xref: /freebsd/sys/kern/uipc_sockbuf.c (revision a8445737e740901f5f2c8d24c12ef7fc8b00134e)
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 1.38 1998/09/04 13:13:18 ache Exp $
35  */
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/domain.h>
40 #include <sys/kernel.h>
41 #include <sys/proc.h>
42 #include <sys/malloc.h>
43 #include <sys/mbuf.h>
44 #include <sys/protosw.h>
45 #include <sys/stat.h>
46 #include <sys/socket.h>
47 #include <sys/socketvar.h>
48 #include <sys/signalvar.h>
49 #include <sys/sysctl.h>
50 
51 /*
52  * Primitive routines for operating on sockets and socket buffers
53  */
54 
55 u_long	sb_max = SB_MAX;		/* XXX should be static */
56 
57 static	u_long sb_efficiency = 8;	/* parameter for sbreserve() */
58 
59 /*
60  * Procedures to manipulate state flags of socket
61  * and do appropriate wakeups.  Normal sequence from the
62  * active (originating) side is that soisconnecting() is
63  * called during processing of connect() call,
64  * resulting in an eventual call to soisconnected() if/when the
65  * connection is established.  When the connection is torn down
66  * soisdisconnecting() is called during processing of disconnect() call,
67  * and soisdisconnected() is called when the connection to the peer
68  * is totally severed.  The semantics of these routines are such that
69  * connectionless protocols can call soisconnected() and soisdisconnected()
70  * only, bypassing the in-progress calls when setting up a ``connection''
71  * takes no time.
72  *
73  * From the passive side, a socket is created with
74  * two queues of sockets: so_q0 for connections in progress
75  * and so_q for connections already made and awaiting user acceptance.
76  * As a protocol is preparing incoming connections, it creates a socket
77  * structure queued on so_q0 by calling sonewconn().  When the connection
78  * is established, soisconnected() is called, and transfers the
79  * socket structure to so_q, making it available to accept().
80  *
81  * If a socket is closed with sockets on either
82  * so_q0 or so_q, these sockets are dropped.
83  *
84  * If higher level protocols are implemented in
85  * the kernel, the wakeups done here will sometimes
86  * cause software-interrupt process scheduling.
87  */
88 
89 void
90 soisconnecting(so)
91 	register struct socket *so;
92 {
93 
94 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
95 	so->so_state |= SS_ISCONNECTING;
96 }
97 
98 void
99 soisconnected(so)
100 	register struct socket *so;
101 {
102 	register struct socket *head = so->so_head;
103 
104 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
105 	so->so_state |= SS_ISCONNECTED;
106 	if (head && (so->so_state & SS_INCOMP)) {
107 		TAILQ_REMOVE(&head->so_incomp, so, so_list);
108 		head->so_incqlen--;
109 		so->so_state &= ~SS_INCOMP;
110 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
111 		so->so_state |= SS_COMP;
112 		sorwakeup(head);
113 		wakeup_one(&head->so_timeo);
114 	} else {
115 		wakeup(&so->so_timeo);
116 		sorwakeup(so);
117 		sowwakeup(so);
118 	}
119 }
120 
121 void
122 soisdisconnecting(so)
123 	register struct socket *so;
124 {
125 
126 	so->so_state &= ~SS_ISCONNECTING;
127 	so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
128 	wakeup((caddr_t)&so->so_timeo);
129 	sowwakeup(so);
130 	sorwakeup(so);
131 }
132 
133 void
134 soisdisconnected(so)
135 	register struct socket *so;
136 {
137 
138 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
139 	so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE);
140 	wakeup((caddr_t)&so->so_timeo);
141 	sowwakeup(so);
142 	sorwakeup(so);
143 }
144 
145 /*
146  * Return a random connection that hasn't been serviced yet and
147  * is eligible for discard.  There is a one in qlen chance that
148  * we will return a null, saying that there are no dropable
149  * requests.  In this case, the protocol specific code should drop
150  * the new request.  This insures fairness.
151  *
152  * This may be used in conjunction with protocol specific queue
153  * congestion routines.
154  */
155 struct socket *
156 sodropablereq(head)
157 	register struct socket *head;
158 {
159 	register struct socket *so;
160 	unsigned int i, j, qlen;
161 	static int rnd;
162 	static struct timeval old_runtime;
163 	static unsigned int cur_cnt, old_cnt;
164 	struct timeval tv;
165 
166 	getmicrouptime(&tv);
167 	if ((i = (tv.tv_sec - old_runtime.tv_sec)) != 0) {
168 		old_runtime = tv;
169 		old_cnt = cur_cnt / i;
170 		cur_cnt = 0;
171 	}
172 
173 	so = TAILQ_FIRST(&head->so_incomp);
174 	if (!so)
175 		return (so);
176 
177 	qlen = head->so_incqlen;
178 	if (++cur_cnt > qlen || old_cnt > qlen) {
179 		rnd = (314159 * rnd + 66329) & 0xffff;
180 		j = ((qlen + 1) * rnd) >> 16;
181 
182 		while (j-- && so)
183 		    so = TAILQ_NEXT(so, so_list);
184 	}
185 
186 	return (so);
187 }
188 
189 /*
190  * When an attempt at a new connection is noted on a socket
191  * which accepts connections, sonewconn is called.  If the
192  * connection is possible (subject to space constraints, etc.)
193  * then we allocate a new structure, propoerly linked into the
194  * data structure of the original socket, and return this.
195  * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
196  */
197 struct socket *
198 sonewconn(head, connstatus)
199 	register struct socket *head;
200 	int connstatus;
201 {
202 	register struct socket *so;
203 
204 	if (head->so_qlen > 3 * head->so_qlimit / 2)
205 		return ((struct socket *)0);
206 	so = soalloc(0);
207 	if (so == NULL)
208 		return ((struct socket *)0);
209 	so->so_head = head;
210 	so->so_type = head->so_type;
211 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
212 	so->so_linger = head->so_linger;
213 	so->so_state = head->so_state | SS_NOFDREF;
214 	so->so_proto = head->so_proto;
215 	so->so_timeo = head->so_timeo;
216 	so->so_pgid = head->so_pgid;
217 	so->so_uid = head->so_uid;
218 	(void) soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat);
219 
220 	if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
221 		sodealloc(so);
222 		return ((struct socket *)0);
223 	}
224 
225 	if (connstatus) {
226 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
227 		so->so_state |= SS_COMP;
228 	} else {
229 		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
230 		so->so_state |= SS_INCOMP;
231 		head->so_incqlen++;
232 	}
233 	head->so_qlen++;
234 	if (connstatus) {
235 		sorwakeup(head);
236 		wakeup((caddr_t)&head->so_timeo);
237 		so->so_state |= connstatus;
238 	}
239 	return (so);
240 }
241 
242 /*
243  * Socantsendmore indicates that no more data will be sent on the
244  * socket; it would normally be applied to a socket when the user
245  * informs the system that no more data is to be sent, by the protocol
246  * code (in case PRU_SHUTDOWN).  Socantrcvmore indicates that no more data
247  * will be received, and will normally be applied to the socket by a
248  * protocol when it detects that the peer will send no more data.
249  * Data queued for reading in the socket may yet be read.
250  */
251 
252 void
253 socantsendmore(so)
254 	struct socket *so;
255 {
256 
257 	so->so_state |= SS_CANTSENDMORE;
258 	sowwakeup(so);
259 }
260 
261 void
262 socantrcvmore(so)
263 	struct socket *so;
264 {
265 
266 	so->so_state |= SS_CANTRCVMORE;
267 	sorwakeup(so);
268 }
269 
270 /*
271  * Wait for data to arrive at/drain from a socket buffer.
272  */
273 int
274 sbwait(sb)
275 	struct sockbuf *sb;
276 {
277 
278 	sb->sb_flags |= SB_WAIT;
279 	return (tsleep((caddr_t)&sb->sb_cc,
280 	    (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
281 	    sb->sb_timeo));
282 }
283 
284 /*
285  * Lock a sockbuf already known to be locked;
286  * return any error returned from sleep (EINTR).
287  */
288 int
289 sb_lock(sb)
290 	register struct sockbuf *sb;
291 {
292 	int error;
293 
294 	while (sb->sb_flags & SB_LOCK) {
295 		sb->sb_flags |= SB_WANT;
296 		error = tsleep((caddr_t)&sb->sb_flags,
297 		    (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH,
298 		    "sblock", 0);
299 		if (error)
300 			return (error);
301 	}
302 	sb->sb_flags |= SB_LOCK;
303 	return (0);
304 }
305 
306 /*
307  * Wakeup processes waiting on a socket buffer.
308  * Do asynchronous notification via SIGIO
309  * if the socket has the SS_ASYNC flag set.
310  */
311 void
312 sowakeup(so, sb)
313 	register struct socket *so;
314 	register struct sockbuf *sb;
315 {
316 	struct proc *p;
317 
318 	selwakeup(&sb->sb_sel);
319 	sb->sb_flags &= ~SB_SEL;
320 	if (sb->sb_flags & SB_WAIT) {
321 		sb->sb_flags &= ~SB_WAIT;
322 		wakeup((caddr_t)&sb->sb_cc);
323 	}
324 	if (so->so_state & SS_ASYNC) {
325 		if (so->so_pgid < 0)
326 			gsignal(-so->so_pgid, SIGIO);
327 		else if (so->so_pgid > 0 && (p = pfind(so->so_pgid)) != 0)
328 			psignal(p, SIGIO);
329 	}
330 	if (sb->sb_flags & SB_UPCALL)
331 		(*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT);
332 }
333 
334 /*
335  * Socket buffer (struct sockbuf) utility routines.
336  *
337  * Each socket contains two socket buffers: one for sending data and
338  * one for receiving data.  Each buffer contains a queue of mbufs,
339  * information about the number of mbufs and amount of data in the
340  * queue, and other fields allowing select() statements and notification
341  * on data availability to be implemented.
342  *
343  * Data stored in a socket buffer is maintained as a list of records.
344  * Each record is a list of mbufs chained together with the m_next
345  * field.  Records are chained together with the m_nextpkt field. The upper
346  * level routine soreceive() expects the following conventions to be
347  * observed when placing information in the receive buffer:
348  *
349  * 1. If the protocol requires each message be preceded by the sender's
350  *    name, then a record containing that name must be present before
351  *    any associated data (mbuf's must be of type MT_SONAME).
352  * 2. If the protocol supports the exchange of ``access rights'' (really
353  *    just additional data associated with the message), and there are
354  *    ``rights'' to be received, then a record containing this data
355  *    should be present (mbuf's must be of type MT_RIGHTS).
356  * 3. If a name or rights record exists, then it must be followed by
357  *    a data record, perhaps of zero length.
358  *
359  * Before using a new socket structure it is first necessary to reserve
360  * buffer space to the socket, by calling sbreserve().  This should commit
361  * some of the available buffer space in the system buffer pool for the
362  * socket (currently, it does nothing but enforce limits).  The space
363  * should be released by calling sbrelease() when the socket is destroyed.
364  */
365 
366 int
367 soreserve(so, sndcc, rcvcc)
368 	register struct socket *so;
369 	u_long sndcc, rcvcc;
370 {
371 
372 	if (sbreserve(&so->so_snd, sndcc) == 0)
373 		goto bad;
374 	if (sbreserve(&so->so_rcv, rcvcc) == 0)
375 		goto bad2;
376 	if (so->so_rcv.sb_lowat == 0)
377 		so->so_rcv.sb_lowat = 1;
378 	if (so->so_snd.sb_lowat == 0)
379 		so->so_snd.sb_lowat = MCLBYTES;
380 	if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
381 		so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
382 	return (0);
383 bad2:
384 	sbrelease(&so->so_snd);
385 bad:
386 	return (ENOBUFS);
387 }
388 
389 /*
390  * Allot mbufs to a sockbuf.
391  * Attempt to scale mbmax so that mbcnt doesn't become limiting
392  * if buffering efficiency is near the normal case.
393  */
394 int
395 sbreserve(sb, cc)
396 	struct sockbuf *sb;
397 	u_long cc;
398 {
399 	if ((u_quad_t)cc > (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES))
400 		return (0);
401 	sb->sb_hiwat = cc;
402 	sb->sb_mbmax = min(cc * sb_efficiency, sb_max);
403 	if (sb->sb_lowat > sb->sb_hiwat)
404 		sb->sb_lowat = sb->sb_hiwat;
405 	return (1);
406 }
407 
408 /*
409  * Free mbufs held by a socket, and reserved mbuf space.
410  */
411 void
412 sbrelease(sb)
413 	struct sockbuf *sb;
414 {
415 
416 	sbflush(sb);
417 	sb->sb_hiwat = sb->sb_mbmax = 0;
418 }
419 
420 /*
421  * Routines to add and remove
422  * data from an mbuf queue.
423  *
424  * The routines sbappend() or sbappendrecord() are normally called to
425  * append new mbufs to a socket buffer, after checking that adequate
426  * space is available, comparing the function sbspace() with the amount
427  * of data to be added.  sbappendrecord() differs from sbappend() in
428  * that data supplied is treated as the beginning of a new record.
429  * To place a sender's address, optional access rights, and data in a
430  * socket receive buffer, sbappendaddr() should be used.  To place
431  * access rights and data in a socket receive buffer, sbappendrights()
432  * should be used.  In either case, the new data begins a new record.
433  * Note that unlike sbappend() and sbappendrecord(), these routines check
434  * for the caller that there will be enough space to store the data.
435  * Each fails if there is not enough space, or if it cannot find mbufs
436  * to store additional information in.
437  *
438  * Reliable protocols may use the socket send buffer to hold data
439  * awaiting acknowledgement.  Data is normally copied from a socket
440  * send buffer in a protocol with m_copy for output to a peer,
441  * and then removing the data from the socket buffer with sbdrop()
442  * or sbdroprecord() when the data is acknowledged by the peer.
443  */
444 
445 /*
446  * Append mbuf chain m to the last record in the
447  * socket buffer sb.  The additional space associated
448  * the mbuf chain is recorded in sb.  Empty mbufs are
449  * discarded and mbufs are compacted where possible.
450  */
451 void
452 sbappend(sb, m)
453 	struct sockbuf *sb;
454 	struct mbuf *m;
455 {
456 	register struct mbuf *n;
457 
458 	if (m == 0)
459 		return;
460 	n = sb->sb_mb;
461 	if (n) {
462 		while (n->m_nextpkt)
463 			n = n->m_nextpkt;
464 		do {
465 			if (n->m_flags & M_EOR) {
466 				sbappendrecord(sb, m); /* XXXXXX!!!! */
467 				return;
468 			}
469 		} while (n->m_next && (n = n->m_next));
470 	}
471 	sbcompress(sb, m, n);
472 }
473 
474 #ifdef SOCKBUF_DEBUG
475 void
476 sbcheck(sb)
477 	register struct sockbuf *sb;
478 {
479 	register struct mbuf *m;
480 	register int len = 0, mbcnt = 0;
481 
482 	for (m = sb->sb_mb; m; m = m->m_next) {
483 		len += m->m_len;
484 		mbcnt += MSIZE;
485 		if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
486 			mbcnt += m->m_ext.ext_size;
487 		if (m->m_nextpkt)
488 			panic("sbcheck nextpkt");
489 	}
490 	if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
491 		printf("cc %d != %d || mbcnt %d != %d\n", len, sb->sb_cc,
492 		    mbcnt, sb->sb_mbcnt);
493 		panic("sbcheck");
494 	}
495 }
496 #endif
497 
498 /*
499  * As above, except the mbuf chain
500  * begins a new record.
501  */
502 void
503 sbappendrecord(sb, m0)
504 	register struct sockbuf *sb;
505 	register struct mbuf *m0;
506 {
507 	register struct mbuf *m;
508 
509 	if (m0 == 0)
510 		return;
511 	m = sb->sb_mb;
512 	if (m)
513 		while (m->m_nextpkt)
514 			m = m->m_nextpkt;
515 	/*
516 	 * Put the first mbuf on the queue.
517 	 * Note this permits zero length records.
518 	 */
519 	sballoc(sb, m0);
520 	if (m)
521 		m->m_nextpkt = m0;
522 	else
523 		sb->sb_mb = m0;
524 	m = m0->m_next;
525 	m0->m_next = 0;
526 	if (m && (m0->m_flags & M_EOR)) {
527 		m0->m_flags &= ~M_EOR;
528 		m->m_flags |= M_EOR;
529 	}
530 	sbcompress(sb, m, m0);
531 }
532 
533 /*
534  * As above except that OOB data
535  * is inserted at the beginning of the sockbuf,
536  * but after any other OOB data.
537  */
538 void
539 sbinsertoob(sb, m0)
540 	register struct sockbuf *sb;
541 	register struct mbuf *m0;
542 {
543 	register struct mbuf *m;
544 	register struct mbuf **mp;
545 
546 	if (m0 == 0)
547 		return;
548 	for (mp = &sb->sb_mb; *mp ; mp = &((*mp)->m_nextpkt)) {
549 	    m = *mp;
550 	    again:
551 		switch (m->m_type) {
552 
553 		case MT_OOBDATA:
554 			continue;		/* WANT next train */
555 
556 		case MT_CONTROL:
557 			m = m->m_next;
558 			if (m)
559 				goto again;	/* inspect THIS train further */
560 		}
561 		break;
562 	}
563 	/*
564 	 * Put the first mbuf on the queue.
565 	 * Note this permits zero length records.
566 	 */
567 	sballoc(sb, m0);
568 	m0->m_nextpkt = *mp;
569 	*mp = m0;
570 	m = m0->m_next;
571 	m0->m_next = 0;
572 	if (m && (m0->m_flags & M_EOR)) {
573 		m0->m_flags &= ~M_EOR;
574 		m->m_flags |= M_EOR;
575 	}
576 	sbcompress(sb, m, m0);
577 }
578 
579 /*
580  * Append address and data, and optionally, control (ancillary) data
581  * to the receive queue of a socket.  If present,
582  * m0 must include a packet header with total length.
583  * Returns 0 if no space in sockbuf or insufficient mbufs.
584  */
585 int
586 sbappendaddr(sb, asa, m0, control)
587 	register struct sockbuf *sb;
588 	struct sockaddr *asa;
589 	struct mbuf *m0, *control;
590 {
591 	register struct mbuf *m, *n;
592 	int space = asa->sa_len;
593 
594 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
595 panic("sbappendaddr");
596 	if (m0)
597 		space += m0->m_pkthdr.len;
598 	for (n = control; n; n = n->m_next) {
599 		space += n->m_len;
600 		if (n->m_next == 0)	/* keep pointer to last control buf */
601 			break;
602 	}
603 	if (space > sbspace(sb))
604 		return (0);
605 	if (asa->sa_len > MLEN)
606 		return (0);
607 	MGET(m, M_DONTWAIT, MT_SONAME);
608 	if (m == 0)
609 		return (0);
610 	m->m_len = asa->sa_len;
611 	bcopy((caddr_t)asa, mtod(m, caddr_t), asa->sa_len);
612 	if (n)
613 		n->m_next = m0;		/* concatenate data to control */
614 	else
615 		control = m0;
616 	m->m_next = control;
617 	for (n = m; n; n = n->m_next)
618 		sballoc(sb, n);
619 	n = sb->sb_mb;
620 	if (n) {
621 		while (n->m_nextpkt)
622 			n = n->m_nextpkt;
623 		n->m_nextpkt = m;
624 	} else
625 		sb->sb_mb = m;
626 	return (1);
627 }
628 
629 int
630 sbappendcontrol(sb, m0, control)
631 	struct sockbuf *sb;
632 	struct mbuf *control, *m0;
633 {
634 	register struct mbuf *m, *n;
635 	int space = 0;
636 
637 	if (control == 0)
638 		panic("sbappendcontrol");
639 	for (m = control; ; m = m->m_next) {
640 		space += m->m_len;
641 		if (m->m_next == 0)
642 			break;
643 	}
644 	n = m;			/* save pointer to last control buffer */
645 	for (m = m0; m; m = m->m_next)
646 		space += m->m_len;
647 	if (space > sbspace(sb))
648 		return (0);
649 	n->m_next = m0;			/* concatenate data to control */
650 	for (m = control; m; m = m->m_next)
651 		sballoc(sb, m);
652 	n = sb->sb_mb;
653 	if (n) {
654 		while (n->m_nextpkt)
655 			n = n->m_nextpkt;
656 		n->m_nextpkt = control;
657 	} else
658 		sb->sb_mb = control;
659 	return (1);
660 }
661 
662 /*
663  * Compress mbuf chain m into the socket
664  * buffer sb following mbuf n.  If n
665  * is null, the buffer is presumed empty.
666  */
667 void
668 sbcompress(sb, m, n)
669 	register struct sockbuf *sb;
670 	register struct mbuf *m, *n;
671 {
672 	register int eor = 0;
673 	register struct mbuf *o;
674 
675 	while (m) {
676 		eor |= m->m_flags & M_EOR;
677 		if (m->m_len == 0 &&
678 		    (eor == 0 ||
679 		     (((o = m->m_next) || (o = n)) &&
680 		      o->m_type == m->m_type))) {
681 			m = m_free(m);
682 			continue;
683 		}
684 		if (n && (n->m_flags & (M_EXT | M_EOR)) == 0 &&
685 		    (n->m_data + n->m_len + m->m_len) < &n->m_dat[MLEN] &&
686 		    n->m_type == m->m_type) {
687 			bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
688 			    (unsigned)m->m_len);
689 			n->m_len += m->m_len;
690 			sb->sb_cc += m->m_len;
691 			m = m_free(m);
692 			continue;
693 		}
694 		if (n)
695 			n->m_next = m;
696 		else
697 			sb->sb_mb = m;
698 		sballoc(sb, m);
699 		n = m;
700 		m->m_flags &= ~M_EOR;
701 		m = m->m_next;
702 		n->m_next = 0;
703 	}
704 	if (eor) {
705 		if (n)
706 			n->m_flags |= eor;
707 		else
708 			printf("semi-panic: sbcompress\n");
709 	}
710 }
711 
712 /*
713  * Free all mbufs in a sockbuf.
714  * Check that all resources are reclaimed.
715  */
716 void
717 sbflush(sb)
718 	register struct sockbuf *sb;
719 {
720 
721 	if (sb->sb_flags & SB_LOCK)
722 		panic("sbflush: locked");
723 	while (sb->sb_mbcnt)
724 		sbdrop(sb, (int)sb->sb_cc);
725 	if (sb->sb_cc || sb->sb_mb)
726 		panic("sbflush: cc %ld || mb %p", sb->sb_cc, (void *)sb->sb_mb);
727 }
728 
729 /*
730  * Drop data from (the front of) a sockbuf.
731  */
732 void
733 sbdrop(sb, len)
734 	register struct sockbuf *sb;
735 	register int len;
736 {
737 	register struct mbuf *m, *mn;
738 	struct mbuf *next;
739 
740 	next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
741 	while (len > 0) {
742 		if (m == 0) {
743 			if (next == 0)
744 				panic("sbdrop");
745 			m = next;
746 			next = m->m_nextpkt;
747 			continue;
748 		}
749 		if (m->m_len > len) {
750 			m->m_len -= len;
751 			m->m_data += len;
752 			sb->sb_cc -= len;
753 			break;
754 		}
755 		len -= m->m_len;
756 		sbfree(sb, m);
757 		MFREE(m, mn);
758 		m = mn;
759 	}
760 	while (m && m->m_len == 0) {
761 		sbfree(sb, m);
762 		MFREE(m, mn);
763 		m = mn;
764 	}
765 	if (m) {
766 		sb->sb_mb = m;
767 		m->m_nextpkt = next;
768 	} else
769 		sb->sb_mb = next;
770 }
771 
772 /*
773  * Drop a record off the front of a sockbuf
774  * and move the next record to the front.
775  */
776 void
777 sbdroprecord(sb)
778 	register struct sockbuf *sb;
779 {
780 	register struct mbuf *m, *mn;
781 
782 	m = sb->sb_mb;
783 	if (m) {
784 		sb->sb_mb = m->m_nextpkt;
785 		do {
786 			sbfree(sb, m);
787 			MFREE(m, mn);
788 			m = mn;
789 		} while (m);
790 	}
791 }
792 
793 /*
794  * Create a "control" mbuf containing the specified data
795  * with the specified type for presentation on a socket buffer.
796  */
797 struct mbuf *
798 sbcreatecontrol(p, size, type, level)
799 	caddr_t p;
800 	register int size;
801 	int type, level;
802 {
803 	register struct cmsghdr *cp;
804 	struct mbuf *m;
805 
806 	if ((m = m_get(M_DONTWAIT, MT_CONTROL)) == NULL)
807 		return ((struct mbuf *) NULL);
808 	cp = mtod(m, struct cmsghdr *);
809 	/* XXX check size? */
810 	(void)memcpy(CMSG_DATA(cp), p, size);
811 	size += sizeof(*cp);
812 	m->m_len = size;
813 	cp->cmsg_len = size;
814 	cp->cmsg_level = level;
815 	cp->cmsg_type = type;
816 	return (m);
817 }
818 
819 /*
820  * Some routines that return EOPNOTSUPP for entry points that are not
821  * supported by a protocol.  Fill in as needed.
822  */
823 int
824 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
825 {
826 	return EOPNOTSUPP;
827 }
828 
829 int
830 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct proc *p)
831 {
832 	return EOPNOTSUPP;
833 }
834 
835 int
836 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
837 {
838 	return EOPNOTSUPP;
839 }
840 
841 int
842 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
843 		    struct ifnet *ifp, struct proc *p)
844 {
845 	return EOPNOTSUPP;
846 }
847 
848 int
849 pru_listen_notsupp(struct socket *so, struct proc *p)
850 {
851 	return EOPNOTSUPP;
852 }
853 
854 int
855 pru_rcvd_notsupp(struct socket *so, int flags)
856 {
857 	return EOPNOTSUPP;
858 }
859 
860 int
861 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
862 {
863 	return EOPNOTSUPP;
864 }
865 
866 /*
867  * This isn't really a ``null'' operation, but it's the default one
868  * and doesn't do anything destructive.
869  */
870 int
871 pru_sense_null(struct socket *so, struct stat *sb)
872 {
873 	sb->st_blksize = so->so_snd.sb_hiwat;
874 	return 0;
875 }
876 
877 /*
878  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
879  */
880 struct sockaddr *
881 dup_sockaddr(sa, canwait)
882 	struct sockaddr *sa;
883 	int canwait;
884 {
885 	struct sockaddr *sa2;
886 
887 	MALLOC(sa2, struct sockaddr *, sa->sa_len, M_SONAME,
888 	       canwait ? M_WAITOK : M_NOWAIT);
889 	if (sa2)
890 		bcopy(sa, sa2, sa->sa_len);
891 	return sa2;
892 }
893 
894 /*
895  * Create an external-format (``xsocket'') structure using the information
896  * in the kernel-format socket structure pointed to by so.  This is done
897  * to reduce the spew of irrelevant information over this interface,
898  * to isolate user code from changes in the kernel structure, and
899  * potentially to provide information-hiding if we decide that
900  * some of this information should be hidden from users.
901  */
902 void
903 sotoxsocket(struct socket *so, struct xsocket *xso)
904 {
905 	xso->xso_len = sizeof *xso;
906 	xso->xso_so = so;
907 	xso->so_type = so->so_type;
908 	xso->so_options = so->so_options;
909 	xso->so_linger = so->so_linger;
910 	xso->so_state = so->so_state;
911 	xso->so_pcb = so->so_pcb;
912 	xso->xso_protocol = so->so_proto->pr_protocol;
913 	xso->xso_family = so->so_proto->pr_domain->dom_family;
914 	xso->so_qlen = so->so_qlen;
915 	xso->so_incqlen = so->so_incqlen;
916 	xso->so_qlimit = so->so_qlimit;
917 	xso->so_timeo = so->so_timeo;
918 	xso->so_error = so->so_error;
919 	xso->so_pgid = so->so_pgid;
920 	xso->so_oobmark = so->so_oobmark;
921 	sbtoxsockbuf(&so->so_snd, &xso->so_snd);
922 	sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
923 	xso->so_uid = so->so_uid;
924 }
925 
926 /*
927  * This does the same for sockbufs.  Note that the xsockbuf structure,
928  * since it is always embedded in a socket, does not include a self
929  * pointer nor a length.  We make this entry point public in case
930  * some other mechanism needs it.
931  */
932 void
933 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb)
934 {
935 	xsb->sb_cc = sb->sb_cc;
936 	xsb->sb_hiwat = sb->sb_hiwat;
937 	xsb->sb_mbcnt = sb->sb_mbcnt;
938 	xsb->sb_mbmax = sb->sb_mbmax;
939 	xsb->sb_lowat = sb->sb_lowat;
940 	xsb->sb_flags = sb->sb_flags;
941 	xsb->sb_timeo = sb->sb_timeo;
942 }
943 
944 /*
945  * Here is the definition of some of the basic objects in the kern.ipc
946  * branch of the MIB.
947  */
948 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
949 
950 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
951 static int dummy;
952 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
953 
954 SYSCTL_INT(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLFLAG_RW, &sb_max, 0, "");
955 SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RD, &maxsockets, 0, "");
956 SYSCTL_INT(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
957 	   &sb_efficiency, 0, "");
958 SYSCTL_INT(_kern_ipc, KIPC_NMBCLUSTERS, nmbclusters, CTLFLAG_RD, &nmbclusters, 0, "");
959 
960