tcp_output.c revision 1.126
1 /* $NetBSD: tcp_output.c,v 1.126 2005/03/12 07:53:08 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 1.126 2005/03/12 07:53:08 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 if (len <= txsegsize) { 855 use_tso = 0; 856 } 857 } else 858 len = txsegsize; 859 flags &= ~TH_FIN; 860 sendalot = 1; 861 } else 862 use_tso = 0; 863 if (sack_rxmit) { 864 if (SEQ_LT(p->rxmit + len, tp->snd_una + so->so_snd.sb_cc)) 865 flags &= ~TH_FIN; 866 } 867 868 win = sbspace(&so->so_rcv); 869 870 /* 871 * Sender silly window avoidance. If connection is idle 872 * and can send all data, a maximum segment, 873 * at least a maximum default-size segment do it, 874 * or are forced, do it; otherwise don't bother. 875 * If peer's buffer is tiny, then send 876 * when window is at least half open. 877 * If retransmitting (possibly after persist timer forced us 878 * to send into a small window), then must resend. 879 */ 880 if (len) { 881 if (len >= txsegsize) 882 goto send; 883 if ((so->so_state & SS_MORETOCOME) == 0 && 884 ((idle || tp->t_flags & TF_NODELAY) && 885 len + off >= so->so_snd.sb_cc)) 886 goto send; 887 if (tp->t_force) 888 goto send; 889 if (len >= tp->max_sndwnd / 2) 890 goto send; 891 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) 892 goto send; 893 if (sack_rxmit) 894 goto send; 895 } 896 897 /* 898 * Compare available window to amount of window known to peer 899 * (as advertised window less next expected input). If the 900 * difference is at least twice the size of the largest segment 901 * we expect to receive (i.e. two segments) or at least 50% of 902 * the maximum possible window, then want to send a window update 903 * to peer. 904 */ 905 if (win > 0) { 906 /* 907 * "adv" is the amount we can increase the window, 908 * taking into account that we are limited by 909 * TCP_MAXWIN << tp->rcv_scale. 910 */ 911 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) - 912 (tp->rcv_adv - tp->rcv_nxt); 913 914 if (adv >= (long) (2 * rxsegsize)) 915 goto send; 916 if (2 * adv >= (long) so->so_rcv.sb_hiwat) 917 goto send; 918 } 919 920 /* 921 * Send if we owe peer an ACK. 922 */ 923 if (tp->t_flags & TF_ACKNOW) 924 goto send; 925 if (flags & (TH_SYN|TH_FIN|TH_RST)) 926 goto send; 927 if (SEQ_GT(tp->snd_up, tp->snd_una)) 928 goto send; 929 /* 930 * In SACK, it is possible for tcp_output to fail to send a segment 931 * after the retransmission timer has been turned off. Make sure 932 * that the retransmission timer is set. 933 */ 934 if (TCP_SACK_ENABLED(tp) && SEQ_GT(tp->snd_max, tp->snd_una) && 935 !TCP_TIMER_ISARMED(tp, TCPT_REXMT) && 936 !TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) { 937 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 938 goto just_return; 939 } 940 941 /* 942 * TCP window updates are not reliable, rather a polling protocol 943 * using ``persist'' packets is used to insure receipt of window 944 * updates. The three ``states'' for the output side are: 945 * idle not doing retransmits or persists 946 * persisting to move a small or zero window 947 * (re)transmitting and thereby not persisting 948 * 949 * tp->t_timer[TCPT_PERSIST] 950 * is set when we are in persist state. 951 * tp->t_force 952 * is set when we are called to send a persist packet. 953 * tp->t_timer[TCPT_REXMT] 954 * is set when we are retransmitting 955 * The output side is idle when both timers are zero. 956 * 957 * If send window is too small, there is data to transmit, and no 958 * retransmit or persist is pending, then go to persist state. 959 * If nothing happens soon, send when timer expires: 960 * if window is nonzero, transmit what we can, 961 * otherwise force out a byte. 962 */ 963 if (so->so_snd.sb_cc && TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 964 TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 965 tp->t_rxtshift = 0; 966 tcp_setpersist(tp); 967 } 968 969 /* 970 * No reason to send a segment, just return. 971 */ 972 just_return: 973 return (0); 974 975 send: 976 /* 977 * Before ESTABLISHED, force sending of initial options 978 * unless TCP set not to do any options. 979 * NOTE: we assume that the IP/TCP header plus TCP options 980 * always fit in a single mbuf, leaving room for a maximum 981 * link header, i.e. 982 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES 983 */ 984 optlen = 0; 985 switch (af) { 986 #ifdef INET 987 case AF_INET: 988 iphdrlen = sizeof(struct ip) + sizeof(struct tcphdr); 989 break; 990 #endif 991 #ifdef INET6 992 case AF_INET6: 993 iphdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 994 break; 995 #endif 996 default: /*pacify gcc*/ 997 iphdrlen = 0; 998 break; 999 } 1000 hdrlen = iphdrlen; 1001 if (flags & TH_SYN) { 1002 struct rtentry *rt; 1003 1004 rt = NULL; 1005 #ifdef INET 1006 if (tp->t_inpcb) 1007 rt = in_pcbrtentry(tp->t_inpcb); 1008 #endif 1009 #ifdef INET6 1010 if (tp->t_in6pcb) 1011 rt = in6_pcbrtentry(tp->t_in6pcb); 1012 #endif 1013 1014 tp->snd_nxt = tp->iss; 1015 tp->t_ourmss = tcp_mss_to_advertise(rt != NULL ? 1016 rt->rt_ifp : NULL, af); 1017 if ((tp->t_flags & TF_NOOPT) == 0) { 1018 opt[0] = TCPOPT_MAXSEG; 1019 opt[1] = 4; 1020 opt[2] = (tp->t_ourmss >> 8) & 0xff; 1021 opt[3] = tp->t_ourmss & 0xff; 1022 optlen = 4; 1023 1024 if ((tp->t_flags & TF_REQ_SCALE) && 1025 ((flags & TH_ACK) == 0 || 1026 (tp->t_flags & TF_RCVD_SCALE))) { 1027 *((u_int32_t *) (opt + optlen)) = htonl( 1028 TCPOPT_NOP << 24 | 1029 TCPOPT_WINDOW << 16 | 1030 TCPOLEN_WINDOW << 8 | 1031 tp->request_r_scale); 1032 optlen += 4; 1033 } 1034 if (tcp_do_sack) { 1035 u_int8_t *p = (u_int8_t *)(opt + optlen); 1036 1037 p[0] = TCPOPT_SACK_PERMITTED; 1038 p[1] = 2; 1039 p[2] = TCPOPT_NOP; 1040 p[3] = TCPOPT_NOP; 1041 optlen += 4; 1042 } 1043 } 1044 } 1045 1046 /* 1047 * Send a timestamp and echo-reply if this is a SYN and our side 1048 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side 1049 * and our peer have sent timestamps in our SYN's. 1050 */ 1051 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1052 (flags & TH_RST) == 0 && 1053 ((flags & (TH_SYN|TH_ACK)) == TH_SYN || 1054 (tp->t_flags & TF_RCVD_TSTMP))) { 1055 u_int32_t *lp = (u_int32_t *)(opt + optlen); 1056 1057 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1058 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1059 *lp++ = htonl(TCP_TIMESTAMP(tp)); 1060 *lp = htonl(tp->ts_recent); 1061 optlen += TCPOLEN_TSTAMP_APPA; 1062 } 1063 1064 /* 1065 * Tack on the SACK block if it is necessary. 1066 */ 1067 if (TCP_SACK_ENABLED(tp) && (tp->t_flags & TF_ACKNOW) 1068 && (tp->rcv_sack_num > 0)) { 1069 int sack_len, i; 1070 u_char *bp = (u_char *)(opt + optlen); 1071 u_int32_t *lp = (u_int32_t *)(bp + 4); 1072 1073 sack_len = tp->rcv_sack_num * 8 + 2; 1074 bp[0] = TCPOPT_NOP; 1075 bp[1] = TCPOPT_NOP; 1076 bp[2] = TCPOPT_SACK; 1077 bp[3] = sack_len; 1078 for (i = 0; i < tp->rcv_sack_num; i++) { 1079 *lp++ = htonl(tp->rcv_sack_block[i].left); 1080 *lp++ = htonl(tp->rcv_sack_block[i].right); 1081 } 1082 tp->rcv_sack_num = 0; 1083 optlen += sack_len + 2; 1084 } 1085 1086 #ifdef TCP_SIGNATURE 1087 #if defined(INET6) && defined(FAST_IPSEC) 1088 if (tp->t_family == AF_INET) 1089 #endif 1090 if (tp->t_flags & TF_SIGNATURE) { 1091 u_char *bp; 1092 /* 1093 * Initialize TCP-MD5 option (RFC2385) 1094 */ 1095 bp = (u_char *)opt + optlen; 1096 *bp++ = TCPOPT_SIGNATURE; 1097 *bp++ = TCPOLEN_SIGNATURE; 1098 sigoff = optlen + 2; 1099 bzero(bp, TCP_SIGLEN); 1100 bp += TCP_SIGLEN; 1101 optlen += TCPOLEN_SIGNATURE; 1102 /* 1103 * Terminate options list and maintain 32-bit alignment. 1104 */ 1105 *bp++ = TCPOPT_NOP; 1106 *bp++ = TCPOPT_EOL; 1107 optlen += 2; 1108 } 1109 #endif /* TCP_SIGNATURE */ 1110 1111 hdrlen += optlen; 1112 1113 #ifdef DIAGNOSTIC 1114 if (!use_tso && len > txsegsize) 1115 panic("tcp data to be sent is larger than segment"); 1116 else if (use_tso && len > IP_MAXPACKET) 1117 panic("tcp data to be sent is larger than max TSO size"); 1118 if (max_linkhdr + hdrlen > MCLBYTES) 1119 panic("tcphdr too big"); 1120 #endif 1121 1122 /* 1123 * Grab a header mbuf, attaching a copy of data to 1124 * be transmitted, and initialize the header from 1125 * the template for sends on this connection. 1126 */ 1127 if (len) { 1128 error = tcp_build_datapkt(tp, so, off, len, hdrlen, &m); 1129 if (error) 1130 goto out; 1131 /* 1132 * If we're sending everything we've got, set PUSH. 1133 * (This will keep happy those implementations which only 1134 * give data to the user when a buffer fills or 1135 * a PUSH comes in.) 1136 */ 1137 if (off + len == so->so_snd.sb_cc) 1138 flags |= TH_PUSH; 1139 } else { 1140 if (tp->t_flags & TF_ACKNOW) 1141 tcpstat.tcps_sndacks++; 1142 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 1143 tcpstat.tcps_sndctrl++; 1144 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 1145 tcpstat.tcps_sndurg++; 1146 else 1147 tcpstat.tcps_sndwinup++; 1148 1149 MGETHDR(m, M_DONTWAIT, MT_HEADER); 1150 if (m != NULL && max_linkhdr + hdrlen > MHLEN) { 1151 MCLGET(m, M_DONTWAIT); 1152 if ((m->m_flags & M_EXT) == 0) { 1153 m_freem(m); 1154 m = NULL; 1155 } 1156 } 1157 if (m == NULL) { 1158 error = ENOBUFS; 1159 goto out; 1160 } 1161 MCLAIM(m, &tcp_tx_mowner); 1162 m->m_data += max_linkhdr; 1163 m->m_len = hdrlen; 1164 } 1165 m->m_pkthdr.rcvif = (struct ifnet *)0; 1166 switch (af) { 1167 #ifdef INET 1168 case AF_INET: 1169 ip = mtod(m, struct ip *); 1170 #ifdef INET6 1171 ip6 = NULL; 1172 #endif 1173 th = (struct tcphdr *)(ip + 1); 1174 break; 1175 #endif 1176 #ifdef INET6 1177 case AF_INET6: 1178 ip = NULL; 1179 ip6 = mtod(m, struct ip6_hdr *); 1180 th = (struct tcphdr *)(ip6 + 1); 1181 break; 1182 #endif 1183 default: /*pacify gcc*/ 1184 ip = NULL; 1185 #ifdef INET6 1186 ip6 = NULL; 1187 #endif 1188 th = NULL; 1189 break; 1190 } 1191 if (tp->t_template == 0) 1192 panic("tcp_output"); 1193 if (tp->t_template->m_len < iphdrlen) 1194 panic("tcp_output"); 1195 bcopy(mtod(tp->t_template, caddr_t), mtod(m, caddr_t), iphdrlen); 1196 1197 /* 1198 * If we are doing retransmissions, then snd_nxt will 1199 * not reflect the first unsent octet. For ACK only 1200 * packets, we do not want the sequence number of the 1201 * retransmitted packet, we want the sequence number 1202 * of the next unsent octet. So, if there is no data 1203 * (and no SYN or FIN), use snd_max instead of snd_nxt 1204 * when filling in ti_seq. But if we are in persist 1205 * state, snd_max might reflect one byte beyond the 1206 * right edge of the window, so use snd_nxt in that 1207 * case, since we know we aren't doing a retransmission. 1208 * (retransmit and persist are mutually exclusive...) 1209 */ 1210 if (TCP_SACK_ENABLED(tp) && sack_rxmit) { 1211 th->th_seq = htonl(p->rxmit); 1212 p->rxmit += len; 1213 } else { 1214 if (len || (flags & (TH_SYN|TH_FIN)) || 1215 TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) 1216 th->th_seq = htonl(tp->snd_nxt); 1217 else 1218 th->th_seq = htonl(tp->snd_max); 1219 } 1220 th->th_ack = htonl(tp->rcv_nxt); 1221 if (optlen) { 1222 bcopy((caddr_t)opt, (caddr_t)(th + 1), optlen); 1223 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1224 } 1225 th->th_flags = flags; 1226 /* 1227 * Calculate receive window. Don't shrink window, 1228 * but avoid silly window syndrome. 1229 */ 1230 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)rxsegsize) 1231 win = 0; 1232 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 1233 win = (long)TCP_MAXWIN << tp->rcv_scale; 1234 if (win < (long)(int32_t)(tp->rcv_adv - tp->rcv_nxt)) 1235 win = (long)(int32_t)(tp->rcv_adv - tp->rcv_nxt); 1236 th->th_win = htons((u_int16_t) (win>>tp->rcv_scale)); 1237 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 1238 u_int32_t urp = tp->snd_up - tp->snd_nxt; 1239 if (urp > IP_MAXPACKET) 1240 urp = IP_MAXPACKET; 1241 th->th_urp = htons((u_int16_t)urp); 1242 th->th_flags |= TH_URG; 1243 } else 1244 /* 1245 * If no urgent pointer to send, then we pull 1246 * the urgent pointer to the left edge of the send window 1247 * so that it doesn't drift into the send window on sequence 1248 * number wraparound. 1249 */ 1250 tp->snd_up = tp->snd_una; /* drag it along */ 1251 1252 #ifdef TCP_SIGNATURE 1253 #if defined(INET6) && defined(FAST_IPSEC) 1254 if (tp->t_family == AF_INET) /* XXX */ 1255 #endif 1256 if (sigoff && (tp->t_flags & TF_SIGNATURE)) { 1257 struct secasvar *sav; 1258 u_int8_t *sigp; 1259 1260 sav = tcp_signature_getsav(m, th); 1261 1262 if (sav == NULL) { 1263 if (m) 1264 m_freem(m); 1265 return (EPERM); 1266 } 1267 1268 m->m_pkthdr.len = hdrlen + len; 1269 sigp = (caddr_t)th + sizeof(*th) + sigoff; 1270 tcp_signature(m, th, (caddr_t)th - mtod(m, caddr_t), sav, sigp); 1271 1272 key_sa_recordxfer(sav, m); 1273 #ifdef FAST_IPSEC 1274 KEY_FREESAV(&sav); 1275 #else 1276 key_freesav(sav); 1277 #endif 1278 } 1279 #endif 1280 1281 /* 1282 * Set ourselves up to be checksummed just before the packet 1283 * hits the wire. Maybe skip checksums on loopback interfaces. 1284 */ 1285 switch (af) { 1286 #ifdef INET 1287 case AF_INET: 1288 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1289 if (use_tso) { 1290 m->m_pkthdr.segsz = txsegsize; 1291 m->m_pkthdr.csum_flags = M_CSUM_TSOv4; 1292 } else { 1293 if (__predict_true(ro->ro_rt == NULL || 1294 !(ro->ro_rt->rt_ifp->if_flags & 1295 IFF_LOOPBACK) || 1296 tcp_do_loopback_cksum)) 1297 m->m_pkthdr.csum_flags = M_CSUM_TCPv4; 1298 else 1299 m->m_pkthdr.csum_flags = 0; 1300 if (len + optlen) { 1301 /* Fixup the pseudo-header checksum. */ 1302 /* XXXJRT Not IP Jumbogram safe. */ 1303 th->th_sum = in_cksum_addword(th->th_sum, 1304 htons((u_int16_t) (len + optlen))); 1305 } 1306 } 1307 break; 1308 #endif 1309 #ifdef INET6 1310 case AF_INET6: 1311 /* 1312 * XXX Actually delaying the checksum is Hard 1313 * XXX (well, maybe not for Itojun, but it is 1314 * XXX for me), but we can still take advantage 1315 * XXX of the cached pseudo-header checksum. 1316 */ 1317 /* equals to hdrlen + len */ 1318 m->m_pkthdr.len = sizeof(struct ip6_hdr) 1319 + sizeof(struct tcphdr) + optlen + len; 1320 #ifdef notyet 1321 if (__predict_true(ro->ro_rt == NULL || 1322 !(ro->ro_rt->rt_ifp->if_flags & 1323 IFF_LOOPBACK) || 1324 tcp_do_loopback_cksum)) 1325 m->m_pkthdr.csum_flags = M_CSUM_TCPv6; 1326 else 1327 m->m_pkthdr.csum_flags = 0; 1328 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1329 #endif 1330 if (len + optlen) { 1331 /* Fixup the pseudo-header checksum. */ 1332 /* XXXJRT: Not IPv6 Jumbogram safe. */ 1333 th->th_sum = in_cksum_addword(th->th_sum, 1334 htons((u_int16_t) (len + optlen))); 1335 } 1336 #ifndef notyet 1337 if (__predict_true(ro->ro_rt == NULL || 1338 !(ro->ro_rt->rt_ifp->if_flags & 1339 IFF_LOOPBACK) || 1340 tcp_do_loopback_cksum)) 1341 th->th_sum = in6_cksum(m, 0, sizeof(struct ip6_hdr), 1342 sizeof(struct tcphdr) + optlen + len); 1343 #endif 1344 break; 1345 #endif 1346 } 1347 1348 /* 1349 * In transmit state, time the transmission and arrange for 1350 * the retransmit. In persist state, just set snd_max. 1351 */ 1352 if (tp->t_force == 0 || TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 1353 tcp_seq startseq = tp->snd_nxt; 1354 1355 /* 1356 * Advance snd_nxt over sequence space of this segment. 1357 * There are no states in which we send both a SYN and a FIN, 1358 * so we collapse the tests for these flags. 1359 */ 1360 if (flags & (TH_SYN|TH_FIN)) 1361 tp->snd_nxt++; 1362 if (sack_rxmit) 1363 goto timer; 1364 tp->snd_nxt += len; 1365 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 1366 tp->snd_max = tp->snd_nxt; 1367 /* 1368 * Time this transmission if not a retransmission and 1369 * not currently timing anything. 1370 */ 1371 if (tp->t_rtttime == 0) { 1372 tp->t_rtttime = tcp_now; 1373 tp->t_rtseq = startseq; 1374 tcpstat.tcps_segstimed++; 1375 } 1376 } 1377 1378 /* 1379 * Set retransmit timer if not currently set, 1380 * and not doing an ack or a keep-alive probe. 1381 * Initial value for retransmit timer is smoothed 1382 * round-trip time + 2 * round-trip time variance. 1383 * Initialize shift counter which is used for backoff 1384 * of retransmit time. 1385 */ 1386 timer: 1387 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 1388 ((sack_rxmit && tp->snd_nxt != tp->snd_max) || 1389 tp->snd_nxt != tp->snd_una)) { 1390 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) { 1391 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 1392 tp->t_rxtshift = 0; 1393 } 1394 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1395 } 1396 } else 1397 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max)) 1398 tp->snd_max = tp->snd_nxt + len; 1399 1400 #ifdef TCP_DEBUG 1401 /* 1402 * Trace. 1403 */ 1404 if (so->so_options & SO_DEBUG) 1405 tcp_trace(TA_OUTPUT, tp->t_state, tp, m, 0); 1406 #endif 1407 1408 /* 1409 * Fill in IP length and desired time to live and 1410 * send to IP level. There should be a better way 1411 * to handle ttl and tos; we could keep them in 1412 * the template, but need a way to checksum without them. 1413 */ 1414 m->m_pkthdr.len = hdrlen + len; 1415 1416 switch (af) { 1417 #ifdef INET 1418 case AF_INET: 1419 ip->ip_len = htons(m->m_pkthdr.len); 1420 if (tp->t_inpcb) { 1421 ip->ip_ttl = tp->t_inpcb->inp_ip.ip_ttl; 1422 ip->ip_tos = tp->t_inpcb->inp_ip.ip_tos; 1423 } 1424 #ifdef INET6 1425 else if (tp->t_in6pcb) { 1426 ip->ip_ttl = in6_selecthlim(tp->t_in6pcb, NULL); /*XXX*/ 1427 ip->ip_tos = 0; /*XXX*/ 1428 } 1429 #endif 1430 break; 1431 #endif 1432 #ifdef INET6 1433 case AF_INET6: 1434 ip6->ip6_nxt = IPPROTO_TCP; 1435 if (tp->t_in6pcb) { 1436 /* 1437 * we separately set hoplimit for every segment, since 1438 * the user might want to change the value via 1439 * setsockopt. Also, desired default hop limit might 1440 * be changed via Neighbor Discovery. 1441 */ 1442 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, 1443 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 1444 } 1445 /* ip6->ip6_flow = ??? */ 1446 /* ip6_plen will be filled in ip6_output(). */ 1447 break; 1448 #endif 1449 } 1450 1451 switch (af) { 1452 #ifdef INET 1453 case AF_INET: 1454 { 1455 struct mbuf *opts; 1456 1457 if (tp->t_inpcb) 1458 opts = tp->t_inpcb->inp_options; 1459 else 1460 opts = NULL; 1461 error = ip_output(m, opts, ro, 1462 (tp->t_mtudisc ? IP_MTUDISC : 0) | 1463 (so->so_options & SO_DONTROUTE), 1464 (struct ip_moptions *)0, so); 1465 break; 1466 } 1467 #endif 1468 #ifdef INET6 1469 case AF_INET6: 1470 { 1471 struct ip6_pktopts *opts; 1472 1473 if (tp->t_in6pcb) 1474 opts = tp->t_in6pcb->in6p_outputopts; 1475 else 1476 opts = NULL; 1477 error = ip6_output(m, opts, (struct route_in6 *)ro, 1478 so->so_options & SO_DONTROUTE, 1479 (struct ip6_moptions *)0, so, NULL); 1480 break; 1481 } 1482 #endif 1483 default: 1484 error = EAFNOSUPPORT; 1485 break; 1486 } 1487 if (error) { 1488 out: 1489 if (error == ENOBUFS) { 1490 tcpstat.tcps_selfquench++; 1491 #ifdef INET 1492 if (tp->t_inpcb) 1493 tcp_quench(tp->t_inpcb, 0); 1494 #endif 1495 #ifdef INET6 1496 if (tp->t_in6pcb) 1497 tcp6_quench(tp->t_in6pcb, 0); 1498 #endif 1499 error = 0; 1500 } else if ((error == EHOSTUNREACH || error == ENETDOWN) && 1501 TCPS_HAVERCVDSYN(tp->t_state)) { 1502 tp->t_softerror = error; 1503 error = 0; 1504 } 1505 1506 /* Back out the seqence number advance. */ 1507 if (sack_rxmit) 1508 p->rxmit -= len; 1509 1510 /* Restart the delayed ACK timer, if necessary. */ 1511 if (tp->t_flags & TF_DELACK) 1512 TCP_RESTART_DELACK(tp); 1513 1514 return (error); 1515 } 1516 tcpstat.tcps_sndtotal++; 1517 if (tp->t_flags & TF_DELACK) 1518 tcpstat.tcps_delack++; 1519 1520 /* 1521 * Data sent (as far as we can tell). 1522 * If this advertises a larger window than any other segment, 1523 * then remember the size of the advertised window. 1524 * Any pending ACK has now been sent. 1525 */ 1526 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 1527 tp->rcv_adv = tp->rcv_nxt + win; 1528 tp->last_ack_sent = tp->rcv_nxt; 1529 tp->t_flags &= ~TF_ACKNOW; 1530 TCP_CLEAR_DELACK(tp); 1531 #ifdef DIAGNOSTIC 1532 if (maxburst < 0) 1533 printf("tcp_output: maxburst exceeded by %d\n", -maxburst); 1534 #endif 1535 if (sendalot && (!tcp_do_newreno || --maxburst)) 1536 goto again; 1537 return (0); 1538 } 1539 1540 void 1541 tcp_setpersist(struct tcpcb *tp) 1542 { 1543 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2); 1544 int nticks; 1545 1546 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT)) 1547 panic("tcp_output REXMT"); 1548 /* 1549 * Start/restart persistance timer. 1550 */ 1551 if (t < tp->t_rttmin) 1552 t = tp->t_rttmin; 1553 TCPT_RANGESET(nticks, t * tcp_backoff[tp->t_rxtshift], 1554 TCPTV_PERSMIN, TCPTV_PERSMAX); 1555 TCP_TIMER_ARM(tp, TCPT_PERSIST, nticks); 1556 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 1557 tp->t_rxtshift++; 1558 } 1559
Indexes created Tue Sep 23 14:10:03 GMT 2025