1 /* 2 * Copyright (c) 1982, 1986, 1988 Regents of the University of California. 3 * All rights reserved. 4 * 5 * %sccs.include.redist.c% 6 * 7 * @(#)tcp_usrreq.c 7.18 (Berkeley) 01/08/93 8 */ 9 10 #include <sys/param.h> 11 #include <sys/systm.h> 12 #include <sys/malloc.h> 13 #include <sys/mbuf.h> 14 #include <sys/socket.h> 15 #include <sys/socketvar.h> 16 #include <sys/protosw.h> 17 #include <sys/errno.h> 18 #include <sys/stat.h> 19 20 #include <net/if.h> 21 #include <net/route.h> 22 23 #include <netinet/in.h> 24 #include <netinet/in_systm.h> 25 #include <netinet/ip.h> 26 #include <netinet/in_pcb.h> 27 #include <netinet/ip_var.h> 28 #include <netinet/tcp.h> 29 #include <netinet/tcp_fsm.h> 30 #include <netinet/tcp_seq.h> 31 #include <netinet/tcp_timer.h> 32 #include <netinet/tcp_var.h> 33 #include <netinet/tcpip.h> 34 #include <netinet/tcp_debug.h> 35 36 /* 37 * TCP protocol interface to socket abstraction. 38 */ 39 extern char *tcpstates[]; 40 struct tcpcb *tcp_newtcpcb(); 41 42 /* 43 * Process a TCP user request for TCP tb. If this is a send request 44 * then m is the mbuf chain of send data. If this is a timer expiration 45 * (called from the software clock routine), then timertype tells which timer. 46 */ 47 /*ARGSUSED*/ 48 tcp_usrreq(so, req, m, nam, control) 49 struct socket *so; 50 int req; 51 struct mbuf *m, *nam, *control; 52 { 53 register struct inpcb *inp; 54 register struct tcpcb *tp; 55 int s; 56 int error = 0; 57 int ostate; 58 59 if (req == PRU_CONTROL) 60 return (in_control(so, (int)m, (caddr_t)nam, 61 (struct ifnet *)control)); 62 if (control && control->m_len) { 63 m_freem(control); 64 if (m) 65 m_freem(m); 66 return (EINVAL); 67 } 68 69 s = splnet(); 70 inp = sotoinpcb(so); 71 /* 72 * When a TCP is attached to a socket, then there will be 73 * a (struct inpcb) pointed at by the socket, and this 74 * structure will point at a subsidary (struct tcpcb). 75 */ 76 if (inp == 0 && req != PRU_ATTACH) { 77 splx(s); 78 return (EINVAL); /* XXX */ 79 } 80 if (inp) { 81 tp = intotcpcb(inp); 82 /* WHAT IF TP IS 0? */ 83 #ifdef KPROF 84 tcp_acounts[tp->t_state][req]++; 85 #endif 86 ostate = tp->t_state; 87 } else 88 ostate = 0; 89 switch (req) { 90 91 /* 92 * TCP attaches to socket via PRU_ATTACH, reserving space, 93 * and an internet control block. 94 */ 95 case PRU_ATTACH: 96 if (inp) { 97 error = EISCONN; 98 break; 99 } 100 error = tcp_attach(so); 101 if (error) 102 break; 103 if ((so->so_options & SO_LINGER) && so->so_linger == 0) 104 so->so_linger = TCP_LINGERTIME; 105 tp = sototcpcb(so); 106 break; 107 108 /* 109 * PRU_DETACH detaches the TCP protocol from the socket. 110 * If the protocol state is non-embryonic, then can't 111 * do this directly: have to initiate a PRU_DISCONNECT, 112 * which may finish later; embryonic TCB's can just 113 * be discarded here. 114 */ 115 case PRU_DETACH: 116 if (tp->t_state > TCPS_LISTEN) 117 tp = tcp_disconnect(tp); 118 else 119 tp = tcp_close(tp); 120 break; 121 122 /* 123 * Give the socket an address. 124 */ 125 case PRU_BIND: 126 error = in_pcbbind(inp, nam); 127 if (error) 128 break; 129 break; 130 131 /* 132 * Prepare to accept connections. 133 */ 134 case PRU_LISTEN: 135 if (inp->inp_lport == 0) 136 error = in_pcbbind(inp, (struct mbuf *)0); 137 if (error == 0) 138 tp->t_state = TCPS_LISTEN; 139 break; 140 141 /* 142 * Initiate connection to peer. 143 * Create a template for use in transmissions on this connection. 144 * Enter SYN_SENT state, and mark socket as connecting. 145 * Start keep-alive timer, and seed output sequence space. 146 * Send initial segment on connection. 147 */ 148 case PRU_CONNECT: 149 if (inp->inp_lport == 0) { 150 error = in_pcbbind(inp, (struct mbuf *)0); 151 if (error) 152 break; 153 } 154 error = in_pcbconnect(inp, nam); 155 if (error) 156 break; 157 tp->t_template = tcp_template(tp); 158 if (tp->t_template == 0) { 159 in_pcbdisconnect(inp); 160 error = ENOBUFS; 161 break; 162 } 163 /* Compute window scaling to request. */ 164 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 165 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat) 166 tp->request_r_scale++; 167 soisconnecting(so); 168 tcpstat.tcps_connattempt++; 169 tp->t_state = TCPS_SYN_SENT; 170 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 171 tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2; 172 tcp_sendseqinit(tp); 173 error = tcp_output(tp); 174 break; 175 176 /* 177 * Create a TCP connection between two sockets. 178 */ 179 case PRU_CONNECT2: 180 error = EOPNOTSUPP; 181 break; 182 183 /* 184 * Initiate disconnect from peer. 185 * If connection never passed embryonic stage, just drop; 186 * else if don't need to let data drain, then can just drop anyways, 187 * else have to begin TCP shutdown process: mark socket disconnecting, 188 * drain unread data, state switch to reflect user close, and 189 * send segment (e.g. FIN) to peer. Socket will be really disconnected 190 * when peer sends FIN and acks ours. 191 * 192 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 193 */ 194 case PRU_DISCONNECT: 195 tp = tcp_disconnect(tp); 196 break; 197 198 /* 199 * Accept a connection. Essentially all the work is 200 * done at higher levels; just return the address 201 * of the peer, storing through addr. 202 */ 203 case PRU_ACCEPT: { 204 struct sockaddr_in *sin = mtod(nam, struct sockaddr_in *); 205 206 nam->m_len = sizeof (struct sockaddr_in); 207 sin->sin_family = AF_INET; 208 sin->sin_len = sizeof(*sin); 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 so->so_options & SO_OOBINLINE || 256 tp->t_oobflags & TCPOOB_HADDATA) { 257 error = EINVAL; 258 break; 259 } 260 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 261 error = EWOULDBLOCK; 262 break; 263 } 264 m->m_len = 1; 265 *mtod(m, caddr_t) = tp->t_iobc; 266 if (((int)nam & MSG_PEEK) == 0) 267 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 268 break; 269 270 case PRU_SENDOOB: 271 if (sbspace(&so->so_snd) < -512) { 272 m_freem(m); 273 error = ENOBUFS; 274 break; 275 } 276 /* 277 * According to RFC961 (Assigned Protocols), 278 * the urgent pointer points to the last octet 279 * of urgent data. We continue, however, 280 * to consider it to indicate the first octet 281 * of data past the urgent section. 282 * Otherwise, snd_up should be one lower. 283 */ 284 sbappend(&so->so_snd, m); 285 tp->snd_up = tp->snd_una + so->so_snd.sb_cc; 286 tp->t_force = 1; 287 error = tcp_output(tp); 288 tp->t_force = 0; 289 break; 290 291 case PRU_SOCKADDR: 292 in_setsockaddr(inp, nam); 293 break; 294 295 case PRU_PEERADDR: 296 in_setpeeraddr(inp, nam); 297 break; 298 299 /* 300 * TCP slow timer went off; going through this 301 * routine for tracing's sake. 302 */ 303 case PRU_SLOWTIMO: 304 tp = tcp_timers(tp, (int)nam); 305 req |= (int)nam << 8; /* for debug's sake */ 306 break; 307 308 default: 309 panic("tcp_usrreq"); 310 } 311 if (tp && (so->so_options & SO_DEBUG)) 312 tcp_trace(TA_USER, ostate, tp, (struct tcpiphdr *)0, req); 313 splx(s); 314 return (error); 315 } 316 317 tcp_ctloutput(op, so, level, optname, mp) 318 int op; 319 struct socket *so; 320 int level, optname; 321 struct mbuf **mp; 322 { 323 int error = 0, s; 324 struct inpcb *inp; 325 register struct tcpcb *tp; 326 register struct mbuf *m; 327 register int i; 328 329 s = splnet(); 330 inp = sotoinpcb(so); 331 if (inp == NULL) { 332 splx(s); 333 return (ECONNRESET); 334 } 335 if (level != IPPROTO_TCP) { 336 error = ip_ctloutput(op, so, level, optname, mp); 337 splx(s); 338 return (error); 339 } 340 tp = intotcpcb(inp); 341 342 switch (op) { 343 344 case PRCO_SETOPT: 345 m = *mp; 346 switch (optname) { 347 348 case TCP_NODELAY: 349 if (m == NULL || m->m_len < sizeof (int)) 350 error = EINVAL; 351 else if (*mtod(m, int *)) 352 tp->t_flags |= TF_NODELAY; 353 else 354 tp->t_flags &= ~TF_NODELAY; 355 break; 356 357 case TCP_MAXSEG: 358 if (m && (i = *mtod(m, int *)) > 0 && i <= tp->t_maxseg) 359 tp->t_maxseg = i; 360 else 361 error = EINVAL; 362 break; 363 364 default: 365 error = EINVAL; 366 break; 367 } 368 if (m) 369 (void) m_free(m); 370 break; 371 372 case PRCO_GETOPT: 373 *mp = m = m_get(M_WAIT, MT_SOOPTS); 374 m->m_len = sizeof(int); 375 376 switch (optname) { 377 case TCP_NODELAY: 378 *mtod(m, int *) = tp->t_flags & TF_NODELAY; 379 break; 380 case TCP_MAXSEG: 381 *mtod(m, int *) = tp->t_maxseg; 382 break; 383 default: 384 error = EINVAL; 385 break; 386 } 387 break; 388 } 389 splx(s); 390 return (error); 391 } 392 393 u_long tcp_sendspace = 1024*8; 394 u_long tcp_recvspace = 1024*8; 395 396 /* 397 * Attach TCP protocol to socket, allocating 398 * internet protocol control block, tcp control block, 399 * bufer space, and entering LISTEN state if to accept connections. 400 */ 401 tcp_attach(so) 402 struct socket *so; 403 { 404 register struct tcpcb *tp; 405 struct inpcb *inp; 406 int error; 407 408 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 409 error = soreserve(so, tcp_sendspace, tcp_recvspace); 410 if (error) 411 return (error); 412 } 413 error = in_pcballoc(so, &tcb); 414 if (error) 415 return (error); 416 inp = sotoinpcb(so); 417 tp = tcp_newtcpcb(inp); 418 if (tp == 0) { 419 int nofd = so->so_state & SS_NOFDREF; /* XXX */ 420 421 so->so_state &= ~SS_NOFDREF; /* don't free the socket yet */ 422 in_pcbdetach(inp); 423 so->so_state |= nofd; 424 return (ENOBUFS); 425 } 426 tp->t_state = TCPS_CLOSED; 427 return (0); 428 } 429 430 /* 431 * Initiate (or continue) disconnect. 432 * If embryonic state, just send reset (once). 433 * If in ``let data drain'' option and linger null, just drop. 434 * Otherwise (hard), mark socket disconnecting and drop 435 * current input data; switch states based on user close, and 436 * send segment to peer (with FIN). 437 */ 438 struct tcpcb * 439 tcp_disconnect(tp) 440 register struct tcpcb *tp; 441 { 442 struct socket *so = tp->t_inpcb->inp_socket; 443 444 if (tp->t_state < TCPS_ESTABLISHED) 445 tp = tcp_close(tp); 446 else if ((so->so_options & SO_LINGER) && so->so_linger == 0) 447 tp = tcp_drop(tp, 0); 448 else { 449 soisdisconnecting(so); 450 sbflush(&so->so_rcv); 451 tp = tcp_usrclosed(tp); 452 if (tp) 453 (void) tcp_output(tp); 454 } 455 return (tp); 456 } 457 458 /* 459 * User issued close, and wish to trail through shutdown states: 460 * if never received SYN, just forget it. If got a SYN from peer, 461 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 462 * If already got a FIN from peer, then almost done; go to LAST_ACK 463 * state. In all other cases, have already sent FIN to peer (e.g. 464 * after PRU_SHUTDOWN), and just have to play tedious game waiting 465 * for peer to send FIN or not respond to keep-alives, etc. 466 * We can let the user exit from the close as soon as the FIN is acked. 467 */ 468 struct tcpcb * 469 tcp_usrclosed(tp) 470 register struct tcpcb *tp; 471 { 472 473 switch (tp->t_state) { 474 475 case TCPS_CLOSED: 476 case TCPS_LISTEN: 477 case TCPS_SYN_SENT: 478 tp->t_state = TCPS_CLOSED; 479 tp = tcp_close(tp); 480 break; 481 482 case TCPS_SYN_RECEIVED: 483 case TCPS_ESTABLISHED: 484 tp->t_state = TCPS_FIN_WAIT_1; 485 break; 486 487 case TCPS_CLOSE_WAIT: 488 tp->t_state = TCPS_LAST_ACK; 489 break; 490 } 491 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) 492 soisdisconnected(tp->t_inpcb->inp_socket); 493 return (tp); 494 } 495