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 1.20 2004/11/14 00:49:08 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 = 1; 140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, path_mtu_discovery, CTLFLAG_RW, 141 &path_mtu_discovery, 1, "Enable Path MTU Discovery"); 142 143 int ss_fltsz = 1; 144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, slowstart_flightsize, CTLFLAG_RW, 145 &ss_fltsz, 1, "Slow start flight size"); 146 147 int ss_fltsz_local = 4; 148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, local_slowstart_flightsize, CTLFLAG_RW, 149 &ss_fltsz_local, 1, "Slow start flight size for local networks"); 150 151 static int avoid_pure_win_update = 1; 152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, avoid_pure_win_update, CTLFLAG_RW, 153 &avoid_pure_win_update, 1, "Avoid pure window updates when possible"); 154 155 /* 156 * Tcp output routine: figure out what should be sent and send it. 157 */ 158 int 159 tcp_output(tp) 160 struct tcpcb *tp; 161 { 162 struct inpcb * const inp = tp->t_inpcb; 163 struct socket *so = inp->inp_socket; 164 long len, recvwin, sendwin; 165 int off, flags, error; 166 struct mbuf *m; 167 struct ip *ip = NULL; 168 struct ipovly *ipov = NULL; 169 struct tcphdr *th; 170 u_char opt[TCP_MAXOLEN]; 171 unsigned int ipoptlen, optlen, hdrlen; 172 int idle; 173 boolean_t sendalot; 174 struct ip6_hdr *ip6 = NULL; 175 #ifdef INET6 176 const boolean_t isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 177 #else 178 const boolean_t isipv6 = FALSE; 179 #endif 180 struct rmxp_tao *taop; 181 182 /* 183 * Determine length of data that should be transmitted, 184 * and flags that will be used. 185 * If there is some data or critical controls (SYN, RST) 186 * to send, then transmit; otherwise, investigate further. 187 */ 188 if ((tp->snd_max == tp->snd_una) && 189 (ticks - tp->t_rcvtime) >= tp->t_rxtcur) { 190 /* 191 * We have been idle for "a while" and no acks are 192 * expected to clock out any data we send -- 193 * slow start to get ack "clock" running again. 194 * 195 * Set the slow-start flight size depending on whether 196 * this is a local network or not. 197 */ 198 if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 199 (!isipv6 && in_localaddr(inp->inp_faddr))) 200 tp->snd_cwnd = tp->t_maxseg * ss_fltsz_local; 201 else 202 tp->snd_cwnd = tp->t_maxseg * ss_fltsz; 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 again: 211 /* Make use of SACK information when slow-starting after a RTO. */ 212 if (TCP_DO_SACK(tp) && tp->snd_nxt != tp->snd_max && 213 !IN_FASTRECOVERY(tp)) 214 tcp_sack_skip_sacked(&tp->scb, &tp->snd_nxt); 215 216 sendalot = FALSE; 217 off = tp->snd_nxt - tp->snd_una; 218 sendwin = min(tp->snd_wnd, tp->snd_cwnd); 219 sendwin = min(sendwin, tp->snd_bwnd); 220 221 flags = tcp_outflags[tp->t_state]; 222 /* 223 * Get standard flags, and add SYN or FIN if requested by 'hidden' 224 * state flags. 225 */ 226 if (tp->t_flags & TF_NEEDFIN) 227 flags |= TH_FIN; 228 if (tp->t_flags & TF_NEEDSYN) 229 flags |= TH_SYN; 230 231 /* 232 * If in persist timeout with window of 0, send 1 byte. 233 * Otherwise, if window is small but nonzero 234 * and timer expired, we will send what we can 235 * and go to transmit state. 236 */ 237 if (tp->t_flags & TF_FORCE) { 238 if (sendwin == 0) { 239 /* 240 * If we still have some data to send, then 241 * clear the FIN bit. Usually this would 242 * happen below when it realizes that we 243 * aren't sending all the data. However, 244 * if we have exactly 1 byte of unsent data, 245 * then it won't clear the FIN bit below, 246 * and if we are in persist state, we wind 247 * up sending the packet without recording 248 * that we sent the FIN bit. 249 * 250 * We can't just blindly clear the FIN bit, 251 * because if we don't have any more data 252 * to send then the probe will be the FIN 253 * itself. 254 */ 255 if (off < so->so_snd.sb_cc) 256 flags &= ~TH_FIN; 257 sendwin = 1; 258 } else { 259 callout_stop(tp->tt_persist); 260 tp->t_rxtshift = 0; 261 } 262 } 263 264 /* 265 * If snd_nxt == snd_max and we have transmitted a FIN, the 266 * offset will be > 0 even if so_snd.sb_cc is 0, resulting in 267 * a negative length. This can also occur when TCP opens up 268 * its congestion window while receiving additional duplicate 269 * acks after fast-retransmit because TCP will reset snd_nxt 270 * to snd_max after the fast-retransmit. 271 * 272 * In the normal retransmit-FIN-only case, however, snd_nxt will 273 * be set to snd_una, the offset will be 0, and the length may 274 * wind up 0. 275 */ 276 len = (long)ulmin(so->so_snd.sb_cc, sendwin) - off; 277 278 /* 279 * Lop off SYN bit if it has already been sent. However, if this 280 * is SYN-SENT state and if segment contains data and if we don't 281 * know that foreign host supports TAO, suppress sending segment. 282 */ 283 if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) { 284 flags &= ~TH_SYN; 285 off--, len++; 286 if (len > 0 && tp->t_state == TCPS_SYN_SENT && 287 ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL || 288 taop->tao_ccsent == 0)) 289 return 0; 290 } 291 292 /* 293 * Be careful not to send data and/or FIN on SYN segments 294 * in cases when no CC option will be sent. 295 * This measure is needed to prevent interoperability problems 296 * with not fully conformant TCP implementations. 297 */ 298 if ((flags & TH_SYN) && 299 ((tp->t_flags & TF_NOOPT) || !(tp->t_flags & TF_REQ_CC) || 300 ((flags & TH_ACK) && !(tp->t_flags & TF_RCVD_CC)))) { 301 len = 0; 302 flags &= ~TH_FIN; 303 } 304 305 if (len < 0) { 306 /* 307 * If FIN has been sent but not acked, 308 * but we haven't been called to retransmit, 309 * len will be < 0. Otherwise, window shrank 310 * after we sent into it. If window shrank to 0, 311 * cancel pending retransmit, pull snd_nxt back 312 * to (closed) window, and set the persist timer 313 * if it isn't already going. If the window didn't 314 * close completely, just wait for an ACK. 315 */ 316 len = 0; 317 if (sendwin == 0) { 318 callout_stop(tp->tt_rexmt); 319 tp->t_rxtshift = 0; 320 tp->snd_nxt = tp->snd_una; 321 if (!callout_active(tp->tt_persist)) 322 tcp_setpersist(tp); 323 } 324 } 325 326 /* 327 * len will be >= 0 after this point. Truncate to the maximum 328 * segment length and ensure that FIN is removed if the length 329 * no longer contains the last data byte. 330 */ 331 if (len > tp->t_maxseg) { 332 len = tp->t_maxseg; 333 sendalot = TRUE; 334 } 335 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc)) 336 flags &= ~TH_FIN; 337 338 recvwin = sbspace(&so->so_rcv); 339 340 /* 341 * Sender silly window avoidance. We transmit under the following 342 * conditions when len is non-zero: 343 * 344 * - We have a full segment 345 * - This is the last buffer in a write()/send() and we are 346 * either idle or running NODELAY 347 * - we've timed out (e.g. persist timer) 348 * - we have more then 1/2 the maximum send window's worth of 349 * data (receiver may be limited the window size) 350 * - we need to retransmit 351 */ 352 if (len) { 353 if (len == tp->t_maxseg) 354 goto send; 355 /* 356 * NOTE! on localhost connections an 'ack' from the remote 357 * end may occur synchronously with the output and cause 358 * us to flush a buffer queued with moretocome. XXX 359 * 360 * note: the len + off check is almost certainly unnecessary. 361 */ 362 if (!(tp->t_flags & TF_MORETOCOME) && /* normal case */ 363 (idle || (tp->t_flags & TF_NODELAY)) && 364 len + off >= so->so_snd.sb_cc && 365 !(tp->t_flags & TF_NOPUSH)) { 366 goto send; 367 } 368 if (tp->t_flags & TF_FORCE) /* typ. timeout case */ 369 goto send; 370 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0) 371 goto send; 372 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */ 373 goto send; 374 } 375 376 /* 377 * Compare available window to amount of window 378 * known to peer (as advertised window less 379 * next expected input). If the difference is at least two 380 * max size segments, or at least 50% of the maximum possible 381 * window, then want to send a window update to peer. 382 */ 383 if (recvwin > 0) { 384 /* 385 * "adv" is the amount we can increase the window, 386 * taking into account that we are limited by 387 * TCP_MAXWIN << tp->rcv_scale. 388 */ 389 long adv = min(recvwin, (long)TCP_MAXWIN << tp->rcv_scale) - 390 (tp->rcv_adv - tp->rcv_nxt); 391 392 /* 393 * This ack case typically occurs when the user has drained 394 * the TCP socket buffer sufficiently to warrent an ack 395 * containing a 'pure window update'... that is, an ack that 396 * ONLY updates the tcp window. 397 * 398 * It is unclear why we would need to do a pure window update 399 * past 2 segments if we are going to do one at 1/2 the high 400 * water mark anyway, especially since under normal conditions 401 * the user program will drain the socket buffer quickly. 402 * The 2-segment pure window update will often add a large 403 * number of extra, unnecessary acks to the stream. 404 * 405 * avoid_pure_win_update now defaults to 1. 406 */ 407 if (avoid_pure_win_update == 0) { 408 if (adv >= (long) (2 * tp->t_maxseg)) 409 goto send; 410 } 411 if (2 * adv >= (long) so->so_rcv.sb_hiwat) 412 goto send; 413 } 414 415 /* 416 * Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW 417 * is also a catch-all for the retransmit timer timeout case. 418 */ 419 if (tp->t_flags & TF_ACKNOW) 420 goto send; 421 if ((flags & TH_RST) || 422 ((flags & TH_SYN) && !(tp->t_flags & TF_NEEDSYN))) 423 goto send; 424 if (SEQ_GT(tp->snd_up, tp->snd_una)) 425 goto send; 426 /* 427 * If our state indicates that FIN should be sent 428 * and we have not yet done so, then we need to send. 429 */ 430 if (flags & TH_FIN && 431 (!(tp->t_flags & TF_SENTFIN) || tp->snd_nxt == tp->snd_una)) 432 goto send; 433 434 /* 435 * TCP window updates are not reliable, rather a polling protocol 436 * using ``persist'' packets is used to insure receipt of window 437 * updates. The three ``states'' for the output side are: 438 * idle not doing retransmits or persists 439 * persisting to move a small or zero window 440 * (re)transmitting and thereby not persisting 441 * 442 * callout_active(tp->tt_persist) 443 * is true when we are in persist state. 444 * The TF_FORCE flag in tp->t_flags 445 * is set when we are called to send a persist packet. 446 * callout_active(tp->tt_rexmt) 447 * is set when we are retransmitting 448 * The output side is idle when both timers are zero. 449 * 450 * If send window is too small, there is data to transmit, and no 451 * retransmit or persist is pending, then go to persist state. 452 * If nothing happens soon, send when timer expires: 453 * if window is nonzero, transmit what we can, 454 * otherwise force out a byte. 455 */ 456 if (so->so_snd.sb_cc > 0 && 457 !callout_active(tp->tt_rexmt) && !callout_active(tp->tt_persist)) { 458 tp->t_rxtshift = 0; 459 tcp_setpersist(tp); 460 } 461 462 /* 463 * No reason to send a segment, just return. 464 */ 465 return (0); 466 467 send: 468 /* 469 * Before ESTABLISHED, force sending of initial options 470 * unless TCP set not to do any options. 471 * NOTE: we assume that the IP/TCP header plus TCP options 472 * always fit in a single mbuf, leaving room for a maximum 473 * link header, i.e. 474 * max_linkhdr + sizeof(struct tcpiphdr) + optlen <= MCLBYTES 475 */ 476 optlen = 0; 477 if (isipv6) 478 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 479 else 480 hdrlen = sizeof(struct tcpiphdr); 481 if (flags & TH_SYN) { 482 tp->snd_nxt = tp->iss; 483 if (!(tp->t_flags & TF_NOOPT)) { 484 u_short mss; 485 486 opt[0] = TCPOPT_MAXSEG; 487 opt[1] = TCPOLEN_MAXSEG; 488 mss = htons((u_short) tcp_mssopt(tp)); 489 (void)memcpy(opt + 2, &mss, sizeof(mss)); 490 optlen = TCPOLEN_MAXSEG; 491 492 if ((tp->t_flags & TF_REQ_SCALE) && 493 (!(flags & TH_ACK) || 494 (tp->t_flags & TF_RCVD_SCALE))) { 495 *((u_int32_t *)(opt + optlen)) = htonl( 496 TCPOPT_NOP << 24 | 497 TCPOPT_WINDOW << 16 | 498 TCPOLEN_WINDOW << 8 | 499 tp->request_r_scale); 500 optlen += 4; 501 } 502 503 if ((tcp_do_sack && !(flags & TH_ACK)) || 504 tp->t_flags & TF_SACK_PERMITTED) { 505 uint32_t *lp = (uint32_t *)(opt + optlen); 506 507 *lp = htonl(TCPOPT_SACK_PERMITTED_ALIGNED); 508 optlen += TCPOLEN_SACK_PERMITTED_ALIGNED; 509 } 510 } 511 } 512 513 /* 514 * Send a timestamp and echo-reply if this is a SYN and our side 515 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side 516 * and our peer have sent timestamps in our SYN's. 517 */ 518 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP && 519 !(flags & TH_RST) && 520 (!(flags & TH_ACK) || (tp->t_flags & TF_RCVD_TSTMP))) { 521 u_int32_t *lp = (u_int32_t *)(opt + optlen); 522 523 /* Form timestamp option as shown in appendix A of RFC 1323. */ 524 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 525 *lp++ = htonl(ticks); 526 *lp = htonl(tp->ts_recent); 527 optlen += TCPOLEN_TSTAMP_APPA; 528 } 529 530 /* 531 * Send `CC-family' options if our side wants to use them (TF_REQ_CC), 532 * options are allowed (!TF_NOOPT) and it's not a RST. 533 */ 534 if ((tp->t_flags & (TF_REQ_CC | TF_NOOPT)) == TF_REQ_CC && 535 !(flags & TH_RST)) { 536 switch (flags & (TH_SYN | TH_ACK)) { 537 /* 538 * This is a normal ACK, send CC if we received CC before 539 * from our peer. 540 */ 541 case TH_ACK: 542 if (!(tp->t_flags & TF_RCVD_CC)) 543 break; 544 /*FALLTHROUGH*/ 545 546 /* 547 * We can only get here in T/TCP's SYN_SENT* state, when 548 * we're a sending a non-SYN segment without waiting for 549 * the ACK of our SYN. A check above assures that we only 550 * do this if our peer understands T/TCP. 551 */ 552 case 0: 553 opt[optlen++] = TCPOPT_NOP; 554 opt[optlen++] = TCPOPT_NOP; 555 opt[optlen++] = TCPOPT_CC; 556 opt[optlen++] = TCPOLEN_CC; 557 *(u_int32_t *)&opt[optlen] = htonl(tp->cc_send); 558 optlen += 4; 559 break; 560 561 /* 562 * This is our initial SYN, check whether we have to use 563 * CC or CC.new. 564 */ 565 case TH_SYN: 566 opt[optlen++] = TCPOPT_NOP; 567 opt[optlen++] = TCPOPT_NOP; 568 opt[optlen++] = tp->t_flags & TF_SENDCCNEW ? 569 TCPOPT_CCNEW : TCPOPT_CC; 570 opt[optlen++] = TCPOLEN_CC; 571 *(u_int32_t *)&opt[optlen] = htonl(tp->cc_send); 572 optlen += 4; 573 break; 574 575 /* 576 * This is a SYN,ACK; send CC and CC.echo if we received 577 * CC from our peer. 578 */ 579 case (TH_SYN | TH_ACK): 580 if (tp->t_flags & TF_RCVD_CC) { 581 opt[optlen++] = TCPOPT_NOP; 582 opt[optlen++] = TCPOPT_NOP; 583 opt[optlen++] = TCPOPT_CC; 584 opt[optlen++] = TCPOLEN_CC; 585 *(u_int32_t *)&opt[optlen] = htonl(tp->cc_send); 586 optlen += 4; 587 opt[optlen++] = TCPOPT_NOP; 588 opt[optlen++] = TCPOPT_NOP; 589 opt[optlen++] = TCPOPT_CCECHO; 590 opt[optlen++] = TCPOLEN_CC; 591 *(u_int32_t *)&opt[optlen] = htonl(tp->cc_recv); 592 optlen += 4; 593 } 594 break; 595 } 596 } 597 598 /* 599 * If this is a SACK connection and we have a block to report, 600 * fill in the SACK blocks in the TCP options. 601 */ 602 if ((tp->t_flags & (TF_SACK_PERMITTED | TF_NOOPT)) == 603 TF_SACK_PERMITTED && 604 (!LIST_EMPTY(&tp->t_segq) || 605 tp->reportblk.rblk_start != tp->reportblk.rblk_end)) 606 tcp_sack_fill_report(tp, opt, &optlen); 607 608 KASSERT(optlen <= TCP_MAXOLEN, ("too many TCP options")); 609 hdrlen += optlen; 610 611 if (isipv6) { 612 ipoptlen = ip6_optlen(inp); 613 } else { 614 if (inp->inp_options) { 615 ipoptlen = inp->inp_options->m_len - 616 offsetof(struct ipoption, ipopt_list); 617 } else { 618 ipoptlen = 0; 619 } 620 } 621 #ifdef IPSEC 622 ipoptlen += ipsec_hdrsiz_tcp(tp); 623 #endif 624 625 /* 626 * Adjust data length if insertion of options will 627 * bump the packet length beyond the t_maxopd length. 628 * Clear the FIN bit because we cut off the tail of 629 * the segment. 630 */ 631 if (len + optlen + ipoptlen > tp->t_maxopd) { 632 /* 633 * If there is still more to send, don't close the connection. 634 */ 635 flags &= ~TH_FIN; 636 len = tp->t_maxopd - optlen - ipoptlen; 637 sendalot = TRUE; 638 } 639 640 #ifdef INET6 641 KASSERT(max_linkhdr + hdrlen <= MCLBYTES, ("tcphdr too big")); 642 #else 643 KASSERT(max_linkhdr + hdrlen <= MHLEN, ("tcphdr too big")); 644 #endif 645 646 /* 647 * Grab a header mbuf, attaching a copy of data to 648 * be transmitted, and initialize the header from 649 * the template for sends on this connection. 650 */ 651 if (len) { 652 if ((tp->t_flags & TF_FORCE) && len == 1) 653 tcpstat.tcps_sndprobe++; 654 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { 655 if (tp->snd_nxt == tp->snd_una) 656 tp->snd_max_rexmt = tp->snd_max; 657 tcpstat.tcps_sndrexmitpack++; 658 tcpstat.tcps_sndrexmitbyte += len; 659 } else { 660 tcpstat.tcps_sndpack++; 661 tcpstat.tcps_sndbyte += len; 662 } 663 #ifdef notyet 664 if ((m = m_copypack(so->so_snd.sb_mb, off, (int)len, 665 max_linkhdr + hdrlen)) == NULL) { 666 error = ENOBUFS; 667 goto out; 668 } 669 /* 670 * m_copypack left space for our hdr; use it. 671 */ 672 m->m_len += hdrlen; 673 m->m_data -= hdrlen; 674 #else 675 MGETHDR(m, MB_DONTWAIT, MT_HEADER); 676 if (m == NULL) { 677 error = ENOBUFS; 678 goto out; 679 } 680 #ifdef INET6 681 if (MHLEN < hdrlen + max_linkhdr) { 682 MCLGET(m, MB_DONTWAIT); 683 if (!(m->m_flags & M_EXT)) { 684 m_freem(m); 685 error = ENOBUFS; 686 goto out; 687 } 688 } 689 #endif 690 m->m_data += max_linkhdr; 691 m->m_len = hdrlen; 692 if (len <= MHLEN - hdrlen - max_linkhdr) { 693 m_copydata(so->so_snd.sb_mb, off, (int) len, 694 mtod(m, caddr_t) + hdrlen); 695 m->m_len += len; 696 } else { 697 m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len); 698 if (m->m_next == NULL) { 699 (void) m_free(m); 700 error = ENOBUFS; 701 goto out; 702 } 703 } 704 #endif 705 /* 706 * If we're sending everything we've got, set PUSH. 707 * (This will keep happy those implementations which only 708 * give data to the user when a buffer fills or 709 * a PUSH comes in.) 710 */ 711 if (off + len == so->so_snd.sb_cc) 712 flags |= TH_PUSH; 713 } else { 714 if (tp->t_flags & TF_ACKNOW) 715 tcpstat.tcps_sndacks++; 716 else if (flags & (TH_SYN | TH_FIN | TH_RST)) 717 tcpstat.tcps_sndctrl++; 718 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 719 tcpstat.tcps_sndurg++; 720 else 721 tcpstat.tcps_sndwinup++; 722 723 MGETHDR(m, MB_DONTWAIT, MT_HEADER); 724 if (m == NULL) { 725 error = ENOBUFS; 726 goto out; 727 } 728 if (isipv6 && 729 (hdrlen + max_linkhdr > MHLEN) && hdrlen <= MHLEN) 730 MH_ALIGN(m, hdrlen); 731 else 732 m->m_data += max_linkhdr; 733 m->m_len = hdrlen; 734 } 735 m->m_pkthdr.rcvif = (struct ifnet *)0; 736 if (isipv6) { 737 ip6 = mtod(m, struct ip6_hdr *); 738 th = (struct tcphdr *)(ip6 + 1); 739 tcp_fillheaders(tp, ip6, th); 740 } else { 741 ip = mtod(m, struct ip *); 742 ipov = (struct ipovly *)ip; 743 th = (struct tcphdr *)(ip + 1); 744 /* this picks up the pseudo header (w/o the length) */ 745 tcp_fillheaders(tp, ip, th); 746 } 747 748 /* 749 * Fill in fields, remembering maximum advertised 750 * window for use in delaying messages about window sizes. 751 * If resending a FIN, be sure not to use a new sequence number. 752 */ 753 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN && 754 tp->snd_nxt == tp->snd_max) 755 tp->snd_nxt--; 756 /* 757 * If we are doing retransmissions, then snd_nxt will 758 * not reflect the first unsent octet. For ACK only 759 * packets, we do not want the sequence number of the 760 * retransmitted packet, we want the sequence number 761 * of the next unsent octet. So, if there is no data 762 * (and no SYN or FIN), use snd_max instead of snd_nxt 763 * when filling in ti_seq. But if we are in persist 764 * state, snd_max might reflect one byte beyond the 765 * right edge of the window, so use snd_nxt in that 766 * case, since we know we aren't doing a retransmission. 767 * (retransmit and persist are mutually exclusive...) 768 */ 769 if (len || (flags & (TH_SYN|TH_FIN)) || callout_active(tp->tt_persist)) 770 th->th_seq = htonl(tp->snd_nxt); 771 else 772 th->th_seq = htonl(tp->snd_max); 773 th->th_ack = htonl(tp->rcv_nxt); 774 if (optlen) { 775 bcopy(opt, th + 1, optlen); 776 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 777 } 778 th->th_flags = flags; 779 /* 780 * Calculate receive window. Don't shrink window, 781 * but avoid silly window syndrome. 782 */ 783 if (recvwin < (long)(so->so_rcv.sb_hiwat / 4) && 784 recvwin < (long)tp->t_maxseg) 785 recvwin = 0; 786 if (recvwin < (long)(tp->rcv_adv - tp->rcv_nxt)) 787 recvwin = (long)(tp->rcv_adv - tp->rcv_nxt); 788 if (recvwin > (long)TCP_MAXWIN << tp->rcv_scale) 789 recvwin = (long)TCP_MAXWIN << tp->rcv_scale; 790 th->th_win = htons((u_short) (recvwin>>tp->rcv_scale)); 791 792 /* 793 * Adjust the RXWIN0SENT flag - indicate that we have advertised 794 * a 0 window. This may cause the remote transmitter to stall. This 795 * flag tells soreceive() to disable delayed acknowledgements when 796 * draining the buffer. This can occur if the receiver is attempting 797 * to read more data then can be buffered prior to transmitting on 798 * the connection. 799 */ 800 if (recvwin == 0) 801 tp->t_flags |= TF_RXWIN0SENT; 802 else 803 tp->t_flags &= ~TF_RXWIN0SENT; 804 805 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 806 th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt)); 807 th->th_flags |= TH_URG; 808 } else { 809 /* 810 * If no urgent pointer to send, then we pull 811 * the urgent pointer to the left edge of the send window 812 * so that it doesn't drift into the send window on sequence 813 * number wraparound. 814 */ 815 tp->snd_up = tp->snd_una; /* drag it along */ 816 } 817 818 /* 819 * Put TCP length in extended header, and then 820 * checksum extended header and data. 821 */ 822 m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */ 823 if (isipv6) { 824 /* 825 * ip6_plen is not need to be filled now, and will be filled 826 * in ip6_output. 827 */ 828 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 829 sizeof(struct tcphdr) + optlen + len); 830 } else { 831 m->m_pkthdr.csum_flags = CSUM_TCP; 832 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 833 if (len + optlen) 834 th->th_sum = in_addword(th->th_sum, 835 htons((u_short)(optlen + len))); 836 837 /* IP version must be set here for ipv4/ipv6 checking later */ 838 KASSERT(ip->ip_v == IPVERSION, 839 ("%s: IP version incorrect: %d", __func__, ip->ip_v)); 840 } 841 842 /* 843 * In transmit state, time the transmission and arrange for 844 * the retransmit. In persist state, just set snd_max. 845 */ 846 if (!(tp->t_flags & TF_FORCE) || !callout_active(tp->tt_persist)) { 847 tcp_seq startseq = tp->snd_nxt; 848 849 /* 850 * Advance snd_nxt over sequence space of this segment. 851 */ 852 if (flags & (TH_SYN | TH_FIN)) { 853 if (flags & TH_SYN) 854 tp->snd_nxt++; 855 if (flags & TH_FIN) { 856 tp->snd_nxt++; 857 tp->t_flags |= TF_SENTFIN; 858 } 859 } 860 tp->snd_nxt += len; 861 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 862 tp->snd_max = tp->snd_nxt; 863 /* 864 * Time this transmission if not a retransmission and 865 * not currently timing anything. 866 */ 867 if (tp->t_rtttime == 0) { 868 tp->t_rtttime = ticks; 869 tp->t_rtseq = startseq; 870 tcpstat.tcps_segstimed++; 871 } 872 } 873 874 /* 875 * Set retransmit timer if not currently set, 876 * and not doing a pure ack or a keep-alive probe. 877 * Initial value for retransmit timer is smoothed 878 * round-trip time + 2 * round-trip time variance. 879 * Initialize shift counter which is used for backoff 880 * of retransmit time. 881 */ 882 if (!callout_active(tp->tt_rexmt) && 883 tp->snd_nxt != tp->snd_una) { 884 if (callout_active(tp->tt_persist)) { 885 callout_stop(tp->tt_persist); 886 tp->t_rxtshift = 0; 887 } 888 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 889 tcp_timer_rexmt, tp); 890 } 891 } else { 892 /* 893 * Persist case, update snd_max but since we are in 894 * persist mode (no window) we do not update snd_nxt. 895 */ 896 int xlen = len; 897 if (flags & TH_SYN) 898 ++xlen; 899 if (flags & TH_FIN) { 900 ++xlen; 901 tp->t_flags |= TF_SENTFIN; 902 } 903 if (SEQ_GT(tp->snd_nxt + xlen, tp->snd_max)) 904 tp->snd_max = tp->snd_nxt + xlen; 905 } 906 907 #ifdef TCPDEBUG 908 /* 909 * Trace. 910 */ 911 if (so->so_options & SO_DEBUG) 912 tcp_trace(TA_OUTPUT, tp->t_state, tp, mtod(m, void *), th, 0); 913 #endif 914 915 /* 916 * Fill in IP length and desired time to live and 917 * send to IP level. There should be a better way 918 * to handle ttl and tos; we could keep them in 919 * the template, but need a way to checksum without them. 920 */ 921 /* 922 * m->m_pkthdr.len should have been set before cksum calcuration, 923 * because in6_cksum() need it. 924 */ 925 if (isipv6) { 926 /* 927 * we separately set hoplimit for every segment, since the 928 * user might want to change the value via setsockopt. 929 * Also, desired default hop limit might be changed via 930 * Neighbor Discovery. 931 */ 932 ip6->ip6_hlim = in6_selecthlim(inp, 933 (inp->in6p_route.ro_rt ? 934 inp->in6p_route.ro_rt->rt_ifp : NULL)); 935 936 /* TODO: IPv6 IP6TOS_ECT bit on */ 937 error = ip6_output(m, inp->in6p_outputopts, &inp->in6p_route, 938 (so->so_options & SO_DONTROUTE), NULL, NULL, inp); 939 } else { 940 struct rtentry *rt; 941 ip->ip_len = m->m_pkthdr.len; 942 #ifdef INET6 943 if (INP_CHECK_SOCKAF(so, AF_INET6)) 944 ip->ip_ttl = in6_selecthlim(inp, 945 (inp->in6p_route.ro_rt ? 946 inp->in6p_route.ro_rt->rt_ifp : NULL)); 947 else 948 #endif 949 ip->ip_ttl = inp->inp_ip_ttl; /* XXX */ 950 951 ip->ip_tos = inp->inp_ip_tos; /* XXX */ 952 /* 953 * See if we should do MTU discovery. 954 * We do it only if the following are true: 955 * 1) we have a valid route to the destination 956 * 2) the MTU is not locked (if it is, 957 * then discovery has been disabled) 958 */ 959 if (path_mtu_discovery && 960 (rt = inp->inp_route.ro_rt) && (rt->rt_flags & RTF_UP) && 961 !(rt->rt_rmx.rmx_locks & RTV_MTU)) 962 ip->ip_off |= IP_DF; 963 964 error = ip_output(m, inp->inp_options, &inp->inp_route, 965 (so->so_options & SO_DONTROUTE), NULL, inp); 966 } 967 if (error) { 968 969 /* 970 * We know that the packet was lost, so back out the 971 * sequence number advance, if any. 972 */ 973 if (!(tp->t_flags & TF_FORCE) || 974 !callout_active(tp->tt_persist)) { 975 /* 976 * No need to check for TH_FIN here because 977 * the TF_SENTFIN flag handles that case. 978 */ 979 if (!(flags & TH_SYN)) 980 tp->snd_nxt -= len; 981 } 982 983 out: 984 if (error == ENOBUFS) { 985 /* 986 * If we can't send, make sure there is something 987 * to get us going again later. Persist state 988 * is not necessarily right, but it is close enough. 989 */ 990 if (!callout_active(tp->tt_rexmt) && 991 !callout_active(tp->tt_persist)) { 992 tp->t_rxtshift = 0; 993 tcp_setpersist(tp); 994 } 995 tcp_quench(inp, 0); 996 return (0); 997 } 998 if (error == EMSGSIZE) { 999 /* 1000 * ip_output() will have already fixed the route 1001 * for us. tcp_mtudisc() will, as its last action, 1002 * initiate retransmission, so it is important to 1003 * not do so here. 1004 */ 1005 tcp_mtudisc(inp, 0); 1006 return 0; 1007 } 1008 if ((error == EHOSTUNREACH || error == ENETDOWN) && 1009 TCPS_HAVERCVDSYN(tp->t_state)) { 1010 tp->t_softerror = error; 1011 return (0); 1012 } 1013 return (error); 1014 } 1015 tcpstat.tcps_sndtotal++; 1016 1017 /* 1018 * Data sent (as far as we can tell). 1019 * If this advertises a larger window than any other segment, 1020 * then remember the size of the advertised window. 1021 * Any pending ACK has now been sent. 1022 */ 1023 if (recvwin > 0 && SEQ_GT(tp->rcv_nxt + recvwin, tp->rcv_adv)) 1024 tp->rcv_adv = tp->rcv_nxt + recvwin; 1025 tp->last_ack_sent = tp->rcv_nxt; 1026 tp->t_flags &= ~TF_ACKNOW; 1027 if (tcp_delack_enabled) 1028 callout_stop(tp->tt_delack); 1029 if (sendalot) 1030 goto again; 1031 return (0); 1032 } 1033 1034 void 1035 tcp_setpersist(tp) 1036 struct tcpcb *tp; 1037 { 1038 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1; 1039 int tt; 1040 1041 if (callout_active(tp->tt_rexmt)) 1042 panic("tcp_setpersist: retransmit pending"); 1043 /* 1044 * Start/restart persistance timer. 1045 */ 1046 TCPT_RANGESET(tt, t * tcp_backoff[tp->t_rxtshift], 1047 TCPTV_PERSMIN, TCPTV_PERSMAX); 1048 callout_reset(tp->tt_persist, tt, tcp_timer_persist, tp); 1049 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 1050 tp->t_rxtshift++; 1051 } 1052