tcp_subr.c revision 1.58
1 /* $NetBSD: tcp_subr.c,v 1.58 1998/09/04 22:29:54 mycroft Exp $ */ 2 3 /*- 4 * Copyright (c) 1997, 1998 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 9 * Facility, NASA Ames Research Center. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. All advertising materials mentioning features or use of this software 20 * must display the following acknowledgement: 21 * This product includes software developed by the NetBSD 22 * Foundation, Inc. and its contributors. 23 * 4. Neither the name of The NetBSD Foundation nor the names of its 24 * contributors may be used to endorse or promote products derived 25 * from this software without specific prior written permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 28 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 29 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 30 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 31 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 32 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 33 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 34 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 35 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 36 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 37 * POSSIBILITY OF SUCH DAMAGE. 38 */ 39 40 /* 41 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 42 * The Regents of the University of California. All rights reserved. 43 * 44 * Redistribution and use in source and binary forms, with or without 45 * modification, are permitted provided that the following conditions 46 * are met: 47 * 1. Redistributions of source code must retain the above copyright 48 * notice, this list of conditions and the following disclaimer. 49 * 2. Redistributions in binary form must reproduce the above copyright 50 * notice, this list of conditions and the following disclaimer in the 51 * documentation and/or other materials provided with the distribution. 52 * 3. All advertising materials mentioning features or use of this software 53 * must display the following acknowledgement: 54 * This product includes software developed by the University of 55 * California, Berkeley and its contributors. 56 * 4. Neither the name of the University nor the names of its contributors 57 * may be used to endorse or promote products derived from this software 58 * without specific prior written permission. 59 * 60 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 61 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 62 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 63 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 64 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 65 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 66 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 67 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 68 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 69 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 70 * SUCH DAMAGE. 71 * 72 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 73 */ 74 75 #include "opt_tcp_compat_42.h" 76 #include "rnd.h" 77 78 #include <sys/param.h> 79 #include <sys/proc.h> 80 #include <sys/systm.h> 81 #include <sys/malloc.h> 82 #include <sys/mbuf.h> 83 #include <sys/socket.h> 84 #include <sys/socketvar.h> 85 #include <sys/protosw.h> 86 #include <sys/errno.h> 87 #include <sys/kernel.h> 88 #include <sys/pool.h> 89 #if NRND > 0 90 #include <sys/rnd.h> 91 #endif 92 93 #include <net/route.h> 94 #include <net/if.h> 95 96 #include <netinet/in.h> 97 #include <netinet/in_systm.h> 98 #include <netinet/ip.h> 99 #include <netinet/in_pcb.h> 100 #include <netinet/ip_var.h> 101 #include <netinet/ip_icmp.h> 102 #include <netinet/tcp.h> 103 #include <netinet/tcp_fsm.h> 104 #include <netinet/tcp_seq.h> 105 #include <netinet/tcp_timer.h> 106 #include <netinet/tcp_var.h> 107 #include <netinet/tcpip.h> 108 109 /* patchable/settable parameters for tcp */ 110 int tcp_mssdflt = TCP_MSS; 111 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 112 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 113 int tcp_do_sack = 1; /* selective acknowledgement */ 114 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 115 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 116 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 117 int tcp_init_win = 1; 118 int tcp_mss_ifmtu = 0; 119 #ifdef TCP_COMPAT_42 120 int tcp_compat_42 = 1; 121 #else 122 int tcp_compat_42 = 0; 123 #endif 124 125 #ifndef TCBHASHSIZE 126 #define TCBHASHSIZE 128 127 #endif 128 int tcbhashsize = TCBHASHSIZE; 129 130 int tcp_freeq __P((struct tcpcb *)); 131 132 struct pool tcpcb_pool; 133 134 /* 135 * Tcp initialization 136 */ 137 void 138 tcp_init() 139 { 140 141 pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", 142 0, NULL, NULL, M_PCB); 143 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 144 LIST_INIT(&tcp_delacks); 145 if (max_protohdr < sizeof(struct tcpiphdr)) 146 max_protohdr = sizeof(struct tcpiphdr); 147 if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN) 148 panic("tcp_init"); 149 150 /* Initialize the compressed state engine. */ 151 syn_cache_init(); 152 } 153 154 /* 155 * Create template to be used to send tcp packets on a connection. 156 * Call after host entry created, allocates an mbuf and fills 157 * in a skeletal tcp/ip header, minimizing the amount of work 158 * necessary when the connection is used. 159 */ 160 struct tcpiphdr * 161 tcp_template(tp) 162 struct tcpcb *tp; 163 { 164 register struct inpcb *inp = tp->t_inpcb; 165 register struct tcpiphdr *n; 166 167 if ((n = tp->t_template) == 0) { 168 MALLOC(n, struct tcpiphdr *, sizeof (struct tcpiphdr), 169 M_MBUF, M_NOWAIT); 170 if (n == NULL) 171 return (0); 172 } 173 bzero(n->ti_x1, sizeof n->ti_x1); 174 n->ti_pr = IPPROTO_TCP; 175 n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip)); 176 n->ti_src = inp->inp_laddr; 177 n->ti_dst = inp->inp_faddr; 178 n->ti_sport = inp->inp_lport; 179 n->ti_dport = inp->inp_fport; 180 n->ti_seq = 0; 181 n->ti_ack = 0; 182 n->ti_x2 = 0; 183 n->ti_off = 5; 184 n->ti_flags = 0; 185 n->ti_win = 0; 186 n->ti_sum = 0; 187 n->ti_urp = 0; 188 return (n); 189 } 190 191 /* 192 * Send a single message to the TCP at address specified by 193 * the given TCP/IP header. If m == 0, then we make a copy 194 * of the tcpiphdr at ti and send directly to the addressed host. 195 * This is used to force keep alive messages out using the TCP 196 * template for a connection tp->t_template. If flags are given 197 * then we send a message back to the TCP which originated the 198 * segment ti, and discard the mbuf containing it and any other 199 * attached mbufs. 200 * 201 * In any case the ack and sequence number of the transmitted 202 * segment are as specified by the parameters. 203 */ 204 int 205 tcp_respond(tp, ti, m, ack, seq, flags) 206 struct tcpcb *tp; 207 register struct tcpiphdr *ti; 208 register struct mbuf *m; 209 tcp_seq ack, seq; 210 int flags; 211 { 212 register int tlen; 213 int win = 0; 214 struct route *ro = 0; 215 216 if (tp) { 217 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 218 ro = &tp->t_inpcb->inp_route; 219 } 220 if (m == 0) { 221 m = m_gethdr(M_DONTWAIT, MT_HEADER); 222 if (m == NULL) 223 return (ENOBUFS); 224 225 if (tcp_compat_42) 226 tlen = 1; 227 else 228 tlen = 0; 229 230 m->m_data += max_linkhdr; 231 *mtod(m, struct tcpiphdr *) = *ti; 232 ti = mtod(m, struct tcpiphdr *); 233 flags = TH_ACK; 234 } else { 235 m_freem(m->m_next); 236 m->m_next = 0; 237 m->m_data = (caddr_t)ti; 238 m->m_len = sizeof (struct tcpiphdr); 239 tlen = 0; 240 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 241 xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_int32_t); 242 xchg(ti->ti_dport, ti->ti_sport, u_int16_t); 243 #undef xchg 244 } 245 bzero(ti->ti_x1, sizeof ti->ti_x1); 246 ti->ti_seq = htonl(seq); 247 ti->ti_ack = htonl(ack); 248 ti->ti_x2 = 0; 249 if ((flags & TH_SYN) == 0) { 250 if (tp) 251 ti->ti_win = htons((u_int16_t) (win >> tp->rcv_scale)); 252 else 253 ti->ti_win = htons((u_int16_t)win); 254 ti->ti_off = sizeof (struct tcphdr) >> 2; 255 tlen += sizeof (struct tcphdr); 256 } else 257 tlen += ti->ti_off << 2; 258 ti->ti_len = htons((u_int16_t)tlen); 259 tlen += sizeof (struct ip); 260 m->m_len = tlen; 261 m->m_pkthdr.len = tlen; 262 m->m_pkthdr.rcvif = (struct ifnet *) 0; 263 ti->ti_flags = flags; 264 ti->ti_urp = 0; 265 ti->ti_sum = 0; 266 ti->ti_sum = in_cksum(m, tlen); 267 ((struct ip *)ti)->ip_len = tlen; 268 ((struct ip *)ti)->ip_ttl = ip_defttl; 269 return ip_output(m, NULL, ro, 0, NULL); 270 } 271 272 /* 273 * Create a new TCP control block, making an 274 * empty reassembly queue and hooking it to the argument 275 * protocol control block. 276 */ 277 struct tcpcb * 278 tcp_newtcpcb(inp) 279 struct inpcb *inp; 280 { 281 register struct tcpcb *tp; 282 283 tp = pool_get(&tcpcb_pool, PR_NOWAIT); 284 if (tp == NULL) 285 return (NULL); 286 bzero((caddr_t)tp, sizeof(struct tcpcb)); 287 LIST_INIT(&tp->segq); 288 LIST_INIT(&tp->timeq); 289 tp->t_peermss = tcp_mssdflt; 290 tp->t_ourmss = tcp_mssdflt; 291 tp->t_segsz = tcp_mssdflt; 292 293 tp->t_flags = 0; 294 if (tcp_do_rfc1323 && tcp_do_win_scale) 295 tp->t_flags |= TF_REQ_SCALE; 296 if (tcp_do_rfc1323 && tcp_do_timestamps) 297 tp->t_flags |= TF_REQ_TSTMP; 298 if (tcp_do_sack == 2) 299 tp->t_flags |= TF_WILL_SACK; 300 else if (tcp_do_sack == 1) 301 tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK; 302 tp->t_flags |= TF_CANT_TXSACK; 303 tp->t_inpcb = inp; 304 /* 305 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 306 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 307 * reasonable initial retransmit time. 308 */ 309 tp->t_srtt = TCPTV_SRTTBASE; 310 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 311 tp->t_rttmin = TCPTV_MIN; 312 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 313 TCPTV_MIN, TCPTV_REXMTMAX); 314 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 315 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 316 inp->inp_ip.ip_ttl = ip_defttl; 317 inp->inp_ppcb = (caddr_t)tp; 318 return (tp); 319 } 320 321 /* 322 * Drop a TCP connection, reporting 323 * the specified error. If connection is synchronized, 324 * then send a RST to peer. 325 */ 326 struct tcpcb * 327 tcp_drop(tp, errno) 328 register struct tcpcb *tp; 329 int errno; 330 { 331 struct socket *so = tp->t_inpcb->inp_socket; 332 333 if (TCPS_HAVERCVDSYN(tp->t_state)) { 334 tp->t_state = TCPS_CLOSED; 335 (void) tcp_output(tp); 336 tcpstat.tcps_drops++; 337 } else 338 tcpstat.tcps_conndrops++; 339 if (errno == ETIMEDOUT && tp->t_softerror) 340 errno = tp->t_softerror; 341 so->so_error = errno; 342 return (tcp_close(tp)); 343 } 344 345 /* 346 * Close a TCP control block: 347 * discard all space held by the tcp 348 * discard internet protocol block 349 * wake up any sleepers 350 */ 351 struct tcpcb * 352 tcp_close(tp) 353 register struct tcpcb *tp; 354 { 355 struct inpcb *inp = tp->t_inpcb; 356 struct socket *so = inp->inp_socket; 357 #ifdef RTV_RTT 358 register struct rtentry *rt; 359 360 /* 361 * If we sent enough data to get some meaningful characteristics, 362 * save them in the routing entry. 'Enough' is arbitrarily 363 * defined as the sendpipesize (default 4K) * 16. This would 364 * give us 16 rtt samples assuming we only get one sample per 365 * window (the usual case on a long haul net). 16 samples is 366 * enough for the srtt filter to converge to within 5% of the correct 367 * value; fewer samples and we could save a very bogus rtt. 368 * 369 * Don't update the default route's characteristics and don't 370 * update anything that the user "locked". 371 */ 372 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 373 (rt = inp->inp_route.ro_rt) && 374 !in_nullhost(satosin(rt_key(rt))->sin_addr)) { 375 register u_long i = 0; 376 377 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 378 i = tp->t_srtt * 379 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 380 if (rt->rt_rmx.rmx_rtt && i) 381 /* 382 * filter this update to half the old & half 383 * the new values, converting scale. 384 * See route.h and tcp_var.h for a 385 * description of the scaling constants. 386 */ 387 rt->rt_rmx.rmx_rtt = 388 (rt->rt_rmx.rmx_rtt + i) / 2; 389 else 390 rt->rt_rmx.rmx_rtt = i; 391 } 392 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 393 i = tp->t_rttvar * 394 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 395 if (rt->rt_rmx.rmx_rttvar && i) 396 rt->rt_rmx.rmx_rttvar = 397 (rt->rt_rmx.rmx_rttvar + i) / 2; 398 else 399 rt->rt_rmx.rmx_rttvar = i; 400 } 401 /* 402 * update the pipelimit (ssthresh) if it has been updated 403 * already or if a pipesize was specified & the threshhold 404 * got below half the pipesize. I.e., wait for bad news 405 * before we start updating, then update on both good 406 * and bad news. 407 */ 408 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 409 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 410 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 411 /* 412 * convert the limit from user data bytes to 413 * packets then to packet data bytes. 414 */ 415 i = (i + tp->t_segsz / 2) / tp->t_segsz; 416 if (i < 2) 417 i = 2; 418 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 419 if (rt->rt_rmx.rmx_ssthresh) 420 rt->rt_rmx.rmx_ssthresh = 421 (rt->rt_rmx.rmx_ssthresh + i) / 2; 422 else 423 rt->rt_rmx.rmx_ssthresh = i; 424 } 425 } 426 #endif /* RTV_RTT */ 427 /* free the reassembly queue, if any */ 428 (void) tcp_freeq(tp); 429 TCP_CLEAR_DELACK(tp); 430 431 if (tp->t_template) 432 FREE(tp->t_template, M_MBUF); 433 pool_put(&tcpcb_pool, tp); 434 inp->inp_ppcb = 0; 435 soisdisconnected(so); 436 in_pcbdetach(inp); 437 tcpstat.tcps_closed++; 438 return ((struct tcpcb *)0); 439 } 440 441 int 442 tcp_freeq(tp) 443 struct tcpcb *tp; 444 { 445 register struct ipqent *qe; 446 int rv = 0; 447 #ifdef TCPREASS_DEBUG 448 int i = 0; 449 #endif 450 451 while ((qe = tp->segq.lh_first) != NULL) { 452 #ifdef TCPREASS_DEBUG 453 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 454 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 455 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 456 #endif 457 LIST_REMOVE(qe, ipqe_q); 458 LIST_REMOVE(qe, ipqe_timeq); 459 m_freem(qe->ipqe_m); 460 FREE(qe, M_IPQ); 461 rv = 1; 462 } 463 return (rv); 464 } 465 466 /* 467 * Protocol drain routine. Called when memory is in short supply. 468 */ 469 void 470 tcp_drain() 471 { 472 register struct inpcb *inp; 473 register struct tcpcb *tp; 474 475 /* 476 * Free the sequence queue of all TCP connections. 477 */ 478 inp = tcbtable.inpt_queue.cqh_first; 479 if (inp) /* XXX */ 480 for (; inp != (struct inpcb *)&tcbtable.inpt_queue; 481 inp = inp->inp_queue.cqe_next) { 482 if ((tp = intotcpcb(inp)) != NULL) { 483 if (tcp_freeq(tp)) 484 tcpstat.tcps_connsdrained++; 485 } 486 } 487 } 488 489 /* 490 * Notify a tcp user of an asynchronous error; 491 * store error as soft error, but wake up user 492 * (for now, won't do anything until can select for soft error). 493 */ 494 void 495 tcp_notify(inp, error) 496 struct inpcb *inp; 497 int error; 498 { 499 register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 500 register struct socket *so = inp->inp_socket; 501 502 /* 503 * Ignore some errors if we are hooked up. 504 * If connection hasn't completed, has retransmitted several times, 505 * and receives a second error, give up now. This is better 506 * than waiting a long time to establish a connection that 507 * can never complete. 508 */ 509 if (tp->t_state == TCPS_ESTABLISHED && 510 (error == EHOSTUNREACH || error == ENETUNREACH || 511 error == EHOSTDOWN)) { 512 return; 513 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 514 tp->t_rxtshift > 3 && tp->t_softerror) 515 so->so_error = error; 516 else 517 tp->t_softerror = error; 518 wakeup((caddr_t) &so->so_timeo); 519 sorwakeup(so); 520 sowwakeup(so); 521 } 522 523 void * 524 tcp_ctlinput(cmd, sa, v) 525 int cmd; 526 struct sockaddr *sa; 527 register void *v; 528 { 529 register struct ip *ip = v; 530 register struct tcphdr *th; 531 extern int inetctlerrmap[]; 532 void (*notify) __P((struct inpcb *, int)) = tcp_notify; 533 int errno; 534 int nmatch; 535 536 if ((unsigned)cmd >= PRC_NCMDS) 537 return NULL; 538 errno = inetctlerrmap[cmd]; 539 if (cmd == PRC_QUENCH) 540 notify = tcp_quench; 541 else if (PRC_IS_REDIRECT(cmd)) 542 notify = in_rtchange, ip = 0; 543 else if (cmd == PRC_MSGSIZE && ip_mtudisc) 544 notify = tcp_mtudisc, ip = 0; 545 else if (cmd == PRC_HOSTDEAD) 546 ip = 0; 547 else if (errno == 0) 548 return NULL; 549 if (ip) { 550 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 551 nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr, 552 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 553 if (nmatch == 0 && syn_cache_count && 554 (inetctlerrmap[cmd] == EHOSTUNREACH || 555 inetctlerrmap[cmd] == ENETUNREACH || 556 inetctlerrmap[cmd] == EHOSTDOWN)) 557 syn_cache_unreach(ip, th); 558 } else 559 (void)in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno, 560 notify); 561 return NULL; 562 } 563 564 /* 565 * When a source quence is received, we are being notifed of congestion. 566 * Close the congestion window down to the Loss Window (one segment). 567 * We will gradually open it again as we proceed. 568 */ 569 void 570 tcp_quench(inp, errno) 571 struct inpcb *inp; 572 int errno; 573 { 574 struct tcpcb *tp = intotcpcb(inp); 575 576 if (tp) 577 tp->snd_cwnd = tp->t_segsz; 578 } 579 580 /* 581 * On receipt of path MTU corrections, flush old route and replace it 582 * with the new one. Retransmit all unacknowledged packets, to ensure 583 * that all packets will be received. 584 */ 585 void 586 tcp_mtudisc(inp, errno) 587 struct inpcb *inp; 588 int errno; 589 { 590 struct tcpcb *tp = intotcpcb(inp); 591 struct rtentry *rt = in_pcbrtentry(inp); 592 593 if (tp != 0) { 594 if (rt != 0) { 595 /* 596 * If this was not a host route, remove and realloc. 597 */ 598 if ((rt->rt_flags & RTF_HOST) == 0) { 599 in_rtchange(inp, errno); 600 if ((rt = in_pcbrtentry(inp)) == 0) 601 return; 602 } 603 604 /* 605 * Slow start out of the error condition. We 606 * use the MTU because we know it's smaller 607 * than the previously transmitted segment. 608 * 609 * Note: This is more conservative than the 610 * suggestion in draft-floyd-incr-init-win-03. 611 */ 612 if (rt->rt_rmx.rmx_mtu != 0) 613 tp->snd_cwnd = 614 TCP_INITIAL_WINDOW(tcp_init_win, 615 rt->rt_rmx.rmx_mtu); 616 } 617 618 /* 619 * Resend unacknowledged packets. 620 */ 621 tp->snd_nxt = tp->snd_una; 622 tcp_output(tp); 623 } 624 } 625 626 627 /* 628 * Compute the MSS to advertise to the peer. Called only during 629 * the 3-way handshake. If we are the server (peer initiated 630 * connection), we are called with a pointer to the interface 631 * on which the SYN packet arrived. If we are the client (we 632 * initiated connection), we are called with a pointer to the 633 * interface out which this connection should go. 634 */ 635 u_long 636 tcp_mss_to_advertise(ifp) 637 const struct ifnet *ifp; 638 { 639 extern u_long in_maxmtu; 640 u_long mss = 0; 641 642 /* 643 * In order to avoid defeating path MTU discovery on the peer, 644 * we advertise the max MTU of all attached networks as our MSS, 645 * per RFC 1191, section 3.1. 646 * 647 * We provide the option to advertise just the MTU of 648 * the interface on which we hope this connection will 649 * be receiving. If we are responding to a SYN, we 650 * will have a pretty good idea about this, but when 651 * initiating a connection there is a bit more doubt. 652 * 653 * We also need to ensure that loopback has a large enough 654 * MSS, as the loopback MTU is never included in in_maxmtu. 655 */ 656 657 if (ifp != NULL) 658 mss = ifp->if_mtu; 659 660 if (tcp_mss_ifmtu == 0) 661 mss = max(in_maxmtu, mss); 662 663 if (mss > sizeof(struct tcpiphdr)) 664 mss -= sizeof(struct tcpiphdr); 665 666 mss = max(tcp_mssdflt, mss); 667 return (mss); 668 } 669 670 /* 671 * Set connection variables based on the peer's advertised MSS. 672 * We are passed the TCPCB for the actual connection. If we 673 * are the server, we are called by the compressed state engine 674 * when the 3-way handshake is complete. If we are the client, 675 * we are called when we recieve the SYN,ACK from the server. 676 * 677 * NOTE: Our advertised MSS value must be initialized in the TCPCB 678 * before this routine is called! 679 */ 680 void 681 tcp_mss_from_peer(tp, offer) 682 struct tcpcb *tp; 683 int offer; 684 { 685 struct inpcb *inp = tp->t_inpcb; 686 struct socket *so = inp->inp_socket; 687 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 688 struct rtentry *rt = in_pcbrtentry(inp); 689 #endif 690 u_long bufsize; 691 int mss; 692 693 /* 694 * As per RFC1122, use the default MSS value, unless they 695 * sent us an offer. Do not accept offers less than 32 bytes. 696 */ 697 mss = tcp_mssdflt; 698 if (offer) 699 mss = offer; 700 mss = max(mss, 32); /* sanity */ 701 tp->t_peermss = mss; 702 mss -= (tcp_optlen(tp) + ip_optlen(tp->t_inpcb)); 703 704 /* 705 * If there's a pipesize, change the socket buffer to that size. 706 * Make the socket buffer an integral number of MSS units. If 707 * the MSS is larger than the socket buffer, artificially decrease 708 * the MSS. 709 */ 710 #ifdef RTV_SPIPE 711 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 712 bufsize = rt->rt_rmx.rmx_sendpipe; 713 else 714 #endif 715 bufsize = so->so_snd.sb_hiwat; 716 if (bufsize < mss) 717 mss = bufsize; 718 else { 719 bufsize = roundup(bufsize, mss); 720 if (bufsize > sb_max) 721 bufsize = sb_max; 722 (void) sbreserve(&so->so_snd, bufsize); 723 } 724 tp->t_segsz = mss; 725 726 #ifdef RTV_SSTHRESH 727 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 728 /* 729 * There's some sort of gateway or interface buffer 730 * limit on the path. Use this to set the slow 731 * start threshold, but set the threshold to no less 732 * than 2 * MSS. 733 */ 734 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 735 } 736 #endif 737 } 738 739 /* 740 * Processing necessary when a TCP connection is established. 741 */ 742 void 743 tcp_established(tp) 744 struct tcpcb *tp; 745 { 746 struct inpcb *inp = tp->t_inpcb; 747 struct socket *so = inp->inp_socket; 748 #ifdef RTV_RPIPE 749 struct rtentry *rt = in_pcbrtentry(inp); 750 #endif 751 u_long bufsize; 752 753 tp->t_state = TCPS_ESTABLISHED; 754 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 755 756 #ifdef RTV_RPIPE 757 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 758 bufsize = rt->rt_rmx.rmx_recvpipe; 759 else 760 #endif 761 bufsize = so->so_rcv.sb_hiwat; 762 if (bufsize > tp->t_ourmss) { 763 bufsize = roundup(bufsize, tp->t_ourmss); 764 if (bufsize > sb_max) 765 bufsize = sb_max; 766 (void) sbreserve(&so->so_rcv, bufsize); 767 } 768 } 769 770 /* 771 * Check if there's an initial rtt or rttvar. Convert from the 772 * route-table units to scaled multiples of the slow timeout timer. 773 * Called only during the 3-way handshake. 774 */ 775 void 776 tcp_rmx_rtt(tp) 777 struct tcpcb *tp; 778 { 779 #ifdef RTV_RTT 780 struct rtentry *rt; 781 int rtt; 782 783 if ((rt = in_pcbrtentry(tp->t_inpcb)) == NULL) 784 return; 785 786 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 787 /* 788 * XXX The lock bit for MTU indicates that the value 789 * is also a minimum value; this is subject to time. 790 */ 791 if (rt->rt_rmx.rmx_locks & RTV_RTT) 792 TCPT_RANGESET(tp->t_rttmin, 793 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 794 TCPTV_MIN, TCPTV_REXMTMAX); 795 tp->t_srtt = rtt / 796 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 797 if (rt->rt_rmx.rmx_rttvar) { 798 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 799 ((RTM_RTTUNIT / PR_SLOWHZ) >> 800 (TCP_RTTVAR_SHIFT + 2)); 801 } else { 802 /* Default variation is +- 1 rtt */ 803 tp->t_rttvar = 804 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 805 } 806 TCPT_RANGESET(tp->t_rxtcur, 807 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 808 tp->t_rttmin, TCPTV_REXMTMAX); 809 } 810 #endif 811 } 812 813 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 814 815 /* 816 * Get a new sequence value given a tcp control block 817 */ 818 tcp_seq 819 tcp_new_iss(tp, len, addin) 820 void *tp; 821 u_long len; 822 tcp_seq addin; 823 { 824 tcp_seq tcp_iss; 825 826 /* 827 * add randomness about this connection, but do not estimate 828 * entropy from the timing, since the physical device driver would 829 * have done that for us. 830 */ 831 #if NRND > 0 832 if (tp != NULL) 833 rnd_add_data(NULL, tp, len, 0); 834 #endif 835 836 /* 837 * randomize. 838 */ 839 #if NRND > 0 840 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 841 #else 842 tcp_iss = random(); 843 #endif 844 845 /* 846 * If we were asked to add some amount to a known value, 847 * we will take a random value obtained above, mask off the upper 848 * bits, and add in the known value. We also add in a constant to 849 * ensure that we are at least a certain distance from the original 850 * value. 851 * 852 * This is used when an old connection is in timed wait 853 * and we have a new one coming in, for instance. 854 */ 855 if (addin != 0) { 856 #ifdef TCPISS_DEBUG 857 printf("Random %08x, ", tcp_iss); 858 #endif 859 tcp_iss &= TCP_ISS_RANDOM_MASK; 860 tcp_iss += addin + TCP_ISSINCR; 861 #ifdef TCPISS_DEBUG 862 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 863 #endif 864 } else { 865 tcp_iss &= TCP_ISS_RANDOM_MASK; 866 tcp_iss += tcp_iss_seq; 867 tcp_iss_seq += TCP_ISSINCR; 868 #ifdef TCPISS_DEBUG 869 printf("ISS %08x\n", tcp_iss); 870 #endif 871 } 872 873 if (tcp_compat_42) { 874 /* 875 * Limit it to the positive range for really old TCP 876 * implementations. 877 */ 878 if (tcp_iss >= 0x80000000) 879 tcp_iss &= 0x7fffffff; /* XXX */ 880 } 881 882 return tcp_iss; 883 } 884 885 886 /* 887 * Determine the length of the TCP options for this connection. 888 * 889 * XXX: What do we do for SACK, when we add that? Just reserve 890 * all of the space? Otherwise we can't exactly be incrementing 891 * cwnd by an amount that varies depending on the amount we last 892 * had to SACK! 893 */ 894 895 u_int 896 tcp_optlen(tp) 897 struct tcpcb *tp; 898 { 899 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 900 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 901 return TCPOLEN_TSTAMP_APPA; 902 else 903 return 0; 904 } 905
Indexes created Sun Sep 21 20:09:37 GMT 2025