1 /* $NetBSD: tcp_output.c,v 1.180 2015/02/14 12:57:53 he 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, 2006 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 * This code is derived from software contributed to The NetBSD Foundation 82 * by Rui Paulo. 83 * 84 * Redistribution and use in source and binary forms, with or without 85 * modification, are permitted provided that the following conditions 86 * are met: 87 * 1. Redistributions of source code must retain the above copyright 88 * notice, this list of conditions and the following disclaimer. 89 * 2. Redistributions in binary form must reproduce the above copyright 90 * notice, this list of conditions and the following disclaimer in the 91 * documentation and/or other materials provided with the distribution. 92 * 93 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 94 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 95 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 96 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 97 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 98 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 99 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 100 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 101 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 102 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 103 * POSSIBILITY OF SUCH DAMAGE. 104 */ 105 106 /* 107 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 108 * The Regents of the University of California. All rights reserved. 109 * 110 * Redistribution and use in source and binary forms, with or without 111 * modification, are permitted provided that the following conditions 112 * are met: 113 * 1. Redistributions of source code must retain the above copyright 114 * notice, this list of conditions and the following disclaimer. 115 * 2. Redistributions in binary form must reproduce the above copyright 116 * notice, this list of conditions and the following disclaimer in the 117 * documentation and/or other materials provided with the distribution. 118 * 3. Neither the name of the University nor the names of its contributors 119 * may be used to endorse or promote products derived from this software 120 * without specific prior written permission. 121 * 122 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 123 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 124 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 125 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 126 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 127 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 128 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 129 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 130 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 131 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 132 * SUCH DAMAGE. 133 * 134 * @(#)tcp_output.c 8.4 (Berkeley) 5/24/95 135 */ 136 137 #include <sys/cdefs.h> 138 __KERNEL_RCSID(0, "$NetBSD: tcp_output.c,v 1.180 2015/02/14 12:57:53 he Exp $"); 139 140 #include "opt_inet.h" 141 #include "opt_ipsec.h" 142 #include "opt_tcp_debug.h" 143 144 #include <sys/param.h> 145 #include <sys/systm.h> 146 #include <sys/mbuf.h> 147 #include <sys/protosw.h> 148 #include <sys/socket.h> 149 #include <sys/socketvar.h> 150 #include <sys/errno.h> 151 #include <sys/domain.h> 152 #include <sys/kernel.h> 153 #ifdef TCP_SIGNATURE 154 #include <sys/md5.h> 155 #endif 156 157 #include <net/if.h> 158 #include <net/route.h> 159 160 #include <netinet/in.h> 161 #include <netinet/in_systm.h> 162 #include <netinet/ip.h> 163 #include <netinet/in_pcb.h> 164 #include <netinet/ip_var.h> 165 166 #ifdef INET6 167 #ifndef INET 168 #include <netinet/in.h> 169 #endif 170 #include <netinet/ip6.h> 171 #include <netinet6/in6_var.h> 172 #include <netinet6/ip6_var.h> 173 #include <netinet6/in6_pcb.h> 174 #include <netinet6/nd6.h> 175 #endif 176 177 #ifdef IPSEC 178 #include <netipsec/ipsec.h> 179 #include <netipsec/key.h> 180 #ifdef INET6 181 #include <netipsec/ipsec6.h> 182 #endif 183 #endif /* IPSEC*/ 184 185 #include <netinet/tcp.h> 186 #define TCPOUTFLAGS 187 #include <netinet/tcp_fsm.h> 188 #include <netinet/tcp_seq.h> 189 #include <netinet/tcp_timer.h> 190 #include <netinet/tcp_var.h> 191 #include <netinet/tcp_private.h> 192 #include <netinet/tcp_congctl.h> 193 #include <netinet/tcpip.h> 194 #include <netinet/tcp_debug.h> 195 #include <netinet/in_offload.h> 196 #include <netinet6/in6_offload.h> 197 198 #ifdef notyet 199 extern struct mbuf *m_copypack(); 200 #endif 201 202 /* 203 * Knob to enable Congestion Window Monitoring, and control 204 * the burst size it allows. Default burst is 4 packets, per 205 * the Internet draft. 206 */ 207 int tcp_cwm = 0; 208 int tcp_cwm_burstsize = 4; 209 210 int tcp_do_autosndbuf = 1; 211 int tcp_autosndbuf_inc = 8 * 1024; 212 int tcp_autosndbuf_max = 256 * 1024; 213 214 #ifdef TCP_OUTPUT_COUNTERS 215 #include <sys/device.h> 216 217 extern struct evcnt tcp_output_bigheader; 218 extern struct evcnt tcp_output_predict_hit; 219 extern struct evcnt tcp_output_predict_miss; 220 extern struct evcnt tcp_output_copysmall; 221 extern struct evcnt tcp_output_copybig; 222 extern struct evcnt tcp_output_refbig; 223 224 #define TCP_OUTPUT_COUNTER_INCR(ev) (ev)->ev_count++ 225 #else 226 227 #define TCP_OUTPUT_COUNTER_INCR(ev) /* nothing */ 228 229 #endif /* TCP_OUTPUT_COUNTERS */ 230 231 static 232 #ifndef GPROF 233 inline 234 #endif 235 int 236 tcp_segsize(struct tcpcb *tp, int *txsegsizep, int *rxsegsizep, 237 bool *alwaysfragp) 238 { 239 #ifdef INET 240 struct inpcb *inp = tp->t_inpcb; 241 #endif 242 #ifdef INET6 243 struct in6pcb *in6p = tp->t_in6pcb; 244 #endif 245 struct socket *so = NULL; 246 struct rtentry *rt; 247 struct ifnet *ifp; 248 int size; 249 int hdrlen; 250 int optlen; 251 252 *alwaysfragp = false; 253 254 #ifdef DIAGNOSTIC 255 if (tp->t_inpcb && tp->t_in6pcb) 256 panic("tcp_segsize: both t_inpcb and t_in6pcb are set"); 257 #endif 258 switch (tp->t_family) { 259 #ifdef INET 260 case AF_INET: 261 hdrlen = sizeof(struct ip) + sizeof(struct tcphdr); 262 break; 263 #endif 264 #ifdef INET6 265 case AF_INET6: 266 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 267 break; 268 #endif 269 default: 270 size = tcp_mssdflt; 271 goto out; 272 } 273 274 rt = NULL; 275 #ifdef INET 276 if (inp) { 277 rt = in_pcbrtentry(inp); 278 so = inp->inp_socket; 279 } 280 #endif 281 #ifdef INET6 282 if (in6p) { 283 rt = in6_pcbrtentry(in6p); 284 so = in6p->in6p_socket; 285 } 286 #endif 287 if (rt == NULL) { 288 size = tcp_mssdflt; 289 goto out; 290 } 291 292 ifp = rt->rt_ifp; 293 294 size = tcp_mssdflt; 295 if (tp->t_mtudisc && rt->rt_rmx.rmx_mtu != 0) { 296 #ifdef INET6 297 if (in6p && rt->rt_rmx.rmx_mtu < IPV6_MMTU) { 298 /* 299 * RFC2460 section 5, last paragraph: if path MTU is 300 * smaller than 1280, use 1280 as packet size and 301 * attach fragment header. 302 */ 303 size = IPV6_MMTU - hdrlen - sizeof(struct ip6_frag); 304 *alwaysfragp = true; 305 } else 306 size = rt->rt_rmx.rmx_mtu - hdrlen; 307 #else 308 size = rt->rt_rmx.rmx_mtu - hdrlen; 309 #endif 310 } else if (ifp->if_flags & IFF_LOOPBACK) 311 size = ifp->if_mtu - hdrlen; 312 #ifdef INET 313 else if (inp && tp->t_mtudisc) 314 size = ifp->if_mtu - hdrlen; 315 else if (inp && in_localaddr(inp->inp_faddr)) 316 size = ifp->if_mtu - hdrlen; 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 - hdrlen; 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 -= hdrlen; 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) 353 if (ipsec_used && 354 !IPSEC_PCB_SKIP_IPSEC(inp->inp_sp, IPSEC_DIR_OUTBOUND)) 355 optlen += ipsec4_hdrsiz_tcp(tp); 356 #endif 357 optlen += ip_optlen(inp); 358 } 359 #endif 360 #ifdef INET6 361 #ifdef INET 362 if (in6p && tp->t_family == AF_INET) { 363 #if defined(IPSEC) 364 if (ipsec_used && 365 !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 #if defined(IPSEC) 373 if (ipsec_used && 374 !IPSEC_PCB_SKIP_IPSEC(in6p->in6p_sp, IPSEC_DIR_OUTBOUND)) 375 optlen += ipsec6_hdrsiz_tcp(tp); 376 #endif 377 optlen += ip6_optlen(in6p); 378 } 379 #endif 380 size -= optlen; 381 382 /* there may not be any room for data if mtu is too small */ 383 if (size < 0) 384 return (EMSGSIZE); 385 386 /* 387 * *rxsegsizep holds *estimated* inbound segment size (estimation 388 * assumes that path MTU is the same for both ways). this is only 389 * for silly window avoidance, do not use the value for other purposes. 390 * 391 * ipseclen is subtracted from both sides, this may not be right. 392 * I'm not quite sure about this (could someone comment). 393 */ 394 *txsegsizep = min(tp->t_peermss - optlen, size); 395 /* 396 * Never send more than half a buffer full. This insures that we can 397 * always keep 2 packets on the wire, no matter what SO_SNDBUF is, and 398 * therefore acks will never be delayed unless we run out of data to 399 * transmit. 400 */ 401 if (so) 402 *txsegsizep = min(so->so_snd.sb_hiwat >> 1, *txsegsizep); 403 *rxsegsizep = min(tp->t_ourmss - optlen, size); 404 405 if (*txsegsizep != tp->t_segsz) { 406 /* 407 * If the new segment size is larger, we don't want to 408 * mess up the congestion window, but if it is smaller 409 * we'll have to reduce the congestion window to ensure 410 * that we don't get into trouble with initial windows 411 * and the rest. In any case, if the segment size 412 * has changed, chances are the path has, too, and 413 * our congestion window will be different. 414 */ 415 if (*txsegsizep < tp->t_segsz) { 416 tp->snd_cwnd = max((tp->snd_cwnd / tp->t_segsz) 417 * *txsegsizep, *txsegsizep); 418 tp->snd_ssthresh = max((tp->snd_ssthresh / tp->t_segsz) 419 * *txsegsizep, *txsegsizep); 420 } 421 tp->t_segsz = *txsegsizep; 422 } 423 424 return (0); 425 } 426 427 static 428 #ifndef GPROF 429 inline 430 #endif 431 int 432 tcp_build_datapkt(struct tcpcb *tp, struct socket *so, int off, 433 long len, int hdrlen, struct mbuf **mp) 434 { 435 struct mbuf *m, *m0; 436 uint64_t *tcps; 437 438 tcps = TCP_STAT_GETREF(); 439 if (tp->t_force && len == 1) 440 tcps[TCP_STAT_SNDPROBE]++; 441 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { 442 tp->t_sndrexmitpack++; 443 tcps[TCP_STAT_SNDREXMITPACK]++; 444 tcps[TCP_STAT_SNDREXMITBYTE] += len; 445 } else { 446 tcps[TCP_STAT_SNDPACK]++; 447 tcps[TCP_STAT_SNDBYTE] += len; 448 } 449 TCP_STAT_PUTREF(); 450 #ifdef notyet 451 if ((m = m_copypack(so->so_snd.sb_mb, off, 452 (int)len, max_linkhdr + hdrlen)) == 0) 453 return (ENOBUFS); 454 /* 455 * m_copypack left space for our hdr; use it. 456 */ 457 m->m_len += hdrlen; 458 m->m_data -= hdrlen; 459 #else 460 MGETHDR(m, M_DONTWAIT, MT_HEADER); 461 if (__predict_false(m == NULL)) 462 return (ENOBUFS); 463 MCLAIM(m, &tcp_tx_mowner); 464 465 /* 466 * XXX Because other code assumes headers will fit in 467 * XXX one header mbuf. 468 * 469 * (This code should almost *never* be run.) 470 */ 471 if (__predict_false((max_linkhdr + hdrlen) > MHLEN)) { 472 TCP_OUTPUT_COUNTER_INCR(&tcp_output_bigheader); 473 MCLGET(m, M_DONTWAIT); 474 if ((m->m_flags & M_EXT) == 0) { 475 m_freem(m); 476 return (ENOBUFS); 477 } 478 } 479 480 m->m_data += max_linkhdr; 481 m->m_len = hdrlen; 482 483 /* 484 * To avoid traversing the whole sb_mb chain for correct 485 * data to send, remember last sent mbuf, its offset and 486 * the sent size. When called the next time, see if the 487 * data to send is directly following the previous transfer. 488 * This is important for large TCP windows. 489 */ 490 if (off == 0 || tp->t_lastm == NULL || 491 (tp->t_lastoff + tp->t_lastlen) != off) { 492 TCP_OUTPUT_COUNTER_INCR(&tcp_output_predict_miss); 493 /* 494 * Either a new packet or a retransmit. 495 * Start from the beginning. 496 */ 497 tp->t_lastm = so->so_snd.sb_mb; 498 tp->t_inoff = off; 499 } else { 500 TCP_OUTPUT_COUNTER_INCR(&tcp_output_predict_hit); 501 tp->t_inoff += tp->t_lastlen; 502 } 503 504 /* Traverse forward to next packet */ 505 while (tp->t_inoff > 0) { 506 if (tp->t_lastm == NULL) 507 panic("tp->t_lastm == NULL"); 508 if (tp->t_inoff < tp->t_lastm->m_len) 509 break; 510 tp->t_inoff -= tp->t_lastm->m_len; 511 tp->t_lastm = tp->t_lastm->m_next; 512 } 513 514 tp->t_lastoff = off; 515 tp->t_lastlen = len; 516 m0 = tp->t_lastm; 517 off = tp->t_inoff; 518 519 if (len <= M_TRAILINGSPACE(m)) { 520 m_copydata(m0, off, (int) len, mtod(m, char *) + hdrlen); 521 m->m_len += len; 522 TCP_OUTPUT_COUNTER_INCR(&tcp_output_copysmall); 523 } else { 524 m->m_next = m_copym(m0, off, (int) len, M_DONTWAIT); 525 if (m->m_next == NULL) { 526 m_freem(m); 527 return (ENOBUFS); 528 } 529 #ifdef TCP_OUTPUT_COUNTERS 530 if (m->m_next->m_flags & M_EXT) 531 TCP_OUTPUT_COUNTER_INCR(&tcp_output_refbig); 532 else 533 TCP_OUTPUT_COUNTER_INCR(&tcp_output_copybig); 534 #endif /* TCP_OUTPUT_COUNTERS */ 535 } 536 #endif 537 538 *mp = m; 539 return (0); 540 } 541 542 /* 543 * Tcp output routine: figure out what should be sent and send it. 544 */ 545 int 546 tcp_output(struct tcpcb *tp) 547 { 548 struct rtentry *rt; 549 struct socket *so; 550 struct route *ro; 551 long len, win; 552 int off, flags, error; 553 struct mbuf *m; 554 struct ip *ip; 555 #ifdef INET6 556 struct ip6_hdr *ip6; 557 #endif 558 struct tcphdr *th; 559 u_char opt[MAX_TCPOPTLEN]; 560 #define OPT_FITS(more) ((optlen + (more)) < sizeof(opt)) 561 unsigned optlen, hdrlen, packetlen; 562 unsigned int sack_numblks; 563 int idle, sendalot, txsegsize, rxsegsize; 564 int txsegsize_nosack; 565 int maxburst = TCP_MAXBURST; 566 int af; /* address family on the wire */ 567 int iphdrlen; 568 int has_tso4, has_tso6; 569 int has_tso, use_tso; 570 bool alwaysfrag; 571 int sack_rxmit; 572 int sack_bytes_rxmt; 573 int ecn_tos; 574 struct sackhole *p; 575 #ifdef TCP_SIGNATURE 576 int sigoff = 0; 577 #endif 578 uint64_t *tcps; 579 580 #ifdef DIAGNOSTIC 581 if (tp->t_inpcb && tp->t_in6pcb) 582 panic("tcp_output: both t_inpcb and t_in6pcb are set"); 583 #endif 584 so = NULL; 585 ro = NULL; 586 if (tp->t_inpcb) { 587 so = tp->t_inpcb->inp_socket; 588 ro = &tp->t_inpcb->inp_route; 589 } 590 #ifdef INET6 591 else if (tp->t_in6pcb) { 592 so = tp->t_in6pcb->in6p_socket; 593 ro = &tp->t_in6pcb->in6p_route; 594 } 595 #endif 596 597 switch (af = tp->t_family) { 598 #ifdef INET 599 case AF_INET: 600 if (tp->t_inpcb) 601 break; 602 #ifdef INET6 603 /* mapped addr case */ 604 if (tp->t_in6pcb) 605 break; 606 #endif 607 return (EINVAL); 608 #endif 609 #ifdef INET6 610 case AF_INET6: 611 if (tp->t_in6pcb) 612 break; 613 return (EINVAL); 614 #endif 615 default: 616 return (EAFNOSUPPORT); 617 } 618 619 if (tcp_segsize(tp, &txsegsize, &rxsegsize, &alwaysfrag)) 620 return (EMSGSIZE); 621 622 idle = (tp->snd_max == tp->snd_una); 623 624 /* 625 * Determine if we can use TCP segmentation offload: 626 * - If we're using IPv4 627 * - If there is not an IPsec policy that prevents it 628 * - If the interface can do it 629 */ 630 has_tso4 = has_tso6 = false; 631 #if defined(INET) 632 has_tso4 = tp->t_inpcb != NULL && 633 #if defined(IPSEC) 634 (!ipsec_used || IPSEC_PCB_SKIP_IPSEC(tp->t_inpcb->inp_sp, 635 IPSEC_DIR_OUTBOUND)) && 636 #endif 637 (rt = rtcache_validate(&tp->t_inpcb->inp_route)) != NULL && 638 (rt->rt_ifp->if_capenable & IFCAP_TSOv4) != 0; 639 #endif /* defined(INET) */ 640 #if defined(INET6) 641 has_tso6 = tp->t_in6pcb != NULL && 642 #if defined(IPSEC) 643 (!ipsec_used || IPSEC_PCB_SKIP_IPSEC(tp->t_in6pcb->in6p_sp, 644 IPSEC_DIR_OUTBOUND)) && 645 #endif 646 (rt = rtcache_validate(&tp->t_in6pcb->in6p_route)) != NULL && 647 (rt->rt_ifp->if_capenable & IFCAP_TSOv6) != 0; 648 #endif /* defined(INET6) */ 649 has_tso = (has_tso4 || has_tso6) && !alwaysfrag; 650 651 /* 652 * Restart Window computation. From draft-floyd-incr-init-win-03: 653 * 654 * Optionally, a TCP MAY set the restart window to the 655 * minimum of the value used for the initial window and 656 * the current value of cwnd (in other words, using a 657 * larger value for the restart window should never increase 658 * the size of cwnd). 659 */ 660 if (tcp_cwm) { 661 /* 662 * Hughes/Touch/Heidemann Congestion Window Monitoring. 663 * Count the number of packets currently pending 664 * acknowledgement, and limit our congestion window 665 * to a pre-determined allowed burst size plus that count. 666 * This prevents bursting once all pending packets have 667 * been acknowledged (i.e. transmission is idle). 668 * 669 * XXX Link this to Initial Window? 670 */ 671 tp->snd_cwnd = min(tp->snd_cwnd, 672 (tcp_cwm_burstsize * txsegsize) + 673 (tp->snd_nxt - tp->snd_una)); 674 } else { 675 if (idle && (tcp_now - tp->t_rcvtime) >= tp->t_rxtcur) { 676 /* 677 * We have been idle for "a while" and no acks are 678 * expected to clock out any data we send -- 679 * slow start to get ack "clock" running again. 680 */ 681 int ss = tcp_init_win; 682 #ifdef INET 683 if (tp->t_inpcb && 684 in_localaddr(tp->t_inpcb->inp_faddr)) 685 ss = tcp_init_win_local; 686 #endif 687 #ifdef INET6 688 if (tp->t_in6pcb && 689 in6_localaddr(&tp->t_in6pcb->in6p_faddr)) 690 ss = tcp_init_win_local; 691 #endif 692 tp->snd_cwnd = min(tp->snd_cwnd, 693 TCP_INITIAL_WINDOW(ss, txsegsize)); 694 } 695 } 696 697 txsegsize_nosack = txsegsize; 698 again: 699 ecn_tos = 0; 700 use_tso = has_tso; 701 if ((tp->t_flags & (TF_ECN_SND_CWR|TF_ECN_SND_ECE)) != 0) { 702 /* don't duplicate CWR/ECE. */ 703 use_tso = 0; 704 } 705 TCP_REASS_LOCK(tp); 706 sack_numblks = tcp_sack_numblks(tp); 707 if (sack_numblks) { 708 int sackoptlen; 709 710 sackoptlen = TCP_SACK_OPTLEN(sack_numblks); 711 if (sackoptlen > txsegsize_nosack) { 712 sack_numblks = 0; /* give up SACK */ 713 txsegsize = txsegsize_nosack; 714 } else { 715 if ((tp->rcv_sack_flags & TCPSACK_HAVED) != 0) { 716 /* don't duplicate D-SACK. */ 717 use_tso = 0; 718 } 719 txsegsize = txsegsize_nosack - sackoptlen; 720 } 721 } else { 722 txsegsize = txsegsize_nosack; 723 } 724 725 /* 726 * Determine length of data that should be transmitted, and 727 * flags that should be used. If there is some data or critical 728 * controls (SYN, RST) to send, then transmit; otherwise, 729 * investigate further. 730 * 731 * Readjust SACK information to avoid resending duplicate data. 732 */ 733 if (TCP_SACK_ENABLED(tp) && SEQ_LT(tp->snd_nxt, tp->snd_max)) 734 tcp_sack_adjust(tp); 735 sendalot = 0; 736 off = tp->snd_nxt - tp->snd_una; 737 win = min(tp->snd_wnd, tp->snd_cwnd); 738 739 flags = tcp_outflags[tp->t_state]; 740 741 /* 742 * Send any SACK-generated retransmissions. If we're explicitly trying 743 * to send out new data (when sendalot is 1), bypass this function. 744 * If we retransmit in fast recovery mode, decrement snd_cwnd, since 745 * we're replacing a (future) new transmission with a retransmission 746 * now, and we previously incremented snd_cwnd in tcp_input(). 747 */ 748 /* 749 * Still in sack recovery , reset rxmit flag to zero. 750 */ 751 sack_rxmit = 0; 752 sack_bytes_rxmt = 0; 753 len = 0; 754 p = NULL; 755 do { 756 long cwin; 757 if (!TCP_SACK_ENABLED(tp)) 758 break; 759 if (tp->t_partialacks < 0) 760 break; 761 p = tcp_sack_output(tp, &sack_bytes_rxmt); 762 if (p == NULL) 763 break; 764 765 cwin = min(tp->snd_wnd, tp->snd_cwnd) - sack_bytes_rxmt; 766 if (cwin < 0) 767 cwin = 0; 768 /* Do not retransmit SACK segments beyond snd_recover */ 769 if (SEQ_GT(p->end, tp->snd_recover)) { 770 /* 771 * (At least) part of sack hole extends beyond 772 * snd_recover. Check to see if we can rexmit data 773 * for this hole. 774 */ 775 if (SEQ_GEQ(p->rxmit, tp->snd_recover)) { 776 /* 777 * Can't rexmit any more data for this hole. 778 * That data will be rexmitted in the next 779 * sack recovery episode, when snd_recover 780 * moves past p->rxmit. 781 */ 782 p = NULL; 783 break; 784 } 785 /* Can rexmit part of the current hole */ 786 len = ((long)ulmin(cwin, tp->snd_recover - p->rxmit)); 787 } else 788 len = ((long)ulmin(cwin, p->end - p->rxmit)); 789 off = p->rxmit - tp->snd_una; 790 if (off + len > so->so_snd.sb_cc) { 791 /* 1 for TH_FIN */ 792 KASSERT(off + len == so->so_snd.sb_cc + 1); 793 KASSERT(p->rxmit + len == tp->snd_max); 794 len = so->so_snd.sb_cc - off; 795 } 796 if (len > 0) { 797 sack_rxmit = 1; 798 sendalot = 1; 799 } 800 } while (/*CONSTCOND*/0); 801 802 /* 803 * If in persist timeout with window of 0, send 1 byte. 804 * Otherwise, if window is small but nonzero 805 * and timer expired, we will send what we can 806 * and go to transmit state. 807 */ 808 if (tp->t_force) { 809 if (win == 0) { 810 /* 811 * If we still have some data to send, then 812 * clear the FIN bit. Usually this would 813 * happen below when it realizes that we 814 * aren't sending all the data. However, 815 * if we have exactly 1 byte of unset data, 816 * then it won't clear the FIN bit below, 817 * and if we are in persist state, we wind 818 * up sending the packet without recording 819 * that we sent the FIN bit. 820 * 821 * We can't just blindly clear the FIN bit, 822 * because if we don't have any more data 823 * to send then the probe will be the FIN 824 * itself. 825 */ 826 if (off < so->so_snd.sb_cc) 827 flags &= ~TH_FIN; 828 win = 1; 829 } else { 830 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 831 tp->t_rxtshift = 0; 832 } 833 } 834 835 if (sack_rxmit == 0) { 836 if (TCP_SACK_ENABLED(tp) && tp->t_partialacks >= 0) { 837 long cwin; 838 839 /* 840 * We are inside of a SACK recovery episode and are 841 * sending new data, having retransmitted all the 842 * data possible in the scoreboard. 843 */ 844 if (tp->snd_wnd < so->so_snd.sb_cc) { 845 len = tp->snd_wnd - off; 846 flags &= ~TH_FIN; 847 } else { 848 len = so->so_snd.sb_cc - off; 849 } 850 851 /* 852 * From FreeBSD: 853 * Don't remove this (len > 0) check ! 854 * We explicitly check for len > 0 here (although it 855 * isn't really necessary), to work around a gcc 856 * optimization issue - to force gcc to compute 857 * len above. Without this check, the computation 858 * of len is bungled by the optimizer. 859 */ 860 if (len > 0) { 861 cwin = tp->snd_cwnd - 862 (tp->snd_nxt - tp->sack_newdata) - 863 sack_bytes_rxmt; 864 if (cwin < 0) 865 cwin = 0; 866 if (cwin < len) { 867 len = cwin; 868 flags &= ~TH_FIN; 869 } 870 } 871 } else if (win < so->so_snd.sb_cc) { 872 len = win - off; 873 flags &= ~TH_FIN; 874 } else { 875 len = so->so_snd.sb_cc - off; 876 } 877 } 878 879 if (len < 0) { 880 /* 881 * If FIN has been sent but not acked, 882 * but we haven't been called to retransmit, 883 * len will be -1. Otherwise, window shrank 884 * after we sent into it. If window shrank to 0, 885 * cancel pending retransmit, pull snd_nxt back 886 * to (closed) window, and set the persist timer 887 * if it isn't already going. If the window didn't 888 * close completely, just wait for an ACK. 889 * 890 * If we have a pending FIN, either it has already been 891 * transmitted or it is outside the window, so drop it. 892 * If the FIN has been transmitted, but this is not a 893 * retransmission, then len must be -1. Therefore we also 894 * prevent here the sending of `gratuitous FINs'. This 895 * eliminates the need to check for that case below (e.g. 896 * to back up snd_nxt before the FIN so that the sequence 897 * number is correct). 898 */ 899 len = 0; 900 flags &= ~TH_FIN; 901 if (win == 0) { 902 TCP_TIMER_DISARM(tp, TCPT_REXMT); 903 tp->t_rxtshift = 0; 904 tp->snd_nxt = tp->snd_una; 905 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 906 tcp_setpersist(tp); 907 } 908 } 909 910 /* 911 * Automatic sizing enables the performance of large buffers 912 * and most of the efficiency of small ones by only allocating 913 * space when it is needed. 914 * 915 * The criteria to step up the send buffer one notch are: 916 * 1. receive window of remote host is larger than send buffer 917 * (with a fudge factor of 5/4th); 918 * 2. send buffer is filled to 7/8th with data (so we actually 919 * have data to make use of it); 920 * 3. send buffer fill has not hit maximal automatic size; 921 * 4. our send window (slow start and cogestion controlled) is 922 * larger than sent but unacknowledged data in send buffer. 923 * 924 * The remote host receive window scaling factor may limit the 925 * growing of the send buffer before it reaches its allowed 926 * maximum. 927 * 928 * It scales directly with slow start or congestion window 929 * and does at most one step per received ACK. This fast 930 * scaling has the drawback of growing the send buffer beyond 931 * what is strictly necessary to make full use of a given 932 * delay*bandwith product. However testing has shown this not 933 * to be much of an problem. At worst we are trading wasting 934 * of available bandwith (the non-use of it) for wasting some 935 * socket buffer memory. 936 * 937 * TODO: Shrink send buffer during idle periods together 938 * with congestion window. Requires another timer. 939 */ 940 if (tcp_do_autosndbuf && so->so_snd.sb_flags & SB_AUTOSIZE) { 941 if ((tp->snd_wnd / 4 * 5) >= so->so_snd.sb_hiwat && 942 so->so_snd.sb_cc >= (so->so_snd.sb_hiwat / 8 * 7) && 943 so->so_snd.sb_cc < tcp_autosndbuf_max && 944 win >= (so->so_snd.sb_cc - (tp->snd_nxt - tp->snd_una))) { 945 if (!sbreserve(&so->so_snd, 946 min(so->so_snd.sb_hiwat + tcp_autosndbuf_inc, 947 tcp_autosndbuf_max), so)) 948 so->so_snd.sb_flags &= ~SB_AUTOSIZE; 949 } 950 } 951 952 if (len > txsegsize) { 953 if (use_tso) { 954 /* 955 * Truncate TSO transfers to IP_MAXPACKET, and make 956 * sure that we send equal size transfers down the 957 * stack (rather than big-small-big-small-...). 958 */ 959 #ifdef INET6 960 CTASSERT(IPV6_MAXPACKET == IP_MAXPACKET); 961 #endif 962 len = (min(len, IP_MAXPACKET) / txsegsize) * txsegsize; 963 if (len <= txsegsize) { 964 use_tso = 0; 965 } 966 } else 967 len = txsegsize; 968 flags &= ~TH_FIN; 969 sendalot = 1; 970 } else 971 use_tso = 0; 972 if (sack_rxmit) { 973 if (SEQ_LT(p->rxmit + len, tp->snd_una + so->so_snd.sb_cc)) 974 flags &= ~TH_FIN; 975 } 976 977 win = sbspace(&so->so_rcv); 978 979 /* 980 * Sender silly window avoidance. If connection is idle 981 * and can send all data, a maximum segment, 982 * at least a maximum default-size segment do it, 983 * or are forced, do it; otherwise don't bother. 984 * If peer's buffer is tiny, then send 985 * when window is at least half open. 986 * If retransmitting (possibly after persist timer forced us 987 * to send into a small window), then must resend. 988 */ 989 if (len) { 990 if (len >= txsegsize) 991 goto send; 992 if ((so->so_state & SS_MORETOCOME) == 0 && 993 ((idle || tp->t_flags & TF_NODELAY) && 994 len + off >= so->so_snd.sb_cc)) 995 goto send; 996 if (tp->t_force) 997 goto send; 998 if (len >= tp->max_sndwnd / 2) 999 goto send; 1000 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) 1001 goto send; 1002 if (sack_rxmit) 1003 goto send; 1004 } 1005 1006 /* 1007 * Compare available window to amount of window known to peer 1008 * (as advertised window less next expected input). If the 1009 * difference is at least twice the size of the largest segment 1010 * we expect to receive (i.e. two segments) or at least 50% of 1011 * the maximum possible window, then want to send a window update 1012 * to peer. 1013 */ 1014 if (win > 0) { 1015 /* 1016 * "adv" is the amount we can increase the window, 1017 * taking into account that we are limited by 1018 * TCP_MAXWIN << tp->rcv_scale. 1019 */ 1020 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) - 1021 (tp->rcv_adv - tp->rcv_nxt); 1022 1023 if (adv >= (long) (2 * rxsegsize)) 1024 goto send; 1025 if (2 * adv >= (long) so->so_rcv.sb_hiwat) 1026 goto send; 1027 } 1028 1029 /* 1030 * Send if we owe peer an ACK. 1031 */ 1032 if (tp->t_flags & TF_ACKNOW) 1033 goto send; 1034 if (flags & (TH_SYN|TH_FIN|TH_RST)) 1035 goto send; 1036 if (SEQ_GT(tp->snd_up, tp->snd_una)) 1037 goto send; 1038 /* 1039 * In SACK, it is possible for tcp_output to fail to send a segment 1040 * after the retransmission timer has been turned off. Make sure 1041 * that the retransmission timer is set. 1042 */ 1043 if (TCP_SACK_ENABLED(tp) && SEQ_GT(tp->snd_max, tp->snd_una) && 1044 !TCP_TIMER_ISARMED(tp, TCPT_REXMT) && 1045 !TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) { 1046 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1047 goto just_return; 1048 } 1049 1050 /* 1051 * TCP window updates are not reliable, rather a polling protocol 1052 * using ``persist'' packets is used to insure receipt of window 1053 * updates. The three ``states'' for the output side are: 1054 * idle not doing retransmits or persists 1055 * persisting to move a small or zero window 1056 * (re)transmitting and thereby not persisting 1057 * 1058 * tp->t_timer[TCPT_PERSIST] 1059 * is set when we are in persist state. 1060 * tp->t_force 1061 * is set when we are called to send a persist packet. 1062 * tp->t_timer[TCPT_REXMT] 1063 * is set when we are retransmitting 1064 * The output side is idle when both timers are zero. 1065 * 1066 * If send window is too small, there is data to transmit, and no 1067 * retransmit or persist is pending, then go to persist state. 1068 * If nothing happens soon, send when timer expires: 1069 * if window is nonzero, transmit what we can, 1070 * otherwise force out a byte. 1071 */ 1072 if (so->so_snd.sb_cc && TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 1073 TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 1074 tp->t_rxtshift = 0; 1075 tcp_setpersist(tp); 1076 } 1077 1078 /* 1079 * No reason to send a segment, just return. 1080 */ 1081 just_return: 1082 TCP_REASS_UNLOCK(tp); 1083 return (0); 1084 1085 send: 1086 /* 1087 * Before ESTABLISHED, force sending of initial options 1088 * unless TCP set not to do any options. 1089 * NOTE: we assume that the IP/TCP header plus TCP options 1090 * always fit in a single mbuf, leaving room for a maximum 1091 * link header, i.e. 1092 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES 1093 */ 1094 optlen = 0; 1095 switch (af) { 1096 #ifdef INET 1097 case AF_INET: 1098 iphdrlen = sizeof(struct ip) + sizeof(struct tcphdr); 1099 break; 1100 #endif 1101 #ifdef INET6 1102 case AF_INET6: 1103 iphdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1104 break; 1105 #endif 1106 default: /*pacify gcc*/ 1107 iphdrlen = 0; 1108 break; 1109 } 1110 hdrlen = iphdrlen; 1111 if (flags & TH_SYN) { 1112 struct rtentry *synrt; 1113 1114 synrt = NULL; 1115 #ifdef INET 1116 if (tp->t_inpcb) 1117 synrt = in_pcbrtentry(tp->t_inpcb); 1118 #endif 1119 #ifdef INET6 1120 if (tp->t_in6pcb) 1121 synrt = in6_pcbrtentry(tp->t_in6pcb); 1122 #endif 1123 1124 tp->snd_nxt = tp->iss; 1125 tp->t_ourmss = tcp_mss_to_advertise(synrt != NULL ? 1126 synrt->rt_ifp : NULL, af); 1127 if ((tp->t_flags & TF_NOOPT) == 0 && OPT_FITS(4)) { 1128 opt[0] = TCPOPT_MAXSEG; 1129 opt[1] = 4; 1130 opt[2] = (tp->t_ourmss >> 8) & 0xff; 1131 opt[3] = tp->t_ourmss & 0xff; 1132 optlen = 4; 1133 1134 if ((tp->t_flags & TF_REQ_SCALE) && 1135 ((flags & TH_ACK) == 0 || 1136 (tp->t_flags & TF_RCVD_SCALE)) && 1137 OPT_FITS(4)) { 1138 *((u_int32_t *) (opt + optlen)) = htonl( 1139 TCPOPT_NOP << 24 | 1140 TCPOPT_WINDOW << 16 | 1141 TCPOLEN_WINDOW << 8 | 1142 tp->request_r_scale); 1143 optlen += 4; 1144 } 1145 if (tcp_do_sack && OPT_FITS(4)) { 1146 u_int8_t *cp = (u_int8_t *)(opt + optlen); 1147 1148 cp[0] = TCPOPT_SACK_PERMITTED; 1149 cp[1] = 2; 1150 cp[2] = TCPOPT_NOP; 1151 cp[3] = TCPOPT_NOP; 1152 optlen += 4; 1153 } 1154 } 1155 } 1156 1157 /* 1158 * Send a timestamp and echo-reply if this is a SYN and our side 1159 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side 1160 * and our peer have sent timestamps in our SYN's. 1161 */ 1162 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1163 (flags & TH_RST) == 0 && 1164 ((flags & (TH_SYN|TH_ACK)) == TH_SYN || 1165 (tp->t_flags & TF_RCVD_TSTMP)) && OPT_FITS(TCPOLEN_TSTAMP_APPA)) { 1166 u_int32_t *lp = (u_int32_t *)(opt + optlen); 1167 1168 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1169 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1170 *lp++ = htonl(TCP_TIMESTAMP(tp)); 1171 *lp = htonl(tp->ts_recent); 1172 optlen += TCPOLEN_TSTAMP_APPA; 1173 1174 /* Set receive buffer autosizing timestamp. */ 1175 if (tp->rfbuf_ts == 0 && (so->so_rcv.sb_flags & SB_AUTOSIZE)) 1176 tp->rfbuf_ts = TCP_TIMESTAMP(tp); 1177 } 1178 1179 /* 1180 * Tack on the SACK block if it is necessary. 1181 */ 1182 if (sack_numblks) { 1183 int sack_len; 1184 u_char *bp = (u_char *)(opt + optlen); 1185 u_int32_t *lp = (u_int32_t *)(bp + 4); 1186 struct ipqent *tiqe; 1187 1188 sack_len = sack_numblks * 8 + 2; 1189 if (OPT_FITS(sack_len + 2)) { 1190 bp[0] = TCPOPT_NOP; 1191 bp[1] = TCPOPT_NOP; 1192 bp[2] = TCPOPT_SACK; 1193 bp[3] = sack_len; 1194 if ((tp->rcv_sack_flags & TCPSACK_HAVED) != 0) { 1195 sack_numblks--; 1196 *lp++ = htonl(tp->rcv_dsack_block.left); 1197 *lp++ = htonl(tp->rcv_dsack_block.right); 1198 tp->rcv_sack_flags &= ~TCPSACK_HAVED; 1199 } 1200 for (tiqe = TAILQ_FIRST(&tp->timeq); 1201 sack_numblks > 0; 1202 tiqe = TAILQ_NEXT(tiqe, ipqe_timeq)) { 1203 KASSERT(tiqe != NULL); 1204 sack_numblks--; 1205 *lp++ = htonl(tiqe->ipqe_seq); 1206 *lp++ = htonl(tiqe->ipqe_seq + tiqe->ipqe_len + 1207 ((tiqe->ipqe_flags & TH_FIN) != 0 ? 1 : 0)); 1208 } 1209 optlen += sack_len + 2; 1210 } 1211 } 1212 TCP_REASS_UNLOCK(tp); 1213 1214 #ifdef TCP_SIGNATURE 1215 if ((tp->t_flags & TF_SIGNATURE) && OPT_FITS(TCPOLEN_SIGNATURE + 2)) { 1216 u_char *bp; 1217 /* 1218 * Initialize TCP-MD5 option (RFC2385) 1219 */ 1220 bp = (u_char *)opt + optlen; 1221 *bp++ = TCPOPT_SIGNATURE; 1222 *bp++ = TCPOLEN_SIGNATURE; 1223 sigoff = optlen + 2; 1224 memset(bp, 0, TCP_SIGLEN); 1225 bp += TCP_SIGLEN; 1226 optlen += TCPOLEN_SIGNATURE; 1227 /* 1228 * Terminate options list and maintain 32-bit alignment. 1229 */ 1230 *bp++ = TCPOPT_NOP; 1231 *bp++ = TCPOPT_EOL; 1232 optlen += 2; 1233 } 1234 #endif /* TCP_SIGNATURE */ 1235 1236 hdrlen += optlen; 1237 1238 #ifdef DIAGNOSTIC 1239 if (!use_tso && len > txsegsize) 1240 panic("tcp data to be sent is larger than segment"); 1241 else if (use_tso && len > IP_MAXPACKET) 1242 panic("tcp data to be sent is larger than max TSO size"); 1243 if (max_linkhdr + hdrlen > MCLBYTES) 1244 panic("tcphdr too big"); 1245 #endif 1246 1247 /* 1248 * Grab a header mbuf, attaching a copy of data to 1249 * be transmitted, and initialize the header from 1250 * the template for sends on this connection. 1251 */ 1252 if (len) { 1253 error = tcp_build_datapkt(tp, so, off, len, hdrlen, &m); 1254 if (error) 1255 goto out; 1256 /* 1257 * If we're sending everything we've got, set PUSH. 1258 * (This will keep happy those implementations which only 1259 * give data to the user when a buffer fills or 1260 * a PUSH comes in.) 1261 */ 1262 if (off + len == so->so_snd.sb_cc) 1263 flags |= TH_PUSH; 1264 } else { 1265 tcps = TCP_STAT_GETREF(); 1266 if (tp->t_flags & TF_ACKNOW) 1267 tcps[TCP_STAT_SNDACKS]++; 1268 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 1269 tcps[TCP_STAT_SNDCTRL]++; 1270 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 1271 tcps[TCP_STAT_SNDURG]++; 1272 else 1273 tcps[TCP_STAT_SNDWINUP]++; 1274 TCP_STAT_PUTREF(); 1275 1276 MGETHDR(m, M_DONTWAIT, MT_HEADER); 1277 if (m != NULL && max_linkhdr + hdrlen > MHLEN) { 1278 MCLGET(m, M_DONTWAIT); 1279 if ((m->m_flags & M_EXT) == 0) { 1280 m_freem(m); 1281 m = NULL; 1282 } 1283 } 1284 if (m == NULL) { 1285 error = ENOBUFS; 1286 goto out; 1287 } 1288 MCLAIM(m, &tcp_tx_mowner); 1289 m->m_data += max_linkhdr; 1290 m->m_len = hdrlen; 1291 } 1292 m->m_pkthdr.rcvif = NULL; 1293 switch (af) { 1294 #ifdef INET 1295 case AF_INET: 1296 ip = mtod(m, struct ip *); 1297 #ifdef INET6 1298 ip6 = NULL; 1299 #endif 1300 th = (struct tcphdr *)(ip + 1); 1301 break; 1302 #endif 1303 #ifdef INET6 1304 case AF_INET6: 1305 ip = NULL; 1306 ip6 = mtod(m, struct ip6_hdr *); 1307 th = (struct tcphdr *)(ip6 + 1); 1308 break; 1309 #endif 1310 default: /*pacify gcc*/ 1311 ip = NULL; 1312 #ifdef INET6 1313 ip6 = NULL; 1314 #endif 1315 th = NULL; 1316 break; 1317 } 1318 if (tp->t_template == 0) 1319 panic("tcp_output"); 1320 if (tp->t_template->m_len < iphdrlen) 1321 panic("tcp_output"); 1322 bcopy(mtod(tp->t_template, void *), mtod(m, void *), iphdrlen); 1323 1324 /* 1325 * If we are starting a connection, send ECN setup 1326 * SYN packet. If we are on a retransmit, we may 1327 * resend those bits a number of times as per 1328 * RFC 3168. 1329 */ 1330 if (tp->t_state == TCPS_SYN_SENT && tcp_do_ecn) { 1331 if (tp->t_flags & TF_SYN_REXMT) { 1332 if (tp->t_ecn_retries--) 1333 flags |= TH_ECE|TH_CWR; 1334 } else { 1335 flags |= TH_ECE|TH_CWR; 1336 tp->t_ecn_retries = tcp_ecn_maxretries; 1337 } 1338 } 1339 1340 if (TCP_ECN_ALLOWED(tp)) { 1341 /* 1342 * If the peer has ECN, mark data packets 1343 * ECN capable. Ignore pure ack packets, retransmissions 1344 * and window probes. 1345 */ 1346 if (len > 0 && SEQ_GEQ(tp->snd_nxt, tp->snd_max) && 1347 !(tp->t_force && len == 1)) { 1348 ecn_tos = IPTOS_ECN_ECT0; 1349 TCP_STATINC(TCP_STAT_ECN_ECT); 1350 } 1351 1352 /* 1353 * Reply with proper ECN notifications. 1354 */ 1355 if (tp->t_flags & TF_ECN_SND_CWR) { 1356 flags |= TH_CWR; 1357 tp->t_flags &= ~TF_ECN_SND_CWR; 1358 } 1359 if (tp->t_flags & TF_ECN_SND_ECE) { 1360 flags |= TH_ECE; 1361 } 1362 } 1363 1364 1365 /* 1366 * If we are doing retransmissions, then snd_nxt will 1367 * not reflect the first unsent octet. For ACK only 1368 * packets, we do not want the sequence number of the 1369 * retransmitted packet, we want the sequence number 1370 * of the next unsent octet. So, if there is no data 1371 * (and no SYN or FIN), use snd_max instead of snd_nxt 1372 * when filling in ti_seq. But if we are in persist 1373 * state, snd_max might reflect one byte beyond the 1374 * right edge of the window, so use snd_nxt in that 1375 * case, since we know we aren't doing a retransmission. 1376 * (retransmit and persist are mutually exclusive...) 1377 */ 1378 if (TCP_SACK_ENABLED(tp) && sack_rxmit) { 1379 th->th_seq = htonl(p->rxmit); 1380 p->rxmit += len; 1381 } else { 1382 if (len || (flags & (TH_SYN|TH_FIN)) || 1383 TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) 1384 th->th_seq = htonl(tp->snd_nxt); 1385 else 1386 th->th_seq = htonl(tp->snd_max); 1387 } 1388 th->th_ack = htonl(tp->rcv_nxt); 1389 if (optlen) { 1390 bcopy((void *)opt, (void *)(th + 1), optlen); 1391 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1392 } 1393 th->th_flags = flags; 1394 /* 1395 * Calculate receive window. Don't shrink window, 1396 * but avoid silly window syndrome. 1397 */ 1398 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)rxsegsize) 1399 win = 0; 1400 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 1401 win = (long)TCP_MAXWIN << tp->rcv_scale; 1402 if (win < (long)(int32_t)(tp->rcv_adv - tp->rcv_nxt)) 1403 win = (long)(int32_t)(tp->rcv_adv - tp->rcv_nxt); 1404 th->th_win = htons((u_int16_t) (win>>tp->rcv_scale)); 1405 if (th->th_win == 0) { 1406 tp->t_sndzerowin++; 1407 } 1408 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 1409 u_int32_t urp = tp->snd_up - tp->snd_nxt; 1410 if (urp > IP_MAXPACKET) 1411 urp = IP_MAXPACKET; 1412 th->th_urp = htons((u_int16_t)urp); 1413 th->th_flags |= TH_URG; 1414 } else 1415 /* 1416 * If no urgent pointer to send, then we pull 1417 * the urgent pointer to the left edge of the send window 1418 * so that it doesn't drift into the send window on sequence 1419 * number wraparound. 1420 */ 1421 tp->snd_up = tp->snd_una; /* drag it along */ 1422 1423 #ifdef TCP_SIGNATURE 1424 if (sigoff && (tp->t_flags & TF_SIGNATURE)) { 1425 struct secasvar *sav; 1426 u_int8_t *sigp; 1427 1428 sav = tcp_signature_getsav(m, th); 1429 1430 if (sav == NULL) { 1431 if (m) 1432 m_freem(m); 1433 return (EPERM); 1434 } 1435 1436 m->m_pkthdr.len = hdrlen + len; 1437 sigp = (char *)th + sizeof(*th) + sigoff; 1438 tcp_signature(m, th, (char *)th - mtod(m, char *), sav, sigp); 1439 1440 key_sa_recordxfer(sav, m); 1441 KEY_FREESAV(&sav); 1442 } 1443 #endif 1444 1445 /* 1446 * Set ourselves up to be checksummed just before the packet 1447 * hits the wire. 1448 */ 1449 switch (af) { 1450 #ifdef INET 1451 case AF_INET: 1452 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1453 if (use_tso) { 1454 m->m_pkthdr.segsz = txsegsize; 1455 m->m_pkthdr.csum_flags = M_CSUM_TSOv4; 1456 } else { 1457 m->m_pkthdr.csum_flags = M_CSUM_TCPv4; 1458 if (len + optlen) { 1459 /* Fixup the pseudo-header checksum. */ 1460 /* XXXJRT Not IP Jumbogram safe. */ 1461 th->th_sum = in_cksum_addword(th->th_sum, 1462 htons((u_int16_t) (len + optlen))); 1463 } 1464 } 1465 break; 1466 #endif 1467 #ifdef INET6 1468 case AF_INET6: 1469 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1470 if (use_tso) { 1471 m->m_pkthdr.segsz = txsegsize; 1472 m->m_pkthdr.csum_flags = M_CSUM_TSOv6; 1473 } else { 1474 m->m_pkthdr.csum_flags = M_CSUM_TCPv6; 1475 if (len + optlen) { 1476 /* Fixup the pseudo-header checksum. */ 1477 /* XXXJRT: Not IPv6 Jumbogram safe. */ 1478 th->th_sum = in_cksum_addword(th->th_sum, 1479 htons((u_int16_t) (len + optlen))); 1480 } 1481 } 1482 break; 1483 #endif 1484 } 1485 1486 /* 1487 * In transmit state, time the transmission and arrange for 1488 * the retransmit. In persist state, just set snd_max. 1489 */ 1490 if (tp->t_force == 0 || TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) { 1491 tcp_seq startseq = tp->snd_nxt; 1492 1493 /* 1494 * Advance snd_nxt over sequence space of this segment. 1495 * There are no states in which we send both a SYN and a FIN, 1496 * so we collapse the tests for these flags. 1497 */ 1498 if (flags & (TH_SYN|TH_FIN)) 1499 tp->snd_nxt++; 1500 if (sack_rxmit) 1501 goto timer; 1502 tp->snd_nxt += len; 1503 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 1504 tp->snd_max = tp->snd_nxt; 1505 /* 1506 * Time this transmission if not a retransmission and 1507 * not currently timing anything. 1508 */ 1509 if (tp->t_rtttime == 0) { 1510 tp->t_rtttime = tcp_now; 1511 tp->t_rtseq = startseq; 1512 TCP_STATINC(TCP_STAT_SEGSTIMED); 1513 } 1514 } 1515 1516 /* 1517 * Set retransmit timer if not currently set, 1518 * and not doing an ack or a keep-alive probe. 1519 * Initial value for retransmit timer is smoothed 1520 * round-trip time + 2 * round-trip time variance. 1521 * Initialize shift counter which is used for backoff 1522 * of retransmit time. 1523 */ 1524 timer: 1525 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 && 1526 ((sack_rxmit && tp->snd_nxt != tp->snd_max) || 1527 tp->snd_nxt != tp->snd_una)) { 1528 if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST)) { 1529 TCP_TIMER_DISARM(tp, TCPT_PERSIST); 1530 tp->t_rxtshift = 0; 1531 } 1532 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1533 } 1534 } else 1535 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max)) 1536 tp->snd_max = tp->snd_nxt + len; 1537 1538 #ifdef TCP_DEBUG 1539 /* 1540 * Trace. 1541 */ 1542 if (so->so_options & SO_DEBUG) 1543 tcp_trace(TA_OUTPUT, tp->t_state, tp, m, 0); 1544 #endif 1545 1546 /* 1547 * Fill in IP length and desired time to live and 1548 * send to IP level. There should be a better way 1549 * to handle ttl and tos; we could keep them in 1550 * the template, but need a way to checksum without them. 1551 */ 1552 m->m_pkthdr.len = hdrlen + len; 1553 1554 switch (af) { 1555 #ifdef INET 1556 case AF_INET: 1557 ip->ip_len = htons(m->m_pkthdr.len); 1558 packetlen = m->m_pkthdr.len; 1559 if (tp->t_inpcb) { 1560 ip->ip_ttl = tp->t_inpcb->inp_ip.ip_ttl; 1561 ip->ip_tos = tp->t_inpcb->inp_ip.ip_tos | ecn_tos; 1562 } 1563 #ifdef INET6 1564 else if (tp->t_in6pcb) { 1565 ip->ip_ttl = in6_selecthlim(tp->t_in6pcb, NULL); /*XXX*/ 1566 ip->ip_tos = ecn_tos; /*XXX*/ 1567 } 1568 #endif 1569 break; 1570 #endif 1571 #ifdef INET6 1572 case AF_INET6: 1573 packetlen = m->m_pkthdr.len; 1574 ip6->ip6_nxt = IPPROTO_TCP; 1575 if (tp->t_in6pcb) { 1576 /* 1577 * we separately set hoplimit for every segment, since 1578 * the user might want to change the value via 1579 * setsockopt. Also, desired default hop limit might 1580 * be changed via Neighbor Discovery. 1581 */ 1582 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, 1583 (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp 1584 : NULL); 1585 } 1586 ip6->ip6_flow |= htonl(ecn_tos << 20); 1587 /* ip6->ip6_flow = ??? (from template) */ 1588 /* ip6_plen will be filled in ip6_output(). */ 1589 break; 1590 #endif 1591 default: /*pacify gcc*/ 1592 packetlen = 0; 1593 break; 1594 } 1595 1596 switch (af) { 1597 #ifdef INET 1598 case AF_INET: 1599 { 1600 struct mbuf *opts; 1601 1602 if (tp->t_inpcb) 1603 opts = tp->t_inpcb->inp_options; 1604 else 1605 opts = NULL; 1606 error = ip_output(m, opts, ro, 1607 (tp->t_mtudisc ? IP_MTUDISC : 0) | 1608 (so->so_options & SO_DONTROUTE), NULL, so); 1609 break; 1610 } 1611 #endif 1612 #ifdef INET6 1613 case AF_INET6: 1614 { 1615 struct ip6_pktopts *opts; 1616 1617 if (tp->t_in6pcb) 1618 opts = tp->t_in6pcb->in6p_outputopts; 1619 else 1620 opts = NULL; 1621 error = ip6_output(m, opts, ro, so->so_options & SO_DONTROUTE, 1622 NULL, so, NULL); 1623 break; 1624 } 1625 #endif 1626 default: 1627 error = EAFNOSUPPORT; 1628 break; 1629 } 1630 if (error) { 1631 out: 1632 if (error == ENOBUFS) { 1633 TCP_STATINC(TCP_STAT_SELFQUENCH); 1634 #ifdef INET 1635 if (tp->t_inpcb) 1636 tcp_quench(tp->t_inpcb, 0); 1637 #endif 1638 #ifdef INET6 1639 if (tp->t_in6pcb) 1640 tcp6_quench(tp->t_in6pcb, 0); 1641 #endif 1642 error = 0; 1643 } else if ((error == EHOSTUNREACH || error == ENETDOWN) && 1644 TCPS_HAVERCVDSYN(tp->t_state)) { 1645 tp->t_softerror = error; 1646 error = 0; 1647 } 1648 1649 /* Back out the seqence number advance. */ 1650 if (sack_rxmit) 1651 p->rxmit -= len; 1652 1653 /* Restart the delayed ACK timer, if necessary. */ 1654 if (tp->t_flags & TF_DELACK) 1655 TCP_RESTART_DELACK(tp); 1656 1657 return (error); 1658 } 1659 1660 if (packetlen > tp->t_pmtud_mtu_sent) 1661 tp->t_pmtud_mtu_sent = packetlen; 1662 1663 tcps = TCP_STAT_GETREF(); 1664 tcps[TCP_STAT_SNDTOTAL]++; 1665 if (tp->t_flags & TF_DELACK) 1666 tcps[TCP_STAT_DELACK]++; 1667 TCP_STAT_PUTREF(); 1668 1669 /* 1670 * Data sent (as far as we can tell). 1671 * If this advertises a larger window than any other segment, 1672 * then remember the size of the advertised window. 1673 * Any pending ACK has now been sent. 1674 */ 1675 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 1676 tp->rcv_adv = tp->rcv_nxt + win; 1677 tp->last_ack_sent = tp->rcv_nxt; 1678 tp->t_flags &= ~TF_ACKNOW; 1679 TCP_CLEAR_DELACK(tp); 1680 #ifdef DIAGNOSTIC 1681 if (maxburst < 0) 1682 printf("tcp_output: maxburst exceeded by %d\n", -maxburst); 1683 #endif 1684 if (sendalot && (tp->t_congctl == &tcp_reno_ctl || --maxburst)) 1685 goto again; 1686 return (0); 1687 } 1688 1689 void 1690 tcp_setpersist(struct tcpcb *tp) 1691 { 1692 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2); 1693 int nticks; 1694 1695 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT)) 1696 panic("tcp_output REXMT"); 1697 /* 1698 * Start/restart persistance timer. 1699 */ 1700 if (t < tp->t_rttmin) 1701 t = tp->t_rttmin; 1702 TCPT_RANGESET(nticks, t * tcp_backoff[tp->t_rxtshift], 1703 TCPTV_PERSMIN, TCPTV_PERSMAX); 1704 TCP_TIMER_ARM(tp, TCPT_PERSIST, nticks); 1705 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 1706 tp->t_rxtshift++; 1707 } 1708