longer TTL's
[unix-history] / usr / src / sys / netinet / tcp_usrreq.c
CommitLineData
8ae0e4b4
KM
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 *
8011f5df 6 * @(#)tcp_usrreq.c 6.10 (Berkeley) %G%
8ae0e4b4 7 */
72f24d7d 8
20666ad3
JB
9#include "param.h"
10#include "systm.h"
11#include "mbuf.h"
12#include "socket.h"
13#include "socketvar.h"
14#include "protosw.h"
15#include "errno.h"
16#include "stat.h"
6e7edb25
BJ
17
18#include "../net/if.h"
c124e997 19#include "../net/route.h"
f4d55810 20
20666ad3
JB
21#include "in.h"
22#include "in_pcb.h"
23#include "in_systm.h"
24#include "ip.h"
25#include "ip_var.h"
26#include "tcp.h"
27#include "tcp_fsm.h"
28#include "tcp_seq.h"
29#include "tcp_timer.h"
30#include "tcp_var.h"
31#include "tcpip.h"
32#include "tcp_debug.h"
eee3ab16 33
290e0b0a
BJ
34/*
35 * TCP protocol interface to socket abstraction.
36 */
37extern char *tcpstates[];
4ad99bae 38struct tcpcb *tcp_newtcpcb();
ab85b059 39int tcpsenderrors;
290e0b0a 40
9c5022e3 41/*
290e0b0a 42 * Process a TCP user request for TCP tb. If this is a send request
9c5022e3
BJ
43 * then m is the mbuf chain of send data. If this is a timer expiration
44 * (called from the software clock routine), then timertype tells which timer.
45 */
a8d3bf7f 46/*ARGSUSED*/
ab85b059 47tcp_usrreq(so, req, m, nam, rights)
eee3ab16
BJ
48 struct socket *so;
49 int req;
ab85b059 50 struct mbuf *m, *nam, *rights;
4eb5d593 51{
53a5409e 52 register struct inpcb *inp = sotoinpcb(so);
cdad2eb1 53 register struct tcpcb *tp;
72f24d7d 54 int s = splnet();
eee3ab16 55 int error = 0;
17b82ed4 56 int ostate;
72f24d7d 57
1d14d351
MK
58 if (req == PRU_CONTROL)
59 return (in_control(so, (int)m, (caddr_t)nam,
60 (struct ifnet *)rights));
ab85b059
SL
61 if (rights && rights->m_len) {
62 splx(s);
63 return (EINVAL);
64 }
53a5409e 65 /*
290e0b0a
BJ
66 * When a TCP is attached to a socket, then there will be
67 * a (struct inpcb) pointed at by the socket, and this
68 * structure will point at a subsidary (struct tcpcb).
53a5409e 69 */
0974b45c 70 if (inp == 0 && req != PRU_ATTACH) {
a6503abf 71 splx(s);
290e0b0a 72 return (EINVAL); /* XXX */
a6503abf
BJ
73 }
74 if (inp) {
cdad2eb1 75 tp = intotcpcb(inp);
8075bb0e 76 /* WHAT IF TP IS 0? */
9c5022e3 77#ifdef KPROF
a6503abf 78 tcp_acounts[tp->t_state][req]++;
9c5022e3 79#endif
17b82ed4 80 ostate = tp->t_state;
ebf42a75
BJ
81 } else
82 ostate = 0;
eee3ab16 83 switch (req) {
4eb5d593 84
290e0b0a
BJ
85 /*
86 * TCP attaches to socket via PRU_ATTACH, reserving space,
8075bb0e 87 * and an internet control block.
290e0b0a 88 */
eee3ab16 89 case PRU_ATTACH:
4ad99bae 90 if (inp) {
eee3ab16 91 error = EISCONN;
cdad2eb1 92 break;
53a5409e 93 }
a1edc12b 94 error = tcp_attach(so);
a6503abf 95 if (error)
4ad99bae 96 break;
0e3936fa 97 if ((so->so_options & SO_LINGER) && so->so_linger == 0)
8e65fd66 98 so->so_linger = TCP_LINGERTIME;
290e0b0a 99 tp = sototcpcb(so);
72f24d7d 100 break;
4eb5d593 101
290e0b0a
BJ
102 /*
103 * PRU_DETACH detaches the TCP protocol from the socket.
104 * If the protocol state is non-embryonic, then can't
105 * do this directly: have to initiate a PRU_DISCONNECT,
106 * which may finish later; embryonic TCB's can just
107 * be discarded here.
108 */
eee3ab16 109 case PRU_DETACH:
290e0b0a 110 if (tp->t_state > TCPS_LISTEN)
0e3936fa
SL
111 tp = tcp_disconnect(tp);
112 else
113 tp = tcp_close(tp);
eee3ab16
BJ
114 break;
115
8075bb0e
BJ
116 /*
117 * Give the socket an address.
118 */
119 case PRU_BIND:
120 error = in_pcbbind(inp, nam);
121 if (error)
122 break;
123 break;
124
125 /*
126 * Prepare to accept connections.
127 */
128 case PRU_LISTEN:
129 if (inp->inp_lport == 0)
130 error = in_pcbbind(inp, (struct mbuf *)0);
131 if (error == 0)
132 tp->t_state = TCPS_LISTEN;
133 break;
134
290e0b0a
BJ
135 /*
136 * Initiate connection to peer.
137 * Create a template for use in transmissions on this connection.
138 * Enter SYN_SENT state, and mark socket as connecting.
139 * Start keep-alive timer, and seed output sequence space.
140 * Send initial segment on connection.
141 */
eee3ab16 142 case PRU_CONNECT:
8075bb0e
BJ
143 if (inp->inp_lport == 0) {
144 error = in_pcbbind(inp, (struct mbuf *)0);
145 if (error)
146 break;
147 }
148 error = in_pcbconnect(inp, nam);
4ad99bae 149 if (error)
53a5409e 150 break;
b454c3ea 151 tp->t_template = tcp_template(tp);
290e0b0a
BJ
152 if (tp->t_template == 0) {
153 in_pcbdisconnect(inp);
154 error = ENOBUFS;
155 break;
156 }
53a5409e 157 soisconnecting(so);
a6503abf 158 tp->t_state = TCPS_SYN_SENT;
4aed14e3
BJ
159 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP;
160 tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2;
161 tcp_sendseqinit(tp);
8a2f82db 162 error = tcp_output(tp);
72f24d7d 163 break;
4eb5d593 164
4945768c
SL
165 /*
166 * Create a TCP connection between two sockets.
167 */
168 case PRU_CONNECT2:
169 error = EOPNOTSUPP;
170 break;
171
290e0b0a
BJ
172 /*
173 * Initiate disconnect from peer.
174 * If connection never passed embryonic stage, just drop;
175 * else if don't need to let data drain, then can just drop anyways,
176 * else have to begin TCP shutdown process: mark socket disconnecting,
177 * drain unread data, state switch to reflect user close, and
178 * send segment (e.g. FIN) to peer. Socket will be really disconnected
179 * when peer sends FIN and acks ours.
180 *
181 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
182 */
183 case PRU_DISCONNECT:
0e3936fa 184 tp = tcp_disconnect(tp);
4aed14e3
BJ
185 break;
186
290e0b0a
BJ
187 /*
188 * Accept a connection. Essentially all the work is
189 * done at higher levels; just return the address
190 * of the peer, storing through addr.
191 */
1acff8ec 192 case PRU_ACCEPT: {
8075bb0e 193 struct sockaddr_in *sin = mtod(nam, struct sockaddr_in *);
1acff8ec 194
8075bb0e
BJ
195 nam->m_len = sizeof (struct sockaddr_in);
196 sin->sin_family = AF_INET;
197 sin->sin_port = inp->inp_fport;
198 sin->sin_addr = inp->inp_faddr;
eee3ab16 199 break;
8075bb0e 200 }
eee3ab16 201
290e0b0a
BJ
202 /*
203 * Mark the connection as being incapable of further output.
204 */
eee3ab16 205 case PRU_SHUTDOWN:
0974b45c 206 socantsendmore(so);
0e3936fa
SL
207 tp = tcp_usrclosed(tp);
208 if (tp)
209 error = tcp_output(tp);
72f24d7d
BJ
210 break;
211
290e0b0a
BJ
212 /*
213 * After a receive, possibly send window update to peer.
214 */
eee3ab16 215 case PRU_RCVD:
f1b2fa5b 216 (void) tcp_output(tp);
72f24d7d
BJ
217 break;
218
290e0b0a
BJ
219 /*
220 * Do a send by putting data in output queue and updating urgent
221 * marker if URG set. Possibly send more data.
222 */
eee3ab16 223 case PRU_SEND:
a6503abf 224 sbappend(&so->so_snd, m);
8a2f82db 225#ifdef notdef
0974b45c 226 if (tp->t_flags & TF_PUSH)
a6503abf 227 tp->snd_end = tp->snd_una + so->so_snd.sb_cc;
8a2f82db
SL
228#endif
229 error = tcp_output(tp);
ab85b059
SL
230 if (error) { /* XXX fix to use other path */
231 if (error == ENOBUFS) /* XXX */
232 error = 0; /* XXX */
233 tcpsenderrors++;
234 }
72f24d7d
BJ
235 break;
236
290e0b0a
BJ
237 /*
238 * Abort the TCP.
239 */
eee3ab16 240 case PRU_ABORT:
0e3936fa 241 tp = tcp_drop(tp, ECONNABORTED);
72f24d7d
BJ
242 break;
243
f1b2fa5b 244 case PRU_SENSE:
74040e68
MK
245 ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat;
246 return (0);
f1b2fa5b
BJ
247
248 case PRU_RCVOOB:
01234a7d
MK
249 if ((so->so_oobmark == 0 &&
250 (so->so_state & SS_RCVATMARK) == 0) ||
251 tp->t_oobflags & TCPOOB_HADDATA) {
0244dbc7
BJ
252 error = EINVAL;
253 break;
254 }
b2db9217 255 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
8b5a83bb 256 error = EWOULDBLOCK;
b2db9217 257 break;
8b5a83bb 258 }
283ea225 259 m->m_len = 1;
b2db9217 260 *mtod(m, caddr_t) = tp->t_iobc;
01234a7d
MK
261 if (((int)nam & MSG_PEEK) == 0)
262 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
f1b2fa5b
BJ
263 break;
264
265 case PRU_SENDOOB:
8b5a83bb 266 if (sbspace(&so->so_snd) < -512) {
37279c1b 267 m_freem(m);
8b5a83bb
BJ
268 error = ENOBUFS;
269 break;
270 }
0244dbc7
BJ
271 tp->snd_up = tp->snd_una + so->so_snd.sb_cc + 1;
272 sbappend(&so->so_snd, m);
b2db9217 273 tp->t_force = 1;
8a2f82db 274 error = tcp_output(tp);
b2db9217 275 tp->t_force = 0;
f1b2fa5b
BJ
276 break;
277
126472ab 278 case PRU_SOCKADDR:
8075bb0e 279 in_setsockaddr(inp, nam);
126472ab
SL
280 break;
281
a7343092
SL
282 case PRU_PEERADDR:
283 in_setpeeraddr(inp, nam);
284 break;
285
290e0b0a
BJ
286 /*
287 * TCP slow timer went off; going through this
288 * routine for tracing's sake.
289 */
eee3ab16 290 case PRU_SLOWTIMO:
0e3936fa 291 tp = tcp_timers(tp, (int)nam);
8075bb0e 292 req |= (int)nam << 8; /* for debug's sake */
eee3ab16
BJ
293 break;
294
9c5022e3
BJ
295 default:
296 panic("tcp_usrreq");
72f24d7d 297 }
17b82ed4
BJ
298 if (tp && (so->so_options & SO_DEBUG))
299 tcp_trace(TA_USER, ostate, tp, (struct tcpiphdr *)0, req);
72f24d7d 300 splx(s);
53a5409e 301 return (error);
4eb5d593 302}
4aed14e3 303
54c84456 304tcp_ctloutput(op, so, level, optname, mp)
01234a7d
MK
305 int op;
306 struct socket *so;
307 int level, optname;
54c84456 308 struct mbuf **mp;
01234a7d 309{
54c84456
MK
310 int error = 0;
311 struct inpcb *inp = sotoinpcb(so);
312 register struct tcpcb *tp = intotcpcb(inp);
313 register struct mbuf *m;
314
01234a7d 315 if (level != IPPROTO_TCP)
b2a3d559 316 return (ip_ctloutput(op, so, level, optname, mp));
54c84456
MK
317
318 switch (op) {
319
320 case PRCO_SETOPT:
321 m = *mp;
322 switch (optname) {
323
324 case TCP_NODELAY:
325 if (m == NULL || m->m_len < sizeof (int))
326 error = EINVAL;
327 else if (*mtod(m, int *))
328 tp->t_flags |= TF_NODELAY;
329 else
330 tp->t_flags &= ~TF_NODELAY;
331 break;
332
333 case TCP_MAXSEG: /* not yet */
334 default:
335 error = EINVAL;
336 break;
337 }
8011f5df 338 (void)m_free(m);
54c84456
MK
339 break;
340
341 case PRCO_GETOPT:
342 *mp = m = m_get(M_WAIT, MT_SOOPTS);
343 m->m_len = sizeof(int);
344
345 switch (optname) {
346 case TCP_NODELAY:
347 *mtod(m, int *) = tp->t_flags & TF_NODELAY;
348 break;
349 case TCP_MAXSEG:
350 *mtod(m, int *) = tp->t_maxseg;
351 break;
352 default:
353 error = EINVAL;
354 break;
355 }
356 break;
357 }
358 return (error);
01234a7d
MK
359}
360
1d14d351
MK
361int tcp_sendspace = 1024*4;
362int tcp_recvspace = 1024*4;
290e0b0a
BJ
363/*
364 * Attach TCP protocol to socket, allocating
365 * internet protocol control block, tcp control block,
366 * bufer space, and entering LISTEN state if to accept connections.
367 */
8075bb0e 368tcp_attach(so)
290e0b0a 369 struct socket *so;
290e0b0a
BJ
370{
371 register struct tcpcb *tp;
372 struct inpcb *inp;
373 int error;
374
59965020 375 error = soreserve(so, tcp_sendspace, tcp_recvspace);
ebf42a75 376 if (error)
054054fd 377 return (error);
ebf42a75 378 error = in_pcballoc(so, &tcb);
290e0b0a 379 if (error)
054054fd 380 return (error);
8075bb0e 381 inp = sotoinpcb(so);
290e0b0a 382 tp = tcp_newtcpcb(inp);
ebf42a75 383 if (tp == 0) {
054054fd
MK
384 int nofd = so->so_state & SS_NOFDREF; /* XXX */
385
386 so->so_state &= ~SS_NOFDREF; /* don't free the socket yet */
387 in_pcbdetach(inp);
388 so->so_state |= nofd;
389 return (ENOBUFS);
ebf42a75 390 }
8075bb0e 391 tp->t_state = TCPS_CLOSED;
290e0b0a
BJ
392 return (0);
393}
394
395/*
396 * Initiate (or continue) disconnect.
397 * If embryonic state, just send reset (once).
f9e4ec68 398 * If in ``let data drain'' option and linger null, just drop.
290e0b0a
BJ
399 * Otherwise (hard), mark socket disconnecting and drop
400 * current input data; switch states based on user close, and
401 * send segment to peer (with FIN).
402 */
0e3936fa 403struct tcpcb *
290e0b0a 404tcp_disconnect(tp)
0e3936fa 405 register struct tcpcb *tp;
290e0b0a
BJ
406{
407 struct socket *so = tp->t_inpcb->inp_socket;
408
409 if (tp->t_state < TCPS_ESTABLISHED)
0e3936fa 410 tp = tcp_close(tp);
f9e4ec68 411 else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
0e3936fa 412 tp = tcp_drop(tp, 0);
290e0b0a
BJ
413 else {
414 soisdisconnecting(so);
415 sbflush(&so->so_rcv);
0e3936fa
SL
416 tp = tcp_usrclosed(tp);
417 if (tp)
418 (void) tcp_output(tp);
290e0b0a 419 }
0e3936fa 420 return (tp);
290e0b0a
BJ
421}
422
423/*
424 * User issued close, and wish to trail through shutdown states:
425 * if never received SYN, just forget it. If got a SYN from peer,
426 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
427 * If already got a FIN from peer, then almost done; go to LAST_ACK
428 * state. In all other cases, have already sent FIN to peer (e.g.
429 * after PRU_SHUTDOWN), and just have to play tedious game waiting
430 * for peer to send FIN or not respond to keep-alives, etc.
085a0b90 431 * We can let the user exit from the close as soon as the FIN is acked.
290e0b0a 432 */
0e3936fa 433struct tcpcb *
4aed14e3 434tcp_usrclosed(tp)
0e3936fa 435 register struct tcpcb *tp;
4aed14e3
BJ
436{
437
4aed14e3
BJ
438 switch (tp->t_state) {
439
815b24e1 440 case TCPS_CLOSED:
4aed14e3
BJ
441 case TCPS_LISTEN:
442 case TCPS_SYN_SENT:
443 tp->t_state = TCPS_CLOSED;
0e3936fa 444 tp = tcp_close(tp);
4aed14e3
BJ
445 break;
446
447 case TCPS_SYN_RECEIVED:
448 case TCPS_ESTABLISHED:
449 tp->t_state = TCPS_FIN_WAIT_1;
450 break;
451
452 case TCPS_CLOSE_WAIT:
453 tp->t_state = TCPS_LAST_ACK;
454 break;
455 }
0e3936fa 456 if (tp && tp->t_state >= TCPS_FIN_WAIT_2)
085a0b90 457 soisdisconnected(tp->t_inpcb->inp_socket);
0e3936fa 458 return (tp);
4aed14e3 459}