1 /* $NetBSD: tcp_output.c,v 1.81 2002/05/29 07:53:39 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 1.81 2002/05/29 07:53:39 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_var.h> 177 #include <netinet6/ip6_var.h> 178 #include <netinet6/in6_pcb.h> 179 #include <netinet6/nd6.h> 180 #endif 181 182 #include <netinet/tcp.h> 183 #define TCPOUTFLAGS 184 #include <netinet/tcp_fsm.h> 185 #include <netinet/tcp_seq.h> 186 #include <netinet/tcp_timer.h> 187 #include <netinet/tcp_var.h> 188 #include <netinet/tcpip.h> 189 #include <netinet/tcp_debug.h> 190 191 #ifdef notyet 192 extern struct mbuf *m_copypack(); 193 #endif 194 195 #define MAX_TCPOPTLEN 32 /* max # bytes that go in options */ 196 197 /* 198 * Knob to enable Congestion Window Monitoring, and control the 199 * the burst size it allows. Default burst is 4 packets, per 200 * the Internet draft. 201 */ 202 int tcp_cwm = 1; 203 int tcp_cwm_burstsize = 4; 204 205 #ifdef TCP_OUTPUT_COUNTERS 206 #include <sys/device.h> 207 208 extern struct evcnt tcp_output_bigheader; 209 extern struct evcnt tcp_output_copysmall; 210 extern struct evcnt tcp_output_copybig; 211 extern struct evcnt tcp_output_refbig; 212 213 #define TCP_OUTPUT_COUNTER_INCR(ev) (ev)->ev_count++ 214 #else 215 216 #define TCP_OUTPUT_COUNTER_INCR(ev) /* nothing */ 217 218 #endif /* TCP_OUTPUT_COUNTERS */ 219 220 static 221 #ifndef GPROF 222 __inline 223 #endif 224 void 225 tcp_segsize(struct tcpcb *tp, int *txsegsizep, int *rxsegsizep) 226 { 227 #ifdef INET 228 struct inpcb *inp = tp->t_inpcb; 229 #endif 230 #ifdef INET6 231 struct in6pcb *in6p = tp->t_in6pcb; 232 #endif 233 struct rtentry *rt; 234 struct ifnet *ifp; 235 int size; 236 int iphlen; 237 int optlen; 238 239 #ifdef DIAGNOSTIC 240 if (tp->t_inpcb && tp->t_in6pcb) 241 panic("tcp_segsize: both t_inpcb and t_in6pcb are set"); 242 #endif 243 switch (tp->t_family) { 244 #ifdef INET 245 case AF_INET: 246 iphlen = sizeof(struct ip); 247 break; 248 #endif 249 #ifdef INET6 250 case AF_INET6: 251 iphlen = sizeof(struct ip6_hdr); 252 break; 253 #endif 254 default: 255 size = tcp_mssdflt; 256 goto out; 257 } 258 259 rt = NULL; 260 #ifdef INET 261 if (inp) 262 rt = in_pcbrtentry(inp); 263 #endif 264 #ifdef INET6 265 if (in6p) 266 rt = in6_pcbrtentry(in6p); 267 #endif 268 if (rt == NULL) { 269 size = tcp_mssdflt; 270 goto out; 271 } 272 273 ifp = rt->rt_ifp; 274 275 size = tcp_mssdflt; 276 if (tp->t_mtudisc && rt->rt_rmx.rmx_mtu != 0) 277 size = rt->rt_rmx.rmx_mtu - iphlen - sizeof(struct tcphdr); 278 else if (ifp->if_flags & IFF_LOOPBACK) 279 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 280 #ifdef INET 281 else if (inp && tp->t_mtudisc) 282 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 283 else if (inp && in_localaddr(inp->inp_faddr)) 284 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 285 #endif 286 #ifdef INET6 287 else if (in6p) { 288 #ifdef INET 289 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) { 290 /* mapped addr case */ 291 struct in_addr d; 292 bcopy(&in6p->in6p_faddr.s6_addr32[3], &d, sizeof(d)); 293 if (tp->t_mtudisc || in_localaddr(d)) 294 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 295 } else 296 #endif 297 { 298 /* 299 * for IPv6, path MTU discovery is always turned on, 300 * or the node must use packet size <= 1280. 301 */ 302 size = tp->t_mtudisc ? IN6_LINKMTU(ifp) : IPV6_MMTU; 303 size -= (iphlen + sizeof(struct tcphdr)); 304 } 305 } 306 #endif 307 out: 308 /* 309 * Now we must make room for whatever extra TCP/IP options are in 310 * the packet. 311 */ 312 optlen = tcp_optlen(tp); 313 314 /* 315 * XXX tp->t_ourmss should have the right size, but without this code 316 * fragmentation will occur... need more investigation 317 */ 318 #ifdef INET 319 if (inp) { 320 #ifdef IPSEC 321 optlen += ipsec4_hdrsiz_tcp(tp); 322 #endif 323 optlen += ip_optlen(inp); 324 } 325 #endif 326 #ifdef INET6 327 #ifdef INET 328 if (in6p && tp->t_family == AF_INET) { 329 #ifdef IPSEC 330 optlen += ipsec4_hdrsiz_tcp(tp); 331 #endif 332 /* XXX size -= ip_optlen(in6p); */ 333 } else 334 #endif 335 if (in6p && tp->t_family == AF_INET6) { 336 #ifdef IPSEC 337 optlen += ipsec6_hdrsiz_tcp(tp); 338 #endif 339 optlen += ip6_optlen(in6p); 340 } 341 #endif 342 size -= optlen; 343 344 /* 345 * *rxsegsizep holds *estimated* inbound segment size (estimation 346 * assumes that path MTU is the same for both ways). this is only 347 * for silly window avoidance, do not use the value for other purposes. 348 * 349 * ipseclen is subtracted from both sides, this may not be right. 350 * I'm not quite sure about this (could someone comment). 351 */ 352 *txsegsizep = min(tp->t_peermss - optlen, size); 353 *rxsegsizep = min(tp->t_ourmss - optlen, size); 354 355 if (*txsegsizep != tp->t_segsz) { 356 /* 357 * If the new segment size is larger, we don't want to 358 * mess up the congestion window, but if it is smaller 359 * we'll have to reduce the congestion window to ensure 360 * that we don't get into trouble with initial windows 361 * and the rest. In any case, if the segment size 362 * has changed, chances are the path has, too, and 363 * our congestion window will be different. 364 */ 365 if (*txsegsizep < tp->t_segsz) { 366 tp->snd_cwnd = max((tp->snd_cwnd / tp->t_segsz) 367 * *txsegsizep, *txsegsizep); 368 tp->snd_ssthresh = max((tp->snd_ssthresh / tp->t_segsz) 369 * *txsegsizep, *txsegsizep); 370 } 371 tp->t_segsz = *txsegsizep; 372 } 373 } 374 375 static 376 #ifndef GPROF 377 __inline 378 #endif 379 int 380 tcp_build_datapkt(struct tcpcb *tp, struct socket *so, int off, 381 long len, int hdrlen, struct mbuf **mp) 382 { 383 struct mbuf *m; 384 385 if (tp->t_force && len == 1) 386 tcpstat.tcps_sndprobe++; 387 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { 388 tcpstat.tcps_sndrexmitpack++; 389 tcpstat.tcps_sndrexmitbyte += len; 390 } else { 391 tcpstat.tcps_sndpack++; 392 tcpstat.tcps_sndbyte += len; 393 } 394 #ifdef notyet 395 if ((m = m_copypack(so->so_snd.sb_mb, off, 396 (int)len, max_linkhdr + hdrlen)) == 0) 397 return (ENOBUFS); 398 /* 399 * m_copypack left space for our hdr; use it. 400 */ 401 m->m_len += hdrlen; 402 m->m_data -= hdrlen; 403 #else 404 MGETHDR(m, M_DONTWAIT, MT_HEADER); 405 if (__predict_false(m == NULL)) 406 return (ENOBUFS); 407 408 /* 409 * XXX Because other code assumes headers will fit in 410 * XXX one header mbuf. 411 * 412 * (This code should almost *never* be run.) 413 */ 414 if (__predict_false((max_linkhdr + hdrlen) > MHLEN)) { 415 TCP_OUTPUT_COUNTER_INCR(&tcp_output_bigheader); 416 MCLGET(m, M_DONTWAIT); 417 if ((m->m_flags & M_EXT) == 0) { 418 m_freem(m); 419 return (ENOBUFS); 420 } 421 } 422 423 m->m_data += max_linkhdr; 424 m->m_len = hdrlen; 425 if (len <= M_TRAILINGSPACE(m)) { 426 m_copydata(so->so_snd.sb_mb, off, (int) len, 427 mtod(m, caddr_t) + hdrlen); 428 m->m_len += len; 429 TCP_OUTPUT_COUNTER_INCR(&tcp_output_copysmall); 430 } else { 431 m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len); 432 if (m->m_next == NULL) { 433 m_freem(m); 434 return (ENOBUFS); 435 } 436 #ifdef TCP_OUTPUT_COUNTERS 437 if (m->m_next->m_flags & M_EXT) 438 TCP_OUTPUT_COUNTER_INCR(&tcp_output_refbig); 439 else 440 TCP_OUTPUT_COUNTER_INCR(&tcp_output_copybig); 441 #endif /* TCP_OUTPUT_COUNTERS */ 442 } 443 #endif 444 445 *mp = m; 446 return (0); 447 } 448 449 /* 450 * Tcp output routine: figure out what should be sent and send it. 451 */ 452 int 453 tcp_output(tp) 454 struct tcpcb *tp; 455 { 456 struct socket *so; 457 struct route *ro; 458 long len, win; 459 int off, flags, error; 460 struct mbuf *m; 461 struct ip *ip; 462 #ifdef INET6 463 struct ip6_hdr *ip6; 464 #endif 465 struct tcphdr *th; 466 u_char opt[MAX_TCPOPTLEN]; 467 unsigned optlen, hdrlen; 468 int idle, sendalot, txsegsize, rxsegsize; 469 int maxburst = TCP_MAXBURST; 470 int af; /* address family on the wire */ 471 int iphdrlen; 472 473 #ifdef DIAGNOSTIC 474 if (tp->t_inpcb && tp->t_in6pcb) 475 panic("tcp_output: both t_inpcb and t_in6pcb are set"); 476 #endif 477 so = NULL; 478 ro = NULL; 479 if (tp->t_inpcb) { 480 so = tp->t_inpcb->inp_socket; 481 ro = &tp->t_inpcb->inp_route; 482 } 483 #ifdef INET6 484 else if (tp->t_in6pcb) { 485 so = tp->t_in6pcb->in6p_socket; 486 ro = (struct route *)&tp->t_in6pcb->in6p_route; 487 } 488 #endif 489 490 switch (af = tp->t_family) { 491 #ifdef INET 492 case AF_INET: 493 if (tp->t_inpcb) 494 break; 495 #ifdef INET6 496 /* mapped addr case */ 497 if (tp->t_in6pcb) 498 break; 499 #endif 500 return EINVAL; 501 #endif 502 #ifdef INET6 503 case AF_INET6: 504 if (tp->t_in6pcb) 505 break; 506 return EINVAL; 507 #endif 508 default: 509 return EAFNOSUPPORT; 510 } 511 512 tcp_segsize(tp, &txsegsize, &rxsegsize); 513 514 idle = (tp->snd_max == tp->snd_una); 515 516 /* 517 * Restart Window computation. From draft-floyd-incr-init-win-03: 518 * 519 * Optionally, a TCP MAY set the restart window to the 520 * minimum of the value used for the initial window and 521 * the current value of cwnd (in other words, using a 522 * larger value for the restart window should never increase 523 * the size of cwnd). 524 */ 525 if (tcp_cwm) { 526 /* 527 * Hughes/Touch/Heidemann Congestion Window Monitoring. 528 * Count the number of packets currently pending 529 * acknowledgement, and limit our congestion window 530 * to a pre-determined allowed burst size plus that count. 531 * This prevents bursting once all pending packets have 532 * been acknowledged (i.e. transmission is idle). 533 * 534 * XXX Link this to Initial Window? 535 */ 536 tp->snd_cwnd = min(tp->snd_cwnd, 537 (tcp_cwm_burstsize * txsegsize) + 538 (tp->snd_nxt - tp->snd_una)); 539 } else { 540 if (idle && (tcp_now - tp->t_rcvtime) >= tp->t_rxtcur) { 541 /* 542 * We have been idle for "a while" and no acks are 543 * expected to clock out any data we send -- 544 * slow start to get ack "clock" running again. 545 */ 546 tp->snd_cwnd = min(tp->snd_cwnd, 547 TCP_INITIAL_WINDOW(tcp_init_win, txsegsize)); 548 } 549 } 550 551 again: 552 /* 553 * Determine length of data that should be transmitted, and 554 * flags that should be used. If there is some data or critical 555 * controls (SYN, RST) to send, then transmit; otherwise, 556 * investigate further. 557 */ 558 sendalot = 0; 559 off = tp->snd_nxt - tp->snd_una; 560 win = min(tp->snd_wnd, tp->snd_cwnd); 561 562 flags = tcp_outflags[tp->t_state]; 563 /* 564 * If in persist timeout with window of 0, send 1 byte. 565 * Otherwise, if window is small but nonzero 566 * and timer expired, we will send what we can 567 * and go to transmit state. 568 */ 569 if (tp->t_force) { 570 if (win == 0) { 571 /* 572 * If we still have some data to send, then 573 * clear the FIN bit. Usually this would 574 * happen below when it realizes that we 575 * aren't sending all the data. However, 576 * if we have exactly 1 byte of unset data, 577 * then it won't clear the FIN bit below, 578 * and if we are in persist state, we wind 579 * up sending the packet without recording 580 * that we sent the FIN bit. 581 * 582 * We can't just blindly clear the FIN bit, 583 * because if we don't have any more data 584 * to send then the probe will be the FIN 585 * itself. 586 */ 587 if (off < so->so_snd.sb_cc) 588 flags &= ~TH_FIN; 589 win = 1; 590 } else { 591 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 592 tp->t_rxtshift = 0; 593 } 594 } 595 596 if (win < so->so_snd.sb_cc) { 597 len = win - off; 598 flags &= ~TH_FIN; 599 } else 600 len = so->so_snd.sb_cc - off; 601 602 if (len < 0) { 603 /* 604 * If FIN has been sent but not acked, 605 * but we haven't been called to retransmit, 606 * len will be -1. Otherwise, window shrank 607 * after we sent into it. If window shrank to 0, 608 * cancel pending retransmit, pull snd_nxt back 609 * to (closed) window, and set the persist timer 610 * if it isn't already going. If the window didn't 611 * close completely, just wait for an ACK. 612 * 613 * If we have a pending FIN, either it has already been 614 * transmitted or it is outside the window, so drop it. 615 * If the FIN has been transmitted, but this is not a 616 * retransmission, then len must be -1. Therefore we also 617 * prevent here the sending of `gratuitous FINs'. This 618 * eliminates the need to check for that case below (e.g. 619 * to back up snd_nxt before the FIN so that the sequence 620 * number is correct). 621 */ 622 len = 0; 623 flags &= ~TH_FIN; 624 if (win == 0) { 625 TCP_TIMER_DISARM(tp, TCPT_REXMT); 626 tp->t_rxtshift = 0; 627 tp->snd_nxt = tp->snd_una; 628 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 629 tcp_setpersist(tp); 630 } 631 } 632 if (len > txsegsize) { 633 len = txsegsize; 634 flags &= ~TH_FIN; 635 sendalot = 1; 636 } 637 638 win = sbspace(&so->so_rcv); 639 640 /* 641 * Sender silly window avoidance. If connection is idle 642 * and can send all data, a maximum segment, 643 * at least a maximum default-size segment do it, 644 * or are forced, do it; otherwise don't bother. 645 * If peer's buffer is tiny, then send 646 * when window is at least half open. 647 * If retransmitting (possibly after persist timer forced us 648 * to send into a small window), then must resend. 649 */ 650 if (len) { 651 if (len == txsegsize) 652 goto send; 653 if ((so->so_state & SS_MORETOCOME) == 0 && 654 ((idle || tp->t_flags & TF_NODELAY) && 655 len + off >= so->so_snd.sb_cc)) 656 goto send; 657 if (tp->t_force) 658 goto send; 659 if (len >= tp->max_sndwnd / 2) 660 goto send; 661 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) 662 goto send; 663 } 664 665 /* 666 * Compare available window to amount of window known to peer 667 * (as advertised window less next expected input). If the 668 * difference is at least twice the size of the largest segment 669 * we expect to receive (i.e. two segments) or at least 50% of 670 * the maximum possible window, then want to send a window update 671 * to peer. 672 */ 673 if (win > 0) { 674 /* 675 * "adv" is the amount we can increase the window, 676 * taking into account that we are limited by 677 * TCP_MAXWIN << tp->rcv_scale. 678 */ 679 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) - 680 (tp->rcv_adv - tp->rcv_nxt); 681 682 if (adv >= (long) (2 * rxsegsize)) 683 goto send; 684 if (2 * adv >= (long) so->so_rcv.sb_hiwat) 685 goto send; 686 } 687 688 /* 689 * Send if we owe peer an ACK. 690 */ 691 if (tp->t_flags & TF_ACKNOW) 692 goto send; 693 if (flags & (TH_SYN|TH_FIN|TH_RST)) 694 goto send; 695 if (SEQ_GT(tp->snd_up, tp->snd_una)) 696 goto send; 697 698 /* 699 * TCP window updates are not reliable, rather a polling protocol 700 * using ``persist'' packets is used to insure receipt of window 701 * updates. The three ``states'' for the output side are: 702 * idle not doing retransmits or persists 703 * persisting to move a small or zero window 704 * (re)transmitting and thereby not persisting 705 * 706 * tp->t_timer[TCPT_PERSIST] 707 * is set when we are in persist state. 708 * tp->t_force 709 * is set when we are called to send a persist packet. 710 * tp->t_timer[TCPT_REXMT] 711 * is set when we are retransmitting 712 * The output side is idle when both timers are zero. 713 * 714 * If send window is too small, there is data to transmit, and no 715 * retransmit or persist is pending, then go to persist state. 716 * If nothing happens soon, send when timer expires: 717 * if window is nonzero, transmit what we can, 718 * otherwise force out a byte. 719 */ 720 if (so->so_snd.sb_cc && TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 721 TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 722 tp->t_rxtshift = 0; 723 tcp_setpersist(tp); 724 } 725 726 /* 727 * No reason to send a segment, just return. 728 */ 729 return (0); 730 731 send: 732 /* 733 * Before ESTABLISHED, force sending of initial options 734 * unless TCP set not to do any options. 735 * NOTE: we assume that the IP/TCP header plus TCP options 736 * always fit in a single mbuf, leaving room for a maximum 737 * link header, i.e. 738 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES 739 */ 740 optlen = 0; 741 switch (af) { 742 #ifdef INET 743 case AF_INET: 744 iphdrlen = sizeof(struct ip) + sizeof(struct tcphdr); 745 break; 746 #endif 747 #ifdef INET6 748 case AF_INET6: 749 iphdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 750 break; 751 #endif 752 default: /*pacify gcc*/ 753 iphdrlen = 0; 754 break; 755 } 756 hdrlen = iphdrlen; 757 if (flags & TH_SYN) { 758 struct rtentry *rt; 759 760 rt = NULL; 761 #ifdef INET 762 if (tp->t_inpcb) 763 rt = in_pcbrtentry(tp->t_inpcb); 764 #endif 765 #ifdef INET6 766 if (tp->t_in6pcb) 767 rt = in6_pcbrtentry(tp->t_in6pcb); 768 #endif 769 770 tp->snd_nxt = tp->iss; 771 tp->t_ourmss = tcp_mss_to_advertise(rt != NULL ? 772 rt->rt_ifp : NULL, af); 773 if ((tp->t_flags & TF_NOOPT) == 0) { 774 opt[0] = TCPOPT_MAXSEG; 775 opt[1] = 4; 776 opt[2] = (tp->t_ourmss >> 8) & 0xff; 777 opt[3] = tp->t_ourmss & 0xff; 778 optlen = 4; 779 780 if ((tp->t_flags & TF_REQ_SCALE) && 781 ((flags & TH_ACK) == 0 || 782 (tp->t_flags & TF_RCVD_SCALE))) { 783 *((u_int32_t *) (opt + optlen)) = htonl( 784 TCPOPT_NOP << 24 | 785 TCPOPT_WINDOW << 16 | 786 TCPOLEN_WINDOW << 8 | 787 tp->request_r_scale); 788 optlen += 4; 789 } 790 } 791 } 792 793 /* 794 * Send a timestamp and echo-reply if this is a SYN and our side 795 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side 796 * and our peer have sent timestamps in our SYN's. 797 */ 798 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 799 (flags & TH_RST) == 0 && 800 ((flags & (TH_SYN|TH_ACK)) == TH_SYN || 801 (tp->t_flags & TF_RCVD_TSTMP))) { 802 u_int32_t *lp = (u_int32_t *)(opt + optlen); 803 804 /* Form timestamp option as shown in appendix A of RFC 1323. */ 805 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 806 *lp++ = htonl(TCP_TIMESTAMP(tp)); 807 *lp = htonl(tp->ts_recent); 808 optlen += TCPOLEN_TSTAMP_APPA; 809 } 810 811 hdrlen += optlen; 812 813 #ifdef DIAGNOSTIC 814 if (len > txsegsize) 815 panic("tcp data to be sent is larger than segment"); 816 if (max_linkhdr + hdrlen > MCLBYTES) 817 panic("tcphdr too big"); 818 #endif 819 820 /* 821 * Grab a header mbuf, attaching a copy of data to 822 * be transmitted, and initialize the header from 823 * the template for sends on this connection. 824 */ 825 if (len) { 826 error = tcp_build_datapkt(tp, so, off, len, hdrlen, &m); 827 if (error) 828 goto out; 829 /* 830 * If we're sending everything we've got, set PUSH. 831 * (This will keep happy those implementations which only 832 * give data to the user when a buffer fills or 833 * a PUSH comes in.) 834 */ 835 if (off + len == so->so_snd.sb_cc) 836 flags |= TH_PUSH; 837 } else { 838 if (tp->t_flags & TF_ACKNOW) 839 tcpstat.tcps_sndacks++; 840 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 841 tcpstat.tcps_sndctrl++; 842 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 843 tcpstat.tcps_sndurg++; 844 else 845 tcpstat.tcps_sndwinup++; 846 847 MGETHDR(m, M_DONTWAIT, MT_HEADER); 848 if (m != NULL && max_linkhdr + hdrlen > MHLEN) { 849 MCLGET(m, M_DONTWAIT); 850 if ((m->m_flags & M_EXT) == 0) { 851 m_freem(m); 852 m = NULL; 853 } 854 } 855 if (m == NULL) { 856 error = ENOBUFS; 857 goto out; 858 } 859 m->m_data += max_linkhdr; 860 m->m_len = hdrlen; 861 } 862 m->m_pkthdr.rcvif = (struct ifnet *)0; 863 switch (af) { 864 #ifdef INET 865 case AF_INET: 866 ip = mtod(m, struct ip *); 867 #ifdef INET6 868 ip6 = NULL; 869 #endif 870 th = (struct tcphdr *)(ip + 1); 871 break; 872 #endif 873 #ifdef INET6 874 case AF_INET6: 875 ip = NULL; 876 ip6 = mtod(m, struct ip6_hdr *); 877 th = (struct tcphdr *)(ip6 + 1); 878 break; 879 #endif 880 default: /*pacify gcc*/ 881 ip = NULL; 882 #ifdef INET6 883 ip6 = NULL; 884 #endif 885 th = NULL; 886 break; 887 } 888 if (tp->t_template == 0) 889 panic("tcp_output"); 890 if (tp->t_template->m_len < iphdrlen) 891 panic("tcp_output"); 892 bcopy(mtod(tp->t_template, caddr_t), mtod(m, caddr_t), iphdrlen); 893 894 /* 895 * If we are doing retransmissions, then snd_nxt will 896 * not reflect the first unsent octet. For ACK only 897 * packets, we do not want the sequence number of the 898 * retransmitted packet, we want the sequence number 899 * of the next unsent octet. So, if there is no data 900 * (and no SYN or FIN), use snd_max instead of snd_nxt 901 * when filling in ti_seq. But if we are in persist 902 * state, snd_max might reflect one byte beyond the 903 * right edge of the window, so use snd_nxt in that 904 * case, since we know we aren't doing a retransmission. 905 * (retransmit and persist are mutually exclusive...) 906 */ 907 if (len || (flags & (TH_SYN|TH_FIN)) || 908 TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) 909 th->th_seq = htonl(tp->snd_nxt); 910 else 911 th->th_seq = htonl(tp->snd_max); 912 th->th_ack = htonl(tp->rcv_nxt); 913 if (optlen) { 914 bcopy((caddr_t)opt, (caddr_t)(th + 1), optlen); 915 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 916 } 917 th->th_flags = flags; 918 /* 919 * Calculate receive window. Don't shrink window, 920 * but avoid silly window syndrome. 921 */ 922 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)rxsegsize) 923 win = 0; 924 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 925 win = (long)TCP_MAXWIN << tp->rcv_scale; 926 if (win < (long)(tp->rcv_adv - tp->rcv_nxt)) 927 win = (long)(tp->rcv_adv - tp->rcv_nxt); 928 th->th_win = htons((u_int16_t) (win>>tp->rcv_scale)); 929 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 930 u_int32_t urp = tp->snd_up - tp->snd_nxt; 931 if (urp > IP_MAXPACKET) 932 urp = IP_MAXPACKET; 933 th->th_urp = htons((u_int16_t)urp); 934 th->th_flags |= TH_URG; 935 } else 936 /* 937 * If no urgent pointer to send, then we pull 938 * the urgent pointer to the left edge of the send window 939 * so that it doesn't drift into the send window on sequence 940 * number wraparound. 941 */ 942 tp->snd_up = tp->snd_una; /* drag it along */ 943 944 /* 945 * Set ourselves up to be checksummed just before the packet 946 * hits the wire. 947 */ 948 switch (af) { 949 #ifdef INET 950 case AF_INET: 951 m->m_pkthdr.csum_flags = M_CSUM_TCPv4; 952 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 953 if (len + optlen) { 954 /* Fixup the pseudo-header checksum. */ 955 /* XXXJRT Not IP Jumbogram safe. */ 956 th->th_sum = in_cksum_addword(th->th_sum, 957 htons((u_int16_t) (len + optlen))); 958 } 959 break; 960 #endif 961 #ifdef INET6 962 case AF_INET6: 963 /* 964 * XXX Actually delaying the checksum is Hard 965 * XXX (well, maybe not for Itojun, but it is 966 * XXX for me), but we can still take advantage 967 * XXX of the cached pseudo-header checksum. 968 */ 969 /* equals to hdrlen + len */ 970 m->m_pkthdr.len = sizeof(struct ip6_hdr) 971 + sizeof(struct tcphdr) + optlen + len; 972 #ifdef notyet 973 m->m_pkthdr.csum_flags = M_CSUM_TCPv6; 974 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 975 #endif 976 if (len + optlen) { 977 /* Fixup the pseudo-header checksum. */ 978 /* XXXJRT: Not IPv6 Jumbogram safe. */ 979 th->th_sum = in_cksum_addword(th->th_sum, 980 htons((u_int16_t) (len + optlen))); 981 } 982 #ifndef notyet 983 th->th_sum = in6_cksum(m, 0, sizeof(struct ip6_hdr), 984 sizeof(struct tcphdr) + optlen + len); 985 #endif 986 break; 987 #endif 988 } 989 990 /* 991 * In transmit state, time the transmission and arrange for 992 * the retransmit. In persist state, just set snd_max. 993 */ 994 if (tp->t_force == 0 || TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 995 tcp_seq startseq = tp->snd_nxt; 996 997 /* 998 * Advance snd_nxt over sequence space of this segment. 999 * There are no states in which we send both a SYN and a FIN, 1000 * so we collapse the tests for these flags. 1001 */ 1002 if (flags & (TH_SYN|TH_FIN)) 1003 tp->snd_nxt++; 1004 tp->snd_nxt += len; 1005 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 1006 tp->snd_max = tp->snd_nxt; 1007 /* 1008 * Time this transmission if not a retransmission and 1009 * not currently timing anything. 1010 */ 1011 if (tp->t_rtttime == 0) { 1012 tp->t_rtttime = tcp_now; 1013 tp->t_rtseq = startseq; 1014 tcpstat.tcps_segstimed++; 1015 } 1016 } 1017 1018 /* 1019 * Set retransmit timer if not currently set, 1020 * and not doing an ack or a keep-alive probe. 1021 * Initial value for retransmit timer is smoothed 1022 * round-trip time + 2 * round-trip time variance. 1023 * Initialize shift counter which is used for backoff 1024 * of retransmit time. 1025 */ 1026 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 1027 tp->snd_nxt != tp->snd_una) { 1028 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1029 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) { 1030 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 1031 tp->t_rxtshift = 0; 1032 } 1033 } 1034 } else 1035 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max)) 1036 tp->snd_max = tp->snd_nxt + len; 1037 1038 #ifdef TCP_DEBUG 1039 /* 1040 * Trace. 1041 */ 1042 if (so->so_options & SO_DEBUG) { 1043 /* 1044 * need to recover version # field, which was overwritten 1045 * on ip_cksum computation. 1046 */ 1047 struct ip *sip; 1048 sip = mtod(m, struct ip *); 1049 switch (af) { 1050 #ifdef INET 1051 case AF_INET: 1052 sip->ip_v = 4; 1053 break; 1054 #endif 1055 #ifdef INET6 1056 case AF_INET6: 1057 sip->ip_v = 6; 1058 break; 1059 #endif 1060 } 1061 tcp_trace(TA_OUTPUT, tp->t_state, tp, m, 0); 1062 } 1063 #endif 1064 1065 /* 1066 * Fill in IP length and desired time to live and 1067 * send to IP level. There should be a better way 1068 * to handle ttl and tos; we could keep them in 1069 * the template, but need a way to checksum without them. 1070 */ 1071 m->m_pkthdr.len = hdrlen + len; 1072 1073 switch (af) { 1074 #ifdef INET 1075 case AF_INET: 1076 ip->ip_len = m->m_pkthdr.len; 1077 if (tp->t_inpcb) { 1078 ip->ip_ttl = tp->t_inpcb->inp_ip.ip_ttl; 1079 ip->ip_tos = tp->t_inpcb->inp_ip.ip_tos; 1080 } 1081 #ifdef INET6 1082 else if (tp->t_in6pcb) { 1083 ip->ip_ttl = in6_selecthlim(tp->t_in6pcb, NULL); /*XXX*/ 1084 ip->ip_tos = 0; /*XXX*/ 1085 } 1086 #endif 1087 break; 1088 #endif 1089 #ifdef INET6 1090 case AF_INET6: 1091 ip6->ip6_nxt = IPPROTO_TCP; 1092 if (tp->t_in6pcb) { 1093 /* 1094 * we separately set hoplimit for every segment, since 1095 * the user might want to change the value via 1096 * setsockopt. Also, desired default hop limit might 1097 * be changed via Neighbor Discovery. 1098 */ 1099 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, 1100 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 1101 } 1102 /* ip6->ip6_flow = ??? */ 1103 /* ip6_plen will be filled in ip6_output(). */ 1104 break; 1105 #endif 1106 } 1107 1108 #ifdef IPSEC 1109 if (ipsec_setsocket(m, so) != 0) { 1110 m_freem(m); 1111 error = ENOBUFS; 1112 goto out; 1113 } 1114 #endif /*IPSEC*/ 1115 1116 switch (af) { 1117 #ifdef INET 1118 case AF_INET: 1119 { 1120 struct mbuf *opts; 1121 1122 if (tp->t_inpcb) 1123 opts = tp->t_inpcb->inp_options; 1124 else 1125 opts = NULL; 1126 error = ip_output(m, opts, ro, 1127 (tp->t_mtudisc ? IP_MTUDISC : 0) | 1128 (so->so_options & SO_DONTROUTE), 1129 0); 1130 break; 1131 } 1132 #endif 1133 #ifdef INET6 1134 case AF_INET6: 1135 { 1136 struct ip6_pktopts *opts; 1137 1138 if (tp->t_in6pcb) 1139 opts = tp->t_in6pcb->in6p_outputopts; 1140 else 1141 opts = NULL; 1142 error = ip6_output(m, opts, (struct route_in6 *)ro, 1143 so->so_options & SO_DONTROUTE, 0, NULL); 1144 break; 1145 } 1146 #endif 1147 default: 1148 error = EAFNOSUPPORT; 1149 break; 1150 } 1151 if (error) { 1152 out: 1153 if (error == ENOBUFS) { 1154 tcpstat.tcps_selfquench++; 1155 #ifdef INET 1156 if (tp->t_inpcb) 1157 tcp_quench(tp->t_inpcb, 0); 1158 #endif 1159 #ifdef INET6 1160 if (tp->t_in6pcb) 1161 tcp6_quench(tp->t_in6pcb, 0); 1162 #endif 1163 error = 0; 1164 } else if ((error == EHOSTUNREACH || error == ENETDOWN) && 1165 TCPS_HAVERCVDSYN(tp->t_state)) { 1166 tp->t_softerror = error; 1167 error = 0; 1168 } 1169 1170 /* Restart the delayed ACK timer, if necessary. */ 1171 if (tp->t_flags & TF_DELACK) 1172 TCP_RESTART_DELACK(tp); 1173 1174 return (error); 1175 } 1176 tcpstat.tcps_sndtotal++; 1177 if (tp->t_flags & TF_DELACK) 1178 tcpstat.tcps_delack++; 1179 1180 /* 1181 * Data sent (as far as we can tell). 1182 * If this advertises a larger window than any other segment, 1183 * then remember the size of the advertised window. 1184 * Any pending ACK has now been sent. 1185 */ 1186 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 1187 tp->rcv_adv = tp->rcv_nxt + win; 1188 tp->last_ack_sent = tp->rcv_nxt; 1189 tp->t_flags &= ~TF_ACKNOW; 1190 TCP_CLEAR_DELACK(tp); 1191 #ifdef DIAGNOSTIC 1192 if (maxburst < 0) 1193 printf("tcp_output: maxburst exceeded by %d\n", -maxburst); 1194 #endif 1195 if (sendalot && (!tcp_do_newreno || --maxburst)) 1196 goto again; 1197 return (0); 1198 } 1199 1200 void 1201 tcp_setpersist(tp) 1202 struct tcpcb *tp; 1203 { 1204 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2); 1205 int nticks; 1206 1207 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT)) 1208 panic("tcp_output REXMT"); 1209 /* 1210 * Start/restart persistance timer. 1211 */ 1212 if (t < tp->t_rttmin) 1213 t = tp->t_rttmin; 1214 TCPT_RANGESET(nticks, t * tcp_backoff[tp->t_rxtshift], 1215 TCPTV_PERSMIN, TCPTV_PERSMAX); 1216 TCP_TIMER_ARM(tp, TCPT_PERSIST, nticks); 1217 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 1218 tp->t_rxtshift++; 1219 } 1220