tcp_output.c revision 1.124
1 /* $NetBSD: tcp_output.c,v 1.124 2005/03/07 09:32:51 yamt 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 1.124 2005/03/07 09:32:51 yamt 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 /* 209 * Knob to enable Congestion Window Monitoring, and control the 210 * the burst size it allows. Default burst is 4 packets, per 211 * the Internet draft. 212 */ 213 int tcp_cwm = 0; 214 int tcp_cwm_burstsize = 4; 215 216 #ifdef TCP_OUTPUT_COUNTERS 217 #include <sys/device.h> 218 219 extern struct evcnt tcp_output_bigheader; 220 extern struct evcnt tcp_output_predict_hit; 221 extern struct evcnt tcp_output_predict_miss; 222 extern struct evcnt tcp_output_copysmall; 223 extern struct evcnt tcp_output_copybig; 224 extern struct evcnt tcp_output_refbig; 225 226 #define TCP_OUTPUT_COUNTER_INCR(ev) (ev)->ev_count++ 227 #else 228 229 #define TCP_OUTPUT_COUNTER_INCR(ev) /* nothing */ 230 231 #endif /* TCP_OUTPUT_COUNTERS */ 232 233 static 234 #ifndef GPROF 235 __inline 236 #endif 237 int 238 tcp_segsize(struct tcpcb *tp, int *txsegsizep, int *rxsegsizep) 239 { 240 #ifdef INET 241 struct inpcb *inp = tp->t_inpcb; 242 #endif 243 #ifdef INET6 244 struct in6pcb *in6p = tp->t_in6pcb; 245 #endif 246 struct socket *so = NULL; 247 struct rtentry *rt; 248 struct ifnet *ifp; 249 int size; 250 int iphlen; 251 int optlen; 252 253 #ifdef DIAGNOSTIC 254 if (tp->t_inpcb && tp->t_in6pcb) 255 panic("tcp_segsize: both t_inpcb and t_in6pcb are set"); 256 #endif 257 switch (tp->t_family) { 258 #ifdef INET 259 case AF_INET: 260 iphlen = sizeof(struct ip); 261 break; 262 #endif 263 #ifdef INET6 264 case AF_INET6: 265 iphlen = sizeof(struct ip6_hdr); 266 break; 267 #endif 268 default: 269 size = tcp_mssdflt; 270 goto out; 271 } 272 273 rt = NULL; 274 #ifdef INET 275 if (inp) { 276 rt = in_pcbrtentry(inp); 277 so = inp->inp_socket; 278 } 279 #endif 280 #ifdef INET6 281 if (in6p) { 282 rt = in6_pcbrtentry(in6p); 283 so = in6p->in6p_socket; 284 } 285 #endif 286 if (rt == NULL) { 287 size = tcp_mssdflt; 288 goto out; 289 } 290 291 ifp = rt->rt_ifp; 292 293 size = tcp_mssdflt; 294 if (tp->t_mtudisc && rt->rt_rmx.rmx_mtu != 0) { 295 #ifdef INET6 296 if (in6p && rt->rt_rmx.rmx_mtu < IPV6_MMTU) { 297 /* 298 * RFC2460 section 5, last paragraph: if path MTU is 299 * smaller than 1280, use 1280 as packet size and 300 * attach fragment header. 301 */ 302 size = IPV6_MMTU - iphlen - sizeof(struct ip6_frag) - 303 sizeof(struct tcphdr); 304 } else 305 size = rt->rt_rmx.rmx_mtu - iphlen - 306 sizeof(struct tcphdr); 307 #else 308 size = rt->rt_rmx.rmx_mtu - iphlen - sizeof(struct tcphdr); 309 #endif 310 } else if (ifp->if_flags & IFF_LOOPBACK) 311 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 312 #ifdef INET 313 else if (inp && tp->t_mtudisc) 314 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 315 else if (inp && in_localaddr(inp->inp_faddr)) 316 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 317 #endif 318 #ifdef INET6 319 else if (in6p) { 320 #ifdef INET 321 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) { 322 /* mapped addr case */ 323 struct in_addr d; 324 bcopy(&in6p->in6p_faddr.s6_addr32[3], &d, sizeof(d)); 325 if (tp->t_mtudisc || in_localaddr(d)) 326 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 327 } else 328 #endif 329 { 330 /* 331 * for IPv6, path MTU discovery is always turned on, 332 * or the node must use packet size <= 1280. 333 */ 334 size = tp->t_mtudisc ? IN6_LINKMTU(ifp) : IPV6_MMTU; 335 size -= (iphlen + sizeof(struct tcphdr)); 336 } 337 } 338 #endif 339 out: 340 /* 341 * Now we must make room for whatever extra TCP/IP options are in 342 * the packet. 343 */ 344 optlen = tcp_optlen(tp); 345 346 /* 347 * XXX tp->t_ourmss should have the right size, but without this code 348 * fragmentation will occur... need more investigation 349 */ 350 #ifdef INET 351 if (inp) { 352 #if defined(IPSEC) || defined(FAST_IPSEC) 353 if (! IPSEC_PCB_SKIP_IPSEC(inp->inp_sp, IPSEC_DIR_OUTBOUND)) 354 optlen += ipsec4_hdrsiz_tcp(tp); 355 #endif 356 optlen += ip_optlen(inp); 357 } 358 #endif 359 #ifdef INET6 360 #ifdef INET 361 if (in6p && tp->t_family == AF_INET) { 362 #if defined(IPSEC) || defined(FAST_IPSEC) 363 if (! IPSEC_PCB_SKIP_IPSEC(in6p->in6p_sp, IPSEC_DIR_OUTBOUND)) 364 optlen += ipsec4_hdrsiz_tcp(tp); 365 #endif 366 /* XXX size -= ip_optlen(in6p); */ 367 } else 368 #endif 369 if (in6p && tp->t_family == AF_INET6) { 370 #ifdef IPSEC 371 if (! IPSEC_PCB_SKIP_IPSEC(in6p->in6p_sp, IPSEC_DIR_OUTBOUND)) 372 optlen += ipsec6_hdrsiz_tcp(tp); 373 #endif 374 optlen += ip6_optlen(in6p); 375 } 376 #endif 377 size -= optlen; 378 379 /* there may not be any room for data if mtu is too small */ 380 if (size < 0) 381 return (EMSGSIZE); 382 383 /* 384 * *rxsegsizep holds *estimated* inbound segment size (estimation 385 * assumes that path MTU is the same for both ways). this is only 386 * for silly window avoidance, do not use the value for other purposes. 387 * 388 * ipseclen is subtracted from both sides, this may not be right. 389 * I'm not quite sure about this (could someone comment). 390 */ 391 *txsegsizep = min(tp->t_peermss - optlen, size); 392 /* 393 * Never send more than half a buffer full. This insures that we can 394 * always keep 2 packets on the wire, no matter what SO_SNDBUF is, and 395 * therefore acks will never be delayed unless we run out of data to 396 * transmit. 397 */ 398 if (so) 399 *txsegsizep = min(so->so_snd.sb_hiwat >> 1, *txsegsizep); 400 *rxsegsizep = min(tp->t_ourmss - optlen, size); 401 402 if (*txsegsizep != tp->t_segsz) { 403 /* 404 * If the new segment size is larger, we don't want to 405 * mess up the congestion window, but if it is smaller 406 * we'll have to reduce the congestion window to ensure 407 * that we don't get into trouble with initial windows 408 * and the rest. In any case, if the segment size 409 * has changed, chances are the path has, too, and 410 * our congestion window will be different. 411 */ 412 if (*txsegsizep < tp->t_segsz) { 413 tp->snd_cwnd = max((tp->snd_cwnd / tp->t_segsz) 414 * *txsegsizep, *txsegsizep); 415 tp->snd_ssthresh = max((tp->snd_ssthresh / tp->t_segsz) 416 * *txsegsizep, *txsegsizep); 417 } 418 tp->t_segsz = *txsegsizep; 419 } 420 421 return (0); 422 } 423 424 static 425 #ifndef GPROF 426 __inline 427 #endif 428 int 429 tcp_build_datapkt(struct tcpcb *tp, struct socket *so, int off, 430 long len, int hdrlen, struct mbuf **mp) 431 { 432 struct mbuf *m, *m0; 433 434 if (tp->t_force && len == 1) 435 tcpstat.tcps_sndprobe++; 436 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { 437 tcpstat.tcps_sndrexmitpack++; 438 tcpstat.tcps_sndrexmitbyte += len; 439 } else { 440 tcpstat.tcps_sndpack++; 441 tcpstat.tcps_sndbyte += len; 442 } 443 #ifdef notyet 444 if ((m = m_copypack(so->so_snd.sb_mb, off, 445 (int)len, max_linkhdr + hdrlen)) == 0) 446 return (ENOBUFS); 447 /* 448 * m_copypack left space for our hdr; use it. 449 */ 450 m->m_len += hdrlen; 451 m->m_data -= hdrlen; 452 #else 453 MGETHDR(m, M_DONTWAIT, MT_HEADER); 454 if (__predict_false(m == NULL)) 455 return (ENOBUFS); 456 MCLAIM(m, &tcp_tx_mowner); 457 458 /* 459 * XXX Because other code assumes headers will fit in 460 * XXX one header mbuf. 461 * 462 * (This code should almost *never* be run.) 463 */ 464 if (__predict_false((max_linkhdr + hdrlen) > MHLEN)) { 465 TCP_OUTPUT_COUNTER_INCR(&tcp_output_bigheader); 466 MCLGET(m, M_DONTWAIT); 467 if ((m->m_flags & M_EXT) == 0) { 468 m_freem(m); 469 return (ENOBUFS); 470 } 471 } 472 473 m->m_data += max_linkhdr; 474 m->m_len = hdrlen; 475 476 /* 477 * To avoid traversing the whole sb_mb chain for correct 478 * data to send, remember last sent mbuf, its offset and 479 * the sent size. When called the next time, see if the 480 * data to send is directly following the previous transfer. 481 * This is important for large TCP windows. 482 */ 483 if (off == 0 || tp->t_lastm == NULL || 484 (tp->t_lastoff + tp->t_lastlen) != off) { 485 TCP_OUTPUT_COUNTER_INCR(&tcp_output_predict_miss); 486 /* 487 * Either a new packet or a retransmit. 488 * Start from the beginning. 489 */ 490 tp->t_lastm = so->so_snd.sb_mb; 491 tp->t_inoff = off; 492 } else { 493 TCP_OUTPUT_COUNTER_INCR(&tcp_output_predict_hit); 494 tp->t_inoff += tp->t_lastlen; 495 } 496 497 /* Traverse forward to next packet */ 498 while (tp->t_inoff > 0) { 499 if (tp->t_lastm == NULL) 500 panic("tp->t_lastm == NULL"); 501 if (tp->t_inoff < tp->t_lastm->m_len) 502 break; 503 tp->t_inoff -= tp->t_lastm->m_len; 504 tp->t_lastm = tp->t_lastm->m_next; 505 } 506 507 tp->t_lastoff = off; 508 tp->t_lastlen = len; 509 m0 = tp->t_lastm; 510 off = tp->t_inoff; 511 512 if (len <= M_TRAILINGSPACE(m)) { 513 m_copydata(m0, off, (int) len, mtod(m, caddr_t) + hdrlen); 514 m->m_len += len; 515 TCP_OUTPUT_COUNTER_INCR(&tcp_output_copysmall); 516 } else { 517 m->m_next = m_copy(m0, off, (int) len); 518 if (m->m_next == NULL) { 519 m_freem(m); 520 return (ENOBUFS); 521 } 522 #ifdef TCP_OUTPUT_COUNTERS 523 if (m->m_next->m_flags & M_EXT) 524 TCP_OUTPUT_COUNTER_INCR(&tcp_output_refbig); 525 else 526 TCP_OUTPUT_COUNTER_INCR(&tcp_output_copybig); 527 #endif /* TCP_OUTPUT_COUNTERS */ 528 } 529 #endif 530 531 *mp = m; 532 return (0); 533 } 534 535 /* 536 * Tcp output routine: figure out what should be sent and send it. 537 */ 538 int 539 tcp_output(struct tcpcb *tp) 540 { 541 struct socket *so; 542 struct route *ro; 543 long len, win; 544 int off, flags, error; 545 struct mbuf *m; 546 struct ip *ip; 547 #ifdef INET6 548 struct ip6_hdr *ip6; 549 #endif 550 struct tcphdr *th; 551 u_char opt[MAX_TCPOPTLEN]; 552 unsigned optlen, hdrlen; 553 int idle, sendalot, txsegsize, rxsegsize; 554 int maxburst = TCP_MAXBURST; 555 int af; /* address family on the wire */ 556 int iphdrlen; 557 int use_tso; 558 int sack_rxmit; 559 int sack_bytes_rxmt; 560 struct sackhole *p; 561 #ifdef TCP_SIGNATURE 562 int sigoff = 0; 563 #endif 564 565 #ifdef DIAGNOSTIC 566 if (tp->t_inpcb && tp->t_in6pcb) 567 panic("tcp_output: both t_inpcb and t_in6pcb are set"); 568 #endif 569 so = NULL; 570 ro = NULL; 571 if (tp->t_inpcb) { 572 so = tp->t_inpcb->inp_socket; 573 ro = &tp->t_inpcb->inp_route; 574 } 575 #ifdef INET6 576 else if (tp->t_in6pcb) { 577 so = tp->t_in6pcb->in6p_socket; 578 ro = (struct route *)&tp->t_in6pcb->in6p_route; 579 } 580 #endif 581 582 switch (af = tp->t_family) { 583 #ifdef INET 584 case AF_INET: 585 if (tp->t_inpcb) 586 break; 587 #ifdef INET6 588 /* mapped addr case */ 589 if (tp->t_in6pcb) 590 break; 591 #endif 592 return (EINVAL); 593 #endif 594 #ifdef INET6 595 case AF_INET6: 596 if (tp->t_in6pcb) 597 break; 598 return (EINVAL); 599 #endif 600 default: 601 return (EAFNOSUPPORT); 602 } 603 604 if (tcp_segsize(tp, &txsegsize, &rxsegsize)) 605 return (EMSGSIZE); 606 607 idle = (tp->snd_max == tp->snd_una); 608 609 /* 610 * Determine if we can use TCP segmentation offload: 611 * - If we're using IPv4 612 * - If there is not an IPsec policy that prevents it 613 * - If the interface can do it 614 */ 615 use_tso = tp->t_inpcb != NULL && 616 #if defined(IPSEC) || defined(FAST_IPSEC) 617 IPSEC_PCB_SKIP_IPSEC(tp->t_inpcb->inp_sp, 618 IPSEC_DIR_OUTBOUND) && 619 #endif 620 tp->t_inpcb->inp_route.ro_rt != NULL && 621 (tp->t_inpcb->inp_route.ro_rt->rt_ifp->if_capenable & 622 IFCAP_TSOv4) != 0; 623 624 /* 625 * Restart Window computation. From draft-floyd-incr-init-win-03: 626 * 627 * Optionally, a TCP MAY set the restart window to the 628 * minimum of the value used for the initial window and 629 * the current value of cwnd (in other words, using a 630 * larger value for the restart window should never increase 631 * the size of cwnd). 632 */ 633 if (tcp_cwm) { 634 /* 635 * Hughes/Touch/Heidemann Congestion Window Monitoring. 636 * Count the number of packets currently pending 637 * acknowledgement, and limit our congestion window 638 * to a pre-determined allowed burst size plus that count. 639 * This prevents bursting once all pending packets have 640 * been acknowledged (i.e. transmission is idle). 641 * 642 * XXX Link this to Initial Window? 643 */ 644 tp->snd_cwnd = min(tp->snd_cwnd, 645 (tcp_cwm_burstsize * txsegsize) + 646 (tp->snd_nxt - tp->snd_una)); 647 } else { 648 if (idle && (tcp_now - tp->t_rcvtime) >= tp->t_rxtcur) { 649 /* 650 * We have been idle for "a while" and no acks are 651 * expected to clock out any data we send -- 652 * slow start to get ack "clock" running again. 653 */ 654 int ss = tcp_init_win; 655 #ifdef INET 656 if (tp->t_inpcb && 657 in_localaddr(tp->t_inpcb->inp_faddr)) 658 ss = tcp_init_win_local; 659 #endif 660 #ifdef INET6 661 if (tp->t_in6pcb && 662 in6_localaddr(&tp->t_in6pcb->in6p_faddr)) 663 ss = tcp_init_win_local; 664 #endif 665 tp->snd_cwnd = min(tp->snd_cwnd, 666 TCP_INITIAL_WINDOW(ss, txsegsize)); 667 } 668 } 669 670 again: 671 /* 672 * Determine length of data that should be transmitted, and 673 * flags that should be used. If there is some data or critical 674 * controls (SYN, RST) to send, then transmit; otherwise, 675 * investigate further. 676 * 677 * Readjust SACK information to avoid resending duplicate data. 678 */ 679 if (TCP_SACK_ENABLED(tp) && SEQ_LT(tp->snd_nxt, tp->snd_max)) 680 tcp_sack_adjust(tp); 681 sendalot = 0; 682 off = tp->snd_nxt - tp->snd_una; 683 win = min(tp->snd_wnd, tp->snd_cwnd); 684 685 flags = tcp_outflags[tp->t_state]; 686 687 /* 688 * Send any SACK-generated retransmissions. If we're explicitly trying 689 * to send out new data (when sendalot is 1), bypass this function. 690 * If we retransmit in fast recovery mode, decrement snd_cwnd, since 691 * we're replacing a (future) new transmission with a retransmission 692 * now, and we previously incremented snd_cwnd in tcp_input(). 693 */ 694 /* 695 * Still in sack recovery , reset rxmit flag to zero. 696 */ 697 sack_rxmit = 0; 698 sack_bytes_rxmt = 0; 699 len = 0; 700 p = NULL; 701 do { 702 long cwin; 703 if (!TCP_SACK_ENABLED(tp)) 704 break; 705 if (tp->t_partialacks < 0) 706 break; 707 p = tcp_sack_output(tp, &sack_bytes_rxmt); 708 if (p == NULL) 709 break; 710 711 cwin = min(tp->snd_wnd, tp->snd_cwnd) - sack_bytes_rxmt; 712 if (cwin < 0) 713 cwin = 0; 714 /* Do not retransmit SACK segments beyond snd_recover */ 715 if (SEQ_GT(p->end, tp->snd_recover)) { 716 /* 717 * (At least) part of sack hole extends beyond 718 * snd_recover. Check to see if we can rexmit data 719 * for this hole. 720 */ 721 if (SEQ_GEQ(p->rxmit, tp->snd_recover)) { 722 /* 723 * Can't rexmit any more data for this hole. 724 * That data will be rexmitted in the next 725 * sack recovery episode, when snd_recover 726 * moves past p->rxmit. 727 */ 728 p = NULL; 729 break; 730 } 731 /* Can rexmit part of the current hole */ 732 len = ((long)ulmin(cwin, tp->snd_recover - p->rxmit)); 733 } else 734 len = ((long)ulmin(cwin, p->end - p->rxmit)); 735 off = p->rxmit - tp->snd_una; 736 if (len > 0) { 737 sack_rxmit = 1; 738 sendalot = 1; 739 } 740 } while (/*CONSTCOND*/0); 741 742 /* 743 * If in persist timeout with window of 0, send 1 byte. 744 * Otherwise, if window is small but nonzero 745 * and timer expired, we will send what we can 746 * and go to transmit state. 747 */ 748 if (tp->t_force) { 749 if (win == 0) { 750 /* 751 * If we still have some data to send, then 752 * clear the FIN bit. Usually this would 753 * happen below when it realizes that we 754 * aren't sending all the data. However, 755 * if we have exactly 1 byte of unset data, 756 * then it won't clear the FIN bit below, 757 * and if we are in persist state, we wind 758 * up sending the packet without recording 759 * that we sent the FIN bit. 760 * 761 * We can't just blindly clear the FIN bit, 762 * because if we don't have any more data 763 * to send then the probe will be the FIN 764 * itself. 765 */ 766 if (off < so->so_snd.sb_cc) 767 flags &= ~TH_FIN; 768 win = 1; 769 } else { 770 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 771 tp->t_rxtshift = 0; 772 } 773 } 774 775 if (!TCP_SACK_ENABLED(tp)) { 776 if (win < so->so_snd.sb_cc) { 777 len = win - off; 778 flags &= ~TH_FIN; 779 } else 780 len = so->so_snd.sb_cc - off; 781 } else if (sack_rxmit == 0) { 782 if (sack_bytes_rxmt != 0) { 783 long cwin; 784 785 /* 786 * We are inside of a SACK recovery episode and are 787 * sending new data, having retransmitted all the 788 * data possible in the scoreboard. 789 */ 790 len = ((long)ulmin(so->so_snd.sb_cc, tp->snd_wnd) 791 - off); 792 /* 793 * From FreeBSD: 794 * Don't remove this (len > 0) check ! 795 * We explicitly check for len > 0 here (although it 796 * isn't really necessary), to work around a gcc 797 * optimization issue - to force gcc to compute 798 * len above. Without this check, the computation 799 * of len is bungled by the optimizer. 800 */ 801 if (len > 0) { 802 cwin = tp->snd_cwnd - 803 (tp->snd_nxt - tp->sack_newdata) - 804 sack_bytes_rxmt; 805 if (cwin < 0) 806 cwin = 0; 807 len = lmin(len, cwin); 808 } 809 } else if (win < so->so_snd.sb_cc) { 810 len = win - off; 811 flags &= ~TH_FIN; 812 } else 813 len = so->so_snd.sb_cc - off; 814 } 815 816 if (len < 0) { 817 /* 818 * If FIN has been sent but not acked, 819 * but we haven't been called to retransmit, 820 * len will be -1. Otherwise, window shrank 821 * after we sent into it. If window shrank to 0, 822 * cancel pending retransmit, pull snd_nxt back 823 * to (closed) window, and set the persist timer 824 * if it isn't already going. If the window didn't 825 * close completely, just wait for an ACK. 826 * 827 * If we have a pending FIN, either it has already been 828 * transmitted or it is outside the window, so drop it. 829 * If the FIN has been transmitted, but this is not a 830 * retransmission, then len must be -1. Therefore we also 831 * prevent here the sending of `gratuitous FINs'. This 832 * eliminates the need to check for that case below (e.g. 833 * to back up snd_nxt before the FIN so that the sequence 834 * number is correct). 835 */ 836 len = 0; 837 flags &= ~TH_FIN; 838 if (win == 0) { 839 TCP_TIMER_DISARM(tp, TCPT_REXMT); 840 tp->t_rxtshift = 0; 841 tp->snd_nxt = tp->snd_una; 842 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 843 tcp_setpersist(tp); 844 } 845 } 846 if (len > txsegsize) { 847 if (use_tso) { 848 /* 849 * Truncate TSO transfers to IP_MAXPACKET, and make 850 * sure that we send equal size transfers down the 851 * stack (rather than big-small-big-small-...). 852 */ 853 len = (min(len, IP_MAXPACKET) / txsegsize) * txsegsize; 854 } else 855 len = txsegsize; 856 flags &= ~TH_FIN; 857 sendalot = 1; 858 } else 859 use_tso = 0; 860 if (sack_rxmit) { 861 if (SEQ_LT(p->rxmit + len, tp->snd_una + so->so_snd.sb_cc)) 862 flags &= ~TH_FIN; 863 } 864 865 win = sbspace(&so->so_rcv); 866 867 /* 868 * Sender silly window avoidance. If connection is idle 869 * and can send all data, a maximum segment, 870 * at least a maximum default-size segment do it, 871 * or are forced, do it; otherwise don't bother. 872 * If peer's buffer is tiny, then send 873 * when window is at least half open. 874 * If retransmitting (possibly after persist timer forced us 875 * to send into a small window), then must resend. 876 */ 877 if (len) { 878 if (len >= txsegsize) 879 goto send; 880 if ((so->so_state & SS_MORETOCOME) == 0 && 881 ((idle || tp->t_flags & TF_NODELAY) && 882 len + off >= so->so_snd.sb_cc)) 883 goto send; 884 if (tp->t_force) 885 goto send; 886 if (len >= tp->max_sndwnd / 2) 887 goto send; 888 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) 889 goto send; 890 if (sack_rxmit) 891 goto send; 892 } 893 894 /* 895 * Compare available window to amount of window known to peer 896 * (as advertised window less next expected input). If the 897 * difference is at least twice the size of the largest segment 898 * we expect to receive (i.e. two segments) or at least 50% of 899 * the maximum possible window, then want to send a window update 900 * to peer. 901 */ 902 if (win > 0) { 903 /* 904 * "adv" is the amount we can increase the window, 905 * taking into account that we are limited by 906 * TCP_MAXWIN << tp->rcv_scale. 907 */ 908 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) - 909 (tp->rcv_adv - tp->rcv_nxt); 910 911 if (adv >= (long) (2 * rxsegsize)) 912 goto send; 913 if (2 * adv >= (long) so->so_rcv.sb_hiwat) 914 goto send; 915 } 916 917 /* 918 * Send if we owe peer an ACK. 919 */ 920 if (tp->t_flags & TF_ACKNOW) 921 goto send; 922 if (flags & (TH_SYN|TH_FIN|TH_RST)) 923 goto send; 924 if (SEQ_GT(tp->snd_up, tp->snd_una)) 925 goto send; 926 /* 927 * In SACK, it is possible for tcp_output to fail to send a segment 928 * after the retransmission timer has been turned off. Make sure 929 * that the retransmission timer is set. 930 */ 931 if (TCP_SACK_ENABLED(tp) && SEQ_GT(tp->snd_max, tp->snd_una) && 932 !TCP_TIMER_ISARMED(tp, TCPT_REXMT) && 933 !TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) { 934 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 935 goto just_return; 936 } 937 938 /* 939 * TCP window updates are not reliable, rather a polling protocol 940 * using ``persist'' packets is used to insure receipt of window 941 * updates. The three ``states'' for the output side are: 942 * idle not doing retransmits or persists 943 * persisting to move a small or zero window 944 * (re)transmitting and thereby not persisting 945 * 946 * tp->t_timer[TCPT_PERSIST] 947 * is set when we are in persist state. 948 * tp->t_force 949 * is set when we are called to send a persist packet. 950 * tp->t_timer[TCPT_REXMT] 951 * is set when we are retransmitting 952 * The output side is idle when both timers are zero. 953 * 954 * If send window is too small, there is data to transmit, and no 955 * retransmit or persist is pending, then go to persist state. 956 * If nothing happens soon, send when timer expires: 957 * if window is nonzero, transmit what we can, 958 * otherwise force out a byte. 959 */ 960 if (so->so_snd.sb_cc && TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 961 TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 962 tp->t_rxtshift = 0; 963 tcp_setpersist(tp); 964 } 965 966 /* 967 * No reason to send a segment, just return. 968 */ 969 just_return: 970 return (0); 971 972 send: 973 /* 974 * Before ESTABLISHED, force sending of initial options 975 * unless TCP set not to do any options. 976 * NOTE: we assume that the IP/TCP header plus TCP options 977 * always fit in a single mbuf, leaving room for a maximum 978 * link header, i.e. 979 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES 980 */ 981 optlen = 0; 982 switch (af) { 983 #ifdef INET 984 case AF_INET: 985 iphdrlen = sizeof(struct ip) + sizeof(struct tcphdr); 986 break; 987 #endif 988 #ifdef INET6 989 case AF_INET6: 990 iphdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 991 break; 992 #endif 993 default: /*pacify gcc*/ 994 iphdrlen = 0; 995 break; 996 } 997 hdrlen = iphdrlen; 998 if (flags & TH_SYN) { 999 struct rtentry *rt; 1000 1001 rt = NULL; 1002 #ifdef INET 1003 if (tp->t_inpcb) 1004 rt = in_pcbrtentry(tp->t_inpcb); 1005 #endif 1006 #ifdef INET6 1007 if (tp->t_in6pcb) 1008 rt = in6_pcbrtentry(tp->t_in6pcb); 1009 #endif 1010 1011 tp->snd_nxt = tp->iss; 1012 tp->t_ourmss = tcp_mss_to_advertise(rt != NULL ? 1013 rt->rt_ifp : NULL, af); 1014 if ((tp->t_flags & TF_NOOPT) == 0) { 1015 opt[0] = TCPOPT_MAXSEG; 1016 opt[1] = 4; 1017 opt[2] = (tp->t_ourmss >> 8) & 0xff; 1018 opt[3] = tp->t_ourmss & 0xff; 1019 optlen = 4; 1020 1021 if ((tp->t_flags & TF_REQ_SCALE) && 1022 ((flags & TH_ACK) == 0 || 1023 (tp->t_flags & TF_RCVD_SCALE))) { 1024 *((u_int32_t *) (opt + optlen)) = htonl( 1025 TCPOPT_NOP << 24 | 1026 TCPOPT_WINDOW << 16 | 1027 TCPOLEN_WINDOW << 8 | 1028 tp->request_r_scale); 1029 optlen += 4; 1030 } 1031 if (tcp_do_sack) { 1032 u_int8_t *p = (u_int8_t *)(opt + optlen); 1033 1034 p[0] = TCPOPT_SACK_PERMITTED; 1035 p[1] = 2; 1036 p[2] = TCPOPT_NOP; 1037 p[3] = TCPOPT_NOP; 1038 optlen += 4; 1039 } 1040 } 1041 } 1042 1043 /* 1044 * Send a timestamp and echo-reply if this is a SYN and our side 1045 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side 1046 * and our peer have sent timestamps in our SYN's. 1047 */ 1048 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1049 (flags & TH_RST) == 0 && 1050 ((flags & (TH_SYN|TH_ACK)) == TH_SYN || 1051 (tp->t_flags & TF_RCVD_TSTMP))) { 1052 u_int32_t *lp = (u_int32_t *)(opt + optlen); 1053 1054 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1055 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1056 *lp++ = htonl(TCP_TIMESTAMP(tp)); 1057 *lp = htonl(tp->ts_recent); 1058 optlen += TCPOLEN_TSTAMP_APPA; 1059 } 1060 1061 /* 1062 * Tack on the SACK block if it is necessary. 1063 */ 1064 if (TCP_SACK_ENABLED(tp) && (tp->t_flags & TF_ACKNOW) 1065 && (tp->rcv_sack_num > 0)) { 1066 int sack_len, i; 1067 u_char *bp = (u_char *)(opt + optlen); 1068 u_int32_t *lp = (u_int32_t *)(bp + 4); 1069 1070 sack_len = tp->rcv_sack_num * 8 + 2; 1071 bp[0] = TCPOPT_NOP; 1072 bp[1] = TCPOPT_NOP; 1073 bp[2] = TCPOPT_SACK; 1074 bp[3] = sack_len; 1075 for (i = 0; i < tp->rcv_sack_num; i++) { 1076 *lp++ = htonl(tp->rcv_sack_block[i].left); 1077 *lp++ = htonl(tp->rcv_sack_block[i].right); 1078 } 1079 tp->rcv_sack_num = 0; 1080 optlen += sack_len + 2; 1081 } 1082 1083 #ifdef TCP_SIGNATURE 1084 #if defined(INET6) && defined(FAST_IPSEC) 1085 if (tp->t_family == AF_INET) 1086 #endif 1087 if (tp->t_flags & TF_SIGNATURE) { 1088 u_char *bp; 1089 /* 1090 * Initialize TCP-MD5 option (RFC2385) 1091 */ 1092 bp = (u_char *)opt + optlen; 1093 *bp++ = TCPOPT_SIGNATURE; 1094 *bp++ = TCPOLEN_SIGNATURE; 1095 sigoff = optlen + 2; 1096 bzero(bp, TCP_SIGLEN); 1097 bp += TCP_SIGLEN; 1098 optlen += TCPOLEN_SIGNATURE; 1099 /* 1100 * Terminate options list and maintain 32-bit alignment. 1101 */ 1102 *bp++ = TCPOPT_NOP; 1103 *bp++ = TCPOPT_EOL; 1104 optlen += 2; 1105 } 1106 #endif /* TCP_SIGNATURE */ 1107 1108 hdrlen += optlen; 1109 1110 #ifdef DIAGNOSTIC 1111 if (!use_tso && len > txsegsize) 1112 panic("tcp data to be sent is larger than segment"); 1113 else if (use_tso && len > IP_MAXPACKET) 1114 panic("tcp data to be sent is larger than max TSO size"); 1115 if (max_linkhdr + hdrlen > MCLBYTES) 1116 panic("tcphdr too big"); 1117 #endif 1118 1119 /* 1120 * Grab a header mbuf, attaching a copy of data to 1121 * be transmitted, and initialize the header from 1122 * the template for sends on this connection. 1123 */ 1124 if (len) { 1125 error = tcp_build_datapkt(tp, so, off, len, hdrlen, &m); 1126 if (error) 1127 goto out; 1128 /* 1129 * If we're sending everything we've got, set PUSH. 1130 * (This will keep happy those implementations which only 1131 * give data to the user when a buffer fills or 1132 * a PUSH comes in.) 1133 */ 1134 if (off + len == so->so_snd.sb_cc) 1135 flags |= TH_PUSH; 1136 } else { 1137 if (tp->t_flags & TF_ACKNOW) 1138 tcpstat.tcps_sndacks++; 1139 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 1140 tcpstat.tcps_sndctrl++; 1141 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 1142 tcpstat.tcps_sndurg++; 1143 else 1144 tcpstat.tcps_sndwinup++; 1145 1146 MGETHDR(m, M_DONTWAIT, MT_HEADER); 1147 if (m != NULL && max_linkhdr + hdrlen > MHLEN) { 1148 MCLGET(m, M_DONTWAIT); 1149 if ((m->m_flags & M_EXT) == 0) { 1150 m_freem(m); 1151 m = NULL; 1152 } 1153 } 1154 if (m == NULL) { 1155 error = ENOBUFS; 1156 goto out; 1157 } 1158 MCLAIM(m, &tcp_tx_mowner); 1159 m->m_data += max_linkhdr; 1160 m->m_len = hdrlen; 1161 } 1162 m->m_pkthdr.rcvif = (struct ifnet *)0; 1163 switch (af) { 1164 #ifdef INET 1165 case AF_INET: 1166 ip = mtod(m, struct ip *); 1167 #ifdef INET6 1168 ip6 = NULL; 1169 #endif 1170 th = (struct tcphdr *)(ip + 1); 1171 break; 1172 #endif 1173 #ifdef INET6 1174 case AF_INET6: 1175 ip = NULL; 1176 ip6 = mtod(m, struct ip6_hdr *); 1177 th = (struct tcphdr *)(ip6 + 1); 1178 break; 1179 #endif 1180 default: /*pacify gcc*/ 1181 ip = NULL; 1182 #ifdef INET6 1183 ip6 = NULL; 1184 #endif 1185 th = NULL; 1186 break; 1187 } 1188 if (tp->t_template == 0) 1189 panic("tcp_output"); 1190 if (tp->t_template->m_len < iphdrlen) 1191 panic("tcp_output"); 1192 bcopy(mtod(tp->t_template, caddr_t), mtod(m, caddr_t), iphdrlen); 1193 1194 /* 1195 * If we are doing retransmissions, then snd_nxt will 1196 * not reflect the first unsent octet. For ACK only 1197 * packets, we do not want the sequence number of the 1198 * retransmitted packet, we want the sequence number 1199 * of the next unsent octet. So, if there is no data 1200 * (and no SYN or FIN), use snd_max instead of snd_nxt 1201 * when filling in ti_seq. But if we are in persist 1202 * state, snd_max might reflect one byte beyond the 1203 * right edge of the window, so use snd_nxt in that 1204 * case, since we know we aren't doing a retransmission. 1205 * (retransmit and persist are mutually exclusive...) 1206 */ 1207 if (TCP_SACK_ENABLED(tp) && sack_rxmit) { 1208 th->th_seq = htonl(p->rxmit); 1209 p->rxmit += len; 1210 } else { 1211 if (len || (flags & (TH_SYN|TH_FIN)) || 1212 TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) 1213 th->th_seq = htonl(tp->snd_nxt); 1214 else 1215 th->th_seq = htonl(tp->snd_max); 1216 } 1217 th->th_ack = htonl(tp->rcv_nxt); 1218 if (optlen) { 1219 bcopy((caddr_t)opt, (caddr_t)(th + 1), optlen); 1220 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1221 } 1222 th->th_flags = flags; 1223 /* 1224 * Calculate receive window. Don't shrink window, 1225 * but avoid silly window syndrome. 1226 */ 1227 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)rxsegsize) 1228 win = 0; 1229 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 1230 win = (long)TCP_MAXWIN << tp->rcv_scale; 1231 if (win < (long)(int32_t)(tp->rcv_adv - tp->rcv_nxt)) 1232 win = (long)(int32_t)(tp->rcv_adv - tp->rcv_nxt); 1233 th->th_win = htons((u_int16_t) (win>>tp->rcv_scale)); 1234 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 1235 u_int32_t urp = tp->snd_up - tp->snd_nxt; 1236 if (urp > IP_MAXPACKET) 1237 urp = IP_MAXPACKET; 1238 th->th_urp = htons((u_int16_t)urp); 1239 th->th_flags |= TH_URG; 1240 } else 1241 /* 1242 * If no urgent pointer to send, then we pull 1243 * the urgent pointer to the left edge of the send window 1244 * so that it doesn't drift into the send window on sequence 1245 * number wraparound. 1246 */ 1247 tp->snd_up = tp->snd_una; /* drag it along */ 1248 1249 #ifdef TCP_SIGNATURE 1250 #if defined(INET6) && defined(FAST_IPSEC) 1251 if (tp->t_family == AF_INET) /* XXX */ 1252 #endif 1253 if (sigoff && (tp->t_flags & TF_SIGNATURE)) { 1254 struct secasvar *sav; 1255 u_int8_t *sigp; 1256 1257 sav = tcp_signature_getsav(m, th); 1258 1259 if (sav == NULL) { 1260 if (m) 1261 m_freem(m); 1262 return (EPERM); 1263 } 1264 1265 m->m_pkthdr.len = hdrlen + len; 1266 sigp = (caddr_t)th + sizeof(*th) + sigoff; 1267 tcp_signature(m, th, (caddr_t)th - mtod(m, caddr_t), sav, sigp); 1268 1269 key_sa_recordxfer(sav, m); 1270 #ifdef FAST_IPSEC 1271 KEY_FREESAV(&sav); 1272 #else 1273 key_freesav(sav); 1274 #endif 1275 } 1276 #endif 1277 1278 /* 1279 * Set ourselves up to be checksummed just before the packet 1280 * hits the wire. Maybe skip checksums on loopback interfaces. 1281 */ 1282 switch (af) { 1283 #ifdef INET 1284 case AF_INET: 1285 if (use_tso) { 1286 m->m_pkthdr.segsz = txsegsize; 1287 m->m_pkthdr.csum_flags |= M_CSUM_TSOv4; 1288 } else { 1289 if (__predict_true(ro->ro_rt == NULL || 1290 !(ro->ro_rt->rt_ifp->if_flags & 1291 IFF_LOOPBACK) || 1292 tcp_do_loopback_cksum)) 1293 m->m_pkthdr.csum_flags = M_CSUM_TCPv4; 1294 else 1295 m->m_pkthdr.csum_flags = 0; 1296 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1297 if (len + optlen) { 1298 /* Fixup the pseudo-header checksum. */ 1299 /* XXXJRT Not IP Jumbogram safe. */ 1300 th->th_sum = in_cksum_addword(th->th_sum, 1301 htons((u_int16_t) (len + optlen))); 1302 } 1303 } 1304 break; 1305 #endif 1306 #ifdef INET6 1307 case AF_INET6: 1308 /* 1309 * XXX Actually delaying the checksum is Hard 1310 * XXX (well, maybe not for Itojun, but it is 1311 * XXX for me), but we can still take advantage 1312 * XXX of the cached pseudo-header checksum. 1313 */ 1314 /* equals to hdrlen + len */ 1315 m->m_pkthdr.len = sizeof(struct ip6_hdr) 1316 + sizeof(struct tcphdr) + optlen + len; 1317 #ifdef notyet 1318 if (__predict_true(ro->ro_rt == NULL || 1319 !(ro->ro_rt->rt_ifp->if_flags & 1320 IFF_LOOPBACK) || 1321 tcp_do_loopback_cksum)) 1322 m->m_pkthdr.csum_flags = M_CSUM_TCPv6; 1323 else 1324 m->m_pkthdr.csum_flags = 0; 1325 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1326 #endif 1327 if (len + optlen) { 1328 /* Fixup the pseudo-header checksum. */ 1329 /* XXXJRT: Not IPv6 Jumbogram safe. */ 1330 th->th_sum = in_cksum_addword(th->th_sum, 1331 htons((u_int16_t) (len + optlen))); 1332 } 1333 #ifndef notyet 1334 if (__predict_true(ro->ro_rt == NULL || 1335 !(ro->ro_rt->rt_ifp->if_flags & 1336 IFF_LOOPBACK) || 1337 tcp_do_loopback_cksum)) 1338 th->th_sum = in6_cksum(m, 0, sizeof(struct ip6_hdr), 1339 sizeof(struct tcphdr) + optlen + len); 1340 #endif 1341 break; 1342 #endif 1343 } 1344 1345 /* 1346 * In transmit state, time the transmission and arrange for 1347 * the retransmit. In persist state, just set snd_max. 1348 */ 1349 if (tp->t_force == 0 || TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 1350 tcp_seq startseq = tp->snd_nxt; 1351 1352 /* 1353 * Advance snd_nxt over sequence space of this segment. 1354 * There are no states in which we send both a SYN and a FIN, 1355 * so we collapse the tests for these flags. 1356 */ 1357 if (flags & (TH_SYN|TH_FIN)) 1358 tp->snd_nxt++; 1359 if (sack_rxmit) 1360 goto timer; 1361 tp->snd_nxt += len; 1362 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 1363 tp->snd_max = tp->snd_nxt; 1364 /* 1365 * Time this transmission if not a retransmission and 1366 * not currently timing anything. 1367 */ 1368 if (tp->t_rtttime == 0) { 1369 tp->t_rtttime = tcp_now; 1370 tp->t_rtseq = startseq; 1371 tcpstat.tcps_segstimed++; 1372 } 1373 } 1374 1375 /* 1376 * Set retransmit timer if not currently set, 1377 * and not doing an ack or a keep-alive probe. 1378 * Initial value for retransmit timer is smoothed 1379 * round-trip time + 2 * round-trip time variance. 1380 * Initialize shift counter which is used for backoff 1381 * of retransmit time. 1382 */ 1383 timer: 1384 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 1385 ((sack_rxmit && tp->snd_nxt != tp->snd_max) || 1386 tp->snd_nxt != tp->snd_una)) { 1387 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) { 1388 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 1389 tp->t_rxtshift = 0; 1390 } 1391 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1392 } 1393 } else 1394 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max)) 1395 tp->snd_max = tp->snd_nxt + len; 1396 1397 #ifdef TCP_DEBUG 1398 /* 1399 * Trace. 1400 */ 1401 if (so->so_options & SO_DEBUG) 1402 tcp_trace(TA_OUTPUT, tp->t_state, tp, m, 0); 1403 #endif 1404 1405 /* 1406 * Fill in IP length and desired time to live and 1407 * send to IP level. There should be a better way 1408 * to handle ttl and tos; we could keep them in 1409 * the template, but need a way to checksum without them. 1410 */ 1411 m->m_pkthdr.len = hdrlen + len; 1412 1413 switch (af) { 1414 #ifdef INET 1415 case AF_INET: 1416 ip->ip_len = htons(m->m_pkthdr.len); 1417 if (tp->t_inpcb) { 1418 ip->ip_ttl = tp->t_inpcb->inp_ip.ip_ttl; 1419 ip->ip_tos = tp->t_inpcb->inp_ip.ip_tos; 1420 } 1421 #ifdef INET6 1422 else if (tp->t_in6pcb) { 1423 ip->ip_ttl = in6_selecthlim(tp->t_in6pcb, NULL); /*XXX*/ 1424 ip->ip_tos = 0; /*XXX*/ 1425 } 1426 #endif 1427 break; 1428 #endif 1429 #ifdef INET6 1430 case AF_INET6: 1431 ip6->ip6_nxt = IPPROTO_TCP; 1432 if (tp->t_in6pcb) { 1433 /* 1434 * we separately set hoplimit for every segment, since 1435 * the user might want to change the value via 1436 * setsockopt. Also, desired default hop limit might 1437 * be changed via Neighbor Discovery. 1438 */ 1439 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, 1440 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 1441 } 1442 /* ip6->ip6_flow = ??? */ 1443 /* ip6_plen will be filled in ip6_output(). */ 1444 break; 1445 #endif 1446 } 1447 1448 switch (af) { 1449 #ifdef INET 1450 case AF_INET: 1451 { 1452 struct mbuf *opts; 1453 1454 if (tp->t_inpcb) 1455 opts = tp->t_inpcb->inp_options; 1456 else 1457 opts = NULL; 1458 error = ip_output(m, opts, ro, 1459 (tp->t_mtudisc ? IP_MTUDISC : 0) | 1460 (so->so_options & SO_DONTROUTE), 1461 (struct ip_moptions *)0, so); 1462 break; 1463 } 1464 #endif 1465 #ifdef INET6 1466 case AF_INET6: 1467 { 1468 struct ip6_pktopts *opts; 1469 1470 if (tp->t_in6pcb) 1471 opts = tp->t_in6pcb->in6p_outputopts; 1472 else 1473 opts = NULL; 1474 error = ip6_output(m, opts, (struct route_in6 *)ro, 1475 so->so_options & SO_DONTROUTE, 1476 (struct ip6_moptions *)0, so, NULL); 1477 break; 1478 } 1479 #endif 1480 default: 1481 error = EAFNOSUPPORT; 1482 break; 1483 } 1484 if (error) { 1485 out: 1486 if (error == ENOBUFS) { 1487 tcpstat.tcps_selfquench++; 1488 #ifdef INET 1489 if (tp->t_inpcb) 1490 tcp_quench(tp->t_inpcb, 0); 1491 #endif 1492 #ifdef INET6 1493 if (tp->t_in6pcb) 1494 tcp6_quench(tp->t_in6pcb, 0); 1495 #endif 1496 error = 0; 1497 } else if ((error == EHOSTUNREACH || error == ENETDOWN) && 1498 TCPS_HAVERCVDSYN(tp->t_state)) { 1499 tp->t_softerror = error; 1500 error = 0; 1501 } 1502 1503 /* Back out the seqence number advance. */ 1504 if (sack_rxmit) 1505 p->rxmit -= len; 1506 1507 /* Restart the delayed ACK timer, if necessary. */ 1508 if (tp->t_flags & TF_DELACK) 1509 TCP_RESTART_DELACK(tp); 1510 1511 return (error); 1512 } 1513 tcpstat.tcps_sndtotal++; 1514 if (tp->t_flags & TF_DELACK) 1515 tcpstat.tcps_delack++; 1516 1517 /* 1518 * Data sent (as far as we can tell). 1519 * If this advertises a larger window than any other segment, 1520 * then remember the size of the advertised window. 1521 * Any pending ACK has now been sent. 1522 */ 1523 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 1524 tp->rcv_adv = tp->rcv_nxt + win; 1525 tp->last_ack_sent = tp->rcv_nxt; 1526 tp->t_flags &= ~TF_ACKNOW; 1527 TCP_CLEAR_DELACK(tp); 1528 #ifdef DIAGNOSTIC 1529 if (maxburst < 0) 1530 printf("tcp_output: maxburst exceeded by %d\n", -maxburst); 1531 #endif 1532 if (sendalot && (!tcp_do_newreno || --maxburst)) 1533 goto again; 1534 return (0); 1535 } 1536 1537 void 1538 tcp_setpersist(struct tcpcb *tp) 1539 { 1540 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2); 1541 int nticks; 1542 1543 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT)) 1544 panic("tcp_output REXMT"); 1545 /* 1546 * Start/restart persistance timer. 1547 */ 1548 if (t < tp->t_rttmin) 1549 t = tp->t_rttmin; 1550 TCPT_RANGESET(nticks, t * tcp_backoff[tp->t_rxtshift], 1551 TCPTV_PERSMIN, TCPTV_PERSMAX); 1552 TCP_TIMER_ARM(tp, TCPT_PERSIST, nticks); 1553 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 1554 tp->t_rxtshift++; 1555 } 1556
Indexes created Tue Sep 23 14:10:03 GMT 2025