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 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_output.c 8.4 (Berkeley) 5/24/95 30 * $FreeBSD$ 31 */ 32 33 #include "opt_inet.h" 34 #include "opt_inet6.h" 35 #include "opt_ipsec.h" 36 #include "opt_mac.h" 37 #include "opt_tcpdebug.h" 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/domain.h> 42 #include <sys/kernel.h> 43 #include <sys/lock.h> 44 #include <sys/mbuf.h> 45 #include <sys/mutex.h> 46 #include <sys/protosw.h> 47 #include <sys/socket.h> 48 #include <sys/socketvar.h> 49 #include <sys/sysctl.h> 50 51 #include <net/route.h> 52 53 #include <netinet/in.h> 54 #include <netinet/in_systm.h> 55 #include <netinet/ip.h> 56 #include <netinet/in_pcb.h> 57 #include <netinet/ip_var.h> 58 #include <netinet/ip_options.h> 59 #ifdef INET6 60 #include <netinet6/in6_pcb.h> 61 #include <netinet/ip6.h> 62 #include <netinet6/ip6_var.h> 63 #endif 64 #include <netinet/tcp.h> 65 #define TCPOUTFLAGS 66 #include <netinet/tcp_fsm.h> 67 #include <netinet/tcp_seq.h> 68 #include <netinet/tcp_timer.h> 69 #include <netinet/tcp_var.h> 70 #include <netinet/tcpip.h> 71 #ifdef TCPDEBUG 72 #include <netinet/tcp_debug.h> 73 #endif 74 75 #ifdef IPSEC 76 #include <netinet6/ipsec.h> 77 #endif /*IPSEC*/ 78 79 #ifdef FAST_IPSEC 80 #include <netipsec/ipsec.h> 81 #define IPSEC 82 #endif /*FAST_IPSEC*/ 83 84 #include <machine/in_cksum.h> 85 86 #include <security/mac/mac_framework.h> 87 88 #ifdef notyet 89 extern struct mbuf *m_copypack(); 90 #endif 91 92 int path_mtu_discovery = 1; 93 SYSCTL_INT(_net_inet_tcp, OID_AUTO, path_mtu_discovery, CTLFLAG_RW, 94 &path_mtu_discovery, 1, "Enable Path MTU Discovery"); 95 96 int ss_fltsz = 1; 97 SYSCTL_INT(_net_inet_tcp, OID_AUTO, slowstart_flightsize, CTLFLAG_RW, 98 &ss_fltsz, 1, "Slow start flight size"); 99 100 int ss_fltsz_local = 4; 101 SYSCTL_INT(_net_inet_tcp, OID_AUTO, local_slowstart_flightsize, CTLFLAG_RW, 102 &ss_fltsz_local, 1, "Slow start flight size for local networks"); 103 104 int tcp_do_newreno = 1; 105 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newreno, CTLFLAG_RW, 106 &tcp_do_newreno, 0, "Enable NewReno Algorithms"); 107 108 int tcp_do_tso = 1; 109 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tso, CTLFLAG_RW, 110 &tcp_do_tso, 0, "Enable TCP Segmentation Offload"); 111 112 int tcp_do_autosndbuf = 1; 113 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sendbuf_auto, CTLFLAG_RW, 114 &tcp_do_autosndbuf, 0, "Enable automatic send buffer sizing"); 115 116 int tcp_autosndbuf_inc = 8*1024; 117 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sendbuf_inc, CTLFLAG_RW, 118 &tcp_autosndbuf_inc, 0, "Incrementor step size of automatic send buffer"); 119 120 int tcp_autosndbuf_max = 256*1024; 121 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sendbuf_max, CTLFLAG_RW, 122 &tcp_autosndbuf_max, 0, "Max size of automatic send buffer"); 123 124 125 /* 126 * Tcp output routine: figure out what should be sent and send it. 127 */ 128 int 129 tcp_output(struct tcpcb *tp) 130 { 131 struct socket *so = tp->t_inpcb->inp_socket; 132 long len, recwin, sendwin; 133 int off, flags, error; 134 #ifdef TCP_SIGNATURE 135 int sigoff = 0; 136 #endif 137 struct mbuf *m; 138 struct ip *ip = NULL; 139 struct ipovly *ipov = NULL; 140 struct tcphdr *th; 141 u_char opt[TCP_MAXOLEN]; 142 unsigned ipoptlen, optlen, hdrlen; 143 int idle, sendalot; 144 int sack_rxmit, sack_bytes_rxmt; 145 struct sackhole *p; 146 int tso = 0; 147 struct tcpopt to; 148 #if 0 149 int maxburst = TCP_MAXBURST; 150 #endif 151 #ifdef INET6 152 struct ip6_hdr *ip6 = NULL; 153 int isipv6; 154 155 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 156 #endif 157 158 INP_LOCK_ASSERT(tp->t_inpcb); 159 160 /* 161 * Determine length of data that should be transmitted, 162 * and flags that will be used. 163 * If there is some data or critical controls (SYN, RST) 164 * to send, then transmit; otherwise, investigate further. 165 */ 166 idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una); 167 if (idle && (ticks - tp->t_rcvtime) >= tp->t_rxtcur) { 168 /* 169 * We have been idle for "a while" and no acks are 170 * expected to clock out any data we send -- 171 * slow start to get ack "clock" running again. 172 * 173 * Set the slow-start flight size depending on whether 174 * this is a local network or not. 175 */ 176 int ss = ss_fltsz; 177 #ifdef INET6 178 if (isipv6) { 179 if (in6_localaddr(&tp->t_inpcb->in6p_faddr)) 180 ss = ss_fltsz_local; 181 } else 182 #endif /* INET6 */ 183 if (in_localaddr(tp->t_inpcb->inp_faddr)) 184 ss = ss_fltsz_local; 185 tp->snd_cwnd = tp->t_maxseg * ss; 186 } 187 tp->t_flags &= ~TF_LASTIDLE; 188 if (idle) { 189 if (tp->t_flags & TF_MORETOCOME) { 190 tp->t_flags |= TF_LASTIDLE; 191 idle = 0; 192 } 193 } 194 again: 195 /* 196 * If we've recently taken a timeout, snd_max will be greater than 197 * snd_nxt. There may be SACK information that allows us to avoid 198 * resending already delivered data. Adjust snd_nxt accordingly. 199 */ 200 if (tp->sack_enable && SEQ_LT(tp->snd_nxt, tp->snd_max)) 201 tcp_sack_adjust(tp); 202 sendalot = 0; 203 off = tp->snd_nxt - tp->snd_una; 204 sendwin = min(tp->snd_wnd, tp->snd_cwnd); 205 sendwin = min(sendwin, tp->snd_bwnd); 206 207 flags = tcp_outflags[tp->t_state]; 208 /* 209 * Send any SACK-generated retransmissions. If we're explicitly trying 210 * to send out new data (when sendalot is 1), bypass this function. 211 * If we retransmit in fast recovery mode, decrement snd_cwnd, since 212 * we're replacing a (future) new transmission with a retransmission 213 * now, and we previously incremented snd_cwnd in tcp_input(). 214 */ 215 /* 216 * Still in sack recovery , reset rxmit flag to zero. 217 */ 218 sack_rxmit = 0; 219 sack_bytes_rxmt = 0; 220 len = 0; 221 p = NULL; 222 if (tp->sack_enable && IN_FASTRECOVERY(tp) && 223 (p = tcp_sack_output(tp, &sack_bytes_rxmt))) { 224 long cwin; 225 226 cwin = min(tp->snd_wnd, tp->snd_cwnd) - sack_bytes_rxmt; 227 if (cwin < 0) 228 cwin = 0; 229 /* Do not retransmit SACK segments beyond snd_recover */ 230 if (SEQ_GT(p->end, tp->snd_recover)) { 231 /* 232 * (At least) part of sack hole extends beyond 233 * snd_recover. Check to see if we can rexmit data 234 * for this hole. 235 */ 236 if (SEQ_GEQ(p->rxmit, tp->snd_recover)) { 237 /* 238 * Can't rexmit any more data for this hole. 239 * That data will be rexmitted in the next 240 * sack recovery episode, when snd_recover 241 * moves past p->rxmit. 242 */ 243 p = NULL; 244 goto after_sack_rexmit; 245 } else 246 /* Can rexmit part of the current hole */ 247 len = ((long)ulmin(cwin, 248 tp->snd_recover - p->rxmit)); 249 } else 250 len = ((long)ulmin(cwin, p->end - p->rxmit)); 251 off = p->rxmit - tp->snd_una; 252 KASSERT(off >= 0,("%s: sack block to the left of una : %d", 253 __func__, off)); 254 if (len > 0) { 255 sack_rxmit = 1; 256 sendalot = 1; 257 tcpstat.tcps_sack_rexmits++; 258 tcpstat.tcps_sack_rexmit_bytes += 259 min(len, tp->t_maxseg); 260 } 261 } 262 after_sack_rexmit: 263 /* 264 * Get standard flags, and add SYN or FIN if requested by 'hidden' 265 * state flags. 266 */ 267 if (tp->t_flags & TF_NEEDFIN) 268 flags |= TH_FIN; 269 if (tp->t_flags & TF_NEEDSYN) 270 flags |= TH_SYN; 271 272 SOCKBUF_LOCK(&so->so_snd); 273 /* 274 * If in persist timeout with window of 0, send 1 byte. 275 * Otherwise, if window is small but nonzero 276 * and timer expired, we will send what we can 277 * and go to transmit state. 278 */ 279 if (tp->t_flags & TF_FORCEDATA) { 280 if (sendwin == 0) { 281 /* 282 * If we still have some data to send, then 283 * clear the FIN bit. Usually this would 284 * happen below when it realizes that we 285 * aren't sending all the data. However, 286 * if we have exactly 1 byte of unsent data, 287 * then it won't clear the FIN bit below, 288 * and if we are in persist state, we wind 289 * up sending the packet without recording 290 * that we sent the FIN bit. 291 * 292 * We can't just blindly clear the FIN bit, 293 * because if we don't have any more data 294 * to send then the probe will be the FIN 295 * itself. 296 */ 297 if (off < so->so_snd.sb_cc) 298 flags &= ~TH_FIN; 299 sendwin = 1; 300 } else { 301 callout_stop(tp->tt_persist); 302 tp->t_rxtshift = 0; 303 } 304 } 305 306 /* 307 * If snd_nxt == snd_max and we have transmitted a FIN, the 308 * offset will be > 0 even if so_snd.sb_cc is 0, resulting in 309 * a negative length. This can also occur when TCP opens up 310 * its congestion window while receiving additional duplicate 311 * acks after fast-retransmit because TCP will reset snd_nxt 312 * to snd_max after the fast-retransmit. 313 * 314 * In the normal retransmit-FIN-only case, however, snd_nxt will 315 * be set to snd_una, the offset will be 0, and the length may 316 * wind up 0. 317 * 318 * If sack_rxmit is true we are retransmitting from the scoreboard 319 * in which case len is already set. 320 */ 321 if (sack_rxmit == 0) { 322 if (sack_bytes_rxmt == 0) 323 len = ((long)ulmin(so->so_snd.sb_cc, sendwin) - off); 324 else { 325 long cwin; 326 327 /* 328 * We are inside of a SACK recovery episode and are 329 * sending new data, having retransmitted all the 330 * data possible in the scoreboard. 331 */ 332 len = ((long)ulmin(so->so_snd.sb_cc, tp->snd_wnd) 333 - off); 334 /* 335 * Don't remove this (len > 0) check ! 336 * We explicitly check for len > 0 here (although it 337 * isn't really necessary), to work around a gcc 338 * optimization issue - to force gcc to compute 339 * len above. Without this check, the computation 340 * of len is bungled by the optimizer. 341 */ 342 if (len > 0) { 343 cwin = tp->snd_cwnd - 344 (tp->snd_nxt - tp->sack_newdata) - 345 sack_bytes_rxmt; 346 if (cwin < 0) 347 cwin = 0; 348 len = lmin(len, cwin); 349 } 350 } 351 } 352 353 /* 354 * Lop off SYN bit if it has already been sent. However, if this 355 * is SYN-SENT state and if segment contains data and if we don't 356 * know that foreign host supports TAO, suppress sending segment. 357 */ 358 if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) { 359 if (tp->t_state != TCPS_SYN_RECEIVED) 360 flags &= ~TH_SYN; 361 off--, len++; 362 } 363 364 /* 365 * Be careful not to send data and/or FIN on SYN segments. 366 * This measure is needed to prevent interoperability problems 367 * with not fully conformant TCP implementations. 368 */ 369 if ((flags & TH_SYN) && (tp->t_flags & TF_NOOPT)) { 370 len = 0; 371 flags &= ~TH_FIN; 372 } 373 374 if (len < 0) { 375 /* 376 * If FIN has been sent but not acked, 377 * but we haven't been called to retransmit, 378 * len will be < 0. Otherwise, window shrank 379 * after we sent into it. If window shrank to 0, 380 * cancel pending retransmit, pull snd_nxt back 381 * to (closed) window, and set the persist timer 382 * if it isn't already going. If the window didn't 383 * close completely, just wait for an ACK. 384 */ 385 len = 0; 386 if (sendwin == 0) { 387 callout_stop(tp->tt_rexmt); 388 tp->t_rxtshift = 0; 389 tp->snd_nxt = tp->snd_una; 390 if (!callout_active(tp->tt_persist)) 391 tcp_setpersist(tp); 392 } 393 } 394 395 /* len will be >= 0 after this point. */ 396 KASSERT(len >= 0, ("%s: len < 0", __func__)); 397 398 /* 399 * Automatic sizing of send socket buffer. Often the send buffer 400 * size is not optimally adjusted to the actual network conditions 401 * at hand (delay bandwidth product). Setting the buffer size too 402 * small limits throughput on links with high bandwidth and high 403 * delay (eg. trans-continental/oceanic links). Setting the 404 * buffer size too big consumes too much real kernel memory, 405 * especially with many connections on busy servers. 406 * 407 * The criteria to step up the send buffer one notch are: 408 * 1. receive window of remote host is larger than send buffer 409 * (with a fudge factor of 5/4th); 410 * 2. send buffer is filled to 7/8th with data (so we actually 411 * have data to make use of it); 412 * 3. send buffer fill has not hit maximal automatic size; 413 * 4. our send window (slow start and cogestion controlled) is 414 * larger than sent but unacknowledged data in send buffer. 415 * 416 * The remote host receive window scaling factor may limit the 417 * growing of the send buffer before it reaches its allowed 418 * maximum. 419 * 420 * It scales directly with slow start or congestion window 421 * and does at most one step per received ACK. This fast 422 * scaling has the drawback of growing the send buffer beyond 423 * what is strictly necessary to make full use of a given 424 * delay*bandwith product. However testing has shown this not 425 * to be much of an problem. At worst we are trading wasting 426 * of available bandwith (the non-use of it) for wasting some 427 * socket buffer memory. 428 * 429 * TODO: Shrink send buffer during idle periods together 430 * with congestion window. Requires another timer. Has to 431 * wait for upcoming tcp timer rewrite. 432 */ 433 if (tcp_do_autosndbuf && so->so_snd.sb_flags & SB_AUTOSIZE) { 434 if ((tp->snd_wnd / 4 * 5) >= so->so_snd.sb_hiwat && 435 so->so_snd.sb_cc >= (so->so_snd.sb_hiwat / 8 * 7) && 436 so->so_snd.sb_cc < tcp_autosndbuf_max && 437 sendwin >= (so->so_snd.sb_cc - (tp->snd_nxt - tp->snd_una))) { 438 if (!sbreserve_locked(&so->so_snd, 439 min(so->so_snd.sb_hiwat + tcp_autosndbuf_inc, 440 tcp_autosndbuf_max), so, curthread)) 441 so->so_snd.sb_flags &= ~SB_AUTOSIZE; 442 } 443 } 444 445 /* 446 * Truncate to the maximum segment length or enable TCP Segmentation 447 * Offloading (if supported by hardware) and ensure that FIN is removed 448 * if the length no longer contains the last data byte. 449 * 450 * TSO may only be used if we are in a pure bulk sending state. The 451 * presence of TCP-MD5, SACK retransmits, SACK advertizements and 452 * IP options prevent using TSO. With TSO the TCP header is the same 453 * (except for the sequence number) for all generated packets. This 454 * makes it impossible to transmit any options which vary per generated 455 * segment or packet. 456 * 457 * The length of TSO bursts is limited to TCP_MAXWIN. That limit and 458 * removal of FIN (if not already catched here) are handled later after 459 * the exact length of the TCP options are known. 460 */ 461 if (len > tp->t_maxseg) { 462 if ((tp->t_flags & TF_TSO) && tcp_do_tso && 463 ((tp->t_flags & TF_SIGNATURE) == 0) && 464 tp->rcv_numsacks == 0 && sack_rxmit == 0 && 465 tp->t_inpcb->inp_options == NULL && 466 tp->t_inpcb->in6p_options == NULL && 467 tp->t_inpcb->inp_sp == NULL) { 468 tso = 1; 469 } else { 470 len = tp->t_maxseg; 471 sendalot = 1; 472 tso = 0; 473 } 474 } 475 if (sack_rxmit) { 476 if (SEQ_LT(p->rxmit + len, tp->snd_una + so->so_snd.sb_cc)) 477 flags &= ~TH_FIN; 478 } else { 479 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc)) 480 flags &= ~TH_FIN; 481 } 482 483 recwin = sbspace(&so->so_rcv); 484 485 /* 486 * Sender silly window avoidance. We transmit under the following 487 * conditions when len is non-zero: 488 * 489 * - We have a full segment (or more with TSO) 490 * - This is the last buffer in a write()/send() and we are 491 * either idle or running NODELAY 492 * - we've timed out (e.g. persist timer) 493 * - we have more then 1/2 the maximum send window's worth of 494 * data (receiver may be limited the window size) 495 * - we need to retransmit 496 */ 497 if (len) { 498 if (len >= tp->t_maxseg) 499 goto send; 500 /* 501 * NOTE! on localhost connections an 'ack' from the remote 502 * end may occur synchronously with the output and cause 503 * us to flush a buffer queued with moretocome. XXX 504 * 505 * note: the len + off check is almost certainly unnecessary. 506 */ 507 if (!(tp->t_flags & TF_MORETOCOME) && /* normal case */ 508 (idle || (tp->t_flags & TF_NODELAY)) && 509 len + off >= so->so_snd.sb_cc && 510 (tp->t_flags & TF_NOPUSH) == 0) { 511 goto send; 512 } 513 if (tp->t_flags & TF_FORCEDATA) /* typ. timeout case */ 514 goto send; 515 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0) 516 goto send; 517 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */ 518 goto send; 519 if (sack_rxmit) 520 goto send; 521 } 522 523 /* 524 * Compare available window to amount of window 525 * known to peer (as advertised window less 526 * next expected input). If the difference is at least two 527 * max size segments, or at least 50% of the maximum possible 528 * window, then want to send a window update to peer. 529 * Skip this if the connection is in T/TCP half-open state. 530 */ 531 if (recwin > 0 && !(tp->t_flags & TF_NEEDSYN)) { 532 /* 533 * "adv" is the amount we can increase the window, 534 * taking into account that we are limited by 535 * TCP_MAXWIN << tp->rcv_scale. 536 */ 537 long adv = min(recwin, (long)TCP_MAXWIN << tp->rcv_scale) - 538 (tp->rcv_adv - tp->rcv_nxt); 539 540 if (adv >= (long) (2 * tp->t_maxseg)) 541 goto send; 542 if (2 * adv >= (long) so->so_rcv.sb_hiwat) 543 goto send; 544 } 545 546 /* 547 * Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW 548 * is also a catch-all for the retransmit timer timeout case. 549 */ 550 if (tp->t_flags & TF_ACKNOW) 551 goto send; 552 if ((flags & TH_RST) || 553 ((flags & TH_SYN) && (tp->t_flags & TF_NEEDSYN) == 0)) 554 goto send; 555 if (SEQ_GT(tp->snd_up, tp->snd_una)) 556 goto send; 557 /* 558 * If our state indicates that FIN should be sent 559 * and we have not yet done so, then we need to send. 560 */ 561 if (flags & TH_FIN && 562 ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una)) 563 goto send; 564 /* 565 * In SACK, it is possible for tcp_output to fail to send a segment 566 * after the retransmission timer has been turned off. Make sure 567 * that the retransmission timer is set. 568 */ 569 if (tp->sack_enable && SEQ_GT(tp->snd_max, tp->snd_una) && 570 !callout_active(tp->tt_rexmt) && 571 !callout_active(tp->tt_persist)) { 572 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 573 tcp_timer_rexmt, tp); 574 goto just_return; 575 } 576 /* 577 * TCP window updates are not reliable, rather a polling protocol 578 * using ``persist'' packets is used to insure receipt of window 579 * updates. The three ``states'' for the output side are: 580 * idle not doing retransmits or persists 581 * persisting to move a small or zero window 582 * (re)transmitting and thereby not persisting 583 * 584 * callout_active(tp->tt_persist) 585 * is true when we are in persist state. 586 * (tp->t_flags & TF_FORCEDATA) 587 * is set when we are called to send a persist packet. 588 * callout_active(tp->tt_rexmt) 589 * is set when we are retransmitting 590 * The output side is idle when both timers are zero. 591 * 592 * If send window is too small, there is data to transmit, and no 593 * retransmit or persist is pending, then go to persist state. 594 * If nothing happens soon, send when timer expires: 595 * if window is nonzero, transmit what we can, 596 * otherwise force out a byte. 597 */ 598 if (so->so_snd.sb_cc && !callout_active(tp->tt_rexmt) && 599 !callout_active(tp->tt_persist)) { 600 tp->t_rxtshift = 0; 601 tcp_setpersist(tp); 602 } 603 604 /* 605 * No reason to send a segment, just return. 606 */ 607 just_return: 608 SOCKBUF_UNLOCK(&so->so_snd); 609 return (0); 610 611 send: 612 SOCKBUF_LOCK_ASSERT(&so->so_snd); 613 /* 614 * Before ESTABLISHED, force sending of initial options 615 * unless TCP set not to do any options. 616 * NOTE: we assume that the IP/TCP header plus TCP options 617 * always fit in a single mbuf, leaving room for a maximum 618 * link header, i.e. 619 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES 620 */ 621 optlen = 0; 622 #ifdef INET6 623 if (isipv6) 624 hdrlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 625 else 626 #endif 627 hdrlen = sizeof (struct tcpiphdr); 628 629 /* 630 * Compute options for segment. 631 * We only have to care about SYN and established connection 632 * segments. Options for SYN-ACK segments are handled in TCP 633 * syncache. 634 */ 635 if ((tp->t_flags & TF_NOOPT) == 0) { 636 to.to_flags = 0; 637 /* Maximum segment size. */ 638 if (flags & TH_SYN) { 639 tp->snd_nxt = tp->iss; 640 to.to_mss = tcp_mssopt(&tp->t_inpcb->inp_inc); 641 to.to_flags |= TOF_MSS; 642 } 643 /* Window scaling. */ 644 if ((flags & TH_SYN) && (tp->t_flags & TF_REQ_SCALE)) { 645 to.to_wscale = tp->request_r_scale; 646 to.to_flags |= TOF_SCALE; 647 } 648 /* Timestamps. */ 649 if ((tp->t_flags & TF_RCVD_TSTMP) || 650 ((flags & TH_SYN) && (tp->t_flags & TF_REQ_TSTMP))) { 651 to.to_tsval = ticks + tp->ts_offset; 652 to.to_tsecr = tp->ts_recent; 653 to.to_flags |= TOF_TS; 654 /* Set receive buffer autosizing timestamp. */ 655 if (tp->rfbuf_ts == 0 && 656 (so->so_rcv.sb_flags & SB_AUTOSIZE)) 657 tp->rfbuf_ts = ticks; 658 } 659 /* Selective ACK's. */ 660 if (tp->sack_enable) { 661 if (flags & TH_SYN) 662 to.to_flags |= TOF_SACKPERM; 663 else if (TCPS_HAVEESTABLISHED(tp->t_state) && 664 (tp->t_flags & TF_SACK_PERMIT) && 665 tp->rcv_numsacks > 0) { 666 to.to_flags |= TOF_SACK; 667 to.to_nsacks = tp->rcv_numsacks; 668 to.to_sacks = (u_char *)tp->sackblks; 669 } 670 } 671 #ifdef TCP_SIGNATURE 672 /* TCP-MD5 (RFC2385). */ 673 #ifdef INET6 674 if (!isipv6 && (tp->t_flags & TF_SIGNATURE)) 675 #else 676 if (tp->t_flags & TF_SIGNATURE) 677 #endif /* INET6 */ 678 to.to_flags |= TOF_SIGNATURE; 679 #endif /* TCP_SIGNATURE */ 680 681 /* Processing the options. */ 682 hdrlen += optlen = tcp_addoptions(&to, (u_char *)&opt); 683 684 #ifdef TCP_SIGNATURE 685 sigoff = to.to_signature - (u_char *)&to; 686 #endif /* TCP_SIGNATURE */ 687 } 688 689 #ifdef INET6 690 if (isipv6) 691 ipoptlen = ip6_optlen(tp->t_inpcb); 692 else 693 #endif 694 if (tp->t_inpcb->inp_options) 695 ipoptlen = tp->t_inpcb->inp_options->m_len - 696 offsetof(struct ipoption, ipopt_list); 697 else 698 ipoptlen = 0; 699 #ifdef IPSEC 700 ipoptlen += ipsec_hdrsiz_tcp(tp); 701 #endif 702 703 /* 704 * Adjust data length if insertion of options will 705 * bump the packet length beyond the t_maxopd length. 706 * Clear the FIN bit because we cut off the tail of 707 * the segment. 708 * 709 * When doing TSO limit a burst to TCP_MAXWIN minus the 710 * IP, TCP and Options length to keep ip->ip_len from 711 * overflowing. Prevent the last segment from being 712 * fractional thus making them all equal sized and set 713 * the flag to continue sending. TSO is disabled when 714 * IP options or IPSEC are present. 715 */ 716 if (len + optlen + ipoptlen > tp->t_maxopd) { 717 flags &= ~TH_FIN; 718 if (tso) { 719 if (len > TCP_MAXWIN - hdrlen - optlen) { 720 len = TCP_MAXWIN - hdrlen - optlen; 721 len = len - (len % (tp->t_maxopd - optlen)); 722 sendalot = 1; 723 } else if (tp->t_flags & TF_NEEDFIN) 724 sendalot = 1; 725 } else { 726 len = tp->t_maxopd - optlen - ipoptlen; 727 sendalot = 1; 728 } 729 } 730 731 /*#ifdef DIAGNOSTIC*/ 732 #ifdef INET6 733 if (max_linkhdr + hdrlen > MCLBYTES) 734 #else 735 if (max_linkhdr + hdrlen > MHLEN) 736 #endif 737 panic("tcphdr too big"); 738 /*#endif*/ 739 740 /* 741 * Grab a header mbuf, attaching a copy of data to 742 * be transmitted, and initialize the header from 743 * the template for sends on this connection. 744 */ 745 if (len) { 746 struct mbuf *mb; 747 u_int moff; 748 749 if ((tp->t_flags & TF_FORCEDATA) && len == 1) 750 tcpstat.tcps_sndprobe++; 751 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { 752 tcpstat.tcps_sndrexmitpack++; 753 tcpstat.tcps_sndrexmitbyte += len; 754 } else { 755 tcpstat.tcps_sndpack++; 756 tcpstat.tcps_sndbyte += len; 757 } 758 #ifdef notyet 759 if ((m = m_copypack(so->so_snd.sb_mb, off, 760 (int)len, max_linkhdr + hdrlen)) == 0) { 761 SOCKBUF_UNLOCK(&so->so_snd); 762 error = ENOBUFS; 763 goto out; 764 } 765 /* 766 * m_copypack left space for our hdr; use it. 767 */ 768 m->m_len += hdrlen; 769 m->m_data -= hdrlen; 770 #else 771 MGETHDR(m, M_DONTWAIT, MT_DATA); 772 if (m == NULL) { 773 SOCKBUF_UNLOCK(&so->so_snd); 774 error = ENOBUFS; 775 goto out; 776 } 777 #ifdef INET6 778 if (MHLEN < hdrlen + max_linkhdr) { 779 MCLGET(m, M_DONTWAIT); 780 if ((m->m_flags & M_EXT) == 0) { 781 SOCKBUF_UNLOCK(&so->so_snd); 782 m_freem(m); 783 error = ENOBUFS; 784 goto out; 785 } 786 } 787 #endif 788 m->m_data += max_linkhdr; 789 m->m_len = hdrlen; 790 791 /* 792 * Start the m_copy functions from the closest mbuf 793 * to the offset in the socket buffer chain. 794 */ 795 mb = sbsndptr(&so->so_snd, off, len, &moff); 796 797 if (len <= MHLEN - hdrlen - max_linkhdr) { 798 m_copydata(mb, moff, (int)len, 799 mtod(m, caddr_t) + hdrlen); 800 m->m_len += len; 801 } else { 802 m->m_next = m_copy(mb, moff, (int)len); 803 if (m->m_next == NULL) { 804 SOCKBUF_UNLOCK(&so->so_snd); 805 (void) m_free(m); 806 error = ENOBUFS; 807 goto out; 808 } 809 } 810 #endif 811 /* 812 * If we're sending everything we've got, set PUSH. 813 * (This will keep happy those implementations which only 814 * give data to the user when a buffer fills or 815 * a PUSH comes in.) 816 */ 817 if (off + len == so->so_snd.sb_cc) 818 flags |= TH_PUSH; 819 SOCKBUF_UNLOCK(&so->so_snd); 820 } else { 821 SOCKBUF_UNLOCK(&so->so_snd); 822 if (tp->t_flags & TF_ACKNOW) 823 tcpstat.tcps_sndacks++; 824 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 825 tcpstat.tcps_sndctrl++; 826 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 827 tcpstat.tcps_sndurg++; 828 else 829 tcpstat.tcps_sndwinup++; 830 831 MGETHDR(m, M_DONTWAIT, MT_DATA); 832 if (m == NULL) { 833 error = ENOBUFS; 834 goto out; 835 } 836 #ifdef INET6 837 if (isipv6 && (MHLEN < hdrlen + max_linkhdr) && 838 MHLEN >= hdrlen) { 839 MH_ALIGN(m, hdrlen); 840 } else 841 #endif 842 m->m_data += max_linkhdr; 843 m->m_len = hdrlen; 844 } 845 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); 846 m->m_pkthdr.rcvif = (struct ifnet *)0; 847 #ifdef MAC 848 mac_create_mbuf_from_inpcb(tp->t_inpcb, m); 849 #endif 850 #ifdef INET6 851 if (isipv6) { 852 ip6 = mtod(m, struct ip6_hdr *); 853 th = (struct tcphdr *)(ip6 + 1); 854 tcpip_fillheaders(tp->t_inpcb, ip6, th); 855 } else 856 #endif /* INET6 */ 857 { 858 ip = mtod(m, struct ip *); 859 ipov = (struct ipovly *)ip; 860 th = (struct tcphdr *)(ip + 1); 861 tcpip_fillheaders(tp->t_inpcb, ip, th); 862 } 863 864 /* 865 * Fill in fields, remembering maximum advertised 866 * window for use in delaying messages about window sizes. 867 * If resending a FIN, be sure not to use a new sequence number. 868 */ 869 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN && 870 tp->snd_nxt == tp->snd_max) 871 tp->snd_nxt--; 872 /* 873 * If we are doing retransmissions, then snd_nxt will 874 * not reflect the first unsent octet. For ACK only 875 * packets, we do not want the sequence number of the 876 * retransmitted packet, we want the sequence number 877 * of the next unsent octet. So, if there is no data 878 * (and no SYN or FIN), use snd_max instead of snd_nxt 879 * when filling in ti_seq. But if we are in persist 880 * state, snd_max might reflect one byte beyond the 881 * right edge of the window, so use snd_nxt in that 882 * case, since we know we aren't doing a retransmission. 883 * (retransmit and persist are mutually exclusive...) 884 */ 885 if (sack_rxmit == 0) { 886 if (len || (flags & (TH_SYN|TH_FIN)) 887 || callout_active(tp->tt_persist)) 888 th->th_seq = htonl(tp->snd_nxt); 889 else 890 th->th_seq = htonl(tp->snd_max); 891 } else { 892 th->th_seq = htonl(p->rxmit); 893 p->rxmit += len; 894 tp->sackhint.sack_bytes_rexmit += len; 895 } 896 th->th_ack = htonl(tp->rcv_nxt); 897 if (optlen) { 898 bcopy(opt, th + 1, optlen); 899 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 900 } 901 th->th_flags = flags; 902 /* 903 * Calculate receive window. Don't shrink window, 904 * but avoid silly window syndrome. 905 * 906 * XXX: RFC1323: The Window field in a SYN (i.e., a <SYN> or 907 * <SYN,ACK>) segment itself is never scaled. 908 */ 909 if (recwin < (long)(so->so_rcv.sb_hiwat / 4) && 910 recwin < (long)tp->t_maxseg) 911 recwin = 0; 912 if (recwin < (long)(tp->rcv_adv - tp->rcv_nxt)) 913 recwin = (long)(tp->rcv_adv - tp->rcv_nxt); 914 if (recwin > (long)TCP_MAXWIN << tp->rcv_scale) 915 recwin = (long)TCP_MAXWIN << tp->rcv_scale; 916 th->th_win = htons((u_short) (recwin >> tp->rcv_scale)); 917 918 919 /* 920 * Adjust the RXWIN0SENT flag - indicate that we have advertised 921 * a 0 window. This may cause the remote transmitter to stall. This 922 * flag tells soreceive() to disable delayed acknowledgements when 923 * draining the buffer. This can occur if the receiver is attempting 924 * to read more data then can be buffered prior to transmitting on 925 * the connection. 926 */ 927 if (recwin == 0) 928 tp->t_flags |= TF_RXWIN0SENT; 929 else 930 tp->t_flags &= ~TF_RXWIN0SENT; 931 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 932 th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt)); 933 th->th_flags |= TH_URG; 934 } else 935 /* 936 * If no urgent pointer to send, then we pull 937 * the urgent pointer to the left edge of the send window 938 * so that it doesn't drift into the send window on sequence 939 * number wraparound. 940 */ 941 tp->snd_up = tp->snd_una; /* drag it along */ 942 943 #ifdef TCP_SIGNATURE 944 #ifdef INET6 945 if (!isipv6) 946 #endif 947 if (tp->t_flags & TF_SIGNATURE) 948 tcp_signature_compute(m, sizeof(struct ip), len, optlen, 949 (u_char *)(th + 1) + sigoff, IPSEC_DIR_OUTBOUND); 950 #endif 951 952 /* 953 * Put TCP length in extended header, and then 954 * checksum extended header and data. 955 */ 956 m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */ 957 #ifdef INET6 958 if (isipv6) 959 /* 960 * ip6_plen is not need to be filled now, and will be filled 961 * in ip6_output. 962 */ 963 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 964 sizeof(struct tcphdr) + optlen + len); 965 else 966 #endif /* INET6 */ 967 { 968 m->m_pkthdr.csum_flags = CSUM_TCP; 969 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 970 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 971 htons(sizeof(struct tcphdr) + IPPROTO_TCP + len + optlen)); 972 973 /* IP version must be set here for ipv4/ipv6 checking later */ 974 KASSERT(ip->ip_v == IPVERSION, 975 ("%s: IP version incorrect: %d", __func__, ip->ip_v)); 976 } 977 978 /* 979 * Enable TSO and specify the size of the segments. 980 * The TCP pseudo header checksum is always provided. 981 * XXX: Fixme: This is currently not the case for IPv6. 982 */ 983 if (tso) { 984 m->m_pkthdr.csum_flags = CSUM_TSO; 985 m->m_pkthdr.tso_segsz = tp->t_maxopd - optlen; 986 } 987 988 /* 989 * In transmit state, time the transmission and arrange for 990 * the retransmit. In persist state, just set snd_max. 991 */ 992 if ((tp->t_flags & TF_FORCEDATA) == 0 || 993 !callout_active(tp->tt_persist)) { 994 tcp_seq startseq = tp->snd_nxt; 995 996 /* 997 * Advance snd_nxt over sequence space of this segment. 998 */ 999 if (flags & (TH_SYN|TH_FIN)) { 1000 if (flags & TH_SYN) 1001 tp->snd_nxt++; 1002 if (flags & TH_FIN) { 1003 tp->snd_nxt++; 1004 tp->t_flags |= TF_SENTFIN; 1005 } 1006 } 1007 if (sack_rxmit) 1008 goto timer; 1009 tp->snd_nxt += len; 1010 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 1011 tp->snd_max = tp->snd_nxt; 1012 /* 1013 * Time this transmission if not a retransmission and 1014 * not currently timing anything. 1015 */ 1016 if (tp->t_rtttime == 0) { 1017 tp->t_rtttime = ticks; 1018 tp->t_rtseq = startseq; 1019 tcpstat.tcps_segstimed++; 1020 } 1021 } 1022 1023 /* 1024 * Set retransmit timer if not currently set, 1025 * and not doing a pure ack or a keep-alive probe. 1026 * Initial value for retransmit timer is smoothed 1027 * round-trip time + 2 * round-trip time variance. 1028 * Initialize shift counter which is used for backoff 1029 * of retransmit time. 1030 */ 1031 timer: 1032 if (!callout_active(tp->tt_rexmt) && 1033 ((sack_rxmit && tp->snd_nxt != tp->snd_max) || 1034 (tp->snd_nxt != tp->snd_una))) { 1035 if (callout_active(tp->tt_persist)) { 1036 callout_stop(tp->tt_persist); 1037 tp->t_rxtshift = 0; 1038 } 1039 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1040 tcp_timer_rexmt, tp); 1041 } 1042 } else { 1043 /* 1044 * Persist case, update snd_max but since we are in 1045 * persist mode (no window) we do not update snd_nxt. 1046 */ 1047 int xlen = len; 1048 if (flags & TH_SYN) 1049 ++xlen; 1050 if (flags & TH_FIN) { 1051 ++xlen; 1052 tp->t_flags |= TF_SENTFIN; 1053 } 1054 if (SEQ_GT(tp->snd_nxt + xlen, tp->snd_max)) 1055 tp->snd_max = tp->snd_nxt + len; 1056 } 1057 1058 #ifdef TCPDEBUG 1059 /* 1060 * Trace. 1061 */ 1062 if (so->so_options & SO_DEBUG) { 1063 u_short save = 0; 1064 #ifdef INET6 1065 if (!isipv6) 1066 #endif 1067 { 1068 save = ipov->ih_len; 1069 ipov->ih_len = htons(m->m_pkthdr.len /* - hdrlen + (th->th_off << 2) */); 1070 } 1071 tcp_trace(TA_OUTPUT, tp->t_state, tp, mtod(m, void *), th, 0); 1072 #ifdef INET6 1073 if (!isipv6) 1074 #endif 1075 ipov->ih_len = save; 1076 } 1077 #endif 1078 1079 /* 1080 * Fill in IP length and desired time to live and 1081 * send to IP level. There should be a better way 1082 * to handle ttl and tos; we could keep them in 1083 * the template, but need a way to checksum without them. 1084 */ 1085 /* 1086 * m->m_pkthdr.len should have been set before cksum calcuration, 1087 * because in6_cksum() need it. 1088 */ 1089 #ifdef INET6 1090 if (isipv6) { 1091 /* 1092 * we separately set hoplimit for every segment, since the 1093 * user might want to change the value via setsockopt. 1094 * Also, desired default hop limit might be changed via 1095 * Neighbor Discovery. 1096 */ 1097 ip6->ip6_hlim = in6_selecthlim(tp->t_inpcb, NULL); 1098 1099 /* TODO: IPv6 IP6TOS_ECT bit on */ 1100 error = ip6_output(m, 1101 tp->t_inpcb->in6p_outputopts, NULL, 1102 ((so->so_options & SO_DONTROUTE) ? 1103 IP_ROUTETOIF : 0), NULL, NULL, tp->t_inpcb); 1104 } else 1105 #endif /* INET6 */ 1106 { 1107 ip->ip_len = m->m_pkthdr.len; 1108 #ifdef INET6 1109 if (INP_CHECK_SOCKAF(so, AF_INET6)) 1110 ip->ip_ttl = in6_selecthlim(tp->t_inpcb, NULL); 1111 #endif /* INET6 */ 1112 /* 1113 * If we do path MTU discovery, then we set DF on every packet. 1114 * This might not be the best thing to do according to RFC3390 1115 * Section 2. However the tcp hostcache migitates the problem 1116 * so it affects only the first tcp connection with a host. 1117 */ 1118 if (path_mtu_discovery) 1119 ip->ip_off |= IP_DF; 1120 1121 error = ip_output(m, tp->t_inpcb->inp_options, NULL, 1122 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 0, 1123 tp->t_inpcb); 1124 } 1125 if (error) { 1126 1127 /* 1128 * We know that the packet was lost, so back out the 1129 * sequence number advance, if any. 1130 * 1131 * If the error is EPERM the packet got blocked by the 1132 * local firewall. Normally we should terminate the 1133 * connection but the blocking may have been spurious 1134 * due to a firewall reconfiguration cycle. So we treat 1135 * it like a packet loss and let the retransmit timer and 1136 * timeouts do their work over time. 1137 * XXX: It is a POLA question whether calling tcp_drop right 1138 * away would be the really correct behavior instead. 1139 */ 1140 if (((tp->t_flags & TF_FORCEDATA) == 0 || 1141 !callout_active(tp->tt_persist)) && 1142 ((flags & TH_SYN) == 0) && 1143 (error != EPERM)) { 1144 if (sack_rxmit) { 1145 p->rxmit -= len; 1146 tp->sackhint.sack_bytes_rexmit -= len; 1147 KASSERT(tp->sackhint.sack_bytes_rexmit >= 0, 1148 ("sackhint bytes rtx >= 0")); 1149 } else 1150 tp->snd_nxt -= len; 1151 } 1152 out: 1153 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); /* Check gotos. */ 1154 switch (error) { 1155 case EPERM: 1156 tp->t_softerror = error; 1157 return (error); 1158 case ENOBUFS: 1159 if (!callout_active(tp->tt_rexmt) && 1160 !callout_active(tp->tt_persist)) 1161 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1162 tcp_timer_rexmt, tp); 1163 tp->snd_cwnd = tp->t_maxseg; 1164 return (0); 1165 case EMSGSIZE: 1166 /* 1167 * For some reason the interface we used initially 1168 * to send segments changed to another or lowered 1169 * its MTU. 1170 * 1171 * tcp_mtudisc() will find out the new MTU and as 1172 * its last action, initiate retransmission, so it 1173 * is important to not do so here. 1174 * 1175 * If TSO was active we either got an interface 1176 * without TSO capabilits or TSO was turned off. 1177 * Disable it for this connection as too and 1178 * immediatly retry with MSS sized segments generated 1179 * by this function. 1180 */ 1181 if (tso) 1182 tp->t_flags &= ~TF_TSO; 1183 tcp_mtudisc(tp->t_inpcb, 0); 1184 return (0); 1185 case EHOSTDOWN: 1186 case EHOSTUNREACH: 1187 case ENETDOWN: 1188 case ENETUNREACH: 1189 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1190 tp->t_softerror = error; 1191 return (0); 1192 } 1193 /* FALLTHROUGH */ 1194 default: 1195 return (error); 1196 } 1197 } 1198 tcpstat.tcps_sndtotal++; 1199 1200 /* 1201 * Data sent (as far as we can tell). 1202 * If this advertises a larger window than any other segment, 1203 * then remember the size of the advertised window. 1204 * Any pending ACK has now been sent. 1205 */ 1206 if (recwin > 0 && SEQ_GT(tp->rcv_nxt + recwin, tp->rcv_adv)) 1207 tp->rcv_adv = tp->rcv_nxt + recwin; 1208 tp->last_ack_sent = tp->rcv_nxt; 1209 tp->t_flags &= ~(TF_ACKNOW | TF_DELACK); 1210 if (callout_active(tp->tt_delack)) 1211 callout_stop(tp->tt_delack); 1212 #if 0 1213 /* 1214 * This completely breaks TCP if newreno is turned on. What happens 1215 * is that if delayed-acks are turned on on the receiver, this code 1216 * on the transmitter effectively destroys the TCP window, forcing 1217 * it to four packets (1.5Kx4 = 6K window). 1218 */ 1219 if (sendalot && (!tcp_do_newreno || --maxburst)) 1220 goto again; 1221 #endif 1222 if (sendalot) 1223 goto again; 1224 return (0); 1225 } 1226 1227 void 1228 tcp_setpersist(struct tcpcb *tp) 1229 { 1230 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1; 1231 int tt; 1232 1233 if (callout_active(tp->tt_rexmt)) 1234 panic("tcp_setpersist: retransmit pending"); 1235 /* 1236 * Start/restart persistance timer. 1237 */ 1238 TCPT_RANGESET(tt, t * tcp_backoff[tp->t_rxtshift], 1239 TCPTV_PERSMIN, TCPTV_PERSMAX); 1240 callout_reset(tp->tt_persist, tt, tcp_timer_persist, tp); 1241 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 1242 tp->t_rxtshift++; 1243 } 1244 1245 /* 1246 * Insert TCP options according to the supplied parameters to the place 1247 * optp in a consistent way. Can handle unaligned destinations. 1248 * 1249 * The order of the option processing is crucial for optimal packing and 1250 * alignment for the scarce option space. 1251 * 1252 * The optimal order for a SYN/SYN-ACK segment is: 1253 * MSS (4) + NOP (1) + Window scale (3) + SACK permitted (2) + 1254 * Timestamp (10) + Signature (18) = 38 bytes out of a maximum of 40. 1255 * 1256 * The SACK options should be last. SACK blocks consume 8*n+2 bytes. 1257 * So a full size SACK blocks option is 34 bytes (with 4 SACK blocks). 1258 * At minimum we need 10 bytes (to generate 1 SACK block). If both 1259 * TCP Timestamps (12 bytes) and TCP Signatures (18 bytes) are present, 1260 * we only have 10 bytes for SACK options (40 - (12 + 18)). 1261 */ 1262 int 1263 tcp_addoptions(struct tcpopt *to, u_char *optp) 1264 { 1265 u_int mask, optlen = 0; 1266 1267 for (mask = 1; mask < TOF_MAXOPT; mask <<= 1) { 1268 if ((to->to_flags & mask) != mask) 1269 continue; 1270 switch (to->to_flags & mask) { 1271 case TOF_MSS: 1272 while (optlen % 4) { 1273 optlen += TCPOLEN_NOP; 1274 *optp++ = TCPOPT_NOP; 1275 } 1276 optlen += TCPOLEN_MAXSEG; 1277 *optp++ = TCPOPT_MAXSEG; 1278 *optp++ = TCPOLEN_MAXSEG; 1279 to->to_mss = htons(to->to_mss); 1280 bcopy((u_char *)&to->to_mss, optp, sizeof(to->to_mss)); 1281 optp += sizeof(to->to_mss); 1282 break; 1283 case TOF_SCALE: 1284 while (!optlen || optlen % 2 != 1) { 1285 optlen += TCPOLEN_NOP; 1286 *optp++ = TCPOPT_NOP; 1287 } 1288 optlen += TCPOLEN_WINDOW; 1289 *optp++ = TCPOPT_WINDOW; 1290 *optp++ = TCPOLEN_WINDOW; 1291 *optp++ = to->to_wscale; 1292 break; 1293 case TOF_SACKPERM: 1294 while (optlen % 2) { 1295 optlen += TCPOLEN_NOP; 1296 *optp++ = TCPOPT_NOP; 1297 } 1298 optlen += TCPOLEN_SACK_PERMITTED; 1299 *optp++ = TCPOPT_SACK_PERMITTED; 1300 *optp++ = TCPOLEN_SACK_PERMITTED; 1301 break; 1302 case TOF_TS: 1303 while (!optlen || optlen % 4 != 2) { 1304 optlen += TCPOLEN_NOP; 1305 *optp++ = TCPOPT_NOP; 1306 } 1307 optlen += TCPOLEN_TIMESTAMP; 1308 *optp++ = TCPOPT_TIMESTAMP; 1309 *optp++ = TCPOLEN_TIMESTAMP; 1310 to->to_tsval = htonl(to->to_tsval); 1311 to->to_tsecr = htonl(to->to_tsecr); 1312 bcopy((u_char *)&to->to_tsval, optp, sizeof(to->to_tsval)); 1313 optp += sizeof(to->to_tsval); 1314 bcopy((u_char *)&to->to_tsecr, optp, sizeof(to->to_tsecr)); 1315 optp += sizeof(to->to_tsecr); 1316 break; 1317 case TOF_SIGNATURE: 1318 { 1319 int siglen = TCPOLEN_SIGNATURE - 2; 1320 1321 while (!optlen || optlen % 4 != 2) { 1322 optlen += TCPOLEN_NOP; 1323 *optp++ = TCPOPT_NOP; 1324 } 1325 if (MAX_TCPOPTLEN - optlen < TCPOLEN_SIGNATURE) 1326 continue; 1327 optlen += TCPOLEN_SIGNATURE; 1328 *optp++ = TCPOPT_SIGNATURE; 1329 *optp++ = TCPOLEN_SIGNATURE; 1330 to->to_signature = optp; 1331 while (siglen--) 1332 *optp++ = 0; 1333 break; 1334 } 1335 case TOF_SACK: 1336 { 1337 int sackblks = 0; 1338 struct sackblk *sack = (struct sackblk *)to->to_sacks; 1339 tcp_seq sack_seq; 1340 1341 while (!optlen || optlen % 4 != 2) { 1342 optlen += TCPOLEN_NOP; 1343 *optp++ = TCPOPT_NOP; 1344 } 1345 if (MAX_TCPOPTLEN - optlen < 2 + TCPOLEN_SACK) 1346 continue; 1347 optlen += TCPOLEN_SACKHDR; 1348 *optp++ = TCPOPT_SACK; 1349 sackblks = min(to->to_nsacks, 1350 (MAX_TCPOPTLEN - optlen) / TCPOLEN_SACK); 1351 *optp++ = TCPOLEN_SACKHDR + sackblks * TCPOLEN_SACK; 1352 while (sackblks--) { 1353 sack_seq = htonl(sack->start); 1354 bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq)); 1355 optp += sizeof(sack_seq); 1356 sack_seq = htonl(sack->end); 1357 bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq)); 1358 optp += sizeof(sack_seq); 1359 optlen += TCPOLEN_SACK; 1360 sack++; 1361 } 1362 tcpstat.tcps_sack_send_blocks++; 1363 break; 1364 } 1365 default: 1366 panic("%s: unknown TCP option type", __func__); 1367 break; 1368 } 1369 } 1370 1371 /* Terminate and pad TCP options to a 4 byte boundary. */ 1372 if (optlen % 4) { 1373 optlen += TCPOLEN_EOL; 1374 *optp++ = TCPOPT_EOL; 1375 } 1376 while (optlen % 4) { 1377 optlen += TCPOLEN_NOP; 1378 *optp++ = TCPOPT_NOP; 1379 } 1380 1381 KASSERT(optlen <= MAX_TCPOPTLEN, ("%s: TCP options too long", __func__)); 1382 return (optlen); 1383 } 1384