xref: /csrg-svn/sys/kern/uipc_socket2.c (revision 23431)
1 /*
2  * Copyright (c) 1982 Regents of the University of California.
3  * All rights reserved.  The Berkeley software License Agreement
4  * specifies the terms and conditions for redistribution.
5  *
6  *	@(#)uipc_socket2.c	6.10 (Berkeley) 06/08/85
7  */
8 
9 #include "param.h"
10 #include "systm.h"
11 #include "dir.h"
12 #include "user.h"
13 #include "proc.h"
14 #include "file.h"
15 #include "inode.h"
16 #include "buf.h"
17 #include "mbuf.h"
18 #include "protosw.h"
19 #include "socket.h"
20 #include "socketvar.h"
21 
22 /*
23  * Primitive routines for operating on sockets and socket buffers
24  */
25 
26 /*
27  * Procedures to manipulate state flags of socket
28  * and do appropriate wakeups.  Normal sequence from the
29  * active (originating) side is that soisconnecting() is
30  * called during processing of connect() call,
31  * resulting in an eventual call to soisconnected() if/when the
32  * connection is established.  When the connection is torn down
33  * soisdisconnecting() is called during processing of disconnect() call,
34  * and soisdisconnected() is called when the connection to the peer
35  * is totally severed.  The semantics of these routines are such that
36  * connectionless protocols can call soisconnected() and soisdisconnected()
37  * only, bypassing the in-progress calls when setting up a ``connection''
38  * takes no time.
39  *
40  * From the passive side, a socket is created with
41  * two queues of sockets: so_q0 for connections in progress
42  * and so_q for connections already made and awaiting user acceptance.
43  * As a protocol is preparing incoming connections, it creates a socket
44  * structure queued on so_q0 by calling sonewconn().  When the connection
45  * is established, soisconnected() is called, and transfers the
46  * socket structure to so_q, making it available to accept().
47  *
48  * If a socket is closed with sockets on either
49  * so_q0 or so_q, these sockets are dropped.
50  *
51  * If higher level protocols are implemented in
52  * the kernel, the wakeups done here will sometimes
53  * cause software-interrupt process scheduling.
54  */
55 
56 soisconnecting(so)
57 	register struct socket *so;
58 {
59 
60 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
61 	so->so_state |= SS_ISCONNECTING;
62 	wakeup((caddr_t)&so->so_timeo);
63 }
64 
65 soisconnected(so)
66 	register struct socket *so;
67 {
68 	register struct socket *head = so->so_head;
69 
70 	if (head) {
71 		if (soqremque(so, 0) == 0)
72 			panic("soisconnected");
73 		soqinsque(head, so, 1);
74 		sorwakeup(head);
75 		wakeup((caddr_t)&head->so_timeo);
76 	}
77 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING);
78 	so->so_state |= SS_ISCONNECTED;
79 	wakeup((caddr_t)&so->so_timeo);
80 	sorwakeup(so);
81 	sowwakeup(so);
82 }
83 
84 soisdisconnecting(so)
85 	register struct socket *so;
86 {
87 
88 	so->so_state &= ~SS_ISCONNECTING;
89 	so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
90 	wakeup((caddr_t)&so->so_timeo);
91 	sowwakeup(so);
92 	sorwakeup(so);
93 }
94 
95 soisdisconnected(so)
96 	register struct socket *so;
97 {
98 
99 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
100 	so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE);
101 	wakeup((caddr_t)&so->so_timeo);
102 	sowwakeup(so);
103 	sorwakeup(so);
104 }
105 
106 /*
107  * When an attempt at a new connection is noted on a socket
108  * which accepts connections, sonewconn is called.  If the
109  * connection is possible (subject to space constraints, etc.)
110  * then we allocate a new structure, propoerly linked into the
111  * data structure of the original socket, and return this.
112  */
113 struct socket *
114 sonewconn(head)
115 	register struct socket *head;
116 {
117 	register struct socket *so;
118 	register struct mbuf *m;
119 
120 	if (head->so_qlen + head->so_q0len > 3 * head->so_qlimit / 2)
121 		goto bad;
122 	m = m_getclr(M_DONTWAIT, MT_SOCKET);
123 	if (m == NULL)
124 		goto bad;
125 	so = mtod(m, struct socket *);
126 	so->so_type = head->so_type;
127 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
128 	so->so_linger = head->so_linger;
129 	so->so_state = head->so_state | SS_NOFDREF;
130 	so->so_proto = head->so_proto;
131 	so->so_timeo = head->so_timeo;
132 	so->so_pgrp = head->so_pgrp;
133 	soqinsque(head, so, 0);
134 	if ((*so->so_proto->pr_usrreq)(so, PRU_ATTACH,
135 	    (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0)) {
136 		(void) soqremque(so, 0);
137 		(void) m_free(m);
138 		goto bad;
139 	}
140 	return (so);
141 bad:
142 	return ((struct socket *)0);
143 }
144 
145 soqinsque(head, so, q)
146 	register struct socket *head, *so;
147 	int q;
148 {
149 
150 	so->so_head = head;
151 	if (q == 0) {
152 		head->so_q0len++;
153 		so->so_q0 = head->so_q0;
154 		head->so_q0 = so;
155 	} else {
156 		head->so_qlen++;
157 		so->so_q = head->so_q;
158 		head->so_q = so;
159 	}
160 }
161 
162 soqremque(so, q)
163 	register struct socket *so;
164 	int q;
165 {
166 	register struct socket *head, *prev, *next;
167 
168 	head = so->so_head;
169 	prev = head;
170 	for (;;) {
171 		next = q ? prev->so_q : prev->so_q0;
172 		if (next == so)
173 			break;
174 		if (next == head)
175 			return (0);
176 		prev = next;
177 	}
178 	if (q == 0) {
179 		prev->so_q0 = next->so_q0;
180 		head->so_q0len--;
181 	} else {
182 		prev->so_q = next->so_q;
183 		head->so_qlen--;
184 	}
185 	next->so_q0 = next->so_q = 0;
186 	next->so_head = 0;
187 	return (1);
188 }
189 
190 /*
191  * Socantsendmore indicates that no more data will be sent on the
192  * socket; it would normally be applied to a socket when the user
193  * informs the system that no more data is to be sent, by the protocol
194  * code (in case PRU_SHUTDOWN).  Socantrcvmore indicates that no more data
195  * will be received, and will normally be applied to the socket by a
196  * protocol when it detects that the peer will send no more data.
197  * Data queued for reading in the socket may yet be read.
198  */
199 
200 socantsendmore(so)
201 	struct socket *so;
202 {
203 
204 	so->so_state |= SS_CANTSENDMORE;
205 	sowwakeup(so);
206 }
207 
208 socantrcvmore(so)
209 	struct socket *so;
210 {
211 
212 	so->so_state |= SS_CANTRCVMORE;
213 	sorwakeup(so);
214 }
215 
216 /*
217  * Socket select/wakeup routines.
218  */
219 
220 /*
221  * Queue a process for a select on a socket buffer.
222  */
223 sbselqueue(sb)
224 	struct sockbuf *sb;
225 {
226 	register struct proc *p;
227 
228 	if ((p = sb->sb_sel) && p->p_wchan == (caddr_t)&selwait)
229 		sb->sb_flags |= SB_COLL;
230 	else
231 		sb->sb_sel = u.u_procp;
232 }
233 
234 /*
235  * Wait for data to arrive at/drain from a socket buffer.
236  */
237 sbwait(sb)
238 	struct sockbuf *sb;
239 {
240 
241 	sb->sb_flags |= SB_WAIT;
242 	sleep((caddr_t)&sb->sb_cc, PZERO+1);
243 }
244 
245 /*
246  * Wakeup processes waiting on a socket buffer.
247  */
248 sbwakeup(sb)
249 	register struct sockbuf *sb;
250 {
251 
252 	if (sb->sb_sel) {
253 		selwakeup(sb->sb_sel, sb->sb_flags & SB_COLL);
254 		sb->sb_sel = 0;
255 		sb->sb_flags &= ~SB_COLL;
256 	}
257 	if (sb->sb_flags & SB_WAIT) {
258 		sb->sb_flags &= ~SB_WAIT;
259 		wakeup((caddr_t)&sb->sb_cc);
260 	}
261 }
262 
263 /*
264  * Wakeup socket readers and writers.
265  * Do asynchronous notification via SIGIO
266  * if the socket has the SS_ASYNC flag set.
267  */
268 sowakeup(so, sb)
269 	register struct socket *so;
270 	struct sockbuf *sb;
271 {
272 	register struct proc *p;
273 
274 	sbwakeup(sb);
275 	if (so->so_state & SS_ASYNC) {
276 		if (so->so_pgrp < 0)
277 			gsignal(-so->so_pgrp, SIGIO);
278 		else if (so->so_pgrp > 0 && (p = pfind(so->so_pgrp)) != 0)
279 			psignal(p, SIGIO);
280 	}
281 }
282 
283 /*
284  * Socket buffer (struct sockbuf) utility routines.
285  *
286  * Each socket contains two socket buffers: one for sending data and
287  * one for receiving data.  Each buffer contains a queue of mbufs,
288  * information about the number of mbufs and amount of data in the
289  * queue, and other fields allowing select() statements and notification
290  * on data availability to be implemented.
291  *
292  * Data stored in a socket buffer is maintained as a list of records.
293  * Each record is a list of mbufs chained together with the m_next
294  * field.  Records are chained together with the m_act field. The upper
295  * level routine soreceive() expects the following conventions to be
296  * observed when placing information in the receive buffer:
297  *
298  * 1. If the protocol requires each message be preceded by the sender's
299  *    name, then a record containing that name must be present before
300  *    any associated data (mbuf's must be of type MT_SONAME).
301  * 2. If the protocol supports the exchange of ``access rights'' (really
302  *    just additional data associated with the message), and there are
303  *    ``rights'' to be received, then a record containing this data
304  *    should be present (mbuf's must be of type MT_RIGHTS).
305  * 3. If a name or rights record exists, then it must be followed by
306  *    a data record, perhaps of zero length.
307  *
308  * Before using a new socket structure it is first necessary to reserve
309  * buffer space to the socket, by calling sbreserve().  This commits
310  * some of the available buffer space in the system buffer pool for the
311  * socket.  The space should be released by calling sbrelease() when the
312  * socket is destroyed.
313  *
314  * The routines sbappend() or sbappendrecord() are normally called to
315  * append new mbufs to a socket buffer, after checking that adequate
316  * space is available, comparing the function sbspace() with the amount
317  * of data to be added.  sbappendrecord() differs from sbappend() in
318  * that data supplied is treated as the beginning of a new record.
319  * Data is normally removed from a socket buffer in a protocol by
320  * first calling m_copy on the socket buffer mbuf chain and sending this
321  * to a peer, and then removing the data from the socket buffer with
322  * sbdrop() or sbdroprecord() when the data is acknowledged by the peer
323  * (or immediately in the case of unreliable protocols.)
324  *
325  * To place a sender's name, optionally, access rights, and data in a
326  * socket buffer sbappendaddr() should be used.  To place access rights
327  * and data in a socket buffer sbappendrights() should be used.  Note
328  * that unlike sbappend() and sbappendrecord(), these routines check
329  * for the caller that there will be enough space to store the data.
330  * Each fails if there is not enough space, or if it cannot find mbufs
331  * to store additional information in.
332  */
333 
334 soreserve(so, sndcc, rcvcc)
335 	register struct socket *so;
336 	int sndcc, rcvcc;
337 {
338 
339 	if (sbreserve(&so->so_snd, sndcc) == 0)
340 		goto bad;
341 	if (sbreserve(&so->so_rcv, rcvcc) == 0)
342 		goto bad2;
343 	return (0);
344 bad2:
345 	sbrelease(&so->so_snd);
346 bad:
347 	return (ENOBUFS);
348 }
349 
350 /*
351  * Allot mbufs to a sockbuf.
352  */
353 sbreserve(sb, cc)
354 	struct sockbuf *sb;
355 {
356 
357 	if ((unsigned) cc > SB_MAX)
358 		return (0);
359 	/* someday maybe this routine will fail... */
360 	sb->sb_hiwat = cc;
361 	/* * 2 implies names can be no more than 1 mbuf each */
362 	sb->sb_mbmax = MIN(cc * 2, SB_MAX);
363 	return (1);
364 }
365 
366 /*
367  * Free mbufs held by a socket, and reserved mbuf space.
368  */
369 sbrelease(sb)
370 	struct sockbuf *sb;
371 {
372 
373 	sbflush(sb);
374 	sb->sb_hiwat = sb->sb_mbmax = 0;
375 }
376 
377 /*
378  * Routines to add and remove
379  * data from an mbuf queue.
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_act)
398 			n = n->m_act;
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_act)
419 			m = m->m_act;
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_act = m0;
427 	else
428 		sb->sb_mb = m0;
429 	m = m0->m_next;
430 	m0->m_next = 0;
431 	sbcompress(sb, m, m0);
432 }
433 
434 /*
435  * Append address and data, and optionally, rights
436  * to the receive queue of a socket.  Return 0 if
437  * no space in sockbuf or insufficient mbufs.
438  */
439 sbappendaddr(sb, asa, m0, rights0)		/* XXX */
440 	register struct sockbuf *sb;
441 	struct sockaddr *asa;
442 	struct mbuf *rights0, *m0;
443 {
444 	register struct mbuf *m, *n;
445 	int space = sizeof (*asa);
446 
447 	m = m0;
448 	if (m == 0)
449 		panic("sbappendaddr");
450 	do {
451 		space += m->m_len;
452 		m = m->m_next;
453 	} while (m);
454 	if (rights0)
455 		space += rights0->m_len;
456 	if (space > sbspace(sb))
457 		return (0);
458 	m = m_get(M_DONTWAIT, MT_SONAME);
459 	if (m == 0)
460 		return (0);
461 	*mtod(m, struct sockaddr *) = *asa;
462 	m->m_len = sizeof (*asa);
463 	if (rights0) {
464 		m->m_act = m_copy(rights0, 0, rights0->m_len);
465 		if (m->m_act == 0) {
466 			m_freem(m);
467 			return (0);
468 		}
469 		sballoc(sb, m->m_act);
470 	}
471 	sballoc(sb, m);
472 	if (n = sb->sb_mb) {
473 		while (n->m_act)
474 			n = n->m_act;
475 		n->m_act = m;
476 	} else
477 		sb->sb_mb = m;
478 	if (m->m_act)
479 		m = m->m_act;
480 	sballoc(sb, m0);
481 	m->m_act = m0;
482 	m = m0->m_next;
483 	m0->m_next = 0;
484 	if (m)
485 		sbcompress(sb, m, m0);
486 	return (1);
487 }
488 
489 #ifdef notdef
490 sbappendrights(sb, rights, m0)
491 	struct sockbuf *sb;
492 	struct mbuf *rights, *m;
493 {
494 	register struct mbuf *m, *n;
495 	int space = 0;
496 
497 	m = m0;
498 	if (m == 0 || rights == 0)
499 		panic("sbappendrights");
500 	do {
501 		space += m->m_len;
502 		m = m->m_next;
503 	} while (m);
504 	space += rights->m_len;
505 	if (space > sbspace(sb))
506 		return (0);
507 	m = m_copy(rights, 0, rights->m_len);
508 	if (m == 0)
509 		return (0);
510 	sballoc(sb, m);
511 	if (n = sb->sb_mb) {
512 		while (n->m_act)
513 			n = n->m_act;
514 		n->m_act = m;
515 	} else
516 		n->m_act = m;
517 	sballoc(sb, m0);
518 	m->m_act = m0;
519 	m = m0->m_next;
520 	m0->m_next = 0;
521 	if (m)
522 		sbcompress(sb, m, m0);
523 	return (1);
524 }
525 #endif
526 
527 /*
528  * Compress mbuf chain m into the socket
529  * buffer sb following mbuf n.  If n
530  * is null, the buffer is presumed empty.
531  */
532 sbcompress(sb, m, n)
533 	register struct sockbuf *sb;
534 	register struct mbuf *m, *n;
535 {
536 
537 	while (m) {
538 		if (m->m_len == 0) {
539 			m = m_free(m);
540 			continue;
541 		}
542 		if (n && n->m_off <= MMAXOFF && m->m_off <= MMAXOFF &&
543 		    (n->m_off + n->m_len + m->m_len) <= MMAXOFF) {
544 			bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
545 			    (unsigned)m->m_len);
546 			n->m_len += m->m_len;
547 			sb->sb_cc += m->m_len;
548 			m = m_free(m);
549 			continue;
550 		}
551 		sballoc(sb, m);
552 		if (n)
553 			n->m_next = m;
554 		else
555 			sb->sb_mb = m;
556 		n = m;
557 		m = m->m_next;
558 		n->m_next = 0;
559 	}
560 }
561 
562 /*
563  * Free all mbufs in a sockbuf.
564  * Check that all resources are reclaimed.
565  */
566 sbflush(sb)
567 	register struct sockbuf *sb;
568 {
569 
570 	if (sb->sb_flags & SB_LOCK)
571 		panic("sbflush");
572 	if (sb->sb_cc)
573 		sbdrop(sb, sb->sb_cc);
574 	if (sb->sb_cc || sb->sb_mbcnt || sb->sb_mb)
575 		panic("sbflush 2");
576 }
577 
578 /*
579  * Drop data from (the front of) a sockbuf.
580  */
581 struct mbuf *
582 sbdrop(sb, len)
583 	register struct sockbuf *sb;
584 	register int len;
585 {
586 	register struct mbuf *m, *mn;
587 	struct mbuf *next;
588 
589 	next = (m = sb->sb_mb) ? m->m_act : 0;
590 	while (len > 0) {
591 		if (m == 0) {
592 			if (next == 0)
593 				panic("sbdrop");
594 			m = next;
595 			next = m->m_act;
596 			continue;
597 		}
598 		if (m->m_len > len) {
599 			m->m_len -= len;
600 			m->m_off += len;
601 			sb->sb_cc -= len;
602 			break;
603 		}
604 		len -= m->m_len;
605 		sbfree(sb, m);
606 		MFREE(m, mn);
607 		m = mn;
608 	}
609 	while (m && m->m_len == 0) {
610 		sbfree(sb, m);
611 		MFREE(m, mn);
612 		m = mn;
613 	}
614 	if (m) {
615 		sb->sb_mb = m;
616 		m->m_act = next;
617 	} else
618 		sb->sb_mb = next;
619 	return (sb->sb_mb);
620 }
621 
622 /*
623  * Drop a record off the front of a sockbuf
624  * and move the next record to the front.
625  */
626 struct mbuf *
627 sbdroprecord(sb)
628 	register struct sockbuf *sb;
629 {
630 	register struct mbuf *m, *mn;
631 
632 	m = sb->sb_mb;
633 	if (m) {
634 		sb->sb_mb = m->m_act;
635 		do {
636 			sbfree(sb, m);
637 			MFREE(m, mn);
638 		} while (m = mn);
639 	}
640 	return (sb->sb_mb);
641 }
642