sysv_msg.c revision 46116
155714Skris/*	$Id: sysv_msg.c,v 1.21 1999/04/27 11:16:16 phk Exp $ */
255714Skris
3238405Sjkim/*
455714Skris * Implementation of SVID messages
555714Skris *
655714Skris * Author:  Daniel Boulet
755714Skris *
855714Skris * Copyright 1993 Daniel Boulet and RTMX Inc.
955714Skris *
10280297Sjkim * This system call was implemented by Daniel Boulet under contract from RTMX.
1155714Skris *
1255714Skris * Redistribution and use in source forms, with and without modification,
1355714Skris * are permitted provided that this entire comment appears intact.
1455714Skris *
1555714Skris * Redistribution in binary form may occur without any restrictions.
1655714Skris * Obviously, it would be nice if you gave credit where credit is due
1755714Skris * but requiring it would be too onerous.
1855714Skris *
1955714Skris * This software is provided ``AS IS'' without any warranties of any kind.
2055714Skris */
2155714Skris
2255714Skris#include <sys/param.h>
2355714Skris#include <sys/systm.h>
2455714Skris#include <sys/sysproto.h>
2555714Skris#include <sys/kernel.h>
2655714Skris#include <sys/proc.h>
2755714Skris#include <sys/msg.h>
2855714Skris#include <sys/sysent.h>
2955714Skris
3055714Skrisstatic void msginit __P((void *));
3155714SkrisSYSINIT(sysv_msg, SI_SUB_SYSV_MSG, SI_ORDER_FIRST, msginit, NULL)
3255714Skris
3355714Skris#define MSG_DEBUG
3455714Skris#undef MSG_DEBUG_OK
3555714Skris
3655714Skris#ifndef _SYS_SYSPROTO_H_
3755714Skrisstruct msgctl_args;
3855714Skrisint msgctl __P((struct proc *p, struct msgctl_args *uap));
3955714Skrisstruct msgget_args;
4055714Skrisint msgget __P((struct proc *p, struct msgget_args *uap));
4155714Skrisstruct msgsnd_args;
4255714Skrisint msgsnd __P((struct proc *p, struct msgsnd_args *uap));
4355714Skrisstruct msgrcv_args;
4455714Skrisint msgrcv __P((struct proc *p, struct msgrcv_args *uap));
4555714Skris#endif
4655714Skrisstatic void msg_freehdr __P((struct msg *msghdr));
4755714Skris
4855714Skris/* XXX casting to (sy_call_t *) is bogus, as usual. */
4955714Skrisstatic sy_call_t *msgcalls[] = {
5055714Skris	(sy_call_t *)msgctl, (sy_call_t *)msgget,
5155714Skris	(sy_call_t *)msgsnd, (sy_call_t *)msgrcv
5255714Skris};
5355714Skris
5455714Skrisstatic int nfree_msgmaps;	/* # of free map entries */
5555714Skrisstatic short free_msgmaps;	/* head of linked list of free map entries */
56280297Sjkimstatic struct msg *free_msghdrs;	/* list of free msg headers */
57280297Sjkimchar *msgpool;			/* MSGMAX byte long msg buffer pool */
5859191Skrisstruct msgmap *msgmaps;		/* MSGSEG msgmap structures */
5959191Skrisstruct msg *msghdrs;		/* MSGTQL msg headers */
6055714Skrisstruct msqid_ds *msqids;	/* MSGMNI msqid_ds struct's */
6155714Skris
6255714Skrisvoid
6355714Skrismsginit(dummy)
6455714Skris	void *dummy;
6555714Skris{
6655714Skris	register int i;
67109998Smarkm
68160814Ssimon	/*
69280297Sjkim	 * msginfo.msgssz should be a power of two for efficiency reasons.
70280297Sjkim	 * It is also pretty silly if msginfo.msgssz is less than 8
71160814Ssimon	 * or greater than about 256 so ...
72280297Sjkim	 */
73280297Sjkim
74280297Sjkim	i = 8;
75280297Sjkim	while (i < 1024 && i != msginfo.msgssz)
76280297Sjkim		i <<= 1;
77280297Sjkim    	if (i != msginfo.msgssz) {
78280297Sjkim		printf("msginfo.msgssz=%d (0x%x)\n", msginfo.msgssz,
79280297Sjkim		    msginfo.msgssz);
80280297Sjkim		panic("msginfo.msgssz not a small power of 2");
81280297Sjkim	}
82280297Sjkim
83280297Sjkim	if (msginfo.msgseg > 32767) {
84280297Sjkim		printf("msginfo.msgseg=%d\n", msginfo.msgseg);
8555714Skris		panic("msginfo.msgseg > 32767");
86280297Sjkim	}
87280297Sjkim
88280297Sjkim	if (msgmaps == NULL)
89280297Sjkim		panic("msgmaps is NULL");
90280297Sjkim
91280297Sjkim	for (i = 0; i < msginfo.msgseg; i++) {
9255714Skris		if (i > 0)
9355714Skris			msgmaps[i-1].next = i;
9455714Skris		msgmaps[i].next = -1;	/* implies entry is available */
9555714Skris	}
96280297Sjkim	free_msgmaps = 0;
97167612Ssimon	nfree_msgmaps = msginfo.msgseg;
9855714Skris
99280297Sjkim	if (msghdrs == NULL)
100280297Sjkim		panic("msghdrs is NULL");
101280297Sjkim
102280297Sjkim	for (i = 0; i < msginfo.msgtql; i++) {
10355714Skris		msghdrs[i].msg_type = 0;
104280297Sjkim		if (i > 0)
105			msghdrs[i-1].msg_next = &msghdrs[i];
106		msghdrs[i].msg_next = NULL;
107    	}
108	free_msghdrs = &msghdrs[0];
109
110	if (msqids == NULL)
111		panic("msqids is NULL");
112
113	for (i = 0; i < msginfo.msgmni; i++) {
114		msqids[i].msg_qbytes = 0;	/* implies entry is available */
115		msqids[i].msg_perm.seq = 0;	/* reset to a known value */
116	}
117}
118
119/*
120 * Entry point for all MSG calls
121 */
122int
123msgsys(p, uap)
124	struct proc *p;
125	/* XXX actually varargs. */
126	struct msgsys_args /* {
127		u_int	which;
128		int	a2;
129		int	a3;
130		int	a4;
131		int	a5;
132		int	a6;
133	} */ *uap;
134{
135
136	if (uap->which >= sizeof(msgcalls)/sizeof(msgcalls[0]))
137		return (EINVAL);
138	return ((*msgcalls[uap->which])(p, &uap->a2));
139}
140
141static void
142msg_freehdr(msghdr)
143	struct msg *msghdr;
144{
145	while (msghdr->msg_ts > 0) {
146		short next;
147		if (msghdr->msg_spot < 0 || msghdr->msg_spot >= msginfo.msgseg)
148			panic("msghdr->msg_spot out of range");
149		next = msgmaps[msghdr->msg_spot].next;
150		msgmaps[msghdr->msg_spot].next = free_msgmaps;
151		free_msgmaps = msghdr->msg_spot;
152		nfree_msgmaps++;
153		msghdr->msg_spot = next;
154		if (msghdr->msg_ts >= msginfo.msgssz)
155			msghdr->msg_ts -= msginfo.msgssz;
156		else
157			msghdr->msg_ts = 0;
158	}
159	if (msghdr->msg_spot != -1)
160		panic("msghdr->msg_spot != -1");
161	msghdr->msg_next = free_msghdrs;
162	free_msghdrs = msghdr;
163}
164
165#ifndef _SYS_SYSPROTO_H_
166struct msgctl_args {
167	int	msqid;
168	int	cmd;
169	struct	msqid_ds *buf;
170};
171#endif
172
173int
174msgctl(p, uap)
175	struct proc *p;
176	register struct msgctl_args *uap;
177{
178	int msqid = uap->msqid;
179	int cmd = uap->cmd;
180	struct msqid_ds *user_msqptr = uap->buf;
181	int rval, eval;
182	struct msqid_ds msqbuf;
183	register struct msqid_ds *msqptr;
184
185#ifdef MSG_DEBUG_OK
186	printf("call to msgctl(%d, %d, 0x%x)\n", msqid, cmd, user_msqptr);
187#endif
188
189	msqid = IPCID_TO_IX(msqid);
190
191	if (msqid < 0 || msqid >= msginfo.msgmni) {
192#ifdef MSG_DEBUG_OK
193		printf("msqid (%d) out of range (0<=msqid<%d)\n", msqid,
194		    msginfo.msgmni);
195#endif
196		return(EINVAL);
197	}
198
199	msqptr = &msqids[msqid];
200
201	if (msqptr->msg_qbytes == 0) {
202#ifdef MSG_DEBUG_OK
203		printf("no such msqid\n");
204#endif
205		return(EINVAL);
206	}
207	if (msqptr->msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) {
208#ifdef MSG_DEBUG_OK
209		printf("wrong sequence number\n");
210#endif
211		return(EINVAL);
212	}
213
214	eval = 0;
215	rval = 0;
216
217	switch (cmd) {
218
219	case IPC_RMID:
220	{
221		struct msg *msghdr;
222		if ((eval = ipcperm(p, &msqptr->msg_perm, IPC_M)))
223			return(eval);
224		/* Free the message headers */
225		msghdr = msqptr->msg_first;
226		while (msghdr != NULL) {
227			struct msg *msghdr_tmp;
228
229			/* Free the segments of each message */
230			msqptr->msg_cbytes -= msghdr->msg_ts;
231			msqptr->msg_qnum--;
232			msghdr_tmp = msghdr;
233			msghdr = msghdr->msg_next;
234			msg_freehdr(msghdr_tmp);
235		}
236
237		if (msqptr->msg_cbytes != 0)
238			panic("msg_cbytes is screwed up");
239		if (msqptr->msg_qnum != 0)
240			panic("msg_qnum is screwed up");
241
242		msqptr->msg_qbytes = 0;	/* Mark it as free */
243
244		wakeup((caddr_t)msqptr);
245	}
246
247		break;
248
249	case IPC_SET:
250		if ((eval = ipcperm(p, &msqptr->msg_perm, IPC_M)))
251			return(eval);
252		if ((eval = copyin(user_msqptr, &msqbuf, sizeof(msqbuf))) != 0)
253			return(eval);
254		if (msqbuf.msg_qbytes > msqptr->msg_qbytes) {
255			eval = suser(p);
256			if (eval)
257				return(eval);
258		}
259		if (msqbuf.msg_qbytes > msginfo.msgmnb) {
260#ifdef MSG_DEBUG_OK
261			printf("can't increase msg_qbytes beyond %d (truncating)\n",
262			    msginfo.msgmnb);
263#endif
264			msqbuf.msg_qbytes = msginfo.msgmnb;	/* silently restrict qbytes to system limit */
265		}
266		if (msqbuf.msg_qbytes == 0) {
267#ifdef MSG_DEBUG_OK
268			printf("can't reduce msg_qbytes to 0\n");
269#endif
270			return(EINVAL);		/* non-standard errno! */
271		}
272		msqptr->msg_perm.uid = msqbuf.msg_perm.uid;	/* change the owner */
273		msqptr->msg_perm.gid = msqbuf.msg_perm.gid;	/* change the owner */
274		msqptr->msg_perm.mode = (msqptr->msg_perm.mode & ~0777) |
275		    (msqbuf.msg_perm.mode & 0777);
276		msqptr->msg_qbytes = msqbuf.msg_qbytes;
277		msqptr->msg_ctime = time_second;
278		break;
279
280	case IPC_STAT:
281		if ((eval = ipcperm(p, &msqptr->msg_perm, IPC_R))) {
282#ifdef MSG_DEBUG_OK
283			printf("requester doesn't have read access\n");
284#endif
285			return(eval);
286		}
287		eval = copyout((caddr_t)msqptr, user_msqptr,
288		    sizeof(struct msqid_ds));
289		break;
290
291	default:
292#ifdef MSG_DEBUG_OK
293		printf("invalid command %d\n", cmd);
294#endif
295		return(EINVAL);
296	}
297
298	if (eval == 0)
299		p->p_retval[0] = rval;
300	return(eval);
301}
302
303#ifndef _SYS_SYSPROTO_H_
304struct msgget_args {
305	key_t	key;
306	int	msgflg;
307};
308#endif
309
310int
311msgget(p, uap)
312	struct proc *p;
313	register struct msgget_args *uap;
314{
315	int msqid, eval;
316	int key = uap->key;
317	int msgflg = uap->msgflg;
318	struct ucred *cred = p->p_ucred;
319	register struct msqid_ds *msqptr = NULL;
320
321#ifdef MSG_DEBUG_OK
322	printf("msgget(0x%x, 0%o)\n", key, msgflg);
323#endif
324
325	if (key != IPC_PRIVATE) {
326		for (msqid = 0; msqid < msginfo.msgmni; msqid++) {
327			msqptr = &msqids[msqid];
328			if (msqptr->msg_qbytes != 0 &&
329			    msqptr->msg_perm.key == key)
330				break;
331		}
332		if (msqid < msginfo.msgmni) {
333#ifdef MSG_DEBUG_OK
334			printf("found public key\n");
335#endif
336			if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) {
337#ifdef MSG_DEBUG_OK
338				printf("not exclusive\n");
339#endif
340				return(EEXIST);
341			}
342			if ((eval = ipcperm(p, &msqptr->msg_perm, msgflg & 0700 ))) {
343#ifdef MSG_DEBUG_OK
344				printf("requester doesn't have 0%o access\n",
345				    msgflg & 0700);
346#endif
347				return(eval);
348			}
349			goto found;
350		}
351	}
352
353#ifdef MSG_DEBUG_OK
354	printf("need to allocate the msqid_ds\n");
355#endif
356	if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) {
357		for (msqid = 0; msqid < msginfo.msgmni; msqid++) {
358			/*
359			 * Look for an unallocated and unlocked msqid_ds.
360			 * msqid_ds's can be locked by msgsnd or msgrcv while
361			 * they are copying the message in/out.  We can't
362			 * re-use the entry until they release it.
363			 */
364			msqptr = &msqids[msqid];
365			if (msqptr->msg_qbytes == 0 &&
366			    (msqptr->msg_perm.mode & MSG_LOCKED) == 0)
367				break;
368		}
369		if (msqid == msginfo.msgmni) {
370#ifdef MSG_DEBUG_OK
371			printf("no more msqid_ds's available\n");
372#endif
373			return(ENOSPC);
374		}
375#ifdef MSG_DEBUG_OK
376		printf("msqid %d is available\n", msqid);
377#endif
378		msqptr->msg_perm.key = key;
379		msqptr->msg_perm.cuid = cred->cr_uid;
380		msqptr->msg_perm.uid = cred->cr_uid;
381		msqptr->msg_perm.cgid = cred->cr_gid;
382		msqptr->msg_perm.gid = cred->cr_gid;
383		msqptr->msg_perm.mode = (msgflg & 0777);
384		/* Make sure that the returned msqid is unique */
385		msqptr->msg_perm.seq++;
386		msqptr->msg_first = NULL;
387		msqptr->msg_last = NULL;
388		msqptr->msg_cbytes = 0;
389		msqptr->msg_qnum = 0;
390		msqptr->msg_qbytes = msginfo.msgmnb;
391		msqptr->msg_lspid = 0;
392		msqptr->msg_lrpid = 0;
393		msqptr->msg_stime = 0;
394		msqptr->msg_rtime = 0;
395		msqptr->msg_ctime = time_second;
396	} else {
397#ifdef MSG_DEBUG_OK
398		printf("didn't find it and wasn't asked to create it\n");
399#endif
400		return(ENOENT);
401	}
402
403found:
404	/* Construct the unique msqid */
405	p->p_retval[0] = IXSEQ_TO_IPCID(msqid, msqptr->msg_perm);
406	return(0);
407}
408
409#ifndef _SYS_SYSPROTO_H_
410struct msgsnd_args {
411	int	msqid;
412	void	*msgp;
413	size_t	msgsz;
414	int	msgflg;
415};
416#endif
417
418int
419msgsnd(p, uap)
420	struct proc *p;
421	register struct msgsnd_args *uap;
422{
423	int msqid = uap->msqid;
424	void *user_msgp = uap->msgp;
425	size_t msgsz = uap->msgsz;
426	int msgflg = uap->msgflg;
427	int segs_needed, eval;
428	register struct msqid_ds *msqptr;
429	register struct msg *msghdr;
430	short next;
431
432#ifdef MSG_DEBUG_OK
433	printf("call to msgsnd(%d, 0x%x, %d, %d)\n", msqid, user_msgp, msgsz,
434	    msgflg);
435#endif
436
437	msqid = IPCID_TO_IX(msqid);
438
439	if (msqid < 0 || msqid >= msginfo.msgmni) {
440#ifdef MSG_DEBUG_OK
441		printf("msqid (%d) out of range (0<=msqid<%d)\n", msqid,
442		    msginfo.msgmni);
443#endif
444		return(EINVAL);
445	}
446
447	msqptr = &msqids[msqid];
448	if (msqptr->msg_qbytes == 0) {
449#ifdef MSG_DEBUG_OK
450		printf("no such message queue id\n");
451#endif
452		return(EINVAL);
453	}
454	if (msqptr->msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) {
455#ifdef MSG_DEBUG_OK
456		printf("wrong sequence number\n");
457#endif
458		return(EINVAL);
459	}
460
461	if ((eval = ipcperm(p, &msqptr->msg_perm, IPC_W))) {
462#ifdef MSG_DEBUG_OK
463		printf("requester doesn't have write access\n");
464#endif
465		return(eval);
466	}
467
468	segs_needed = (msgsz + msginfo.msgssz - 1) / msginfo.msgssz;
469#ifdef MSG_DEBUG_OK
470	printf("msgsz=%d, msgssz=%d, segs_needed=%d\n", msgsz, msginfo.msgssz,
471	    segs_needed);
472#endif
473	for (;;) {
474		int need_more_resources = 0;
475
476		/*
477		 * check msgsz
478		 * (inside this loop in case msg_qbytes changes while we sleep)
479		 */
480
481		if (msgsz > msqptr->msg_qbytes) {
482#ifdef MSG_DEBUG_OK
483			printf("msgsz > msqptr->msg_qbytes\n");
484#endif
485			return(EINVAL);
486		}
487
488		if (msqptr->msg_perm.mode & MSG_LOCKED) {
489#ifdef MSG_DEBUG_OK
490			printf("msqid is locked\n");
491#endif
492			need_more_resources = 1;
493		}
494		if (msgsz + msqptr->msg_cbytes > msqptr->msg_qbytes) {
495#ifdef MSG_DEBUG_OK
496			printf("msgsz + msg_cbytes > msg_qbytes\n");
497#endif
498			need_more_resources = 1;
499		}
500		if (segs_needed > nfree_msgmaps) {
501#ifdef MSG_DEBUG_OK
502			printf("segs_needed > nfree_msgmaps\n");
503#endif
504			need_more_resources = 1;
505		}
506		if (free_msghdrs == NULL) {
507#ifdef MSG_DEBUG_OK
508			printf("no more msghdrs\n");
509#endif
510			need_more_resources = 1;
511		}
512
513		if (need_more_resources) {
514			int we_own_it;
515
516			if ((msgflg & IPC_NOWAIT) != 0) {
517#ifdef MSG_DEBUG_OK
518				printf("need more resources but caller doesn't want to wait\n");
519#endif
520				return(EAGAIN);
521			}
522
523			if ((msqptr->msg_perm.mode & MSG_LOCKED) != 0) {
524#ifdef MSG_DEBUG_OK
525				printf("we don't own the msqid_ds\n");
526#endif
527				we_own_it = 0;
528			} else {
529				/* Force later arrivals to wait for our
530				   request */
531#ifdef MSG_DEBUG_OK
532				printf("we own the msqid_ds\n");
533#endif
534				msqptr->msg_perm.mode |= MSG_LOCKED;
535				we_own_it = 1;
536			}
537#ifdef MSG_DEBUG_OK
538			printf("goodnight\n");
539#endif
540			eval = tsleep((caddr_t)msqptr, (PZERO - 4) | PCATCH,
541			    "msgwait", 0);
542#ifdef MSG_DEBUG_OK
543			printf("good morning, eval=%d\n", eval);
544#endif
545			if (we_own_it)
546				msqptr->msg_perm.mode &= ~MSG_LOCKED;
547			if (eval != 0) {
548#ifdef MSG_DEBUG_OK
549				printf("msgsnd:  interrupted system call\n");
550#endif
551				return(EINTR);
552			}
553
554			/*
555			 * Make sure that the msq queue still exists
556			 */
557
558			if (msqptr->msg_qbytes == 0) {
559#ifdef MSG_DEBUG_OK
560				printf("msqid deleted\n");
561#endif
562				/* The SVID says to return EIDRM. */
563#ifdef EIDRM
564				return(EIDRM);
565#else
566				/* Unfortunately, BSD doesn't define that code
567				   yet! */
568				return(EINVAL);
569#endif
570			}
571
572		} else {
573#ifdef MSG_DEBUG_OK
574			printf("got all the resources that we need\n");
575#endif
576			break;
577		}
578	}
579
580	/*
581	 * We have the resources that we need.
582	 * Make sure!
583	 */
584
585	if (msqptr->msg_perm.mode & MSG_LOCKED)
586		panic("msg_perm.mode & MSG_LOCKED");
587	if (segs_needed > nfree_msgmaps)
588		panic("segs_needed > nfree_msgmaps");
589	if (msgsz + msqptr->msg_cbytes > msqptr->msg_qbytes)
590		panic("msgsz + msg_cbytes > msg_qbytes");
591	if (free_msghdrs == NULL)
592		panic("no more msghdrs");
593
594	/*
595	 * Re-lock the msqid_ds in case we page-fault when copying in the
596	 * message
597	 */
598
599	if ((msqptr->msg_perm.mode & MSG_LOCKED) != 0)
600		panic("msqid_ds is already locked");
601	msqptr->msg_perm.mode |= MSG_LOCKED;
602
603	/*
604	 * Allocate a message header
605	 */
606
607	msghdr = free_msghdrs;
608	free_msghdrs = msghdr->msg_next;
609	msghdr->msg_spot = -1;
610	msghdr->msg_ts = msgsz;
611
612	/*
613	 * Allocate space for the message
614	 */
615
616	while (segs_needed > 0) {
617		if (nfree_msgmaps <= 0)
618			panic("not enough msgmaps");
619		if (free_msgmaps == -1)
620			panic("nil free_msgmaps");
621		next = free_msgmaps;
622		if (next <= -1)
623			panic("next too low #1");
624		if (next >= msginfo.msgseg)
625			panic("next out of range #1");
626#ifdef MSG_DEBUG_OK
627		printf("allocating segment %d to message\n", next);
628#endif
629		free_msgmaps = msgmaps[next].next;
630		nfree_msgmaps--;
631		msgmaps[next].next = msghdr->msg_spot;
632		msghdr->msg_spot = next;
633		segs_needed--;
634	}
635
636	/*
637	 * Copy in the message type
638	 */
639
640	if ((eval = copyin(user_msgp, &msghdr->msg_type,
641	    sizeof(msghdr->msg_type))) != 0) {
642#ifdef MSG_DEBUG_OK
643		printf("error %d copying the message type\n", eval);
644#endif
645		msg_freehdr(msghdr);
646		msqptr->msg_perm.mode &= ~MSG_LOCKED;
647		wakeup((caddr_t)msqptr);
648		return(eval);
649	}
650	user_msgp = (char *)user_msgp + sizeof(msghdr->msg_type);
651
652	/*
653	 * Validate the message type
654	 */
655
656	if (msghdr->msg_type < 1) {
657		msg_freehdr(msghdr);
658		msqptr->msg_perm.mode &= ~MSG_LOCKED;
659		wakeup((caddr_t)msqptr);
660#ifdef MSG_DEBUG_OK
661		printf("mtype (%d) < 1\n", msghdr->msg_type);
662#endif
663		return(EINVAL);
664	}
665
666	/*
667	 * Copy in the message body
668	 */
669
670	next = msghdr->msg_spot;
671	while (msgsz > 0) {
672		size_t tlen;
673		if (msgsz > msginfo.msgssz)
674			tlen = msginfo.msgssz;
675		else
676			tlen = msgsz;
677		if (next <= -1)
678			panic("next too low #2");
679		if (next >= msginfo.msgseg)
680			panic("next out of range #2");
681		if ((eval = copyin(user_msgp, &msgpool[next * msginfo.msgssz],
682		    tlen)) != 0) {
683#ifdef MSG_DEBUG_OK
684			printf("error %d copying in message segment\n", eval);
685#endif
686			msg_freehdr(msghdr);
687			msqptr->msg_perm.mode &= ~MSG_LOCKED;
688			wakeup((caddr_t)msqptr);
689			return(eval);
690		}
691		msgsz -= tlen;
692		user_msgp = (char *)user_msgp + tlen;
693		next = msgmaps[next].next;
694	}
695	if (next != -1)
696		panic("didn't use all the msg segments");
697
698	/*
699	 * We've got the message.  Unlock the msqid_ds.
700	 */
701
702	msqptr->msg_perm.mode &= ~MSG_LOCKED;
703
704	/*
705	 * Make sure that the msqid_ds is still allocated.
706	 */
707
708	if (msqptr->msg_qbytes == 0) {
709		msg_freehdr(msghdr);
710		wakeup((caddr_t)msqptr);
711		/* The SVID says to return EIDRM. */
712#ifdef EIDRM
713		return(EIDRM);
714#else
715		/* Unfortunately, BSD doesn't define that code yet! */
716		return(EINVAL);
717#endif
718	}
719
720	/*
721	 * Put the message into the queue
722	 */
723
724	if (msqptr->msg_first == NULL) {
725		msqptr->msg_first = msghdr;
726		msqptr->msg_last = msghdr;
727	} else {
728		msqptr->msg_last->msg_next = msghdr;
729		msqptr->msg_last = msghdr;
730	}
731	msqptr->msg_last->msg_next = NULL;
732
733	msqptr->msg_cbytes += msghdr->msg_ts;
734	msqptr->msg_qnum++;
735	msqptr->msg_lspid = p->p_pid;
736	msqptr->msg_stime = time_second;
737
738	wakeup((caddr_t)msqptr);
739	p->p_retval[0] = 0;
740	return(0);
741}
742
743#ifndef _SYS_SYSPROTO_H_
744struct msgrcv_args {
745	int	msqid;
746	void	*msgp;
747	size_t	msgsz;
748	long	msgtyp;
749	int	msgflg;
750};
751#endif
752
753int
754msgrcv(p, uap)
755	struct proc *p;
756	register struct msgrcv_args *uap;
757{
758	int msqid = uap->msqid;
759	void *user_msgp = uap->msgp;
760	size_t msgsz = uap->msgsz;
761	long msgtyp = uap->msgtyp;
762	int msgflg = uap->msgflg;
763	size_t len;
764	register struct msqid_ds *msqptr;
765	register struct msg *msghdr;
766	int eval;
767	short next;
768
769#ifdef MSG_DEBUG_OK
770	printf("call to msgrcv(%d, 0x%x, %d, %ld, %d)\n", msqid, user_msgp,
771	    msgsz, msgtyp, msgflg);
772#endif
773
774	msqid = IPCID_TO_IX(msqid);
775
776	if (msqid < 0 || msqid >= msginfo.msgmni) {
777#ifdef MSG_DEBUG_OK
778		printf("msqid (%d) out of range (0<=msqid<%d)\n", msqid,
779		    msginfo.msgmni);
780#endif
781		return(EINVAL);
782	}
783
784	msqptr = &msqids[msqid];
785	if (msqptr->msg_qbytes == 0) {
786#ifdef MSG_DEBUG_OK
787		printf("no such message queue id\n");
788#endif
789		return(EINVAL);
790	}
791	if (msqptr->msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) {
792#ifdef MSG_DEBUG_OK
793		printf("wrong sequence number\n");
794#endif
795		return(EINVAL);
796	}
797
798	if ((eval = ipcperm(p, &msqptr->msg_perm, IPC_R))) {
799#ifdef MSG_DEBUG_OK
800		printf("requester doesn't have read access\n");
801#endif
802		return(eval);
803	}
804
805	msghdr = NULL;
806	while (msghdr == NULL) {
807		if (msgtyp == 0) {
808			msghdr = msqptr->msg_first;
809			if (msghdr != NULL) {
810				if (msgsz < msghdr->msg_ts &&
811				    (msgflg & MSG_NOERROR) == 0) {
812#ifdef MSG_DEBUG_OK
813					printf("first message on the queue is too big (want %d, got %d)\n",
814					    msgsz, msghdr->msg_ts);
815#endif
816					return(E2BIG);
817				}
818				if (msqptr->msg_first == msqptr->msg_last) {
819					msqptr->msg_first = NULL;
820					msqptr->msg_last = NULL;
821				} else {
822					msqptr->msg_first = msghdr->msg_next;
823					if (msqptr->msg_first == NULL)
824						panic("msg_first/last screwed up #1");
825				}
826			}
827		} else {
828			struct msg *previous;
829			struct msg **prev;
830
831			previous = NULL;
832			prev = &(msqptr->msg_first);
833			while ((msghdr = *prev) != NULL) {
834				/*
835				 * Is this message's type an exact match or is
836				 * this message's type less than or equal to
837				 * the absolute value of a negative msgtyp?
838				 * Note that the second half of this test can
839				 * NEVER be true if msgtyp is positive since
840				 * msg_type is always positive!
841				 */
842
843				if (msgtyp == msghdr->msg_type ||
844				    msghdr->msg_type <= -msgtyp) {
845#ifdef MSG_DEBUG_OK
846					printf("found message type %d, requested %d\n",
847					    msghdr->msg_type, msgtyp);
848#endif
849					if (msgsz < msghdr->msg_ts &&
850					    (msgflg & MSG_NOERROR) == 0) {
851#ifdef MSG_DEBUG_OK
852						printf("requested message on the queue is too big (want %d, got %d)\n",
853						    msgsz, msghdr->msg_ts);
854#endif
855						return(E2BIG);
856					}
857					*prev = msghdr->msg_next;
858					if (msghdr == msqptr->msg_last) {
859						if (previous == NULL) {
860							if (prev !=
861							    &msqptr->msg_first)
862								panic("msg_first/last screwed up #2");
863							msqptr->msg_first =
864							    NULL;
865							msqptr->msg_last =
866							    NULL;
867						} else {
868							if (prev ==
869							    &msqptr->msg_first)
870								panic("msg_first/last screwed up #3");
871							msqptr->msg_last =
872							    previous;
873						}
874					}
875					break;
876				}
877				previous = msghdr;
878				prev = &(msghdr->msg_next);
879			}
880		}
881
882		/*
883		 * We've either extracted the msghdr for the appropriate
884		 * message or there isn't one.
885		 * If there is one then bail out of this loop.
886		 */
887
888		if (msghdr != NULL)
889			break;
890
891		/*
892		 * Hmph!  No message found.  Does the user want to wait?
893		 */
894
895		if ((msgflg & IPC_NOWAIT) != 0) {
896#ifdef MSG_DEBUG_OK
897			printf("no appropriate message found (msgtyp=%d)\n",
898			    msgtyp);
899#endif
900			/* The SVID says to return ENOMSG. */
901#ifdef ENOMSG
902			return(ENOMSG);
903#else
904			/* Unfortunately, BSD doesn't define that code yet! */
905			return(EAGAIN);
906#endif
907		}
908
909		/*
910		 * Wait for something to happen
911		 */
912
913#ifdef MSG_DEBUG_OK
914		printf("msgrcv:  goodnight\n");
915#endif
916		eval = tsleep((caddr_t)msqptr, (PZERO - 4) | PCATCH, "msgwait",
917		    0);
918#ifdef MSG_DEBUG_OK
919		printf("msgrcv:  good morning (eval=%d)\n", eval);
920#endif
921
922		if (eval != 0) {
923#ifdef MSG_DEBUG_OK
924			printf("msgsnd:  interrupted system call\n");
925#endif
926			return(EINTR);
927		}
928
929		/*
930		 * Make sure that the msq queue still exists
931		 */
932
933		if (msqptr->msg_qbytes == 0 ||
934		    msqptr->msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) {
935#ifdef MSG_DEBUG_OK
936			printf("msqid deleted\n");
937#endif
938			/* The SVID says to return EIDRM. */
939#ifdef EIDRM
940			return(EIDRM);
941#else
942			/* Unfortunately, BSD doesn't define that code yet! */
943			return(EINVAL);
944#endif
945		}
946	}
947
948	/*
949	 * Return the message to the user.
950	 *
951	 * First, do the bookkeeping (before we risk being interrupted).
952	 */
953
954	msqptr->msg_cbytes -= msghdr->msg_ts;
955	msqptr->msg_qnum--;
956	msqptr->msg_lrpid = p->p_pid;
957	msqptr->msg_rtime = time_second;
958
959	/*
960	 * Make msgsz the actual amount that we'll be returning.
961	 * Note that this effectively truncates the message if it is too long
962	 * (since msgsz is never increased).
963	 */
964
965#ifdef MSG_DEBUG_OK
966	printf("found a message, msgsz=%d, msg_ts=%d\n", msgsz,
967	    msghdr->msg_ts);
968#endif
969	if (msgsz > msghdr->msg_ts)
970		msgsz = msghdr->msg_ts;
971
972	/*
973	 * Return the type to the user.
974	 */
975
976	eval = copyout((caddr_t)&(msghdr->msg_type), user_msgp,
977	    sizeof(msghdr->msg_type));
978	if (eval != 0) {
979#ifdef MSG_DEBUG_OK
980		printf("error (%d) copying out message type\n", eval);
981#endif
982		msg_freehdr(msghdr);
983		wakeup((caddr_t)msqptr);
984		return(eval);
985	}
986	user_msgp = (char *)user_msgp + sizeof(msghdr->msg_type);
987
988	/*
989	 * Return the segments to the user
990	 */
991
992	next = msghdr->msg_spot;
993	for (len = 0; len < msgsz; len += msginfo.msgssz) {
994		size_t tlen;
995
996		if (msgsz - len > msginfo.msgssz)
997			tlen = msginfo.msgssz;
998		else
999			tlen = msgsz - len;
1000		if (next <= -1)
1001			panic("next too low #3");
1002		if (next >= msginfo.msgseg)
1003			panic("next out of range #3");
1004		eval = copyout((caddr_t)&msgpool[next * msginfo.msgssz],
1005		    user_msgp, tlen);
1006		if (eval != 0) {
1007#ifdef MSG_DEBUG_OK
1008			printf("error (%d) copying out message segment\n",
1009			    eval);
1010#endif
1011			msg_freehdr(msghdr);
1012			wakeup((caddr_t)msqptr);
1013			return(eval);
1014		}
1015		user_msgp = (char *)user_msgp + tlen;
1016		next = msgmaps[next].next;
1017	}
1018
1019	/*
1020	 * Done, return the actual number of bytes copied out.
1021	 */
1022
1023	msg_freehdr(msghdr);
1024	wakeup((caddr_t)msqptr);
1025	p->p_retval[0] = msgsz;
1026	return(0);
1027}
1028