xref: /freebsd/sys/kern/uipc_sockbuf.c (revision b52b9d56d4e96089873a75f9e29062eec19fabba)
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/aio.h> /* for aio_swake proto */
40 #include <sys/domain.h>
41 #include <sys/event.h>
42 #include <sys/file.h>	/* for maxfiles */
43 #include <sys/kernel.h>
44 #include <sys/lock.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/mutex.h>
48 #include <sys/proc.h>
49 #include <sys/protosw.h>
50 #include <sys/resourcevar.h>
51 #include <sys/signalvar.h>
52 #include <sys/socket.h>
53 #include <sys/socketvar.h>
54 #include <sys/stat.h>
55 #include <sys/sysctl.h>
56 #include <sys/systm.h>
57 
58 int	maxsockets;
59 
60 void (*aio_swake)(struct socket *, struct sockbuf *);
61 
62 /*
63  * Primitive routines for operating on sockets and socket buffers
64  */
65 
66 u_long	sb_max = SB_MAX;		/* XXX should be static */
67 
68 static	u_long sb_efficiency = 8;	/* parameter for sbreserve() */
69 
70 /*
71  * Procedures to manipulate state flags of socket
72  * and do appropriate wakeups.  Normal sequence from the
73  * active (originating) side is that soisconnecting() is
74  * called during processing of connect() call,
75  * resulting in an eventual call to soisconnected() if/when the
76  * connection is established.  When the connection is torn down
77  * soisdisconnecting() is called during processing of disconnect() call,
78  * and soisdisconnected() is called when the connection to the peer
79  * is totally severed.  The semantics of these routines are such that
80  * connectionless protocols can call soisconnected() and soisdisconnected()
81  * only, bypassing the in-progress calls when setting up a ``connection''
82  * takes no time.
83  *
84  * From the passive side, a socket is created with
85  * two queues of sockets: so_incomp for connections in progress
86  * and so_comp for connections already made and awaiting user acceptance.
87  * As a protocol is preparing incoming connections, it creates a socket
88  * structure queued on so_incomp by calling sonewconn().  When the connection
89  * is established, soisconnected() is called, and transfers the
90  * socket structure to so_comp, making it available to accept().
91  *
92  * If a socket is closed with sockets on either
93  * so_incomp or so_comp, these sockets are dropped.
94  *
95  * If higher level protocols are implemented in
96  * the kernel, the wakeups done here will sometimes
97  * cause software-interrupt process scheduling.
98  */
99 
100 void
101 soisconnecting(so)
102 	register struct socket *so;
103 {
104 
105 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
106 	so->so_state |= SS_ISCONNECTING;
107 }
108 
109 void
110 soisconnected(so)
111 	struct socket *so;
112 {
113 	struct socket *head = so->so_head;
114 
115 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
116 	so->so_state |= SS_ISCONNECTED;
117 	if (head && (so->so_state & SS_INCOMP)) {
118 		if ((so->so_options & SO_ACCEPTFILTER) != 0) {
119 			so->so_upcall = head->so_accf->so_accept_filter->accf_callback;
120 			so->so_upcallarg = head->so_accf->so_accept_filter_arg;
121 			so->so_rcv.sb_flags |= SB_UPCALL;
122 			so->so_options &= ~SO_ACCEPTFILTER;
123 			so->so_upcall(so, so->so_upcallarg, 0);
124 			return;
125 		}
126 		TAILQ_REMOVE(&head->so_incomp, so, so_list);
127 		head->so_incqlen--;
128 		so->so_state &= ~SS_INCOMP;
129 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
130 		head->so_qlen++;
131 		so->so_state |= SS_COMP;
132 		sorwakeup(head);
133 		wakeup_one(&head->so_timeo);
134 	} else {
135 		wakeup(&so->so_timeo);
136 		sorwakeup(so);
137 		sowwakeup(so);
138 	}
139 }
140 
141 void
142 soisdisconnecting(so)
143 	register struct socket *so;
144 {
145 
146 	so->so_state &= ~SS_ISCONNECTING;
147 	so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
148 	wakeup(&so->so_timeo);
149 	sowwakeup(so);
150 	sorwakeup(so);
151 }
152 
153 void
154 soisdisconnected(so)
155 	register struct socket *so;
156 {
157 
158 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
159 	so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED);
160 	wakeup(&so->so_timeo);
161 	sbdrop(&so->so_snd, so->so_snd.sb_cc);
162 	sowwakeup(so);
163 	sorwakeup(so);
164 }
165 
166 /*
167  * When an attempt at a new connection is noted on a socket
168  * which accepts connections, sonewconn is called.  If the
169  * connection is possible (subject to space constraints, etc.)
170  * then we allocate a new structure, propoerly linked into the
171  * data structure of the original socket, and return this.
172  * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
173  *
174  * note: the ref count on the socket is 0 on return
175  */
176 struct socket *
177 sonewconn(head, connstatus)
178 	register struct socket *head;
179 	int connstatus;
180 {
181 	register struct socket *so;
182 
183 	if (head->so_qlen > 3 * head->so_qlimit / 2)
184 		return ((struct socket *)0);
185 	so = soalloc(0);
186 	if (so == NULL)
187 		return ((struct socket *)0);
188 	if ((head->so_options & SO_ACCEPTFILTER) != 0)
189 		connstatus = 0;
190 	so->so_head = head;
191 	so->so_type = head->so_type;
192 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
193 	so->so_linger = head->so_linger;
194 	so->so_state = head->so_state | SS_NOFDREF;
195 	so->so_proto = head->so_proto;
196 	so->so_timeo = head->so_timeo;
197 	so->so_cred = crhold(head->so_cred);
198 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
199 	    (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
200 		sotryfree(so);
201 		return ((struct socket *)0);
202 	}
203 
204 	if (connstatus) {
205 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
206 		so->so_state |= SS_COMP;
207 		head->so_qlen++;
208 	} else {
209 		if (head->so_incqlen > head->so_qlimit) {
210 			struct socket *sp;
211 			sp = TAILQ_FIRST(&head->so_incomp);
212 			(void) soabort(sp);
213 		}
214 		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
215 		so->so_state |= SS_INCOMP;
216 		head->so_incqlen++;
217 	}
218 	if (connstatus) {
219 		sorwakeup(head);
220 		wakeup(&head->so_timeo);
221 		so->so_state |= connstatus;
222 	}
223 	return (so);
224 }
225 
226 /*
227  * Socantsendmore indicates that no more data will be sent on the
228  * socket; it would normally be applied to a socket when the user
229  * informs the system that no more data is to be sent, by the protocol
230  * code (in case PRU_SHUTDOWN).  Socantrcvmore indicates that no more data
231  * will be received, and will normally be applied to the socket by a
232  * protocol when it detects that the peer will send no more data.
233  * Data queued for reading in the socket may yet be read.
234  */
235 
236 void
237 socantsendmore(so)
238 	struct socket *so;
239 {
240 
241 	so->so_state |= SS_CANTSENDMORE;
242 	sowwakeup(so);
243 }
244 
245 void
246 socantrcvmore(so)
247 	struct socket *so;
248 {
249 
250 	so->so_state |= SS_CANTRCVMORE;
251 	sorwakeup(so);
252 }
253 
254 /*
255  * Wait for data to arrive at/drain from a socket buffer.
256  */
257 int
258 sbwait(sb)
259 	struct sockbuf *sb;
260 {
261 
262 	sb->sb_flags |= SB_WAIT;
263 	return (tsleep(&sb->sb_cc,
264 	    (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
265 	    sb->sb_timeo));
266 }
267 
268 /*
269  * Lock a sockbuf already known to be locked;
270  * return any error returned from sleep (EINTR).
271  */
272 int
273 sb_lock(sb)
274 	register struct sockbuf *sb;
275 {
276 	int error;
277 
278 	while (sb->sb_flags & SB_LOCK) {
279 		sb->sb_flags |= SB_WANT;
280 		error = tsleep(&sb->sb_flags,
281 		    (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH,
282 		    "sblock", 0);
283 		if (error)
284 			return (error);
285 	}
286 	sb->sb_flags |= SB_LOCK;
287 	return (0);
288 }
289 
290 /*
291  * Wakeup processes waiting on a socket buffer.
292  * Do asynchronous notification via SIGIO
293  * if the socket has the SS_ASYNC flag set.
294  */
295 void
296 sowakeup(so, sb)
297 	register struct socket *so;
298 	register struct sockbuf *sb;
299 {
300 
301 	selwakeup(&sb->sb_sel);
302 	sb->sb_flags &= ~SB_SEL;
303 	if (sb->sb_flags & SB_WAIT) {
304 		sb->sb_flags &= ~SB_WAIT;
305 		wakeup(&sb->sb_cc);
306 	}
307 	if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
308 		pgsigio(&so->so_sigio, SIGIO, 0);
309 	if (sb->sb_flags & SB_UPCALL)
310 		(*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT);
311 	if (sb->sb_flags & SB_AIO)
312 		aio_swake(so, sb);
313 	KNOTE(&sb->sb_sel.si_note, 0);
314 }
315 
316 /*
317  * Socket buffer (struct sockbuf) utility routines.
318  *
319  * Each socket contains two socket buffers: one for sending data and
320  * one for receiving data.  Each buffer contains a queue of mbufs,
321  * information about the number of mbufs and amount of data in the
322  * queue, and other fields allowing select() statements and notification
323  * on data availability to be implemented.
324  *
325  * Data stored in a socket buffer is maintained as a list of records.
326  * Each record is a list of mbufs chained together with the m_next
327  * field.  Records are chained together with the m_nextpkt field. The upper
328  * level routine soreceive() expects the following conventions to be
329  * observed when placing information in the receive buffer:
330  *
331  * 1. If the protocol requires each message be preceded by the sender's
332  *    name, then a record containing that name must be present before
333  *    any associated data (mbuf's must be of type MT_SONAME).
334  * 2. If the protocol supports the exchange of ``access rights'' (really
335  *    just additional data associated with the message), and there are
336  *    ``rights'' to be received, then a record containing this data
337  *    should be present (mbuf's must be of type MT_RIGHTS).
338  * 3. If a name or rights record exists, then it must be followed by
339  *    a data record, perhaps of zero length.
340  *
341  * Before using a new socket structure it is first necessary to reserve
342  * buffer space to the socket, by calling sbreserve().  This should commit
343  * some of the available buffer space in the system buffer pool for the
344  * socket (currently, it does nothing but enforce limits).  The space
345  * should be released by calling sbrelease() when the socket is destroyed.
346  */
347 
348 int
349 soreserve(so, sndcc, rcvcc)
350 	register struct socket *so;
351 	u_long sndcc, rcvcc;
352 {
353 	struct thread *td = curthread;
354 
355 	if (sbreserve(&so->so_snd, sndcc, so, td) == 0)
356 		goto bad;
357 	if (sbreserve(&so->so_rcv, rcvcc, so, td) == 0)
358 		goto bad2;
359 	if (so->so_rcv.sb_lowat == 0)
360 		so->so_rcv.sb_lowat = 1;
361 	if (so->so_snd.sb_lowat == 0)
362 		so->so_snd.sb_lowat = MCLBYTES;
363 	if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
364 		so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
365 	return (0);
366 bad2:
367 	sbrelease(&so->so_snd, so);
368 bad:
369 	return (ENOBUFS);
370 }
371 
372 /*
373  * Allot mbufs to a sockbuf.
374  * Attempt to scale mbmax so that mbcnt doesn't become limiting
375  * if buffering efficiency is near the normal case.
376  */
377 int
378 sbreserve(sb, cc, so, td)
379 	struct sockbuf *sb;
380 	u_long cc;
381 	struct socket *so;
382 	struct thread *td;
383 {
384 
385 	/*
386 	 * td will only be NULL when we're in an interrupt
387 	 * (e.g. in tcp_input())
388 	 */
389 	if ((u_quad_t)cc > (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES))
390 		return (0);
391 	if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
392 	    td ? td->td_proc->p_rlimit[RLIMIT_SBSIZE].rlim_cur : RLIM_INFINITY)) {
393 		return (0);
394 	}
395 	sb->sb_mbmax = min(cc * sb_efficiency, sb_max);
396 	if (sb->sb_lowat > sb->sb_hiwat)
397 		sb->sb_lowat = sb->sb_hiwat;
398 	return (1);
399 }
400 
401 /*
402  * Free mbufs held by a socket, and reserved mbuf space.
403  */
404 void
405 sbrelease(sb, so)
406 	struct sockbuf *sb;
407 	struct socket *so;
408 {
409 
410 	sbflush(sb);
411 	(void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
412 	    RLIM_INFINITY);
413 	sb->sb_mbmax = 0;
414 }
415 
416 /*
417  * Routines to add and remove
418  * data from an mbuf queue.
419  *
420  * The routines sbappend() or sbappendrecord() are normally called to
421  * append new mbufs to a socket buffer, after checking that adequate
422  * space is available, comparing the function sbspace() with the amount
423  * of data to be added.  sbappendrecord() differs from sbappend() in
424  * that data supplied is treated as the beginning of a new record.
425  * To place a sender's address, optional access rights, and data in a
426  * socket receive buffer, sbappendaddr() should be used.  To place
427  * access rights and data in a socket receive buffer, sbappendrights()
428  * should be used.  In either case, the new data begins a new record.
429  * Note that unlike sbappend() and sbappendrecord(), these routines check
430  * for the caller that there will be enough space to store the data.
431  * Each fails if there is not enough space, or if it cannot find mbufs
432  * to store additional information in.
433  *
434  * Reliable protocols may use the socket send buffer to hold data
435  * awaiting acknowledgement.  Data is normally copied from a socket
436  * send buffer in a protocol with m_copy for output to a peer,
437  * and then removing the data from the socket buffer with sbdrop()
438  * or sbdroprecord() when the data is acknowledged by the peer.
439  */
440 
441 /*
442  * Append mbuf chain m to the last record in the
443  * socket buffer sb.  The additional space associated
444  * the mbuf chain is recorded in sb.  Empty mbufs are
445  * discarded and mbufs are compacted where possible.
446  */
447 void
448 sbappend(sb, m)
449 	struct sockbuf *sb;
450 	struct mbuf *m;
451 {
452 	register struct mbuf *n;
453 
454 	if (m == 0)
455 		return;
456 	n = sb->sb_mb;
457 	if (n) {
458 		while (n->m_nextpkt)
459 			n = n->m_nextpkt;
460 		do {
461 			if (n->m_flags & M_EOR) {
462 				sbappendrecord(sb, m); /* XXXXXX!!!! */
463 				return;
464 			}
465 		} while (n->m_next && (n = n->m_next));
466 	}
467 	sbcompress(sb, m, n);
468 }
469 
470 #ifdef SOCKBUF_DEBUG
471 void
472 sbcheck(sb)
473 	register struct sockbuf *sb;
474 {
475 	register struct mbuf *m;
476 	register struct mbuf *n = 0;
477 	register u_long len = 0, mbcnt = 0;
478 
479 	for (m = sb->sb_mb; m; m = n) {
480 	    n = m->m_nextpkt;
481 	    for (; m; m = m->m_next) {
482 		len += m->m_len;
483 		mbcnt += MSIZE;
484 		if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
485 			mbcnt += m->m_ext.ext_size;
486 	    }
487 	}
488 	if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
489 		printf("cc %ld != %ld || mbcnt %ld != %ld\n", len, sb->sb_cc,
490 		    mbcnt, sb->sb_mbcnt);
491 		panic("sbcheck");
492 	}
493 }
494 #endif
495 
496 /*
497  * As above, except the mbuf chain
498  * begins a new record.
499  */
500 void
501 sbappendrecord(sb, m0)
502 	register struct sockbuf *sb;
503 	register struct mbuf *m0;
504 {
505 	register struct mbuf *m;
506 
507 	if (m0 == 0)
508 		return;
509 	m = sb->sb_mb;
510 	if (m)
511 		while (m->m_nextpkt)
512 			m = m->m_nextpkt;
513 	/*
514 	 * Put the first mbuf on the queue.
515 	 * Note this permits zero length records.
516 	 */
517 	sballoc(sb, m0);
518 	if (m)
519 		m->m_nextpkt = m0;
520 	else
521 		sb->sb_mb = m0;
522 	m = m0->m_next;
523 	m0->m_next = 0;
524 	if (m && (m0->m_flags & M_EOR)) {
525 		m0->m_flags &= ~M_EOR;
526 		m->m_flags |= M_EOR;
527 	}
528 	sbcompress(sb, m, m0);
529 }
530 
531 /*
532  * As above except that OOB data
533  * is inserted at the beginning of the sockbuf,
534  * but after any other OOB data.
535  */
536 void
537 sbinsertoob(sb, m0)
538 	register struct sockbuf *sb;
539 	register struct mbuf *m0;
540 {
541 	register struct mbuf *m;
542 	register struct mbuf **mp;
543 
544 	if (m0 == 0)
545 		return;
546 	for (mp = &sb->sb_mb; *mp ; mp = &((*mp)->m_nextpkt)) {
547 	    m = *mp;
548 	    again:
549 		switch (m->m_type) {
550 
551 		case MT_OOBDATA:
552 			continue;		/* WANT next train */
553 
554 		case MT_CONTROL:
555 			m = m->m_next;
556 			if (m)
557 				goto again;	/* inspect THIS train further */
558 		}
559 		break;
560 	}
561 	/*
562 	 * Put the first mbuf on the queue.
563 	 * Note this permits zero length records.
564 	 */
565 	sballoc(sb, m0);
566 	m0->m_nextpkt = *mp;
567 	*mp = m0;
568 	m = m0->m_next;
569 	m0->m_next = 0;
570 	if (m && (m0->m_flags & M_EOR)) {
571 		m0->m_flags &= ~M_EOR;
572 		m->m_flags |= M_EOR;
573 	}
574 	sbcompress(sb, m, m0);
575 }
576 
577 /*
578  * Append address and data, and optionally, control (ancillary) data
579  * to the receive queue of a socket.  If present,
580  * m0 must include a packet header with total length.
581  * Returns 0 if no space in sockbuf or insufficient mbufs.
582  */
583 int
584 sbappendaddr(sb, asa, m0, control)
585 	register struct sockbuf *sb;
586 	struct sockaddr *asa;
587 	struct mbuf *m0, *control;
588 {
589 	register struct mbuf *m, *n;
590 	int space = asa->sa_len;
591 
592 	if (m0 && (m0->m_flags & M_PKTHDR) == 0)
593 		panic("sbappendaddr");
594 	if (m0)
595 		space += m0->m_pkthdr.len;
596 	for (n = control; n; n = n->m_next) {
597 		space += n->m_len;
598 		if (n->m_next == 0)	/* keep pointer to last control buf */
599 			break;
600 	}
601 	if (space > sbspace(sb))
602 		return (0);
603 	if (asa->sa_len > MLEN)
604 		return (0);
605 	MGET(m, M_DONTWAIT, MT_SONAME);
606 	if (m == 0)
607 		return (0);
608 	m->m_len = asa->sa_len;
609 	bcopy(asa, mtod(m, caddr_t), asa->sa_len);
610 	if (n)
611 		n->m_next = m0;		/* concatenate data to control */
612 	else
613 		control = m0;
614 	m->m_next = control;
615 	for (n = m; n; n = n->m_next)
616 		sballoc(sb, n);
617 	n = sb->sb_mb;
618 	if (n) {
619 		while (n->m_nextpkt)
620 			n = n->m_nextpkt;
621 		n->m_nextpkt = m;
622 	} else
623 		sb->sb_mb = m;
624 	return (1);
625 }
626 
627 int
628 sbappendcontrol(sb, m0, control)
629 	struct sockbuf *sb;
630 	struct mbuf *control, *m0;
631 {
632 	register struct mbuf *m, *n;
633 	int space = 0;
634 
635 	if (control == 0)
636 		panic("sbappendcontrol");
637 	for (m = control; ; m = m->m_next) {
638 		space += m->m_len;
639 		if (m->m_next == 0)
640 			break;
641 	}
642 	n = m;			/* save pointer to last control buffer */
643 	for (m = m0; m; m = m->m_next)
644 		space += m->m_len;
645 	if (space > sbspace(sb))
646 		return (0);
647 	n->m_next = m0;			/* concatenate data to control */
648 	for (m = control; m; m = m->m_next)
649 		sballoc(sb, m);
650 	n = sb->sb_mb;
651 	if (n) {
652 		while (n->m_nextpkt)
653 			n = n->m_nextpkt;
654 		n->m_nextpkt = control;
655 	} else
656 		sb->sb_mb = control;
657 	return (1);
658 }
659 
660 /*
661  * Compress mbuf chain m into the socket
662  * buffer sb following mbuf n.  If n
663  * is null, the buffer is presumed empty.
664  */
665 void
666 sbcompress(sb, m, n)
667 	register struct sockbuf *sb;
668 	register struct mbuf *m, *n;
669 {
670 	register int eor = 0;
671 	register struct mbuf *o;
672 
673 	while (m) {
674 		eor |= m->m_flags & M_EOR;
675 		if (m->m_len == 0 &&
676 		    (eor == 0 ||
677 		     (((o = m->m_next) || (o = n)) &&
678 		      o->m_type == m->m_type))) {
679 			m = m_free(m);
680 			continue;
681 		}
682 		if (n && (n->m_flags & M_EOR) == 0 &&
683 		    M_WRITABLE(n) &&
684 		    m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
685 		    m->m_len <= M_TRAILINGSPACE(n) &&
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 		/*
725 		 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty:
726 		 * we would loop forever. Panic instead.
727 		 */
728 		if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len))
729 			break;
730 		sbdrop(sb, (int)sb->sb_cc);
731 	}
732 	if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt)
733 		panic("sbflush: cc %u || mb %p || mbcnt %u", sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt);
734 }
735 
736 /*
737  * Drop data from (the front of) a sockbuf.
738  */
739 void
740 sbdrop(sb, len)
741 	register struct sockbuf *sb;
742 	register int len;
743 {
744 	register struct mbuf *m;
745 	struct mbuf *next;
746 
747 	next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
748 	while (len > 0) {
749 		if (m == 0) {
750 			if (next == 0)
751 				panic("sbdrop");
752 			m = next;
753 			next = m->m_nextpkt;
754 			continue;
755 		}
756 		if (m->m_len > len) {
757 			m->m_len -= len;
758 			m->m_data += len;
759 			sb->sb_cc -= len;
760 			break;
761 		}
762 		len -= m->m_len;
763 		sbfree(sb, m);
764 		m = m_free(m);
765 	}
766 	while (m && m->m_len == 0) {
767 		sbfree(sb, m);
768 		m = m_free(m);
769 	}
770 	if (m) {
771 		sb->sb_mb = m;
772 		m->m_nextpkt = next;
773 	} else
774 		sb->sb_mb = next;
775 }
776 
777 /*
778  * Drop a record off the front of a sockbuf
779  * and move the next record to the front.
780  */
781 void
782 sbdroprecord(sb)
783 	register struct sockbuf *sb;
784 {
785 	register struct mbuf *m;
786 
787 	m = sb->sb_mb;
788 	if (m) {
789 		sb->sb_mb = m->m_nextpkt;
790 		do {
791 			sbfree(sb, m);
792 			m = m_free(m);
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 (CMSG_SPACE((u_int)size) > MCLBYTES)
811 		return ((struct mbuf *) NULL);
812 	if ((m = m_get(M_DONTWAIT, MT_CONTROL)) == NULL)
813 		return ((struct mbuf *) NULL);
814 	if (CMSG_SPACE((u_int)size) > MLEN) {
815 		MCLGET(m, M_DONTWAIT);
816 		if ((m->m_flags & M_EXT) == 0) {
817 			m_free(m);
818 			return ((struct mbuf *) NULL);
819 		}
820 	}
821 	cp = mtod(m, struct cmsghdr *);
822 	m->m_len = 0;
823 	KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m),
824 	    ("sbcreatecontrol: short mbuf"));
825 	if (p != NULL)
826 		(void)memcpy(CMSG_DATA(cp), p, size);
827 	m->m_len = CMSG_SPACE(size);
828 	cp->cmsg_len = CMSG_LEN(size);
829 	cp->cmsg_level = level;
830 	cp->cmsg_type = type;
831 	return (m);
832 }
833 
834 /*
835  * Some routines that return EOPNOTSUPP for entry points that are not
836  * supported by a protocol.  Fill in as needed.
837  */
838 int
839 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
840 {
841 	return EOPNOTSUPP;
842 }
843 
844 int
845 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
846 {
847 	return EOPNOTSUPP;
848 }
849 
850 int
851 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
852 {
853 	return EOPNOTSUPP;
854 }
855 
856 int
857 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
858 		    struct ifnet *ifp, struct thread *td)
859 {
860 	return EOPNOTSUPP;
861 }
862 
863 int
864 pru_listen_notsupp(struct socket *so, struct thread *td)
865 {
866 	return EOPNOTSUPP;
867 }
868 
869 int
870 pru_rcvd_notsupp(struct socket *so, int flags)
871 {
872 	return EOPNOTSUPP;
873 }
874 
875 int
876 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
877 {
878 	return EOPNOTSUPP;
879 }
880 
881 /*
882  * This isn't really a ``null'' operation, but it's the default one
883  * and doesn't do anything destructive.
884  */
885 int
886 pru_sense_null(struct socket *so, struct stat *sb)
887 {
888 	sb->st_blksize = so->so_snd.sb_hiwat;
889 	return 0;
890 }
891 
892 /*
893  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
894  */
895 struct sockaddr *
896 dup_sockaddr(sa, canwait)
897 	struct sockaddr *sa;
898 	int canwait;
899 {
900 	struct sockaddr *sa2;
901 
902 	MALLOC(sa2, struct sockaddr *, sa->sa_len, M_SONAME,
903 	       canwait ? M_WAITOK : M_NOWAIT);
904 	if (sa2)
905 		bcopy(sa, sa2, sa->sa_len);
906 	return sa2;
907 }
908 
909 /*
910  * Create an external-format (``xsocket'') structure using the information
911  * in the kernel-format socket structure pointed to by so.  This is done
912  * to reduce the spew of irrelevant information over this interface,
913  * to isolate user code from changes in the kernel structure, and
914  * potentially to provide information-hiding if we decide that
915  * some of this information should be hidden from users.
916  */
917 void
918 sotoxsocket(struct socket *so, struct xsocket *xso)
919 {
920 	xso->xso_len = sizeof *xso;
921 	xso->xso_so = so;
922 	xso->so_type = so->so_type;
923 	xso->so_options = so->so_options;
924 	xso->so_linger = so->so_linger;
925 	xso->so_state = so->so_state;
926 	xso->so_pcb = so->so_pcb;
927 	xso->xso_protocol = so->so_proto->pr_protocol;
928 	xso->xso_family = so->so_proto->pr_domain->dom_family;
929 	xso->so_qlen = so->so_qlen;
930 	xso->so_incqlen = so->so_incqlen;
931 	xso->so_qlimit = so->so_qlimit;
932 	xso->so_timeo = so->so_timeo;
933 	xso->so_error = so->so_error;
934 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
935 	xso->so_oobmark = so->so_oobmark;
936 	sbtoxsockbuf(&so->so_snd, &xso->so_snd);
937 	sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
938 	xso->so_uid = so->so_cred->cr_uid;
939 }
940 
941 /*
942  * This does the same for sockbufs.  Note that the xsockbuf structure,
943  * since it is always embedded in a socket, does not include a self
944  * pointer nor a length.  We make this entry point public in case
945  * some other mechanism needs it.
946  */
947 void
948 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb)
949 {
950 	xsb->sb_cc = sb->sb_cc;
951 	xsb->sb_hiwat = sb->sb_hiwat;
952 	xsb->sb_mbcnt = sb->sb_mbcnt;
953 	xsb->sb_mbmax = sb->sb_mbmax;
954 	xsb->sb_lowat = sb->sb_lowat;
955 	xsb->sb_flags = sb->sb_flags;
956 	xsb->sb_timeo = sb->sb_timeo;
957 }
958 
959 /*
960  * Here is the definition of some of the basic objects in the kern.ipc
961  * branch of the MIB.
962  */
963 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
964 
965 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
966 static int dummy;
967 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
968 
969 SYSCTL_INT(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLFLAG_RW,
970     &sb_max, 0, "Maximum socket buffer size");
971 SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RD,
972     &maxsockets, 0, "Maximum number of sockets avaliable");
973 SYSCTL_INT(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
974     &sb_efficiency, 0, "");
975 
976 /*
977  * Initialise maxsockets
978  */
979 static void init_maxsockets(void *ignored)
980 {
981 	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
982 	maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
983 }
984 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
985