xref: /freebsd/sys/kern/uipc_sockbuf.c (revision f9218d3d4fd34f082473b3a021c6d4d109fb47cf)
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_mac.h"
38 #include "opt_param.h"
39 
40 #include <sys/param.h>
41 #include <sys/aio.h> /* for aio_swake proto */
42 #include <sys/domain.h>
43 #include <sys/event.h>
44 #include <sys/file.h>	/* for maxfiles */
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/mac.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/mutex.h>
51 #include <sys/proc.h>
52 #include <sys/protosw.h>
53 #include <sys/resourcevar.h>
54 #include <sys/signalvar.h>
55 #include <sys/socket.h>
56 #include <sys/socketvar.h>
57 #include <sys/stat.h>
58 #include <sys/sysctl.h>
59 #include <sys/systm.h>
60 
61 int	maxsockets;
62 
63 void (*aio_swake)(struct socket *, struct sockbuf *);
64 
65 /*
66  * Primitive routines for operating on sockets and socket buffers
67  */
68 
69 u_long	sb_max = SB_MAX;
70 u_long	sb_max_adj =
71     SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */
72 
73 static	u_long sb_efficiency = 8;	/* parameter for sbreserve() */
74 
75 /*
76  * Procedures to manipulate state flags of socket
77  * and do appropriate wakeups.  Normal sequence from the
78  * active (originating) side is that soisconnecting() is
79  * called during processing of connect() call,
80  * resulting in an eventual call to soisconnected() if/when the
81  * connection is established.  When the connection is torn down
82  * soisdisconnecting() is called during processing of disconnect() call,
83  * and soisdisconnected() is called when the connection to the peer
84  * is totally severed.  The semantics of these routines are such that
85  * connectionless protocols can call soisconnected() and soisdisconnected()
86  * only, bypassing the in-progress calls when setting up a ``connection''
87  * takes no time.
88  *
89  * From the passive side, a socket is created with
90  * two queues of sockets: so_incomp for connections in progress
91  * and so_comp for connections already made and awaiting user acceptance.
92  * As a protocol is preparing incoming connections, it creates a socket
93  * structure queued on so_incomp by calling sonewconn().  When the connection
94  * is established, soisconnected() is called, and transfers the
95  * socket structure to so_comp, making it available to accept().
96  *
97  * If a socket is closed with sockets on either
98  * so_incomp or so_comp, these sockets are dropped.
99  *
100  * If higher level protocols are implemented in
101  * the kernel, the wakeups done here will sometimes
102  * cause software-interrupt process scheduling.
103  */
104 
105 void
106 soisconnecting(so)
107 	register struct socket *so;
108 {
109 
110 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
111 	so->so_state |= SS_ISCONNECTING;
112 }
113 
114 void
115 soisconnected(so)
116 	struct socket *so;
117 {
118 	struct socket *head = so->so_head;
119 
120 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
121 	so->so_state |= SS_ISCONNECTED;
122 	if (head && (so->so_state & SS_INCOMP)) {
123 		if ((so->so_options & SO_ACCEPTFILTER) != 0) {
124 			so->so_upcall = head->so_accf->so_accept_filter->accf_callback;
125 			so->so_upcallarg = head->so_accf->so_accept_filter_arg;
126 			so->so_rcv.sb_flags |= SB_UPCALL;
127 			so->so_options &= ~SO_ACCEPTFILTER;
128 			so->so_upcall(so, so->so_upcallarg, M_TRYWAIT);
129 			return;
130 		}
131 		TAILQ_REMOVE(&head->so_incomp, so, so_list);
132 		head->so_incqlen--;
133 		so->so_state &= ~SS_INCOMP;
134 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
135 		head->so_qlen++;
136 		so->so_state |= SS_COMP;
137 		sorwakeup(head);
138 		wakeup_one(&head->so_timeo);
139 	} else {
140 		wakeup(&so->so_timeo);
141 		sorwakeup(so);
142 		sowwakeup(so);
143 	}
144 }
145 
146 void
147 soisdisconnecting(so)
148 	register struct socket *so;
149 {
150 
151 	so->so_state &= ~SS_ISCONNECTING;
152 	so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
153 	wakeup(&so->so_timeo);
154 	sowwakeup(so);
155 	sorwakeup(so);
156 }
157 
158 void
159 soisdisconnected(so)
160 	register struct socket *so;
161 {
162 
163 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
164 	so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED);
165 	wakeup(&so->so_timeo);
166 	sbdrop(&so->so_snd, so->so_snd.sb_cc);
167 	sowwakeup(so);
168 	sorwakeup(so);
169 }
170 
171 /*
172  * When an attempt at a new connection is noted on a socket
173  * which accepts connections, sonewconn is called.  If the
174  * connection is possible (subject to space constraints, etc.)
175  * then we allocate a new structure, propoerly linked into the
176  * data structure of the original socket, and return this.
177  * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
178  *
179  * note: the ref count on the socket is 0 on return
180  */
181 struct socket *
182 sonewconn(head, connstatus)
183 	register struct socket *head;
184 	int connstatus;
185 {
186 	register struct socket *so;
187 
188 	if (head->so_qlen > 3 * head->so_qlimit / 2)
189 		return ((struct socket *)0);
190 	so = soalloc(0);
191 	if (so == NULL)
192 		return ((struct socket *)0);
193 	if ((head->so_options & SO_ACCEPTFILTER) != 0)
194 		connstatus = 0;
195 	so->so_head = head;
196 	so->so_type = head->so_type;
197 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
198 	so->so_linger = head->so_linger;
199 	so->so_state = head->so_state | SS_NOFDREF;
200 	so->so_proto = head->so_proto;
201 	so->so_timeo = head->so_timeo;
202 	so->so_cred = crhold(head->so_cred);
203 #ifdef MAC
204 	mac_create_socket_from_socket(head, so);
205 #endif
206 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
207 	    (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
208 		sodealloc(so);
209 		return ((struct socket *)0);
210 	}
211 
212 	if (connstatus) {
213 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
214 		so->so_state |= SS_COMP;
215 		head->so_qlen++;
216 	} else {
217 		if (head->so_incqlen > head->so_qlimit) {
218 			struct socket *sp;
219 			sp = TAILQ_FIRST(&head->so_incomp);
220 			(void) soabort(sp);
221 		}
222 		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
223 		so->so_state |= SS_INCOMP;
224 		head->so_incqlen++;
225 	}
226 	if (connstatus) {
227 		sorwakeup(head);
228 		wakeup(&head->so_timeo);
229 		so->so_state |= connstatus;
230 	}
231 	return (so);
232 }
233 
234 /*
235  * Socantsendmore indicates that no more data will be sent on the
236  * socket; it would normally be applied to a socket when the user
237  * informs the system that no more data is to be sent, by the protocol
238  * code (in case PRU_SHUTDOWN).  Socantrcvmore indicates that no more data
239  * will be received, and will normally be applied to the socket by a
240  * protocol when it detects that the peer will send no more data.
241  * Data queued for reading in the socket may yet be read.
242  */
243 
244 void
245 socantsendmore(so)
246 	struct socket *so;
247 {
248 
249 	so->so_state |= SS_CANTSENDMORE;
250 	sowwakeup(so);
251 }
252 
253 void
254 socantrcvmore(so)
255 	struct socket *so;
256 {
257 
258 	so->so_state |= SS_CANTRCVMORE;
259 	sorwakeup(so);
260 }
261 
262 /*
263  * Wait for data to arrive at/drain from a socket buffer.
264  */
265 int
266 sbwait(sb)
267 	struct sockbuf *sb;
268 {
269 
270 	sb->sb_flags |= SB_WAIT;
271 	return (tsleep(&sb->sb_cc,
272 	    (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
273 	    sb->sb_timeo));
274 }
275 
276 /*
277  * Lock a sockbuf already known to be locked;
278  * return any error returned from sleep (EINTR).
279  */
280 int
281 sb_lock(sb)
282 	register struct sockbuf *sb;
283 {
284 	int error;
285 
286 	while (sb->sb_flags & SB_LOCK) {
287 		sb->sb_flags |= SB_WANT;
288 		error = tsleep(&sb->sb_flags,
289 		    (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH,
290 		    "sblock", 0);
291 		if (error)
292 			return (error);
293 	}
294 	sb->sb_flags |= SB_LOCK;
295 	return (0);
296 }
297 
298 /*
299  * Wakeup processes waiting on a socket buffer.
300  * Do asynchronous notification via SIGIO
301  * if the socket has the SS_ASYNC flag set.
302  */
303 void
304 sowakeup(so, sb)
305 	register struct socket *so;
306 	register struct sockbuf *sb;
307 {
308 
309 	selwakeup(&sb->sb_sel);
310 	sb->sb_flags &= ~SB_SEL;
311 	if (sb->sb_flags & SB_WAIT) {
312 		sb->sb_flags &= ~SB_WAIT;
313 		wakeup(&sb->sb_cc);
314 	}
315 	if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
316 		pgsigio(&so->so_sigio, SIGIO, 0);
317 	if (sb->sb_flags & SB_UPCALL)
318 		(*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT);
319 	if (sb->sb_flags & SB_AIO)
320 		aio_swake(so, sb);
321 	KNOTE(&sb->sb_sel.si_note, 0);
322 }
323 
324 /*
325  * Socket buffer (struct sockbuf) utility routines.
326  *
327  * Each socket contains two socket buffers: one for sending data and
328  * one for receiving data.  Each buffer contains a queue of mbufs,
329  * information about the number of mbufs and amount of data in the
330  * queue, and other fields allowing select() statements and notification
331  * on data availability to be implemented.
332  *
333  * Data stored in a socket buffer is maintained as a list of records.
334  * Each record is a list of mbufs chained together with the m_next
335  * field.  Records are chained together with the m_nextpkt field. The upper
336  * level routine soreceive() expects the following conventions to be
337  * observed when placing information in the receive buffer:
338  *
339  * 1. If the protocol requires each message be preceded by the sender's
340  *    name, then a record containing that name must be present before
341  *    any associated data (mbuf's must be of type MT_SONAME).
342  * 2. If the protocol supports the exchange of ``access rights'' (really
343  *    just additional data associated with the message), and there are
344  *    ``rights'' to be received, then a record containing this data
345  *    should be present (mbuf's must be of type MT_RIGHTS).
346  * 3. If a name or rights record exists, then it must be followed by
347  *    a data record, perhaps of zero length.
348  *
349  * Before using a new socket structure it is first necessary to reserve
350  * buffer space to the socket, by calling sbreserve().  This should commit
351  * some of the available buffer space in the system buffer pool for the
352  * socket (currently, it does nothing but enforce limits).  The space
353  * should be released by calling sbrelease() when the socket is destroyed.
354  */
355 
356 int
357 soreserve(so, sndcc, rcvcc)
358 	register struct socket *so;
359 	u_long sndcc, rcvcc;
360 {
361 	struct thread *td = curthread;
362 
363 	if (sbreserve(&so->so_snd, sndcc, so, td) == 0)
364 		goto bad;
365 	if (sbreserve(&so->so_rcv, rcvcc, so, td) == 0)
366 		goto bad2;
367 	if (so->so_rcv.sb_lowat == 0)
368 		so->so_rcv.sb_lowat = 1;
369 	if (so->so_snd.sb_lowat == 0)
370 		so->so_snd.sb_lowat = MCLBYTES;
371 	if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
372 		so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
373 	return (0);
374 bad2:
375 	sbrelease(&so->so_snd, so);
376 bad:
377 	return (ENOBUFS);
378 }
379 
380 static int
381 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS)
382 {
383 	int error = 0;
384 	u_long old_sb_max = sb_max;
385 
386 	error = SYSCTL_OUT(req, arg1, sizeof(u_long));
387 	if (error || !req->newptr)
388 		return (error);
389 	error = SYSCTL_IN(req, arg1, sizeof(u_long));
390 	if (error)
391 		return (error);
392 	if (sb_max < MSIZE + MCLBYTES) {
393 		sb_max = old_sb_max;
394 		return (EINVAL);
395 	}
396 	sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES);
397 	return (0);
398 }
399 
400 /*
401  * Allot mbufs to a sockbuf.
402  * Attempt to scale mbmax so that mbcnt doesn't become limiting
403  * if buffering efficiency is near the normal case.
404  */
405 int
406 sbreserve(sb, cc, so, td)
407 	struct sockbuf *sb;
408 	u_long cc;
409 	struct socket *so;
410 	struct thread *td;
411 {
412 
413 	/*
414 	 * td will only be NULL when we're in an interrupt
415 	 * (e.g. in tcp_input())
416 	 */
417 	if (cc > sb_max_adj)
418 		return (0);
419 	if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
420 	    td ? td->td_proc->p_rlimit[RLIMIT_SBSIZE].rlim_cur : RLIM_INFINITY)) {
421 		return (0);
422 	}
423 	sb->sb_mbmax = min(cc * sb_efficiency, sb_max);
424 	if (sb->sb_lowat > sb->sb_hiwat)
425 		sb->sb_lowat = sb->sb_hiwat;
426 	return (1);
427 }
428 
429 /*
430  * Free mbufs held by a socket, and reserved mbuf space.
431  */
432 void
433 sbrelease(sb, so)
434 	struct sockbuf *sb;
435 	struct socket *so;
436 {
437 
438 	sbflush(sb);
439 	(void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
440 	    RLIM_INFINITY);
441 	sb->sb_mbmax = 0;
442 }
443 
444 /*
445  * Routines to add and remove
446  * data from an mbuf queue.
447  *
448  * The routines sbappend() or sbappendrecord() are normally called to
449  * append new mbufs to a socket buffer, after checking that adequate
450  * space is available, comparing the function sbspace() with the amount
451  * of data to be added.  sbappendrecord() differs from sbappend() in
452  * that data supplied is treated as the beginning of a new record.
453  * To place a sender's address, optional access rights, and data in a
454  * socket receive buffer, sbappendaddr() should be used.  To place
455  * access rights and data in a socket receive buffer, sbappendrights()
456  * should be used.  In either case, the new data begins a new record.
457  * Note that unlike sbappend() and sbappendrecord(), these routines check
458  * for the caller that there will be enough space to store the data.
459  * Each fails if there is not enough space, or if it cannot find mbufs
460  * to store additional information in.
461  *
462  * Reliable protocols may use the socket send buffer to hold data
463  * awaiting acknowledgement.  Data is normally copied from a socket
464  * send buffer in a protocol with m_copy for output to a peer,
465  * and then removing the data from the socket buffer with sbdrop()
466  * or sbdroprecord() when the data is acknowledged by the peer.
467  */
468 
469 /*
470  * Append mbuf chain m to the last record in the
471  * socket buffer sb.  The additional space associated
472  * the mbuf chain is recorded in sb.  Empty mbufs are
473  * discarded and mbufs are compacted where possible.
474  */
475 void
476 sbappend(sb, m)
477 	struct sockbuf *sb;
478 	struct mbuf *m;
479 {
480 	register struct mbuf *n;
481 
482 	if (m == 0)
483 		return;
484 	n = sb->sb_mb;
485 	if (n) {
486 		while (n->m_nextpkt)
487 			n = n->m_nextpkt;
488 		do {
489 			if (n->m_flags & M_EOR) {
490 				sbappendrecord(sb, m); /* XXXXXX!!!! */
491 				return;
492 			}
493 		} while (n->m_next && (n = n->m_next));
494 	}
495 	sbcompress(sb, m, n);
496 }
497 
498 #ifdef SOCKBUF_DEBUG
499 void
500 sbcheck(sb)
501 	struct sockbuf *sb;
502 {
503 	struct mbuf *m;
504 	struct mbuf *n = 0;
505 	u_long len = 0, mbcnt = 0;
506 
507 	for (m = sb->sb_mb; m; m = n) {
508 	    n = m->m_nextpkt;
509 	    for (; m; m = m->m_next) {
510 		len += m->m_len;
511 		mbcnt += MSIZE;
512 		if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
513 			mbcnt += m->m_ext.ext_size;
514 	    }
515 	}
516 	if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
517 		printf("cc %ld != %ld || mbcnt %ld != %ld\n", len, sb->sb_cc,
518 		    mbcnt, sb->sb_mbcnt);
519 		panic("sbcheck");
520 	}
521 }
522 #endif
523 
524 /*
525  * As above, except the mbuf chain
526  * begins a new record.
527  */
528 void
529 sbappendrecord(sb, m0)
530 	register struct sockbuf *sb;
531 	register struct mbuf *m0;
532 {
533 	register struct mbuf *m;
534 
535 	if (m0 == 0)
536 		return;
537 	m = sb->sb_mb;
538 	if (m)
539 		while (m->m_nextpkt)
540 			m = m->m_nextpkt;
541 	/*
542 	 * Put the first mbuf on the queue.
543 	 * Note this permits zero length records.
544 	 */
545 	sballoc(sb, m0);
546 	if (m)
547 		m->m_nextpkt = m0;
548 	else
549 		sb->sb_mb = m0;
550 	m = m0->m_next;
551 	m0->m_next = 0;
552 	if (m && (m0->m_flags & M_EOR)) {
553 		m0->m_flags &= ~M_EOR;
554 		m->m_flags |= M_EOR;
555 	}
556 	sbcompress(sb, m, m0);
557 }
558 
559 /*
560  * As above except that OOB data
561  * is inserted at the beginning of the sockbuf,
562  * but after any other OOB data.
563  */
564 void
565 sbinsertoob(sb, m0)
566 	register struct sockbuf *sb;
567 	register struct mbuf *m0;
568 {
569 	register struct mbuf *m;
570 	register struct mbuf **mp;
571 
572 	if (m0 == 0)
573 		return;
574 	for (mp = &sb->sb_mb; *mp ; mp = &((*mp)->m_nextpkt)) {
575 	    m = *mp;
576 	    again:
577 		switch (m->m_type) {
578 
579 		case MT_OOBDATA:
580 			continue;		/* WANT next train */
581 
582 		case MT_CONTROL:
583 			m = m->m_next;
584 			if (m)
585 				goto again;	/* inspect THIS train further */
586 		}
587 		break;
588 	}
589 	/*
590 	 * Put the first mbuf on the queue.
591 	 * Note this permits zero length records.
592 	 */
593 	sballoc(sb, m0);
594 	m0->m_nextpkt = *mp;
595 	*mp = m0;
596 	m = m0->m_next;
597 	m0->m_next = 0;
598 	if (m && (m0->m_flags & M_EOR)) {
599 		m0->m_flags &= ~M_EOR;
600 		m->m_flags |= M_EOR;
601 	}
602 	sbcompress(sb, m, m0);
603 }
604 
605 /*
606  * Append address and data, and optionally, control (ancillary) data
607  * to the receive queue of a socket.  If present,
608  * m0 must include a packet header with total length.
609  * Returns 0 if no space in sockbuf or insufficient mbufs.
610  */
611 int
612 sbappendaddr(sb, asa, m0, control)
613 	struct sockbuf *sb;
614 	struct sockaddr *asa;
615 	struct mbuf *m0, *control;
616 {
617 	struct mbuf *m, *n;
618 	int space = asa->sa_len;
619 
620 	if (m0 && (m0->m_flags & M_PKTHDR) == 0)
621 		panic("sbappendaddr");
622 	if (m0)
623 		space += m0->m_pkthdr.len;
624 	space += m_length(control, &n);
625 	if (space > sbspace(sb))
626 		return (0);
627 	if (asa->sa_len > MLEN)
628 		return (0);
629 	MGET(m, M_DONTWAIT, MT_SONAME);
630 	if (m == 0)
631 		return (0);
632 	m->m_len = asa->sa_len;
633 	bcopy(asa, mtod(m, caddr_t), asa->sa_len);
634 	if (n)
635 		n->m_next = m0;		/* concatenate data to control */
636 	else
637 		control = m0;
638 	m->m_next = control;
639 	for (n = m; n; n = n->m_next)
640 		sballoc(sb, n);
641 	n = sb->sb_mb;
642 	if (n) {
643 		while (n->m_nextpkt)
644 			n = n->m_nextpkt;
645 		n->m_nextpkt = m;
646 	} else
647 		sb->sb_mb = m;
648 	return (1);
649 }
650 
651 int
652 sbappendcontrol(sb, m0, control)
653 	struct sockbuf *sb;
654 	struct mbuf *control, *m0;
655 {
656 	struct mbuf *m, *n;
657 	int space;
658 
659 	if (control == 0)
660 		panic("sbappendcontrol");
661 	space = m_length(control, &n) + m_length(m0, NULL);
662 	if (space > sbspace(sb))
663 		return (0);
664 	n->m_next = m0;			/* concatenate data to control */
665 	for (m = control; m; m = m->m_next)
666 		sballoc(sb, m);
667 	n = sb->sb_mb;
668 	if (n) {
669 		while (n->m_nextpkt)
670 			n = n->m_nextpkt;
671 		n->m_nextpkt = control;
672 	} else
673 		sb->sb_mb = control;
674 	return (1);
675 }
676 
677 /*
678  * Compress mbuf chain m into the socket
679  * buffer sb following mbuf n.  If n
680  * is null, the buffer is presumed empty.
681  */
682 void
683 sbcompress(sb, m, n)
684 	register struct sockbuf *sb;
685 	register struct mbuf *m, *n;
686 {
687 	register int eor = 0;
688 	register struct mbuf *o;
689 
690 	while (m) {
691 		eor |= m->m_flags & M_EOR;
692 		if (m->m_len == 0 &&
693 		    (eor == 0 ||
694 		     (((o = m->m_next) || (o = n)) &&
695 		      o->m_type == m->m_type))) {
696 			m = m_free(m);
697 			continue;
698 		}
699 		if (n && (n->m_flags & M_EOR) == 0 &&
700 		    M_WRITABLE(n) &&
701 		    m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
702 		    m->m_len <= M_TRAILINGSPACE(n) &&
703 		    n->m_type == m->m_type) {
704 			bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
705 			    (unsigned)m->m_len);
706 			n->m_len += m->m_len;
707 			sb->sb_cc += m->m_len;
708 			if (m->m_type != MT_DATA && m->m_type != MT_HEADER &&
709 			    m->m_type != MT_OOBDATA)
710 				/* XXX: Probably don't need.*/
711 				sb->sb_ctl += m->m_len;
712 			m = m_free(m);
713 			continue;
714 		}
715 		if (n)
716 			n->m_next = m;
717 		else
718 			sb->sb_mb = m;
719 		sballoc(sb, m);
720 		n = m;
721 		m->m_flags &= ~M_EOR;
722 		m = m->m_next;
723 		n->m_next = 0;
724 	}
725 	if (eor) {
726 		if (n)
727 			n->m_flags |= eor;
728 		else
729 			printf("semi-panic: sbcompress\n");
730 	}
731 }
732 
733 /*
734  * Free all mbufs in a sockbuf.
735  * Check that all resources are reclaimed.
736  */
737 void
738 sbflush(sb)
739 	register struct sockbuf *sb;
740 {
741 
742 	if (sb->sb_flags & SB_LOCK)
743 		panic("sbflush: locked");
744 	while (sb->sb_mbcnt) {
745 		/*
746 		 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty:
747 		 * we would loop forever. Panic instead.
748 		 */
749 		if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len))
750 			break;
751 		sbdrop(sb, (int)sb->sb_cc);
752 	}
753 	if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt)
754 		panic("sbflush: cc %u || mb %p || mbcnt %u", sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt);
755 }
756 
757 /*
758  * Drop data from (the front of) a sockbuf.
759  */
760 void
761 sbdrop(sb, len)
762 	register struct sockbuf *sb;
763 	register int len;
764 {
765 	register struct mbuf *m;
766 	struct mbuf *next;
767 
768 	next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
769 	while (len > 0) {
770 		if (m == 0) {
771 			if (next == 0)
772 				panic("sbdrop");
773 			m = next;
774 			next = m->m_nextpkt;
775 			continue;
776 		}
777 		if (m->m_len > len) {
778 			m->m_len -= len;
779 			m->m_data += len;
780 			sb->sb_cc -= len;
781 			if (m->m_type != MT_DATA && m->m_type != MT_HEADER &&
782 			    m->m_type != MT_OOBDATA)
783 				sb->sb_ctl -= len;
784 			break;
785 		}
786 		len -= m->m_len;
787 		sbfree(sb, m);
788 		m = m_free(m);
789 	}
790 	while (m && m->m_len == 0) {
791 		sbfree(sb, m);
792 		m = m_free(m);
793 	}
794 	if (m) {
795 		sb->sb_mb = m;
796 		m->m_nextpkt = next;
797 	} else
798 		sb->sb_mb = next;
799 }
800 
801 /*
802  * Drop a record off the front of a sockbuf
803  * and move the next record to the front.
804  */
805 void
806 sbdroprecord(sb)
807 	register struct sockbuf *sb;
808 {
809 	register struct mbuf *m;
810 
811 	m = sb->sb_mb;
812 	if (m) {
813 		sb->sb_mb = m->m_nextpkt;
814 		do {
815 			sbfree(sb, m);
816 			m = m_free(m);
817 		} while (m);
818 	}
819 }
820 
821 /*
822  * Create a "control" mbuf containing the specified data
823  * with the specified type for presentation on a socket buffer.
824  */
825 struct mbuf *
826 sbcreatecontrol(p, size, type, level)
827 	caddr_t p;
828 	register int size;
829 	int type, level;
830 {
831 	register struct cmsghdr *cp;
832 	struct mbuf *m;
833 
834 	if (CMSG_SPACE((u_int)size) > MCLBYTES)
835 		return ((struct mbuf *) NULL);
836 	if ((m = m_get(M_DONTWAIT, MT_CONTROL)) == NULL)
837 		return ((struct mbuf *) NULL);
838 	if (CMSG_SPACE((u_int)size) > MLEN) {
839 		MCLGET(m, M_DONTWAIT);
840 		if ((m->m_flags & M_EXT) == 0) {
841 			m_free(m);
842 			return ((struct mbuf *) NULL);
843 		}
844 	}
845 	cp = mtod(m, struct cmsghdr *);
846 	m->m_len = 0;
847 	KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m),
848 	    ("sbcreatecontrol: short mbuf"));
849 	if (p != NULL)
850 		(void)memcpy(CMSG_DATA(cp), p, size);
851 	m->m_len = CMSG_SPACE(size);
852 	cp->cmsg_len = CMSG_LEN(size);
853 	cp->cmsg_level = level;
854 	cp->cmsg_type = type;
855 	return (m);
856 }
857 
858 /*
859  * Some routines that return EOPNOTSUPP for entry points that are not
860  * supported by a protocol.  Fill in as needed.
861  */
862 int
863 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
864 {
865 	return EOPNOTSUPP;
866 }
867 
868 int
869 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
870 {
871 	return EOPNOTSUPP;
872 }
873 
874 int
875 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
876 {
877 	return EOPNOTSUPP;
878 }
879 
880 int
881 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
882 		    struct ifnet *ifp, struct thread *td)
883 {
884 	return EOPNOTSUPP;
885 }
886 
887 int
888 pru_listen_notsupp(struct socket *so, struct thread *td)
889 {
890 	return EOPNOTSUPP;
891 }
892 
893 int
894 pru_rcvd_notsupp(struct socket *so, int flags)
895 {
896 	return EOPNOTSUPP;
897 }
898 
899 int
900 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
901 {
902 	return EOPNOTSUPP;
903 }
904 
905 /*
906  * This isn't really a ``null'' operation, but it's the default one
907  * and doesn't do anything destructive.
908  */
909 int
910 pru_sense_null(struct socket *so, struct stat *sb)
911 {
912 	sb->st_blksize = so->so_snd.sb_hiwat;
913 	return 0;
914 }
915 
916 /*
917  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
918  */
919 struct sockaddr *
920 dup_sockaddr(sa, canwait)
921 	struct sockaddr *sa;
922 	int canwait;
923 {
924 	struct sockaddr *sa2;
925 
926 	MALLOC(sa2, struct sockaddr *, sa->sa_len, M_SONAME,
927 	       canwait ? M_WAITOK : M_NOWAIT);
928 	if (sa2)
929 		bcopy(sa, sa2, sa->sa_len);
930 	return sa2;
931 }
932 
933 /*
934  * Create an external-format (``xsocket'') structure using the information
935  * in the kernel-format socket structure pointed to by so.  This is done
936  * to reduce the spew of irrelevant information over this interface,
937  * to isolate user code from changes in the kernel structure, and
938  * potentially to provide information-hiding if we decide that
939  * some of this information should be hidden from users.
940  */
941 void
942 sotoxsocket(struct socket *so, struct xsocket *xso)
943 {
944 	xso->xso_len = sizeof *xso;
945 	xso->xso_so = so;
946 	xso->so_type = so->so_type;
947 	xso->so_options = so->so_options;
948 	xso->so_linger = so->so_linger;
949 	xso->so_state = so->so_state;
950 	xso->so_pcb = so->so_pcb;
951 	xso->xso_protocol = so->so_proto->pr_protocol;
952 	xso->xso_family = so->so_proto->pr_domain->dom_family;
953 	xso->so_qlen = so->so_qlen;
954 	xso->so_incqlen = so->so_incqlen;
955 	xso->so_qlimit = so->so_qlimit;
956 	xso->so_timeo = so->so_timeo;
957 	xso->so_error = so->so_error;
958 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
959 	xso->so_oobmark = so->so_oobmark;
960 	sbtoxsockbuf(&so->so_snd, &xso->so_snd);
961 	sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
962 	xso->so_uid = so->so_cred->cr_uid;
963 }
964 
965 /*
966  * This does the same for sockbufs.  Note that the xsockbuf structure,
967  * since it is always embedded in a socket, does not include a self
968  * pointer nor a length.  We make this entry point public in case
969  * some other mechanism needs it.
970  */
971 void
972 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb)
973 {
974 	xsb->sb_cc = sb->sb_cc;
975 	xsb->sb_hiwat = sb->sb_hiwat;
976 	xsb->sb_mbcnt = sb->sb_mbcnt;
977 	xsb->sb_mbmax = sb->sb_mbmax;
978 	xsb->sb_lowat = sb->sb_lowat;
979 	xsb->sb_flags = sb->sb_flags;
980 	xsb->sb_timeo = sb->sb_timeo;
981 }
982 
983 /*
984  * Here is the definition of some of the basic objects in the kern.ipc
985  * branch of the MIB.
986  */
987 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
988 
989 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
990 static int dummy;
991 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
992 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_ULONG|CTLFLAG_RW,
993     &sb_max, 0, sysctl_handle_sb_max, "LU", "Maximum socket buffer size");
994 SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RD,
995     &maxsockets, 0, "Maximum number of sockets avaliable");
996 SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
997     &sb_efficiency, 0, "");
998 
999 /*
1000  * Initialise maxsockets
1001  */
1002 static void init_maxsockets(void *ignored)
1003 {
1004 	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
1005 	maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
1006 }
1007 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
1008