tcp_output.c revision 1.110
1 /* $NetBSD: tcp_output.c,v 1.110 2004/04/25 22:25:03 jonathan 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. Neither the name of the University nor the names of its contributors 122 * may be used to endorse or promote products derived from this software 123 * without specific prior written permission. 124 * 125 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 126 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 127 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 128 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 129 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 130 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 131 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 132 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 133 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 134 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 135 * SUCH DAMAGE. 136 * 137 * @(#)tcp_output.c 8.4 (Berkeley) 5/24/95 138 */ 139 140 #include <sys/cdefs.h> 141 __KERNEL_RCSID(0, "$NetBSD: tcp_output.c,v 1.110 2004/04/25 22:25:03 jonathan Exp $"); 142 143 #include "opt_inet.h" 144 #include "opt_ipsec.h" 145 #include "opt_tcp_debug.h" 146 147 #include <sys/param.h> 148 #include <sys/systm.h> 149 #include <sys/malloc.h> 150 #include <sys/mbuf.h> 151 #include <sys/protosw.h> 152 #include <sys/socket.h> 153 #include <sys/socketvar.h> 154 #include <sys/errno.h> 155 #include <sys/domain.h> 156 #include <sys/kernel.h> 157 158 #include <net/if.h> 159 #include <net/route.h> 160 161 #include <netinet/in.h> 162 #include <netinet/in_systm.h> 163 #include <netinet/ip.h> 164 #include <netinet/in_pcb.h> 165 #include <netinet/ip_var.h> 166 167 #ifdef INET6 168 #ifndef INET 169 #include <netinet/in.h> 170 #endif 171 #include <netinet/ip6.h> 172 #include <netinet6/in6_var.h> 173 #include <netinet6/ip6_var.h> 174 #include <netinet6/in6_pcb.h> 175 #include <netinet6/nd6.h> 176 #endif 177 178 #ifdef FAST_IPSEC 179 #include <netipsec/ipsec.h> 180 #endif /* FAST_IPSEC*/ 181 #ifdef IPSEC 182 #include <netinet6/ipsec.h> 183 #endif 184 185 #include <netinet/tcp.h> 186 #define TCPOUTFLAGS 187 #include <netinet/tcp_fsm.h> 188 #include <netinet/tcp_seq.h> 189 #include <netinet/tcp_timer.h> 190 #include <netinet/tcp_var.h> 191 #include <netinet/tcpip.h> 192 #include <netinet/tcp_debug.h> 193 194 #ifdef notyet 195 extern struct mbuf *m_copypack(); 196 #endif 197 198 #define MAX_TCPOPTLEN 32 /* max # bytes that go in options */ 199 200 /* 201 * Knob to enable Congestion Window Monitoring, and control the 202 * the burst size it allows. Default burst is 4 packets, per 203 * the Internet draft. 204 */ 205 int tcp_cwm = 0; 206 int tcp_cwm_burstsize = 4; 207 208 #ifdef TCP_OUTPUT_COUNTERS 209 #include <sys/device.h> 210 211 extern struct evcnt tcp_output_bigheader; 212 extern struct evcnt tcp_output_predict_hit; 213 extern struct evcnt tcp_output_predict_miss; 214 extern struct evcnt tcp_output_copysmall; 215 extern struct evcnt tcp_output_copybig; 216 extern struct evcnt tcp_output_refbig; 217 218 #define TCP_OUTPUT_COUNTER_INCR(ev) (ev)->ev_count++ 219 #else 220 221 #define TCP_OUTPUT_COUNTER_INCR(ev) /* nothing */ 222 223 #endif /* TCP_OUTPUT_COUNTERS */ 224 225 static 226 #ifndef GPROF 227 __inline 228 #endif 229 int 230 tcp_segsize(struct tcpcb *tp, int *txsegsizep, int *rxsegsizep) 231 { 232 #ifdef INET 233 struct inpcb *inp = tp->t_inpcb; 234 #endif 235 #ifdef INET6 236 struct in6pcb *in6p = tp->t_in6pcb; 237 #endif 238 struct socket *so = NULL; 239 struct rtentry *rt; 240 struct ifnet *ifp; 241 int size; 242 int iphlen; 243 int optlen; 244 245 #ifdef DIAGNOSTIC 246 if (tp->t_inpcb && tp->t_in6pcb) 247 panic("tcp_segsize: both t_inpcb and t_in6pcb are set"); 248 #endif 249 switch (tp->t_family) { 250 #ifdef INET 251 case AF_INET: 252 iphlen = sizeof(struct ip); 253 break; 254 #endif 255 #ifdef INET6 256 case AF_INET6: 257 iphlen = sizeof(struct ip6_hdr); 258 break; 259 #endif 260 default: 261 size = tcp_mssdflt; 262 goto out; 263 } 264 265 rt = NULL; 266 #ifdef INET 267 if (inp) { 268 rt = in_pcbrtentry(inp); 269 so = inp->inp_socket; 270 } 271 #endif 272 #ifdef INET6 273 if (in6p) { 274 rt = in6_pcbrtentry(in6p); 275 so = in6p->in6p_socket; 276 } 277 #endif 278 if (rt == NULL) { 279 size = tcp_mssdflt; 280 goto out; 281 } 282 283 ifp = rt->rt_ifp; 284 285 size = tcp_mssdflt; 286 if (tp->t_mtudisc && rt->rt_rmx.rmx_mtu != 0) { 287 #ifdef INET6 288 if (in6p && rt->rt_rmx.rmx_mtu < IPV6_MMTU) { 289 /* 290 * RFC2460 section 5, last paragraph: if path MTU is 291 * smaller than 1280, use 1280 as packet size and 292 * attach fragment header. 293 */ 294 size = IPV6_MMTU - iphlen - sizeof(struct ip6_frag) - 295 sizeof(struct tcphdr); 296 } else 297 size = rt->rt_rmx.rmx_mtu - iphlen - 298 sizeof(struct tcphdr); 299 #else 300 size = rt->rt_rmx.rmx_mtu - iphlen - sizeof(struct tcphdr); 301 #endif 302 } else if (ifp->if_flags & IFF_LOOPBACK) 303 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 304 #ifdef INET 305 else if (inp && tp->t_mtudisc) 306 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 307 else if (inp && in_localaddr(inp->inp_faddr)) 308 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 309 #endif 310 #ifdef INET6 311 else if (in6p) { 312 #ifdef INET 313 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) { 314 /* mapped addr case */ 315 struct in_addr d; 316 bcopy(&in6p->in6p_faddr.s6_addr32[3], &d, sizeof(d)); 317 if (tp->t_mtudisc || in_localaddr(d)) 318 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 319 } else 320 #endif 321 { 322 /* 323 * for IPv6, path MTU discovery is always turned on, 324 * or the node must use packet size <= 1280. 325 */ 326 size = tp->t_mtudisc ? IN6_LINKMTU(ifp) : IPV6_MMTU; 327 size -= (iphlen + sizeof(struct tcphdr)); 328 } 329 } 330 #endif 331 out: 332 /* 333 * Now we must make room for whatever extra TCP/IP options are in 334 * the packet. 335 */ 336 optlen = tcp_optlen(tp); 337 338 /* 339 * XXX tp->t_ourmss should have the right size, but without this code 340 * fragmentation will occur... need more investigation 341 */ 342 #ifdef INET 343 if (inp) { 344 #if defined(IPSEC) || defined(FAST_IPSEC) 345 if (! IPSEC_PCB_SKIP_IPSEC(inp->inp_sp, IPSEC_DIR_OUTBOUND)) 346 optlen += ipsec4_hdrsiz_tcp(tp); 347 #endif 348 optlen += ip_optlen(inp); 349 } 350 #endif 351 #ifdef INET6 352 #ifdef INET 353 if (in6p && tp->t_family == AF_INET) { 354 #if defined(IPSEC) || defined(FAST_IPSEC) 355 if (! IPSEC_PCB_SKIP_IPSEC(in6p->in6p_sp, IPSEC_DIR_OUTBOUND)) 356 optlen += ipsec4_hdrsiz_tcp(tp); 357 #endif 358 /* XXX size -= ip_optlen(in6p); */ 359 } else 360 #endif 361 if (in6p && tp->t_family == AF_INET6) { 362 #ifdef IPSEC 363 if (! IPSEC_PCB_SKIP_IPSEC(in6p->in6p_sp, IPSEC_DIR_OUTBOUND)) 364 optlen += ipsec6_hdrsiz_tcp(tp); 365 #endif 366 optlen += ip6_optlen(in6p); 367 } 368 #endif 369 size -= optlen; 370 371 /* there may not be any room for data if mtu is too small */ 372 if (size < 0) 373 return (EMSGSIZE); 374 375 /* 376 * *rxsegsizep holds *estimated* inbound segment size (estimation 377 * assumes that path MTU is the same for both ways). this is only 378 * for silly window avoidance, do not use the value for other purposes. 379 * 380 * ipseclen is subtracted from both sides, this may not be right. 381 * I'm not quite sure about this (could someone comment). 382 */ 383 *txsegsizep = min(tp->t_peermss - optlen, size); 384 /* 385 * Never send more than half a buffer full. This insures that we can 386 * always keep 2 packets on the wire, no matter what SO_SNDBUF is, and 387 * therefore acks will never be delayed unless we run out of data to 388 * transmit. 389 */ 390 if (so) 391 *txsegsizep = min(so->so_snd.sb_hiwat >> 1, *txsegsizep); 392 *rxsegsizep = min(tp->t_ourmss - optlen, size); 393 394 if (*txsegsizep != tp->t_segsz) { 395 /* 396 * If the new segment size is larger, we don't want to 397 * mess up the congestion window, but if it is smaller 398 * we'll have to reduce the congestion window to ensure 399 * that we don't get into trouble with initial windows 400 * and the rest. In any case, if the segment size 401 * has changed, chances are the path has, too, and 402 * our congestion window will be different. 403 */ 404 if (*txsegsizep < tp->t_segsz) { 405 tp->snd_cwnd = max((tp->snd_cwnd / tp->t_segsz) 406 * *txsegsizep, *txsegsizep); 407 tp->snd_ssthresh = max((tp->snd_ssthresh / tp->t_segsz) 408 * *txsegsizep, *txsegsizep); 409 } 410 tp->t_segsz = *txsegsizep; 411 } 412 413 return (0); 414 } 415 416 static 417 #ifndef GPROF 418 __inline 419 #endif 420 int 421 tcp_build_datapkt(struct tcpcb *tp, struct socket *so, int off, 422 long len, int hdrlen, struct mbuf **mp) 423 { 424 struct mbuf *m, *m0; 425 426 if (tp->t_force && len == 1) 427 tcpstat.tcps_sndprobe++; 428 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { 429 tcpstat.tcps_sndrexmitpack++; 430 tcpstat.tcps_sndrexmitbyte += len; 431 } else { 432 tcpstat.tcps_sndpack++; 433 tcpstat.tcps_sndbyte += len; 434 } 435 #ifdef notyet 436 if ((m = m_copypack(so->so_snd.sb_mb, off, 437 (int)len, max_linkhdr + hdrlen)) == 0) 438 return (ENOBUFS); 439 /* 440 * m_copypack left space for our hdr; use it. 441 */ 442 m->m_len += hdrlen; 443 m->m_data -= hdrlen; 444 #else 445 MGETHDR(m, M_DONTWAIT, MT_HEADER); 446 if (__predict_false(m == NULL)) 447 return (ENOBUFS); 448 MCLAIM(m, &tcp_tx_mowner); 449 450 /* 451 * XXX Because other code assumes headers will fit in 452 * XXX one header mbuf. 453 * 454 * (This code should almost *never* be run.) 455 */ 456 if (__predict_false((max_linkhdr + hdrlen) > MHLEN)) { 457 TCP_OUTPUT_COUNTER_INCR(&tcp_output_bigheader); 458 MCLGET(m, M_DONTWAIT); 459 if ((m->m_flags & M_EXT) == 0) { 460 m_freem(m); 461 return (ENOBUFS); 462 } 463 } 464 465 m->m_data += max_linkhdr; 466 m->m_len = hdrlen; 467 468 /* 469 * To avoid traversing the whole sb_mb chain for correct 470 * data to send, remember last sent mbuf, its offset and 471 * the sent size. When called the next time, see if the 472 * data to send is directly following the previous transfer. 473 * This is important for large TCP windows. 474 */ 475 if (off == 0 || tp->t_lastm == NULL || 476 (tp->t_lastoff + tp->t_lastlen) != off) { 477 TCP_OUTPUT_COUNTER_INCR(&tcp_output_predict_miss); 478 /* 479 * Either a new packet or a retransmit. 480 * Start from the beginning. 481 */ 482 tp->t_lastm = so->so_snd.sb_mb; 483 tp->t_inoff = off; 484 } else { 485 TCP_OUTPUT_COUNTER_INCR(&tcp_output_predict_hit); 486 tp->t_inoff += tp->t_lastlen; 487 } 488 489 /* Traverse forward to next packet */ 490 while (tp->t_inoff > 0) { 491 if (tp->t_lastm == NULL) 492 panic("tp->t_lastm == NULL"); 493 if (tp->t_inoff < tp->t_lastm->m_len) 494 break; 495 tp->t_inoff -= tp->t_lastm->m_len; 496 tp->t_lastm = tp->t_lastm->m_next; 497 } 498 499 tp->t_lastoff = off; 500 tp->t_lastlen = len; 501 m0 = tp->t_lastm; 502 off = tp->t_inoff; 503 504 if (len <= M_TRAILINGSPACE(m)) { 505 m_copydata(m0, off, (int) len, mtod(m, caddr_t) + hdrlen); 506 m->m_len += len; 507 TCP_OUTPUT_COUNTER_INCR(&tcp_output_copysmall); 508 } else { 509 m->m_next = m_copy(m0, off, (int) len); 510 if (m->m_next == NULL) { 511 m_freem(m); 512 return (ENOBUFS); 513 } 514 #ifdef TCP_OUTPUT_COUNTERS 515 if (m->m_next->m_flags & M_EXT) 516 TCP_OUTPUT_COUNTER_INCR(&tcp_output_refbig); 517 else 518 TCP_OUTPUT_COUNTER_INCR(&tcp_output_copybig); 519 #endif /* TCP_OUTPUT_COUNTERS */ 520 } 521 #endif 522 523 *mp = m; 524 return (0); 525 } 526 527 /* 528 * Tcp output routine: figure out what should be sent and send it. 529 */ 530 int 531 tcp_output(tp) 532 struct tcpcb *tp; 533 { 534 struct socket *so; 535 struct route *ro; 536 long len, win; 537 int off, flags, error; 538 struct mbuf *m; 539 struct ip *ip; 540 #ifdef INET6 541 struct ip6_hdr *ip6; 542 #endif 543 struct tcphdr *th; 544 u_char opt[MAX_TCPOPTLEN]; 545 unsigned optlen, hdrlen; 546 int idle, sendalot, txsegsize, rxsegsize; 547 int maxburst = TCP_MAXBURST; 548 int af; /* address family on the wire */ 549 int iphdrlen; 550 #ifdef TCP_SIGNATURE 551 int sigoff = 0; 552 #endif 553 554 #ifdef DIAGNOSTIC 555 if (tp->t_inpcb && tp->t_in6pcb) 556 panic("tcp_output: both t_inpcb and t_in6pcb are set"); 557 #endif 558 so = NULL; 559 ro = NULL; 560 if (tp->t_inpcb) { 561 so = tp->t_inpcb->inp_socket; 562 ro = &tp->t_inpcb->inp_route; 563 } 564 #ifdef INET6 565 else if (tp->t_in6pcb) { 566 so = tp->t_in6pcb->in6p_socket; 567 ro = (struct route *)&tp->t_in6pcb->in6p_route; 568 } 569 #endif 570 571 switch (af = tp->t_family) { 572 #ifdef INET 573 case AF_INET: 574 if (tp->t_inpcb) 575 break; 576 #ifdef INET6 577 /* mapped addr case */ 578 if (tp->t_in6pcb) 579 break; 580 #endif 581 return (EINVAL); 582 #endif 583 #ifdef INET6 584 case AF_INET6: 585 if (tp->t_in6pcb) 586 break; 587 return (EINVAL); 588 #endif 589 default: 590 return (EAFNOSUPPORT); 591 } 592 593 if (tcp_segsize(tp, &txsegsize, &rxsegsize)) 594 return (EMSGSIZE); 595 596 idle = (tp->snd_max == tp->snd_una); 597 598 /* 599 * Restart Window computation. From draft-floyd-incr-init-win-03: 600 * 601 * Optionally, a TCP MAY set the restart window to the 602 * minimum of the value used for the initial window and 603 * the current value of cwnd (in other words, using a 604 * larger value for the restart window should never increase 605 * the size of cwnd). 606 */ 607 if (tcp_cwm) { 608 /* 609 * Hughes/Touch/Heidemann Congestion Window Monitoring. 610 * Count the number of packets currently pending 611 * acknowledgement, and limit our congestion window 612 * to a pre-determined allowed burst size plus that count. 613 * This prevents bursting once all pending packets have 614 * been acknowledged (i.e. transmission is idle). 615 * 616 * XXX Link this to Initial Window? 617 */ 618 tp->snd_cwnd = min(tp->snd_cwnd, 619 (tcp_cwm_burstsize * txsegsize) + 620 (tp->snd_nxt - tp->snd_una)); 621 } else { 622 if (idle && (tcp_now - tp->t_rcvtime) >= tp->t_rxtcur) { 623 /* 624 * We have been idle for "a while" and no acks are 625 * expected to clock out any data we send -- 626 * slow start to get ack "clock" running again. 627 */ 628 int ss = tcp_init_win; 629 #ifdef INET 630 if (tp->t_inpcb && 631 in_localaddr(tp->t_inpcb->inp_faddr)) 632 ss = tcp_init_win_local; 633 #endif 634 #ifdef INET6 635 if (tp->t_in6pcb && 636 in6_localaddr(&tp->t_in6pcb->in6p_faddr)) 637 ss = tcp_init_win_local; 638 #endif 639 tp->snd_cwnd = min(tp->snd_cwnd, 640 TCP_INITIAL_WINDOW(ss, txsegsize)); 641 } 642 } 643 644 again: 645 /* 646 * Determine length of data that should be transmitted, and 647 * flags that should be used. If there is some data or critical 648 * controls (SYN, RST) to send, then transmit; otherwise, 649 * investigate further. 650 */ 651 sendalot = 0; 652 off = tp->snd_nxt - tp->snd_una; 653 win = min(tp->snd_wnd, tp->snd_cwnd); 654 655 flags = tcp_outflags[tp->t_state]; 656 /* 657 * If in persist timeout with window of 0, send 1 byte. 658 * Otherwise, if window is small but nonzero 659 * and timer expired, we will send what we can 660 * and go to transmit state. 661 */ 662 if (tp->t_force) { 663 if (win == 0) { 664 /* 665 * If we still have some data to send, then 666 * clear the FIN bit. Usually this would 667 * happen below when it realizes that we 668 * aren't sending all the data. However, 669 * if we have exactly 1 byte of unset data, 670 * then it won't clear the FIN bit below, 671 * and if we are in persist state, we wind 672 * up sending the packet without recording 673 * that we sent the FIN bit. 674 * 675 * We can't just blindly clear the FIN bit, 676 * because if we don't have any more data 677 * to send then the probe will be the FIN 678 * itself. 679 */ 680 if (off < so->so_snd.sb_cc) 681 flags &= ~TH_FIN; 682 win = 1; 683 } else { 684 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 685 tp->t_rxtshift = 0; 686 } 687 } 688 689 if (win < so->so_snd.sb_cc) { 690 len = win - off; 691 flags &= ~TH_FIN; 692 } else 693 len = so->so_snd.sb_cc - off; 694 695 if (len < 0) { 696 /* 697 * If FIN has been sent but not acked, 698 * but we haven't been called to retransmit, 699 * len will be -1. Otherwise, window shrank 700 * after we sent into it. If window shrank to 0, 701 * cancel pending retransmit, pull snd_nxt back 702 * to (closed) window, and set the persist timer 703 * if it isn't already going. If the window didn't 704 * close completely, just wait for an ACK. 705 * 706 * If we have a pending FIN, either it has already been 707 * transmitted or it is outside the window, so drop it. 708 * If the FIN has been transmitted, but this is not a 709 * retransmission, then len must be -1. Therefore we also 710 * prevent here the sending of `gratuitous FINs'. This 711 * eliminates the need to check for that case below (e.g. 712 * to back up snd_nxt before the FIN so that the sequence 713 * number is correct). 714 */ 715 len = 0; 716 flags &= ~TH_FIN; 717 if (win == 0) { 718 TCP_TIMER_DISARM(tp, TCPT_REXMT); 719 tp->t_rxtshift = 0; 720 tp->snd_nxt = tp->snd_una; 721 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 722 tcp_setpersist(tp); 723 } 724 } 725 if (len > txsegsize) { 726 len = txsegsize; 727 flags &= ~TH_FIN; 728 sendalot = 1; 729 } 730 731 win = sbspace(&so->so_rcv); 732 733 /* 734 * Sender silly window avoidance. If connection is idle 735 * and can send all data, a maximum segment, 736 * at least a maximum default-size segment do it, 737 * or are forced, do it; otherwise don't bother. 738 * If peer's buffer is tiny, then send 739 * when window is at least half open. 740 * If retransmitting (possibly after persist timer forced us 741 * to send into a small window), then must resend. 742 */ 743 if (len) { 744 if (len == txsegsize) 745 goto send; 746 if ((so->so_state & SS_MORETOCOME) == 0 && 747 ((idle || tp->t_flags & TF_NODELAY) && 748 len + off >= so->so_snd.sb_cc)) 749 goto send; 750 if (tp->t_force) 751 goto send; 752 if (len >= tp->max_sndwnd / 2) 753 goto send; 754 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) 755 goto send; 756 } 757 758 /* 759 * Compare available window to amount of window known to peer 760 * (as advertised window less next expected input). If the 761 * difference is at least twice the size of the largest segment 762 * we expect to receive (i.e. two segments) or at least 50% of 763 * the maximum possible window, then want to send a window update 764 * to peer. 765 */ 766 if (win > 0) { 767 /* 768 * "adv" is the amount we can increase the window, 769 * taking into account that we are limited by 770 * TCP_MAXWIN << tp->rcv_scale. 771 */ 772 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) - 773 (tp->rcv_adv - tp->rcv_nxt); 774 775 if (adv >= (long) (2 * rxsegsize)) 776 goto send; 777 if (2 * adv >= (long) so->so_rcv.sb_hiwat) 778 goto send; 779 } 780 781 /* 782 * Send if we owe peer an ACK. 783 */ 784 if (tp->t_flags & TF_ACKNOW) 785 goto send; 786 if (flags & (TH_SYN|TH_FIN|TH_RST)) 787 goto send; 788 if (SEQ_GT(tp->snd_up, tp->snd_una)) 789 goto send; 790 791 /* 792 * TCP window updates are not reliable, rather a polling protocol 793 * using ``persist'' packets is used to insure receipt of window 794 * updates. The three ``states'' for the output side are: 795 * idle not doing retransmits or persists 796 * persisting to move a small or zero window 797 * (re)transmitting and thereby not persisting 798 * 799 * tp->t_timer[TCPT_PERSIST] 800 * is set when we are in persist state. 801 * tp->t_force 802 * is set when we are called to send a persist packet. 803 * tp->t_timer[TCPT_REXMT] 804 * is set when we are retransmitting 805 * The output side is idle when both timers are zero. 806 * 807 * If send window is too small, there is data to transmit, and no 808 * retransmit or persist is pending, then go to persist state. 809 * If nothing happens soon, send when timer expires: 810 * if window is nonzero, transmit what we can, 811 * otherwise force out a byte. 812 */ 813 if (so->so_snd.sb_cc && TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 814 TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 815 tp->t_rxtshift = 0; 816 tcp_setpersist(tp); 817 } 818 819 /* 820 * No reason to send a segment, just return. 821 */ 822 return (0); 823 824 send: 825 /* 826 * Before ESTABLISHED, force sending of initial options 827 * unless TCP set not to do any options. 828 * NOTE: we assume that the IP/TCP header plus TCP options 829 * always fit in a single mbuf, leaving room for a maximum 830 * link header, i.e. 831 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES 832 */ 833 optlen = 0; 834 switch (af) { 835 #ifdef INET 836 case AF_INET: 837 iphdrlen = sizeof(struct ip) + sizeof(struct tcphdr); 838 break; 839 #endif 840 #ifdef INET6 841 case AF_INET6: 842 iphdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 843 break; 844 #endif 845 default: /*pacify gcc*/ 846 iphdrlen = 0; 847 break; 848 } 849 hdrlen = iphdrlen; 850 if (flags & TH_SYN) { 851 struct rtentry *rt; 852 853 rt = NULL; 854 #ifdef INET 855 if (tp->t_inpcb) 856 rt = in_pcbrtentry(tp->t_inpcb); 857 #endif 858 #ifdef INET6 859 if (tp->t_in6pcb) 860 rt = in6_pcbrtentry(tp->t_in6pcb); 861 #endif 862 863 tp->snd_nxt = tp->iss; 864 tp->t_ourmss = tcp_mss_to_advertise(rt != NULL ? 865 rt->rt_ifp : NULL, af); 866 if ((tp->t_flags & TF_NOOPT) == 0) { 867 opt[0] = TCPOPT_MAXSEG; 868 opt[1] = 4; 869 opt[2] = (tp->t_ourmss >> 8) & 0xff; 870 opt[3] = tp->t_ourmss & 0xff; 871 optlen = 4; 872 873 if ((tp->t_flags & TF_REQ_SCALE) && 874 ((flags & TH_ACK) == 0 || 875 (tp->t_flags & TF_RCVD_SCALE))) { 876 *((u_int32_t *) (opt + optlen)) = htonl( 877 TCPOPT_NOP << 24 | 878 TCPOPT_WINDOW << 16 | 879 TCPOLEN_WINDOW << 8 | 880 tp->request_r_scale); 881 optlen += 4; 882 } 883 } 884 } 885 886 /* 887 * Send a timestamp and echo-reply if this is a SYN and our side 888 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side 889 * and our peer have sent timestamps in our SYN's. 890 */ 891 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 892 (flags & TH_RST) == 0 && 893 ((flags & (TH_SYN|TH_ACK)) == TH_SYN || 894 (tp->t_flags & TF_RCVD_TSTMP))) { 895 u_int32_t *lp = (u_int32_t *)(opt + optlen); 896 897 /* Form timestamp option as shown in appendix A of RFC 1323. */ 898 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 899 *lp++ = htonl(TCP_TIMESTAMP(tp)); 900 *lp = htonl(tp->ts_recent); 901 optlen += TCPOLEN_TSTAMP_APPA; 902 } 903 904 #ifdef TCP_SIGNATURE 905 #ifdef INET6 906 /* XXX: FreeBSD uses ((tp->t_inpcb->inp_vflag & INP_IPV6) == 0) */ 907 if (tp->t_family == AF_INET) 908 #endif 909 if (tp->t_flags & TF_SIGNATURE) { 910 int i; 911 u_char *bp; 912 /* 913 * Initialize TCP-MD5 option (RFC2385) 914 */ 915 bp = (u_char *)opt + optlen; 916 *bp++ = TCPOPT_SIGNATURE; 917 *bp++ = TCPOLEN_SIGNATURE; 918 sigoff = optlen + 2; 919 for (i = 0; i < TCP_SIGLEN; i++) 920 *bp++ = 0; 921 optlen += TCPOLEN_SIGNATURE; 922 /* 923 * Terminate options list and maintain 32-bit alignment. 924 */ 925 *bp++ = TCPOPT_NOP; 926 *bp++ = TCPOPT_EOL; 927 optlen += 2; 928 } 929 #endif /* TCP_SIGNATURE */ 930 931 hdrlen += optlen; 932 933 #ifdef DIAGNOSTIC 934 if (len > txsegsize) 935 panic("tcp data to be sent is larger than segment"); 936 if (max_linkhdr + hdrlen > MCLBYTES) 937 panic("tcphdr too big"); 938 #endif 939 940 /* 941 * Grab a header mbuf, attaching a copy of data to 942 * be transmitted, and initialize the header from 943 * the template for sends on this connection. 944 */ 945 if (len) { 946 error = tcp_build_datapkt(tp, so, off, len, hdrlen, &m); 947 if (error) 948 goto out; 949 /* 950 * If we're sending everything we've got, set PUSH. 951 * (This will keep happy those implementations which only 952 * give data to the user when a buffer fills or 953 * a PUSH comes in.) 954 */ 955 if (off + len == so->so_snd.sb_cc) 956 flags |= TH_PUSH; 957 } else { 958 if (tp->t_flags & TF_ACKNOW) 959 tcpstat.tcps_sndacks++; 960 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 961 tcpstat.tcps_sndctrl++; 962 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 963 tcpstat.tcps_sndurg++; 964 else 965 tcpstat.tcps_sndwinup++; 966 967 MGETHDR(m, M_DONTWAIT, MT_HEADER); 968 if (m != NULL && max_linkhdr + hdrlen > MHLEN) { 969 MCLGET(m, M_DONTWAIT); 970 if ((m->m_flags & M_EXT) == 0) { 971 m_freem(m); 972 m = NULL; 973 } 974 } 975 if (m == NULL) { 976 error = ENOBUFS; 977 goto out; 978 } 979 MCLAIM(m, &tcp_tx_mowner); 980 m->m_data += max_linkhdr; 981 m->m_len = hdrlen; 982 } 983 m->m_pkthdr.rcvif = (struct ifnet *)0; 984 switch (af) { 985 #ifdef INET 986 case AF_INET: 987 ip = mtod(m, struct ip *); 988 #ifdef INET6 989 ip6 = NULL; 990 #endif 991 th = (struct tcphdr *)(ip + 1); 992 break; 993 #endif 994 #ifdef INET6 995 case AF_INET6: 996 ip = NULL; 997 ip6 = mtod(m, struct ip6_hdr *); 998 th = (struct tcphdr *)(ip6 + 1); 999 break; 1000 #endif 1001 default: /*pacify gcc*/ 1002 ip = NULL; 1003 #ifdef INET6 1004 ip6 = NULL; 1005 #endif 1006 th = NULL; 1007 break; 1008 } 1009 if (tp->t_template == 0) 1010 panic("tcp_output"); 1011 if (tp->t_template->m_len < iphdrlen) 1012 panic("tcp_output"); 1013 bcopy(mtod(tp->t_template, caddr_t), mtod(m, caddr_t), iphdrlen); 1014 1015 /* 1016 * If we are doing retransmissions, then snd_nxt will 1017 * not reflect the first unsent octet. For ACK only 1018 * packets, we do not want the sequence number of the 1019 * retransmitted packet, we want the sequence number 1020 * of the next unsent octet. So, if there is no data 1021 * (and no SYN or FIN), use snd_max instead of snd_nxt 1022 * when filling in ti_seq. But if we are in persist 1023 * state, snd_max might reflect one byte beyond the 1024 * right edge of the window, so use snd_nxt in that 1025 * case, since we know we aren't doing a retransmission. 1026 * (retransmit and persist are mutually exclusive...) 1027 */ 1028 if (len || (flags & (TH_SYN|TH_FIN)) || 1029 TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) 1030 th->th_seq = htonl(tp->snd_nxt); 1031 else 1032 th->th_seq = htonl(tp->snd_max); 1033 th->th_ack = htonl(tp->rcv_nxt); 1034 if (optlen) { 1035 bcopy((caddr_t)opt, (caddr_t)(th + 1), optlen); 1036 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1037 } 1038 th->th_flags = flags; 1039 /* 1040 * Calculate receive window. Don't shrink window, 1041 * but avoid silly window syndrome. 1042 */ 1043 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)rxsegsize) 1044 win = 0; 1045 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 1046 win = (long)TCP_MAXWIN << tp->rcv_scale; 1047 if (win < (long)(tp->rcv_adv - tp->rcv_nxt)) 1048 win = (long)(tp->rcv_adv - tp->rcv_nxt); 1049 th->th_win = htons((u_int16_t) (win>>tp->rcv_scale)); 1050 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 1051 u_int32_t urp = tp->snd_up - tp->snd_nxt; 1052 if (urp > IP_MAXPACKET) 1053 urp = IP_MAXPACKET; 1054 th->th_urp = htons((u_int16_t)urp); 1055 th->th_flags |= TH_URG; 1056 } else 1057 /* 1058 * If no urgent pointer to send, then we pull 1059 * the urgent pointer to the left edge of the send window 1060 * so that it doesn't drift into the send window on sequence 1061 * number wraparound. 1062 */ 1063 tp->snd_up = tp->snd_una; /* drag it along */ 1064 1065 #ifdef TCP_SIGNATURE 1066 #ifdef INET6 1067 /* XXX: FreeBSD uses ((tp->t_inpcb->inp_vflag & INP_IPV6) == 0) */ 1068 if (tp->t_family == AF_INET) /* XXX */ 1069 #endif 1070 if (tp->t_flags & TF_SIGNATURE) 1071 tcp_signature_compute(m, sizeof(struct ip), len, optlen, 1072 (u_char *)(th + 1) + sigoff, IPSEC_DIR_OUTBOUND); 1073 #endif 1074 1075 /* 1076 * Set ourselves up to be checksummed just before the packet 1077 * hits the wire. 1078 */ 1079 switch (af) { 1080 #ifdef INET 1081 case AF_INET: 1082 m->m_pkthdr.csum_flags = M_CSUM_TCPv4; 1083 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1084 if (len + optlen) { 1085 /* Fixup the pseudo-header checksum. */ 1086 /* XXXJRT Not IP Jumbogram safe. */ 1087 th->th_sum = in_cksum_addword(th->th_sum, 1088 htons((u_int16_t) (len + optlen))); 1089 } 1090 break; 1091 #endif 1092 #ifdef INET6 1093 case AF_INET6: 1094 /* 1095 * XXX Actually delaying the checksum is Hard 1096 * XXX (well, maybe not for Itojun, but it is 1097 * XXX for me), but we can still take advantage 1098 * XXX of the cached pseudo-header checksum. 1099 */ 1100 /* equals to hdrlen + len */ 1101 m->m_pkthdr.len = sizeof(struct ip6_hdr) 1102 + sizeof(struct tcphdr) + optlen + len; 1103 #ifdef notyet 1104 m->m_pkthdr.csum_flags = M_CSUM_TCPv6; 1105 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1106 #endif 1107 if (len + optlen) { 1108 /* Fixup the pseudo-header checksum. */ 1109 /* XXXJRT: Not IPv6 Jumbogram safe. */ 1110 th->th_sum = in_cksum_addword(th->th_sum, 1111 htons((u_int16_t) (len + optlen))); 1112 } 1113 #ifndef notyet 1114 th->th_sum = in6_cksum(m, 0, sizeof(struct ip6_hdr), 1115 sizeof(struct tcphdr) + optlen + len); 1116 #endif 1117 break; 1118 #endif 1119 } 1120 1121 /* 1122 * In transmit state, time the transmission and arrange for 1123 * the retransmit. In persist state, just set snd_max. 1124 */ 1125 if (tp->t_force == 0 || TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 1126 tcp_seq startseq = tp->snd_nxt; 1127 1128 /* 1129 * Advance snd_nxt over sequence space of this segment. 1130 * There are no states in which we send both a SYN and a FIN, 1131 * so we collapse the tests for these flags. 1132 */ 1133 if (flags & (TH_SYN|TH_FIN)) 1134 tp->snd_nxt++; 1135 tp->snd_nxt += len; 1136 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 1137 tp->snd_max = tp->snd_nxt; 1138 /* 1139 * Time this transmission if not a retransmission and 1140 * not currently timing anything. 1141 */ 1142 if (tp->t_rtttime == 0) { 1143 tp->t_rtttime = tcp_now; 1144 tp->t_rtseq = startseq; 1145 tcpstat.tcps_segstimed++; 1146 } 1147 } 1148 1149 /* 1150 * Set retransmit timer if not currently set, 1151 * and not doing an ack or a keep-alive probe. 1152 * Initial value for retransmit timer is smoothed 1153 * round-trip time + 2 * round-trip time variance. 1154 * Initialize shift counter which is used for backoff 1155 * of retransmit time. 1156 */ 1157 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 1158 tp->snd_nxt != tp->snd_una) { 1159 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) { 1160 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 1161 tp->t_rxtshift = 0; 1162 } 1163 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1164 } 1165 } else 1166 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max)) 1167 tp->snd_max = tp->snd_nxt + len; 1168 1169 #ifdef TCP_DEBUG 1170 /* 1171 * Trace. 1172 */ 1173 if (so->so_options & SO_DEBUG) 1174 tcp_trace(TA_OUTPUT, tp->t_state, tp, m, 0); 1175 #endif 1176 1177 /* 1178 * Fill in IP length and desired time to live and 1179 * send to IP level. There should be a better way 1180 * to handle ttl and tos; we could keep them in 1181 * the template, but need a way to checksum without them. 1182 */ 1183 m->m_pkthdr.len = hdrlen + len; 1184 1185 switch (af) { 1186 #ifdef INET 1187 case AF_INET: 1188 ip->ip_len = htons(m->m_pkthdr.len); 1189 if (tp->t_inpcb) { 1190 ip->ip_ttl = tp->t_inpcb->inp_ip.ip_ttl; 1191 ip->ip_tos = tp->t_inpcb->inp_ip.ip_tos; 1192 } 1193 #ifdef INET6 1194 else if (tp->t_in6pcb) { 1195 ip->ip_ttl = in6_selecthlim(tp->t_in6pcb, NULL); /*XXX*/ 1196 ip->ip_tos = 0; /*XXX*/ 1197 } 1198 #endif 1199 break; 1200 #endif 1201 #ifdef INET6 1202 case AF_INET6: 1203 ip6->ip6_nxt = IPPROTO_TCP; 1204 if (tp->t_in6pcb) { 1205 /* 1206 * we separately set hoplimit for every segment, since 1207 * the user might want to change the value via 1208 * setsockopt. Also, desired default hop limit might 1209 * be changed via Neighbor Discovery. 1210 */ 1211 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, 1212 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 1213 } 1214 /* ip6->ip6_flow = ??? */ 1215 /* ip6_plen will be filled in ip6_output(). */ 1216 break; 1217 #endif 1218 } 1219 1220 switch (af) { 1221 #ifdef INET 1222 case AF_INET: 1223 { 1224 struct mbuf *opts; 1225 1226 if (tp->t_inpcb) 1227 opts = tp->t_inpcb->inp_options; 1228 else 1229 opts = NULL; 1230 error = ip_output(m, opts, ro, 1231 (tp->t_mtudisc ? IP_MTUDISC : 0) | 1232 (so->so_options & SO_DONTROUTE), 1233 (struct ip_moptions *)0, so); 1234 break; 1235 } 1236 #endif 1237 #ifdef INET6 1238 case AF_INET6: 1239 { 1240 struct ip6_pktopts *opts; 1241 1242 if (tp->t_in6pcb) 1243 opts = tp->t_in6pcb->in6p_outputopts; 1244 else 1245 opts = NULL; 1246 error = ip6_output(m, opts, (struct route_in6 *)ro, 1247 so->so_options & SO_DONTROUTE, 1248 (struct ip6_moptions *)0, so, NULL); 1249 break; 1250 } 1251 #endif 1252 default: 1253 error = EAFNOSUPPORT; 1254 break; 1255 } 1256 if (error) { 1257 out: 1258 if (error == ENOBUFS) { 1259 tcpstat.tcps_selfquench++; 1260 #ifdef INET 1261 if (tp->t_inpcb) 1262 tcp_quench(tp->t_inpcb, 0); 1263 #endif 1264 #ifdef INET6 1265 if (tp->t_in6pcb) 1266 tcp6_quench(tp->t_in6pcb, 0); 1267 #endif 1268 error = 0; 1269 } else if ((error == EHOSTUNREACH || error == ENETDOWN) && 1270 TCPS_HAVERCVDSYN(tp->t_state)) { 1271 tp->t_softerror = error; 1272 error = 0; 1273 } 1274 1275 /* Restart the delayed ACK timer, if necessary. */ 1276 if (tp->t_flags & TF_DELACK) 1277 TCP_RESTART_DELACK(tp); 1278 1279 return (error); 1280 } 1281 tcpstat.tcps_sndtotal++; 1282 if (tp->t_flags & TF_DELACK) 1283 tcpstat.tcps_delack++; 1284 1285 /* 1286 * Data sent (as far as we can tell). 1287 * If this advertises a larger window than any other segment, 1288 * then remember the size of the advertised window. 1289 * Any pending ACK has now been sent. 1290 */ 1291 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 1292 tp->rcv_adv = tp->rcv_nxt + win; 1293 tp->last_ack_sent = tp->rcv_nxt; 1294 tp->t_flags &= ~TF_ACKNOW; 1295 TCP_CLEAR_DELACK(tp); 1296 #ifdef DIAGNOSTIC 1297 if (maxburst < 0) 1298 printf("tcp_output: maxburst exceeded by %d\n", -maxburst); 1299 #endif 1300 if (sendalot && (!tcp_do_newreno || --maxburst)) 1301 goto again; 1302 return (0); 1303 } 1304 1305 void 1306 tcp_setpersist(tp) 1307 struct tcpcb *tp; 1308 { 1309 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2); 1310 int nticks; 1311 1312 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT)) 1313 panic("tcp_output REXMT"); 1314 /* 1315 * Start/restart persistance timer. 1316 */ 1317 if (t < tp->t_rttmin) 1318 t = tp->t_rttmin; 1319 TCPT_RANGESET(nticks, t * tcp_backoff[tp->t_rxtshift], 1320 TCPTV_PERSMIN, TCPTV_PERSMAX); 1321 TCP_TIMER_ARM(tp, TCPT_PERSIST, nticks); 1322 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 1323 tp->t_rxtshift++; 1324 } 1325
Indexes created Tue Sep 23 14:10:03 GMT 2025