tcp_output.c revision 1.121
1 /* $NetBSD: tcp_output.c,v 1.121 2005/03/06 00:48:52 matt 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 1.121 2005/03/06 00:48:52 matt Exp $"); 144 145 #include "opt_inet.h" 146 #include "opt_ipsec.h" 147 #include "opt_tcp_debug.h" 148 149 #include <sys/param.h> 150 #include <sys/systm.h> 151 #include <sys/malloc.h> 152 #include <sys/mbuf.h> 153 #include <sys/protosw.h> 154 #include <sys/socket.h> 155 #include <sys/socketvar.h> 156 #include <sys/errno.h> 157 #include <sys/domain.h> 158 #include <sys/kernel.h> 159 #ifdef TCP_SIGNATURE 160 #include <sys/md5.h> 161 #endif 162 163 #include <net/if.h> 164 #include <net/route.h> 165 166 #include <netinet/in.h> 167 #include <netinet/in_systm.h> 168 #include <netinet/ip.h> 169 #include <netinet/in_pcb.h> 170 #include <netinet/ip_var.h> 171 172 #ifdef INET6 173 #ifndef INET 174 #include <netinet/in.h> 175 #endif 176 #include <netinet/ip6.h> 177 #include <netinet6/in6_var.h> 178 #include <netinet6/ip6_var.h> 179 #include <netinet6/in6_pcb.h> 180 #include <netinet6/nd6.h> 181 #endif 182 183 #ifdef FAST_IPSEC 184 #include <netipsec/ipsec.h> 185 #include <netipsec/key.h> 186 #endif /* FAST_IPSEC*/ 187 #ifdef IPSEC 188 #include <netinet6/ipsec.h> 189 #endif 190 191 #include <netinet/tcp.h> 192 #define TCPOUTFLAGS 193 #include <netinet/tcp_fsm.h> 194 #include <netinet/tcp_seq.h> 195 #include <netinet/tcp_timer.h> 196 #include <netinet/tcp_var.h> 197 #include <netinet/tcpip.h> 198 #include <netinet/tcp_debug.h> 199 200 #ifdef IPSEC 201 #include <netkey/key.h> 202 #endif 203 204 #ifdef notyet 205 extern struct mbuf *m_copypack(); 206 #endif 207 208 #define MAX_TCPOPTLEN 40 /* max # bytes that go in options */ 209 210 /* 211 * Knob to enable Congestion Window Monitoring, and control the 212 * the burst size it allows. Default burst is 4 packets, per 213 * the Internet draft. 214 */ 215 int tcp_cwm = 0; 216 int tcp_cwm_burstsize = 4; 217 218 #ifdef TCP_OUTPUT_COUNTERS 219 #include <sys/device.h> 220 221 extern struct evcnt tcp_output_bigheader; 222 extern struct evcnt tcp_output_predict_hit; 223 extern struct evcnt tcp_output_predict_miss; 224 extern struct evcnt tcp_output_copysmall; 225 extern struct evcnt tcp_output_copybig; 226 extern struct evcnt tcp_output_refbig; 227 228 #define TCP_OUTPUT_COUNTER_INCR(ev) (ev)->ev_count++ 229 #else 230 231 #define TCP_OUTPUT_COUNTER_INCR(ev) /* nothing */ 232 233 #endif /* TCP_OUTPUT_COUNTERS */ 234 235 static 236 #ifndef GPROF 237 __inline 238 #endif 239 int 240 tcp_segsize(struct tcpcb *tp, int *txsegsizep, int *rxsegsizep) 241 { 242 #ifdef INET 243 struct inpcb *inp = tp->t_inpcb; 244 #endif 245 #ifdef INET6 246 struct in6pcb *in6p = tp->t_in6pcb; 247 #endif 248 struct socket *so = NULL; 249 struct rtentry *rt; 250 struct ifnet *ifp; 251 int size; 252 int iphlen; 253 int optlen; 254 255 #ifdef DIAGNOSTIC 256 if (tp->t_inpcb && tp->t_in6pcb) 257 panic("tcp_segsize: both t_inpcb and t_in6pcb are set"); 258 #endif 259 switch (tp->t_family) { 260 #ifdef INET 261 case AF_INET: 262 iphlen = sizeof(struct ip); 263 break; 264 #endif 265 #ifdef INET6 266 case AF_INET6: 267 iphlen = sizeof(struct ip6_hdr); 268 break; 269 #endif 270 default: 271 size = tcp_mssdflt; 272 goto out; 273 } 274 275 rt = NULL; 276 #ifdef INET 277 if (inp) { 278 rt = in_pcbrtentry(inp); 279 so = inp->inp_socket; 280 } 281 #endif 282 #ifdef INET6 283 if (in6p) { 284 rt = in6_pcbrtentry(in6p); 285 so = in6p->in6p_socket; 286 } 287 #endif 288 if (rt == NULL) { 289 size = tcp_mssdflt; 290 goto out; 291 } 292 293 ifp = rt->rt_ifp; 294 295 size = tcp_mssdflt; 296 if (tp->t_mtudisc && rt->rt_rmx.rmx_mtu != 0) { 297 #ifdef INET6 298 if (in6p && rt->rt_rmx.rmx_mtu < IPV6_MMTU) { 299 /* 300 * RFC2460 section 5, last paragraph: if path MTU is 301 * smaller than 1280, use 1280 as packet size and 302 * attach fragment header. 303 */ 304 size = IPV6_MMTU - iphlen - sizeof(struct ip6_frag) - 305 sizeof(struct tcphdr); 306 } else 307 size = rt->rt_rmx.rmx_mtu - iphlen - 308 sizeof(struct tcphdr); 309 #else 310 size = rt->rt_rmx.rmx_mtu - iphlen - sizeof(struct tcphdr); 311 #endif 312 } else if (ifp->if_flags & IFF_LOOPBACK) 313 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 314 #ifdef INET 315 else if (inp && tp->t_mtudisc) 316 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 317 else if (inp && in_localaddr(inp->inp_faddr)) 318 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 319 #endif 320 #ifdef INET6 321 else if (in6p) { 322 #ifdef INET 323 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) { 324 /* mapped addr case */ 325 struct in_addr d; 326 bcopy(&in6p->in6p_faddr.s6_addr32[3], &d, sizeof(d)); 327 if (tp->t_mtudisc || in_localaddr(d)) 328 size = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 329 } else 330 #endif 331 { 332 /* 333 * for IPv6, path MTU discovery is always turned on, 334 * or the node must use packet size <= 1280. 335 */ 336 size = tp->t_mtudisc ? IN6_LINKMTU(ifp) : IPV6_MMTU; 337 size -= (iphlen + sizeof(struct tcphdr)); 338 } 339 } 340 #endif 341 out: 342 /* 343 * Now we must make room for whatever extra TCP/IP options are in 344 * the packet. 345 */ 346 optlen = tcp_optlen(tp); 347 348 /* 349 * XXX tp->t_ourmss should have the right size, but without this code 350 * fragmentation will occur... need more investigation 351 */ 352 #ifdef INET 353 if (inp) { 354 #if defined(IPSEC) || defined(FAST_IPSEC) 355 if (! IPSEC_PCB_SKIP_IPSEC(inp->inp_sp, IPSEC_DIR_OUTBOUND)) 356 optlen += ipsec4_hdrsiz_tcp(tp); 357 #endif 358 optlen += ip_optlen(inp); 359 } 360 #endif 361 #ifdef INET6 362 #ifdef INET 363 if (in6p && tp->t_family == AF_INET) { 364 #if defined(IPSEC) || defined(FAST_IPSEC) 365 if (! IPSEC_PCB_SKIP_IPSEC(in6p->in6p_sp, IPSEC_DIR_OUTBOUND)) 366 optlen += ipsec4_hdrsiz_tcp(tp); 367 #endif 368 /* XXX size -= ip_optlen(in6p); */ 369 } else 370 #endif 371 if (in6p && tp->t_family == AF_INET6) { 372 #ifdef IPSEC 373 if (! IPSEC_PCB_SKIP_IPSEC(in6p->in6p_sp, IPSEC_DIR_OUTBOUND)) 374 optlen += ipsec6_hdrsiz_tcp(tp); 375 #endif 376 optlen += ip6_optlen(in6p); 377 } 378 #endif 379 size -= optlen; 380 381 /* there may not be any room for data if mtu is too small */ 382 if (size < 0) 383 return (EMSGSIZE); 384 385 /* 386 * *rxsegsizep holds *estimated* inbound segment size (estimation 387 * assumes that path MTU is the same for both ways). this is only 388 * for silly window avoidance, do not use the value for other purposes. 389 * 390 * ipseclen is subtracted from both sides, this may not be right. 391 * I'm not quite sure about this (could someone comment). 392 */ 393 *txsegsizep = min(tp->t_peermss - optlen, size); 394 /* 395 * Never send more than half a buffer full. This insures that we can 396 * always keep 2 packets on the wire, no matter what SO_SNDBUF is, and 397 * therefore acks will never be delayed unless we run out of data to 398 * transmit. 399 */ 400 if (so) 401 *txsegsizep = min(so->so_snd.sb_hiwat >> 1, *txsegsizep); 402 *rxsegsizep = min(tp->t_ourmss - optlen, size); 403 404 if (*txsegsizep != tp->t_segsz) { 405 /* 406 * If the new segment size is larger, we don't want to 407 * mess up the congestion window, but if it is smaller 408 * we'll have to reduce the congestion window to ensure 409 * that we don't get into trouble with initial windows 410 * and the rest. In any case, if the segment size 411 * has changed, chances are the path has, too, and 412 * our congestion window will be different. 413 */ 414 if (*txsegsizep < tp->t_segsz) { 415 tp->snd_cwnd = max((tp->snd_cwnd / tp->t_segsz) 416 * *txsegsizep, *txsegsizep); 417 tp->snd_ssthresh = max((tp->snd_ssthresh / tp->t_segsz) 418 * *txsegsizep, *txsegsizep); 419 } 420 tp->t_segsz = *txsegsizep; 421 } 422 423 return (0); 424 } 425 426 static 427 #ifndef GPROF 428 __inline 429 #endif 430 int 431 tcp_build_datapkt(struct tcpcb *tp, struct socket *so, int off, 432 long len, int hdrlen, struct mbuf **mp) 433 { 434 struct mbuf *m, *m0; 435 436 if (tp->t_force && len == 1) 437 tcpstat.tcps_sndprobe++; 438 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { 439 tcpstat.tcps_sndrexmitpack++; 440 tcpstat.tcps_sndrexmitbyte += len; 441 } else { 442 tcpstat.tcps_sndpack++; 443 tcpstat.tcps_sndbyte += len; 444 } 445 #ifdef notyet 446 if ((m = m_copypack(so->so_snd.sb_mb, off, 447 (int)len, max_linkhdr + hdrlen)) == 0) 448 return (ENOBUFS); 449 /* 450 * m_copypack left space for our hdr; use it. 451 */ 452 m->m_len += hdrlen; 453 m->m_data -= hdrlen; 454 #else 455 MGETHDR(m, M_DONTWAIT, MT_HEADER); 456 if (__predict_false(m == NULL)) 457 return (ENOBUFS); 458 MCLAIM(m, &tcp_tx_mowner); 459 460 /* 461 * XXX Because other code assumes headers will fit in 462 * XXX one header mbuf. 463 * 464 * (This code should almost *never* be run.) 465 */ 466 if (__predict_false((max_linkhdr + hdrlen) > MHLEN)) { 467 TCP_OUTPUT_COUNTER_INCR(&tcp_output_bigheader); 468 MCLGET(m, M_DONTWAIT); 469 if ((m->m_flags & M_EXT) == 0) { 470 m_freem(m); 471 return (ENOBUFS); 472 } 473 } 474 475 m->m_data += max_linkhdr; 476 m->m_len = hdrlen; 477 478 /* 479 * To avoid traversing the whole sb_mb chain for correct 480 * data to send, remember last sent mbuf, its offset and 481 * the sent size. When called the next time, see if the 482 * data to send is directly following the previous transfer. 483 * This is important for large TCP windows. 484 */ 485 if (off == 0 || tp->t_lastm == NULL || 486 (tp->t_lastoff + tp->t_lastlen) != off) { 487 TCP_OUTPUT_COUNTER_INCR(&tcp_output_predict_miss); 488 /* 489 * Either a new packet or a retransmit. 490 * Start from the beginning. 491 */ 492 tp->t_lastm = so->so_snd.sb_mb; 493 tp->t_inoff = off; 494 } else { 495 TCP_OUTPUT_COUNTER_INCR(&tcp_output_predict_hit); 496 tp->t_inoff += tp->t_lastlen; 497 } 498 499 /* Traverse forward to next packet */ 500 while (tp->t_inoff > 0) { 501 if (tp->t_lastm == NULL) 502 panic("tp->t_lastm == NULL"); 503 if (tp->t_inoff < tp->t_lastm->m_len) 504 break; 505 tp->t_inoff -= tp->t_lastm->m_len; 506 tp->t_lastm = tp->t_lastm->m_next; 507 } 508 509 tp->t_lastoff = off; 510 tp->t_lastlen = len; 511 m0 = tp->t_lastm; 512 off = tp->t_inoff; 513 514 if (len <= M_TRAILINGSPACE(m)) { 515 m_copydata(m0, off, (int) len, mtod(m, caddr_t) + hdrlen); 516 m->m_len += len; 517 TCP_OUTPUT_COUNTER_INCR(&tcp_output_copysmall); 518 } else { 519 m->m_next = m_copy(m0, off, (int) len); 520 if (m->m_next == NULL) { 521 m_freem(m); 522 return (ENOBUFS); 523 } 524 #ifdef TCP_OUTPUT_COUNTERS 525 if (m->m_next->m_flags & M_EXT) 526 TCP_OUTPUT_COUNTER_INCR(&tcp_output_refbig); 527 else 528 TCP_OUTPUT_COUNTER_INCR(&tcp_output_copybig); 529 #endif /* TCP_OUTPUT_COUNTERS */ 530 } 531 #endif 532 533 *mp = m; 534 return (0); 535 } 536 537 /* 538 * Tcp output routine: figure out what should be sent and send it. 539 */ 540 int 541 tcp_output(struct tcpcb *tp) 542 { 543 struct socket *so; 544 struct route *ro; 545 long len, win; 546 int off, flags, error; 547 struct mbuf *m; 548 struct ip *ip; 549 #ifdef INET6 550 struct ip6_hdr *ip6; 551 #endif 552 struct tcphdr *th; 553 u_char opt[MAX_TCPOPTLEN]; 554 unsigned optlen, hdrlen; 555 int idle, sendalot, txsegsize, rxsegsize; 556 int maxburst = TCP_MAXBURST; 557 int af; /* address family on the wire */ 558 int iphdrlen; 559 int 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 use_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 again: 673 /* 674 * Determine length of data that should be transmitted, and 675 * flags that should be used. If there is some data or critical 676 * controls (SYN, RST) to send, then transmit; otherwise, 677 * investigate further. 678 * 679 * Readjust SACK information to avoid resending duplicate data. 680 */ 681 if (TCP_SACK_ENABLED(tp) && SEQ_LT(tp->snd_nxt, tp->snd_max)) 682 tcp_sack_adjust(tp); 683 sendalot = 0; 684 off = tp->snd_nxt - tp->snd_una; 685 win = min(tp->snd_wnd, tp->snd_cwnd); 686 687 flags = tcp_outflags[tp->t_state]; 688 689 /* 690 * Send any SACK-generated retransmissions. If we're explicitly trying 691 * to send out new data (when sendalot is 1), bypass this function. 692 * If we retransmit in fast recovery mode, decrement snd_cwnd, since 693 * we're replacing a (future) new transmission with a retransmission 694 * now, and we previously incremented snd_cwnd in tcp_input(). 695 */ 696 /* 697 * Still in sack recovery , reset rxmit flag to zero. 698 */ 699 sack_rxmit = 0; 700 sack_bytes_rxmt = 0; 701 len = 0; 702 p = NULL; 703 do { 704 long cwin; 705 if (!TCP_SACK_ENABLED(tp)) 706 break; 707 if (tp->t_partialacks == 0) 708 break; 709 p = tcp_sack_output(tp, &sack_bytes_rxmt); 710 if (p == NULL) 711 break; 712 713 cwin = min(tp->snd_wnd, tp->snd_cwnd) - sack_bytes_rxmt; 714 if (cwin < 0) 715 cwin = 0; 716 /* Do not retransmit SACK segments beyond snd_recover */ 717 if (SEQ_GT(p->end, tp->snd_recover)) { 718 /* 719 * (At least) part of sack hole extends beyond 720 * snd_recover. Check to see if we can rexmit data 721 * for this hole. 722 */ 723 if (SEQ_GEQ(p->rxmit, tp->snd_recover)) { 724 /* 725 * Can't rexmit any more data for this hole. 726 * That data will be rexmitted in the next 727 * sack recovery episode, when snd_recover 728 * moves past p->rxmit. 729 */ 730 p = NULL; 731 break; 732 } 733 /* Can rexmit part of the current hole */ 734 len = ((long)ulmin(cwin, tp->snd_recover - p->rxmit)); 735 } else 736 len = ((long)ulmin(cwin, p->end - p->rxmit)); 737 off = p->rxmit - tp->snd_una; 738 if (len > 0) { 739 sack_rxmit = 1; 740 sendalot = 1; 741 } 742 } while (0); 743 744 /* 745 * If in persist timeout with window of 0, send 1 byte. 746 * Otherwise, if window is small but nonzero 747 * and timer expired, we will send what we can 748 * and go to transmit state. 749 */ 750 if (tp->t_force) { 751 if (win == 0) { 752 /* 753 * If we still have some data to send, then 754 * clear the FIN bit. Usually this would 755 * happen below when it realizes that we 756 * aren't sending all the data. However, 757 * if we have exactly 1 byte of unset data, 758 * then it won't clear the FIN bit below, 759 * and if we are in persist state, we wind 760 * up sending the packet without recording 761 * that we sent the FIN bit. 762 * 763 * We can't just blindly clear the FIN bit, 764 * because if we don't have any more data 765 * to send then the probe will be the FIN 766 * itself. 767 */ 768 if (off < so->so_snd.sb_cc) 769 flags &= ~TH_FIN; 770 win = 1; 771 } else { 772 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 773 tp->t_rxtshift = 0; 774 } 775 } 776 777 if (!TCP_SACK_ENABLED(tp)) { 778 if (win < so->so_snd.sb_cc) { 779 len = win - off; 780 flags &= ~TH_FIN; 781 } else 782 len = so->so_snd.sb_cc - off; 783 } else if (sack_rxmit == 0) { 784 if (sack_bytes_rxmt != 0) { 785 long cwin; 786 787 /* 788 * We are inside of a SACK recovery episode and are 789 * sending new data, having retransmitted all the 790 * data possible in the scoreboard. 791 */ 792 len = ((long)ulmin(so->so_snd.sb_cc, tp->snd_wnd) 793 - off); 794 /* 795 * From FreeBSD: 796 * Don't remove this (len > 0) check ! 797 * We explicitly check for len > 0 here (although it 798 * isn't really necessary), to work around a gcc 799 * optimization issue - to force gcc to compute 800 * len above. Without this check, the computation 801 * of len is bungled by the optimizer. 802 */ 803 if (len > 0) { 804 cwin = tp->snd_cwnd - 805 (tp->snd_nxt - tp->sack_newdata) - 806 sack_bytes_rxmt; 807 if (cwin < 0) 808 cwin = 0; 809 len = lmin(len, cwin); 810 } 811 } else if (win < so->so_snd.sb_cc) { 812 len = win - off; 813 flags &= ~TH_FIN; 814 } else 815 len = so->so_snd.sb_cc - off; 816 } 817 818 if (len < 0) { 819 /* 820 * If FIN has been sent but not acked, 821 * but we haven't been called to retransmit, 822 * len will be -1. Otherwise, window shrank 823 * after we sent into it. If window shrank to 0, 824 * cancel pending retransmit, pull snd_nxt back 825 * to (closed) window, and set the persist timer 826 * if it isn't already going. If the window didn't 827 * close completely, just wait for an ACK. 828 * 829 * If we have a pending FIN, either it has already been 830 * transmitted or it is outside the window, so drop it. 831 * If the FIN has been transmitted, but this is not a 832 * retransmission, then len must be -1. Therefore we also 833 * prevent here the sending of `gratuitous FINs'. This 834 * eliminates the need to check for that case below (e.g. 835 * to back up snd_nxt before the FIN so that the sequence 836 * number is correct). 837 */ 838 len = 0; 839 flags &= ~TH_FIN; 840 if (win == 0) { 841 TCP_TIMER_DISARM(tp, TCPT_REXMT); 842 tp->t_rxtshift = 0; 843 tp->snd_nxt = tp->snd_una; 844 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 845 tcp_setpersist(tp); 846 } 847 } 848 if (len > txsegsize) { 849 if (use_tso) { 850 /* 851 * Truncate TSO transfers to IP_MAXPACKET, and make 852 * sure that we send equal size transfers down the 853 * stack (rather than big-small-big-small-...). 854 */ 855 len = (min(len, IP_MAXPACKET) / txsegsize) * txsegsize; 856 } else 857 len = txsegsize; 858 flags &= ~TH_FIN; 859 sendalot = 1; 860 } else 861 use_tso = 0; 862 if (sack_rxmit) { 863 if (SEQ_LT(p->rxmit + len, tp->snd_una + so->so_snd.sb_cc)) 864 flags &= ~TH_FIN; 865 } 866 867 win = sbspace(&so->so_rcv); 868 869 /* 870 * Sender silly window avoidance. If connection is idle 871 * and can send all data, a maximum segment, 872 * at least a maximum default-size segment do it, 873 * or are forced, do it; otherwise don't bother. 874 * If peer's buffer is tiny, then send 875 * when window is at least half open. 876 * If retransmitting (possibly after persist timer forced us 877 * to send into a small window), then must resend. 878 */ 879 if (len) { 880 if (len >= txsegsize) 881 goto send; 882 if ((so->so_state & SS_MORETOCOME) == 0 && 883 ((idle || tp->t_flags & TF_NODELAY) && 884 len + off >= so->so_snd.sb_cc)) 885 goto send; 886 if (tp->t_force) 887 goto send; 888 if (len >= tp->max_sndwnd / 2) 889 goto send; 890 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) 891 goto send; 892 if (sack_rxmit) 893 goto send; 894 } 895 896 /* 897 * Compare available window to amount of window known to peer 898 * (as advertised window less next expected input). If the 899 * difference is at least twice the size of the largest segment 900 * we expect to receive (i.e. two segments) or at least 50% of 901 * the maximum possible window, then want to send a window update 902 * to peer. 903 */ 904 if (win > 0) { 905 /* 906 * "adv" is the amount we can increase the window, 907 * taking into account that we are limited by 908 * TCP_MAXWIN << tp->rcv_scale. 909 */ 910 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) - 911 (tp->rcv_adv - tp->rcv_nxt); 912 913 if (adv >= (long) (2 * rxsegsize)) 914 goto send; 915 if (2 * adv >= (long) so->so_rcv.sb_hiwat) 916 goto send; 917 } 918 919 /* 920 * Send if we owe peer an ACK. 921 */ 922 if (tp->t_flags & TF_ACKNOW) 923 goto send; 924 if (flags & (TH_SYN|TH_FIN|TH_RST)) 925 goto send; 926 if (SEQ_GT(tp->snd_up, tp->snd_una)) 927 goto send; 928 /* 929 * In SACK, it is possible for tcp_output to fail to send a segment 930 * after the retransmission timer has been turned off. Make sure 931 * that the retransmission timer is set. 932 */ 933 if (TCP_SACK_ENABLED(tp) && SEQ_GT(tp->snd_max, tp->snd_una) && 934 !TCP_TIMER_ISARMED(tp, TCPT_REXMT) && 935 !TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) { 936 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 937 goto just_return; 938 } 939 940 /* 941 * TCP window updates are not reliable, rather a polling protocol 942 * using ``persist'' packets is used to insure receipt of window 943 * updates. The three ``states'' for the output side are: 944 * idle not doing retransmits or persists 945 * persisting to move a small or zero window 946 * (re)transmitting and thereby not persisting 947 * 948 * tp->t_timer[TCPT_PERSIST] 949 * is set when we are in persist state. 950 * tp->t_force 951 * is set when we are called to send a persist packet. 952 * tp->t_timer[TCPT_REXMT] 953 * is set when we are retransmitting 954 * The output side is idle when both timers are zero. 955 * 956 * If send window is too small, there is data to transmit, and no 957 * retransmit or persist is pending, then go to persist state. 958 * If nothing happens soon, send when timer expires: 959 * if window is nonzero, transmit what we can, 960 * otherwise force out a byte. 961 */ 962 if (so->so_snd.sb_cc && TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 963 TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 964 tp->t_rxtshift = 0; 965 tcp_setpersist(tp); 966 } 967 968 /* 969 * No reason to send a segment, just return. 970 */ 971 just_return: 972 return (0); 973 974 send: 975 /* 976 * Before ESTABLISHED, force sending of initial options 977 * unless TCP set not to do any options. 978 * NOTE: we assume that the IP/TCP header plus TCP options 979 * always fit in a single mbuf, leaving room for a maximum 980 * link header, i.e. 981 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES 982 */ 983 optlen = 0; 984 switch (af) { 985 #ifdef INET 986 case AF_INET: 987 iphdrlen = sizeof(struct ip) + sizeof(struct tcphdr); 988 break; 989 #endif 990 #ifdef INET6 991 case AF_INET6: 992 iphdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 993 break; 994 #endif 995 default: /*pacify gcc*/ 996 iphdrlen = 0; 997 break; 998 } 999 hdrlen = iphdrlen; 1000 if (flags & TH_SYN) { 1001 struct rtentry *rt; 1002 1003 rt = NULL; 1004 #ifdef INET 1005 if (tp->t_inpcb) 1006 rt = in_pcbrtentry(tp->t_inpcb); 1007 #endif 1008 #ifdef INET6 1009 if (tp->t_in6pcb) 1010 rt = in6_pcbrtentry(tp->t_in6pcb); 1011 #endif 1012 1013 tp->snd_nxt = tp->iss; 1014 tp->t_ourmss = tcp_mss_to_advertise(rt != NULL ? 1015 rt->rt_ifp : NULL, af); 1016 if ((tp->t_flags & TF_NOOPT) == 0) { 1017 opt[0] = TCPOPT_MAXSEG; 1018 opt[1] = 4; 1019 opt[2] = (tp->t_ourmss >> 8) & 0xff; 1020 opt[3] = tp->t_ourmss & 0xff; 1021 optlen = 4; 1022 1023 if ((tp->t_flags & TF_REQ_SCALE) && 1024 ((flags & TH_ACK) == 0 || 1025 (tp->t_flags & TF_RCVD_SCALE))) { 1026 *((u_int32_t *) (opt + optlen)) = htonl( 1027 TCPOPT_NOP << 24 | 1028 TCPOPT_WINDOW << 16 | 1029 TCPOLEN_WINDOW << 8 | 1030 tp->request_r_scale); 1031 optlen += 4; 1032 } 1033 if (tcp_do_sack) { 1034 u_int8_t *p = (u_int8_t *)(opt + optlen); 1035 1036 p[0] = TCPOPT_SACK_PERMITTED; 1037 p[1] = 2; 1038 p[2] = TCPOPT_NOP; 1039 p[3] = TCPOPT_NOP; 1040 optlen += 4; 1041 } 1042 } 1043 } 1044 1045 /* 1046 * Send a timestamp and echo-reply if this is a SYN and our side 1047 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side 1048 * and our peer have sent timestamps in our SYN's. 1049 */ 1050 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1051 (flags & TH_RST) == 0 && 1052 ((flags & (TH_SYN|TH_ACK)) == TH_SYN || 1053 (tp->t_flags & TF_RCVD_TSTMP))) { 1054 u_int32_t *lp = (u_int32_t *)(opt + optlen); 1055 1056 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1057 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1058 *lp++ = htonl(TCP_TIMESTAMP(tp)); 1059 *lp = htonl(tp->ts_recent); 1060 optlen += TCPOLEN_TSTAMP_APPA; 1061 } 1062 1063 /* 1064 * Tack on the SACK block if it is necessary. 1065 */ 1066 if (TCP_SACK_ENABLED(tp) && (tp->t_flags & TF_ACKNOW) 1067 && (tp->rcv_sack_num > 0)) { 1068 int sack_len, i; 1069 u_char *bp = (u_char *)(opt + optlen); 1070 u_int32_t *lp = (u_int32_t *)(bp + 4); 1071 1072 sack_len = tp->rcv_sack_num * 8 + 2; 1073 bp[0] = TCPOPT_NOP; 1074 bp[1] = TCPOPT_NOP; 1075 bp[2] = TCPOPT_SACK; 1076 bp[3] = sack_len; 1077 for (i = 0; i < tp->rcv_sack_num; i++) { 1078 *lp++ = htonl(tp->rcv_sack_block[i].left); 1079 *lp++ = htonl(tp->rcv_sack_block[i].right); 1080 } 1081 tp->rcv_sack_num = 0; 1082 optlen += sack_len + 2; 1083 } 1084 1085 #ifdef TCP_SIGNATURE 1086 #if defined(INET6) && defined(FAST_IPSEC) 1087 if (tp->t_family == AF_INET) 1088 #endif 1089 if (tp->t_flags & TF_SIGNATURE) { 1090 u_char *bp; 1091 /* 1092 * Initialize TCP-MD5 option (RFC2385) 1093 */ 1094 bp = (u_char *)opt + optlen; 1095 *bp++ = TCPOPT_SIGNATURE; 1096 *bp++ = TCPOLEN_SIGNATURE; 1097 sigoff = optlen + 2; 1098 bzero(bp, TCP_SIGLEN); 1099 bp += TCP_SIGLEN; 1100 optlen += TCPOLEN_SIGNATURE; 1101 /* 1102 * Terminate options list and maintain 32-bit alignment. 1103 */ 1104 *bp++ = TCPOPT_NOP; 1105 *bp++ = TCPOPT_EOL; 1106 optlen += 2; 1107 } 1108 #endif /* TCP_SIGNATURE */ 1109 1110 hdrlen += optlen; 1111 1112 #ifdef DIAGNOSTIC 1113 if (!use_tso && len > txsegsize) 1114 panic("tcp data to be sent is larger than segment"); 1115 else if (use_tso && len > IP_MAXPACKET) 1116 panic("tcp data to be sent is larger than max TSO size"); 1117 if (max_linkhdr + hdrlen > MCLBYTES) 1118 panic("tcphdr too big"); 1119 #endif 1120 1121 /* 1122 * Grab a header mbuf, attaching a copy of data to 1123 * be transmitted, and initialize the header from 1124 * the template for sends on this connection. 1125 */ 1126 if (len) { 1127 error = tcp_build_datapkt(tp, so, off, len, hdrlen, &m); 1128 if (error) 1129 goto out; 1130 /* 1131 * If we're sending everything we've got, set PUSH. 1132 * (This will keep happy those implementations which only 1133 * give data to the user when a buffer fills or 1134 * a PUSH comes in.) 1135 */ 1136 if (off + len == so->so_snd.sb_cc) 1137 flags |= TH_PUSH; 1138 } else { 1139 if (tp->t_flags & TF_ACKNOW) 1140 tcpstat.tcps_sndacks++; 1141 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 1142 tcpstat.tcps_sndctrl++; 1143 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 1144 tcpstat.tcps_sndurg++; 1145 else 1146 tcpstat.tcps_sndwinup++; 1147 1148 MGETHDR(m, M_DONTWAIT, MT_HEADER); 1149 if (m != NULL && max_linkhdr + hdrlen > MHLEN) { 1150 MCLGET(m, M_DONTWAIT); 1151 if ((m->m_flags & M_EXT) == 0) { 1152 m_freem(m); 1153 m = NULL; 1154 } 1155 } 1156 if (m == NULL) { 1157 error = ENOBUFS; 1158 goto out; 1159 } 1160 MCLAIM(m, &tcp_tx_mowner); 1161 m->m_data += max_linkhdr; 1162 m->m_len = hdrlen; 1163 } 1164 m->m_pkthdr.rcvif = (struct ifnet *)0; 1165 switch (af) { 1166 #ifdef INET 1167 case AF_INET: 1168 ip = mtod(m, struct ip *); 1169 #ifdef INET6 1170 ip6 = NULL; 1171 #endif 1172 th = (struct tcphdr *)(ip + 1); 1173 break; 1174 #endif 1175 #ifdef INET6 1176 case AF_INET6: 1177 ip = NULL; 1178 ip6 = mtod(m, struct ip6_hdr *); 1179 th = (struct tcphdr *)(ip6 + 1); 1180 break; 1181 #endif 1182 default: /*pacify gcc*/ 1183 ip = NULL; 1184 #ifdef INET6 1185 ip6 = NULL; 1186 #endif 1187 th = NULL; 1188 break; 1189 } 1190 if (tp->t_template == 0) 1191 panic("tcp_output"); 1192 if (tp->t_template->m_len < iphdrlen) 1193 panic("tcp_output"); 1194 bcopy(mtod(tp->t_template, caddr_t), mtod(m, caddr_t), iphdrlen); 1195 1196 /* 1197 * If we are doing retransmissions, then snd_nxt will 1198 * not reflect the first unsent octet. For ACK only 1199 * packets, we do not want the sequence number of the 1200 * retransmitted packet, we want the sequence number 1201 * of the next unsent octet. So, if there is no data 1202 * (and no SYN or FIN), use snd_max instead of snd_nxt 1203 * when filling in ti_seq. But if we are in persist 1204 * state, snd_max might reflect one byte beyond the 1205 * right edge of the window, so use snd_nxt in that 1206 * case, since we know we aren't doing a retransmission. 1207 * (retransmit and persist are mutually exclusive...) 1208 */ 1209 if (TCP_SACK_ENABLED(tp) && sack_rxmit) { 1210 th->th_seq = htonl(p->rxmit); 1211 p->rxmit += len; 1212 } else { 1213 if (len || (flags & (TH_SYN|TH_FIN)) || 1214 TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) 1215 th->th_seq = htonl(tp->snd_nxt); 1216 else 1217 th->th_seq = htonl(tp->snd_max); 1218 } 1219 th->th_ack = htonl(tp->rcv_nxt); 1220 if (optlen) { 1221 bcopy((caddr_t)opt, (caddr_t)(th + 1), optlen); 1222 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1223 } 1224 th->th_flags = flags; 1225 /* 1226 * Calculate receive window. Don't shrink window, 1227 * but avoid silly window syndrome. 1228 */ 1229 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)rxsegsize) 1230 win = 0; 1231 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 1232 win = (long)TCP_MAXWIN << tp->rcv_scale; 1233 if (win < (long)(int32_t)(tp->rcv_adv - tp->rcv_nxt)) 1234 win = (long)(int32_t)(tp->rcv_adv - tp->rcv_nxt); 1235 th->th_win = htons((u_int16_t) (win>>tp->rcv_scale)); 1236 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 1237 u_int32_t urp = tp->snd_up - tp->snd_nxt; 1238 if (urp > IP_MAXPACKET) 1239 urp = IP_MAXPACKET; 1240 th->th_urp = htons((u_int16_t)urp); 1241 th->th_flags |= TH_URG; 1242 } else 1243 /* 1244 * If no urgent pointer to send, then we pull 1245 * the urgent pointer to the left edge of the send window 1246 * so that it doesn't drift into the send window on sequence 1247 * number wraparound. 1248 */ 1249 tp->snd_up = tp->snd_una; /* drag it along */ 1250 1251 #ifdef TCP_SIGNATURE 1252 #if defined(INET6) && defined(FAST_IPSEC) 1253 if (tp->t_family == AF_INET) /* XXX */ 1254 #endif 1255 if (sigoff && (tp->t_flags & TF_SIGNATURE)) { 1256 struct secasvar *sav; 1257 u_int8_t *sigp; 1258 1259 sav = tcp_signature_getsav(m, th); 1260 1261 if (sav == NULL) { 1262 if (m) 1263 m_freem(m); 1264 return (EPERM); 1265 } 1266 1267 m->m_pkthdr.len = hdrlen + len; 1268 sigp = (caddr_t)th + sizeof(*th) + sigoff; 1269 tcp_signature(m, th, (caddr_t)th - mtod(m, caddr_t), sav, sigp); 1270 1271 key_sa_recordxfer(sav, m); 1272 #ifdef FAST_IPSEC 1273 KEY_FREESAV(&sav); 1274 #else 1275 key_freesav(sav); 1276 #endif 1277 } 1278 #endif 1279 1280 /* 1281 * Set ourselves up to be checksummed just before the packet 1282 * hits the wire. Maybe skip checksums on loopback interfaces. 1283 */ 1284 switch (af) { 1285 #ifdef INET 1286 case AF_INET: 1287 if (use_tso) { 1288 m->m_pkthdr.segsz = txsegsize; 1289 m->m_pkthdr.csum_flags |= M_CSUM_TSOv4; 1290 } else { 1291 if (__predict_true(ro->ro_rt == NULL || 1292 !(ro->ro_rt->rt_ifp->if_flags & 1293 IFF_LOOPBACK) || 1294 tcp_do_loopback_cksum)) 1295 m->m_pkthdr.csum_flags = M_CSUM_TCPv4; 1296 else 1297 m->m_pkthdr.csum_flags = 0; 1298 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1299 if (len + optlen) { 1300 /* Fixup the pseudo-header checksum. */ 1301 /* XXXJRT Not IP Jumbogram safe. */ 1302 th->th_sum = in_cksum_addword(th->th_sum, 1303 htons((u_int16_t) (len + optlen))); 1304 } 1305 } 1306 break; 1307 #endif 1308 #ifdef INET6 1309 case AF_INET6: 1310 /* 1311 * XXX Actually delaying the checksum is Hard 1312 * XXX (well, maybe not for Itojun, but it is 1313 * XXX for me), but we can still take advantage 1314 * XXX of the cached pseudo-header checksum. 1315 */ 1316 /* equals to hdrlen + len */ 1317 m->m_pkthdr.len = sizeof(struct ip6_hdr) 1318 + sizeof(struct tcphdr) + optlen + len; 1319 #ifdef notyet 1320 if (__predict_true(ro->ro_rt == NULL || 1321 !(ro->ro_rt->rt_ifp->if_flags & 1322 IFF_LOOPBACK) || 1323 tcp_do_loopback_cksum)) 1324 m->m_pkthdr.csum_flags = M_CSUM_TCPv6; 1325 else 1326 m->m_pkthdr.csum_flags = 0; 1327 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1328 #endif 1329 if (len + optlen) { 1330 /* Fixup the pseudo-header checksum. */ 1331 /* XXXJRT: Not IPv6 Jumbogram safe. */ 1332 th->th_sum = in_cksum_addword(th->th_sum, 1333 htons((u_int16_t) (len + optlen))); 1334 } 1335 #ifndef notyet 1336 if (__predict_true(ro->ro_rt == NULL || 1337 !(ro->ro_rt->rt_ifp->if_flags & 1338 IFF_LOOPBACK) || 1339 tcp_do_loopback_cksum)) 1340 th->th_sum = in6_cksum(m, 0, sizeof(struct ip6_hdr), 1341 sizeof(struct tcphdr) + optlen + len); 1342 #endif 1343 break; 1344 #endif 1345 } 1346 1347 /* 1348 * In transmit state, time the transmission and arrange for 1349 * the retransmit. In persist state, just set snd_max. 1350 */ 1351 if (tp->t_force == 0 || TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 1352 tcp_seq startseq = tp->snd_nxt; 1353 1354 /* 1355 * Advance snd_nxt over sequence space of this segment. 1356 * There are no states in which we send both a SYN and a FIN, 1357 * so we collapse the tests for these flags. 1358 */ 1359 if (flags & (TH_SYN|TH_FIN)) 1360 tp->snd_nxt++; 1361 if (sack_rxmit) 1362 goto timer; 1363 tp->snd_nxt += len; 1364 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 1365 tp->snd_max = tp->snd_nxt; 1366 /* 1367 * Time this transmission if not a retransmission and 1368 * not currently timing anything. 1369 */ 1370 if (tp->t_rtttime == 0) { 1371 tp->t_rtttime = tcp_now; 1372 tp->t_rtseq = startseq; 1373 tcpstat.tcps_segstimed++; 1374 } 1375 } 1376 1377 /* 1378 * Set retransmit timer if not currently set, 1379 * and not doing an ack or a keep-alive probe. 1380 * Initial value for retransmit timer is smoothed 1381 * round-trip time + 2 * round-trip time variance. 1382 * Initialize shift counter which is used for backoff 1383 * of retransmit time. 1384 */ 1385 timer: 1386 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 1387 ((sack_rxmit && tp->snd_nxt != tp->snd_max) || 1388 tp->snd_nxt != tp->snd_una)) { 1389 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) { 1390 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 1391 tp->t_rxtshift = 0; 1392 } 1393 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1394 } 1395 } else 1396 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max)) 1397 tp->snd_max = tp->snd_nxt + len; 1398 1399 #ifdef TCP_DEBUG 1400 /* 1401 * Trace. 1402 */ 1403 if (so->so_options & SO_DEBUG) 1404 tcp_trace(TA_OUTPUT, tp->t_state, tp, m, 0); 1405 #endif 1406 1407 /* 1408 * Fill in IP length and desired time to live and 1409 * send to IP level. There should be a better way 1410 * to handle ttl and tos; we could keep them in 1411 * the template, but need a way to checksum without them. 1412 */ 1413 m->m_pkthdr.len = hdrlen + len; 1414 1415 switch (af) { 1416 #ifdef INET 1417 case AF_INET: 1418 ip->ip_len = htons(m->m_pkthdr.len); 1419 if (tp->t_inpcb) { 1420 ip->ip_ttl = tp->t_inpcb->inp_ip.ip_ttl; 1421 ip->ip_tos = tp->t_inpcb->inp_ip.ip_tos; 1422 } 1423 #ifdef INET6 1424 else if (tp->t_in6pcb) { 1425 ip->ip_ttl = in6_selecthlim(tp->t_in6pcb, NULL); /*XXX*/ 1426 ip->ip_tos = 0; /*XXX*/ 1427 } 1428 #endif 1429 break; 1430 #endif 1431 #ifdef INET6 1432 case AF_INET6: 1433 ip6->ip6_nxt = IPPROTO_TCP; 1434 if (tp->t_in6pcb) { 1435 /* 1436 * we separately set hoplimit for every segment, since 1437 * the user might want to change the value via 1438 * setsockopt. Also, desired default hop limit might 1439 * be changed via Neighbor Discovery. 1440 */ 1441 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, 1442 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 1443 } 1444 /* ip6->ip6_flow = ??? */ 1445 /* ip6_plen will be filled in ip6_output(). */ 1446 break; 1447 #endif 1448 } 1449 1450 switch (af) { 1451 #ifdef INET 1452 case AF_INET: 1453 { 1454 struct mbuf *opts; 1455 1456 if (tp->t_inpcb) 1457 opts = tp->t_inpcb->inp_options; 1458 else 1459 opts = NULL; 1460 error = ip_output(m, opts, ro, 1461 (tp->t_mtudisc ? IP_MTUDISC : 0) | 1462 (so->so_options & SO_DONTROUTE), 1463 (struct ip_moptions *)0, so); 1464 break; 1465 } 1466 #endif 1467 #ifdef INET6 1468 case AF_INET6: 1469 { 1470 struct ip6_pktopts *opts; 1471 1472 if (tp->t_in6pcb) 1473 opts = tp->t_in6pcb->in6p_outputopts; 1474 else 1475 opts = NULL; 1476 error = ip6_output(m, opts, (struct route_in6 *)ro, 1477 so->so_options & SO_DONTROUTE, 1478 (struct ip6_moptions *)0, so, NULL); 1479 break; 1480 } 1481 #endif 1482 default: 1483 error = EAFNOSUPPORT; 1484 break; 1485 } 1486 if (error) { 1487 out: 1488 if (error == ENOBUFS) { 1489 tcpstat.tcps_selfquench++; 1490 #ifdef INET 1491 if (tp->t_inpcb) 1492 tcp_quench(tp->t_inpcb, 0); 1493 #endif 1494 #ifdef INET6 1495 if (tp->t_in6pcb) 1496 tcp6_quench(tp->t_in6pcb, 0); 1497 #endif 1498 error = 0; 1499 } else if ((error == EHOSTUNREACH || error == ENETDOWN) && 1500 TCPS_HAVERCVDSYN(tp->t_state)) { 1501 tp->t_softerror = error; 1502 error = 0; 1503 } 1504 1505 /* Back out the seqence number advance. */ 1506 if (sack_rxmit) 1507 p->rxmit -= len; 1508 1509 /* Restart the delayed ACK timer, if necessary. */ 1510 if (tp->t_flags & TF_DELACK) 1511 TCP_RESTART_DELACK(tp); 1512 1513 return (error); 1514 } 1515 tcpstat.tcps_sndtotal++; 1516 if (tp->t_flags & TF_DELACK) 1517 tcpstat.tcps_delack++; 1518 1519 /* 1520 * Data sent (as far as we can tell). 1521 * If this advertises a larger window than any other segment, 1522 * then remember the size of the advertised window. 1523 * Any pending ACK has now been sent. 1524 */ 1525 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 1526 tp->rcv_adv = tp->rcv_nxt + win; 1527 tp->last_ack_sent = tp->rcv_nxt; 1528 tp->t_flags &= ~TF_ACKNOW; 1529 TCP_CLEAR_DELACK(tp); 1530 #ifdef DIAGNOSTIC 1531 if (maxburst < 0) 1532 printf("tcp_output: maxburst exceeded by %d\n", -maxburst); 1533 #endif 1534 if (sendalot && (!tcp_do_newreno || --maxburst)) 1535 goto again; 1536 return (0); 1537 } 1538 1539 void 1540 tcp_setpersist(struct tcpcb *tp) 1541 { 1542 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2); 1543 int nticks; 1544 1545 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT)) 1546 panic("tcp_output REXMT"); 1547 /* 1548 * Start/restart persistance timer. 1549 */ 1550 if (t < tp->t_rttmin) 1551 t = tp->t_rttmin; 1552 TCPT_RANGESET(nticks, t * tcp_backoff[tp->t_rxtshift], 1553 TCPTV_PERSMIN, TCPTV_PERSMAX); 1554 TCP_TIMER_ARM(tp, TCPT_PERSIST, nticks); 1555 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 1556 tp->t_rxtshift++; 1557 } 1558
Indexes created Tue Sep 23 14:10:03 GMT 2025