tcp_output.c revision 1.79.4.3
1 /* $NetBSD: tcp_output.c,v 1.79.4.3 2002/11/30 14:31:59 he 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 1.79.4.3 2002/11/30 14:31:59 he 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 (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 && ip_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 (ip_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 = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 306 } 307 } 308 #endif 309 out: 310 /* 311 * Now we must make room for whatever extra TCP/IP options are in 312 * the packet. 313 */ 314 optlen = tcp_optlen(tp); 315 316 /* 317 * XXX tp->t_ourmss should have the right size, but without this code 318 * fragmentation will occur... need more investigation 319 */ 320 #ifdef INET 321 if (inp) { 322 #ifdef IPSEC 323 optlen += ipsec4_hdrsiz_tcp(tp); 324 #endif 325 optlen += ip_optlen(inp); 326 } 327 #endif 328 #ifdef INET6 329 #ifdef INET 330 if (in6p && tp->t_family == AF_INET) { 331 #ifdef IPSEC 332 optlen += ipsec4_hdrsiz_tcp(tp); 333 #endif 334 /* XXX size -= ip_optlen(in6p); */ 335 } else 336 #endif 337 if (in6p && tp->t_family == AF_INET6) { 338 #ifdef IPSEC 339 optlen += ipsec6_hdrsiz_tcp(tp); 340 #endif 341 optlen += ip6_optlen(in6p); 342 } 343 #endif 344 size -= optlen; 345 346 /* 347 * *rxsegsizep holds *estimated* inbound segment size (estimation 348 * assumes that path MTU is the same for both ways). this is only 349 * for silly window avoidance, do not use the value for other purposes. 350 * 351 * ipseclen is subtracted from both sides, this may not be right. 352 * I'm not quite sure about this (could someone comment). 353 */ 354 *txsegsizep = min(tp->t_peermss - optlen, size); 355 /* 356 * Never send more than half a buffer full. This insures that we can 357 * always keep 2 packets on the wire, no matter what SO_SNDBUF is, and 358 * therefore acks will never be delayed unless we run out of data to 359 * transmit. 360 */ 361 if (so) 362 *txsegsizep = min(so->so_snd.sb_hiwat >> 1, *txsegsizep); 363 *rxsegsizep = min(tp->t_ourmss - optlen, size); 364 365 if (*txsegsizep != tp->t_segsz) { 366 /* 367 * If the new segment size is larger, we don't want to 368 * mess up the congestion window, but if it is smaller 369 * we'll have to reduce the congestion window to ensure 370 * that we don't get into trouble with initial windows 371 * and the rest. In any case, if the segment size 372 * has changed, chances are the path has, too, and 373 * our congestion window will be different. 374 */ 375 if (*txsegsizep < tp->t_segsz) { 376 tp->snd_cwnd = max((tp->snd_cwnd / tp->t_segsz) 377 * *txsegsizep, *txsegsizep); 378 tp->snd_ssthresh = max((tp->snd_ssthresh / tp->t_segsz) 379 * *txsegsizep, *txsegsizep); 380 } 381 tp->t_segsz = *txsegsizep; 382 } 383 } 384 385 static 386 #ifndef GPROF 387 __inline 388 #endif 389 int 390 tcp_build_datapkt(struct tcpcb *tp, struct socket *so, int off, 391 long len, int hdrlen, struct mbuf **mp) 392 { 393 struct mbuf *m; 394 395 if (tp->t_force && len == 1) 396 tcpstat.tcps_sndprobe++; 397 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { 398 tcpstat.tcps_sndrexmitpack++; 399 tcpstat.tcps_sndrexmitbyte += len; 400 } else { 401 tcpstat.tcps_sndpack++; 402 tcpstat.tcps_sndbyte += len; 403 } 404 #ifdef notyet 405 if ((m = m_copypack(so->so_snd.sb_mb, off, 406 (int)len, max_linkhdr + hdrlen)) == 0) 407 return (ENOBUFS); 408 /* 409 * m_copypack left space for our hdr; use it. 410 */ 411 m->m_len += hdrlen; 412 m->m_data -= hdrlen; 413 #else 414 MGETHDR(m, M_DONTWAIT, MT_HEADER); 415 if (__predict_false(m == NULL)) 416 return (ENOBUFS); 417 418 /* 419 * XXX Because other code assumes headers will fit in 420 * XXX one header mbuf. 421 * 422 * (This code should almost *never* be run.) 423 */ 424 if (__predict_false((max_linkhdr + hdrlen) > MHLEN)) { 425 TCP_OUTPUT_COUNTER_INCR(&tcp_output_bigheader); 426 MCLGET(m, M_DONTWAIT); 427 if ((m->m_flags & M_EXT) == 0) { 428 m_freem(m); 429 return (ENOBUFS); 430 } 431 } 432 433 m->m_data += max_linkhdr; 434 m->m_len = hdrlen; 435 if (len <= M_TRAILINGSPACE(m)) { 436 m_copydata(so->so_snd.sb_mb, off, (int) len, 437 mtod(m, caddr_t) + hdrlen); 438 m->m_len += len; 439 TCP_OUTPUT_COUNTER_INCR(&tcp_output_copysmall); 440 } else { 441 m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len); 442 if (m->m_next == NULL) { 443 m_freem(m); 444 return (ENOBUFS); 445 } 446 #ifdef TCP_OUTPUT_COUNTERS 447 if (m->m_next->m_flags & M_EXT) 448 TCP_OUTPUT_COUNTER_INCR(&tcp_output_refbig); 449 else 450 TCP_OUTPUT_COUNTER_INCR(&tcp_output_copybig); 451 #endif /* TCP_OUTPUT_COUNTERS */ 452 } 453 #endif 454 455 *mp = m; 456 return (0); 457 } 458 459 /* 460 * Tcp output routine: figure out what should be sent and send it. 461 */ 462 int 463 tcp_output(tp) 464 struct tcpcb *tp; 465 { 466 struct socket *so; 467 struct route *ro; 468 long len, win; 469 int off, flags, error; 470 struct mbuf *m; 471 struct ip *ip; 472 #ifdef INET6 473 struct ip6_hdr *ip6; 474 #endif 475 struct tcphdr *th; 476 u_char opt[MAX_TCPOPTLEN]; 477 unsigned optlen, hdrlen; 478 int idle, sendalot, txsegsize, rxsegsize; 479 int maxburst = TCP_MAXBURST; 480 int af; /* address family on the wire */ 481 int iphdrlen; 482 483 #ifdef DIAGNOSTIC 484 if (tp->t_inpcb && tp->t_in6pcb) 485 panic("tcp_output: both t_inpcb and t_in6pcb are set"); 486 #endif 487 so = NULL; 488 ro = NULL; 489 if (tp->t_inpcb) { 490 so = tp->t_inpcb->inp_socket; 491 ro = &tp->t_inpcb->inp_route; 492 } 493 #ifdef INET6 494 else if (tp->t_in6pcb) { 495 so = tp->t_in6pcb->in6p_socket; 496 ro = (struct route *)&tp->t_in6pcb->in6p_route; 497 } 498 #endif 499 500 switch (af = tp->t_family) { 501 #ifdef INET 502 case AF_INET: 503 if (tp->t_inpcb) 504 break; 505 #ifdef INET6 506 /* mapped addr case */ 507 if (tp->t_in6pcb) 508 break; 509 #endif 510 return EINVAL; 511 #endif 512 #ifdef INET6 513 case AF_INET6: 514 if (tp->t_in6pcb) 515 break; 516 return EINVAL; 517 #endif 518 default: 519 return EAFNOSUPPORT; 520 } 521 522 tcp_segsize(tp, &txsegsize, &rxsegsize); 523 524 idle = (tp->snd_max == tp->snd_una); 525 526 /* 527 * Restart Window computation. From draft-floyd-incr-init-win-03: 528 * 529 * Optionally, a TCP MAY set the restart window to the 530 * minimum of the value used for the initial window and 531 * the current value of cwnd (in other words, using a 532 * larger value for the restart window should never increase 533 * the size of cwnd). 534 */ 535 if (tcp_cwm) { 536 /* 537 * Hughes/Touch/Heidemann Congestion Window Monitoring. 538 * Count the number of packets currently pending 539 * acknowledgement, and limit our congestion window 540 * to a pre-determined allowed burst size plus that count. 541 * This prevents bursting once all pending packets have 542 * been acknowledged (i.e. transmission is idle). 543 * 544 * XXX Link this to Initial Window? 545 */ 546 tp->snd_cwnd = min(tp->snd_cwnd, 547 (tcp_cwm_burstsize * txsegsize) + 548 (tp->snd_nxt - tp->snd_una)); 549 } else { 550 if (idle && (tcp_now - tp->t_rcvtime) >= tp->t_rxtcur) { 551 /* 552 * We have been idle for "a while" and no acks are 553 * expected to clock out any data we send -- 554 * slow start to get ack "clock" running again. 555 */ 556 tp->snd_cwnd = min(tp->snd_cwnd, 557 TCP_INITIAL_WINDOW(tcp_init_win, txsegsize)); 558 } 559 } 560 561 again: 562 /* 563 * Determine length of data that should be transmitted, and 564 * flags that should be used. If there is some data or critical 565 * controls (SYN, RST) to send, then transmit; otherwise, 566 * investigate further. 567 */ 568 sendalot = 0; 569 off = tp->snd_nxt - tp->snd_una; 570 win = min(tp->snd_wnd, tp->snd_cwnd); 571 572 flags = tcp_outflags[tp->t_state]; 573 /* 574 * If in persist timeout with window of 0, send 1 byte. 575 * Otherwise, if window is small but nonzero 576 * and timer expired, we will send what we can 577 * and go to transmit state. 578 */ 579 if (tp->t_force) { 580 if (win == 0) { 581 /* 582 * If we still have some data to send, then 583 * clear the FIN bit. Usually this would 584 * happen below when it realizes that we 585 * aren't sending all the data. However, 586 * if we have exactly 1 byte of unset data, 587 * then it won't clear the FIN bit below, 588 * and if we are in persist state, we wind 589 * up sending the packet without recording 590 * that we sent the FIN bit. 591 * 592 * We can't just blindly clear the FIN bit, 593 * because if we don't have any more data 594 * to send then the probe will be the FIN 595 * itself. 596 */ 597 if (off < so->so_snd.sb_cc) 598 flags &= ~TH_FIN; 599 win = 1; 600 } else { 601 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 602 tp->t_rxtshift = 0; 603 } 604 } 605 606 if (win < so->so_snd.sb_cc) { 607 len = win - off; 608 flags &= ~TH_FIN; 609 } else 610 len = so->so_snd.sb_cc - off; 611 612 if (len < 0) { 613 /* 614 * If FIN has been sent but not acked, 615 * but we haven't been called to retransmit, 616 * len will be -1. Otherwise, window shrank 617 * after we sent into it. If window shrank to 0, 618 * cancel pending retransmit, pull snd_nxt back 619 * to (closed) window, and set the persist timer 620 * if it isn't already going. If the window didn't 621 * close completely, just wait for an ACK. 622 * 623 * If we have a pending FIN, either it has already been 624 * transmitted or it is outside the window, so drop it. 625 * If the FIN has been transmitted, but this is not a 626 * retransmission, then len must be -1. Therefore we also 627 * prevent here the sending of `gratuitous FINs'. This 628 * eliminates the need to check for that case below (e.g. 629 * to back up snd_nxt before the FIN so that the sequence 630 * number is correct). 631 */ 632 len = 0; 633 flags &= ~TH_FIN; 634 if (win == 0) { 635 TCP_TIMER_DISARM(tp, TCPT_REXMT); 636 tp->t_rxtshift = 0; 637 tp->snd_nxt = tp->snd_una; 638 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 639 tcp_setpersist(tp); 640 } 641 } 642 if (len > txsegsize) { 643 len = txsegsize; 644 flags &= ~TH_FIN; 645 sendalot = 1; 646 } 647 648 win = sbspace(&so->so_rcv); 649 650 /* 651 * Sender silly window avoidance. If connection is idle 652 * and can send all data, a maximum segment, 653 * at least a maximum default-size segment do it, 654 * or are forced, do it; otherwise don't bother. 655 * If peer's buffer is tiny, then send 656 * when window is at least half open. 657 * If retransmitting (possibly after persist timer forced us 658 * to send into a small window), then must resend. 659 */ 660 if (len) { 661 if (len == txsegsize) 662 goto send; 663 if ((so->so_state & SS_MORETOCOME) == 0 && 664 ((idle || tp->t_flags & TF_NODELAY) && 665 len + off >= so->so_snd.sb_cc)) 666 goto send; 667 if (tp->t_force) 668 goto send; 669 if (len >= tp->max_sndwnd / 2) 670 goto send; 671 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) 672 goto send; 673 } 674 675 /* 676 * Compare available window to amount of window known to peer 677 * (as advertised window less next expected input). If the 678 * difference is at least twice the size of the largest segment 679 * we expect to receive (i.e. two segments) or at least 50% of 680 * the maximum possible window, then want to send a window update 681 * to peer. 682 */ 683 if (win > 0) { 684 /* 685 * "adv" is the amount we can increase the window, 686 * taking into account that we are limited by 687 * TCP_MAXWIN << tp->rcv_scale. 688 */ 689 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) - 690 (tp->rcv_adv - tp->rcv_nxt); 691 692 if (adv >= (long) (2 * rxsegsize)) 693 goto send; 694 if (2 * adv >= (long) so->so_rcv.sb_hiwat) 695 goto send; 696 } 697 698 /* 699 * Send if we owe peer an ACK. 700 */ 701 if (tp->t_flags & TF_ACKNOW) 702 goto send; 703 if (flags & (TH_SYN|TH_FIN|TH_RST)) 704 goto send; 705 if (SEQ_GT(tp->snd_up, tp->snd_una)) 706 goto send; 707 708 /* 709 * TCP window updates are not reliable, rather a polling protocol 710 * using ``persist'' packets is used to insure receipt of window 711 * updates. The three ``states'' for the output side are: 712 * idle not doing retransmits or persists 713 * persisting to move a small or zero window 714 * (re)transmitting and thereby not persisting 715 * 716 * tp->t_timer[TCPT_PERSIST] 717 * is set when we are in persist state. 718 * tp->t_force 719 * is set when we are called to send a persist packet. 720 * tp->t_timer[TCPT_REXMT] 721 * is set when we are retransmitting 722 * The output side is idle when both timers are zero. 723 * 724 * If send window is too small, there is data to transmit, and no 725 * retransmit or persist is pending, then go to persist state. 726 * If nothing happens soon, send when timer expires: 727 * if window is nonzero, transmit what we can, 728 * otherwise force out a byte. 729 */ 730 if (so->so_snd.sb_cc && TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 731 TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 732 tp->t_rxtshift = 0; 733 tcp_setpersist(tp); 734 } 735 736 /* 737 * No reason to send a segment, just return. 738 */ 739 return (0); 740 741 send: 742 /* 743 * Before ESTABLISHED, force sending of initial options 744 * unless TCP set not to do any options. 745 * NOTE: we assume that the IP/TCP header plus TCP options 746 * always fit in a single mbuf, leaving room for a maximum 747 * link header, i.e. 748 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES 749 */ 750 optlen = 0; 751 switch (af) { 752 #ifdef INET 753 case AF_INET: 754 iphdrlen = sizeof(struct ip) + sizeof(struct tcphdr); 755 break; 756 #endif 757 #ifdef INET6 758 case AF_INET6: 759 iphdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 760 break; 761 #endif 762 default: /*pacify gcc*/ 763 iphdrlen = 0; 764 break; 765 } 766 hdrlen = iphdrlen; 767 if (flags & TH_SYN) { 768 struct rtentry *rt; 769 770 rt = NULL; 771 #ifdef INET 772 if (tp->t_inpcb) 773 rt = in_pcbrtentry(tp->t_inpcb); 774 #endif 775 #ifdef INET6 776 if (tp->t_in6pcb) 777 rt = in6_pcbrtentry(tp->t_in6pcb); 778 #endif 779 780 tp->snd_nxt = tp->iss; 781 tp->t_ourmss = tcp_mss_to_advertise(rt != NULL ? 782 rt->rt_ifp : NULL, af); 783 if ((tp->t_flags & TF_NOOPT) == 0) { 784 opt[0] = TCPOPT_MAXSEG; 785 opt[1] = 4; 786 opt[2] = (tp->t_ourmss >> 8) & 0xff; 787 opt[3] = tp->t_ourmss & 0xff; 788 optlen = 4; 789 790 if ((tp->t_flags & TF_REQ_SCALE) && 791 ((flags & TH_ACK) == 0 || 792 (tp->t_flags & TF_RCVD_SCALE))) { 793 *((u_int32_t *) (opt + optlen)) = htonl( 794 TCPOPT_NOP << 24 | 795 TCPOPT_WINDOW << 16 | 796 TCPOLEN_WINDOW << 8 | 797 tp->request_r_scale); 798 optlen += 4; 799 } 800 } 801 } 802 803 /* 804 * Send a timestamp and echo-reply if this is a SYN and our side 805 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side 806 * and our peer have sent timestamps in our SYN's. 807 */ 808 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 809 (flags & TH_RST) == 0 && 810 ((flags & (TH_SYN|TH_ACK)) == TH_SYN || 811 (tp->t_flags & TF_RCVD_TSTMP))) { 812 u_int32_t *lp = (u_int32_t *)(opt + optlen); 813 814 /* Form timestamp option as shown in appendix A of RFC 1323. */ 815 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 816 *lp++ = htonl(TCP_TIMESTAMP(tp)); 817 *lp = htonl(tp->ts_recent); 818 optlen += TCPOLEN_TSTAMP_APPA; 819 } 820 821 hdrlen += optlen; 822 823 #ifdef DIAGNOSTIC 824 if (len > txsegsize) 825 panic("tcp data to be sent is larger than segment"); 826 if (max_linkhdr + hdrlen > MCLBYTES) 827 panic("tcphdr too big"); 828 #endif 829 830 /* 831 * Grab a header mbuf, attaching a copy of data to 832 * be transmitted, and initialize the header from 833 * the template for sends on this connection. 834 */ 835 if (len) { 836 error = tcp_build_datapkt(tp, so, off, len, hdrlen, &m); 837 if (error) 838 goto out; 839 /* 840 * If we're sending everything we've got, set PUSH. 841 * (This will keep happy those implementations which only 842 * give data to the user when a buffer fills or 843 * a PUSH comes in.) 844 */ 845 if (off + len == so->so_snd.sb_cc) 846 flags |= TH_PUSH; 847 } else { 848 if (tp->t_flags & TF_ACKNOW) 849 tcpstat.tcps_sndacks++; 850 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 851 tcpstat.tcps_sndctrl++; 852 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 853 tcpstat.tcps_sndurg++; 854 else 855 tcpstat.tcps_sndwinup++; 856 857 MGETHDR(m, M_DONTWAIT, MT_HEADER); 858 if (m != NULL && max_linkhdr + hdrlen > MHLEN) { 859 MCLGET(m, M_DONTWAIT); 860 if ((m->m_flags & M_EXT) == 0) { 861 m_freem(m); 862 m = NULL; 863 } 864 } 865 if (m == NULL) { 866 error = ENOBUFS; 867 goto out; 868 } 869 m->m_data += max_linkhdr; 870 m->m_len = hdrlen; 871 } 872 m->m_pkthdr.rcvif = (struct ifnet *)0; 873 switch (af) { 874 #ifdef INET 875 case AF_INET: 876 ip = mtod(m, struct ip *); 877 #ifdef INET6 878 ip6 = NULL; 879 #endif 880 th = (struct tcphdr *)(ip + 1); 881 break; 882 #endif 883 #ifdef INET6 884 case AF_INET6: 885 ip = NULL; 886 ip6 = mtod(m, struct ip6_hdr *); 887 th = (struct tcphdr *)(ip6 + 1); 888 break; 889 #endif 890 default: /*pacify gcc*/ 891 ip = NULL; 892 #ifdef INET6 893 ip6 = NULL; 894 #endif 895 th = NULL; 896 break; 897 } 898 if (tp->t_template == 0) 899 panic("tcp_output"); 900 if (tp->t_template->m_len < iphdrlen) 901 panic("tcp_output"); 902 bcopy(mtod(tp->t_template, caddr_t), mtod(m, caddr_t), iphdrlen); 903 904 /* 905 * If we are doing retransmissions, then snd_nxt will 906 * not reflect the first unsent octet. For ACK only 907 * packets, we do not want the sequence number of the 908 * retransmitted packet, we want the sequence number 909 * of the next unsent octet. So, if there is no data 910 * (and no SYN or FIN), use snd_max instead of snd_nxt 911 * when filling in ti_seq. But if we are in persist 912 * state, snd_max might reflect one byte beyond the 913 * right edge of the window, so use snd_nxt in that 914 * case, since we know we aren't doing a retransmission. 915 * (retransmit and persist are mutually exclusive...) 916 */ 917 if (len || (flags & (TH_SYN|TH_FIN)) || 918 TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) 919 th->th_seq = htonl(tp->snd_nxt); 920 else 921 th->th_seq = htonl(tp->snd_max); 922 th->th_ack = htonl(tp->rcv_nxt); 923 if (optlen) { 924 bcopy((caddr_t)opt, (caddr_t)(th + 1), optlen); 925 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 926 } 927 th->th_flags = flags; 928 /* 929 * Calculate receive window. Don't shrink window, 930 * but avoid silly window syndrome. 931 */ 932 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)rxsegsize) 933 win = 0; 934 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 935 win = (long)TCP_MAXWIN << tp->rcv_scale; 936 if (win < (long)(tp->rcv_adv - tp->rcv_nxt)) 937 win = (long)(tp->rcv_adv - tp->rcv_nxt); 938 th->th_win = htons((u_int16_t) (win>>tp->rcv_scale)); 939 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 940 u_int32_t urp = tp->snd_up - tp->snd_nxt; 941 if (urp > IP_MAXPACKET) 942 urp = IP_MAXPACKET; 943 th->th_urp = htons((u_int16_t)urp); 944 th->th_flags |= TH_URG; 945 } else 946 /* 947 * If no urgent pointer to send, then we pull 948 * the urgent pointer to the left edge of the send window 949 * so that it doesn't drift into the send window on sequence 950 * number wraparound. 951 */ 952 tp->snd_up = tp->snd_una; /* drag it along */ 953 954 /* 955 * Set ourselves up to be checksummed just before the packet 956 * hits the wire. 957 */ 958 switch (af) { 959 #ifdef INET 960 case AF_INET: 961 m->m_pkthdr.csum_flags = M_CSUM_TCPv4; 962 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 963 if (len + optlen) { 964 /* Fixup the pseudo-header checksum. */ 965 /* XXXJRT Not IP Jumbogram safe. */ 966 th->th_sum = in_cksum_addword(th->th_sum, 967 htons((u_int16_t) (len + optlen))); 968 } 969 break; 970 #endif 971 #ifdef INET6 972 case AF_INET6: 973 /* 974 * XXX Actually delaying the checksum is Hard 975 * XXX (well, maybe not for Itojun, but it is 976 * XXX for me), but we can still take advantage 977 * XXX of the cached pseudo-header checksum. 978 */ 979 /* equals to hdrlen + len */ 980 m->m_pkthdr.len = sizeof(struct ip6_hdr) 981 + sizeof(struct tcphdr) + optlen + len; 982 #ifdef notyet 983 m->m_pkthdr.csum_flags = M_CSUM_TCPv6; 984 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 985 #endif 986 if (len + optlen) { 987 /* Fixup the pseudo-header checksum. */ 988 /* XXXJRT: Not IPv6 Jumbogram safe. */ 989 th->th_sum = in_cksum_addword(th->th_sum, 990 htons((u_int16_t) (len + optlen))); 991 } 992 #ifndef notyet 993 th->th_sum = in6_cksum(m, 0, sizeof(struct ip6_hdr), 994 sizeof(struct tcphdr) + optlen + len); 995 #endif 996 break; 997 #endif 998 } 999 1000 /* 1001 * In transmit state, time the transmission and arrange for 1002 * the retransmit. In persist state, just set snd_max. 1003 */ 1004 if (tp->t_force == 0 || TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 1005 tcp_seq startseq = tp->snd_nxt; 1006 1007 /* 1008 * Advance snd_nxt over sequence space of this segment. 1009 * There are no states in which we send both a SYN and a FIN, 1010 * so we collapse the tests for these flags. 1011 */ 1012 if (flags & (TH_SYN|TH_FIN)) 1013 tp->snd_nxt++; 1014 tp->snd_nxt += len; 1015 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 1016 tp->snd_max = tp->snd_nxt; 1017 /* 1018 * Time this transmission if not a retransmission and 1019 * not currently timing anything. 1020 */ 1021 if (tp->t_rtttime == 0) { 1022 tp->t_rtttime = tcp_now; 1023 tp->t_rtseq = startseq; 1024 tcpstat.tcps_segstimed++; 1025 } 1026 } 1027 1028 /* 1029 * Set retransmit timer if not currently set, 1030 * and not doing an ack or a keep-alive probe. 1031 * Initial value for retransmit timer is smoothed 1032 * round-trip time + 2 * round-trip time variance. 1033 * Initialize shift counter which is used for backoff 1034 * of retransmit time. 1035 */ 1036 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 1037 tp->snd_nxt != tp->snd_una) { 1038 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1039 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) { 1040 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 1041 tp->t_rxtshift = 0; 1042 } 1043 } 1044 } else 1045 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max)) 1046 tp->snd_max = tp->snd_nxt + len; 1047 1048 #ifdef TCP_DEBUG 1049 /* 1050 * Trace. 1051 */ 1052 if (so->so_options & SO_DEBUG) { 1053 /* 1054 * need to recover version # field, which was overwritten 1055 * on ip_cksum computation. 1056 */ 1057 struct ip *sip; 1058 sip = mtod(m, struct ip *); 1059 switch (af) { 1060 #ifdef INET 1061 case AF_INET: 1062 sip->ip_v = 4; 1063 break; 1064 #endif 1065 #ifdef INET6 1066 case AF_INET6: 1067 sip->ip_v = 6; 1068 break; 1069 #endif 1070 } 1071 tcp_trace(TA_OUTPUT, tp->t_state, tp, m, 0); 1072 } 1073 #endif 1074 1075 /* 1076 * Fill in IP length and desired time to live and 1077 * send to IP level. There should be a better way 1078 * to handle ttl and tos; we could keep them in 1079 * the template, but need a way to checksum without them. 1080 */ 1081 m->m_pkthdr.len = hdrlen + len; 1082 1083 switch (af) { 1084 #ifdef INET 1085 case AF_INET: 1086 ip->ip_len = m->m_pkthdr.len; 1087 if (tp->t_inpcb) { 1088 ip->ip_ttl = tp->t_inpcb->inp_ip.ip_ttl; 1089 ip->ip_tos = tp->t_inpcb->inp_ip.ip_tos; 1090 } 1091 #ifdef INET6 1092 else if (tp->t_in6pcb) { 1093 ip->ip_ttl = in6_selecthlim(tp->t_in6pcb, NULL); /*XXX*/ 1094 ip->ip_tos = 0; /*XXX*/ 1095 } 1096 #endif 1097 break; 1098 #endif 1099 #ifdef INET6 1100 case AF_INET6: 1101 ip6->ip6_nxt = IPPROTO_TCP; 1102 if (tp->t_in6pcb) { 1103 /* 1104 * we separately set hoplimit for every segment, since 1105 * the user might want to change the value via 1106 * setsockopt. Also, desired default hop limit might 1107 * be changed via Neighbor Discovery. 1108 */ 1109 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, 1110 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 1111 } 1112 /* ip6->ip6_flow = ??? */ 1113 /* ip6_plen will be filled in ip6_output(). */ 1114 break; 1115 #endif 1116 } 1117 1118 #ifdef IPSEC 1119 if (ipsec_setsocket(m, so) != 0) { 1120 m_freem(m); 1121 error = ENOBUFS; 1122 goto out; 1123 } 1124 #endif /*IPSEC*/ 1125 1126 switch (af) { 1127 #ifdef INET 1128 case AF_INET: 1129 { 1130 struct mbuf *opts; 1131 1132 if (tp->t_inpcb) 1133 opts = tp->t_inpcb->inp_options; 1134 else 1135 opts = NULL; 1136 error = ip_output(m, opts, ro, 1137 (ip_mtudisc ? IP_MTUDISC : 0) | 1138 (so->so_options & SO_DONTROUTE), 1139 0); 1140 break; 1141 } 1142 #endif 1143 #ifdef INET6 1144 case AF_INET6: 1145 { 1146 struct ip6_pktopts *opts; 1147 1148 if (tp->t_in6pcb) 1149 opts = tp->t_in6pcb->in6p_outputopts; 1150 else 1151 opts = NULL; 1152 error = ip6_output(m, opts, (struct route_in6 *)ro, 1153 so->so_options & SO_DONTROUTE, 0, NULL); 1154 break; 1155 } 1156 #endif 1157 default: 1158 error = EAFNOSUPPORT; 1159 break; 1160 } 1161 if (error) { 1162 out: 1163 if (error == ENOBUFS) { 1164 tcpstat.tcps_selfquench++; 1165 #ifdef INET 1166 if (tp->t_inpcb) 1167 tcp_quench(tp->t_inpcb, 0); 1168 #endif 1169 #ifdef INET6 1170 if (tp->t_in6pcb) 1171 tcp6_quench(tp->t_in6pcb, 0); 1172 #endif 1173 error = 0; 1174 } else if ((error == EHOSTUNREACH || error == ENETDOWN) && 1175 TCPS_HAVERCVDSYN(tp->t_state)) { 1176 tp->t_softerror = error; 1177 error = 0; 1178 } 1179 1180 /* Restart the delayed ACK timer, if necessary. */ 1181 if (tp->t_flags & TF_DELACK) 1182 TCP_RESTART_DELACK(tp); 1183 1184 return (error); 1185 } 1186 tcpstat.tcps_sndtotal++; 1187 if (tp->t_flags & TF_DELACK) 1188 tcpstat.tcps_delack++; 1189 1190 /* 1191 * Data sent (as far as we can tell). 1192 * If this advertises a larger window than any other segment, 1193 * then remember the size of the advertised window. 1194 * Any pending ACK has now been sent. 1195 */ 1196 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 1197 tp->rcv_adv = tp->rcv_nxt + win; 1198 tp->last_ack_sent = tp->rcv_nxt; 1199 tp->t_flags &= ~TF_ACKNOW; 1200 TCP_CLEAR_DELACK(tp); 1201 #ifdef DIAGNOSTIC 1202 if (maxburst < 0) 1203 printf("tcp_output: maxburst exceeded by %d\n", -maxburst); 1204 #endif 1205 if (sendalot && (!tcp_do_newreno || --maxburst)) 1206 goto again; 1207 return (0); 1208 } 1209 1210 void 1211 tcp_setpersist(tp) 1212 struct tcpcb *tp; 1213 { 1214 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2); 1215 int nticks; 1216 1217 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT)) 1218 panic("tcp_output REXMT"); 1219 /* 1220 * Start/restart persistance timer. 1221 */ 1222 if (t < tp->t_rttmin) 1223 t = tp->t_rttmin; 1224 TCPT_RANGESET(nticks, t * tcp_backoff[tp->t_rxtshift], 1225 TCPTV_PERSMIN, TCPTV_PERSMAX); 1226 TCP_TIMER_ARM(tp, TCPT_PERSIST, nticks); 1227 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 1228 tp->t_rxtshift++; 1229 } 1230
Indexes created Tue Sep 23 07:09:52 GMT 2025