tcp_output.c revision 1.63.2.3
1 /* $NetBSD: tcp_output.c,v 1.63.2.3 2001/08/24 00:12:30 nathanw Exp $ */ 2 3 /* 4 %%% portions-copyright-nrl-95 5 Portions of this software are Copyright 1995-1998 by Randall Atkinson, 6 Ronald Lee, Daniel McDonald, Bao Phan, and Chris Winters. All Rights 7 Reserved. All rights under this copyright have been assigned to the US 8 Naval Research Laboratory (NRL). The NRL Copyright Notice and License 9 Agreement Version 1.1 (January 17, 1995) applies to these portions of the 10 software. 11 You should have received a copy of the license with this software. If you 12 didn't get a copy, you may request one from <license (at) ipv6.nrl.navy.mil>. 13 14 */ 15 16 /* 17 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 18 * All rights reserved. 19 * 20 * Redistribution and use in source and binary forms, with or without 21 * modification, are permitted provided that the following conditions 22 * are met: 23 * 1. Redistributions of source code must retain the above copyright 24 * notice, this list of conditions and the following disclaimer. 25 * 2. Redistributions in binary form must reproduce the above copyright 26 * notice, this list of conditions and the following disclaimer in the 27 * documentation and/or other materials provided with the distribution. 28 * 3. Neither the name of the project nor the names of its contributors 29 * may be used to endorse or promote products derived from this software 30 * without specific prior written permission. 31 * 32 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 33 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 34 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 35 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 36 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 37 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 38 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 39 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 40 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 41 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 42 * SUCH DAMAGE. 43 */ 44 45 /*- 46 * Copyright (c) 1997, 1998, 2001 The NetBSD Foundation, Inc. 47 * All rights reserved. 48 * 49 * This code is derived from software contributed to The NetBSD Foundation 50 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 51 * Facility, NASA Ames Research Center. 52 * 53 * Redistribution and use in source and binary forms, with or without 54 * modification, are permitted provided that the following conditions 55 * are met: 56 * 1. Redistributions of source code must retain the above copyright 57 * notice, this list of conditions and the following disclaimer. 58 * 2. Redistributions in binary form must reproduce the above copyright 59 * notice, this list of conditions and the following disclaimer in the 60 * documentation and/or other materials provided with the distribution. 61 * 3. All advertising materials mentioning features or use of this software 62 * must display the following acknowledgement: 63 * This product includes software developed by the NetBSD 64 * Foundation, Inc. and its contributors. 65 * 4. Neither the name of The NetBSD Foundation nor the names of its 66 * contributors may be used to endorse or promote products derived 67 * from this software without specific prior written permission. 68 * 69 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 70 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 71 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 72 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 73 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 74 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 75 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 76 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 77 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 78 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 79 * POSSIBILITY OF SUCH DAMAGE. 80 */ 81 82 /* 83 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 84 * The Regents of the University of California. All rights reserved. 85 * 86 * Redistribution and use in source and binary forms, with or without 87 * modification, are permitted provided that the following conditions 88 * are met: 89 * 1. Redistributions of source code must retain the above copyright 90 * notice, this list of conditions and the following disclaimer. 91 * 2. Redistributions in binary form must reproduce the above copyright 92 * notice, this list of conditions and the following disclaimer in the 93 * documentation and/or other materials provided with the distribution. 94 * 3. All advertising materials mentioning features or use of this software 95 * must display the following acknowledgement: 96 * This product includes software developed by the University of 97 * California, Berkeley and its contributors. 98 * 4. Neither the name of the University nor the names of its contributors 99 * may be used to endorse or promote products derived from this software 100 * without specific prior written permission. 101 * 102 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 103 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 104 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 105 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 106 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 107 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 108 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 109 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 110 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 111 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 112 * SUCH DAMAGE. 113 * 114 * @(#)tcp_output.c 8.4 (Berkeley) 5/24/95 115 */ 116 117 #include "opt_inet.h" 118 #include "opt_ipsec.h" 119 #include "opt_tcp_debug.h" 120 121 #include <sys/param.h> 122 #include <sys/systm.h> 123 #include <sys/malloc.h> 124 #include <sys/mbuf.h> 125 #include <sys/protosw.h> 126 #include <sys/socket.h> 127 #include <sys/socketvar.h> 128 #include <sys/errno.h> 129 #include <sys/domain.h> 130 131 #include <net/if.h> 132 #include <net/route.h> 133 134 #include <netinet/in.h> 135 #include <netinet/in_systm.h> 136 #include <netinet/ip.h> 137 #include <netinet/in_pcb.h> 138 #include <netinet/ip_var.h> 139 140 #ifdef INET6 141 #ifndef INET 142 #include <netinet/in.h> 143 #endif 144 #include <netinet/ip6.h> 145 #include <netinet6/in6_pcb.h> 146 #include <netinet6/ip6_var.h> 147 #endif 148 149 #include <netinet/tcp.h> 150 #define TCPOUTFLAGS 151 #include <netinet/tcp_fsm.h> 152 #include <netinet/tcp_seq.h> 153 #include <netinet/tcp_timer.h> 154 #include <netinet/tcp_var.h> 155 #include <netinet/tcpip.h> 156 #include <netinet/tcp_debug.h> 157 158 #ifdef notyet 159 extern struct mbuf *m_copypack(); 160 #endif 161 162 #define MAX_TCPOPTLEN 32 /* max # bytes that go in options */ 163 164 /* 165 * Knob to enable Congestion Window Monitoring, and control the 166 * the burst size it allows. Default burst is 4 packets, per 167 * the Internet draft. 168 */ 169 int tcp_cwm = 0; 170 int tcp_cwm_burstsize = 4; 171 172 static 173 #ifndef GPROF 174 __inline 175 #endif 176 void 177 tcp_segsize(struct tcpcb *tp, int *txsegsizep, int *rxsegsizep) 178 { 179 #ifdef INET 180 struct inpcb *inp = tp->t_inpcb; 181 #endif 182 #ifdef INET6 183 struct in6pcb *in6p = tp->t_in6pcb; 184 #endif 185 struct rtentry *rt; 186 struct ifnet *ifp; 187 int size; 188 int iphlen; 189 190 #ifdef DIAGNOSTIC 191 if (tp->t_inpcb && tp->t_in6pcb) 192 panic("tcp_segsize: both t_inpcb and t_in6pcb are set"); 193 #endif 194 switch (tp->t_family) { 195 #ifdef INET 196 case AF_INET: 197 iphlen = sizeof(struct ip); 198 break; 199 #endif 200 #ifdef INET6 201 case AF_INET6: 202 iphlen = sizeof(struct ip6_hdr); 203 break; 204 #endif 205 default: 206 size = tcp_mssdflt; 207 goto out; 208 } 209 210 rt = NULL; 211 #ifdef INET 212 if (inp) 213 rt = in_pcbrtentry(inp); 214 #endif 215 #ifdef INET6 216 if (in6p) 217 rt = in6_pcbrtentry(in6p); 218 #endif 219 if (rt == NULL) { 220 size = tcp_mssdflt; 221 goto out; 222 } 223 224 ifp = rt->rt_ifp; 225 226 size = tcp_mssdflt; 227 if (rt->rt_rmx.rmx_mtu != 0) 228 size = rt->rt_rmx.rmx_mtu - iphlen - sizeof(struct tcphdr); 229 else if (ifp->if_flags & IFF_LOOPBACK) 230 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 231 #ifdef INET 232 else if (inp && ip_mtudisc) 233 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 234 else if (inp && in_localaddr(inp->inp_faddr)) 235 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 236 #endif 237 #ifdef INET6 238 else if (in6p) { 239 #ifdef INET 240 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) { 241 /* mapped addr case */ 242 struct in_addr d; 243 bcopy(&in6p->in6p_faddr.s6_addr32[3], &d, sizeof(d)); 244 if (ip_mtudisc || in_localaddr(d)) 245 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 246 } else 247 #endif 248 { 249 /* 250 * for IPv6, path MTU discovery is always turned on, 251 * or the node must use packet size <= 1280. 252 */ 253 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 254 } 255 } 256 #endif 257 size -= tcp_optlen(tp); 258 /* 259 * XXX tp->t_ourmss should have the right size, but without this code 260 * fragmentation will occur... need more investigation 261 */ 262 #ifdef INET 263 if (inp) { 264 #ifdef IPSEC 265 size -= ipsec4_hdrsiz_tcp(tp); 266 #endif 267 size -= ip_optlen(inp); 268 } 269 #endif 270 #ifdef INET6 271 #ifdef INET 272 if (in6p && tp->t_family == AF_INET) { 273 #ifdef IPSEC 274 size -= ipsec4_hdrsiz_tcp(tp); 275 #endif 276 /* XXX size -= ip_optlen(in6p); */ 277 } else 278 #endif 279 if (in6p && tp->t_family == AF_INET6) { 280 #ifdef IPSEC 281 size -= ipsec6_hdrsiz_tcp(tp); 282 #endif 283 size -= ip6_optlen(in6p); 284 } 285 #endif 286 287 out: 288 /* 289 * *rxsegsizep holds *estimated* inbound segment size (estimation 290 * assumes that path MTU is the same for both ways). this is only 291 * for silly window avoidance, do not use the value for other purposes. 292 * 293 * ipseclen is subtracted from both sides, this may not be right. 294 * I'm not quite sure about this (could someone comment). 295 */ 296 *txsegsizep = min(tp->t_peermss, size); 297 *rxsegsizep = min(tp->t_ourmss, size); 298 299 if (*txsegsizep != tp->t_segsz) { 300 /* 301 * If the new segment size is larger, we don't want to 302 * mess up the congestion window, but if it is smaller 303 * we'll have to reduce the congestion window to ensure 304 * that we don't get into trouble with initial windows 305 * and the rest. In any case, if the segment size 306 * has changed, chances are the path has, too, and 307 * our congestion window will be different. 308 */ 309 if (*txsegsizep < tp->t_segsz) { 310 tp->snd_cwnd = max((tp->snd_cwnd / tp->t_segsz) 311 * *txsegsizep, *txsegsizep); 312 tp->snd_ssthresh = max((tp->snd_ssthresh / tp->t_segsz) 313 * *txsegsizep, *txsegsizep); 314 } 315 tp->t_segsz = *txsegsizep; 316 } 317 } 318 319 static 320 #ifndef GPROF 321 __inline 322 #endif 323 int 324 tcp_build_datapkt(struct tcpcb *tp, struct socket *so, int off, 325 long len, int hdrlen, struct mbuf **mp) 326 { 327 struct mbuf *m; 328 329 if (tp->t_force && len == 1) 330 tcpstat.tcps_sndprobe++; 331 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { 332 tcpstat.tcps_sndrexmitpack++; 333 tcpstat.tcps_sndrexmitbyte += len; 334 } else { 335 tcpstat.tcps_sndpack++; 336 tcpstat.tcps_sndbyte += len; 337 } 338 #ifdef notyet 339 if ((m = m_copypack(so->so_snd.sb_mb, off, 340 (int)len, max_linkhdr + hdrlen)) == 0) 341 return (ENOBUFS); 342 /* 343 * m_copypack left space for our hdr; use it. 344 */ 345 m->m_len += hdrlen; 346 m->m_data -= hdrlen; 347 #else 348 MGETHDR(m, M_DONTWAIT, MT_HEADER); 349 if (m != NULL && 350 (max_linkhdr + hdrlen > MHLEN || 351 max_linkhdr + hdrlen + len <= MCLBYTES)) { 352 MCLGET(m, M_DONTWAIT); 353 if ((m->m_flags & M_EXT) == 0) { 354 m_freem(m); 355 m = NULL; 356 } 357 } 358 if (m == NULL) 359 return (ENOBUFS); 360 m->m_data += max_linkhdr; 361 m->m_len = hdrlen; 362 if (len <= M_TRAILINGSPACE(m)) { 363 m_copydata(so->so_snd.sb_mb, off, (int) len, 364 mtod(m, caddr_t) + hdrlen); 365 m->m_len += len; 366 } else { 367 m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len); 368 if (m->m_next == NULL) { 369 m_freem(m); 370 return (ENOBUFS); 371 } 372 } 373 #endif 374 375 *mp = m; 376 return (0); 377 } 378 379 /* 380 * Tcp output routine: figure out what should be sent and send it. 381 */ 382 int 383 tcp_output(tp) 384 struct tcpcb *tp; 385 { 386 struct socket *so; 387 struct route *ro; 388 long len, win; 389 int off, flags, error; 390 struct mbuf *m; 391 struct ip *ip; 392 #ifdef INET6 393 struct ip6_hdr *ip6; 394 #endif 395 struct tcphdr *th; 396 u_char opt[MAX_TCPOPTLEN]; 397 unsigned optlen, hdrlen; 398 int idle, sendalot, txsegsize, rxsegsize; 399 int maxburst = TCP_MAXBURST; 400 int af; /* address family on the wire */ 401 int iphdrlen; 402 403 #ifdef DIAGNOSTIC 404 if (tp->t_inpcb && tp->t_in6pcb) 405 panic("tcp_output: both t_inpcb and t_in6pcb are set"); 406 #endif 407 so = NULL; 408 ro = NULL; 409 if (tp->t_inpcb) { 410 so = tp->t_inpcb->inp_socket; 411 ro = &tp->t_inpcb->inp_route; 412 } 413 #ifdef INET6 414 else if (tp->t_in6pcb) { 415 so = tp->t_in6pcb->in6p_socket; 416 ro = (struct route *)&tp->t_in6pcb->in6p_route; 417 } 418 #endif 419 420 switch (af = tp->t_family) { 421 #ifdef INET 422 case AF_INET: 423 if (tp->t_inpcb) 424 break; 425 #ifdef INET6 426 /* mapped addr case */ 427 if (tp->t_in6pcb) 428 break; 429 #endif 430 return EINVAL; 431 #endif 432 #ifdef INET6 433 case AF_INET6: 434 if (tp->t_in6pcb) 435 break; 436 return EINVAL; 437 #endif 438 default: 439 return EAFNOSUPPORT; 440 } 441 442 tcp_segsize(tp, &txsegsize, &rxsegsize); 443 444 idle = (tp->snd_max == tp->snd_una); 445 446 /* 447 * Restart Window computation. From draft-floyd-incr-init-win-03: 448 * 449 * Optionally, a TCP MAY set the restart window to the 450 * minimum of the value used for the initial window and 451 * the current value of cwnd (in other words, using a 452 * larger value for the restart window should never increase 453 * the size of cwnd). 454 */ 455 if (tcp_cwm) { 456 /* 457 * Hughes/Touch/Heidemann Congestion Window Monitoring. 458 * Count the number of packets currently pending 459 * acknowledgement, and limit our congestion window 460 * to a pre-determined allowed burst size plus that count. 461 * This prevents bursting once all pending packets have 462 * been acknowledged (i.e. transmission is idle). 463 * 464 * XXX Link this to Initial Window? 465 */ 466 tp->snd_cwnd = min(tp->snd_cwnd, 467 (tcp_cwm_burstsize * txsegsize) + 468 (tp->snd_nxt - tp->snd_una)); 469 } else { 470 if (idle && tp->t_idle >= tp->t_rxtcur) { 471 /* 472 * We have been idle for "a while" and no acks are 473 * expected to clock out any data we send -- 474 * slow start to get ack "clock" running again. 475 */ 476 tp->snd_cwnd = min(tp->snd_cwnd, 477 TCP_INITIAL_WINDOW(tcp_init_win, txsegsize)); 478 } 479 } 480 481 again: 482 /* 483 * Determine length of data that should be transmitted, and 484 * flags that should be used. If there is some data or critical 485 * controls (SYN, RST) to send, then transmit; otherwise, 486 * investigate further. 487 */ 488 sendalot = 0; 489 off = tp->snd_nxt - tp->snd_una; 490 win = min(tp->snd_wnd, tp->snd_cwnd); 491 492 flags = tcp_outflags[tp->t_state]; 493 /* 494 * If in persist timeout with window of 0, send 1 byte. 495 * Otherwise, if window is small but nonzero 496 * and timer expired, we will send what we can 497 * and go to transmit state. 498 */ 499 if (tp->t_force) { 500 if (win == 0) { 501 /* 502 * If we still have some data to send, then 503 * clear the FIN bit. Usually this would 504 * happen below when it realizes that we 505 * aren't sending all the data. However, 506 * if we have exactly 1 byte of unset data, 507 * then it won't clear the FIN bit below, 508 * and if we are in persist state, we wind 509 * up sending the packet without recording 510 * that we sent the FIN bit. 511 * 512 * We can't just blindly clear the FIN bit, 513 * because if we don't have any more data 514 * to send then the probe will be the FIN 515 * itself. 516 */ 517 if (off < so->so_snd.sb_cc) 518 flags &= ~TH_FIN; 519 win = 1; 520 } else { 521 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 522 tp->t_rxtshift = 0; 523 } 524 } 525 526 if (win < so->so_snd.sb_cc) { 527 len = win - off; 528 flags &= ~TH_FIN; 529 } else 530 len = so->so_snd.sb_cc - off; 531 532 if (len < 0) { 533 /* 534 * If FIN has been sent but not acked, 535 * but we haven't been called to retransmit, 536 * len will be -1. Otherwise, window shrank 537 * after we sent into it. If window shrank to 0, 538 * cancel pending retransmit, pull snd_nxt back 539 * to (closed) window, and set the persist timer 540 * if it isn't already going. If the window didn't 541 * close completely, just wait for an ACK. 542 * 543 * If we have a pending FIN, either it has already been 544 * transmitted or it is outside the window, so drop it. 545 * If the FIN has been transmitted, but this is not a 546 * retransmission, then len must be -1. Therefore we also 547 * prevent here the sending of `gratuitous FINs'. This 548 * eliminates the need to check for that case below (e.g. 549 * to back up snd_nxt before the FIN so that the sequence 550 * number is correct). 551 */ 552 len = 0; 553 flags &= ~TH_FIN; 554 if (win == 0) { 555 TCP_TIMER_DISARM(tp, TCPT_REXMT); 556 tp->t_rxtshift = 0; 557 tp->snd_nxt = tp->snd_una; 558 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 559 tcp_setpersist(tp); 560 } 561 } 562 if (len > txsegsize) { 563 len = txsegsize; 564 flags &= ~TH_FIN; 565 sendalot = 1; 566 } 567 568 win = sbspace(&so->so_rcv); 569 570 /* 571 * Sender silly window avoidance. If connection is idle 572 * and can send all data, a maximum segment, 573 * at least a maximum default-size segment do it, 574 * or are forced, do it; otherwise don't bother. 575 * If peer's buffer is tiny, then send 576 * when window is at least half open. 577 * If retransmitting (possibly after persist timer forced us 578 * to send into a small window), then must resend. 579 */ 580 if (len) { 581 if (len == txsegsize) 582 goto send; 583 if ((so->so_state & SS_MORETOCOME) == 0 && 584 ((idle || tp->t_flags & TF_NODELAY) && 585 len + off >= so->so_snd.sb_cc)) 586 goto send; 587 if (tp->t_force) 588 goto send; 589 if (len >= tp->max_sndwnd / 2) 590 goto send; 591 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) 592 goto send; 593 } 594 595 /* 596 * Compare available window to amount of window known to peer 597 * (as advertised window less next expected input). If the 598 * difference is at least twice the size of the largest segment 599 * we expect to receive (i.e. two segments) or at least 50% of 600 * the maximum possible window, then want to send a window update 601 * to peer. 602 */ 603 if (win > 0) { 604 /* 605 * "adv" is the amount we can increase the window, 606 * taking into account that we are limited by 607 * TCP_MAXWIN << tp->rcv_scale. 608 */ 609 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) - 610 (tp->rcv_adv - tp->rcv_nxt); 611 612 if (adv >= (long) (2 * rxsegsize)) 613 goto send; 614 if (2 * adv >= (long) so->so_rcv.sb_hiwat) 615 goto send; 616 } 617 618 /* 619 * Send if we owe peer an ACK. 620 */ 621 if (tp->t_flags & TF_ACKNOW) 622 goto send; 623 if (flags & (TH_SYN|TH_FIN|TH_RST)) 624 goto send; 625 if (SEQ_GT(tp->snd_up, tp->snd_una)) 626 goto send; 627 628 /* 629 * TCP window updates are not reliable, rather a polling protocol 630 * using ``persist'' packets is used to insure receipt of window 631 * updates. The three ``states'' for the output side are: 632 * idle not doing retransmits or persists 633 * persisting to move a small or zero window 634 * (re)transmitting and thereby not persisting 635 * 636 * tp->t_timer[TCPT_PERSIST] 637 * is set when we are in persist state. 638 * tp->t_force 639 * is set when we are called to send a persist packet. 640 * tp->t_timer[TCPT_REXMT] 641 * is set when we are retransmitting 642 * The output side is idle when both timers are zero. 643 * 644 * If send window is too small, there is data to transmit, and no 645 * retransmit or persist is pending, then go to persist state. 646 * If nothing happens soon, send when timer expires: 647 * if window is nonzero, transmit what we can, 648 * otherwise force out a byte. 649 */ 650 if (so->so_snd.sb_cc && TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 651 TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 652 tp->t_rxtshift = 0; 653 tcp_setpersist(tp); 654 } 655 656 /* 657 * No reason to send a segment, just return. 658 */ 659 return (0); 660 661 send: 662 /* 663 * Before ESTABLISHED, force sending of initial options 664 * unless TCP set not to do any options. 665 * NOTE: we assume that the IP/TCP header plus TCP options 666 * always fit in a single mbuf, leaving room for a maximum 667 * link header, i.e. 668 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES 669 */ 670 optlen = 0; 671 switch (af) { 672 #ifdef INET 673 case AF_INET: 674 iphdrlen = sizeof(struct ip) + sizeof(struct tcphdr); 675 break; 676 #endif 677 #ifdef INET6 678 case AF_INET6: 679 iphdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 680 break; 681 #endif 682 default: /*pacify gcc*/ 683 iphdrlen = 0; 684 break; 685 } 686 hdrlen = iphdrlen; 687 if (flags & TH_SYN) { 688 struct rtentry *rt; 689 690 rt = NULL; 691 #ifdef INET 692 if (tp->t_inpcb) 693 rt = in_pcbrtentry(tp->t_inpcb); 694 #endif 695 #ifdef INET6 696 if (tp->t_in6pcb) 697 rt = in6_pcbrtentry(tp->t_in6pcb); 698 #endif 699 700 tp->snd_nxt = tp->iss; 701 tp->t_ourmss = tcp_mss_to_advertise(rt != NULL ? 702 rt->rt_ifp : NULL, af); 703 if ((tp->t_flags & TF_NOOPT) == 0) { 704 opt[0] = TCPOPT_MAXSEG; 705 opt[1] = 4; 706 opt[2] = (tp->t_ourmss >> 8) & 0xff; 707 opt[3] = tp->t_ourmss & 0xff; 708 optlen = 4; 709 710 if ((tp->t_flags & TF_REQ_SCALE) && 711 ((flags & TH_ACK) == 0 || 712 (tp->t_flags & TF_RCVD_SCALE))) { 713 *((u_int32_t *) (opt + optlen)) = htonl( 714 TCPOPT_NOP << 24 | 715 TCPOPT_WINDOW << 16 | 716 TCPOLEN_WINDOW << 8 | 717 tp->request_r_scale); 718 optlen += 4; 719 } 720 } 721 } 722 723 /* 724 * Send a timestamp and echo-reply if this is a SYN and our side 725 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side 726 * and our peer have sent timestamps in our SYN's. 727 */ 728 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 729 (flags & TH_RST) == 0 && 730 ((flags & (TH_SYN|TH_ACK)) == TH_SYN || 731 (tp->t_flags & TF_RCVD_TSTMP))) { 732 u_int32_t *lp = (u_int32_t *)(opt + optlen); 733 734 /* Form timestamp option as shown in appendix A of RFC 1323. */ 735 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 736 *lp++ = htonl(TCP_TIMESTAMP(tp)); 737 *lp = htonl(tp->ts_recent); 738 optlen += TCPOLEN_TSTAMP_APPA; 739 } 740 741 hdrlen += optlen; 742 743 #ifdef DIAGNOSTIC 744 if (len > txsegsize) 745 panic("tcp data to be sent is larger than segment"); 746 if (max_linkhdr + hdrlen > MCLBYTES) 747 panic("tcphdr too big"); 748 #endif 749 750 /* 751 * Grab a header mbuf, attaching a copy of data to 752 * be transmitted, and initialize the header from 753 * the template for sends on this connection. 754 */ 755 if (len) { 756 error = tcp_build_datapkt(tp, so, off, len, hdrlen, &m); 757 if (error) 758 goto out; 759 /* 760 * If we're sending everything we've got, set PUSH. 761 * (This will keep happy those implementations which only 762 * give data to the user when a buffer fills or 763 * a PUSH comes in.) 764 */ 765 if (off + len == so->so_snd.sb_cc) 766 flags |= TH_PUSH; 767 } else { 768 if (tp->t_flags & TF_ACKNOW) 769 tcpstat.tcps_sndacks++; 770 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 771 tcpstat.tcps_sndctrl++; 772 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 773 tcpstat.tcps_sndurg++; 774 else 775 tcpstat.tcps_sndwinup++; 776 777 MGETHDR(m, M_DONTWAIT, MT_HEADER); 778 if (m != NULL && max_linkhdr + hdrlen > MHLEN) { 779 MCLGET(m, M_DONTWAIT); 780 if ((m->m_flags & M_EXT) == 0) { 781 m_freem(m); 782 m = NULL; 783 } 784 } 785 if (m == NULL) { 786 error = ENOBUFS; 787 goto out; 788 } 789 m->m_data += max_linkhdr; 790 m->m_len = hdrlen; 791 } 792 m->m_pkthdr.rcvif = (struct ifnet *)0; 793 switch (af) { 794 #ifdef INET 795 case AF_INET: 796 ip = mtod(m, struct ip *); 797 #ifdef INET6 798 ip6 = NULL; 799 #endif 800 th = (struct tcphdr *)(ip + 1); 801 break; 802 #endif 803 #ifdef INET6 804 case AF_INET6: 805 ip = NULL; 806 ip6 = mtod(m, struct ip6_hdr *); 807 th = (struct tcphdr *)(ip6 + 1); 808 break; 809 #endif 810 default: /*pacify gcc*/ 811 ip = NULL; 812 #ifdef INET6 813 ip6 = NULL; 814 #endif 815 th = NULL; 816 break; 817 } 818 if (tp->t_template == 0) 819 panic("tcp_output"); 820 if (tp->t_template->m_len < iphdrlen) 821 panic("tcp_output"); 822 bcopy(mtod(tp->t_template, caddr_t), mtod(m, caddr_t), iphdrlen); 823 824 /* 825 * If we are doing retransmissions, then snd_nxt will 826 * not reflect the first unsent octet. For ACK only 827 * packets, we do not want the sequence number of the 828 * retransmitted packet, we want the sequence number 829 * of the next unsent octet. So, if there is no data 830 * (and no SYN or FIN), use snd_max instead of snd_nxt 831 * when filling in ti_seq. But if we are in persist 832 * state, snd_max might reflect one byte beyond the 833 * right edge of the window, so use snd_nxt in that 834 * case, since we know we aren't doing a retransmission. 835 * (retransmit and persist are mutually exclusive...) 836 */ 837 if (len || (flags & (TH_SYN|TH_FIN)) || 838 TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) 839 th->th_seq = htonl(tp->snd_nxt); 840 else 841 th->th_seq = htonl(tp->snd_max); 842 th->th_ack = htonl(tp->rcv_nxt); 843 if (optlen) { 844 bcopy((caddr_t)opt, (caddr_t)(th + 1), optlen); 845 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 846 } 847 th->th_flags = flags; 848 /* 849 * Calculate receive window. Don't shrink window, 850 * but avoid silly window syndrome. 851 */ 852 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)rxsegsize) 853 win = 0; 854 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 855 win = (long)TCP_MAXWIN << tp->rcv_scale; 856 if (win < (long)(tp->rcv_adv - tp->rcv_nxt)) 857 win = (long)(tp->rcv_adv - tp->rcv_nxt); 858 th->th_win = htons((u_int16_t) (win>>tp->rcv_scale)); 859 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 860 u_int32_t urp = tp->snd_up - tp->snd_nxt; 861 if (urp > IP_MAXPACKET) 862 urp = IP_MAXPACKET; 863 th->th_urp = htons((u_int16_t)urp); 864 th->th_flags |= TH_URG; 865 } else 866 /* 867 * If no urgent pointer to send, then we pull 868 * the urgent pointer to the left edge of the send window 869 * so that it doesn't drift into the send window on sequence 870 * number wraparound. 871 */ 872 tp->snd_up = tp->snd_una; /* drag it along */ 873 874 /* 875 * Set ourselves up to be checksummed just before the packet 876 * hits the wire. 877 */ 878 switch (af) { 879 #ifdef INET 880 case AF_INET: 881 m->m_pkthdr.csum_flags = M_CSUM_TCPv4; 882 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 883 if (len + optlen) { 884 /* Fixup the pseudo-header checksum. */ 885 /* XXXJRT Not IP Jumbogram safe. */ 886 th->th_sum = in_cksum_addword(th->th_sum, 887 htons((u_int16_t) (len + optlen))); 888 } 889 break; 890 #endif 891 #ifdef INET6 892 case AF_INET6: 893 /* 894 * XXX Actually delaying the checksum is Hard 895 * XXX (well, maybe not for Itojun, but it is 896 * XXX for me), but we can still take advantage 897 * XXX of the cached pseudo-header checksum. 898 */ 899 /* equals to hdrlen + len */ 900 m->m_pkthdr.len = sizeof(struct ip6_hdr) 901 + sizeof(struct tcphdr) + optlen + len; 902 #ifdef notyet 903 m->m_pkthdr.csum_flags = M_CSUM_TCPv6; 904 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 905 #endif 906 if (len + optlen) { 907 /* Fixup the pseudo-header checksum. */ 908 /* XXXJRT: Not IPv6 Jumbogram safe. */ 909 th->th_sum = in_cksum_addword(th->th_sum, 910 htons((u_int16_t) (len + optlen))); 911 } 912 #ifndef notyet 913 th->th_sum = in6_cksum(m, 0, sizeof(struct ip6_hdr), 914 sizeof(struct tcphdr) + optlen + len); 915 #endif 916 break; 917 #endif 918 } 919 920 /* 921 * In transmit state, time the transmission and arrange for 922 * the retransmit. In persist state, just set snd_max. 923 */ 924 if (tp->t_force == 0 || TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 925 tcp_seq startseq = tp->snd_nxt; 926 927 /* 928 * Advance snd_nxt over sequence space of this segment. 929 * There are no states in which we send both a SYN and a FIN, 930 * so we collapse the tests for these flags. 931 */ 932 if (flags & (TH_SYN|TH_FIN)) 933 tp->snd_nxt++; 934 tp->snd_nxt += len; 935 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 936 tp->snd_max = tp->snd_nxt; 937 /* 938 * Time this transmission if not a retransmission and 939 * not currently timing anything. 940 */ 941 if (tp->t_rtt == 0) { 942 tp->t_rtt = 1; 943 tp->t_rtseq = startseq; 944 tcpstat.tcps_segstimed++; 945 } 946 } 947 948 /* 949 * Set retransmit timer if not currently set, 950 * and not doing an ack or a keep-alive probe. 951 * Initial value for retransmit timer is smoothed 952 * round-trip time + 2 * round-trip time variance. 953 * Initialize shift counter which is used for backoff 954 * of retransmit time. 955 */ 956 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 957 tp->snd_nxt != tp->snd_una) { 958 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 959 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) { 960 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 961 tp->t_rxtshift = 0; 962 } 963 } 964 } else 965 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max)) 966 tp->snd_max = tp->snd_nxt + len; 967 968 #ifdef TCP_DEBUG 969 /* 970 * Trace. 971 */ 972 if (so->so_options & SO_DEBUG) { 973 /* 974 * need to recover version # field, which was overwritten 975 * on ip_cksum computation. 976 */ 977 struct ip *sip; 978 sip = mtod(m, struct ip *); 979 switch (af) { 980 #ifdef INET 981 case AF_INET: 982 sip->ip_v = 4; 983 break; 984 #endif 985 #ifdef INET6 986 case AF_INET6: 987 sip->ip_v = 6; 988 break; 989 #endif 990 } 991 tcp_trace(TA_OUTPUT, tp->t_state, tp, m, 0); 992 } 993 #endif 994 995 /* 996 * Fill in IP length and desired time to live and 997 * send to IP level. There should be a better way 998 * to handle ttl and tos; we could keep them in 999 * the template, but need a way to checksum without them. 1000 */ 1001 m->m_pkthdr.len = hdrlen + len; 1002 1003 switch (af) { 1004 #ifdef INET 1005 case AF_INET: 1006 ip->ip_len = m->m_pkthdr.len; 1007 if (tp->t_inpcb) { 1008 ip->ip_ttl = tp->t_inpcb->inp_ip.ip_ttl; 1009 ip->ip_tos = tp->t_inpcb->inp_ip.ip_tos; 1010 } 1011 #ifdef INET6 1012 else if (tp->t_in6pcb) { 1013 ip->ip_ttl = in6_selecthlim(tp->t_in6pcb, NULL); /*XXX*/ 1014 ip->ip_tos = 0; /*XXX*/ 1015 } 1016 #endif 1017 break; 1018 #endif 1019 #ifdef INET6 1020 case AF_INET6: 1021 ip6->ip6_nxt = IPPROTO_TCP; 1022 if (tp->t_in6pcb) { 1023 /* 1024 * we separately set hoplimit for every segment, since 1025 * the user might want to change the value via 1026 * setsockopt. Also, desired default hop limit might 1027 * be changed via Neighbor Discovery. 1028 */ 1029 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, 1030 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 1031 } 1032 /* ip6->ip6_flow = ??? */ 1033 /* ip6_plen will be filled in ip6_output(). */ 1034 break; 1035 #endif 1036 } 1037 1038 #ifdef IPSEC 1039 if (ipsec_setsocket(m, so) != 0) { 1040 m_freem(m); 1041 error = ENOBUFS; 1042 goto out; 1043 } 1044 #endif /*IPSEC*/ 1045 1046 switch (af) { 1047 #ifdef INET 1048 case AF_INET: 1049 { 1050 struct mbuf *opts; 1051 1052 if (tp->t_inpcb) 1053 opts = tp->t_inpcb->inp_options; 1054 else 1055 opts = NULL; 1056 error = ip_output(m, opts, ro, 1057 (ip_mtudisc ? IP_MTUDISC : 0) | 1058 (so->so_options & SO_DONTROUTE), 1059 0); 1060 break; 1061 } 1062 #endif 1063 #ifdef INET6 1064 case AF_INET6: 1065 { 1066 struct ip6_pktopts *opts; 1067 1068 if (tp->t_in6pcb) 1069 opts = tp->t_in6pcb->in6p_outputopts; 1070 else 1071 opts = NULL; 1072 error = ip6_output(m, opts, (struct route_in6 *)ro, 1073 so->so_options & SO_DONTROUTE, 0, NULL); 1074 break; 1075 } 1076 #endif 1077 default: 1078 error = EAFNOSUPPORT; 1079 break; 1080 } 1081 if (error) { 1082 out: 1083 if (error == ENOBUFS) { 1084 tcpstat.tcps_selfquench++; 1085 #ifdef INET 1086 if (tp->t_inpcb) 1087 tcp_quench(tp->t_inpcb, 0); 1088 #endif 1089 #ifdef INET6 1090 if (tp->t_in6pcb) 1091 tcp6_quench(tp->t_in6pcb, 0); 1092 #endif 1093 return (0); 1094 } 1095 if ((error == EHOSTUNREACH || error == ENETDOWN) 1096 && TCPS_HAVERCVDSYN(tp->t_state)) { 1097 tp->t_softerror = error; 1098 return (0); 1099 } 1100 return (error); 1101 } 1102 tcpstat.tcps_sndtotal++; 1103 if (tp->t_flags & TF_DELACK) 1104 tcpstat.tcps_delack++; 1105 1106 /* 1107 * Data sent (as far as we can tell). 1108 * If this advertises a larger window than any other segment, 1109 * then remember the size of the advertised window. 1110 * Any pending ACK has now been sent. 1111 */ 1112 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 1113 tp->rcv_adv = tp->rcv_nxt + win; 1114 tp->last_ack_sent = tp->rcv_nxt; 1115 tp->t_flags &= ~TF_ACKNOW; 1116 TCP_CLEAR_DELACK(tp); 1117 #ifdef DIAGNOSTIC 1118 if (maxburst < 0) 1119 printf("tcp_output: maxburst exceeded by %d\n", -maxburst); 1120 #endif 1121 if (sendalot && (!tcp_do_newreno || --maxburst)) 1122 goto again; 1123 return (0); 1124 } 1125 1126 void 1127 tcp_setpersist(tp) 1128 struct tcpcb *tp; 1129 { 1130 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2); 1131 int nticks; 1132 1133 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT)) 1134 panic("tcp_output REXMT"); 1135 /* 1136 * Start/restart persistance timer. 1137 */ 1138 if (t < tp->t_rttmin) 1139 t = tp->t_rttmin; 1140 TCPT_RANGESET(nticks, t * tcp_backoff[tp->t_rxtshift], 1141 TCPTV_PERSMIN, TCPTV_PERSMAX); 1142 TCP_TIMER_ARM(tp, TCPT_PERSIST, nticks); 1143 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 1144 tp->t_rxtshift++; 1145 } 1146
Indexes created Tue Sep 23 14:10:03 GMT 2025