xref: /original-bsd/sys/netinet/tcp_usrreq.c (revision f82e54c4)
1 /*	tcp_usrreq.c	6.4	84/08/29	*/
2 
3 #include "param.h"
4 #include "systm.h"
5 #include "mbuf.h"
6 #include "socket.h"
7 #include "socketvar.h"
8 #include "protosw.h"
9 #include "errno.h"
10 #include "stat.h"
11 
12 #include "../net/if.h"
13 #include "../net/route.h"
14 
15 #include "in.h"
16 #include "in_pcb.h"
17 #include "in_systm.h"
18 #include "ip.h"
19 #include "ip_var.h"
20 #include "tcp.h"
21 #include "tcp_fsm.h"
22 #include "tcp_seq.h"
23 #include "tcp_timer.h"
24 #include "tcp_var.h"
25 #include "tcpip.h"
26 #include "tcp_debug.h"
27 
28 /*
29  * TCP protocol interface to socket abstraction.
30  */
31 extern	char *tcpstates[];
32 struct	tcpcb *tcp_newtcpcb();
33 int	tcpsenderrors;
34 
35 /*
36  * Process a TCP user request for TCP tb.  If this is a send request
37  * then m is the mbuf chain of send data.  If this is a timer expiration
38  * (called from the software clock routine), then timertype tells which timer.
39  */
40 /*ARGSUSED*/
41 tcp_usrreq(so, req, m, nam, rights)
42 	struct socket *so;
43 	int req;
44 	struct mbuf *m, *nam, *rights;
45 {
46 	register struct inpcb *inp = sotoinpcb(so);
47 	register struct tcpcb *tp;
48 	int s = splnet();
49 	int error = 0;
50 	int ostate;
51 
52 	if (rights && rights->m_len) {
53 		splx(s);
54 		return (EINVAL);
55 	}
56 	/*
57 	 * When a TCP is attached to a socket, then there will be
58 	 * a (struct inpcb) pointed at by the socket, and this
59 	 * structure will point at a subsidary (struct tcpcb).
60 	 */
61 	if (inp == 0 && req != PRU_ATTACH) {
62 		splx(s);
63 		return (EINVAL);		/* XXX */
64 	}
65 	if (inp) {
66 		tp = intotcpcb(inp);
67 		/* WHAT IF TP IS 0? */
68 #ifdef KPROF
69 		tcp_acounts[tp->t_state][req]++;
70 #endif
71 		ostate = tp->t_state;
72 	} else
73 		ostate = 0;
74 	switch (req) {
75 
76 	/*
77 	 * TCP attaches to socket via PRU_ATTACH, reserving space,
78 	 * and an internet control block.
79 	 */
80 	case PRU_ATTACH:
81 		if (inp) {
82 			error = EISCONN;
83 			break;
84 		}
85 		error = tcp_attach(so);
86 		if (error)
87 			break;
88 		if ((so->so_options & SO_LINGER) && so->so_linger == 0)
89 			so->so_linger = TCP_LINGERTIME;
90 		tp = sototcpcb(so);
91 		break;
92 
93 	/*
94 	 * PRU_DETACH detaches the TCP protocol from the socket.
95 	 * If the protocol state is non-embryonic, then can't
96 	 * do this directly: have to initiate a PRU_DISCONNECT,
97 	 * which may finish later; embryonic TCB's can just
98 	 * be discarded here.
99 	 */
100 	case PRU_DETACH:
101 		if (tp->t_state > TCPS_LISTEN)
102 			tp = tcp_disconnect(tp);
103 		else
104 			tp = tcp_close(tp);
105 		break;
106 
107 	/*
108 	 * Give the socket an address.
109 	 */
110 	case PRU_BIND:
111 		error = in_pcbbind(inp, nam);
112 		if (error)
113 			break;
114 		break;
115 
116 	/*
117 	 * Prepare to accept connections.
118 	 */
119 	case PRU_LISTEN:
120 		if (inp->inp_lport == 0)
121 			error = in_pcbbind(inp, (struct mbuf *)0);
122 		if (error == 0)
123 			tp->t_state = TCPS_LISTEN;
124 		break;
125 
126 	/*
127 	 * Initiate connection to peer.
128 	 * Create a template for use in transmissions on this connection.
129 	 * Enter SYN_SENT state, and mark socket as connecting.
130 	 * Start keep-alive timer, and seed output sequence space.
131 	 * Send initial segment on connection.
132 	 */
133 	case PRU_CONNECT:
134 		if (inp->inp_lport == 0) {
135 			error = in_pcbbind(inp, (struct mbuf *)0);
136 			if (error)
137 				break;
138 		}
139 		error = in_pcbconnect(inp, nam);
140 		if (error)
141 			break;
142 		tp->t_template = tcp_template(tp);
143 		if (tp->t_template == 0) {
144 			in_pcbdisconnect(inp);
145 			error = ENOBUFS;
146 			break;
147 		}
148 		soisconnecting(so);
149 		tp->t_state = TCPS_SYN_SENT;
150 		tp->t_timer[TCPT_KEEP] = TCPTV_KEEP;
151 		tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2;
152 		tcp_sendseqinit(tp);
153 		error = tcp_output(tp);
154 		break;
155 
156 	/*
157 	 * Create a TCP connection between two sockets.
158 	 */
159 	case PRU_CONNECT2:
160 		error = EOPNOTSUPP;
161 		break;
162 
163 	/*
164 	 * Initiate disconnect from peer.
165 	 * If connection never passed embryonic stage, just drop;
166 	 * else if don't need to let data drain, then can just drop anyways,
167 	 * else have to begin TCP shutdown process: mark socket disconnecting,
168 	 * drain unread data, state switch to reflect user close, and
169 	 * send segment (e.g. FIN) to peer.  Socket will be really disconnected
170 	 * when peer sends FIN and acks ours.
171 	 *
172 	 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
173 	 */
174 	case PRU_DISCONNECT:
175 		tp = tcp_disconnect(tp);
176 		break;
177 
178 	/*
179 	 * Accept a connection.  Essentially all the work is
180 	 * done at higher levels; just return the address
181 	 * of the peer, storing through addr.
182 	 */
183 	case PRU_ACCEPT: {
184 		struct sockaddr_in *sin = mtod(nam, struct sockaddr_in *);
185 
186 		nam->m_len = sizeof (struct sockaddr_in);
187 		sin->sin_family = AF_INET;
188 		sin->sin_port = inp->inp_fport;
189 		sin->sin_addr = inp->inp_faddr;
190 		break;
191 		}
192 
193 	/*
194 	 * Mark the connection as being incapable of further output.
195 	 */
196 	case PRU_SHUTDOWN:
197 		socantsendmore(so);
198 		tp = tcp_usrclosed(tp);
199 		if (tp)
200 			error = tcp_output(tp);
201 		break;
202 
203 	/*
204 	 * After a receive, possibly send window update to peer.
205 	 */
206 	case PRU_RCVD:
207 		(void) tcp_output(tp);
208 		break;
209 
210 	/*
211 	 * Do a send by putting data in output queue and updating urgent
212 	 * marker if URG set.  Possibly send more data.
213 	 */
214 	case PRU_SEND:
215 		sbappend(&so->so_snd, m);
216 #ifdef notdef
217 		if (tp->t_flags & TF_PUSH)
218 			tp->snd_end = tp->snd_una + so->so_snd.sb_cc;
219 #endif
220 		error = tcp_output(tp);
221 		if (error) {		/* XXX fix to use other path */
222 			if (error == ENOBUFS)		/* XXX */
223 				error = 0;		/* XXX */
224 			tcpsenderrors++;
225 		}
226 		break;
227 
228 	/*
229 	 * Abort the TCP.
230 	 */
231 	case PRU_ABORT:
232 		tp = tcp_drop(tp, ECONNABORTED);
233 		break;
234 
235 /* SOME AS YET UNIMPLEMENTED HOOKS */
236 	case PRU_CONTROL:
237 		error = EOPNOTSUPP;
238 		break;
239 /* END UNIMPLEMENTED HOOKS */
240 
241 	case PRU_SENSE:
242 		((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat;
243 		return (0);
244 
245 	case PRU_RCVOOB:
246 		if (so->so_oobmark == 0 &&
247 		    (so->so_state & SS_RCVATMARK) == 0) {
248 			error = EINVAL;
249 			break;
250 		}
251 		if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
252 			error = EWOULDBLOCK;
253 			break;
254 		}
255 		m->m_len = 1;
256 		*mtod(m, caddr_t) = tp->t_iobc;
257 		break;
258 
259 	case PRU_SENDOOB:
260 		if (sbspace(&so->so_snd) < -512) {
261 			m_freem(m);
262 			error = ENOBUFS;
263 			break;
264 		}
265 		tp->snd_up = tp->snd_una + so->so_snd.sb_cc + 1;
266 		sbappend(&so->so_snd, m);
267 		tp->t_force = 1;
268 		error = tcp_output(tp);
269 		tp->t_force = 0;
270 		break;
271 
272 	case PRU_SOCKADDR:
273 		in_setsockaddr(inp, nam);
274 		break;
275 
276 	case PRU_PEERADDR:
277 		in_setpeeraddr(inp, nam);
278 		break;
279 
280 	/*
281 	 * TCP slow timer went off; going through this
282 	 * routine for tracing's sake.
283 	 */
284 	case PRU_SLOWTIMO:
285 		tp = tcp_timers(tp, (int)nam);
286 		req |= (int)nam << 8;		/* for debug's sake */
287 		break;
288 
289 	default:
290 		panic("tcp_usrreq");
291 	}
292 	if (tp && (so->so_options & SO_DEBUG))
293 		tcp_trace(TA_USER, ostate, tp, (struct tcpiphdr *)0, req);
294 	splx(s);
295 	return (error);
296 }
297 
298 int	tcp_sendspace = 1024*2;
299 int	tcp_recvspace = 1024*2;
300 /*
301  * Attach TCP protocol to socket, allocating
302  * internet protocol control block, tcp control block,
303  * bufer space, and entering LISTEN state if to accept connections.
304  */
305 tcp_attach(so)
306 	struct socket *so;
307 {
308 	register struct tcpcb *tp;
309 	struct inpcb *inp;
310 	int error;
311 
312 	error = soreserve(so, tcp_sendspace, tcp_recvspace);
313 	if (error)
314 		return (error);
315 	error = in_pcballoc(so, &tcb);
316 	if (error)
317 		return (error);
318 	inp = sotoinpcb(so);
319 	tp = tcp_newtcpcb(inp);
320 	if (tp == 0) {
321 		int nofd = so->so_state & SS_NOFDREF;	/* XXX */
322 
323 		so->so_state &= ~SS_NOFDREF;	/* don't free the socket yet */
324 		in_pcbdetach(inp);
325 		so->so_state |= nofd;
326 		return (ENOBUFS);
327 	}
328 	tp->t_state = TCPS_CLOSED;
329 	return (0);
330 }
331 
332 /*
333  * Initiate (or continue) disconnect.
334  * If embryonic state, just send reset (once).
335  * If in ``let data drain'' option and linger null, just drop.
336  * Otherwise (hard), mark socket disconnecting and drop
337  * current input data; switch states based on user close, and
338  * send segment to peer (with FIN).
339  */
340 struct tcpcb *
341 tcp_disconnect(tp)
342 	register struct tcpcb *tp;
343 {
344 	struct socket *so = tp->t_inpcb->inp_socket;
345 
346 	if (tp->t_state < TCPS_ESTABLISHED)
347 		tp = tcp_close(tp);
348 	else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
349 		tp = tcp_drop(tp, 0);
350 	else {
351 		soisdisconnecting(so);
352 		sbflush(&so->so_rcv);
353 		tp = tcp_usrclosed(tp);
354 		if (tp)
355 			(void) tcp_output(tp);
356 	}
357 	return (tp);
358 }
359 
360 /*
361  * User issued close, and wish to trail through shutdown states:
362  * if never received SYN, just forget it.  If got a SYN from peer,
363  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
364  * If already got a FIN from peer, then almost done; go to LAST_ACK
365  * state.  In all other cases, have already sent FIN to peer (e.g.
366  * after PRU_SHUTDOWN), and just have to play tedious game waiting
367  * for peer to send FIN or not respond to keep-alives, etc.
368  * We can let the user exit from the close as soon as the FIN is acked.
369  */
370 struct tcpcb *
371 tcp_usrclosed(tp)
372 	register struct tcpcb *tp;
373 {
374 
375 	switch (tp->t_state) {
376 
377 	case TCPS_CLOSED:
378 	case TCPS_LISTEN:
379 	case TCPS_SYN_SENT:
380 		tp->t_state = TCPS_CLOSED;
381 		tp = tcp_close(tp);
382 		break;
383 
384 	case TCPS_SYN_RECEIVED:
385 	case TCPS_ESTABLISHED:
386 		tp->t_state = TCPS_FIN_WAIT_1;
387 		break;
388 
389 	case TCPS_CLOSE_WAIT:
390 		tp->t_state = TCPS_LAST_ACK;
391 		break;
392 	}
393 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2)
394 		soisdisconnected(tp->t_inpcb->inp_socket);
395 	return (tp);
396 }
397