xref: /csrg-svn/sys/netinet/tcp_usrreq.c (revision 5953)
1 /* tcp_usrreq.c 1.51 82/02/25 */
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 "../net/in.h"
10 #include "../net/in_pcb.h"
11 #include "../net/in_systm.h"
12 #include "../net/if.h"
13 #include "../net/ip.h"
14 #include "../net/ip_var.h"
15 #include "../net/tcp.h"
16 #include "../net/tcp_fsm.h"
17 #include "../net/tcp_seq.h"
18 #include "../net/tcp_timer.h"
19 #include "../net/tcp_var.h"
20 #include "../net/tcpip.h"
21 #include "../net/tcp_debug.h"
22 #include "../errno.h"
23 
24 /*
25  * TCP protocol interface to socket abstraction.
26  */
27 extern	char *tcpstates[];
28 struct	tcpcb *tcp_newtcpcb();
29 
30 /*
31  * Process a TCP user request for TCP tb.  If this is a send request
32  * then m is the mbuf chain of send data.  If this is a timer expiration
33  * (called from the software clock routine), then timertype tells which timer.
34  */
35 tcp_usrreq(so, req, m, addr)
36 	struct socket *so;
37 	int req;
38 	struct mbuf *m;
39 	caddr_t addr;
40 {
41 	register struct inpcb *inp = sotoinpcb(so);
42 	register struct tcpcb *tp;
43 	int s = splnet();
44 	int error = 0;
45 	int ostate;
46 COUNT(TCP_USRREQ);
47 
48 	/*
49 	 * When a TCP is attached to a socket, then there will be
50 	 * a (struct inpcb) pointed at by the socket, and this
51 	 * structure will point at a subsidary (struct tcpcb).
52 	 * The normal sequence of events is:
53 	 *	PRU_ATTACH		creating these structures
54 	 *	PRU_CONNECT		connecting to a remote peer
55 	 *	(PRU_SEND|PRU_RCVD)*	exchanging data
56 	 *	PRU_DISCONNECT		disconnecting from remote peer
57 	 *	PRU_DETACH		deleting the structures
58 	 * With the operations from PRU_CONNECT through PRU_DISCONNECT
59 	 * possible repeated several times.
60 	 *
61 	 * MULTIPLE CONNECTS ARE NOT YET IMPLEMENTED.
62 	 */
63 	if (inp == 0 && req != PRU_ATTACH) {
64 		splx(s);
65 		return (EINVAL);		/* XXX */
66 	}
67 	if (inp) {
68 		tp = intotcpcb(inp);
69 #ifdef KPROF
70 		tcp_acounts[tp->t_state][req]++;
71 #endif
72 		ostate = tp->t_state;
73 	}
74 	switch (req) {
75 
76 	/*
77 	 * TCP attaches to socket via PRU_ATTACH, reserving space,
78 	 * and internet and TCP control blocks.
79 	 * If the socket is to receive connections,
80 	 * then the LISTEN state is entered.
81 	 */
82 	case PRU_ATTACH:
83 		if (inp) {
84 			error = EISCONN;
85 			break;
86 		}
87 		error = tcp_attach(so, (struct sockaddr *)addr);
88 		if (error)
89 			break;
90 		if ((so->so_options & SO_DONTLINGER) == 0)
91 			so->so_linger = TCP_LINGERTIME;
92 		tp = sototcpcb(so);
93 		break;
94 
95 	/*
96 	 * PRU_DETACH detaches the TCP protocol from the socket.
97 	 * If the protocol state is non-embryonic, then can't
98 	 * do this directly: have to initiate a PRU_DISCONNECT,
99 	 * which may finish later; embryonic TCB's can just
100 	 * be discarded here.
101 	 */
102 	case PRU_DETACH:
103 		if (tp->t_state > TCPS_LISTEN)
104 			tcp_disconnect(tp);
105 		else {
106 			tcp_close(tp);
107 			tp = 0;
108 		}
109 		break;
110 
111 	/*
112 	 * Initiate connection to peer.
113 	 * Create a template for use in transmissions on this connection.
114 	 * Enter SYN_SENT state, and mark socket as connecting.
115 	 * Start keep-alive timer, and seed output sequence space.
116 	 * Send initial segment on connection.
117 	 */
118 	case PRU_CONNECT:
119 		error = in_pcbconnect(inp, (struct sockaddr_in *)addr);
120 		if (error)
121 			break;
122 		tp->t_template = tcp_template(tp);
123 		if (tp->t_template == 0) {
124 			in_pcbdisconnect(inp);
125 			error = ENOBUFS;
126 			break;
127 		}
128 		soisconnecting(so);
129 		tp->t_state = TCPS_SYN_SENT;
130 		tp->t_timer[TCPT_KEEP] = TCPTV_KEEP;
131 		tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2;
132 		tcp_sendseqinit(tp);
133 		(void) tcp_output(tp);
134 		break;
135 
136 	/*
137 	 * Initiate disconnect from peer.
138 	 * If connection never passed embryonic stage, just drop;
139 	 * else if don't need to let data drain, then can just drop anyways,
140 	 * else have to begin TCP shutdown process: mark socket disconnecting,
141 	 * drain unread data, state switch to reflect user close, and
142 	 * send segment (e.g. FIN) to peer.  Socket will be really disconnected
143 	 * when peer sends FIN and acks ours.
144 	 *
145 	 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
146 	 */
147 	case PRU_DISCONNECT:
148 		tcp_disconnect(tp);
149 		break;
150 
151 	/*
152 	 * Accept a connection.  Essentially all the work is
153 	 * done at higher levels; just return the address
154 	 * of the peer, storing through addr.
155 	 */
156 	case PRU_ACCEPT:
157 		in_pcbconnaddr(inp, (struct sockaddr *)addr);
158 		break;
159 
160 	/*
161 	 * Mark the connection as being incapable of further output.
162 	 */
163 	case PRU_SHUTDOWN:
164 		socantsendmore(so);
165 		tcp_usrclosed(tp);
166 		(void) tcp_output(tp);
167 		break;
168 
169 	/*
170 	 * After a receive, possibly send window update to peer.
171 	 */
172 	case PRU_RCVD:
173 		(void) tcp_output(tp);
174 		break;
175 
176 	/*
177 	 * Do a send by putting data in output queue and updating urgent
178 	 * marker if URG set.  Possibly send more data.
179 	 */
180 	case PRU_SEND:
181 		sbappend(&so->so_snd, m);
182 /*
183 		if (tp->t_flags & TF_PUSH)
184 			tp->snd_end = tp->snd_una + so->so_snd.sb_cc;
185  */
186 		(void) tcp_output(tp);
187 		break;
188 
189 	/*
190 	 * Abort the TCP.
191 	 */
192 	case PRU_ABORT:
193 		tcp_drop(tp, ECONNABORTED);
194 		break;
195 
196 /* SOME AS YET UNIMPLEMENTED HOOKS */
197 	case PRU_CONTROL:
198 		error = EOPNOTSUPP;
199 		break;
200 
201 	case PRU_SENSE:
202 		error = EOPNOTSUPP;
203 		break;
204 /* END UNIMPLEMENTED HOOKS */
205 
206 	case PRU_RCVOOB:
207 		if (so->so_oobmark == 0 &&
208 		    (so->so_state & SS_RCVATMARK) == 0) {
209 			error = EINVAL;
210 			break;
211 		}
212 		if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
213 			error = EWOULDBLOCK;
214 			break;
215 		}
216 		*mtod(m, caddr_t) = tp->t_iobc;
217 		break;
218 
219 	case PRU_SENDOOB:
220 #ifdef TCPTRUEOOB
221 		if (tp->t_flags & TF_DOOOB) {
222 			tp->t_oobseq++;
223 			tp->t_oobc = *mtod(m, caddr_t);
224 			tp->t_oobmark = tp->snd_una + so->so_snd.sb_cc;
225 printf("sendoob seq now %x oobc %x\n", tp->t_oobseq, tp->t_oobc);
226 			tp->t_oobflags |= TCPOOB_NEEDACK;
227 			(void) tcp_output(tp);
228 		}
229 #endif
230 		if (sbspace(&so->so_snd) < -512) {
231 			error = ENOBUFS;
232 			break;
233 		}
234 		tp->snd_up = tp->snd_una + so->so_snd.sb_cc + 1;
235 		sbappend(&so->so_snd, m);
236 /*
237 		if (tp->t_flags & TF_PUSH)
238 			tp->snd_end = tp->snd_una + so->so_snd.sb_cc;
239  */
240 		tp->t_force = 1;
241 		(void) tcp_output(tp);
242 		tp->t_force = 0;
243 		break;
244 
245 	/*
246 	 * TCP slow timer went off; going through this
247 	 * routine for tracing's sake.
248 	 */
249 	case PRU_SLOWTIMO:
250 		tcp_timers(tp, (int)addr);
251 		req |= (int)addr << 8;		/* for debug's sake */
252 		break;
253 
254 	default:
255 		panic("tcp_usrreq");
256 	}
257 	if (tp && (so->so_options & SO_DEBUG))
258 		tcp_trace(TA_USER, ostate, tp, (struct tcpiphdr *)0, req);
259 	splx(s);
260 	return (error);
261 }
262 
263 int	tcp_sendspace = 1024*2;
264 int	tcp_recvspace = 1024*3;
265 /*
266  * Attach TCP protocol to socket, allocating
267  * internet protocol control block, tcp control block,
268  * bufer space, and entering LISTEN state if to accept connections.
269  */
270 tcp_attach(so, sa)
271 	struct socket *so;
272 	struct sockaddr *sa;
273 {
274 	register struct tcpcb *tp;
275 	struct inpcb *inp;
276 	int error;
277 
278 	error = in_pcbattach(so, &tcb,
279 	    tcp_sendspace, tcp_recvspace, (struct sockaddr_in *)sa);
280 	if (error)
281 		return (error);
282 	inp = (struct inpcb *)so->so_pcb;
283 	tp = tcp_newtcpcb(inp);
284 	if (so->so_options & SO_ACCEPTCONN) {
285 		if (tp == 0) {
286 			in_pcbdetach(inp);
287 			return (ENOBUFS);
288 		}
289 		tp->t_state = TCPS_LISTEN;
290 	} else
291 		tp->t_state = TCPS_CLOSED;
292 	return (0);
293 }
294 
295 /*
296  * Initiate (or continue) disconnect.
297  * If embryonic state, just send reset (once).
298  * If not in ``let data drain'' option, just drop.
299  * Otherwise (hard), mark socket disconnecting and drop
300  * current input data; switch states based on user close, and
301  * send segment to peer (with FIN).
302  */
303 tcp_disconnect(tp)
304 	struct tcpcb *tp;
305 {
306 	struct socket *so = tp->t_inpcb->inp_socket;
307 
308 	if (tp->t_state < TCPS_ESTABLISHED)
309 		tcp_close(tp);
310 	else if (so->so_linger == 0)
311 		tcp_drop(tp, 0);
312 	else {
313 		soisdisconnecting(so);
314 		sbflush(&so->so_rcv);
315 		tcp_usrclosed(tp);
316 		(void) tcp_output(tp);
317 	}
318 }
319 
320 /*
321  * User issued close, and wish to trail through shutdown states:
322  * if never received SYN, just forget it.  If got a SYN from peer,
323  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
324  * If already got a FIN from peer, then almost done; go to LAST_ACK
325  * state.  In all other cases, have already sent FIN to peer (e.g.
326  * after PRU_SHUTDOWN), and just have to play tedious game waiting
327  * for peer to send FIN or not respond to keep-alives, etc.
328  * We can let the user exit from the close as soon as the FIN is acked.
329  */
330 tcp_usrclosed(tp)
331 	struct tcpcb *tp;
332 {
333 
334 	switch (tp->t_state) {
335 
336 	case TCPS_LISTEN:
337 	case TCPS_SYN_SENT:
338 		tp->t_state = TCPS_CLOSED;
339 		tcp_close(tp);
340 		break;
341 
342 	case TCPS_SYN_RECEIVED:
343 	case TCPS_ESTABLISHED:
344 		tp->t_state = TCPS_FIN_WAIT_1;
345 		break;
346 
347 	case TCPS_CLOSE_WAIT:
348 		tp->t_state = TCPS_LAST_ACK;
349 		break;
350 	}
351 	if (tp->t_state >= TCPS_FIN_WAIT_2)
352 		soisdisconnected(tp->t_inpcb->inp_socket);
353 }
354