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("$FreeBSD$"); 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 #ifdef TCP_HHOOK 44 #include <sys/hhook.h> 45 #endif 46 #include <sys/kernel.h> 47 #include <sys/lock.h> 48 #include <sys/mbuf.h> 49 #include <sys/mutex.h> 50 #include <sys/protosw.h> 51 #include <sys/sdt.h> 52 #include <sys/socket.h> 53 #include <sys/socketvar.h> 54 #include <sys/sysctl.h> 55 56 #include <net/if.h> 57 #include <net/route.h> 58 #include <net/vnet.h> 59 60 #include <netinet/in.h> 61 #include <netinet/in_kdtrace.h> 62 #include <netinet/in_systm.h> 63 #include <netinet/ip.h> 64 #include <netinet/in_pcb.h> 65 #include <netinet/ip_var.h> 66 #include <netinet/ip_options.h> 67 #ifdef INET6 68 #include <netinet6/in6_pcb.h> 69 #include <netinet/ip6.h> 70 #include <netinet6/ip6_var.h> 71 #endif 72 #ifdef TCP_RFC7413 73 #include <netinet/tcp_fastopen.h> 74 #endif 75 #include <netinet/tcp.h> 76 #define TCPOUTFLAGS 77 #include <netinet/tcp_fsm.h> 78 #include <netinet/tcp_seq.h> 79 #include <netinet/tcp_timer.h> 80 #include <netinet/tcp_var.h> 81 #include <netinet/tcpip.h> 82 #include <netinet/cc/cc.h> 83 #ifdef TCPPCAP 84 #include <netinet/tcp_pcap.h> 85 #endif 86 #ifdef TCPDEBUG 87 #include <netinet/tcp_debug.h> 88 #endif 89 #ifdef TCP_OFFLOAD 90 #include <netinet/tcp_offload.h> 91 #endif 92 93 #include <netipsec/ipsec_support.h> 94 95 #include <machine/in_cksum.h> 96 97 #include <security/mac/mac_framework.h> 98 99 VNET_DEFINE(int, path_mtu_discovery) = 1; 100 SYSCTL_INT(_net_inet_tcp, OID_AUTO, path_mtu_discovery, CTLFLAG_VNET | CTLFLAG_RW, 101 &VNET_NAME(path_mtu_discovery), 1, 102 "Enable Path MTU Discovery"); 103 104 VNET_DEFINE(int, tcp_do_tso) = 1; 105 #define V_tcp_do_tso VNET(tcp_do_tso) 106 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tso, CTLFLAG_VNET | CTLFLAG_RW, 107 &VNET_NAME(tcp_do_tso), 0, 108 "Enable TCP Segmentation Offload"); 109 110 VNET_DEFINE(int, tcp_sendspace) = 1024*32; 111 #define V_tcp_sendspace VNET(tcp_sendspace) 112 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_VNET | CTLFLAG_RW, 113 &VNET_NAME(tcp_sendspace), 0, "Initial send socket buffer size"); 114 115 VNET_DEFINE(int, tcp_do_autosndbuf) = 1; 116 #define V_tcp_do_autosndbuf VNET(tcp_do_autosndbuf) 117 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sendbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 118 &VNET_NAME(tcp_do_autosndbuf), 0, 119 "Enable automatic send buffer sizing"); 120 121 VNET_DEFINE(int, tcp_autosndbuf_inc) = 8*1024; 122 #define V_tcp_autosndbuf_inc VNET(tcp_autosndbuf_inc) 123 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sendbuf_inc, CTLFLAG_VNET | CTLFLAG_RW, 124 &VNET_NAME(tcp_autosndbuf_inc), 0, 125 "Incrementor step size of automatic send buffer"); 126 127 VNET_DEFINE(int, tcp_autosndbuf_max) = 2*1024*1024; 128 #define V_tcp_autosndbuf_max VNET(tcp_autosndbuf_max) 129 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sendbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 130 &VNET_NAME(tcp_autosndbuf_max), 0, 131 "Max size of automatic send buffer"); 132 133 /* 134 * Make sure that either retransmit or persist timer is set for SYN, FIN and 135 * non-ACK. 136 */ 137 #define TCP_XMIT_TIMER_ASSERT(tp, len, th_flags) \ 138 KASSERT(((len) == 0 && ((th_flags) & (TH_SYN | TH_FIN)) == 0) ||\ 139 tcp_timer_active((tp), TT_REXMT) || \ 140 tcp_timer_active((tp), TT_PERSIST), \ 141 ("neither rexmt nor persist timer is set")) 142 143 #ifdef TCP_HHOOK 144 static void inline hhook_run_tcp_est_out(struct tcpcb *tp, 145 struct tcphdr *th, struct tcpopt *to, 146 uint32_t len, int tso); 147 #endif 148 static void inline cc_after_idle(struct tcpcb *tp); 149 150 #ifdef TCP_HHOOK 151 /* 152 * Wrapper for the TCP established output helper hook. 153 */ 154 static void inline 155 hhook_run_tcp_est_out(struct tcpcb *tp, struct tcphdr *th, 156 struct tcpopt *to, uint32_t len, int tso) 157 { 158 struct tcp_hhook_data hhook_data; 159 160 if (V_tcp_hhh[HHOOK_TCP_EST_OUT]->hhh_nhooks > 0) { 161 hhook_data.tp = tp; 162 hhook_data.th = th; 163 hhook_data.to = to; 164 hhook_data.len = len; 165 hhook_data.tso = tso; 166 167 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_OUT], &hhook_data, 168 tp->osd); 169 } 170 } 171 #endif 172 173 /* 174 * CC wrapper hook functions 175 */ 176 static void inline 177 cc_after_idle(struct tcpcb *tp) 178 { 179 INP_WLOCK_ASSERT(tp->t_inpcb); 180 181 if (CC_ALGO(tp)->after_idle != NULL) 182 CC_ALGO(tp)->after_idle(tp->ccv); 183 } 184 185 /* 186 * Tcp output routine: figure out what should be sent and send it. 187 */ 188 int 189 tcp_output(struct tcpcb *tp) 190 { 191 struct socket *so = tp->t_inpcb->inp_socket; 192 int32_t len; 193 uint32_t recwin, sendwin; 194 int off, flags, error = 0; /* Keep compiler happy */ 195 struct mbuf *m; 196 struct ip *ip = NULL; 197 struct ipovly *ipov = NULL; 198 struct tcphdr *th; 199 u_char opt[TCP_MAXOLEN]; 200 unsigned ipoptlen, optlen, hdrlen; 201 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 202 unsigned ipsec_optlen = 0; 203 #endif 204 int idle, sendalot; 205 int sack_rxmit, sack_bytes_rxmt; 206 struct sackhole *p; 207 int tso, mtu; 208 struct tcpopt to; 209 #if 0 210 int maxburst = TCP_MAXBURST; 211 #endif 212 #ifdef INET6 213 struct ip6_hdr *ip6 = NULL; 214 int isipv6; 215 216 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 217 #endif 218 219 INP_WLOCK_ASSERT(tp->t_inpcb); 220 221 #ifdef TCP_OFFLOAD 222 if (tp->t_flags & TF_TOE) 223 return (tcp_offload_output(tp)); 224 #endif 225 226 #ifdef TCP_RFC7413 227 /* 228 * For TFO connections in SYN_RECEIVED, only allow the initial 229 * SYN|ACK and those sent by the retransmit timer. 230 */ 231 if (IS_FASTOPEN(tp->t_flags) && 232 (tp->t_state == TCPS_SYN_RECEIVED) && 233 SEQ_GT(tp->snd_max, tp->snd_una) && /* initial SYN|ACK sent */ 234 (tp->snd_nxt != tp->snd_una)) /* not a retransmit */ 235 return (0); 236 #endif 237 /* 238 * Determine length of data that should be transmitted, 239 * and flags that will be used. 240 * If there is some data or critical controls (SYN, RST) 241 * to send, then transmit; otherwise, investigate further. 242 */ 243 idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una); 244 if (idle && ticks - tp->t_rcvtime >= tp->t_rxtcur) 245 cc_after_idle(tp); 246 tp->t_flags &= ~TF_LASTIDLE; 247 if (idle) { 248 if (tp->t_flags & TF_MORETOCOME) { 249 tp->t_flags |= TF_LASTIDLE; 250 idle = 0; 251 } 252 } 253 again: 254 /* 255 * If we've recently taken a timeout, snd_max will be greater than 256 * snd_nxt. There may be SACK information that allows us to avoid 257 * resending already delivered data. Adjust snd_nxt accordingly. 258 */ 259 if ((tp->t_flags & TF_SACK_PERMIT) && 260 SEQ_LT(tp->snd_nxt, tp->snd_max)) 261 tcp_sack_adjust(tp); 262 sendalot = 0; 263 tso = 0; 264 mtu = 0; 265 off = tp->snd_nxt - tp->snd_una; 266 sendwin = min(tp->snd_wnd, tp->snd_cwnd); 267 268 flags = tcp_outflags[tp->t_state]; 269 /* 270 * Send any SACK-generated retransmissions. If we're explicitly trying 271 * to send out new data (when sendalot is 1), bypass this function. 272 * If we retransmit in fast recovery mode, decrement snd_cwnd, since 273 * we're replacing a (future) new transmission with a retransmission 274 * now, and we previously incremented snd_cwnd in tcp_input(). 275 */ 276 /* 277 * Still in sack recovery , reset rxmit flag to zero. 278 */ 279 sack_rxmit = 0; 280 sack_bytes_rxmt = 0; 281 len = 0; 282 p = NULL; 283 if ((tp->t_flags & TF_SACK_PERMIT) && IN_FASTRECOVERY(tp->t_flags) && 284 (p = tcp_sack_output(tp, &sack_bytes_rxmt))) { 285 uint32_t cwin; 286 287 cwin = 288 imax(min(tp->snd_wnd, tp->snd_cwnd) - sack_bytes_rxmt, 0); 289 /* Do not retransmit SACK segments beyond snd_recover */ 290 if (SEQ_GT(p->end, tp->snd_recover)) { 291 /* 292 * (At least) part of sack hole extends beyond 293 * snd_recover. Check to see if we can rexmit data 294 * for this hole. 295 */ 296 if (SEQ_GEQ(p->rxmit, tp->snd_recover)) { 297 /* 298 * Can't rexmit any more data for this hole. 299 * That data will be rexmitted in the next 300 * sack recovery episode, when snd_recover 301 * moves past p->rxmit. 302 */ 303 p = NULL; 304 goto after_sack_rexmit; 305 } else 306 /* Can rexmit part of the current hole */ 307 len = ((int32_t)ulmin(cwin, 308 tp->snd_recover - p->rxmit)); 309 } else 310 len = ((int32_t)ulmin(cwin, p->end - p->rxmit)); 311 off = p->rxmit - tp->snd_una; 312 KASSERT(off >= 0,("%s: sack block to the left of una : %d", 313 __func__, off)); 314 if (len > 0) { 315 sack_rxmit = 1; 316 sendalot = 1; 317 TCPSTAT_INC(tcps_sack_rexmits); 318 TCPSTAT_ADD(tcps_sack_rexmit_bytes, 319 min(len, tp->t_maxseg)); 320 } 321 } 322 after_sack_rexmit: 323 /* 324 * Get standard flags, and add SYN or FIN if requested by 'hidden' 325 * state flags. 326 */ 327 if (tp->t_flags & TF_NEEDFIN) 328 flags |= TH_FIN; 329 if (tp->t_flags & TF_NEEDSYN) 330 flags |= TH_SYN; 331 332 SOCKBUF_LOCK(&so->so_snd); 333 /* 334 * If in persist timeout with window of 0, send 1 byte. 335 * Otherwise, if window is small but nonzero 336 * and timer expired, we will send what we can 337 * and go to transmit state. 338 */ 339 if (tp->t_flags & TF_FORCEDATA) { 340 if (sendwin == 0) { 341 /* 342 * If we still have some data to send, then 343 * clear the FIN bit. Usually this would 344 * happen below when it realizes that we 345 * aren't sending all the data. However, 346 * if we have exactly 1 byte of unsent data, 347 * then it won't clear the FIN bit below, 348 * and if we are in persist state, we wind 349 * up sending the packet without recording 350 * that we sent the FIN bit. 351 * 352 * We can't just blindly clear the FIN bit, 353 * because if we don't have any more data 354 * to send then the probe will be the FIN 355 * itself. 356 */ 357 if (off < sbused(&so->so_snd)) 358 flags &= ~TH_FIN; 359 sendwin = 1; 360 } else { 361 tcp_timer_activate(tp, TT_PERSIST, 0); 362 tp->t_rxtshift = 0; 363 } 364 } 365 366 /* 367 * If snd_nxt == snd_max and we have transmitted a FIN, the 368 * offset will be > 0 even if so_snd.sb_cc is 0, resulting in 369 * a negative length. This can also occur when TCP opens up 370 * its congestion window while receiving additional duplicate 371 * acks after fast-retransmit because TCP will reset snd_nxt 372 * to snd_max after the fast-retransmit. 373 * 374 * In the normal retransmit-FIN-only case, however, snd_nxt will 375 * be set to snd_una, the offset will be 0, and the length may 376 * wind up 0. 377 * 378 * If sack_rxmit is true we are retransmitting from the scoreboard 379 * in which case len is already set. 380 */ 381 if (sack_rxmit == 0) { 382 if (sack_bytes_rxmt == 0) 383 len = ((int32_t)ulmin(sbavail(&so->so_snd), sendwin) - 384 off); 385 else { 386 int32_t cwin; 387 388 /* 389 * We are inside of a SACK recovery episode and are 390 * sending new data, having retransmitted all the 391 * data possible in the scoreboard. 392 */ 393 len = ((int32_t)min(sbavail(&so->so_snd), tp->snd_wnd) - 394 off); 395 /* 396 * Don't remove this (len > 0) check ! 397 * We explicitly check for len > 0 here (although it 398 * isn't really necessary), to work around a gcc 399 * optimization issue - to force gcc to compute 400 * len above. Without this check, the computation 401 * of len is bungled by the optimizer. 402 */ 403 if (len > 0) { 404 cwin = tp->snd_cwnd - 405 (tp->snd_nxt - tp->sack_newdata) - 406 sack_bytes_rxmt; 407 if (cwin < 0) 408 cwin = 0; 409 len = imin(len, cwin); 410 } 411 } 412 } 413 414 /* 415 * Lop off SYN bit if it has already been sent. However, if this 416 * is SYN-SENT state and if segment contains data and if we don't 417 * know that foreign host supports TAO, suppress sending segment. 418 */ 419 if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) { 420 if (tp->t_state != TCPS_SYN_RECEIVED) 421 flags &= ~TH_SYN; 422 #ifdef TCP_RFC7413 423 /* 424 * When sending additional segments following a TFO SYN|ACK, 425 * do not include the SYN bit. 426 */ 427 if (IS_FASTOPEN(tp->t_flags) && 428 (tp->t_state == TCPS_SYN_RECEIVED)) 429 flags &= ~TH_SYN; 430 #endif 431 off--, len++; 432 } 433 434 /* 435 * Be careful not to send data and/or FIN on SYN segments. 436 * This measure is needed to prevent interoperability problems 437 * with not fully conformant TCP implementations. 438 */ 439 if ((flags & TH_SYN) && (tp->t_flags & TF_NOOPT)) { 440 len = 0; 441 flags &= ~TH_FIN; 442 } 443 444 #ifdef TCP_RFC7413 445 /* 446 * When retransmitting SYN|ACK on a passively-created TFO socket, 447 * don't include data, as the presence of data may have caused the 448 * original SYN|ACK to have been dropped by a middlebox. 449 */ 450 if (IS_FASTOPEN(tp->t_flags) && 451 (((tp->t_state == TCPS_SYN_RECEIVED) && (tp->t_rxtshift > 0)) || 452 (flags & TH_RST))) 453 len = 0; 454 #endif 455 if (len <= 0) { 456 /* 457 * If FIN has been sent but not acked, 458 * but we haven't been called to retransmit, 459 * len will be < 0. Otherwise, window shrank 460 * after we sent into it. If window shrank to 0, 461 * cancel pending retransmit, pull snd_nxt back 462 * to (closed) window, and set the persist timer 463 * if it isn't already going. If the window didn't 464 * close completely, just wait for an ACK. 465 * 466 * We also do a general check here to ensure that 467 * we will set the persist timer when we have data 468 * to send, but a 0-byte window. This makes sure 469 * the persist timer is set even if the packet 470 * hits one of the "goto send" lines below. 471 */ 472 len = 0; 473 if ((sendwin == 0) && (TCPS_HAVEESTABLISHED(tp->t_state)) && 474 (off < (int) sbavail(&so->so_snd))) { 475 tcp_timer_activate(tp, TT_REXMT, 0); 476 tp->t_rxtshift = 0; 477 tp->snd_nxt = tp->snd_una; 478 if (!tcp_timer_active(tp, TT_PERSIST)) 479 tcp_setpersist(tp); 480 } 481 } 482 483 /* len will be >= 0 after this point. */ 484 KASSERT(len >= 0, ("[%s:%d]: len < 0", __func__, __LINE__)); 485 486 /* 487 * Automatic sizing of send socket buffer. Often the send buffer 488 * size is not optimally adjusted to the actual network conditions 489 * at hand (delay bandwidth product). Setting the buffer size too 490 * small limits throughput on links with high bandwidth and high 491 * delay (eg. trans-continental/oceanic links). Setting the 492 * buffer size too big consumes too much real kernel memory, 493 * especially with many connections on busy servers. 494 * 495 * The criteria to step up the send buffer one notch are: 496 * 1. receive window of remote host is larger than send buffer 497 * (with a fudge factor of 5/4th); 498 * 2. send buffer is filled to 7/8th with data (so we actually 499 * have data to make use of it); 500 * 3. send buffer fill has not hit maximal automatic size; 501 * 4. our send window (slow start and cogestion controlled) is 502 * larger than sent but unacknowledged data in send buffer. 503 * 504 * The remote host receive window scaling factor may limit the 505 * growing of the send buffer before it reaches its allowed 506 * maximum. 507 * 508 * It scales directly with slow start or congestion window 509 * and does at most one step per received ACK. This fast 510 * scaling has the drawback of growing the send buffer beyond 511 * what is strictly necessary to make full use of a given 512 * delay*bandwidth product. However testing has shown this not 513 * to be much of an problem. At worst we are trading wasting 514 * of available bandwidth (the non-use of it) for wasting some 515 * socket buffer memory. 516 * 517 * TODO: Shrink send buffer during idle periods together 518 * with congestion window. Requires another timer. Has to 519 * wait for upcoming tcp timer rewrite. 520 * 521 * XXXGL: should there be used sbused() or sbavail()? 522 */ 523 if (V_tcp_do_autosndbuf && so->so_snd.sb_flags & SB_AUTOSIZE) { 524 if ((tp->snd_wnd / 4 * 5) >= so->so_snd.sb_hiwat && 525 sbused(&so->so_snd) >= (so->so_snd.sb_hiwat / 8 * 7) && 526 sbused(&so->so_snd) < V_tcp_autosndbuf_max && 527 sendwin >= (sbused(&so->so_snd) - 528 (tp->snd_nxt - tp->snd_una))) { 529 if (!sbreserve_locked(&so->so_snd, 530 min(so->so_snd.sb_hiwat + V_tcp_autosndbuf_inc, 531 V_tcp_autosndbuf_max), so, curthread)) 532 so->so_snd.sb_flags &= ~SB_AUTOSIZE; 533 } 534 } 535 536 /* 537 * Decide if we can use TCP Segmentation Offloading (if supported by 538 * hardware). 539 * 540 * TSO may only be used if we are in a pure bulk sending state. The 541 * presence of TCP-MD5, SACK retransmits, SACK advertizements and 542 * IP options prevent using TSO. With TSO the TCP header is the same 543 * (except for the sequence number) for all generated packets. This 544 * makes it impossible to transmit any options which vary per generated 545 * segment or packet. 546 * 547 * IPv4 handling has a clear separation of ip options and ip header 548 * flags while IPv6 combines both in in6p_outputopts. ip6_optlen() does 549 * the right thing below to provide length of just ip options and thus 550 * checking for ipoptlen is enough to decide if ip options are present. 551 */ 552 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 553 /* 554 * Pre-calculate here as we save another lookup into the darknesses 555 * of IPsec that way and can actually decide if TSO is ok. 556 */ 557 #ifdef INET6 558 if (isipv6 && IPSEC_ENABLED(ipv6)) 559 ipsec_optlen = IPSEC_HDRSIZE(ipv6, tp->t_inpcb); 560 #ifdef INET 561 else 562 #endif 563 #endif /* INET6 */ 564 #ifdef INET 565 if (IPSEC_ENABLED(ipv4)) 566 ipsec_optlen = IPSEC_HDRSIZE(ipv4, tp->t_inpcb); 567 #endif /* INET */ 568 #endif /* IPSEC */ 569 #ifdef INET6 570 if (isipv6) 571 ipoptlen = ip6_optlen(tp->t_inpcb); 572 else 573 #endif 574 if (tp->t_inpcb->inp_options) 575 ipoptlen = tp->t_inpcb->inp_options->m_len - 576 offsetof(struct ipoption, ipopt_list); 577 else 578 ipoptlen = 0; 579 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 580 ipoptlen += ipsec_optlen; 581 #endif 582 583 if ((tp->t_flags & TF_TSO) && V_tcp_do_tso && len > tp->t_maxseg && 584 ((tp->t_flags & TF_SIGNATURE) == 0) && 585 tp->rcv_numsacks == 0 && sack_rxmit == 0 && 586 ipoptlen == 0) 587 tso = 1; 588 589 if (sack_rxmit) { 590 if (SEQ_LT(p->rxmit + len, tp->snd_una + sbused(&so->so_snd))) 591 flags &= ~TH_FIN; 592 } else { 593 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + 594 sbused(&so->so_snd))) 595 flags &= ~TH_FIN; 596 } 597 598 recwin = lmin(lmax(sbspace(&so->so_rcv), 0), 599 (long)TCP_MAXWIN << tp->rcv_scale); 600 601 /* 602 * Sender silly window avoidance. We transmit under the following 603 * conditions when len is non-zero: 604 * 605 * - We have a full segment (or more with TSO) 606 * - This is the last buffer in a write()/send() and we are 607 * either idle or running NODELAY 608 * - we've timed out (e.g. persist timer) 609 * - we have more then 1/2 the maximum send window's worth of 610 * data (receiver may be limited the window size) 611 * - we need to retransmit 612 */ 613 if (len) { 614 if (len >= tp->t_maxseg) 615 goto send; 616 /* 617 * NOTE! on localhost connections an 'ack' from the remote 618 * end may occur synchronously with the output and cause 619 * us to flush a buffer queued with moretocome. XXX 620 * 621 * note: the len + off check is almost certainly unnecessary. 622 */ 623 if (!(tp->t_flags & TF_MORETOCOME) && /* normal case */ 624 (idle || (tp->t_flags & TF_NODELAY)) && 625 (uint32_t)len + (uint32_t)off >= sbavail(&so->so_snd) && 626 (tp->t_flags & TF_NOPUSH) == 0) { 627 goto send; 628 } 629 if (tp->t_flags & TF_FORCEDATA) /* typ. timeout case */ 630 goto send; 631 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0) 632 goto send; 633 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */ 634 goto send; 635 if (sack_rxmit) 636 goto send; 637 } 638 639 /* 640 * Sending of standalone window updates. 641 * 642 * Window updates are important when we close our window due to a 643 * full socket buffer and are opening it again after the application 644 * reads data from it. Once the window has opened again and the 645 * remote end starts to send again the ACK clock takes over and 646 * provides the most current window information. 647 * 648 * We must avoid the silly window syndrome whereas every read 649 * from the receive buffer, no matter how small, causes a window 650 * update to be sent. We also should avoid sending a flurry of 651 * window updates when the socket buffer had queued a lot of data 652 * and the application is doing small reads. 653 * 654 * Prevent a flurry of pointless window updates by only sending 655 * an update when we can increase the advertized window by more 656 * than 1/4th of the socket buffer capacity. When the buffer is 657 * getting full or is very small be more aggressive and send an 658 * update whenever we can increase by two mss sized segments. 659 * In all other situations the ACK's to new incoming data will 660 * carry further window increases. 661 * 662 * Don't send an independent window update if a delayed 663 * ACK is pending (it will get piggy-backed on it) or the 664 * remote side already has done a half-close and won't send 665 * more data. Skip this if the connection is in T/TCP 666 * half-open state. 667 */ 668 if (recwin > 0 && !(tp->t_flags & TF_NEEDSYN) && 669 !(tp->t_flags & TF_DELACK) && 670 !TCPS_HAVERCVDFIN(tp->t_state)) { 671 /* 672 * "adv" is the amount we could increase the window, 673 * taking into account that we are limited by 674 * TCP_MAXWIN << tp->rcv_scale. 675 */ 676 int32_t adv; 677 int oldwin; 678 679 adv = recwin; 680 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) { 681 oldwin = (tp->rcv_adv - tp->rcv_nxt); 682 adv -= oldwin; 683 } else 684 oldwin = 0; 685 686 /* 687 * If the new window size ends up being the same as or less 688 * than the old size when it is scaled, then don't force 689 * a window update. 690 */ 691 if (oldwin >> tp->rcv_scale >= (adv + oldwin) >> tp->rcv_scale) 692 goto dontupdate; 693 694 if (adv >= (int32_t)(2 * tp->t_maxseg) && 695 (adv >= (int32_t)(so->so_rcv.sb_hiwat / 4) || 696 recwin <= (so->so_rcv.sb_hiwat / 8) || 697 so->so_rcv.sb_hiwat <= 8 * tp->t_maxseg)) 698 goto send; 699 } 700 dontupdate: 701 702 /* 703 * Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW 704 * is also a catch-all for the retransmit timer timeout case. 705 */ 706 if (tp->t_flags & TF_ACKNOW) 707 goto send; 708 if ((flags & TH_RST) || 709 ((flags & TH_SYN) && (tp->t_flags & TF_NEEDSYN) == 0)) 710 goto send; 711 if (SEQ_GT(tp->snd_up, tp->snd_una)) 712 goto send; 713 /* 714 * If our state indicates that FIN should be sent 715 * and we have not yet done so, then we need to send. 716 */ 717 if (flags & TH_FIN && 718 ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una)) 719 goto send; 720 /* 721 * In SACK, it is possible for tcp_output to fail to send a segment 722 * after the retransmission timer has been turned off. Make sure 723 * that the retransmission timer is set. 724 */ 725 if ((tp->t_flags & TF_SACK_PERMIT) && 726 SEQ_GT(tp->snd_max, tp->snd_una) && 727 !tcp_timer_active(tp, TT_REXMT) && 728 !tcp_timer_active(tp, TT_PERSIST)) { 729 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 730 goto just_return; 731 } 732 /* 733 * TCP window updates are not reliable, rather a polling protocol 734 * using ``persist'' packets is used to insure receipt of window 735 * updates. The three ``states'' for the output side are: 736 * idle not doing retransmits or persists 737 * persisting to move a small or zero window 738 * (re)transmitting and thereby not persisting 739 * 740 * tcp_timer_active(tp, TT_PERSIST) 741 * is true when we are in persist state. 742 * (tp->t_flags & TF_FORCEDATA) 743 * is set when we are called to send a persist packet. 744 * tcp_timer_active(tp, TT_REXMT) 745 * is set when we are retransmitting 746 * The output side is idle when both timers are zero. 747 * 748 * If send window is too small, there is data to transmit, and no 749 * retransmit or persist is pending, then go to persist state. 750 * If nothing happens soon, send when timer expires: 751 * if window is nonzero, transmit what we can, 752 * otherwise force out a byte. 753 */ 754 if (sbavail(&so->so_snd) && !tcp_timer_active(tp, TT_REXMT) && 755 !tcp_timer_active(tp, TT_PERSIST)) { 756 tp->t_rxtshift = 0; 757 tcp_setpersist(tp); 758 } 759 760 /* 761 * No reason to send a segment, just return. 762 */ 763 just_return: 764 SOCKBUF_UNLOCK(&so->so_snd); 765 return (0); 766 767 send: 768 SOCKBUF_LOCK_ASSERT(&so->so_snd); 769 if (len > 0) { 770 if (len >= tp->t_maxseg) 771 tp->t_flags2 |= TF2_PLPMTU_MAXSEGSNT; 772 else 773 tp->t_flags2 &= ~TF2_PLPMTU_MAXSEGSNT; 774 } 775 /* 776 * Before ESTABLISHED, force sending of initial options 777 * unless TCP set not to do any options. 778 * NOTE: we assume that the IP/TCP header plus TCP options 779 * always fit in a single mbuf, leaving room for a maximum 780 * link header, i.e. 781 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES 782 */ 783 optlen = 0; 784 #ifdef INET6 785 if (isipv6) 786 hdrlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 787 else 788 #endif 789 hdrlen = sizeof (struct tcpiphdr); 790 791 /* 792 * Compute options for segment. 793 * We only have to care about SYN and established connection 794 * segments. Options for SYN-ACK segments are handled in TCP 795 * syncache. 796 */ 797 to.to_flags = 0; 798 if ((tp->t_flags & TF_NOOPT) == 0) { 799 /* Maximum segment size. */ 800 if (flags & TH_SYN) { 801 tp->snd_nxt = tp->iss; 802 to.to_mss = tcp_mssopt(&tp->t_inpcb->inp_inc); 803 to.to_flags |= TOF_MSS; 804 #ifdef TCP_RFC7413 805 /* 806 * Only include the TFO option on the first 807 * transmission of the SYN|ACK on a 808 * passively-created TFO socket, as the presence of 809 * the TFO option may have caused the original 810 * SYN|ACK to have been dropped by a middlebox. 811 */ 812 if (IS_FASTOPEN(tp->t_flags) && 813 (tp->t_state == TCPS_SYN_RECEIVED) && 814 (tp->t_rxtshift == 0)) { 815 to.to_tfo_len = TCP_FASTOPEN_COOKIE_LEN; 816 to.to_tfo_cookie = (u_char *)&tp->t_tfo_cookie; 817 to.to_flags |= TOF_FASTOPEN; 818 } 819 #endif 820 } 821 /* Window scaling. */ 822 if ((flags & TH_SYN) && (tp->t_flags & TF_REQ_SCALE)) { 823 to.to_wscale = tp->request_r_scale; 824 to.to_flags |= TOF_SCALE; 825 } 826 /* Timestamps. */ 827 if ((tp->t_flags & TF_RCVD_TSTMP) || 828 ((flags & TH_SYN) && (tp->t_flags & TF_REQ_TSTMP))) { 829 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 830 to.to_tsecr = tp->ts_recent; 831 to.to_flags |= TOF_TS; 832 /* Set receive buffer autosizing timestamp. */ 833 if (tp->rfbuf_ts == 0 && 834 (so->so_rcv.sb_flags & SB_AUTOSIZE)) 835 tp->rfbuf_ts = tcp_ts_getticks(); 836 } 837 /* Selective ACK's. */ 838 if (tp->t_flags & TF_SACK_PERMIT) { 839 if (flags & TH_SYN) 840 to.to_flags |= TOF_SACKPERM; 841 else if (TCPS_HAVEESTABLISHED(tp->t_state) && 842 (tp->t_flags & TF_SACK_PERMIT) && 843 tp->rcv_numsacks > 0) { 844 to.to_flags |= TOF_SACK; 845 to.to_nsacks = tp->rcv_numsacks; 846 to.to_sacks = (u_char *)tp->sackblks; 847 } 848 } 849 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 850 /* TCP-MD5 (RFC2385). */ 851 /* 852 * Check that TCP_MD5SIG is enabled in tcpcb to 853 * account the size needed to set this TCP option. 854 */ 855 if (tp->t_flags & TF_SIGNATURE) 856 to.to_flags |= TOF_SIGNATURE; 857 #endif /* TCP_SIGNATURE */ 858 859 /* Processing the options. */ 860 hdrlen += optlen = tcp_addoptions(&to, opt); 861 } 862 863 /* 864 * Adjust data length if insertion of options will 865 * bump the packet length beyond the t_maxseg length. 866 * Clear the FIN bit because we cut off the tail of 867 * the segment. 868 */ 869 if (len + optlen + ipoptlen > tp->t_maxseg) { 870 flags &= ~TH_FIN; 871 872 if (tso) { 873 u_int if_hw_tsomax; 874 u_int if_hw_tsomaxsegcount; 875 u_int if_hw_tsomaxsegsize; 876 struct mbuf *mb; 877 u_int moff; 878 int max_len; 879 880 /* extract TSO information */ 881 if_hw_tsomax = tp->t_tsomax; 882 if_hw_tsomaxsegcount = tp->t_tsomaxsegcount; 883 if_hw_tsomaxsegsize = tp->t_tsomaxsegsize; 884 885 /* 886 * Limit a TSO burst to prevent it from 887 * overflowing or exceeding the maximum length 888 * allowed by the network interface: 889 */ 890 KASSERT(ipoptlen == 0, 891 ("%s: TSO can't do IP options", __func__)); 892 893 /* 894 * Check if we should limit by maximum payload 895 * length: 896 */ 897 if (if_hw_tsomax != 0) { 898 /* compute maximum TSO length */ 899 max_len = (if_hw_tsomax - hdrlen - 900 max_linkhdr); 901 if (max_len <= 0) { 902 len = 0; 903 } else if (len > max_len) { 904 sendalot = 1; 905 len = max_len; 906 } 907 } 908 909 /* 910 * Check if we should limit by maximum segment 911 * size and count: 912 */ 913 if (if_hw_tsomaxsegcount != 0 && 914 if_hw_tsomaxsegsize != 0) { 915 /* 916 * Subtract one segment for the LINK 917 * and TCP/IP headers mbuf that will 918 * be prepended to this mbuf chain 919 * after the code in this section 920 * limits the number of mbufs in the 921 * chain to if_hw_tsomaxsegcount. 922 */ 923 if_hw_tsomaxsegcount -= 1; 924 max_len = 0; 925 mb = sbsndmbuf(&so->so_snd, off, &moff); 926 927 while (mb != NULL && max_len < len) { 928 u_int mlen; 929 u_int frags; 930 931 /* 932 * Get length of mbuf fragment 933 * and how many hardware frags, 934 * rounded up, it would use: 935 */ 936 mlen = (mb->m_len - moff); 937 frags = howmany(mlen, 938 if_hw_tsomaxsegsize); 939 940 /* Handle special case: Zero Length Mbuf */ 941 if (frags == 0) 942 frags = 1; 943 944 /* 945 * Check if the fragment limit 946 * will be reached or exceeded: 947 */ 948 if (frags >= if_hw_tsomaxsegcount) { 949 max_len += min(mlen, 950 if_hw_tsomaxsegcount * 951 if_hw_tsomaxsegsize); 952 break; 953 } 954 max_len += mlen; 955 if_hw_tsomaxsegcount -= frags; 956 moff = 0; 957 mb = mb->m_next; 958 } 959 if (max_len <= 0) { 960 len = 0; 961 } else if (len > max_len) { 962 sendalot = 1; 963 len = max_len; 964 } 965 } 966 967 /* 968 * Prevent the last segment from being 969 * fractional unless the send sockbuf can be 970 * emptied: 971 */ 972 max_len = (tp->t_maxseg - optlen); 973 if (((uint32_t)off + (uint32_t)len) < 974 sbavail(&so->so_snd)) { 975 moff = len % max_len; 976 if (moff != 0) { 977 len -= moff; 978 sendalot = 1; 979 } 980 } 981 982 /* 983 * In case there are too many small fragments 984 * don't use TSO: 985 */ 986 if (len <= max_len) { 987 len = max_len; 988 sendalot = 1; 989 tso = 0; 990 } 991 992 /* 993 * Send the FIN in a separate segment 994 * after the bulk sending is done. 995 * We don't trust the TSO implementations 996 * to clear the FIN flag on all but the 997 * last segment. 998 */ 999 if (tp->t_flags & TF_NEEDFIN) 1000 sendalot = 1; 1001 1002 } else { 1003 len = tp->t_maxseg - optlen - ipoptlen; 1004 sendalot = 1; 1005 } 1006 } else 1007 tso = 0; 1008 1009 KASSERT(len + hdrlen + ipoptlen <= IP_MAXPACKET, 1010 ("%s: len > IP_MAXPACKET", __func__)); 1011 1012 /*#ifdef DIAGNOSTIC*/ 1013 #ifdef INET6 1014 if (max_linkhdr + hdrlen > MCLBYTES) 1015 #else 1016 if (max_linkhdr + hdrlen > MHLEN) 1017 #endif 1018 panic("tcphdr too big"); 1019 /*#endif*/ 1020 1021 /* 1022 * This KASSERT is here to catch edge cases at a well defined place. 1023 * Before, those had triggered (random) panic conditions further down. 1024 */ 1025 KASSERT(len >= 0, ("[%s:%d]: len < 0", __func__, __LINE__)); 1026 1027 /* 1028 * Grab a header mbuf, attaching a copy of data to 1029 * be transmitted, and initialize the header from 1030 * the template for sends on this connection. 1031 */ 1032 if (len) { 1033 struct mbuf *mb; 1034 u_int moff; 1035 1036 if ((tp->t_flags & TF_FORCEDATA) && len == 1) 1037 TCPSTAT_INC(tcps_sndprobe); 1038 else if (SEQ_LT(tp->snd_nxt, tp->snd_max) || sack_rxmit) { 1039 tp->t_sndrexmitpack++; 1040 TCPSTAT_INC(tcps_sndrexmitpack); 1041 TCPSTAT_ADD(tcps_sndrexmitbyte, len); 1042 } else { 1043 TCPSTAT_INC(tcps_sndpack); 1044 TCPSTAT_ADD(tcps_sndbyte, len); 1045 } 1046 #ifdef INET6 1047 if (MHLEN < hdrlen + max_linkhdr) 1048 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 1049 else 1050 #endif 1051 m = m_gethdr(M_NOWAIT, MT_DATA); 1052 1053 if (m == NULL) { 1054 SOCKBUF_UNLOCK(&so->so_snd); 1055 error = ENOBUFS; 1056 sack_rxmit = 0; 1057 goto out; 1058 } 1059 1060 m->m_data += max_linkhdr; 1061 m->m_len = hdrlen; 1062 1063 /* 1064 * Start the m_copy functions from the closest mbuf 1065 * to the offset in the socket buffer chain. 1066 */ 1067 mb = sbsndptr(&so->so_snd, off, len, &moff); 1068 1069 if (len <= MHLEN - hdrlen - max_linkhdr) { 1070 m_copydata(mb, moff, len, 1071 mtod(m, caddr_t) + hdrlen); 1072 m->m_len += len; 1073 } else { 1074 m->m_next = m_copym(mb, moff, len, M_NOWAIT); 1075 if (m->m_next == NULL) { 1076 SOCKBUF_UNLOCK(&so->so_snd); 1077 (void) m_free(m); 1078 error = ENOBUFS; 1079 sack_rxmit = 0; 1080 goto out; 1081 } 1082 } 1083 1084 /* 1085 * If we're sending everything we've got, set PUSH. 1086 * (This will keep happy those implementations which only 1087 * give data to the user when a buffer fills or 1088 * a PUSH comes in.) 1089 */ 1090 if (((uint32_t)off + (uint32_t)len == sbused(&so->so_snd)) && 1091 !(flags & TH_SYN)) 1092 flags |= TH_PUSH; 1093 SOCKBUF_UNLOCK(&so->so_snd); 1094 } else { 1095 SOCKBUF_UNLOCK(&so->so_snd); 1096 if (tp->t_flags & TF_ACKNOW) 1097 TCPSTAT_INC(tcps_sndacks); 1098 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 1099 TCPSTAT_INC(tcps_sndctrl); 1100 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 1101 TCPSTAT_INC(tcps_sndurg); 1102 else 1103 TCPSTAT_INC(tcps_sndwinup); 1104 1105 m = m_gethdr(M_NOWAIT, MT_DATA); 1106 if (m == NULL) { 1107 error = ENOBUFS; 1108 sack_rxmit = 0; 1109 goto out; 1110 } 1111 #ifdef INET6 1112 if (isipv6 && (MHLEN < hdrlen + max_linkhdr) && 1113 MHLEN >= hdrlen) { 1114 M_ALIGN(m, hdrlen); 1115 } else 1116 #endif 1117 m->m_data += max_linkhdr; 1118 m->m_len = hdrlen; 1119 } 1120 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); 1121 m->m_pkthdr.rcvif = (struct ifnet *)0; 1122 #ifdef MAC 1123 mac_inpcb_create_mbuf(tp->t_inpcb, m); 1124 #endif 1125 #ifdef INET6 1126 if (isipv6) { 1127 ip6 = mtod(m, struct ip6_hdr *); 1128 th = (struct tcphdr *)(ip6 + 1); 1129 tcpip_fillheaders(tp->t_inpcb, ip6, th); 1130 } else 1131 #endif /* INET6 */ 1132 { 1133 ip = mtod(m, struct ip *); 1134 ipov = (struct ipovly *)ip; 1135 th = (struct tcphdr *)(ip + 1); 1136 tcpip_fillheaders(tp->t_inpcb, ip, th); 1137 } 1138 1139 /* 1140 * Fill in fields, remembering maximum advertised 1141 * window for use in delaying messages about window sizes. 1142 * If resending a FIN, be sure not to use a new sequence number. 1143 */ 1144 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN && 1145 tp->snd_nxt == tp->snd_max) 1146 tp->snd_nxt--; 1147 /* 1148 * If we are starting a connection, send ECN setup 1149 * SYN packet. If we are on a retransmit, we may 1150 * resend those bits a number of times as per 1151 * RFC 3168. 1152 */ 1153 if (tp->t_state == TCPS_SYN_SENT && V_tcp_do_ecn == 1) { 1154 if (tp->t_rxtshift >= 1) { 1155 if (tp->t_rxtshift <= V_tcp_ecn_maxretries) 1156 flags |= TH_ECE|TH_CWR; 1157 } else 1158 flags |= TH_ECE|TH_CWR; 1159 } 1160 1161 if (tp->t_state == TCPS_ESTABLISHED && 1162 (tp->t_flags & TF_ECN_PERMIT)) { 1163 /* 1164 * If the peer has ECN, mark data packets with 1165 * ECN capable transmission (ECT). 1166 * Ignore pure ack packets, retransmissions and window probes. 1167 */ 1168 if (len > 0 && SEQ_GEQ(tp->snd_nxt, tp->snd_max) && 1169 !((tp->t_flags & TF_FORCEDATA) && len == 1)) { 1170 #ifdef INET6 1171 if (isipv6) 1172 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 1173 else 1174 #endif 1175 ip->ip_tos |= IPTOS_ECN_ECT0; 1176 TCPSTAT_INC(tcps_ecn_ect0); 1177 } 1178 1179 /* 1180 * Reply with proper ECN notifications. 1181 */ 1182 if (tp->t_flags & TF_ECN_SND_CWR) { 1183 flags |= TH_CWR; 1184 tp->t_flags &= ~TF_ECN_SND_CWR; 1185 } 1186 if (tp->t_flags & TF_ECN_SND_ECE) 1187 flags |= TH_ECE; 1188 } 1189 1190 /* 1191 * If we are doing retransmissions, then snd_nxt will 1192 * not reflect the first unsent octet. For ACK only 1193 * packets, we do not want the sequence number of the 1194 * retransmitted packet, we want the sequence number 1195 * of the next unsent octet. So, if there is no data 1196 * (and no SYN or FIN), use snd_max instead of snd_nxt 1197 * when filling in ti_seq. But if we are in persist 1198 * state, snd_max might reflect one byte beyond the 1199 * right edge of the window, so use snd_nxt in that 1200 * case, since we know we aren't doing a retransmission. 1201 * (retransmit and persist are mutually exclusive...) 1202 */ 1203 if (sack_rxmit == 0) { 1204 if (len || (flags & (TH_SYN|TH_FIN)) || 1205 tcp_timer_active(tp, TT_PERSIST)) 1206 th->th_seq = htonl(tp->snd_nxt); 1207 else 1208 th->th_seq = htonl(tp->snd_max); 1209 } else { 1210 th->th_seq = htonl(p->rxmit); 1211 p->rxmit += len; 1212 tp->sackhint.sack_bytes_rexmit += len; 1213 } 1214 th->th_ack = htonl(tp->rcv_nxt); 1215 if (optlen) { 1216 bcopy(opt, th + 1, optlen); 1217 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1218 } 1219 th->th_flags = flags; 1220 /* 1221 * Calculate receive window. Don't shrink window, 1222 * but avoid silly window syndrome. 1223 */ 1224 if (recwin < (so->so_rcv.sb_hiwat / 4) && 1225 recwin < tp->t_maxseg) 1226 recwin = 0; 1227 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt) && 1228 recwin < (tp->rcv_adv - tp->rcv_nxt)) 1229 recwin = (tp->rcv_adv - tp->rcv_nxt); 1230 1231 /* 1232 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1233 * or <SYN,ACK>) segment itself is never scaled. The <SYN,ACK> 1234 * case is handled in syncache. 1235 */ 1236 if (flags & TH_SYN) 1237 th->th_win = htons((u_short) 1238 (min(sbspace(&so->so_rcv), TCP_MAXWIN))); 1239 else 1240 th->th_win = htons((u_short)(recwin >> tp->rcv_scale)); 1241 1242 /* 1243 * Adjust the RXWIN0SENT flag - indicate that we have advertised 1244 * a 0 window. This may cause the remote transmitter to stall. This 1245 * flag tells soreceive() to disable delayed acknowledgements when 1246 * draining the buffer. This can occur if the receiver is attempting 1247 * to read more data than can be buffered prior to transmitting on 1248 * the connection. 1249 */ 1250 if (th->th_win == 0) { 1251 tp->t_sndzerowin++; 1252 tp->t_flags |= TF_RXWIN0SENT; 1253 } else 1254 tp->t_flags &= ~TF_RXWIN0SENT; 1255 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 1256 th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt)); 1257 th->th_flags |= TH_URG; 1258 } else 1259 /* 1260 * If no urgent pointer to send, then we pull 1261 * the urgent pointer to the left edge of the send window 1262 * so that it doesn't drift into the send window on sequence 1263 * number wraparound. 1264 */ 1265 tp->snd_up = tp->snd_una; /* drag it along */ 1266 1267 /* 1268 * Put TCP length in extended header, and then 1269 * checksum extended header and data. 1270 */ 1271 m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */ 1272 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1273 1274 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1275 if (to.to_flags & TOF_SIGNATURE) { 1276 /* 1277 * Calculate MD5 signature and put it into the place 1278 * determined before. 1279 * NOTE: since TCP options buffer doesn't point into 1280 * mbuf's data, calculate offset and use it. 1281 */ 1282 if (!TCPMD5_ENABLED() || TCPMD5_OUTPUT(m, th, 1283 (u_char *)(th + 1) + (to.to_signature - opt)) != 0) { 1284 /* 1285 * Do not send segment if the calculation of MD5 1286 * digest has failed. 1287 */ 1288 goto out; 1289 } 1290 } 1291 #endif 1292 #ifdef INET6 1293 if (isipv6) { 1294 /* 1295 * There is no need to fill in ip6_plen right now. 1296 * It will be filled later by ip6_output. 1297 */ 1298 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 1299 th->th_sum = in6_cksum_pseudo(ip6, sizeof(struct tcphdr) + 1300 optlen + len, IPPROTO_TCP, 0); 1301 } 1302 #endif 1303 #if defined(INET6) && defined(INET) 1304 else 1305 #endif 1306 #ifdef INET 1307 { 1308 m->m_pkthdr.csum_flags = CSUM_TCP; 1309 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1310 htons(sizeof(struct tcphdr) + IPPROTO_TCP + len + optlen)); 1311 1312 /* IP version must be set here for ipv4/ipv6 checking later */ 1313 KASSERT(ip->ip_v == IPVERSION, 1314 ("%s: IP version incorrect: %d", __func__, ip->ip_v)); 1315 } 1316 #endif 1317 1318 /* 1319 * Enable TSO and specify the size of the segments. 1320 * The TCP pseudo header checksum is always provided. 1321 */ 1322 if (tso) { 1323 KASSERT(len > tp->t_maxseg - optlen, 1324 ("%s: len <= tso_segsz", __func__)); 1325 m->m_pkthdr.csum_flags |= CSUM_TSO; 1326 m->m_pkthdr.tso_segsz = tp->t_maxseg - optlen; 1327 } 1328 1329 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1330 KASSERT(len + hdrlen + ipoptlen - ipsec_optlen == m_length(m, NULL), 1331 ("%s: mbuf chain shorter than expected: %d + %u + %u - %u != %u", 1332 __func__, len, hdrlen, ipoptlen, ipsec_optlen, m_length(m, NULL))); 1333 #else 1334 KASSERT(len + hdrlen + ipoptlen == m_length(m, NULL), 1335 ("%s: mbuf chain shorter than expected: %d + %u + %u != %u", 1336 __func__, len, hdrlen, ipoptlen, m_length(m, NULL))); 1337 #endif 1338 1339 #ifdef TCP_HHOOK 1340 /* Run HHOOK_TCP_ESTABLISHED_OUT helper hooks. */ 1341 hhook_run_tcp_est_out(tp, th, &to, len, tso); 1342 #endif 1343 1344 #ifdef TCPDEBUG 1345 /* 1346 * Trace. 1347 */ 1348 if (so->so_options & SO_DEBUG) { 1349 u_short save = 0; 1350 #ifdef INET6 1351 if (!isipv6) 1352 #endif 1353 { 1354 save = ipov->ih_len; 1355 ipov->ih_len = htons(m->m_pkthdr.len /* - hdrlen + (th->th_off << 2) */); 1356 } 1357 tcp_trace(TA_OUTPUT, tp->t_state, tp, mtod(m, void *), th, 0); 1358 #ifdef INET6 1359 if (!isipv6) 1360 #endif 1361 ipov->ih_len = save; 1362 } 1363 #endif /* TCPDEBUG */ 1364 TCP_PROBE3(debug__output, tp, th, m); 1365 1366 /* 1367 * Fill in IP length and desired time to live and 1368 * send to IP level. There should be a better way 1369 * to handle ttl and tos; we could keep them in 1370 * the template, but need a way to checksum without them. 1371 */ 1372 /* 1373 * m->m_pkthdr.len should have been set before checksum calculation, 1374 * because in6_cksum() need it. 1375 */ 1376 #ifdef INET6 1377 if (isipv6) { 1378 struct route_in6 ro; 1379 1380 bzero(&ro, sizeof(ro)); 1381 /* 1382 * we separately set hoplimit for every segment, since the 1383 * user might want to change the value via setsockopt. 1384 * Also, desired default hop limit might be changed via 1385 * Neighbor Discovery. 1386 */ 1387 ip6->ip6_hlim = in6_selecthlim(tp->t_inpcb, NULL); 1388 1389 /* 1390 * Set the packet size here for the benefit of DTrace probes. 1391 * ip6_output() will set it properly; it's supposed to include 1392 * the option header lengths as well. 1393 */ 1394 ip6->ip6_plen = htons(m->m_pkthdr.len - sizeof(*ip6)); 1395 1396 if (V_path_mtu_discovery && tp->t_maxseg > V_tcp_minmss) 1397 tp->t_flags2 |= TF2_PLPMTU_PMTUD; 1398 else 1399 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD; 1400 1401 if (tp->t_state == TCPS_SYN_SENT) 1402 TCP_PROBE5(connect__request, NULL, tp, ip6, tp, th); 1403 1404 TCP_PROBE5(send, NULL, tp, ip6, tp, th); 1405 1406 #ifdef TCPPCAP 1407 /* Save packet, if requested. */ 1408 tcp_pcap_add(th, m, &(tp->t_outpkts)); 1409 #endif 1410 1411 /* TODO: IPv6 IP6TOS_ECT bit on */ 1412 error = ip6_output(m, tp->t_inpcb->in6p_outputopts, &ro, 1413 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 1414 NULL, NULL, tp->t_inpcb); 1415 1416 if (error == EMSGSIZE && ro.ro_rt != NULL) 1417 mtu = ro.ro_rt->rt_mtu; 1418 RO_RTFREE(&ro); 1419 } 1420 #endif /* INET6 */ 1421 #if defined(INET) && defined(INET6) 1422 else 1423 #endif 1424 #ifdef INET 1425 { 1426 ip->ip_len = htons(m->m_pkthdr.len); 1427 #ifdef INET6 1428 if (tp->t_inpcb->inp_vflag & INP_IPV6PROTO) 1429 ip->ip_ttl = in6_selecthlim(tp->t_inpcb, NULL); 1430 #endif /* INET6 */ 1431 /* 1432 * If we do path MTU discovery, then we set DF on every packet. 1433 * This might not be the best thing to do according to RFC3390 1434 * Section 2. However the tcp hostcache migitates the problem 1435 * so it affects only the first tcp connection with a host. 1436 * 1437 * NB: Don't set DF on small MTU/MSS to have a safe fallback. 1438 */ 1439 if (V_path_mtu_discovery && tp->t_maxseg > V_tcp_minmss) { 1440 ip->ip_off |= htons(IP_DF); 1441 tp->t_flags2 |= TF2_PLPMTU_PMTUD; 1442 } else { 1443 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD; 1444 } 1445 1446 if (tp->t_state == TCPS_SYN_SENT) 1447 TCP_PROBE5(connect__request, NULL, tp, ip, tp, th); 1448 1449 TCP_PROBE5(send, NULL, tp, ip, tp, th); 1450 1451 #ifdef TCPPCAP 1452 /* Save packet, if requested. */ 1453 tcp_pcap_add(th, m, &(tp->t_outpkts)); 1454 #endif 1455 1456 error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route, 1457 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 0, 1458 tp->t_inpcb); 1459 1460 if (error == EMSGSIZE && tp->t_inpcb->inp_route.ro_rt != NULL) 1461 mtu = tp->t_inpcb->inp_route.ro_rt->rt_mtu; 1462 } 1463 #endif /* INET */ 1464 1465 out: 1466 /* 1467 * In transmit state, time the transmission and arrange for 1468 * the retransmit. In persist state, just set snd_max. 1469 */ 1470 if ((tp->t_flags & TF_FORCEDATA) == 0 || 1471 !tcp_timer_active(tp, TT_PERSIST)) { 1472 tcp_seq startseq = tp->snd_nxt; 1473 1474 /* 1475 * Advance snd_nxt over sequence space of this segment. 1476 */ 1477 if (flags & (TH_SYN|TH_FIN)) { 1478 if (flags & TH_SYN) 1479 tp->snd_nxt++; 1480 if (flags & TH_FIN) { 1481 tp->snd_nxt++; 1482 tp->t_flags |= TF_SENTFIN; 1483 } 1484 } 1485 if (sack_rxmit) 1486 goto timer; 1487 tp->snd_nxt += len; 1488 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 1489 tp->snd_max = tp->snd_nxt; 1490 /* 1491 * Time this transmission if not a retransmission and 1492 * not currently timing anything. 1493 */ 1494 if (tp->t_rtttime == 0) { 1495 tp->t_rtttime = ticks; 1496 tp->t_rtseq = startseq; 1497 TCPSTAT_INC(tcps_segstimed); 1498 } 1499 } 1500 1501 /* 1502 * Set retransmit timer if not currently set, 1503 * and not doing a pure ack or a keep-alive probe. 1504 * Initial value for retransmit timer is smoothed 1505 * round-trip time + 2 * round-trip time variance. 1506 * Initialize shift counter which is used for backoff 1507 * of retransmit time. 1508 */ 1509 timer: 1510 if (!tcp_timer_active(tp, TT_REXMT) && 1511 ((sack_rxmit && tp->snd_nxt != tp->snd_max) || 1512 (tp->snd_nxt != tp->snd_una))) { 1513 if (tcp_timer_active(tp, TT_PERSIST)) { 1514 tcp_timer_activate(tp, TT_PERSIST, 0); 1515 tp->t_rxtshift = 0; 1516 } 1517 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 1518 } else if (len == 0 && sbavail(&so->so_snd) && 1519 !tcp_timer_active(tp, TT_REXMT) && 1520 !tcp_timer_active(tp, TT_PERSIST)) { 1521 /* 1522 * Avoid a situation where we do not set persist timer 1523 * after a zero window condition. For example: 1524 * 1) A -> B: packet with enough data to fill the window 1525 * 2) B -> A: ACK for #1 + new data (0 window 1526 * advertisement) 1527 * 3) A -> B: ACK for #2, 0 len packet 1528 * 1529 * In this case, A will not activate the persist timer, 1530 * because it chose to send a packet. Unless tcp_output 1531 * is called for some other reason (delayed ack timer, 1532 * another input packet from B, socket syscall), A will 1533 * not send zero window probes. 1534 * 1535 * So, if you send a 0-length packet, but there is data 1536 * in the socket buffer, and neither the rexmt or 1537 * persist timer is already set, then activate the 1538 * persist timer. 1539 */ 1540 tp->t_rxtshift = 0; 1541 tcp_setpersist(tp); 1542 } 1543 } else { 1544 /* 1545 * Persist case, update snd_max but since we are in 1546 * persist mode (no window) we do not update snd_nxt. 1547 */ 1548 int xlen = len; 1549 if (flags & TH_SYN) 1550 ++xlen; 1551 if (flags & TH_FIN) { 1552 ++xlen; 1553 tp->t_flags |= TF_SENTFIN; 1554 } 1555 if (SEQ_GT(tp->snd_nxt + xlen, tp->snd_max)) 1556 tp->snd_max = tp->snd_nxt + xlen; 1557 } 1558 1559 if (error) { 1560 1561 /* 1562 * We know that the packet was lost, so back out the 1563 * sequence number advance, if any. 1564 * 1565 * If the error is EPERM the packet got blocked by the 1566 * local firewall. Normally we should terminate the 1567 * connection but the blocking may have been spurious 1568 * due to a firewall reconfiguration cycle. So we treat 1569 * it like a packet loss and let the retransmit timer and 1570 * timeouts do their work over time. 1571 * XXX: It is a POLA question whether calling tcp_drop right 1572 * away would be the really correct behavior instead. 1573 */ 1574 if (((tp->t_flags & TF_FORCEDATA) == 0 || 1575 !tcp_timer_active(tp, TT_PERSIST)) && 1576 ((flags & TH_SYN) == 0) && 1577 (error != EPERM)) { 1578 if (sack_rxmit) { 1579 p->rxmit -= len; 1580 tp->sackhint.sack_bytes_rexmit -= len; 1581 KASSERT(tp->sackhint.sack_bytes_rexmit >= 0, 1582 ("sackhint bytes rtx >= 0")); 1583 } else 1584 tp->snd_nxt -= len; 1585 } 1586 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); /* Check gotos. */ 1587 switch (error) { 1588 case EACCES: 1589 tp->t_softerror = error; 1590 return (0); 1591 case EPERM: 1592 tp->t_softerror = error; 1593 return (error); 1594 case ENOBUFS: 1595 TCP_XMIT_TIMER_ASSERT(tp, len, flags); 1596 tp->snd_cwnd = tp->t_maxseg; 1597 return (0); 1598 case EMSGSIZE: 1599 /* 1600 * For some reason the interface we used initially 1601 * to send segments changed to another or lowered 1602 * its MTU. 1603 * If TSO was active we either got an interface 1604 * without TSO capabilits or TSO was turned off. 1605 * If we obtained mtu from ip_output() then update 1606 * it and try again. 1607 */ 1608 if (tso) 1609 tp->t_flags &= ~TF_TSO; 1610 if (mtu != 0) { 1611 tcp_mss_update(tp, -1, mtu, NULL, NULL); 1612 goto again; 1613 } 1614 return (error); 1615 case EHOSTDOWN: 1616 case EHOSTUNREACH: 1617 case ENETDOWN: 1618 case ENETUNREACH: 1619 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1620 tp->t_softerror = error; 1621 return (0); 1622 } 1623 /* FALLTHROUGH */ 1624 default: 1625 return (error); 1626 } 1627 } 1628 TCPSTAT_INC(tcps_sndtotal); 1629 1630 /* 1631 * Data sent (as far as we can tell). 1632 * If this advertises a larger window than any other segment, 1633 * then remember the size of the advertised window. 1634 * Any pending ACK has now been sent. 1635 */ 1636 if (SEQ_GT(tp->rcv_nxt + recwin, tp->rcv_adv)) 1637 tp->rcv_adv = tp->rcv_nxt + recwin; 1638 tp->last_ack_sent = tp->rcv_nxt; 1639 tp->t_flags &= ~(TF_ACKNOW | TF_DELACK); 1640 if (tcp_timer_active(tp, TT_DELACK)) 1641 tcp_timer_activate(tp, TT_DELACK, 0); 1642 #if 0 1643 /* 1644 * This completely breaks TCP if newreno is turned on. What happens 1645 * is that if delayed-acks are turned on on the receiver, this code 1646 * on the transmitter effectively destroys the TCP window, forcing 1647 * it to four packets (1.5Kx4 = 6K window). 1648 */ 1649 if (sendalot && --maxburst) 1650 goto again; 1651 #endif 1652 if (sendalot) 1653 goto again; 1654 return (0); 1655 } 1656 1657 void 1658 tcp_setpersist(struct tcpcb *tp) 1659 { 1660 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1; 1661 int tt; 1662 1663 tp->t_flags &= ~TF_PREVVALID; 1664 if (tcp_timer_active(tp, TT_REXMT)) 1665 panic("tcp_setpersist: retransmit pending"); 1666 /* 1667 * Start/restart persistence timer. 1668 */ 1669 TCPT_RANGESET(tt, t * tcp_backoff[tp->t_rxtshift], 1670 tcp_persmin, tcp_persmax); 1671 tcp_timer_activate(tp, TT_PERSIST, tt); 1672 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 1673 tp->t_rxtshift++; 1674 } 1675 1676 /* 1677 * Insert TCP options according to the supplied parameters to the place 1678 * optp in a consistent way. Can handle unaligned destinations. 1679 * 1680 * The order of the option processing is crucial for optimal packing and 1681 * alignment for the scarce option space. 1682 * 1683 * The optimal order for a SYN/SYN-ACK segment is: 1684 * MSS (4) + NOP (1) + Window scale (3) + SACK permitted (2) + 1685 * Timestamp (10) + Signature (18) = 38 bytes out of a maximum of 40. 1686 * 1687 * The SACK options should be last. SACK blocks consume 8*n+2 bytes. 1688 * So a full size SACK blocks option is 34 bytes (with 4 SACK blocks). 1689 * At minimum we need 10 bytes (to generate 1 SACK block). If both 1690 * TCP Timestamps (12 bytes) and TCP Signatures (18 bytes) are present, 1691 * we only have 10 bytes for SACK options (40 - (12 + 18)). 1692 */ 1693 int 1694 tcp_addoptions(struct tcpopt *to, u_char *optp) 1695 { 1696 u_int32_t mask, optlen = 0; 1697 1698 for (mask = 1; mask < TOF_MAXOPT; mask <<= 1) { 1699 if ((to->to_flags & mask) != mask) 1700 continue; 1701 if (optlen == TCP_MAXOLEN) 1702 break; 1703 switch (to->to_flags & mask) { 1704 case TOF_MSS: 1705 while (optlen % 4) { 1706 optlen += TCPOLEN_NOP; 1707 *optp++ = TCPOPT_NOP; 1708 } 1709 if (TCP_MAXOLEN - optlen < TCPOLEN_MAXSEG) 1710 continue; 1711 optlen += TCPOLEN_MAXSEG; 1712 *optp++ = TCPOPT_MAXSEG; 1713 *optp++ = TCPOLEN_MAXSEG; 1714 to->to_mss = htons(to->to_mss); 1715 bcopy((u_char *)&to->to_mss, optp, sizeof(to->to_mss)); 1716 optp += sizeof(to->to_mss); 1717 break; 1718 case TOF_SCALE: 1719 while (!optlen || optlen % 2 != 1) { 1720 optlen += TCPOLEN_NOP; 1721 *optp++ = TCPOPT_NOP; 1722 } 1723 if (TCP_MAXOLEN - optlen < TCPOLEN_WINDOW) 1724 continue; 1725 optlen += TCPOLEN_WINDOW; 1726 *optp++ = TCPOPT_WINDOW; 1727 *optp++ = TCPOLEN_WINDOW; 1728 *optp++ = to->to_wscale; 1729 break; 1730 case TOF_SACKPERM: 1731 while (optlen % 2) { 1732 optlen += TCPOLEN_NOP; 1733 *optp++ = TCPOPT_NOP; 1734 } 1735 if (TCP_MAXOLEN - optlen < TCPOLEN_SACK_PERMITTED) 1736 continue; 1737 optlen += TCPOLEN_SACK_PERMITTED; 1738 *optp++ = TCPOPT_SACK_PERMITTED; 1739 *optp++ = TCPOLEN_SACK_PERMITTED; 1740 break; 1741 case TOF_TS: 1742 while (!optlen || optlen % 4 != 2) { 1743 optlen += TCPOLEN_NOP; 1744 *optp++ = TCPOPT_NOP; 1745 } 1746 if (TCP_MAXOLEN - optlen < TCPOLEN_TIMESTAMP) 1747 continue; 1748 optlen += TCPOLEN_TIMESTAMP; 1749 *optp++ = TCPOPT_TIMESTAMP; 1750 *optp++ = TCPOLEN_TIMESTAMP; 1751 to->to_tsval = htonl(to->to_tsval); 1752 to->to_tsecr = htonl(to->to_tsecr); 1753 bcopy((u_char *)&to->to_tsval, optp, sizeof(to->to_tsval)); 1754 optp += sizeof(to->to_tsval); 1755 bcopy((u_char *)&to->to_tsecr, optp, sizeof(to->to_tsecr)); 1756 optp += sizeof(to->to_tsecr); 1757 break; 1758 case TOF_SIGNATURE: 1759 { 1760 int siglen = TCPOLEN_SIGNATURE - 2; 1761 1762 while (!optlen || optlen % 4 != 2) { 1763 optlen += TCPOLEN_NOP; 1764 *optp++ = TCPOPT_NOP; 1765 } 1766 if (TCP_MAXOLEN - optlen < TCPOLEN_SIGNATURE) { 1767 to->to_flags &= ~TOF_SIGNATURE; 1768 continue; 1769 } 1770 optlen += TCPOLEN_SIGNATURE; 1771 *optp++ = TCPOPT_SIGNATURE; 1772 *optp++ = TCPOLEN_SIGNATURE; 1773 to->to_signature = optp; 1774 while (siglen--) 1775 *optp++ = 0; 1776 break; 1777 } 1778 case TOF_SACK: 1779 { 1780 int sackblks = 0; 1781 struct sackblk *sack = (struct sackblk *)to->to_sacks; 1782 tcp_seq sack_seq; 1783 1784 while (!optlen || optlen % 4 != 2) { 1785 optlen += TCPOLEN_NOP; 1786 *optp++ = TCPOPT_NOP; 1787 } 1788 if (TCP_MAXOLEN - optlen < TCPOLEN_SACKHDR + TCPOLEN_SACK) 1789 continue; 1790 optlen += TCPOLEN_SACKHDR; 1791 *optp++ = TCPOPT_SACK; 1792 sackblks = min(to->to_nsacks, 1793 (TCP_MAXOLEN - optlen) / TCPOLEN_SACK); 1794 *optp++ = TCPOLEN_SACKHDR + sackblks * TCPOLEN_SACK; 1795 while (sackblks--) { 1796 sack_seq = htonl(sack->start); 1797 bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq)); 1798 optp += sizeof(sack_seq); 1799 sack_seq = htonl(sack->end); 1800 bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq)); 1801 optp += sizeof(sack_seq); 1802 optlen += TCPOLEN_SACK; 1803 sack++; 1804 } 1805 TCPSTAT_INC(tcps_sack_send_blocks); 1806 break; 1807 } 1808 #ifdef TCP_RFC7413 1809 case TOF_FASTOPEN: 1810 { 1811 int total_len; 1812 1813 /* XXX is there any point to aligning this option? */ 1814 total_len = TCPOLEN_FAST_OPEN_EMPTY + to->to_tfo_len; 1815 if (TCP_MAXOLEN - optlen < total_len) 1816 continue; 1817 *optp++ = TCPOPT_FAST_OPEN; 1818 *optp++ = total_len; 1819 if (to->to_tfo_len > 0) { 1820 bcopy(to->to_tfo_cookie, optp, to->to_tfo_len); 1821 optp += to->to_tfo_len; 1822 } 1823 optlen += total_len; 1824 break; 1825 } 1826 #endif 1827 default: 1828 panic("%s: unknown TCP option type", __func__); 1829 break; 1830 } 1831 } 1832 1833 /* Terminate and pad TCP options to a 4 byte boundary. */ 1834 if (optlen % 4) { 1835 optlen += TCPOLEN_EOL; 1836 *optp++ = TCPOPT_EOL; 1837 } 1838 /* 1839 * According to RFC 793 (STD0007): 1840 * "The content of the header beyond the End-of-Option option 1841 * must be header padding (i.e., zero)." 1842 * and later: "The padding is composed of zeros." 1843 */ 1844 while (optlen % 4) { 1845 optlen += TCPOLEN_PAD; 1846 *optp++ = TCPOPT_PAD; 1847 } 1848 1849 KASSERT(optlen <= TCP_MAXOLEN, ("%s: TCP options too long", __func__)); 1850 return (optlen); 1851 } 1852