xref: /netbsd-src/sys/kern/uipc_socket2.c (revision 76dfffe33547c37f8bdd446e3e4ab0f3c16cea4b)
1 /*	$NetBSD: uipc_socket2.c,v 1.16 1996/11/10 05:58:37 thorpej Exp $	*/
2 
3 /*
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. All advertising materials mentioning features or use of this software
16  *    must display the following acknowledgement:
17  *	This product includes software developed by the University of
18  *	California, Berkeley and its contributors.
19  * 4. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	@(#)uipc_socket2.c	8.1 (Berkeley) 6/10/93
36  */
37 
38 #include <sys/param.h>
39 #include <sys/systm.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/socket.h>
47 #include <sys/socketvar.h>
48 #include <sys/signalvar.h>
49 
50 /*
51  * Primitive routines for operating on sockets and socket buffers
52  */
53 
54 /* strings for sleep message: */
55 char	netio[] = "netio";
56 char	netcon[] = "netcon";
57 char	netcls[] = "netcls";
58 
59 u_long	sb_max = SB_MAX;		/* patchable */
60 
61 /*
62  * Procedures to manipulate state flags of socket
63  * and do appropriate wakeups.  Normal sequence from the
64  * active (originating) side is that soisconnecting() is
65  * called during processing of connect() call,
66  * resulting in an eventual call to soisconnected() if/when the
67  * connection is established.  When the connection is torn down
68  * soisdisconnecting() is called during processing of disconnect() call,
69  * and soisdisconnected() is called when the connection to the peer
70  * is totally severed.  The semantics of these routines are such that
71  * connectionless protocols can call soisconnected() and soisdisconnected()
72  * only, bypassing the in-progress calls when setting up a ``connection''
73  * takes no time.
74  *
75  * From the passive side, a socket is created with
76  * two queues of sockets: so_q0 for connections in progress
77  * and so_q for connections already made and awaiting user acceptance.
78  * As a protocol is preparing incoming connections, it creates a socket
79  * structure queued on so_q0 by calling sonewconn().  When the connection
80  * is established, soisconnected() is called, and transfers the
81  * socket structure to so_q, making it available to accept().
82  *
83  * If a socket is closed with sockets on either
84  * so_q0 or so_q, these sockets are dropped.
85  *
86  * If higher level protocols are implemented in
87  * the kernel, the wakeups done here will sometimes
88  * cause software-interrupt process scheduling.
89  */
90 
91 void
92 soisconnecting(so)
93 	register struct socket *so;
94 {
95 
96 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
97 	so->so_state |= SS_ISCONNECTING;
98 }
99 
100 void
101 soisconnected(so)
102 	register struct socket *so;
103 {
104 	register struct socket *head = so->so_head;
105 
106 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
107 	so->so_state |= SS_ISCONNECTED;
108 	if (head && soqremque(so, 0)) {
109 		soqinsque(head, so, 1);
110 		sorwakeup(head);
111 		wakeup((caddr_t)&head->so_timeo);
112 	} else {
113 		wakeup((caddr_t)&so->so_timeo);
114 		sorwakeup(so);
115 		sowwakeup(so);
116 	}
117 }
118 
119 void
120 soisdisconnecting(so)
121 	register struct socket *so;
122 {
123 
124 	so->so_state &= ~SS_ISCONNECTING;
125 	so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
126 	wakeup((caddr_t)&so->so_timeo);
127 	sowwakeup(so);
128 	sorwakeup(so);
129 }
130 
131 void
132 soisdisconnected(so)
133 	register struct socket *so;
134 {
135 
136 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
137 	so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE);
138 	wakeup((caddr_t)&so->so_timeo);
139 	sowwakeup(so);
140 	sorwakeup(so);
141 }
142 
143 /*
144  * When an attempt at a new connection is noted on a socket
145  * which accepts connections, sonewconn is called.  If the
146  * connection is possible (subject to space constraints, etc.)
147  * then we allocate a new structure, propoerly linked into the
148  * data structure of the original socket, and return this.
149  * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
150  *
151  * Currently, sonewconn() is defined as sonewconn1() in socketvar.h
152  * to catch calls that are missing the (new) second parameter.
153  */
154 struct socket *
155 sonewconn1(head, connstatus)
156 	register struct socket *head;
157 	int connstatus;
158 {
159 	register struct socket *so;
160 	int soqueue = connstatus ? 1 : 0;
161 
162 	if (head->so_qlen + head->so_q0len > 3 * head->so_qlimit / 2)
163 		return ((struct socket *)0);
164 	MALLOC(so, struct socket *, sizeof(*so), M_SOCKET, M_DONTWAIT);
165 	if (so == NULL)
166 		return ((struct socket *)0);
167 	bzero((caddr_t)so, sizeof(*so));
168 	so->so_type = head->so_type;
169 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
170 	so->so_linger = head->so_linger;
171 	so->so_state = head->so_state | SS_NOFDREF;
172 	so->so_proto = head->so_proto;
173 	so->so_timeo = head->so_timeo;
174 	so->so_pgid = head->so_pgid;
175 	(void) soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat);
176 	soqinsque(head, so, soqueue);
177 	if ((*so->so_proto->pr_usrreq)(so, PRU_ATTACH,
178 	    (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0,
179 	    (struct proc *)0)) {
180 		(void) soqremque(so, soqueue);
181 		(void) free((caddr_t)so, M_SOCKET);
182 		return ((struct socket *)0);
183 	}
184 	if (connstatus) {
185 		sorwakeup(head);
186 		wakeup((caddr_t)&head->so_timeo);
187 		so->so_state |= connstatus;
188 	}
189 	return (so);
190 }
191 
192 /*
193  * The following two routines (soqinsque & soqremque) have been changed
194  * to keep the queue of pending sockets in a double linked list.
195  * (Previously a singly linked list was used.  This gave O(N)
196  * insertion/deletion times and was a major time consumer for sockets
197  * with large pending socket queues).  The doublely-linked list gives
198  * constant insertion/deletion times with only small cost in complexity.
199  *
200  * Since a socket can be on, at most, one queue at a time both so_q and
201  * so_q0 can safely be used as (forward and backward, respectively) queue
202  * pointers.
203  *
204  * Unlike traditional doublely linked lists, the queue head is not present
205  * in the list.  Instead only a single pointer to the first element is kept.
206  * Only when this first element is modified (either adding to an empty list
207  * or removing the first element) does the pointer change.  If the list is
208  * empty, the pointer will be NULL.
209  *
210  * The back pointer of the first entry points to the last entry (instead of
211  * the queue head since there isn't a queue head).
212  */
213 
214 void
215 soqinsque(head, so, q)
216 	register struct socket *head, *so;
217 	int q;
218 {
219 	register struct socket **qh;
220 
221 #ifdef DIAGNOSTIC
222 	if (so->so_head != NULL)
223 		panic("soqinsque");
224 #endif
225 
226 	so->so_head = head;
227 	if (q == 0) {
228 		head->so_q0len++;
229 		qh = &head->so_q0;
230 	} else {
231 		head->so_qlen++;
232 		qh = &head->so_q;
233 	}
234 	if ((*qh) == NULL) {
235 		so->so_q = so->so_q0 = so;
236 		(*qh) = so;
237 	} else {
238 		/* insert at tail */
239 		so->so_q = (*qh);
240 		so->so_q0 = (*qh)->so_q0;
241 		so->so_q0->so_q = so;
242 		(*qh)->so_q0 = so;
243 	}
244 }
245 
246 int
247 soqremque(so, q)
248 	register struct socket *so;
249 	int q;
250 {
251 	register struct socket *head = so->so_head;
252 	register struct socket **qh;
253 
254 	if (head == NULL) {
255 #ifdef DIAGNOSTIC
256 		if (so->so_q != NULL || so->so_q0 != NULL)
257 			panic("soqremque 1");
258 #endif
259 		return (0);
260 	}
261 	if (q == 0) {
262 		head->so_q0len--;
263 		qh = &head->so_q0;
264 	} else {
265 		head->so_qlen--;
266 		qh = &head->so_q;
267 	}
268 
269 #ifdef DIAGNOSTIC
270 	if ((*qh) == NULL || so->so_q == NULL || so->so_q0 == NULL)
271 		panic("soqremque 2");
272 #endif
273 
274 	if ((*qh) == so) {
275 		/* first */
276 		if (so->so_q == so) {
277 			/* single entry; don't remove it from itself */
278 			(*qh) = NULL;
279 		} else {
280 			so->so_q0->so_q = so->so_q;
281 			so->so_q->so_q0 = so->so_q0;
282 			(*qh) = so->so_q;
283 		}
284 	} else {
285 		/* in the middle (or last) but not first */
286 		so->so_q0->so_q = so->so_q;
287 		so->so_q->so_q0 = so->so_q0;
288 	}
289 	so->so_q = so->so_q0 = NULL;
290 	so->so_head = NULL;
291 	return (1);
292 }
293 
294 /*
295  * Socantsendmore indicates that no more data will be sent on the
296  * socket; it would normally be applied to a socket when the user
297  * informs the system that no more data is to be sent, by the protocol
298  * code (in case PRU_SHUTDOWN).  Socantrcvmore indicates that no more data
299  * will be received, and will normally be applied to the socket by a
300  * protocol when it detects that the peer will send no more data.
301  * Data queued for reading in the socket may yet be read.
302  */
303 
304 void
305 socantsendmore(so)
306 	struct socket *so;
307 {
308 
309 	so->so_state |= SS_CANTSENDMORE;
310 	sowwakeup(so);
311 }
312 
313 void
314 socantrcvmore(so)
315 	struct socket *so;
316 {
317 
318 	so->so_state |= SS_CANTRCVMORE;
319 	sorwakeup(so);
320 }
321 
322 /*
323  * Wait for data to arrive at/drain from a socket buffer.
324  */
325 int
326 sbwait(sb)
327 	struct sockbuf *sb;
328 {
329 
330 	sb->sb_flags |= SB_WAIT;
331 	return (tsleep((caddr_t)&sb->sb_cc,
332 	    (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, netio,
333 	    sb->sb_timeo));
334 }
335 
336 /*
337  * Lock a sockbuf already known to be locked;
338  * return any error returned from sleep (EINTR).
339  */
340 int
341 sb_lock(sb)
342 	register struct sockbuf *sb;
343 {
344 	int error;
345 
346 	while (sb->sb_flags & SB_LOCK) {
347 		sb->sb_flags |= SB_WANT;
348 		error = tsleep((caddr_t)&sb->sb_flags,
349 			       (sb->sb_flags & SB_NOINTR) ?
350 					PSOCK : PSOCK|PCATCH, netio, 0);
351 		if (error)
352 			return (error);
353 	}
354 	sb->sb_flags |= SB_LOCK;
355 	return (0);
356 }
357 
358 /*
359  * Wakeup processes waiting on a socket buffer.
360  * Do asynchronous notification via SIGIO
361  * if the socket has the SS_ASYNC flag set.
362  */
363 void
364 sowakeup(so, sb)
365 	register struct socket *so;
366 	register struct sockbuf *sb;
367 {
368 	struct proc *p;
369 
370 	selwakeup(&sb->sb_sel);
371 	sb->sb_flags &= ~SB_SEL;
372 	if (sb->sb_flags & SB_WAIT) {
373 		sb->sb_flags &= ~SB_WAIT;
374 		wakeup((caddr_t)&sb->sb_cc);
375 	}
376 	if (so->so_state & SS_ASYNC) {
377 		if (so->so_pgid < 0)
378 			gsignal(-so->so_pgid, SIGIO);
379 		else if (so->so_pgid > 0 && (p = pfind(so->so_pgid)) != 0)
380 			psignal(p, SIGIO);
381 	}
382 }
383 
384 /*
385  * Socket buffer (struct sockbuf) utility routines.
386  *
387  * Each socket contains two socket buffers: one for sending data and
388  * one for receiving data.  Each buffer contains a queue of mbufs,
389  * information about the number of mbufs and amount of data in the
390  * queue, and other fields allowing poll() statements and notification
391  * on data availability to be implemented.
392  *
393  * Data stored in a socket buffer is maintained as a list of records.
394  * Each record is a list of mbufs chained together with the m_next
395  * field.  Records are chained together with the m_nextpkt field. The upper
396  * level routine soreceive() expects the following conventions to be
397  * observed when placing information in the receive buffer:
398  *
399  * 1. If the protocol requires each message be preceded by the sender's
400  *    name, then a record containing that name must be present before
401  *    any associated data (mbuf's must be of type MT_SONAME).
402  * 2. If the protocol supports the exchange of ``access rights'' (really
403  *    just additional data associated with the message), and there are
404  *    ``rights'' to be received, then a record containing this data
405  *    should be present (mbuf's must be of type MT_CONTROL).
406  * 3. If a name or rights record exists, then it must be followed by
407  *    a data record, perhaps of zero length.
408  *
409  * Before using a new socket structure it is first necessary to reserve
410  * buffer space to the socket, by calling sbreserve().  This should commit
411  * some of the available buffer space in the system buffer pool for the
412  * socket (currently, it does nothing but enforce limits).  The space
413  * should be released by calling sbrelease() when the socket is destroyed.
414  */
415 
416 int
417 soreserve(so, sndcc, rcvcc)
418 	register struct socket *so;
419 	u_long sndcc, rcvcc;
420 {
421 
422 	if (sbreserve(&so->so_snd, sndcc) == 0)
423 		goto bad;
424 	if (sbreserve(&so->so_rcv, rcvcc) == 0)
425 		goto bad2;
426 	if (so->so_rcv.sb_lowat == 0)
427 		so->so_rcv.sb_lowat = 1;
428 	if (so->so_snd.sb_lowat == 0)
429 		so->so_snd.sb_lowat = MCLBYTES;
430 	if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
431 		so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
432 	return (0);
433 bad2:
434 	sbrelease(&so->so_snd);
435 bad:
436 	return (ENOBUFS);
437 }
438 
439 /*
440  * Allot mbufs to a sockbuf.
441  * Attempt to scale mbmax so that mbcnt doesn't become limiting
442  * if buffering efficiency is near the normal case.
443  */
444 int
445 sbreserve(sb, cc)
446 	struct sockbuf *sb;
447 	u_long cc;
448 {
449 
450 	if (cc > sb_max * MCLBYTES / (MSIZE + MCLBYTES))
451 		return (0);
452 	sb->sb_hiwat = cc;
453 	sb->sb_mbmax = min(cc * 2, sb_max);
454 	if (sb->sb_lowat > sb->sb_hiwat)
455 		sb->sb_lowat = sb->sb_hiwat;
456 	return (1);
457 }
458 
459 /*
460  * Free mbufs held by a socket, and reserved mbuf space.
461  */
462 void
463 sbrelease(sb)
464 	struct sockbuf *sb;
465 {
466 
467 	sbflush(sb);
468 	sb->sb_hiwat = sb->sb_mbmax = 0;
469 }
470 
471 /*
472  * Routines to add and remove
473  * data from an mbuf queue.
474  *
475  * The routines sbappend() or sbappendrecord() are normally called to
476  * append new mbufs to a socket buffer, after checking that adequate
477  * space is available, comparing the function sbspace() with the amount
478  * of data to be added.  sbappendrecord() differs from sbappend() in
479  * that data supplied is treated as the beginning of a new record.
480  * To place a sender's address, optional access rights, and data in a
481  * socket receive buffer, sbappendaddr() should be used.  To place
482  * access rights and data in a socket receive buffer, sbappendrights()
483  * should be used.  In either case, the new data begins a new record.
484  * Note that unlike sbappend() and sbappendrecord(), these routines check
485  * for the caller that there will be enough space to store the data.
486  * Each fails if there is not enough space, or if it cannot find mbufs
487  * to store additional information in.
488  *
489  * Reliable protocols may use the socket send buffer to hold data
490  * awaiting acknowledgement.  Data is normally copied from a socket
491  * send buffer in a protocol with m_copy for output to a peer,
492  * and then removing the data from the socket buffer with sbdrop()
493  * or sbdroprecord() when the data is acknowledged by the peer.
494  */
495 
496 /*
497  * Append mbuf chain m to the last record in the
498  * socket buffer sb.  The additional space associated
499  * the mbuf chain is recorded in sb.  Empty mbufs are
500  * discarded and mbufs are compacted where possible.
501  */
502 void
503 sbappend(sb, m)
504 	struct sockbuf *sb;
505 	struct mbuf *m;
506 {
507 	register struct mbuf *n;
508 
509 	if (m == 0)
510 		return;
511 	if ((n = sb->sb_mb) != NULL) {
512 		while (n->m_nextpkt)
513 			n = n->m_nextpkt;
514 		do {
515 			if (n->m_flags & M_EOR) {
516 				sbappendrecord(sb, m); /* XXXXXX!!!! */
517 				return;
518 			}
519 		} while (n->m_next && (n = n->m_next));
520 	}
521 	sbcompress(sb, m, n);
522 }
523 
524 #ifdef SOCKBUF_DEBUG
525 void
526 sbcheck(sb)
527 	register struct sockbuf *sb;
528 {
529 	register struct mbuf *m;
530 	register int len = 0, mbcnt = 0;
531 
532 	for (m = sb->sb_mb; m; m = m->m_next) {
533 		len += m->m_len;
534 		mbcnt += MSIZE;
535 		if (m->m_flags & M_EXT)
536 			mbcnt += m->m_ext.ext_size;
537 		if (m->m_nextpkt)
538 			panic("sbcheck nextpkt");
539 	}
540 	if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
541 		printf("cc %d != %d || mbcnt %d != %d\n", len, sb->sb_cc,
542 		    mbcnt, sb->sb_mbcnt);
543 		panic("sbcheck");
544 	}
545 }
546 #endif
547 
548 /*
549  * As above, except the mbuf chain
550  * begins a new record.
551  */
552 void
553 sbappendrecord(sb, m0)
554 	register struct sockbuf *sb;
555 	register struct mbuf *m0;
556 {
557 	register struct mbuf *m;
558 
559 	if (m0 == 0)
560 		return;
561 	if ((m = sb->sb_mb) != NULL)
562 		while (m->m_nextpkt)
563 			m = m->m_nextpkt;
564 	/*
565 	 * Put the first mbuf on the queue.
566 	 * Note this permits zero length records.
567 	 */
568 	sballoc(sb, m0);
569 	if (m)
570 		m->m_nextpkt = m0;
571 	else
572 		sb->sb_mb = m0;
573 	m = m0->m_next;
574 	m0->m_next = 0;
575 	if (m && (m0->m_flags & M_EOR)) {
576 		m0->m_flags &= ~M_EOR;
577 		m->m_flags |= M_EOR;
578 	}
579 	sbcompress(sb, m, m0);
580 }
581 
582 /*
583  * As above except that OOB data
584  * is inserted at the beginning of the sockbuf,
585  * but after any other OOB data.
586  */
587 void
588 sbinsertoob(sb, m0)
589 	register struct sockbuf *sb;
590 	register struct mbuf *m0;
591 {
592 	register struct mbuf *m;
593 	register struct mbuf **mp;
594 
595 	if (m0 == 0)
596 		return;
597 	for (mp = &sb->sb_mb; (m = *mp) != NULL; mp = &((*mp)->m_nextpkt)) {
598 	    again:
599 		switch (m->m_type) {
600 
601 		case MT_OOBDATA:
602 			continue;		/* WANT next train */
603 
604 		case MT_CONTROL:
605 			if ((m = m->m_next) != NULL)
606 				goto again;	/* inspect THIS train further */
607 		}
608 		break;
609 	}
610 	/*
611 	 * Put the first mbuf on the queue.
612 	 * Note this permits zero length records.
613 	 */
614 	sballoc(sb, m0);
615 	m0->m_nextpkt = *mp;
616 	*mp = m0;
617 	m = m0->m_next;
618 	m0->m_next = 0;
619 	if (m && (m0->m_flags & M_EOR)) {
620 		m0->m_flags &= ~M_EOR;
621 		m->m_flags |= M_EOR;
622 	}
623 	sbcompress(sb, m, m0);
624 }
625 
626 /*
627  * Append address and data, and optionally, control (ancillary) data
628  * to the receive queue of a socket.  If present,
629  * m0 must include a packet header with total length.
630  * Returns 0 if no space in sockbuf or insufficient mbufs.
631  */
632 int
633 sbappendaddr(sb, asa, m0, control)
634 	register struct sockbuf *sb;
635 	struct sockaddr *asa;
636 	struct mbuf *m0, *control;
637 {
638 	register struct mbuf *m, *n;
639 	int space = asa->sa_len;
640 
641 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
642 panic("sbappendaddr");
643 	if (m0)
644 		space += m0->m_pkthdr.len;
645 	for (n = control; n; n = n->m_next) {
646 		space += n->m_len;
647 		if (n->m_next == 0)	/* keep pointer to last control buf */
648 			break;
649 	}
650 	if (space > sbspace(sb))
651 		return (0);
652 	if (asa->sa_len > MLEN)
653 		return (0);
654 	MGET(m, M_DONTWAIT, MT_SONAME);
655 	if (m == 0)
656 		return (0);
657 	m->m_len = asa->sa_len;
658 	bcopy((caddr_t)asa, mtod(m, caddr_t), asa->sa_len);
659 	if (n)
660 		n->m_next = m0;		/* concatenate data to control */
661 	else
662 		control = m0;
663 	m->m_next = control;
664 	for (n = m; n; n = n->m_next)
665 		sballoc(sb, n);
666 	if ((n = sb->sb_mb) != NULL) {
667 		while (n->m_nextpkt)
668 			n = n->m_nextpkt;
669 		n->m_nextpkt = m;
670 	} else
671 		sb->sb_mb = m;
672 	return (1);
673 }
674 
675 int
676 sbappendcontrol(sb, m0, control)
677 	struct sockbuf *sb;
678 	struct mbuf *m0, *control;
679 {
680 	register struct mbuf *m, *n;
681 	int space = 0;
682 
683 	if (control == 0)
684 		panic("sbappendcontrol");
685 	for (m = control; ; m = m->m_next) {
686 		space += m->m_len;
687 		if (m->m_next == 0)
688 			break;
689 	}
690 	n = m;			/* save pointer to last control buffer */
691 	for (m = m0; m; m = m->m_next)
692 		space += m->m_len;
693 	if (space > sbspace(sb))
694 		return (0);
695 	n->m_next = m0;			/* concatenate data to control */
696 	for (m = control; m; m = m->m_next)
697 		sballoc(sb, m);
698 	if ((n = sb->sb_mb) != NULL) {
699 		while (n->m_nextpkt)
700 			n = n->m_nextpkt;
701 		n->m_nextpkt = control;
702 	} else
703 		sb->sb_mb = control;
704 	return (1);
705 }
706 
707 /*
708  * Compress mbuf chain m into the socket
709  * buffer sb following mbuf n.  If n
710  * is null, the buffer is presumed empty.
711  */
712 void
713 sbcompress(sb, m, n)
714 	register struct sockbuf *sb;
715 	register struct mbuf *m, *n;
716 {
717 	register int eor = 0;
718 	register struct mbuf *o;
719 
720 	while (m) {
721 		eor |= m->m_flags & M_EOR;
722 		if (m->m_len == 0 &&
723 		    (eor == 0 ||
724 		     (((o = m->m_next) || (o = n)) &&
725 		      o->m_type == m->m_type))) {
726 			m = m_free(m);
727 			continue;
728 		}
729 		if (n && (n->m_flags & (M_EXT | M_EOR)) == 0 &&
730 		    (n->m_data + n->m_len + m->m_len) < &n->m_dat[MLEN] &&
731 		    n->m_type == m->m_type) {
732 			bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
733 			    (unsigned)m->m_len);
734 			n->m_len += m->m_len;
735 			sb->sb_cc += m->m_len;
736 			m = m_free(m);
737 			continue;
738 		}
739 		if (n)
740 			n->m_next = m;
741 		else
742 			sb->sb_mb = m;
743 		sballoc(sb, m);
744 		n = m;
745 		m->m_flags &= ~M_EOR;
746 		m = m->m_next;
747 		n->m_next = 0;
748 	}
749 	if (eor) {
750 		if (n)
751 			n->m_flags |= eor;
752 		else
753 			printf("semi-panic: sbcompress\n");
754 	}
755 }
756 
757 /*
758  * Free all mbufs in a sockbuf.
759  * Check that all resources are reclaimed.
760  */
761 void
762 sbflush(sb)
763 	register struct sockbuf *sb;
764 {
765 
766 	if (sb->sb_flags & SB_LOCK)
767 		panic("sbflush");
768 	while (sb->sb_mbcnt)
769 		sbdrop(sb, (int)sb->sb_cc);
770 	if (sb->sb_cc || sb->sb_mb)
771 		panic("sbflush 2");
772 }
773 
774 /*
775  * Drop data from (the front of) a sockbuf.
776  */
777 void
778 sbdrop(sb, len)
779 	register struct sockbuf *sb;
780 	register int len;
781 {
782 	register struct mbuf *m, *mn;
783 	struct mbuf *next;
784 
785 	next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
786 	while (len > 0) {
787 		if (m == 0) {
788 			if (next == 0)
789 				panic("sbdrop");
790 			m = next;
791 			next = m->m_nextpkt;
792 			continue;
793 		}
794 		if (m->m_len > len) {
795 			m->m_len -= len;
796 			m->m_data += len;
797 			sb->sb_cc -= len;
798 			break;
799 		}
800 		len -= m->m_len;
801 		sbfree(sb, m);
802 		MFREE(m, mn);
803 		m = mn;
804 	}
805 	while (m && m->m_len == 0) {
806 		sbfree(sb, m);
807 		MFREE(m, mn);
808 		m = mn;
809 	}
810 	if (m) {
811 		sb->sb_mb = m;
812 		m->m_nextpkt = next;
813 	} else
814 		sb->sb_mb = next;
815 }
816 
817 /*
818  * Drop a record off the front of a sockbuf
819  * and move the next record to the front.
820  */
821 void
822 sbdroprecord(sb)
823 	register struct sockbuf *sb;
824 {
825 	register struct mbuf *m, *mn;
826 
827 	m = sb->sb_mb;
828 	if (m) {
829 		sb->sb_mb = m->m_nextpkt;
830 		do {
831 			sbfree(sb, m);
832 			MFREE(m, mn);
833 		} while ((m = mn) != NULL);
834 	}
835 }
836