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 * $FreeBSD$ 35 */ 36 37 #include "opt_inet6.h" 38 #include "opt_ipsec.h" 39 #include "opt_mac.h" 40 #include "opt_tcpdebug.h" 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/domain.h> 45 #include <sys/kernel.h> 46 #include <sys/lock.h> 47 #include <sys/mac.h> 48 #include <sys/mbuf.h> 49 #include <sys/mutex.h> 50 #include <sys/protosw.h> 51 #include <sys/socket.h> 52 #include <sys/socketvar.h> 53 #include <sys/sysctl.h> 54 55 #include <net/route.h> 56 57 #include <netinet/in.h> 58 #include <netinet/in_systm.h> 59 #include <netinet/ip.h> 60 #include <netinet/in_pcb.h> 61 #include <netinet/ip_var.h> 62 #ifdef INET6 63 #include <netinet6/in6_pcb.h> 64 #include <netinet/ip6.h> 65 #include <netinet6/ip6_var.h> 66 #endif 67 #include <netinet/tcp.h> 68 #define TCPOUTFLAGS 69 #include <netinet/tcp_fsm.h> 70 #include <netinet/tcp_seq.h> 71 #include <netinet/tcp_timer.h> 72 #include <netinet/tcp_var.h> 73 #include <netinet/tcpip.h> 74 #ifdef TCPDEBUG 75 #include <netinet/tcp_debug.h> 76 #endif 77 78 #ifdef IPSEC 79 #include <netinet6/ipsec.h> 80 #endif /*IPSEC*/ 81 82 #ifdef FAST_IPSEC 83 #include <netipsec/ipsec.h> 84 #define IPSEC 85 #endif /*FAST_IPSEC*/ 86 87 #include <machine/in_cksum.h> 88 89 #ifdef notyet 90 extern struct mbuf *m_copypack(); 91 #endif 92 93 int path_mtu_discovery = 1; 94 SYSCTL_INT(_net_inet_tcp, OID_AUTO, path_mtu_discovery, CTLFLAG_RW, 95 &path_mtu_discovery, 1, "Enable Path MTU Discovery"); 96 97 int ss_fltsz = 1; 98 SYSCTL_INT(_net_inet_tcp, OID_AUTO, slowstart_flightsize, CTLFLAG_RW, 99 &ss_fltsz, 1, "Slow start flight size"); 100 101 int ss_fltsz_local = 4; 102 SYSCTL_INT(_net_inet_tcp, OID_AUTO, local_slowstart_flightsize, CTLFLAG_RW, 103 &ss_fltsz_local, 1, "Slow start flight size for local networks"); 104 105 int tcp_do_newreno = 1; 106 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newreno, CTLFLAG_RW, &tcp_do_newreno, 107 0, "Enable NewReno Algorithms"); 108 109 /* 110 * Tcp output routine: figure out what should be sent and send it. 111 */ 112 int 113 tcp_output(struct tcpcb *tp) 114 { 115 struct socket *so = tp->t_inpcb->inp_socket; 116 long len, win; 117 int off, flags, error; 118 struct mbuf *m; 119 struct ip *ip = NULL; 120 struct ipovly *ipov = NULL; 121 struct tcphdr *th; 122 u_char opt[TCP_MAXOLEN]; 123 unsigned ipoptlen, optlen, hdrlen; 124 int idle, sendalot; 125 #if 0 126 int maxburst = TCP_MAXBURST; 127 #endif 128 struct rmxp_tao *taop; 129 #ifdef INET6 130 struct ip6_hdr *ip6 = NULL; 131 int isipv6; 132 133 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 134 #endif 135 136 #ifndef INET6 137 mtx_assert(&tp->t_inpcb->inp_mtx, MA_OWNED); 138 #endif 139 140 /* 141 * Determine length of data that should be transmitted, 142 * and flags that will be used. 143 * If there is some data or critical controls (SYN, RST) 144 * to send, then transmit; otherwise, investigate further. 145 */ 146 idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una); 147 if (idle && (ticks - tp->t_rcvtime) >= tp->t_rxtcur) { 148 /* 149 * We have been idle for "a while" and no acks are 150 * expected to clock out any data we send -- 151 * slow start to get ack "clock" running again. 152 * 153 * Set the slow-start flight size depending on whether 154 * this is a local network or not. 155 */ 156 int ss = ss_fltsz; 157 #ifdef INET6 158 if (isipv6) { 159 if (in6_localaddr(&tp->t_inpcb->in6p_faddr)) 160 ss = ss_fltsz_local; 161 } else 162 #endif /* INET6 */ 163 if (in_localaddr(tp->t_inpcb->inp_faddr)) 164 ss = ss_fltsz_local; 165 tp->snd_cwnd = tp->t_maxseg * ss; 166 } 167 tp->t_flags &= ~TF_LASTIDLE; 168 if (idle) { 169 if (tp->t_flags & TF_MORETOCOME) { 170 tp->t_flags |= TF_LASTIDLE; 171 idle = 0; 172 } 173 } 174 again: 175 sendalot = 0; 176 off = tp->snd_nxt - tp->snd_una; 177 win = min(tp->snd_wnd, tp->snd_cwnd); 178 win = min(win, tp->snd_bwnd); 179 180 flags = tcp_outflags[tp->t_state]; 181 /* 182 * Get standard flags, and add SYN or FIN if requested by 'hidden' 183 * state flags. 184 */ 185 if (tp->t_flags & TF_NEEDFIN) 186 flags |= TH_FIN; 187 if (tp->t_flags & TF_NEEDSYN) 188 flags |= TH_SYN; 189 190 /* 191 * If in persist timeout with window of 0, send 1 byte. 192 * Otherwise, if window is small but nonzero 193 * and timer expired, we will send what we can 194 * and go to transmit state. 195 */ 196 if (tp->t_force) { 197 if (win == 0) { 198 /* 199 * If we still have some data to send, then 200 * clear the FIN bit. Usually this would 201 * happen below when it realizes that we 202 * aren't sending all the data. However, 203 * if we have exactly 1 byte of unsent data, 204 * then it won't clear the FIN bit below, 205 * and if we are in persist state, we wind 206 * up sending the packet without recording 207 * that we sent the FIN bit. 208 * 209 * We can't just blindly clear the FIN bit, 210 * because if we don't have any more data 211 * to send then the probe will be the FIN 212 * itself. 213 */ 214 if (off < so->so_snd.sb_cc) 215 flags &= ~TH_FIN; 216 win = 1; 217 } else { 218 callout_stop(tp->tt_persist); 219 tp->t_rxtshift = 0; 220 } 221 } 222 223 /* 224 * If snd_nxt == snd_max and we have transmitted a FIN, the 225 * offset will be > 0 even if so_snd.sb_cc is 0, resulting in 226 * a negative length. This can also occur when tcp opens up 227 * its congestion window while receiving additional duplicate 228 * acks after fast-retransmit because TCP will reset snd_nxt 229 * to snd_max after the fast-retransmit. 230 * 231 * In the normal retransmit-FIN-only case, however, snd_nxt will 232 * be set to snd_una, the offset will be 0, and the length may 233 * wind up 0. 234 */ 235 len = (long)ulmin(so->so_snd.sb_cc, win) - off; 236 237 taop = tcp_gettaocache(&tp->t_inpcb->inp_inc); 238 239 /* 240 * Lop off SYN bit if it has already been sent. However, if this 241 * is SYN-SENT state and if segment contains data and if we don't 242 * know that foreign host supports TAO, suppress sending segment. 243 */ 244 if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) { 245 flags &= ~TH_SYN; 246 off--, len++; 247 if (len > 0 && tp->t_state == TCPS_SYN_SENT && 248 (taop == NULL || taop->tao_ccsent == 0)) 249 return 0; 250 } 251 252 /* 253 * Be careful not to send data and/or FIN on SYN segments 254 * in cases when no CC option will be sent. 255 * This measure is needed to prevent interoperability problems 256 * with not fully conformant TCP implementations. 257 */ 258 if ((flags & TH_SYN) && 259 ((tp->t_flags & TF_NOOPT) || !(tp->t_flags & TF_REQ_CC) || 260 ((flags & TH_ACK) && !(tp->t_flags & TF_RCVD_CC)))) { 261 len = 0; 262 flags &= ~TH_FIN; 263 } 264 265 if (len < 0) { 266 /* 267 * If FIN has been sent but not acked, 268 * but we haven't been called to retransmit, 269 * len will be < 0. Otherwise, window shrank 270 * after we sent into it. If window shrank to 0, 271 * cancel pending retransmit, pull snd_nxt back 272 * to (closed) window, and set the persist timer 273 * if it isn't already going. If the window didn't 274 * close completely, just wait for an ACK. 275 */ 276 len = 0; 277 if (win == 0) { 278 callout_stop(tp->tt_rexmt); 279 tp->t_rxtshift = 0; 280 tp->snd_nxt = tp->snd_una; 281 if (!callout_active(tp->tt_persist)) 282 tcp_setpersist(tp); 283 } 284 } 285 286 /* 287 * len will be >= 0 after this point. Truncate to the maximum 288 * segment length and ensure that FIN is removed if the length 289 * no longer contains the last data byte. 290 */ 291 if (len > tp->t_maxseg) { 292 len = tp->t_maxseg; 293 sendalot = 1; 294 } 295 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc)) 296 flags &= ~TH_FIN; 297 298 win = sbspace(&so->so_rcv); 299 300 /* 301 * Sender silly window avoidance. We transmit under the following 302 * conditions when len is non-zero: 303 * 304 * - We have a full segment 305 * - This is the last buffer in a write()/send() and we are 306 * either idle or running NODELAY 307 * - we've timed out (e.g. persist timer) 308 * - we have more then 1/2 the maximum send window's worth of 309 * data (receiver may be limited the window size) 310 * - we need to retransmit 311 */ 312 if (len) { 313 if (len == tp->t_maxseg) 314 goto send; 315 /* 316 * NOTE! on localhost connections an 'ack' from the remote 317 * end may occur synchronously with the output and cause 318 * us to flush a buffer queued with moretocome. XXX 319 * 320 * note: the len + off check is almost certainly unnecessary. 321 */ 322 if (!(tp->t_flags & TF_MORETOCOME) && /* normal case */ 323 (idle || (tp->t_flags & TF_NODELAY)) && 324 len + off >= so->so_snd.sb_cc && 325 (tp->t_flags & TF_NOPUSH) == 0) { 326 goto send; 327 } 328 if (tp->t_force) /* typ. timeout case */ 329 goto send; 330 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0) 331 goto send; 332 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */ 333 goto send; 334 } 335 336 /* 337 * Compare available window to amount of window 338 * known to peer (as advertised window less 339 * next expected input). If the difference is at least two 340 * max size segments, or at least 50% of the maximum possible 341 * window, then want to send a window update to peer. 342 */ 343 if (win > 0) { 344 /* 345 * "adv" is the amount we can increase the window, 346 * taking into account that we are limited by 347 * TCP_MAXWIN << tp->rcv_scale. 348 */ 349 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) - 350 (tp->rcv_adv - tp->rcv_nxt); 351 352 if (adv >= (long) (2 * tp->t_maxseg)) 353 goto send; 354 if (2 * adv >= (long) so->so_rcv.sb_hiwat) 355 goto send; 356 } 357 358 /* 359 * Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW 360 * is also a catch-all for the retransmit timer timeout case. 361 */ 362 if (tp->t_flags & TF_ACKNOW) 363 goto send; 364 if ((flags & TH_RST) || 365 ((flags & TH_SYN) && (tp->t_flags & TF_NEEDSYN) == 0)) 366 goto send; 367 if (SEQ_GT(tp->snd_up, tp->snd_una)) 368 goto send; 369 /* 370 * If our state indicates that FIN should be sent 371 * and we have not yet done so, then we need to send. 372 */ 373 if (flags & TH_FIN && 374 ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una)) 375 goto send; 376 377 /* 378 * TCP window updates are not reliable, rather a polling protocol 379 * using ``persist'' packets is used to insure receipt of window 380 * updates. The three ``states'' for the output side are: 381 * idle not doing retransmits or persists 382 * persisting to move a small or zero window 383 * (re)transmitting and thereby not persisting 384 * 385 * callout_active(tp->tt_persist) 386 * is true when we are in persist state. 387 * tp->t_force 388 * is set when we are called to send a persist packet. 389 * callout_active(tp->tt_rexmt) 390 * is set when we are retransmitting 391 * The output side is idle when both timers are zero. 392 * 393 * If send window is too small, there is data to transmit, and no 394 * retransmit or persist is pending, then go to persist state. 395 * If nothing happens soon, send when timer expires: 396 * if window is nonzero, transmit what we can, 397 * otherwise force out a byte. 398 */ 399 if (so->so_snd.sb_cc && !callout_active(tp->tt_rexmt) && 400 !callout_active(tp->tt_persist)) { 401 tp->t_rxtshift = 0; 402 tcp_setpersist(tp); 403 } 404 405 /* 406 * No reason to send a segment, just return. 407 */ 408 return (0); 409 410 send: 411 /* 412 * Before ESTABLISHED, force sending of initial options 413 * unless TCP set not to do any options. 414 * NOTE: we assume that the IP/TCP header plus TCP options 415 * always fit in a single mbuf, leaving room for a maximum 416 * link header, i.e. 417 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES 418 */ 419 optlen = 0; 420 #ifdef INET6 421 if (isipv6) 422 hdrlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 423 else 424 #endif 425 hdrlen = sizeof (struct tcpiphdr); 426 if (flags & TH_SYN) { 427 tp->snd_nxt = tp->iss; 428 if ((tp->t_flags & TF_NOOPT) == 0) { 429 u_short mss; 430 431 opt[0] = TCPOPT_MAXSEG; 432 opt[1] = TCPOLEN_MAXSEG; 433 mss = htons((u_short) tcp_mssopt(tp)); 434 (void)memcpy(opt + 2, &mss, sizeof(mss)); 435 optlen = TCPOLEN_MAXSEG; 436 437 if ((tp->t_flags & TF_REQ_SCALE) && 438 ((flags & TH_ACK) == 0 || 439 (tp->t_flags & TF_RCVD_SCALE))) { 440 *((u_int32_t *)(opt + optlen)) = htonl( 441 TCPOPT_NOP << 24 | 442 TCPOPT_WINDOW << 16 | 443 TCPOLEN_WINDOW << 8 | 444 tp->request_r_scale); 445 optlen += 4; 446 } 447 } 448 } 449 450 /* 451 * Send a timestamp and echo-reply if this is a SYN and our side 452 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side 453 * and our peer have sent timestamps in our SYN's. 454 */ 455 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 456 (flags & TH_RST) == 0 && 457 ((flags & TH_ACK) == 0 || 458 (tp->t_flags & TF_RCVD_TSTMP))) { 459 u_int32_t *lp = (u_int32_t *)(opt + optlen); 460 461 /* Form timestamp option as shown in appendix A of RFC 1323. */ 462 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 463 *lp++ = htonl(ticks); 464 *lp = htonl(tp->ts_recent); 465 optlen += TCPOLEN_TSTAMP_APPA; 466 } 467 468 /* 469 * Send `CC-family' options if our side wants to use them (TF_REQ_CC), 470 * options are allowed (!TF_NOOPT) and it's not a RST. 471 */ 472 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 473 (flags & TH_RST) == 0) { 474 switch (flags & (TH_SYN|TH_ACK)) { 475 /* 476 * This is a normal ACK, send CC if we received CC before 477 * from our peer. 478 */ 479 case TH_ACK: 480 if (!(tp->t_flags & TF_RCVD_CC)) 481 break; 482 /*FALLTHROUGH*/ 483 484 /* 485 * We can only get here in T/TCP's SYN_SENT* state, when 486 * we're a sending a non-SYN segment without waiting for 487 * the ACK of our SYN. A check above assures that we only 488 * do this if our peer understands T/TCP. 489 */ 490 case 0: 491 opt[optlen++] = TCPOPT_NOP; 492 opt[optlen++] = TCPOPT_NOP; 493 opt[optlen++] = TCPOPT_CC; 494 opt[optlen++] = TCPOLEN_CC; 495 *(u_int32_t *)&opt[optlen] = htonl(tp->cc_send); 496 497 optlen += 4; 498 break; 499 500 /* 501 * This is our initial SYN, check whether we have to use 502 * CC or CC.new. 503 */ 504 case TH_SYN: 505 opt[optlen++] = TCPOPT_NOP; 506 opt[optlen++] = TCPOPT_NOP; 507 opt[optlen++] = tp->t_flags & TF_SENDCCNEW ? 508 TCPOPT_CCNEW : TCPOPT_CC; 509 opt[optlen++] = TCPOLEN_CC; 510 *(u_int32_t *)&opt[optlen] = htonl(tp->cc_send); 511 optlen += 4; 512 break; 513 514 /* 515 * This is a SYN,ACK; send CC and CC.echo if we received 516 * CC from our peer. 517 */ 518 case (TH_SYN|TH_ACK): 519 if (tp->t_flags & TF_RCVD_CC) { 520 opt[optlen++] = TCPOPT_NOP; 521 opt[optlen++] = TCPOPT_NOP; 522 opt[optlen++] = TCPOPT_CC; 523 opt[optlen++] = TCPOLEN_CC; 524 *(u_int32_t *)&opt[optlen] = 525 htonl(tp->cc_send); 526 optlen += 4; 527 opt[optlen++] = TCPOPT_NOP; 528 opt[optlen++] = TCPOPT_NOP; 529 opt[optlen++] = TCPOPT_CCECHO; 530 opt[optlen++] = TCPOLEN_CC; 531 *(u_int32_t *)&opt[optlen] = 532 htonl(tp->cc_recv); 533 optlen += 4; 534 } 535 break; 536 } 537 } 538 539 hdrlen += optlen; 540 541 #ifdef INET6 542 if (isipv6) 543 ipoptlen = ip6_optlen(tp->t_inpcb); 544 else 545 #endif 546 if (tp->t_inpcb->inp_options) 547 ipoptlen = tp->t_inpcb->inp_options->m_len - 548 offsetof(struct ipoption, ipopt_list); 549 else 550 ipoptlen = 0; 551 #ifdef IPSEC 552 ipoptlen += ipsec_hdrsiz_tcp(tp); 553 #endif 554 555 /* 556 * Adjust data length if insertion of options will 557 * bump the packet length beyond the t_maxopd length. 558 * Clear the FIN bit because we cut off the tail of 559 * the segment. 560 */ 561 if (len + optlen + ipoptlen > tp->t_maxopd) { 562 /* 563 * If there is still more to send, don't close the connection. 564 */ 565 flags &= ~TH_FIN; 566 len = tp->t_maxopd - optlen - ipoptlen; 567 sendalot = 1; 568 } 569 570 /*#ifdef DIAGNOSTIC*/ 571 #ifdef INET6 572 if (max_linkhdr + hdrlen > MCLBYTES) 573 #else 574 if (max_linkhdr + hdrlen > MHLEN) 575 #endif 576 panic("tcphdr too big"); 577 /*#endif*/ 578 579 /* 580 * Grab a header mbuf, attaching a copy of data to 581 * be transmitted, and initialize the header from 582 * the template for sends on this connection. 583 */ 584 if (len) { 585 if (tp->t_force && len == 1) 586 tcpstat.tcps_sndprobe++; 587 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { 588 tcpstat.tcps_sndrexmitpack++; 589 tcpstat.tcps_sndrexmitbyte += len; 590 } else { 591 tcpstat.tcps_sndpack++; 592 tcpstat.tcps_sndbyte += len; 593 } 594 #ifdef notyet 595 if ((m = m_copypack(so->so_snd.sb_mb, off, 596 (int)len, max_linkhdr + hdrlen)) == 0) { 597 error = ENOBUFS; 598 goto out; 599 } 600 /* 601 * m_copypack left space for our hdr; use it. 602 */ 603 m->m_len += hdrlen; 604 m->m_data -= hdrlen; 605 #else 606 MGETHDR(m, M_NOWAIT, MT_HEADER); 607 if (m == NULL) { 608 error = ENOBUFS; 609 goto out; 610 } 611 #ifdef INET6 612 if (MHLEN < hdrlen + max_linkhdr) { 613 MCLGET(m, M_NOWAIT); 614 if ((m->m_flags & M_EXT) == 0) { 615 m_freem(m); 616 error = ENOBUFS; 617 goto out; 618 } 619 } 620 #endif 621 m->m_data += max_linkhdr; 622 m->m_len = hdrlen; 623 if (len <= MHLEN - hdrlen - max_linkhdr) { 624 m_copydata(so->so_snd.sb_mb, off, (int) len, 625 mtod(m, caddr_t) + hdrlen); 626 m->m_len += len; 627 } else { 628 m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len); 629 if (m->m_next == 0) { 630 (void) m_free(m); 631 error = ENOBUFS; 632 goto out; 633 } 634 } 635 #endif 636 /* 637 * If we're sending everything we've got, set PUSH. 638 * (This will keep happy those implementations which only 639 * give data to the user when a buffer fills or 640 * a PUSH comes in.) 641 */ 642 if (off + len == so->so_snd.sb_cc) 643 flags |= TH_PUSH; 644 } else { 645 if (tp->t_flags & TF_ACKNOW) 646 tcpstat.tcps_sndacks++; 647 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 648 tcpstat.tcps_sndctrl++; 649 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 650 tcpstat.tcps_sndurg++; 651 else 652 tcpstat.tcps_sndwinup++; 653 654 MGETHDR(m, M_NOWAIT, MT_HEADER); 655 if (m == NULL) { 656 error = ENOBUFS; 657 goto out; 658 } 659 #ifdef INET6 660 if (isipv6 && (MHLEN < hdrlen + max_linkhdr) && 661 MHLEN >= hdrlen) { 662 MH_ALIGN(m, hdrlen); 663 } else 664 #endif 665 m->m_data += max_linkhdr; 666 m->m_len = hdrlen; 667 } 668 m->m_pkthdr.rcvif = (struct ifnet *)0; 669 #ifdef MAC 670 mac_create_mbuf_from_socket(so, m); 671 #endif 672 #ifdef INET6 673 if (isipv6) { 674 ip6 = mtod(m, struct ip6_hdr *); 675 th = (struct tcphdr *)(ip6 + 1); 676 tcp_fillheaders(tp, ip6, th); 677 } else 678 #endif /* INET6 */ 679 { 680 ip = mtod(m, struct ip *); 681 ipov = (struct ipovly *)ip; 682 th = (struct tcphdr *)(ip + 1); 683 /* this picks up the pseudo header (w/o the length) */ 684 tcp_fillheaders(tp, ip, th); 685 } 686 687 /* 688 * Fill in fields, remembering maximum advertised 689 * window for use in delaying messages about window sizes. 690 * If resending a FIN, be sure not to use a new sequence number. 691 */ 692 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN && 693 tp->snd_nxt == tp->snd_max) 694 tp->snd_nxt--; 695 /* 696 * If we are doing retransmissions, then snd_nxt will 697 * not reflect the first unsent octet. For ACK only 698 * packets, we do not want the sequence number of the 699 * retransmitted packet, we want the sequence number 700 * of the next unsent octet. So, if there is no data 701 * (and no SYN or FIN), use snd_max instead of snd_nxt 702 * when filling in ti_seq. But if we are in persist 703 * state, snd_max might reflect one byte beyond the 704 * right edge of the window, so use snd_nxt in that 705 * case, since we know we aren't doing a retransmission. 706 * (retransmit and persist are mutually exclusive...) 707 */ 708 if (len || (flags & (TH_SYN|TH_FIN)) 709 || callout_active(tp->tt_persist)) 710 th->th_seq = htonl(tp->snd_nxt); 711 else 712 th->th_seq = htonl(tp->snd_max); 713 th->th_ack = htonl(tp->rcv_nxt); 714 if (optlen) { 715 bcopy(opt, th + 1, optlen); 716 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 717 } 718 th->th_flags = flags; 719 /* 720 * Calculate receive window. Don't shrink window, 721 * but avoid silly window syndrome. 722 */ 723 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)tp->t_maxseg) 724 win = 0; 725 if (win < (long)(tp->rcv_adv - tp->rcv_nxt)) 726 win = (long)(tp->rcv_adv - tp->rcv_nxt); 727 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 728 win = (long)TCP_MAXWIN << tp->rcv_scale; 729 th->th_win = htons((u_short) (win>>tp->rcv_scale)); 730 731 732 /* 733 * Adjust the RXWIN0SENT flag - indicate that we have advertised 734 * a 0 window. This may cause the remote transmitter to stall. This 735 * flag tells soreceive() to disable delayed acknowledgements when 736 * draining the buffer. This can occur if the receiver is attempting 737 * to read more data then can be buffered prior to transmitting on 738 * the connection. 739 */ 740 if (win == 0) 741 tp->t_flags |= TF_RXWIN0SENT; 742 else 743 tp->t_flags &= ~TF_RXWIN0SENT; 744 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 745 th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt)); 746 th->th_flags |= TH_URG; 747 } else 748 /* 749 * If no urgent pointer to send, then we pull 750 * the urgent pointer to the left edge of the send window 751 * so that it doesn't drift into the send window on sequence 752 * number wraparound. 753 */ 754 tp->snd_up = tp->snd_una; /* drag it along */ 755 756 /* 757 * Put TCP length in extended header, and then 758 * checksum extended header and data. 759 */ 760 m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */ 761 #ifdef INET6 762 if (isipv6) 763 /* 764 * ip6_plen is not need to be filled now, and will be filled 765 * in ip6_output. 766 */ 767 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 768 sizeof(struct tcphdr) + optlen + len); 769 else 770 #endif /* INET6 */ 771 { 772 m->m_pkthdr.csum_flags = CSUM_TCP; 773 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 774 if (len + optlen) 775 th->th_sum = in_addword(th->th_sum, 776 htons((u_short)(optlen + len))); 777 778 /* IP version must be set here for ipv4/ipv6 checking later */ 779 KASSERT(ip->ip_v == IPVERSION, 780 ("%s: IP version incorrect: %d", __func__, ip->ip_v)); 781 } 782 783 /* 784 * In transmit state, time the transmission and arrange for 785 * the retransmit. In persist state, just set snd_max. 786 */ 787 if (tp->t_force == 0 || !callout_active(tp->tt_persist)) { 788 tcp_seq startseq = tp->snd_nxt; 789 790 /* 791 * Advance snd_nxt over sequence space of this segment. 792 */ 793 if (flags & (TH_SYN|TH_FIN)) { 794 if (flags & TH_SYN) 795 tp->snd_nxt++; 796 if (flags & TH_FIN) { 797 tp->snd_nxt++; 798 tp->t_flags |= TF_SENTFIN; 799 } 800 } 801 tp->snd_nxt += len; 802 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 803 tp->snd_max = tp->snd_nxt; 804 /* 805 * Time this transmission if not a retransmission and 806 * not currently timing anything. 807 */ 808 if (tp->t_rtttime == 0) { 809 tp->t_rtttime = ticks; 810 tp->t_rtseq = startseq; 811 tcpstat.tcps_segstimed++; 812 } 813 } 814 815 /* 816 * Set retransmit timer if not currently set, 817 * and not doing a pure ack or a keep-alive probe. 818 * Initial value for retransmit timer is smoothed 819 * round-trip time + 2 * round-trip time variance. 820 * Initialize shift counter which is used for backoff 821 * of retransmit time. 822 */ 823 if (!callout_active(tp->tt_rexmt) && 824 tp->snd_nxt != tp->snd_una) { 825 if (callout_active(tp->tt_persist)) { 826 callout_stop(tp->tt_persist); 827 tp->t_rxtshift = 0; 828 } 829 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 830 tcp_timer_rexmt, tp); 831 } 832 } else { 833 /* 834 * Persist case, update snd_max but since we are in 835 * persist mode (no window) we do not update snd_nxt. 836 */ 837 int xlen = len; 838 if (flags & TH_SYN) 839 ++xlen; 840 if (flags & TH_FIN) { 841 ++xlen; 842 tp->t_flags |= TF_SENTFIN; 843 } 844 if (SEQ_GT(tp->snd_nxt + xlen, tp->snd_max)) 845 tp->snd_max = tp->snd_nxt + len; 846 } 847 848 #ifdef TCPDEBUG 849 /* 850 * Trace. 851 */ 852 if (so->so_options & SO_DEBUG) 853 tcp_trace(TA_OUTPUT, tp->t_state, tp, mtod(m, void *), th, 0); 854 #endif 855 856 /* 857 * Fill in IP length and desired time to live and 858 * send to IP level. There should be a better way 859 * to handle ttl and tos; we could keep them in 860 * the template, but need a way to checksum without them. 861 */ 862 /* 863 * m->m_pkthdr.len should have been set before cksum calcuration, 864 * because in6_cksum() need it. 865 */ 866 #ifdef INET6 867 if (isipv6) { 868 /* 869 * we separately set hoplimit for every segment, since the 870 * user might want to change the value via setsockopt. 871 * Also, desired default hop limit might be changed via 872 * Neighbor Discovery. 873 */ 874 ip6->ip6_hlim = in6_selecthlim(tp->t_inpcb, 875 tp->t_inpcb->in6p_route.ro_rt ? 876 tp->t_inpcb->in6p_route.ro_rt->rt_ifp 877 : NULL); 878 879 /* TODO: IPv6 IP6TOS_ECT bit on */ 880 error = ip6_output(m, 881 tp->t_inpcb->in6p_outputopts, 882 &tp->t_inpcb->in6p_route, 883 (so->so_options & SO_DONTROUTE), NULL, NULL, 884 tp->t_inpcb); 885 } else 886 #endif /* INET6 */ 887 { 888 struct rtentry *rt; 889 ip->ip_len = m->m_pkthdr.len; 890 #ifdef INET6 891 if (INP_CHECK_SOCKAF(so, AF_INET6)) 892 ip->ip_ttl = in6_selecthlim(tp->t_inpcb, 893 tp->t_inpcb->in6p_route.ro_rt ? 894 tp->t_inpcb->in6p_route.ro_rt->rt_ifp 895 : NULL); 896 else 897 #endif /* INET6 */ 898 ip->ip_ttl = tp->t_inpcb->inp_ip_ttl; /* XXX */ 899 ip->ip_tos = tp->t_inpcb->inp_ip_tos; /* XXX */ 900 /* 901 * See if we should do MTU discovery. We do it only if the following 902 * are true: 903 * 1) we have a valid route to the destination 904 * 2) the MTU is not locked (if it is, then discovery has been 905 * disabled) 906 */ 907 if (path_mtu_discovery 908 && (rt = tp->t_inpcb->inp_route.ro_rt) 909 && rt->rt_flags & RTF_UP 910 && !(rt->rt_rmx.rmx_locks & RTV_MTU)) { 911 ip->ip_off |= IP_DF; 912 } 913 error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route, 914 (so->so_options & SO_DONTROUTE), 0, tp->t_inpcb); 915 } 916 if (error) { 917 918 /* 919 * We know that the packet was lost, so back out the 920 * sequence number advance, if any. 921 */ 922 if (tp->t_force == 0 || !callout_active(tp->tt_persist)) { 923 /* 924 * No need to check for TH_FIN here because 925 * the TF_SENTFIN flag handles that case. 926 */ 927 if ((flags & TH_SYN) == 0) 928 tp->snd_nxt -= len; 929 } 930 931 out: 932 if (error == ENOBUFS) { 933 if (!callout_active(tp->tt_rexmt) && 934 !callout_active(tp->tt_persist)) 935 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 936 tcp_timer_rexmt, tp); 937 tcp_quench(tp->t_inpcb, 0); 938 return (0); 939 } 940 if (error == EMSGSIZE) { 941 /* 942 * ip_output() will have already fixed the route 943 * for us. tcp_mtudisc() will, as its last action, 944 * initiate retransmission, so it is important to 945 * not do so here. 946 */ 947 tcp_mtudisc(tp->t_inpcb, 0); 948 return 0; 949 } 950 if ((error == EHOSTUNREACH || error == ENETDOWN) 951 && TCPS_HAVERCVDSYN(tp->t_state)) { 952 tp->t_softerror = error; 953 return (0); 954 } 955 return (error); 956 } 957 tcpstat.tcps_sndtotal++; 958 959 /* 960 * Data sent (as far as we can tell). 961 * If this advertises a larger window than any other segment, 962 * then remember the size of the advertised window. 963 * Any pending ACK has now been sent. 964 */ 965 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 966 tp->rcv_adv = tp->rcv_nxt + win; 967 tp->last_ack_sent = tp->rcv_nxt; 968 tp->t_flags &= ~TF_ACKNOW; 969 if (tcp_delack_enabled) 970 callout_stop(tp->tt_delack); 971 #if 0 972 /* 973 * This completely breaks TCP if newreno is turned on. What happens 974 * is that if delayed-acks are turned on on the receiver, this code 975 * on the transmitter effectively destroys the TCP window, forcing 976 * it to four packets (1.5Kx4 = 6K window). 977 */ 978 if (sendalot && (!tcp_do_newreno || --maxburst)) 979 goto again; 980 #endif 981 if (sendalot) 982 goto again; 983 return (0); 984 } 985 986 void 987 tcp_setpersist(tp) 988 register struct tcpcb *tp; 989 { 990 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1; 991 int tt; 992 993 if (callout_active(tp->tt_rexmt)) 994 panic("tcp_setpersist: retransmit pending"); 995 /* 996 * Start/restart persistance timer. 997 */ 998 TCPT_RANGESET(tt, t * tcp_backoff[tp->t_rxtshift], 999 TCPTV_PERSMIN, TCPTV_PERSMAX); 1000 callout_reset(tp->tt_persist, tt, tcp_timer_persist, tp); 1001 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 1002 tp->t_rxtshift++; 1003 } 1004