tcp_output.c revision 315514
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 */ 31 32#include <sys/cdefs.h> 33__FBSDID("$FreeBSD: stable/11/sys/netinet/tcp_output.c 315514 2017-03-18 22:04:20Z ae $"); 34 35#include "opt_inet.h" 36#include "opt_inet6.h" 37#include "opt_ipsec.h" 38#include "opt_tcpdebug.h" 39 40#include <sys/param.h> 41#include <sys/systm.h> 42#include <sys/domain.h> 43#include <sys/hhook.h> 44#include <sys/kernel.h> 45#include <sys/lock.h> 46#include <sys/mbuf.h> 47#include <sys/mutex.h> 48#include <sys/protosw.h> 49#include <sys/sdt.h> 50#include <sys/socket.h> 51#include <sys/socketvar.h> 52#include <sys/sysctl.h> 53 54#include <net/if.h> 55#include <net/route.h> 56#include <net/vnet.h> 57 58#include <netinet/in.h> 59#include <netinet/in_kdtrace.h> 60#include <netinet/in_systm.h> 61#include <netinet/ip.h> 62#include <netinet/in_pcb.h> 63#include <netinet/ip_var.h> 64#include <netinet/ip_options.h> 65#ifdef INET6 66#include <netinet6/in6_pcb.h> 67#include <netinet/ip6.h> 68#include <netinet6/ip6_var.h> 69#endif 70#ifdef TCP_RFC7413 71#include <netinet/tcp_fastopen.h> 72#endif 73#include <netinet/tcp.h> 74#define TCPOUTFLAGS 75#include <netinet/tcp_fsm.h> 76#include <netinet/tcp_seq.h> 77#include <netinet/tcp_timer.h> 78#include <netinet/tcp_var.h> 79#include <netinet/tcpip.h> 80#include <netinet/cc/cc.h> 81#ifdef TCPPCAP 82#include <netinet/tcp_pcap.h> 83#endif 84#ifdef TCPDEBUG 85#include <netinet/tcp_debug.h> 86#endif 87#ifdef TCP_OFFLOAD 88#include <netinet/tcp_offload.h> 89#endif 90 91#include <netipsec/ipsec_support.h> 92 93#include <machine/in_cksum.h> 94 95#include <security/mac/mac_framework.h> 96 97VNET_DEFINE(int, path_mtu_discovery) = 1; 98SYSCTL_INT(_net_inet_tcp, OID_AUTO, path_mtu_discovery, CTLFLAG_VNET | CTLFLAG_RW, 99 &VNET_NAME(path_mtu_discovery), 1, 100 "Enable Path MTU Discovery"); 101 102VNET_DEFINE(int, tcp_do_tso) = 1; 103#define V_tcp_do_tso VNET(tcp_do_tso) 104SYSCTL_INT(_net_inet_tcp, OID_AUTO, tso, CTLFLAG_VNET | CTLFLAG_RW, 105 &VNET_NAME(tcp_do_tso), 0, 106 "Enable TCP Segmentation Offload"); 107 108VNET_DEFINE(int, tcp_sendspace) = 1024*32; 109#define V_tcp_sendspace VNET(tcp_sendspace) 110SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_VNET | CTLFLAG_RW, 111 &VNET_NAME(tcp_sendspace), 0, "Initial send socket buffer size"); 112 113VNET_DEFINE(int, tcp_do_autosndbuf) = 1; 114#define V_tcp_do_autosndbuf VNET(tcp_do_autosndbuf) 115SYSCTL_INT(_net_inet_tcp, OID_AUTO, sendbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 116 &VNET_NAME(tcp_do_autosndbuf), 0, 117 "Enable automatic send buffer sizing"); 118 119VNET_DEFINE(int, tcp_autosndbuf_inc) = 8*1024; 120#define V_tcp_autosndbuf_inc VNET(tcp_autosndbuf_inc) 121SYSCTL_INT(_net_inet_tcp, OID_AUTO, sendbuf_inc, CTLFLAG_VNET | CTLFLAG_RW, 122 &VNET_NAME(tcp_autosndbuf_inc), 0, 123 "Incrementor step size of automatic send buffer"); 124 125VNET_DEFINE(int, tcp_autosndbuf_max) = 2*1024*1024; 126#define V_tcp_autosndbuf_max VNET(tcp_autosndbuf_max) 127SYSCTL_INT(_net_inet_tcp, OID_AUTO, sendbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 128 &VNET_NAME(tcp_autosndbuf_max), 0, 129 "Max size of automatic send buffer"); 130 131/* 132 * Make sure that either retransmit or persist timer is set for SYN, FIN and 133 * non-ACK. 134 */ 135#define TCP_XMIT_TIMER_ASSERT(tp, len, th_flags) \ 136 KASSERT(((len) == 0 && ((th_flags) & (TH_SYN | TH_FIN)) == 0) ||\ 137 tcp_timer_active((tp), TT_REXMT) || \ 138 tcp_timer_active((tp), TT_PERSIST), \ 139 ("neither rexmt nor persist timer is set")) 140 141static void inline hhook_run_tcp_est_out(struct tcpcb *tp, 142 struct tcphdr *th, struct tcpopt *to, 143 long len, int tso); 144static void inline cc_after_idle(struct tcpcb *tp); 145 146/* 147 * Wrapper for the TCP established output helper hook. 148 */ 149static void inline 150hhook_run_tcp_est_out(struct tcpcb *tp, struct tcphdr *th, 151 struct tcpopt *to, long len, int tso) 152{ 153 struct tcp_hhook_data hhook_data; 154 155 if (V_tcp_hhh[HHOOK_TCP_EST_OUT]->hhh_nhooks > 0) { 156 hhook_data.tp = tp; 157 hhook_data.th = th; 158 hhook_data.to = to; 159 hhook_data.len = len; 160 hhook_data.tso = tso; 161 162 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_OUT], &hhook_data, 163 tp->osd); 164 } 165} 166 167/* 168 * CC wrapper hook functions 169 */ 170static void inline 171cc_after_idle(struct tcpcb *tp) 172{ 173 INP_WLOCK_ASSERT(tp->t_inpcb); 174 175 if (CC_ALGO(tp)->after_idle != NULL) 176 CC_ALGO(tp)->after_idle(tp->ccv); 177} 178 179/* 180 * Tcp output routine: figure out what should be sent and send it. 181 */ 182int 183tcp_output(struct tcpcb *tp) 184{ 185 struct socket *so = tp->t_inpcb->inp_socket; 186 long len, recwin, sendwin; 187 int off, flags, error = 0; /* Keep compiler happy */ 188 struct mbuf *m; 189 struct ip *ip = NULL; 190 struct ipovly *ipov = NULL; 191 struct tcphdr *th; 192 u_char opt[TCP_MAXOLEN]; 193 unsigned ipoptlen, optlen, hdrlen; 194#if defined(IPSEC) || defined(IPSEC_SUPPORT) 195 unsigned ipsec_optlen = 0; 196#endif 197 int idle, sendalot; 198 int sack_rxmit, sack_bytes_rxmt; 199 struct sackhole *p; 200 int tso, mtu; 201 struct tcpopt to; 202#if 0 203 int maxburst = TCP_MAXBURST; 204#endif 205#ifdef INET6 206 struct ip6_hdr *ip6 = NULL; 207 int isipv6; 208 209 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 210#endif 211 212 INP_WLOCK_ASSERT(tp->t_inpcb); 213 214#ifdef TCP_OFFLOAD 215 if (tp->t_flags & TF_TOE) 216 return (tcp_offload_output(tp)); 217#endif 218 219#ifdef TCP_RFC7413 220 /* 221 * For TFO connections in SYN_RECEIVED, only allow the initial 222 * SYN|ACK and those sent by the retransmit timer. 223 */ 224 if ((tp->t_flags & TF_FASTOPEN) && 225 (tp->t_state == TCPS_SYN_RECEIVED) && 226 SEQ_GT(tp->snd_max, tp->snd_una) && /* initial SYN|ACK sent */ 227 (tp->snd_nxt != tp->snd_una)) /* not a retransmit */ 228 return (0); 229#endif 230 /* 231 * Determine length of data that should be transmitted, 232 * and flags that will be used. 233 * If there is some data or critical controls (SYN, RST) 234 * to send, then transmit; otherwise, investigate further. 235 */ 236 idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una); 237 if (idle && ticks - tp->t_rcvtime >= tp->t_rxtcur) 238 cc_after_idle(tp); 239 tp->t_flags &= ~TF_LASTIDLE; 240 if (idle) { 241 if (tp->t_flags & TF_MORETOCOME) { 242 tp->t_flags |= TF_LASTIDLE; 243 idle = 0; 244 } 245 } 246again: 247 /* 248 * If we've recently taken a timeout, snd_max will be greater than 249 * snd_nxt. There may be SACK information that allows us to avoid 250 * resending already delivered data. Adjust snd_nxt accordingly. 251 */ 252 if ((tp->t_flags & TF_SACK_PERMIT) && 253 SEQ_LT(tp->snd_nxt, tp->snd_max)) 254 tcp_sack_adjust(tp); 255 sendalot = 0; 256 tso = 0; 257 mtu = 0; 258 off = tp->snd_nxt - tp->snd_una; 259 sendwin = min(tp->snd_wnd, tp->snd_cwnd); 260 261 flags = tcp_outflags[tp->t_state]; 262 /* 263 * Send any SACK-generated retransmissions. If we're explicitly trying 264 * to send out new data (when sendalot is 1), bypass this function. 265 * If we retransmit in fast recovery mode, decrement snd_cwnd, since 266 * we're replacing a (future) new transmission with a retransmission 267 * now, and we previously incremented snd_cwnd in tcp_input(). 268 */ 269 /* 270 * Still in sack recovery , reset rxmit flag to zero. 271 */ 272 sack_rxmit = 0; 273 sack_bytes_rxmt = 0; 274 len = 0; 275 p = NULL; 276 if ((tp->t_flags & TF_SACK_PERMIT) && IN_FASTRECOVERY(tp->t_flags) && 277 (p = tcp_sack_output(tp, &sack_bytes_rxmt))) { 278 long cwin; 279 280 cwin = min(tp->snd_wnd, tp->snd_cwnd) - sack_bytes_rxmt; 281 if (cwin < 0) 282 cwin = 0; 283 /* Do not retransmit SACK segments beyond snd_recover */ 284 if (SEQ_GT(p->end, tp->snd_recover)) { 285 /* 286 * (At least) part of sack hole extends beyond 287 * snd_recover. Check to see if we can rexmit data 288 * for this hole. 289 */ 290 if (SEQ_GEQ(p->rxmit, tp->snd_recover)) { 291 /* 292 * Can't rexmit any more data for this hole. 293 * That data will be rexmitted in the next 294 * sack recovery episode, when snd_recover 295 * moves past p->rxmit. 296 */ 297 p = NULL; 298 goto after_sack_rexmit; 299 } else 300 /* Can rexmit part of the current hole */ 301 len = ((long)ulmin(cwin, 302 tp->snd_recover - p->rxmit)); 303 } else 304 len = ((long)ulmin(cwin, p->end - p->rxmit)); 305 off = p->rxmit - tp->snd_una; 306 KASSERT(off >= 0,("%s: sack block to the left of una : %d", 307 __func__, off)); 308 if (len > 0) { 309 sack_rxmit = 1; 310 sendalot = 1; 311 TCPSTAT_INC(tcps_sack_rexmits); 312 TCPSTAT_ADD(tcps_sack_rexmit_bytes, 313 min(len, tp->t_maxseg)); 314 } 315 } 316after_sack_rexmit: 317 /* 318 * Get standard flags, and add SYN or FIN if requested by 'hidden' 319 * state flags. 320 */ 321 if (tp->t_flags & TF_NEEDFIN) 322 flags |= TH_FIN; 323 if (tp->t_flags & TF_NEEDSYN) 324 flags |= TH_SYN; 325 326 SOCKBUF_LOCK(&so->so_snd); 327 /* 328 * If in persist timeout with window of 0, send 1 byte. 329 * Otherwise, if window is small but nonzero 330 * and timer expired, we will send what we can 331 * and go to transmit state. 332 */ 333 if (tp->t_flags & TF_FORCEDATA) { 334 if (sendwin == 0) { 335 /* 336 * If we still have some data to send, then 337 * clear the FIN bit. Usually this would 338 * happen below when it realizes that we 339 * aren't sending all the data. However, 340 * if we have exactly 1 byte of unsent data, 341 * then it won't clear the FIN bit below, 342 * and if we are in persist state, we wind 343 * up sending the packet without recording 344 * that we sent the FIN bit. 345 * 346 * We can't just blindly clear the FIN bit, 347 * because if we don't have any more data 348 * to send then the probe will be the FIN 349 * itself. 350 */ 351 if (off < sbused(&so->so_snd)) 352 flags &= ~TH_FIN; 353 sendwin = 1; 354 } else { 355 tcp_timer_activate(tp, TT_PERSIST, 0); 356 tp->t_rxtshift = 0; 357 } 358 } 359 360 /* 361 * If snd_nxt == snd_max and we have transmitted a FIN, the 362 * offset will be > 0 even if so_snd.sb_cc is 0, resulting in 363 * a negative length. This can also occur when TCP opens up 364 * its congestion window while receiving additional duplicate 365 * acks after fast-retransmit because TCP will reset snd_nxt 366 * to snd_max after the fast-retransmit. 367 * 368 * In the normal retransmit-FIN-only case, however, snd_nxt will 369 * be set to snd_una, the offset will be 0, and the length may 370 * wind up 0. 371 * 372 * If sack_rxmit is true we are retransmitting from the scoreboard 373 * in which case len is already set. 374 */ 375 if (sack_rxmit == 0) { 376 if (sack_bytes_rxmt == 0) 377 len = ((long)ulmin(sbavail(&so->so_snd), sendwin) - 378 off); 379 else { 380 long cwin; 381 382 /* 383 * We are inside of a SACK recovery episode and are 384 * sending new data, having retransmitted all the 385 * data possible in the scoreboard. 386 */ 387 len = ((long)ulmin(sbavail(&so->so_snd), tp->snd_wnd) - 388 off); 389 /* 390 * Don't remove this (len > 0) check ! 391 * We explicitly check for len > 0 here (although it 392 * isn't really necessary), to work around a gcc 393 * optimization issue - to force gcc to compute 394 * len above. Without this check, the computation 395 * of len is bungled by the optimizer. 396 */ 397 if (len > 0) { 398 cwin = tp->snd_cwnd - 399 (tp->snd_nxt - tp->sack_newdata) - 400 sack_bytes_rxmt; 401 if (cwin < 0) 402 cwin = 0; 403 len = lmin(len, cwin); 404 } 405 } 406 } 407 408 /* 409 * Lop off SYN bit if it has already been sent. However, if this 410 * is SYN-SENT state and if segment contains data and if we don't 411 * know that foreign host supports TAO, suppress sending segment. 412 */ 413 if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) { 414 if (tp->t_state != TCPS_SYN_RECEIVED) 415 flags &= ~TH_SYN; 416#ifdef TCP_RFC7413 417 /* 418 * When sending additional segments following a TFO SYN|ACK, 419 * do not include the SYN bit. 420 */ 421 if ((tp->t_flags & TF_FASTOPEN) && 422 (tp->t_state == TCPS_SYN_RECEIVED)) 423 flags &= ~TH_SYN; 424#endif 425 off--, len++; 426 } 427 428 /* 429 * Be careful not to send data and/or FIN on SYN segments. 430 * This measure is needed to prevent interoperability problems 431 * with not fully conformant TCP implementations. 432 */ 433 if ((flags & TH_SYN) && (tp->t_flags & TF_NOOPT)) { 434 len = 0; 435 flags &= ~TH_FIN; 436 } 437 438#ifdef TCP_RFC7413 439 /* 440 * When retransmitting SYN|ACK on a passively-created TFO socket, 441 * don't include data, as the presence of data may have caused the 442 * original SYN|ACK to have been dropped by a middlebox. 443 */ 444 if ((tp->t_flags & TF_FASTOPEN) && 445 (((tp->t_state == TCPS_SYN_RECEIVED) && (tp->t_rxtshift > 0)) || 446 (flags & TH_RST))) 447 len = 0; 448#endif 449 if (len <= 0) { 450 /* 451 * If FIN has been sent but not acked, 452 * but we haven't been called to retransmit, 453 * len will be < 0. Otherwise, window shrank 454 * after we sent into it. If window shrank to 0, 455 * cancel pending retransmit, pull snd_nxt back 456 * to (closed) window, and set the persist timer 457 * if it isn't already going. If the window didn't 458 * close completely, just wait for an ACK. 459 * 460 * We also do a general check here to ensure that 461 * we will set the persist timer when we have data 462 * to send, but a 0-byte window. This makes sure 463 * the persist timer is set even if the packet 464 * hits one of the "goto send" lines below. 465 */ 466 len = 0; 467 if ((sendwin == 0) && (TCPS_HAVEESTABLISHED(tp->t_state)) && 468 (off < (int) sbavail(&so->so_snd))) { 469 tcp_timer_activate(tp, TT_REXMT, 0); 470 tp->t_rxtshift = 0; 471 tp->snd_nxt = tp->snd_una; 472 if (!tcp_timer_active(tp, TT_PERSIST)) 473 tcp_setpersist(tp); 474 } 475 } 476 477 /* len will be >= 0 after this point. */ 478 KASSERT(len >= 0, ("[%s:%d]: len < 0", __func__, __LINE__)); 479 480 /* 481 * Automatic sizing of send socket buffer. Often the send buffer 482 * size is not optimally adjusted to the actual network conditions 483 * at hand (delay bandwidth product). Setting the buffer size too 484 * small limits throughput on links with high bandwidth and high 485 * delay (eg. trans-continental/oceanic links). Setting the 486 * buffer size too big consumes too much real kernel memory, 487 * especially with many connections on busy servers. 488 * 489 * The criteria to step up the send buffer one notch are: 490 * 1. receive window of remote host is larger than send buffer 491 * (with a fudge factor of 5/4th); 492 * 2. send buffer is filled to 7/8th with data (so we actually 493 * have data to make use of it); 494 * 3. send buffer fill has not hit maximal automatic size; 495 * 4. our send window (slow start and cogestion controlled) is 496 * larger than sent but unacknowledged data in send buffer. 497 * 498 * The remote host receive window scaling factor may limit the 499 * growing of the send buffer before it reaches its allowed 500 * maximum. 501 * 502 * It scales directly with slow start or congestion window 503 * and does at most one step per received ACK. This fast 504 * scaling has the drawback of growing the send buffer beyond 505 * what is strictly necessary to make full use of a given 506 * delay*bandwidth product. However testing has shown this not 507 * to be much of an problem. At worst we are trading wasting 508 * of available bandwidth (the non-use of it) for wasting some 509 * socket buffer memory. 510 * 511 * TODO: Shrink send buffer during idle periods together 512 * with congestion window. Requires another timer. Has to 513 * wait for upcoming tcp timer rewrite. 514 * 515 * XXXGL: should there be used sbused() or sbavail()? 516 */ 517 if (V_tcp_do_autosndbuf && so->so_snd.sb_flags & SB_AUTOSIZE) { 518 if ((tp->snd_wnd / 4 * 5) >= so->so_snd.sb_hiwat && 519 sbused(&so->so_snd) >= (so->so_snd.sb_hiwat / 8 * 7) && 520 sbused(&so->so_snd) < V_tcp_autosndbuf_max && 521 sendwin >= (sbused(&so->so_snd) - 522 (tp->snd_nxt - tp->snd_una))) { 523 if (!sbreserve_locked(&so->so_snd, 524 min(so->so_snd.sb_hiwat + V_tcp_autosndbuf_inc, 525 V_tcp_autosndbuf_max), so, curthread)) 526 so->so_snd.sb_flags &= ~SB_AUTOSIZE; 527 } 528 } 529 530 /* 531 * Decide if we can use TCP Segmentation Offloading (if supported by 532 * hardware). 533 * 534 * TSO may only be used if we are in a pure bulk sending state. The 535 * presence of TCP-MD5, SACK retransmits, SACK advertizements and 536 * IP options prevent using TSO. With TSO the TCP header is the same 537 * (except for the sequence number) for all generated packets. This 538 * makes it impossible to transmit any options which vary per generated 539 * segment or packet. 540 * 541 * IPv4 handling has a clear separation of ip options and ip header 542 * flags while IPv6 combines both in in6p_outputopts. ip6_optlen() does 543 * the right thing below to provide length of just ip options and thus 544 * checking for ipoptlen is enough to decide if ip options are present. 545 */ 546#if defined(IPSEC) || defined(IPSEC_SUPPORT) 547 /* 548 * Pre-calculate here as we save another lookup into the darknesses 549 * of IPsec that way and can actually decide if TSO is ok. 550 */ 551#ifdef INET6 552 if (isipv6 && IPSEC_ENABLED(ipv6)) 553 ipsec_optlen = IPSEC_HDRSIZE(ipv6, tp->t_inpcb); 554#ifdef INET 555 else 556#endif 557#endif /* INET6 */ 558#ifdef INET 559 if (IPSEC_ENABLED(ipv4)) 560 ipsec_optlen = IPSEC_HDRSIZE(ipv4, tp->t_inpcb); 561#endif /* INET */ 562#endif /* IPSEC */ 563#ifdef INET6 564 if (isipv6) 565 ipoptlen = ip6_optlen(tp->t_inpcb); 566 else 567#endif 568 if (tp->t_inpcb->inp_options) 569 ipoptlen = tp->t_inpcb->inp_options->m_len - 570 offsetof(struct ipoption, ipopt_list); 571 else 572 ipoptlen = 0; 573#if defined(IPSEC) || defined(IPSEC_SUPPORT) 574 ipoptlen += ipsec_optlen; 575#endif 576 577 if ((tp->t_flags & TF_TSO) && V_tcp_do_tso && len > tp->t_maxseg && 578 ((tp->t_flags & TF_SIGNATURE) == 0) && 579 tp->rcv_numsacks == 0 && sack_rxmit == 0 && 580 ipoptlen == 0) 581 tso = 1; 582 583 if (sack_rxmit) { 584 if (SEQ_LT(p->rxmit + len, tp->snd_una + sbused(&so->so_snd))) 585 flags &= ~TH_FIN; 586 } else { 587 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + 588 sbused(&so->so_snd))) 589 flags &= ~TH_FIN; 590 } 591 592 recwin = sbspace(&so->so_rcv); 593 594 /* 595 * Sender silly window avoidance. We transmit under the following 596 * conditions when len is non-zero: 597 * 598 * - We have a full segment (or more with TSO) 599 * - This is the last buffer in a write()/send() and we are 600 * either idle or running NODELAY 601 * - we've timed out (e.g. persist timer) 602 * - we have more then 1/2 the maximum send window's worth of 603 * data (receiver may be limited the window size) 604 * - we need to retransmit 605 */ 606 if (len) { 607 if (len >= tp->t_maxseg) 608 goto send; 609 /* 610 * NOTE! on localhost connections an 'ack' from the remote 611 * end may occur synchronously with the output and cause 612 * us to flush a buffer queued with moretocome. XXX 613 * 614 * note: the len + off check is almost certainly unnecessary. 615 */ 616 if (!(tp->t_flags & TF_MORETOCOME) && /* normal case */ 617 (idle || (tp->t_flags & TF_NODELAY)) && 618 len + off >= sbavail(&so->so_snd) && 619 (tp->t_flags & TF_NOPUSH) == 0) { 620 goto send; 621 } 622 if (tp->t_flags & TF_FORCEDATA) /* typ. timeout case */ 623 goto send; 624 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0) 625 goto send; 626 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */ 627 goto send; 628 if (sack_rxmit) 629 goto send; 630 } 631 632 /* 633 * Sending of standalone window updates. 634 * 635 * Window updates are important when we close our window due to a 636 * full socket buffer and are opening it again after the application 637 * reads data from it. Once the window has opened again and the 638 * remote end starts to send again the ACK clock takes over and 639 * provides the most current window information. 640 * 641 * We must avoid the silly window syndrome whereas every read 642 * from the receive buffer, no matter how small, causes a window 643 * update to be sent. We also should avoid sending a flurry of 644 * window updates when the socket buffer had queued a lot of data 645 * and the application is doing small reads. 646 * 647 * Prevent a flurry of pointless window updates by only sending 648 * an update when we can increase the advertized window by more 649 * than 1/4th of the socket buffer capacity. When the buffer is 650 * getting full or is very small be more aggressive and send an 651 * update whenever we can increase by two mss sized segments. 652 * In all other situations the ACK's to new incoming data will 653 * carry further window increases. 654 * 655 * Don't send an independent window update if a delayed 656 * ACK is pending (it will get piggy-backed on it) or the 657 * remote side already has done a half-close and won't send 658 * more data. Skip this if the connection is in T/TCP 659 * half-open state. 660 */ 661 if (recwin > 0 && !(tp->t_flags & TF_NEEDSYN) && 662 !(tp->t_flags & TF_DELACK) && 663 !TCPS_HAVERCVDFIN(tp->t_state)) { 664 /* 665 * "adv" is the amount we could increase the window, 666 * taking into account that we are limited by 667 * TCP_MAXWIN << tp->rcv_scale. 668 */ 669 long adv; 670 int oldwin; 671 672 adv = min(recwin, (long)TCP_MAXWIN << tp->rcv_scale); 673 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) { 674 oldwin = (tp->rcv_adv - tp->rcv_nxt); 675 adv -= oldwin; 676 } else 677 oldwin = 0; 678 679 /* 680 * If the new window size ends up being the same as the old 681 * size when it is scaled, then don't force a window update. 682 */ 683 if (oldwin >> tp->rcv_scale == (adv + oldwin) >> tp->rcv_scale) 684 goto dontupdate; 685 686 if (adv >= (long)(2 * tp->t_maxseg) && 687 (adv >= (long)(so->so_rcv.sb_hiwat / 4) || 688 recwin <= (long)(so->so_rcv.sb_hiwat / 8) || 689 so->so_rcv.sb_hiwat <= 8 * tp->t_maxseg)) 690 goto send; 691 } 692dontupdate: 693 694 /* 695 * Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW 696 * is also a catch-all for the retransmit timer timeout case. 697 */ 698 if (tp->t_flags & TF_ACKNOW) 699 goto send; 700 if ((flags & TH_RST) || 701 ((flags & TH_SYN) && (tp->t_flags & TF_NEEDSYN) == 0)) 702 goto send; 703 if (SEQ_GT(tp->snd_up, tp->snd_una)) 704 goto send; 705 /* 706 * If our state indicates that FIN should be sent 707 * and we have not yet done so, then we need to send. 708 */ 709 if (flags & TH_FIN && 710 ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una)) 711 goto send; 712 /* 713 * In SACK, it is possible for tcp_output to fail to send a segment 714 * after the retransmission timer has been turned off. Make sure 715 * that the retransmission timer is set. 716 */ 717 if ((tp->t_flags & TF_SACK_PERMIT) && 718 SEQ_GT(tp->snd_max, tp->snd_una) && 719 !tcp_timer_active(tp, TT_REXMT) && 720 !tcp_timer_active(tp, TT_PERSIST)) { 721 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 722 goto just_return; 723 } 724 /* 725 * TCP window updates are not reliable, rather a polling protocol 726 * using ``persist'' packets is used to insure receipt of window 727 * updates. The three ``states'' for the output side are: 728 * idle not doing retransmits or persists 729 * persisting to move a small or zero window 730 * (re)transmitting and thereby not persisting 731 * 732 * tcp_timer_active(tp, TT_PERSIST) 733 * is true when we are in persist state. 734 * (tp->t_flags & TF_FORCEDATA) 735 * is set when we are called to send a persist packet. 736 * tcp_timer_active(tp, TT_REXMT) 737 * is set when we are retransmitting 738 * The output side is idle when both timers are zero. 739 * 740 * If send window is too small, there is data to transmit, and no 741 * retransmit or persist is pending, then go to persist state. 742 * If nothing happens soon, send when timer expires: 743 * if window is nonzero, transmit what we can, 744 * otherwise force out a byte. 745 */ 746 if (sbavail(&so->so_snd) && !tcp_timer_active(tp, TT_REXMT) && 747 !tcp_timer_active(tp, TT_PERSIST)) { 748 tp->t_rxtshift = 0; 749 tcp_setpersist(tp); 750 } 751 752 /* 753 * No reason to send a segment, just return. 754 */ 755just_return: 756 SOCKBUF_UNLOCK(&so->so_snd); 757 return (0); 758 759send: 760 SOCKBUF_LOCK_ASSERT(&so->so_snd); 761 if (len > 0) { 762 if (len >= tp->t_maxseg) 763 tp->t_flags2 |= TF2_PLPMTU_MAXSEGSNT; 764 else 765 tp->t_flags2 &= ~TF2_PLPMTU_MAXSEGSNT; 766 } 767 /* 768 * Before ESTABLISHED, force sending of initial options 769 * unless TCP set not to do any options. 770 * NOTE: we assume that the IP/TCP header plus TCP options 771 * always fit in a single mbuf, leaving room for a maximum 772 * link header, i.e. 773 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES 774 */ 775 optlen = 0; 776#ifdef INET6 777 if (isipv6) 778 hdrlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 779 else 780#endif 781 hdrlen = sizeof (struct tcpiphdr); 782 783 /* 784 * Compute options for segment. 785 * We only have to care about SYN and established connection 786 * segments. Options for SYN-ACK segments are handled in TCP 787 * syncache. 788 */ 789 to.to_flags = 0; 790 if ((tp->t_flags & TF_NOOPT) == 0) { 791 /* Maximum segment size. */ 792 if (flags & TH_SYN) { 793 tp->snd_nxt = tp->iss; 794 to.to_mss = tcp_mssopt(&tp->t_inpcb->inp_inc); 795 to.to_flags |= TOF_MSS; 796#ifdef TCP_RFC7413 797 /* 798 * Only include the TFO option on the first 799 * transmission of the SYN|ACK on a 800 * passively-created TFO socket, as the presence of 801 * the TFO option may have caused the original 802 * SYN|ACK to have been dropped by a middlebox. 803 */ 804 if ((tp->t_flags & TF_FASTOPEN) && 805 (tp->t_state == TCPS_SYN_RECEIVED) && 806 (tp->t_rxtshift == 0)) { 807 to.to_tfo_len = TCP_FASTOPEN_COOKIE_LEN; 808 to.to_tfo_cookie = (u_char *)&tp->t_tfo_cookie; 809 to.to_flags |= TOF_FASTOPEN; 810 } 811#endif 812 } 813 /* Window scaling. */ 814 if ((flags & TH_SYN) && (tp->t_flags & TF_REQ_SCALE)) { 815 to.to_wscale = tp->request_r_scale; 816 to.to_flags |= TOF_SCALE; 817 } 818 /* Timestamps. */ 819 if ((tp->t_flags & TF_RCVD_TSTMP) || 820 ((flags & TH_SYN) && (tp->t_flags & TF_REQ_TSTMP))) { 821 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 822 to.to_tsecr = tp->ts_recent; 823 to.to_flags |= TOF_TS; 824 /* Set receive buffer autosizing timestamp. */ 825 if (tp->rfbuf_ts == 0 && 826 (so->so_rcv.sb_flags & SB_AUTOSIZE)) 827 tp->rfbuf_ts = tcp_ts_getticks(); 828 } 829 /* Selective ACK's. */ 830 if (tp->t_flags & TF_SACK_PERMIT) { 831 if (flags & TH_SYN) 832 to.to_flags |= TOF_SACKPERM; 833 else if (TCPS_HAVEESTABLISHED(tp->t_state) && 834 (tp->t_flags & TF_SACK_PERMIT) && 835 tp->rcv_numsacks > 0) { 836 to.to_flags |= TOF_SACK; 837 to.to_nsacks = tp->rcv_numsacks; 838 to.to_sacks = (u_char *)tp->sackblks; 839 } 840 } 841#if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 842 /* TCP-MD5 (RFC2385). */ 843 /* 844 * Check that TCP_MD5SIG is enabled in tcpcb to 845 * account the size needed to set this TCP option. 846 */ 847 if (tp->t_flags & TF_SIGNATURE) 848 to.to_flags |= TOF_SIGNATURE; 849#endif /* TCP_SIGNATURE */ 850 851 /* Processing the options. */ 852 hdrlen += optlen = tcp_addoptions(&to, opt); 853 } 854 855 /* 856 * Adjust data length if insertion of options will 857 * bump the packet length beyond the t_maxseg length. 858 * Clear the FIN bit because we cut off the tail of 859 * the segment. 860 */ 861 if (len + optlen + ipoptlen > tp->t_maxseg) { 862 flags &= ~TH_FIN; 863 864 if (tso) { 865 u_int if_hw_tsomax; 866 u_int if_hw_tsomaxsegcount; 867 u_int if_hw_tsomaxsegsize; 868 struct mbuf *mb; 869 u_int moff; 870 int max_len; 871 872 /* extract TSO information */ 873 if_hw_tsomax = tp->t_tsomax; 874 if_hw_tsomaxsegcount = tp->t_tsomaxsegcount; 875 if_hw_tsomaxsegsize = tp->t_tsomaxsegsize; 876 877 /* 878 * Limit a TSO burst to prevent it from 879 * overflowing or exceeding the maximum length 880 * allowed by the network interface: 881 */ 882 KASSERT(ipoptlen == 0, 883 ("%s: TSO can't do IP options", __func__)); 884 885 /* 886 * Check if we should limit by maximum payload 887 * length: 888 */ 889 if (if_hw_tsomax != 0) { 890 /* compute maximum TSO length */ 891 max_len = (if_hw_tsomax - hdrlen - 892 max_linkhdr); 893 if (max_len <= 0) { 894 len = 0; 895 } else if (len > max_len) { 896 sendalot = 1; 897 len = max_len; 898 } 899 } 900 901 /* 902 * Check if we should limit by maximum segment 903 * size and count: 904 */ 905 if (if_hw_tsomaxsegcount != 0 && 906 if_hw_tsomaxsegsize != 0) { 907 /* 908 * Subtract one segment for the LINK 909 * and TCP/IP headers mbuf that will 910 * be prepended to this mbuf chain 911 * after the code in this section 912 * limits the number of mbufs in the 913 * chain to if_hw_tsomaxsegcount. 914 */ 915 if_hw_tsomaxsegcount -= 1; 916 max_len = 0; 917 mb = sbsndmbuf(&so->so_snd, off, &moff); 918 919 while (mb != NULL && max_len < len) { 920 u_int mlen; 921 u_int frags; 922 923 /* 924 * Get length of mbuf fragment 925 * and how many hardware frags, 926 * rounded up, it would use: 927 */ 928 mlen = (mb->m_len - moff); 929 frags = howmany(mlen, 930 if_hw_tsomaxsegsize); 931 932 /* Handle special case: Zero Length Mbuf */ 933 if (frags == 0) 934 frags = 1; 935 936 /* 937 * Check if the fragment limit 938 * will be reached or exceeded: 939 */ 940 if (frags >= if_hw_tsomaxsegcount) { 941 max_len += min(mlen, 942 if_hw_tsomaxsegcount * 943 if_hw_tsomaxsegsize); 944 break; 945 } 946 max_len += mlen; 947 if_hw_tsomaxsegcount -= frags; 948 moff = 0; 949 mb = mb->m_next; 950 } 951 if (max_len <= 0) { 952 len = 0; 953 } else if (len > max_len) { 954 sendalot = 1; 955 len = max_len; 956 } 957 } 958 959 /* 960 * Prevent the last segment from being 961 * fractional unless the send sockbuf can be 962 * emptied: 963 */ 964 max_len = (tp->t_maxseg - optlen); 965 if ((off + len) < sbavail(&so->so_snd)) { 966 moff = len % max_len; 967 if (moff != 0) { 968 len -= moff; 969 sendalot = 1; 970 } 971 } 972 973 /* 974 * In case there are too many small fragments 975 * don't use TSO: 976 */ 977 if (len <= max_len) { 978 len = max_len; 979 sendalot = 1; 980 tso = 0; 981 } 982 983 /* 984 * Send the FIN in a separate segment 985 * after the bulk sending is done. 986 * We don't trust the TSO implementations 987 * to clear the FIN flag on all but the 988 * last segment. 989 */ 990 if (tp->t_flags & TF_NEEDFIN) 991 sendalot = 1; 992 993 } else { 994 len = tp->t_maxseg - optlen - ipoptlen; 995 sendalot = 1; 996 } 997 } else 998 tso = 0; 999 1000 KASSERT(len + hdrlen + ipoptlen <= IP_MAXPACKET, 1001 ("%s: len > IP_MAXPACKET", __func__)); 1002 1003/*#ifdef DIAGNOSTIC*/ 1004#ifdef INET6 1005 if (max_linkhdr + hdrlen > MCLBYTES) 1006#else 1007 if (max_linkhdr + hdrlen > MHLEN) 1008#endif 1009 panic("tcphdr too big"); 1010/*#endif*/ 1011 1012 /* 1013 * This KASSERT is here to catch edge cases at a well defined place. 1014 * Before, those had triggered (random) panic conditions further down. 1015 */ 1016 KASSERT(len >= 0, ("[%s:%d]: len < 0", __func__, __LINE__)); 1017 1018 /* 1019 * Grab a header mbuf, attaching a copy of data to 1020 * be transmitted, and initialize the header from 1021 * the template for sends on this connection. 1022 */ 1023 if (len) { 1024 struct mbuf *mb; 1025 u_int moff; 1026 1027 if ((tp->t_flags & TF_FORCEDATA) && len == 1) 1028 TCPSTAT_INC(tcps_sndprobe); 1029 else if (SEQ_LT(tp->snd_nxt, tp->snd_max) || sack_rxmit) { 1030 tp->t_sndrexmitpack++; 1031 TCPSTAT_INC(tcps_sndrexmitpack); 1032 TCPSTAT_ADD(tcps_sndrexmitbyte, len); 1033 } else { 1034 TCPSTAT_INC(tcps_sndpack); 1035 TCPSTAT_ADD(tcps_sndbyte, len); 1036 } 1037#ifdef INET6 1038 if (MHLEN < hdrlen + max_linkhdr) 1039 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 1040 else 1041#endif 1042 m = m_gethdr(M_NOWAIT, MT_DATA); 1043 1044 if (m == NULL) { 1045 SOCKBUF_UNLOCK(&so->so_snd); 1046 error = ENOBUFS; 1047 sack_rxmit = 0; 1048 goto out; 1049 } 1050 1051 m->m_data += max_linkhdr; 1052 m->m_len = hdrlen; 1053 1054 /* 1055 * Start the m_copy functions from the closest mbuf 1056 * to the offset in the socket buffer chain. 1057 */ 1058 mb = sbsndptr(&so->so_snd, off, len, &moff); 1059 1060 if (len <= MHLEN - hdrlen - max_linkhdr) { 1061 m_copydata(mb, moff, (int)len, 1062 mtod(m, caddr_t) + hdrlen); 1063 m->m_len += len; 1064 } else { 1065 m->m_next = m_copy(mb, moff, (int)len); 1066 if (m->m_next == NULL) { 1067 SOCKBUF_UNLOCK(&so->so_snd); 1068 (void) m_free(m); 1069 error = ENOBUFS; 1070 sack_rxmit = 0; 1071 goto out; 1072 } 1073 } 1074 1075 /* 1076 * If we're sending everything we've got, set PUSH. 1077 * (This will keep happy those implementations which only 1078 * give data to the user when a buffer fills or 1079 * a PUSH comes in.) 1080 */ 1081 if ((off + len == sbused(&so->so_snd)) && !(flags & TH_SYN)) 1082 flags |= TH_PUSH; 1083 SOCKBUF_UNLOCK(&so->so_snd); 1084 } else { 1085 SOCKBUF_UNLOCK(&so->so_snd); 1086 if (tp->t_flags & TF_ACKNOW) 1087 TCPSTAT_INC(tcps_sndacks); 1088 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 1089 TCPSTAT_INC(tcps_sndctrl); 1090 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 1091 TCPSTAT_INC(tcps_sndurg); 1092 else 1093 TCPSTAT_INC(tcps_sndwinup); 1094 1095 m = m_gethdr(M_NOWAIT, MT_DATA); 1096 if (m == NULL) { 1097 error = ENOBUFS; 1098 sack_rxmit = 0; 1099 goto out; 1100 } 1101#ifdef INET6 1102 if (isipv6 && (MHLEN < hdrlen + max_linkhdr) && 1103 MHLEN >= hdrlen) { 1104 M_ALIGN(m, hdrlen); 1105 } else 1106#endif 1107 m->m_data += max_linkhdr; 1108 m->m_len = hdrlen; 1109 } 1110 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); 1111 m->m_pkthdr.rcvif = (struct ifnet *)0; 1112#ifdef MAC 1113 mac_inpcb_create_mbuf(tp->t_inpcb, m); 1114#endif 1115#ifdef INET6 1116 if (isipv6) { 1117 ip6 = mtod(m, struct ip6_hdr *); 1118 th = (struct tcphdr *)(ip6 + 1); 1119 tcpip_fillheaders(tp->t_inpcb, ip6, th); 1120 } else 1121#endif /* INET6 */ 1122 { 1123 ip = mtod(m, struct ip *); 1124 ipov = (struct ipovly *)ip; 1125 th = (struct tcphdr *)(ip + 1); 1126 tcpip_fillheaders(tp->t_inpcb, ip, th); 1127 } 1128 1129 /* 1130 * Fill in fields, remembering maximum advertised 1131 * window for use in delaying messages about window sizes. 1132 * If resending a FIN, be sure not to use a new sequence number. 1133 */ 1134 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN && 1135 tp->snd_nxt == tp->snd_max) 1136 tp->snd_nxt--; 1137 /* 1138 * If we are starting a connection, send ECN setup 1139 * SYN packet. If we are on a retransmit, we may 1140 * resend those bits a number of times as per 1141 * RFC 3168. 1142 */ 1143 if (tp->t_state == TCPS_SYN_SENT && V_tcp_do_ecn == 1) { 1144 if (tp->t_rxtshift >= 1) { 1145 if (tp->t_rxtshift <= V_tcp_ecn_maxretries) 1146 flags |= TH_ECE|TH_CWR; 1147 } else 1148 flags |= TH_ECE|TH_CWR; 1149 } 1150 1151 if (tp->t_state == TCPS_ESTABLISHED && 1152 (tp->t_flags & TF_ECN_PERMIT)) { 1153 /* 1154 * If the peer has ECN, mark data packets with 1155 * ECN capable transmission (ECT). 1156 * Ignore pure ack packets, retransmissions and window probes. 1157 */ 1158 if (len > 0 && SEQ_GEQ(tp->snd_nxt, tp->snd_max) && 1159 !((tp->t_flags & TF_FORCEDATA) && len == 1)) { 1160#ifdef INET6 1161 if (isipv6) 1162 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 1163 else 1164#endif 1165 ip->ip_tos |= IPTOS_ECN_ECT0; 1166 TCPSTAT_INC(tcps_ecn_ect0); 1167 } 1168 1169 /* 1170 * Reply with proper ECN notifications. 1171 */ 1172 if (tp->t_flags & TF_ECN_SND_CWR) { 1173 flags |= TH_CWR; 1174 tp->t_flags &= ~TF_ECN_SND_CWR; 1175 } 1176 if (tp->t_flags & TF_ECN_SND_ECE) 1177 flags |= TH_ECE; 1178 } 1179 1180 /* 1181 * If we are doing retransmissions, then snd_nxt will 1182 * not reflect the first unsent octet. For ACK only 1183 * packets, we do not want the sequence number of the 1184 * retransmitted packet, we want the sequence number 1185 * of the next unsent octet. So, if there is no data 1186 * (and no SYN or FIN), use snd_max instead of snd_nxt 1187 * when filling in ti_seq. But if we are in persist 1188 * state, snd_max might reflect one byte beyond the 1189 * right edge of the window, so use snd_nxt in that 1190 * case, since we know we aren't doing a retransmission. 1191 * (retransmit and persist are mutually exclusive...) 1192 */ 1193 if (sack_rxmit == 0) { 1194 if (len || (flags & (TH_SYN|TH_FIN)) || 1195 tcp_timer_active(tp, TT_PERSIST)) 1196 th->th_seq = htonl(tp->snd_nxt); 1197 else 1198 th->th_seq = htonl(tp->snd_max); 1199 } else { 1200 th->th_seq = htonl(p->rxmit); 1201 p->rxmit += len; 1202 tp->sackhint.sack_bytes_rexmit += len; 1203 } 1204 th->th_ack = htonl(tp->rcv_nxt); 1205 if (optlen) { 1206 bcopy(opt, th + 1, optlen); 1207 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1208 } 1209 th->th_flags = flags; 1210 /* 1211 * Calculate receive window. Don't shrink window, 1212 * but avoid silly window syndrome. 1213 */ 1214 if (recwin < (long)(so->so_rcv.sb_hiwat / 4) && 1215 recwin < (long)tp->t_maxseg) 1216 recwin = 0; 1217 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt) && 1218 recwin < (long)(tp->rcv_adv - tp->rcv_nxt)) 1219 recwin = (long)(tp->rcv_adv - tp->rcv_nxt); 1220 if (recwin > (long)TCP_MAXWIN << tp->rcv_scale) 1221 recwin = (long)TCP_MAXWIN << tp->rcv_scale; 1222 1223 /* 1224 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1225 * or <SYN,ACK>) segment itself is never scaled. The <SYN,ACK> 1226 * case is handled in syncache. 1227 */ 1228 if (flags & TH_SYN) 1229 th->th_win = htons((u_short) 1230 (min(sbspace(&so->so_rcv), TCP_MAXWIN))); 1231 else 1232 th->th_win = htons((u_short)(recwin >> tp->rcv_scale)); 1233 1234 /* 1235 * Adjust the RXWIN0SENT flag - indicate that we have advertised 1236 * a 0 window. This may cause the remote transmitter to stall. This 1237 * flag tells soreceive() to disable delayed acknowledgements when 1238 * draining the buffer. This can occur if the receiver is attempting 1239 * to read more data than can be buffered prior to transmitting on 1240 * the connection. 1241 */ 1242 if (th->th_win == 0) { 1243 tp->t_sndzerowin++; 1244 tp->t_flags |= TF_RXWIN0SENT; 1245 } else 1246 tp->t_flags &= ~TF_RXWIN0SENT; 1247 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 1248 th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt)); 1249 th->th_flags |= TH_URG; 1250 } else 1251 /* 1252 * If no urgent pointer to send, then we pull 1253 * the urgent pointer to the left edge of the send window 1254 * so that it doesn't drift into the send window on sequence 1255 * number wraparound. 1256 */ 1257 tp->snd_up = tp->snd_una; /* drag it along */ 1258 1259 /* 1260 * Put TCP length in extended header, and then 1261 * checksum extended header and data. 1262 */ 1263 m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */ 1264 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1265 1266#if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1267 if (to.to_flags & TOF_SIGNATURE) { 1268 /* 1269 * Calculate MD5 signature and put it into the place 1270 * determined before. 1271 * NOTE: since TCP options buffer doesn't point into 1272 * mbuf's data, calculate offset and use it. 1273 */ 1274 if (!TCPMD5_ENABLED() || TCPMD5_OUTPUT(m, th, 1275 (u_char *)(th + 1) + (to.to_signature - opt)) != 0) { 1276 /* 1277 * Do not send segment if the calculation of MD5 1278 * digest has failed. 1279 */ 1280 goto out; 1281 } 1282 } 1283#endif 1284#ifdef INET6 1285 if (isipv6) { 1286 /* 1287 * There is no need to fill in ip6_plen right now. 1288 * It will be filled later by ip6_output. 1289 */ 1290 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 1291 th->th_sum = in6_cksum_pseudo(ip6, sizeof(struct tcphdr) + 1292 optlen + len, IPPROTO_TCP, 0); 1293 } 1294#endif 1295#if defined(INET6) && defined(INET) 1296 else 1297#endif 1298#ifdef INET 1299 { 1300 m->m_pkthdr.csum_flags = CSUM_TCP; 1301 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1302 htons(sizeof(struct tcphdr) + IPPROTO_TCP + len + optlen)); 1303 1304 /* IP version must be set here for ipv4/ipv6 checking later */ 1305 KASSERT(ip->ip_v == IPVERSION, 1306 ("%s: IP version incorrect: %d", __func__, ip->ip_v)); 1307 } 1308#endif 1309 1310 /* 1311 * Enable TSO and specify the size of the segments. 1312 * The TCP pseudo header checksum is always provided. 1313 */ 1314 if (tso) { 1315 KASSERT(len > tp->t_maxseg - optlen, 1316 ("%s: len <= tso_segsz", __func__)); 1317 m->m_pkthdr.csum_flags |= CSUM_TSO; 1318 m->m_pkthdr.tso_segsz = tp->t_maxseg - optlen; 1319 } 1320 1321#if defined(IPSEC) || defined(IPSEC_SUPPORT) 1322 KASSERT(len + hdrlen + ipoptlen - ipsec_optlen == m_length(m, NULL), 1323 ("%s: mbuf chain shorter than expected: %ld + %u + %u - %u != %u", 1324 __func__, len, hdrlen, ipoptlen, ipsec_optlen, m_length(m, NULL))); 1325#else 1326 KASSERT(len + hdrlen + ipoptlen == m_length(m, NULL), 1327 ("%s: mbuf chain shorter than expected: %ld + %u + %u != %u", 1328 __func__, len, hdrlen, ipoptlen, m_length(m, NULL))); 1329#endif 1330 1331 /* Run HHOOK_TCP_ESTABLISHED_OUT helper hooks. */ 1332 hhook_run_tcp_est_out(tp, th, &to, len, tso); 1333 1334#ifdef TCPDEBUG 1335 /* 1336 * Trace. 1337 */ 1338 if (so->so_options & SO_DEBUG) { 1339 u_short save = 0; 1340#ifdef INET6 1341 if (!isipv6) 1342#endif 1343 { 1344 save = ipov->ih_len; 1345 ipov->ih_len = htons(m->m_pkthdr.len /* - hdrlen + (th->th_off << 2) */); 1346 } 1347 tcp_trace(TA_OUTPUT, tp->t_state, tp, mtod(m, void *), th, 0); 1348#ifdef INET6 1349 if (!isipv6) 1350#endif 1351 ipov->ih_len = save; 1352 } 1353#endif /* TCPDEBUG */ 1354 TCP_PROBE3(debug__output, tp, th, mtod(m, const char *)); 1355 1356 /* 1357 * Fill in IP length and desired time to live and 1358 * send to IP level. There should be a better way 1359 * to handle ttl and tos; we could keep them in 1360 * the template, but need a way to checksum without them. 1361 */ 1362 /* 1363 * m->m_pkthdr.len should have been set before checksum calculation, 1364 * because in6_cksum() need it. 1365 */ 1366#ifdef INET6 1367 if (isipv6) { 1368 struct route_in6 ro; 1369 1370 bzero(&ro, sizeof(ro)); 1371 /* 1372 * we separately set hoplimit for every segment, since the 1373 * user might want to change the value via setsockopt. 1374 * Also, desired default hop limit might be changed via 1375 * Neighbor Discovery. 1376 */ 1377 ip6->ip6_hlim = in6_selecthlim(tp->t_inpcb, NULL); 1378 1379 /* 1380 * Set the packet size here for the benefit of DTrace probes. 1381 * ip6_output() will set it properly; it's supposed to include 1382 * the option header lengths as well. 1383 */ 1384 ip6->ip6_plen = htons(m->m_pkthdr.len - sizeof(*ip6)); 1385 1386 if (V_path_mtu_discovery && tp->t_maxseg > V_tcp_minmss) 1387 tp->t_flags2 |= TF2_PLPMTU_PMTUD; 1388 else 1389 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD; 1390 1391 if (tp->t_state == TCPS_SYN_SENT) 1392 TCP_PROBE5(connect__request, NULL, tp, ip6, tp, th); 1393 1394 TCP_PROBE5(send, NULL, tp, ip6, tp, th); 1395 1396#ifdef TCPPCAP 1397 /* Save packet, if requested. */ 1398 tcp_pcap_add(th, m, &(tp->t_outpkts)); 1399#endif 1400 1401 /* TODO: IPv6 IP6TOS_ECT bit on */ 1402 error = ip6_output(m, tp->t_inpcb->in6p_outputopts, &ro, 1403 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 1404 NULL, NULL, tp->t_inpcb); 1405 1406 if (error == EMSGSIZE && ro.ro_rt != NULL) 1407 mtu = ro.ro_rt->rt_mtu; 1408 RO_RTFREE(&ro); 1409 } 1410#endif /* INET6 */ 1411#if defined(INET) && defined(INET6) 1412 else 1413#endif 1414#ifdef INET 1415 { 1416 ip->ip_len = htons(m->m_pkthdr.len); 1417#ifdef INET6 1418 if (tp->t_inpcb->inp_vflag & INP_IPV6PROTO) 1419 ip->ip_ttl = in6_selecthlim(tp->t_inpcb, NULL); 1420#endif /* INET6 */ 1421 /* 1422 * If we do path MTU discovery, then we set DF on every packet. 1423 * This might not be the best thing to do according to RFC3390 1424 * Section 2. However the tcp hostcache migitates the problem 1425 * so it affects only the first tcp connection with a host. 1426 * 1427 * NB: Don't set DF on small MTU/MSS to have a safe fallback. 1428 */ 1429 if (V_path_mtu_discovery && tp->t_maxseg > V_tcp_minmss) { 1430 ip->ip_off |= htons(IP_DF); 1431 tp->t_flags2 |= TF2_PLPMTU_PMTUD; 1432 } else { 1433 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD; 1434 } 1435 1436 if (tp->t_state == TCPS_SYN_SENT) 1437 TCP_PROBE5(connect__request, NULL, tp, ip, tp, th); 1438 1439 TCP_PROBE5(send, NULL, tp, ip, tp, th); 1440 1441#ifdef TCPPCAP 1442 /* Save packet, if requested. */ 1443 tcp_pcap_add(th, m, &(tp->t_outpkts)); 1444#endif 1445 1446 error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route, 1447 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 0, 1448 tp->t_inpcb); 1449 1450 if (error == EMSGSIZE && tp->t_inpcb->inp_route.ro_rt != NULL) 1451 mtu = tp->t_inpcb->inp_route.ro_rt->rt_mtu; 1452 } 1453#endif /* INET */ 1454 1455out: 1456 /* 1457 * In transmit state, time the transmission and arrange for 1458 * the retransmit. In persist state, just set snd_max. 1459 */ 1460 if ((tp->t_flags & TF_FORCEDATA) == 0 || 1461 !tcp_timer_active(tp, TT_PERSIST)) { 1462 tcp_seq startseq = tp->snd_nxt; 1463 1464 /* 1465 * Advance snd_nxt over sequence space of this segment. 1466 */ 1467 if (flags & (TH_SYN|TH_FIN)) { 1468 if (flags & TH_SYN) 1469 tp->snd_nxt++; 1470 if (flags & TH_FIN) { 1471 tp->snd_nxt++; 1472 tp->t_flags |= TF_SENTFIN; 1473 } 1474 } 1475 if (sack_rxmit) 1476 goto timer; 1477 tp->snd_nxt += len; 1478 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 1479 tp->snd_max = tp->snd_nxt; 1480 /* 1481 * Time this transmission if not a retransmission and 1482 * not currently timing anything. 1483 */ 1484 if (tp->t_rtttime == 0) { 1485 tp->t_rtttime = ticks; 1486 tp->t_rtseq = startseq; 1487 TCPSTAT_INC(tcps_segstimed); 1488 } 1489 } 1490 1491 /* 1492 * Set retransmit timer if not currently set, 1493 * and not doing a pure ack or a keep-alive probe. 1494 * Initial value for retransmit timer is smoothed 1495 * round-trip time + 2 * round-trip time variance. 1496 * Initialize shift counter which is used for backoff 1497 * of retransmit time. 1498 */ 1499timer: 1500 if (!tcp_timer_active(tp, TT_REXMT) && 1501 ((sack_rxmit && tp->snd_nxt != tp->snd_max) || 1502 (tp->snd_nxt != tp->snd_una))) { 1503 if (tcp_timer_active(tp, TT_PERSIST)) { 1504 tcp_timer_activate(tp, TT_PERSIST, 0); 1505 tp->t_rxtshift = 0; 1506 } 1507 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 1508 } else if (len == 0 && sbavail(&so->so_snd) && 1509 !tcp_timer_active(tp, TT_REXMT) && 1510 !tcp_timer_active(tp, TT_PERSIST)) { 1511 /* 1512 * Avoid a situation where we do not set persist timer 1513 * after a zero window condition. For example: 1514 * 1) A -> B: packet with enough data to fill the window 1515 * 2) B -> A: ACK for #1 + new data (0 window 1516 * advertisement) 1517 * 3) A -> B: ACK for #2, 0 len packet 1518 * 1519 * In this case, A will not activate the persist timer, 1520 * because it chose to send a packet. Unless tcp_output 1521 * is called for some other reason (delayed ack timer, 1522 * another input packet from B, socket syscall), A will 1523 * not send zero window probes. 1524 * 1525 * So, if you send a 0-length packet, but there is data 1526 * in the socket buffer, and neither the rexmt or 1527 * persist timer is already set, then activate the 1528 * persist timer. 1529 */ 1530 tp->t_rxtshift = 0; 1531 tcp_setpersist(tp); 1532 } 1533 } else { 1534 /* 1535 * Persist case, update snd_max but since we are in 1536 * persist mode (no window) we do not update snd_nxt. 1537 */ 1538 int xlen = len; 1539 if (flags & TH_SYN) 1540 ++xlen; 1541 if (flags & TH_FIN) { 1542 ++xlen; 1543 tp->t_flags |= TF_SENTFIN; 1544 } 1545 if (SEQ_GT(tp->snd_nxt + xlen, tp->snd_max)) 1546 tp->snd_max = tp->snd_nxt + len; 1547 } 1548 1549 if (error) { 1550 1551 /* 1552 * We know that the packet was lost, so back out the 1553 * sequence number advance, if any. 1554 * 1555 * If the error is EPERM the packet got blocked by the 1556 * local firewall. Normally we should terminate the 1557 * connection but the blocking may have been spurious 1558 * due to a firewall reconfiguration cycle. So we treat 1559 * it like a packet loss and let the retransmit timer and 1560 * timeouts do their work over time. 1561 * XXX: It is a POLA question whether calling tcp_drop right 1562 * away would be the really correct behavior instead. 1563 */ 1564 if (((tp->t_flags & TF_FORCEDATA) == 0 || 1565 !tcp_timer_active(tp, TT_PERSIST)) && 1566 ((flags & TH_SYN) == 0) && 1567 (error != EPERM)) { 1568 if (sack_rxmit) { 1569 p->rxmit -= len; 1570 tp->sackhint.sack_bytes_rexmit -= len; 1571 KASSERT(tp->sackhint.sack_bytes_rexmit >= 0, 1572 ("sackhint bytes rtx >= 0")); 1573 } else 1574 tp->snd_nxt -= len; 1575 } 1576 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); /* Check gotos. */ 1577 switch (error) { 1578 case EACCES: 1579 tp->t_softerror = error; 1580 return (0); 1581 case EPERM: 1582 tp->t_softerror = error; 1583 return (error); 1584 case ENOBUFS: 1585 TCP_XMIT_TIMER_ASSERT(tp, len, flags); 1586 tp->snd_cwnd = tp->t_maxseg; 1587 return (0); 1588 case EMSGSIZE: 1589 /* 1590 * For some reason the interface we used initially 1591 * to send segments changed to another or lowered 1592 * its MTU. 1593 * If TSO was active we either got an interface 1594 * without TSO capabilits or TSO was turned off. 1595 * If we obtained mtu from ip_output() then update 1596 * it and try again. 1597 */ 1598 if (tso) 1599 tp->t_flags &= ~TF_TSO; 1600 if (mtu != 0) { 1601 tcp_mss_update(tp, -1, mtu, NULL, NULL); 1602 goto again; 1603 } 1604 return (error); 1605 case EHOSTDOWN: 1606 case EHOSTUNREACH: 1607 case ENETDOWN: 1608 case ENETUNREACH: 1609 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1610 tp->t_softerror = error; 1611 return (0); 1612 } 1613 /* FALLTHROUGH */ 1614 default: 1615 return (error); 1616 } 1617 } 1618 TCPSTAT_INC(tcps_sndtotal); 1619 1620 /* 1621 * Data sent (as far as we can tell). 1622 * If this advertises a larger window than any other segment, 1623 * then remember the size of the advertised window. 1624 * Any pending ACK has now been sent. 1625 */ 1626 if (recwin >= 0 && SEQ_GT(tp->rcv_nxt + recwin, tp->rcv_adv)) 1627 tp->rcv_adv = tp->rcv_nxt + recwin; 1628 tp->last_ack_sent = tp->rcv_nxt; 1629 tp->t_flags &= ~(TF_ACKNOW | TF_DELACK); 1630 if (tcp_timer_active(tp, TT_DELACK)) 1631 tcp_timer_activate(tp, TT_DELACK, 0); 1632#if 0 1633 /* 1634 * This completely breaks TCP if newreno is turned on. What happens 1635 * is that if delayed-acks are turned on on the receiver, this code 1636 * on the transmitter effectively destroys the TCP window, forcing 1637 * it to four packets (1.5Kx4 = 6K window). 1638 */ 1639 if (sendalot && --maxburst) 1640 goto again; 1641#endif 1642 if (sendalot) 1643 goto again; 1644 return (0); 1645} 1646 1647void 1648tcp_setpersist(struct tcpcb *tp) 1649{ 1650 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1; 1651 int tt; 1652 1653 tp->t_flags &= ~TF_PREVVALID; 1654 if (tcp_timer_active(tp, TT_REXMT)) 1655 panic("tcp_setpersist: retransmit pending"); 1656 /* 1657 * Start/restart persistence timer. 1658 */ 1659 TCPT_RANGESET(tt, t * tcp_backoff[tp->t_rxtshift], 1660 tcp_persmin, tcp_persmax); 1661 tcp_timer_activate(tp, TT_PERSIST, tt); 1662 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 1663 tp->t_rxtshift++; 1664} 1665 1666/* 1667 * Insert TCP options according to the supplied parameters to the place 1668 * optp in a consistent way. Can handle unaligned destinations. 1669 * 1670 * The order of the option processing is crucial for optimal packing and 1671 * alignment for the scarce option space. 1672 * 1673 * The optimal order for a SYN/SYN-ACK segment is: 1674 * MSS (4) + NOP (1) + Window scale (3) + SACK permitted (2) + 1675 * Timestamp (10) + Signature (18) = 38 bytes out of a maximum of 40. 1676 * 1677 * The SACK options should be last. SACK blocks consume 8*n+2 bytes. 1678 * So a full size SACK blocks option is 34 bytes (with 4 SACK blocks). 1679 * At minimum we need 10 bytes (to generate 1 SACK block). If both 1680 * TCP Timestamps (12 bytes) and TCP Signatures (18 bytes) are present, 1681 * we only have 10 bytes for SACK options (40 - (12 + 18)). 1682 */ 1683int 1684tcp_addoptions(struct tcpopt *to, u_char *optp) 1685{ 1686 u_int32_t mask, optlen = 0; 1687 1688 for (mask = 1; mask < TOF_MAXOPT; mask <<= 1) { 1689 if ((to->to_flags & mask) != mask) 1690 continue; 1691 if (optlen == TCP_MAXOLEN) 1692 break; 1693 switch (to->to_flags & mask) { 1694 case TOF_MSS: 1695 while (optlen % 4) { 1696 optlen += TCPOLEN_NOP; 1697 *optp++ = TCPOPT_NOP; 1698 } 1699 if (TCP_MAXOLEN - optlen < TCPOLEN_MAXSEG) 1700 continue; 1701 optlen += TCPOLEN_MAXSEG; 1702 *optp++ = TCPOPT_MAXSEG; 1703 *optp++ = TCPOLEN_MAXSEG; 1704 to->to_mss = htons(to->to_mss); 1705 bcopy((u_char *)&to->to_mss, optp, sizeof(to->to_mss)); 1706 optp += sizeof(to->to_mss); 1707 break; 1708 case TOF_SCALE: 1709 while (!optlen || optlen % 2 != 1) { 1710 optlen += TCPOLEN_NOP; 1711 *optp++ = TCPOPT_NOP; 1712 } 1713 if (TCP_MAXOLEN - optlen < TCPOLEN_WINDOW) 1714 continue; 1715 optlen += TCPOLEN_WINDOW; 1716 *optp++ = TCPOPT_WINDOW; 1717 *optp++ = TCPOLEN_WINDOW; 1718 *optp++ = to->to_wscale; 1719 break; 1720 case TOF_SACKPERM: 1721 while (optlen % 2) { 1722 optlen += TCPOLEN_NOP; 1723 *optp++ = TCPOPT_NOP; 1724 } 1725 if (TCP_MAXOLEN - optlen < TCPOLEN_SACK_PERMITTED) 1726 continue; 1727 optlen += TCPOLEN_SACK_PERMITTED; 1728 *optp++ = TCPOPT_SACK_PERMITTED; 1729 *optp++ = TCPOLEN_SACK_PERMITTED; 1730 break; 1731 case TOF_TS: 1732 while (!optlen || optlen % 4 != 2) { 1733 optlen += TCPOLEN_NOP; 1734 *optp++ = TCPOPT_NOP; 1735 } 1736 if (TCP_MAXOLEN - optlen < TCPOLEN_TIMESTAMP) 1737 continue; 1738 optlen += TCPOLEN_TIMESTAMP; 1739 *optp++ = TCPOPT_TIMESTAMP; 1740 *optp++ = TCPOLEN_TIMESTAMP; 1741 to->to_tsval = htonl(to->to_tsval); 1742 to->to_tsecr = htonl(to->to_tsecr); 1743 bcopy((u_char *)&to->to_tsval, optp, sizeof(to->to_tsval)); 1744 optp += sizeof(to->to_tsval); 1745 bcopy((u_char *)&to->to_tsecr, optp, sizeof(to->to_tsecr)); 1746 optp += sizeof(to->to_tsecr); 1747 break; 1748 case TOF_SIGNATURE: 1749 { 1750 int siglen = TCPOLEN_SIGNATURE - 2; 1751 1752 while (!optlen || optlen % 4 != 2) { 1753 optlen += TCPOLEN_NOP; 1754 *optp++ = TCPOPT_NOP; 1755 } 1756 if (TCP_MAXOLEN - optlen < TCPOLEN_SIGNATURE) { 1757 to->to_flags &= ~TOF_SIGNATURE; 1758 continue; 1759 } 1760 optlen += TCPOLEN_SIGNATURE; 1761 *optp++ = TCPOPT_SIGNATURE; 1762 *optp++ = TCPOLEN_SIGNATURE; 1763 to->to_signature = optp; 1764 while (siglen--) 1765 *optp++ = 0; 1766 break; 1767 } 1768 case TOF_SACK: 1769 { 1770 int sackblks = 0; 1771 struct sackblk *sack = (struct sackblk *)to->to_sacks; 1772 tcp_seq sack_seq; 1773 1774 while (!optlen || optlen % 4 != 2) { 1775 optlen += TCPOLEN_NOP; 1776 *optp++ = TCPOPT_NOP; 1777 } 1778 if (TCP_MAXOLEN - optlen < TCPOLEN_SACKHDR + TCPOLEN_SACK) 1779 continue; 1780 optlen += TCPOLEN_SACKHDR; 1781 *optp++ = TCPOPT_SACK; 1782 sackblks = min(to->to_nsacks, 1783 (TCP_MAXOLEN - optlen) / TCPOLEN_SACK); 1784 *optp++ = TCPOLEN_SACKHDR + sackblks * TCPOLEN_SACK; 1785 while (sackblks--) { 1786 sack_seq = htonl(sack->start); 1787 bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq)); 1788 optp += sizeof(sack_seq); 1789 sack_seq = htonl(sack->end); 1790 bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq)); 1791 optp += sizeof(sack_seq); 1792 optlen += TCPOLEN_SACK; 1793 sack++; 1794 } 1795 TCPSTAT_INC(tcps_sack_send_blocks); 1796 break; 1797 } 1798#ifdef TCP_RFC7413 1799 case TOF_FASTOPEN: 1800 { 1801 int total_len; 1802 1803 /* XXX is there any point to aligning this option? */ 1804 total_len = TCPOLEN_FAST_OPEN_EMPTY + to->to_tfo_len; 1805 if (TCP_MAXOLEN - optlen < total_len) 1806 continue; 1807 *optp++ = TCPOPT_FAST_OPEN; 1808 *optp++ = total_len; 1809 if (to->to_tfo_len > 0) { 1810 bcopy(to->to_tfo_cookie, optp, to->to_tfo_len); 1811 optp += to->to_tfo_len; 1812 } 1813 optlen += total_len; 1814 break; 1815 } 1816#endif 1817 default: 1818 panic("%s: unknown TCP option type", __func__); 1819 break; 1820 } 1821 } 1822 1823 /* Terminate and pad TCP options to a 4 byte boundary. */ 1824 if (optlen % 4) { 1825 optlen += TCPOLEN_EOL; 1826 *optp++ = TCPOPT_EOL; 1827 } 1828 /* 1829 * According to RFC 793 (STD0007): 1830 * "The content of the header beyond the End-of-Option option 1831 * must be header padding (i.e., zero)." 1832 * and later: "The padding is composed of zeros." 1833 */ 1834 while (optlen % 4) { 1835 optlen += TCPOLEN_PAD; 1836 *optp++ = TCPOPT_PAD; 1837 } 1838 1839 KASSERT(optlen <= TCP_MAXOLEN, ("%s: TCP options too long", __func__)); 1840 return (optlen); 1841} 1842