xref: /csrg-svn/sys/kern/uipc_socket2.c (revision 8301)
1 /*	uipc_socket2.c	4.26	82/10/03	*/
2 
3 #include "../h/param.h"
4 #include "../h/systm.h"
5 #include "../h/dir.h"
6 #include "../h/user.h"
7 #include "../h/proc.h"
8 #include "../h/file.h"
9 #include "../h/inode.h"
10 #include "../h/buf.h"
11 #include "../h/mbuf.h"
12 #include "../h/protosw.h"
13 #include "../h/socket.h"
14 #include "../h/socketvar.h"
15 #include "../net/in.h"
16 #include "../net/in_systm.h"
17 
18 /*
19  * Primitive routines for operating on sockets and socket buffers
20  */
21 
22 /*
23  * Procedures to manipulate state flags of socket
24  * and do appropriate wakeups.  Normal sequence from the
25  * active (originating) side is that soisconnecting() is
26  * called during processing of connect() call,
27  * resulting in an eventual call to soisconnected() if/when the
28  * connection is established.  When the connection is torn down
29  * soisdisconnecting() is called during processing of disconnect() call,
30  * and soisdisconnected() is called when the connection to the peer
31  * is totally severed.  The semantics of these routines are such that
32  * connectionless protocols can call soisconnected() and soisdisconnected()
33  * only, bypassing the in-progress calls when setting up a ``connection''
34  * takes no time.
35  *
36  * From the passive side, a socket is created with SO_ACCEPTCONN
37  * creating two queues of sockets: so_q0 for connections in progress
38  * and so_q for connections already made and awaiting user acceptance.
39  * As a protocol is preparing incoming connections, it creates a socket
40  * structure queued on so_q0 by calling sonewconn().  When the connection
41  * is established, soisconnected() is called, and transfers the
42  * socket structure to so_q, making it available to accept().
43  *
44  * If a SO_ACCEPTCONN socket is closed with sockets on either
45  * so_q0 or so_q, these sockets are dropped.
46  *
47  * If and when higher level protocols are implemented in
48  * the kernel, the wakeups done here will sometimes
49  * be implemented as software-interrupt process scheduling.
50  */
51 
52 soisconnecting(so)
53 	struct socket *so;
54 {
55 
56 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
57 	so->so_state |= SS_ISCONNECTING;
58 	wakeup((caddr_t)&so->so_timeo);
59 }
60 
61 soisconnected(so)
62 	struct socket *so;
63 {
64 	register struct socket *head = so->so_head;
65 
66 	if (head) {
67 		if (soqremque(so, 0) == 0)
68 			panic("soisconnected");
69 		soqinsque(head, so, 1);
70 		wakeup((caddr_t)&head->so_timeo);
71 	}
72 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING);
73 	so->so_state |= SS_ISCONNECTED;
74 	wakeup((caddr_t)&so->so_timeo);
75 	sorwakeup(so);
76 	sowwakeup(so);
77 }
78 
79 soisdisconnecting(so)
80 	struct socket *so;
81 {
82 
83 	so->so_state &= ~SS_ISCONNECTING;
84 	so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
85 	wakeup((caddr_t)&so->so_timeo);
86 	sowwakeup(so);
87 	sorwakeup(so);
88 }
89 
90 soisdisconnected(so)
91 	struct socket *so;
92 {
93 
94 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
95 	so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE);
96 	wakeup((caddr_t)&so->so_timeo);
97 	sowwakeup(so);
98 	sorwakeup(so);
99 }
100 
101 /*
102  * When an attempt at a new connection is noted on a socket
103  * which accepts connections, sonewconn is called.  If the
104  * connection is possible (subject to space constraints, etc.)
105  * then we allocate a new structure, propoerly linked into the
106  * data structure of the original socket, and return this.
107  */
108 struct socket *
109 sonewconn(head)
110 	register struct socket *head;
111 {
112 	register struct socket *so;
113 	struct mbuf *m;
114 
115 	if (head->so_qlen + head->so_q0len > 3 * head->so_qlimit / 2)
116 		goto bad;
117 	m = m_getclr(M_DONTWAIT);
118 	if (m == 0)
119 		goto bad;
120 	so = mtod(m, struct socket *);
121 	so->so_type = head->so_type;
122 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
123 	so->so_linger = head->so_linger;
124 	so->so_state = head->so_state;
125 	so->so_proto = head->so_proto;
126 	so->so_timeo = head->so_timeo;
127 	so->so_pgrp = head->so_pgrp;
128 	soqinsque(head, so, 0);
129 	if ((*so->so_proto->pr_usrreq)(so, PRU_ATTACH, 0, 0, 0)) {
130 		(void) soqremque(so, 0);
131 		m_free(m);
132 		goto bad;
133 	}
134 	return (so);
135 bad:
136 	return ((struct socket *)0);
137 }
138 
139 soqinsque(head, so, q)
140 	register struct socket *head, *so;
141 	int q;
142 {
143 
144 	so->so_head = head;
145 	if (q == 0) {
146 		head->so_q0len++;
147 		so->so_q0 = head->so_q0;
148 		head->so_q0 = so;
149 	} else {
150 		head->so_qlen++;
151 		so->so_q = head->so_q;
152 		head->so_q = so;
153 	}
154 }
155 
156 soqremque(so, q)
157 	register struct socket *so;
158 	int q;
159 {
160 	register struct socket *head, *prev, *next;
161 
162 	head = so->so_head;
163 	prev = head;
164 	for (;;) {
165 		next = q ? prev->so_q : prev->so_q0;
166 		if (next == so)
167 			break;
168 		if (next == head)
169 			return (0);
170 		prev = next;
171 	}
172 	if (q == 0) {
173 		prev->so_q0 = next->so_q0;
174 		head->so_q0len--;
175 	} else {
176 		prev->so_q = next->so_q;
177 		head->so_qlen--;
178 	}
179 	next->so_q0 = next->so_q = 0;
180 	next->so_head = 0;
181 	return (1);
182 }
183 
184 /*
185  * Socantsendmore indicates that no more data will be sent on the
186  * socket; it would normally be applied to a socket when the user
187  * informs the system that no more data is to be sent, by the protocol
188  * code (in case PRU_SHUTDOWN).  Socantrcvmore indicates that no more data
189  * will be received, and will normally be applied to the socket by a
190  * protocol when it detects that the peer will send no more data.
191  * Data queued for reading in the socket may yet be read.
192  */
193 
194 socantsendmore(so)
195 	struct socket *so;
196 {
197 
198 	so->so_state |= SS_CANTSENDMORE;
199 	sowwakeup(so);
200 }
201 
202 socantrcvmore(so)
203 	struct socket *so;
204 {
205 
206 	so->so_state |= SS_CANTRCVMORE;
207 	sorwakeup(so);
208 }
209 
210 /*
211  * Socket select/wakeup routines.
212  */
213 
214 /*
215  * Interface routine to select() system
216  * call for sockets.
217  */
218 soselect(so, rw)
219 	register struct socket *so;
220 	int rw;
221 {
222 	int s = splnet();
223 
224 	switch (rw) {
225 
226 	case FREAD:
227 		if (soreadable(so)) {
228 			splx(s);
229 			return (1);
230 		}
231 		sbselqueue(&so->so_rcv);
232 		break;
233 
234 	case FWRITE:
235 		if (sowriteable(so)) {
236 			splx(s);
237 			return (1);
238 		}
239 		sbselqueue(&so->so_snd);
240 		break;
241 	}
242 	splx(s);
243 	return (0);
244 }
245 
246 /*
247  * Queue a process for a select on a socket buffer.
248  */
249 sbselqueue(sb)
250 	struct sockbuf *sb;
251 {
252 	register struct proc *p;
253 
254 	if ((p = sb->sb_sel) && p->p_wchan == (caddr_t)&selwait)
255 		sb->sb_flags |= SB_COLL;
256 	else
257 		sb->sb_sel = u.u_procp;
258 }
259 
260 /*
261  * Wait for data to arrive at/drain from a socket buffer.
262  */
263 sbwait(sb)
264 	struct sockbuf *sb;
265 {
266 
267 	sb->sb_flags |= SB_WAIT;
268 	sleep((caddr_t)&sb->sb_cc, PZERO+1);
269 }
270 
271 /*
272  * Wakeup processes waiting on a socket buffer.
273  */
274 sbwakeup(sb)
275 	struct sockbuf *sb;
276 {
277 
278 	if (sb->sb_sel) {
279 		selwakeup(sb->sb_sel, sb->sb_flags & SB_COLL);
280 		sb->sb_sel = 0;
281 		sb->sb_flags &= ~SB_COLL;
282 	}
283 	if (sb->sb_flags & SB_WAIT) {
284 		sb->sb_flags &= ~SB_WAIT;
285 		wakeup((caddr_t)&sb->sb_cc);
286 	}
287 }
288 
289 /*
290  * Socket buffer (struct sockbuf) utility routines.
291  *
292  * Each socket contains two socket buffers: one for sending data and
293  * one for receiving data.  Each buffer contains a queue of mbufs,
294  * information about the number of mbufs and amount of data in the
295  * queue, and other fields allowing select() statements and notification
296  * on data availability to be implemented.
297  *
298  * Before using a new socket structure it is first necessary to reserve
299  * buffer space to the socket, by calling sbreserve.  This commits
300  * some of the available buffer space in the system buffer pool for the
301  * socket.  The space should be released by calling sbrelease when the
302  * socket is destroyed.
303  *
304  * The routine sbappend() is normally called to append new mbufs
305  * to a socket buffer, after checking that adequate space is available
306  * comparing the function spspace() with the amount of data to be added.
307  * Data is normally removed from a socket buffer in a protocol by
308  * first calling m_copy on the socket buffer mbuf chain and sending this
309  * to a peer, and then removing the data from the socket buffer with
310  * sbdrop when the data is acknowledged by the peer (or immediately
311  * in the case of unreliable protocols.)
312  *
313  * Protocols which do not require connections place both source address
314  * and data information in socket buffer queues.  The source addresses
315  * are stored in single mbufs after each data item, and are easily found
316  * as the data items are all marked with end of record markers.  The
317  * sbappendaddr() routine stores a datum and associated address in
318  * a socket buffer.  Note that, unlike sbappend(), this routine checks
319  * for the caller that there will be enough space to store the data.
320  * It fails if there is not enough space, or if it cannot find
321  * a mbuf to store the address in.
322  *
323  * The higher-level routines sosend and soreceive (in socket.c)
324  * also add data to, and remove data from socket buffers repectively.
325  */
326 
327 /*
328  * Allot mbufs to a sockbuf.
329  */
330 sbreserve(sb, cc)
331 	struct sockbuf *sb;
332 {
333 
334 	/* someday maybe this routine will fail... */
335 	sb->sb_hiwat = cc;
336 	sb->sb_mbmax = cc*2;
337 	return (1);
338 }
339 
340 /*
341  * Free mbufs held by a socket, and reserved mbuf space.
342  */
343 sbrelease(sb)
344 	struct sockbuf *sb;
345 {
346 
347 	sbflush(sb);
348 	sb->sb_hiwat = sb->sb_mbmax = 0;
349 }
350 
351 /*
352  * Routines to add (at the end) and remove (from the beginning)
353  * data from a mbuf queue.
354  */
355 
356 /*
357  * Append mbuf queue m to sockbuf sb.
358  */
359 sbappend(sb, m)
360 	register struct mbuf *m;
361 	register struct sockbuf *sb;
362 {
363 	register struct mbuf *n;
364 
365 	n = sb->sb_mb;
366 	if (n)
367 		while (n->m_next)
368 			n = n->m_next;
369 	while (m) {
370 		if (m->m_len == 0 && (int)m->m_act == 0) {
371 			m = m_free(m);
372 			continue;
373 		}
374 		if (n && n->m_off <= MMAXOFF && m->m_off <= MMAXOFF &&
375 		   (int)n->m_act == 0 && (int)m->m_act == 0 &&
376 		   (n->m_off + n->m_len + m->m_len) <= MMAXOFF) {
377 			bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
378 			    (unsigned)m->m_len);
379 			n->m_len += m->m_len;
380 			sb->sb_cc += m->m_len;
381 			m = m_free(m);
382 			continue;
383 		}
384 		sballoc(sb, m);
385 		if (n == 0)
386 			sb->sb_mb = m;
387 		else
388 			n->m_next = m;
389 		n = m;
390 		m = m->m_next;
391 		n->m_next = 0;
392 	}
393 }
394 
395 /*
396  * Append data and address.
397  * Return 0 if no space in sockbuf or if
398  * can't get mbuf to stuff address in.
399  */
400 sbappendaddr(sb, asa, m0)
401 	struct sockbuf *sb;
402 	struct sockaddr *asa;
403 	struct mbuf *m0;
404 {
405 	struct sockaddr *msa;
406 	register struct mbuf *m;
407 	register int len = sizeof (struct sockaddr);
408 
409 	m = m0;
410 	if (m == 0)
411 		panic("sbappendaddr");
412 	for (;;) {
413 		len += m->m_len;
414 		if (m->m_next == 0) {
415 			m->m_act = (struct mbuf *)1;
416 			break;
417 		}
418 		m = m->m_next;
419 	}
420 	if (len > sbspace(sb))
421 		return (0);
422 	m = m_get(M_DONTWAIT);
423 	if (m == 0)
424 		return (0);
425 	m->m_off = MMINOFF;
426 	m->m_len = sizeof (struct sockaddr);
427 	msa = mtod(m, struct sockaddr *);
428 	*msa = *asa;
429 	m->m_act = (struct mbuf *)1;
430 	sbappend(sb, m);
431 	sbappend(sb, m0);
432 	return (1);
433 }
434 
435 /*
436  * Free all mbufs on a sockbuf mbuf chain.
437  * Check that resource allocations return to 0.
438  */
439 sbflush(sb)
440 	struct sockbuf *sb;
441 {
442 
443 	if (sb->sb_flags & SB_LOCK)
444 		panic("sbflush");
445 	if (sb->sb_cc)
446 		sbdrop(sb, sb->sb_cc);
447 	if (sb->sb_cc || sb->sb_mbcnt || sb->sb_mb)
448 		panic("sbflush 2");
449 }
450 
451 /*
452  * Drop data from (the front of) a sockbuf chain.
453  */
454 sbdrop(sb, len)
455 	register struct sockbuf *sb;
456 	register int len;
457 {
458 	register struct mbuf *m = sb->sb_mb, *mn;
459 
460 	while (len > 0) {
461 		if (m == 0)
462 			panic("sbdrop");
463 		if (m->m_len > len) {
464 			m->m_len -= len;
465 			m->m_off += len;
466 			sb->sb_cc -= len;
467 			break;
468 		}
469 		len -= m->m_len;
470 		sbfree(sb, m);
471 		MFREE(m, mn);
472 		m = mn;
473 	}
474 	sb->sb_mb = m;
475 }
476