1 /* $NetBSD: tcp_output.c,v 1.77 2002/01/24 02:12:29 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 1.77 2002/01/24 02:12:29 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 static 204 #ifndef GPROF 205 __inline 206 #endif 207 void 208 tcp_segsize(struct tcpcb *tp, int *txsegsizep, int *rxsegsizep) 209 { 210 #ifdef INET 211 struct inpcb *inp = tp->t_inpcb; 212 #endif 213 #ifdef INET6 214 struct in6pcb *in6p = tp->t_in6pcb; 215 #endif 216 struct rtentry *rt; 217 struct ifnet *ifp; 218 int size; 219 int iphlen; 220 int optlen; 221 222 #ifdef DIAGNOSTIC 223 if (tp->t_inpcb && tp->t_in6pcb) 224 panic("tcp_segsize: both t_inpcb and t_in6pcb are set"); 225 #endif 226 switch (tp->t_family) { 227 #ifdef INET 228 case AF_INET: 229 iphlen = sizeof(struct ip); 230 break; 231 #endif 232 #ifdef INET6 233 case AF_INET6: 234 iphlen = sizeof(struct ip6_hdr); 235 break; 236 #endif 237 default: 238 size = tcp_mssdflt; 239 goto out; 240 } 241 242 rt = NULL; 243 #ifdef INET 244 if (inp) 245 rt = in_pcbrtentry(inp); 246 #endif 247 #ifdef INET6 248 if (in6p) 249 rt = in6_pcbrtentry(in6p); 250 #endif 251 if (rt == NULL) { 252 size = tcp_mssdflt; 253 goto out; 254 } 255 256 ifp = rt->rt_ifp; 257 258 size = tcp_mssdflt; 259 if (rt->rt_rmx.rmx_mtu != 0) 260 size = rt->rt_rmx.rmx_mtu - iphlen - sizeof(struct tcphdr); 261 else if (ifp->if_flags & IFF_LOOPBACK) 262 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 263 #ifdef INET 264 else if (inp && ip_mtudisc) 265 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 266 else if (inp && in_localaddr(inp->inp_faddr)) 267 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 268 #endif 269 #ifdef INET6 270 else if (in6p) { 271 #ifdef INET 272 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) { 273 /* mapped addr case */ 274 struct in_addr d; 275 bcopy(&in6p->in6p_faddr.s6_addr32[3], &d, sizeof(d)); 276 if (ip_mtudisc || in_localaddr(d)) 277 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 278 } else 279 #endif 280 { 281 /* 282 * for IPv6, path MTU discovery is always turned on, 283 * or the node must use packet size <= 1280. 284 */ 285 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 286 } 287 } 288 #endif 289 /* 290 * Now we must make room for whatever extra TCP/IP options are in 291 * the packet. 292 */ 293 optlen = tcp_optlen(tp); 294 295 /* 296 * XXX tp->t_ourmss should have the right size, but without this code 297 * fragmentation will occur... need more investigation 298 */ 299 #ifdef INET 300 if (inp) { 301 #ifdef IPSEC 302 optlen += ipsec4_hdrsiz_tcp(tp); 303 #endif 304 optlen += ip_optlen(inp); 305 } 306 #endif 307 #ifdef INET6 308 #ifdef INET 309 if (in6p && tp->t_family == AF_INET) { 310 #ifdef IPSEC 311 optlen += ipsec4_hdrsiz_tcp(tp); 312 #endif 313 /* XXX size -= ip_optlen(in6p); */ 314 } else 315 #endif 316 if (in6p && tp->t_family == AF_INET6) { 317 #ifdef IPSEC 318 optlen += ipsec6_hdrsiz_tcp(tp); 319 #endif 320 optlen += ip6_optlen(in6p); 321 } 322 #endif 323 size -= optlen; 324 325 out: 326 /* 327 * *rxsegsizep holds *estimated* inbound segment size (estimation 328 * assumes that path MTU is the same for both ways). this is only 329 * for silly window avoidance, do not use the value for other purposes. 330 * 331 * ipseclen is subtracted from both sides, this may not be right. 332 * I'm not quite sure about this (could someone comment). 333 */ 334 *txsegsizep = min(tp->t_peermss - optlen, size); 335 *rxsegsizep = min(tp->t_ourmss - optlen, size); 336 337 if (*txsegsizep != tp->t_segsz) { 338 /* 339 * If the new segment size is larger, we don't want to 340 * mess up the congestion window, but if it is smaller 341 * we'll have to reduce the congestion window to ensure 342 * that we don't get into trouble with initial windows 343 * and the rest. In any case, if the segment size 344 * has changed, chances are the path has, too, and 345 * our congestion window will be different. 346 */ 347 if (*txsegsizep < tp->t_segsz) { 348 tp->snd_cwnd = max((tp->snd_cwnd / tp->t_segsz) 349 * *txsegsizep, *txsegsizep); 350 tp->snd_ssthresh = max((tp->snd_ssthresh / tp->t_segsz) 351 * *txsegsizep, *txsegsizep); 352 } 353 tp->t_segsz = *txsegsizep; 354 } 355 } 356 357 static 358 #ifndef GPROF 359 __inline 360 #endif 361 int 362 tcp_build_datapkt(struct tcpcb *tp, struct socket *so, int off, 363 long len, int hdrlen, struct mbuf **mp) 364 { 365 struct mbuf *m; 366 367 if (tp->t_force && len == 1) 368 tcpstat.tcps_sndprobe++; 369 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { 370 tcpstat.tcps_sndrexmitpack++; 371 tcpstat.tcps_sndrexmitbyte += len; 372 } else { 373 tcpstat.tcps_sndpack++; 374 tcpstat.tcps_sndbyte += len; 375 } 376 #ifdef notyet 377 if ((m = m_copypack(so->so_snd.sb_mb, off, 378 (int)len, max_linkhdr + hdrlen)) == 0) 379 return (ENOBUFS); 380 /* 381 * m_copypack left space for our hdr; use it. 382 */ 383 m->m_len += hdrlen; 384 m->m_data -= hdrlen; 385 #else 386 MGETHDR(m, M_DONTWAIT, MT_HEADER); 387 if (m != NULL && 388 (max_linkhdr + hdrlen > MHLEN || 389 max_linkhdr + hdrlen + len <= MCLBYTES)) { 390 MCLGET(m, M_DONTWAIT); 391 if ((m->m_flags & M_EXT) == 0) { 392 m_freem(m); 393 m = NULL; 394 } 395 } 396 if (m == NULL) 397 return (ENOBUFS); 398 m->m_data += max_linkhdr; 399 m->m_len = hdrlen; 400 if (len <= M_TRAILINGSPACE(m)) { 401 m_copydata(so->so_snd.sb_mb, off, (int) len, 402 mtod(m, caddr_t) + hdrlen); 403 m->m_len += len; 404 } else { 405 m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len); 406 if (m->m_next == NULL) { 407 m_freem(m); 408 return (ENOBUFS); 409 } 410 } 411 #endif 412 413 *mp = m; 414 return (0); 415 } 416 417 /* 418 * Tcp output routine: figure out what should be sent and send it. 419 */ 420 int 421 tcp_output(tp) 422 struct tcpcb *tp; 423 { 424 struct socket *so; 425 struct route *ro; 426 long len, win; 427 int off, flags, error; 428 struct mbuf *m; 429 struct ip *ip; 430 #ifdef INET6 431 struct ip6_hdr *ip6; 432 #endif 433 struct tcphdr *th; 434 u_char opt[MAX_TCPOPTLEN]; 435 unsigned optlen, hdrlen; 436 int idle, sendalot, txsegsize, rxsegsize; 437 int maxburst = TCP_MAXBURST; 438 int af; /* address family on the wire */ 439 int iphdrlen; 440 441 #ifdef DIAGNOSTIC 442 if (tp->t_inpcb && tp->t_in6pcb) 443 panic("tcp_output: both t_inpcb and t_in6pcb are set"); 444 #endif 445 so = NULL; 446 ro = NULL; 447 if (tp->t_inpcb) { 448 so = tp->t_inpcb->inp_socket; 449 ro = &tp->t_inpcb->inp_route; 450 } 451 #ifdef INET6 452 else if (tp->t_in6pcb) { 453 so = tp->t_in6pcb->in6p_socket; 454 ro = (struct route *)&tp->t_in6pcb->in6p_route; 455 } 456 #endif 457 458 switch (af = tp->t_family) { 459 #ifdef INET 460 case AF_INET: 461 if (tp->t_inpcb) 462 break; 463 #ifdef INET6 464 /* mapped addr case */ 465 if (tp->t_in6pcb) 466 break; 467 #endif 468 return EINVAL; 469 #endif 470 #ifdef INET6 471 case AF_INET6: 472 if (tp->t_in6pcb) 473 break; 474 return EINVAL; 475 #endif 476 default: 477 return EAFNOSUPPORT; 478 } 479 480 tcp_segsize(tp, &txsegsize, &rxsegsize); 481 482 idle = (tp->snd_max == tp->snd_una); 483 484 /* 485 * Restart Window computation. From draft-floyd-incr-init-win-03: 486 * 487 * Optionally, a TCP MAY set the restart window to the 488 * minimum of the value used for the initial window and 489 * the current value of cwnd (in other words, using a 490 * larger value for the restart window should never increase 491 * the size of cwnd). 492 */ 493 if (tcp_cwm) { 494 /* 495 * Hughes/Touch/Heidemann Congestion Window Monitoring. 496 * Count the number of packets currently pending 497 * acknowledgement, and limit our congestion window 498 * to a pre-determined allowed burst size plus that count. 499 * This prevents bursting once all pending packets have 500 * been acknowledged (i.e. transmission is idle). 501 * 502 * XXX Link this to Initial Window? 503 */ 504 tp->snd_cwnd = min(tp->snd_cwnd, 505 (tcp_cwm_burstsize * txsegsize) + 506 (tp->snd_nxt - tp->snd_una)); 507 } else { 508 if (idle && (tcp_now - tp->t_rcvtime) >= tp->t_rxtcur) { 509 /* 510 * We have been idle for "a while" and no acks are 511 * expected to clock out any data we send -- 512 * slow start to get ack "clock" running again. 513 */ 514 tp->snd_cwnd = min(tp->snd_cwnd, 515 TCP_INITIAL_WINDOW(tcp_init_win, txsegsize)); 516 } 517 } 518 519 again: 520 /* 521 * Determine length of data that should be transmitted, and 522 * flags that should be used. If there is some data or critical 523 * controls (SYN, RST) to send, then transmit; otherwise, 524 * investigate further. 525 */ 526 sendalot = 0; 527 off = tp->snd_nxt - tp->snd_una; 528 win = min(tp->snd_wnd, tp->snd_cwnd); 529 530 flags = tcp_outflags[tp->t_state]; 531 /* 532 * If in persist timeout with window of 0, send 1 byte. 533 * Otherwise, if window is small but nonzero 534 * and timer expired, we will send what we can 535 * and go to transmit state. 536 */ 537 if (tp->t_force) { 538 if (win == 0) { 539 /* 540 * If we still have some data to send, then 541 * clear the FIN bit. Usually this would 542 * happen below when it realizes that we 543 * aren't sending all the data. However, 544 * if we have exactly 1 byte of unset data, 545 * then it won't clear the FIN bit below, 546 * and if we are in persist state, we wind 547 * up sending the packet without recording 548 * that we sent the FIN bit. 549 * 550 * We can't just blindly clear the FIN bit, 551 * because if we don't have any more data 552 * to send then the probe will be the FIN 553 * itself. 554 */ 555 if (off < so->so_snd.sb_cc) 556 flags &= ~TH_FIN; 557 win = 1; 558 } else { 559 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 560 tp->t_rxtshift = 0; 561 } 562 } 563 564 if (win < so->so_snd.sb_cc) { 565 len = win - off; 566 flags &= ~TH_FIN; 567 } else 568 len = so->so_snd.sb_cc - off; 569 570 if (len < 0) { 571 /* 572 * If FIN has been sent but not acked, 573 * but we haven't been called to retransmit, 574 * len will be -1. Otherwise, window shrank 575 * after we sent into it. If window shrank to 0, 576 * cancel pending retransmit, pull snd_nxt back 577 * to (closed) window, and set the persist timer 578 * if it isn't already going. If the window didn't 579 * close completely, just wait for an ACK. 580 * 581 * If we have a pending FIN, either it has already been 582 * transmitted or it is outside the window, so drop it. 583 * If the FIN has been transmitted, but this is not a 584 * retransmission, then len must be -1. Therefore we also 585 * prevent here the sending of `gratuitous FINs'. This 586 * eliminates the need to check for that case below (e.g. 587 * to back up snd_nxt before the FIN so that the sequence 588 * number is correct). 589 */ 590 len = 0; 591 flags &= ~TH_FIN; 592 if (win == 0) { 593 TCP_TIMER_DISARM(tp, TCPT_REXMT); 594 tp->t_rxtshift = 0; 595 tp->snd_nxt = tp->snd_una; 596 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 597 tcp_setpersist(tp); 598 } 599 } 600 if (len > txsegsize) { 601 len = txsegsize; 602 flags &= ~TH_FIN; 603 sendalot = 1; 604 } 605 606 win = sbspace(&so->so_rcv); 607 608 /* 609 * Sender silly window avoidance. If connection is idle 610 * and can send all data, a maximum segment, 611 * at least a maximum default-size segment do it, 612 * or are forced, do it; otherwise don't bother. 613 * If peer's buffer is tiny, then send 614 * when window is at least half open. 615 * If retransmitting (possibly after persist timer forced us 616 * to send into a small window), then must resend. 617 */ 618 if (len) { 619 if (len == txsegsize) 620 goto send; 621 if ((so->so_state & SS_MORETOCOME) == 0 && 622 ((idle || tp->t_flags & TF_NODELAY) && 623 len + off >= so->so_snd.sb_cc)) 624 goto send; 625 if (tp->t_force) 626 goto send; 627 if (len >= tp->max_sndwnd / 2) 628 goto send; 629 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) 630 goto send; 631 } 632 633 /* 634 * Compare available window to amount of window known to peer 635 * (as advertised window less next expected input). If the 636 * difference is at least twice the size of the largest segment 637 * we expect to receive (i.e. two segments) or at least 50% of 638 * the maximum possible window, then want to send a window update 639 * to peer. 640 */ 641 if (win > 0) { 642 /* 643 * "adv" is the amount we can increase the window, 644 * taking into account that we are limited by 645 * TCP_MAXWIN << tp->rcv_scale. 646 */ 647 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) - 648 (tp->rcv_adv - tp->rcv_nxt); 649 650 if (adv >= (long) (2 * rxsegsize)) 651 goto send; 652 if (2 * adv >= (long) so->so_rcv.sb_hiwat) 653 goto send; 654 } 655 656 /* 657 * Send if we owe peer an ACK. 658 */ 659 if (tp->t_flags & TF_ACKNOW) 660 goto send; 661 if (flags & (TH_SYN|TH_FIN|TH_RST)) 662 goto send; 663 if (SEQ_GT(tp->snd_up, tp->snd_una)) 664 goto send; 665 666 /* 667 * TCP window updates are not reliable, rather a polling protocol 668 * using ``persist'' packets is used to insure receipt of window 669 * updates. The three ``states'' for the output side are: 670 * idle not doing retransmits or persists 671 * persisting to move a small or zero window 672 * (re)transmitting and thereby not persisting 673 * 674 * tp->t_timer[TCPT_PERSIST] 675 * is set when we are in persist state. 676 * tp->t_force 677 * is set when we are called to send a persist packet. 678 * tp->t_timer[TCPT_REXMT] 679 * is set when we are retransmitting 680 * The output side is idle when both timers are zero. 681 * 682 * If send window is too small, there is data to transmit, and no 683 * retransmit or persist is pending, then go to persist state. 684 * If nothing happens soon, send when timer expires: 685 * if window is nonzero, transmit what we can, 686 * otherwise force out a byte. 687 */ 688 if (so->so_snd.sb_cc && TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 689 TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 690 tp->t_rxtshift = 0; 691 tcp_setpersist(tp); 692 } 693 694 /* 695 * No reason to send a segment, just return. 696 */ 697 return (0); 698 699 send: 700 /* 701 * Before ESTABLISHED, force sending of initial options 702 * unless TCP set not to do any options. 703 * NOTE: we assume that the IP/TCP header plus TCP options 704 * always fit in a single mbuf, leaving room for a maximum 705 * link header, i.e. 706 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES 707 */ 708 optlen = 0; 709 switch (af) { 710 #ifdef INET 711 case AF_INET: 712 iphdrlen = sizeof(struct ip) + sizeof(struct tcphdr); 713 break; 714 #endif 715 #ifdef INET6 716 case AF_INET6: 717 iphdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 718 break; 719 #endif 720 default: /*pacify gcc*/ 721 iphdrlen = 0; 722 break; 723 } 724 hdrlen = iphdrlen; 725 if (flags & TH_SYN) { 726 struct rtentry *rt; 727 728 rt = NULL; 729 #ifdef INET 730 if (tp->t_inpcb) 731 rt = in_pcbrtentry(tp->t_inpcb); 732 #endif 733 #ifdef INET6 734 if (tp->t_in6pcb) 735 rt = in6_pcbrtentry(tp->t_in6pcb); 736 #endif 737 738 tp->snd_nxt = tp->iss; 739 tp->t_ourmss = tcp_mss_to_advertise(rt != NULL ? 740 rt->rt_ifp : NULL, af); 741 if ((tp->t_flags & TF_NOOPT) == 0) { 742 opt[0] = TCPOPT_MAXSEG; 743 opt[1] = 4; 744 opt[2] = (tp->t_ourmss >> 8) & 0xff; 745 opt[3] = tp->t_ourmss & 0xff; 746 optlen = 4; 747 748 if ((tp->t_flags & TF_REQ_SCALE) && 749 ((flags & TH_ACK) == 0 || 750 (tp->t_flags & TF_RCVD_SCALE))) { 751 *((u_int32_t *) (opt + optlen)) = htonl( 752 TCPOPT_NOP << 24 | 753 TCPOPT_WINDOW << 16 | 754 TCPOLEN_WINDOW << 8 | 755 tp->request_r_scale); 756 optlen += 4; 757 } 758 } 759 } 760 761 /* 762 * Send a timestamp and echo-reply if this is a SYN and our side 763 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side 764 * and our peer have sent timestamps in our SYN's. 765 */ 766 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 767 (flags & TH_RST) == 0 && 768 ((flags & (TH_SYN|TH_ACK)) == TH_SYN || 769 (tp->t_flags & TF_RCVD_TSTMP))) { 770 u_int32_t *lp = (u_int32_t *)(opt + optlen); 771 772 /* Form timestamp option as shown in appendix A of RFC 1323. */ 773 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 774 *lp++ = htonl(TCP_TIMESTAMP(tp)); 775 *lp = htonl(tp->ts_recent); 776 optlen += TCPOLEN_TSTAMP_APPA; 777 } 778 779 hdrlen += optlen; 780 781 #ifdef DIAGNOSTIC 782 if (len > txsegsize) 783 panic("tcp data to be sent is larger than segment"); 784 if (max_linkhdr + hdrlen > MCLBYTES) 785 panic("tcphdr too big"); 786 #endif 787 788 /* 789 * Grab a header mbuf, attaching a copy of data to 790 * be transmitted, and initialize the header from 791 * the template for sends on this connection. 792 */ 793 if (len) { 794 error = tcp_build_datapkt(tp, so, off, len, hdrlen, &m); 795 if (error) 796 goto out; 797 /* 798 * If we're sending everything we've got, set PUSH. 799 * (This will keep happy those implementations which only 800 * give data to the user when a buffer fills or 801 * a PUSH comes in.) 802 */ 803 if (off + len == so->so_snd.sb_cc) 804 flags |= TH_PUSH; 805 } else { 806 if (tp->t_flags & TF_ACKNOW) 807 tcpstat.tcps_sndacks++; 808 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 809 tcpstat.tcps_sndctrl++; 810 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 811 tcpstat.tcps_sndurg++; 812 else 813 tcpstat.tcps_sndwinup++; 814 815 MGETHDR(m, M_DONTWAIT, MT_HEADER); 816 if (m != NULL && max_linkhdr + hdrlen > MHLEN) { 817 MCLGET(m, M_DONTWAIT); 818 if ((m->m_flags & M_EXT) == 0) { 819 m_freem(m); 820 m = NULL; 821 } 822 } 823 if (m == NULL) { 824 error = ENOBUFS; 825 goto out; 826 } 827 m->m_data += max_linkhdr; 828 m->m_len = hdrlen; 829 } 830 m->m_pkthdr.rcvif = (struct ifnet *)0; 831 switch (af) { 832 #ifdef INET 833 case AF_INET: 834 ip = mtod(m, struct ip *); 835 #ifdef INET6 836 ip6 = NULL; 837 #endif 838 th = (struct tcphdr *)(ip + 1); 839 break; 840 #endif 841 #ifdef INET6 842 case AF_INET6: 843 ip = NULL; 844 ip6 = mtod(m, struct ip6_hdr *); 845 th = (struct tcphdr *)(ip6 + 1); 846 break; 847 #endif 848 default: /*pacify gcc*/ 849 ip = NULL; 850 #ifdef INET6 851 ip6 = NULL; 852 #endif 853 th = NULL; 854 break; 855 } 856 if (tp->t_template == 0) 857 panic("tcp_output"); 858 if (tp->t_template->m_len < iphdrlen) 859 panic("tcp_output"); 860 bcopy(mtod(tp->t_template, caddr_t), mtod(m, caddr_t), iphdrlen); 861 862 /* 863 * If we are doing retransmissions, then snd_nxt will 864 * not reflect the first unsent octet. For ACK only 865 * packets, we do not want the sequence number of the 866 * retransmitted packet, we want the sequence number 867 * of the next unsent octet. So, if there is no data 868 * (and no SYN or FIN), use snd_max instead of snd_nxt 869 * when filling in ti_seq. But if we are in persist 870 * state, snd_max might reflect one byte beyond the 871 * right edge of the window, so use snd_nxt in that 872 * case, since we know we aren't doing a retransmission. 873 * (retransmit and persist are mutually exclusive...) 874 */ 875 if (len || (flags & (TH_SYN|TH_FIN)) || 876 TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) 877 th->th_seq = htonl(tp->snd_nxt); 878 else 879 th->th_seq = htonl(tp->snd_max); 880 th->th_ack = htonl(tp->rcv_nxt); 881 if (optlen) { 882 bcopy((caddr_t)opt, (caddr_t)(th + 1), optlen); 883 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 884 } 885 th->th_flags = flags; 886 /* 887 * Calculate receive window. Don't shrink window, 888 * but avoid silly window syndrome. 889 */ 890 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)rxsegsize) 891 win = 0; 892 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 893 win = (long)TCP_MAXWIN << tp->rcv_scale; 894 if (win < (long)(tp->rcv_adv - tp->rcv_nxt)) 895 win = (long)(tp->rcv_adv - tp->rcv_nxt); 896 th->th_win = htons((u_int16_t) (win>>tp->rcv_scale)); 897 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 898 u_int32_t urp = tp->snd_up - tp->snd_nxt; 899 if (urp > IP_MAXPACKET) 900 urp = IP_MAXPACKET; 901 th->th_urp = htons((u_int16_t)urp); 902 th->th_flags |= TH_URG; 903 } else 904 /* 905 * If no urgent pointer to send, then we pull 906 * the urgent pointer to the left edge of the send window 907 * so that it doesn't drift into the send window on sequence 908 * number wraparound. 909 */ 910 tp->snd_up = tp->snd_una; /* drag it along */ 911 912 /* 913 * Set ourselves up to be checksummed just before the packet 914 * hits the wire. 915 */ 916 switch (af) { 917 #ifdef INET 918 case AF_INET: 919 m->m_pkthdr.csum_flags = M_CSUM_TCPv4; 920 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 921 if (len + optlen) { 922 /* Fixup the pseudo-header checksum. */ 923 /* XXXJRT Not IP Jumbogram safe. */ 924 th->th_sum = in_cksum_addword(th->th_sum, 925 htons((u_int16_t) (len + optlen))); 926 } 927 break; 928 #endif 929 #ifdef INET6 930 case AF_INET6: 931 /* 932 * XXX Actually delaying the checksum is Hard 933 * XXX (well, maybe not for Itojun, but it is 934 * XXX for me), but we can still take advantage 935 * XXX of the cached pseudo-header checksum. 936 */ 937 /* equals to hdrlen + len */ 938 m->m_pkthdr.len = sizeof(struct ip6_hdr) 939 + sizeof(struct tcphdr) + optlen + len; 940 #ifdef notyet 941 m->m_pkthdr.csum_flags = M_CSUM_TCPv6; 942 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 943 #endif 944 if (len + optlen) { 945 /* Fixup the pseudo-header checksum. */ 946 /* XXXJRT: Not IPv6 Jumbogram safe. */ 947 th->th_sum = in_cksum_addword(th->th_sum, 948 htons((u_int16_t) (len + optlen))); 949 } 950 #ifndef notyet 951 th->th_sum = in6_cksum(m, 0, sizeof(struct ip6_hdr), 952 sizeof(struct tcphdr) + optlen + len); 953 #endif 954 break; 955 #endif 956 } 957 958 /* 959 * In transmit state, time the transmission and arrange for 960 * the retransmit. In persist state, just set snd_max. 961 */ 962 if (tp->t_force == 0 || TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 963 tcp_seq startseq = tp->snd_nxt; 964 965 /* 966 * Advance snd_nxt over sequence space of this segment. 967 * There are no states in which we send both a SYN and a FIN, 968 * so we collapse the tests for these flags. 969 */ 970 if (flags & (TH_SYN|TH_FIN)) 971 tp->snd_nxt++; 972 tp->snd_nxt += len; 973 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 974 tp->snd_max = tp->snd_nxt; 975 /* 976 * Time this transmission if not a retransmission and 977 * not currently timing anything. 978 */ 979 if (tp->t_rtttime == 0) { 980 tp->t_rtttime = tcp_now; 981 tp->t_rtseq = startseq; 982 tcpstat.tcps_segstimed++; 983 } 984 } 985 986 /* 987 * Set retransmit timer if not currently set, 988 * and not doing an ack or a keep-alive probe. 989 * Initial value for retransmit timer is smoothed 990 * round-trip time + 2 * round-trip time variance. 991 * Initialize shift counter which is used for backoff 992 * of retransmit time. 993 */ 994 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 995 tp->snd_nxt != tp->snd_una) { 996 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 997 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) { 998 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 999 tp->t_rxtshift = 0; 1000 } 1001 } 1002 } else 1003 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max)) 1004 tp->snd_max = tp->snd_nxt + len; 1005 1006 #ifdef TCP_DEBUG 1007 /* 1008 * Trace. 1009 */ 1010 if (so->so_options & SO_DEBUG) { 1011 /* 1012 * need to recover version # field, which was overwritten 1013 * on ip_cksum computation. 1014 */ 1015 struct ip *sip; 1016 sip = mtod(m, struct ip *); 1017 switch (af) { 1018 #ifdef INET 1019 case AF_INET: 1020 sip->ip_v = 4; 1021 break; 1022 #endif 1023 #ifdef INET6 1024 case AF_INET6: 1025 sip->ip_v = 6; 1026 break; 1027 #endif 1028 } 1029 tcp_trace(TA_OUTPUT, tp->t_state, tp, m, 0); 1030 } 1031 #endif 1032 1033 /* 1034 * Fill in IP length and desired time to live and 1035 * send to IP level. There should be a better way 1036 * to handle ttl and tos; we could keep them in 1037 * the template, but need a way to checksum without them. 1038 */ 1039 m->m_pkthdr.len = hdrlen + len; 1040 1041 switch (af) { 1042 #ifdef INET 1043 case AF_INET: 1044 ip->ip_len = m->m_pkthdr.len; 1045 if (tp->t_inpcb) { 1046 ip->ip_ttl = tp->t_inpcb->inp_ip.ip_ttl; 1047 ip->ip_tos = tp->t_inpcb->inp_ip.ip_tos; 1048 } 1049 #ifdef INET6 1050 else if (tp->t_in6pcb) { 1051 ip->ip_ttl = in6_selecthlim(tp->t_in6pcb, NULL); /*XXX*/ 1052 ip->ip_tos = 0; /*XXX*/ 1053 } 1054 #endif 1055 break; 1056 #endif 1057 #ifdef INET6 1058 case AF_INET6: 1059 ip6->ip6_nxt = IPPROTO_TCP; 1060 if (tp->t_in6pcb) { 1061 /* 1062 * we separately set hoplimit for every segment, since 1063 * the user might want to change the value via 1064 * setsockopt. Also, desired default hop limit might 1065 * be changed via Neighbor Discovery. 1066 */ 1067 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, 1068 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 1069 } 1070 /* ip6->ip6_flow = ??? */ 1071 /* ip6_plen will be filled in ip6_output(). */ 1072 break; 1073 #endif 1074 } 1075 1076 #ifdef IPSEC 1077 if (ipsec_setsocket(m, so) != 0) { 1078 m_freem(m); 1079 error = ENOBUFS; 1080 goto out; 1081 } 1082 #endif /*IPSEC*/ 1083 1084 switch (af) { 1085 #ifdef INET 1086 case AF_INET: 1087 { 1088 struct mbuf *opts; 1089 1090 if (tp->t_inpcb) 1091 opts = tp->t_inpcb->inp_options; 1092 else 1093 opts = NULL; 1094 error = ip_output(m, opts, ro, 1095 (ip_mtudisc ? IP_MTUDISC : 0) | 1096 (so->so_options & SO_DONTROUTE), 1097 0); 1098 break; 1099 } 1100 #endif 1101 #ifdef INET6 1102 case AF_INET6: 1103 { 1104 struct ip6_pktopts *opts; 1105 1106 if (tp->t_in6pcb) 1107 opts = tp->t_in6pcb->in6p_outputopts; 1108 else 1109 opts = NULL; 1110 error = ip6_output(m, opts, (struct route_in6 *)ro, 1111 so->so_options & SO_DONTROUTE, 0, NULL); 1112 break; 1113 } 1114 #endif 1115 default: 1116 error = EAFNOSUPPORT; 1117 break; 1118 } 1119 if (error) { 1120 out: 1121 if (error == ENOBUFS) { 1122 tcpstat.tcps_selfquench++; 1123 #ifdef INET 1124 if (tp->t_inpcb) 1125 tcp_quench(tp->t_inpcb, 0); 1126 #endif 1127 #ifdef INET6 1128 if (tp->t_in6pcb) 1129 tcp6_quench(tp->t_in6pcb, 0); 1130 #endif 1131 error = 0; 1132 } else if ((error == EHOSTUNREACH || error == ENETDOWN) && 1133 TCPS_HAVERCVDSYN(tp->t_state)) { 1134 tp->t_softerror = error; 1135 error = 0; 1136 } 1137 1138 /* Restart the delayed ACK timer, if necessary. */ 1139 if (tp->t_flags & TF_DELACK) 1140 TCP_RESTART_DELACK(tp); 1141 1142 return (error); 1143 } 1144 tcpstat.tcps_sndtotal++; 1145 if (tp->t_flags & TF_DELACK) 1146 tcpstat.tcps_delack++; 1147 1148 /* 1149 * Data sent (as far as we can tell). 1150 * If this advertises a larger window than any other segment, 1151 * then remember the size of the advertised window. 1152 * Any pending ACK has now been sent. 1153 */ 1154 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 1155 tp->rcv_adv = tp->rcv_nxt + win; 1156 tp->last_ack_sent = tp->rcv_nxt; 1157 tp->t_flags &= ~TF_ACKNOW; 1158 TCP_CLEAR_DELACK(tp); 1159 #ifdef DIAGNOSTIC 1160 if (maxburst < 0) 1161 printf("tcp_output: maxburst exceeded by %d\n", -maxburst); 1162 #endif 1163 if (sendalot && (!tcp_do_newreno || --maxburst)) 1164 goto again; 1165 return (0); 1166 } 1167 1168 void 1169 tcp_setpersist(tp) 1170 struct tcpcb *tp; 1171 { 1172 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2); 1173 int nticks; 1174 1175 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT)) 1176 panic("tcp_output REXMT"); 1177 /* 1178 * Start/restart persistance timer. 1179 */ 1180 if (t < tp->t_rttmin) 1181 t = tp->t_rttmin; 1182 TCPT_RANGESET(nticks, t * tcp_backoff[tp->t_rxtshift], 1183 TCPTV_PERSMIN, TCPTV_PERSMAX); 1184 TCP_TIMER_ARM(tp, TCPT_PERSIST, nticks); 1185 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 1186 tp->t_rxtshift++; 1187 } 1188