1 /* 2 * Copyright (c) 1982, 1986 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 this notice is preserved and that due credit is given 7 * to the University of California at Berkeley. The name of the University 8 * may not be used to endorse or promote products derived from this 9 * software without specific prior written permission. This software 10 * is provided ``as is'' without express or implied warranty. 11 * 12 * @(#)tcp_usrreq.c 7.7.1.1 (Berkeley) 02/07/88 13 */ 14 15 #include "param.h" 16 #include "systm.h" 17 #include "mbuf.h" 18 #include "socket.h" 19 #include "socketvar.h" 20 #include "protosw.h" 21 #include "errno.h" 22 #include "stat.h" 23 24 #include "../net/if.h" 25 #include "../net/route.h" 26 27 #include "in.h" 28 #include "in_pcb.h" 29 #include "in_systm.h" 30 #include "ip.h" 31 #include "ip_var.h" 32 #include "tcp.h" 33 #include "tcp_fsm.h" 34 #include "tcp_seq.h" 35 #include "tcp_timer.h" 36 #include "tcp_var.h" 37 #include "tcpip.h" 38 #include "tcp_debug.h" 39 40 /* 41 * TCP protocol interface to socket abstraction. 42 */ 43 extern char *tcpstates[]; 44 struct tcpcb *tcp_newtcpcb(); 45 46 /* 47 * Process a TCP user request for TCP tb. If this is a send request 48 * then m is the mbuf chain of send data. If this is a timer expiration 49 * (called from the software clock routine), then timertype tells which timer. 50 */ 51 /*ARGSUSED*/ 52 tcp_usrreq(so, req, m, nam, rights) 53 struct socket *so; 54 int req; 55 struct mbuf *m, *nam, *rights; 56 { 57 register struct inpcb *inp; 58 register struct tcpcb *tp; 59 int s; 60 int error = 0; 61 int ostate; 62 63 #if BSD>=43 64 if (req == PRU_CONTROL) 65 return (in_control(so, (int)m, (caddr_t)nam, 66 (struct ifnet *)rights)); 67 #else 68 if (req == PRU_CONTROL) 69 return(EOPNOTSUPP); 70 #endif 71 if (rights && rights->m_len) 72 return (EINVAL); 73 74 s = splnet(); 75 inp = sotoinpcb(so); 76 /* 77 * When a TCP is attached to a socket, then there will be 78 * a (struct inpcb) pointed at by the socket, and this 79 * structure will point at a subsidary (struct tcpcb). 80 */ 81 if (inp == 0 && req != PRU_ATTACH) { 82 splx(s); 83 return (EINVAL); /* XXX */ 84 } 85 if (inp) { 86 tp = intotcpcb(inp); 87 /* WHAT IF TP IS 0? */ 88 #ifdef KPROF 89 tcp_acounts[tp->t_state][req]++; 90 #endif 91 ostate = tp->t_state; 92 } else 93 ostate = 0; 94 switch (req) { 95 96 /* 97 * TCP attaches to socket via PRU_ATTACH, reserving space, 98 * and an internet control block. 99 */ 100 case PRU_ATTACH: 101 if (inp) { 102 error = EISCONN; 103 break; 104 } 105 error = tcp_attach(so); 106 if (error) 107 break; 108 if ((so->so_options & SO_LINGER) && so->so_linger == 0) 109 so->so_linger = TCP_LINGERTIME; 110 tp = sototcpcb(so); 111 break; 112 113 /* 114 * PRU_DETACH detaches the TCP protocol from the socket. 115 * If the protocol state is non-embryonic, then can't 116 * do this directly: have to initiate a PRU_DISCONNECT, 117 * which may finish later; embryonic TCB's can just 118 * be discarded here. 119 */ 120 case PRU_DETACH: 121 if (tp->t_state > TCPS_LISTEN) 122 tp = tcp_disconnect(tp); 123 else 124 tp = tcp_close(tp); 125 break; 126 127 /* 128 * Give the socket an address. 129 */ 130 case PRU_BIND: 131 error = in_pcbbind(inp, nam); 132 if (error) 133 break; 134 break; 135 136 /* 137 * Prepare to accept connections. 138 */ 139 case PRU_LISTEN: 140 if (inp->inp_lport == 0) 141 error = in_pcbbind(inp, (struct mbuf *)0); 142 if (error == 0) 143 tp->t_state = TCPS_LISTEN; 144 break; 145 146 /* 147 * Initiate connection to peer. 148 * Create a template for use in transmissions on this connection. 149 * Enter SYN_SENT state, and mark socket as connecting. 150 * Start keep-alive timer, and seed output sequence space. 151 * Send initial segment on connection. 152 */ 153 case PRU_CONNECT: 154 if (inp->inp_lport == 0) { 155 error = in_pcbbind(inp, (struct mbuf *)0); 156 if (error) 157 break; 158 } 159 error = in_pcbconnect(inp, nam); 160 if (error) 161 break; 162 tp->t_template = tcp_template(tp); 163 if (tp->t_template == 0) { 164 in_pcbdisconnect(inp); 165 error = ENOBUFS; 166 break; 167 } 168 soisconnecting(so); 169 tcpstat.tcps_connattempt++; 170 tp->t_state = TCPS_SYN_SENT; 171 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP; 172 tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2; 173 tcp_sendseqinit(tp); 174 error = tcp_output(tp); 175 break; 176 177 /* 178 * Create a TCP connection between two sockets. 179 */ 180 case PRU_CONNECT2: 181 error = EOPNOTSUPP; 182 break; 183 184 /* 185 * Initiate disconnect from peer. 186 * If connection never passed embryonic stage, just drop; 187 * else if don't need to let data drain, then can just drop anyways, 188 * else have to begin TCP shutdown process: mark socket disconnecting, 189 * drain unread data, state switch to reflect user close, and 190 * send segment (e.g. FIN) to peer. Socket will be really disconnected 191 * when peer sends FIN and acks ours. 192 * 193 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 194 */ 195 case PRU_DISCONNECT: 196 tp = tcp_disconnect(tp); 197 break; 198 199 /* 200 * Accept a connection. Essentially all the work is 201 * done at higher levels; just return the address 202 * of the peer, storing through addr. 203 */ 204 case PRU_ACCEPT: { 205 struct sockaddr_in *sin = mtod(nam, struct sockaddr_in *); 206 207 nam->m_len = sizeof (struct sockaddr_in); 208 sin->sin_family = AF_INET; 209 sin->sin_port = inp->inp_fport; 210 sin->sin_addr = inp->inp_faddr; 211 break; 212 } 213 214 /* 215 * Mark the connection as being incapable of further output. 216 */ 217 case PRU_SHUTDOWN: 218 socantsendmore(so); 219 tp = tcp_usrclosed(tp); 220 if (tp) 221 error = tcp_output(tp); 222 break; 223 224 /* 225 * After a receive, possibly send window update to peer. 226 */ 227 case PRU_RCVD: 228 (void) tcp_output(tp); 229 break; 230 231 /* 232 * Do a send by putting data in output queue and updating urgent 233 * marker if URG set. Possibly send more data. 234 */ 235 case PRU_SEND: 236 sbappend(&so->so_snd, m); 237 error = tcp_output(tp); 238 break; 239 240 /* 241 * Abort the TCP. 242 */ 243 case PRU_ABORT: 244 tp = tcp_drop(tp, ECONNABORTED); 245 break; 246 247 case PRU_SENSE: 248 ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat; 249 (void) splx(s); 250 return (0); 251 252 case PRU_RCVOOB: 253 if ((so->so_oobmark == 0 && 254 (so->so_state & SS_RCVATMARK) == 0) || 255 #ifdef SO_OOBINLINE 256 so->so_options & SO_OOBINLINE || 257 #endif 258 tp->t_oobflags & TCPOOB_HADDATA) { 259 error = EINVAL; 260 break; 261 } 262 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 263 error = EWOULDBLOCK; 264 break; 265 } 266 m->m_len = 1; 267 *mtod(m, caddr_t) = tp->t_iobc; 268 if (((int)nam & MSG_PEEK) == 0) 269 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 270 break; 271 272 case PRU_SENDOOB: 273 if (sbspace(&so->so_snd) < -512) { 274 m_freem(m); 275 error = ENOBUFS; 276 break; 277 } 278 /* 279 * According to RFC961 (Assigned Protocols), 280 * the urgent pointer points to the last octet 281 * of urgent data. We continue, however, 282 * to consider it to indicate the first octet 283 * of data past the urgent section. 284 * Otherwise, snd_up should be one lower. 285 */ 286 sbappend(&so->so_snd, m); 287 tp->snd_up = tp->snd_una + so->so_snd.sb_cc; 288 tp->t_force = 1; 289 error = tcp_output(tp); 290 tp->t_force = 0; 291 break; 292 293 case PRU_SOCKADDR: 294 in_setsockaddr(inp, nam); 295 break; 296 297 case PRU_PEERADDR: 298 in_setpeeraddr(inp, nam); 299 break; 300 301 /* 302 * TCP slow timer went off; going through this 303 * routine for tracing's sake. 304 */ 305 case PRU_SLOWTIMO: 306 tp = tcp_timers(tp, (int)nam); 307 req |= (int)nam << 8; /* for debug's sake */ 308 break; 309 310 default: 311 panic("tcp_usrreq"); 312 } 313 if (tp && (so->so_options & SO_DEBUG)) 314 tcp_trace(TA_USER, ostate, tp, (struct tcpiphdr *)0, req); 315 splx(s); 316 return (error); 317 } 318 319 #if BSD>=43 320 tcp_ctloutput(op, so, level, optname, mp) 321 int op; 322 struct socket *so; 323 int level, optname; 324 struct mbuf **mp; 325 { 326 int error = 0; 327 struct inpcb *inp = sotoinpcb(so); 328 register struct tcpcb *tp = intotcpcb(inp); 329 register struct mbuf *m; 330 331 if (level != IPPROTO_TCP) 332 return (ip_ctloutput(op, so, level, optname, mp)); 333 334 switch (op) { 335 336 case PRCO_SETOPT: 337 m = *mp; 338 switch (optname) { 339 340 case TCP_NODELAY: 341 if (m == NULL || m->m_len < sizeof (int)) 342 error = EINVAL; 343 else if (*mtod(m, int *)) 344 tp->t_flags |= TF_NODELAY; 345 else 346 tp->t_flags &= ~TF_NODELAY; 347 break; 348 349 case TCP_MAXSEG: /* not yet */ 350 default: 351 error = EINVAL; 352 break; 353 } 354 if (m) 355 (void) m_free(m); 356 break; 357 358 case PRCO_GETOPT: 359 *mp = m = m_get(M_WAIT, MT_SOOPTS); 360 m->m_len = sizeof(int); 361 362 switch (optname) { 363 case TCP_NODELAY: 364 *mtod(m, int *) = tp->t_flags & TF_NODELAY; 365 break; 366 case TCP_MAXSEG: 367 *mtod(m, int *) = tp->t_maxseg; 368 break; 369 default: 370 error = EINVAL; 371 break; 372 } 373 break; 374 } 375 return (error); 376 } 377 #endif 378 379 int tcp_sendspace = 1024*4; 380 int tcp_recvspace = 1024*4; 381 /* 382 * Attach TCP protocol to socket, allocating 383 * internet protocol control block, tcp control block, 384 * bufer space, and entering LISTEN state if to accept connections. 385 */ 386 tcp_attach(so) 387 struct socket *so; 388 { 389 register struct tcpcb *tp; 390 struct inpcb *inp; 391 int error; 392 393 error = soreserve(so, tcp_sendspace, tcp_recvspace); 394 if (error) 395 return (error); 396 error = in_pcballoc(so, &tcb); 397 if (error) 398 return (error); 399 inp = sotoinpcb(so); 400 tp = tcp_newtcpcb(inp); 401 if (tp == 0) { 402 int nofd = so->so_state & SS_NOFDREF; /* XXX */ 403 404 so->so_state &= ~SS_NOFDREF; /* don't free the socket yet */ 405 in_pcbdetach(inp); 406 so->so_state |= nofd; 407 return (ENOBUFS); 408 } 409 tp->t_state = TCPS_CLOSED; 410 return (0); 411 } 412 413 /* 414 * Initiate (or continue) disconnect. 415 * If embryonic state, just send reset (once). 416 * If in ``let data drain'' option and linger null, just drop. 417 * Otherwise (hard), mark socket disconnecting and drop 418 * current input data; switch states based on user close, and 419 * send segment to peer (with FIN). 420 */ 421 struct tcpcb * 422 tcp_disconnect(tp) 423 register struct tcpcb *tp; 424 { 425 struct socket *so = tp->t_inpcb->inp_socket; 426 427 if (tp->t_state < TCPS_ESTABLISHED) 428 tp = tcp_close(tp); 429 else if ((so->so_options & SO_LINGER) && so->so_linger == 0) 430 tp = tcp_drop(tp, 0); 431 else { 432 soisdisconnecting(so); 433 sbflush(&so->so_rcv); 434 tp = tcp_usrclosed(tp); 435 if (tp) 436 (void) tcp_output(tp); 437 } 438 return (tp); 439 } 440 441 /* 442 * User issued close, and wish to trail through shutdown states: 443 * if never received SYN, just forget it. If got a SYN from peer, 444 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 445 * If already got a FIN from peer, then almost done; go to LAST_ACK 446 * state. In all other cases, have already sent FIN to peer (e.g. 447 * after PRU_SHUTDOWN), and just have to play tedious game waiting 448 * for peer to send FIN or not respond to keep-alives, etc. 449 * We can let the user exit from the close as soon as the FIN is acked. 450 */ 451 struct tcpcb * 452 tcp_usrclosed(tp) 453 register struct tcpcb *tp; 454 { 455 456 switch (tp->t_state) { 457 458 case TCPS_CLOSED: 459 case TCPS_LISTEN: 460 case TCPS_SYN_SENT: 461 tp->t_state = TCPS_CLOSED; 462 tp = tcp_close(tp); 463 break; 464 465 case TCPS_SYN_RECEIVED: 466 case TCPS_ESTABLISHED: 467 tp->t_state = TCPS_FIN_WAIT_1; 468 break; 469 470 case TCPS_CLOSE_WAIT: 471 tp->t_state = TCPS_LAST_ACK; 472 break; 473 } 474 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) 475 soisdisconnected(tp->t_inpcb->inp_socket); 476 return (tp); 477 } 478