1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)tcp_output.c 8.4 (Berkeley) 5/24/95 34 * $Id: tcp_output.c,v 1.15 1995/10/16 18:21:12 wollman Exp $ 35 */ 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/malloc.h> 40 #include <sys/mbuf.h> 41 #include <sys/protosw.h> 42 #include <sys/socket.h> 43 #include <sys/socketvar.h> 44 #include <sys/errno.h> 45 #include <sys/queue.h> 46 47 #include <net/route.h> 48 49 #include <netinet/in.h> 50 #include <netinet/in_systm.h> 51 #include <netinet/ip.h> 52 #include <netinet/in_pcb.h> 53 #include <netinet/ip_var.h> 54 #include <netinet/tcp.h> 55 #define TCPOUTFLAGS 56 #include <netinet/tcp_fsm.h> 57 #include <netinet/tcp_seq.h> 58 #include <netinet/tcp_timer.h> 59 #include <netinet/tcp_var.h> 60 #include <netinet/tcpip.h> 61 #ifdef TCPDEBUG 62 #include <netinet/tcp_debug.h> 63 #endif 64 65 #ifdef notyet 66 extern struct mbuf *m_copypack(); 67 #endif 68 69 70 /* 71 * Tcp output routine: figure out what should be sent and send it. 72 */ 73 int 74 tcp_output(tp) 75 register struct tcpcb *tp; 76 { 77 register struct socket *so = tp->t_inpcb->inp_socket; 78 register long len, win; 79 int off, flags, error; 80 register struct mbuf *m; 81 register struct tcpiphdr *ti; 82 u_char opt[TCP_MAXOLEN]; 83 unsigned optlen, hdrlen; 84 int idle, sendalot; 85 struct rmxp_tao *taop; 86 struct rmxp_tao tao_noncached; 87 88 /* 89 * Determine length of data that should be transmitted, 90 * and flags that will be used. 91 * If there is some data or critical controls (SYN, RST) 92 * to send, then transmit; otherwise, investigate further. 93 */ 94 idle = (tp->snd_max == tp->snd_una); 95 if (idle && tp->t_idle >= tp->t_rxtcur) 96 /* 97 * We have been idle for "a while" and no acks are 98 * expected to clock out any data we send -- 99 * slow start to get ack "clock" running again. 100 */ 101 tp->snd_cwnd = tp->t_maxseg; 102 again: 103 sendalot = 0; 104 off = tp->snd_nxt - tp->snd_una; 105 win = min(tp->snd_wnd, tp->snd_cwnd); 106 107 flags = tcp_outflags[tp->t_state]; 108 /* 109 * Get standard flags, and add SYN or FIN if requested by 'hidden' 110 * state flags. 111 */ 112 if (tp->t_flags & TF_NEEDFIN) 113 flags |= TH_FIN; 114 if (tp->t_flags & TF_NEEDSYN) 115 flags |= TH_SYN; 116 117 /* 118 * If in persist timeout with window of 0, send 1 byte. 119 * Otherwise, if window is small but nonzero 120 * and timer expired, we will send what we can 121 * and go to transmit state. 122 */ 123 if (tp->t_force) { 124 if (win == 0) { 125 /* 126 * If we still have some data to send, then 127 * clear the FIN bit. Usually this would 128 * happen below when it realizes that we 129 * aren't sending all the data. However, 130 * if we have exactly 1 byte of unset data, 131 * then it won't clear the FIN bit below, 132 * and if we are in persist state, we wind 133 * up sending the packet without recording 134 * that we sent the FIN bit. 135 * 136 * We can't just blindly clear the FIN bit, 137 * because if we don't have any more data 138 * to send then the probe will be the FIN 139 * itself. 140 */ 141 if (off < so->so_snd.sb_cc) 142 flags &= ~TH_FIN; 143 win = 1; 144 } else { 145 tp->t_timer[TCPT_PERSIST] = 0; 146 tp->t_rxtshift = 0; 147 } 148 } 149 150 len = min(so->so_snd.sb_cc, win) - off; 151 152 if ((taop = tcp_gettaocache(tp->t_inpcb)) == NULL) { 153 taop = &tao_noncached; 154 bzero(taop, sizeof(*taop)); 155 } 156 157 /* 158 * Lop off SYN bit if it has already been sent. However, if this 159 * is SYN-SENT state and if segment contains data and if we don't 160 * know that foreign host supports TAO, suppress sending segment. 161 */ 162 if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) { 163 flags &= ~TH_SYN; 164 off--, len++; 165 if (len > 0 && tp->t_state == TCPS_SYN_SENT && 166 taop->tao_ccsent == 0) 167 return 0; 168 } 169 170 if (len < 0) { 171 /* 172 * If FIN has been sent but not acked, 173 * but we haven't been called to retransmit, 174 * len will be -1. Otherwise, window shrank 175 * after we sent into it. If window shrank to 0, 176 * cancel pending retransmit and pull snd_nxt 177 * back to (closed) window. We will enter persist 178 * state below. If the window didn't close completely, 179 * just wait for an ACK. 180 */ 181 len = 0; 182 if (win == 0) { 183 tp->t_timer[TCPT_REXMT] = 0; 184 tp->snd_nxt = tp->snd_una; 185 } 186 } 187 if (len > tp->t_maxseg) { 188 len = tp->t_maxseg; 189 sendalot = 1; 190 } 191 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc)) 192 flags &= ~TH_FIN; 193 194 win = sbspace(&so->so_rcv); 195 196 /* 197 * Sender silly window avoidance. If connection is idle 198 * and can send all data, a maximum segment, 199 * at least a maximum default-size segment do it, 200 * or are forced, do it; otherwise don't bother. 201 * If peer's buffer is tiny, then send 202 * when window is at least half open. 203 * If retransmitting (possibly after persist timer forced us 204 * to send into a small window), then must resend. 205 */ 206 if (len) { 207 if (len == tp->t_maxseg) 208 goto send; 209 if ((idle || tp->t_flags & TF_NODELAY) && 210 (tp->t_flags & TF_NOPUSH) == 0 && 211 len + off >= so->so_snd.sb_cc) 212 goto send; 213 if (tp->t_force) 214 goto send; 215 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0) 216 goto send; 217 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) 218 goto send; 219 } 220 221 /* 222 * Compare available window to amount of window 223 * known to peer (as advertised window less 224 * next expected input). If the difference is at least two 225 * max size segments, or at least 50% of the maximum possible 226 * window, then want to send a window update to peer. 227 */ 228 if (win > 0) { 229 /* 230 * "adv" is the amount we can increase the window, 231 * taking into account that we are limited by 232 * TCP_MAXWIN << tp->rcv_scale. 233 */ 234 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) - 235 (tp->rcv_adv - tp->rcv_nxt); 236 237 if (adv >= (long) (2 * tp->t_maxseg)) 238 goto send; 239 if (2 * adv >= (long) so->so_rcv.sb_hiwat) 240 goto send; 241 } 242 243 /* 244 * Send if we owe peer an ACK. 245 */ 246 if (tp->t_flags & TF_ACKNOW) 247 goto send; 248 if ((flags & TH_RST) || 249 ((flags & TH_SYN) && (tp->t_flags & TF_NEEDSYN) == 0)) 250 goto send; 251 if (SEQ_GT(tp->snd_up, tp->snd_una)) 252 goto send; 253 /* 254 * If our state indicates that FIN should be sent 255 * and we have not yet done so, or we're retransmitting the FIN, 256 * then we need to send. 257 */ 258 if (flags & TH_FIN && 259 ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una)) 260 goto send; 261 262 /* 263 * TCP window updates are not reliable, rather a polling protocol 264 * using ``persist'' packets is used to insure receipt of window 265 * updates. The three ``states'' for the output side are: 266 * idle not doing retransmits or persists 267 * persisting to move a small or zero window 268 * (re)transmitting and thereby not persisting 269 * 270 * tp->t_timer[TCPT_PERSIST] 271 * is set when we are in persist state. 272 * tp->t_force 273 * is set when we are called to send a persist packet. 274 * tp->t_timer[TCPT_REXMT] 275 * is set when we are retransmitting 276 * The output side is idle when both timers are zero. 277 * 278 * If send window is too small, there is data to transmit, and no 279 * retransmit or persist is pending, then go to persist state. 280 * If nothing happens soon, send when timer expires: 281 * if window is nonzero, transmit what we can, 282 * otherwise force out a byte. 283 */ 284 if (so->so_snd.sb_cc && tp->t_timer[TCPT_REXMT] == 0 && 285 tp->t_timer[TCPT_PERSIST] == 0) { 286 tp->t_rxtshift = 0; 287 tcp_setpersist(tp); 288 } 289 290 /* 291 * No reason to send a segment, just return. 292 */ 293 return (0); 294 295 send: 296 /* 297 * Before ESTABLISHED, force sending of initial options 298 * unless TCP set not to do any options. 299 * NOTE: we assume that the IP/TCP header plus TCP options 300 * always fit in a single mbuf, leaving room for a maximum 301 * link header, i.e. 302 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MHLEN 303 */ 304 optlen = 0; 305 hdrlen = sizeof (struct tcpiphdr); 306 if (flags & TH_SYN) { 307 tp->snd_nxt = tp->iss; 308 if ((tp->t_flags & TF_NOOPT) == 0) { 309 u_short mss; 310 311 opt[0] = TCPOPT_MAXSEG; 312 opt[1] = TCPOLEN_MAXSEG; 313 mss = htons((u_short) tcp_mssopt(tp)); 314 (void)memcpy(opt + 2, &mss, sizeof(mss)); 315 optlen = TCPOLEN_MAXSEG; 316 317 if ((tp->t_flags & TF_REQ_SCALE) && 318 ((flags & TH_ACK) == 0 || 319 (tp->t_flags & TF_RCVD_SCALE))) { 320 *((u_long *) (opt + optlen)) = htonl( 321 TCPOPT_NOP << 24 | 322 TCPOPT_WINDOW << 16 | 323 TCPOLEN_WINDOW << 8 | 324 tp->request_r_scale); 325 optlen += 4; 326 } 327 } 328 } 329 330 /* 331 * Send a timestamp and echo-reply if this is a SYN and our side 332 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side 333 * and our peer have sent timestamps in our SYN's. 334 */ 335 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 336 (flags & TH_RST) == 0 && 337 ((flags & TH_ACK) == 0 || 338 (tp->t_flags & TF_RCVD_TSTMP))) { 339 u_long *lp = (u_long *)(opt + optlen); 340 341 /* Form timestamp option as shown in appendix A of RFC 1323. */ 342 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 343 *lp++ = htonl(tcp_now); 344 *lp = htonl(tp->ts_recent); 345 optlen += TCPOLEN_TSTAMP_APPA; 346 } 347 348 /* 349 * Send `CC-family' options if our side wants to use them (TF_REQ_CC), 350 * options are allowed (!TF_NOOPT) and it's not a RST. 351 */ 352 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 353 (flags & TH_RST) == 0) { 354 switch (flags & (TH_SYN|TH_ACK)) { 355 /* 356 * This is a normal ACK, send CC if we received CC before 357 * from our peer. 358 */ 359 case TH_ACK: 360 if (!(tp->t_flags & TF_RCVD_CC)) 361 break; 362 /*FALLTHROUGH*/ 363 364 /* 365 * We can only get here in T/TCP's SYN_SENT* state, when 366 * we're a sending a non-SYN segment without waiting for 367 * the ACK of our SYN. A check above assures that we only 368 * do this if our peer understands T/TCP. 369 */ 370 case 0: 371 opt[optlen++] = TCPOPT_NOP; 372 opt[optlen++] = TCPOPT_NOP; 373 opt[optlen++] = TCPOPT_CC; 374 opt[optlen++] = TCPOLEN_CC; 375 *(u_int32_t *)&opt[optlen] = htonl(tp->cc_send); 376 377 optlen += 4; 378 break; 379 380 /* 381 * This is our initial SYN, check whether we have to use 382 * CC or CC.new. 383 */ 384 case TH_SYN: 385 opt[optlen++] = TCPOPT_NOP; 386 opt[optlen++] = TCPOPT_NOP; 387 opt[optlen++] = tp->t_flags & TF_SENDCCNEW ? 388 TCPOPT_CCNEW : TCPOPT_CC; 389 opt[optlen++] = TCPOLEN_CC; 390 *(u_int32_t *)&opt[optlen] = htonl(tp->cc_send); 391 optlen += 4; 392 break; 393 394 /* 395 * This is a SYN,ACK; send CC and CC.echo if we received 396 * CC from our peer. 397 */ 398 case (TH_SYN|TH_ACK): 399 if (tp->t_flags & TF_RCVD_CC) { 400 opt[optlen++] = TCPOPT_NOP; 401 opt[optlen++] = TCPOPT_NOP; 402 opt[optlen++] = TCPOPT_CC; 403 opt[optlen++] = TCPOLEN_CC; 404 *(u_int32_t *)&opt[optlen] = 405 htonl(tp->cc_send); 406 optlen += 4; 407 opt[optlen++] = TCPOPT_NOP; 408 opt[optlen++] = TCPOPT_NOP; 409 opt[optlen++] = TCPOPT_CCECHO; 410 opt[optlen++] = TCPOLEN_CC; 411 *(u_int32_t *)&opt[optlen] = 412 htonl(tp->cc_recv); 413 optlen += 4; 414 } 415 break; 416 } 417 } 418 419 hdrlen += optlen; 420 421 /* 422 * Adjust data length if insertion of options will 423 * bump the packet length beyond the t_maxopd length. 424 * Clear the FIN bit because we cut off the tail of 425 * the segment. 426 */ 427 if (len + optlen > tp->t_maxopd) { 428 /* 429 * If there is still more to send, don't close the connection. 430 */ 431 flags &= ~TH_FIN; 432 len = tp->t_maxopd - optlen; 433 sendalot = 1; 434 } 435 436 /*#ifdef DIAGNOSTIC*/ 437 if (max_linkhdr + hdrlen > MHLEN) 438 panic("tcphdr too big"); 439 /*#endif*/ 440 441 /* 442 * Grab a header mbuf, attaching a copy of data to 443 * be transmitted, and initialize the header from 444 * the template for sends on this connection. 445 */ 446 if (len) { 447 if (tp->t_force && len == 1) 448 tcpstat.tcps_sndprobe++; 449 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { 450 tcpstat.tcps_sndrexmitpack++; 451 tcpstat.tcps_sndrexmitbyte += len; 452 } else { 453 tcpstat.tcps_sndpack++; 454 tcpstat.tcps_sndbyte += len; 455 } 456 #ifdef notyet 457 if ((m = m_copypack(so->so_snd.sb_mb, off, 458 (int)len, max_linkhdr + hdrlen)) == 0) { 459 error = ENOBUFS; 460 goto out; 461 } 462 /* 463 * m_copypack left space for our hdr; use it. 464 */ 465 m->m_len += hdrlen; 466 m->m_data -= hdrlen; 467 #else 468 MGETHDR(m, M_DONTWAIT, MT_HEADER); 469 if (m == NULL) { 470 error = ENOBUFS; 471 goto out; 472 } 473 m->m_data += max_linkhdr; 474 m->m_len = hdrlen; 475 if (len <= MHLEN - hdrlen - max_linkhdr) { 476 m_copydata(so->so_snd.sb_mb, off, (int) len, 477 mtod(m, caddr_t) + hdrlen); 478 m->m_len += len; 479 } else { 480 m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len); 481 if (m->m_next == 0) { 482 (void) m_free(m); 483 error = ENOBUFS; 484 goto out; 485 } 486 } 487 #endif 488 /* 489 * If we're sending everything we've got, set PUSH. 490 * (This will keep happy those implementations which only 491 * give data to the user when a buffer fills or 492 * a PUSH comes in.) 493 */ 494 if (off + len == so->so_snd.sb_cc) 495 flags |= TH_PUSH; 496 } else { 497 if (tp->t_flags & TF_ACKNOW) 498 tcpstat.tcps_sndacks++; 499 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 500 tcpstat.tcps_sndctrl++; 501 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 502 tcpstat.tcps_sndurg++; 503 else 504 tcpstat.tcps_sndwinup++; 505 506 MGETHDR(m, M_DONTWAIT, MT_HEADER); 507 if (m == NULL) { 508 error = ENOBUFS; 509 goto out; 510 } 511 m->m_data += max_linkhdr; 512 m->m_len = hdrlen; 513 } 514 m->m_pkthdr.rcvif = (struct ifnet *)0; 515 ti = mtod(m, struct tcpiphdr *); 516 if (tp->t_template == 0) 517 panic("tcp_output"); 518 (void)memcpy(ti, tp->t_template, sizeof (struct tcpiphdr)); 519 520 /* 521 * Fill in fields, remembering maximum advertised 522 * window for use in delaying messages about window sizes. 523 * If resending a FIN, be sure not to use a new sequence number. 524 */ 525 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN && 526 tp->snd_nxt == tp->snd_max) 527 tp->snd_nxt--; 528 /* 529 * If we are doing retransmissions, then snd_nxt will 530 * not reflect the first unsent octet. For ACK only 531 * packets, we do not want the sequence number of the 532 * retransmitted packet, we want the sequence number 533 * of the next unsent octet. So, if there is no data 534 * (and no SYN or FIN), use snd_max instead of snd_nxt 535 * when filling in ti_seq. But if we are in persist 536 * state, snd_max might reflect one byte beyond the 537 * right edge of the window, so use snd_nxt in that 538 * case, since we know we aren't doing a retransmission. 539 * (retransmit and persist are mutually exclusive...) 540 */ 541 if (len || (flags & (TH_SYN|TH_FIN)) || tp->t_timer[TCPT_PERSIST]) 542 ti->ti_seq = htonl(tp->snd_nxt); 543 else 544 ti->ti_seq = htonl(tp->snd_max); 545 ti->ti_ack = htonl(tp->rcv_nxt); 546 if (optlen) { 547 (void)memcpy(ti + 1, opt, optlen); 548 ti->ti_off = (sizeof (struct tcphdr) + optlen) >> 2; 549 } 550 ti->ti_flags = flags; 551 /* 552 * Calculate receive window. Don't shrink window, 553 * but avoid silly window syndrome. 554 */ 555 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)tp->t_maxseg) 556 win = 0; 557 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 558 win = (long)TCP_MAXWIN << tp->rcv_scale; 559 if (win < (long)(tp->rcv_adv - tp->rcv_nxt)) 560 win = (long)(tp->rcv_adv - tp->rcv_nxt); 561 ti->ti_win = htons((u_short) (win>>tp->rcv_scale)); 562 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 563 ti->ti_urp = htons((u_short)(tp->snd_up - tp->snd_nxt)); 564 ti->ti_flags |= TH_URG; 565 } else 566 /* 567 * If no urgent pointer to send, then we pull 568 * the urgent pointer to the left edge of the send window 569 * so that it doesn't drift into the send window on sequence 570 * number wraparound. 571 */ 572 tp->snd_up = tp->snd_una; /* drag it along */ 573 574 /* 575 * Put TCP length in extended header, and then 576 * checksum extended header and data. 577 */ 578 if (len + optlen) 579 ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + 580 optlen + len)); 581 ti->ti_sum = in_cksum(m, (int)(hdrlen + len)); 582 583 /* 584 * In transmit state, time the transmission and arrange for 585 * the retransmit. In persist state, just set snd_max. 586 */ 587 if (tp->t_force == 0 || tp->t_timer[TCPT_PERSIST] == 0) { 588 tcp_seq startseq = tp->snd_nxt; 589 590 /* 591 * Advance snd_nxt over sequence space of this segment. 592 */ 593 if (flags & (TH_SYN|TH_FIN)) { 594 if (flags & TH_SYN) 595 tp->snd_nxt++; 596 if (flags & TH_FIN) { 597 tp->snd_nxt++; 598 tp->t_flags |= TF_SENTFIN; 599 } 600 } 601 tp->snd_nxt += len; 602 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 603 tp->snd_max = tp->snd_nxt; 604 /* 605 * Time this transmission if not a retransmission and 606 * not currently timing anything. 607 */ 608 if (tp->t_rtt == 0) { 609 tp->t_rtt = 1; 610 tp->t_rtseq = startseq; 611 tcpstat.tcps_segstimed++; 612 } 613 } 614 615 /* 616 * Set retransmit timer if not currently set, 617 * and not doing an ack or a keep-alive probe. 618 * Initial value for retransmit timer is smoothed 619 * round-trip time + 2 * round-trip time variance. 620 * Initialize shift counter which is used for backoff 621 * of retransmit time. 622 */ 623 if (tp->t_timer[TCPT_REXMT] == 0 && 624 tp->snd_nxt != tp->snd_una) { 625 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 626 if (tp->t_timer[TCPT_PERSIST]) { 627 tp->t_timer[TCPT_PERSIST] = 0; 628 tp->t_rxtshift = 0; 629 } 630 } 631 } else 632 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max)) 633 tp->snd_max = tp->snd_nxt + len; 634 635 #ifdef TCPDEBUG 636 /* 637 * Trace. 638 */ 639 if (so->so_options & SO_DEBUG) 640 tcp_trace(TA_OUTPUT, tp->t_state, tp, ti, 0); 641 #endif 642 643 /* 644 * Fill in IP length and desired time to live and 645 * send to IP level. There should be a better way 646 * to handle ttl and tos; we could keep them in 647 * the template, but need a way to checksum without them. 648 */ 649 m->m_pkthdr.len = hdrlen + len; 650 #ifdef TUBA 651 if (tp->t_tuba_pcb) 652 error = tuba_output(m, tp); 653 else 654 #endif 655 { 656 #ifdef MTUDISC 657 struct rtentry *rt; 658 #endif 659 ((struct ip *)ti)->ip_len = m->m_pkthdr.len; 660 ((struct ip *)ti)->ip_ttl = tp->t_inpcb->inp_ip.ip_ttl; /* XXX */ 661 ((struct ip *)ti)->ip_tos = tp->t_inpcb->inp_ip.ip_tos; /* XXX */ 662 #ifdef MTUDISC 663 /* 664 * See if we should do MTU discovery. We do it only if the following 665 * are true: 666 * 1) we have a valid route to the destination 667 * 2) the MTU is not locked (if it is, then discovery has been 668 * disabled) 669 */ 670 if ((rt = tp->t_inpcb->inp_route.ro_rt) 671 && rt->rt_flags & RTF_UP 672 && !(rt->rt_rmx.rmx_locks & RTV_MTU)) { 673 ((struct ip *)ti)->ip_off |= IP_DF; 674 } 675 #endif /* MTUDISC */ 676 error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route, 677 so->so_options & SO_DONTROUTE, 0); 678 } 679 if (error) { 680 out: 681 if (error == ENOBUFS) { 682 tcp_quench(tp->t_inpcb, 0); 683 return (0); 684 } 685 #ifdef MTUDISC 686 if (error == EMSGSIZE) { 687 /* 688 * ip_output() will have already fixed the route 689 * for us. tcp_mtudisc() will, as its last action, 690 * initiate retransmission, so it is important to 691 * not do so here. 692 */ 693 tcp_mtudisc(tp->t_inpcb, 0); 694 return 0; 695 } 696 #endif /* MTUDISC */. 697 if ((error == EHOSTUNREACH || error == ENETDOWN) 698 && TCPS_HAVERCVDSYN(tp->t_state)) { 699 tp->t_softerror = error; 700 return (0); 701 } 702 return (error); 703 } 704 tcpstat.tcps_sndtotal++; 705 706 /* 707 * Data sent (as far as we can tell). 708 * If this advertises a larger window than any other segment, 709 * then remember the size of the advertised window. 710 * Any pending ACK has now been sent. 711 */ 712 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 713 tp->rcv_adv = tp->rcv_nxt + win; 714 tp->last_ack_sent = tp->rcv_nxt; 715 tp->t_flags &= ~(TF_ACKNOW|TF_DELACK); 716 if (sendalot) 717 goto again; 718 return (0); 719 } 720 721 void 722 tcp_setpersist(tp) 723 register struct tcpcb *tp; 724 { 725 register t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1; 726 727 if (tp->t_timer[TCPT_REXMT]) 728 panic("tcp_output REXMT"); 729 /* 730 * Start/restart persistance timer. 731 */ 732 TCPT_RANGESET(tp->t_timer[TCPT_PERSIST], 733 t * tcp_backoff[tp->t_rxtshift], 734 TCPTV_PERSMIN, TCPTV_PERSMAX); 735 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 736 tp->t_rxtshift++; 737 } 738