1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. 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 VNET_DEFINE(int, tcp_sendbuf_auto_lowat) = 0; 134 #define V_tcp_sendbuf_auto_lowat VNET(tcp_sendbuf_auto_lowat) 135 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sendbuf_auto_lowat, CTLFLAG_VNET | CTLFLAG_RW, 136 &VNET_NAME(tcp_sendbuf_auto_lowat), 0, 137 "Modify threshold for auto send buffer growth to account for SO_SNDLOWAT"); 138 139 /* 140 * Make sure that either retransmit or persist timer is set for SYN, FIN and 141 * non-ACK. 142 */ 143 #define TCP_XMIT_TIMER_ASSERT(tp, len, th_flags) \ 144 KASSERT(((len) == 0 && ((th_flags) & (TH_SYN | TH_FIN)) == 0) ||\ 145 tcp_timer_active((tp), TT_REXMT) || \ 146 tcp_timer_active((tp), TT_PERSIST), \ 147 ("neither rexmt nor persist timer is set")) 148 149 #ifdef TCP_HHOOK 150 static void inline hhook_run_tcp_est_out(struct tcpcb *tp, 151 struct tcphdr *th, struct tcpopt *to, 152 uint32_t len, int tso); 153 #endif 154 static void inline cc_after_idle(struct tcpcb *tp); 155 156 #ifdef TCP_HHOOK 157 /* 158 * Wrapper for the TCP established output helper hook. 159 */ 160 static void inline 161 hhook_run_tcp_est_out(struct tcpcb *tp, struct tcphdr *th, 162 struct tcpopt *to, uint32_t len, int tso) 163 { 164 struct tcp_hhook_data hhook_data; 165 166 if (V_tcp_hhh[HHOOK_TCP_EST_OUT]->hhh_nhooks > 0) { 167 hhook_data.tp = tp; 168 hhook_data.th = th; 169 hhook_data.to = to; 170 hhook_data.len = len; 171 hhook_data.tso = tso; 172 173 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_OUT], &hhook_data, 174 tp->osd); 175 } 176 } 177 #endif 178 179 /* 180 * CC wrapper hook functions 181 */ 182 static void inline 183 cc_after_idle(struct tcpcb *tp) 184 { 185 INP_WLOCK_ASSERT(tp->t_inpcb); 186 187 if (CC_ALGO(tp)->after_idle != NULL) 188 CC_ALGO(tp)->after_idle(tp->ccv); 189 } 190 191 /* 192 * Tcp output routine: figure out what should be sent and send it. 193 */ 194 int 195 tcp_output(struct tcpcb *tp) 196 { 197 struct socket *so = tp->t_inpcb->inp_socket; 198 int32_t len; 199 uint32_t recwin, sendwin; 200 int off, flags, error = 0; /* Keep compiler happy */ 201 struct mbuf *m; 202 struct ip *ip = NULL; 203 struct ipovly *ipov = NULL; 204 struct tcphdr *th; 205 u_char opt[TCP_MAXOLEN]; 206 unsigned ipoptlen, optlen, hdrlen; 207 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 208 unsigned ipsec_optlen = 0; 209 #endif 210 int idle, sendalot; 211 int sack_rxmit, sack_bytes_rxmt; 212 struct sackhole *p; 213 int tso, mtu; 214 struct tcpopt to; 215 #if 0 216 int maxburst = TCP_MAXBURST; 217 #endif 218 #ifdef INET6 219 struct ip6_hdr *ip6 = NULL; 220 int isipv6; 221 222 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 223 #endif 224 225 INP_WLOCK_ASSERT(tp->t_inpcb); 226 227 #ifdef TCP_OFFLOAD 228 if (tp->t_flags & TF_TOE) 229 return (tcp_offload_output(tp)); 230 #endif 231 232 #ifdef TCP_RFC7413 233 /* 234 * For TFO connections in SYN_RECEIVED, only allow the initial 235 * SYN|ACK and those sent by the retransmit timer. 236 */ 237 if (IS_FASTOPEN(tp->t_flags) && 238 (tp->t_state == TCPS_SYN_RECEIVED) && 239 SEQ_GT(tp->snd_max, tp->snd_una) && /* initial SYN|ACK sent */ 240 (tp->snd_nxt != tp->snd_una)) /* not a retransmit */ 241 return (0); 242 #endif 243 /* 244 * Determine length of data that should be transmitted, 245 * and flags that will be used. 246 * If there is some data or critical controls (SYN, RST) 247 * to send, then transmit; otherwise, investigate further. 248 */ 249 idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una); 250 if (idle && ticks - tp->t_rcvtime >= tp->t_rxtcur) 251 cc_after_idle(tp); 252 tp->t_flags &= ~TF_LASTIDLE; 253 if (idle) { 254 if (tp->t_flags & TF_MORETOCOME) { 255 tp->t_flags |= TF_LASTIDLE; 256 idle = 0; 257 } 258 } 259 again: 260 /* 261 * If we've recently taken a timeout, snd_max will be greater than 262 * snd_nxt. There may be SACK information that allows us to avoid 263 * resending already delivered data. Adjust snd_nxt accordingly. 264 */ 265 if ((tp->t_flags & TF_SACK_PERMIT) && 266 SEQ_LT(tp->snd_nxt, tp->snd_max)) 267 tcp_sack_adjust(tp); 268 sendalot = 0; 269 tso = 0; 270 mtu = 0; 271 off = tp->snd_nxt - tp->snd_una; 272 sendwin = min(tp->snd_wnd, tp->snd_cwnd); 273 274 flags = tcp_outflags[tp->t_state]; 275 /* 276 * Send any SACK-generated retransmissions. If we're explicitly trying 277 * to send out new data (when sendalot is 1), bypass this function. 278 * If we retransmit in fast recovery mode, decrement snd_cwnd, since 279 * we're replacing a (future) new transmission with a retransmission 280 * now, and we previously incremented snd_cwnd in tcp_input(). 281 */ 282 /* 283 * Still in sack recovery , reset rxmit flag to zero. 284 */ 285 sack_rxmit = 0; 286 sack_bytes_rxmt = 0; 287 len = 0; 288 p = NULL; 289 if ((tp->t_flags & TF_SACK_PERMIT) && IN_FASTRECOVERY(tp->t_flags) && 290 (p = tcp_sack_output(tp, &sack_bytes_rxmt))) { 291 uint32_t cwin; 292 293 cwin = 294 imax(min(tp->snd_wnd, tp->snd_cwnd) - sack_bytes_rxmt, 0); 295 /* Do not retransmit SACK segments beyond snd_recover */ 296 if (SEQ_GT(p->end, tp->snd_recover)) { 297 /* 298 * (At least) part of sack hole extends beyond 299 * snd_recover. Check to see if we can rexmit data 300 * for this hole. 301 */ 302 if (SEQ_GEQ(p->rxmit, tp->snd_recover)) { 303 /* 304 * Can't rexmit any more data for this hole. 305 * That data will be rexmitted in the next 306 * sack recovery episode, when snd_recover 307 * moves past p->rxmit. 308 */ 309 p = NULL; 310 goto after_sack_rexmit; 311 } else 312 /* Can rexmit part of the current hole */ 313 len = ((int32_t)ulmin(cwin, 314 tp->snd_recover - p->rxmit)); 315 } else 316 len = ((int32_t)ulmin(cwin, p->end - p->rxmit)); 317 off = p->rxmit - tp->snd_una; 318 KASSERT(off >= 0,("%s: sack block to the left of una : %d", 319 __func__, off)); 320 if (len > 0) { 321 sack_rxmit = 1; 322 sendalot = 1; 323 TCPSTAT_INC(tcps_sack_rexmits); 324 TCPSTAT_ADD(tcps_sack_rexmit_bytes, 325 min(len, tp->t_maxseg)); 326 } 327 } 328 after_sack_rexmit: 329 /* 330 * Get standard flags, and add SYN or FIN if requested by 'hidden' 331 * state flags. 332 */ 333 if (tp->t_flags & TF_NEEDFIN) 334 flags |= TH_FIN; 335 if (tp->t_flags & TF_NEEDSYN) 336 flags |= TH_SYN; 337 338 SOCKBUF_LOCK(&so->so_snd); 339 /* 340 * If in persist timeout with window of 0, send 1 byte. 341 * Otherwise, if window is small but nonzero 342 * and timer expired, we will send what we can 343 * and go to transmit state. 344 */ 345 if (tp->t_flags & TF_FORCEDATA) { 346 if (sendwin == 0) { 347 /* 348 * If we still have some data to send, then 349 * clear the FIN bit. Usually this would 350 * happen below when it realizes that we 351 * aren't sending all the data. However, 352 * if we have exactly 1 byte of unsent data, 353 * then it won't clear the FIN bit below, 354 * and if we are in persist state, we wind 355 * up sending the packet without recording 356 * that we sent the FIN bit. 357 * 358 * We can't just blindly clear the FIN bit, 359 * because if we don't have any more data 360 * to send then the probe will be the FIN 361 * itself. 362 */ 363 if (off < sbused(&so->so_snd)) 364 flags &= ~TH_FIN; 365 sendwin = 1; 366 } else { 367 tcp_timer_activate(tp, TT_PERSIST, 0); 368 tp->t_rxtshift = 0; 369 } 370 } 371 372 /* 373 * If snd_nxt == snd_max and we have transmitted a FIN, the 374 * offset will be > 0 even if so_snd.sb_cc is 0, resulting in 375 * a negative length. This can also occur when TCP opens up 376 * its congestion window while receiving additional duplicate 377 * acks after fast-retransmit because TCP will reset snd_nxt 378 * to snd_max after the fast-retransmit. 379 * 380 * In the normal retransmit-FIN-only case, however, snd_nxt will 381 * be set to snd_una, the offset will be 0, and the length may 382 * wind up 0. 383 * 384 * If sack_rxmit is true we are retransmitting from the scoreboard 385 * in which case len is already set. 386 */ 387 if (sack_rxmit == 0) { 388 if (sack_bytes_rxmt == 0) 389 len = ((int32_t)min(sbavail(&so->so_snd), sendwin) - 390 off); 391 else { 392 int32_t cwin; 393 394 /* 395 * We are inside of a SACK recovery episode and are 396 * sending new data, having retransmitted all the 397 * data possible in the scoreboard. 398 */ 399 len = ((int32_t)min(sbavail(&so->so_snd), tp->snd_wnd) - 400 off); 401 /* 402 * Don't remove this (len > 0) check ! 403 * We explicitly check for len > 0 here (although it 404 * isn't really necessary), to work around a gcc 405 * optimization issue - to force gcc to compute 406 * len above. Without this check, the computation 407 * of len is bungled by the optimizer. 408 */ 409 if (len > 0) { 410 cwin = tp->snd_cwnd - 411 (tp->snd_nxt - tp->sack_newdata) - 412 sack_bytes_rxmt; 413 if (cwin < 0) 414 cwin = 0; 415 len = imin(len, cwin); 416 } 417 } 418 } 419 420 /* 421 * Lop off SYN bit if it has already been sent. However, if this 422 * is SYN-SENT state and if segment contains data and if we don't 423 * know that foreign host supports TAO, suppress sending segment. 424 */ 425 if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) { 426 if (tp->t_state != TCPS_SYN_RECEIVED) 427 flags &= ~TH_SYN; 428 #ifdef TCP_RFC7413 429 /* 430 * When sending additional segments following a TFO SYN|ACK, 431 * do not include the SYN bit. 432 */ 433 if (IS_FASTOPEN(tp->t_flags) && 434 (tp->t_state == TCPS_SYN_RECEIVED)) 435 flags &= ~TH_SYN; 436 #endif 437 off--, len++; 438 } 439 440 /* 441 * Be careful not to send data and/or FIN on SYN segments. 442 * This measure is needed to prevent interoperability problems 443 * with not fully conformant TCP implementations. 444 */ 445 if ((flags & TH_SYN) && (tp->t_flags & TF_NOOPT)) { 446 len = 0; 447 flags &= ~TH_FIN; 448 } 449 450 #ifdef TCP_RFC7413 451 /* 452 * When retransmitting SYN|ACK on a passively-created TFO socket, 453 * don't include data, as the presence of data may have caused the 454 * original SYN|ACK to have been dropped by a middlebox. 455 */ 456 if (IS_FASTOPEN(tp->t_flags) && 457 (((tp->t_state == TCPS_SYN_RECEIVED) && (tp->t_rxtshift > 0)) || 458 (flags & TH_RST))) 459 len = 0; 460 #endif 461 if (len <= 0) { 462 /* 463 * If FIN has been sent but not acked, 464 * but we haven't been called to retransmit, 465 * len will be < 0. Otherwise, window shrank 466 * after we sent into it. If window shrank to 0, 467 * cancel pending retransmit, pull snd_nxt back 468 * to (closed) window, and set the persist timer 469 * if it isn't already going. If the window didn't 470 * close completely, just wait for an ACK. 471 * 472 * We also do a general check here to ensure that 473 * we will set the persist timer when we have data 474 * to send, but a 0-byte window. This makes sure 475 * the persist timer is set even if the packet 476 * hits one of the "goto send" lines below. 477 */ 478 len = 0; 479 if ((sendwin == 0) && (TCPS_HAVEESTABLISHED(tp->t_state)) && 480 (off < (int) sbavail(&so->so_snd))) { 481 tcp_timer_activate(tp, TT_REXMT, 0); 482 tp->t_rxtshift = 0; 483 tp->snd_nxt = tp->snd_una; 484 if (!tcp_timer_active(tp, TT_PERSIST)) 485 tcp_setpersist(tp); 486 } 487 } 488 489 /* len will be >= 0 after this point. */ 490 KASSERT(len >= 0, ("[%s:%d]: len < 0", __func__, __LINE__)); 491 492 /* 493 * Automatic sizing of send socket buffer. Often the send buffer 494 * size is not optimally adjusted to the actual network conditions 495 * at hand (delay bandwidth product). Setting the buffer size too 496 * small limits throughput on links with high bandwidth and high 497 * delay (eg. trans-continental/oceanic links). Setting the 498 * buffer size too big consumes too much real kernel memory, 499 * especially with many connections on busy servers. 500 * 501 * The criteria to step up the send buffer one notch are: 502 * 1. receive window of remote host is larger than send buffer 503 * (with a fudge factor of 5/4th); 504 * 2. send buffer is filled to 7/8th with data (so we actually 505 * have data to make use of it); 506 * 3. send buffer fill has not hit maximal automatic size; 507 * 4. our send window (slow start and cogestion controlled) is 508 * larger than sent but unacknowledged data in send buffer. 509 * 510 * The remote host receive window scaling factor may limit the 511 * growing of the send buffer before it reaches its allowed 512 * maximum. 513 * 514 * It scales directly with slow start or congestion window 515 * and does at most one step per received ACK. This fast 516 * scaling has the drawback of growing the send buffer beyond 517 * what is strictly necessary to make full use of a given 518 * delay*bandwidth product. However testing has shown this not 519 * to be much of an problem. At worst we are trading wasting 520 * of available bandwidth (the non-use of it) for wasting some 521 * socket buffer memory. 522 * 523 * TODO: Shrink send buffer during idle periods together 524 * with congestion window. Requires another timer. Has to 525 * wait for upcoming tcp timer rewrite. 526 * 527 * XXXGL: should there be used sbused() or sbavail()? 528 */ 529 if (V_tcp_do_autosndbuf && so->so_snd.sb_flags & SB_AUTOSIZE) { 530 int autosndbuf_mod = 0; 531 if (V_tcp_sendbuf_auto_lowat) 532 autosndbuf_mod = so->so_snd.sb_lowat; 533 534 if ((tp->snd_wnd / 4 * 5) >= so->so_snd.sb_hiwat - autosndbuf_mod && 535 sbused(&so->so_snd) >= (so->so_snd.sb_hiwat / 8 * 7) - autosndbuf_mod && 536 sbused(&so->so_snd) < V_tcp_autosndbuf_max && 537 sendwin >= (sbused(&so->so_snd) - 538 (tp->snd_nxt - tp->snd_una))) { 539 if (!sbreserve_locked(&so->so_snd, 540 min(so->so_snd.sb_hiwat + V_tcp_autosndbuf_inc, 541 V_tcp_autosndbuf_max), so, curthread)) 542 so->so_snd.sb_flags &= ~SB_AUTOSIZE; 543 } 544 } 545 546 /* 547 * Decide if we can use TCP Segmentation Offloading (if supported by 548 * hardware). 549 * 550 * TSO may only be used if we are in a pure bulk sending state. The 551 * presence of TCP-MD5, SACK retransmits, SACK advertizements and 552 * IP options prevent using TSO. With TSO the TCP header is the same 553 * (except for the sequence number) for all generated packets. This 554 * makes it impossible to transmit any options which vary per generated 555 * segment or packet. 556 * 557 * IPv4 handling has a clear separation of ip options and ip header 558 * flags while IPv6 combines both in in6p_outputopts. ip6_optlen() does 559 * the right thing below to provide length of just ip options and thus 560 * checking for ipoptlen is enough to decide if ip options are present. 561 */ 562 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 563 /* 564 * Pre-calculate here as we save another lookup into the darknesses 565 * of IPsec that way and can actually decide if TSO is ok. 566 */ 567 #ifdef INET6 568 if (isipv6 && IPSEC_ENABLED(ipv6)) 569 ipsec_optlen = IPSEC_HDRSIZE(ipv6, tp->t_inpcb); 570 #ifdef INET 571 else 572 #endif 573 #endif /* INET6 */ 574 #ifdef INET 575 if (IPSEC_ENABLED(ipv4)) 576 ipsec_optlen = IPSEC_HDRSIZE(ipv4, tp->t_inpcb); 577 #endif /* INET */ 578 #endif /* IPSEC */ 579 #ifdef INET6 580 if (isipv6) 581 ipoptlen = ip6_optlen(tp->t_inpcb); 582 else 583 #endif 584 if (tp->t_inpcb->inp_options) 585 ipoptlen = tp->t_inpcb->inp_options->m_len - 586 offsetof(struct ipoption, ipopt_list); 587 else 588 ipoptlen = 0; 589 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 590 ipoptlen += ipsec_optlen; 591 #endif 592 593 if ((tp->t_flags & TF_TSO) && V_tcp_do_tso && len > tp->t_maxseg && 594 ((tp->t_flags & TF_SIGNATURE) == 0) && 595 tp->rcv_numsacks == 0 && sack_rxmit == 0 && 596 ipoptlen == 0) 597 tso = 1; 598 599 if (sack_rxmit) { 600 if (SEQ_LT(p->rxmit + len, tp->snd_una + sbused(&so->so_snd))) 601 flags &= ~TH_FIN; 602 } else { 603 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + 604 sbused(&so->so_snd))) 605 flags &= ~TH_FIN; 606 } 607 608 recwin = lmin(lmax(sbspace(&so->so_rcv), 0), 609 (long)TCP_MAXWIN << tp->rcv_scale); 610 611 /* 612 * Sender silly window avoidance. We transmit under the following 613 * conditions when len is non-zero: 614 * 615 * - We have a full segment (or more with TSO) 616 * - This is the last buffer in a write()/send() and we are 617 * either idle or running NODELAY 618 * - we've timed out (e.g. persist timer) 619 * - we have more then 1/2 the maximum send window's worth of 620 * data (receiver may be limited the window size) 621 * - we need to retransmit 622 */ 623 if (len) { 624 if (len >= tp->t_maxseg) 625 goto send; 626 /* 627 * NOTE! on localhost connections an 'ack' from the remote 628 * end may occur synchronously with the output and cause 629 * us to flush a buffer queued with moretocome. XXX 630 * 631 * note: the len + off check is almost certainly unnecessary. 632 */ 633 if (!(tp->t_flags & TF_MORETOCOME) && /* normal case */ 634 (idle || (tp->t_flags & TF_NODELAY)) && 635 (uint32_t)len + (uint32_t)off >= sbavail(&so->so_snd) && 636 (tp->t_flags & TF_NOPUSH) == 0) { 637 goto send; 638 } 639 if (tp->t_flags & TF_FORCEDATA) /* typ. timeout case */ 640 goto send; 641 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0) 642 goto send; 643 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */ 644 goto send; 645 if (sack_rxmit) 646 goto send; 647 } 648 649 /* 650 * Sending of standalone window updates. 651 * 652 * Window updates are important when we close our window due to a 653 * full socket buffer and are opening it again after the application 654 * reads data from it. Once the window has opened again and the 655 * remote end starts to send again the ACK clock takes over and 656 * provides the most current window information. 657 * 658 * We must avoid the silly window syndrome whereas every read 659 * from the receive buffer, no matter how small, causes a window 660 * update to be sent. We also should avoid sending a flurry of 661 * window updates when the socket buffer had queued a lot of data 662 * and the application is doing small reads. 663 * 664 * Prevent a flurry of pointless window updates by only sending 665 * an update when we can increase the advertized window by more 666 * than 1/4th of the socket buffer capacity. When the buffer is 667 * getting full or is very small be more aggressive and send an 668 * update whenever we can increase by two mss sized segments. 669 * In all other situations the ACK's to new incoming data will 670 * carry further window increases. 671 * 672 * Don't send an independent window update if a delayed 673 * ACK is pending (it will get piggy-backed on it) or the 674 * remote side already has done a half-close and won't send 675 * more data. Skip this if the connection is in T/TCP 676 * half-open state. 677 */ 678 if (recwin > 0 && !(tp->t_flags & TF_NEEDSYN) && 679 !(tp->t_flags & TF_DELACK) && 680 !TCPS_HAVERCVDFIN(tp->t_state)) { 681 /* 682 * "adv" is the amount we could increase the window, 683 * taking into account that we are limited by 684 * TCP_MAXWIN << tp->rcv_scale. 685 */ 686 int32_t adv; 687 int oldwin; 688 689 adv = recwin; 690 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) { 691 oldwin = (tp->rcv_adv - tp->rcv_nxt); 692 adv -= oldwin; 693 } else 694 oldwin = 0; 695 696 /* 697 * If the new window size ends up being the same as or less 698 * than the old size when it is scaled, then don't force 699 * a window update. 700 */ 701 if (oldwin >> tp->rcv_scale >= (adv + oldwin) >> tp->rcv_scale) 702 goto dontupdate; 703 704 if (adv >= (int32_t)(2 * tp->t_maxseg) && 705 (adv >= (int32_t)(so->so_rcv.sb_hiwat / 4) || 706 recwin <= (so->so_rcv.sb_hiwat / 8) || 707 so->so_rcv.sb_hiwat <= 8 * tp->t_maxseg)) 708 goto send; 709 if (2 * adv >= (int32_t)so->so_rcv.sb_hiwat) 710 goto send; 711 } 712 dontupdate: 713 714 /* 715 * Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW 716 * is also a catch-all for the retransmit timer timeout case. 717 */ 718 if (tp->t_flags & TF_ACKNOW) 719 goto send; 720 if ((flags & TH_RST) || 721 ((flags & TH_SYN) && (tp->t_flags & TF_NEEDSYN) == 0)) 722 goto send; 723 if (SEQ_GT(tp->snd_up, tp->snd_una)) 724 goto send; 725 /* 726 * If our state indicates that FIN should be sent 727 * and we have not yet done so, then we need to send. 728 */ 729 if (flags & TH_FIN && 730 ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una)) 731 goto send; 732 /* 733 * In SACK, it is possible for tcp_output to fail to send a segment 734 * after the retransmission timer has been turned off. Make sure 735 * that the retransmission timer is set. 736 */ 737 if ((tp->t_flags & TF_SACK_PERMIT) && 738 SEQ_GT(tp->snd_max, tp->snd_una) && 739 !tcp_timer_active(tp, TT_REXMT) && 740 !tcp_timer_active(tp, TT_PERSIST)) { 741 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 742 goto just_return; 743 } 744 /* 745 * TCP window updates are not reliable, rather a polling protocol 746 * using ``persist'' packets is used to insure receipt of window 747 * updates. The three ``states'' for the output side are: 748 * idle not doing retransmits or persists 749 * persisting to move a small or zero window 750 * (re)transmitting and thereby not persisting 751 * 752 * tcp_timer_active(tp, TT_PERSIST) 753 * is true when we are in persist state. 754 * (tp->t_flags & TF_FORCEDATA) 755 * is set when we are called to send a persist packet. 756 * tcp_timer_active(tp, TT_REXMT) 757 * is set when we are retransmitting 758 * The output side is idle when both timers are zero. 759 * 760 * If send window is too small, there is data to transmit, and no 761 * retransmit or persist is pending, then go to persist state. 762 * If nothing happens soon, send when timer expires: 763 * if window is nonzero, transmit what we can, 764 * otherwise force out a byte. 765 */ 766 if (sbavail(&so->so_snd) && !tcp_timer_active(tp, TT_REXMT) && 767 !tcp_timer_active(tp, TT_PERSIST)) { 768 tp->t_rxtshift = 0; 769 tcp_setpersist(tp); 770 } 771 772 /* 773 * No reason to send a segment, just return. 774 */ 775 just_return: 776 SOCKBUF_UNLOCK(&so->so_snd); 777 return (0); 778 779 send: 780 SOCKBUF_LOCK_ASSERT(&so->so_snd); 781 if (len > 0) { 782 if (len >= tp->t_maxseg) 783 tp->t_flags2 |= TF2_PLPMTU_MAXSEGSNT; 784 else 785 tp->t_flags2 &= ~TF2_PLPMTU_MAXSEGSNT; 786 } 787 /* 788 * Before ESTABLISHED, force sending of initial options 789 * unless TCP set not to do any options. 790 * NOTE: we assume that the IP/TCP header plus TCP options 791 * always fit in a single mbuf, leaving room for a maximum 792 * link header, i.e. 793 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES 794 */ 795 optlen = 0; 796 #ifdef INET6 797 if (isipv6) 798 hdrlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 799 else 800 #endif 801 hdrlen = sizeof (struct tcpiphdr); 802 803 /* 804 * Compute options for segment. 805 * We only have to care about SYN and established connection 806 * segments. Options for SYN-ACK segments are handled in TCP 807 * syncache. 808 */ 809 to.to_flags = 0; 810 if ((tp->t_flags & TF_NOOPT) == 0) { 811 /* Maximum segment size. */ 812 if (flags & TH_SYN) { 813 tp->snd_nxt = tp->iss; 814 to.to_mss = tcp_mssopt(&tp->t_inpcb->inp_inc); 815 to.to_flags |= TOF_MSS; 816 #ifdef TCP_RFC7413 817 /* 818 * Only include the TFO option on the first 819 * transmission of the SYN|ACK on a 820 * passively-created TFO socket, as the presence of 821 * the TFO option may have caused the original 822 * SYN|ACK to have been dropped by a middlebox. 823 */ 824 if (IS_FASTOPEN(tp->t_flags) && 825 (tp->t_state == TCPS_SYN_RECEIVED) && 826 (tp->t_rxtshift == 0)) { 827 to.to_tfo_len = TCP_FASTOPEN_COOKIE_LEN; 828 to.to_tfo_cookie = (u_char *)&tp->t_tfo_cookie; 829 to.to_flags |= TOF_FASTOPEN; 830 } 831 #endif 832 } 833 /* Window scaling. */ 834 if ((flags & TH_SYN) && (tp->t_flags & TF_REQ_SCALE)) { 835 to.to_wscale = tp->request_r_scale; 836 to.to_flags |= TOF_SCALE; 837 } 838 /* Timestamps. */ 839 if ((tp->t_flags & TF_RCVD_TSTMP) || 840 ((flags & TH_SYN) && (tp->t_flags & TF_REQ_TSTMP))) { 841 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 842 to.to_tsecr = tp->ts_recent; 843 to.to_flags |= TOF_TS; 844 } 845 846 /* Set receive buffer autosizing timestamp. */ 847 if (tp->rfbuf_ts == 0 && 848 (so->so_rcv.sb_flags & SB_AUTOSIZE)) 849 tp->rfbuf_ts = tcp_ts_getticks(); 850 851 /* Selective ACK's. */ 852 if (tp->t_flags & TF_SACK_PERMIT) { 853 if (flags & TH_SYN) 854 to.to_flags |= TOF_SACKPERM; 855 else if (TCPS_HAVEESTABLISHED(tp->t_state) && 856 (tp->t_flags & TF_SACK_PERMIT) && 857 tp->rcv_numsacks > 0) { 858 to.to_flags |= TOF_SACK; 859 to.to_nsacks = tp->rcv_numsacks; 860 to.to_sacks = (u_char *)tp->sackblks; 861 } 862 } 863 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 864 /* TCP-MD5 (RFC2385). */ 865 /* 866 * Check that TCP_MD5SIG is enabled in tcpcb to 867 * account the size needed to set this TCP option. 868 */ 869 if (tp->t_flags & TF_SIGNATURE) 870 to.to_flags |= TOF_SIGNATURE; 871 #endif /* TCP_SIGNATURE */ 872 873 /* Processing the options. */ 874 hdrlen += optlen = tcp_addoptions(&to, opt); 875 } 876 877 /* 878 * Adjust data length if insertion of options will 879 * bump the packet length beyond the t_maxseg length. 880 * Clear the FIN bit because we cut off the tail of 881 * the segment. 882 */ 883 if (len + optlen + ipoptlen > tp->t_maxseg) { 884 flags &= ~TH_FIN; 885 886 if (tso) { 887 u_int if_hw_tsomax; 888 u_int if_hw_tsomaxsegcount; 889 u_int if_hw_tsomaxsegsize; 890 struct mbuf *mb; 891 u_int moff; 892 int max_len; 893 894 /* extract TSO information */ 895 if_hw_tsomax = tp->t_tsomax; 896 if_hw_tsomaxsegcount = tp->t_tsomaxsegcount; 897 if_hw_tsomaxsegsize = tp->t_tsomaxsegsize; 898 899 /* 900 * Limit a TSO burst to prevent it from 901 * overflowing or exceeding the maximum length 902 * allowed by the network interface: 903 */ 904 KASSERT(ipoptlen == 0, 905 ("%s: TSO can't do IP options", __func__)); 906 907 /* 908 * Check if we should limit by maximum payload 909 * length: 910 */ 911 if (if_hw_tsomax != 0) { 912 /* compute maximum TSO length */ 913 max_len = (if_hw_tsomax - hdrlen - 914 max_linkhdr); 915 if (max_len <= 0) { 916 len = 0; 917 } else if (len > max_len) { 918 sendalot = 1; 919 len = max_len; 920 } 921 } 922 923 /* 924 * Check if we should limit by maximum segment 925 * size and count: 926 */ 927 if (if_hw_tsomaxsegcount != 0 && 928 if_hw_tsomaxsegsize != 0) { 929 /* 930 * Subtract one segment for the LINK 931 * and TCP/IP headers mbuf that will 932 * be prepended to this mbuf chain 933 * after the code in this section 934 * limits the number of mbufs in the 935 * chain to if_hw_tsomaxsegcount. 936 */ 937 if_hw_tsomaxsegcount -= 1; 938 max_len = 0; 939 mb = sbsndmbuf(&so->so_snd, off, &moff); 940 941 while (mb != NULL && max_len < len) { 942 u_int mlen; 943 u_int frags; 944 945 /* 946 * Get length of mbuf fragment 947 * and how many hardware frags, 948 * rounded up, it would use: 949 */ 950 mlen = (mb->m_len - moff); 951 frags = howmany(mlen, 952 if_hw_tsomaxsegsize); 953 954 /* Handle special case: Zero Length Mbuf */ 955 if (frags == 0) 956 frags = 1; 957 958 /* 959 * Check if the fragment limit 960 * will be reached or exceeded: 961 */ 962 if (frags >= if_hw_tsomaxsegcount) { 963 max_len += min(mlen, 964 if_hw_tsomaxsegcount * 965 if_hw_tsomaxsegsize); 966 break; 967 } 968 max_len += mlen; 969 if_hw_tsomaxsegcount -= frags; 970 moff = 0; 971 mb = mb->m_next; 972 } 973 if (max_len <= 0) { 974 len = 0; 975 } else if (len > max_len) { 976 sendalot = 1; 977 len = max_len; 978 } 979 } 980 981 /* 982 * Prevent the last segment from being 983 * fractional unless the send sockbuf can be 984 * emptied: 985 */ 986 max_len = (tp->t_maxseg - optlen); 987 if (((uint32_t)off + (uint32_t)len) < 988 sbavail(&so->so_snd)) { 989 moff = len % max_len; 990 if (moff != 0) { 991 len -= moff; 992 sendalot = 1; 993 } 994 } 995 996 /* 997 * In case there are too many small fragments 998 * don't use TSO: 999 */ 1000 if (len <= max_len) { 1001 len = max_len; 1002 sendalot = 1; 1003 tso = 0; 1004 } 1005 1006 /* 1007 * Send the FIN in a separate segment 1008 * after the bulk sending is done. 1009 * We don't trust the TSO implementations 1010 * to clear the FIN flag on all but the 1011 * last segment. 1012 */ 1013 if (tp->t_flags & TF_NEEDFIN) 1014 sendalot = 1; 1015 1016 } else { 1017 len = tp->t_maxseg - optlen - ipoptlen; 1018 sendalot = 1; 1019 } 1020 } else 1021 tso = 0; 1022 1023 KASSERT(len + hdrlen + ipoptlen <= IP_MAXPACKET, 1024 ("%s: len > IP_MAXPACKET", __func__)); 1025 1026 /*#ifdef DIAGNOSTIC*/ 1027 #ifdef INET6 1028 if (max_linkhdr + hdrlen > MCLBYTES) 1029 #else 1030 if (max_linkhdr + hdrlen > MHLEN) 1031 #endif 1032 panic("tcphdr too big"); 1033 /*#endif*/ 1034 1035 /* 1036 * This KASSERT is here to catch edge cases at a well defined place. 1037 * Before, those had triggered (random) panic conditions further down. 1038 */ 1039 KASSERT(len >= 0, ("[%s:%d]: len < 0", __func__, __LINE__)); 1040 1041 /* 1042 * Grab a header mbuf, attaching a copy of data to 1043 * be transmitted, and initialize the header from 1044 * the template for sends on this connection. 1045 */ 1046 if (len) { 1047 struct mbuf *mb; 1048 u_int moff; 1049 1050 if ((tp->t_flags & TF_FORCEDATA) && len == 1) 1051 TCPSTAT_INC(tcps_sndprobe); 1052 else if (SEQ_LT(tp->snd_nxt, tp->snd_max) || sack_rxmit) { 1053 tp->t_sndrexmitpack++; 1054 TCPSTAT_INC(tcps_sndrexmitpack); 1055 TCPSTAT_ADD(tcps_sndrexmitbyte, len); 1056 } else { 1057 TCPSTAT_INC(tcps_sndpack); 1058 TCPSTAT_ADD(tcps_sndbyte, len); 1059 } 1060 #ifdef INET6 1061 if (MHLEN < hdrlen + max_linkhdr) 1062 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 1063 else 1064 #endif 1065 m = m_gethdr(M_NOWAIT, MT_DATA); 1066 1067 if (m == NULL) { 1068 SOCKBUF_UNLOCK(&so->so_snd); 1069 error = ENOBUFS; 1070 sack_rxmit = 0; 1071 goto out; 1072 } 1073 1074 m->m_data += max_linkhdr; 1075 m->m_len = hdrlen; 1076 1077 /* 1078 * Start the m_copy functions from the closest mbuf 1079 * to the offset in the socket buffer chain. 1080 */ 1081 mb = sbsndptr(&so->so_snd, off, len, &moff); 1082 1083 if (len <= MHLEN - hdrlen - max_linkhdr) { 1084 m_copydata(mb, moff, len, 1085 mtod(m, caddr_t) + hdrlen); 1086 m->m_len += len; 1087 } else { 1088 m->m_next = m_copym(mb, moff, len, M_NOWAIT); 1089 if (m->m_next == NULL) { 1090 SOCKBUF_UNLOCK(&so->so_snd); 1091 (void) m_free(m); 1092 error = ENOBUFS; 1093 sack_rxmit = 0; 1094 goto out; 1095 } 1096 } 1097 1098 /* 1099 * If we're sending everything we've got, set PUSH. 1100 * (This will keep happy those implementations which only 1101 * give data to the user when a buffer fills or 1102 * a PUSH comes in.) 1103 */ 1104 if (((uint32_t)off + (uint32_t)len == sbused(&so->so_snd)) && 1105 !(flags & TH_SYN)) 1106 flags |= TH_PUSH; 1107 SOCKBUF_UNLOCK(&so->so_snd); 1108 } else { 1109 SOCKBUF_UNLOCK(&so->so_snd); 1110 if (tp->t_flags & TF_ACKNOW) 1111 TCPSTAT_INC(tcps_sndacks); 1112 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 1113 TCPSTAT_INC(tcps_sndctrl); 1114 else if (SEQ_GT(tp->snd_up, tp->snd_una)) 1115 TCPSTAT_INC(tcps_sndurg); 1116 else 1117 TCPSTAT_INC(tcps_sndwinup); 1118 1119 m = m_gethdr(M_NOWAIT, MT_DATA); 1120 if (m == NULL) { 1121 error = ENOBUFS; 1122 sack_rxmit = 0; 1123 goto out; 1124 } 1125 #ifdef INET6 1126 if (isipv6 && (MHLEN < hdrlen + max_linkhdr) && 1127 MHLEN >= hdrlen) { 1128 M_ALIGN(m, hdrlen); 1129 } else 1130 #endif 1131 m->m_data += max_linkhdr; 1132 m->m_len = hdrlen; 1133 } 1134 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); 1135 m->m_pkthdr.rcvif = (struct ifnet *)0; 1136 #ifdef MAC 1137 mac_inpcb_create_mbuf(tp->t_inpcb, m); 1138 #endif 1139 #ifdef INET6 1140 if (isipv6) { 1141 ip6 = mtod(m, struct ip6_hdr *); 1142 th = (struct tcphdr *)(ip6 + 1); 1143 tcpip_fillheaders(tp->t_inpcb, ip6, th); 1144 } else 1145 #endif /* INET6 */ 1146 { 1147 ip = mtod(m, struct ip *); 1148 ipov = (struct ipovly *)ip; 1149 th = (struct tcphdr *)(ip + 1); 1150 tcpip_fillheaders(tp->t_inpcb, ip, th); 1151 } 1152 1153 /* 1154 * Fill in fields, remembering maximum advertised 1155 * window for use in delaying messages about window sizes. 1156 * If resending a FIN, be sure not to use a new sequence number. 1157 */ 1158 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN && 1159 tp->snd_nxt == tp->snd_max) 1160 tp->snd_nxt--; 1161 /* 1162 * If we are starting a connection, send ECN setup 1163 * SYN packet. If we are on a retransmit, we may 1164 * resend those bits a number of times as per 1165 * RFC 3168. 1166 */ 1167 if (tp->t_state == TCPS_SYN_SENT && V_tcp_do_ecn == 1) { 1168 if (tp->t_rxtshift >= 1) { 1169 if (tp->t_rxtshift <= V_tcp_ecn_maxretries) 1170 flags |= TH_ECE|TH_CWR; 1171 } else 1172 flags |= TH_ECE|TH_CWR; 1173 } 1174 1175 if (tp->t_state == TCPS_ESTABLISHED && 1176 (tp->t_flags & TF_ECN_PERMIT)) { 1177 /* 1178 * If the peer has ECN, mark data packets with 1179 * ECN capable transmission (ECT). 1180 * Ignore pure ack packets, retransmissions and window probes. 1181 */ 1182 if (len > 0 && SEQ_GEQ(tp->snd_nxt, tp->snd_max) && 1183 !((tp->t_flags & TF_FORCEDATA) && len == 1)) { 1184 #ifdef INET6 1185 if (isipv6) 1186 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 1187 else 1188 #endif 1189 ip->ip_tos |= IPTOS_ECN_ECT0; 1190 TCPSTAT_INC(tcps_ecn_ect0); 1191 } 1192 1193 /* 1194 * Reply with proper ECN notifications. 1195 */ 1196 if (tp->t_flags & TF_ECN_SND_CWR) { 1197 flags |= TH_CWR; 1198 tp->t_flags &= ~TF_ECN_SND_CWR; 1199 } 1200 if (tp->t_flags & TF_ECN_SND_ECE) 1201 flags |= TH_ECE; 1202 } 1203 1204 /* 1205 * If we are doing retransmissions, then snd_nxt will 1206 * not reflect the first unsent octet. For ACK only 1207 * packets, we do not want the sequence number of the 1208 * retransmitted packet, we want the sequence number 1209 * of the next unsent octet. So, if there is no data 1210 * (and no SYN or FIN), use snd_max instead of snd_nxt 1211 * when filling in ti_seq. But if we are in persist 1212 * state, snd_max might reflect one byte beyond the 1213 * right edge of the window, so use snd_nxt in that 1214 * case, since we know we aren't doing a retransmission. 1215 * (retransmit and persist are mutually exclusive...) 1216 */ 1217 if (sack_rxmit == 0) { 1218 if (len || (flags & (TH_SYN|TH_FIN)) || 1219 tcp_timer_active(tp, TT_PERSIST)) 1220 th->th_seq = htonl(tp->snd_nxt); 1221 else 1222 th->th_seq = htonl(tp->snd_max); 1223 } else { 1224 th->th_seq = htonl(p->rxmit); 1225 p->rxmit += len; 1226 tp->sackhint.sack_bytes_rexmit += len; 1227 } 1228 th->th_ack = htonl(tp->rcv_nxt); 1229 if (optlen) { 1230 bcopy(opt, th + 1, optlen); 1231 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1232 } 1233 th->th_flags = flags; 1234 /* 1235 * Calculate receive window. Don't shrink window, 1236 * but avoid silly window syndrome. 1237 */ 1238 if (recwin < (so->so_rcv.sb_hiwat / 4) && 1239 recwin < tp->t_maxseg) 1240 recwin = 0; 1241 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt) && 1242 recwin < (tp->rcv_adv - tp->rcv_nxt)) 1243 recwin = (tp->rcv_adv - tp->rcv_nxt); 1244 1245 /* 1246 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1247 * or <SYN,ACK>) segment itself is never scaled. The <SYN,ACK> 1248 * case is handled in syncache. 1249 */ 1250 if (flags & TH_SYN) 1251 th->th_win = htons((u_short) 1252 (min(sbspace(&so->so_rcv), TCP_MAXWIN))); 1253 else 1254 th->th_win = htons((u_short)(recwin >> tp->rcv_scale)); 1255 1256 /* 1257 * Adjust the RXWIN0SENT flag - indicate that we have advertised 1258 * a 0 window. This may cause the remote transmitter to stall. This 1259 * flag tells soreceive() to disable delayed acknowledgements when 1260 * draining the buffer. This can occur if the receiver is attempting 1261 * to read more data than can be buffered prior to transmitting on 1262 * the connection. 1263 */ 1264 if (th->th_win == 0) { 1265 tp->t_sndzerowin++; 1266 tp->t_flags |= TF_RXWIN0SENT; 1267 } else 1268 tp->t_flags &= ~TF_RXWIN0SENT; 1269 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { 1270 th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt)); 1271 th->th_flags |= TH_URG; 1272 } else 1273 /* 1274 * If no urgent pointer to send, then we pull 1275 * the urgent pointer to the left edge of the send window 1276 * so that it doesn't drift into the send window on sequence 1277 * number wraparound. 1278 */ 1279 tp->snd_up = tp->snd_una; /* drag it along */ 1280 1281 /* 1282 * Put TCP length in extended header, and then 1283 * checksum extended header and data. 1284 */ 1285 m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */ 1286 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1287 1288 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1289 if (to.to_flags & TOF_SIGNATURE) { 1290 /* 1291 * Calculate MD5 signature and put it into the place 1292 * determined before. 1293 * NOTE: since TCP options buffer doesn't point into 1294 * mbuf's data, calculate offset and use it. 1295 */ 1296 if (!TCPMD5_ENABLED() || TCPMD5_OUTPUT(m, th, 1297 (u_char *)(th + 1) + (to.to_signature - opt)) != 0) { 1298 /* 1299 * Do not send segment if the calculation of MD5 1300 * digest has failed. 1301 */ 1302 goto out; 1303 } 1304 } 1305 #endif 1306 #ifdef INET6 1307 if (isipv6) { 1308 /* 1309 * There is no need to fill in ip6_plen right now. 1310 * It will be filled later by ip6_output. 1311 */ 1312 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 1313 th->th_sum = in6_cksum_pseudo(ip6, sizeof(struct tcphdr) + 1314 optlen + len, IPPROTO_TCP, 0); 1315 } 1316 #endif 1317 #if defined(INET6) && defined(INET) 1318 else 1319 #endif 1320 #ifdef INET 1321 { 1322 m->m_pkthdr.csum_flags = CSUM_TCP; 1323 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1324 htons(sizeof(struct tcphdr) + IPPROTO_TCP + len + optlen)); 1325 1326 /* IP version must be set here for ipv4/ipv6 checking later */ 1327 KASSERT(ip->ip_v == IPVERSION, 1328 ("%s: IP version incorrect: %d", __func__, ip->ip_v)); 1329 } 1330 #endif 1331 1332 /* 1333 * Enable TSO and specify the size of the segments. 1334 * The TCP pseudo header checksum is always provided. 1335 */ 1336 if (tso) { 1337 KASSERT(len > tp->t_maxseg - optlen, 1338 ("%s: len <= tso_segsz", __func__)); 1339 m->m_pkthdr.csum_flags |= CSUM_TSO; 1340 m->m_pkthdr.tso_segsz = tp->t_maxseg - optlen; 1341 } 1342 1343 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1344 KASSERT(len + hdrlen + ipoptlen - ipsec_optlen == m_length(m, NULL), 1345 ("%s: mbuf chain shorter than expected: %d + %u + %u - %u != %u", 1346 __func__, len, hdrlen, ipoptlen, ipsec_optlen, m_length(m, NULL))); 1347 #else 1348 KASSERT(len + hdrlen + ipoptlen == m_length(m, NULL), 1349 ("%s: mbuf chain shorter than expected: %d + %u + %u != %u", 1350 __func__, len, hdrlen, ipoptlen, m_length(m, NULL))); 1351 #endif 1352 1353 #ifdef TCP_HHOOK 1354 /* Run HHOOK_TCP_ESTABLISHED_OUT helper hooks. */ 1355 hhook_run_tcp_est_out(tp, th, &to, len, tso); 1356 #endif 1357 1358 #ifdef TCPDEBUG 1359 /* 1360 * Trace. 1361 */ 1362 if (so->so_options & SO_DEBUG) { 1363 u_short save = 0; 1364 #ifdef INET6 1365 if (!isipv6) 1366 #endif 1367 { 1368 save = ipov->ih_len; 1369 ipov->ih_len = htons(m->m_pkthdr.len /* - hdrlen + (th->th_off << 2) */); 1370 } 1371 tcp_trace(TA_OUTPUT, tp->t_state, tp, mtod(m, void *), th, 0); 1372 #ifdef INET6 1373 if (!isipv6) 1374 #endif 1375 ipov->ih_len = save; 1376 } 1377 #endif /* TCPDEBUG */ 1378 TCP_PROBE3(debug__output, tp, th, m); 1379 1380 /* 1381 * Fill in IP length and desired time to live and 1382 * send to IP level. There should be a better way 1383 * to handle ttl and tos; we could keep them in 1384 * the template, but need a way to checksum without them. 1385 */ 1386 /* 1387 * m->m_pkthdr.len should have been set before checksum calculation, 1388 * because in6_cksum() need it. 1389 */ 1390 #ifdef INET6 1391 if (isipv6) { 1392 /* 1393 * we separately set hoplimit for every segment, since the 1394 * user might want to change the value via setsockopt. 1395 * Also, desired default hop limit might be changed via 1396 * Neighbor Discovery. 1397 */ 1398 ip6->ip6_hlim = in6_selecthlim(tp->t_inpcb, NULL); 1399 1400 /* 1401 * Set the packet size here for the benefit of DTrace probes. 1402 * ip6_output() will set it properly; it's supposed to include 1403 * the option header lengths as well. 1404 */ 1405 ip6->ip6_plen = htons(m->m_pkthdr.len - sizeof(*ip6)); 1406 1407 if (V_path_mtu_discovery && tp->t_maxseg > V_tcp_minmss) 1408 tp->t_flags2 |= TF2_PLPMTU_PMTUD; 1409 else 1410 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD; 1411 1412 if (tp->t_state == TCPS_SYN_SENT) 1413 TCP_PROBE5(connect__request, NULL, tp, ip6, tp, th); 1414 1415 TCP_PROBE5(send, NULL, tp, ip6, tp, th); 1416 1417 #ifdef TCPPCAP 1418 /* Save packet, if requested. */ 1419 tcp_pcap_add(th, m, &(tp->t_outpkts)); 1420 #endif 1421 1422 /* TODO: IPv6 IP6TOS_ECT bit on */ 1423 error = ip6_output(m, tp->t_inpcb->in6p_outputopts, 1424 &tp->t_inpcb->inp_route6, 1425 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 1426 NULL, NULL, tp->t_inpcb); 1427 1428 if (error == EMSGSIZE && tp->t_inpcb->inp_route6.ro_rt != NULL) 1429 mtu = tp->t_inpcb->inp_route6.ro_rt->rt_mtu; 1430 } 1431 #endif /* INET6 */ 1432 #if defined(INET) && defined(INET6) 1433 else 1434 #endif 1435 #ifdef INET 1436 { 1437 ip->ip_len = htons(m->m_pkthdr.len); 1438 #ifdef INET6 1439 if (tp->t_inpcb->inp_vflag & INP_IPV6PROTO) 1440 ip->ip_ttl = in6_selecthlim(tp->t_inpcb, NULL); 1441 #endif /* INET6 */ 1442 /* 1443 * If we do path MTU discovery, then we set DF on every packet. 1444 * This might not be the best thing to do according to RFC3390 1445 * Section 2. However the tcp hostcache migitates the problem 1446 * so it affects only the first tcp connection with a host. 1447 * 1448 * NB: Don't set DF on small MTU/MSS to have a safe fallback. 1449 */ 1450 if (V_path_mtu_discovery && tp->t_maxseg > V_tcp_minmss) { 1451 ip->ip_off |= htons(IP_DF); 1452 tp->t_flags2 |= TF2_PLPMTU_PMTUD; 1453 } else { 1454 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD; 1455 } 1456 1457 if (tp->t_state == TCPS_SYN_SENT) 1458 TCP_PROBE5(connect__request, NULL, tp, ip, tp, th); 1459 1460 TCP_PROBE5(send, NULL, tp, ip, tp, th); 1461 1462 #ifdef TCPPCAP 1463 /* Save packet, if requested. */ 1464 tcp_pcap_add(th, m, &(tp->t_outpkts)); 1465 #endif 1466 1467 error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route, 1468 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 0, 1469 tp->t_inpcb); 1470 1471 if (error == EMSGSIZE && tp->t_inpcb->inp_route.ro_rt != NULL) 1472 mtu = tp->t_inpcb->inp_route.ro_rt->rt_mtu; 1473 } 1474 #endif /* INET */ 1475 1476 out: 1477 /* 1478 * In transmit state, time the transmission and arrange for 1479 * the retransmit. In persist state, just set snd_max. 1480 */ 1481 if ((tp->t_flags & TF_FORCEDATA) == 0 || 1482 !tcp_timer_active(tp, TT_PERSIST)) { 1483 tcp_seq startseq = tp->snd_nxt; 1484 1485 /* 1486 * Advance snd_nxt over sequence space of this segment. 1487 */ 1488 if (flags & (TH_SYN|TH_FIN)) { 1489 if (flags & TH_SYN) 1490 tp->snd_nxt++; 1491 if (flags & TH_FIN) { 1492 tp->snd_nxt++; 1493 tp->t_flags |= TF_SENTFIN; 1494 } 1495 } 1496 if (sack_rxmit) 1497 goto timer; 1498 tp->snd_nxt += len; 1499 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { 1500 tp->snd_max = tp->snd_nxt; 1501 /* 1502 * Time this transmission if not a retransmission and 1503 * not currently timing anything. 1504 */ 1505 if (tp->t_rtttime == 0) { 1506 tp->t_rtttime = ticks; 1507 tp->t_rtseq = startseq; 1508 TCPSTAT_INC(tcps_segstimed); 1509 } 1510 } 1511 1512 /* 1513 * Set retransmit timer if not currently set, 1514 * and not doing a pure ack or a keep-alive probe. 1515 * Initial value for retransmit timer is smoothed 1516 * round-trip time + 2 * round-trip time variance. 1517 * Initialize shift counter which is used for backoff 1518 * of retransmit time. 1519 */ 1520 timer: 1521 if (!tcp_timer_active(tp, TT_REXMT) && 1522 ((sack_rxmit && tp->snd_nxt != tp->snd_max) || 1523 (tp->snd_nxt != tp->snd_una))) { 1524 if (tcp_timer_active(tp, TT_PERSIST)) { 1525 tcp_timer_activate(tp, TT_PERSIST, 0); 1526 tp->t_rxtshift = 0; 1527 } 1528 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 1529 } else if (len == 0 && sbavail(&so->so_snd) && 1530 !tcp_timer_active(tp, TT_REXMT) && 1531 !tcp_timer_active(tp, TT_PERSIST)) { 1532 /* 1533 * Avoid a situation where we do not set persist timer 1534 * after a zero window condition. For example: 1535 * 1) A -> B: packet with enough data to fill the window 1536 * 2) B -> A: ACK for #1 + new data (0 window 1537 * advertisement) 1538 * 3) A -> B: ACK for #2, 0 len packet 1539 * 1540 * In this case, A will not activate the persist timer, 1541 * because it chose to send a packet. Unless tcp_output 1542 * is called for some other reason (delayed ack timer, 1543 * another input packet from B, socket syscall), A will 1544 * not send zero window probes. 1545 * 1546 * So, if you send a 0-length packet, but there is data 1547 * in the socket buffer, and neither the rexmt or 1548 * persist timer is already set, then activate the 1549 * persist timer. 1550 */ 1551 tp->t_rxtshift = 0; 1552 tcp_setpersist(tp); 1553 } 1554 } else { 1555 /* 1556 * Persist case, update snd_max but since we are in 1557 * persist mode (no window) we do not update snd_nxt. 1558 */ 1559 int xlen = len; 1560 if (flags & TH_SYN) 1561 ++xlen; 1562 if (flags & TH_FIN) { 1563 ++xlen; 1564 tp->t_flags |= TF_SENTFIN; 1565 } 1566 if (SEQ_GT(tp->snd_nxt + xlen, tp->snd_max)) 1567 tp->snd_max = tp->snd_nxt + xlen; 1568 } 1569 1570 if (error) { 1571 1572 /* 1573 * We know that the packet was lost, so back out the 1574 * sequence number advance, if any. 1575 * 1576 * If the error is EPERM the packet got blocked by the 1577 * local firewall. Normally we should terminate the 1578 * connection but the blocking may have been spurious 1579 * due to a firewall reconfiguration cycle. So we treat 1580 * it like a packet loss and let the retransmit timer and 1581 * timeouts do their work over time. 1582 * XXX: It is a POLA question whether calling tcp_drop right 1583 * away would be the really correct behavior instead. 1584 */ 1585 if (((tp->t_flags & TF_FORCEDATA) == 0 || 1586 !tcp_timer_active(tp, TT_PERSIST)) && 1587 ((flags & TH_SYN) == 0) && 1588 (error != EPERM)) { 1589 if (sack_rxmit) { 1590 p->rxmit -= len; 1591 tp->sackhint.sack_bytes_rexmit -= len; 1592 KASSERT(tp->sackhint.sack_bytes_rexmit >= 0, 1593 ("sackhint bytes rtx >= 0")); 1594 } else 1595 tp->snd_nxt -= len; 1596 } 1597 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); /* Check gotos. */ 1598 switch (error) { 1599 case EACCES: 1600 tp->t_softerror = error; 1601 return (0); 1602 case EPERM: 1603 tp->t_softerror = error; 1604 return (error); 1605 case ENOBUFS: 1606 TCP_XMIT_TIMER_ASSERT(tp, len, flags); 1607 tp->snd_cwnd = tp->t_maxseg; 1608 return (0); 1609 case EMSGSIZE: 1610 /* 1611 * For some reason the interface we used initially 1612 * to send segments changed to another or lowered 1613 * its MTU. 1614 * If TSO was active we either got an interface 1615 * without TSO capabilits or TSO was turned off. 1616 * If we obtained mtu from ip_output() then update 1617 * it and try again. 1618 */ 1619 if (tso) 1620 tp->t_flags &= ~TF_TSO; 1621 if (mtu != 0) { 1622 tcp_mss_update(tp, -1, mtu, NULL, NULL); 1623 goto again; 1624 } 1625 return (error); 1626 case EHOSTDOWN: 1627 case EHOSTUNREACH: 1628 case ENETDOWN: 1629 case ENETUNREACH: 1630 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1631 tp->t_softerror = error; 1632 return (0); 1633 } 1634 /* FALLTHROUGH */ 1635 default: 1636 return (error); 1637 } 1638 } 1639 TCPSTAT_INC(tcps_sndtotal); 1640 1641 /* 1642 * Data sent (as far as we can tell). 1643 * If this advertises a larger window than any other segment, 1644 * then remember the size of the advertised window. 1645 * Any pending ACK has now been sent. 1646 */ 1647 if (SEQ_GT(tp->rcv_nxt + recwin, tp->rcv_adv)) 1648 tp->rcv_adv = tp->rcv_nxt + recwin; 1649 tp->last_ack_sent = tp->rcv_nxt; 1650 tp->t_flags &= ~(TF_ACKNOW | TF_DELACK); 1651 if (tcp_timer_active(tp, TT_DELACK)) 1652 tcp_timer_activate(tp, TT_DELACK, 0); 1653 #if 0 1654 /* 1655 * This completely breaks TCP if newreno is turned on. What happens 1656 * is that if delayed-acks are turned on on the receiver, this code 1657 * on the transmitter effectively destroys the TCP window, forcing 1658 * it to four packets (1.5Kx4 = 6K window). 1659 */ 1660 if (sendalot && --maxburst) 1661 goto again; 1662 #endif 1663 if (sendalot) 1664 goto again; 1665 return (0); 1666 } 1667 1668 void 1669 tcp_setpersist(struct tcpcb *tp) 1670 { 1671 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1; 1672 int tt; 1673 1674 tp->t_flags &= ~TF_PREVVALID; 1675 if (tcp_timer_active(tp, TT_REXMT)) 1676 panic("tcp_setpersist: retransmit pending"); 1677 /* 1678 * Start/restart persistence timer. 1679 */ 1680 TCPT_RANGESET(tt, t * tcp_backoff[tp->t_rxtshift], 1681 tcp_persmin, tcp_persmax); 1682 tcp_timer_activate(tp, TT_PERSIST, tt); 1683 if (tp->t_rxtshift < TCP_MAXRXTSHIFT) 1684 tp->t_rxtshift++; 1685 } 1686 1687 /* 1688 * Insert TCP options according to the supplied parameters to the place 1689 * optp in a consistent way. Can handle unaligned destinations. 1690 * 1691 * The order of the option processing is crucial for optimal packing and 1692 * alignment for the scarce option space. 1693 * 1694 * The optimal order for a SYN/SYN-ACK segment is: 1695 * MSS (4) + NOP (1) + Window scale (3) + SACK permitted (2) + 1696 * Timestamp (10) + Signature (18) = 38 bytes out of a maximum of 40. 1697 * 1698 * The SACK options should be last. SACK blocks consume 8*n+2 bytes. 1699 * So a full size SACK blocks option is 34 bytes (with 4 SACK blocks). 1700 * At minimum we need 10 bytes (to generate 1 SACK block). If both 1701 * TCP Timestamps (12 bytes) and TCP Signatures (18 bytes) are present, 1702 * we only have 10 bytes for SACK options (40 - (12 + 18)). 1703 */ 1704 int 1705 tcp_addoptions(struct tcpopt *to, u_char *optp) 1706 { 1707 u_int32_t mask, optlen = 0; 1708 1709 for (mask = 1; mask < TOF_MAXOPT; mask <<= 1) { 1710 if ((to->to_flags & mask) != mask) 1711 continue; 1712 if (optlen == TCP_MAXOLEN) 1713 break; 1714 switch (to->to_flags & mask) { 1715 case TOF_MSS: 1716 while (optlen % 4) { 1717 optlen += TCPOLEN_NOP; 1718 *optp++ = TCPOPT_NOP; 1719 } 1720 if (TCP_MAXOLEN - optlen < TCPOLEN_MAXSEG) 1721 continue; 1722 optlen += TCPOLEN_MAXSEG; 1723 *optp++ = TCPOPT_MAXSEG; 1724 *optp++ = TCPOLEN_MAXSEG; 1725 to->to_mss = htons(to->to_mss); 1726 bcopy((u_char *)&to->to_mss, optp, sizeof(to->to_mss)); 1727 optp += sizeof(to->to_mss); 1728 break; 1729 case TOF_SCALE: 1730 while (!optlen || optlen % 2 != 1) { 1731 optlen += TCPOLEN_NOP; 1732 *optp++ = TCPOPT_NOP; 1733 } 1734 if (TCP_MAXOLEN - optlen < TCPOLEN_WINDOW) 1735 continue; 1736 optlen += TCPOLEN_WINDOW; 1737 *optp++ = TCPOPT_WINDOW; 1738 *optp++ = TCPOLEN_WINDOW; 1739 *optp++ = to->to_wscale; 1740 break; 1741 case TOF_SACKPERM: 1742 while (optlen % 2) { 1743 optlen += TCPOLEN_NOP; 1744 *optp++ = TCPOPT_NOP; 1745 } 1746 if (TCP_MAXOLEN - optlen < TCPOLEN_SACK_PERMITTED) 1747 continue; 1748 optlen += TCPOLEN_SACK_PERMITTED; 1749 *optp++ = TCPOPT_SACK_PERMITTED; 1750 *optp++ = TCPOLEN_SACK_PERMITTED; 1751 break; 1752 case TOF_TS: 1753 while (!optlen || optlen % 4 != 2) { 1754 optlen += TCPOLEN_NOP; 1755 *optp++ = TCPOPT_NOP; 1756 } 1757 if (TCP_MAXOLEN - optlen < TCPOLEN_TIMESTAMP) 1758 continue; 1759 optlen += TCPOLEN_TIMESTAMP; 1760 *optp++ = TCPOPT_TIMESTAMP; 1761 *optp++ = TCPOLEN_TIMESTAMP; 1762 to->to_tsval = htonl(to->to_tsval); 1763 to->to_tsecr = htonl(to->to_tsecr); 1764 bcopy((u_char *)&to->to_tsval, optp, sizeof(to->to_tsval)); 1765 optp += sizeof(to->to_tsval); 1766 bcopy((u_char *)&to->to_tsecr, optp, sizeof(to->to_tsecr)); 1767 optp += sizeof(to->to_tsecr); 1768 break; 1769 case TOF_SIGNATURE: 1770 { 1771 int siglen = TCPOLEN_SIGNATURE - 2; 1772 1773 while (!optlen || optlen % 4 != 2) { 1774 optlen += TCPOLEN_NOP; 1775 *optp++ = TCPOPT_NOP; 1776 } 1777 if (TCP_MAXOLEN - optlen < TCPOLEN_SIGNATURE) { 1778 to->to_flags &= ~TOF_SIGNATURE; 1779 continue; 1780 } 1781 optlen += TCPOLEN_SIGNATURE; 1782 *optp++ = TCPOPT_SIGNATURE; 1783 *optp++ = TCPOLEN_SIGNATURE; 1784 to->to_signature = optp; 1785 while (siglen--) 1786 *optp++ = 0; 1787 break; 1788 } 1789 case TOF_SACK: 1790 { 1791 int sackblks = 0; 1792 struct sackblk *sack = (struct sackblk *)to->to_sacks; 1793 tcp_seq sack_seq; 1794 1795 while (!optlen || optlen % 4 != 2) { 1796 optlen += TCPOLEN_NOP; 1797 *optp++ = TCPOPT_NOP; 1798 } 1799 if (TCP_MAXOLEN - optlen < TCPOLEN_SACKHDR + TCPOLEN_SACK) 1800 continue; 1801 optlen += TCPOLEN_SACKHDR; 1802 *optp++ = TCPOPT_SACK; 1803 sackblks = min(to->to_nsacks, 1804 (TCP_MAXOLEN - optlen) / TCPOLEN_SACK); 1805 *optp++ = TCPOLEN_SACKHDR + sackblks * TCPOLEN_SACK; 1806 while (sackblks--) { 1807 sack_seq = htonl(sack->start); 1808 bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq)); 1809 optp += sizeof(sack_seq); 1810 sack_seq = htonl(sack->end); 1811 bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq)); 1812 optp += sizeof(sack_seq); 1813 optlen += TCPOLEN_SACK; 1814 sack++; 1815 } 1816 TCPSTAT_INC(tcps_sack_send_blocks); 1817 break; 1818 } 1819 #ifdef TCP_RFC7413 1820 case TOF_FASTOPEN: 1821 { 1822 int total_len; 1823 1824 /* XXX is there any point to aligning this option? */ 1825 total_len = TCPOLEN_FAST_OPEN_EMPTY + to->to_tfo_len; 1826 if (TCP_MAXOLEN - optlen < total_len) 1827 continue; 1828 *optp++ = TCPOPT_FAST_OPEN; 1829 *optp++ = total_len; 1830 if (to->to_tfo_len > 0) { 1831 bcopy(to->to_tfo_cookie, optp, to->to_tfo_len); 1832 optp += to->to_tfo_len; 1833 } 1834 optlen += total_len; 1835 break; 1836 } 1837 #endif 1838 default: 1839 panic("%s: unknown TCP option type", __func__); 1840 break; 1841 } 1842 } 1843 1844 /* Terminate and pad TCP options to a 4 byte boundary. */ 1845 if (optlen % 4) { 1846 optlen += TCPOLEN_EOL; 1847 *optp++ = TCPOPT_EOL; 1848 } 1849 /* 1850 * According to RFC 793 (STD0007): 1851 * "The content of the header beyond the End-of-Option option 1852 * must be header padding (i.e., zero)." 1853 * and later: "The padding is composed of zeros." 1854 */ 1855 while (optlen % 4) { 1856 optlen += TCPOLEN_PAD; 1857 *optp++ = TCPOPT_PAD; 1858 } 1859 1860 KASSERT(optlen <= TCP_MAXOLEN, ("%s: TCP options too long", __func__)); 1861 return (optlen); 1862 } 1863