xref: /freebsd/sys/kern/uipc_sockbuf.c (revision 4a0f765fbf09711e612e86fce8bb09ec43f482d9)
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  *	$Id$
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 /*
64  * Procedures to manipulate state flags of socket
65  * and do appropriate wakeups.  Normal sequence from the
66  * active (originating) side is that soisconnecting() is
67  * called during processing of connect() call,
68  * resulting in an eventual call to soisconnected() if/when the
69  * connection is established.  When the connection is torn down
70  * soisdisconnecting() is called during processing of disconnect() call,
71  * and soisdisconnected() is called when the connection to the peer
72  * is totally severed.  The semantics of these routines are such that
73  * connectionless protocols can call soisconnected() and soisdisconnected()
74  * only, bypassing the in-progress calls when setting up a ``connection''
75  * takes no time.
76  *
77  * From the passive side, a socket is created with
78  * two queues of sockets: so_q0 for connections in progress
79  * and so_q for connections already made and awaiting user acceptance.
80  * As a protocol is preparing incoming connections, it creates a socket
81  * structure queued on so_q0 by calling sonewconn().  When the connection
82  * is established, soisconnected() is called, and transfers the
83  * socket structure to so_q, making it available to accept().
84  *
85  * If a socket is closed with sockets on either
86  * so_q0 or so_q, these sockets are dropped.
87  *
88  * If higher level protocols are implemented in
89  * the kernel, the wakeups done here will sometimes
90  * cause software-interrupt process scheduling.
91  */
92 
93 void
94 soisconnecting(so)
95 	register struct socket *so;
96 {
97 
98 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
99 	so->so_state |= SS_ISCONNECTING;
100 }
101 
102 void
103 soisconnected(so)
104 	register struct socket *so;
105 {
106 	register struct socket *head = so->so_head;
107 
108 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
109 	so->so_state |= SS_ISCONNECTED;
110 	if (head && (so->so_state & SS_INCOMP)) {
111 		TAILQ_REMOVE(&head->so_incomp, so, so_list);
112 		head->so_incqlen--;
113 		so->so_state &= ~SS_INCOMP;
114 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
115 		so->so_state |= SS_COMP;
116 		sorwakeup(head);
117 		wakeup((caddr_t)&head->so_timeo);
118 	} else {
119 		wakeup((caddr_t)&so->so_timeo);
120 		sorwakeup(so);
121 		sowwakeup(so);
122 	}
123 }
124 
125 void
126 soisdisconnecting(so)
127 	register struct socket *so;
128 {
129 
130 	so->so_state &= ~SS_ISCONNECTING;
131 	so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
132 	wakeup((caddr_t)&so->so_timeo);
133 	sowwakeup(so);
134 	sorwakeup(so);
135 }
136 
137 void
138 soisdisconnected(so)
139 	register struct socket *so;
140 {
141 
142 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
143 	so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE);
144 	wakeup((caddr_t)&so->so_timeo);
145 	sowwakeup(so);
146 	sorwakeup(so);
147 }
148 
149 /*
150  * Return a random connection that hasn't been serviced yet and
151  * is eligible for discard.  There is a one in qlen chance that
152  * we will return a null, saying that there are no dropable
153  * requests.  In this case, the protocol specific code should drop
154  * the new request.  This insures fairness.
155  *
156  * This may be used in conjunction with protocol specific queue
157  * congestion routines.
158  */
159 struct socket *
160 sodropablereq(head)
161 	register struct socket *head;
162 {
163 	register struct socket *so;
164 	unsigned int i, j, qlen;
165 
166 	static int rnd;
167 	static long old_mono_secs;
168 	static unsigned int cur_cnt, old_cnt;
169 
170 	if ((i = (mono_time.tv_sec - old_mono_secs)) != 0) {
171 		old_mono_secs = mono_time.tv_sec;
172 		old_cnt = cur_cnt / i;
173 		cur_cnt = 0;
174 	}
175 
176 	so = TAILQ_FIRST(&head->so_incomp);
177 	if (!so)
178 		return (so);
179 
180 	qlen = head->so_incqlen;
181 	if (++cur_cnt > qlen || old_cnt > qlen) {
182 		rnd = (314159 * rnd + 66329) & 0xffff;
183 		j = ((qlen + 1) * rnd) >> 16;
184 
185 		while (j-- && so)
186 		    so = TAILQ_NEXT(so, so_list);
187 	}
188 
189 	return (so);
190 }
191 
192 /*
193  * When an attempt at a new connection is noted on a socket
194  * which accepts connections, sonewconn is called.  If the
195  * connection is possible (subject to space constraints, etc.)
196  * then we allocate a new structure, propoerly linked into the
197  * data structure of the original socket, and return this.
198  * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
199  *
200  * Currently, sonewconn() is defined as sonewconn1() in socketvar.h
201  * to catch calls that are missing the (new) second parameter.
202  */
203 struct socket *
204 sonewconn1(head, connstatus)
205 	register struct socket *head;
206 	int connstatus;
207 {
208 	register struct socket *so;
209 
210 	if (head->so_qlen > 3 * head->so_qlimit / 2)
211 		return ((struct socket *)0);
212 	MALLOC(so, struct socket *, sizeof(*so), M_SOCKET, M_DONTWAIT);
213 	if (so == NULL)
214 		return ((struct socket *)0);
215 	bzero((caddr_t)so, sizeof(*so));
216 	so->so_head = head;
217 	so->so_type = head->so_type;
218 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
219 	so->so_linger = head->so_linger;
220 	so->so_state = head->so_state | SS_NOFDREF;
221 	so->so_proto = head->so_proto;
222 	so->so_timeo = head->so_timeo;
223 	so->so_pgid = head->so_pgid;
224 	(void) soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat);
225 
226 	if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0)) {
227 		(void) free((caddr_t)so, M_SOCKET);
228 		return ((struct socket *)0);
229 	}
230 
231 	if (connstatus) {
232 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
233 		so->so_state |= SS_COMP;
234 	} else {
235 		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
236 		so->so_state |= SS_INCOMP;
237 		head->so_incqlen++;
238 	}
239 	head->so_qlen++;
240 	if (connstatus) {
241 		sorwakeup(head);
242 		wakeup((caddr_t)&head->so_timeo);
243 		so->so_state |= connstatus;
244 	}
245 	return (so);
246 }
247 
248 /*
249  * Socantsendmore indicates that no more data will be sent on the
250  * socket; it would normally be applied to a socket when the user
251  * informs the system that no more data is to be sent, by the protocol
252  * code (in case PRU_SHUTDOWN).  Socantrcvmore indicates that no more data
253  * will be received, and will normally be applied to the socket by a
254  * protocol when it detects that the peer will send no more data.
255  * Data queued for reading in the socket may yet be read.
256  */
257 
258 void
259 socantsendmore(so)
260 	struct socket *so;
261 {
262 
263 	so->so_state |= SS_CANTSENDMORE;
264 	sowwakeup(so);
265 }
266 
267 void
268 socantrcvmore(so)
269 	struct socket *so;
270 {
271 
272 	so->so_state |= SS_CANTRCVMORE;
273 	sorwakeup(so);
274 }
275 
276 /*
277  * Wait for data to arrive at/drain from a socket buffer.
278  */
279 int
280 sbwait(sb)
281 	struct sockbuf *sb;
282 {
283 
284 	sb->sb_flags |= SB_WAIT;
285 	return (tsleep((caddr_t)&sb->sb_cc,
286 	    (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
287 	    sb->sb_timeo));
288 }
289 
290 /*
291  * Lock a sockbuf already known to be locked;
292  * return any error returned from sleep (EINTR).
293  */
294 int
295 sb_lock(sb)
296 	register struct sockbuf *sb;
297 {
298 	int error;
299 
300 	while (sb->sb_flags & SB_LOCK) {
301 		sb->sb_flags |= SB_WANT;
302 		error = tsleep((caddr_t)&sb->sb_flags,
303 		    (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH,
304 		    "sblock", 0);
305 		if (error)
306 			return (error);
307 	}
308 	sb->sb_flags |= SB_LOCK;
309 	return (0);
310 }
311 
312 /*
313  * Wakeup processes waiting on a socket buffer.
314  * Do asynchronous notification via SIGIO
315  * if the socket has the SS_ASYNC flag set.
316  */
317 void
318 sowakeup(so, sb)
319 	register struct socket *so;
320 	register struct sockbuf *sb;
321 {
322 	struct proc *p;
323 
324 	selwakeup(&sb->sb_sel);
325 	sb->sb_flags &= ~SB_SEL;
326 	if (sb->sb_flags & SB_WAIT) {
327 		sb->sb_flags &= ~SB_WAIT;
328 		wakeup((caddr_t)&sb->sb_cc);
329 	}
330 	if (so->so_state & SS_ASYNC) {
331 		if (so->so_pgid < 0)
332 			gsignal(-so->so_pgid, SIGIO);
333 		else if (so->so_pgid > 0 && (p = pfind(so->so_pgid)) != 0)
334 			psignal(p, SIGIO);
335 	}
336 }
337 
338 /*
339  * Socket buffer (struct sockbuf) utility routines.
340  *
341  * Each socket contains two socket buffers: one for sending data and
342  * one for receiving data.  Each buffer contains a queue of mbufs,
343  * information about the number of mbufs and amount of data in the
344  * queue, and other fields allowing select() statements and notification
345  * on data availability to be implemented.
346  *
347  * Data stored in a socket buffer is maintained as a list of records.
348  * Each record is a list of mbufs chained together with the m_next
349  * field.  Records are chained together with the m_nextpkt field. The upper
350  * level routine soreceive() expects the following conventions to be
351  * observed when placing information in the receive buffer:
352  *
353  * 1. If the protocol requires each message be preceded by the sender's
354  *    name, then a record containing that name must be present before
355  *    any associated data (mbuf's must be of type MT_SONAME).
356  * 2. If the protocol supports the exchange of ``access rights'' (really
357  *    just additional data associated with the message), and there are
358  *    ``rights'' to be received, then a record containing this data
359  *    should be present (mbuf's must be of type MT_RIGHTS).
360  * 3. If a name or rights record exists, then it must be followed by
361  *    a data record, perhaps of zero length.
362  *
363  * Before using a new socket structure it is first necessary to reserve
364  * buffer space to the socket, by calling sbreserve().  This should commit
365  * some of the available buffer space in the system buffer pool for the
366  * socket (currently, it does nothing but enforce limits).  The space
367  * should be released by calling sbrelease() when the socket is destroyed.
368  */
369 
370 int
371 soreserve(so, sndcc, rcvcc)
372 	register struct socket *so;
373 	u_long sndcc, rcvcc;
374 {
375 
376 	if (sbreserve(&so->so_snd, sndcc) == 0)
377 		goto bad;
378 	if (sbreserve(&so->so_rcv, rcvcc) == 0)
379 		goto bad2;
380 	if (so->so_rcv.sb_lowat == 0)
381 		so->so_rcv.sb_lowat = 1;
382 	if (so->so_snd.sb_lowat == 0)
383 		so->so_snd.sb_lowat = MCLBYTES;
384 	if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
385 		so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
386 	return (0);
387 bad2:
388 	sbrelease(&so->so_snd);
389 bad:
390 	return (ENOBUFS);
391 }
392 
393 /*
394  * Allot mbufs to a sockbuf.
395  * Attempt to scale mbmax so that mbcnt doesn't become limiting
396  * if buffering efficiency is near the normal case.
397  */
398 int
399 sbreserve(sb, cc)
400 	struct sockbuf *sb;
401 	u_long cc;
402 {
403 	if ((u_quad_t)cc > (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES))
404 		return (0);
405 	sb->sb_hiwat = cc;
406 	sb->sb_mbmax = min(cc * sb_efficiency, sb_max);
407 	if (sb->sb_lowat > sb->sb_hiwat)
408 		sb->sb_lowat = sb->sb_hiwat;
409 	return (1);
410 }
411 
412 /*
413  * Free mbufs held by a socket, and reserved mbuf space.
414  */
415 void
416 sbrelease(sb)
417 	struct sockbuf *sb;
418 {
419 
420 	sbflush(sb);
421 	sb->sb_hiwat = sb->sb_mbmax = 0;
422 }
423 
424 /*
425  * Routines to add and remove
426  * data from an mbuf queue.
427  *
428  * The routines sbappend() or sbappendrecord() are normally called to
429  * append new mbufs to a socket buffer, after checking that adequate
430  * space is available, comparing the function sbspace() with the amount
431  * of data to be added.  sbappendrecord() differs from sbappend() in
432  * that data supplied is treated as the beginning of a new record.
433  * To place a sender's address, optional access rights, and data in a
434  * socket receive buffer, sbappendaddr() should be used.  To place
435  * access rights and data in a socket receive buffer, sbappendrights()
436  * should be used.  In either case, the new data begins a new record.
437  * Note that unlike sbappend() and sbappendrecord(), these routines check
438  * for the caller that there will be enough space to store the data.
439  * Each fails if there is not enough space, or if it cannot find mbufs
440  * to store additional information in.
441  *
442  * Reliable protocols may use the socket send buffer to hold data
443  * awaiting acknowledgement.  Data is normally copied from a socket
444  * send buffer in a protocol with m_copy for output to a peer,
445  * and then removing the data from the socket buffer with sbdrop()
446  * or sbdroprecord() when the data is acknowledged by the peer.
447  */
448 
449 /*
450  * Append mbuf chain m to the last record in the
451  * socket buffer sb.  The additional space associated
452  * the mbuf chain is recorded in sb.  Empty mbufs are
453  * discarded and mbufs are compacted where possible.
454  */
455 void
456 sbappend(sb, m)
457 	struct sockbuf *sb;
458 	struct mbuf *m;
459 {
460 	register struct mbuf *n;
461 
462 	if (m == 0)
463 		return;
464 	n = sb->sb_mb;
465 	if (n) {
466 		while (n->m_nextpkt)
467 			n = n->m_nextpkt;
468 		do {
469 			if (n->m_flags & M_EOR) {
470 				sbappendrecord(sb, m); /* XXXXXX!!!! */
471 				return;
472 			}
473 		} while (n->m_next && (n = n->m_next));
474 	}
475 	sbcompress(sb, m, n);
476 }
477 
478 #ifdef SOCKBUF_DEBUG
479 void
480 sbcheck(sb)
481 	register struct sockbuf *sb;
482 {
483 	register struct mbuf *m;
484 	register int len = 0, mbcnt = 0;
485 
486 	for (m = sb->sb_mb; m; m = m->m_next) {
487 		len += m->m_len;
488 		mbcnt += MSIZE;
489 		if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
490 			mbcnt += m->m_ext.ext_size;
491 		if (m->m_nextpkt)
492 			panic("sbcheck nextpkt");
493 	}
494 	if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
495 		printf("cc %d != %d || mbcnt %d != %d\n", len, sb->sb_cc,
496 		    mbcnt, sb->sb_mbcnt);
497 		panic("sbcheck");
498 	}
499 }
500 #endif
501 
502 /*
503  * As above, except the mbuf chain
504  * begins a new record.
505  */
506 void
507 sbappendrecord(sb, m0)
508 	register struct sockbuf *sb;
509 	register struct mbuf *m0;
510 {
511 	register struct mbuf *m;
512 
513 	if (m0 == 0)
514 		return;
515 	m = sb->sb_mb;
516 	if (m)
517 		while (m->m_nextpkt)
518 			m = m->m_nextpkt;
519 	/*
520 	 * Put the first mbuf on the queue.
521 	 * Note this permits zero length records.
522 	 */
523 	sballoc(sb, m0);
524 	if (m)
525 		m->m_nextpkt = m0;
526 	else
527 		sb->sb_mb = m0;
528 	m = m0->m_next;
529 	m0->m_next = 0;
530 	if (m && (m0->m_flags & M_EOR)) {
531 		m0->m_flags &= ~M_EOR;
532 		m->m_flags |= M_EOR;
533 	}
534 	sbcompress(sb, m, m0);
535 }
536 
537 /*
538  * As above except that OOB data
539  * is inserted at the beginning of the sockbuf,
540  * but after any other OOB data.
541  */
542 void
543 sbinsertoob(sb, m0)
544 	register struct sockbuf *sb;
545 	register struct mbuf *m0;
546 {
547 	register struct mbuf *m;
548 	register struct mbuf **mp;
549 
550 	if (m0 == 0)
551 		return;
552 	for (mp = &sb->sb_mb; *mp ; mp = &((*mp)->m_nextpkt)) {
553 	    m = *mp;
554 	    again:
555 		switch (m->m_type) {
556 
557 		case MT_OOBDATA:
558 			continue;		/* WANT next train */
559 
560 		case MT_CONTROL:
561 			m = m->m_next;
562 			if (m)
563 				goto again;	/* inspect THIS train further */
564 		}
565 		break;
566 	}
567 	/*
568 	 * Put the first mbuf on the queue.
569 	 * Note this permits zero length records.
570 	 */
571 	sballoc(sb, m0);
572 	m0->m_nextpkt = *mp;
573 	*mp = m0;
574 	m = m0->m_next;
575 	m0->m_next = 0;
576 	if (m && (m0->m_flags & M_EOR)) {
577 		m0->m_flags &= ~M_EOR;
578 		m->m_flags |= M_EOR;
579 	}
580 	sbcompress(sb, m, m0);
581 }
582 
583 /*
584  * Append address and data, and optionally, control (ancillary) data
585  * to the receive queue of a socket.  If present,
586  * m0 must include a packet header with total length.
587  * Returns 0 if no space in sockbuf or insufficient mbufs.
588  */
589 int
590 sbappendaddr(sb, asa, m0, control)
591 	register struct sockbuf *sb;
592 	struct sockaddr *asa;
593 	struct mbuf *m0, *control;
594 {
595 	register struct mbuf *m, *n;
596 	int space = asa->sa_len;
597 
598 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
599 panic("sbappendaddr");
600 	if (m0)
601 		space += m0->m_pkthdr.len;
602 	for (n = control; n; n = n->m_next) {
603 		space += n->m_len;
604 		if (n->m_next == 0)	/* keep pointer to last control buf */
605 			break;
606 	}
607 	if (space > sbspace(sb))
608 		return (0);
609 	if (asa->sa_len > MLEN)
610 		return (0);
611 	MGET(m, M_DONTWAIT, MT_SONAME);
612 	if (m == 0)
613 		return (0);
614 	m->m_len = asa->sa_len;
615 	bcopy((caddr_t)asa, mtod(m, caddr_t), asa->sa_len);
616 	if (n)
617 		n->m_next = m0;		/* concatenate data to control */
618 	else
619 		control = m0;
620 	m->m_next = control;
621 	for (n = m; n; n = n->m_next)
622 		sballoc(sb, n);
623 	n = sb->sb_mb;
624 	if (n) {
625 		while (n->m_nextpkt)
626 			n = n->m_nextpkt;
627 		n->m_nextpkt = m;
628 	} else
629 		sb->sb_mb = m;
630 	return (1);
631 }
632 
633 int
634 sbappendcontrol(sb, m0, control)
635 	struct sockbuf *sb;
636 	struct mbuf *control, *m0;
637 {
638 	register struct mbuf *m, *n;
639 	int space = 0;
640 
641 	if (control == 0)
642 		panic("sbappendcontrol");
643 	for (m = control; ; m = m->m_next) {
644 		space += m->m_len;
645 		if (m->m_next == 0)
646 			break;
647 	}
648 	n = m;			/* save pointer to last control buffer */
649 	for (m = m0; m; m = m->m_next)
650 		space += m->m_len;
651 	if (space > sbspace(sb))
652 		return (0);
653 	n->m_next = m0;			/* concatenate data to control */
654 	for (m = control; m; m = m->m_next)
655 		sballoc(sb, m);
656 	n = sb->sb_mb;
657 	if (n) {
658 		while (n->m_nextpkt)
659 			n = n->m_nextpkt;
660 		n->m_nextpkt = control;
661 	} else
662 		sb->sb_mb = control;
663 	return (1);
664 }
665 
666 /*
667  * Compress mbuf chain m into the socket
668  * buffer sb following mbuf n.  If n
669  * is null, the buffer is presumed empty.
670  */
671 void
672 sbcompress(sb, m, n)
673 	register struct sockbuf *sb;
674 	register struct mbuf *m, *n;
675 {
676 	register int eor = 0;
677 	register struct mbuf *o;
678 
679 	while (m) {
680 		eor |= m->m_flags & M_EOR;
681 		if (m->m_len == 0 &&
682 		    (eor == 0 ||
683 		     (((o = m->m_next) || (o = n)) &&
684 		      o->m_type == m->m_type))) {
685 			m = m_free(m);
686 			continue;
687 		}
688 		if (n && (n->m_flags & (M_EXT | M_EOR)) == 0 &&
689 		    (n->m_data + n->m_len + m->m_len) < &n->m_dat[MLEN] &&
690 		    n->m_type == m->m_type) {
691 			bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
692 			    (unsigned)m->m_len);
693 			n->m_len += m->m_len;
694 			sb->sb_cc += m->m_len;
695 			m = m_free(m);
696 			continue;
697 		}
698 		if (n)
699 			n->m_next = m;
700 		else
701 			sb->sb_mb = m;
702 		sballoc(sb, m);
703 		n = m;
704 		m->m_flags &= ~M_EOR;
705 		m = m->m_next;
706 		n->m_next = 0;
707 	}
708 	if (eor) {
709 		if (n)
710 			n->m_flags |= eor;
711 		else
712 			printf("semi-panic: sbcompress\n");
713 	}
714 }
715 
716 /*
717  * Free all mbufs in a sockbuf.
718  * Check that all resources are reclaimed.
719  */
720 void
721 sbflush(sb)
722 	register struct sockbuf *sb;
723 {
724 
725 	if (sb->sb_flags & SB_LOCK)
726 		panic("sbflush");
727 	while (sb->sb_mbcnt)
728 		sbdrop(sb, (int)sb->sb_cc);
729 	if (sb->sb_cc || sb->sb_mb)
730 		panic("sbflush 2");
731 }
732 
733 /*
734  * Drop data from (the front of) a sockbuf.
735  */
736 void
737 sbdrop(sb, len)
738 	register struct sockbuf *sb;
739 	register int len;
740 {
741 	register struct mbuf *m, *mn;
742 	struct mbuf *next;
743 
744 	next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
745 	while (len > 0) {
746 		if (m == 0) {
747 			if (next == 0)
748 				panic("sbdrop");
749 			m = next;
750 			next = m->m_nextpkt;
751 			continue;
752 		}
753 		if (m->m_len > len) {
754 			m->m_len -= len;
755 			m->m_data += len;
756 			sb->sb_cc -= len;
757 			break;
758 		}
759 		len -= m->m_len;
760 		sbfree(sb, m);
761 		MFREE(m, mn);
762 		m = mn;
763 	}
764 	while (m && m->m_len == 0) {
765 		sbfree(sb, m);
766 		MFREE(m, mn);
767 		m = mn;
768 	}
769 	if (m) {
770 		sb->sb_mb = m;
771 		m->m_nextpkt = next;
772 	} else
773 		sb->sb_mb = next;
774 }
775 
776 /*
777  * Drop a record off the front of a sockbuf
778  * and move the next record to the front.
779  */
780 void
781 sbdroprecord(sb)
782 	register struct sockbuf *sb;
783 {
784 	register struct mbuf *m, *mn;
785 
786 	m = sb->sb_mb;
787 	if (m) {
788 		sb->sb_mb = m->m_nextpkt;
789 		do {
790 			sbfree(sb, m);
791 			MFREE(m, mn);
792 			m = mn;
793 		} while (m);
794 	}
795 }
796 
797 /*
798  * Create a "control" mbuf containing the specified data
799  * with the specified type for presentation on a socket buffer.
800  */
801 struct mbuf *
802 sbcreatecontrol(p, size, type, level)
803 	caddr_t p;
804 	register int size;
805 	int type, level;
806 {
807 	register struct cmsghdr *cp;
808 	struct mbuf *m;
809 
810 	if ((m = m_get(M_DONTWAIT, MT_CONTROL)) == NULL)
811 		return ((struct mbuf *) NULL);
812 	cp = mtod(m, struct cmsghdr *);
813 	/* XXX check size? */
814 	(void)memcpy(CMSG_DATA(cp), p, size);
815 	size += sizeof(*cp);
816 	m->m_len = size;
817 	cp->cmsg_len = size;
818 	cp->cmsg_level = level;
819 	cp->cmsg_type = type;
820 	return (m);
821 }
822 
823 #ifdef PRU_OLDSTYLE
824 /*
825  * The following routines mediate between the old-style `pr_usrreq'
826  * protocol implementations and the new-style `struct pr_usrreqs'
827  * calling convention.
828  */
829 
830 /* syntactic sugar */
831 #define	nomb	(struct mbuf *)0
832 
833 static int
834 old_abort(struct socket *so)
835 {
836 	return so->so_proto->pr_ousrreq(so, PRU_ABORT, nomb, nomb, nomb);
837 }
838 
839 static int
840 old_accept(struct socket *so, struct mbuf *nam)
841 {
842 	return so->so_proto->pr_ousrreq(so, PRU_ACCEPT, nomb,  nam, nomb);
843 }
844 
845 static int
846 old_attach(struct socket *so, int proto)
847 {
848 	return so->so_proto->pr_ousrreq(so, PRU_ATTACH, nomb,
849 				       (struct mbuf *)proto, /* XXX */
850 				       nomb);
851 }
852 
853 static int
854 old_bind(struct socket *so, struct mbuf *nam)
855 {
856 	return so->so_proto->pr_ousrreq(so, PRU_BIND, nomb, nam, nomb);
857 }
858 
859 static int
860 old_connect(struct socket *so, struct mbuf *nam)
861 {
862 	return so->so_proto->pr_ousrreq(so, PRU_CONNECT, nomb, nam, nomb);
863 }
864 
865 static int
866 old_connect2(struct socket *so1, struct socket *so2)
867 {
868 	return so1->so_proto->pr_ousrreq(so1, PRU_CONNECT2, nomb,
869 				       (struct mbuf *)so2, nomb);
870 }
871 
872 static int
873 old_control(struct socket *so, int cmd, caddr_t data, struct ifnet *ifp)
874 {
875 	return so->so_proto->pr_ousrreq(so, PRU_CONTROL, (struct mbuf *)cmd,
876 				       (struct mbuf *)data,
877 				       (struct mbuf *)ifp);
878 }
879 
880 static int
881 old_detach(struct socket *so)
882 {
883 	return so->so_proto->pr_ousrreq(so, PRU_DETACH, nomb, nomb, nomb);
884 }
885 
886 static int
887 old_disconnect(struct socket *so)
888 {
889 	return so->so_proto->pr_ousrreq(so, PRU_DISCONNECT, nomb, nomb, nomb);
890 }
891 
892 static int
893 old_listen(struct socket *so)
894 {
895 	return so->so_proto->pr_ousrreq(so, PRU_LISTEN, nomb, nomb, nomb);
896 }
897 
898 static int
899 old_peeraddr(struct socket *so, struct mbuf *nam)
900 {
901 	return so->so_proto->pr_ousrreq(so, PRU_PEERADDR, nomb, nam, nomb);
902 }
903 
904 static int
905 old_rcvd(struct socket *so, int flags)
906 {
907 	return so->so_proto->pr_ousrreq(so, PRU_RCVD, nomb,
908 				       (struct mbuf *)flags, /* XXX */
909 				       nomb);
910 }
911 
912 static int
913 old_rcvoob(struct socket *so, struct mbuf *m, int flags)
914 {
915 	return so->so_proto->pr_ousrreq(so, PRU_RCVOOB, m,
916 				       (struct mbuf *)flags, /* XXX */
917 				       nomb);
918 }
919 
920 static int
921 old_send(struct socket *so, int flags, struct mbuf *m, struct mbuf *addr,
922 	 struct mbuf *control)
923 {
924 	int req;
925 
926 	if (flags & PRUS_OOB) {
927 		req = PRU_SENDOOB;
928 	} else if(flags & PRUS_EOF) {
929 		req = PRU_SEND_EOF;
930 	} else {
931 		req = PRU_SEND;
932 	}
933 	return so->so_proto->pr_ousrreq(so, req, m, addr, control);
934 }
935 
936 static int
937 old_sense(struct socket *so, struct stat *sb)
938 {
939 	return so->so_proto->pr_ousrreq(so, PRU_SENSE, (struct mbuf *)sb,
940 				       nomb, nomb);
941 }
942 
943 static int
944 old_shutdown(struct socket *so)
945 {
946 	return so->so_proto->pr_ousrreq(so, PRU_SHUTDOWN, nomb, nomb, nomb);
947 }
948 
949 static int
950 old_sockaddr(struct socket *so, struct mbuf *nam)
951 {
952 	return so->so_proto->pr_ousrreq(so, PRU_SOCKADDR, nomb, nam, nomb);
953 }
954 
955 struct pr_usrreqs pru_oldstyle = {
956 	old_abort, old_accept, old_attach, old_bind, old_connect,
957 	old_connect2, old_control, old_detach, old_disconnect,
958 	old_listen, old_peeraddr, old_rcvd, old_rcvoob, old_send,
959 	old_sense, old_shutdown, old_sockaddr
960 };
961 
962 #endif /* PRU_OLDSTYLE */
963 
964 /*
965  * Some routines that return EOPNOTSUPP for entry points that are not
966  * supported by a protocol.  Fill in as needed.
967  */
968 int
969 pru_accept_notsupp(struct socket *so, struct mbuf *nam)
970 {
971 	return EOPNOTSUPP;
972 }
973 
974 int
975 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
976 {
977 	return EOPNOTSUPP;
978 }
979 
980 int
981 pru_listen_notsupp(struct socket *so)
982 {
983 	return EOPNOTSUPP;
984 }
985 
986 int
987 pru_rcvd_notsupp(struct socket *so, int flags)
988 {
989 	return EOPNOTSUPP;
990 }
991 
992 int
993 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
994 {
995 	return EOPNOTSUPP;
996 }
997 
998 /*
999  * This isn't really a ``null'' operation, but it's the default one
1000  * and doesn't do anything destructive.
1001  */
1002 int
1003 pru_sense_null(struct socket *so, struct stat *sb)
1004 {
1005 	sb->st_blksize = so->so_snd.sb_hiwat;
1006 	return 0;
1007 }
1008