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