tcp_output.c revision 1.128.2.2
1 /* $NetBSD: tcp_output.c,v 1.128.2.2 2005/05/06 08:40:05 tron 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 1.128.2.2 2005/05/06 08:40:05 tron 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 unsigned int sack_numblks; 554 int idle, sendalot, txsegsize, rxsegsize; 555 int txsegsize_nosack; 556 int maxburst = TCP_MAXBURST; 557 int af; /* address family on the wire */ 558 int iphdrlen; 559 int has_tso, use_tso; 560 int sack_rxmit; 561 int sack_bytes_rxmt; 562 struct sackhole *p; 563 #ifdef TCP_SIGNATURE 564 int sigoff = 0; 565 #endif 566 567 #ifdef DIAGNOSTIC 568 if (tp->t_inpcb && tp->t_in6pcb) 569 panic("tcp_output: both t_inpcb and t_in6pcb are set"); 570 #endif 571 so = NULL; 572 ro = NULL; 573 if (tp->t_inpcb) { 574 so = tp->t_inpcb->inp_socket; 575 ro = &tp->t_inpcb->inp_route; 576 } 577 #ifdef INET6 578 else if (tp->t_in6pcb) { 579 so = tp->t_in6pcb->in6p_socket; 580 ro = (struct route *)&tp->t_in6pcb->in6p_route; 581 } 582 #endif 583 584 switch (af = tp->t_family) { 585 #ifdef INET 586 case AF_INET: 587 if (tp->t_inpcb) 588 break; 589 #ifdef INET6 590 /* mapped addr case */ 591 if (tp->t_in6pcb) 592 break; 593 #endif 594 return (EINVAL); 595 #endif 596 #ifdef INET6 597 case AF_INET6: 598 if (tp->t_in6pcb) 599 break; 600 return (EINVAL); 601 #endif 602 default: 603 return (EAFNOSUPPORT); 604 } 605 606 if (tcp_segsize(tp, &txsegsize, &rxsegsize)) 607 return (EMSGSIZE); 608 609 idle = (tp->snd_max == tp->snd_una); 610 611 /* 612 * Determine if we can use TCP segmentation offload: 613 * - If we're using IPv4 614 * - If there is not an IPsec policy that prevents it 615 * - If the interface can do it 616 */ 617 has_tso = tp->t_inpcb != NULL && 618 #if defined(IPSEC) || defined(FAST_IPSEC) 619 IPSEC_PCB_SKIP_IPSEC(tp->t_inpcb->inp_sp, 620 IPSEC_DIR_OUTBOUND) && 621 #endif 622 tp->t_inpcb->inp_route.ro_rt != NULL && 623 (tp->t_inpcb->inp_route.ro_rt->rt_ifp->if_capenable & 624 IFCAP_TSOv4) != 0; 625 626 /* 627 * Restart Window computation. From draft-floyd-incr-init-win-03: 628 * 629 * Optionally, a TCP MAY set the restart window to the 630 * minimum of the value used for the initial window and 631 * the current value of cwnd (in other words, using a 632 * larger value for the restart window should never increase 633 * the size of cwnd). 634 */ 635 if (tcp_cwm) { 636 /* 637 * Hughes/Touch/Heidemann Congestion Window Monitoring. 638 * Count the number of packets currently pending 639 * acknowledgement, and limit our congestion window 640 * to a pre-determined allowed burst size plus that count. 641 * This prevents bursting once all pending packets have 642 * been acknowledged (i.e. transmission is idle). 643 * 644 * XXX Link this to Initial Window? 645 */ 646 tp->snd_cwnd = min(tp->snd_cwnd, 647 (tcp_cwm_burstsize * txsegsize) + 648 (tp->snd_nxt - tp->snd_una)); 649 } else { 650 if (idle && (tcp_now - tp->t_rcvtime) >= tp->t_rxtcur) { 651 /* 652 * We have been idle for "a while" and no acks are 653 * expected to clock out any data we send -- 654 * slow start to get ack "clock" running again. 655 */ 656 int ss = tcp_init_win; 657 #ifdef INET 658 if (tp->t_inpcb && 659 in_localaddr(tp->t_inpcb->inp_faddr)) 660 ss = tcp_init_win_local; 661 #endif 662 #ifdef INET6 663 if (tp->t_in6pcb && 664 in6_localaddr(&tp->t_in6pcb->in6p_faddr)) 665 ss = tcp_init_win_local; 666 #endif 667 tp->snd_cwnd = min(tp->snd_cwnd, 668 TCP_INITIAL_WINDOW(ss, txsegsize)); 669 } 670 } 671 672 txsegsize_nosack = txsegsize; 673 again: 674 use_tso = has_tso; 675 TCP_REASS_LOCK(tp); 676 sack_numblks = tcp_sack_numblks(tp); 677 if (sack_numblks) { 678 if ((tp->rcv_sack_flags & TCPSACK_HAVED) != 0) { 679 /* don't duplicate D-SACK. */ 680 use_tso = 0; 681 } 682 txsegsize = txsegsize_nosack - TCP_SACK_OPTLEN(sack_numblks); 683 } else { 684 txsegsize = txsegsize_nosack; 685 } 686 687 /* 688 * Determine length of data that should be transmitted, and 689 * flags that should be used. If there is some data or critical 690 * controls (SYN, RST) to send, then transmit; otherwise, 691 * investigate further. 692 * 693 * Readjust SACK information to avoid resending duplicate data. 694 */ 695 if (TCP_SACK_ENABLED(tp) && SEQ_LT(tp->snd_nxt, tp->snd_max)) 696 tcp_sack_adjust(tp); 697 sendalot = 0; 698 off = tp->snd_nxt - tp->snd_una; 699 win = min(tp->snd_wnd, tp->snd_cwnd); 700 701 flags = tcp_outflags[tp->t_state]; 702 703 /* 704 * Send any SACK-generated retransmissions. If we're explicitly trying 705 * to send out new data (when sendalot is 1), bypass this function. 706 * If we retransmit in fast recovery mode, decrement snd_cwnd, since 707 * we're replacing a (future) new transmission with a retransmission 708 * now, and we previously incremented snd_cwnd in tcp_input(). 709 */ 710 /* 711 * Still in sack recovery , reset rxmit flag to zero. 712 */ 713 sack_rxmit = 0; 714 sack_bytes_rxmt = 0; 715 len = 0; 716 p = NULL; 717 do { 718 long cwin; 719 if (!TCP_SACK_ENABLED(tp)) 720 break; 721 if (tp->t_partialacks < 0) 722 break; 723 p = tcp_sack_output(tp, &sack_bytes_rxmt); 724 if (p == NULL) 725 break; 726 727 cwin = min(tp->snd_wnd, tp->snd_cwnd) - sack_bytes_rxmt; 728 if (cwin < 0) 729 cwin = 0; 730 /* Do not retransmit SACK segments beyond snd_recover */ 731 if (SEQ_GT(p->end, tp->snd_recover)) { 732 /* 733 * (At least) part of sack hole extends beyond 734 * snd_recover. Check to see if we can rexmit data 735 * for this hole. 736 */ 737 if (SEQ_GEQ(p->rxmit, tp->snd_recover)) { 738 /* 739 * Can't rexmit any more data for this hole. 740 * That data will be rexmitted in the next 741 * sack recovery episode, when snd_recover 742 * moves past p->rxmit. 743 */ 744 p = NULL; 745 break; 746 } 747 /* Can rexmit part of the current hole */ 748 len = ((long)ulmin(cwin, tp->snd_recover - p->rxmit)); 749 } else 750 len = ((long)ulmin(cwin, p->end - p->rxmit)); 751 off = p->rxmit - tp->snd_una; 752 if (len > 0) { 753 sack_rxmit = 1; 754 sendalot = 1; 755 } 756 } while (/*CONSTCOND*/0); 757 758 /* 759 * If in persist timeout with window of 0, send 1 byte. 760 * Otherwise, if window is small but nonzero 761 * and timer expired, we will send what we can 762 * and go to transmit state. 763 */ 764 if (tp->t_force) { 765 if (win == 0) { 766 /* 767 * If we still have some data to send, then 768 * clear the FIN bit. Usually this would 769 * happen below when it realizes that we 770 * aren't sending all the data. However, 771 * if we have exactly 1 byte of unset data, 772 * then it won't clear the FIN bit below, 773 * and if we are in persist state, we wind 774 * up sending the packet without recording 775 * that we sent the FIN bit. 776 * 777 * We can't just blindly clear the FIN bit, 778 * because if we don't have any more data 779 * to send then the probe will be the FIN 780 * itself. 781 */ 782 if (off < so->so_snd.sb_cc) 783 flags &= ~TH_FIN; 784 win = 1; 785 } else { 786 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 787 tp->t_rxtshift = 0; 788 } 789 } 790 791 if (!TCP_SACK_ENABLED(tp)) { 792 if (win < so->so_snd.sb_cc) { 793 len = win - off; 794 flags &= ~TH_FIN; 795 } else 796 len = so->so_snd.sb_cc - off; 797 } else if (sack_rxmit == 0) { 798 if (sack_bytes_rxmt != 0) { 799 long cwin; 800 801 /* 802 * We are inside of a SACK recovery episode and are 803 * sending new data, having retransmitted all the 804 * data possible in the scoreboard. 805 */ 806 len = ((long)ulmin(so->so_snd.sb_cc, tp->snd_wnd) 807 - off); 808 /* 809 * From FreeBSD: 810 * Don't remove this (len > 0) check ! 811 * We explicitly check for len > 0 here (although it 812 * isn't really necessary), to work around a gcc 813 * optimization issue - to force gcc to compute 814 * len above. Without this check, the computation 815 * of len is bungled by the optimizer. 816 */ 817 if (len > 0) { 818 cwin = tp->snd_cwnd - 819 (tp->snd_nxt - tp->sack_newdata) - 820 sack_bytes_rxmt; 821 if (cwin < 0) 822 cwin = 0; 823 len = lmin(len, cwin); 824 } 825 } else if (win < so->so_snd.sb_cc) { 826 len = win - off; 827 flags &= ~TH_FIN; 828 } else 829 len = so->so_snd.sb_cc - off; 830 } 831 832 if (len < 0) { 833 /* 834 * If FIN has been sent but not acked, 835 * but we haven't been called to retransmit, 836 * len will be -1. Otherwise, window shrank 837 * after we sent into it. If window shrank to 0, 838 * cancel pending retransmit, pull snd_nxt back 839 * to (closed) window, and set the persist timer 840 * if it isn't already going. If the window didn't 841 * close completely, just wait for an ACK. 842 * 843 * If we have a pending FIN, either it has already been 844 * transmitted or it is outside the window, so drop it. 845 * If the FIN has been transmitted, but this is not a 846 * retransmission, then len must be -1. Therefore we also 847 * prevent here the sending of `gratuitous FINs'. This 848 * eliminates the need to check for that case below (e.g. 849 * to back up snd_nxt before the FIN so that the sequence 850 * number is correct). 851 */ 852 len = 0; 853 flags &= ~TH_FIN; 854 if (win == 0) { 855 TCP_TIMER_DISARM(tp, TCPT_REXMT); 856 tp->t_rxtshift = 0; 857 tp->snd_nxt = tp->snd_una; 858 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 859 tcp_setpersist(tp); 860 } 861 } 862 if (len > txsegsize) { 863 if (use_tso) { 864 /* 865 * Truncate TSO transfers to IP_MAXPACKET, and make 866 * sure that we send equal size transfers down the 867 * stack (rather than big-small-big-small-...). 868 */ 869 len = (min(len, IP_MAXPACKET) / txsegsize) * txsegsize; 870 if (len <= txsegsize) { 871 use_tso = 0; 872 } 873 } else 874 len = txsegsize; 875 flags &= ~TH_FIN; 876 sendalot = 1; 877 } else 878 use_tso = 0; 879 if (sack_rxmit) { 880 if (SEQ_LT(p->rxmit + len, tp->snd_una + so->so_snd.sb_cc)) 881 flags &= ~TH_FIN; 882 } 883 884 win = sbspace(&so->so_rcv); 885 886 /* 887 * Sender silly window avoidance. If connection is idle 888 * and can send all data, a maximum segment, 889 * at least a maximum default-size segment do it, 890 * or are forced, do it; otherwise don't bother. 891 * If peer's buffer is tiny, then send 892 * when window is at least half open. 893 * If retransmitting (possibly after persist timer forced us 894 * to send into a small window), then must resend. 895 */ 896 if (len) { 897 if (len >= txsegsize) 898 goto send; 899 if ((so->so_state & SS_MORETOCOME) == 0 && 900 ((idle || tp->t_flags & TF_NODELAY) && 901 len + off >= so->so_snd.sb_cc)) 902 goto send; 903 if (tp->t_force) 904 goto send; 905 if (len >= tp->max_sndwnd / 2) 906 goto send; 907 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) 908 goto send; 909 if (sack_rxmit) 910 goto send; 911 } 912 913 /* 914 * Compare available window to amount of window known to peer 915 * (as advertised window less next expected input). If the 916 * difference is at least twice the size of the largest segment 917 * we expect to receive (i.e. two segments) or at least 50% of 918 * the maximum possible window, then want to send a window update 919 * to peer. 920 */ 921 if (win > 0) { 922 /* 923 * "adv" is the amount we can increase the window, 924 * taking into account that we are limited by 925 * TCP_MAXWIN << tp->rcv_scale. 926 */ 927 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) - 928 (tp->rcv_adv - tp->rcv_nxt); 929 930 if (adv >= (long) (2 * rxsegsize)) 931 goto send; 932 if (2 * adv >= (long) so->so_rcv.sb_hiwat) 933 goto send; 934 } 935 936 /* 937 * Send if we owe peer an ACK. 938 */ 939 if (tp->t_flags & TF_ACKNOW) 940 goto send; 941 if (flags & (TH_SYN|TH_FIN|TH_RST)) 942 goto send; 943 if (SEQ_GT(tp->snd_up, tp->snd_una)) 944 goto send; 945 /* 946 * In SACK, it is possible for tcp_output to fail to send a segment 947 * after the retransmission timer has been turned off. Make sure 948 * that the retransmission timer is set. 949 */ 950 if (TCP_SACK_ENABLED(tp) && SEQ_GT(tp->snd_max, tp->snd_una) && 951 !TCP_TIMER_ISARMED(tp, TCPT_REXMT) && 952 !TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) { 953 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 954 goto just_return; 955 } 956 957 /* 958 * TCP window updates are not reliable, rather a polling protocol 959 * using ``persist'' packets is used to insure receipt of window 960 * updates. The three ``states'' for the output side are: 961 * idle not doing retransmits or persists 962 * persisting to move a small or zero window 963 * (re)transmitting and thereby not persisting 964 * 965 * tp->t_timer[TCPT_PERSIST] 966 * is set when we are in persist state. 967 * tp->t_force 968 * is set when we are called to send a persist packet. 969 * tp->t_timer[TCPT_REXMT] 970 * is set when we are retransmitting 971 * The output side is idle when both timers are zero. 972 * 973 * If send window is too small, there is data to transmit, and no 974 * retransmit or persist is pending, then go to persist state. 975 * If nothing happens soon, send when timer expires: 976 * if window is nonzero, transmit what we can, 977 * otherwise force out a byte. 978 */ 979 if (so->so_snd.sb_cc && TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 980 TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 981 tp->t_rxtshift = 0; 982 tcp_setpersist(tp); 983 } 984 985 /* 986 * No reason to send a segment, just return. 987 */ 988 just_return: 989 TCP_REASS_UNLOCK(tp); 990 return (0); 991 992 send: 993 /* 994 * Before ESTABLISHED, force sending of initial options 995 * unless TCP set not to do any options. 996 * NOTE: we assume that the IP/TCP header plus TCP options 997 * always fit in a single mbuf, leaving room for a maximum 998 * link header, i.e. 999 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES 1000 */ 1001 optlen = 0; 1002 switch (af) { 1003 #ifdef INET 1004 case AF_INET: 1005 iphdrlen = sizeof(struct ip) + sizeof(struct tcphdr); 1006 break; 1007 #endif 1008 #ifdef INET6 1009 case AF_INET6: 1010 iphdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1011 break; 1012 #endif 1013 default: /*pacify gcc*/ 1014 iphdrlen = 0; 1015 break; 1016 } 1017 hdrlen = iphdrlen; 1018 if (flags & TH_SYN) { 1019 struct rtentry *rt; 1020 1021 rt = NULL; 1022 #ifdef INET 1023 if (tp->t_inpcb) 1024 rt = in_pcbrtentry(tp->t_inpcb); 1025 #endif 1026 #ifdef INET6 1027 if (tp->t_in6pcb) 1028 rt = in6_pcbrtentry(tp->t_in6pcb); 1029 #endif 1030 1031 tp->snd_nxt = tp->iss; 1032 tp->t_ourmss = tcp_mss_to_advertise(rt != NULL ? 1033 rt->rt_ifp : NULL, af); 1034 if ((tp->t_flags & TF_NOOPT) == 0) { 1035 opt[0] = TCPOPT_MAXSEG; 1036 opt[1] = 4; 1037 opt[2] = (tp->t_ourmss >> 8) & 0xff; 1038 opt[3] = tp->t_ourmss & 0xff; 1039 optlen = 4; 1040 1041 if ((tp->t_flags & TF_REQ_SCALE) && 1042 ((flags & TH_ACK) == 0 || 1043 (tp->t_flags & TF_RCVD_SCALE))) { 1044 *((u_int32_t *) (opt + optlen)) = htonl( 1045 TCPOPT_NOP << 24 | 1046 TCPOPT_WINDOW << 16 | 1047 TCPOLEN_WINDOW << 8 | 1048 tp->request_r_scale); 1049 optlen += 4; 1050 } 1051 if (tcp_do_sack) { 1052 u_int8_t *p = (u_int8_t *)(opt + optlen); 1053 1054 p[0] = TCPOPT_SACK_PERMITTED; 1055 p[1] = 2; 1056 p[2] = TCPOPT_NOP; 1057 p[3] = TCPOPT_NOP; 1058 optlen += 4; 1059 } 1060 } 1061 } 1062 1063 /* 1064 * Send a timestamp and echo-reply if this is a SYN and our side 1065 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side 1066 * and our peer have sent timestamps in our SYN's. 1067 */ 1068 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1069 (flags & TH_RST) == 0 && 1070 ((flags & (TH_SYN|TH_ACK)) == TH_SYN || 1071 (tp->t_flags & TF_RCVD_TSTMP))) { 1072 u_int32_t *lp = (u_int32_t *)(opt + optlen); 1073 1074 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1075 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1076 *lp++ = htonl(TCP_TIMESTAMP(tp)); 1077 *lp = htonl(tp->ts_recent); 1078 optlen += TCPOLEN_TSTAMP_APPA; 1079 } 1080 1081 /* 1082 * Tack on the SACK block if it is necessary. 1083 */ 1084 if (sack_numblks) { 1085 int sack_len; 1086 u_char *bp = (u_char *)(opt + optlen); 1087 u_int32_t *lp = (u_int32_t *)(bp + 4); 1088 struct ipqent *tiqe; 1089 1090 sack_len = sack_numblks * 8 + 2; 1091 bp[0] = TCPOPT_NOP; 1092 bp[1] = TCPOPT_NOP; 1093 bp[2] = TCPOPT_SACK; 1094 bp[3] = sack_len; 1095 if ((tp->rcv_sack_flags & TCPSACK_HAVED) != 0) { 1096 sack_numblks--; 1097 *lp++ = htonl(tp->rcv_dsack_block.left); 1098 *lp++ = htonl(tp->rcv_dsack_block.right); 1099 tp->rcv_sack_flags &= ~TCPSACK_HAVED; 1100 } 1101 for (tiqe = TAILQ_FIRST(&tp->timeq); 1102 sack_numblks > 0; tiqe = TAILQ_NEXT(tiqe, ipqe_timeq)) { 1103 KASSERT(tiqe != NULL); 1104 sack_numblks--; 1105 *lp++ = htonl(tiqe->ipqe_seq); 1106 *lp++ = htonl(tiqe->ipqe_seq + tiqe->ipqe_len); 1107 } 1108 optlen += sack_len + 2; 1109 } 1110 TCP_REASS_UNLOCK(tp); 1111 1112 #ifdef TCP_SIGNATURE 1113 #if defined(INET6) && defined(FAST_IPSEC) 1114 if (tp->t_family == AF_INET) 1115 #endif 1116 if (tp->t_flags & TF_SIGNATURE) { 1117 u_char *bp; 1118 /* 1119 * Initialize TCP-MD5 option (RFC2385) 1120 */ 1121 bp = (u_char *)opt + optlen; 1122 *bp++ = TCPOPT_SIGNATURE; 1123 *bp++ = TCPOLEN_SIGNATURE; 1124 sigoff = optlen + 2; 1125 bzero(bp, TCP_SIGLEN); 1126 bp += TCP_SIGLEN; 1127 optlen += TCPOLEN_SIGNATURE; 1128 /* 1129 * Terminate options list and maintain 32-bit alignment. 1130 */ 1131 *bp++ = TCPOPT_NOP; 1132 *bp++ = TCPOPT_EOL; 1133 optlen += 2; 1134 } 1135 #endif /* TCP_SIGNATURE */ 1136 1137 hdrlen += optlen; 1138 1139 #ifdef DIAGNOSTIC 1140 if (!use_tso && len > txsegsize) 1141 panic("tcp data to be sent is larger than segment"); 1142 else if (use_tso && len > IP_MAXPACKET) 1143 panic("tcp data to be sent is larger than max TSO size"); 1144 if (max_linkhdr + hdrlen > MCLBYTES) 1145 panic("tcphdr too big"); 1146 #endif 1147 1148 /* 1149 * Grab a header mbuf, attaching a copy of data to 1150 * be transmitted, and initialize the header from 1151 * the template for sends on this connection. 1152 */ 1153 if (len) { 1154 error = tcp_build_datapkt(tp, so, off, len, hdrlen, &m); 1155 if (error) 1156 goto out; 1157 /* 1158 * If we're sending everything we've got, set PUSH. 1159 * (This will keep happy those implementations which only 1160 * give data to the user when a buffer fills or 1161 * a PUSH comes in.) 1162 */ 1163 if (off + len == so->so_snd.sb_cc) 1164 flags |= TH_PUSH; 1165 } else { 1166 if (tp->t_flags & TF_ACKNOW) 1167 tcpstat.tcps_sndacks++; 1168 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 1169 tcpstat.tcps_sndctrl++; 1170 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 1171 tcpstat.tcps_sndurg++; 1172 else 1173 tcpstat.tcps_sndwinup++; 1174 1175 MGETHDR(m, M_DONTWAIT, MT_HEADER); 1176 if (m != NULL && max_linkhdr + hdrlen > MHLEN) { 1177 MCLGET(m, M_DONTWAIT); 1178 if ((m->m_flags & M_EXT) == 0) { 1179 m_freem(m); 1180 m = NULL; 1181 } 1182 } 1183 if (m == NULL) { 1184 error = ENOBUFS; 1185 goto out; 1186 } 1187 MCLAIM(m, &tcp_tx_mowner); 1188 m->m_data += max_linkhdr; 1189 m->m_len = hdrlen; 1190 } 1191 m->m_pkthdr.rcvif = (struct ifnet *)0; 1192 switch (af) { 1193 #ifdef INET 1194 case AF_INET: 1195 ip = mtod(m, struct ip *); 1196 #ifdef INET6 1197 ip6 = NULL; 1198 #endif 1199 th = (struct tcphdr *)(ip + 1); 1200 break; 1201 #endif 1202 #ifdef INET6 1203 case AF_INET6: 1204 ip = NULL; 1205 ip6 = mtod(m, struct ip6_hdr *); 1206 th = (struct tcphdr *)(ip6 + 1); 1207 break; 1208 #endif 1209 default: /*pacify gcc*/ 1210 ip = NULL; 1211 #ifdef INET6 1212 ip6 = NULL; 1213 #endif 1214 th = NULL; 1215 break; 1216 } 1217 if (tp->t_template == 0) 1218 panic("tcp_output"); 1219 if (tp->t_template->m_len < iphdrlen) 1220 panic("tcp_output"); 1221 bcopy(mtod(tp->t_template, caddr_t), mtod(m, caddr_t), iphdrlen); 1222 1223 /* 1224 * If we are doing retransmissions, then snd_nxt will 1225 * not reflect the first unsent octet. For ACK only 1226 * packets, we do not want the sequence number of the 1227 * retransmitted packet, we want the sequence number 1228 * of the next unsent octet. So, if there is no data 1229 * (and no SYN or FIN), use snd_max instead of snd_nxt 1230 * when filling in ti_seq. But if we are in persist 1231 * state, snd_max might reflect one byte beyond the 1232 * right edge of the window, so use snd_nxt in that 1233 * case, since we know we aren't doing a retransmission. 1234 * (retransmit and persist are mutually exclusive...) 1235 */ 1236 if (TCP_SACK_ENABLED(tp) && sack_rxmit) { 1237 th->th_seq = htonl(p->rxmit); 1238 p->rxmit += len; 1239 } else { 1240 if (len || (flags & (TH_SYN|TH_FIN)) || 1241 TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) 1242 th->th_seq = htonl(tp->snd_nxt); 1243 else 1244 th->th_seq = htonl(tp->snd_max); 1245 } 1246 th->th_ack = htonl(tp->rcv_nxt); 1247 if (optlen) { 1248 bcopy((caddr_t)opt, (caddr_t)(th + 1), optlen); 1249 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1250 } 1251 th->th_flags = flags; 1252 /* 1253 * Calculate receive window. Don't shrink window, 1254 * but avoid silly window syndrome. 1255 */ 1256 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)rxsegsize) 1257 win = 0; 1258 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 1259 win = (long)TCP_MAXWIN << tp->rcv_scale; 1260 if (win < (long)(int32_t)(tp->rcv_adv - tp->rcv_nxt)) 1261 win = (long)(int32_t)(tp->rcv_adv - tp->rcv_nxt); 1262 th->th_win = htons((u_int16_t) (win>>tp->rcv_scale)); 1263 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 1264 u_int32_t urp = tp->snd_up - tp->snd_nxt; 1265 if (urp > IP_MAXPACKET) 1266 urp = IP_MAXPACKET; 1267 th->th_urp = htons((u_int16_t)urp); 1268 th->th_flags |= TH_URG; 1269 } else 1270 /* 1271 * If no urgent pointer to send, then we pull 1272 * the urgent pointer to the left edge of the send window 1273 * so that it doesn't drift into the send window on sequence 1274 * number wraparound. 1275 */ 1276 tp->snd_up = tp->snd_una; /* drag it along */ 1277 1278 #ifdef TCP_SIGNATURE 1279 #if defined(INET6) && defined(FAST_IPSEC) 1280 if (tp->t_family == AF_INET) /* XXX */ 1281 #endif 1282 if (sigoff && (tp->t_flags & TF_SIGNATURE)) { 1283 struct secasvar *sav; 1284 u_int8_t *sigp; 1285 1286 sav = tcp_signature_getsav(m, th); 1287 1288 if (sav == NULL) { 1289 if (m) 1290 m_freem(m); 1291 return (EPERM); 1292 } 1293 1294 m->m_pkthdr.len = hdrlen + len; 1295 sigp = (caddr_t)th + sizeof(*th) + sigoff; 1296 tcp_signature(m, th, (caddr_t)th - mtod(m, caddr_t), sav, sigp); 1297 1298 key_sa_recordxfer(sav, m); 1299 #ifdef FAST_IPSEC 1300 KEY_FREESAV(&sav); 1301 #else 1302 key_freesav(sav); 1303 #endif 1304 } 1305 #endif 1306 1307 /* 1308 * Set ourselves up to be checksummed just before the packet 1309 * hits the wire. 1310 */ 1311 switch (af) { 1312 #ifdef INET 1313 case AF_INET: 1314 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1315 if (use_tso) { 1316 m->m_pkthdr.segsz = txsegsize; 1317 m->m_pkthdr.csum_flags = M_CSUM_TSOv4; 1318 } else { 1319 m->m_pkthdr.csum_flags = M_CSUM_TCPv4; 1320 if (len + optlen) { 1321 /* Fixup the pseudo-header checksum. */ 1322 /* XXXJRT Not IP Jumbogram safe. */ 1323 th->th_sum = in_cksum_addword(th->th_sum, 1324 htons((u_int16_t) (len + optlen))); 1325 } 1326 } 1327 break; 1328 #endif 1329 #ifdef INET6 1330 case AF_INET6: 1331 /* 1332 * XXX Actually delaying the checksum is Hard 1333 * XXX (well, maybe not for Itojun, but it is 1334 * XXX for me), but we can still take advantage 1335 * XXX of the cached pseudo-header checksum. 1336 */ 1337 /* equals to hdrlen + len */ 1338 m->m_pkthdr.len = sizeof(struct ip6_hdr) 1339 + sizeof(struct tcphdr) + optlen + len; 1340 #ifdef notyet 1341 m->m_pkthdr.csum_flags = M_CSUM_TCPv6; 1342 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1343 #endif 1344 if (len + optlen) { 1345 /* Fixup the pseudo-header checksum. */ 1346 /* XXXJRT: Not IPv6 Jumbogram safe. */ 1347 th->th_sum = in_cksum_addword(th->th_sum, 1348 htons((u_int16_t) (len + optlen))); 1349 } 1350 th->th_sum = in6_cksum(m, 0, sizeof(struct ip6_hdr), 1351 sizeof(struct tcphdr) + optlen + len); 1352 break; 1353 #endif 1354 } 1355 1356 /* 1357 * In transmit state, time the transmission and arrange for 1358 * the retransmit. In persist state, just set snd_max. 1359 */ 1360 if (tp->t_force == 0 || TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 1361 tcp_seq startseq = tp->snd_nxt; 1362 1363 /* 1364 * Advance snd_nxt over sequence space of this segment. 1365 * There are no states in which we send both a SYN and a FIN, 1366 * so we collapse the tests for these flags. 1367 */ 1368 if (flags & (TH_SYN|TH_FIN)) 1369 tp->snd_nxt++; 1370 if (sack_rxmit) 1371 goto timer; 1372 tp->snd_nxt += len; 1373 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 1374 tp->snd_max = tp->snd_nxt; 1375 /* 1376 * Time this transmission if not a retransmission and 1377 * not currently timing anything. 1378 */ 1379 if (tp->t_rtttime == 0) { 1380 tp->t_rtttime = tcp_now; 1381 tp->t_rtseq = startseq; 1382 tcpstat.tcps_segstimed++; 1383 } 1384 } 1385 1386 /* 1387 * Set retransmit timer if not currently set, 1388 * and not doing an ack or a keep-alive probe. 1389 * Initial value for retransmit timer is smoothed 1390 * round-trip time + 2 * round-trip time variance. 1391 * Initialize shift counter which is used for backoff 1392 * of retransmit time. 1393 */ 1394 timer: 1395 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 1396 ((sack_rxmit && tp->snd_nxt != tp->snd_max) || 1397 tp->snd_nxt != tp->snd_una)) { 1398 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) { 1399 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 1400 tp->t_rxtshift = 0; 1401 } 1402 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1403 } 1404 } else 1405 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max)) 1406 tp->snd_max = tp->snd_nxt + len; 1407 1408 #ifdef TCP_DEBUG 1409 /* 1410 * Trace. 1411 */ 1412 if (so->so_options & SO_DEBUG) 1413 tcp_trace(TA_OUTPUT, tp->t_state, tp, m, 0); 1414 #endif 1415 1416 /* 1417 * Fill in IP length and desired time to live and 1418 * send to IP level. There should be a better way 1419 * to handle ttl and tos; we could keep them in 1420 * the template, but need a way to checksum without them. 1421 */ 1422 m->m_pkthdr.len = hdrlen + len; 1423 1424 switch (af) { 1425 #ifdef INET 1426 case AF_INET: 1427 ip->ip_len = htons(m->m_pkthdr.len); 1428 if (tp->t_inpcb) { 1429 ip->ip_ttl = tp->t_inpcb->inp_ip.ip_ttl; 1430 ip->ip_tos = tp->t_inpcb->inp_ip.ip_tos; 1431 } 1432 #ifdef INET6 1433 else if (tp->t_in6pcb) { 1434 ip->ip_ttl = in6_selecthlim(tp->t_in6pcb, NULL); /*XXX*/ 1435 ip->ip_tos = 0; /*XXX*/ 1436 } 1437 #endif 1438 break; 1439 #endif 1440 #ifdef INET6 1441 case AF_INET6: 1442 ip6->ip6_nxt = IPPROTO_TCP; 1443 if (tp->t_in6pcb) { 1444 /* 1445 * we separately set hoplimit for every segment, since 1446 * the user might want to change the value via 1447 * setsockopt. Also, desired default hop limit might 1448 * be changed via Neighbor Discovery. 1449 */ 1450 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, 1451 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 1452 } 1453 /* ip6->ip6_flow = ??? */ 1454 /* ip6_plen will be filled in ip6_output(). */ 1455 break; 1456 #endif 1457 } 1458 1459 switch (af) { 1460 #ifdef INET 1461 case AF_INET: 1462 { 1463 struct mbuf *opts; 1464 1465 if (tp->t_inpcb) 1466 opts = tp->t_inpcb->inp_options; 1467 else 1468 opts = NULL; 1469 error = ip_output(m, opts, ro, 1470 (tp->t_mtudisc ? IP_MTUDISC : 0) | 1471 (so->so_options & SO_DONTROUTE), 1472 (struct ip_moptions *)0, so); 1473 break; 1474 } 1475 #endif 1476 #ifdef INET6 1477 case AF_INET6: 1478 { 1479 struct ip6_pktopts *opts; 1480 1481 if (tp->t_in6pcb) 1482 opts = tp->t_in6pcb->in6p_outputopts; 1483 else 1484 opts = NULL; 1485 error = ip6_output(m, opts, (struct route_in6 *)ro, 1486 so->so_options & SO_DONTROUTE, 1487 (struct ip6_moptions *)0, so, NULL); 1488 break; 1489 } 1490 #endif 1491 default: 1492 error = EAFNOSUPPORT; 1493 break; 1494 } 1495 if (error) { 1496 out: 1497 if (error == ENOBUFS) { 1498 tcpstat.tcps_selfquench++; 1499 #ifdef INET 1500 if (tp->t_inpcb) 1501 tcp_quench(tp->t_inpcb, 0); 1502 #endif 1503 #ifdef INET6 1504 if (tp->t_in6pcb) 1505 tcp6_quench(tp->t_in6pcb, 0); 1506 #endif 1507 error = 0; 1508 } else if ((error == EHOSTUNREACH || error == ENETDOWN) && 1509 TCPS_HAVERCVDSYN(tp->t_state)) { 1510 tp->t_softerror = error; 1511 error = 0; 1512 } 1513 1514 /* Back out the seqence number advance. */ 1515 if (sack_rxmit) 1516 p->rxmit -= len; 1517 1518 /* Restart the delayed ACK timer, if necessary. */ 1519 if (tp->t_flags & TF_DELACK) 1520 TCP_RESTART_DELACK(tp); 1521 1522 return (error); 1523 } 1524 tcpstat.tcps_sndtotal++; 1525 if (tp->t_flags & TF_DELACK) 1526 tcpstat.tcps_delack++; 1527 1528 /* 1529 * Data sent (as far as we can tell). 1530 * If this advertises a larger window than any other segment, 1531 * then remember the size of the advertised window. 1532 * Any pending ACK has now been sent. 1533 */ 1534 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 1535 tp->rcv_adv = tp->rcv_nxt + win; 1536 tp->last_ack_sent = tp->rcv_nxt; 1537 tp->t_flags &= ~TF_ACKNOW; 1538 TCP_CLEAR_DELACK(tp); 1539 #ifdef DIAGNOSTIC 1540 if (maxburst < 0) 1541 printf("tcp_output: maxburst exceeded by %d\n", -maxburst); 1542 #endif 1543 if (sendalot && (!tcp_do_newreno || --maxburst)) 1544 goto again; 1545 return (0); 1546 } 1547 1548 void 1549 tcp_setpersist(struct tcpcb *tp) 1550 { 1551 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2); 1552 int nticks; 1553 1554 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT)) 1555 panic("tcp_output REXMT"); 1556 /* 1557 * Start/restart persistance timer. 1558 */ 1559 if (t < tp->t_rttmin) 1560 t = tp->t_rttmin; 1561 TCPT_RANGESET(nticks, t * tcp_backoff[tp->t_rxtshift], 1562 TCPTV_PERSMIN, TCPTV_PERSMAX); 1563 TCP_TIMER_ARM(tp, TCPT_PERSIST, nticks); 1564 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 1565 tp->t_rxtshift++; 1566 } 1567
Indexes created Wed Sep 24 05:09:52 GMT 2025