1 /* 2 * Copyright (c) 1982, 1986, 1988 Regents of the University of California. 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms are permitted 6 * provided that the above copyright notice and this paragraph are 7 * duplicated in all such forms and that any documentation, 8 * advertising materials, and other materials related to such 9 * distribution and use acknowledge that the software was developed 10 * by the University of California, Berkeley. The name of the 11 * University may not be used to endorse or promote products derived 12 * from this software without specific prior written permission. 13 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR 14 * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED 15 * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE. 16 * 17 * @(#)tcp_usrreq.c 7.14 (Berkeley) 05/17/90 18 */ 19 20 #include "param.h" 21 #include "systm.h" 22 #include "malloc.h" 23 #include "mbuf.h" 24 #include "socket.h" 25 #include "socketvar.h" 26 #include "protosw.h" 27 #include "errno.h" 28 #include "stat.h" 29 30 #include "../net/if.h" 31 #include "../net/route.h" 32 33 #include "in.h" 34 #include "in_systm.h" 35 #include "ip.h" 36 #include "in_pcb.h" 37 #include "ip_var.h" 38 #include "tcp.h" 39 #include "tcp_fsm.h" 40 #include "tcp_seq.h" 41 #include "tcp_timer.h" 42 #include "tcp_var.h" 43 #include "tcpip.h" 44 #include "tcp_debug.h" 45 46 /* 47 * TCP protocol interface to socket abstraction. 48 */ 49 extern char *tcpstates[]; 50 struct tcpcb *tcp_newtcpcb(); 51 52 /* 53 * Process a TCP user request for TCP tb. If this is a send request 54 * then m is the mbuf chain of send data. If this is a timer expiration 55 * (called from the software clock routine), then timertype tells which timer. 56 */ 57 /*ARGSUSED*/ 58 tcp_usrreq(so, req, m, nam, control) 59 struct socket *so; 60 int req; 61 struct mbuf *m, *nam, *control; 62 { 63 register struct inpcb *inp; 64 register struct tcpcb *tp; 65 int s; 66 int error = 0; 67 int ostate; 68 69 if (req == PRU_CONTROL) 70 return (in_control(so, (int)m, (caddr_t)nam, 71 (struct ifnet *)control)); 72 if (control && control->m_len) { 73 m_freem(control); 74 if (m) 75 m_freem(m); 76 return (EINVAL); 77 } 78 79 s = splnet(); 80 inp = sotoinpcb(so); 81 /* 82 * When a TCP is attached to a socket, then there will be 83 * a (struct inpcb) pointed at by the socket, and this 84 * structure will point at a subsidary (struct tcpcb). 85 */ 86 if (inp == 0 && req != PRU_ATTACH) { 87 splx(s); 88 return (EINVAL); /* XXX */ 89 } 90 if (inp) { 91 tp = intotcpcb(inp); 92 /* WHAT IF TP IS 0? */ 93 #ifdef KPROF 94 tcp_acounts[tp->t_state][req]++; 95 #endif 96 ostate = tp->t_state; 97 } else 98 ostate = 0; 99 switch (req) { 100 101 /* 102 * TCP attaches to socket via PRU_ATTACH, reserving space, 103 * and an internet control block. 104 */ 105 case PRU_ATTACH: 106 if (inp) { 107 error = EISCONN; 108 break; 109 } 110 error = tcp_attach(so); 111 if (error) 112 break; 113 if ((so->so_options & SO_LINGER) && so->so_linger == 0) 114 so->so_linger = TCP_LINGERTIME; 115 tp = sototcpcb(so); 116 break; 117 118 /* 119 * PRU_DETACH detaches the TCP protocol from the socket. 120 * If the protocol state is non-embryonic, then can't 121 * do this directly: have to initiate a PRU_DISCONNECT, 122 * which may finish later; embryonic TCB's can just 123 * be discarded here. 124 */ 125 case PRU_DETACH: 126 if (tp->t_state > TCPS_LISTEN) 127 tp = tcp_disconnect(tp); 128 else 129 tp = tcp_close(tp); 130 break; 131 132 /* 133 * Give the socket an address. 134 */ 135 case PRU_BIND: 136 error = in_pcbbind(inp, nam); 137 if (error) 138 break; 139 break; 140 141 /* 142 * Prepare to accept connections. 143 */ 144 case PRU_LISTEN: 145 if (inp->inp_lport == 0) 146 error = in_pcbbind(inp, (struct mbuf *)0); 147 if (error == 0) 148 tp->t_state = TCPS_LISTEN; 149 break; 150 151 /* 152 * Initiate connection to peer. 153 * Create a template for use in transmissions on this connection. 154 * Enter SYN_SENT state, and mark socket as connecting. 155 * Start keep-alive timer, and seed output sequence space. 156 * Send initial segment on connection. 157 */ 158 case PRU_CONNECT: 159 if (inp->inp_lport == 0) { 160 error = in_pcbbind(inp, (struct mbuf *)0); 161 if (error) 162 break; 163 } 164 error = in_pcbconnect(inp, nam); 165 if (error) 166 break; 167 tp->t_template = tcp_template(tp); 168 if (tp->t_template == 0) { 169 in_pcbdisconnect(inp); 170 error = ENOBUFS; 171 break; 172 } 173 soisconnecting(so); 174 tcpstat.tcps_connattempt++; 175 tp->t_state = TCPS_SYN_SENT; 176 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 177 tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2; 178 tcp_sendseqinit(tp); 179 error = tcp_output(tp); 180 break; 181 182 /* 183 * Create a TCP connection between two sockets. 184 */ 185 case PRU_CONNECT2: 186 error = EOPNOTSUPP; 187 break; 188 189 /* 190 * Initiate disconnect from peer. 191 * If connection never passed embryonic stage, just drop; 192 * else if don't need to let data drain, then can just drop anyways, 193 * else have to begin TCP shutdown process: mark socket disconnecting, 194 * drain unread data, state switch to reflect user close, and 195 * send segment (e.g. FIN) to peer. Socket will be really disconnected 196 * when peer sends FIN and acks ours. 197 * 198 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 199 */ 200 case PRU_DISCONNECT: 201 tp = tcp_disconnect(tp); 202 break; 203 204 /* 205 * Accept a connection. Essentially all the work is 206 * done at higher levels; just return the address 207 * of the peer, storing through addr. 208 */ 209 case PRU_ACCEPT: { 210 struct sockaddr_in *sin = mtod(nam, struct sockaddr_in *); 211 212 nam->m_len = sizeof (struct sockaddr_in); 213 sin->sin_family = AF_INET; 214 sin->sin_len = sizeof(*sin); 215 sin->sin_port = inp->inp_fport; 216 sin->sin_addr = inp->inp_faddr; 217 break; 218 } 219 220 /* 221 * Mark the connection as being incapable of further output. 222 */ 223 case PRU_SHUTDOWN: 224 socantsendmore(so); 225 tp = tcp_usrclosed(tp); 226 if (tp) 227 error = tcp_output(tp); 228 break; 229 230 /* 231 * After a receive, possibly send window update to peer. 232 */ 233 case PRU_RCVD: 234 (void) tcp_output(tp); 235 break; 236 237 /* 238 * Do a send by putting data in output queue and updating urgent 239 * marker if URG set. Possibly send more data. 240 */ 241 case PRU_SEND: 242 sbappend(&so->so_snd, m); 243 error = tcp_output(tp); 244 break; 245 246 /* 247 * Abort the TCP. 248 */ 249 case PRU_ABORT: 250 tp = tcp_drop(tp, ECONNABORTED); 251 break; 252 253 case PRU_SENSE: 254 ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat; 255 (void) splx(s); 256 return (0); 257 258 case PRU_RCVOOB: 259 if ((so->so_oobmark == 0 && 260 (so->so_state & SS_RCVATMARK) == 0) || 261 so->so_options & SO_OOBINLINE || 262 tp->t_oobflags & TCPOOB_HADDATA) { 263 error = EINVAL; 264 break; 265 } 266 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 267 error = EWOULDBLOCK; 268 break; 269 } 270 m->m_len = 1; 271 *mtod(m, caddr_t) = tp->t_iobc; 272 if (((int)nam & MSG_PEEK) == 0) 273 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 274 break; 275 276 case PRU_SENDOOB: 277 if (sbspace(&so->so_snd) < -512) { 278 m_freem(m); 279 error = ENOBUFS; 280 break; 281 } 282 /* 283 * According to RFC961 (Assigned Protocols), 284 * the urgent pointer points to the last octet 285 * of urgent data. We continue, however, 286 * to consider it to indicate the first octet 287 * of data past the urgent section. 288 * Otherwise, snd_up should be one lower. 289 */ 290 sbappend(&so->so_snd, m); 291 tp->snd_up = tp->snd_una + so->so_snd.sb_cc; 292 tp->t_force = 1; 293 error = tcp_output(tp); 294 tp->t_force = 0; 295 break; 296 297 case PRU_SOCKADDR: 298 in_setsockaddr(inp, nam); 299 break; 300 301 case PRU_PEERADDR: 302 in_setpeeraddr(inp, nam); 303 break; 304 305 /* 306 * TCP slow timer went off; going through this 307 * routine for tracing's sake. 308 */ 309 case PRU_SLOWTIMO: 310 tp = tcp_timers(tp, (int)nam); 311 req |= (int)nam << 8; /* for debug's sake */ 312 break; 313 314 default: 315 panic("tcp_usrreq"); 316 } 317 if (tp && (so->so_options & SO_DEBUG)) 318 tcp_trace(TA_USER, ostate, tp, (struct tcpiphdr *)0, req); 319 splx(s); 320 return (error); 321 } 322 323 tcp_ctloutput(op, so, level, optname, mp) 324 int op; 325 struct socket *so; 326 int level, optname; 327 struct mbuf **mp; 328 { 329 int error = 0; 330 struct inpcb *inp = sotoinpcb(so); 331 register struct tcpcb *tp = intotcpcb(inp); 332 register struct mbuf *m; 333 334 if (level != IPPROTO_TCP) 335 return (ip_ctloutput(op, so, level, optname, mp)); 336 337 switch (op) { 338 339 case PRCO_SETOPT: 340 m = *mp; 341 switch (optname) { 342 343 case TCP_NODELAY: 344 if (m == NULL || m->m_len < sizeof (int)) 345 error = EINVAL; 346 else if (*mtod(m, int *)) 347 tp->t_flags |= TF_NODELAY; 348 else 349 tp->t_flags &= ~TF_NODELAY; 350 break; 351 352 case TCP_MAXSEG: /* not yet */ 353 default: 354 error = EINVAL; 355 break; 356 } 357 if (m) 358 (void) m_free(m); 359 break; 360 361 case PRCO_GETOPT: 362 *mp = m = m_get(M_WAIT, MT_SOOPTS); 363 m->m_len = sizeof(int); 364 365 switch (optname) { 366 case TCP_NODELAY: 367 *mtod(m, int *) = tp->t_flags & TF_NODELAY; 368 break; 369 case TCP_MAXSEG: 370 *mtod(m, int *) = tp->t_maxseg; 371 break; 372 default: 373 error = EINVAL; 374 break; 375 } 376 break; 377 } 378 return (error); 379 } 380 381 u_long tcp_sendspace = 1024*4; 382 u_long tcp_recvspace = 1024*4; 383 384 /* 385 * Attach TCP protocol to socket, allocating 386 * internet protocol control block, tcp control block, 387 * bufer space, and entering LISTEN state if to accept connections. 388 */ 389 tcp_attach(so) 390 struct socket *so; 391 { 392 register struct tcpcb *tp; 393 struct inpcb *inp; 394 int error; 395 396 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 397 error = soreserve(so, tcp_sendspace, tcp_recvspace); 398 if (error) 399 return (error); 400 } 401 error = in_pcballoc(so, &tcb); 402 if (error) 403 return (error); 404 inp = sotoinpcb(so); 405 tp = tcp_newtcpcb(inp); 406 if (tp == 0) { 407 int nofd = so->so_state & SS_NOFDREF; /* XXX */ 408 409 so->so_state &= ~SS_NOFDREF; /* don't free the socket yet */ 410 in_pcbdetach(inp); 411 so->so_state |= nofd; 412 return (ENOBUFS); 413 } 414 tp->t_state = TCPS_CLOSED; 415 return (0); 416 } 417 418 /* 419 * Initiate (or continue) disconnect. 420 * If embryonic state, just send reset (once). 421 * If in ``let data drain'' option and linger null, just drop. 422 * Otherwise (hard), mark socket disconnecting and drop 423 * current input data; switch states based on user close, and 424 * send segment to peer (with FIN). 425 */ 426 struct tcpcb * 427 tcp_disconnect(tp) 428 register struct tcpcb *tp; 429 { 430 struct socket *so = tp->t_inpcb->inp_socket; 431 432 if (tp->t_state < TCPS_ESTABLISHED) 433 tp = tcp_close(tp); 434 else if ((so->so_options & SO_LINGER) && so->so_linger == 0) 435 tp = tcp_drop(tp, 0); 436 else { 437 soisdisconnecting(so); 438 sbflush(&so->so_rcv); 439 tp = tcp_usrclosed(tp); 440 if (tp) 441 (void) tcp_output(tp); 442 } 443 return (tp); 444 } 445 446 /* 447 * User issued close, and wish to trail through shutdown states: 448 * if never received SYN, just forget it. If got a SYN from peer, 449 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 450 * If already got a FIN from peer, then almost done; go to LAST_ACK 451 * state. In all other cases, have already sent FIN to peer (e.g. 452 * after PRU_SHUTDOWN), and just have to play tedious game waiting 453 * for peer to send FIN or not respond to keep-alives, etc. 454 * We can let the user exit from the close as soon as the FIN is acked. 455 */ 456 struct tcpcb * 457 tcp_usrclosed(tp) 458 register struct tcpcb *tp; 459 { 460 461 switch (tp->t_state) { 462 463 case TCPS_CLOSED: 464 case TCPS_LISTEN: 465 case TCPS_SYN_SENT: 466 tp->t_state = TCPS_CLOSED; 467 tp = tcp_close(tp); 468 break; 469 470 case TCPS_SYN_RECEIVED: 471 case TCPS_ESTABLISHED: 472 tp->t_state = TCPS_FIN_WAIT_1; 473 break; 474 475 case TCPS_CLOSE_WAIT: 476 tp->t_state = TCPS_LAST_ACK; 477 break; 478 } 479 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) 480 soisdisconnected(tp->t_inpcb->inp_socket); 481 return (tp); 482 } 483