xref: /freebsd/sys/kern/uipc_sockbuf.c (revision 7660b554bc59a07be0431c17e0e33815818baa69)
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  */
35 
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include "opt_mac.h"
40 #include "opt_param.h"
41 
42 #include <sys/param.h>
43 #include <sys/aio.h> /* for aio_swake proto */
44 #include <sys/domain.h>
45 #include <sys/event.h>
46 #include <sys/file.h>	/* for maxfiles */
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/mac.h>
50 #include <sys/malloc.h>
51 #include <sys/mbuf.h>
52 #include <sys/mutex.h>
53 #include <sys/proc.h>
54 #include <sys/protosw.h>
55 #include <sys/resourcevar.h>
56 #include <sys/signalvar.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/stat.h>
60 #include <sys/sysctl.h>
61 #include <sys/systm.h>
62 
63 int	maxsockets;
64 
65 void (*aio_swake)(struct socket *, struct sockbuf *);
66 
67 /*
68  * Primitive routines for operating on sockets and socket buffers
69  */
70 
71 u_long	sb_max = SB_MAX;
72 static	u_long sb_max_adj =
73     SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */
74 
75 static	u_long sb_efficiency = 8;	/* parameter for sbreserve() */
76 
77 /*
78  * Procedures to manipulate state flags of socket
79  * and do appropriate wakeups.  Normal sequence from the
80  * active (originating) side is that soisconnecting() is
81  * called during processing of connect() call,
82  * resulting in an eventual call to soisconnected() if/when the
83  * connection is established.  When the connection is torn down
84  * soisdisconnecting() is called during processing of disconnect() call,
85  * and soisdisconnected() is called when the connection to the peer
86  * is totally severed.  The semantics of these routines are such that
87  * connectionless protocols can call soisconnected() and soisdisconnected()
88  * only, bypassing the in-progress calls when setting up a ``connection''
89  * takes no time.
90  *
91  * From the passive side, a socket is created with
92  * two queues of sockets: so_incomp for connections in progress
93  * and so_comp for connections already made and awaiting user acceptance.
94  * As a protocol is preparing incoming connections, it creates a socket
95  * structure queued on so_incomp by calling sonewconn().  When the connection
96  * is established, soisconnected() is called, and transfers the
97  * socket structure to so_comp, making it available to accept().
98  *
99  * If a socket is closed with sockets on either
100  * so_incomp or so_comp, these sockets are dropped.
101  *
102  * If higher level protocols are implemented in
103  * the kernel, the wakeups done here will sometimes
104  * cause software-interrupt process scheduling.
105  */
106 
107 void
108 soisconnecting(so)
109 	register struct socket *so;
110 {
111 
112 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
113 	so->so_state |= SS_ISCONNECTING;
114 }
115 
116 void
117 soisconnected(so)
118 	struct socket *so;
119 {
120 	struct socket *head = so->so_head;
121 
122 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
123 	so->so_state |= SS_ISCONNECTED;
124 	if (head && (so->so_state & SS_INCOMP)) {
125 		if ((so->so_options & SO_ACCEPTFILTER) != 0) {
126 			so->so_upcall = head->so_accf->so_accept_filter->accf_callback;
127 			so->so_upcallarg = head->so_accf->so_accept_filter_arg;
128 			so->so_rcv.sb_flags |= SB_UPCALL;
129 			so->so_options &= ~SO_ACCEPTFILTER;
130 			so->so_upcall(so, so->so_upcallarg, M_TRYWAIT);
131 			return;
132 		}
133 		TAILQ_REMOVE(&head->so_incomp, so, so_list);
134 		head->so_incqlen--;
135 		so->so_state &= ~SS_INCOMP;
136 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
137 		head->so_qlen++;
138 		so->so_state |= SS_COMP;
139 		sorwakeup(head);
140 		wakeup_one(&head->so_timeo);
141 	} else {
142 		wakeup(&so->so_timeo);
143 		sorwakeup(so);
144 		sowwakeup(so);
145 	}
146 }
147 
148 void
149 soisdisconnecting(so)
150 	register struct socket *so;
151 {
152 
153 	so->so_state &= ~SS_ISCONNECTING;
154 	so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
155 	wakeup(&so->so_timeo);
156 	sowwakeup(so);
157 	sorwakeup(so);
158 }
159 
160 void
161 soisdisconnected(so)
162 	register struct socket *so;
163 {
164 
165 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
166 	so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED);
167 	wakeup(&so->so_timeo);
168 	sbdrop(&so->so_snd, so->so_snd.sb_cc);
169 	sowwakeup(so);
170 	sorwakeup(so);
171 }
172 
173 /*
174  * When an attempt at a new connection is noted on a socket
175  * which accepts connections, sonewconn is called.  If the
176  * connection is possible (subject to space constraints, etc.)
177  * then we allocate a new structure, propoerly linked into the
178  * data structure of the original socket, and return this.
179  * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
180  *
181  * note: the ref count on the socket is 0 on return
182  */
183 struct socket *
184 sonewconn(head, connstatus)
185 	register struct socket *head;
186 	int connstatus;
187 {
188 	register struct socket *so;
189 
190 	if (head->so_qlen > 3 * head->so_qlimit / 2)
191 		return ((struct socket *)0);
192 	so = soalloc(0);
193 	if (so == NULL)
194 		return ((struct socket *)0);
195 	if ((head->so_options & SO_ACCEPTFILTER) != 0)
196 		connstatus = 0;
197 	so->so_head = head;
198 	so->so_type = head->so_type;
199 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
200 	so->so_linger = head->so_linger;
201 	so->so_state = head->so_state | SS_NOFDREF;
202 	so->so_proto = head->so_proto;
203 	so->so_timeo = head->so_timeo;
204 	so->so_cred = crhold(head->so_cred);
205 #ifdef MAC
206 	mac_create_socket_from_socket(head, so);
207 #endif
208 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
209 	    (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
210 		sodealloc(so);
211 		return ((struct socket *)0);
212 	}
213 
214 	if (connstatus) {
215 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
216 		so->so_state |= SS_COMP;
217 		head->so_qlen++;
218 	} else {
219 		if (head->so_incqlen > head->so_qlimit) {
220 			struct socket *sp;
221 			sp = TAILQ_FIRST(&head->so_incomp);
222 			(void) soabort(sp);
223 		}
224 		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
225 		so->so_state |= SS_INCOMP;
226 		head->so_incqlen++;
227 	}
228 	if (connstatus) {
229 		sorwakeup(head);
230 		wakeup(&head->so_timeo);
231 		so->so_state |= connstatus;
232 	}
233 	return (so);
234 }
235 
236 /*
237  * Socantsendmore indicates that no more data will be sent on the
238  * socket; it would normally be applied to a socket when the user
239  * informs the system that no more data is to be sent, by the protocol
240  * code (in case PRU_SHUTDOWN).  Socantrcvmore indicates that no more data
241  * will be received, and will normally be applied to the socket by a
242  * protocol when it detects that the peer will send no more data.
243  * Data queued for reading in the socket may yet be read.
244  */
245 
246 void
247 socantsendmore(so)
248 	struct socket *so;
249 {
250 
251 	so->so_state |= SS_CANTSENDMORE;
252 	sowwakeup(so);
253 }
254 
255 void
256 socantrcvmore(so)
257 	struct socket *so;
258 {
259 
260 	so->so_state |= SS_CANTRCVMORE;
261 	sorwakeup(so);
262 }
263 
264 /*
265  * Wait for data to arrive at/drain from a socket buffer.
266  */
267 int
268 sbwait(sb)
269 	struct sockbuf *sb;
270 {
271 
272 	sb->sb_flags |= SB_WAIT;
273 	return (tsleep(&sb->sb_cc,
274 	    (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
275 	    sb->sb_timeo));
276 }
277 
278 /*
279  * Lock a sockbuf already known to be locked;
280  * return any error returned from sleep (EINTR).
281  */
282 int
283 sb_lock(sb)
284 	register struct sockbuf *sb;
285 {
286 	int error;
287 
288 	while (sb->sb_flags & SB_LOCK) {
289 		sb->sb_flags |= SB_WANT;
290 		error = tsleep(&sb->sb_flags,
291 		    (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH,
292 		    "sblock", 0);
293 		if (error)
294 			return (error);
295 	}
296 	sb->sb_flags |= SB_LOCK;
297 	return (0);
298 }
299 
300 /*
301  * Wakeup processes waiting on a socket buffer.
302  * Do asynchronous notification via SIGIO
303  * if the socket has the SS_ASYNC flag set.
304  */
305 void
306 sowakeup(so, sb)
307 	register struct socket *so;
308 	register struct sockbuf *sb;
309 {
310 
311 	selwakeup(&sb->sb_sel);
312 	sb->sb_flags &= ~SB_SEL;
313 	if (sb->sb_flags & SB_WAIT) {
314 		sb->sb_flags &= ~SB_WAIT;
315 		wakeup(&sb->sb_cc);
316 	}
317 	if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
318 		pgsigio(&so->so_sigio, SIGIO, 0);
319 	if (sb->sb_flags & SB_UPCALL)
320 		(*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT);
321 	if (sb->sb_flags & SB_AIO)
322 		aio_swake(so, sb);
323 	KNOTE(&sb->sb_sel.si_note, 0);
324 }
325 
326 /*
327  * Socket buffer (struct sockbuf) utility routines.
328  *
329  * Each socket contains two socket buffers: one for sending data and
330  * one for receiving data.  Each buffer contains a queue of mbufs,
331  * information about the number of mbufs and amount of data in the
332  * queue, and other fields allowing select() statements and notification
333  * on data availability to be implemented.
334  *
335  * Data stored in a socket buffer is maintained as a list of records.
336  * Each record is a list of mbufs chained together with the m_next
337  * field.  Records are chained together with the m_nextpkt field. The upper
338  * level routine soreceive() expects the following conventions to be
339  * observed when placing information in the receive buffer:
340  *
341  * 1. If the protocol requires each message be preceded by the sender's
342  *    name, then a record containing that name must be present before
343  *    any associated data (mbuf's must be of type MT_SONAME).
344  * 2. If the protocol supports the exchange of ``access rights'' (really
345  *    just additional data associated with the message), and there are
346  *    ``rights'' to be received, then a record containing this data
347  *    should be present (mbuf's must be of type MT_RIGHTS).
348  * 3. If a name or rights record exists, then it must be followed by
349  *    a data record, perhaps of zero length.
350  *
351  * Before using a new socket structure it is first necessary to reserve
352  * buffer space to the socket, by calling sbreserve().  This should commit
353  * some of the available buffer space in the system buffer pool for the
354  * socket (currently, it does nothing but enforce limits).  The space
355  * should be released by calling sbrelease() when the socket is destroyed.
356  */
357 
358 int
359 soreserve(so, sndcc, rcvcc)
360 	register struct socket *so;
361 	u_long sndcc, rcvcc;
362 {
363 	struct thread *td = curthread;
364 
365 	if (sbreserve(&so->so_snd, sndcc, so, td) == 0)
366 		goto bad;
367 	if (sbreserve(&so->so_rcv, rcvcc, so, td) == 0)
368 		goto bad2;
369 	if (so->so_rcv.sb_lowat == 0)
370 		so->so_rcv.sb_lowat = 1;
371 	if (so->so_snd.sb_lowat == 0)
372 		so->so_snd.sb_lowat = MCLBYTES;
373 	if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
374 		so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
375 	return (0);
376 bad2:
377 	sbrelease(&so->so_snd, so);
378 bad:
379 	return (ENOBUFS);
380 }
381 
382 static int
383 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS)
384 {
385 	int error = 0;
386 	u_long old_sb_max = sb_max;
387 
388 	error = SYSCTL_OUT(req, arg1, sizeof(u_long));
389 	if (error || !req->newptr)
390 		return (error);
391 	error = SYSCTL_IN(req, arg1, sizeof(u_long));
392 	if (error)
393 		return (error);
394 	if (sb_max < MSIZE + MCLBYTES) {
395 		sb_max = old_sb_max;
396 		return (EINVAL);
397 	}
398 	sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES);
399 	return (0);
400 }
401 
402 /*
403  * Allot mbufs to a sockbuf.
404  * Attempt to scale mbmax so that mbcnt doesn't become limiting
405  * if buffering efficiency is near the normal case.
406  */
407 int
408 sbreserve(sb, cc, so, td)
409 	struct sockbuf *sb;
410 	u_long cc;
411 	struct socket *so;
412 	struct thread *td;
413 {
414 
415 	/*
416 	 * td will only be NULL when we're in an interrupt
417 	 * (e.g. in tcp_input())
418 	 */
419 	if (cc > sb_max_adj)
420 		return (0);
421 	if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
422 	    td ? td->td_proc->p_rlimit[RLIMIT_SBSIZE].rlim_cur : RLIM_INFINITY)) {
423 		return (0);
424 	}
425 	sb->sb_mbmax = min(cc * sb_efficiency, sb_max);
426 	if (sb->sb_lowat > sb->sb_hiwat)
427 		sb->sb_lowat = sb->sb_hiwat;
428 	return (1);
429 }
430 
431 /*
432  * Free mbufs held by a socket, and reserved mbuf space.
433  */
434 void
435 sbrelease(sb, so)
436 	struct sockbuf *sb;
437 	struct socket *so;
438 {
439 
440 	sbflush(sb);
441 	(void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
442 	    RLIM_INFINITY);
443 	sb->sb_mbmax = 0;
444 }
445 
446 /*
447  * Routines to add and remove
448  * data from an mbuf queue.
449  *
450  * The routines sbappend() or sbappendrecord() are normally called to
451  * append new mbufs to a socket buffer, after checking that adequate
452  * space is available, comparing the function sbspace() with the amount
453  * of data to be added.  sbappendrecord() differs from sbappend() in
454  * that data supplied is treated as the beginning of a new record.
455  * To place a sender's address, optional access rights, and data in a
456  * socket receive buffer, sbappendaddr() should be used.  To place
457  * access rights and data in a socket receive buffer, sbappendrights()
458  * should be used.  In either case, the new data begins a new record.
459  * Note that unlike sbappend() and sbappendrecord(), these routines check
460  * for the caller that there will be enough space to store the data.
461  * Each fails if there is not enough space, or if it cannot find mbufs
462  * to store additional information in.
463  *
464  * Reliable protocols may use the socket send buffer to hold data
465  * awaiting acknowledgement.  Data is normally copied from a socket
466  * send buffer in a protocol with m_copy for output to a peer,
467  * and then removing the data from the socket buffer with sbdrop()
468  * or sbdroprecord() when the data is acknowledged by the peer.
469  */
470 
471 /*
472  * Append mbuf chain m to the last record in the
473  * socket buffer sb.  The additional space associated
474  * the mbuf chain is recorded in sb.  Empty mbufs are
475  * discarded and mbufs are compacted where possible.
476  */
477 void
478 sbappend(sb, m)
479 	struct sockbuf *sb;
480 	struct mbuf *m;
481 {
482 	register struct mbuf *n;
483 
484 	if (m == 0)
485 		return;
486 	n = sb->sb_mb;
487 	if (n) {
488 		while (n->m_nextpkt)
489 			n = n->m_nextpkt;
490 		do {
491 			if (n->m_flags & M_EOR) {
492 				sbappendrecord(sb, m); /* XXXXXX!!!! */
493 				return;
494 			}
495 		} while (n->m_next && (n = n->m_next));
496 	}
497 	sbcompress(sb, m, n);
498 }
499 
500 #ifdef SOCKBUF_DEBUG
501 void
502 sbcheck(sb)
503 	struct sockbuf *sb;
504 {
505 	struct mbuf *m;
506 	struct mbuf *n = 0;
507 	u_long len = 0, mbcnt = 0;
508 
509 	for (m = sb->sb_mb; m; m = n) {
510 	    n = m->m_nextpkt;
511 	    for (; m; m = m->m_next) {
512 		len += m->m_len;
513 		mbcnt += MSIZE;
514 		if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
515 			mbcnt += m->m_ext.ext_size;
516 	    }
517 	}
518 	if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
519 		printf("cc %ld != %ld || mbcnt %ld != %ld\n", len, sb->sb_cc,
520 		    mbcnt, sb->sb_mbcnt);
521 		panic("sbcheck");
522 	}
523 }
524 #endif
525 
526 /*
527  * As above, except the mbuf chain
528  * begins a new record.
529  */
530 void
531 sbappendrecord(sb, m0)
532 	register struct sockbuf *sb;
533 	register struct mbuf *m0;
534 {
535 	register struct mbuf *m;
536 
537 	if (m0 == 0)
538 		return;
539 	m = sb->sb_mb;
540 	if (m)
541 		while (m->m_nextpkt)
542 			m = m->m_nextpkt;
543 	/*
544 	 * Put the first mbuf on the queue.
545 	 * Note this permits zero length records.
546 	 */
547 	sballoc(sb, m0);
548 	if (m)
549 		m->m_nextpkt = m0;
550 	else
551 		sb->sb_mb = m0;
552 	m = m0->m_next;
553 	m0->m_next = 0;
554 	if (m && (m0->m_flags & M_EOR)) {
555 		m0->m_flags &= ~M_EOR;
556 		m->m_flags |= M_EOR;
557 	}
558 	sbcompress(sb, m, m0);
559 }
560 
561 /*
562  * As above except that OOB data
563  * is inserted at the beginning of the sockbuf,
564  * but after any other OOB data.
565  */
566 void
567 sbinsertoob(sb, m0)
568 	register struct sockbuf *sb;
569 	register struct mbuf *m0;
570 {
571 	register struct mbuf *m;
572 	register struct mbuf **mp;
573 
574 	if (m0 == 0)
575 		return;
576 	for (mp = &sb->sb_mb; *mp ; mp = &((*mp)->m_nextpkt)) {
577 	    m = *mp;
578 	    again:
579 		switch (m->m_type) {
580 
581 		case MT_OOBDATA:
582 			continue;		/* WANT next train */
583 
584 		case MT_CONTROL:
585 			m = m->m_next;
586 			if (m)
587 				goto again;	/* inspect THIS train further */
588 		}
589 		break;
590 	}
591 	/*
592 	 * Put the first mbuf on the queue.
593 	 * Note this permits zero length records.
594 	 */
595 	sballoc(sb, m0);
596 	m0->m_nextpkt = *mp;
597 	*mp = m0;
598 	m = m0->m_next;
599 	m0->m_next = 0;
600 	if (m && (m0->m_flags & M_EOR)) {
601 		m0->m_flags &= ~M_EOR;
602 		m->m_flags |= M_EOR;
603 	}
604 	sbcompress(sb, m, m0);
605 }
606 
607 /*
608  * Append address and data, and optionally, control (ancillary) data
609  * to the receive queue of a socket.  If present,
610  * m0 must include a packet header with total length.
611  * Returns 0 if no space in sockbuf or insufficient mbufs.
612  */
613 int
614 sbappendaddr(sb, asa, m0, control)
615 	struct sockbuf *sb;
616 	struct sockaddr *asa;
617 	struct mbuf *m0, *control;
618 {
619 	struct mbuf *m, *n;
620 	int space = asa->sa_len;
621 
622 	if (m0 && (m0->m_flags & M_PKTHDR) == 0)
623 		panic("sbappendaddr");
624 	if (m0)
625 		space += m0->m_pkthdr.len;
626 	space += m_length(control, &n);
627 	if (space > sbspace(sb))
628 		return (0);
629 #if MSIZE <= 256
630 	if (asa->sa_len > MLEN)
631 		return (0);
632 #endif
633 	MGET(m, M_DONTWAIT, MT_SONAME);
634 	if (m == 0)
635 		return (0);
636 	m->m_len = asa->sa_len;
637 	bcopy(asa, mtod(m, caddr_t), asa->sa_len);
638 	if (n)
639 		n->m_next = m0;		/* concatenate data to control */
640 	else
641 		control = m0;
642 	m->m_next = control;
643 	for (n = m; n; n = n->m_next)
644 		sballoc(sb, n);
645 	n = sb->sb_mb;
646 	if (n) {
647 		while (n->m_nextpkt)
648 			n = n->m_nextpkt;
649 		n->m_nextpkt = m;
650 	} else
651 		sb->sb_mb = m;
652 	return (1);
653 }
654 
655 int
656 sbappendcontrol(sb, m0, control)
657 	struct sockbuf *sb;
658 	struct mbuf *control, *m0;
659 {
660 	struct mbuf *m, *n;
661 	int space;
662 
663 	if (control == 0)
664 		panic("sbappendcontrol");
665 	space = m_length(control, &n) + m_length(m0, NULL);
666 	if (space > sbspace(sb))
667 		return (0);
668 	n->m_next = m0;			/* concatenate data to control */
669 	for (m = control; m; m = m->m_next)
670 		sballoc(sb, m);
671 	n = sb->sb_mb;
672 	if (n) {
673 		while (n->m_nextpkt)
674 			n = n->m_nextpkt;
675 		n->m_nextpkt = control;
676 	} else
677 		sb->sb_mb = control;
678 	return (1);
679 }
680 
681 /*
682  * Compress mbuf chain m into the socket
683  * buffer sb following mbuf n.  If n
684  * is null, the buffer is presumed empty.
685  */
686 void
687 sbcompress(sb, m, n)
688 	register struct sockbuf *sb;
689 	register struct mbuf *m, *n;
690 {
691 	register int eor = 0;
692 	register struct mbuf *o;
693 
694 	while (m) {
695 		eor |= m->m_flags & M_EOR;
696 		if (m->m_len == 0 &&
697 		    (eor == 0 ||
698 		     (((o = m->m_next) || (o = n)) &&
699 		      o->m_type == m->m_type))) {
700 			m = m_free(m);
701 			continue;
702 		}
703 		if (n && (n->m_flags & M_EOR) == 0 &&
704 		    M_WRITABLE(n) &&
705 		    m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
706 		    m->m_len <= M_TRAILINGSPACE(n) &&
707 		    n->m_type == m->m_type) {
708 			bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
709 			    (unsigned)m->m_len);
710 			n->m_len += m->m_len;
711 			sb->sb_cc += m->m_len;
712 			if (m->m_type != MT_DATA && m->m_type != MT_HEADER &&
713 			    m->m_type != MT_OOBDATA)
714 				/* XXX: Probably don't need.*/
715 				sb->sb_ctl += m->m_len;
716 			m = m_free(m);
717 			continue;
718 		}
719 		if (n)
720 			n->m_next = m;
721 		else
722 			sb->sb_mb = m;
723 		sballoc(sb, m);
724 		n = m;
725 		m->m_flags &= ~M_EOR;
726 		m = m->m_next;
727 		n->m_next = 0;
728 	}
729 	if (eor) {
730 		if (n)
731 			n->m_flags |= eor;
732 		else
733 			printf("semi-panic: sbcompress\n");
734 	}
735 }
736 
737 /*
738  * Free all mbufs in a sockbuf.
739  * Check that all resources are reclaimed.
740  */
741 void
742 sbflush(sb)
743 	register struct sockbuf *sb;
744 {
745 
746 	if (sb->sb_flags & SB_LOCK)
747 		panic("sbflush: locked");
748 	while (sb->sb_mbcnt) {
749 		/*
750 		 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty:
751 		 * we would loop forever. Panic instead.
752 		 */
753 		if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len))
754 			break;
755 		sbdrop(sb, (int)sb->sb_cc);
756 	}
757 	if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt)
758 		panic("sbflush: cc %u || mb %p || mbcnt %u", sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt);
759 }
760 
761 /*
762  * Drop data from (the front of) a sockbuf.
763  */
764 void
765 sbdrop(sb, len)
766 	register struct sockbuf *sb;
767 	register int len;
768 {
769 	register struct mbuf *m;
770 	struct mbuf *next;
771 
772 	next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
773 	while (len > 0) {
774 		if (m == 0) {
775 			if (next == 0)
776 				panic("sbdrop");
777 			m = next;
778 			next = m->m_nextpkt;
779 			continue;
780 		}
781 		if (m->m_len > len) {
782 			m->m_len -= len;
783 			m->m_data += len;
784 			sb->sb_cc -= len;
785 			if (m->m_type != MT_DATA && m->m_type != MT_HEADER &&
786 			    m->m_type != MT_OOBDATA)
787 				sb->sb_ctl -= len;
788 			break;
789 		}
790 		len -= m->m_len;
791 		sbfree(sb, m);
792 		m = m_free(m);
793 	}
794 	while (m && m->m_len == 0) {
795 		sbfree(sb, m);
796 		m = m_free(m);
797 	}
798 	if (m) {
799 		sb->sb_mb = m;
800 		m->m_nextpkt = next;
801 	} else
802 		sb->sb_mb = next;
803 }
804 
805 /*
806  * Drop a record off the front of a sockbuf
807  * and move the next record to the front.
808  */
809 void
810 sbdroprecord(sb)
811 	register struct sockbuf *sb;
812 {
813 	register struct mbuf *m;
814 
815 	m = sb->sb_mb;
816 	if (m) {
817 		sb->sb_mb = m->m_nextpkt;
818 		do {
819 			sbfree(sb, m);
820 			m = m_free(m);
821 		} while (m);
822 	}
823 }
824 
825 /*
826  * Create a "control" mbuf containing the specified data
827  * with the specified type for presentation on a socket buffer.
828  */
829 struct mbuf *
830 sbcreatecontrol(p, size, type, level)
831 	caddr_t p;
832 	register int size;
833 	int type, level;
834 {
835 	register struct cmsghdr *cp;
836 	struct mbuf *m;
837 
838 	if (CMSG_SPACE((u_int)size) > MCLBYTES)
839 		return ((struct mbuf *) NULL);
840 	if ((m = m_get(M_DONTWAIT, MT_CONTROL)) == NULL)
841 		return ((struct mbuf *) NULL);
842 	if (CMSG_SPACE((u_int)size) > MLEN) {
843 		MCLGET(m, M_DONTWAIT);
844 		if ((m->m_flags & M_EXT) == 0) {
845 			m_free(m);
846 			return ((struct mbuf *) NULL);
847 		}
848 	}
849 	cp = mtod(m, struct cmsghdr *);
850 	m->m_len = 0;
851 	KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m),
852 	    ("sbcreatecontrol: short mbuf"));
853 	if (p != NULL)
854 		(void)memcpy(CMSG_DATA(cp), p, size);
855 	m->m_len = CMSG_SPACE(size);
856 	cp->cmsg_len = CMSG_LEN(size);
857 	cp->cmsg_level = level;
858 	cp->cmsg_type = type;
859 	return (m);
860 }
861 
862 /*
863  * Some routines that return EOPNOTSUPP for entry points that are not
864  * supported by a protocol.  Fill in as needed.
865  */
866 int
867 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
868 {
869 	return EOPNOTSUPP;
870 }
871 
872 int
873 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
874 {
875 	return EOPNOTSUPP;
876 }
877 
878 int
879 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
880 {
881 	return EOPNOTSUPP;
882 }
883 
884 int
885 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
886 		    struct ifnet *ifp, struct thread *td)
887 {
888 	return EOPNOTSUPP;
889 }
890 
891 int
892 pru_listen_notsupp(struct socket *so, struct thread *td)
893 {
894 	return EOPNOTSUPP;
895 }
896 
897 int
898 pru_rcvd_notsupp(struct socket *so, int flags)
899 {
900 	return EOPNOTSUPP;
901 }
902 
903 int
904 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
905 {
906 	return EOPNOTSUPP;
907 }
908 
909 /*
910  * This isn't really a ``null'' operation, but it's the default one
911  * and doesn't do anything destructive.
912  */
913 int
914 pru_sense_null(struct socket *so, struct stat *sb)
915 {
916 	sb->st_blksize = so->so_snd.sb_hiwat;
917 	return 0;
918 }
919 
920 /*
921  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
922  */
923 struct sockaddr *
924 dup_sockaddr(sa, canwait)
925 	struct sockaddr *sa;
926 	int canwait;
927 {
928 	struct sockaddr *sa2;
929 
930 	MALLOC(sa2, struct sockaddr *, sa->sa_len, M_SONAME,
931 	       canwait ? M_WAITOK : M_NOWAIT);
932 	if (sa2)
933 		bcopy(sa, sa2, sa->sa_len);
934 	return sa2;
935 }
936 
937 /*
938  * Create an external-format (``xsocket'') structure using the information
939  * in the kernel-format socket structure pointed to by so.  This is done
940  * to reduce the spew of irrelevant information over this interface,
941  * to isolate user code from changes in the kernel structure, and
942  * potentially to provide information-hiding if we decide that
943  * some of this information should be hidden from users.
944  */
945 void
946 sotoxsocket(struct socket *so, struct xsocket *xso)
947 {
948 	xso->xso_len = sizeof *xso;
949 	xso->xso_so = so;
950 	xso->so_type = so->so_type;
951 	xso->so_options = so->so_options;
952 	xso->so_linger = so->so_linger;
953 	xso->so_state = so->so_state;
954 	xso->so_pcb = so->so_pcb;
955 	xso->xso_protocol = so->so_proto->pr_protocol;
956 	xso->xso_family = so->so_proto->pr_domain->dom_family;
957 	xso->so_qlen = so->so_qlen;
958 	xso->so_incqlen = so->so_incqlen;
959 	xso->so_qlimit = so->so_qlimit;
960 	xso->so_timeo = so->so_timeo;
961 	xso->so_error = so->so_error;
962 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
963 	xso->so_oobmark = so->so_oobmark;
964 	sbtoxsockbuf(&so->so_snd, &xso->so_snd);
965 	sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
966 	xso->so_uid = so->so_cred->cr_uid;
967 }
968 
969 /*
970  * This does the same for sockbufs.  Note that the xsockbuf structure,
971  * since it is always embedded in a socket, does not include a self
972  * pointer nor a length.  We make this entry point public in case
973  * some other mechanism needs it.
974  */
975 void
976 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb)
977 {
978 	xsb->sb_cc = sb->sb_cc;
979 	xsb->sb_hiwat = sb->sb_hiwat;
980 	xsb->sb_mbcnt = sb->sb_mbcnt;
981 	xsb->sb_mbmax = sb->sb_mbmax;
982 	xsb->sb_lowat = sb->sb_lowat;
983 	xsb->sb_flags = sb->sb_flags;
984 	xsb->sb_timeo = sb->sb_timeo;
985 }
986 
987 /*
988  * Here is the definition of some of the basic objects in the kern.ipc
989  * branch of the MIB.
990  */
991 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
992 
993 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
994 static int dummy;
995 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
996 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_ULONG|CTLFLAG_RW,
997     &sb_max, 0, sysctl_handle_sb_max, "LU", "Maximum socket buffer size");
998 SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RD,
999     &maxsockets, 0, "Maximum number of sockets avaliable");
1000 SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
1001     &sb_efficiency, 0, "");
1002 
1003 /*
1004  * Initialise maxsockets
1005  */
1006 static void init_maxsockets(void *ignored)
1007 {
1008 	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
1009 	maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
1010 }
1011 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
1012