xref: /csrg-svn/sys/netinet/tcp_usrreq.c (revision 8640)
1 /*	tcp_usrreq.c	1.67	82/10/17	*/
2 
3 #include "../h/param.h"
4 #include "../h/systm.h"
5 #include "../h/mbuf.h"
6 #include "../h/socket.h"
7 #include "../h/socketvar.h"
8 #include "../h/protosw.h"
9 #include "../netinet/in.h"
10 #include "../net/route.h"
11 #include "../netinet/in_pcb.h"
12 #include "../netinet/in_systm.h"
13 #include "../net/if.h"
14 #include "../netinet/ip.h"
15 #include "../netinet/ip_var.h"
16 #include "../netinet/tcp.h"
17 #include "../netinet/tcp_fsm.h"
18 #include "../netinet/tcp_seq.h"
19 #include "../netinet/tcp_timer.h"
20 #include "../netinet/tcp_var.h"
21 #include "../netinet/tcpip.h"
22 #include "../netinet/tcp_debug.h"
23 #include <errno.h>
24 
25 /*
26  * TCP protocol interface to socket abstraction.
27  */
28 extern	char *tcpstates[];
29 struct	tcpcb *tcp_newtcpcb();
30 
31 /*
32  * Process a TCP user request for TCP tb.  If this is a send request
33  * then m is the mbuf chain of send data.  If this is a timer expiration
34  * (called from the software clock routine), then timertype tells which timer.
35  */
36 /*ARGSUSED*/
37 tcp_usrreq(so, req, m, nam, opt)
38 	struct socket *so;
39 	int req;
40 	struct mbuf *m, *nam;
41 	struct socketopt *opt;
42 {
43 	register struct inpcb *inp = sotoinpcb(so);
44 	register struct tcpcb *tp;
45 	int s = splnet();
46 	int error = 0;
47 	int ostate;
48 
49 	/*
50 	 * When a TCP is attached to a socket, then there will be
51 	 * a (struct inpcb) pointed at by the socket, and this
52 	 * structure will point at a subsidary (struct tcpcb).
53 	 */
54 	if (inp == 0 && req != PRU_ATTACH) {
55 		splx(s);
56 		return (EINVAL);		/* XXX */
57 	}
58 	if (inp) {
59 		tp = intotcpcb(inp);
60 		/* WHAT IF TP IS 0? */
61 #ifdef KPROF
62 		tcp_acounts[tp->t_state][req]++;
63 #endif
64 		ostate = tp->t_state;
65 	} else
66 		ostate = 0;
67 	switch (req) {
68 
69 	/*
70 	 * TCP attaches to socket via PRU_ATTACH, reserving space,
71 	 * and an internet control block.
72 	 */
73 	case PRU_ATTACH:
74 		if (inp) {
75 			error = EISCONN;
76 			break;
77 		}
78 		error = tcp_attach(so);
79 		if (error)
80 			break;
81 		if ((so->so_options & SO_DONTLINGER) == 0)
82 			so->so_linger = TCP_LINGERTIME;
83 		tp = sototcpcb(so);
84 		break;
85 
86 	/*
87 	 * PRU_DETACH detaches the TCP protocol from the socket.
88 	 * If the protocol state is non-embryonic, then can't
89 	 * do this directly: have to initiate a PRU_DISCONNECT,
90 	 * which may finish later; embryonic TCB's can just
91 	 * be discarded here.
92 	 */
93 	case PRU_DETACH:
94 		if (tp->t_state > TCPS_LISTEN)
95 			tcp_disconnect(tp);
96 		else {
97 			tcp_close(tp);
98 			tp = 0;
99 		}
100 		break;
101 
102 	/*
103 	 * Give the socket an address.
104 	 */
105 	case PRU_BIND:
106 		error = in_pcbbind(inp, nam);
107 		if (error)
108 			break;
109 		break;
110 
111 	/*
112 	 * Prepare to accept connections.
113 	 */
114 	case PRU_LISTEN:
115 		if (inp->inp_lport == 0)
116 			error = in_pcbbind(inp, (struct mbuf *)0);
117 		if (error == 0)
118 			tp->t_state = TCPS_LISTEN;
119 		break;
120 
121 	/*
122 	 * Initiate connection to peer.
123 	 * Create a template for use in transmissions on this connection.
124 	 * Enter SYN_SENT state, and mark socket as connecting.
125 	 * Start keep-alive timer, and seed output sequence space.
126 	 * Send initial segment on connection.
127 	 */
128 	case PRU_CONNECT:
129 		if (inp->inp_lport == 0) {
130 			error = in_pcbbind(inp, (struct mbuf *)0);
131 			if (error)
132 				break;
133 		}
134 		error = in_pcbconnect(inp, nam);
135 		if (error)
136 			break;
137 		tp->t_template = tcp_template(tp);
138 		if (tp->t_template == 0) {
139 			in_pcbdisconnect(inp);
140 			error = ENOBUFS;
141 			break;
142 		}
143 		soisconnecting(so);
144 		tp->t_state = TCPS_SYN_SENT;
145 		tp->t_timer[TCPT_KEEP] = TCPTV_KEEP;
146 		tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2;
147 		tcp_sendseqinit(tp);
148 		error = tcp_output(tp);
149 		break;
150 
151 	/*
152 	 * Initiate disconnect from peer.
153 	 * If connection never passed embryonic stage, just drop;
154 	 * else if don't need to let data drain, then can just drop anyways,
155 	 * else have to begin TCP shutdown process: mark socket disconnecting,
156 	 * drain unread data, state switch to reflect user close, and
157 	 * send segment (e.g. FIN) to peer.  Socket will be really disconnected
158 	 * when peer sends FIN and acks ours.
159 	 *
160 	 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
161 	 */
162 	case PRU_DISCONNECT:
163 		tcp_disconnect(tp);
164 		break;
165 
166 	/*
167 	 * Accept a connection.  Essentially all the work is
168 	 * done at higher levels; just return the address
169 	 * of the peer, storing through addr.
170 	 */
171 	case PRU_ACCEPT: {
172 		struct sockaddr_in *sin = mtod(nam, struct sockaddr_in *);
173 
174 		nam->m_len = sizeof (struct sockaddr_in);
175 		sin->sin_family = AF_INET;
176 		sin->sin_port = inp->inp_fport;
177 		sin->sin_addr = inp->inp_faddr;
178 		break;
179 		}
180 
181 	/*
182 	 * Mark the connection as being incapable of further output.
183 	 */
184 	case PRU_SHUTDOWN:
185 		socantsendmore(so);
186 		tcp_usrclosed(tp);
187 		error = tcp_output(tp);
188 		break;
189 
190 	/*
191 	 * After a receive, possibly send window update to peer.
192 	 */
193 	case PRU_RCVD:
194 		(void) tcp_output(tp);
195 		break;
196 
197 	/*
198 	 * Do a send by putting data in output queue and updating urgent
199 	 * marker if URG set.  Possibly send more data.
200 	 */
201 	case PRU_SEND:
202 		sbappend(&so->so_snd, m);
203 #ifdef notdef
204 		if (tp->t_flags & TF_PUSH)
205 			tp->snd_end = tp->snd_una + so->so_snd.sb_cc;
206 #endif
207 		error = tcp_output(tp);
208 		break;
209 
210 	/*
211 	 * Abort the TCP.
212 	 */
213 	case PRU_ABORT:
214 		tcp_drop(tp, ECONNABORTED);
215 		break;
216 
217 /* SOME AS YET UNIMPLEMENTED HOOKS */
218 	case PRU_CONTROL:
219 		error = EOPNOTSUPP;
220 		break;
221 
222 	case PRU_SENSE:
223 		error = EOPNOTSUPP;
224 		break;
225 /* END UNIMPLEMENTED HOOKS */
226 
227 	case PRU_RCVOOB:
228 		if (so->so_oobmark == 0 &&
229 		    (so->so_state & SS_RCVATMARK) == 0) {
230 			error = EINVAL;
231 			break;
232 		}
233 		if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
234 			error = EWOULDBLOCK;
235 			break;
236 		}
237 		m->m_len = 1;
238 		*mtod(m, caddr_t) = tp->t_iobc;
239 		break;
240 
241 	case PRU_SENDOOB:
242 		if (sbspace(&so->so_snd) < -512) {
243 			error = ENOBUFS;
244 			break;
245 		}
246 		tp->snd_up = tp->snd_una + so->so_snd.sb_cc + 1;
247 		sbappend(&so->so_snd, m);
248 		tp->t_force = 1;
249 		error = tcp_output(tp);
250 		tp->t_force = 0;
251 		break;
252 
253 	case PRU_SOCKADDR:
254 		in_setsockaddr(inp, nam);
255 		break;
256 
257 	/*
258 	 * TCP slow timer went off; going through this
259 	 * routine for tracing's sake.
260 	 */
261 	case PRU_SLOWTIMO:
262 		tcp_timers(tp, (int)nam);
263 		req |= (int)nam << 8;		/* for debug's sake */
264 		break;
265 
266 	default:
267 		panic("tcp_usrreq");
268 	}
269 	if (tp && (so->so_options & SO_DEBUG))
270 		tcp_trace(TA_USER, ostate, tp, (struct tcpiphdr *)0, req);
271 	splx(s);
272 	return (error);
273 }
274 
275 int	tcp_sendspace = 1024*2;
276 int	tcp_recvspace = 1024*2;
277 /*
278  * Attach TCP protocol to socket, allocating
279  * internet protocol control block, tcp control block,
280  * bufer space, and entering LISTEN state if to accept connections.
281  */
282 tcp_attach(so)
283 	struct socket *so;
284 {
285 	register struct tcpcb *tp;
286 	struct inpcb *inp;
287 	int error;
288 
289 	error = in_pcbreserve(so, tcp_sendspace, tcp_recvspace);
290 	if (error)
291 		goto bad;
292 	error = in_pcballoc(so, &tcb);
293 	if (error)
294 		goto bad;
295 	inp = sotoinpcb(so);
296 	tp = tcp_newtcpcb(inp);
297 	if (tp == 0) {
298 		error = ENOBUFS;
299 		goto bad2;
300 	}
301 	tp->t_state = TCPS_CLOSED;
302 	return (0);
303 bad2:
304 	in_pcbdetach(inp);
305 bad:
306 	return (error);
307 }
308 
309 /*
310  * Initiate (or continue) disconnect.
311  * If embryonic state, just send reset (once).
312  * If not in ``let data drain'' option, just drop.
313  * Otherwise (hard), mark socket disconnecting and drop
314  * current input data; switch states based on user close, and
315  * send segment to peer (with FIN).
316  */
317 tcp_disconnect(tp)
318 	struct tcpcb *tp;
319 {
320 	struct socket *so = tp->t_inpcb->inp_socket;
321 
322 	if (tp->t_state < TCPS_ESTABLISHED)
323 		tcp_close(tp);
324 	else if (so->so_linger == 0)
325 		tcp_drop(tp, 0);
326 	else {
327 		soisdisconnecting(so);
328 		sbflush(&so->so_rcv);
329 		tcp_usrclosed(tp);
330 		(void) tcp_output(tp);
331 	}
332 }
333 
334 /*
335  * User issued close, and wish to trail through shutdown states:
336  * if never received SYN, just forget it.  If got a SYN from peer,
337  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
338  * If already got a FIN from peer, then almost done; go to LAST_ACK
339  * state.  In all other cases, have already sent FIN to peer (e.g.
340  * after PRU_SHUTDOWN), and just have to play tedious game waiting
341  * for peer to send FIN or not respond to keep-alives, etc.
342  * We can let the user exit from the close as soon as the FIN is acked.
343  */
344 tcp_usrclosed(tp)
345 	struct tcpcb *tp;
346 {
347 
348 	switch (tp->t_state) {
349 
350 	case TCPS_LISTEN:
351 	case TCPS_SYN_SENT:
352 		tp->t_state = TCPS_CLOSED;
353 		tcp_close(tp);
354 		break;
355 
356 	case TCPS_SYN_RECEIVED:
357 	case TCPS_ESTABLISHED:
358 		tp->t_state = TCPS_FIN_WAIT_1;
359 		break;
360 
361 	case TCPS_CLOSE_WAIT:
362 		tp->t_state = TCPS_LAST_ACK;
363 		break;
364 	}
365 	if (tp->t_state >= TCPS_FIN_WAIT_2)
366 		soisdisconnected(tp->t_inpcb->inp_socket);
367 }
368