tcp_output.c revision 1.79.4.4
1 /* $NetBSD: tcp_output.c,v 1.79.4.4 2003/09/05 13:42:39 tron Exp $ */ 2 3 /* 4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the project nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 /* 33 * @(#)COPYRIGHT 1.1 (NRL) 17 January 1995 34 * 35 * NRL grants permission for redistribution and use in source and binary 36 * forms, with or without modification, of the software and documentation 37 * created at NRL provided that the following conditions are met: 38 * 39 * 1. Redistributions of source code must retain the above copyright 40 * notice, this list of conditions and the following disclaimer. 41 * 2. Redistributions in binary form must reproduce the above copyright 42 * notice, this list of conditions and the following disclaimer in the 43 * documentation and/or other materials provided with the distribution. 44 * 3. All advertising materials mentioning features or use of this software 45 * must display the following acknowledgements: 46 * This product includes software developed by the University of 47 * California, Berkeley and its contributors. 48 * This product includes software developed at the Information 49 * Technology Division, US Naval Research Laboratory. 50 * 4. Neither the name of the NRL nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS 55 * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 56 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A 57 * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NRL OR 58 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 59 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 60 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 61 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 62 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 63 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 64 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 65 * 66 * The views and conclusions contained in the software and documentation 67 * are those of the authors and should not be interpreted as representing 68 * official policies, either expressed or implied, of the US Naval 69 * Research Laboratory (NRL). 70 */ 71 72 /*- 73 * Copyright (c) 1997, 1998, 2001 The NetBSD Foundation, Inc. 74 * All rights reserved. 75 * 76 * This code is derived from software contributed to The NetBSD Foundation 77 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 78 * Facility, NASA Ames Research Center. 79 * 80 * Redistribution and use in source and binary forms, with or without 81 * modification, are permitted provided that the following conditions 82 * are met: 83 * 1. Redistributions of source code must retain the above copyright 84 * notice, this list of conditions and the following disclaimer. 85 * 2. Redistributions in binary form must reproduce the above copyright 86 * notice, this list of conditions and the following disclaimer in the 87 * documentation and/or other materials provided with the distribution. 88 * 3. All advertising materials mentioning features or use of this software 89 * must display the following acknowledgement: 90 * This product includes software developed by the NetBSD 91 * Foundation, Inc. and its contributors. 92 * 4. Neither the name of The NetBSD Foundation nor the names of its 93 * contributors may be used to endorse or promote products derived 94 * from this software without specific prior written permission. 95 * 96 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 97 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 98 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 99 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 100 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 101 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 102 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 103 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 104 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 105 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 106 * POSSIBILITY OF SUCH DAMAGE. 107 */ 108 109 /* 110 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 111 * The Regents of the University of California. All rights reserved. 112 * 113 * Redistribution and use in source and binary forms, with or without 114 * modification, are permitted provided that the following conditions 115 * are met: 116 * 1. Redistributions of source code must retain the above copyright 117 * notice, this list of conditions and the following disclaimer. 118 * 2. Redistributions in binary form must reproduce the above copyright 119 * notice, this list of conditions and the following disclaimer in the 120 * documentation and/or other materials provided with the distribution. 121 * 3. All advertising materials mentioning features or use of this software 122 * must display the following acknowledgement: 123 * This product includes software developed by the University of 124 * California, Berkeley and its contributors. 125 * 4. Neither the name of the University nor the names of its contributors 126 * may be used to endorse or promote products derived from this software 127 * without specific prior written permission. 128 * 129 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 130 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 131 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 132 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 133 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 134 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 135 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 136 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 137 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 138 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 139 * SUCH DAMAGE. 140 * 141 * @(#)tcp_output.c 8.4 (Berkeley) 5/24/95 142 */ 143 144 #include <sys/cdefs.h> 145 __KERNEL_RCSID(0, "$NetBSD: tcp_output.c,v 1.79.4.4 2003/09/05 13:42:39 tron Exp $"); 146 147 #include "opt_inet.h" 148 #include "opt_ipsec.h" 149 #include "opt_tcp_debug.h" 150 151 #include <sys/param.h> 152 #include <sys/systm.h> 153 #include <sys/malloc.h> 154 #include <sys/mbuf.h> 155 #include <sys/protosw.h> 156 #include <sys/socket.h> 157 #include <sys/socketvar.h> 158 #include <sys/errno.h> 159 #include <sys/domain.h> 160 #include <sys/kernel.h> 161 162 #include <net/if.h> 163 #include <net/route.h> 164 165 #include <netinet/in.h> 166 #include <netinet/in_systm.h> 167 #include <netinet/ip.h> 168 #include <netinet/in_pcb.h> 169 #include <netinet/ip_var.h> 170 171 #ifdef INET6 172 #ifndef INET 173 #include <netinet/in.h> 174 #endif 175 #include <netinet/ip6.h> 176 #include <netinet6/in6_pcb.h> 177 #include <netinet6/ip6_var.h> 178 #endif 179 180 #include <netinet/tcp.h> 181 #define TCPOUTFLAGS 182 #include <netinet/tcp_fsm.h> 183 #include <netinet/tcp_seq.h> 184 #include <netinet/tcp_timer.h> 185 #include <netinet/tcp_var.h> 186 #include <netinet/tcpip.h> 187 #include <netinet/tcp_debug.h> 188 189 #ifdef notyet 190 extern struct mbuf *m_copypack(); 191 #endif 192 193 #define MAX_TCPOPTLEN 32 /* max # bytes that go in options */ 194 195 /* 196 * Knob to enable Congestion Window Monitoring, and control the 197 * the burst size it allows. Default burst is 4 packets, per 198 * the Internet draft. 199 */ 200 int tcp_cwm = 0; 201 int tcp_cwm_burstsize = 4; 202 203 #ifdef TCP_OUTPUT_COUNTERS 204 #include <sys/device.h> 205 206 extern struct evcnt tcp_output_bigheader; 207 extern struct evcnt tcp_output_copysmall; 208 extern struct evcnt tcp_output_copybig; 209 extern struct evcnt tcp_output_refbig; 210 211 #define TCP_OUTPUT_COUNTER_INCR(ev) (ev)->ev_count++ 212 #else 213 214 #define TCP_OUTPUT_COUNTER_INCR(ev) /* nothing */ 215 216 #endif /* TCP_OUTPUT_COUNTERS */ 217 218 static 219 #ifndef GPROF 220 __inline 221 #endif 222 void 223 tcp_segsize(struct tcpcb *tp, int *txsegsizep, int *rxsegsizep) 224 { 225 #ifdef INET 226 struct inpcb *inp = tp->t_inpcb; 227 #endif 228 #ifdef INET6 229 struct in6pcb *in6p = tp->t_in6pcb; 230 #endif 231 struct socket *so; 232 struct rtentry *rt; 233 struct ifnet *ifp; 234 int size; 235 int iphlen; 236 int optlen; 237 238 #ifdef DIAGNOSTIC 239 if (tp->t_inpcb && tp->t_in6pcb) 240 panic("tcp_segsize: both t_inpcb and t_in6pcb are set"); 241 #endif 242 switch (tp->t_family) { 243 #ifdef INET 244 case AF_INET: 245 iphlen = sizeof(struct ip); 246 break; 247 #endif 248 #ifdef INET6 249 case AF_INET6: 250 iphlen = sizeof(struct ip6_hdr); 251 break; 252 #endif 253 default: 254 size = tcp_mssdflt; 255 goto out; 256 } 257 258 rt = NULL; 259 #ifdef INET 260 if (inp) { 261 rt = in_pcbrtentry(inp); 262 so = inp->inp_socket; 263 } 264 #endif 265 #ifdef INET6 266 if (in6p) { 267 rt = in6_pcbrtentry(in6p); 268 so = in6p->in6p_socket; 269 } 270 #endif 271 if (rt == NULL) { 272 size = tcp_mssdflt; 273 goto out; 274 } 275 276 ifp = rt->rt_ifp; 277 278 size = tcp_mssdflt; 279 if (tp->t_mtudisc && rt->rt_rmx.rmx_mtu != 0) 280 size = rt->rt_rmx.rmx_mtu - iphlen - sizeof(struct tcphdr); 281 else if (ifp->if_flags & IFF_LOOPBACK) 282 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 283 #ifdef INET 284 else if (inp && tp->t_mtudisc) 285 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 286 else if (inp && in_localaddr(inp->inp_faddr)) 287 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 288 #endif 289 #ifdef INET6 290 else if (in6p) { 291 #ifdef INET 292 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) { 293 /* mapped addr case */ 294 struct in_addr d; 295 bcopy(&in6p->in6p_faddr.s6_addr32[3], &d, sizeof(d)); 296 if (tp->t_mtudisc || in_localaddr(d)) 297 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 298 } else 299 #endif 300 { 301 /* 302 * for IPv6, path MTU discovery is always turned on, 303 * or the node must use packet size <= 1280. 304 */ 305 size = tp->t_mtudisc ? ifp->if_mtu : IPV6_MMTU; 306 size -= (iphlen + sizeof(struct tcphdr)); 307 } 308 } 309 #endif 310 out: 311 /* 312 * Now we must make room for whatever extra TCP/IP options are in 313 * the packet. 314 */ 315 optlen = tcp_optlen(tp); 316 317 /* 318 * XXX tp->t_ourmss should have the right size, but without this code 319 * fragmentation will occur... need more investigation 320 */ 321 #ifdef INET 322 if (inp) { 323 #ifdef IPSEC 324 optlen += ipsec4_hdrsiz_tcp(tp); 325 #endif 326 optlen += ip_optlen(inp); 327 } 328 #endif 329 #ifdef INET6 330 #ifdef INET 331 if (in6p && tp->t_family == AF_INET) { 332 #ifdef IPSEC 333 optlen += ipsec4_hdrsiz_tcp(tp); 334 #endif 335 /* XXX size -= ip_optlen(in6p); */ 336 } else 337 #endif 338 if (in6p && tp->t_family == AF_INET6) { 339 #ifdef IPSEC 340 optlen += ipsec6_hdrsiz_tcp(tp); 341 #endif 342 optlen += ip6_optlen(in6p); 343 } 344 #endif 345 size -= optlen; 346 347 /* 348 * *rxsegsizep holds *estimated* inbound segment size (estimation 349 * assumes that path MTU is the same for both ways). this is only 350 * for silly window avoidance, do not use the value for other purposes. 351 * 352 * ipseclen is subtracted from both sides, this may not be right. 353 * I'm not quite sure about this (could someone comment). 354 */ 355 *txsegsizep = min(tp->t_peermss - optlen, size); 356 /* 357 * Never send more than half a buffer full. This insures that we can 358 * always keep 2 packets on the wire, no matter what SO_SNDBUF is, and 359 * therefore acks will never be delayed unless we run out of data to 360 * transmit. 361 */ 362 if (so) 363 *txsegsizep = min(so->so_snd.sb_hiwat >> 1, *txsegsizep); 364 *rxsegsizep = min(tp->t_ourmss - optlen, size); 365 366 if (*txsegsizep != tp->t_segsz) { 367 /* 368 * If the new segment size is larger, we don't want to 369 * mess up the congestion window, but if it is smaller 370 * we'll have to reduce the congestion window to ensure 371 * that we don't get into trouble with initial windows 372 * and the rest. In any case, if the segment size 373 * has changed, chances are the path has, too, and 374 * our congestion window will be different. 375 */ 376 if (*txsegsizep < tp->t_segsz) { 377 tp->snd_cwnd = max((tp->snd_cwnd / tp->t_segsz) 378 * *txsegsizep, *txsegsizep); 379 tp->snd_ssthresh = max((tp->snd_ssthresh / tp->t_segsz) 380 * *txsegsizep, *txsegsizep); 381 } 382 tp->t_segsz = *txsegsizep; 383 } 384 } 385 386 static 387 #ifndef GPROF 388 __inline 389 #endif 390 int 391 tcp_build_datapkt(struct tcpcb *tp, struct socket *so, int off, 392 long len, int hdrlen, struct mbuf **mp) 393 { 394 struct mbuf *m; 395 396 if (tp->t_force && len == 1) 397 tcpstat.tcps_sndprobe++; 398 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { 399 tcpstat.tcps_sndrexmitpack++; 400 tcpstat.tcps_sndrexmitbyte += len; 401 } else { 402 tcpstat.tcps_sndpack++; 403 tcpstat.tcps_sndbyte += len; 404 } 405 #ifdef notyet 406 if ((m = m_copypack(so->so_snd.sb_mb, off, 407 (int)len, max_linkhdr + hdrlen)) == 0) 408 return (ENOBUFS); 409 /* 410 * m_copypack left space for our hdr; use it. 411 */ 412 m->m_len += hdrlen; 413 m->m_data -= hdrlen; 414 #else 415 MGETHDR(m, M_DONTWAIT, MT_HEADER); 416 if (__predict_false(m == NULL)) 417 return (ENOBUFS); 418 419 /* 420 * XXX Because other code assumes headers will fit in 421 * XXX one header mbuf. 422 * 423 * (This code should almost *never* be run.) 424 */ 425 if (__predict_false((max_linkhdr + hdrlen) > MHLEN)) { 426 TCP_OUTPUT_COUNTER_INCR(&tcp_output_bigheader); 427 MCLGET(m, M_DONTWAIT); 428 if ((m->m_flags & M_EXT) == 0) { 429 m_freem(m); 430 return (ENOBUFS); 431 } 432 } 433 434 m->m_data += max_linkhdr; 435 m->m_len = hdrlen; 436 if (len <= M_TRAILINGSPACE(m)) { 437 m_copydata(so->so_snd.sb_mb, off, (int) len, 438 mtod(m, caddr_t) + hdrlen); 439 m->m_len += len; 440 TCP_OUTPUT_COUNTER_INCR(&tcp_output_copysmall); 441 } else { 442 m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len); 443 if (m->m_next == NULL) { 444 m_freem(m); 445 return (ENOBUFS); 446 } 447 #ifdef TCP_OUTPUT_COUNTERS 448 if (m->m_next->m_flags & M_EXT) 449 TCP_OUTPUT_COUNTER_INCR(&tcp_output_refbig); 450 else 451 TCP_OUTPUT_COUNTER_INCR(&tcp_output_copybig); 452 #endif /* TCP_OUTPUT_COUNTERS */ 453 } 454 #endif 455 456 *mp = m; 457 return (0); 458 } 459 460 /* 461 * Tcp output routine: figure out what should be sent and send it. 462 */ 463 int 464 tcp_output(tp) 465 struct tcpcb *tp; 466 { 467 struct socket *so; 468 struct route *ro; 469 long len, win; 470 int off, flags, error; 471 struct mbuf *m; 472 struct ip *ip; 473 #ifdef INET6 474 struct ip6_hdr *ip6; 475 #endif 476 struct tcphdr *th; 477 u_char opt[MAX_TCPOPTLEN]; 478 unsigned optlen, hdrlen; 479 int idle, sendalot, txsegsize, rxsegsize; 480 int maxburst = TCP_MAXBURST; 481 int af; /* address family on the wire */ 482 int iphdrlen; 483 484 #ifdef DIAGNOSTIC 485 if (tp->t_inpcb && tp->t_in6pcb) 486 panic("tcp_output: both t_inpcb and t_in6pcb are set"); 487 #endif 488 so = NULL; 489 ro = NULL; 490 if (tp->t_inpcb) { 491 so = tp->t_inpcb->inp_socket; 492 ro = &tp->t_inpcb->inp_route; 493 } 494 #ifdef INET6 495 else if (tp->t_in6pcb) { 496 so = tp->t_in6pcb->in6p_socket; 497 ro = (struct route *)&tp->t_in6pcb->in6p_route; 498 } 499 #endif 500 501 switch (af = tp->t_family) { 502 #ifdef INET 503 case AF_INET: 504 if (tp->t_inpcb) 505 break; 506 #ifdef INET6 507 /* mapped addr case */ 508 if (tp->t_in6pcb) 509 break; 510 #endif 511 return EINVAL; 512 #endif 513 #ifdef INET6 514 case AF_INET6: 515 if (tp->t_in6pcb) 516 break; 517 return EINVAL; 518 #endif 519 default: 520 return EAFNOSUPPORT; 521 } 522 523 tcp_segsize(tp, &txsegsize, &rxsegsize); 524 525 idle = (tp->snd_max == tp->snd_una); 526 527 /* 528 * Restart Window computation. From draft-floyd-incr-init-win-03: 529 * 530 * Optionally, a TCP MAY set the restart window to the 531 * minimum of the value used for the initial window and 532 * the current value of cwnd (in other words, using a 533 * larger value for the restart window should never increase 534 * the size of cwnd). 535 */ 536 if (tcp_cwm) { 537 /* 538 * Hughes/Touch/Heidemann Congestion Window Monitoring. 539 * Count the number of packets currently pending 540 * acknowledgement, and limit our congestion window 541 * to a pre-determined allowed burst size plus that count. 542 * This prevents bursting once all pending packets have 543 * been acknowledged (i.e. transmission is idle). 544 * 545 * XXX Link this to Initial Window? 546 */ 547 tp->snd_cwnd = min(tp->snd_cwnd, 548 (tcp_cwm_burstsize * txsegsize) + 549 (tp->snd_nxt - tp->snd_una)); 550 } else { 551 if (idle && (tcp_now - tp->t_rcvtime) >= tp->t_rxtcur) { 552 /* 553 * We have been idle for "a while" and no acks are 554 * expected to clock out any data we send -- 555 * slow start to get ack "clock" running again. 556 */ 557 tp->snd_cwnd = min(tp->snd_cwnd, 558 TCP_INITIAL_WINDOW(tcp_init_win, txsegsize)); 559 } 560 } 561 562 again: 563 /* 564 * Determine length of data that should be transmitted, and 565 * flags that should be used. If there is some data or critical 566 * controls (SYN, RST) to send, then transmit; otherwise, 567 * investigate further. 568 */ 569 sendalot = 0; 570 off = tp->snd_nxt - tp->snd_una; 571 win = min(tp->snd_wnd, tp->snd_cwnd); 572 573 flags = tcp_outflags[tp->t_state]; 574 /* 575 * If in persist timeout with window of 0, send 1 byte. 576 * Otherwise, if window is small but nonzero 577 * and timer expired, we will send what we can 578 * and go to transmit state. 579 */ 580 if (tp->t_force) { 581 if (win == 0) { 582 /* 583 * If we still have some data to send, then 584 * clear the FIN bit. Usually this would 585 * happen below when it realizes that we 586 * aren't sending all the data. However, 587 * if we have exactly 1 byte of unset data, 588 * then it won't clear the FIN bit below, 589 * and if we are in persist state, we wind 590 * up sending the packet without recording 591 * that we sent the FIN bit. 592 * 593 * We can't just blindly clear the FIN bit, 594 * because if we don't have any more data 595 * to send then the probe will be the FIN 596 * itself. 597 */ 598 if (off < so->so_snd.sb_cc) 599 flags &= ~TH_FIN; 600 win = 1; 601 } else { 602 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 603 tp->t_rxtshift = 0; 604 } 605 } 606 607 if (win < so->so_snd.sb_cc) { 608 len = win - off; 609 flags &= ~TH_FIN; 610 } else 611 len = so->so_snd.sb_cc - off; 612 613 if (len < 0) { 614 /* 615 * If FIN has been sent but not acked, 616 * but we haven't been called to retransmit, 617 * len will be -1. Otherwise, window shrank 618 * after we sent into it. If window shrank to 0, 619 * cancel pending retransmit, pull snd_nxt back 620 * to (closed) window, and set the persist timer 621 * if it isn't already going. If the window didn't 622 * close completely, just wait for an ACK. 623 * 624 * If we have a pending FIN, either it has already been 625 * transmitted or it is outside the window, so drop it. 626 * If the FIN has been transmitted, but this is not a 627 * retransmission, then len must be -1. Therefore we also 628 * prevent here the sending of `gratuitous FINs'. This 629 * eliminates the need to check for that case below (e.g. 630 * to back up snd_nxt before the FIN so that the sequence 631 * number is correct). 632 */ 633 len = 0; 634 flags &= ~TH_FIN; 635 if (win == 0) { 636 TCP_TIMER_DISARM(tp, TCPT_REXMT); 637 tp->t_rxtshift = 0; 638 tp->snd_nxt = tp->snd_una; 639 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 640 tcp_setpersist(tp); 641 } 642 } 643 if (len > txsegsize) { 644 len = txsegsize; 645 flags &= ~TH_FIN; 646 sendalot = 1; 647 } 648 649 win = sbspace(&so->so_rcv); 650 651 /* 652 * Sender silly window avoidance. If connection is idle 653 * and can send all data, a maximum segment, 654 * at least a maximum default-size segment do it, 655 * or are forced, do it; otherwise don't bother. 656 * If peer's buffer is tiny, then send 657 * when window is at least half open. 658 * If retransmitting (possibly after persist timer forced us 659 * to send into a small window), then must resend. 660 */ 661 if (len) { 662 if (len == txsegsize) 663 goto send; 664 if ((so->so_state & SS_MORETOCOME) == 0 && 665 ((idle || tp->t_flags & TF_NODELAY) && 666 len + off >= so->so_snd.sb_cc)) 667 goto send; 668 if (tp->t_force) 669 goto send; 670 if (len >= tp->max_sndwnd / 2) 671 goto send; 672 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) 673 goto send; 674 } 675 676 /* 677 * Compare available window to amount of window known to peer 678 * (as advertised window less next expected input). If the 679 * difference is at least twice the size of the largest segment 680 * we expect to receive (i.e. two segments) or at least 50% of 681 * the maximum possible window, then want to send a window update 682 * to peer. 683 */ 684 if (win > 0) { 685 /* 686 * "adv" is the amount we can increase the window, 687 * taking into account that we are limited by 688 * TCP_MAXWIN << tp->rcv_scale. 689 */ 690 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) - 691 (tp->rcv_adv - tp->rcv_nxt); 692 693 if (adv >= (long) (2 * rxsegsize)) 694 goto send; 695 if (2 * adv >= (long) so->so_rcv.sb_hiwat) 696 goto send; 697 } 698 699 /* 700 * Send if we owe peer an ACK. 701 */ 702 if (tp->t_flags & TF_ACKNOW) 703 goto send; 704 if (flags & (TH_SYN|TH_FIN|TH_RST)) 705 goto send; 706 if (SEQ_GT(tp->snd_up, tp->snd_una)) 707 goto send; 708 709 /* 710 * TCP window updates are not reliable, rather a polling protocol 711 * using ``persist'' packets is used to insure receipt of window 712 * updates. The three ``states'' for the output side are: 713 * idle not doing retransmits or persists 714 * persisting to move a small or zero window 715 * (re)transmitting and thereby not persisting 716 * 717 * tp->t_timer[TCPT_PERSIST] 718 * is set when we are in persist state. 719 * tp->t_force 720 * is set when we are called to send a persist packet. 721 * tp->t_timer[TCPT_REXMT] 722 * is set when we are retransmitting 723 * The output side is idle when both timers are zero. 724 * 725 * If send window is too small, there is data to transmit, and no 726 * retransmit or persist is pending, then go to persist state. 727 * If nothing happens soon, send when timer expires: 728 * if window is nonzero, transmit what we can, 729 * otherwise force out a byte. 730 */ 731 if (so->so_snd.sb_cc && TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 732 TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 733 tp->t_rxtshift = 0; 734 tcp_setpersist(tp); 735 } 736 737 /* 738 * No reason to send a segment, just return. 739 */ 740 return (0); 741 742 send: 743 /* 744 * Before ESTABLISHED, force sending of initial options 745 * unless TCP set not to do any options. 746 * NOTE: we assume that the IP/TCP header plus TCP options 747 * always fit in a single mbuf, leaving room for a maximum 748 * link header, i.e. 749 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES 750 */ 751 optlen = 0; 752 switch (af) { 753 #ifdef INET 754 case AF_INET: 755 iphdrlen = sizeof(struct ip) + sizeof(struct tcphdr); 756 break; 757 #endif 758 #ifdef INET6 759 case AF_INET6: 760 iphdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 761 break; 762 #endif 763 default: /*pacify gcc*/ 764 iphdrlen = 0; 765 break; 766 } 767 hdrlen = iphdrlen; 768 if (flags & TH_SYN) { 769 struct rtentry *rt; 770 771 rt = NULL; 772 #ifdef INET 773 if (tp->t_inpcb) 774 rt = in_pcbrtentry(tp->t_inpcb); 775 #endif 776 #ifdef INET6 777 if (tp->t_in6pcb) 778 rt = in6_pcbrtentry(tp->t_in6pcb); 779 #endif 780 781 tp->snd_nxt = tp->iss; 782 tp->t_ourmss = tcp_mss_to_advertise(rt != NULL ? 783 rt->rt_ifp : NULL, af); 784 if ((tp->t_flags & TF_NOOPT) == 0) { 785 opt[0] = TCPOPT_MAXSEG; 786 opt[1] = 4; 787 opt[2] = (tp->t_ourmss >> 8) & 0xff; 788 opt[3] = tp->t_ourmss & 0xff; 789 optlen = 4; 790 791 if ((tp->t_flags & TF_REQ_SCALE) && 792 ((flags & TH_ACK) == 0 || 793 (tp->t_flags & TF_RCVD_SCALE))) { 794 *((u_int32_t *) (opt + optlen)) = htonl( 795 TCPOPT_NOP << 24 | 796 TCPOPT_WINDOW << 16 | 797 TCPOLEN_WINDOW << 8 | 798 tp->request_r_scale); 799 optlen += 4; 800 } 801 } 802 } 803 804 /* 805 * Send a timestamp and echo-reply if this is a SYN and our side 806 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side 807 * and our peer have sent timestamps in our SYN's. 808 */ 809 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 810 (flags & TH_RST) == 0 && 811 ((flags & (TH_SYN|TH_ACK)) == TH_SYN || 812 (tp->t_flags & TF_RCVD_TSTMP))) { 813 u_int32_t *lp = (u_int32_t *)(opt + optlen); 814 815 /* Form timestamp option as shown in appendix A of RFC 1323. */ 816 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 817 *lp++ = htonl(TCP_TIMESTAMP(tp)); 818 *lp = htonl(tp->ts_recent); 819 optlen += TCPOLEN_TSTAMP_APPA; 820 } 821 822 hdrlen += optlen; 823 824 #ifdef DIAGNOSTIC 825 if (len > txsegsize) 826 panic("tcp data to be sent is larger than segment"); 827 if (max_linkhdr + hdrlen > MCLBYTES) 828 panic("tcphdr too big"); 829 #endif 830 831 /* 832 * Grab a header mbuf, attaching a copy of data to 833 * be transmitted, and initialize the header from 834 * the template for sends on this connection. 835 */ 836 if (len) { 837 error = tcp_build_datapkt(tp, so, off, len, hdrlen, &m); 838 if (error) 839 goto out; 840 /* 841 * If we're sending everything we've got, set PUSH. 842 * (This will keep happy those implementations which only 843 * give data to the user when a buffer fills or 844 * a PUSH comes in.) 845 */ 846 if (off + len == so->so_snd.sb_cc) 847 flags |= TH_PUSH; 848 } else { 849 if (tp->t_flags & TF_ACKNOW) 850 tcpstat.tcps_sndacks++; 851 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 852 tcpstat.tcps_sndctrl++; 853 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 854 tcpstat.tcps_sndurg++; 855 else 856 tcpstat.tcps_sndwinup++; 857 858 MGETHDR(m, M_DONTWAIT, MT_HEADER); 859 if (m != NULL && max_linkhdr + hdrlen > MHLEN) { 860 MCLGET(m, M_DONTWAIT); 861 if ((m->m_flags & M_EXT) == 0) { 862 m_freem(m); 863 m = NULL; 864 } 865 } 866 if (m == NULL) { 867 error = ENOBUFS; 868 goto out; 869 } 870 m->m_data += max_linkhdr; 871 m->m_len = hdrlen; 872 } 873 m->m_pkthdr.rcvif = (struct ifnet *)0; 874 switch (af) { 875 #ifdef INET 876 case AF_INET: 877 ip = mtod(m, struct ip *); 878 #ifdef INET6 879 ip6 = NULL; 880 #endif 881 th = (struct tcphdr *)(ip + 1); 882 break; 883 #endif 884 #ifdef INET6 885 case AF_INET6: 886 ip = NULL; 887 ip6 = mtod(m, struct ip6_hdr *); 888 th = (struct tcphdr *)(ip6 + 1); 889 break; 890 #endif 891 default: /*pacify gcc*/ 892 ip = NULL; 893 #ifdef INET6 894 ip6 = NULL; 895 #endif 896 th = NULL; 897 break; 898 } 899 if (tp->t_template == 0) 900 panic("tcp_output"); 901 if (tp->t_template->m_len < iphdrlen) 902 panic("tcp_output"); 903 bcopy(mtod(tp->t_template, caddr_t), mtod(m, caddr_t), iphdrlen); 904 905 /* 906 * If we are doing retransmissions, then snd_nxt will 907 * not reflect the first unsent octet. For ACK only 908 * packets, we do not want the sequence number of the 909 * retransmitted packet, we want the sequence number 910 * of the next unsent octet. So, if there is no data 911 * (and no SYN or FIN), use snd_max instead of snd_nxt 912 * when filling in ti_seq. But if we are in persist 913 * state, snd_max might reflect one byte beyond the 914 * right edge of the window, so use snd_nxt in that 915 * case, since we know we aren't doing a retransmission. 916 * (retransmit and persist are mutually exclusive...) 917 */ 918 if (len || (flags & (TH_SYN|TH_FIN)) || 919 TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) 920 th->th_seq = htonl(tp->snd_nxt); 921 else 922 th->th_seq = htonl(tp->snd_max); 923 th->th_ack = htonl(tp->rcv_nxt); 924 if (optlen) { 925 bcopy((caddr_t)opt, (caddr_t)(th + 1), optlen); 926 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 927 } 928 th->th_flags = flags; 929 /* 930 * Calculate receive window. Don't shrink window, 931 * but avoid silly window syndrome. 932 */ 933 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)rxsegsize) 934 win = 0; 935 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 936 win = (long)TCP_MAXWIN << tp->rcv_scale; 937 if (win < (long)(tp->rcv_adv - tp->rcv_nxt)) 938 win = (long)(tp->rcv_adv - tp->rcv_nxt); 939 th->th_win = htons((u_int16_t) (win>>tp->rcv_scale)); 940 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 941 u_int32_t urp = tp->snd_up - tp->snd_nxt; 942 if (urp > IP_MAXPACKET) 943 urp = IP_MAXPACKET; 944 th->th_urp = htons((u_int16_t)urp); 945 th->th_flags |= TH_URG; 946 } else 947 /* 948 * If no urgent pointer to send, then we pull 949 * the urgent pointer to the left edge of the send window 950 * so that it doesn't drift into the send window on sequence 951 * number wraparound. 952 */ 953 tp->snd_up = tp->snd_una; /* drag it along */ 954 955 /* 956 * Set ourselves up to be checksummed just before the packet 957 * hits the wire. 958 */ 959 switch (af) { 960 #ifdef INET 961 case AF_INET: 962 m->m_pkthdr.csum_flags = M_CSUM_TCPv4; 963 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 964 if (len + optlen) { 965 /* Fixup the pseudo-header checksum. */ 966 /* XXXJRT Not IP Jumbogram safe. */ 967 th->th_sum = in_cksum_addword(th->th_sum, 968 htons((u_int16_t) (len + optlen))); 969 } 970 break; 971 #endif 972 #ifdef INET6 973 case AF_INET6: 974 /* 975 * XXX Actually delaying the checksum is Hard 976 * XXX (well, maybe not for Itojun, but it is 977 * XXX for me), but we can still take advantage 978 * XXX of the cached pseudo-header checksum. 979 */ 980 /* equals to hdrlen + len */ 981 m->m_pkthdr.len = sizeof(struct ip6_hdr) 982 + sizeof(struct tcphdr) + optlen + len; 983 #ifdef notyet 984 m->m_pkthdr.csum_flags = M_CSUM_TCPv6; 985 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 986 #endif 987 if (len + optlen) { 988 /* Fixup the pseudo-header checksum. */ 989 /* XXXJRT: Not IPv6 Jumbogram safe. */ 990 th->th_sum = in_cksum_addword(th->th_sum, 991 htons((u_int16_t) (len + optlen))); 992 } 993 #ifndef notyet 994 th->th_sum = in6_cksum(m, 0, sizeof(struct ip6_hdr), 995 sizeof(struct tcphdr) + optlen + len); 996 #endif 997 break; 998 #endif 999 } 1000 1001 /* 1002 * In transmit state, time the transmission and arrange for 1003 * the retransmit. In persist state, just set snd_max. 1004 */ 1005 if (tp->t_force == 0 || TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 1006 tcp_seq startseq = tp->snd_nxt; 1007 1008 /* 1009 * Advance snd_nxt over sequence space of this segment. 1010 * There are no states in which we send both a SYN and a FIN, 1011 * so we collapse the tests for these flags. 1012 */ 1013 if (flags & (TH_SYN|TH_FIN)) 1014 tp->snd_nxt++; 1015 tp->snd_nxt += len; 1016 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 1017 tp->snd_max = tp->snd_nxt; 1018 /* 1019 * Time this transmission if not a retransmission and 1020 * not currently timing anything. 1021 */ 1022 if (tp->t_rtttime == 0) { 1023 tp->t_rtttime = tcp_now; 1024 tp->t_rtseq = startseq; 1025 tcpstat.tcps_segstimed++; 1026 } 1027 } 1028 1029 /* 1030 * Set retransmit timer if not currently set, 1031 * and not doing an ack or a keep-alive probe. 1032 * Initial value for retransmit timer is smoothed 1033 * round-trip time + 2 * round-trip time variance. 1034 * Initialize shift counter which is used for backoff 1035 * of retransmit time. 1036 */ 1037 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 1038 tp->snd_nxt != tp->snd_una) { 1039 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1040 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) { 1041 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 1042 tp->t_rxtshift = 0; 1043 } 1044 } 1045 } else 1046 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max)) 1047 tp->snd_max = tp->snd_nxt + len; 1048 1049 #ifdef TCP_DEBUG 1050 /* 1051 * Trace. 1052 */ 1053 if (so->so_options & SO_DEBUG) { 1054 /* 1055 * need to recover version # field, which was overwritten 1056 * on ip_cksum computation. 1057 */ 1058 struct ip *sip; 1059 sip = mtod(m, struct ip *); 1060 switch (af) { 1061 #ifdef INET 1062 case AF_INET: 1063 sip->ip_v = 4; 1064 break; 1065 #endif 1066 #ifdef INET6 1067 case AF_INET6: 1068 sip->ip_v = 6; 1069 break; 1070 #endif 1071 } 1072 tcp_trace(TA_OUTPUT, tp->t_state, tp, m, 0); 1073 } 1074 #endif 1075 1076 /* 1077 * Fill in IP length and desired time to live and 1078 * send to IP level. There should be a better way 1079 * to handle ttl and tos; we could keep them in 1080 * the template, but need a way to checksum without them. 1081 */ 1082 m->m_pkthdr.len = hdrlen + len; 1083 1084 switch (af) { 1085 #ifdef INET 1086 case AF_INET: 1087 ip->ip_len = m->m_pkthdr.len; 1088 if (tp->t_inpcb) { 1089 ip->ip_ttl = tp->t_inpcb->inp_ip.ip_ttl; 1090 ip->ip_tos = tp->t_inpcb->inp_ip.ip_tos; 1091 } 1092 #ifdef INET6 1093 else if (tp->t_in6pcb) { 1094 ip->ip_ttl = in6_selecthlim(tp->t_in6pcb, NULL); /*XXX*/ 1095 ip->ip_tos = 0; /*XXX*/ 1096 } 1097 #endif 1098 break; 1099 #endif 1100 #ifdef INET6 1101 case AF_INET6: 1102 ip6->ip6_nxt = IPPROTO_TCP; 1103 if (tp->t_in6pcb) { 1104 /* 1105 * we separately set hoplimit for every segment, since 1106 * the user might want to change the value via 1107 * setsockopt. Also, desired default hop limit might 1108 * be changed via Neighbor Discovery. 1109 */ 1110 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, 1111 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 1112 } 1113 /* ip6->ip6_flow = ??? */ 1114 /* ip6_plen will be filled in ip6_output(). */ 1115 break; 1116 #endif 1117 } 1118 1119 #ifdef IPSEC 1120 if (ipsec_setsocket(m, so) != 0) { 1121 m_freem(m); 1122 error = ENOBUFS; 1123 goto out; 1124 } 1125 #endif /*IPSEC*/ 1126 1127 switch (af) { 1128 #ifdef INET 1129 case AF_INET: 1130 { 1131 struct mbuf *opts; 1132 1133 if (tp->t_inpcb) 1134 opts = tp->t_inpcb->inp_options; 1135 else 1136 opts = NULL; 1137 error = ip_output(m, opts, ro, 1138 (tp->t_mtudisc ? IP_MTUDISC : 0) | 1139 (so->so_options & SO_DONTROUTE), 1140 0); 1141 break; 1142 } 1143 #endif 1144 #ifdef INET6 1145 case AF_INET6: 1146 { 1147 struct ip6_pktopts *opts; 1148 1149 if (tp->t_in6pcb) 1150 opts = tp->t_in6pcb->in6p_outputopts; 1151 else 1152 opts = NULL; 1153 error = ip6_output(m, opts, (struct route_in6 *)ro, 1154 so->so_options & SO_DONTROUTE, 0, NULL); 1155 break; 1156 } 1157 #endif 1158 default: 1159 error = EAFNOSUPPORT; 1160 break; 1161 } 1162 if (error) { 1163 out: 1164 if (error == ENOBUFS) { 1165 tcpstat.tcps_selfquench++; 1166 #ifdef INET 1167 if (tp->t_inpcb) 1168 tcp_quench(tp->t_inpcb, 0); 1169 #endif 1170 #ifdef INET6 1171 if (tp->t_in6pcb) 1172 tcp6_quench(tp->t_in6pcb, 0); 1173 #endif 1174 error = 0; 1175 } else if ((error == EHOSTUNREACH || error == ENETDOWN) && 1176 TCPS_HAVERCVDSYN(tp->t_state)) { 1177 tp->t_softerror = error; 1178 error = 0; 1179 } 1180 1181 /* Restart the delayed ACK timer, if necessary. */ 1182 if (tp->t_flags & TF_DELACK) 1183 TCP_RESTART_DELACK(tp); 1184 1185 return (error); 1186 } 1187 tcpstat.tcps_sndtotal++; 1188 if (tp->t_flags & TF_DELACK) 1189 tcpstat.tcps_delack++; 1190 1191 /* 1192 * Data sent (as far as we can tell). 1193 * If this advertises a larger window than any other segment, 1194 * then remember the size of the advertised window. 1195 * Any pending ACK has now been sent. 1196 */ 1197 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 1198 tp->rcv_adv = tp->rcv_nxt + win; 1199 tp->last_ack_sent = tp->rcv_nxt; 1200 tp->t_flags &= ~TF_ACKNOW; 1201 TCP_CLEAR_DELACK(tp); 1202 #ifdef DIAGNOSTIC 1203 if (maxburst < 0) 1204 printf("tcp_output: maxburst exceeded by %d\n", -maxburst); 1205 #endif 1206 if (sendalot && (!tcp_do_newreno || --maxburst)) 1207 goto again; 1208 return (0); 1209 } 1210 1211 void 1212 tcp_setpersist(tp) 1213 struct tcpcb *tp; 1214 { 1215 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2); 1216 int nticks; 1217 1218 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT)) 1219 panic("tcp_output REXMT"); 1220 /* 1221 * Start/restart persistance timer. 1222 */ 1223 if (t < tp->t_rttmin) 1224 t = tp->t_rttmin; 1225 TCPT_RANGESET(nticks, t * tcp_backoff[tp->t_rxtshift], 1226 TCPTV_PERSMIN, TCPTV_PERSMAX); 1227 TCP_TIMER_ARM(tp, TCPT_PERSIST, nticks); 1228 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 1229 tp->t_rxtshift++; 1230 } 1231
Indexes created Tue Sep 23 07:09:52 GMT 2025