1 /* 2 * Copyright (c) 1982, 1986, 1988, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * %sccs.include.redist.c% 6 * 7 * @(#)tcp_usrreq.c 8.3 (Berkeley) 01/09/95 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 41 /* 42 * Process a TCP user request for TCP tb. If this is a send request 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 */ 46 /*ARGSUSED*/ 47 int 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, (u_long)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 in_setpeeraddr(inp, nam); 205 break; 206 207 /* 208 * Mark the connection as being incapable of further output. 209 */ 210 case PRU_SHUTDOWN: 211 socantsendmore(so); 212 tp = tcp_usrclosed(tp); 213 if (tp) 214 error = tcp_output(tp); 215 break; 216 217 /* 218 * After a receive, possibly send window update to peer. 219 */ 220 case PRU_RCVD: 221 (void) tcp_output(tp); 222 break; 223 224 /* 225 * Do a send by putting data in output queue and updating urgent 226 * marker if URG set. Possibly send more data. 227 */ 228 case PRU_SEND: 229 sbappend(&so->so_snd, m); 230 error = tcp_output(tp); 231 break; 232 233 /* 234 * Abort the TCP. 235 */ 236 case PRU_ABORT: 237 tp = tcp_drop(tp, ECONNABORTED); 238 break; 239 240 case PRU_SENSE: 241 ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat; 242 (void) splx(s); 243 return (0); 244 245 case PRU_RCVOOB: 246 if ((so->so_oobmark == 0 && 247 (so->so_state & SS_RCVATMARK) == 0) || 248 so->so_options & SO_OOBINLINE || 249 tp->t_oobflags & TCPOOB_HADDATA) { 250 error = EINVAL; 251 break; 252 } 253 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 254 error = EWOULDBLOCK; 255 break; 256 } 257 m->m_len = 1; 258 *mtod(m, caddr_t) = tp->t_iobc; 259 if (((int)nam & MSG_PEEK) == 0) 260 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 261 break; 262 263 case PRU_SENDOOB: 264 if (sbspace(&so->so_snd) < -512) { 265 m_freem(m); 266 error = ENOBUFS; 267 break; 268 } 269 /* 270 * According to RFC961 (Assigned Protocols), 271 * the urgent pointer points to the last octet 272 * of urgent data. We continue, however, 273 * to consider it to indicate the first octet 274 * of data past the urgent section. 275 * Otherwise, snd_up should be one lower. 276 */ 277 sbappend(&so->so_snd, m); 278 tp->snd_up = tp->snd_una + so->so_snd.sb_cc; 279 tp->t_force = 1; 280 error = tcp_output(tp); 281 tp->t_force = 0; 282 break; 283 284 case PRU_SOCKADDR: 285 in_setsockaddr(inp, nam); 286 break; 287 288 case PRU_PEERADDR: 289 in_setpeeraddr(inp, nam); 290 break; 291 292 /* 293 * TCP slow timer went off; going through this 294 * routine for tracing's sake. 295 */ 296 case PRU_SLOWTIMO: 297 tp = tcp_timers(tp, (int)nam); 298 req |= (int)nam << 8; /* for debug's sake */ 299 break; 300 301 default: 302 panic("tcp_usrreq"); 303 } 304 if (tp && (so->so_options & SO_DEBUG)) 305 tcp_trace(TA_USER, ostate, tp, (struct tcpiphdr *)0, req); 306 splx(s); 307 return (error); 308 } 309 310 int 311 tcp_ctloutput(op, so, level, optname, mp) 312 int op; 313 struct socket *so; 314 int level, optname; 315 struct mbuf **mp; 316 { 317 int error = 0, s; 318 struct inpcb *inp; 319 register struct tcpcb *tp; 320 register struct mbuf *m; 321 register int i; 322 323 s = splnet(); 324 inp = sotoinpcb(so); 325 if (inp == NULL) { 326 splx(s); 327 if (op == PRCO_SETOPT && *mp) 328 (void) m_free(*mp); 329 return (ECONNRESET); 330 } 331 if (level != IPPROTO_TCP) { 332 error = ip_ctloutput(op, so, level, optname, mp); 333 splx(s); 334 return (error); 335 } 336 tp = intotcpcb(inp); 337 338 switch (op) { 339 340 case PRCO_SETOPT: 341 m = *mp; 342 switch (optname) { 343 344 case TCP_NODELAY: 345 if (m == NULL || m->m_len < sizeof (int)) 346 error = EINVAL; 347 else if (*mtod(m, int *)) 348 tp->t_flags |= TF_NODELAY; 349 else 350 tp->t_flags &= ~TF_NODELAY; 351 break; 352 353 case TCP_MAXSEG: 354 if (m && (i = *mtod(m, int *)) > 0 && i <= tp->t_maxseg) 355 tp->t_maxseg = i; 356 else 357 error = EINVAL; 358 break; 359 360 default: 361 error = ENOPROTOOPT; 362 break; 363 } 364 if (m) 365 (void) m_free(m); 366 break; 367 368 case PRCO_GETOPT: 369 *mp = m = m_get(M_WAIT, MT_SOOPTS); 370 m->m_len = sizeof(int); 371 372 switch (optname) { 373 case TCP_NODELAY: 374 *mtod(m, int *) = tp->t_flags & TF_NODELAY; 375 break; 376 case TCP_MAXSEG: 377 *mtod(m, int *) = tp->t_maxseg; 378 break; 379 default: 380 error = ENOPROTOOPT; 381 break; 382 } 383 break; 384 } 385 splx(s); 386 return (error); 387 } 388 389 u_long tcp_sendspace = 1024*8; 390 u_long tcp_recvspace = 1024*8; 391 392 /* 393 * Attach TCP protocol to socket, allocating 394 * internet protocol control block, tcp control block, 395 * bufer space, and entering LISTEN state if to accept connections. 396 */ 397 int 398 tcp_attach(so) 399 struct socket *so; 400 { 401 register struct tcpcb *tp; 402 struct inpcb *inp; 403 int error; 404 405 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 406 error = soreserve(so, tcp_sendspace, tcp_recvspace); 407 if (error) 408 return (error); 409 } 410 error = in_pcballoc(so, &tcb); 411 if (error) 412 return (error); 413 inp = sotoinpcb(so); 414 tp = tcp_newtcpcb(inp); 415 if (tp == 0) { 416 int nofd = so->so_state & SS_NOFDREF; /* XXX */ 417 418 so->so_state &= ~SS_NOFDREF; /* don't free the socket yet */ 419 in_pcbdetach(inp); 420 so->so_state |= nofd; 421 return (ENOBUFS); 422 } 423 tp->t_state = TCPS_CLOSED; 424 return (0); 425 } 426 427 /* 428 * Initiate (or continue) disconnect. 429 * If embryonic state, just send reset (once). 430 * If in ``let data drain'' option and linger null, just drop. 431 * Otherwise (hard), mark socket disconnecting and drop 432 * current input data; switch states based on user close, and 433 * send segment to peer (with FIN). 434 */ 435 struct tcpcb * 436 tcp_disconnect(tp) 437 register struct tcpcb *tp; 438 { 439 struct socket *so = tp->t_inpcb->inp_socket; 440 441 if (tp->t_state < TCPS_ESTABLISHED) 442 tp = tcp_close(tp); 443 else if ((so->so_options & SO_LINGER) && so->so_linger == 0) 444 tp = tcp_drop(tp, 0); 445 else { 446 soisdisconnecting(so); 447 sbflush(&so->so_rcv); 448 tp = tcp_usrclosed(tp); 449 if (tp) 450 (void) tcp_output(tp); 451 } 452 return (tp); 453 } 454 455 /* 456 * User issued close, and wish to trail through shutdown states: 457 * if never received SYN, just forget it. If got a SYN from peer, 458 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 459 * If already got a FIN from peer, then almost done; go to LAST_ACK 460 * state. In all other cases, have already sent FIN to peer (e.g. 461 * after PRU_SHUTDOWN), and just have to play tedious game waiting 462 * for peer to send FIN or not respond to keep-alives, etc. 463 * We can let the user exit from the close as soon as the FIN is acked. 464 */ 465 struct tcpcb * 466 tcp_usrclosed(tp) 467 register struct tcpcb *tp; 468 { 469 470 switch (tp->t_state) { 471 472 case TCPS_CLOSED: 473 case TCPS_LISTEN: 474 case TCPS_SYN_SENT: 475 tp->t_state = TCPS_CLOSED; 476 tp = tcp_close(tp); 477 break; 478 479 case TCPS_SYN_RECEIVED: 480 case TCPS_ESTABLISHED: 481 tp->t_state = TCPS_FIN_WAIT_1; 482 break; 483 484 case TCPS_CLOSE_WAIT: 485 tp->t_state = TCPS_LAST_ACK; 486 break; 487 } 488 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) 489 soisdisconnected(tp->t_inpcb->inp_socket); 490 return (tp); 491 } 492