sysv_msg.c revision 66085
1/* $FreeBSD: head/sys/kern/sysv_msg.c 66085 2000-09-19 22:59:22Z peter $ */
2
3/*
4 * Implementation of SVID messages
5 *
6 * Author:  Daniel Boulet
7 *
8 * Copyright 1993 Daniel Boulet and RTMX Inc.
9 *
10 * This system call was implemented by Daniel Boulet under contract from RTMX.
11 *
12 * Redistribution and use in source forms, with and without modification,
13 * are permitted provided that this entire comment appears intact.
14 *
15 * Redistribution in binary form may occur without any restrictions.
16 * Obviously, it would be nice if you gave credit where credit is due
17 * but requiring it would be too onerous.
18 *
19 * This software is provided ``AS IS'' without any warranties of any kind.
20 */
21
22#include "opt_sysvipc.h"
23
24#include <sys/param.h>
25#include <sys/systm.h>
26#include <sys/sysproto.h>
27#include <sys/kernel.h>
28#include <sys/proc.h>
29#include <sys/msg.h>
30#include <sys/sysent.h>
31#include <sys/sysctl.h>
32#include <sys/malloc.h>
33
34static MALLOC_DEFINE(M_MSG, "msg", "SVID compatible message queues");
35
36static void msginit __P((void *));
37
38#define MSG_DEBUG
39#undef MSG_DEBUG_OK
40
41static void msg_freehdr __P((struct msg *msghdr));
42
43/* XXX casting to (sy_call_t *) is bogus, as usual. */
44static sy_call_t *msgcalls[] = {
45	(sy_call_t *)msgctl, (sy_call_t *)msgget,
46	(sy_call_t *)msgsnd, (sy_call_t *)msgrcv
47};
48
49struct msg {
50	struct	msg *msg_next;	/* next msg in the chain */
51	long	msg_type;	/* type of this message */
52    				/* >0 -> type of this message */
53    				/* 0 -> free header */
54	u_short	msg_ts;		/* size of this message */
55	short	msg_spot;	/* location of start of msg in buffer */
56};
57
58
59#ifndef MSGSSZ
60#define MSGSSZ	8		/* Each segment must be 2^N long */
61#endif
62#ifndef MSGSEG
63#define MSGSEG	2048		/* must be less than 32767 */
64#endif
65#define MSGMAX	(MSGSSZ*MSGSEG)
66#ifndef MSGMNB
67#define MSGMNB	2048		/* max # of bytes in a queue */
68#endif
69#ifndef MSGMNI
70#define MSGMNI	40
71#endif
72#ifndef MSGTQL
73#define MSGTQL	40
74#endif
75
76/*
77 * Based on the configuration parameters described in an SVR2 (yes, two)
78 * config(1m) man page.
79 *
80 * Each message is broken up and stored in segments that are msgssz bytes
81 * long.  For efficiency reasons, this should be a power of two.  Also,
82 * it doesn't make sense if it is less than 8 or greater than about 256.
83 * Consequently, msginit in kern/sysv_msg.c checks that msgssz is a power of
84 * two between 8 and 1024 inclusive (and panic's if it isn't).
85 */
86struct msginfo msginfo = {
87                MSGMAX,         /* max chars in a message */
88                MSGMNI,         /* # of message queue identifiers */
89                MSGMNB,         /* max chars in a queue */
90                MSGTQL,         /* max messages in system */
91                MSGSSZ,         /* size of a message segment */
92                		/* (must be small power of 2 greater than 4) */
93                MSGSEG          /* number of message segments */
94};
95
96/*
97 * macros to convert between msqid_ds's and msqid's.
98 * (specific to this implementation)
99 */
100#define MSQID(ix,ds)	((ix) & 0xffff | (((ds).msg_perm.seq << 16) & 0xffff0000))
101#define MSQID_IX(id)	((id) & 0xffff)
102#define MSQID_SEQ(id)	(((id) >> 16) & 0xffff)
103
104/*
105 * The rest of this file is specific to this particular implementation.
106 */
107
108struct msgmap {
109	short	next;		/* next segment in buffer */
110    				/* -1 -> available */
111    				/* 0..(MSGSEG-1) -> index of next segment */
112};
113
114#define MSG_LOCKED	01000	/* Is this msqid_ds locked? */
115
116static int nfree_msgmaps;	/* # of free map entries */
117static short free_msgmaps;	/* head of linked list of free map entries */
118static struct msg *free_msghdrs;/* list of free msg headers */
119static char *msgpool;		/* MSGMAX byte long msg buffer pool */
120static struct msgmap *msgmaps;	/* MSGSEG msgmap structures */
121static struct msg *msghdrs;	/* MSGTQL msg headers */
122static struct msqid_ds *msqids;	/* MSGMNI msqid_ds struct's */
123
124static void
125msginit(dummy)
126	void *dummy;
127{
128	register int i;
129
130	msgpool = malloc(msginfo.msgmax, M_MSG, M_WAITOK);
131	if (msgpool == NULL)
132		panic("msgpool is NULL");
133	msgmaps = malloc(sizeof(struct msgmap) * msginfo.msgseg, M_MSG, M_WAITOK);
134	if (msgmaps == NULL)
135		panic("msgmaps is NULL");
136	msghdrs = malloc(sizeof(struct msg) * msginfo.msgtql, M_MSG, M_WAITOK);
137	if (msghdrs == NULL)
138		panic("msghdrs is NULL");
139	msqids = malloc(sizeof(struct msqid_ds) * msginfo.msgmni, M_MSG, M_WAITOK);
140	if (msqids == NULL)
141		panic("msqids is NULL");
142
143	/*
144	 * msginfo.msgssz should be a power of two for efficiency reasons.
145	 * It is also pretty silly if msginfo.msgssz is less than 8
146	 * or greater than about 256 so ...
147	 */
148
149	i = 8;
150	while (i < 1024 && i != msginfo.msgssz)
151		i <<= 1;
152    	if (i != msginfo.msgssz) {
153		printf("msginfo.msgssz=%d (0x%x)\n", msginfo.msgssz,
154		    msginfo.msgssz);
155		panic("msginfo.msgssz not a small power of 2");
156	}
157
158	if (msginfo.msgseg > 32767) {
159		printf("msginfo.msgseg=%d\n", msginfo.msgseg);
160		panic("msginfo.msgseg > 32767");
161	}
162
163	if (msgmaps == NULL)
164		panic("msgmaps is NULL");
165
166	for (i = 0; i < msginfo.msgseg; i++) {
167		if (i > 0)
168			msgmaps[i-1].next = i;
169		msgmaps[i].next = -1;	/* implies entry is available */
170	}
171	free_msgmaps = 0;
172	nfree_msgmaps = msginfo.msgseg;
173
174	if (msghdrs == NULL)
175		panic("msghdrs is NULL");
176
177	for (i = 0; i < msginfo.msgtql; i++) {
178		msghdrs[i].msg_type = 0;
179		if (i > 0)
180			msghdrs[i-1].msg_next = &msghdrs[i];
181		msghdrs[i].msg_next = NULL;
182    	}
183	free_msghdrs = &msghdrs[0];
184
185	if (msqids == NULL)
186		panic("msqids is NULL");
187
188	for (i = 0; i < msginfo.msgmni; i++) {
189		msqids[i].msg_qbytes = 0;	/* implies entry is available */
190		msqids[i].msg_perm.seq = 0;	/* reset to a known value */
191		msqids[i].msg_perm.mode = 0;
192	}
193}
194SYSINIT(sysv_msg, SI_SUB_SYSV_MSG, SI_ORDER_FIRST, msginit, NULL)
195
196/*
197 * Entry point for all MSG calls
198 */
199int
200msgsys(p, uap)
201	struct proc *p;
202	/* XXX actually varargs. */
203	struct msgsys_args /* {
204		u_int	which;
205		int	a2;
206		int	a3;
207		int	a4;
208		int	a5;
209		int	a6;
210	} */ *uap;
211{
212
213	if (uap->which >= sizeof(msgcalls)/sizeof(msgcalls[0]))
214		return (EINVAL);
215	return ((*msgcalls[uap->which])(p, &uap->a2));
216}
217
218static void
219msg_freehdr(msghdr)
220	struct msg *msghdr;
221{
222	while (msghdr->msg_ts > 0) {
223		short next;
224		if (msghdr->msg_spot < 0 || msghdr->msg_spot >= msginfo.msgseg)
225			panic("msghdr->msg_spot out of range");
226		next = msgmaps[msghdr->msg_spot].next;
227		msgmaps[msghdr->msg_spot].next = free_msgmaps;
228		free_msgmaps = msghdr->msg_spot;
229		nfree_msgmaps++;
230		msghdr->msg_spot = next;
231		if (msghdr->msg_ts >= msginfo.msgssz)
232			msghdr->msg_ts -= msginfo.msgssz;
233		else
234			msghdr->msg_ts = 0;
235	}
236	if (msghdr->msg_spot != -1)
237		panic("msghdr->msg_spot != -1");
238	msghdr->msg_next = free_msghdrs;
239	free_msghdrs = msghdr;
240}
241
242#ifndef _SYS_SYSPROTO_H_
243struct msgctl_args {
244	int	msqid;
245	int	cmd;
246	struct	msqid_ds *buf;
247};
248#endif
249
250int
251msgctl(p, uap)
252	struct proc *p;
253	register struct msgctl_args *uap;
254{
255	int msqid = uap->msqid;
256	int cmd = uap->cmd;
257	struct msqid_ds *user_msqptr = uap->buf;
258	int rval, eval;
259	struct msqid_ds msqbuf;
260	register struct msqid_ds *msqptr;
261
262#ifdef MSG_DEBUG_OK
263	printf("call to msgctl(%d, %d, 0x%x)\n", msqid, cmd, user_msqptr);
264#endif
265
266	msqid = IPCID_TO_IX(msqid);
267
268	if (msqid < 0 || msqid >= msginfo.msgmni) {
269#ifdef MSG_DEBUG_OK
270		printf("msqid (%d) out of range (0<=msqid<%d)\n", msqid,
271		    msginfo.msgmni);
272#endif
273		return(EINVAL);
274	}
275
276	msqptr = &msqids[msqid];
277
278	if (msqptr->msg_qbytes == 0) {
279#ifdef MSG_DEBUG_OK
280		printf("no such msqid\n");
281#endif
282		return(EINVAL);
283	}
284	if (msqptr->msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) {
285#ifdef MSG_DEBUG_OK
286		printf("wrong sequence number\n");
287#endif
288		return(EINVAL);
289	}
290
291	eval = 0;
292	rval = 0;
293
294	switch (cmd) {
295
296	case IPC_RMID:
297	{
298		struct msg *msghdr;
299		if ((eval = ipcperm(p, &msqptr->msg_perm, IPC_M)))
300			return(eval);
301		/* Free the message headers */
302		msghdr = msqptr->msg_first;
303		while (msghdr != NULL) {
304			struct msg *msghdr_tmp;
305
306			/* Free the segments of each message */
307			msqptr->msg_cbytes -= msghdr->msg_ts;
308			msqptr->msg_qnum--;
309			msghdr_tmp = msghdr;
310			msghdr = msghdr->msg_next;
311			msg_freehdr(msghdr_tmp);
312		}
313
314		if (msqptr->msg_cbytes != 0)
315			panic("msg_cbytes is screwed up");
316		if (msqptr->msg_qnum != 0)
317			panic("msg_qnum is screwed up");
318
319		msqptr->msg_qbytes = 0;	/* Mark it as free */
320
321		wakeup((caddr_t)msqptr);
322	}
323
324		break;
325
326	case IPC_SET:
327		if ((eval = ipcperm(p, &msqptr->msg_perm, IPC_M)))
328			return(eval);
329		if ((eval = copyin(user_msqptr, &msqbuf, sizeof(msqbuf))) != 0)
330			return(eval);
331		if (msqbuf.msg_qbytes > msqptr->msg_qbytes) {
332			eval = suser(p);
333			if (eval)
334				return(eval);
335		}
336		if (msqbuf.msg_qbytes > msginfo.msgmnb) {
337#ifdef MSG_DEBUG_OK
338			printf("can't increase msg_qbytes beyond %d (truncating)\n",
339			    msginfo.msgmnb);
340#endif
341			msqbuf.msg_qbytes = msginfo.msgmnb;	/* silently restrict qbytes to system limit */
342		}
343		if (msqbuf.msg_qbytes == 0) {
344#ifdef MSG_DEBUG_OK
345			printf("can't reduce msg_qbytes to 0\n");
346#endif
347			return(EINVAL);		/* non-standard errno! */
348		}
349		msqptr->msg_perm.uid = msqbuf.msg_perm.uid;	/* change the owner */
350		msqptr->msg_perm.gid = msqbuf.msg_perm.gid;	/* change the owner */
351		msqptr->msg_perm.mode = (msqptr->msg_perm.mode & ~0777) |
352		    (msqbuf.msg_perm.mode & 0777);
353		msqptr->msg_qbytes = msqbuf.msg_qbytes;
354		msqptr->msg_ctime = time_second;
355		break;
356
357	case IPC_STAT:
358		if ((eval = ipcperm(p, &msqptr->msg_perm, IPC_R))) {
359#ifdef MSG_DEBUG_OK
360			printf("requester doesn't have read access\n");
361#endif
362			return(eval);
363		}
364		eval = copyout((caddr_t)msqptr, user_msqptr,
365		    sizeof(struct msqid_ds));
366		break;
367
368	default:
369#ifdef MSG_DEBUG_OK
370		printf("invalid command %d\n", cmd);
371#endif
372		return(EINVAL);
373	}
374
375	if (eval == 0)
376		p->p_retval[0] = rval;
377	return(eval);
378}
379
380#ifndef _SYS_SYSPROTO_H_
381struct msgget_args {
382	key_t	key;
383	int	msgflg;
384};
385#endif
386
387int
388msgget(p, uap)
389	struct proc *p;
390	register struct msgget_args *uap;
391{
392	int msqid, eval;
393	int key = uap->key;
394	int msgflg = uap->msgflg;
395	struct ucred *cred = p->p_ucred;
396	register struct msqid_ds *msqptr = NULL;
397
398#ifdef MSG_DEBUG_OK
399	printf("msgget(0x%x, 0%o)\n", key, msgflg);
400#endif
401
402	if (key != IPC_PRIVATE) {
403		for (msqid = 0; msqid < msginfo.msgmni; msqid++) {
404			msqptr = &msqids[msqid];
405			if (msqptr->msg_qbytes != 0 &&
406			    msqptr->msg_perm.key == key)
407				break;
408		}
409		if (msqid < msginfo.msgmni) {
410#ifdef MSG_DEBUG_OK
411			printf("found public key\n");
412#endif
413			if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) {
414#ifdef MSG_DEBUG_OK
415				printf("not exclusive\n");
416#endif
417				return(EEXIST);
418			}
419			if ((eval = ipcperm(p, &msqptr->msg_perm, msgflg & 0700 ))) {
420#ifdef MSG_DEBUG_OK
421				printf("requester doesn't have 0%o access\n",
422				    msgflg & 0700);
423#endif
424				return(eval);
425			}
426			goto found;
427		}
428	}
429
430#ifdef MSG_DEBUG_OK
431	printf("need to allocate the msqid_ds\n");
432#endif
433	if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) {
434		for (msqid = 0; msqid < msginfo.msgmni; msqid++) {
435			/*
436			 * Look for an unallocated and unlocked msqid_ds.
437			 * msqid_ds's can be locked by msgsnd or msgrcv while
438			 * they are copying the message in/out.  We can't
439			 * re-use the entry until they release it.
440			 */
441			msqptr = &msqids[msqid];
442			if (msqptr->msg_qbytes == 0 &&
443			    (msqptr->msg_perm.mode & MSG_LOCKED) == 0)
444				break;
445		}
446		if (msqid == msginfo.msgmni) {
447#ifdef MSG_DEBUG_OK
448			printf("no more msqid_ds's available\n");
449#endif
450			return(ENOSPC);
451		}
452#ifdef MSG_DEBUG_OK
453		printf("msqid %d is available\n", msqid);
454#endif
455		msqptr->msg_perm.key = key;
456		msqptr->msg_perm.cuid = cred->cr_uid;
457		msqptr->msg_perm.uid = cred->cr_uid;
458		msqptr->msg_perm.cgid = cred->cr_gid;
459		msqptr->msg_perm.gid = cred->cr_gid;
460		msqptr->msg_perm.mode = (msgflg & 0777);
461		/* Make sure that the returned msqid is unique */
462		msqptr->msg_perm.seq++;
463		msqptr->msg_first = NULL;
464		msqptr->msg_last = NULL;
465		msqptr->msg_cbytes = 0;
466		msqptr->msg_qnum = 0;
467		msqptr->msg_qbytes = msginfo.msgmnb;
468		msqptr->msg_lspid = 0;
469		msqptr->msg_lrpid = 0;
470		msqptr->msg_stime = 0;
471		msqptr->msg_rtime = 0;
472		msqptr->msg_ctime = time_second;
473	} else {
474#ifdef MSG_DEBUG_OK
475		printf("didn't find it and wasn't asked to create it\n");
476#endif
477		return(ENOENT);
478	}
479
480found:
481	/* Construct the unique msqid */
482	p->p_retval[0] = IXSEQ_TO_IPCID(msqid, msqptr->msg_perm);
483	return(0);
484}
485
486#ifndef _SYS_SYSPROTO_H_
487struct msgsnd_args {
488	int	msqid;
489	void	*msgp;
490	size_t	msgsz;
491	int	msgflg;
492};
493#endif
494
495int
496msgsnd(p, uap)
497	struct proc *p;
498	register struct msgsnd_args *uap;
499{
500	int msqid = uap->msqid;
501	void *user_msgp = uap->msgp;
502	size_t msgsz = uap->msgsz;
503	int msgflg = uap->msgflg;
504	int segs_needed, eval;
505	register struct msqid_ds *msqptr;
506	register struct msg *msghdr;
507	short next;
508
509#ifdef MSG_DEBUG_OK
510	printf("call to msgsnd(%d, 0x%x, %d, %d)\n", msqid, user_msgp, msgsz,
511	    msgflg);
512#endif
513
514	msqid = IPCID_TO_IX(msqid);
515
516	if (msqid < 0 || msqid >= msginfo.msgmni) {
517#ifdef MSG_DEBUG_OK
518		printf("msqid (%d) out of range (0<=msqid<%d)\n", msqid,
519		    msginfo.msgmni);
520#endif
521		return(EINVAL);
522	}
523
524	msqptr = &msqids[msqid];
525	if (msqptr->msg_qbytes == 0) {
526#ifdef MSG_DEBUG_OK
527		printf("no such message queue id\n");
528#endif
529		return(EINVAL);
530	}
531	if (msqptr->msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) {
532#ifdef MSG_DEBUG_OK
533		printf("wrong sequence number\n");
534#endif
535		return(EINVAL);
536	}
537
538	if ((eval = ipcperm(p, &msqptr->msg_perm, IPC_W))) {
539#ifdef MSG_DEBUG_OK
540		printf("requester doesn't have write access\n");
541#endif
542		return(eval);
543	}
544
545	segs_needed = (msgsz + msginfo.msgssz - 1) / msginfo.msgssz;
546#ifdef MSG_DEBUG_OK
547	printf("msgsz=%d, msgssz=%d, segs_needed=%d\n", msgsz, msginfo.msgssz,
548	    segs_needed);
549#endif
550	for (;;) {
551		int need_more_resources = 0;
552
553		/*
554		 * check msgsz
555		 * (inside this loop in case msg_qbytes changes while we sleep)
556		 */
557
558		if (msgsz > msqptr->msg_qbytes) {
559#ifdef MSG_DEBUG_OK
560			printf("msgsz > msqptr->msg_qbytes\n");
561#endif
562			return(EINVAL);
563		}
564
565		if (msqptr->msg_perm.mode & MSG_LOCKED) {
566#ifdef MSG_DEBUG_OK
567			printf("msqid is locked\n");
568#endif
569			need_more_resources = 1;
570		}
571		if (msgsz + msqptr->msg_cbytes > msqptr->msg_qbytes) {
572#ifdef MSG_DEBUG_OK
573			printf("msgsz + msg_cbytes > msg_qbytes\n");
574#endif
575			need_more_resources = 1;
576		}
577		if (segs_needed > nfree_msgmaps) {
578#ifdef MSG_DEBUG_OK
579			printf("segs_needed > nfree_msgmaps\n");
580#endif
581			need_more_resources = 1;
582		}
583		if (free_msghdrs == NULL) {
584#ifdef MSG_DEBUG_OK
585			printf("no more msghdrs\n");
586#endif
587			need_more_resources = 1;
588		}
589
590		if (need_more_resources) {
591			int we_own_it;
592
593			if ((msgflg & IPC_NOWAIT) != 0) {
594#ifdef MSG_DEBUG_OK
595				printf("need more resources but caller doesn't want to wait\n");
596#endif
597				return(EAGAIN);
598			}
599
600			if ((msqptr->msg_perm.mode & MSG_LOCKED) != 0) {
601#ifdef MSG_DEBUG_OK
602				printf("we don't own the msqid_ds\n");
603#endif
604				we_own_it = 0;
605			} else {
606				/* Force later arrivals to wait for our
607				   request */
608#ifdef MSG_DEBUG_OK
609				printf("we own the msqid_ds\n");
610#endif
611				msqptr->msg_perm.mode |= MSG_LOCKED;
612				we_own_it = 1;
613			}
614#ifdef MSG_DEBUG_OK
615			printf("goodnight\n");
616#endif
617			eval = tsleep((caddr_t)msqptr, (PZERO - 4) | PCATCH,
618			    "msgwait", 0);
619#ifdef MSG_DEBUG_OK
620			printf("good morning, eval=%d\n", eval);
621#endif
622			if (we_own_it)
623				msqptr->msg_perm.mode &= ~MSG_LOCKED;
624			if (eval != 0) {
625#ifdef MSG_DEBUG_OK
626				printf("msgsnd:  interrupted system call\n");
627#endif
628				return(EINTR);
629			}
630
631			/*
632			 * Make sure that the msq queue still exists
633			 */
634
635			if (msqptr->msg_qbytes == 0) {
636#ifdef MSG_DEBUG_OK
637				printf("msqid deleted\n");
638#endif
639				return(EIDRM);
640			}
641
642		} else {
643#ifdef MSG_DEBUG_OK
644			printf("got all the resources that we need\n");
645#endif
646			break;
647		}
648	}
649
650	/*
651	 * We have the resources that we need.
652	 * Make sure!
653	 */
654
655	if (msqptr->msg_perm.mode & MSG_LOCKED)
656		panic("msg_perm.mode & MSG_LOCKED");
657	if (segs_needed > nfree_msgmaps)
658		panic("segs_needed > nfree_msgmaps");
659	if (msgsz + msqptr->msg_cbytes > msqptr->msg_qbytes)
660		panic("msgsz + msg_cbytes > msg_qbytes");
661	if (free_msghdrs == NULL)
662		panic("no more msghdrs");
663
664	/*
665	 * Re-lock the msqid_ds in case we page-fault when copying in the
666	 * message
667	 */
668
669	if ((msqptr->msg_perm.mode & MSG_LOCKED) != 0)
670		panic("msqid_ds is already locked");
671	msqptr->msg_perm.mode |= MSG_LOCKED;
672
673	/*
674	 * Allocate a message header
675	 */
676
677	msghdr = free_msghdrs;
678	free_msghdrs = msghdr->msg_next;
679	msghdr->msg_spot = -1;
680	msghdr->msg_ts = msgsz;
681
682	/*
683	 * Allocate space for the message
684	 */
685
686	while (segs_needed > 0) {
687		if (nfree_msgmaps <= 0)
688			panic("not enough msgmaps");
689		if (free_msgmaps == -1)
690			panic("nil free_msgmaps");
691		next = free_msgmaps;
692		if (next <= -1)
693			panic("next too low #1");
694		if (next >= msginfo.msgseg)
695			panic("next out of range #1");
696#ifdef MSG_DEBUG_OK
697		printf("allocating segment %d to message\n", next);
698#endif
699		free_msgmaps = msgmaps[next].next;
700		nfree_msgmaps--;
701		msgmaps[next].next = msghdr->msg_spot;
702		msghdr->msg_spot = next;
703		segs_needed--;
704	}
705
706	/*
707	 * Copy in the message type
708	 */
709
710	if ((eval = copyin(user_msgp, &msghdr->msg_type,
711	    sizeof(msghdr->msg_type))) != 0) {
712#ifdef MSG_DEBUG_OK
713		printf("error %d copying the message type\n", eval);
714#endif
715		msg_freehdr(msghdr);
716		msqptr->msg_perm.mode &= ~MSG_LOCKED;
717		wakeup((caddr_t)msqptr);
718		return(eval);
719	}
720	user_msgp = (char *)user_msgp + sizeof(msghdr->msg_type);
721
722	/*
723	 * Validate the message type
724	 */
725
726	if (msghdr->msg_type < 1) {
727		msg_freehdr(msghdr);
728		msqptr->msg_perm.mode &= ~MSG_LOCKED;
729		wakeup((caddr_t)msqptr);
730#ifdef MSG_DEBUG_OK
731		printf("mtype (%d) < 1\n", msghdr->msg_type);
732#endif
733		return(EINVAL);
734	}
735
736	/*
737	 * Copy in the message body
738	 */
739
740	next = msghdr->msg_spot;
741	while (msgsz > 0) {
742		size_t tlen;
743		if (msgsz > msginfo.msgssz)
744			tlen = msginfo.msgssz;
745		else
746			tlen = msgsz;
747		if (next <= -1)
748			panic("next too low #2");
749		if (next >= msginfo.msgseg)
750			panic("next out of range #2");
751		if ((eval = copyin(user_msgp, &msgpool[next * msginfo.msgssz],
752		    tlen)) != 0) {
753#ifdef MSG_DEBUG_OK
754			printf("error %d copying in message segment\n", eval);
755#endif
756			msg_freehdr(msghdr);
757			msqptr->msg_perm.mode &= ~MSG_LOCKED;
758			wakeup((caddr_t)msqptr);
759			return(eval);
760		}
761		msgsz -= tlen;
762		user_msgp = (char *)user_msgp + tlen;
763		next = msgmaps[next].next;
764	}
765	if (next != -1)
766		panic("didn't use all the msg segments");
767
768	/*
769	 * We've got the message.  Unlock the msqid_ds.
770	 */
771
772	msqptr->msg_perm.mode &= ~MSG_LOCKED;
773
774	/*
775	 * Make sure that the msqid_ds is still allocated.
776	 */
777
778	if (msqptr->msg_qbytes == 0) {
779		msg_freehdr(msghdr);
780		wakeup((caddr_t)msqptr);
781		return(EIDRM);
782	}
783
784	/*
785	 * Put the message into the queue
786	 */
787
788	if (msqptr->msg_first == NULL) {
789		msqptr->msg_first = msghdr;
790		msqptr->msg_last = msghdr;
791	} else {
792		msqptr->msg_last->msg_next = msghdr;
793		msqptr->msg_last = msghdr;
794	}
795	msqptr->msg_last->msg_next = NULL;
796
797	msqptr->msg_cbytes += msghdr->msg_ts;
798	msqptr->msg_qnum++;
799	msqptr->msg_lspid = p->p_pid;
800	msqptr->msg_stime = time_second;
801
802	wakeup((caddr_t)msqptr);
803	p->p_retval[0] = 0;
804	return(0);
805}
806
807#ifndef _SYS_SYSPROTO_H_
808struct msgrcv_args {
809	int	msqid;
810	void	*msgp;
811	size_t	msgsz;
812	long	msgtyp;
813	int	msgflg;
814};
815#endif
816
817int
818msgrcv(p, uap)
819	struct proc *p;
820	register struct msgrcv_args *uap;
821{
822	int msqid = uap->msqid;
823	void *user_msgp = uap->msgp;
824	size_t msgsz = uap->msgsz;
825	long msgtyp = uap->msgtyp;
826	int msgflg = uap->msgflg;
827	size_t len;
828	register struct msqid_ds *msqptr;
829	register struct msg *msghdr;
830	int eval;
831	short next;
832
833#ifdef MSG_DEBUG_OK
834	printf("call to msgrcv(%d, 0x%x, %d, %ld, %d)\n", msqid, user_msgp,
835	    msgsz, msgtyp, msgflg);
836#endif
837
838	msqid = IPCID_TO_IX(msqid);
839
840	if (msqid < 0 || msqid >= msginfo.msgmni) {
841#ifdef MSG_DEBUG_OK
842		printf("msqid (%d) out of range (0<=msqid<%d)\n", msqid,
843		    msginfo.msgmni);
844#endif
845		return(EINVAL);
846	}
847
848	msqptr = &msqids[msqid];
849	if (msqptr->msg_qbytes == 0) {
850#ifdef MSG_DEBUG_OK
851		printf("no such message queue id\n");
852#endif
853		return(EINVAL);
854	}
855	if (msqptr->msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) {
856#ifdef MSG_DEBUG_OK
857		printf("wrong sequence number\n");
858#endif
859		return(EINVAL);
860	}
861
862	if ((eval = ipcperm(p, &msqptr->msg_perm, IPC_R))) {
863#ifdef MSG_DEBUG_OK
864		printf("requester doesn't have read access\n");
865#endif
866		return(eval);
867	}
868
869	msghdr = NULL;
870	while (msghdr == NULL) {
871		if (msgtyp == 0) {
872			msghdr = msqptr->msg_first;
873			if (msghdr != NULL) {
874				if (msgsz < msghdr->msg_ts &&
875				    (msgflg & MSG_NOERROR) == 0) {
876#ifdef MSG_DEBUG_OK
877					printf("first message on the queue is too big (want %d, got %d)\n",
878					    msgsz, msghdr->msg_ts);
879#endif
880					return(E2BIG);
881				}
882				if (msqptr->msg_first == msqptr->msg_last) {
883					msqptr->msg_first = NULL;
884					msqptr->msg_last = NULL;
885				} else {
886					msqptr->msg_first = msghdr->msg_next;
887					if (msqptr->msg_first == NULL)
888						panic("msg_first/last screwed up #1");
889				}
890			}
891		} else {
892			struct msg *previous;
893			struct msg **prev;
894
895			previous = NULL;
896			prev = &(msqptr->msg_first);
897			while ((msghdr = *prev) != NULL) {
898				/*
899				 * Is this message's type an exact match or is
900				 * this message's type less than or equal to
901				 * the absolute value of a negative msgtyp?
902				 * Note that the second half of this test can
903				 * NEVER be true if msgtyp is positive since
904				 * msg_type is always positive!
905				 */
906
907				if (msgtyp == msghdr->msg_type ||
908				    msghdr->msg_type <= -msgtyp) {
909#ifdef MSG_DEBUG_OK
910					printf("found message type %d, requested %d\n",
911					    msghdr->msg_type, msgtyp);
912#endif
913					if (msgsz < msghdr->msg_ts &&
914					    (msgflg & MSG_NOERROR) == 0) {
915#ifdef MSG_DEBUG_OK
916						printf("requested message on the queue is too big (want %d, got %d)\n",
917						    msgsz, msghdr->msg_ts);
918#endif
919						return(E2BIG);
920					}
921					*prev = msghdr->msg_next;
922					if (msghdr == msqptr->msg_last) {
923						if (previous == NULL) {
924							if (prev !=
925							    &msqptr->msg_first)
926								panic("msg_first/last screwed up #2");
927							msqptr->msg_first =
928							    NULL;
929							msqptr->msg_last =
930							    NULL;
931						} else {
932							if (prev ==
933							    &msqptr->msg_first)
934								panic("msg_first/last screwed up #3");
935							msqptr->msg_last =
936							    previous;
937						}
938					}
939					break;
940				}
941				previous = msghdr;
942				prev = &(msghdr->msg_next);
943			}
944		}
945
946		/*
947		 * We've either extracted the msghdr for the appropriate
948		 * message or there isn't one.
949		 * If there is one then bail out of this loop.
950		 */
951
952		if (msghdr != NULL)
953			break;
954
955		/*
956		 * Hmph!  No message found.  Does the user want to wait?
957		 */
958
959		if ((msgflg & IPC_NOWAIT) != 0) {
960#ifdef MSG_DEBUG_OK
961			printf("no appropriate message found (msgtyp=%d)\n",
962			    msgtyp);
963#endif
964			/* The SVID says to return ENOMSG. */
965#ifdef ENOMSG
966			return(ENOMSG);
967#else
968			/* Unfortunately, BSD doesn't define that code yet! */
969			return(EAGAIN);
970#endif
971		}
972
973		/*
974		 * Wait for something to happen
975		 */
976
977#ifdef MSG_DEBUG_OK
978		printf("msgrcv:  goodnight\n");
979#endif
980		eval = tsleep((caddr_t)msqptr, (PZERO - 4) | PCATCH, "msgwait",
981		    0);
982#ifdef MSG_DEBUG_OK
983		printf("msgrcv:  good morning (eval=%d)\n", eval);
984#endif
985
986		if (eval != 0) {
987#ifdef MSG_DEBUG_OK
988			printf("msgsnd:  interrupted system call\n");
989#endif
990			return(EINTR);
991		}
992
993		/*
994		 * Make sure that the msq queue still exists
995		 */
996
997		if (msqptr->msg_qbytes == 0 ||
998		    msqptr->msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) {
999#ifdef MSG_DEBUG_OK
1000			printf("msqid deleted\n");
1001#endif
1002			return(EIDRM);
1003		}
1004	}
1005
1006	/*
1007	 * Return the message to the user.
1008	 *
1009	 * First, do the bookkeeping (before we risk being interrupted).
1010	 */
1011
1012	msqptr->msg_cbytes -= msghdr->msg_ts;
1013	msqptr->msg_qnum--;
1014	msqptr->msg_lrpid = p->p_pid;
1015	msqptr->msg_rtime = time_second;
1016
1017	/*
1018	 * Make msgsz the actual amount that we'll be returning.
1019	 * Note that this effectively truncates the message if it is too long
1020	 * (since msgsz is never increased).
1021	 */
1022
1023#ifdef MSG_DEBUG_OK
1024	printf("found a message, msgsz=%d, msg_ts=%d\n", msgsz,
1025	    msghdr->msg_ts);
1026#endif
1027	if (msgsz > msghdr->msg_ts)
1028		msgsz = msghdr->msg_ts;
1029
1030	/*
1031	 * Return the type to the user.
1032	 */
1033
1034	eval = copyout((caddr_t)&(msghdr->msg_type), user_msgp,
1035	    sizeof(msghdr->msg_type));
1036	if (eval != 0) {
1037#ifdef MSG_DEBUG_OK
1038		printf("error (%d) copying out message type\n", eval);
1039#endif
1040		msg_freehdr(msghdr);
1041		wakeup((caddr_t)msqptr);
1042		return(eval);
1043	}
1044	user_msgp = (char *)user_msgp + sizeof(msghdr->msg_type);
1045
1046	/*
1047	 * Return the segments to the user
1048	 */
1049
1050	next = msghdr->msg_spot;
1051	for (len = 0; len < msgsz; len += msginfo.msgssz) {
1052		size_t tlen;
1053
1054		if (msgsz - len > msginfo.msgssz)
1055			tlen = msginfo.msgssz;
1056		else
1057			tlen = msgsz - len;
1058		if (next <= -1)
1059			panic("next too low #3");
1060		if (next >= msginfo.msgseg)
1061			panic("next out of range #3");
1062		eval = copyout((caddr_t)&msgpool[next * msginfo.msgssz],
1063		    user_msgp, tlen);
1064		if (eval != 0) {
1065#ifdef MSG_DEBUG_OK
1066			printf("error (%d) copying out message segment\n",
1067			    eval);
1068#endif
1069			msg_freehdr(msghdr);
1070			wakeup((caddr_t)msqptr);
1071			return(eval);
1072		}
1073		user_msgp = (char *)user_msgp + tlen;
1074		next = msgmaps[next].next;
1075	}
1076
1077	/*
1078	 * Done, return the actual number of bytes copied out.
1079	 */
1080
1081	msg_freehdr(msghdr);
1082	wakeup((caddr_t)msqptr);
1083	p->p_retval[0] = msgsz;
1084	return(0);
1085}
1086