xref: /csrg-svn/sys/kern/uipc_socket2.c (revision 37727)
1 /*
2  * Copyright (c) 1982, 1986, 1988 Regents of the University of California.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms are permitted
6  * provided that the above copyright notice and this paragraph are
7  * duplicated in all such forms and that any documentation,
8  * advertising materials, and other materials related to such
9  * distribution and use acknowledge that the software was developed
10  * by the University of California, Berkeley.  The name of the
11  * University may not be used to endorse or promote products derived
12  * from this software without specific prior written permission.
13  * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
14  * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
15  * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE.
16  *
17  *	@(#)uipc_socket2.c	7.9 (Berkeley) 05/09/89
18  */
19 
20 #include "param.h"
21 #include "systm.h"
22 #include "user.h"
23 #include "proc.h"
24 #include "file.h"
25 #include "buf.h"
26 #include "malloc.h"
27 #include "mbuf.h"
28 #include "protosw.h"
29 #include "socket.h"
30 #include "socketvar.h"
31 
32 /*
33  * Primitive routines for operating on sockets and socket buffers
34  */
35 
36 /*
37  * Procedures to manipulate state flags of socket
38  * and do appropriate wakeups.  Normal sequence from the
39  * active (originating) side is that soisconnecting() is
40  * called during processing of connect() call,
41  * resulting in an eventual call to soisconnected() if/when the
42  * connection is established.  When the connection is torn down
43  * soisdisconnecting() is called during processing of disconnect() call,
44  * and soisdisconnected() is called when the connection to the peer
45  * is totally severed.  The semantics of these routines are such that
46  * connectionless protocols can call soisconnected() and soisdisconnected()
47  * only, bypassing the in-progress calls when setting up a ``connection''
48  * takes no time.
49  *
50  * From the passive side, a socket is created with
51  * two queues of sockets: so_q0 for connections in progress
52  * and so_q for connections already made and awaiting user acceptance.
53  * As a protocol is preparing incoming connections, it creates a socket
54  * structure queued on so_q0 by calling sonewconn().  When the connection
55  * is established, soisconnected() is called, and transfers the
56  * socket structure to so_q, making it available to accept().
57  *
58  * If a socket is closed with sockets on either
59  * so_q0 or so_q, these sockets are dropped.
60  *
61  * If higher level protocols are implemented in
62  * the kernel, the wakeups done here will sometimes
63  * cause software-interrupt process scheduling.
64  */
65 
66 soisconnecting(so)
67 	register struct socket *so;
68 {
69 
70 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
71 	so->so_state |= SS_ISCONNECTING;
72 	wakeup((caddr_t)&so->so_timeo);
73 }
74 
75 soisconnected(so)
76 	register struct socket *so;
77 {
78 	register struct socket *head = so->so_head;
79 
80 	if (head) {
81 		if (soqremque(so, 0) == 0)
82 			panic("soisconnected");
83 		soqinsque(head, so, 1);
84 		sorwakeup(head);
85 		wakeup((caddr_t)&head->so_timeo);
86 	}
87 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING);
88 	so->so_state |= SS_ISCONNECTED;
89 	wakeup((caddr_t)&so->so_timeo);
90 	sorwakeup(so);
91 	sowwakeup(so);
92 }
93 
94 soisdisconnecting(so)
95 	register struct socket *so;
96 {
97 
98 	so->so_state &= ~SS_ISCONNECTING;
99 	so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
100 	wakeup((caddr_t)&so->so_timeo);
101 	sowwakeup(so);
102 	sorwakeup(so);
103 }
104 
105 soisdisconnected(so)
106 	register struct socket *so;
107 {
108 
109 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
110 	so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE);
111 	wakeup((caddr_t)&so->so_timeo);
112 	sowwakeup(so);
113 	sorwakeup(so);
114 }
115 
116 /*
117  * When an attempt at a new connection is noted on a socket
118  * which accepts connections, sonewconn is called.  If the
119  * connection is possible (subject to space constraints, etc.)
120  * then we allocate a new structure, propoerly linked into the
121  * data structure of the original socket, and return this.
122  */
123 struct socket *
124 sonewconn(head)
125 	register struct socket *head;
126 {
127 	register struct socket *so;
128 
129 	if (head->so_qlen + head->so_q0len > 3 * head->so_qlimit / 2)
130 		return ((struct socket *)0);
131 	MALLOC(so, struct socket *, sizeof(*so), M_SOCKET, M_DONTWAIT);
132 	if (so == NULL)
133 		return ((struct socket *)0);
134 	bzero((caddr_t)so, sizeof(*so));
135 	so->so_type = head->so_type;
136 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
137 	so->so_linger = head->so_linger;
138 	so->so_state = head->so_state | SS_NOFDREF;
139 	so->so_proto = head->so_proto;
140 	so->so_timeo = head->so_timeo;
141 	so->so_pgid = head->so_pgid;
142 	(void) soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat);
143 	soqinsque(head, so, 0);
144 	if ((*so->so_proto->pr_usrreq)(so, PRU_ATTACH,
145 	    (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0)) {
146 		(void) soqremque(so, 0);
147 		(void) free((caddr_t)so, M_SOCKET);
148 		return ((struct socket *)0);
149 	}
150 	return (so);
151 }
152 
153 soqinsque(head, so, q)
154 	register struct socket *head, *so;
155 	int q;
156 {
157 
158 	so->so_head = head;
159 	if (q == 0) {
160 		head->so_q0len++;
161 		so->so_q0 = head->so_q0;
162 		head->so_q0 = so;
163 	} else {
164 		head->so_qlen++;
165 		so->so_q = head->so_q;
166 		head->so_q = so;
167 	}
168 }
169 
170 soqremque(so, q)
171 	register struct socket *so;
172 	int q;
173 {
174 	register struct socket *head, *prev, *next;
175 
176 	head = so->so_head;
177 	prev = head;
178 	for (;;) {
179 		next = q ? prev->so_q : prev->so_q0;
180 		if (next == so)
181 			break;
182 		if (next == head)
183 			return (0);
184 		prev = next;
185 	}
186 	if (q == 0) {
187 		prev->so_q0 = next->so_q0;
188 		head->so_q0len--;
189 	} else {
190 		prev->so_q = next->so_q;
191 		head->so_qlen--;
192 	}
193 	next->so_q0 = next->so_q = 0;
194 	next->so_head = 0;
195 	return (1);
196 }
197 
198 /*
199  * Socantsendmore indicates that no more data will be sent on the
200  * socket; it would normally be applied to a socket when the user
201  * informs the system that no more data is to be sent, by the protocol
202  * code (in case PRU_SHUTDOWN).  Socantrcvmore indicates that no more data
203  * will be received, and will normally be applied to the socket by a
204  * protocol when it detects that the peer will send no more data.
205  * Data queued for reading in the socket may yet be read.
206  */
207 
208 socantsendmore(so)
209 	struct socket *so;
210 {
211 
212 	so->so_state |= SS_CANTSENDMORE;
213 	sowwakeup(so);
214 }
215 
216 socantrcvmore(so)
217 	struct socket *so;
218 {
219 
220 	so->so_state |= SS_CANTRCVMORE;
221 	sorwakeup(so);
222 }
223 
224 /*
225  * Socket select/wakeup routines.
226  */
227 
228 /*
229  * Queue a process for a select on a socket buffer.
230  */
231 sbselqueue(sb)
232 	struct sockbuf *sb;
233 {
234 	struct proc *p;
235 
236 	if ((p = sb->sb_sel) && p->p_wchan == (caddr_t)&selwait)
237 		sb->sb_flags |= SB_COLL;
238 	else
239 		sb->sb_sel = u.u_procp;
240 }
241 
242 /*
243  * Wait for data to arrive at/drain from a socket buffer.
244  */
245 sbwait(sb)
246 	struct sockbuf *sb;
247 {
248 
249 	sb->sb_flags |= SB_WAIT;
250 	sleep((caddr_t)&sb->sb_cc, PZERO+1);
251 }
252 
253 /*
254  * Wakeup processes waiting on a socket buffer.
255  * Do asynchronous notification via SIGIO
256  * if the socket has the SS_ASYNC flag set.
257  */
258 sowakeup(so, sb)
259 	register struct socket *so;
260 	register struct sockbuf *sb;
261 {
262 	struct proc *p;
263 
264 	if (sb->sb_sel) {
265 		selwakeup(sb->sb_sel, sb->sb_flags & SB_COLL);
266 		sb->sb_sel = 0;
267 		sb->sb_flags &= ~SB_COLL;
268 	}
269 	if (sb->sb_flags & SB_WAIT) {
270 		sb->sb_flags &= ~SB_WAIT;
271 		wakeup((caddr_t)&sb->sb_cc);
272 	}
273 	if (so->so_state & SS_ASYNC) {
274 		if (so->so_pgid < 0)
275 			gsignal(-so->so_pgid, SIGIO);
276 		else if (so->so_pgid > 0 && (p = pfind(so->so_pgid)) != 0)
277 			psignal(p, SIGIO);
278 	}
279 }
280 
281 /*
282  * Socket buffer (struct sockbuf) utility routines.
283  *
284  * Each socket contains two socket buffers: one for sending data and
285  * one for receiving data.  Each buffer contains a queue of mbufs,
286  * information about the number of mbufs and amount of data in the
287  * queue, and other fields allowing select() statements and notification
288  * on data availability to be implemented.
289  *
290  * Data stored in a socket buffer is maintained as a list of records.
291  * Each record is a list of mbufs chained together with the m_next
292  * field.  Records are chained together with the m_nextpkt field. The upper
293  * level routine soreceive() expects the following conventions to be
294  * observed when placing information in the receive buffer:
295  *
296  * 1. If the protocol requires each message be preceded by the sender's
297  *    name, then a record containing that name must be present before
298  *    any associated data (mbuf's must be of type MT_SONAME).
299  * 2. If the protocol supports the exchange of ``access rights'' (really
300  *    just additional data associated with the message), and there are
301  *    ``rights'' to be received, then a record containing this data
302  *    should be present (mbuf's must be of type MT_RIGHTS).
303  * 3. If a name or rights record exists, then it must be followed by
304  *    a data record, perhaps of zero length.
305  *
306  * Before using a new socket structure it is first necessary to reserve
307  * buffer space to the socket, by calling sbreserve().  This should commit
308  * some of the available buffer space in the system buffer pool for the
309  * socket (currently, it does nothing but enforce limits).  The space
310  * should be released by calling sbrelease() when the socket is destroyed.
311  */
312 
313 soreserve(so, sndcc, rcvcc)
314 	register struct socket *so;
315 	u_long sndcc, rcvcc;
316 {
317 
318 	if (sbreserve(&so->so_snd, sndcc) == 0)
319 		goto bad;
320 	if (sbreserve(&so->so_rcv, rcvcc) == 0)
321 		goto bad2;
322 	return (0);
323 bad2:
324 	sbrelease(&so->so_snd);
325 bad:
326 	return (ENOBUFS);
327 }
328 
329 /*
330  * Allot mbufs to a sockbuf.
331  * Attempt to scale cc so that mbcnt doesn't become limiting
332  * if buffering efficiency is near the normal case.
333  */
334 sbreserve(sb, cc)
335 	struct sockbuf *sb;
336 	u_long cc;
337 {
338 
339 	if (cc > (u_long)SB_MAX * MCLBYTES / (2 * MSIZE + MCLBYTES))
340 		return (0);
341 	sb->sb_hiwat = cc;
342 	sb->sb_mbmax = MIN(cc * 2, SB_MAX);
343 	return (1);
344 }
345 
346 /*
347  * Free mbufs held by a socket, and reserved mbuf space.
348  */
349 sbrelease(sb)
350 	struct sockbuf *sb;
351 {
352 
353 	sbflush(sb);
354 	sb->sb_hiwat = sb->sb_mbmax = 0;
355 }
356 
357 /*
358  * Routines to add and remove
359  * data from an mbuf queue.
360  *
361  * The routines sbappend() or sbappendrecord() are normally called to
362  * append new mbufs to a socket buffer, after checking that adequate
363  * space is available, comparing the function sbspace() with the amount
364  * of data to be added.  sbappendrecord() differs from sbappend() in
365  * that data supplied is treated as the beginning of a new record.
366  * To place a sender's address, optional access rights, and data in a
367  * socket receive buffer, sbappendaddr() should be used.  To place
368  * access rights and data in a socket receive buffer, sbappendrights()
369  * should be used.  In either case, the new data begins a new record.
370  * Note that unlike sbappend() and sbappendrecord(), these routines check
371  * for the caller that there will be enough space to store the data.
372  * Each fails if there is not enough space, or if it cannot find mbufs
373  * to store additional information in.
374  *
375  * Reliable protocols may use the socket send buffer to hold data
376  * awaiting acknowledgement.  Data is normally copied from a socket
377  * send buffer in a protocol with m_copy for output to a peer,
378  * and then removing the data from the socket buffer with sbdrop()
379  * or sbdroprecord() when the data is acknowledged by the peer.
380  */
381 
382 /*
383  * Append mbuf chain m to the last record in the
384  * socket buffer sb.  The additional space associated
385  * the mbuf chain is recorded in sb.  Empty mbufs are
386  * discarded and mbufs are compacted where possible.
387  */
388 sbappend(sb, m)
389 	struct sockbuf *sb;
390 	struct mbuf *m;
391 {
392 	register struct mbuf *n;
393 
394 	if (m == 0)
395 		return;
396 	if (n = sb->sb_mb) {
397 		while (n->m_nextpkt)
398 			n = n->m_nextpkt;
399 		while (n->m_next)
400 			n = n->m_next;
401 	}
402 	sbcompress(sb, m, n);
403 }
404 
405 /*
406  * As above, except the mbuf chain
407  * begins a new record.
408  */
409 sbappendrecord(sb, m0)
410 	register struct sockbuf *sb;
411 	register struct mbuf *m0;
412 {
413 	register struct mbuf *m;
414 
415 	if (m0 == 0)
416 		return;
417 	if (m = sb->sb_mb)
418 		while (m->m_nextpkt)
419 			m = m->m_nextpkt;
420 	/*
421 	 * Put the first mbuf on the queue.
422 	 * Note this permits zero length records.
423 	 */
424 	sballoc(sb, m0);
425 	if (m)
426 		m->m_nextpkt = m0;
427 	else
428 		sb->sb_mb = m0;
429 	m = m0->m_next;
430 	m0->m_next = 0;
431 	if (m && (m0->m_flags & M_EOR)) {
432 		m0->m_flags &= ~M_EOR;
433 		m->m_flags |= M_EOR;
434 	}
435 	sbcompress(sb, m, m0);
436 }
437 
438 /*
439  * As above except that OOB data
440  * is inserted at the beginning of the sockbuf,
441  * but after any other OOB data.
442  */
443 sbinsertoob(sb, m0)
444 	register struct sockbuf *sb;
445 	register struct mbuf *m0;
446 {
447 	register struct mbuf *m;
448 	register struct mbuf **mp;
449 
450 	if (m0 == 0)
451 		return;
452 	for (mp = &sb->sb_mb; m = *mp; mp = &((*mp)->m_nextpkt)) {
453 	    again:
454 		switch (m->m_type) {
455 
456 		case MT_OOBDATA:
457 			continue;		/* WANT next train */
458 
459 		case MT_CONTROL:
460 			if (m = m->m_next)
461 				goto again;	/* inspect THIS train further */
462 		}
463 		break;
464 	}
465 	/*
466 	 * Put the first mbuf on the queue.
467 	 * Note this permits zero length records.
468 	 */
469 	sballoc(sb, m0);
470 	m0->m_nextpkt = *mp;
471 	*mp = m0;
472 	m = m0->m_next;
473 	m0->m_next = 0;
474 	if (m && (m0->m_flags & M_EOR)) {
475 		m0->m_flags &= ~M_EOR;
476 		m->m_flags |= M_EOR;
477 	}
478 	sbcompress(sb, m, m0);
479 }
480 
481 /*
482  * Append address and data, and optionally, rights
483  * to the receive queue of a socket.  If present,
484  * m0 Return 0 if
485  * no space in sockbuf or insufficient mbufs.
486  */
487 sbappendaddr(sb, asa, m0, rights0)
488 	register struct sockbuf *sb;
489 	struct sockaddr *asa;
490 	struct mbuf *m0, *rights0;
491 {
492 	register struct mbuf *m, *n;
493 	int space = asa->sa_len;
494 
495 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
496 panic("sbappendaddr");
497 	if (m0)
498 		space += m0->m_pkthdr.len;
499 	if (rights0)
500 		space += rights0->m_len;
501 	if (space > sbspace(sb))
502 		return (0);
503 	MGET(m, M_DONTWAIT, MT_SONAME);
504 	if (m == 0)
505 		return (0);
506 	if (asa->sa_len > MLEN) {
507 		(void) m_free(m);
508 		return (0);
509 	}
510 	m->m_len = asa->sa_len;
511 	bcopy((caddr_t)asa, mtod(m, caddr_t), asa->sa_len);
512 	if (rights0 && rights0->m_len) {
513 		m->m_next = m_copy(rights0, 0, rights0->m_len);
514 		if (m->m_next == 0) {
515 			m_freem(m);
516 			return (0);
517 		}
518 		sballoc(sb, m->m_next);
519 	}
520 	sballoc(sb, m);
521 	if (n = sb->sb_mb) {
522 		while (n->m_nextpkt)
523 			n = n->m_nextpkt;
524 		n->m_nextpkt = m;
525 	} else
526 		sb->sb_mb = m;
527 	if (m->m_next)
528 		m = m->m_next;
529 	if (m0)
530 		sbcompress(sb, m0, m);
531 	return (1);
532 }
533 
534 sbappendrights(sb, m0, rights)
535 	struct sockbuf *sb;
536 	struct mbuf *rights, *m0;
537 {
538 	register struct mbuf *m, *n;
539 	int space = 0;
540 
541 	if (rights == 0)
542 		panic("sbappendrights");
543 	for (m = m0; m; m = m->m_next)
544 		space += m->m_len;
545 	space += rights->m_len;
546 	if (space > sbspace(sb))
547 		return (0);
548 	m = m_copy(rights, 0, rights->m_len);
549 	if (m == 0)
550 		return (0);
551 	sballoc(sb, m);
552 	if (n = sb->sb_mb) {
553 		while (n->m_nextpkt)
554 			n = n->m_nextpkt;
555 		n->m_nextpkt = m;
556 	} else
557 		sb->sb_mb = m;
558 	if (m0)
559 		sbcompress(sb, m0, m);
560 	return (1);
561 }
562 
563 /*
564  * Compress mbuf chain m into the socket
565  * buffer sb following mbuf n.  If n
566  * is null, the buffer is presumed empty.
567  */
568 sbcompress(sb, m, n)
569 	register struct sockbuf *sb;
570 	register struct mbuf *m, *n;
571 {
572 
573 	register int eor = 0;
574 	while (m) {
575 		eor |= m->m_flags & M_EOR;
576 		if (m->m_len == 0) {
577 			m = m_free(m);
578 			continue;
579 		}
580 		if (n && (n->m_flags & (M_EXT | M_EOR)) == 0 &&
581 		    (n->m_data + n->m_len + m->m_len) < &n->m_dat[MLEN] &&
582 		    n->m_type == m->m_type) {
583 			bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
584 			    (unsigned)m->m_len);
585 			n->m_len += m->m_len;
586 			sb->sb_cc += m->m_len;
587 			m = m_free(m);
588 			continue;
589 		}
590 		if (n)
591 			n->m_next = m;
592 		else
593 			sb->sb_mb = m;
594 		sballoc(sb, m);
595 		n = m;
596 		m->m_flags &= ~M_EOR;
597 		m = m->m_next;
598 		n->m_next = 0;
599 	}
600 	if (n)
601 		n->m_flags |= eor;
602 }
603 
604 /*
605  * Free all mbufs in a sockbuf.
606  * Check that all resources are reclaimed.
607  */
608 sbflush(sb)
609 	register struct sockbuf *sb;
610 {
611 
612 	if (sb->sb_flags & SB_LOCK)
613 		panic("sbflush");
614 	while (sb->sb_mbcnt)
615 		sbdrop(sb, (int)sb->sb_cc);
616 	if (sb->sb_cc || sb->sb_mbcnt || sb->sb_mb)
617 		panic("sbflush 2");
618 }
619 
620 /*
621  * Drop data from (the front of) a sockbuf.
622  */
623 sbdrop(sb, len)
624 	register struct sockbuf *sb;
625 	register int len;
626 {
627 	register struct mbuf *m, *mn;
628 	struct mbuf *next;
629 
630 	next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
631 	while (len > 0) {
632 		if (m == 0) {
633 			if (next == 0)
634 				panic("sbdrop");
635 			m = next;
636 			next = m->m_nextpkt;
637 			continue;
638 		}
639 		if (m->m_len > len) {
640 			m->m_len -= len;
641 			m->m_data += len;
642 			sb->sb_cc -= len;
643 			break;
644 		}
645 		len -= m->m_len;
646 		sbfree(sb, m);
647 		MFREE(m, mn);
648 		m = mn;
649 	}
650 	while (m && m->m_len == 0) {
651 		sbfree(sb, m);
652 		MFREE(m, mn);
653 		m = mn;
654 	}
655 	if (m) {
656 		sb->sb_mb = m;
657 		m->m_nextpkt = next;
658 	} else
659 		sb->sb_mb = next;
660 }
661 
662 /*
663  * Drop a record off the front of a sockbuf
664  * and move the next record to the front.
665  */
666 sbdroprecord(sb)
667 	register struct sockbuf *sb;
668 {
669 	register struct mbuf *m, *mn;
670 
671 	m = sb->sb_mb;
672 	if (m) {
673 		sb->sb_mb = m->m_nextpkt;
674 		do {
675 			sbfree(sb, m);
676 			MFREE(m, mn);
677 		} while (m = mn);
678 	}
679 }
680