1 /* 2 * Copyright (c) 1982, 1986 Regents of the University of California. 3 * All rights reserved. The Berkeley software License Agreement 4 * specifies the terms and conditions for redistribution. 5 * 6 * @(#)tcp_usrreq.c 7.2 (Berkeley) 02/19/87 7 */ 8 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" 17 18 #include "../net/if.h" 19 #include "../net/route.h" 20 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" 33 34 /* 35 * TCP protocol interface to socket abstraction. 36 */ 37 extern char *tcpstates[]; 38 struct tcpcb *tcp_newtcpcb(); 39 int tcpsenderrors; 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 tcp_usrreq(so, req, m, nam, rights) 48 struct socket *so; 49 int req; 50 struct mbuf *m, *nam, *rights; 51 { 52 register struct inpcb *inp = sotoinpcb(so); 53 register struct tcpcb *tp; 54 int s = splnet(); 55 int error = 0; 56 int ostate; 57 58 if (req == PRU_CONTROL) 59 return (in_control(so, (int)m, (caddr_t)nam, 60 (struct ifnet *)rights)); 61 if (rights && rights->m_len) { 62 splx(s); 63 return (EINVAL); 64 } 65 /* 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). 69 */ 70 if (inp == 0 && req != PRU_ATTACH) { 71 splx(s); 72 return (EINVAL); /* XXX */ 73 } 74 if (inp) { 75 tp = intotcpcb(inp); 76 /* WHAT IF TP IS 0? */ 77 #ifdef KPROF 78 tcp_acounts[tp->t_state][req]++; 79 #endif 80 ostate = tp->t_state; 81 } else 82 ostate = 0; 83 switch (req) { 84 85 /* 86 * TCP attaches to socket via PRU_ATTACH, reserving space, 87 * and an internet control block. 88 */ 89 case PRU_ATTACH: 90 if (inp) { 91 error = EISCONN; 92 break; 93 } 94 error = tcp_attach(so); 95 if (error) 96 break; 97 if ((so->so_options & SO_LINGER) && so->so_linger == 0) 98 so->so_linger = TCP_LINGERTIME; 99 tp = sototcpcb(so); 100 break; 101 102 /* 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 */ 109 case PRU_DETACH: 110 if (tp->t_state > TCPS_LISTEN) 111 tp = tcp_disconnect(tp); 112 else 113 tp = tcp_close(tp); 114 break; 115 116 /* 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; 134 135 /* 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 */ 142 case PRU_CONNECT: 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); 149 if (error) 150 break; 151 tp->t_template = tcp_template(tp); 152 if (tp->t_template == 0) { 153 in_pcbdisconnect(inp); 154 error = ENOBUFS; 155 break; 156 } 157 soisconnecting(so); 158 tcpstat.tcps_connattempt++; 159 tp->t_state = TCPS_SYN_SENT; 160 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP; 161 tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2; 162 tcp_sendseqinit(tp); 163 error = tcp_output(tp); 164 break; 165 166 /* 167 * Create a TCP connection between two sockets. 168 */ 169 case PRU_CONNECT2: 170 error = EOPNOTSUPP; 171 break; 172 173 /* 174 * Initiate disconnect from peer. 175 * If connection never passed embryonic stage, just drop; 176 * else if don't need to let data drain, then can just drop anyways, 177 * else have to begin TCP shutdown process: mark socket disconnecting, 178 * drain unread data, state switch to reflect user close, and 179 * send segment (e.g. FIN) to peer. Socket will be really disconnected 180 * when peer sends FIN and acks ours. 181 * 182 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 183 */ 184 case PRU_DISCONNECT: 185 tp = tcp_disconnect(tp); 186 break; 187 188 /* 189 * Accept a connection. Essentially all the work is 190 * done at higher levels; just return the address 191 * of the peer, storing through addr. 192 */ 193 case PRU_ACCEPT: { 194 struct sockaddr_in *sin = mtod(nam, struct sockaddr_in *); 195 196 nam->m_len = sizeof (struct sockaddr_in); 197 sin->sin_family = AF_INET; 198 sin->sin_port = inp->inp_fport; 199 sin->sin_addr = inp->inp_faddr; 200 break; 201 } 202 203 /* 204 * Mark the connection as being incapable of further output. 205 */ 206 case PRU_SHUTDOWN: 207 socantsendmore(so); 208 tp = tcp_usrclosed(tp); 209 if (tp) 210 error = tcp_output(tp); 211 break; 212 213 /* 214 * After a receive, possibly send window update to peer. 215 */ 216 case PRU_RCVD: 217 (void) tcp_output(tp); 218 break; 219 220 /* 221 * Do a send by putting data in output queue and updating urgent 222 * marker if URG set. Possibly send more data. 223 */ 224 case PRU_SEND: 225 sbappend(&so->so_snd, m); 226 #ifdef notdef 227 if (tp->t_flags & TF_PUSH) 228 tp->snd_end = tp->snd_una + so->so_snd.sb_cc; 229 #endif 230 error = tcp_output(tp); 231 if (error) { /* XXX fix to use other path */ 232 if (error == ENOBUFS) /* XXX */ 233 error = 0; /* XXX */ 234 tcpsenderrors++; 235 } 236 break; 237 238 /* 239 * Abort the TCP. 240 */ 241 case PRU_ABORT: 242 tp = tcp_drop(tp, ECONNABORTED); 243 break; 244 245 case PRU_SENSE: 246 ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat; 247 return (0); 248 249 case PRU_RCVOOB: 250 if ((so->so_oobmark == 0 && 251 (so->so_state & SS_RCVATMARK) == 0) || 252 so->so_options & SO_OOBINLINE || 253 tp->t_oobflags & TCPOOB_HADDATA) { 254 error = EINVAL; 255 break; 256 } 257 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 258 error = EWOULDBLOCK; 259 break; 260 } 261 m->m_len = 1; 262 *mtod(m, caddr_t) = tp->t_iobc; 263 if (((int)nam & MSG_PEEK) == 0) 264 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 265 break; 266 267 case PRU_SENDOOB: 268 if (sbspace(&so->so_snd) < -512) { 269 m_freem(m); 270 error = ENOBUFS; 271 break; 272 } 273 /* 274 * According to RFC961 (Assigned Protocols), 275 * the urgent pointer points to the last octet 276 * of urgent data. We continue, however, 277 * to consider it to indicate the first octet 278 * of data past the urgent section. 279 * Otherwise, snd_up should be one lower. 280 */ 281 sbappend(&so->so_snd, m); 282 tp->snd_up = tp->snd_una + so->so_snd.sb_cc; 283 tp->t_force = 1; 284 error = tcp_output(tp); 285 tp->t_force = 0; 286 break; 287 288 case PRU_SOCKADDR: 289 in_setsockaddr(inp, nam); 290 break; 291 292 case PRU_PEERADDR: 293 in_setpeeraddr(inp, nam); 294 break; 295 296 /* 297 * TCP slow timer went off; going through this 298 * routine for tracing's sake. 299 */ 300 case PRU_SLOWTIMO: 301 tp = tcp_timers(tp, (int)nam); 302 req |= (int)nam << 8; /* for debug's sake */ 303 break; 304 305 default: 306 panic("tcp_usrreq"); 307 } 308 if (tp && (so->so_options & SO_DEBUG)) 309 tcp_trace(TA_USER, ostate, tp, (struct tcpiphdr *)0, req); 310 splx(s); 311 return (error); 312 } 313 314 tcp_ctloutput(op, so, level, optname, mp) 315 int op; 316 struct socket *so; 317 int level, optname; 318 struct mbuf **mp; 319 { 320 int error = 0; 321 struct inpcb *inp = sotoinpcb(so); 322 register struct tcpcb *tp = intotcpcb(inp); 323 register struct mbuf *m; 324 325 if (level != IPPROTO_TCP) 326 return (ip_ctloutput(op, so, level, optname, mp)); 327 328 switch (op) { 329 330 case PRCO_SETOPT: 331 m = *mp; 332 switch (optname) { 333 334 case TCP_NODELAY: 335 if (m == NULL || m->m_len < sizeof (int)) 336 error = EINVAL; 337 else if (*mtod(m, int *)) 338 tp->t_flags |= TF_NODELAY; 339 else 340 tp->t_flags &= ~TF_NODELAY; 341 break; 342 343 case TCP_MAXSEG: /* not yet */ 344 default: 345 error = EINVAL; 346 break; 347 } 348 (void)m_free(m); 349 break; 350 351 case PRCO_GETOPT: 352 *mp = m = m_get(M_WAIT, MT_SOOPTS); 353 m->m_len = sizeof(int); 354 355 switch (optname) { 356 case TCP_NODELAY: 357 *mtod(m, int *) = tp->t_flags & TF_NODELAY; 358 break; 359 case TCP_MAXSEG: 360 *mtod(m, int *) = tp->t_maxseg; 361 break; 362 default: 363 error = EINVAL; 364 break; 365 } 366 break; 367 } 368 return (error); 369 } 370 371 int tcp_sendspace = 1024*4; 372 int tcp_recvspace = 1024*4; 373 /* 374 * Attach TCP protocol to socket, allocating 375 * internet protocol control block, tcp control block, 376 * bufer space, and entering LISTEN state if to accept connections. 377 */ 378 tcp_attach(so) 379 struct socket *so; 380 { 381 register struct tcpcb *tp; 382 struct inpcb *inp; 383 int error; 384 385 error = soreserve(so, tcp_sendspace, tcp_recvspace); 386 if (error) 387 return (error); 388 error = in_pcballoc(so, &tcb); 389 if (error) 390 return (error); 391 inp = sotoinpcb(so); 392 tp = tcp_newtcpcb(inp); 393 if (tp == 0) { 394 int nofd = so->so_state & SS_NOFDREF; /* XXX */ 395 396 so->so_state &= ~SS_NOFDREF; /* don't free the socket yet */ 397 in_pcbdetach(inp); 398 so->so_state |= nofd; 399 return (ENOBUFS); 400 } 401 tp->t_state = TCPS_CLOSED; 402 return (0); 403 } 404 405 /* 406 * Initiate (or continue) disconnect. 407 * If embryonic state, just send reset (once). 408 * If in ``let data drain'' option and linger null, just drop. 409 * Otherwise (hard), mark socket disconnecting and drop 410 * current input data; switch states based on user close, and 411 * send segment to peer (with FIN). 412 */ 413 struct tcpcb * 414 tcp_disconnect(tp) 415 register struct tcpcb *tp; 416 { 417 struct socket *so = tp->t_inpcb->inp_socket; 418 419 if (tp->t_state < TCPS_ESTABLISHED) 420 tp = tcp_close(tp); 421 else if ((so->so_options & SO_LINGER) && so->so_linger == 0) 422 tp = tcp_drop(tp, 0); 423 else { 424 soisdisconnecting(so); 425 sbflush(&so->so_rcv); 426 tp = tcp_usrclosed(tp); 427 if (tp) 428 (void) tcp_output(tp); 429 } 430 return (tp); 431 } 432 433 /* 434 * User issued close, and wish to trail through shutdown states: 435 * if never received SYN, just forget it. If got a SYN from peer, 436 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 437 * If already got a FIN from peer, then almost done; go to LAST_ACK 438 * state. In all other cases, have already sent FIN to peer (e.g. 439 * after PRU_SHUTDOWN), and just have to play tedious game waiting 440 * for peer to send FIN or not respond to keep-alives, etc. 441 * We can let the user exit from the close as soon as the FIN is acked. 442 */ 443 struct tcpcb * 444 tcp_usrclosed(tp) 445 register struct tcpcb *tp; 446 { 447 448 switch (tp->t_state) { 449 450 case TCPS_CLOSED: 451 case TCPS_LISTEN: 452 case TCPS_SYN_SENT: 453 tp->t_state = TCPS_CLOSED; 454 tp = tcp_close(tp); 455 break; 456 457 case TCPS_SYN_RECEIVED: 458 case TCPS_ESTABLISHED: 459 tp->t_state = TCPS_FIN_WAIT_1; 460 break; 461 462 case TCPS_CLOSE_WAIT: 463 tp->t_state = TCPS_LAST_ACK; 464 break; 465 } 466 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) 467 soisdisconnected(tp->t_inpcb->inp_socket); 468 return (tp); 469 } 470