xref: /freebsd/sys/kern/uipc_sockbuf.c (revision e627b39baccd1ec9129690167cf5e6d860509655)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	@(#)uipc_socket2.c	8.1 (Berkeley) 6/10/93
34  * $Id: uipc_socket2.c,v 1.13 1996/08/19 19:22:26 julian Exp $
35  */
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.h>
40 #include <sys/proc.h>
41 #include <sys/file.h>
42 #include <sys/buf.h>
43 #include <sys/malloc.h>
44 #include <sys/mbuf.h>
45 #include <sys/protosw.h>
46 #include <sys/stat.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/signalvar.h>
50 #include <sys/sysctl.h>
51 
52 /*
53  * Primitive routines for operating on sockets and socket buffers
54  */
55 
56 u_long	sb_max = SB_MAX;		/* XXX should be static */
57 SYSCTL_INT(_kern, KERN_MAXSOCKBUF, maxsockbuf, CTLFLAG_RW, &sb_max, 0, "")
58 
59 static	u_long sb_efficiency = 8;	/* parameter for sbreserve() */
60 SYSCTL_INT(_kern, OID_AUTO, sockbuf_waste_factor, CTLFLAG_RW, &sb_efficiency,
61 	   0, "");
62 
63 static int sominqueue = 0;
64 SYSCTL_INT(_kern, KERN_SOMINQUEUE, sominqueue, CTLFLAG_RW, &sominqueue, 0, "");
65 
66 /*
67  * Procedures to manipulate state flags of socket
68  * and do appropriate wakeups.  Normal sequence from the
69  * active (originating) side is that soisconnecting() is
70  * called during processing of connect() call,
71  * resulting in an eventual call to soisconnected() if/when the
72  * connection is established.  When the connection is torn down
73  * soisdisconnecting() is called during processing of disconnect() call,
74  * and soisdisconnected() is called when the connection to the peer
75  * is totally severed.  The semantics of these routines are such that
76  * connectionless protocols can call soisconnected() and soisdisconnected()
77  * only, bypassing the in-progress calls when setting up a ``connection''
78  * takes no time.
79  *
80  * From the passive side, a socket is created with
81  * two queues of sockets: so_q0 for connections in progress
82  * and so_q for connections already made and awaiting user acceptance.
83  * As a protocol is preparing incoming connections, it creates a socket
84  * structure queued on so_q0 by calling sonewconn().  When the connection
85  * is established, soisconnected() is called, and transfers the
86  * socket structure to so_q, making it available to accept().
87  *
88  * If a socket is closed with sockets on either
89  * so_q0 or so_q, these sockets are dropped.
90  *
91  * If higher level protocols are implemented in
92  * the kernel, the wakeups done here will sometimes
93  * cause software-interrupt process scheduling.
94  */
95 
96 void
97 soisconnecting(so)
98 	register struct socket *so;
99 {
100 
101 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
102 	so->so_state |= SS_ISCONNECTING;
103 }
104 
105 void
106 soisconnected(so)
107 	register struct socket *so;
108 {
109 	register struct socket *head = so->so_head;
110 
111 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
112 	so->so_state |= SS_ISCONNECTED;
113 	if (head && (so->so_state & SS_INCOMP)) {
114 		TAILQ_REMOVE(&head->so_incomp, so, so_list);
115 		so->so_state &= ~SS_INCOMP;
116 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
117 		so->so_state |= SS_COMP;
118 		sorwakeup(head);
119 		wakeup((caddr_t)&head->so_timeo);
120 	} else {
121 		wakeup((caddr_t)&so->so_timeo);
122 		sorwakeup(so);
123 		sowwakeup(so);
124 	}
125 }
126 
127 void
128 soisdisconnecting(so)
129 	register struct socket *so;
130 {
131 
132 	so->so_state &= ~SS_ISCONNECTING;
133 	so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
134 	wakeup((caddr_t)&so->so_timeo);
135 	sowwakeup(so);
136 	sorwakeup(so);
137 }
138 
139 void
140 soisdisconnected(so)
141 	register struct socket *so;
142 {
143 
144 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
145 	so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE);
146 	wakeup((caddr_t)&so->so_timeo);
147 	sowwakeup(so);
148 	sorwakeup(so);
149 }
150 
151 /*
152  * When an attempt at a new connection is noted on a socket
153  * which accepts connections, sonewconn is called.  If the
154  * connection is possible (subject to space constraints, etc.)
155  * then we allocate a new structure, propoerly linked into the
156  * data structure of the original socket, and return this.
157  * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
158  *
159  * Currently, sonewconn() is defined as sonewconn1() in socketvar.h
160  * to catch calls that are missing the (new) second parameter.
161  */
162 struct socket *
163 sonewconn1(head, connstatus)
164 	register struct socket *head;
165 	int connstatus;
166 {
167 	register struct socket *so;
168 
169 	if ((head->so_qlen > 3 * head->so_qlimit / 2) &&
170 	    (head->so_qlen > sominqueue))
171 		return ((struct socket *)0);
172 	MALLOC(so, struct socket *, sizeof(*so), M_SOCKET, M_DONTWAIT);
173 	if (so == NULL)
174 		return ((struct socket *)0);
175 	bzero((caddr_t)so, sizeof(*so));
176 	so->so_head = head;
177 	so->so_type = head->so_type;
178 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
179 	so->so_linger = head->so_linger;
180 	so->so_state = head->so_state | SS_NOFDREF;
181 	so->so_proto = head->so_proto;
182 	so->so_timeo = head->so_timeo;
183 	so->so_pgid = head->so_pgid;
184 	(void) soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat);
185 	if (connstatus) {
186 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
187 		so->so_state |= SS_COMP;
188 	} else {
189 		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
190 		so->so_state |= SS_INCOMP;
191 	}
192 	head->so_qlen++;
193 	if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0)) {
194 		if (so->so_state & SS_COMP) {
195 			TAILQ_REMOVE(&head->so_comp, so, so_list);
196 		} else {
197 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
198 		}
199 		head->so_qlen--;
200 		(void) free((caddr_t)so, M_SOCKET);
201 		return ((struct socket *)0);
202 	}
203 	if (connstatus) {
204 		sorwakeup(head);
205 		wakeup((caddr_t)&head->so_timeo);
206 		so->so_state |= connstatus;
207 	}
208 	return (so);
209 }
210 
211 /*
212  * Socantsendmore indicates that no more data will be sent on the
213  * socket; it would normally be applied to a socket when the user
214  * informs the system that no more data is to be sent, by the protocol
215  * code (in case PRU_SHUTDOWN).  Socantrcvmore indicates that no more data
216  * will be received, and will normally be applied to the socket by a
217  * protocol when it detects that the peer will send no more data.
218  * Data queued for reading in the socket may yet be read.
219  */
220 
221 void
222 socantsendmore(so)
223 	struct socket *so;
224 {
225 
226 	so->so_state |= SS_CANTSENDMORE;
227 	sowwakeup(so);
228 }
229 
230 void
231 socantrcvmore(so)
232 	struct socket *so;
233 {
234 
235 	so->so_state |= SS_CANTRCVMORE;
236 	sorwakeup(so);
237 }
238 
239 /*
240  * Wait for data to arrive at/drain from a socket buffer.
241  */
242 int
243 sbwait(sb)
244 	struct sockbuf *sb;
245 {
246 
247 	sb->sb_flags |= SB_WAIT;
248 	return (tsleep((caddr_t)&sb->sb_cc,
249 	    (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
250 	    sb->sb_timeo));
251 }
252 
253 /*
254  * Lock a sockbuf already known to be locked;
255  * return any error returned from sleep (EINTR).
256  */
257 int
258 sb_lock(sb)
259 	register struct sockbuf *sb;
260 {
261 	int error;
262 
263 	while (sb->sb_flags & SB_LOCK) {
264 		sb->sb_flags |= SB_WANT;
265 		error = tsleep((caddr_t)&sb->sb_flags,
266 		    (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH,
267 		    "sblock", 0);
268 		if (error)
269 			return (error);
270 	}
271 	sb->sb_flags |= SB_LOCK;
272 	return (0);
273 }
274 
275 /*
276  * Wakeup processes waiting on a socket buffer.
277  * Do asynchronous notification via SIGIO
278  * if the socket has the SS_ASYNC flag set.
279  */
280 void
281 sowakeup(so, sb)
282 	register struct socket *so;
283 	register struct sockbuf *sb;
284 {
285 	struct proc *p;
286 
287 	selwakeup(&sb->sb_sel);
288 	sb->sb_flags &= ~SB_SEL;
289 	if (sb->sb_flags & SB_WAIT) {
290 		sb->sb_flags &= ~SB_WAIT;
291 		wakeup((caddr_t)&sb->sb_cc);
292 	}
293 	if (so->so_state & SS_ASYNC) {
294 		if (so->so_pgid < 0)
295 			gsignal(-so->so_pgid, SIGIO);
296 		else if (so->so_pgid > 0 && (p = pfind(so->so_pgid)) != 0)
297 			psignal(p, SIGIO);
298 	}
299 }
300 
301 /*
302  * Socket buffer (struct sockbuf) utility routines.
303  *
304  * Each socket contains two socket buffers: one for sending data and
305  * one for receiving data.  Each buffer contains a queue of mbufs,
306  * information about the number of mbufs and amount of data in the
307  * queue, and other fields allowing select() statements and notification
308  * on data availability to be implemented.
309  *
310  * Data stored in a socket buffer is maintained as a list of records.
311  * Each record is a list of mbufs chained together with the m_next
312  * field.  Records are chained together with the m_nextpkt field. The upper
313  * level routine soreceive() expects the following conventions to be
314  * observed when placing information in the receive buffer:
315  *
316  * 1. If the protocol requires each message be preceded by the sender's
317  *    name, then a record containing that name must be present before
318  *    any associated data (mbuf's must be of type MT_SONAME).
319  * 2. If the protocol supports the exchange of ``access rights'' (really
320  *    just additional data associated with the message), and there are
321  *    ``rights'' to be received, then a record containing this data
322  *    should be present (mbuf's must be of type MT_RIGHTS).
323  * 3. If a name or rights record exists, then it must be followed by
324  *    a data record, perhaps of zero length.
325  *
326  * Before using a new socket structure it is first necessary to reserve
327  * buffer space to the socket, by calling sbreserve().  This should commit
328  * some of the available buffer space in the system buffer pool for the
329  * socket (currently, it does nothing but enforce limits).  The space
330  * should be released by calling sbrelease() when the socket is destroyed.
331  */
332 
333 int
334 soreserve(so, sndcc, rcvcc)
335 	register struct socket *so;
336 	u_long sndcc, rcvcc;
337 {
338 
339 	if (sbreserve(&so->so_snd, sndcc) == 0)
340 		goto bad;
341 	if (sbreserve(&so->so_rcv, rcvcc) == 0)
342 		goto bad2;
343 	if (so->so_rcv.sb_lowat == 0)
344 		so->so_rcv.sb_lowat = 1;
345 	if (so->so_snd.sb_lowat == 0)
346 		so->so_snd.sb_lowat = MCLBYTES;
347 	if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
348 		so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
349 	return (0);
350 bad2:
351 	sbrelease(&so->so_snd);
352 bad:
353 	return (ENOBUFS);
354 }
355 
356 /*
357  * Allot mbufs to a sockbuf.
358  * Attempt to scale mbmax so that mbcnt doesn't become limiting
359  * if buffering efficiency is near the normal case.
360  */
361 int
362 sbreserve(sb, cc)
363 	struct sockbuf *sb;
364 	u_long cc;
365 {
366 
367 	if (cc > sb_max * MCLBYTES / (MSIZE + MCLBYTES))
368 		return (0);
369 	sb->sb_hiwat = cc;
370 	sb->sb_mbmax = min(cc * sb_efficiency, sb_max);
371 	if (sb->sb_lowat > sb->sb_hiwat)
372 		sb->sb_lowat = sb->sb_hiwat;
373 	return (1);
374 }
375 
376 /*
377  * Free mbufs held by a socket, and reserved mbuf space.
378  */
379 void
380 sbrelease(sb)
381 	struct sockbuf *sb;
382 {
383 
384 	sbflush(sb);
385 	sb->sb_hiwat = sb->sb_mbmax = 0;
386 }
387 
388 /*
389  * Routines to add and remove
390  * data from an mbuf queue.
391  *
392  * The routines sbappend() or sbappendrecord() are normally called to
393  * append new mbufs to a socket buffer, after checking that adequate
394  * space is available, comparing the function sbspace() with the amount
395  * of data to be added.  sbappendrecord() differs from sbappend() in
396  * that data supplied is treated as the beginning of a new record.
397  * To place a sender's address, optional access rights, and data in a
398  * socket receive buffer, sbappendaddr() should be used.  To place
399  * access rights and data in a socket receive buffer, sbappendrights()
400  * should be used.  In either case, the new data begins a new record.
401  * Note that unlike sbappend() and sbappendrecord(), these routines check
402  * for the caller that there will be enough space to store the data.
403  * Each fails if there is not enough space, or if it cannot find mbufs
404  * to store additional information in.
405  *
406  * Reliable protocols may use the socket send buffer to hold data
407  * awaiting acknowledgement.  Data is normally copied from a socket
408  * send buffer in a protocol with m_copy for output to a peer,
409  * and then removing the data from the socket buffer with sbdrop()
410  * or sbdroprecord() when the data is acknowledged by the peer.
411  */
412 
413 /*
414  * Append mbuf chain m to the last record in the
415  * socket buffer sb.  The additional space associated
416  * the mbuf chain is recorded in sb.  Empty mbufs are
417  * discarded and mbufs are compacted where possible.
418  */
419 void
420 sbappend(sb, m)
421 	struct sockbuf *sb;
422 	struct mbuf *m;
423 {
424 	register struct mbuf *n;
425 
426 	if (m == 0)
427 		return;
428 	n = sb->sb_mb;
429 	if (n) {
430 		while (n->m_nextpkt)
431 			n = n->m_nextpkt;
432 		do {
433 			if (n->m_flags & M_EOR) {
434 				sbappendrecord(sb, m); /* XXXXXX!!!! */
435 				return;
436 			}
437 		} while (n->m_next && (n = n->m_next));
438 	}
439 	sbcompress(sb, m, n);
440 }
441 
442 #ifdef SOCKBUF_DEBUG
443 void
444 sbcheck(sb)
445 	register struct sockbuf *sb;
446 {
447 	register struct mbuf *m;
448 	register int len = 0, mbcnt = 0;
449 
450 	for (m = sb->sb_mb; m; m = m->m_next) {
451 		len += m->m_len;
452 		mbcnt += MSIZE;
453 		if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
454 			mbcnt += m->m_ext.ext_size;
455 		if (m->m_nextpkt)
456 			panic("sbcheck nextpkt");
457 	}
458 	if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
459 		printf("cc %d != %d || mbcnt %d != %d\n", len, sb->sb_cc,
460 		    mbcnt, sb->sb_mbcnt);
461 		panic("sbcheck");
462 	}
463 }
464 #endif
465 
466 /*
467  * As above, except the mbuf chain
468  * begins a new record.
469  */
470 void
471 sbappendrecord(sb, m0)
472 	register struct sockbuf *sb;
473 	register struct mbuf *m0;
474 {
475 	register struct mbuf *m;
476 
477 	if (m0 == 0)
478 		return;
479 	m = sb->sb_mb;
480 	if (m)
481 		while (m->m_nextpkt)
482 			m = m->m_nextpkt;
483 	/*
484 	 * Put the first mbuf on the queue.
485 	 * Note this permits zero length records.
486 	 */
487 	sballoc(sb, m0);
488 	if (m)
489 		m->m_nextpkt = m0;
490 	else
491 		sb->sb_mb = m0;
492 	m = m0->m_next;
493 	m0->m_next = 0;
494 	if (m && (m0->m_flags & M_EOR)) {
495 		m0->m_flags &= ~M_EOR;
496 		m->m_flags |= M_EOR;
497 	}
498 	sbcompress(sb, m, m0);
499 }
500 
501 /*
502  * As above except that OOB data
503  * is inserted at the beginning of the sockbuf,
504  * but after any other OOB data.
505  */
506 void
507 sbinsertoob(sb, m0)
508 	register struct sockbuf *sb;
509 	register struct mbuf *m0;
510 {
511 	register struct mbuf *m;
512 	register struct mbuf **mp;
513 
514 	if (m0 == 0)
515 		return;
516 	for (mp = &sb->sb_mb; *mp ; mp = &((*mp)->m_nextpkt)) {
517 	    m = *mp;
518 	    again:
519 		switch (m->m_type) {
520 
521 		case MT_OOBDATA:
522 			continue;		/* WANT next train */
523 
524 		case MT_CONTROL:
525 			m = m->m_next;
526 			if (m)
527 				goto again;	/* inspect THIS train further */
528 		}
529 		break;
530 	}
531 	/*
532 	 * Put the first mbuf on the queue.
533 	 * Note this permits zero length records.
534 	 */
535 	sballoc(sb, m0);
536 	m0->m_nextpkt = *mp;
537 	*mp = m0;
538 	m = m0->m_next;
539 	m0->m_next = 0;
540 	if (m && (m0->m_flags & M_EOR)) {
541 		m0->m_flags &= ~M_EOR;
542 		m->m_flags |= M_EOR;
543 	}
544 	sbcompress(sb, m, m0);
545 }
546 
547 /*
548  * Append address and data, and optionally, control (ancillary) data
549  * to the receive queue of a socket.  If present,
550  * m0 must include a packet header with total length.
551  * Returns 0 if no space in sockbuf or insufficient mbufs.
552  */
553 int
554 sbappendaddr(sb, asa, m0, control)
555 	register struct sockbuf *sb;
556 	struct sockaddr *asa;
557 	struct mbuf *m0, *control;
558 {
559 	register struct mbuf *m, *n;
560 	int space = asa->sa_len;
561 
562 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
563 panic("sbappendaddr");
564 	if (m0)
565 		space += m0->m_pkthdr.len;
566 	for (n = control; n; n = n->m_next) {
567 		space += n->m_len;
568 		if (n->m_next == 0)	/* keep pointer to last control buf */
569 			break;
570 	}
571 	if (space > sbspace(sb))
572 		return (0);
573 	if (asa->sa_len > MLEN)
574 		return (0);
575 	MGET(m, M_DONTWAIT, MT_SONAME);
576 	if (m == 0)
577 		return (0);
578 	m->m_len = asa->sa_len;
579 	bcopy((caddr_t)asa, mtod(m, caddr_t), asa->sa_len);
580 	if (n)
581 		n->m_next = m0;		/* concatenate data to control */
582 	else
583 		control = m0;
584 	m->m_next = control;
585 	for (n = m; n; n = n->m_next)
586 		sballoc(sb, n);
587 	n = sb->sb_mb;
588 	if (n) {
589 		while (n->m_nextpkt)
590 			n = n->m_nextpkt;
591 		n->m_nextpkt = m;
592 	} else
593 		sb->sb_mb = m;
594 	return (1);
595 }
596 
597 int
598 sbappendcontrol(sb, m0, control)
599 	struct sockbuf *sb;
600 	struct mbuf *control, *m0;
601 {
602 	register struct mbuf *m, *n;
603 	int space = 0;
604 
605 	if (control == 0)
606 		panic("sbappendcontrol");
607 	for (m = control; ; m = m->m_next) {
608 		space += m->m_len;
609 		if (m->m_next == 0)
610 			break;
611 	}
612 	n = m;			/* save pointer to last control buffer */
613 	for (m = m0; m; m = m->m_next)
614 		space += m->m_len;
615 	if (space > sbspace(sb))
616 		return (0);
617 	n->m_next = m0;			/* concatenate data to control */
618 	for (m = control; m; m = m->m_next)
619 		sballoc(sb, m);
620 	n = sb->sb_mb;
621 	if (n) {
622 		while (n->m_nextpkt)
623 			n = n->m_nextpkt;
624 		n->m_nextpkt = control;
625 	} else
626 		sb->sb_mb = control;
627 	return (1);
628 }
629 
630 /*
631  * Compress mbuf chain m into the socket
632  * buffer sb following mbuf n.  If n
633  * is null, the buffer is presumed empty.
634  */
635 void
636 sbcompress(sb, m, n)
637 	register struct sockbuf *sb;
638 	register struct mbuf *m, *n;
639 {
640 	register int eor = 0;
641 	register struct mbuf *o;
642 
643 	while (m) {
644 		eor |= m->m_flags & M_EOR;
645 		if (m->m_len == 0 &&
646 		    (eor == 0 ||
647 		     (((o = m->m_next) || (o = n)) &&
648 		      o->m_type == m->m_type))) {
649 			m = m_free(m);
650 			continue;
651 		}
652 		if (n && (n->m_flags & (M_EXT | M_EOR)) == 0 &&
653 		    (n->m_data + n->m_len + m->m_len) < &n->m_dat[MLEN] &&
654 		    n->m_type == m->m_type) {
655 			bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
656 			    (unsigned)m->m_len);
657 			n->m_len += m->m_len;
658 			sb->sb_cc += m->m_len;
659 			m = m_free(m);
660 			continue;
661 		}
662 		if (n)
663 			n->m_next = m;
664 		else
665 			sb->sb_mb = m;
666 		sballoc(sb, m);
667 		n = m;
668 		m->m_flags &= ~M_EOR;
669 		m = m->m_next;
670 		n->m_next = 0;
671 	}
672 	if (eor) {
673 		if (n)
674 			n->m_flags |= eor;
675 		else
676 			printf("semi-panic: sbcompress\n");
677 	}
678 }
679 
680 /*
681  * Free all mbufs in a sockbuf.
682  * Check that all resources are reclaimed.
683  */
684 void
685 sbflush(sb)
686 	register struct sockbuf *sb;
687 {
688 
689 	if (sb->sb_flags & SB_LOCK)
690 		panic("sbflush");
691 	while (sb->sb_mbcnt)
692 		sbdrop(sb, (int)sb->sb_cc);
693 	if (sb->sb_cc || sb->sb_mb)
694 		panic("sbflush 2");
695 }
696 
697 /*
698  * Drop data from (the front of) a sockbuf.
699  */
700 void
701 sbdrop(sb, len)
702 	register struct sockbuf *sb;
703 	register int len;
704 {
705 	register struct mbuf *m, *mn;
706 	struct mbuf *next;
707 
708 	next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
709 	while (len > 0) {
710 		if (m == 0) {
711 			if (next == 0)
712 				panic("sbdrop");
713 			m = next;
714 			next = m->m_nextpkt;
715 			continue;
716 		}
717 		if (m->m_len > len) {
718 			m->m_len -= len;
719 			m->m_data += len;
720 			sb->sb_cc -= len;
721 			break;
722 		}
723 		len -= m->m_len;
724 		sbfree(sb, m);
725 		MFREE(m, mn);
726 		m = mn;
727 	}
728 	while (m && m->m_len == 0) {
729 		sbfree(sb, m);
730 		MFREE(m, mn);
731 		m = mn;
732 	}
733 	if (m) {
734 		sb->sb_mb = m;
735 		m->m_nextpkt = next;
736 	} else
737 		sb->sb_mb = next;
738 }
739 
740 /*
741  * Drop a record off the front of a sockbuf
742  * and move the next record to the front.
743  */
744 void
745 sbdroprecord(sb)
746 	register struct sockbuf *sb;
747 {
748 	register struct mbuf *m, *mn;
749 
750 	m = sb->sb_mb;
751 	if (m) {
752 		sb->sb_mb = m->m_nextpkt;
753 		do {
754 			sbfree(sb, m);
755 			MFREE(m, mn);
756 			m = mn;
757 		} while (m);
758 	}
759 }
760 
761 #ifdef PRU_OLDSTYLE
762 /*
763  * The following routines mediate between the old-style `pr_usrreq'
764  * protocol implementations and the new-style `struct pr_usrreqs'
765  * calling convention.
766  */
767 
768 /* syntactic sugar */
769 #define	nomb	(struct mbuf *)0
770 
771 static int
772 old_abort(struct socket *so)
773 {
774 	return so->so_proto->pr_ousrreq(so, PRU_ABORT, nomb, nomb, nomb);
775 }
776 
777 static int
778 old_accept(struct socket *so, struct mbuf *nam)
779 {
780 	return so->so_proto->pr_ousrreq(so, PRU_ACCEPT, nomb,  nam, nomb);
781 }
782 
783 static int
784 old_attach(struct socket *so, int proto)
785 {
786 	return so->so_proto->pr_ousrreq(so, PRU_ATTACH, nomb,
787 				       (struct mbuf *)proto, /* XXX */
788 				       nomb);
789 }
790 
791 static int
792 old_bind(struct socket *so, struct mbuf *nam)
793 {
794 	return so->so_proto->pr_ousrreq(so, PRU_BIND, nomb, nam, nomb);
795 }
796 
797 static int
798 old_connect(struct socket *so, struct mbuf *nam)
799 {
800 	return so->so_proto->pr_ousrreq(so, PRU_CONNECT, nomb, nam, nomb);
801 }
802 
803 static int
804 old_connect2(struct socket *so1, struct socket *so2)
805 {
806 	return so1->so_proto->pr_ousrreq(so1, PRU_CONNECT2, nomb,
807 				       (struct mbuf *)so2, nomb);
808 }
809 
810 static int
811 old_control(struct socket *so, int cmd, caddr_t data, struct ifnet *ifp)
812 {
813 	return so->so_proto->pr_ousrreq(so, PRU_CONTROL, (struct mbuf *)cmd,
814 				       (struct mbuf *)data,
815 				       (struct mbuf *)ifp);
816 }
817 
818 static int
819 old_detach(struct socket *so)
820 {
821 	return so->so_proto->pr_ousrreq(so, PRU_DETACH, nomb, nomb, nomb);
822 }
823 
824 static int
825 old_disconnect(struct socket *so)
826 {
827 	return so->so_proto->pr_ousrreq(so, PRU_DISCONNECT, nomb, nomb, nomb);
828 }
829 
830 static int
831 old_listen(struct socket *so)
832 {
833 	return so->so_proto->pr_ousrreq(so, PRU_LISTEN, nomb, nomb, nomb);
834 }
835 
836 static int
837 old_peeraddr(struct socket *so, struct mbuf *nam)
838 {
839 	return so->so_proto->pr_ousrreq(so, PRU_PEERADDR, nomb, nam, nomb);
840 }
841 
842 static int
843 old_rcvd(struct socket *so, int flags)
844 {
845 	return so->so_proto->pr_ousrreq(so, PRU_RCVD, nomb,
846 				       (struct mbuf *)flags, /* XXX */
847 				       nomb);
848 }
849 
850 static int
851 old_rcvoob(struct socket *so, struct mbuf *m, int flags)
852 {
853 	return so->so_proto->pr_ousrreq(so, PRU_RCVOOB, m,
854 				       (struct mbuf *)flags, /* XXX */
855 				       nomb);
856 }
857 
858 static int
859 old_send(struct socket *so, int flags, struct mbuf *m, struct mbuf *addr,
860 	 struct mbuf *control)
861 {
862 	int req;
863 
864 	if (flags & PRUS_OOB) {
865 		req = PRU_SENDOOB;
866 	} else if(flags & PRUS_EOF) {
867 		req = PRU_SEND_EOF;
868 	} else {
869 		req = PRU_SEND;
870 	}
871 	return so->so_proto->pr_ousrreq(so, req, m, addr, control);
872 }
873 
874 static int
875 old_sense(struct socket *so, struct stat *sb)
876 {
877 	return so->so_proto->pr_ousrreq(so, PRU_SENSE, (struct mbuf *)sb,
878 				       nomb, nomb);
879 }
880 
881 static int
882 old_shutdown(struct socket *so)
883 {
884 	return so->so_proto->pr_ousrreq(so, PRU_SHUTDOWN, nomb, nomb, nomb);
885 }
886 
887 static int
888 old_sockaddr(struct socket *so, struct mbuf *nam)
889 {
890 	return so->so_proto->pr_ousrreq(so, PRU_SOCKADDR, nomb, nam, nomb);
891 }
892 
893 struct pr_usrreqs pru_oldstyle = {
894 	old_abort, old_accept, old_attach, old_bind, old_connect,
895 	old_connect2, old_control, old_detach, old_disconnect,
896 	old_listen, old_peeraddr, old_rcvd, old_rcvoob, old_send,
897 	old_sense, old_shutdown, old_sockaddr
898 };
899 
900 #endif /* PRU_OLDSTYLE */
901 
902 /*
903  * Some routines that return EOPNOTSUPP for entry points that are not
904  * supported by a protocol.  Fill in as needed.
905  */
906 int
907 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
908 {
909 	return EOPNOTSUPP;
910 }
911