xref: /freebsd/sys/kern/uipc_sockbuf.c (revision 817420dc8eac7df799c78f5309b75092b7f7cd40)
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  * $FreeBSD$
35  */
36 
37 #include "opt_param.h"
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/domain.h>
41 #include <sys/file.h>	/* for maxfiles */
42 #include <sys/kernel.h>
43 #include <sys/proc.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/protosw.h>
47 #include <sys/resourcevar.h>
48 #include <sys/stat.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/signalvar.h>
52 #include <sys/sysctl.h>
53 #include <sys/aio.h> /* for aio_swake proto */
54 #include <sys/event.h>
55 
56 int	maxsockets;
57 
58 /*
59  * Primitive routines for operating on sockets and socket buffers
60  */
61 
62 u_long	sb_max = SB_MAX;		/* XXX should be static */
63 
64 static	u_long sb_efficiency = 8;	/* parameter for sbreserve() */
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_incomp for connections in progress
82  * and so_comp 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_incomp by calling sonewconn().  When the connection
85  * is established, soisconnected() is called, and transfers the
86  * socket structure to so_comp, making it available to accept().
87  *
88  * If a socket is closed with sockets on either
89  * so_incomp or so_comp, 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 	struct socket *so;
108 {
109 	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 		if ((so->so_options & SO_ACCEPTFILTER) != 0) {
115 			so->so_upcall = head->so_accf->so_accept_filter->accf_callback;
116 			so->so_upcallarg = head->so_accf->so_accept_filter_arg;
117 			so->so_rcv.sb_flags |= SB_UPCALL;
118 			so->so_options &= ~SO_ACCEPTFILTER;
119 			so->so_upcall(so, so->so_upcallarg, 0);
120 			return;
121 		}
122 		TAILQ_REMOVE(&head->so_incomp, so, so_list);
123 		head->so_incqlen--;
124 		so->so_state &= ~SS_INCOMP;
125 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
126 		so->so_state |= SS_COMP;
127 		sorwakeup(head);
128 		wakeup_one(&head->so_timeo);
129 	} else {
130 		wakeup(&so->so_timeo);
131 		sorwakeup(so);
132 		sowwakeup(so);
133 	}
134 }
135 
136 void
137 soisdisconnecting(so)
138 	register struct socket *so;
139 {
140 
141 	so->so_state &= ~SS_ISCONNECTING;
142 	so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
143 	wakeup((caddr_t)&so->so_timeo);
144 	sowwakeup(so);
145 	sorwakeup(so);
146 }
147 
148 void
149 soisdisconnected(so)
150 	register struct socket *so;
151 {
152 
153 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
154 	so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED);
155 	wakeup((caddr_t)&so->so_timeo);
156 	sowwakeup(so);
157 	sorwakeup(so);
158 }
159 
160 /*
161  * Return a random connection that hasn't been serviced yet and
162  * is eligible for discard.  There is a one in qlen chance that
163  * we will return a null, saying that there are no dropable
164  * requests.  In this case, the protocol specific code should drop
165  * the new request.  This insures fairness.
166  *
167  * This may be used in conjunction with protocol specific queue
168  * congestion routines.
169  */
170 struct socket *
171 sodropablereq(head)
172 	register struct socket *head;
173 {
174 	register struct socket *so;
175 	unsigned int i, j, qlen;
176 	static int rnd;
177 	static struct timeval old_runtime;
178 	static unsigned int cur_cnt, old_cnt;
179 	struct timeval tv;
180 
181 	getmicrouptime(&tv);
182 	if ((i = (tv.tv_sec - old_runtime.tv_sec)) != 0) {
183 		old_runtime = tv;
184 		old_cnt = cur_cnt / i;
185 		cur_cnt = 0;
186 	}
187 
188 	so = TAILQ_FIRST(&head->so_incomp);
189 	if (!so)
190 		return (so);
191 
192 	qlen = head->so_incqlen;
193 	if (++cur_cnt > qlen || old_cnt > qlen) {
194 		rnd = (314159 * rnd + 66329) & 0xffff;
195 		j = ((qlen + 1) * rnd) >> 16;
196 
197 		while (j-- && so)
198 		    so = TAILQ_NEXT(so, so_list);
199 	}
200 
201 	return (so);
202 }
203 
204 /*
205  * When an attempt at a new connection is noted on a socket
206  * which accepts connections, sonewconn is called.  If the
207  * connection is possible (subject to space constraints, etc.)
208  * then we allocate a new structure, propoerly linked into the
209  * data structure of the original socket, and return this.
210  * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
211  */
212 struct socket *
213 sonewconn(head, connstatus)
214 	register struct socket *head;
215 	int connstatus;
216 {
217 
218 	return (sonewconn3(head, connstatus, NULL));
219 }
220 
221 struct socket *
222 sonewconn3(head, connstatus, p)
223 	register struct socket *head;
224 	int connstatus;
225 	struct proc *p;
226 {
227 	register struct socket *so;
228 
229 	if (head->so_qlen > 3 * head->so_qlimit / 2)
230 		return ((struct socket *)0);
231 	so = soalloc(0);
232 	if (so == NULL)
233 		return ((struct socket *)0);
234 	so->so_head = head;
235 	so->so_type = head->so_type;
236 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
237 	so->so_linger = head->so_linger;
238 	so->so_state = head->so_state | SS_NOFDREF;
239 	so->so_proto = head->so_proto;
240 	so->so_timeo = head->so_timeo;
241 	so->so_cred = p ? p->p_ucred : head->so_cred;
242 	crhold(so->so_cred);
243 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
244 	    (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
245 		sodealloc(so);
246 		return ((struct socket *)0);
247 	}
248 
249 	if (connstatus) {
250 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
251 		so->so_state |= SS_COMP;
252 	} else {
253 		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
254 		so->so_state |= SS_INCOMP;
255 		head->so_incqlen++;
256 	}
257 	head->so_qlen++;
258 	if (connstatus) {
259 		sorwakeup(head);
260 		wakeup((caddr_t)&head->so_timeo);
261 		so->so_state |= connstatus;
262 	}
263 	return (so);
264 }
265 
266 /*
267  * Socantsendmore indicates that no more data will be sent on the
268  * socket; it would normally be applied to a socket when the user
269  * informs the system that no more data is to be sent, by the protocol
270  * code (in case PRU_SHUTDOWN).  Socantrcvmore indicates that no more data
271  * will be received, and will normally be applied to the socket by a
272  * protocol when it detects that the peer will send no more data.
273  * Data queued for reading in the socket may yet be read.
274  */
275 
276 void
277 socantsendmore(so)
278 	struct socket *so;
279 {
280 
281 	so->so_state |= SS_CANTSENDMORE;
282 	sowwakeup(so);
283 }
284 
285 void
286 socantrcvmore(so)
287 	struct socket *so;
288 {
289 
290 	so->so_state |= SS_CANTRCVMORE;
291 	sorwakeup(so);
292 }
293 
294 /*
295  * Wait for data to arrive at/drain from a socket buffer.
296  */
297 int
298 sbwait(sb)
299 	struct sockbuf *sb;
300 {
301 
302 	sb->sb_flags |= SB_WAIT;
303 	return (tsleep((caddr_t)&sb->sb_cc,
304 	    (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
305 	    sb->sb_timeo));
306 }
307 
308 /*
309  * Lock a sockbuf already known to be locked;
310  * return any error returned from sleep (EINTR).
311  */
312 int
313 sb_lock(sb)
314 	register struct sockbuf *sb;
315 {
316 	int error;
317 
318 	while (sb->sb_flags & SB_LOCK) {
319 		sb->sb_flags |= SB_WANT;
320 		error = tsleep((caddr_t)&sb->sb_flags,
321 		    (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH,
322 		    "sblock", 0);
323 		if (error)
324 			return (error);
325 	}
326 	sb->sb_flags |= SB_LOCK;
327 	return (0);
328 }
329 
330 /*
331  * Wakeup processes waiting on a socket buffer.
332  * Do asynchronous notification via SIGIO
333  * if the socket has the SS_ASYNC flag set.
334  */
335 void
336 sowakeup(so, sb)
337 	register struct socket *so;
338 	register struct sockbuf *sb;
339 {
340 	selwakeup(&sb->sb_sel);
341 	sb->sb_flags &= ~SB_SEL;
342 	if (sb->sb_flags & SB_WAIT) {
343 		sb->sb_flags &= ~SB_WAIT;
344 		wakeup((caddr_t)&sb->sb_cc);
345 	}
346 	if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
347 		pgsigio(so->so_sigio, SIGIO, 0);
348 	if (sb->sb_flags & SB_UPCALL)
349 		(*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT);
350 	if (sb->sb_flags & SB_AIO)
351 		aio_swake(so, sb);
352 	KNOTE(&sb->sb_sel.si_note, 0);
353 }
354 
355 /*
356  * Socket buffer (struct sockbuf) utility routines.
357  *
358  * Each socket contains two socket buffers: one for sending data and
359  * one for receiving data.  Each buffer contains a queue of mbufs,
360  * information about the number of mbufs and amount of data in the
361  * queue, and other fields allowing select() statements and notification
362  * on data availability to be implemented.
363  *
364  * Data stored in a socket buffer is maintained as a list of records.
365  * Each record is a list of mbufs chained together with the m_next
366  * field.  Records are chained together with the m_nextpkt field. The upper
367  * level routine soreceive() expects the following conventions to be
368  * observed when placing information in the receive buffer:
369  *
370  * 1. If the protocol requires each message be preceded by the sender's
371  *    name, then a record containing that name must be present before
372  *    any associated data (mbuf's must be of type MT_SONAME).
373  * 2. If the protocol supports the exchange of ``access rights'' (really
374  *    just additional data associated with the message), and there are
375  *    ``rights'' to be received, then a record containing this data
376  *    should be present (mbuf's must be of type MT_RIGHTS).
377  * 3. If a name or rights record exists, then it must be followed by
378  *    a data record, perhaps of zero length.
379  *
380  * Before using a new socket structure it is first necessary to reserve
381  * buffer space to the socket, by calling sbreserve().  This should commit
382  * some of the available buffer space in the system buffer pool for the
383  * socket (currently, it does nothing but enforce limits).  The space
384  * should be released by calling sbrelease() when the socket is destroyed.
385  */
386 
387 int
388 soreserve(so, sndcc, rcvcc)
389 	register struct socket *so;
390 	u_long sndcc, rcvcc;
391 {
392 	struct proc *p = curproc;
393 
394 	if (sbreserve(&so->so_snd, sndcc, so, p) == 0)
395 		goto bad;
396 	if (sbreserve(&so->so_rcv, rcvcc, so, p) == 0)
397 		goto bad2;
398 	if (so->so_rcv.sb_lowat == 0)
399 		so->so_rcv.sb_lowat = 1;
400 	if (so->so_snd.sb_lowat == 0)
401 		so->so_snd.sb_lowat = MCLBYTES;
402 	if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
403 		so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
404 	return (0);
405 bad2:
406 	sbrelease(&so->so_snd, so);
407 bad:
408 	return (ENOBUFS);
409 }
410 
411 /*
412  * Allot mbufs to a sockbuf.
413  * Attempt to scale mbmax so that mbcnt doesn't become limiting
414  * if buffering efficiency is near the normal case.
415  */
416 int
417 sbreserve(sb, cc, so, p)
418 	struct sockbuf *sb;
419 	u_long cc;
420 	struct socket *so;
421 	struct proc *p;
422 {
423 
424 	/*
425 	 * p will only be NULL when we're in an interrupt
426 	 * (e.g. in tcp_input())
427 	 */
428 	if ((u_quad_t)cc > (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES))
429 		return (0);
430 	if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
431 	    p ? p->p_rlimit[RLIMIT_SBSIZE].rlim_cur : RLIM_INFINITY)) {
432 		return (0);
433 	}
434 	sb->sb_mbmax = min(cc * sb_efficiency, sb_max);
435 	if (sb->sb_lowat > sb->sb_hiwat)
436 		sb->sb_lowat = sb->sb_hiwat;
437 	return (1);
438 }
439 
440 /*
441  * Free mbufs held by a socket, and reserved mbuf space.
442  */
443 void
444 sbrelease(sb, so)
445 	struct sockbuf *sb;
446 	struct socket *so;
447 {
448 
449 	sbflush(sb);
450 	(void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
451 	    RLIM_INFINITY);
452 	sb->sb_mbmax = 0;
453 }
454 
455 /*
456  * Routines to add and remove
457  * data from an mbuf queue.
458  *
459  * The routines sbappend() or sbappendrecord() are normally called to
460  * append new mbufs to a socket buffer, after checking that adequate
461  * space is available, comparing the function sbspace() with the amount
462  * of data to be added.  sbappendrecord() differs from sbappend() in
463  * that data supplied is treated as the beginning of a new record.
464  * To place a sender's address, optional access rights, and data in a
465  * socket receive buffer, sbappendaddr() should be used.  To place
466  * access rights and data in a socket receive buffer, sbappendrights()
467  * should be used.  In either case, the new data begins a new record.
468  * Note that unlike sbappend() and sbappendrecord(), these routines check
469  * for the caller that there will be enough space to store the data.
470  * Each fails if there is not enough space, or if it cannot find mbufs
471  * to store additional information in.
472  *
473  * Reliable protocols may use the socket send buffer to hold data
474  * awaiting acknowledgement.  Data is normally copied from a socket
475  * send buffer in a protocol with m_copy for output to a peer,
476  * and then removing the data from the socket buffer with sbdrop()
477  * or sbdroprecord() when the data is acknowledged by the peer.
478  */
479 
480 /*
481  * Append mbuf chain m to the last record in the
482  * socket buffer sb.  The additional space associated
483  * the mbuf chain is recorded in sb.  Empty mbufs are
484  * discarded and mbufs are compacted where possible.
485  */
486 void
487 sbappend(sb, m)
488 	struct sockbuf *sb;
489 	struct mbuf *m;
490 {
491 	register struct mbuf *n;
492 
493 	if (m == 0)
494 		return;
495 	n = sb->sb_mb;
496 	if (n) {
497 		while (n->m_nextpkt)
498 			n = n->m_nextpkt;
499 		do {
500 			if (n->m_flags & M_EOR) {
501 				sbappendrecord(sb, m); /* XXXXXX!!!! */
502 				return;
503 			}
504 		} while (n->m_next && (n = n->m_next));
505 	}
506 	sbcompress(sb, m, n);
507 }
508 
509 #ifdef SOCKBUF_DEBUG
510 void
511 sbcheck(sb)
512 	register struct sockbuf *sb;
513 {
514 	register struct mbuf *m;
515 	register struct mbuf *n = 0;
516 	register u_long len = 0, mbcnt = 0;
517 
518 	for (m = sb->sb_mb; m; m = n) {
519 	    n = m->m_nextpkt;
520 	    for (; m; m = m->m_next) {
521 		len += m->m_len;
522 		mbcnt += MSIZE;
523 		if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
524 			mbcnt += m->m_ext.ext_size;
525 	    }
526 	}
527 	if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
528 		printf("cc %ld != %ld || mbcnt %ld != %ld\n", len, sb->sb_cc,
529 		    mbcnt, sb->sb_mbcnt);
530 		panic("sbcheck");
531 	}
532 }
533 #endif
534 
535 /*
536  * As above, except the mbuf chain
537  * begins a new record.
538  */
539 void
540 sbappendrecord(sb, m0)
541 	register struct sockbuf *sb;
542 	register struct mbuf *m0;
543 {
544 	register struct mbuf *m;
545 
546 	if (m0 == 0)
547 		return;
548 	m = sb->sb_mb;
549 	if (m)
550 		while (m->m_nextpkt)
551 			m = m->m_nextpkt;
552 	/*
553 	 * Put the first mbuf on the queue.
554 	 * Note this permits zero length records.
555 	 */
556 	sballoc(sb, m0);
557 	if (m)
558 		m->m_nextpkt = m0;
559 	else
560 		sb->sb_mb = m0;
561 	m = m0->m_next;
562 	m0->m_next = 0;
563 	if (m && (m0->m_flags & M_EOR)) {
564 		m0->m_flags &= ~M_EOR;
565 		m->m_flags |= M_EOR;
566 	}
567 	sbcompress(sb, m, m0);
568 }
569 
570 /*
571  * As above except that OOB data
572  * is inserted at the beginning of the sockbuf,
573  * but after any other OOB data.
574  */
575 void
576 sbinsertoob(sb, m0)
577 	register struct sockbuf *sb;
578 	register struct mbuf *m0;
579 {
580 	register struct mbuf *m;
581 	register struct mbuf **mp;
582 
583 	if (m0 == 0)
584 		return;
585 	for (mp = &sb->sb_mb; *mp ; mp = &((*mp)->m_nextpkt)) {
586 	    m = *mp;
587 	    again:
588 		switch (m->m_type) {
589 
590 		case MT_OOBDATA:
591 			continue;		/* WANT next train */
592 
593 		case MT_CONTROL:
594 			m = m->m_next;
595 			if (m)
596 				goto again;	/* inspect THIS train further */
597 		}
598 		break;
599 	}
600 	/*
601 	 * Put the first mbuf on the queue.
602 	 * Note this permits zero length records.
603 	 */
604 	sballoc(sb, m0);
605 	m0->m_nextpkt = *mp;
606 	*mp = m0;
607 	m = m0->m_next;
608 	m0->m_next = 0;
609 	if (m && (m0->m_flags & M_EOR)) {
610 		m0->m_flags &= ~M_EOR;
611 		m->m_flags |= M_EOR;
612 	}
613 	sbcompress(sb, m, m0);
614 }
615 
616 /*
617  * Append address and data, and optionally, control (ancillary) data
618  * to the receive queue of a socket.  If present,
619  * m0 must include a packet header with total length.
620  * Returns 0 if no space in sockbuf or insufficient mbufs.
621  */
622 int
623 sbappendaddr(sb, asa, m0, control)
624 	register struct sockbuf *sb;
625 	struct sockaddr *asa;
626 	struct mbuf *m0, *control;
627 {
628 	register struct mbuf *m, *n;
629 	int space = asa->sa_len;
630 
631 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
632 panic("sbappendaddr");
633 	if (m0)
634 		space += m0->m_pkthdr.len;
635 	for (n = control; n; n = n->m_next) {
636 		space += n->m_len;
637 		if (n->m_next == 0)	/* keep pointer to last control buf */
638 			break;
639 	}
640 	if (space > sbspace(sb))
641 		return (0);
642 	if (asa->sa_len > MLEN)
643 		return (0);
644 	MGET(m, M_DONTWAIT, MT_SONAME);
645 	if (m == 0)
646 		return (0);
647 	m->m_len = asa->sa_len;
648 	bcopy((caddr_t)asa, mtod(m, caddr_t), asa->sa_len);
649 	if (n)
650 		n->m_next = m0;		/* concatenate data to control */
651 	else
652 		control = m0;
653 	m->m_next = control;
654 	for (n = m; n; n = n->m_next)
655 		sballoc(sb, n);
656 	n = sb->sb_mb;
657 	if (n) {
658 		while (n->m_nextpkt)
659 			n = n->m_nextpkt;
660 		n->m_nextpkt = m;
661 	} else
662 		sb->sb_mb = m;
663 	return (1);
664 }
665 
666 int
667 sbappendcontrol(sb, m0, control)
668 	struct sockbuf *sb;
669 	struct mbuf *control, *m0;
670 {
671 	register struct mbuf *m, *n;
672 	int space = 0;
673 
674 	if (control == 0)
675 		panic("sbappendcontrol");
676 	for (m = control; ; m = m->m_next) {
677 		space += m->m_len;
678 		if (m->m_next == 0)
679 			break;
680 	}
681 	n = m;			/* save pointer to last control buffer */
682 	for (m = m0; m; m = m->m_next)
683 		space += m->m_len;
684 	if (space > sbspace(sb))
685 		return (0);
686 	n->m_next = m0;			/* concatenate data to control */
687 	for (m = control; m; m = m->m_next)
688 		sballoc(sb, m);
689 	n = sb->sb_mb;
690 	if (n) {
691 		while (n->m_nextpkt)
692 			n = n->m_nextpkt;
693 		n->m_nextpkt = control;
694 	} else
695 		sb->sb_mb = control;
696 	return (1);
697 }
698 
699 /*
700  * Compress mbuf chain m into the socket
701  * buffer sb following mbuf n.  If n
702  * is null, the buffer is presumed empty.
703  */
704 void
705 sbcompress(sb, m, n)
706 	register struct sockbuf *sb;
707 	register struct mbuf *m, *n;
708 {
709 	register int eor = 0;
710 	register struct mbuf *o;
711 
712 	while (m) {
713 		eor |= m->m_flags & M_EOR;
714 		if (m->m_len == 0 &&
715 		    (eor == 0 ||
716 		     (((o = m->m_next) || (o = n)) &&
717 		      o->m_type == m->m_type))) {
718 			m = m_free(m);
719 			continue;
720 		}
721 		if (n && (n->m_flags & M_EOR) == 0 &&
722 		    M_WRITABLE(n) &&
723 		    m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
724 		    m->m_len <= M_TRAILINGSPACE(n) &&
725 		    n->m_type == m->m_type) {
726 			bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
727 			    (unsigned)m->m_len);
728 			n->m_len += m->m_len;
729 			sb->sb_cc += m->m_len;
730 			m = m_free(m);
731 			continue;
732 		}
733 		if (n)
734 			n->m_next = m;
735 		else
736 			sb->sb_mb = m;
737 		sballoc(sb, m);
738 		n = m;
739 		m->m_flags &= ~M_EOR;
740 		m = m->m_next;
741 		n->m_next = 0;
742 	}
743 	if (eor) {
744 		if (n)
745 			n->m_flags |= eor;
746 		else
747 			printf("semi-panic: sbcompress\n");
748 	}
749 }
750 
751 /*
752  * Free all mbufs in a sockbuf.
753  * Check that all resources are reclaimed.
754  */
755 void
756 sbflush(sb)
757 	register struct sockbuf *sb;
758 {
759 
760 	if (sb->sb_flags & SB_LOCK)
761 		panic("sbflush: locked");
762 	while (sb->sb_mbcnt) {
763 		/*
764 		 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty:
765 		 * we would loop forever. Panic instead.
766 		 */
767 		if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len))
768 			break;
769 		sbdrop(sb, (int)sb->sb_cc);
770 	}
771 	if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt)
772 		panic("sbflush: cc %ld || mb %p || mbcnt %ld", sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt);
773 }
774 
775 /*
776  * Drop data from (the front of) a sockbuf.
777  */
778 void
779 sbdrop(sb, len)
780 	register struct sockbuf *sb;
781 	register int len;
782 {
783 	register struct mbuf *m, *mn;
784 	struct mbuf *next;
785 
786 	next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
787 	while (len > 0) {
788 		if (m == 0) {
789 			if (next == 0)
790 				panic("sbdrop");
791 			m = next;
792 			next = m->m_nextpkt;
793 			continue;
794 		}
795 		if (m->m_len > len) {
796 			m->m_len -= len;
797 			m->m_data += len;
798 			sb->sb_cc -= len;
799 			break;
800 		}
801 		len -= m->m_len;
802 		sbfree(sb, m);
803 		MFREE(m, mn);
804 		m = mn;
805 	}
806 	while (m && m->m_len == 0) {
807 		sbfree(sb, m);
808 		MFREE(m, mn);
809 		m = mn;
810 	}
811 	if (m) {
812 		sb->sb_mb = m;
813 		m->m_nextpkt = next;
814 	} else
815 		sb->sb_mb = next;
816 }
817 
818 /*
819  * Drop a record off the front of a sockbuf
820  * and move the next record to the front.
821  */
822 void
823 sbdroprecord(sb)
824 	register struct sockbuf *sb;
825 {
826 	register struct mbuf *m, *mn;
827 
828 	m = sb->sb_mb;
829 	if (m) {
830 		sb->sb_mb = m->m_nextpkt;
831 		do {
832 			sbfree(sb, m);
833 			MFREE(m, mn);
834 			m = mn;
835 		} while (m);
836 	}
837 }
838 
839 /*
840  * Create a "control" mbuf containing the specified data
841  * with the specified type for presentation on a socket buffer.
842  */
843 struct mbuf *
844 sbcreatecontrol(p, size, type, level)
845 	caddr_t p;
846 	register int size;
847 	int type, level;
848 {
849 	register struct cmsghdr *cp;
850 	struct mbuf *m;
851 
852 	if (CMSG_SPACE((u_int)size) > MLEN)
853 		return ((struct mbuf *) NULL);
854 	if ((m = m_get(M_DONTWAIT, MT_CONTROL)) == NULL)
855 		return ((struct mbuf *) NULL);
856 	cp = mtod(m, struct cmsghdr *);
857 	/* XXX check size? */
858 	(void)memcpy(CMSG_DATA(cp), p, size);
859 	m->m_len = CMSG_SPACE(size);
860 	cp->cmsg_len = CMSG_LEN(size);
861 	cp->cmsg_level = level;
862 	cp->cmsg_type = type;
863 	return (m);
864 }
865 
866 /*
867  * Some routines that return EOPNOTSUPP for entry points that are not
868  * supported by a protocol.  Fill in as needed.
869  */
870 int
871 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
872 {
873 	return EOPNOTSUPP;
874 }
875 
876 int
877 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct proc *p)
878 {
879 	return EOPNOTSUPP;
880 }
881 
882 int
883 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
884 {
885 	return EOPNOTSUPP;
886 }
887 
888 int
889 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
890 		    struct ifnet *ifp, struct proc *p)
891 {
892 	return EOPNOTSUPP;
893 }
894 
895 int
896 pru_listen_notsupp(struct socket *so, struct proc *p)
897 {
898 	return EOPNOTSUPP;
899 }
900 
901 int
902 pru_rcvd_notsupp(struct socket *so, int flags)
903 {
904 	return EOPNOTSUPP;
905 }
906 
907 int
908 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
909 {
910 	return EOPNOTSUPP;
911 }
912 
913 /*
914  * This isn't really a ``null'' operation, but it's the default one
915  * and doesn't do anything destructive.
916  */
917 int
918 pru_sense_null(struct socket *so, struct stat *sb)
919 {
920 	sb->st_blksize = so->so_snd.sb_hiwat;
921 	return 0;
922 }
923 
924 /*
925  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
926  */
927 struct sockaddr *
928 dup_sockaddr(sa, canwait)
929 	struct sockaddr *sa;
930 	int canwait;
931 {
932 	struct sockaddr *sa2;
933 
934 	MALLOC(sa2, struct sockaddr *, sa->sa_len, M_SONAME,
935 	       canwait ? M_WAITOK : M_NOWAIT);
936 	if (sa2)
937 		bcopy(sa, sa2, sa->sa_len);
938 	return sa2;
939 }
940 
941 /*
942  * Create an external-format (``xsocket'') structure using the information
943  * in the kernel-format socket structure pointed to by so.  This is done
944  * to reduce the spew of irrelevant information over this interface,
945  * to isolate user code from changes in the kernel structure, and
946  * potentially to provide information-hiding if we decide that
947  * some of this information should be hidden from users.
948  */
949 void
950 sotoxsocket(struct socket *so, struct xsocket *xso)
951 {
952 	xso->xso_len = sizeof *xso;
953 	xso->xso_so = so;
954 	xso->so_type = so->so_type;
955 	xso->so_options = so->so_options;
956 	xso->so_linger = so->so_linger;
957 	xso->so_state = so->so_state;
958 	xso->so_pcb = so->so_pcb;
959 	xso->xso_protocol = so->so_proto->pr_protocol;
960 	xso->xso_family = so->so_proto->pr_domain->dom_family;
961 	xso->so_qlen = so->so_qlen;
962 	xso->so_incqlen = so->so_incqlen;
963 	xso->so_qlimit = so->so_qlimit;
964 	xso->so_timeo = so->so_timeo;
965 	xso->so_error = so->so_error;
966 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
967 	xso->so_oobmark = so->so_oobmark;
968 	sbtoxsockbuf(&so->so_snd, &xso->so_snd);
969 	sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
970 	xso->so_uid = so->so_cred->cr_uid;
971 }
972 
973 /*
974  * This does the same for sockbufs.  Note that the xsockbuf structure,
975  * since it is always embedded in a socket, does not include a self
976  * pointer nor a length.  We make this entry point public in case
977  * some other mechanism needs it.
978  */
979 void
980 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb)
981 {
982 	xsb->sb_cc = sb->sb_cc;
983 	xsb->sb_hiwat = sb->sb_hiwat;
984 	xsb->sb_mbcnt = sb->sb_mbcnt;
985 	xsb->sb_mbmax = sb->sb_mbmax;
986 	xsb->sb_lowat = sb->sb_lowat;
987 	xsb->sb_flags = sb->sb_flags;
988 	xsb->sb_timeo = sb->sb_timeo;
989 }
990 
991 /*
992  * Here is the definition of some of the basic objects in the kern.ipc
993  * branch of the MIB.
994  */
995 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
996 
997 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
998 static int dummy;
999 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
1000 
1001 SYSCTL_INT(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLFLAG_RW,
1002     &sb_max, 0, "Maximum socket buffer size");
1003 SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RD,
1004     &maxsockets, 0, "Maximum number of sockets avaliable");
1005 SYSCTL_INT(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
1006     &sb_efficiency, 0, "");
1007 
1008 /*
1009  * Initialise maxsockets
1010  */
1011 static void init_maxsockets(void *ignored)
1012 {
1013     TUNABLE_INT_FETCH("kern.ipc.maxsockets", 0, maxsockets);
1014     maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
1015 }
1016 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
1017