1 /* 2 * Copyright (c) 2004 Jeffrey M. Hsu. All rights reserved. 3 * Copyright (c) 2004 The DragonFly Project. All rights reserved. 4 * 5 * This code is derived from software contributed to The DragonFly Project 6 * by Jeffrey M. Hsu. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of The DragonFly Project nor the names of its 17 * contributors may be used to endorse or promote products derived 18 * from this software without specific, prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34 /* 35 * Copyright (c) 2004 Jeffrey M. Hsu. All rights reserved. 36 * 37 * License terms: all terms for the DragonFly license above plus the following: 38 * 39 * 4. All advertising materials mentioning features or use of this software 40 * must display the following acknowledgement: 41 * 42 * This product includes software developed by Jeffrey M. Hsu 43 * for the DragonFly Project. 44 * 45 * This requirement may be waived with permission from Jeffrey Hsu. 46 * This requirement will sunset and may be removed on July 8 2005, 47 * after which the standard DragonFly license (as shown above) will 48 * apply. 49 */ 50 51 /* 52 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 53 * The Regents of the University of California. All rights reserved. 54 * 55 * Redistribution and use in source and binary forms, with or without 56 * modification, are permitted provided that the following conditions 57 * are met: 58 * 1. Redistributions of source code must retain the above copyright 59 * notice, this list of conditions and the following disclaimer. 60 * 2. Redistributions in binary form must reproduce the above copyright 61 * notice, this list of conditions and the following disclaimer in the 62 * documentation and/or other materials provided with the distribution. 63 * 3. All advertising materials mentioning features or use of this software 64 * must display the following acknowledgement: 65 * This product includes software developed by the University of 66 * California, Berkeley and its contributors. 67 * 4. Neither the name of the University nor the names of its contributors 68 * may be used to endorse or promote products derived from this software 69 * without specific prior written permission. 70 * 71 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 72 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 73 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 74 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 75 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 76 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 77 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 78 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 79 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 80 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 81 * SUCH DAMAGE. 82 * 83 * @(#)tcp_output.c 8.4 (Berkeley) 5/24/95 84 * $FreeBSD: src/sys/netinet/tcp_output.c,v 1.39.2.20 2003/01/29 22:45:36 hsu Exp $ 85 * $DragonFly: src/sys/netinet/tcp_output.c,v 1.27 2005/04/18 22:41:23 hsu Exp $ 86 */ 87 88 #include "opt_inet6.h" 89 #include "opt_ipsec.h" 90 #include "opt_tcpdebug.h" 91 92 #include <sys/param.h> 93 #include <sys/systm.h> 94 #include <sys/kernel.h> 95 #include <sys/sysctl.h> 96 #include <sys/mbuf.h> 97 #include <sys/domain.h> 98 #include <sys/protosw.h> 99 #include <sys/socket.h> 100 #include <sys/socketvar.h> 101 #include <sys/in_cksum.h> 102 #include <sys/thread.h> 103 #include <sys/globaldata.h> 104 105 #include <net/route.h> 106 107 #include <netinet/in.h> 108 #include <netinet/in_systm.h> 109 #include <netinet/ip.h> 110 #include <netinet/in_pcb.h> 111 #include <netinet/ip_var.h> 112 #include <netinet6/in6_pcb.h> 113 #include <netinet/ip6.h> 114 #include <netinet6/ip6_var.h> 115 #include <netinet/tcp.h> 116 #define TCPOUTFLAGS 117 #include <netinet/tcp_fsm.h> 118 #include <netinet/tcp_seq.h> 119 #include <netinet/tcp_timer.h> 120 #include <netinet/tcp_var.h> 121 #include <netinet/tcpip.h> 122 #ifdef TCPDEBUG 123 #include <netinet/tcp_debug.h> 124 #endif 125 126 #ifdef IPSEC 127 #include <netinet6/ipsec.h> 128 #endif /*IPSEC*/ 129 130 #ifdef FAST_IPSEC 131 #include <netproto/ipsec/ipsec.h> 132 #define IPSEC 133 #endif /*FAST_IPSEC*/ 134 135 #ifdef notyet 136 extern struct mbuf *m_copypack(); 137 #endif 138 139 int path_mtu_discovery = 0; 140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, path_mtu_discovery, CTLFLAG_RW, 141 &path_mtu_discovery, 1, "Enable Path MTU Discovery"); 142 143 static int avoid_pure_win_update = 1; 144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, avoid_pure_win_update, CTLFLAG_RW, 145 &avoid_pure_win_update, 1, "Avoid pure window updates when possible"); 146 147 /* 148 * Tcp output routine: figure out what should be sent and send it. 149 */ 150 int 151 tcp_output(tp) 152 struct tcpcb *tp; 153 { 154 struct inpcb * const inp = tp->t_inpcb; 155 struct socket *so = inp->inp_socket; 156 long len, recvwin, sendwin; 157 int nsacked = 0; 158 int off, flags, error; 159 struct mbuf *m; 160 struct ip *ip = NULL; 161 struct ipovly *ipov = NULL; 162 struct tcphdr *th; 163 u_char opt[TCP_MAXOLEN]; 164 unsigned int ipoptlen, optlen, hdrlen; 165 int idle; 166 boolean_t sendalot; 167 struct ip6_hdr *ip6 = NULL; 168 #ifdef INET6 169 const boolean_t isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 170 #else 171 const boolean_t isipv6 = FALSE; 172 #endif 173 struct rmxp_tao *taop; 174 175 /* 176 * Determine length of data that should be transmitted, 177 * and flags that will be used. 178 * If there is some data or critical controls (SYN, RST) 179 * to send, then transmit; otherwise, investigate further. 180 */ 181 182 /* 183 * If we have been idle for a while, the send congestion window 184 * could be no longer representative of the current state of the link. 185 * So unless we are expecting more acks to come in, slow-start from 186 * scratch to re-determine the send congestion window. 187 */ 188 if (tp->snd_max == tp->snd_una && 189 (ticks - tp->t_rcvtime) >= tp->t_rxtcur) { 190 if (tcp_do_rfc3390) { 191 int initial_cwnd = 192 min(4 * tp->t_maxseg, max(2 * tp->t_maxseg, 4380)); 193 194 tp->snd_cwnd = min(tp->snd_cwnd, initial_cwnd); 195 } else { 196 tp->snd_cwnd = tp->t_maxseg; 197 } 198 } 199 200 /* 201 * Calculate whether the transmit stream was previously idle 202 * and adjust TF_LASTIDLE for the next time. 203 */ 204 idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una); 205 if (idle && (tp->t_flags & TF_MORETOCOME)) 206 tp->t_flags |= TF_LASTIDLE; 207 else 208 tp->t_flags &= ~TF_LASTIDLE; 209 210 if (TCP_DO_SACK(tp) && tp->snd_nxt != tp->snd_max && 211 !IN_FASTRECOVERY(tp)) 212 nsacked = tcp_sack_bytes_below(&tp->scb, tp->snd_nxt); 213 214 again: 215 /* Make use of SACK information when slow-starting after a RTO. */ 216 if (TCP_DO_SACK(tp) && tp->snd_nxt != tp->snd_max && 217 !IN_FASTRECOVERY(tp)) { 218 tcp_seq old_snd_nxt = tp->snd_nxt; 219 220 tcp_sack_skip_sacked(&tp->scb, &tp->snd_nxt); 221 nsacked += tp->snd_nxt - old_snd_nxt; 222 } 223 224 sendalot = FALSE; 225 off = tp->snd_nxt - tp->snd_una; 226 sendwin = min(tp->snd_wnd, tp->snd_cwnd + nsacked); 227 sendwin = min(sendwin, tp->snd_bwnd); 228 229 flags = tcp_outflags[tp->t_state]; 230 /* 231 * Get standard flags, and add SYN or FIN if requested by 'hidden' 232 * state flags. 233 */ 234 if (tp->t_flags & TF_NEEDFIN) 235 flags |= TH_FIN; 236 if (tp->t_flags & TF_NEEDSYN) 237 flags |= TH_SYN; 238 239 /* 240 * If in persist timeout with window of 0, send 1 byte. 241 * Otherwise, if window is small but nonzero 242 * and timer expired, we will send what we can 243 * and go to transmit state. 244 */ 245 if (tp->t_flags & TF_FORCE) { 246 if (sendwin == 0) { 247 /* 248 * If we still have some data to send, then 249 * clear the FIN bit. Usually this would 250 * happen below when it realizes that we 251 * aren't sending all the data. However, 252 * if we have exactly 1 byte of unsent data, 253 * then it won't clear the FIN bit below, 254 * and if we are in persist state, we wind 255 * up sending the packet without recording 256 * that we sent the FIN bit. 257 * 258 * We can't just blindly clear the FIN bit, 259 * because if we don't have any more data 260 * to send then the probe will be the FIN 261 * itself. 262 */ 263 if (off < so->so_snd.sb_cc) 264 flags &= ~TH_FIN; 265 sendwin = 1; 266 } else { 267 callout_stop(tp->tt_persist); 268 tp->t_rxtshift = 0; 269 } 270 } 271 272 /* 273 * If snd_nxt == snd_max and we have transmitted a FIN, the 274 * offset will be > 0 even if so_snd.sb_cc is 0, resulting in 275 * a negative length. This can also occur when TCP opens up 276 * its congestion window while receiving additional duplicate 277 * acks after fast-retransmit because TCP will reset snd_nxt 278 * to snd_max after the fast-retransmit. 279 * 280 * In the normal retransmit-FIN-only case, however, snd_nxt will 281 * be set to snd_una, the offset will be 0, and the length may 282 * wind up 0. 283 */ 284 len = (long)ulmin(so->so_snd.sb_cc, sendwin) - off; 285 286 /* 287 * Lop off SYN bit if it has already been sent. However, if this 288 * is SYN-SENT state and if segment contains data and if we don't 289 * know that foreign host supports TAO, suppress sending segment. 290 */ 291 if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) { 292 flags &= ~TH_SYN; 293 off--, len++; 294 if (len > 0 && tp->t_state == TCPS_SYN_SENT && 295 ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL || 296 taop->tao_ccsent == 0)) 297 return 0; 298 } 299 300 /* 301 * Be careful not to send data and/or FIN on SYN segments 302 * in cases when no CC option will be sent. 303 * This measure is needed to prevent interoperability problems 304 * with not fully conformant TCP implementations. 305 */ 306 if ((flags & TH_SYN) && 307 ((tp->t_flags & TF_NOOPT) || !(tp->t_flags & TF_REQ_CC) || 308 ((flags & TH_ACK) && !(tp->t_flags & TF_RCVD_CC)))) { 309 len = 0; 310 flags &= ~TH_FIN; 311 } 312 313 if (len < 0) { 314 /* 315 * If FIN has been sent but not acked, 316 * but we haven't been called to retransmit, 317 * len will be < 0. Otherwise, window shrank 318 * after we sent into it. If window shrank to 0, 319 * cancel pending retransmit, pull snd_nxt back 320 * to (closed) window, and set the persist timer 321 * if it isn't already going. If the window didn't 322 * close completely, just wait for an ACK. 323 */ 324 len = 0; 325 if (sendwin == 0) { 326 callout_stop(tp->tt_rexmt); 327 tp->t_rxtshift = 0; 328 tp->snd_nxt = tp->snd_una; 329 if (!callout_active(tp->tt_persist)) 330 tcp_setpersist(tp); 331 } 332 } 333 334 /* 335 * len will be >= 0 after this point. Truncate to the maximum 336 * segment length and ensure that FIN is removed if the length 337 * no longer contains the last data byte. 338 */ 339 if (len > tp->t_maxseg) { 340 len = tp->t_maxseg; 341 sendalot = TRUE; 342 } 343 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc)) 344 flags &= ~TH_FIN; 345 346 recvwin = sbspace(&so->so_rcv); 347 348 /* 349 * Sender silly window avoidance. We transmit under the following 350 * conditions when len is non-zero: 351 * 352 * - We have a full segment 353 * - This is the last buffer in a write()/send() and we are 354 * either idle or running NODELAY 355 * - we've timed out (e.g. persist timer) 356 * - we have more then 1/2 the maximum send window's worth of 357 * data (receiver may be limiting the window size) 358 * - we need to retransmit 359 */ 360 if (len) { 361 if (len == tp->t_maxseg) 362 goto send; 363 /* 364 * NOTE! on localhost connections an 'ack' from the remote 365 * end may occur synchronously with the output and cause 366 * us to flush a buffer queued with moretocome. XXX 367 * 368 * note: the len + off check is almost certainly unnecessary. 369 */ 370 if (!(tp->t_flags & TF_MORETOCOME) && /* normal case */ 371 (idle || (tp->t_flags & TF_NODELAY)) && 372 len + off >= so->so_snd.sb_cc && 373 !(tp->t_flags & TF_NOPUSH)) { 374 goto send; 375 } 376 if (tp->t_flags & TF_FORCE) /* typ. timeout case */ 377 goto send; 378 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0) 379 goto send; 380 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */ 381 goto send; 382 } 383 384 /* 385 * Compare available window to amount of window 386 * known to peer (as advertised window less 387 * next expected input). If the difference is at least two 388 * max size segments, or at least 50% of the maximum possible 389 * window, then want to send a window update to peer. 390 */ 391 if (recvwin > 0) { 392 /* 393 * "adv" is the amount we can increase the window, 394 * taking into account that we are limited by 395 * TCP_MAXWIN << tp->rcv_scale. 396 */ 397 long adv = min(recvwin, (long)TCP_MAXWIN << tp->rcv_scale) - 398 (tp->rcv_adv - tp->rcv_nxt); 399 400 /* 401 * This ack case typically occurs when the user has drained 402 * the TCP socket buffer sufficiently to warrent an ack 403 * containing a 'pure window update'... that is, an ack that 404 * ONLY updates the tcp window. 405 * 406 * It is unclear why we would need to do a pure window update 407 * past 2 segments if we are going to do one at 1/2 the high 408 * water mark anyway, especially since under normal conditions 409 * the user program will drain the socket buffer quickly. 410 * The 2-segment pure window update will often add a large 411 * number of extra, unnecessary acks to the stream. 412 * 413 * avoid_pure_win_update now defaults to 1. 414 */ 415 if (avoid_pure_win_update == 0) { 416 if (adv >= (long) (2 * tp->t_maxseg)) 417 goto send; 418 } 419 if (2 * adv >= (long) so->so_rcv.sb_hiwat) 420 goto send; 421 } 422 423 /* 424 * Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW 425 * is also a catch-all for the retransmit timer timeout case. 426 */ 427 if (tp->t_flags & TF_ACKNOW) 428 goto send; 429 if ((flags & TH_RST) || 430 ((flags & TH_SYN) && !(tp->t_flags & TF_NEEDSYN))) 431 goto send; 432 if (SEQ_GT(tp->snd_up, tp->snd_una)) 433 goto send; 434 /* 435 * If our state indicates that FIN should be sent 436 * and we have not yet done so, then we need to send. 437 */ 438 if (flags & TH_FIN && 439 (!(tp->t_flags & TF_SENTFIN) || tp->snd_nxt == tp->snd_una)) 440 goto send; 441 442 /* 443 * TCP window updates are not reliable, rather a polling protocol 444 * using ``persist'' packets is used to insure receipt of window 445 * updates. The three ``states'' for the output side are: 446 * idle not doing retransmits or persists 447 * persisting to move a small or zero window 448 * (re)transmitting and thereby not persisting 449 * 450 * callout_active(tp->tt_persist) 451 * is true when we are in persist state. 452 * The TF_FORCE flag in tp->t_flags 453 * is set when we are called to send a persist packet. 454 * callout_active(tp->tt_rexmt) 455 * is set when we are retransmitting 456 * The output side is idle when both timers are zero. 457 * 458 * If send window is too small, there is data to transmit, and no 459 * retransmit or persist is pending, then go to persist state. 460 * If nothing happens soon, send when timer expires: 461 * if window is nonzero, transmit what we can, 462 * otherwise force out a byte. 463 */ 464 if (so->so_snd.sb_cc > 0 && 465 !callout_active(tp->tt_rexmt) && !callout_active(tp->tt_persist)) { 466 tp->t_rxtshift = 0; 467 tcp_setpersist(tp); 468 } 469 470 /* 471 * No reason to send a segment, just return. 472 */ 473 return (0); 474 475 send: 476 /* 477 * Before ESTABLISHED, force sending of initial options 478 * unless TCP set not to do any options. 479 * NOTE: we assume that the IP/TCP header plus TCP options 480 * always fit in a single mbuf, leaving room for a maximum 481 * link header, i.e. 482 * max_linkhdr + sizeof(struct tcpiphdr) + optlen <= MCLBYTES 483 */ 484 optlen = 0; 485 if (isipv6) 486 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 487 else 488 hdrlen = sizeof(struct tcpiphdr); 489 if (flags & TH_SYN) { 490 tp->snd_nxt = tp->iss; 491 if (!(tp->t_flags & TF_NOOPT)) { 492 u_short mss; 493 494 opt[0] = TCPOPT_MAXSEG; 495 opt[1] = TCPOLEN_MAXSEG; 496 mss = htons((u_short) tcp_mssopt(tp)); 497 memcpy(opt + 2, &mss, sizeof mss); 498 optlen = TCPOLEN_MAXSEG; 499 500 if ((tp->t_flags & TF_REQ_SCALE) && 501 (!(flags & TH_ACK) || 502 (tp->t_flags & TF_RCVD_SCALE))) { 503 *((u_int32_t *)(opt + optlen)) = htonl( 504 TCPOPT_NOP << 24 | 505 TCPOPT_WINDOW << 16 | 506 TCPOLEN_WINDOW << 8 | 507 tp->request_r_scale); 508 optlen += 4; 509 } 510 511 if ((tcp_do_sack && !(flags & TH_ACK)) || 512 tp->t_flags & TF_SACK_PERMITTED) { 513 uint32_t *lp = (uint32_t *)(opt + optlen); 514 515 *lp = htonl(TCPOPT_SACK_PERMITTED_ALIGNED); 516 optlen += TCPOLEN_SACK_PERMITTED_ALIGNED; 517 } 518 } 519 } 520 521 /* 522 * Send a timestamp and echo-reply if this is a SYN and our side 523 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side 524 * and our peer have sent timestamps in our SYN's. 525 */ 526 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP && 527 !(flags & TH_RST) && 528 (!(flags & TH_ACK) || (tp->t_flags & TF_RCVD_TSTMP))) { 529 u_int32_t *lp = (u_int32_t *)(opt + optlen); 530 531 /* Form timestamp option as shown in appendix A of RFC 1323. */ 532 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 533 *lp++ = htonl(ticks); 534 *lp = htonl(tp->ts_recent); 535 optlen += TCPOLEN_TSTAMP_APPA; 536 } 537 538 /* 539 * Send `CC-family' options if our side wants to use them (TF_REQ_CC), 540 * options are allowed (!TF_NOOPT) and it's not a RST. 541 */ 542 if ((tp->t_flags & (TF_REQ_CC | TF_NOOPT)) == TF_REQ_CC && 543 !(flags & TH_RST)) { 544 switch (flags & (TH_SYN | TH_ACK)) { 545 /* 546 * This is a normal ACK, send CC if we received CC before 547 * from our peer. 548 */ 549 case TH_ACK: 550 if (!(tp->t_flags & TF_RCVD_CC)) 551 break; 552 /*FALLTHROUGH*/ 553 554 /* 555 * We can only get here in T/TCP's SYN_SENT* state, when 556 * we're a sending a non-SYN segment without waiting for 557 * the ACK of our SYN. A check above assures that we only 558 * do this if our peer understands T/TCP. 559 */ 560 case 0: 561 opt[optlen++] = TCPOPT_NOP; 562 opt[optlen++] = TCPOPT_NOP; 563 opt[optlen++] = TCPOPT_CC; 564 opt[optlen++] = TCPOLEN_CC; 565 *(u_int32_t *)&opt[optlen] = htonl(tp->cc_send); 566 optlen += 4; 567 break; 568 569 /* 570 * This is our initial SYN, check whether we have to use 571 * CC or CC.new. 572 */ 573 case TH_SYN: 574 opt[optlen++] = TCPOPT_NOP; 575 opt[optlen++] = TCPOPT_NOP; 576 opt[optlen++] = tp->t_flags & TF_SENDCCNEW ? 577 TCPOPT_CCNEW : TCPOPT_CC; 578 opt[optlen++] = TCPOLEN_CC; 579 *(u_int32_t *)&opt[optlen] = htonl(tp->cc_send); 580 optlen += 4; 581 break; 582 583 /* 584 * This is a SYN,ACK; send CC and CC.echo if we received 585 * CC from our peer. 586 */ 587 case (TH_SYN | TH_ACK): 588 if (tp->t_flags & TF_RCVD_CC) { 589 opt[optlen++] = TCPOPT_NOP; 590 opt[optlen++] = TCPOPT_NOP; 591 opt[optlen++] = TCPOPT_CC; 592 opt[optlen++] = TCPOLEN_CC; 593 *(u_int32_t *)&opt[optlen] = htonl(tp->cc_send); 594 optlen += 4; 595 opt[optlen++] = TCPOPT_NOP; 596 opt[optlen++] = TCPOPT_NOP; 597 opt[optlen++] = TCPOPT_CCECHO; 598 opt[optlen++] = TCPOLEN_CC; 599 *(u_int32_t *)&opt[optlen] = htonl(tp->cc_recv); 600 optlen += 4; 601 } 602 break; 603 } 604 } 605 606 /* 607 * If this is a SACK connection and we have a block to report, 608 * fill in the SACK blocks in the TCP options. 609 */ 610 if ((tp->t_flags & (TF_SACK_PERMITTED | TF_NOOPT)) == 611 TF_SACK_PERMITTED && 612 (!LIST_EMPTY(&tp->t_segq) || 613 tp->reportblk.rblk_start != tp->reportblk.rblk_end)) 614 tcp_sack_fill_report(tp, opt, &optlen); 615 616 KASSERT(optlen <= TCP_MAXOLEN, ("too many TCP options")); 617 hdrlen += optlen; 618 619 if (isipv6) { 620 ipoptlen = ip6_optlen(inp); 621 } else { 622 if (inp->inp_options) { 623 ipoptlen = inp->inp_options->m_len - 624 offsetof(struct ipoption, ipopt_list); 625 } else { 626 ipoptlen = 0; 627 } 628 } 629 #ifdef IPSEC 630 ipoptlen += ipsec_hdrsiz_tcp(tp); 631 #endif 632 633 /* 634 * Adjust data length if insertion of options will 635 * bump the packet length beyond the t_maxopd length. 636 * Clear the FIN bit because we cut off the tail of 637 * the segment. 638 */ 639 if (len + optlen + ipoptlen > tp->t_maxopd) { 640 /* 641 * If there is still more to send, don't close the connection. 642 */ 643 flags &= ~TH_FIN; 644 len = tp->t_maxopd - optlen - ipoptlen; 645 sendalot = TRUE; 646 } 647 648 #ifdef INET6 649 KASSERT(max_linkhdr + hdrlen <= MCLBYTES, ("tcphdr too big")); 650 #else 651 KASSERT(max_linkhdr + hdrlen <= MHLEN, ("tcphdr too big")); 652 #endif 653 654 /* 655 * Grab a header mbuf, attaching a copy of data to 656 * be transmitted, and initialize the header from 657 * the template for sends on this connection. 658 */ 659 if (len) { 660 if ((tp->t_flags & TF_FORCE) && len == 1) 661 tcpstat.tcps_sndprobe++; 662 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { 663 if (tp->snd_nxt == tp->snd_una) 664 tp->snd_max_rexmt = tp->snd_max; 665 tcpstat.tcps_sndrexmitpack++; 666 tcpstat.tcps_sndrexmitbyte += len; 667 } else { 668 tcpstat.tcps_sndpack++; 669 tcpstat.tcps_sndbyte += len; 670 } 671 #ifdef notyet 672 if ((m = m_copypack(so->so_snd.sb_mb, off, (int)len, 673 max_linkhdr + hdrlen)) == NULL) { 674 error = ENOBUFS; 675 goto out; 676 } 677 /* 678 * m_copypack left space for our hdr; use it. 679 */ 680 m->m_len += hdrlen; 681 m->m_data -= hdrlen; 682 #else 683 MGETHDR(m, MB_DONTWAIT, MT_HEADER); 684 if (m == NULL) { 685 error = ENOBUFS; 686 goto out; 687 } 688 #ifdef INET6 689 if (MHLEN < hdrlen + max_linkhdr) { 690 MCLGET(m, MB_DONTWAIT); 691 if (!(m->m_flags & M_EXT)) { 692 m_freem(m); 693 error = ENOBUFS; 694 goto out; 695 } 696 } 697 #endif 698 m->m_data += max_linkhdr; 699 m->m_len = hdrlen; 700 if (len <= MHLEN - hdrlen - max_linkhdr) { 701 m_copydata(so->so_snd.sb_mb, off, (int) len, 702 mtod(m, caddr_t) + hdrlen); 703 m->m_len += len; 704 } else { 705 m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len); 706 if (m->m_next == NULL) { 707 m_free(m); 708 error = ENOBUFS; 709 goto out; 710 } 711 } 712 #endif 713 /* 714 * If we're sending everything we've got, set PUSH. 715 * (This will keep happy those implementations which only 716 * give data to the user when a buffer fills or 717 * a PUSH comes in.) 718 */ 719 if (off + len == so->so_snd.sb_cc) 720 flags |= TH_PUSH; 721 } else { 722 if (tp->t_flags & TF_ACKNOW) 723 tcpstat.tcps_sndacks++; 724 else if (flags & (TH_SYN | TH_FIN | TH_RST)) 725 tcpstat.tcps_sndctrl++; 726 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 727 tcpstat.tcps_sndurg++; 728 else 729 tcpstat.tcps_sndwinup++; 730 731 MGETHDR(m, MB_DONTWAIT, MT_HEADER); 732 if (m == NULL) { 733 error = ENOBUFS; 734 goto out; 735 } 736 if (isipv6 && 737 (hdrlen + max_linkhdr > MHLEN) && hdrlen <= MHLEN) 738 MH_ALIGN(m, hdrlen); 739 else 740 m->m_data += max_linkhdr; 741 m->m_len = hdrlen; 742 } 743 m->m_pkthdr.rcvif = (struct ifnet *)0; 744 if (isipv6) { 745 ip6 = mtod(m, struct ip6_hdr *); 746 th = (struct tcphdr *)(ip6 + 1); 747 tcp_fillheaders(tp, ip6, th); 748 } else { 749 ip = mtod(m, struct ip *); 750 ipov = (struct ipovly *)ip; 751 th = (struct tcphdr *)(ip + 1); 752 /* this picks up the pseudo header (w/o the length) */ 753 tcp_fillheaders(tp, ip, th); 754 } 755 756 /* 757 * Fill in fields, remembering maximum advertised 758 * window for use in delaying messages about window sizes. 759 * If resending a FIN, be sure not to use a new sequence number. 760 */ 761 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN && 762 tp->snd_nxt == tp->snd_max) 763 tp->snd_nxt--; 764 /* 765 * If we are doing retransmissions, then snd_nxt will 766 * not reflect the first unsent octet. For ACK only 767 * packets, we do not want the sequence number of the 768 * retransmitted packet, we want the sequence number 769 * of the next unsent octet. So, if there is no data 770 * (and no SYN or FIN), use snd_max instead of snd_nxt 771 * when filling in ti_seq. But if we are in persist 772 * state, snd_max might reflect one byte beyond the 773 * right edge of the window, so use snd_nxt in that 774 * case, since we know we aren't doing a retransmission. 775 * (retransmit and persist are mutually exclusive...) 776 */ 777 if (len || (flags & (TH_SYN|TH_FIN)) || callout_active(tp->tt_persist)) 778 th->th_seq = htonl(tp->snd_nxt); 779 else 780 th->th_seq = htonl(tp->snd_max); 781 th->th_ack = htonl(tp->rcv_nxt); 782 if (optlen) { 783 bcopy(opt, th + 1, optlen); 784 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 785 } 786 th->th_flags = flags; 787 /* 788 * Calculate receive window. Don't shrink window, 789 * but avoid silly window syndrome. 790 */ 791 if (recvwin < (long)(so->so_rcv.sb_hiwat / 4) && 792 recvwin < (long)tp->t_maxseg) 793 recvwin = 0; 794 if (recvwin < (long)(tp->rcv_adv - tp->rcv_nxt)) 795 recvwin = (long)(tp->rcv_adv - tp->rcv_nxt); 796 if (recvwin > (long)TCP_MAXWIN << tp->rcv_scale) 797 recvwin = (long)TCP_MAXWIN << tp->rcv_scale; 798 th->th_win = htons((u_short) (recvwin>>tp->rcv_scale)); 799 800 /* 801 * Adjust the RXWIN0SENT flag - indicate that we have advertised 802 * a 0 window. This may cause the remote transmitter to stall. This 803 * flag tells soreceive() to disable delayed acknowledgements when 804 * draining the buffer. This can occur if the receiver is attempting 805 * to read more data then can be buffered prior to transmitting on 806 * the connection. 807 */ 808 if (recvwin == 0) 809 tp->t_flags |= TF_RXWIN0SENT; 810 else 811 tp->t_flags &= ~TF_RXWIN0SENT; 812 813 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 814 th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt)); 815 th->th_flags |= TH_URG; 816 } else { 817 /* 818 * If no urgent pointer to send, then we pull 819 * the urgent pointer to the left edge of the send window 820 * so that it doesn't drift into the send window on sequence 821 * number wraparound. 822 */ 823 tp->snd_up = tp->snd_una; /* drag it along */ 824 } 825 826 /* 827 * Put TCP length in extended header, and then 828 * checksum extended header and data. 829 */ 830 m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */ 831 if (isipv6) { 832 /* 833 * ip6_plen is not need to be filled now, and will be filled 834 * in ip6_output(). 835 */ 836 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 837 sizeof(struct tcphdr) + optlen + len); 838 } else { 839 m->m_pkthdr.csum_flags = CSUM_TCP; 840 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 841 if (len + optlen) 842 th->th_sum = in_addword(th->th_sum, 843 htons((u_short)(optlen + len))); 844 845 /* IP version must be set here for ipv4/ipv6 checking later */ 846 KASSERT(ip->ip_v == IPVERSION, 847 ("%s: IP version incorrect: %d", __func__, ip->ip_v)); 848 } 849 850 /* 851 * In transmit state, time the transmission and arrange for 852 * the retransmit. In persist state, just set snd_max. 853 */ 854 if (!(tp->t_flags & TF_FORCE) || !callout_active(tp->tt_persist)) { 855 tcp_seq startseq = tp->snd_nxt; 856 857 /* 858 * Advance snd_nxt over sequence space of this segment. 859 */ 860 if (flags & (TH_SYN | TH_FIN)) { 861 if (flags & TH_SYN) 862 tp->snd_nxt++; 863 if (flags & TH_FIN) { 864 tp->snd_nxt++; 865 tp->t_flags |= TF_SENTFIN; 866 } 867 } 868 tp->snd_nxt += len; 869 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 870 tp->snd_max = tp->snd_nxt; 871 /* 872 * Time this transmission if not a retransmission and 873 * not currently timing anything. 874 */ 875 if (tp->t_rtttime == 0) { 876 tp->t_rtttime = ticks; 877 tp->t_rtseq = startseq; 878 tcpstat.tcps_segstimed++; 879 } 880 } 881 882 /* 883 * Set retransmit timer if not currently set, 884 * and not doing a pure ack or a keep-alive probe. 885 * Initial value for retransmit timer is smoothed 886 * round-trip time + 2 * round-trip time variance. 887 * Initialize shift counter which is used for backoff 888 * of retransmit time. 889 */ 890 if (!callout_active(tp->tt_rexmt) && 891 tp->snd_nxt != tp->snd_una) { 892 if (callout_active(tp->tt_persist)) { 893 callout_stop(tp->tt_persist); 894 tp->t_rxtshift = 0; 895 } 896 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 897 tcp_timer_rexmt, tp); 898 } 899 } else { 900 /* 901 * Persist case, update snd_max but since we are in 902 * persist mode (no window) we do not update snd_nxt. 903 */ 904 int xlen = len; 905 if (flags & TH_SYN) 906 ++xlen; 907 if (flags & TH_FIN) { 908 ++xlen; 909 tp->t_flags |= TF_SENTFIN; 910 } 911 if (SEQ_GT(tp->snd_nxt + xlen, tp->snd_max)) 912 tp->snd_max = tp->snd_nxt + xlen; 913 } 914 915 #ifdef TCPDEBUG 916 /* 917 * Trace. 918 */ 919 if (so->so_options & SO_DEBUG) 920 tcp_trace(TA_OUTPUT, tp->t_state, tp, mtod(m, void *), th, 0); 921 #endif 922 923 /* 924 * Fill in IP length and desired time to live and 925 * send to IP level. There should be a better way 926 * to handle ttl and tos; we could keep them in 927 * the template, but need a way to checksum without them. 928 */ 929 /* 930 * m->m_pkthdr.len should have been set before cksum calcuration, 931 * because in6_cksum() need it. 932 */ 933 if (isipv6) { 934 /* 935 * we separately set hoplimit for every segment, since the 936 * user might want to change the value via setsockopt. 937 * Also, desired default hop limit might be changed via 938 * Neighbor Discovery. 939 */ 940 ip6->ip6_hlim = in6_selecthlim(inp, 941 (inp->in6p_route.ro_rt ? 942 inp->in6p_route.ro_rt->rt_ifp : NULL)); 943 944 /* TODO: IPv6 IP6TOS_ECT bit on */ 945 error = ip6_output(m, inp->in6p_outputopts, &inp->in6p_route, 946 (so->so_options & SO_DONTROUTE), NULL, NULL, 947 inp); 948 } else { 949 struct rtentry *rt; 950 ip->ip_len = m->m_pkthdr.len; 951 #ifdef INET6 952 if (INP_CHECK_SOCKAF(so, AF_INET6)) 953 ip->ip_ttl = in6_selecthlim(inp, 954 (inp->in6p_route.ro_rt ? 955 inp->in6p_route.ro_rt->rt_ifp : NULL)); 956 else 957 #endif 958 ip->ip_ttl = inp->inp_ip_ttl; /* XXX */ 959 960 ip->ip_tos = inp->inp_ip_tos; /* XXX */ 961 /* 962 * See if we should do MTU discovery. 963 * We do it only if the following are true: 964 * 1) we have a valid route to the destination 965 * 2) the MTU is not locked (if it is, 966 * then discovery has been disabled) 967 */ 968 if (path_mtu_discovery && 969 (rt = inp->inp_route.ro_rt) && (rt->rt_flags & RTF_UP) && 970 !(rt->rt_rmx.rmx_locks & RTV_MTU)) 971 ip->ip_off |= IP_DF; 972 973 error = ip_output(m, inp->inp_options, &inp->inp_route, 974 (so->so_options & SO_DONTROUTE), NULL, inp); 975 } 976 if (error) { 977 978 /* 979 * We know that the packet was lost, so back out the 980 * sequence number advance, if any. 981 */ 982 if (!(tp->t_flags & TF_FORCE) || 983 !callout_active(tp->tt_persist)) { 984 /* 985 * No need to check for TH_FIN here because 986 * the TF_SENTFIN flag handles that case. 987 */ 988 if (!(flags & TH_SYN)) 989 tp->snd_nxt -= len; 990 } 991 992 out: 993 if (error == ENOBUFS) { 994 /* 995 * If we can't send, make sure there is something 996 * to get us going again later. Persist state 997 * is not necessarily right, but it is close enough. 998 */ 999 if (!callout_active(tp->tt_rexmt) && 1000 !callout_active(tp->tt_persist)) { 1001 tp->t_rxtshift = 0; 1002 tcp_setpersist(tp); 1003 } 1004 tcp_quench(inp, 0); 1005 return (0); 1006 } 1007 if (error == EMSGSIZE) { 1008 /* 1009 * ip_output() will have already fixed the route 1010 * for us. tcp_mtudisc() will, as its last action, 1011 * initiate retransmission, so it is important to 1012 * not do so here. 1013 */ 1014 tcp_mtudisc(inp, 0); 1015 return 0; 1016 } 1017 if ((error == EHOSTUNREACH || error == ENETDOWN) && 1018 TCPS_HAVERCVDSYN(tp->t_state)) { 1019 tp->t_softerror = error; 1020 return (0); 1021 } 1022 return (error); 1023 } 1024 tcpstat.tcps_sndtotal++; 1025 1026 /* 1027 * Data sent (as far as we can tell). 1028 * If this advertises a larger window than any other segment, 1029 * then remember the size of the advertised window. 1030 * Any pending ACK has now been sent. 1031 */ 1032 if (recvwin > 0 && SEQ_GT(tp->rcv_nxt + recvwin, tp->rcv_adv)) 1033 tp->rcv_adv = tp->rcv_nxt + recvwin; 1034 tp->last_ack_sent = tp->rcv_nxt; 1035 tp->t_flags &= ~TF_ACKNOW; 1036 if (tcp_delack_enabled) 1037 callout_stop(tp->tt_delack); 1038 if (sendalot) 1039 goto again; 1040 return (0); 1041 } 1042 1043 void 1044 tcp_setpersist(tp) 1045 struct tcpcb *tp; 1046 { 1047 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1; 1048 int tt; 1049 1050 if (callout_active(tp->tt_rexmt)) 1051 panic("tcp_setpersist: retransmit pending"); 1052 /* 1053 * Start/restart persistance timer. 1054 */ 1055 TCPT_RANGESET(tt, t * tcp_backoff[tp->t_rxtshift], TCPTV_PERSMIN, 1056 TCPTV_PERSMAX); 1057 callout_reset(tp->tt_persist, tt, tcp_timer_persist, tp); 1058 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 1059 tp->t_rxtshift++; 1060 } 1061