319 		 * Make sure that the msq queue still exists
1320 		 */
1321 
1322 		if (msqkptr->u.msg_qbytes == 0 ||
1323 		    msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) {
1324 			DPRINTF(("msqid deleted\n"));
1325 			error = EIDRM;
1326 			goto done2;
1327 		}
1328 	}
1329 
1330 	/*
1331 	 * Return the message to the user.
1332 	 *
1333 	 * First, do the bookkeeping (before we risk being interrupted).
1334 	 */
1335 
1336 	msqkptr->u.msg_cbytes -= msghdr->msg_ts;
1337 	msqkptr->u.msg_qnum--;
1338 	msqkptr->u.msg_lrpid = td->td_proc->p_pid;
1339 	msqkptr->u.msg_rtime = time_second;
1340 
1341 	racct_sub_cred(msqkptr->cred, RACCT_MSGQQUEUED, 1);
1342 	racct_sub_cred(msqkptr->cred, RACCT_MSGQSIZE, msghdr->msg_ts);
1343 
1344 	/*
1345 	 * Make msgsz the actual amount that we'll be returning.
1346 	 * Note that this effectively truncates the message if it is too long
1347 	 * (since msgsz is never increased).
1348 	 */
1349 
1350 	DPRINTF(("found a message, msgsz=%zu, msg_ts=%hu\n", msgsz,
1351 	    msghdr->msg_ts));
1352 	if (msgsz > msghdr->msg_ts)
1353 		msgsz = msghdr->msg_ts;
1354 	*mtype = msghdr->msg_type;
1355 
1356 	/*
1357 	 * Return the segments to the user
1358 	 */
1359 
1360 	next = msghdr->msg_spot;
1361 	for (len = 0; len < msgsz; len += msginfo.msgssz) {
1362 		size_t tlen;
1363 
1364 		if (msgsz - len > msginfo.msgssz)
1365 			tlen = msginfo.msgssz;
1366 		else
1367 			tlen = msgsz - len;
1368 		if (next <= -1)
1369 			panic("next too low #3");
1370 		if (next >= msginfo.msgseg)
1371 			panic("next out of range #3");
1372 		mtx_unlock(&msq_mtx);
1373 		error = copyout(&msgpool[next * msginfo.msgssz], msgp, tlen);
1374 		mtx_lock(&msq_mtx);
1375 		if (error != 0) {
1376 			DPRINTF(("error (%d) copying out message segment\n",
1377 			    error));
1378 			msg_freehdr(msghdr);
1379 			wakeup(msqkptr);
1380 			goto done2;
1381 		}
1382 		msgp = (char *)msgp + tlen;
1383 		next = msgmaps[next].next;
1384 	}
1385 
1386 	/*
1387 	 * Done, return the actual number of bytes copied out.
1388 	 */
1389 
1390 	msg_freehdr(msghdr);
1391 	wakeup(msqkptr);
1392 	td->td_retval[0] = msgsz;
1393 done2:
1394 	mtx_unlock(&msq_mtx);
1395 	return (error);
1396 }
1397 
1398 int
1399 sys_msgrcv(struct thread *td, struct msgrcv_args *uap)
1400 {
1401 	int error;
1402 	long mtype;
1403 
1404 	DPRINTF(("call to msgrcv(%d, %p, %zu, %ld, %d)\n", uap->msqid,
1405 	    uap->msgp, uap->msgsz, uap->msgtyp, uap->msgflg));
1406 
1407 	if ((error = kern_msgrcv(td, uap->msqid,
1408 	    (char *)uap->msgp + sizeof(mtype), uap->msgsz,
1409 	    uap->msgtyp, uap->msgflg, &mtype)) != 0)
1410 		return (error);
1411 	if ((error = copyout(&mtype, uap->msgp, sizeof(mtype))) != 0)
1412 		DPRINTF(("error %d copying the message type\n", error));
1413 	return (error);
1414 }
1415 
1416 static int
1417 sysctl_msqids(SYSCTL_HANDLER_ARGS)
1418 {
1419 	struct msqid_kernel tmsqk;
1420 #ifdef COMPAT_FREEBSD32
1421 	struct msqid_kernel32 tmsqk32;
1422 #endif
1423 	struct prison *pr, *rpr;
1424 	void *outaddr;
1425 	size_t outsize;
1426 	int error, i;
1427 
1428 	pr = req->td->td_ucred->cr_prison;
1429 	rpr = msg_find_prison(req->td->td_ucred);
1430 	error = 0;
1431 	for (i = 0; i < msginfo.msgmni; i++) {
1432 		mtx_lock(&msq_mtx);
1433 		if (msqids[i].u.msg_qbytes == 0 || rpr == NULL ||
1434 		    msq_prison_cansee(rpr, &msqids[i]) != 0)
1435 			bzero(&tmsqk, sizeof(tmsqk));
1436 		else {
1437 			tmsqk = msqids[i];
1438 			if (tmsqk.cred->cr_prison != pr)
1439 				tmsqk.u.msg_perm.key = IPC_PRIVATE;
1440 		}
1441 		mtx_unlock(&msq_mtx);
1442 #ifdef COMPAT_FREEBSD32
1443 		if (SV_CURPROC_FLAG(SV_ILP32)) {
1444 			bzero(&tmsqk32, sizeof(tmsqk32));
1445 			freebsd32_ipcperm_out(&tmsqk.u.msg_perm,
1446 			    &tmsqk32.u.msg_perm);
1447 			/* Don't copy u.msg_first or u.msg_last */
1448 			CP(tmsqk, tmsqk32, u.msg_cbytes);
1449 			CP(tmsqk, tmsqk32, u.msg_qnum);
1450 			CP(tmsqk, tmsqk32, u.msg_qbytes);
1451 			CP(tmsqk, tmsqk32, u.msg_lspid);
1452 			CP(tmsqk, tmsqk32, u.msg_lrpid);
1453 			CP(tmsqk, tmsqk32, u.msg_stime);
1454 			CP(tmsqk, tmsqk32, u.msg_rtime);
1455 			CP(tmsqk, tmsqk32, u.msg_ctime);
1456 			/* Don't copy label or cred */
1457 			outaddr = &tmsqk32;
1458 			outsize = sizeof(tmsqk32);
1459 		} else
1460 #endif
1461 		{
1462 			/* Don't leak kernel pointers */
1463 			tmsqk.u.__msg_first = NULL;
1464 			tmsqk.u.__msg_last = NULL;
1465 			tmsqk.label = NULL;
1466 			tmsqk.cred = NULL;
1467 			/*
1468 			 * XXX: some padding also exists, but we take care to
1469 			 * allocate our pool of msqid_kernel structs with
1470 			 * zeroed memory so this should be OK.
1471 			 */
1472 			outaddr = &tmsqk;
1473 			outsize = sizeof(tmsqk);
1474 		}
1475 		error = SYSCTL_OUT(req, outaddr, outsize);
1476 		if (error != 0)
1477 			break;
1478 	}
1479 	return (error);
1480 }
1481 
1482 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmax, CTLFLAG_RD, &msginfo.msgmax, 0,
1483     "Maximum message size");
1484 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmni, CTLFLAG_RDTUN, &msginfo.msgmni, 0,
1485     "Number of message queue identifiers");
1486 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmnb, CTLFLAG_RDTUN, &msginfo.msgmnb, 0,
1487     "Maximum number of bytes in a queue");
1488 SYSCTL_INT(_kern_ipc, OID_AUTO, msgtql, CTLFLAG_RDTUN, &msginfo.msgtql, 0,
1489     "Maximum number of messages in the system");
1490 SYSCTL_INT(_kern_ipc, OID_AUTO, msgssz, CTLFLAG_RDTUN, &msginfo.msgssz, 0,
1491     "Size of a message segment");
1492 SYSCTL_INT(_kern_ipc, OID_AUTO, msgseg, CTLFLAG_RDTUN, &msginfo.msgseg, 0,
1493     "Number of message segments");
1494 SYSCTL_PROC(_kern_ipc, OID_AUTO, msqids,
1495     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1496     NULL, 0, sysctl_msqids, "",
1497     "Array of struct msqid_kernel for each potential message queue");
1498 
1499 static int
1500 msg_prison_check(void *obj, void *data)
1501 {
1502 	struct prison *pr = obj;
1503 	struct prison *prpr;
1504 	struct vfsoptlist *opts = data;
1505 	int error, jsys;
1506 
1507 	/*
1508 	 * sysvmsg is a jailsys integer.
1509 	 * It must be "disable" if the parent jail is disabled.
1510 	 */
1511 	error = vfs_copyopt(opts, "sysvmsg", &jsys, sizeof(jsys));
1512 	if (error != ENOENT) {
1513 		if (error != 0)
1514 			return (error);
1515 		switch (jsys) {
1516 		case JAIL_SYS_DISABLE:
1517 			break;
1518 		case JAIL_SYS_NEW:
1519 		case JAIL_SYS_INHERIT:
1520 			prison_lock(pr->pr_parent);
1521 			prpr = osd_jail_get(pr->pr_parent, msg_prison_slot);
1522 			prison_unlock(pr->pr_parent);
1523 			if (prpr == NULL)
1524 				return (EPERM);
1525 			break;
1526 		default:
1527 			return (EINVAL);
1528 		}
1529 	}
1530 
1531 	return (0);
1532 }
1533 
1534 static int
1535 msg_prison_set(void *obj, void *data)
1536 {
1537 	struct prison *pr = obj;
1538 	struct prison *tpr, *orpr, *nrpr, *trpr;
1539 	struct vfsoptlist *opts = data;
1540 	void *rsv;
1541 	int jsys, descend;
1542 
1543 	/*
1544 	 * sysvmsg controls which jail is the root of the associated msgs (this
1545 	 * jail or same as the parent), or if the feature is available at all.
1546 	 */
1547 	if (vfs_copyopt(opts, "sysvmsg", &jsys, sizeof(jsys)) == ENOENT)
1548 		jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0)
1549 		    ? JAIL_SYS_INHERIT
1550 		    : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0)
1551 		    ? JAIL_SYS_DISABLE
1552 		    : -1;
1553 	if (jsys == JAIL_SYS_DISABLE) {
1554 		prison_lock(pr);
1555 		orpr = osd_jail_get(pr, msg_prison_slot);
1556 		if (orpr != NULL)
1557 			osd_jail_del(pr, msg_prison_slot);
1558 		prison_unlock(pr);
1559 		if (orpr != NULL) {
1560 			if (orpr == pr)
1561 				msg_prison_cleanup(pr);
1562 			/* Disable all child jails as well. */
1563 			FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1564 				prison_lock(tpr);
1565 				trpr = osd_jail_get(tpr, msg_prison_slot);
1566 				if (trpr != NULL) {
1567 					osd_jail_del(tpr, msg_prison_slot);
1568 					prison_unlock(tpr);
1569 					if (trpr == tpr)
1570 						msg_prison_cleanup(tpr);
1571 				} else {
1572 					prison_unlock(tpr);
1573 					descend = 0;
1574 				}
1575 			}
1576 		}
1577 	} else if (jsys != -1) {
1578 		if (jsys == JAIL_SYS_NEW)
1579 			nrpr = pr;
1580 		else {
1581 			prison_lock(pr->pr_parent);
1582 			nrpr = osd_jail_get(pr->pr_parent, msg_prison_slot);
1583 			prison_unlock(pr->pr_parent);
1584 		}
1585 		rsv = osd_reserve(msg_prison_slot);
1586 		prison_lock(pr);
1587 		orpr = osd_jail_get(pr, msg_prison_slot);
1588 		if (orpr != nrpr)
1589 			(void)osd_jail_set_reserved(pr, msg_prison_slot, rsv,
1590 			    nrpr);
1591 		else
1592 			osd_free_reserved(rsv);
1593 		prison_unlock(pr);
1594 		if (orpr != nrpr) {
1595 			if (orpr == pr)
1596 				msg_prison_cleanup(pr);
1597 			if (orpr != NULL) {
1598 				/* Change child jails matching the old root, */
1599 				FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1600 					prison_lock(tpr);
1601 					trpr = osd_jail_get(tpr,
1602 					    msg_prison_slot);
1603 					if (trpr == orpr) {
1604 						(void)osd_jail_set(tpr,
1605 						    msg_prison_slot, nrpr);
1606 						prison_unlock(tpr);
1607 						if (trpr == tpr)
1608 							msg_prison_cleanup(tpr);
1609 					} else {
1610 						prison_unlock(tpr);
1611 						descend = 0;
1612 					}
1613 				}
1614 			}
1615 		}
1616 	}
1617 
1618 	return (0);
1619 }
1620 
1621 static int
1622 msg_prison_get(void *obj, void *data)
1623 {
1624 	struct prison *pr = obj;
1625 	struct prison *rpr;
1626 	struct vfsoptlist *opts = data;
1627 	int error, jsys;
1628 
1629 	/* Set sysvmsg based on the jail's root prison. */
1630 	prison_lock(pr);
1631 	rpr = osd_jail_get(pr, msg_prison_slot);
1632 	prison_unlock(pr);
1633 	jsys = rpr == NULL ? JAIL_SYS_DISABLE
1634 	    : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT;
1635 	error = vfs_setopt(opts, "sysvmsg", &jsys, sizeof(jsys));
1636 	if (error == ENOENT)
1637 		error = 0;
1638 	return (error);
1639 }
1640 
1641 static int
1642 msg_prison_remove(void *obj, void *data __unused)
1643 {
1644 	struct prison *pr = obj;
1645 	struct prison *rpr;
1646 
1647 	prison_lock(pr);
1648 	rpr = osd_jail_get(pr, msg_prison_slot);
1649 	prison_unlock(pr);
1650 	if (rpr == pr)
1651 		msg_prison_cleanup(pr);
1652 	return (0);
1653 }
1654 
1655 static void
1656 msg_prison_cleanup(struct prison *pr)
1657 {
1658 	struct msqid_kernel *msqkptr;
1659 	int i;
1660 
1661 	/* Remove any msqs that belong to this jail. */
1662 	mtx_lock(&msq_mtx);
1663 	for (i = 0; i < msginfo.msgmni; i++) {
1664 		msqkptr = &msqids[i];
1665 		if (msqkptr->u.msg_qbytes != 0 &&
1666 		    msqkptr->cred != NULL && msqkptr->cred->cr_prison == pr)
1667 			msq_remove(msqkptr);
1668 	}
1669 	mtx_unlock(&msq_mtx);
1670 }
1671 
1672 SYSCTL_JAIL_PARAM_SYS_NODE(sysvmsg, CTLFLAG_RW, "SYSV message queues");
1673 
1674 #ifdef COMPAT_FREEBSD32
1675 int
1676 freebsd32_msgsys(struct thread *td, struct freebsd32_msgsys_args *uap)
1677 {
1678 
1679 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1680     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1681 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1682 	switch (uap->which) {
1683 	case 0:
1684 		return (freebsd7_freebsd32_msgctl(td,
1685 		    (struct freebsd7_freebsd32_msgctl_args *)&uap->a2));
1686 	case 2:
1687 		return (freebsd32_msgsnd(td,
1688 		    (struct freebsd32_msgsnd_args *)&uap->a2));
1689 	case 3:
1690 		return (freebsd32_msgrcv(td,
1691 		    (struct freebsd32_msgrcv_args *)&uap->a2));
1692 	default:
1693 		return (sys_msgsys(td, (struct msgsys_args *)uap));
1694 	}
1695 #else
1696 	return (nosys(td, NULL));
1697 #endif
1698 }
1699 
1700 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1701     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1702 int
1703 freebsd7_freebsd32_msgctl(struct thread *td,
1704     struct freebsd7_freebsd32_msgctl_args *uap)
1705 {
1706 	struct msqid_ds msqbuf;
1707 	struct msqid_ds32_old msqbuf32;
1708 	int error;
1709 
1710 	if (uap->cmd == IPC_SET) {
1711 		error = copyin(uap->buf, &msqbuf32, sizeof(msqbuf32));
1712 		if (error)
1713 			return (error);
1714 		freebsd32_ipcperm_old_in(&msqbuf32.msg_perm, &msqbuf.msg_perm);
1715 		PTRIN_CP(msqbuf32, msqbuf, __msg_first);
1716 		PTRIN_CP(msqbuf32, msqbuf, __msg_last);
1717 		CP(msqbuf32, msqbuf, msg_cbytes);
1718 		CP(msqbuf32, msqbuf, msg_qnum);
1719 		CP(msqbuf32, msqbuf, msg_qbytes);
1720 		CP(msqbuf32, msqbuf, msg_lspid);
1721 		CP(msqbuf32, msqbuf, msg_lrpid);
1722 		CP(msqbuf32, msqbuf, msg_stime);
1723 		CP(msqbuf32, msqbuf, msg_rtime);
1724 		CP(msqbuf32, msqbuf, msg_ctime);
1725 	}
1726 	error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf);
1727 	if (error)
1728 		return (error);
1729 	if (uap->cmd == IPC_STAT) {
1730 		bzero(&msqbuf32, sizeof(msqbuf32));
1731 		freebsd32_ipcperm_old_out(&msqbuf.msg_perm, &msqbuf32.msg_perm);
1732 		PTROUT_CP(msqbuf, msqbuf32, __msg_first);
1733 		PTROUT_CP(msqbuf, msqbuf32, __msg_last);
1734 		CP(msqbuf, msqbuf32, msg_cbytes);
1735 		CP(msqbuf, msqbuf32, msg_qnum);
1736 		CP(msqbuf, msqbuf32, msg_qbytes);
1737 		CP(msqbuf, msqbuf32, msg_lspid);
1738 		CP(msqbuf, msqbuf32, msg_lrpid);
1739 		CP(msqbuf, msqbuf32, msg_stime);
1740 		CP(msqbuf, msqbuf32, msg_rtime);
1741 		CP(msqbuf, msqbuf32, msg_ctime);
1742 		error = copyout(&msqbuf32, uap->buf, sizeof(struct msqid_ds32));
1743 	}
1744 	return (error);
1745 }
1746 #endif
1747 
1748 int
1749 freebsd32_msgctl(struct thread *td, struct freebsd32_msgctl_args *uap)
1750 {
1751 	struct msqid_ds msqbuf;
1752 	struct msqid_ds32 msqbuf32;
1753 	int error;
1754 
1755 	if (uap->cmd == IPC_SET) {
1756 		error = copyin(uap->buf, &msqbuf32, sizeof(msqbuf32));
1757 		if (error)
1758 			return (error);
1759 		freebsd32_ipcperm_in(&msqbuf32.msg_perm, &msqbuf.msg_perm);
1760 		PTRIN_CP(msqbuf32, msqbuf, __msg_first);
1761 		PTRIN_CP(msqbuf32, msqbuf, __msg_last);
1762 		CP(msqbuf32, msqbuf, msg_cbytes);
1763 		CP(msqbuf32, msqbuf, msg_qnum);
1764 		CP(msqbuf32, msqbuf, msg_qbytes);
1765 		CP(msqbuf32, msqbuf, msg_lspid);
1766 		CP(msqbuf32, msqbuf, msg_lrpid);
1767 		CP(msqbuf32, msqbuf, msg_stime);
1768 		CP(msqbuf32, msqbuf, msg_rtime);
1769 		CP(msqbuf32, msqbuf, msg_ctime);
1770 	}
1771 	error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf);
1772 	if (error)
1773 		return (error);
1774 	if (uap->cmd == IPC_STAT) {
1775 		freebsd32_ipcperm_out(&msqbuf.msg_perm, &msqbuf32.msg_perm);
1776 		PTROUT_CP(msqbuf, msqbuf32, __msg_first);
1777 		PTROUT_CP(msqbuf, msqbuf32, __msg_last);
1778 		CP(msqbuf, msqbuf32, msg_cbytes);
1779 		CP(msqbuf, msqbuf32, msg_qnum);
1780 		CP(msqbuf, msqbuf32, msg_qbytes);
1781 		CP(msqbuf, msqbuf32, msg_lspid);
1782 		CP(msqbuf, msqbuf32, msg_lrpid);
1783 		CP(msqbuf, msqbuf32, msg_stime);
1784 		CP(msqbuf, msqbuf32, msg_rtime);
1785 		CP(msqbuf, msqbuf32, msg_ctime);
1786 		error = copyout(&msqbuf32, uap->buf, sizeof(struct msqid_ds32));
1787 	}
1788 	return (error);
1789 }
1790 
1791 int
1792 freebsd32_msgsnd(struct thread *td, struct freebsd32_msgsnd_args *uap)
1793 {
1794 	const void *msgp;
1795 	long mtype;
1796 	int32_t mtype32;
1797 	int error;
1798 
1799 	msgp = PTRIN(uap->msgp);
1800 	if ((error = copyin(msgp, &mtype32, sizeof(mtype32))) != 0)
1801 		return (error);
1802 	mtype = mtype32;
1803 	return (kern_msgsnd(td, uap->msqid,
1804 	    (const char *)msgp + sizeof(mtype32),
1805 	    uap->msgsz, uap->msgflg, mtype));
1806 }
1807 
1808 int
1809 freebsd32_msgrcv(struct thread *td, struct freebsd32_msgrcv_args *uap)
1810 {
1811 	void *msgp;
1812 	long mtype;
1813 	int32_t mtype32;
1814 	int error;
1815 
1816 	msgp = PTRIN(uap->msgp);
1817 	if ((error = kern_msgrcv(td, uap->msqid,
1818 	    (char *)msgp + sizeof(mtype32), uap->msgsz,
1819 	    uap->msgtyp, uap->msgflg, &mtype)) != 0)
1820 		return (error);
1821 	mtype32 = (int32_t)mtype;
1822 	return (copyout(&mtype32, msgp, sizeof(mtype32)));
1823 }
1824 #endif
1825 
1826 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1827     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1828 
1829 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1830 static sy_call_t *msgcalls[] = {
1831 	(sy_call_t *)freebsd7_msgctl, (sy_call_t *)sys_msgget,
1832 	(sy_call_t *)sys_msgsnd, (sy_call_t *)sys_msgrcv
1833 };
1834 
1835 /*
1836  * Entry point for all MSG calls.
1837  *
1838  * XXX actually varargs.
1839  * struct msgsys_args {
1840  *		int	which;
1841  *		int	a2;
1842  *		int	a3;
1843  *		int	a4;
1844  *		int	a5;
1845  *		int	a6;
1846  *	} *uap;
1847  */
1848 int
1849 sys_msgsys(struct thread *td, struct msgsys_args *uap)
1850 {
1851 	int error;
1852 
1853 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1854 	if (uap->which < 0 || uap->which >= nitems(msgcalls))
1855 		return (EINVAL);
1856 	error = (*msgcalls[uap->which])(td, &uap->a2);
1857 	return (error);
1858 }
1859 
1860 #ifndef CP
1861 #define CP(src, dst, fld)	do { (dst).fld = (src).fld; } while (0)
1862 #endif
1863 
1864 #ifndef _SYS_SYSPROTO_H_
1865 struct freebsd7_msgctl_args {
1866 	int	msqid;
1867 	int	cmd;
1868 	struct	msqid_ds_old *buf;
1869 };
1870 #endif
1871 int
1872 freebsd7_msgctl(struct thread *td, struct freebsd7_msgctl_args *uap)
1873 {
1874 	struct msqid_ds_old msqold;
1875 	struct msqid_ds msqbuf;
1876 	int error;
1877 
1878 	DPRINTF(("call to freebsd7_msgctl(%d, %d, %p)\n", uap->msqid, uap->cmd,
1879 	    uap->buf));
1880 	if (uap->cmd == IPC_SET) {
1881 		error = copyin(uap->buf, &msqold, sizeof(msqold));
1882 		if (error)
1883 			return (error);
1884 		ipcperm_old2new(&msqold.msg_perm, &msqbuf.msg_perm);
1885 		CP(msqold, msqbuf, __msg_first);
1886 		CP(msqold, msqbuf, __msg_last);
1887 		CP(msqold, msqbuf, msg_cbytes);
1888 		CP(msqold, msqbuf, msg_qnum);
1889 		CP(msqold, msqbuf, msg_qbytes);
1890 		CP(msqold, msqbuf, msg_lspid);
1891 		CP(msqold, msqbuf, msg_lrpid);
1892 		CP(msqold, msqbuf, msg_stime);
1893 		CP(msqold, msqbuf, msg_rtime);
1894 		CP(msqold, msqbuf, msg_ctime);
1895 	}
1896 	error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf);
1897 	if (error)
1898 		return (error);
1899 	if (uap->cmd == IPC_STAT) {
1900 		bzero(&msqold, sizeof(msqold));
1901 		ipcperm_new2old(&msqbuf.msg_perm, &msqold.msg_perm);
1902 		CP(msqbuf, msqold, __msg_first);
1903 		CP(msqbuf, msqold, __msg_last);
1904 		CP(msqbuf, msqold, msg_cbytes);
1905 		CP(msqbuf, msqold, msg_qnum);
1906 		CP(msqbuf, msqold, msg_qbytes);
1907 		CP(msqbuf, msqold, msg_lspid);
1908 		CP(msqbuf, msqold, msg_lrpid);
1909 		CP(msqbuf, msqold, msg_stime);
1910 		CP(msqbuf, msqold, msg_rtime);
1911 		CP(msqbuf, msqold, msg_ctime);
1912 		error = copyout(&msqold, uap->buf, sizeof(struct msqid_ds_old));
1913 	}
1914 	return (error);
1915 }
1916 
1917 #undef CP
1918 
1919 #endif	/* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 ||
1920 	   COMPAT_FREEBSD7 */
1921