1 /* $NetBSD: tcp_output.c,v 1.80 2002/05/26 16:05:43 itojun 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 1.80 2002/05/26 16:05:43 itojun 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 = 1; 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 rtentry *rt; 232 struct ifnet *ifp; 233 int size; 234 int iphlen; 235 int optlen; 236 237 #ifdef DIAGNOSTIC 238 if (tp->t_inpcb && tp->t_in6pcb) 239 panic("tcp_segsize: both t_inpcb and t_in6pcb are set"); 240 #endif 241 switch (tp->t_family) { 242 #ifdef INET 243 case AF_INET: 244 iphlen = sizeof(struct ip); 245 break; 246 #endif 247 #ifdef INET6 248 case AF_INET6: 249 iphlen = sizeof(struct ip6_hdr); 250 break; 251 #endif 252 default: 253 size = tcp_mssdflt; 254 goto out; 255 } 256 257 rt = NULL; 258 #ifdef INET 259 if (inp) 260 rt = in_pcbrtentry(inp); 261 #endif 262 #ifdef INET6 263 if (in6p) 264 rt = in6_pcbrtentry(in6p); 265 #endif 266 if (rt == NULL) { 267 size = tcp_mssdflt; 268 goto out; 269 } 270 271 ifp = rt->rt_ifp; 272 273 size = tcp_mssdflt; 274 if (tp->t_mtudisc && rt->rt_rmx.rmx_mtu != 0) 275 size = rt->rt_rmx.rmx_mtu - iphlen - sizeof(struct tcphdr); 276 else if (ifp->if_flags & IFF_LOOPBACK) 277 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 278 #ifdef INET 279 else if (inp && tp->t_mtudisc) 280 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 281 else if (inp && in_localaddr(inp->inp_faddr)) 282 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 283 #endif 284 #ifdef INET6 285 else if (in6p) { 286 #ifdef INET 287 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) { 288 /* mapped addr case */ 289 struct in_addr d; 290 bcopy(&in6p->in6p_faddr.s6_addr32[3], &d, sizeof(d)); 291 if (tp->t_mtudisc || in_localaddr(d)) 292 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 293 } else 294 #endif 295 { 296 /* 297 * for IPv6, path MTU discovery is always turned on, 298 * or the node must use packet size <= 1280. 299 */ 300 size = tp->t_mtudisc ? ifp->if_mtu : IPV6_MMTU; 301 size -= (iphlen + sizeof(struct tcphdr)); 302 } 303 } 304 #endif 305 out: 306 /* 307 * Now we must make room for whatever extra TCP/IP options are in 308 * the packet. 309 */ 310 optlen = tcp_optlen(tp); 311 312 /* 313 * XXX tp->t_ourmss should have the right size, but without this code 314 * fragmentation will occur... need more investigation 315 */ 316 #ifdef INET 317 if (inp) { 318 #ifdef IPSEC 319 optlen += ipsec4_hdrsiz_tcp(tp); 320 #endif 321 optlen += ip_optlen(inp); 322 } 323 #endif 324 #ifdef INET6 325 #ifdef INET 326 if (in6p && tp->t_family == AF_INET) { 327 #ifdef IPSEC 328 optlen += ipsec4_hdrsiz_tcp(tp); 329 #endif 330 /* XXX size -= ip_optlen(in6p); */ 331 } else 332 #endif 333 if (in6p && tp->t_family == AF_INET6) { 334 #ifdef IPSEC 335 optlen += ipsec6_hdrsiz_tcp(tp); 336 #endif 337 optlen += ip6_optlen(in6p); 338 } 339 #endif 340 size -= optlen; 341 342 /* 343 * *rxsegsizep holds *estimated* inbound segment size (estimation 344 * assumes that path MTU is the same for both ways). this is only 345 * for silly window avoidance, do not use the value for other purposes. 346 * 347 * ipseclen is subtracted from both sides, this may not be right. 348 * I'm not quite sure about this (could someone comment). 349 */ 350 *txsegsizep = min(tp->t_peermss - optlen, size); 351 *rxsegsizep = min(tp->t_ourmss - optlen, size); 352 353 if (*txsegsizep != tp->t_segsz) { 354 /* 355 * If the new segment size is larger, we don't want to 356 * mess up the congestion window, but if it is smaller 357 * we'll have to reduce the congestion window to ensure 358 * that we don't get into trouble with initial windows 359 * and the rest. In any case, if the segment size 360 * has changed, chances are the path has, too, and 361 * our congestion window will be different. 362 */ 363 if (*txsegsizep < tp->t_segsz) { 364 tp->snd_cwnd = max((tp->snd_cwnd / tp->t_segsz) 365 * *txsegsizep, *txsegsizep); 366 tp->snd_ssthresh = max((tp->snd_ssthresh / tp->t_segsz) 367 * *txsegsizep, *txsegsizep); 368 } 369 tp->t_segsz = *txsegsizep; 370 } 371 } 372 373 static 374 #ifndef GPROF 375 __inline 376 #endif 377 int 378 tcp_build_datapkt(struct tcpcb *tp, struct socket *so, int off, 379 long len, int hdrlen, struct mbuf **mp) 380 { 381 struct mbuf *m; 382 383 if (tp->t_force && len == 1) 384 tcpstat.tcps_sndprobe++; 385 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { 386 tcpstat.tcps_sndrexmitpack++; 387 tcpstat.tcps_sndrexmitbyte += len; 388 } else { 389 tcpstat.tcps_sndpack++; 390 tcpstat.tcps_sndbyte += len; 391 } 392 #ifdef notyet 393 if ((m = m_copypack(so->so_snd.sb_mb, off, 394 (int)len, max_linkhdr + hdrlen)) == 0) 395 return (ENOBUFS); 396 /* 397 * m_copypack left space for our hdr; use it. 398 */ 399 m->m_len += hdrlen; 400 m->m_data -= hdrlen; 401 #else 402 MGETHDR(m, M_DONTWAIT, MT_HEADER); 403 if (__predict_false(m == NULL)) 404 return (ENOBUFS); 405 406 /* 407 * XXX Because other code assumes headers will fit in 408 * XXX one header mbuf. 409 * 410 * (This code should almost *never* be run.) 411 */ 412 if (__predict_false((max_linkhdr + hdrlen) > MHLEN)) { 413 TCP_OUTPUT_COUNTER_INCR(&tcp_output_bigheader); 414 MCLGET(m, M_DONTWAIT); 415 if ((m->m_flags & M_EXT) == 0) { 416 m_freem(m); 417 return (ENOBUFS); 418 } 419 } 420 421 m->m_data += max_linkhdr; 422 m->m_len = hdrlen; 423 if (len <= M_TRAILINGSPACE(m)) { 424 m_copydata(so->so_snd.sb_mb, off, (int) len, 425 mtod(m, caddr_t) + hdrlen); 426 m->m_len += len; 427 TCP_OUTPUT_COUNTER_INCR(&tcp_output_copysmall); 428 } else { 429 m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len); 430 if (m->m_next == NULL) { 431 m_freem(m); 432 return (ENOBUFS); 433 } 434 #ifdef TCP_OUTPUT_COUNTERS 435 if (m->m_next->m_flags & M_EXT) 436 TCP_OUTPUT_COUNTER_INCR(&tcp_output_refbig); 437 else 438 TCP_OUTPUT_COUNTER_INCR(&tcp_output_copybig); 439 #endif /* TCP_OUTPUT_COUNTERS */ 440 } 441 #endif 442 443 *mp = m; 444 return (0); 445 } 446 447 /* 448 * Tcp output routine: figure out what should be sent and send it. 449 */ 450 int 451 tcp_output(tp) 452 struct tcpcb *tp; 453 { 454 struct socket *so; 455 struct route *ro; 456 long len, win; 457 int off, flags, error; 458 struct mbuf *m; 459 struct ip *ip; 460 #ifdef INET6 461 struct ip6_hdr *ip6; 462 #endif 463 struct tcphdr *th; 464 u_char opt[MAX_TCPOPTLEN]; 465 unsigned optlen, hdrlen; 466 int idle, sendalot, txsegsize, rxsegsize; 467 int maxburst = TCP_MAXBURST; 468 int af; /* address family on the wire */ 469 int iphdrlen; 470 471 #ifdef DIAGNOSTIC 472 if (tp->t_inpcb && tp->t_in6pcb) 473 panic("tcp_output: both t_inpcb and t_in6pcb are set"); 474 #endif 475 so = NULL; 476 ro = NULL; 477 if (tp->t_inpcb) { 478 so = tp->t_inpcb->inp_socket; 479 ro = &tp->t_inpcb->inp_route; 480 } 481 #ifdef INET6 482 else if (tp->t_in6pcb) { 483 so = tp->t_in6pcb->in6p_socket; 484 ro = (struct route *)&tp->t_in6pcb->in6p_route; 485 } 486 #endif 487 488 switch (af = tp->t_family) { 489 #ifdef INET 490 case AF_INET: 491 if (tp->t_inpcb) 492 break; 493 #ifdef INET6 494 /* mapped addr case */ 495 if (tp->t_in6pcb) 496 break; 497 #endif 498 return EINVAL; 499 #endif 500 #ifdef INET6 501 case AF_INET6: 502 if (tp->t_in6pcb) 503 break; 504 return EINVAL; 505 #endif 506 default: 507 return EAFNOSUPPORT; 508 } 509 510 tcp_segsize(tp, &txsegsize, &rxsegsize); 511 512 idle = (tp->snd_max == tp->snd_una); 513 514 /* 515 * Restart Window computation. From draft-floyd-incr-init-win-03: 516 * 517 * Optionally, a TCP MAY set the restart window to the 518 * minimum of the value used for the initial window and 519 * the current value of cwnd (in other words, using a 520 * larger value for the restart window should never increase 521 * the size of cwnd). 522 */ 523 if (tcp_cwm) { 524 /* 525 * Hughes/Touch/Heidemann Congestion Window Monitoring. 526 * Count the number of packets currently pending 527 * acknowledgement, and limit our congestion window 528 * to a pre-determined allowed burst size plus that count. 529 * This prevents bursting once all pending packets have 530 * been acknowledged (i.e. transmission is idle). 531 * 532 * XXX Link this to Initial Window? 533 */ 534 tp->snd_cwnd = min(tp->snd_cwnd, 535 (tcp_cwm_burstsize * txsegsize) + 536 (tp->snd_nxt - tp->snd_una)); 537 } else { 538 if (idle && (tcp_now - tp->t_rcvtime) >= tp->t_rxtcur) { 539 /* 540 * We have been idle for "a while" and no acks are 541 * expected to clock out any data we send -- 542 * slow start to get ack "clock" running again. 543 */ 544 tp->snd_cwnd = min(tp->snd_cwnd, 545 TCP_INITIAL_WINDOW(tcp_init_win, txsegsize)); 546 } 547 } 548 549 again: 550 /* 551 * Determine length of data that should be transmitted, and 552 * flags that should be used. If there is some data or critical 553 * controls (SYN, RST) to send, then transmit; otherwise, 554 * investigate further. 555 */ 556 sendalot = 0; 557 off = tp->snd_nxt - tp->snd_una; 558 win = min(tp->snd_wnd, tp->snd_cwnd); 559 560 flags = tcp_outflags[tp->t_state]; 561 /* 562 * If in persist timeout with window of 0, send 1 byte. 563 * Otherwise, if window is small but nonzero 564 * and timer expired, we will send what we can 565 * and go to transmit state. 566 */ 567 if (tp->t_force) { 568 if (win == 0) { 569 /* 570 * If we still have some data to send, then 571 * clear the FIN bit. Usually this would 572 * happen below when it realizes that we 573 * aren't sending all the data. However, 574 * if we have exactly 1 byte of unset data, 575 * then it won't clear the FIN bit below, 576 * and if we are in persist state, we wind 577 * up sending the packet without recording 578 * that we sent the FIN bit. 579 * 580 * We can't just blindly clear the FIN bit, 581 * because if we don't have any more data 582 * to send then the probe will be the FIN 583 * itself. 584 */ 585 if (off < so->so_snd.sb_cc) 586 flags &= ~TH_FIN; 587 win = 1; 588 } else { 589 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 590 tp->t_rxtshift = 0; 591 } 592 } 593 594 if (win < so->so_snd.sb_cc) { 595 len = win - off; 596 flags &= ~TH_FIN; 597 } else 598 len = so->so_snd.sb_cc - off; 599 600 if (len < 0) { 601 /* 602 * If FIN has been sent but not acked, 603 * but we haven't been called to retransmit, 604 * len will be -1. Otherwise, window shrank 605 * after we sent into it. If window shrank to 0, 606 * cancel pending retransmit, pull snd_nxt back 607 * to (closed) window, and set the persist timer 608 * if it isn't already going. If the window didn't 609 * close completely, just wait for an ACK. 610 * 611 * If we have a pending FIN, either it has already been 612 * transmitted or it is outside the window, so drop it. 613 * If the FIN has been transmitted, but this is not a 614 * retransmission, then len must be -1. Therefore we also 615 * prevent here the sending of `gratuitous FINs'. This 616 * eliminates the need to check for that case below (e.g. 617 * to back up snd_nxt before the FIN so that the sequence 618 * number is correct). 619 */ 620 len = 0; 621 flags &= ~TH_FIN; 622 if (win == 0) { 623 TCP_TIMER_DISARM(tp, TCPT_REXMT); 624 tp->t_rxtshift = 0; 625 tp->snd_nxt = tp->snd_una; 626 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 627 tcp_setpersist(tp); 628 } 629 } 630 if (len > txsegsize) { 631 len = txsegsize; 632 flags &= ~TH_FIN; 633 sendalot = 1; 634 } 635 636 win = sbspace(&so->so_rcv); 637 638 /* 639 * Sender silly window avoidance. If connection is idle 640 * and can send all data, a maximum segment, 641 * at least a maximum default-size segment do it, 642 * or are forced, do it; otherwise don't bother. 643 * If peer's buffer is tiny, then send 644 * when window is at least half open. 645 * If retransmitting (possibly after persist timer forced us 646 * to send into a small window), then must resend. 647 */ 648 if (len) { 649 if (len == txsegsize) 650 goto send; 651 if ((so->so_state & SS_MORETOCOME) == 0 && 652 ((idle || tp->t_flags & TF_NODELAY) && 653 len + off >= so->so_snd.sb_cc)) 654 goto send; 655 if (tp->t_force) 656 goto send; 657 if (len >= tp->max_sndwnd / 2) 658 goto send; 659 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) 660 goto send; 661 } 662 663 /* 664 * Compare available window to amount of window known to peer 665 * (as advertised window less next expected input). If the 666 * difference is at least twice the size of the largest segment 667 * we expect to receive (i.e. two segments) or at least 50% of 668 * the maximum possible window, then want to send a window update 669 * to peer. 670 */ 671 if (win > 0) { 672 /* 673 * "adv" is the amount we can increase the window, 674 * taking into account that we are limited by 675 * TCP_MAXWIN << tp->rcv_scale. 676 */ 677 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) - 678 (tp->rcv_adv - tp->rcv_nxt); 679 680 if (adv >= (long) (2 * rxsegsize)) 681 goto send; 682 if (2 * adv >= (long) so->so_rcv.sb_hiwat) 683 goto send; 684 } 685 686 /* 687 * Send if we owe peer an ACK. 688 */ 689 if (tp->t_flags & TF_ACKNOW) 690 goto send; 691 if (flags & (TH_SYN|TH_FIN|TH_RST)) 692 goto send; 693 if (SEQ_GT(tp->snd_up, tp->snd_una)) 694 goto send; 695 696 /* 697 * TCP window updates are not reliable, rather a polling protocol 698 * using ``persist'' packets is used to insure receipt of window 699 * updates. The three ``states'' for the output side are: 700 * idle not doing retransmits or persists 701 * persisting to move a small or zero window 702 * (re)transmitting and thereby not persisting 703 * 704 * tp->t_timer[TCPT_PERSIST] 705 * is set when we are in persist state. 706 * tp->t_force 707 * is set when we are called to send a persist packet. 708 * tp->t_timer[TCPT_REXMT] 709 * is set when we are retransmitting 710 * The output side is idle when both timers are zero. 711 * 712 * If send window is too small, there is data to transmit, and no 713 * retransmit or persist is pending, then go to persist state. 714 * If nothing happens soon, send when timer expires: 715 * if window is nonzero, transmit what we can, 716 * otherwise force out a byte. 717 */ 718 if (so->so_snd.sb_cc && TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 719 TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 720 tp->t_rxtshift = 0; 721 tcp_setpersist(tp); 722 } 723 724 /* 725 * No reason to send a segment, just return. 726 */ 727 return (0); 728 729 send: 730 /* 731 * Before ESTABLISHED, force sending of initial options 732 * unless TCP set not to do any options. 733 * NOTE: we assume that the IP/TCP header plus TCP options 734 * always fit in a single mbuf, leaving room for a maximum 735 * link header, i.e. 736 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES 737 */ 738 optlen = 0; 739 switch (af) { 740 #ifdef INET 741 case AF_INET: 742 iphdrlen = sizeof(struct ip) + sizeof(struct tcphdr); 743 break; 744 #endif 745 #ifdef INET6 746 case AF_INET6: 747 iphdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 748 break; 749 #endif 750 default: /*pacify gcc*/ 751 iphdrlen = 0; 752 break; 753 } 754 hdrlen = iphdrlen; 755 if (flags & TH_SYN) { 756 struct rtentry *rt; 757 758 rt = NULL; 759 #ifdef INET 760 if (tp->t_inpcb) 761 rt = in_pcbrtentry(tp->t_inpcb); 762 #endif 763 #ifdef INET6 764 if (tp->t_in6pcb) 765 rt = in6_pcbrtentry(tp->t_in6pcb); 766 #endif 767 768 tp->snd_nxt = tp->iss; 769 tp->t_ourmss = tcp_mss_to_advertise(rt != NULL ? 770 rt->rt_ifp : NULL, af); 771 if ((tp->t_flags & TF_NOOPT) == 0) { 772 opt[0] = TCPOPT_MAXSEG; 773 opt[1] = 4; 774 opt[2] = (tp->t_ourmss >> 8) & 0xff; 775 opt[3] = tp->t_ourmss & 0xff; 776 optlen = 4; 777 778 if ((tp->t_flags & TF_REQ_SCALE) && 779 ((flags & TH_ACK) == 0 || 780 (tp->t_flags & TF_RCVD_SCALE))) { 781 *((u_int32_t *) (opt + optlen)) = htonl( 782 TCPOPT_NOP << 24 | 783 TCPOPT_WINDOW << 16 | 784 TCPOLEN_WINDOW << 8 | 785 tp->request_r_scale); 786 optlen += 4; 787 } 788 } 789 } 790 791 /* 792 * Send a timestamp and echo-reply if this is a SYN and our side 793 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side 794 * and our peer have sent timestamps in our SYN's. 795 */ 796 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 797 (flags & TH_RST) == 0 && 798 ((flags & (TH_SYN|TH_ACK)) == TH_SYN || 799 (tp->t_flags & TF_RCVD_TSTMP))) { 800 u_int32_t *lp = (u_int32_t *)(opt + optlen); 801 802 /* Form timestamp option as shown in appendix A of RFC 1323. */ 803 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 804 *lp++ = htonl(TCP_TIMESTAMP(tp)); 805 *lp = htonl(tp->ts_recent); 806 optlen += TCPOLEN_TSTAMP_APPA; 807 } 808 809 hdrlen += optlen; 810 811 #ifdef DIAGNOSTIC 812 if (len > txsegsize) 813 panic("tcp data to be sent is larger than segment"); 814 if (max_linkhdr + hdrlen > MCLBYTES) 815 panic("tcphdr too big"); 816 #endif 817 818 /* 819 * Grab a header mbuf, attaching a copy of data to 820 * be transmitted, and initialize the header from 821 * the template for sends on this connection. 822 */ 823 if (len) { 824 error = tcp_build_datapkt(tp, so, off, len, hdrlen, &m); 825 if (error) 826 goto out; 827 /* 828 * If we're sending everything we've got, set PUSH. 829 * (This will keep happy those implementations which only 830 * give data to the user when a buffer fills or 831 * a PUSH comes in.) 832 */ 833 if (off + len == so->so_snd.sb_cc) 834 flags |= TH_PUSH; 835 } else { 836 if (tp->t_flags & TF_ACKNOW) 837 tcpstat.tcps_sndacks++; 838 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 839 tcpstat.tcps_sndctrl++; 840 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 841 tcpstat.tcps_sndurg++; 842 else 843 tcpstat.tcps_sndwinup++; 844 845 MGETHDR(m, M_DONTWAIT, MT_HEADER); 846 if (m != NULL && max_linkhdr + hdrlen > MHLEN) { 847 MCLGET(m, M_DONTWAIT); 848 if ((m->m_flags & M_EXT) == 0) { 849 m_freem(m); 850 m = NULL; 851 } 852 } 853 if (m == NULL) { 854 error = ENOBUFS; 855 goto out; 856 } 857 m->m_data += max_linkhdr; 858 m->m_len = hdrlen; 859 } 860 m->m_pkthdr.rcvif = (struct ifnet *)0; 861 switch (af) { 862 #ifdef INET 863 case AF_INET: 864 ip = mtod(m, struct ip *); 865 #ifdef INET6 866 ip6 = NULL; 867 #endif 868 th = (struct tcphdr *)(ip + 1); 869 break; 870 #endif 871 #ifdef INET6 872 case AF_INET6: 873 ip = NULL; 874 ip6 = mtod(m, struct ip6_hdr *); 875 th = (struct tcphdr *)(ip6 + 1); 876 break; 877 #endif 878 default: /*pacify gcc*/ 879 ip = NULL; 880 #ifdef INET6 881 ip6 = NULL; 882 #endif 883 th = NULL; 884 break; 885 } 886 if (tp->t_template == 0) 887 panic("tcp_output"); 888 if (tp->t_template->m_len < iphdrlen) 889 panic("tcp_output"); 890 bcopy(mtod(tp->t_template, caddr_t), mtod(m, caddr_t), iphdrlen); 891 892 /* 893 * If we are doing retransmissions, then snd_nxt will 894 * not reflect the first unsent octet. For ACK only 895 * packets, we do not want the sequence number of the 896 * retransmitted packet, we want the sequence number 897 * of the next unsent octet. So, if there is no data 898 * (and no SYN or FIN), use snd_max instead of snd_nxt 899 * when filling in ti_seq. But if we are in persist 900 * state, snd_max might reflect one byte beyond the 901 * right edge of the window, so use snd_nxt in that 902 * case, since we know we aren't doing a retransmission. 903 * (retransmit and persist are mutually exclusive...) 904 */ 905 if (len || (flags & (TH_SYN|TH_FIN)) || 906 TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) 907 th->th_seq = htonl(tp->snd_nxt); 908 else 909 th->th_seq = htonl(tp->snd_max); 910 th->th_ack = htonl(tp->rcv_nxt); 911 if (optlen) { 912 bcopy((caddr_t)opt, (caddr_t)(th + 1), optlen); 913 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 914 } 915 th->th_flags = flags; 916 /* 917 * Calculate receive window. Don't shrink window, 918 * but avoid silly window syndrome. 919 */ 920 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)rxsegsize) 921 win = 0; 922 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 923 win = (long)TCP_MAXWIN << tp->rcv_scale; 924 if (win < (long)(tp->rcv_adv - tp->rcv_nxt)) 925 win = (long)(tp->rcv_adv - tp->rcv_nxt); 926 th->th_win = htons((u_int16_t) (win>>tp->rcv_scale)); 927 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 928 u_int32_t urp = tp->snd_up - tp->snd_nxt; 929 if (urp > IP_MAXPACKET) 930 urp = IP_MAXPACKET; 931 th->th_urp = htons((u_int16_t)urp); 932 th->th_flags |= TH_URG; 933 } else 934 /* 935 * If no urgent pointer to send, then we pull 936 * the urgent pointer to the left edge of the send window 937 * so that it doesn't drift into the send window on sequence 938 * number wraparound. 939 */ 940 tp->snd_up = tp->snd_una; /* drag it along */ 941 942 /* 943 * Set ourselves up to be checksummed just before the packet 944 * hits the wire. 945 */ 946 switch (af) { 947 #ifdef INET 948 case AF_INET: 949 m->m_pkthdr.csum_flags = M_CSUM_TCPv4; 950 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 951 if (len + optlen) { 952 /* Fixup the pseudo-header checksum. */ 953 /* XXXJRT Not IP Jumbogram safe. */ 954 th->th_sum = in_cksum_addword(th->th_sum, 955 htons((u_int16_t) (len + optlen))); 956 } 957 break; 958 #endif 959 #ifdef INET6 960 case AF_INET6: 961 /* 962 * XXX Actually delaying the checksum is Hard 963 * XXX (well, maybe not for Itojun, but it is 964 * XXX for me), but we can still take advantage 965 * XXX of the cached pseudo-header checksum. 966 */ 967 /* equals to hdrlen + len */ 968 m->m_pkthdr.len = sizeof(struct ip6_hdr) 969 + sizeof(struct tcphdr) + optlen + len; 970 #ifdef notyet 971 m->m_pkthdr.csum_flags = M_CSUM_TCPv6; 972 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 973 #endif 974 if (len + optlen) { 975 /* Fixup the pseudo-header checksum. */ 976 /* XXXJRT: Not IPv6 Jumbogram safe. */ 977 th->th_sum = in_cksum_addword(th->th_sum, 978 htons((u_int16_t) (len + optlen))); 979 } 980 #ifndef notyet 981 th->th_sum = in6_cksum(m, 0, sizeof(struct ip6_hdr), 982 sizeof(struct tcphdr) + optlen + len); 983 #endif 984 break; 985 #endif 986 } 987 988 /* 989 * In transmit state, time the transmission and arrange for 990 * the retransmit. In persist state, just set snd_max. 991 */ 992 if (tp->t_force == 0 || TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 993 tcp_seq startseq = tp->snd_nxt; 994 995 /* 996 * Advance snd_nxt over sequence space of this segment. 997 * There are no states in which we send both a SYN and a FIN, 998 * so we collapse the tests for these flags. 999 */ 1000 if (flags & (TH_SYN|TH_FIN)) 1001 tp->snd_nxt++; 1002 tp->snd_nxt += len; 1003 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 1004 tp->snd_max = tp->snd_nxt; 1005 /* 1006 * Time this transmission if not a retransmission and 1007 * not currently timing anything. 1008 */ 1009 if (tp->t_rtttime == 0) { 1010 tp->t_rtttime = tcp_now; 1011 tp->t_rtseq = startseq; 1012 tcpstat.tcps_segstimed++; 1013 } 1014 } 1015 1016 /* 1017 * Set retransmit timer if not currently set, 1018 * and not doing an ack or a keep-alive probe. 1019 * Initial value for retransmit timer is smoothed 1020 * round-trip time + 2 * round-trip time variance. 1021 * Initialize shift counter which is used for backoff 1022 * of retransmit time. 1023 */ 1024 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 1025 tp->snd_nxt != tp->snd_una) { 1026 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1027 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) { 1028 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 1029 tp->t_rxtshift = 0; 1030 } 1031 } 1032 } else 1033 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max)) 1034 tp->snd_max = tp->snd_nxt + len; 1035 1036 #ifdef TCP_DEBUG 1037 /* 1038 * Trace. 1039 */ 1040 if (so->so_options & SO_DEBUG) { 1041 /* 1042 * need to recover version # field, which was overwritten 1043 * on ip_cksum computation. 1044 */ 1045 struct ip *sip; 1046 sip = mtod(m, struct ip *); 1047 switch (af) { 1048 #ifdef INET 1049 case AF_INET: 1050 sip->ip_v = 4; 1051 break; 1052 #endif 1053 #ifdef INET6 1054 case AF_INET6: 1055 sip->ip_v = 6; 1056 break; 1057 #endif 1058 } 1059 tcp_trace(TA_OUTPUT, tp->t_state, tp, m, 0); 1060 } 1061 #endif 1062 1063 /* 1064 * Fill in IP length and desired time to live and 1065 * send to IP level. There should be a better way 1066 * to handle ttl and tos; we could keep them in 1067 * the template, but need a way to checksum without them. 1068 */ 1069 m->m_pkthdr.len = hdrlen + len; 1070 1071 switch (af) { 1072 #ifdef INET 1073 case AF_INET: 1074 ip->ip_len = m->m_pkthdr.len; 1075 if (tp->t_inpcb) { 1076 ip->ip_ttl = tp->t_inpcb->inp_ip.ip_ttl; 1077 ip->ip_tos = tp->t_inpcb->inp_ip.ip_tos; 1078 } 1079 #ifdef INET6 1080 else if (tp->t_in6pcb) { 1081 ip->ip_ttl = in6_selecthlim(tp->t_in6pcb, NULL); /*XXX*/ 1082 ip->ip_tos = 0; /*XXX*/ 1083 } 1084 #endif 1085 break; 1086 #endif 1087 #ifdef INET6 1088 case AF_INET6: 1089 ip6->ip6_nxt = IPPROTO_TCP; 1090 if (tp->t_in6pcb) { 1091 /* 1092 * we separately set hoplimit for every segment, since 1093 * the user might want to change the value via 1094 * setsockopt. Also, desired default hop limit might 1095 * be changed via Neighbor Discovery. 1096 */ 1097 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, 1098 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 1099 } 1100 /* ip6->ip6_flow = ??? */ 1101 /* ip6_plen will be filled in ip6_output(). */ 1102 break; 1103 #endif 1104 } 1105 1106 #ifdef IPSEC 1107 if (ipsec_setsocket(m, so) != 0) { 1108 m_freem(m); 1109 error = ENOBUFS; 1110 goto out; 1111 } 1112 #endif /*IPSEC*/ 1113 1114 switch (af) { 1115 #ifdef INET 1116 case AF_INET: 1117 { 1118 struct mbuf *opts; 1119 1120 if (tp->t_inpcb) 1121 opts = tp->t_inpcb->inp_options; 1122 else 1123 opts = NULL; 1124 error = ip_output(m, opts, ro, 1125 (tp->t_mtudisc ? IP_MTUDISC : 0) | 1126 (so->so_options & SO_DONTROUTE), 1127 0); 1128 break; 1129 } 1130 #endif 1131 #ifdef INET6 1132 case AF_INET6: 1133 { 1134 struct ip6_pktopts *opts; 1135 1136 if (tp->t_in6pcb) 1137 opts = tp->t_in6pcb->in6p_outputopts; 1138 else 1139 opts = NULL; 1140 error = ip6_output(m, opts, (struct route_in6 *)ro, 1141 so->so_options & SO_DONTROUTE, 0, NULL); 1142 break; 1143 } 1144 #endif 1145 default: 1146 error = EAFNOSUPPORT; 1147 break; 1148 } 1149 if (error) { 1150 out: 1151 if (error == ENOBUFS) { 1152 tcpstat.tcps_selfquench++; 1153 #ifdef INET 1154 if (tp->t_inpcb) 1155 tcp_quench(tp->t_inpcb, 0); 1156 #endif 1157 #ifdef INET6 1158 if (tp->t_in6pcb) 1159 tcp6_quench(tp->t_in6pcb, 0); 1160 #endif 1161 error = 0; 1162 } else if ((error == EHOSTUNREACH || error == ENETDOWN) && 1163 TCPS_HAVERCVDSYN(tp->t_state)) { 1164 tp->t_softerror = error; 1165 error = 0; 1166 } 1167 1168 /* Restart the delayed ACK timer, if necessary. */ 1169 if (tp->t_flags & TF_DELACK) 1170 TCP_RESTART_DELACK(tp); 1171 1172 return (error); 1173 } 1174 tcpstat.tcps_sndtotal++; 1175 if (tp->t_flags & TF_DELACK) 1176 tcpstat.tcps_delack++; 1177 1178 /* 1179 * Data sent (as far as we can tell). 1180 * If this advertises a larger window than any other segment, 1181 * then remember the size of the advertised window. 1182 * Any pending ACK has now been sent. 1183 */ 1184 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 1185 tp->rcv_adv = tp->rcv_nxt + win; 1186 tp->last_ack_sent = tp->rcv_nxt; 1187 tp->t_flags &= ~TF_ACKNOW; 1188 TCP_CLEAR_DELACK(tp); 1189 #ifdef DIAGNOSTIC 1190 if (maxburst < 0) 1191 printf("tcp_output: maxburst exceeded by %d\n", -maxburst); 1192 #endif 1193 if (sendalot && (!tcp_do_newreno || --maxburst)) 1194 goto again; 1195 return (0); 1196 } 1197 1198 void 1199 tcp_setpersist(tp) 1200 struct tcpcb *tp; 1201 { 1202 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2); 1203 int nticks; 1204 1205 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT)) 1206 panic("tcp_output REXMT"); 1207 /* 1208 * Start/restart persistance timer. 1209 */ 1210 if (t < tp->t_rttmin) 1211 t = tp->t_rttmin; 1212 TCPT_RANGESET(nticks, t * tcp_backoff[tp->t_rxtshift], 1213 TCPTV_PERSMIN, TCPTV_PERSMAX); 1214 TCP_TIMER_ARM(tp, TCPT_PERSIST, nticks); 1215 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 1216 tp->t_rxtshift++; 1217 } 1218