tcp_output.c revision 1.119
1 /* $NetBSD: tcp_output.c,v 1.119 2005/03/02 10:20:18 mycroft 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, 2005 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 * This code is derived from software contributed to The NetBSD Foundation 80 * by Charles M. Hannum. 81 * 82 * Redistribution and use in source and binary forms, with or without 83 * modification, are permitted provided that the following conditions 84 * are met: 85 * 1. Redistributions of source code must retain the above copyright 86 * notice, this list of conditions and the following disclaimer. 87 * 2. Redistributions in binary form must reproduce the above copyright 88 * notice, this list of conditions and the following disclaimer in the 89 * documentation and/or other materials provided with the distribution. 90 * 3. All advertising materials mentioning features or use of this software 91 * must display the following acknowledgement: 92 * This product includes software developed by the NetBSD 93 * Foundation, Inc. and its contributors. 94 * 4. Neither the name of The NetBSD Foundation nor the names of its 95 * contributors may be used to endorse or promote products derived 96 * from this software without specific prior written permission. 97 * 98 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 99 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 100 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 101 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 102 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 103 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 104 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 105 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 106 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 107 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 108 * POSSIBILITY OF SUCH DAMAGE. 109 */ 110 111 /* 112 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 113 * The Regents of the University of California. All rights reserved. 114 * 115 * Redistribution and use in source and binary forms, with or without 116 * modification, are permitted provided that the following conditions 117 * are met: 118 * 1. Redistributions of source code must retain the above copyright 119 * notice, this list of conditions and the following disclaimer. 120 * 2. Redistributions in binary form must reproduce the above copyright 121 * notice, this list of conditions and the following disclaimer in the 122 * documentation and/or other materials provided with the distribution. 123 * 3. Neither the name of the University nor the names of its contributors 124 * may be used to endorse or promote products derived from this software 125 * without specific prior written permission. 126 * 127 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 128 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 129 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 130 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 131 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 132 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 133 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 134 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 135 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 136 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 137 * SUCH DAMAGE. 138 * 139 * @(#)tcp_output.c 8.4 (Berkeley) 5/24/95 140 */ 141 142 #include <sys/cdefs.h> 143 __KERNEL_RCSID(0, "$NetBSD: tcp_output.c,v 1.119 2005/03/02 10:20:18 mycroft Exp $"); 144 145 #include "opt_inet.h" 146 #include "opt_ipsec.h" 147 #include "opt_tcp_debug.h" 148 149 #include <sys/param.h> 150 #include <sys/systm.h> 151 #include <sys/malloc.h> 152 #include <sys/mbuf.h> 153 #include <sys/protosw.h> 154 #include <sys/socket.h> 155 #include <sys/socketvar.h> 156 #include <sys/errno.h> 157 #include <sys/domain.h> 158 #include <sys/kernel.h> 159 #ifdef TCP_SIGNATURE 160 #include <sys/md5.h> 161 #endif 162 163 #include <net/if.h> 164 #include <net/route.h> 165 166 #include <netinet/in.h> 167 #include <netinet/in_systm.h> 168 #include <netinet/ip.h> 169 #include <netinet/in_pcb.h> 170 #include <netinet/ip_var.h> 171 172 #ifdef INET6 173 #ifndef INET 174 #include <netinet/in.h> 175 #endif 176 #include <netinet/ip6.h> 177 #include <netinet6/in6_var.h> 178 #include <netinet6/ip6_var.h> 179 #include <netinet6/in6_pcb.h> 180 #include <netinet6/nd6.h> 181 #endif 182 183 #ifdef FAST_IPSEC 184 #include <netipsec/ipsec.h> 185 #include <netipsec/key.h> 186 #endif /* FAST_IPSEC*/ 187 #ifdef IPSEC 188 #include <netinet6/ipsec.h> 189 #endif 190 191 #include <netinet/tcp.h> 192 #define TCPOUTFLAGS 193 #include <netinet/tcp_fsm.h> 194 #include <netinet/tcp_seq.h> 195 #include <netinet/tcp_timer.h> 196 #include <netinet/tcp_var.h> 197 #include <netinet/tcpip.h> 198 #include <netinet/tcp_debug.h> 199 200 #ifdef IPSEC 201 #include <netkey/key.h> 202 #endif 203 204 #ifdef notyet 205 extern struct mbuf *m_copypack(); 206 #endif 207 208 #define MAX_TCPOPTLEN 40 /* max # bytes that go in options */ 209 210 /* 211 * Knob to enable Congestion Window Monitoring, and control the 212 * the burst size it allows. Default burst is 4 packets, per 213 * the Internet draft. 214 */ 215 int tcp_cwm = 0; 216 int tcp_cwm_burstsize = 4; 217 218 #ifdef TCP_OUTPUT_COUNTERS 219 #include <sys/device.h> 220 221 extern struct evcnt tcp_output_bigheader; 222 extern struct evcnt tcp_output_predict_hit; 223 extern struct evcnt tcp_output_predict_miss; 224 extern struct evcnt tcp_output_copysmall; 225 extern struct evcnt tcp_output_copybig; 226 extern struct evcnt tcp_output_refbig; 227 228 #define TCP_OUTPUT_COUNTER_INCR(ev) (ev)->ev_count++ 229 #else 230 231 #define TCP_OUTPUT_COUNTER_INCR(ev) /* nothing */ 232 233 #endif /* TCP_OUTPUT_COUNTERS */ 234 235 static 236 #ifndef GPROF 237 __inline 238 #endif 239 int 240 tcp_segsize(struct tcpcb *tp, int *txsegsizep, int *rxsegsizep) 241 { 242 #ifdef INET 243 struct inpcb *inp = tp->t_inpcb; 244 #endif 245 #ifdef INET6 246 struct in6pcb *in6p = tp->t_in6pcb; 247 #endif 248 struct socket *so = NULL; 249 struct rtentry *rt; 250 struct ifnet *ifp; 251 int size; 252 int iphlen; 253 int optlen; 254 255 #ifdef DIAGNOSTIC 256 if (tp->t_inpcb && tp->t_in6pcb) 257 panic("tcp_segsize: both t_inpcb and t_in6pcb are set"); 258 #endif 259 switch (tp->t_family) { 260 #ifdef INET 261 case AF_INET: 262 iphlen = sizeof(struct ip); 263 break; 264 #endif 265 #ifdef INET6 266 case AF_INET6: 267 iphlen = sizeof(struct ip6_hdr); 268 break; 269 #endif 270 default: 271 size = tcp_mssdflt; 272 goto out; 273 } 274 275 rt = NULL; 276 #ifdef INET 277 if (inp) { 278 rt = in_pcbrtentry(inp); 279 so = inp->inp_socket; 280 } 281 #endif 282 #ifdef INET6 283 if (in6p) { 284 rt = in6_pcbrtentry(in6p); 285 so = in6p->in6p_socket; 286 } 287 #endif 288 if (rt == NULL) { 289 size = tcp_mssdflt; 290 goto out; 291 } 292 293 ifp = rt->rt_ifp; 294 295 size = tcp_mssdflt; 296 if (tp->t_mtudisc && rt->rt_rmx.rmx_mtu != 0) { 297 #ifdef INET6 298 if (in6p && rt->rt_rmx.rmx_mtu < IPV6_MMTU) { 299 /* 300 * RFC2460 section 5, last paragraph: if path MTU is 301 * smaller than 1280, use 1280 as packet size and 302 * attach fragment header. 303 */ 304 size = IPV6_MMTU - iphlen - sizeof(struct ip6_frag) - 305 sizeof(struct tcphdr); 306 } else 307 size = rt->rt_rmx.rmx_mtu - iphlen - 308 sizeof(struct tcphdr); 309 #else 310 size = rt->rt_rmx.rmx_mtu - iphlen - sizeof(struct tcphdr); 311 #endif 312 } else if (ifp->if_flags & IFF_LOOPBACK) 313 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 314 #ifdef INET 315 else if (inp && tp->t_mtudisc) 316 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 317 else if (inp && in_localaddr(inp->inp_faddr)) 318 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 319 #endif 320 #ifdef INET6 321 else if (in6p) { 322 #ifdef INET 323 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) { 324 /* mapped addr case */ 325 struct in_addr d; 326 bcopy(&in6p->in6p_faddr.s6_addr32[3], &d, sizeof(d)); 327 if (tp->t_mtudisc || in_localaddr(d)) 328 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 329 } else 330 #endif 331 { 332 /* 333 * for IPv6, path MTU discovery is always turned on, 334 * or the node must use packet size <= 1280. 335 */ 336 size = tp->t_mtudisc ? IN6_LINKMTU(ifp) : IPV6_MMTU; 337 size -= (iphlen + sizeof(struct tcphdr)); 338 } 339 } 340 #endif 341 out: 342 /* 343 * Now we must make room for whatever extra TCP/IP options are in 344 * the packet. 345 */ 346 optlen = tcp_optlen(tp); 347 348 /* 349 * XXX tp->t_ourmss should have the right size, but without this code 350 * fragmentation will occur... need more investigation 351 */ 352 #ifdef INET 353 if (inp) { 354 #if defined(IPSEC) || defined(FAST_IPSEC) 355 if (! IPSEC_PCB_SKIP_IPSEC(inp->inp_sp, IPSEC_DIR_OUTBOUND)) 356 optlen += ipsec4_hdrsiz_tcp(tp); 357 #endif 358 optlen += ip_optlen(inp); 359 } 360 #endif 361 #ifdef INET6 362 #ifdef INET 363 if (in6p && tp->t_family == AF_INET) { 364 #if defined(IPSEC) || defined(FAST_IPSEC) 365 if (! IPSEC_PCB_SKIP_IPSEC(in6p->in6p_sp, IPSEC_DIR_OUTBOUND)) 366 optlen += ipsec4_hdrsiz_tcp(tp); 367 #endif 368 /* XXX size -= ip_optlen(in6p); */ 369 } else 370 #endif 371 if (in6p && tp->t_family == AF_INET6) { 372 #ifdef IPSEC 373 if (! IPSEC_PCB_SKIP_IPSEC(in6p->in6p_sp, IPSEC_DIR_OUTBOUND)) 374 optlen += ipsec6_hdrsiz_tcp(tp); 375 #endif 376 optlen += ip6_optlen(in6p); 377 } 378 #endif 379 size -= optlen; 380 381 /* there may not be any room for data if mtu is too small */ 382 if (size < 0) 383 return (EMSGSIZE); 384 385 /* 386 * *rxsegsizep holds *estimated* inbound segment size (estimation 387 * assumes that path MTU is the same for both ways). this is only 388 * for silly window avoidance, do not use the value for other purposes. 389 * 390 * ipseclen is subtracted from both sides, this may not be right. 391 * I'm not quite sure about this (could someone comment). 392 */ 393 *txsegsizep = min(tp->t_peermss - optlen, size); 394 /* 395 * Never send more than half a buffer full. This insures that we can 396 * always keep 2 packets on the wire, no matter what SO_SNDBUF is, and 397 * therefore acks will never be delayed unless we run out of data to 398 * transmit. 399 */ 400 if (so) 401 *txsegsizep = min(so->so_snd.sb_hiwat >> 1, *txsegsizep); 402 *rxsegsizep = min(tp->t_ourmss - optlen, size); 403 404 if (*txsegsizep != tp->t_segsz) { 405 /* 406 * If the new segment size is larger, we don't want to 407 * mess up the congestion window, but if it is smaller 408 * we'll have to reduce the congestion window to ensure 409 * that we don't get into trouble with initial windows 410 * and the rest. In any case, if the segment size 411 * has changed, chances are the path has, too, and 412 * our congestion window will be different. 413 */ 414 if (*txsegsizep < tp->t_segsz) { 415 tp->snd_cwnd = max((tp->snd_cwnd / tp->t_segsz) 416 * *txsegsizep, *txsegsizep); 417 tp->snd_ssthresh = max((tp->snd_ssthresh / tp->t_segsz) 418 * *txsegsizep, *txsegsizep); 419 } 420 tp->t_segsz = *txsegsizep; 421 } 422 423 return (0); 424 } 425 426 static 427 #ifndef GPROF 428 __inline 429 #endif 430 int 431 tcp_build_datapkt(struct tcpcb *tp, struct socket *so, int off, 432 long len, int hdrlen, struct mbuf **mp) 433 { 434 struct mbuf *m, *m0; 435 436 if (tp->t_force && len == 1) 437 tcpstat.tcps_sndprobe++; 438 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { 439 tcpstat.tcps_sndrexmitpack++; 440 tcpstat.tcps_sndrexmitbyte += len; 441 } else { 442 tcpstat.tcps_sndpack++; 443 tcpstat.tcps_sndbyte += len; 444 } 445 #ifdef notyet 446 if ((m = m_copypack(so->so_snd.sb_mb, off, 447 (int)len, max_linkhdr + hdrlen)) == 0) 448 return (ENOBUFS); 449 /* 450 * m_copypack left space for our hdr; use it. 451 */ 452 m->m_len += hdrlen; 453 m->m_data -= hdrlen; 454 #else 455 MGETHDR(m, M_DONTWAIT, MT_HEADER); 456 if (__predict_false(m == NULL)) 457 return (ENOBUFS); 458 MCLAIM(m, &tcp_tx_mowner); 459 460 /* 461 * XXX Because other code assumes headers will fit in 462 * XXX one header mbuf. 463 * 464 * (This code should almost *never* be run.) 465 */ 466 if (__predict_false((max_linkhdr + hdrlen) > MHLEN)) { 467 TCP_OUTPUT_COUNTER_INCR(&tcp_output_bigheader); 468 MCLGET(m, M_DONTWAIT); 469 if ((m->m_flags & M_EXT) == 0) { 470 m_freem(m); 471 return (ENOBUFS); 472 } 473 } 474 475 m->m_data += max_linkhdr; 476 m->m_len = hdrlen; 477 478 /* 479 * To avoid traversing the whole sb_mb chain for correct 480 * data to send, remember last sent mbuf, its offset and 481 * the sent size. When called the next time, see if the 482 * data to send is directly following the previous transfer. 483 * This is important for large TCP windows. 484 */ 485 if (off == 0 || tp->t_lastm == NULL || 486 (tp->t_lastoff + tp->t_lastlen) != off) { 487 TCP_OUTPUT_COUNTER_INCR(&tcp_output_predict_miss); 488 /* 489 * Either a new packet or a retransmit. 490 * Start from the beginning. 491 */ 492 tp->t_lastm = so->so_snd.sb_mb; 493 tp->t_inoff = off; 494 } else { 495 TCP_OUTPUT_COUNTER_INCR(&tcp_output_predict_hit); 496 tp->t_inoff += tp->t_lastlen; 497 } 498 499 /* Traverse forward to next packet */ 500 while (tp->t_inoff > 0) { 501 if (tp->t_lastm == NULL) 502 panic("tp->t_lastm == NULL"); 503 if (tp->t_inoff < tp->t_lastm->m_len) 504 break; 505 tp->t_inoff -= tp->t_lastm->m_len; 506 tp->t_lastm = tp->t_lastm->m_next; 507 } 508 509 tp->t_lastoff = off; 510 tp->t_lastlen = len; 511 m0 = tp->t_lastm; 512 off = tp->t_inoff; 513 514 if (len <= M_TRAILINGSPACE(m)) { 515 m_copydata(m0, off, (int) len, mtod(m, caddr_t) + hdrlen); 516 m->m_len += len; 517 TCP_OUTPUT_COUNTER_INCR(&tcp_output_copysmall); 518 } else { 519 m->m_next = m_copy(m0, off, (int) len); 520 if (m->m_next == NULL) { 521 m_freem(m); 522 return (ENOBUFS); 523 } 524 #ifdef TCP_OUTPUT_COUNTERS 525 if (m->m_next->m_flags & M_EXT) 526 TCP_OUTPUT_COUNTER_INCR(&tcp_output_refbig); 527 else 528 TCP_OUTPUT_COUNTER_INCR(&tcp_output_copybig); 529 #endif /* TCP_OUTPUT_COUNTERS */ 530 } 531 #endif 532 533 *mp = m; 534 return (0); 535 } 536 537 /* 538 * Tcp output routine: figure out what should be sent and send it. 539 */ 540 int 541 tcp_output(struct tcpcb *tp) 542 { 543 struct socket *so; 544 struct route *ro; 545 long len, win; 546 int off, flags, error; 547 struct mbuf *m; 548 struct ip *ip; 549 #ifdef INET6 550 struct ip6_hdr *ip6; 551 #endif 552 struct tcphdr *th; 553 u_char opt[MAX_TCPOPTLEN]; 554 unsigned optlen, hdrlen; 555 int idle, sendalot, txsegsize, rxsegsize; 556 int maxburst = TCP_MAXBURST; 557 int af; /* address family on the wire */ 558 int iphdrlen; 559 int sack_rxmit; 560 int sack_bytes_rxmt; 561 struct sackhole *p; 562 #ifdef TCP_SIGNATURE 563 int sigoff = 0; 564 #endif 565 566 #ifdef DIAGNOSTIC 567 if (tp->t_inpcb && tp->t_in6pcb) 568 panic("tcp_output: both t_inpcb and t_in6pcb are set"); 569 #endif 570 so = NULL; 571 ro = NULL; 572 if (tp->t_inpcb) { 573 so = tp->t_inpcb->inp_socket; 574 ro = &tp->t_inpcb->inp_route; 575 } 576 #ifdef INET6 577 else if (tp->t_in6pcb) { 578 so = tp->t_in6pcb->in6p_socket; 579 ro = (struct route *)&tp->t_in6pcb->in6p_route; 580 } 581 #endif 582 583 switch (af = tp->t_family) { 584 #ifdef INET 585 case AF_INET: 586 if (tp->t_inpcb) 587 break; 588 #ifdef INET6 589 /* mapped addr case */ 590 if (tp->t_in6pcb) 591 break; 592 #endif 593 return (EINVAL); 594 #endif 595 #ifdef INET6 596 case AF_INET6: 597 if (tp->t_in6pcb) 598 break; 599 return (EINVAL); 600 #endif 601 default: 602 return (EAFNOSUPPORT); 603 } 604 605 if (tcp_segsize(tp, &txsegsize, &rxsegsize)) 606 return (EMSGSIZE); 607 608 idle = (tp->snd_max == tp->snd_una); 609 610 /* 611 * Restart Window computation. From draft-floyd-incr-init-win-03: 612 * 613 * Optionally, a TCP MAY set the restart window to the 614 * minimum of the value used for the initial window and 615 * the current value of cwnd (in other words, using a 616 * larger value for the restart window should never increase 617 * the size of cwnd). 618 */ 619 if (tcp_cwm) { 620 /* 621 * Hughes/Touch/Heidemann Congestion Window Monitoring. 622 * Count the number of packets currently pending 623 * acknowledgement, and limit our congestion window 624 * to a pre-determined allowed burst size plus that count. 625 * This prevents bursting once all pending packets have 626 * been acknowledged (i.e. transmission is idle). 627 * 628 * XXX Link this to Initial Window? 629 */ 630 tp->snd_cwnd = min(tp->snd_cwnd, 631 (tcp_cwm_burstsize * txsegsize) + 632 (tp->snd_nxt - tp->snd_una)); 633 } else { 634 if (idle && (tcp_now - tp->t_rcvtime) >= tp->t_rxtcur) { 635 /* 636 * We have been idle for "a while" and no acks are 637 * expected to clock out any data we send -- 638 * slow start to get ack "clock" running again. 639 */ 640 int ss = tcp_init_win; 641 #ifdef INET 642 if (tp->t_inpcb && 643 in_localaddr(tp->t_inpcb->inp_faddr)) 644 ss = tcp_init_win_local; 645 #endif 646 #ifdef INET6 647 if (tp->t_in6pcb && 648 in6_localaddr(&tp->t_in6pcb->in6p_faddr)) 649 ss = tcp_init_win_local; 650 #endif 651 tp->snd_cwnd = min(tp->snd_cwnd, 652 TCP_INITIAL_WINDOW(ss, txsegsize)); 653 } 654 } 655 656 again: 657 /* 658 * Determine length of data that should be transmitted, and 659 * flags that should be used. If there is some data or critical 660 * controls (SYN, RST) to send, then transmit; otherwise, 661 * investigate further. 662 * 663 * Readjust SACK information to avoid resending duplicate data. 664 */ 665 if (TCP_SACK_ENABLED(tp) && SEQ_LT(tp->snd_nxt, tp->snd_max)) 666 tcp_sack_adjust(tp); 667 sendalot = 0; 668 off = tp->snd_nxt - tp->snd_una; 669 win = min(tp->snd_wnd, tp->snd_cwnd); 670 671 flags = tcp_outflags[tp->t_state]; 672 673 /* 674 * Send any SACK-generated retransmissions. If we're explicitly trying 675 * to send out new data (when sendalot is 1), bypass this function. 676 * If we retransmit in fast recovery mode, decrement snd_cwnd, since 677 * we're replacing a (future) new transmission with a retransmission 678 * now, and we previously incremented snd_cwnd in tcp_input(). 679 */ 680 /* 681 * Still in sack recovery , reset rxmit flag to zero. 682 */ 683 sack_rxmit = 0; 684 sack_bytes_rxmt = 0; 685 len = 0; 686 p = NULL; 687 if (!TCP_SACK_ENABLED(tp)) 688 goto after_sack_rexmit; 689 if ((tp->t_partialacks >= 0) && 690 (p = tcp_sack_output(tp, &sack_bytes_rxmt))) { 691 long cwin; 692 693 cwin = min(tp->snd_wnd, tp->snd_cwnd) - sack_bytes_rxmt; 694 if (cwin < 0) 695 cwin = 0; 696 /* Do not retransmit SACK segments beyond snd_recover */ 697 if (SEQ_GT(p->end, tp->snd_recover)) { 698 /* 699 * (At least) part of sack hole extends beyond 700 * snd_recover. Check to see if we can rexmit data 701 * for this hole. 702 */ 703 if (SEQ_GEQ(p->rxmit, tp->snd_recover)) { 704 /* 705 * Can't rexmit any more data for this hole. 706 * That data will be rexmitted in the next 707 * sack recovery episode, when snd_recover 708 * moves past p->rxmit. 709 */ 710 p = NULL; 711 goto after_sack_rexmit; 712 } else 713 /* Can rexmit part of the current hole */ 714 len = ((long)ulmin(cwin, 715 tp->snd_recover - p->rxmit)); 716 } else 717 len = ((long)ulmin(cwin, p->end - p->rxmit)); 718 off = p->rxmit - tp->snd_una; 719 if (len > 0) { 720 sack_rxmit = 1; 721 sendalot = 1; 722 } 723 } 724 after_sack_rexmit: 725 726 /* 727 * If in persist timeout with window of 0, send 1 byte. 728 * Otherwise, if window is small but nonzero 729 * and timer expired, we will send what we can 730 * and go to transmit state. 731 */ 732 if (tp->t_force) { 733 if (win == 0) { 734 /* 735 * If we still have some data to send, then 736 * clear the FIN bit. Usually this would 737 * happen below when it realizes that we 738 * aren't sending all the data. However, 739 * if we have exactly 1 byte of unset data, 740 * then it won't clear the FIN bit below, 741 * and if we are in persist state, we wind 742 * up sending the packet without recording 743 * that we sent the FIN bit. 744 * 745 * We can't just blindly clear the FIN bit, 746 * because if we don't have any more data 747 * to send then the probe will be the FIN 748 * itself. 749 */ 750 if (off < so->so_snd.sb_cc) 751 flags &= ~TH_FIN; 752 win = 1; 753 } else { 754 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 755 tp->t_rxtshift = 0; 756 } 757 } 758 759 if (!TCP_SACK_ENABLED(tp)) { 760 if (win < so->so_snd.sb_cc) { 761 len = win - off; 762 flags &= ~TH_FIN; 763 } else 764 len = so->so_snd.sb_cc - off; 765 } else if (sack_rxmit == 0) { 766 if (sack_bytes_rxmt != 0) { 767 long cwin; 768 769 /* 770 * We are inside of a SACK recovery episode and are 771 * sending new data, having retransmitted all the 772 * data possible in the scoreboard. 773 */ 774 len = ((long)ulmin(so->so_snd.sb_cc, tp->snd_wnd) 775 - off); 776 /* 777 * From FreeBSD: 778 * Don't remove this (len > 0) check ! 779 * We explicitly check for len > 0 here (although it 780 * isn't really necessary), to work around a gcc 781 * optimization issue - to force gcc to compute 782 * len above. Without this check, the computation 783 * of len is bungled by the optimizer. 784 */ 785 if (len > 0) { 786 cwin = tp->snd_cwnd - 787 (tp->snd_nxt - tp->sack_newdata) - 788 sack_bytes_rxmt; 789 if (cwin < 0) 790 cwin = 0; 791 len = lmin(len, cwin); 792 } 793 } else if (win < so->so_snd.sb_cc) { 794 len = win - off; 795 flags &= ~TH_FIN; 796 } else 797 len = so->so_snd.sb_cc - off; 798 } 799 800 if (len < 0) { 801 /* 802 * If FIN has been sent but not acked, 803 * but we haven't been called to retransmit, 804 * len will be -1. Otherwise, window shrank 805 * after we sent into it. If window shrank to 0, 806 * cancel pending retransmit, pull snd_nxt back 807 * to (closed) window, and set the persist timer 808 * if it isn't already going. If the window didn't 809 * close completely, just wait for an ACK. 810 * 811 * If we have a pending FIN, either it has already been 812 * transmitted or it is outside the window, so drop it. 813 * If the FIN has been transmitted, but this is not a 814 * retransmission, then len must be -1. Therefore we also 815 * prevent here the sending of `gratuitous FINs'. This 816 * eliminates the need to check for that case below (e.g. 817 * to back up snd_nxt before the FIN so that the sequence 818 * number is correct). 819 */ 820 len = 0; 821 flags &= ~TH_FIN; 822 if (win == 0) { 823 TCP_TIMER_DISARM(tp, TCPT_REXMT); 824 tp->t_rxtshift = 0; 825 tp->snd_nxt = tp->snd_una; 826 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 827 tcp_setpersist(tp); 828 } 829 } 830 if (len > txsegsize) { 831 len = txsegsize; 832 flags &= ~TH_FIN; 833 sendalot = 1; 834 } 835 if (sack_rxmit) { 836 if (SEQ_LT(p->rxmit + len, tp->snd_una + so->so_snd.sb_cc)) 837 flags &= ~TH_FIN; 838 } 839 840 win = sbspace(&so->so_rcv); 841 842 /* 843 * Sender silly window avoidance. If connection is idle 844 * and can send all data, a maximum segment, 845 * at least a maximum default-size segment do it, 846 * or are forced, do it; otherwise don't bother. 847 * If peer's buffer is tiny, then send 848 * when window is at least half open. 849 * If retransmitting (possibly after persist timer forced us 850 * to send into a small window), then must resend. 851 */ 852 if (len) { 853 if (len == txsegsize) 854 goto send; 855 if ((so->so_state & SS_MORETOCOME) == 0 && 856 ((idle || tp->t_flags & TF_NODELAY) && 857 len + off >= so->so_snd.sb_cc)) 858 goto send; 859 if (tp->t_force) 860 goto send; 861 if (len >= tp->max_sndwnd / 2) 862 goto send; 863 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) 864 goto send; 865 if (sack_rxmit) 866 goto send; 867 } 868 869 /* 870 * Compare available window to amount of window known to peer 871 * (as advertised window less next expected input). If the 872 * difference is at least twice the size of the largest segment 873 * we expect to receive (i.e. two segments) or at least 50% of 874 * the maximum possible window, then want to send a window update 875 * to peer. 876 */ 877 if (win > 0) { 878 /* 879 * "adv" is the amount we can increase the window, 880 * taking into account that we are limited by 881 * TCP_MAXWIN << tp->rcv_scale. 882 */ 883 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) - 884 (tp->rcv_adv - tp->rcv_nxt); 885 886 if (adv >= (long) (2 * rxsegsize)) 887 goto send; 888 if (2 * adv >= (long) so->so_rcv.sb_hiwat) 889 goto send; 890 } 891 892 /* 893 * Send if we owe peer an ACK. 894 */ 895 if (tp->t_flags & TF_ACKNOW) 896 goto send; 897 if (flags & (TH_SYN|TH_FIN|TH_RST)) 898 goto send; 899 if (SEQ_GT(tp->snd_up, tp->snd_una)) 900 goto send; 901 /* 902 * In SACK, it is possible for tcp_output to fail to send a segment 903 * after the retransmission timer has been turned off. Make sure 904 * that the retransmission timer is set. 905 */ 906 if (TCP_SACK_ENABLED(tp) && SEQ_GT(tp->snd_max, tp->snd_una) && 907 !TCP_TIMER_ISARMED(tp, TCPT_REXMT) && 908 !TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) { 909 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 910 goto just_return; 911 } 912 913 /* 914 * TCP window updates are not reliable, rather a polling protocol 915 * using ``persist'' packets is used to insure receipt of window 916 * updates. The three ``states'' for the output side are: 917 * idle not doing retransmits or persists 918 * persisting to move a small or zero window 919 * (re)transmitting and thereby not persisting 920 * 921 * tp->t_timer[TCPT_PERSIST] 922 * is set when we are in persist state. 923 * tp->t_force 924 * is set when we are called to send a persist packet. 925 * tp->t_timer[TCPT_REXMT] 926 * is set when we are retransmitting 927 * The output side is idle when both timers are zero. 928 * 929 * If send window is too small, there is data to transmit, and no 930 * retransmit or persist is pending, then go to persist state. 931 * If nothing happens soon, send when timer expires: 932 * if window is nonzero, transmit what we can, 933 * otherwise force out a byte. 934 */ 935 if (so->so_snd.sb_cc && TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 936 TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 937 tp->t_rxtshift = 0; 938 tcp_setpersist(tp); 939 } 940 941 /* 942 * No reason to send a segment, just return. 943 */ 944 just_return: 945 return (0); 946 947 send: 948 /* 949 * Before ESTABLISHED, force sending of initial options 950 * unless TCP set not to do any options. 951 * NOTE: we assume that the IP/TCP header plus TCP options 952 * always fit in a single mbuf, leaving room for a maximum 953 * link header, i.e. 954 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES 955 */ 956 optlen = 0; 957 switch (af) { 958 #ifdef INET 959 case AF_INET: 960 iphdrlen = sizeof(struct ip) + sizeof(struct tcphdr); 961 break; 962 #endif 963 #ifdef INET6 964 case AF_INET6: 965 iphdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 966 break; 967 #endif 968 default: /*pacify gcc*/ 969 iphdrlen = 0; 970 break; 971 } 972 hdrlen = iphdrlen; 973 if (flags & TH_SYN) { 974 struct rtentry *rt; 975 976 rt = NULL; 977 #ifdef INET 978 if (tp->t_inpcb) 979 rt = in_pcbrtentry(tp->t_inpcb); 980 #endif 981 #ifdef INET6 982 if (tp->t_in6pcb) 983 rt = in6_pcbrtentry(tp->t_in6pcb); 984 #endif 985 986 tp->snd_nxt = tp->iss; 987 tp->t_ourmss = tcp_mss_to_advertise(rt != NULL ? 988 rt->rt_ifp : NULL, af); 989 if ((tp->t_flags & TF_NOOPT) == 0) { 990 opt[0] = TCPOPT_MAXSEG; 991 opt[1] = 4; 992 opt[2] = (tp->t_ourmss >> 8) & 0xff; 993 opt[3] = tp->t_ourmss & 0xff; 994 optlen = 4; 995 996 if ((tp->t_flags & TF_REQ_SCALE) && 997 ((flags & TH_ACK) == 0 || 998 (tp->t_flags & TF_RCVD_SCALE))) { 999 *((u_int32_t *) (opt + optlen)) = htonl( 1000 TCPOPT_NOP << 24 | 1001 TCPOPT_WINDOW << 16 | 1002 TCPOLEN_WINDOW << 8 | 1003 tp->request_r_scale); 1004 optlen += 4; 1005 } 1006 if (tcp_do_sack) { 1007 u_int8_t *p = (u_int8_t *)(opt + optlen); 1008 1009 p[0] = TCPOPT_SACK_PERMITTED; 1010 p[1] = 2; 1011 p[2] = TCPOPT_NOP; 1012 p[3] = TCPOPT_NOP; 1013 optlen += 4; 1014 } 1015 } 1016 } 1017 1018 /* 1019 * Send a timestamp and echo-reply if this is a SYN and our side 1020 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side 1021 * and our peer have sent timestamps in our SYN's. 1022 */ 1023 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1024 (flags & TH_RST) == 0 && 1025 ((flags & (TH_SYN|TH_ACK)) == TH_SYN || 1026 (tp->t_flags & TF_RCVD_TSTMP))) { 1027 u_int32_t *lp = (u_int32_t *)(opt + optlen); 1028 1029 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1030 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1031 *lp++ = htonl(TCP_TIMESTAMP(tp)); 1032 *lp = htonl(tp->ts_recent); 1033 optlen += TCPOLEN_TSTAMP_APPA; 1034 } 1035 1036 /* 1037 * Tack on the SACK block if it is necessary. 1038 */ 1039 if (TCP_SACK_ENABLED(tp) && (tp->t_flags & TF_ACKNOW) 1040 && (tp->rcv_sack_num > 0)) { 1041 int sack_len, i; 1042 u_char *bp = (u_char *)(opt + optlen); 1043 u_int32_t *lp = (u_int32_t *)(bp + 4); 1044 1045 sack_len = tp->rcv_sack_num * 8 + 2; 1046 bp[0] = TCPOPT_NOP; 1047 bp[1] = TCPOPT_NOP; 1048 bp[2] = TCPOPT_SACK; 1049 bp[3] = sack_len; 1050 for (i = 0; i < tp->rcv_sack_num; i++) { 1051 *lp++ = htonl(tp->rcv_sack_block[i].left); 1052 *lp++ = htonl(tp->rcv_sack_block[i].right); 1053 } 1054 tp->rcv_sack_num = 0; 1055 optlen += sack_len + 2; 1056 } 1057 1058 #ifdef TCP_SIGNATURE 1059 #if defined(INET6) && defined(FAST_IPSEC) 1060 if (tp->t_family == AF_INET) 1061 #endif 1062 if (tp->t_flags & TF_SIGNATURE) { 1063 u_char *bp; 1064 /* 1065 * Initialize TCP-MD5 option (RFC2385) 1066 */ 1067 bp = (u_char *)opt + optlen; 1068 *bp++ = TCPOPT_SIGNATURE; 1069 *bp++ = TCPOLEN_SIGNATURE; 1070 sigoff = optlen + 2; 1071 bzero(bp, TCP_SIGLEN); 1072 bp += TCP_SIGLEN; 1073 optlen += TCPOLEN_SIGNATURE; 1074 /* 1075 * Terminate options list and maintain 32-bit alignment. 1076 */ 1077 *bp++ = TCPOPT_NOP; 1078 *bp++ = TCPOPT_EOL; 1079 optlen += 2; 1080 } 1081 #endif /* TCP_SIGNATURE */ 1082 1083 hdrlen += optlen; 1084 1085 #ifdef DIAGNOSTIC 1086 if (len > txsegsize) 1087 panic("tcp data to be sent is larger than segment"); 1088 if (max_linkhdr + hdrlen > MCLBYTES) 1089 panic("tcphdr too big"); 1090 #endif 1091 1092 /* 1093 * Grab a header mbuf, attaching a copy of data to 1094 * be transmitted, and initialize the header from 1095 * the template for sends on this connection. 1096 */ 1097 if (len) { 1098 error = tcp_build_datapkt(tp, so, off, len, hdrlen, &m); 1099 if (error) 1100 goto out; 1101 /* 1102 * If we're sending everything we've got, set PUSH. 1103 * (This will keep happy those implementations which only 1104 * give data to the user when a buffer fills or 1105 * a PUSH comes in.) 1106 */ 1107 if (off + len == so->so_snd.sb_cc) 1108 flags |= TH_PUSH; 1109 } else { 1110 if (tp->t_flags & TF_ACKNOW) 1111 tcpstat.tcps_sndacks++; 1112 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 1113 tcpstat.tcps_sndctrl++; 1114 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 1115 tcpstat.tcps_sndurg++; 1116 else 1117 tcpstat.tcps_sndwinup++; 1118 1119 MGETHDR(m, M_DONTWAIT, MT_HEADER); 1120 if (m != NULL && max_linkhdr + hdrlen > MHLEN) { 1121 MCLGET(m, M_DONTWAIT); 1122 if ((m->m_flags & M_EXT) == 0) { 1123 m_freem(m); 1124 m = NULL; 1125 } 1126 } 1127 if (m == NULL) { 1128 error = ENOBUFS; 1129 goto out; 1130 } 1131 MCLAIM(m, &tcp_tx_mowner); 1132 m->m_data += max_linkhdr; 1133 m->m_len = hdrlen; 1134 } 1135 m->m_pkthdr.rcvif = (struct ifnet *)0; 1136 switch (af) { 1137 #ifdef INET 1138 case AF_INET: 1139 ip = mtod(m, struct ip *); 1140 #ifdef INET6 1141 ip6 = NULL; 1142 #endif 1143 th = (struct tcphdr *)(ip + 1); 1144 break; 1145 #endif 1146 #ifdef INET6 1147 case AF_INET6: 1148 ip = NULL; 1149 ip6 = mtod(m, struct ip6_hdr *); 1150 th = (struct tcphdr *)(ip6 + 1); 1151 break; 1152 #endif 1153 default: /*pacify gcc*/ 1154 ip = NULL; 1155 #ifdef INET6 1156 ip6 = NULL; 1157 #endif 1158 th = NULL; 1159 break; 1160 } 1161 if (tp->t_template == 0) 1162 panic("tcp_output"); 1163 if (tp->t_template->m_len < iphdrlen) 1164 panic("tcp_output"); 1165 bcopy(mtod(tp->t_template, caddr_t), mtod(m, caddr_t), iphdrlen); 1166 1167 /* 1168 * If we are doing retransmissions, then snd_nxt will 1169 * not reflect the first unsent octet. For ACK only 1170 * packets, we do not want the sequence number of the 1171 * retransmitted packet, we want the sequence number 1172 * of the next unsent octet. So, if there is no data 1173 * (and no SYN or FIN), use snd_max instead of snd_nxt 1174 * when filling in ti_seq. But if we are in persist 1175 * state, snd_max might reflect one byte beyond the 1176 * right edge of the window, so use snd_nxt in that 1177 * case, since we know we aren't doing a retransmission. 1178 * (retransmit and persist are mutually exclusive...) 1179 */ 1180 if (TCP_SACK_ENABLED(tp) && sack_rxmit) { 1181 th->th_seq = htonl(p->rxmit); 1182 p->rxmit += len; 1183 } else { 1184 if (len || (flags & (TH_SYN|TH_FIN)) || 1185 TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) 1186 th->th_seq = htonl(tp->snd_nxt); 1187 else 1188 th->th_seq = htonl(tp->snd_max); 1189 } 1190 th->th_ack = htonl(tp->rcv_nxt); 1191 if (optlen) { 1192 bcopy((caddr_t)opt, (caddr_t)(th + 1), optlen); 1193 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1194 } 1195 th->th_flags = flags; 1196 /* 1197 * Calculate receive window. Don't shrink window, 1198 * but avoid silly window syndrome. 1199 */ 1200 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)rxsegsize) 1201 win = 0; 1202 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 1203 win = (long)TCP_MAXWIN << tp->rcv_scale; 1204 if (win < (long)(int32_t)(tp->rcv_adv - tp->rcv_nxt)) 1205 win = (long)(int32_t)(tp->rcv_adv - tp->rcv_nxt); 1206 th->th_win = htons((u_int16_t) (win>>tp->rcv_scale)); 1207 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 1208 u_int32_t urp = tp->snd_up - tp->snd_nxt; 1209 if (urp > IP_MAXPACKET) 1210 urp = IP_MAXPACKET; 1211 th->th_urp = htons((u_int16_t)urp); 1212 th->th_flags |= TH_URG; 1213 } else 1214 /* 1215 * If no urgent pointer to send, then we pull 1216 * the urgent pointer to the left edge of the send window 1217 * so that it doesn't drift into the send window on sequence 1218 * number wraparound. 1219 */ 1220 tp->snd_up = tp->snd_una; /* drag it along */ 1221 1222 #ifdef TCP_SIGNATURE 1223 #if defined(INET6) && defined(FAST_IPSEC) 1224 if (tp->t_family == AF_INET) /* XXX */ 1225 #endif 1226 if (sigoff && (tp->t_flags & TF_SIGNATURE)) { 1227 struct secasvar *sav; 1228 u_int8_t *sigp; 1229 1230 sav = tcp_signature_getsav(m, th); 1231 1232 if (sav == NULL) { 1233 if (m) 1234 m_freem(m); 1235 return (EPERM); 1236 } 1237 1238 m->m_pkthdr.len = hdrlen + len; 1239 sigp = (caddr_t)th + sizeof(*th) + sigoff; 1240 tcp_signature(m, th, (caddr_t)th - mtod(m, caddr_t), sav, sigp); 1241 1242 key_sa_recordxfer(sav, m); 1243 #ifdef FAST_IPSEC 1244 KEY_FREESAV(&sav); 1245 #else 1246 key_freesav(sav); 1247 #endif 1248 } 1249 #endif 1250 1251 /* 1252 * Set ourselves up to be checksummed just before the packet 1253 * hits the wire. Maybe skip checksums on loopback interfaces. 1254 */ 1255 switch (af) { 1256 #ifdef INET 1257 case AF_INET: 1258 if (__predict_true(ro->ro_rt == NULL || 1259 !(ro->ro_rt->rt_ifp->if_flags & 1260 IFF_LOOPBACK) || 1261 tcp_do_loopback_cksum)) 1262 m->m_pkthdr.csum_flags = M_CSUM_TCPv4; 1263 else 1264 m->m_pkthdr.csum_flags = 0; 1265 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1266 if (len + optlen) { 1267 /* Fixup the pseudo-header checksum. */ 1268 /* XXXJRT Not IP Jumbogram safe. */ 1269 th->th_sum = in_cksum_addword(th->th_sum, 1270 htons((u_int16_t) (len + optlen))); 1271 } 1272 break; 1273 #endif 1274 #ifdef INET6 1275 case AF_INET6: 1276 /* 1277 * XXX Actually delaying the checksum is Hard 1278 * XXX (well, maybe not for Itojun, but it is 1279 * XXX for me), but we can still take advantage 1280 * XXX of the cached pseudo-header checksum. 1281 */ 1282 /* equals to hdrlen + len */ 1283 m->m_pkthdr.len = sizeof(struct ip6_hdr) 1284 + sizeof(struct tcphdr) + optlen + len; 1285 #ifdef notyet 1286 if (__predict_true(ro->ro_rt == NULL || 1287 !(ro->ro_rt->rt_ifp->if_flags & 1288 IFF_LOOPBACK) || 1289 tcp_do_loopback_cksum)) 1290 m->m_pkthdr.csum_flags = M_CSUM_TCPv6; 1291 else 1292 m->m_pkthdr.csum_flags = 0; 1293 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1294 #endif 1295 if (len + optlen) { 1296 /* Fixup the pseudo-header checksum. */ 1297 /* XXXJRT: Not IPv6 Jumbogram safe. */ 1298 th->th_sum = in_cksum_addword(th->th_sum, 1299 htons((u_int16_t) (len + optlen))); 1300 } 1301 #ifndef notyet 1302 if (__predict_true(ro->ro_rt == NULL || 1303 !(ro->ro_rt->rt_ifp->if_flags & 1304 IFF_LOOPBACK) || 1305 tcp_do_loopback_cksum)) 1306 th->th_sum = in6_cksum(m, 0, sizeof(struct ip6_hdr), 1307 sizeof(struct tcphdr) + optlen + len); 1308 #endif 1309 break; 1310 #endif 1311 } 1312 1313 /* 1314 * In transmit state, time the transmission and arrange for 1315 * the retransmit. In persist state, just set snd_max. 1316 */ 1317 if (tp->t_force == 0 || TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 1318 tcp_seq startseq = tp->snd_nxt; 1319 1320 /* 1321 * Advance snd_nxt over sequence space of this segment. 1322 * There are no states in which we send both a SYN and a FIN, 1323 * so we collapse the tests for these flags. 1324 */ 1325 if (flags & (TH_SYN|TH_FIN)) 1326 tp->snd_nxt++; 1327 if (sack_rxmit) 1328 goto timer; 1329 tp->snd_nxt += len; 1330 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 1331 tp->snd_max = tp->snd_nxt; 1332 /* 1333 * Time this transmission if not a retransmission and 1334 * not currently timing anything. 1335 */ 1336 if (tp->t_rtttime == 0) { 1337 tp->t_rtttime = tcp_now; 1338 tp->t_rtseq = startseq; 1339 tcpstat.tcps_segstimed++; 1340 } 1341 } 1342 1343 /* 1344 * Set retransmit timer if not currently set, 1345 * and not doing an ack or a keep-alive probe. 1346 * Initial value for retransmit timer is smoothed 1347 * round-trip time + 2 * round-trip time variance. 1348 * Initialize shift counter which is used for backoff 1349 * of retransmit time. 1350 */ 1351 timer: 1352 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 1353 ((sack_rxmit && tp->snd_nxt != tp->snd_max) || 1354 tp->snd_nxt != tp->snd_una)) { 1355 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) { 1356 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 1357 tp->t_rxtshift = 0; 1358 } 1359 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1360 } 1361 } else 1362 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max)) 1363 tp->snd_max = tp->snd_nxt + len; 1364 1365 #ifdef TCP_DEBUG 1366 /* 1367 * Trace. 1368 */ 1369 if (so->so_options & SO_DEBUG) 1370 tcp_trace(TA_OUTPUT, tp->t_state, tp, m, 0); 1371 #endif 1372 1373 /* 1374 * Fill in IP length and desired time to live and 1375 * send to IP level. There should be a better way 1376 * to handle ttl and tos; we could keep them in 1377 * the template, but need a way to checksum without them. 1378 */ 1379 m->m_pkthdr.len = hdrlen + len; 1380 1381 switch (af) { 1382 #ifdef INET 1383 case AF_INET: 1384 ip->ip_len = htons(m->m_pkthdr.len); 1385 if (tp->t_inpcb) { 1386 ip->ip_ttl = tp->t_inpcb->inp_ip.ip_ttl; 1387 ip->ip_tos = tp->t_inpcb->inp_ip.ip_tos; 1388 } 1389 #ifdef INET6 1390 else if (tp->t_in6pcb) { 1391 ip->ip_ttl = in6_selecthlim(tp->t_in6pcb, NULL); /*XXX*/ 1392 ip->ip_tos = 0; /*XXX*/ 1393 } 1394 #endif 1395 break; 1396 #endif 1397 #ifdef INET6 1398 case AF_INET6: 1399 ip6->ip6_nxt = IPPROTO_TCP; 1400 if (tp->t_in6pcb) { 1401 /* 1402 * we separately set hoplimit for every segment, since 1403 * the user might want to change the value via 1404 * setsockopt. Also, desired default hop limit might 1405 * be changed via Neighbor Discovery. 1406 */ 1407 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, 1408 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 1409 } 1410 /* ip6->ip6_flow = ??? */ 1411 /* ip6_plen will be filled in ip6_output(). */ 1412 break; 1413 #endif 1414 } 1415 1416 switch (af) { 1417 #ifdef INET 1418 case AF_INET: 1419 { 1420 struct mbuf *opts; 1421 1422 if (tp->t_inpcb) 1423 opts = tp->t_inpcb->inp_options; 1424 else 1425 opts = NULL; 1426 error = ip_output(m, opts, ro, 1427 (tp->t_mtudisc ? IP_MTUDISC : 0) | 1428 (so->so_options & SO_DONTROUTE), 1429 (struct ip_moptions *)0, so); 1430 break; 1431 } 1432 #endif 1433 #ifdef INET6 1434 case AF_INET6: 1435 { 1436 struct ip6_pktopts *opts; 1437 1438 if (tp->t_in6pcb) 1439 opts = tp->t_in6pcb->in6p_outputopts; 1440 else 1441 opts = NULL; 1442 error = ip6_output(m, opts, (struct route_in6 *)ro, 1443 so->so_options & SO_DONTROUTE, 1444 (struct ip6_moptions *)0, so, NULL); 1445 break; 1446 } 1447 #endif 1448 default: 1449 error = EAFNOSUPPORT; 1450 break; 1451 } 1452 if (error) { 1453 out: 1454 if (error == ENOBUFS) { 1455 tcpstat.tcps_selfquench++; 1456 #ifdef INET 1457 if (tp->t_inpcb) 1458 tcp_quench(tp->t_inpcb, 0); 1459 #endif 1460 #ifdef INET6 1461 if (tp->t_in6pcb) 1462 tcp6_quench(tp->t_in6pcb, 0); 1463 #endif 1464 error = 0; 1465 } else if ((error == EHOSTUNREACH || error == ENETDOWN) && 1466 TCPS_HAVERCVDSYN(tp->t_state)) { 1467 tp->t_softerror = error; 1468 error = 0; 1469 } 1470 1471 /* Back out the seqence number advance. */ 1472 if (sack_rxmit) 1473 p->rxmit -= len; 1474 1475 /* Restart the delayed ACK timer, if necessary. */ 1476 if (tp->t_flags & TF_DELACK) 1477 TCP_RESTART_DELACK(tp); 1478 1479 return (error); 1480 } 1481 tcpstat.tcps_sndtotal++; 1482 if (tp->t_flags & TF_DELACK) 1483 tcpstat.tcps_delack++; 1484 1485 /* 1486 * Data sent (as far as we can tell). 1487 * If this advertises a larger window than any other segment, 1488 * then remember the size of the advertised window. 1489 * Any pending ACK has now been sent. 1490 */ 1491 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 1492 tp->rcv_adv = tp->rcv_nxt + win; 1493 tp->last_ack_sent = tp->rcv_nxt; 1494 tp->t_flags &= ~TF_ACKNOW; 1495 TCP_CLEAR_DELACK(tp); 1496 #ifdef DIAGNOSTIC 1497 if (maxburst < 0) 1498 printf("tcp_output: maxburst exceeded by %d\n", -maxburst); 1499 #endif 1500 if (sendalot && (!tcp_do_newreno || --maxburst)) 1501 goto again; 1502 return (0); 1503 } 1504 1505 void 1506 tcp_setpersist(struct tcpcb *tp) 1507 { 1508 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2); 1509 int nticks; 1510 1511 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT)) 1512 panic("tcp_output REXMT"); 1513 /* 1514 * Start/restart persistance timer. 1515 */ 1516 if (t < tp->t_rttmin) 1517 t = tp->t_rttmin; 1518 TCPT_RANGESET(nticks, t * tcp_backoff[tp->t_rxtshift], 1519 TCPTV_PERSMIN, TCPTV_PERSMAX); 1520 TCP_TIMER_ARM(tp, TCPT_PERSIST, nticks); 1521 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 1522 tp->t_rxtshift++; 1523 } 1524
Indexes created Tue Sep 23 14:10:03 GMT 2025