tcp_input.c revision 1.249
1 /* $NetBSD: tcp_input.c,v 1.249 2006/10/12 01:32:38 christos Exp $ */ 2 3 /* 4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the project nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 /* 33 * @(#)COPYRIGHT 1.1 (NRL) 17 January 1995 34 * 35 * NRL grants permission for redistribution and use in source and binary 36 * forms, with or without modification, of the software and documentation 37 * created at NRL provided that the following conditions are met: 38 * 39 * 1. Redistributions of source code must retain the above copyright 40 * notice, this list of conditions and the following disclaimer. 41 * 2. Redistributions in binary form must reproduce the above copyright 42 * notice, this list of conditions and the following disclaimer in the 43 * documentation and/or other materials provided with the distribution. 44 * 3. All advertising materials mentioning features or use of this software 45 * must display the following acknowledgements: 46 * This product includes software developed by the University of 47 * California, Berkeley and its contributors. 48 * This product includes software developed at the Information 49 * Technology Division, US Naval Research Laboratory. 50 * 4. Neither the name of the NRL nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS 55 * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 56 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A 57 * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NRL OR 58 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 59 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 60 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 61 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 62 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 63 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 64 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 65 * 66 * The views and conclusions contained in the software and documentation 67 * are those of the authors and should not be interpreted as representing 68 * official policies, either expressed or implied, of the US Naval 69 * Research Laboratory (NRL). 70 */ 71 72 /*- 73 * Copyright (c) 1997, 1998, 1999, 2001, 2005, 2006 The NetBSD Foundation, Inc. 74 * All rights reserved. 75 * 76 * This code is derived from software contributed to The NetBSD Foundation 77 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 78 * Facility, NASA Ames Research Center. 79 * This code is derived from software contributed to The NetBSD Foundation 80 * by Charles M. Hannum. 81 * This code is derived from software contributed to The NetBSD Foundation 82 * by Rui Paulo. 83 * 84 * Redistribution and use in source and binary forms, with or without 85 * modification, are permitted provided that the following conditions 86 * are met: 87 * 1. Redistributions of source code must retain the above copyright 88 * notice, this list of conditions and the following disclaimer. 89 * 2. Redistributions in binary form must reproduce the above copyright 90 * notice, this list of conditions and the following disclaimer in the 91 * documentation and/or other materials provided with the distribution. 92 * 3. All advertising materials mentioning features or use of this software 93 * must display the following acknowledgement: 94 * This product includes software developed by the NetBSD 95 * Foundation, Inc. and its contributors. 96 * 4. Neither the name of The NetBSD Foundation nor the names of its 97 * contributors may be used to endorse or promote products derived 98 * from this software without specific prior written permission. 99 * 100 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 101 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 102 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 103 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 104 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 105 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 106 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 107 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 108 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 109 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 110 * POSSIBILITY OF SUCH DAMAGE. 111 */ 112 113 /* 114 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 115 * The Regents of the University of California. All rights reserved. 116 * 117 * Redistribution and use in source and binary forms, with or without 118 * modification, are permitted provided that the following conditions 119 * are met: 120 * 1. Redistributions of source code must retain the above copyright 121 * notice, this list of conditions and the following disclaimer. 122 * 2. Redistributions in binary form must reproduce the above copyright 123 * notice, this list of conditions and the following disclaimer in the 124 * documentation and/or other materials provided with the distribution. 125 * 3. Neither the name of the University nor the names of its contributors 126 * may be used to endorse or promote products derived from this software 127 * without specific prior written permission. 128 * 129 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 130 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 131 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 132 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 133 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 134 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 135 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 136 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 137 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 138 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 139 * SUCH DAMAGE. 140 * 141 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 142 */ 143 144 /* 145 * TODO list for SYN cache stuff: 146 * 147 * Find room for a "state" field, which is needed to keep a 148 * compressed state for TIME_WAIT TCBs. It's been noted already 149 * that this is fairly important for very high-volume web and 150 * mail servers, which use a large number of short-lived 151 * connections. 152 */ 153 154 #include <sys/cdefs.h> 155 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.249 2006/10/12 01:32:38 christos Exp $"); 156 157 #include "opt_inet.h" 158 #include "opt_ipsec.h" 159 #include "opt_inet_csum.h" 160 #include "opt_tcp_debug.h" 161 162 #include <sys/param.h> 163 #include <sys/systm.h> 164 #include <sys/malloc.h> 165 #include <sys/mbuf.h> 166 #include <sys/protosw.h> 167 #include <sys/socket.h> 168 #include <sys/socketvar.h> 169 #include <sys/errno.h> 170 #include <sys/syslog.h> 171 #include <sys/pool.h> 172 #include <sys/domain.h> 173 #include <sys/kernel.h> 174 #ifdef TCP_SIGNATURE 175 #include <sys/md5.h> 176 #endif 177 178 #include <net/if.h> 179 #include <net/route.h> 180 #include <net/if_types.h> 181 182 #include <netinet/in.h> 183 #include <netinet/in_systm.h> 184 #include <netinet/ip.h> 185 #include <netinet/in_pcb.h> 186 #include <netinet/in_var.h> 187 #include <netinet/ip_var.h> 188 #include <netinet/in_offload.h> 189 190 #ifdef INET6 191 #ifndef INET 192 #include <netinet/in.h> 193 #endif 194 #include <netinet/ip6.h> 195 #include <netinet6/ip6_var.h> 196 #include <netinet6/in6_pcb.h> 197 #include <netinet6/ip6_var.h> 198 #include <netinet6/in6_var.h> 199 #include <netinet/icmp6.h> 200 #include <netinet6/nd6.h> 201 #ifdef TCP_SIGNATURE 202 #include <netinet6/scope6_var.h> 203 #endif 204 #endif 205 206 #ifndef INET6 207 /* always need ip6.h for IP6_EXTHDR_GET */ 208 #include <netinet/ip6.h> 209 #endif 210 211 #include <netinet/tcp.h> 212 #include <netinet/tcp_fsm.h> 213 #include <netinet/tcp_seq.h> 214 #include <netinet/tcp_timer.h> 215 #include <netinet/tcp_var.h> 216 #include <netinet/tcpip.h> 217 #include <netinet/tcp_congctl.h> 218 #include <netinet/tcp_debug.h> 219 220 #include <machine/stdarg.h> 221 222 #ifdef IPSEC 223 #include <netinet6/ipsec.h> 224 #include <netkey/key.h> 225 #endif /*IPSEC*/ 226 #ifdef INET6 227 #include "faith.h" 228 #if defined(NFAITH) && NFAITH > 0 229 #include <net/if_faith.h> 230 #endif 231 #endif /* IPSEC */ 232 233 #ifdef FAST_IPSEC 234 #include <netipsec/ipsec.h> 235 #include <netipsec/ipsec_var.h> /* XXX ipsecstat namespace */ 236 #include <netipsec/key.h> 237 #ifdef INET6 238 #include <netipsec/ipsec6.h> 239 #endif 240 #endif /* FAST_IPSEC*/ 241 242 int tcprexmtthresh = 3; 243 int tcp_log_refused; 244 245 static int tcp_rst_ppslim_count = 0; 246 static struct timeval tcp_rst_ppslim_last; 247 static int tcp_ackdrop_ppslim_count = 0; 248 static struct timeval tcp_ackdrop_ppslim_last; 249 250 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 251 252 /* for modulo comparisons of timestamps */ 253 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 254 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 255 256 /* 257 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 258 */ 259 #ifdef INET6 260 #define ND6_HINT(tp) \ 261 do { \ 262 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 && \ 263 tp->t_in6pcb->in6p_route.ro_rt) { \ 264 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \ 265 } \ 266 } while (/*CONSTCOND*/ 0) 267 #else 268 #define ND6_HINT(tp) 269 #endif 270 271 /* 272 * Macro to compute ACK transmission behavior. Delay the ACK unless 273 * we have already delayed an ACK (must send an ACK every two segments). 274 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 275 * option is enabled. 276 */ 277 #define TCP_SETUP_ACK(tp, th) \ 278 do { \ 279 if ((tp)->t_flags & TF_DELACK || \ 280 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 281 tp->t_flags |= TF_ACKNOW; \ 282 else \ 283 TCP_SET_DELACK(tp); \ 284 } while (/*CONSTCOND*/ 0) 285 286 #define ICMP_CHECK(tp, th, acked) \ 287 do { \ 288 /* \ 289 * If we had a pending ICMP message that \ 290 * refers to data that have just been \ 291 * acknowledged, disregard the recorded ICMP \ 292 * message. \ 293 */ \ 294 if (((tp)->t_flags & TF_PMTUD_PEND) && \ 295 SEQ_GT((th)->th_ack, (tp)->t_pmtud_th_seq)) \ 296 (tp)->t_flags &= ~TF_PMTUD_PEND; \ 297 \ 298 /* \ 299 * Keep track of the largest chunk of data \ 300 * acknowledged since last PMTU update \ 301 */ \ 302 if ((tp)->t_pmtud_mss_acked < (acked)) \ 303 (tp)->t_pmtud_mss_acked = (acked); \ 304 } while (/*CONSTCOND*/ 0) 305 306 /* 307 * Convert TCP protocol fields to host order for easier processing. 308 */ 309 #define TCP_FIELDS_TO_HOST(th) \ 310 do { \ 311 NTOHL((th)->th_seq); \ 312 NTOHL((th)->th_ack); \ 313 NTOHS((th)->th_win); \ 314 NTOHS((th)->th_urp); \ 315 } while (/*CONSTCOND*/ 0) 316 317 /* 318 * ... and reverse the above. 319 */ 320 #define TCP_FIELDS_TO_NET(th) \ 321 do { \ 322 HTONL((th)->th_seq); \ 323 HTONL((th)->th_ack); \ 324 HTONS((th)->th_win); \ 325 HTONS((th)->th_urp); \ 326 } while (/*CONSTCOND*/ 0) 327 328 #ifdef TCP_CSUM_COUNTERS 329 #include <sys/device.h> 330 331 #if defined(INET) 332 extern struct evcnt tcp_hwcsum_ok; 333 extern struct evcnt tcp_hwcsum_bad; 334 extern struct evcnt tcp_hwcsum_data; 335 extern struct evcnt tcp_swcsum; 336 #endif /* defined(INET) */ 337 #if defined(INET6) 338 extern struct evcnt tcp6_hwcsum_ok; 339 extern struct evcnt tcp6_hwcsum_bad; 340 extern struct evcnt tcp6_hwcsum_data; 341 extern struct evcnt tcp6_swcsum; 342 #endif /* defined(INET6) */ 343 344 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 345 346 #else 347 348 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 349 350 #endif /* TCP_CSUM_COUNTERS */ 351 352 #ifdef TCP_REASS_COUNTERS 353 #include <sys/device.h> 354 355 extern struct evcnt tcp_reass_; 356 extern struct evcnt tcp_reass_empty; 357 extern struct evcnt tcp_reass_iteration[8]; 358 extern struct evcnt tcp_reass_prependfirst; 359 extern struct evcnt tcp_reass_prepend; 360 extern struct evcnt tcp_reass_insert; 361 extern struct evcnt tcp_reass_inserttail; 362 extern struct evcnt tcp_reass_append; 363 extern struct evcnt tcp_reass_appendtail; 364 extern struct evcnt tcp_reass_overlaptail; 365 extern struct evcnt tcp_reass_overlapfront; 366 extern struct evcnt tcp_reass_segdup; 367 extern struct evcnt tcp_reass_fragdup; 368 369 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 370 371 #else 372 373 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 374 375 #endif /* TCP_REASS_COUNTERS */ 376 377 #ifdef INET 378 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 379 #endif 380 #ifdef INET6 381 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 382 #endif 383 384 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 385 386 POOL_INIT(tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", NULL); 387 388 struct ipqent * 389 tcpipqent_alloc() 390 { 391 struct ipqent *ipqe; 392 int s; 393 394 s = splvm(); 395 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 396 splx(s); 397 398 return ipqe; 399 } 400 401 void 402 tcpipqent_free(struct ipqent *ipqe) 403 { 404 int s; 405 406 s = splvm(); 407 pool_put(&tcpipqent_pool, ipqe); 408 splx(s); 409 } 410 411 int 412 tcp_reass(struct tcpcb *tp, struct tcphdr *th, struct mbuf *m, int *tlen) 413 { 414 struct ipqent *p, *q, *nq, *tiqe = NULL; 415 struct socket *so = NULL; 416 int pkt_flags; 417 tcp_seq pkt_seq; 418 unsigned pkt_len; 419 u_long rcvpartdupbyte = 0; 420 u_long rcvoobyte; 421 #ifdef TCP_REASS_COUNTERS 422 u_int count = 0; 423 #endif 424 425 if (tp->t_inpcb) 426 so = tp->t_inpcb->inp_socket; 427 #ifdef INET6 428 else if (tp->t_in6pcb) 429 so = tp->t_in6pcb->in6p_socket; 430 #endif 431 432 TCP_REASS_LOCK_CHECK(tp); 433 434 /* 435 * Call with th==0 after become established to 436 * force pre-ESTABLISHED data up to user socket. 437 */ 438 if (th == 0) 439 goto present; 440 441 rcvoobyte = *tlen; 442 /* 443 * Copy these to local variables because the tcpiphdr 444 * gets munged while we are collapsing mbufs. 445 */ 446 pkt_seq = th->th_seq; 447 pkt_len = *tlen; 448 pkt_flags = th->th_flags; 449 450 TCP_REASS_COUNTER_INCR(&tcp_reass_); 451 452 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 453 /* 454 * When we miss a packet, the vast majority of time we get 455 * packets that follow it in order. So optimize for that. 456 */ 457 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 458 p->ipqe_len += pkt_len; 459 p->ipqe_flags |= pkt_flags; 460 m_cat(p->ipre_mlast, m); 461 TRAVERSE(p->ipre_mlast); 462 m = NULL; 463 tiqe = p; 464 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 465 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 466 goto skip_replacement; 467 } 468 /* 469 * While we're here, if the pkt is completely beyond 470 * anything we have, just insert it at the tail. 471 */ 472 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 473 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 474 goto insert_it; 475 } 476 } 477 478 q = TAILQ_FIRST(&tp->segq); 479 480 if (q != NULL) { 481 /* 482 * If this segment immediately precedes the first out-of-order 483 * block, simply slap the segment in front of it and (mostly) 484 * skip the complicated logic. 485 */ 486 if (pkt_seq + pkt_len == q->ipqe_seq) { 487 q->ipqe_seq = pkt_seq; 488 q->ipqe_len += pkt_len; 489 q->ipqe_flags |= pkt_flags; 490 m_cat(m, q->ipqe_m); 491 q->ipqe_m = m; 492 q->ipre_mlast = m; /* last mbuf may have changed */ 493 TRAVERSE(q->ipre_mlast); 494 tiqe = q; 495 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 496 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 497 goto skip_replacement; 498 } 499 } else { 500 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 501 } 502 503 /* 504 * Find a segment which begins after this one does. 505 */ 506 for (p = NULL; q != NULL; q = nq) { 507 nq = TAILQ_NEXT(q, ipqe_q); 508 #ifdef TCP_REASS_COUNTERS 509 count++; 510 #endif 511 /* 512 * If the received segment is just right after this 513 * fragment, merge the two together and then check 514 * for further overlaps. 515 */ 516 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 517 #ifdef TCPREASS_DEBUG 518 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 519 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 520 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 521 #endif 522 pkt_len += q->ipqe_len; 523 pkt_flags |= q->ipqe_flags; 524 pkt_seq = q->ipqe_seq; 525 m_cat(q->ipre_mlast, m); 526 TRAVERSE(q->ipre_mlast); 527 m = q->ipqe_m; 528 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 529 goto free_ipqe; 530 } 531 /* 532 * If the received segment is completely past this 533 * fragment, we need to go the next fragment. 534 */ 535 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 536 p = q; 537 continue; 538 } 539 /* 540 * If the fragment is past the received segment, 541 * it (or any following) can't be concatenated. 542 */ 543 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 544 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 545 break; 546 } 547 548 /* 549 * We've received all the data in this segment before. 550 * mark it as a duplicate and return. 551 */ 552 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 553 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 554 tcpstat.tcps_rcvduppack++; 555 tcpstat.tcps_rcvdupbyte += pkt_len; 556 tcp_new_dsack(tp, pkt_seq, pkt_len); 557 m_freem(m); 558 if (tiqe != NULL) { 559 tcpipqent_free(tiqe); 560 } 561 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 562 return (0); 563 } 564 /* 565 * Received segment completely overlaps this fragment 566 * so we drop the fragment (this keeps the temporal 567 * ordering of segments correct). 568 */ 569 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 570 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 571 rcvpartdupbyte += q->ipqe_len; 572 m_freem(q->ipqe_m); 573 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 574 goto free_ipqe; 575 } 576 /* 577 * RX'ed segment extends past the end of the 578 * fragment. Drop the overlapping bytes. Then 579 * merge the fragment and segment then treat as 580 * a longer received packet. 581 */ 582 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 583 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 584 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 585 #ifdef TCPREASS_DEBUG 586 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 587 tp, overlap, 588 pkt_seq, pkt_seq + pkt_len, pkt_len); 589 #endif 590 m_adj(m, overlap); 591 rcvpartdupbyte += overlap; 592 m_cat(q->ipre_mlast, m); 593 TRAVERSE(q->ipre_mlast); 594 m = q->ipqe_m; 595 pkt_seq = q->ipqe_seq; 596 pkt_len += q->ipqe_len - overlap; 597 rcvoobyte -= overlap; 598 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 599 goto free_ipqe; 600 } 601 /* 602 * RX'ed segment extends past the front of the 603 * fragment. Drop the overlapping bytes on the 604 * received packet. The packet will then be 605 * contatentated with this fragment a bit later. 606 */ 607 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 608 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 609 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 610 #ifdef TCPREASS_DEBUG 611 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 612 tp, overlap, 613 pkt_seq, pkt_seq + pkt_len, pkt_len); 614 #endif 615 m_adj(m, -overlap); 616 pkt_len -= overlap; 617 rcvpartdupbyte += overlap; 618 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 619 rcvoobyte -= overlap; 620 } 621 /* 622 * If the received segment immediates precedes this 623 * fragment then tack the fragment onto this segment 624 * and reinsert the data. 625 */ 626 if (q->ipqe_seq == pkt_seq + pkt_len) { 627 #ifdef TCPREASS_DEBUG 628 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 629 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 630 pkt_seq, pkt_seq + pkt_len, pkt_len); 631 #endif 632 pkt_len += q->ipqe_len; 633 pkt_flags |= q->ipqe_flags; 634 m_cat(m, q->ipqe_m); 635 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 636 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 637 tp->t_segqlen--; 638 KASSERT(tp->t_segqlen >= 0); 639 KASSERT(tp->t_segqlen != 0 || 640 (TAILQ_EMPTY(&tp->segq) && 641 TAILQ_EMPTY(&tp->timeq))); 642 if (tiqe == NULL) { 643 tiqe = q; 644 } else { 645 tcpipqent_free(q); 646 } 647 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 648 break; 649 } 650 /* 651 * If the fragment is before the segment, remember it. 652 * When this loop is terminated, p will contain the 653 * pointer to fragment that is right before the received 654 * segment. 655 */ 656 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 657 p = q; 658 659 continue; 660 661 /* 662 * This is a common operation. It also will allow 663 * to save doing a malloc/free in most instances. 664 */ 665 free_ipqe: 666 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 667 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 668 tp->t_segqlen--; 669 KASSERT(tp->t_segqlen >= 0); 670 KASSERT(tp->t_segqlen != 0 || 671 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 672 if (tiqe == NULL) { 673 tiqe = q; 674 } else { 675 tcpipqent_free(q); 676 } 677 } 678 679 #ifdef TCP_REASS_COUNTERS 680 if (count > 7) 681 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 682 else if (count > 0) 683 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 684 #endif 685 686 insert_it: 687 688 /* 689 * Allocate a new queue entry since the received segment did not 690 * collapse onto any other out-of-order block; thus we are allocating 691 * a new block. If it had collapsed, tiqe would not be NULL and 692 * we would be reusing it. 693 * XXX If we can't, just drop the packet. XXX 694 */ 695 if (tiqe == NULL) { 696 tiqe = tcpipqent_alloc(); 697 if (tiqe == NULL) { 698 tcpstat.tcps_rcvmemdrop++; 699 m_freem(m); 700 return (0); 701 } 702 } 703 704 /* 705 * Update the counters. 706 */ 707 tcpstat.tcps_rcvoopack++; 708 tcpstat.tcps_rcvoobyte += rcvoobyte; 709 if (rcvpartdupbyte) { 710 tcpstat.tcps_rcvpartduppack++; 711 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 712 } 713 714 /* 715 * Insert the new fragment queue entry into both queues. 716 */ 717 tiqe->ipqe_m = m; 718 tiqe->ipre_mlast = m; 719 tiqe->ipqe_seq = pkt_seq; 720 tiqe->ipqe_len = pkt_len; 721 tiqe->ipqe_flags = pkt_flags; 722 if (p == NULL) { 723 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 724 #ifdef TCPREASS_DEBUG 725 if (tiqe->ipqe_seq != tp->rcv_nxt) 726 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 727 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 728 #endif 729 } else { 730 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 731 #ifdef TCPREASS_DEBUG 732 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 733 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 734 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 735 #endif 736 } 737 tp->t_segqlen++; 738 739 skip_replacement: 740 741 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 742 743 present: 744 /* 745 * Present data to user, advancing rcv_nxt through 746 * completed sequence space. 747 */ 748 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 749 return (0); 750 q = TAILQ_FIRST(&tp->segq); 751 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 752 return (0); 753 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 754 return (0); 755 756 tp->rcv_nxt += q->ipqe_len; 757 pkt_flags = q->ipqe_flags & TH_FIN; 758 ND6_HINT(tp); 759 760 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 761 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 762 tp->t_segqlen--; 763 KASSERT(tp->t_segqlen >= 0); 764 KASSERT(tp->t_segqlen != 0 || 765 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 766 if (so->so_state & SS_CANTRCVMORE) 767 m_freem(q->ipqe_m); 768 else 769 sbappendstream(&so->so_rcv, q->ipqe_m); 770 tcpipqent_free(q); 771 sorwakeup(so); 772 return (pkt_flags); 773 } 774 775 #ifdef INET6 776 int 777 tcp6_input(struct mbuf **mp, int *offp, int proto) 778 { 779 struct mbuf *m = *mp; 780 781 /* 782 * draft-itojun-ipv6-tcp-to-anycast 783 * better place to put this in? 784 */ 785 if (m->m_flags & M_ANYCAST6) { 786 struct ip6_hdr *ip6; 787 if (m->m_len < sizeof(struct ip6_hdr)) { 788 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 789 tcpstat.tcps_rcvshort++; 790 return IPPROTO_DONE; 791 } 792 } 793 ip6 = mtod(m, struct ip6_hdr *); 794 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 795 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 796 return IPPROTO_DONE; 797 } 798 799 tcp_input(m, *offp, proto); 800 return IPPROTO_DONE; 801 } 802 #endif 803 804 #ifdef INET 805 static void 806 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 807 { 808 char src[4*sizeof "123"]; 809 char dst[4*sizeof "123"]; 810 811 if (ip) { 812 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 813 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 814 } 815 else { 816 strlcpy(src, "(unknown)", sizeof(src)); 817 strlcpy(dst, "(unknown)", sizeof(dst)); 818 } 819 log(LOG_INFO, 820 "Connection attempt to TCP %s:%d from %s:%d\n", 821 dst, ntohs(th->th_dport), 822 src, ntohs(th->th_sport)); 823 } 824 #endif 825 826 #ifdef INET6 827 static void 828 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 829 { 830 char src[INET6_ADDRSTRLEN]; 831 char dst[INET6_ADDRSTRLEN]; 832 833 if (ip6) { 834 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 835 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 836 } 837 else { 838 strlcpy(src, "(unknown v6)", sizeof(src)); 839 strlcpy(dst, "(unknown v6)", sizeof(dst)); 840 } 841 log(LOG_INFO, 842 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 843 dst, ntohs(th->th_dport), 844 src, ntohs(th->th_sport)); 845 } 846 #endif 847 848 /* 849 * Checksum extended TCP header and data. 850 */ 851 int 852 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th __unused, 853 int toff, int off, int tlen) 854 { 855 856 /* 857 * XXX it's better to record and check if this mbuf is 858 * already checked. 859 */ 860 861 switch (af) { 862 #ifdef INET 863 case AF_INET: 864 switch (m->m_pkthdr.csum_flags & 865 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 866 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 867 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 868 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 869 goto badcsum; 870 871 case M_CSUM_TCPv4|M_CSUM_DATA: { 872 u_int32_t hw_csum = m->m_pkthdr.csum_data; 873 874 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 875 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 876 const struct ip *ip = 877 mtod(m, const struct ip *); 878 879 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 880 ip->ip_dst.s_addr, 881 htons(hw_csum + tlen + off + IPPROTO_TCP)); 882 } 883 if ((hw_csum ^ 0xffff) != 0) 884 goto badcsum; 885 break; 886 } 887 888 case M_CSUM_TCPv4: 889 /* Checksum was okay. */ 890 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 891 break; 892 893 default: 894 /* 895 * Must compute it ourselves. Maybe skip checksum 896 * on loopback interfaces. 897 */ 898 if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & 899 IFF_LOOPBACK) || 900 tcp_do_loopback_cksum)) { 901 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 902 if (in4_cksum(m, IPPROTO_TCP, toff, 903 tlen + off) != 0) 904 goto badcsum; 905 } 906 break; 907 } 908 break; 909 #endif /* INET4 */ 910 911 #ifdef INET6 912 case AF_INET6: 913 switch (m->m_pkthdr.csum_flags & 914 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 915 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 916 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 917 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 918 goto badcsum; 919 920 #if 0 /* notyet */ 921 case M_CSUM_TCPv6|M_CSUM_DATA: 922 #endif 923 924 case M_CSUM_TCPv6: 925 /* Checksum was okay. */ 926 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 927 break; 928 929 default: 930 /* 931 * Must compute it ourselves. Maybe skip checksum 932 * on loopback interfaces. 933 */ 934 if (__predict_true((m->m_flags & M_LOOP) == 0 || 935 tcp_do_loopback_cksum)) { 936 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 937 if (in6_cksum(m, IPPROTO_TCP, toff, 938 tlen + off) != 0) 939 goto badcsum; 940 } 941 } 942 break; 943 #endif /* INET6 */ 944 } 945 946 return 0; 947 948 badcsum: 949 tcpstat.tcps_rcvbadsum++; 950 return -1; 951 } 952 953 /* 954 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 955 */ 956 void 957 tcp_input(struct mbuf *m, ...) 958 { 959 struct tcphdr *th; 960 struct ip *ip; 961 struct inpcb *inp; 962 #ifdef INET6 963 struct ip6_hdr *ip6; 964 struct in6pcb *in6p; 965 #endif 966 u_int8_t *optp = NULL; 967 int optlen = 0; 968 int len, tlen, toff, hdroptlen = 0; 969 struct tcpcb *tp = 0; 970 int tiflags; 971 struct socket *so = NULL; 972 int todrop, dupseg, acked, ourfinisacked, needoutput = 0; 973 #ifdef TCP_DEBUG 974 short ostate = 0; 975 #endif 976 u_long tiwin; 977 struct tcp_opt_info opti; 978 int off, iphlen; 979 va_list ap; 980 int af; /* af on the wire */ 981 struct mbuf *tcp_saveti = NULL; 982 uint32_t ts_rtt; 983 uint8_t iptos; 984 985 MCLAIM(m, &tcp_rx_mowner); 986 va_start(ap, m); 987 toff = va_arg(ap, int); 988 (void)va_arg(ap, int); /* ignore value, advance ap */ 989 va_end(ap); 990 991 tcpstat.tcps_rcvtotal++; 992 993 bzero(&opti, sizeof(opti)); 994 opti.ts_present = 0; 995 opti.maxseg = 0; 996 997 /* 998 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 999 * 1000 * TCP is, by definition, unicast, so we reject all 1001 * multicast outright. 1002 * 1003 * Note, there are additional src/dst address checks in 1004 * the AF-specific code below. 1005 */ 1006 if (m->m_flags & (M_BCAST|M_MCAST)) { 1007 /* XXX stat */ 1008 goto drop; 1009 } 1010 #ifdef INET6 1011 if (m->m_flags & M_ANYCAST6) { 1012 /* XXX stat */ 1013 goto drop; 1014 } 1015 #endif 1016 1017 /* 1018 * Get IP and TCP header. 1019 * Note: IP leaves IP header in first mbuf. 1020 */ 1021 ip = mtod(m, struct ip *); 1022 #ifdef INET6 1023 ip6 = NULL; 1024 #endif 1025 switch (ip->ip_v) { 1026 #ifdef INET 1027 case 4: 1028 af = AF_INET; 1029 iphlen = sizeof(struct ip); 1030 ip = mtod(m, struct ip *); 1031 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1032 sizeof(struct tcphdr)); 1033 if (th == NULL) { 1034 tcpstat.tcps_rcvshort++; 1035 return; 1036 } 1037 /* We do the checksum after PCB lookup... */ 1038 len = ntohs(ip->ip_len); 1039 tlen = len - toff; 1040 iptos = ip->ip_tos; 1041 break; 1042 #endif 1043 #ifdef INET6 1044 case 6: 1045 ip = NULL; 1046 iphlen = sizeof(struct ip6_hdr); 1047 af = AF_INET6; 1048 ip6 = mtod(m, struct ip6_hdr *); 1049 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1050 sizeof(struct tcphdr)); 1051 if (th == NULL) { 1052 tcpstat.tcps_rcvshort++; 1053 return; 1054 } 1055 1056 /* Be proactive about malicious use of IPv4 mapped address */ 1057 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 1058 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 1059 /* XXX stat */ 1060 goto drop; 1061 } 1062 1063 /* 1064 * Be proactive about unspecified IPv6 address in source. 1065 * As we use all-zero to indicate unbounded/unconnected pcb, 1066 * unspecified IPv6 address can be used to confuse us. 1067 * 1068 * Note that packets with unspecified IPv6 destination is 1069 * already dropped in ip6_input. 1070 */ 1071 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1072 /* XXX stat */ 1073 goto drop; 1074 } 1075 1076 /* 1077 * Make sure destination address is not multicast. 1078 * Source address checked in ip6_input(). 1079 */ 1080 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1081 /* XXX stat */ 1082 goto drop; 1083 } 1084 1085 /* We do the checksum after PCB lookup... */ 1086 len = m->m_pkthdr.len; 1087 tlen = len - toff; 1088 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1089 break; 1090 #endif 1091 default: 1092 m_freem(m); 1093 return; 1094 } 1095 1096 KASSERT(TCP_HDR_ALIGNED_P(th)); 1097 1098 /* 1099 * Check that TCP offset makes sense, 1100 * pull out TCP options and adjust length. XXX 1101 */ 1102 off = th->th_off << 2; 1103 if (off < sizeof (struct tcphdr) || off > tlen) { 1104 tcpstat.tcps_rcvbadoff++; 1105 goto drop; 1106 } 1107 tlen -= off; 1108 1109 /* 1110 * tcp_input() has been modified to use tlen to mean the TCP data 1111 * length throughout the function. Other functions can use 1112 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1113 * rja 1114 */ 1115 1116 if (off > sizeof (struct tcphdr)) { 1117 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1118 if (th == NULL) { 1119 tcpstat.tcps_rcvshort++; 1120 return; 1121 } 1122 /* 1123 * NOTE: ip/ip6 will not be affected by m_pulldown() 1124 * (as they're before toff) and we don't need to update those. 1125 */ 1126 KASSERT(TCP_HDR_ALIGNED_P(th)); 1127 optlen = off - sizeof (struct tcphdr); 1128 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1129 /* 1130 * Do quick retrieval of timestamp options ("options 1131 * prediction?"). If timestamp is the only option and it's 1132 * formatted as recommended in RFC 1323 appendix A, we 1133 * quickly get the values now and not bother calling 1134 * tcp_dooptions(), etc. 1135 */ 1136 if ((optlen == TCPOLEN_TSTAMP_APPA || 1137 (optlen > TCPOLEN_TSTAMP_APPA && 1138 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1139 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1140 (th->th_flags & TH_SYN) == 0) { 1141 opti.ts_present = 1; 1142 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1143 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1144 optp = NULL; /* we've parsed the options */ 1145 } 1146 } 1147 tiflags = th->th_flags; 1148 1149 /* 1150 * Locate pcb for segment. 1151 */ 1152 findpcb: 1153 inp = NULL; 1154 #ifdef INET6 1155 in6p = NULL; 1156 #endif 1157 switch (af) { 1158 #ifdef INET 1159 case AF_INET: 1160 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1161 ip->ip_dst, th->th_dport); 1162 if (inp == 0) { 1163 ++tcpstat.tcps_pcbhashmiss; 1164 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1165 } 1166 #ifdef INET6 1167 if (inp == 0) { 1168 struct in6_addr s, d; 1169 1170 /* mapped addr case */ 1171 bzero(&s, sizeof(s)); 1172 s.s6_addr16[5] = htons(0xffff); 1173 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1174 bzero(&d, sizeof(d)); 1175 d.s6_addr16[5] = htons(0xffff); 1176 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1177 in6p = in6_pcblookup_connect(&tcbtable, &s, 1178 th->th_sport, &d, th->th_dport, 0); 1179 if (in6p == 0) { 1180 ++tcpstat.tcps_pcbhashmiss; 1181 in6p = in6_pcblookup_bind(&tcbtable, &d, 1182 th->th_dport, 0); 1183 } 1184 } 1185 #endif 1186 #ifndef INET6 1187 if (inp == 0) 1188 #else 1189 if (inp == 0 && in6p == 0) 1190 #endif 1191 { 1192 ++tcpstat.tcps_noport; 1193 if (tcp_log_refused && 1194 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1195 tcp4_log_refused(ip, th); 1196 } 1197 TCP_FIELDS_TO_HOST(th); 1198 goto dropwithreset_ratelim; 1199 } 1200 #if defined(IPSEC) || defined(FAST_IPSEC) 1201 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1202 ipsec4_in_reject(m, inp)) { 1203 ipsecstat.in_polvio++; 1204 goto drop; 1205 } 1206 #ifdef INET6 1207 else if (in6p && 1208 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1209 ipsec4_in_reject_so(m, in6p->in6p_socket)) { 1210 ipsecstat.in_polvio++; 1211 goto drop; 1212 } 1213 #endif 1214 #endif /*IPSEC*/ 1215 break; 1216 #endif /*INET*/ 1217 #ifdef INET6 1218 case AF_INET6: 1219 { 1220 int faith; 1221 1222 #if defined(NFAITH) && NFAITH > 0 1223 faith = faithprefix(&ip6->ip6_dst); 1224 #else 1225 faith = 0; 1226 #endif 1227 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1228 th->th_sport, &ip6->ip6_dst, th->th_dport, faith); 1229 if (in6p == NULL) { 1230 ++tcpstat.tcps_pcbhashmiss; 1231 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1232 th->th_dport, faith); 1233 } 1234 if (in6p == NULL) { 1235 ++tcpstat.tcps_noport; 1236 if (tcp_log_refused && 1237 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1238 tcp6_log_refused(ip6, th); 1239 } 1240 TCP_FIELDS_TO_HOST(th); 1241 goto dropwithreset_ratelim; 1242 } 1243 #if defined(IPSEC) || defined(FAST_IPSEC) 1244 if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1245 ipsec6_in_reject(m, in6p)) { 1246 ipsec6stat.in_polvio++; 1247 goto drop; 1248 } 1249 #endif /*IPSEC*/ 1250 break; 1251 } 1252 #endif 1253 } 1254 1255 /* 1256 * If the state is CLOSED (i.e., TCB does not exist) then 1257 * all data in the incoming segment is discarded. 1258 * If the TCB exists but is in CLOSED state, it is embryonic, 1259 * but should either do a listen or a connect soon. 1260 */ 1261 tp = NULL; 1262 so = NULL; 1263 if (inp) { 1264 tp = intotcpcb(inp); 1265 so = inp->inp_socket; 1266 } 1267 #ifdef INET6 1268 else if (in6p) { 1269 tp = in6totcpcb(in6p); 1270 so = in6p->in6p_socket; 1271 } 1272 #endif 1273 if (tp == 0) { 1274 TCP_FIELDS_TO_HOST(th); 1275 goto dropwithreset_ratelim; 1276 } 1277 if (tp->t_state == TCPS_CLOSED) 1278 goto drop; 1279 1280 /* 1281 * Checksum extended TCP header and data. 1282 */ 1283 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1284 goto badcsum; 1285 1286 TCP_FIELDS_TO_HOST(th); 1287 1288 /* Unscale the window into a 32-bit value. */ 1289 if ((tiflags & TH_SYN) == 0) 1290 tiwin = th->th_win << tp->snd_scale; 1291 else 1292 tiwin = th->th_win; 1293 1294 #ifdef INET6 1295 /* save packet options if user wanted */ 1296 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1297 if (in6p->in6p_options) { 1298 m_freem(in6p->in6p_options); 1299 in6p->in6p_options = 0; 1300 } 1301 KASSERT(ip6 != NULL); 1302 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1303 } 1304 #endif 1305 1306 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1307 union syn_cache_sa src; 1308 union syn_cache_sa dst; 1309 1310 bzero(&src, sizeof(src)); 1311 bzero(&dst, sizeof(dst)); 1312 switch (af) { 1313 #ifdef INET 1314 case AF_INET: 1315 src.sin.sin_len = sizeof(struct sockaddr_in); 1316 src.sin.sin_family = AF_INET; 1317 src.sin.sin_addr = ip->ip_src; 1318 src.sin.sin_port = th->th_sport; 1319 1320 dst.sin.sin_len = sizeof(struct sockaddr_in); 1321 dst.sin.sin_family = AF_INET; 1322 dst.sin.sin_addr = ip->ip_dst; 1323 dst.sin.sin_port = th->th_dport; 1324 break; 1325 #endif 1326 #ifdef INET6 1327 case AF_INET6: 1328 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1329 src.sin6.sin6_family = AF_INET6; 1330 src.sin6.sin6_addr = ip6->ip6_src; 1331 src.sin6.sin6_port = th->th_sport; 1332 1333 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1334 dst.sin6.sin6_family = AF_INET6; 1335 dst.sin6.sin6_addr = ip6->ip6_dst; 1336 dst.sin6.sin6_port = th->th_dport; 1337 break; 1338 #endif /* INET6 */ 1339 default: 1340 goto badsyn; /*sanity*/ 1341 } 1342 1343 if (so->so_options & SO_DEBUG) { 1344 #ifdef TCP_DEBUG 1345 ostate = tp->t_state; 1346 #endif 1347 1348 tcp_saveti = NULL; 1349 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1350 goto nosave; 1351 1352 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1353 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1354 if (!tcp_saveti) 1355 goto nosave; 1356 } else { 1357 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1358 if (!tcp_saveti) 1359 goto nosave; 1360 MCLAIM(m, &tcp_mowner); 1361 tcp_saveti->m_len = iphlen; 1362 m_copydata(m, 0, iphlen, 1363 mtod(tcp_saveti, caddr_t)); 1364 } 1365 1366 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1367 m_freem(tcp_saveti); 1368 tcp_saveti = NULL; 1369 } else { 1370 tcp_saveti->m_len += sizeof(struct tcphdr); 1371 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1372 sizeof(struct tcphdr)); 1373 } 1374 nosave:; 1375 } 1376 if (so->so_options & SO_ACCEPTCONN) { 1377 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1378 if (tiflags & TH_RST) { 1379 syn_cache_reset(&src.sa, &dst.sa, th); 1380 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1381 (TH_ACK|TH_SYN)) { 1382 /* 1383 * Received a SYN,ACK. This should 1384 * never happen while we are in 1385 * LISTEN. Send an RST. 1386 */ 1387 goto badsyn; 1388 } else if (tiflags & TH_ACK) { 1389 so = syn_cache_get(&src.sa, &dst.sa, 1390 th, toff, tlen, so, m); 1391 if (so == NULL) { 1392 /* 1393 * We don't have a SYN for 1394 * this ACK; send an RST. 1395 */ 1396 goto badsyn; 1397 } else if (so == 1398 (struct socket *)(-1)) { 1399 /* 1400 * We were unable to create 1401 * the connection. If the 1402 * 3-way handshake was 1403 * completed, and RST has 1404 * been sent to the peer. 1405 * Since the mbuf might be 1406 * in use for the reply, 1407 * do not free it. 1408 */ 1409 m = NULL; 1410 } else { 1411 /* 1412 * We have created a 1413 * full-blown connection. 1414 */ 1415 tp = NULL; 1416 inp = NULL; 1417 #ifdef INET6 1418 in6p = NULL; 1419 #endif 1420 switch (so->so_proto->pr_domain->dom_family) { 1421 #ifdef INET 1422 case AF_INET: 1423 inp = sotoinpcb(so); 1424 tp = intotcpcb(inp); 1425 break; 1426 #endif 1427 #ifdef INET6 1428 case AF_INET6: 1429 in6p = sotoin6pcb(so); 1430 tp = in6totcpcb(in6p); 1431 break; 1432 #endif 1433 } 1434 if (tp == NULL) 1435 goto badsyn; /*XXX*/ 1436 tiwin <<= tp->snd_scale; 1437 goto after_listen; 1438 } 1439 } else { 1440 /* 1441 * None of RST, SYN or ACK was set. 1442 * This is an invalid packet for a 1443 * TCB in LISTEN state. Send a RST. 1444 */ 1445 goto badsyn; 1446 } 1447 } else { 1448 /* 1449 * Received a SYN. 1450 */ 1451 1452 #ifdef INET6 1453 /* 1454 * If deprecated address is forbidden, we do 1455 * not accept SYN to deprecated interface 1456 * address to prevent any new inbound 1457 * connection from getting established. 1458 * When we do not accept SYN, we send a TCP 1459 * RST, with deprecated source address (instead 1460 * of dropping it). We compromise it as it is 1461 * much better for peer to send a RST, and 1462 * RST will be the final packet for the 1463 * exchange. 1464 * 1465 * If we do not forbid deprecated addresses, we 1466 * accept the SYN packet. RFC2462 does not 1467 * suggest dropping SYN in this case. 1468 * If we decipher RFC2462 5.5.4, it says like 1469 * this: 1470 * 1. use of deprecated addr with existing 1471 * communication is okay - "SHOULD continue 1472 * to be used" 1473 * 2. use of it with new communication: 1474 * (2a) "SHOULD NOT be used if alternate 1475 * address with sufficient scope is 1476 * available" 1477 * (2b) nothing mentioned otherwise. 1478 * Here we fall into (2b) case as we have no 1479 * choice in our source address selection - we 1480 * must obey the peer. 1481 * 1482 * The wording in RFC2462 is confusing, and 1483 * there are multiple description text for 1484 * deprecated address handling - worse, they 1485 * are not exactly the same. I believe 5.5.4 1486 * is the best one, so we follow 5.5.4. 1487 */ 1488 if (af == AF_INET6 && !ip6_use_deprecated) { 1489 struct in6_ifaddr *ia6; 1490 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1491 &ip6->ip6_dst)) && 1492 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1493 tp = NULL; 1494 goto dropwithreset; 1495 } 1496 } 1497 #endif 1498 1499 #ifdef IPSEC 1500 switch (af) { 1501 #ifdef INET 1502 case AF_INET: 1503 if (ipsec4_in_reject_so(m, so)) { 1504 ipsecstat.in_polvio++; 1505 tp = NULL; 1506 goto dropwithreset; 1507 } 1508 break; 1509 #endif 1510 #ifdef INET6 1511 case AF_INET6: 1512 if (ipsec6_in_reject_so(m, so)) { 1513 ipsec6stat.in_polvio++; 1514 tp = NULL; 1515 goto dropwithreset; 1516 } 1517 break; 1518 #endif 1519 } 1520 #endif 1521 1522 /* 1523 * LISTEN socket received a SYN 1524 * from itself? This can't possibly 1525 * be valid; drop the packet. 1526 */ 1527 if (th->th_sport == th->th_dport) { 1528 int i; 1529 1530 switch (af) { 1531 #ifdef INET 1532 case AF_INET: 1533 i = in_hosteq(ip->ip_src, ip->ip_dst); 1534 break; 1535 #endif 1536 #ifdef INET6 1537 case AF_INET6: 1538 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1539 break; 1540 #endif 1541 default: 1542 i = 1; 1543 } 1544 if (i) { 1545 tcpstat.tcps_badsyn++; 1546 goto drop; 1547 } 1548 } 1549 1550 /* 1551 * SYN looks ok; create compressed TCP 1552 * state for it. 1553 */ 1554 if (so->so_qlen <= so->so_qlimit && 1555 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1556 so, m, optp, optlen, &opti)) 1557 m = NULL; 1558 } 1559 goto drop; 1560 } 1561 } 1562 1563 after_listen: 1564 #ifdef DIAGNOSTIC 1565 /* 1566 * Should not happen now that all embryonic connections 1567 * are handled with compressed state. 1568 */ 1569 if (tp->t_state == TCPS_LISTEN) 1570 panic("tcp_input: TCPS_LISTEN"); 1571 #endif 1572 1573 /* 1574 * Segment received on connection. 1575 * Reset idle time and keep-alive timer. 1576 */ 1577 tp->t_rcvtime = tcp_now; 1578 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1579 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1580 1581 /* 1582 * Process options. 1583 */ 1584 #ifdef TCP_SIGNATURE 1585 if (optp || (tp->t_flags & TF_SIGNATURE)) 1586 #else 1587 if (optp) 1588 #endif 1589 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1590 goto drop; 1591 1592 if (TCP_SACK_ENABLED(tp)) { 1593 tcp_del_sackholes(tp, th); 1594 } 1595 1596 if (TCP_ECN_ALLOWED(tp)) { 1597 switch (iptos & IPTOS_ECN_MASK) { 1598 case IPTOS_ECN_CE: 1599 tp->t_flags |= TF_ECN_SND_ECE; 1600 tcpstat.tcps_ecn_ce++; 1601 break; 1602 case IPTOS_ECN_ECT0: 1603 tcpstat.tcps_ecn_ect++; 1604 break; 1605 case IPTOS_ECN_ECT1: 1606 /* XXX: ignore for now -- rpaulo */ 1607 break; 1608 } 1609 1610 if (tiflags & TH_CWR) 1611 tp->t_flags &= ~TF_ECN_SND_ECE; 1612 1613 /* 1614 * Congestion experienced. 1615 * Ignore if we are already trying to recover. 1616 */ 1617 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1618 tcp_reno_congestion_exp(tp); 1619 } 1620 1621 if (opti.ts_present && opti.ts_ecr) { 1622 /* 1623 * Calculate the RTT from the returned time stamp and the 1624 * connection's time base. If the time stamp is later than 1625 * the current time, or is extremely old, fall back to non-1323 1626 * RTT calculation. Since ts_ecr is unsigned, we can test both 1627 * at the same time. 1628 */ 1629 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1630 if (ts_rtt > TCP_PAWS_IDLE) 1631 ts_rtt = 0; 1632 } else { 1633 ts_rtt = 0; 1634 } 1635 1636 /* 1637 * Header prediction: check for the two common cases 1638 * of a uni-directional data xfer. If the packet has 1639 * no control flags, is in-sequence, the window didn't 1640 * change and we're not retransmitting, it's a 1641 * candidate. If the length is zero and the ack moved 1642 * forward, we're the sender side of the xfer. Just 1643 * free the data acked & wake any higher level process 1644 * that was blocked waiting for space. If the length 1645 * is non-zero and the ack didn't move, we're the 1646 * receiver side. If we're getting packets in-order 1647 * (the reassembly queue is empty), add the data to 1648 * the socket buffer and note that we need a delayed ack. 1649 */ 1650 if (tp->t_state == TCPS_ESTABLISHED && 1651 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1652 == TH_ACK && 1653 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1654 th->th_seq == tp->rcv_nxt && 1655 tiwin && tiwin == tp->snd_wnd && 1656 tp->snd_nxt == tp->snd_max) { 1657 1658 /* 1659 * If last ACK falls within this segment's sequence numbers, 1660 * record the timestamp. 1661 * NOTE: 1662 * 1) That the test incorporates suggestions from the latest 1663 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 1664 * 2) That updating only on newer timestamps interferes with 1665 * our earlier PAWS tests, so this check should be solely 1666 * predicated on the sequence space of this segment. 1667 * 3) That we modify the segment boundary check to be 1668 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1669 * instead of RFC1323's 1670 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1671 * This modified check allows us to overcome RFC1323's 1672 * limitations as described in Stevens TCP/IP Illustrated 1673 * Vol. 2 p.869. In such cases, we can still calculate the 1674 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1675 */ 1676 if (opti.ts_present && 1677 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1678 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1679 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1680 tp->ts_recent_age = tcp_now; 1681 tp->ts_recent = opti.ts_val; 1682 } 1683 1684 if (tlen == 0) { 1685 /* Ack prediction. */ 1686 if (SEQ_GT(th->th_ack, tp->snd_una) && 1687 SEQ_LEQ(th->th_ack, tp->snd_max) && 1688 tp->snd_cwnd >= tp->snd_wnd && 1689 tp->t_partialacks < 0) { 1690 /* 1691 * this is a pure ack for outstanding data. 1692 */ 1693 ++tcpstat.tcps_predack; 1694 if (ts_rtt) 1695 tcp_xmit_timer(tp, ts_rtt); 1696 else if (tp->t_rtttime && 1697 SEQ_GT(th->th_ack, tp->t_rtseq)) 1698 tcp_xmit_timer(tp, 1699 tcp_now - tp->t_rtttime); 1700 acked = th->th_ack - tp->snd_una; 1701 tcpstat.tcps_rcvackpack++; 1702 tcpstat.tcps_rcvackbyte += acked; 1703 ND6_HINT(tp); 1704 1705 if (acked > (tp->t_lastoff - tp->t_inoff)) 1706 tp->t_lastm = NULL; 1707 sbdrop(&so->so_snd, acked); 1708 tp->t_lastoff -= acked; 1709 1710 ICMP_CHECK(tp, th, acked); 1711 1712 tp->snd_una = th->th_ack; 1713 tp->snd_fack = tp->snd_una; 1714 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1715 tp->snd_high = tp->snd_una; 1716 m_freem(m); 1717 1718 /* 1719 * If all outstanding data are acked, stop 1720 * retransmit timer, otherwise restart timer 1721 * using current (possibly backed-off) value. 1722 * If process is waiting for space, 1723 * wakeup/selwakeup/signal. If data 1724 * are ready to send, let tcp_output 1725 * decide between more output or persist. 1726 */ 1727 if (tp->snd_una == tp->snd_max) 1728 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1729 else if (TCP_TIMER_ISARMED(tp, 1730 TCPT_PERSIST) == 0) 1731 TCP_TIMER_ARM(tp, TCPT_REXMT, 1732 tp->t_rxtcur); 1733 1734 sowwakeup(so); 1735 if (so->so_snd.sb_cc) 1736 (void) tcp_output(tp); 1737 if (tcp_saveti) 1738 m_freem(tcp_saveti); 1739 return; 1740 } 1741 } else if (th->th_ack == tp->snd_una && 1742 TAILQ_FIRST(&tp->segq) == NULL && 1743 tlen <= sbspace(&so->so_rcv)) { 1744 /* 1745 * this is a pure, in-sequence data packet 1746 * with nothing on the reassembly queue and 1747 * we have enough buffer space to take it. 1748 */ 1749 ++tcpstat.tcps_preddat; 1750 tp->rcv_nxt += tlen; 1751 tcpstat.tcps_rcvpack++; 1752 tcpstat.tcps_rcvbyte += tlen; 1753 ND6_HINT(tp); 1754 /* 1755 * Drop TCP, IP headers and TCP options then add data 1756 * to socket buffer. 1757 */ 1758 if (so->so_state & SS_CANTRCVMORE) 1759 m_freem(m); 1760 else { 1761 m_adj(m, toff + off); 1762 sbappendstream(&so->so_rcv, m); 1763 } 1764 sorwakeup(so); 1765 TCP_SETUP_ACK(tp, th); 1766 if (tp->t_flags & TF_ACKNOW) 1767 (void) tcp_output(tp); 1768 if (tcp_saveti) 1769 m_freem(tcp_saveti); 1770 return; 1771 } 1772 } 1773 1774 /* 1775 * Compute mbuf offset to TCP data segment. 1776 */ 1777 hdroptlen = toff + off; 1778 1779 /* 1780 * Calculate amount of space in receive window, 1781 * and then do TCP input processing. 1782 * Receive window is amount of space in rcv queue, 1783 * but not less than advertised window. 1784 */ 1785 { int win; 1786 1787 win = sbspace(&so->so_rcv); 1788 if (win < 0) 1789 win = 0; 1790 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1791 } 1792 1793 switch (tp->t_state) { 1794 case TCPS_LISTEN: 1795 /* 1796 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1797 */ 1798 if (m->m_flags & (M_BCAST|M_MCAST)) 1799 goto drop; 1800 switch (af) { 1801 #ifdef INET6 1802 case AF_INET6: 1803 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1804 goto drop; 1805 break; 1806 #endif /* INET6 */ 1807 case AF_INET: 1808 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1809 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1810 goto drop; 1811 break; 1812 } 1813 break; 1814 1815 /* 1816 * If the state is SYN_SENT: 1817 * if seg contains an ACK, but not for our SYN, drop the input. 1818 * if seg contains a RST, then drop the connection. 1819 * if seg does not contain SYN, then drop it. 1820 * Otherwise this is an acceptable SYN segment 1821 * initialize tp->rcv_nxt and tp->irs 1822 * if seg contains ack then advance tp->snd_una 1823 * if seg contains a ECE and ECN support is enabled, the stream 1824 * is ECN capable. 1825 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1826 * arrange for segment to be acked (eventually) 1827 * continue processing rest of data/controls, beginning with URG 1828 */ 1829 case TCPS_SYN_SENT: 1830 if ((tiflags & TH_ACK) && 1831 (SEQ_LEQ(th->th_ack, tp->iss) || 1832 SEQ_GT(th->th_ack, tp->snd_max))) 1833 goto dropwithreset; 1834 if (tiflags & TH_RST) { 1835 if (tiflags & TH_ACK) 1836 tp = tcp_drop(tp, ECONNREFUSED); 1837 goto drop; 1838 } 1839 if ((tiflags & TH_SYN) == 0) 1840 goto drop; 1841 if (tiflags & TH_ACK) { 1842 tp->snd_una = th->th_ack; 1843 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1844 tp->snd_nxt = tp->snd_una; 1845 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1846 tp->snd_high = tp->snd_una; 1847 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1848 1849 if ((tiflags & TH_ECE) && tcp_do_ecn) { 1850 tp->t_flags |= TF_ECN_PERMIT; 1851 tcpstat.tcps_ecn_shs++; 1852 } 1853 1854 } 1855 tp->irs = th->th_seq; 1856 tcp_rcvseqinit(tp); 1857 tp->t_flags |= TF_ACKNOW; 1858 tcp_mss_from_peer(tp, opti.maxseg); 1859 1860 /* 1861 * Initialize the initial congestion window. If we 1862 * had to retransmit the SYN, we must initialize cwnd 1863 * to 1 segment (i.e. the Loss Window). 1864 */ 1865 if (tp->t_flags & TF_SYN_REXMT) 1866 tp->snd_cwnd = tp->t_peermss; 1867 else { 1868 int ss = tcp_init_win; 1869 #ifdef INET 1870 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1871 ss = tcp_init_win_local; 1872 #endif 1873 #ifdef INET6 1874 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1875 ss = tcp_init_win_local; 1876 #endif 1877 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1878 } 1879 1880 tcp_rmx_rtt(tp); 1881 if (tiflags & TH_ACK) { 1882 tcpstat.tcps_connects++; 1883 soisconnected(so); 1884 tcp_established(tp); 1885 /* Do window scaling on this connection? */ 1886 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1887 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1888 tp->snd_scale = tp->requested_s_scale; 1889 tp->rcv_scale = tp->request_r_scale; 1890 } 1891 TCP_REASS_LOCK(tp); 1892 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1893 TCP_REASS_UNLOCK(tp); 1894 /* 1895 * if we didn't have to retransmit the SYN, 1896 * use its rtt as our initial srtt & rtt var. 1897 */ 1898 if (tp->t_rtttime) 1899 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1900 } else 1901 tp->t_state = TCPS_SYN_RECEIVED; 1902 1903 /* 1904 * Advance th->th_seq to correspond to first data byte. 1905 * If data, trim to stay within window, 1906 * dropping FIN if necessary. 1907 */ 1908 th->th_seq++; 1909 if (tlen > tp->rcv_wnd) { 1910 todrop = tlen - tp->rcv_wnd; 1911 m_adj(m, -todrop); 1912 tlen = tp->rcv_wnd; 1913 tiflags &= ~TH_FIN; 1914 tcpstat.tcps_rcvpackafterwin++; 1915 tcpstat.tcps_rcvbyteafterwin += todrop; 1916 } 1917 tp->snd_wl1 = th->th_seq - 1; 1918 tp->rcv_up = th->th_seq; 1919 goto step6; 1920 1921 /* 1922 * If the state is SYN_RECEIVED: 1923 * If seg contains an ACK, but not for our SYN, drop the input 1924 * and generate an RST. See page 36, rfc793 1925 */ 1926 case TCPS_SYN_RECEIVED: 1927 if ((tiflags & TH_ACK) && 1928 (SEQ_LEQ(th->th_ack, tp->iss) || 1929 SEQ_GT(th->th_ack, tp->snd_max))) 1930 goto dropwithreset; 1931 break; 1932 } 1933 1934 /* 1935 * States other than LISTEN or SYN_SENT. 1936 * First check timestamp, if present. 1937 * Then check that at least some bytes of segment are within 1938 * receive window. If segment begins before rcv_nxt, 1939 * drop leading data (and SYN); if nothing left, just ack. 1940 * 1941 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1942 * and it's less than ts_recent, drop it. 1943 */ 1944 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1945 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1946 1947 /* Check to see if ts_recent is over 24 days old. */ 1948 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 1949 /* 1950 * Invalidate ts_recent. If this segment updates 1951 * ts_recent, the age will be reset later and ts_recent 1952 * will get a valid value. If it does not, setting 1953 * ts_recent to zero will at least satisfy the 1954 * requirement that zero be placed in the timestamp 1955 * echo reply when ts_recent isn't valid. The 1956 * age isn't reset until we get a valid ts_recent 1957 * because we don't want out-of-order segments to be 1958 * dropped when ts_recent is old. 1959 */ 1960 tp->ts_recent = 0; 1961 } else { 1962 tcpstat.tcps_rcvduppack++; 1963 tcpstat.tcps_rcvdupbyte += tlen; 1964 tcpstat.tcps_pawsdrop++; 1965 tcp_new_dsack(tp, th->th_seq, tlen); 1966 goto dropafterack; 1967 } 1968 } 1969 1970 todrop = tp->rcv_nxt - th->th_seq; 1971 dupseg = FALSE; 1972 if (todrop > 0) { 1973 if (tiflags & TH_SYN) { 1974 tiflags &= ~TH_SYN; 1975 th->th_seq++; 1976 if (th->th_urp > 1) 1977 th->th_urp--; 1978 else { 1979 tiflags &= ~TH_URG; 1980 th->th_urp = 0; 1981 } 1982 todrop--; 1983 } 1984 if (todrop > tlen || 1985 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1986 /* 1987 * Any valid FIN or RST must be to the left of the 1988 * window. At this point the FIN or RST must be a 1989 * duplicate or out of sequence; drop it. 1990 */ 1991 if (tiflags & TH_RST) 1992 goto drop; 1993 tiflags &= ~(TH_FIN|TH_RST); 1994 /* 1995 * Send an ACK to resynchronize and drop any data. 1996 * But keep on processing for RST or ACK. 1997 */ 1998 tp->t_flags |= TF_ACKNOW; 1999 todrop = tlen; 2000 dupseg = TRUE; 2001 tcpstat.tcps_rcvdupbyte += todrop; 2002 tcpstat.tcps_rcvduppack++; 2003 } else if ((tiflags & TH_RST) && 2004 th->th_seq != tp->last_ack_sent) { 2005 /* 2006 * Test for reset before adjusting the sequence 2007 * number for overlapping data. 2008 */ 2009 goto dropafterack_ratelim; 2010 } else { 2011 tcpstat.tcps_rcvpartduppack++; 2012 tcpstat.tcps_rcvpartdupbyte += todrop; 2013 } 2014 tcp_new_dsack(tp, th->th_seq, todrop); 2015 hdroptlen += todrop; /*drop from head afterwards*/ 2016 th->th_seq += todrop; 2017 tlen -= todrop; 2018 if (th->th_urp > todrop) 2019 th->th_urp -= todrop; 2020 else { 2021 tiflags &= ~TH_URG; 2022 th->th_urp = 0; 2023 } 2024 } 2025 2026 /* 2027 * If new data are received on a connection after the 2028 * user processes are gone, then RST the other end. 2029 */ 2030 if ((so->so_state & SS_NOFDREF) && 2031 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2032 tp = tcp_close(tp); 2033 tcpstat.tcps_rcvafterclose++; 2034 goto dropwithreset; 2035 } 2036 2037 /* 2038 * If segment ends after window, drop trailing data 2039 * (and PUSH and FIN); if nothing left, just ACK. 2040 */ 2041 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 2042 if (todrop > 0) { 2043 tcpstat.tcps_rcvpackafterwin++; 2044 if (todrop >= tlen) { 2045 /* 2046 * The segment actually starts after the window. 2047 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2048 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2049 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2050 */ 2051 tcpstat.tcps_rcvbyteafterwin += tlen; 2052 /* 2053 * If a new connection request is received 2054 * while in TIME_WAIT, drop the old connection 2055 * and start over if the sequence numbers 2056 * are above the previous ones. 2057 * 2058 * NOTE: We will checksum the packet again, and 2059 * so we need to put the header fields back into 2060 * network order! 2061 * XXX This kind of sucks, but we don't expect 2062 * XXX this to happen very often, so maybe it 2063 * XXX doesn't matter so much. 2064 */ 2065 if (tiflags & TH_SYN && 2066 tp->t_state == TCPS_TIME_WAIT && 2067 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2068 tp = tcp_close(tp); 2069 TCP_FIELDS_TO_NET(th); 2070 goto findpcb; 2071 } 2072 /* 2073 * If window is closed can only take segments at 2074 * window edge, and have to drop data and PUSH from 2075 * incoming segments. Continue processing, but 2076 * remember to ack. Otherwise, drop segment 2077 * and (if not RST) ack. 2078 */ 2079 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2080 tp->t_flags |= TF_ACKNOW; 2081 tcpstat.tcps_rcvwinprobe++; 2082 } else 2083 goto dropafterack; 2084 } else 2085 tcpstat.tcps_rcvbyteafterwin += todrop; 2086 m_adj(m, -todrop); 2087 tlen -= todrop; 2088 tiflags &= ~(TH_PUSH|TH_FIN); 2089 } 2090 2091 /* 2092 * If last ACK falls within this segment's sequence numbers, 2093 * and the timestamp is newer, record it. 2094 */ 2095 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 2096 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2097 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 2098 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2099 tp->ts_recent_age = tcp_now; 2100 tp->ts_recent = opti.ts_val; 2101 } 2102 2103 /* 2104 * If the RST bit is set examine the state: 2105 * SYN_RECEIVED STATE: 2106 * If passive open, return to LISTEN state. 2107 * If active open, inform user that connection was refused. 2108 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 2109 * Inform user that connection was reset, and close tcb. 2110 * CLOSING, LAST_ACK, TIME_WAIT STATES 2111 * Close the tcb. 2112 */ 2113 if (tiflags & TH_RST) { 2114 if (th->th_seq != tp->last_ack_sent) 2115 goto dropafterack_ratelim; 2116 2117 switch (tp->t_state) { 2118 case TCPS_SYN_RECEIVED: 2119 so->so_error = ECONNREFUSED; 2120 goto close; 2121 2122 case TCPS_ESTABLISHED: 2123 case TCPS_FIN_WAIT_1: 2124 case TCPS_FIN_WAIT_2: 2125 case TCPS_CLOSE_WAIT: 2126 so->so_error = ECONNRESET; 2127 close: 2128 tp->t_state = TCPS_CLOSED; 2129 tcpstat.tcps_drops++; 2130 tp = tcp_close(tp); 2131 goto drop; 2132 2133 case TCPS_CLOSING: 2134 case TCPS_LAST_ACK: 2135 case TCPS_TIME_WAIT: 2136 tp = tcp_close(tp); 2137 goto drop; 2138 } 2139 } 2140 2141 /* 2142 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2143 * we must be in a synchronized state. RFC791 states (under RST 2144 * generation) that any unacceptable segment (an out-of-order SYN 2145 * qualifies) received in a synchronized state must elicit only an 2146 * empty acknowledgment segment ... and the connection remains in 2147 * the same state. 2148 */ 2149 if (tiflags & TH_SYN) { 2150 if (tp->rcv_nxt == th->th_seq) { 2151 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2152 TH_ACK); 2153 if (tcp_saveti) 2154 m_freem(tcp_saveti); 2155 return; 2156 } 2157 2158 goto dropafterack_ratelim; 2159 } 2160 2161 /* 2162 * If the ACK bit is off we drop the segment and return. 2163 */ 2164 if ((tiflags & TH_ACK) == 0) { 2165 if (tp->t_flags & TF_ACKNOW) 2166 goto dropafterack; 2167 else 2168 goto drop; 2169 } 2170 2171 /* 2172 * Ack processing. 2173 */ 2174 switch (tp->t_state) { 2175 2176 /* 2177 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2178 * ESTABLISHED state and continue processing, otherwise 2179 * send an RST. 2180 */ 2181 case TCPS_SYN_RECEIVED: 2182 if (SEQ_GT(tp->snd_una, th->th_ack) || 2183 SEQ_GT(th->th_ack, tp->snd_max)) 2184 goto dropwithreset; 2185 tcpstat.tcps_connects++; 2186 soisconnected(so); 2187 tcp_established(tp); 2188 /* Do window scaling? */ 2189 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2190 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2191 tp->snd_scale = tp->requested_s_scale; 2192 tp->rcv_scale = tp->request_r_scale; 2193 } 2194 TCP_REASS_LOCK(tp); 2195 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2196 TCP_REASS_UNLOCK(tp); 2197 tp->snd_wl1 = th->th_seq - 1; 2198 /* fall into ... */ 2199 2200 /* 2201 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2202 * ACKs. If the ack is in the range 2203 * tp->snd_una < th->th_ack <= tp->snd_max 2204 * then advance tp->snd_una to th->th_ack and drop 2205 * data from the retransmission queue. If this ACK reflects 2206 * more up to date window information we update our window information. 2207 */ 2208 case TCPS_ESTABLISHED: 2209 case TCPS_FIN_WAIT_1: 2210 case TCPS_FIN_WAIT_2: 2211 case TCPS_CLOSE_WAIT: 2212 case TCPS_CLOSING: 2213 case TCPS_LAST_ACK: 2214 case TCPS_TIME_WAIT: 2215 2216 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2217 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2218 tcpstat.tcps_rcvdupack++; 2219 /* 2220 * If we have outstanding data (other than 2221 * a window probe), this is a completely 2222 * duplicate ack (ie, window info didn't 2223 * change), the ack is the biggest we've 2224 * seen and we've seen exactly our rexmt 2225 * threshhold of them, assume a packet 2226 * has been dropped and retransmit it. 2227 * Kludge snd_nxt & the congestion 2228 * window so we send only this one 2229 * packet. 2230 * 2231 * We know we're losing at the current 2232 * window size so do congestion avoidance 2233 * (set ssthresh to half the current window 2234 * and pull our congestion window back to 2235 * the new ssthresh). 2236 * 2237 * Dup acks mean that packets have left the 2238 * network (they're now cached at the receiver) 2239 * so bump cwnd by the amount in the receiver 2240 * to keep a constant cwnd packets in the 2241 * network. 2242 * 2243 * When using TCP ECN, notify the peer that 2244 * we reduced the cwnd. 2245 * 2246 * If we are using TCP/SACK, then enter 2247 * Fast Recovery if the receiver SACKs 2248 * data that is tcprexmtthresh * MSS 2249 * bytes past the last ACKed segment, 2250 * irrespective of the number of DupAcks. 2251 */ 2252 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2253 th->th_ack != tp->snd_una) 2254 tp->t_dupacks = 0; 2255 else if (tp->t_partialacks < 0 && 2256 (++tp->t_dupacks == tcprexmtthresh || 2257 TCP_FACK_FASTRECOV(tp))) { 2258 /* 2259 * Do the fast retransmit, and adjust 2260 * congestion control paramenters. 2261 */ 2262 if (tp->t_congctl->fast_retransmit(tp, th)) { 2263 /* False fast retransmit */ 2264 break; 2265 } else 2266 goto drop; 2267 } else if (tp->t_dupacks > tcprexmtthresh) { 2268 tp->snd_cwnd += tp->t_segsz; 2269 (void) tcp_output(tp); 2270 goto drop; 2271 } 2272 } else { 2273 /* 2274 * If the ack appears to be very old, only 2275 * allow data that is in-sequence. This 2276 * makes it somewhat more difficult to insert 2277 * forged data by guessing sequence numbers. 2278 * Sent an ack to try to update the send 2279 * sequence number on the other side. 2280 */ 2281 if (tlen && th->th_seq != tp->rcv_nxt && 2282 SEQ_LT(th->th_ack, 2283 tp->snd_una - tp->max_sndwnd)) 2284 goto dropafterack; 2285 } 2286 break; 2287 } 2288 /* 2289 * If the congestion window was inflated to account 2290 * for the other side's cached packets, retract it. 2291 */ 2292 /* XXX: make SACK have his own congestion control 2293 * struct -- rpaulo */ 2294 if (TCP_SACK_ENABLED(tp)) 2295 tcp_sack_newack(tp, th); 2296 else 2297 tp->t_congctl->fast_retransmit_newack(tp, th); 2298 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2299 tcpstat.tcps_rcvacktoomuch++; 2300 goto dropafterack; 2301 } 2302 acked = th->th_ack - tp->snd_una; 2303 tcpstat.tcps_rcvackpack++; 2304 tcpstat.tcps_rcvackbyte += acked; 2305 2306 /* 2307 * If we have a timestamp reply, update smoothed 2308 * round trip time. If no timestamp is present but 2309 * transmit timer is running and timed sequence 2310 * number was acked, update smoothed round trip time. 2311 * Since we now have an rtt measurement, cancel the 2312 * timer backoff (cf., Phil Karn's retransmit alg.). 2313 * Recompute the initial retransmit timer. 2314 */ 2315 if (ts_rtt) 2316 tcp_xmit_timer(tp, ts_rtt); 2317 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2318 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2319 2320 /* 2321 * If all outstanding data is acked, stop retransmit 2322 * timer and remember to restart (more output or persist). 2323 * If there is more data to be acked, restart retransmit 2324 * timer, using current (possibly backed-off) value. 2325 */ 2326 if (th->th_ack == tp->snd_max) { 2327 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2328 needoutput = 1; 2329 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2330 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2331 2332 /* 2333 * New data has been acked, adjust the congestion window. 2334 */ 2335 tp->t_congctl->newack(tp, th); 2336 2337 ND6_HINT(tp); 2338 if (acked > so->so_snd.sb_cc) { 2339 tp->snd_wnd -= so->so_snd.sb_cc; 2340 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2341 ourfinisacked = 1; 2342 } else { 2343 if (acked > (tp->t_lastoff - tp->t_inoff)) 2344 tp->t_lastm = NULL; 2345 sbdrop(&so->so_snd, acked); 2346 tp->t_lastoff -= acked; 2347 tp->snd_wnd -= acked; 2348 ourfinisacked = 0; 2349 } 2350 sowwakeup(so); 2351 2352 ICMP_CHECK(tp, th, acked); 2353 2354 tp->snd_una = th->th_ack; 2355 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2356 tp->snd_fack = tp->snd_una; 2357 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2358 tp->snd_nxt = tp->snd_una; 2359 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2360 tp->snd_high = tp->snd_una; 2361 2362 switch (tp->t_state) { 2363 2364 /* 2365 * In FIN_WAIT_1 STATE in addition to the processing 2366 * for the ESTABLISHED state if our FIN is now acknowledged 2367 * then enter FIN_WAIT_2. 2368 */ 2369 case TCPS_FIN_WAIT_1: 2370 if (ourfinisacked) { 2371 /* 2372 * If we can't receive any more 2373 * data, then closing user can proceed. 2374 * Starting the timer is contrary to the 2375 * specification, but if we don't get a FIN 2376 * we'll hang forever. 2377 */ 2378 if (so->so_state & SS_CANTRCVMORE) { 2379 soisdisconnected(so); 2380 if (tcp_maxidle > 0) 2381 TCP_TIMER_ARM(tp, TCPT_2MSL, 2382 tcp_maxidle); 2383 } 2384 tp->t_state = TCPS_FIN_WAIT_2; 2385 } 2386 break; 2387 2388 /* 2389 * In CLOSING STATE in addition to the processing for 2390 * the ESTABLISHED state if the ACK acknowledges our FIN 2391 * then enter the TIME-WAIT state, otherwise ignore 2392 * the segment. 2393 */ 2394 case TCPS_CLOSING: 2395 if (ourfinisacked) { 2396 tp->t_state = TCPS_TIME_WAIT; 2397 tcp_canceltimers(tp); 2398 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2399 soisdisconnected(so); 2400 } 2401 break; 2402 2403 /* 2404 * In LAST_ACK, we may still be waiting for data to drain 2405 * and/or to be acked, as well as for the ack of our FIN. 2406 * If our FIN is now acknowledged, delete the TCB, 2407 * enter the closed state and return. 2408 */ 2409 case TCPS_LAST_ACK: 2410 if (ourfinisacked) { 2411 tp = tcp_close(tp); 2412 goto drop; 2413 } 2414 break; 2415 2416 /* 2417 * In TIME_WAIT state the only thing that should arrive 2418 * is a retransmission of the remote FIN. Acknowledge 2419 * it and restart the finack timer. 2420 */ 2421 case TCPS_TIME_WAIT: 2422 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2423 goto dropafterack; 2424 } 2425 } 2426 2427 step6: 2428 /* 2429 * Update window information. 2430 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2431 */ 2432 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2433 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2434 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2435 /* keep track of pure window updates */ 2436 if (tlen == 0 && 2437 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2438 tcpstat.tcps_rcvwinupd++; 2439 tp->snd_wnd = tiwin; 2440 tp->snd_wl1 = th->th_seq; 2441 tp->snd_wl2 = th->th_ack; 2442 if (tp->snd_wnd > tp->max_sndwnd) 2443 tp->max_sndwnd = tp->snd_wnd; 2444 needoutput = 1; 2445 } 2446 2447 /* 2448 * Process segments with URG. 2449 */ 2450 if ((tiflags & TH_URG) && th->th_urp && 2451 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2452 /* 2453 * This is a kludge, but if we receive and accept 2454 * random urgent pointers, we'll crash in 2455 * soreceive. It's hard to imagine someone 2456 * actually wanting to send this much urgent data. 2457 */ 2458 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2459 th->th_urp = 0; /* XXX */ 2460 tiflags &= ~TH_URG; /* XXX */ 2461 goto dodata; /* XXX */ 2462 } 2463 /* 2464 * If this segment advances the known urgent pointer, 2465 * then mark the data stream. This should not happen 2466 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2467 * a FIN has been received from the remote side. 2468 * In these states we ignore the URG. 2469 * 2470 * According to RFC961 (Assigned Protocols), 2471 * the urgent pointer points to the last octet 2472 * of urgent data. We continue, however, 2473 * to consider it to indicate the first octet 2474 * of data past the urgent section as the original 2475 * spec states (in one of two places). 2476 */ 2477 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2478 tp->rcv_up = th->th_seq + th->th_urp; 2479 so->so_oobmark = so->so_rcv.sb_cc + 2480 (tp->rcv_up - tp->rcv_nxt) - 1; 2481 if (so->so_oobmark == 0) 2482 so->so_state |= SS_RCVATMARK; 2483 sohasoutofband(so); 2484 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2485 } 2486 /* 2487 * Remove out of band data so doesn't get presented to user. 2488 * This can happen independent of advancing the URG pointer, 2489 * but if two URG's are pending at once, some out-of-band 2490 * data may creep in... ick. 2491 */ 2492 if (th->th_urp <= (u_int16_t) tlen 2493 #ifdef SO_OOBINLINE 2494 && (so->so_options & SO_OOBINLINE) == 0 2495 #endif 2496 ) 2497 tcp_pulloutofband(so, th, m, hdroptlen); 2498 } else 2499 /* 2500 * If no out of band data is expected, 2501 * pull receive urgent pointer along 2502 * with the receive window. 2503 */ 2504 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2505 tp->rcv_up = tp->rcv_nxt; 2506 dodata: /* XXX */ 2507 2508 /* 2509 * Process the segment text, merging it into the TCP sequencing queue, 2510 * and arranging for acknowledgement of receipt if necessary. 2511 * This process logically involves adjusting tp->rcv_wnd as data 2512 * is presented to the user (this happens in tcp_usrreq.c, 2513 * case PRU_RCVD). If a FIN has already been received on this 2514 * connection then we just ignore the text. 2515 */ 2516 if ((tlen || (tiflags & TH_FIN)) && 2517 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2518 /* 2519 * Insert segment ti into reassembly queue of tcp with 2520 * control block tp. Return TH_FIN if reassembly now includes 2521 * a segment with FIN. The macro form does the common case 2522 * inline (segment is the next to be received on an 2523 * established connection, and the queue is empty), 2524 * avoiding linkage into and removal from the queue and 2525 * repetition of various conversions. 2526 * Set DELACK for segments received in order, but ack 2527 * immediately when segments are out of order 2528 * (so fast retransmit can work). 2529 */ 2530 /* NOTE: this was TCP_REASS() macro, but used only once */ 2531 TCP_REASS_LOCK(tp); 2532 if (th->th_seq == tp->rcv_nxt && 2533 TAILQ_FIRST(&tp->segq) == NULL && 2534 tp->t_state == TCPS_ESTABLISHED) { 2535 TCP_SETUP_ACK(tp, th); 2536 tp->rcv_nxt += tlen; 2537 tiflags = th->th_flags & TH_FIN; 2538 tcpstat.tcps_rcvpack++; 2539 tcpstat.tcps_rcvbyte += tlen; 2540 ND6_HINT(tp); 2541 if (so->so_state & SS_CANTRCVMORE) 2542 m_freem(m); 2543 else { 2544 m_adj(m, hdroptlen); 2545 sbappendstream(&(so)->so_rcv, m); 2546 } 2547 sorwakeup(so); 2548 } else { 2549 m_adj(m, hdroptlen); 2550 tiflags = tcp_reass(tp, th, m, &tlen); 2551 tp->t_flags |= TF_ACKNOW; 2552 } 2553 TCP_REASS_UNLOCK(tp); 2554 2555 /* 2556 * Note the amount of data that peer has sent into 2557 * our window, in order to estimate the sender's 2558 * buffer size. 2559 */ 2560 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2561 } else { 2562 m_freem(m); 2563 m = NULL; 2564 tiflags &= ~TH_FIN; 2565 } 2566 2567 /* 2568 * If FIN is received ACK the FIN and let the user know 2569 * that the connection is closing. Ignore a FIN received before 2570 * the connection is fully established. 2571 */ 2572 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2573 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2574 socantrcvmore(so); 2575 tp->t_flags |= TF_ACKNOW; 2576 tp->rcv_nxt++; 2577 } 2578 switch (tp->t_state) { 2579 2580 /* 2581 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2582 */ 2583 case TCPS_ESTABLISHED: 2584 tp->t_state = TCPS_CLOSE_WAIT; 2585 break; 2586 2587 /* 2588 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2589 * enter the CLOSING state. 2590 */ 2591 case TCPS_FIN_WAIT_1: 2592 tp->t_state = TCPS_CLOSING; 2593 break; 2594 2595 /* 2596 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2597 * starting the time-wait timer, turning off the other 2598 * standard timers. 2599 */ 2600 case TCPS_FIN_WAIT_2: 2601 tp->t_state = TCPS_TIME_WAIT; 2602 tcp_canceltimers(tp); 2603 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2604 soisdisconnected(so); 2605 break; 2606 2607 /* 2608 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2609 */ 2610 case TCPS_TIME_WAIT: 2611 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2612 break; 2613 } 2614 } 2615 #ifdef TCP_DEBUG 2616 if (so->so_options & SO_DEBUG) 2617 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2618 #endif 2619 2620 /* 2621 * Return any desired output. 2622 */ 2623 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2624 (void) tcp_output(tp); 2625 } 2626 if (tcp_saveti) 2627 m_freem(tcp_saveti); 2628 return; 2629 2630 badsyn: 2631 /* 2632 * Received a bad SYN. Increment counters and dropwithreset. 2633 */ 2634 tcpstat.tcps_badsyn++; 2635 tp = NULL; 2636 goto dropwithreset; 2637 2638 dropafterack: 2639 /* 2640 * Generate an ACK dropping incoming segment if it occupies 2641 * sequence space, where the ACK reflects our state. 2642 */ 2643 if (tiflags & TH_RST) 2644 goto drop; 2645 goto dropafterack2; 2646 2647 dropafterack_ratelim: 2648 /* 2649 * We may want to rate-limit ACKs against SYN/RST attack. 2650 */ 2651 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2652 tcp_ackdrop_ppslim) == 0) { 2653 /* XXX stat */ 2654 goto drop; 2655 } 2656 /* ...fall into dropafterack2... */ 2657 2658 dropafterack2: 2659 m_freem(m); 2660 tp->t_flags |= TF_ACKNOW; 2661 (void) tcp_output(tp); 2662 if (tcp_saveti) 2663 m_freem(tcp_saveti); 2664 return; 2665 2666 dropwithreset_ratelim: 2667 /* 2668 * We may want to rate-limit RSTs in certain situations, 2669 * particularly if we are sending an RST in response to 2670 * an attempt to connect to or otherwise communicate with 2671 * a port for which we have no socket. 2672 */ 2673 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2674 tcp_rst_ppslim) == 0) { 2675 /* XXX stat */ 2676 goto drop; 2677 } 2678 /* ...fall into dropwithreset... */ 2679 2680 dropwithreset: 2681 /* 2682 * Generate a RST, dropping incoming segment. 2683 * Make ACK acceptable to originator of segment. 2684 */ 2685 if (tiflags & TH_RST) 2686 goto drop; 2687 2688 switch (af) { 2689 #ifdef INET6 2690 case AF_INET6: 2691 /* For following calls to tcp_respond */ 2692 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2693 goto drop; 2694 break; 2695 #endif /* INET6 */ 2696 case AF_INET: 2697 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2698 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2699 goto drop; 2700 } 2701 2702 if (tiflags & TH_ACK) 2703 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2704 else { 2705 if (tiflags & TH_SYN) 2706 tlen++; 2707 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2708 TH_RST|TH_ACK); 2709 } 2710 if (tcp_saveti) 2711 m_freem(tcp_saveti); 2712 return; 2713 2714 badcsum: 2715 drop: 2716 /* 2717 * Drop space held by incoming segment and return. 2718 */ 2719 if (tp) { 2720 if (tp->t_inpcb) 2721 so = tp->t_inpcb->inp_socket; 2722 #ifdef INET6 2723 else if (tp->t_in6pcb) 2724 so = tp->t_in6pcb->in6p_socket; 2725 #endif 2726 else 2727 so = NULL; 2728 #ifdef TCP_DEBUG 2729 if (so && (so->so_options & SO_DEBUG) != 0) 2730 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2731 #endif 2732 } 2733 if (tcp_saveti) 2734 m_freem(tcp_saveti); 2735 m_freem(m); 2736 return; 2737 } 2738 2739 #ifdef TCP_SIGNATURE 2740 int 2741 tcp_signature_apply(void *fstate, caddr_t data, u_int len) 2742 { 2743 2744 MD5Update(fstate, (u_char *)data, len); 2745 return (0); 2746 } 2747 2748 struct secasvar * 2749 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 2750 { 2751 struct secasvar *sav; 2752 #ifdef FAST_IPSEC 2753 union sockaddr_union dst; 2754 #endif 2755 struct ip *ip; 2756 struct ip6_hdr *ip6; 2757 2758 ip = mtod(m, struct ip *); 2759 switch (ip->ip_v) { 2760 case 4: 2761 ip = mtod(m, struct ip *); 2762 ip6 = NULL; 2763 break; 2764 case 6: 2765 ip = NULL; 2766 ip6 = mtod(m, struct ip6_hdr *); 2767 break; 2768 default: 2769 return (NULL); 2770 } 2771 2772 #ifdef FAST_IPSEC 2773 /* Extract the destination from the IP header in the mbuf. */ 2774 bzero(&dst, sizeof(union sockaddr_union)); 2775 dst.sa.sa_len = sizeof(struct sockaddr_in); 2776 dst.sa.sa_family = AF_INET; 2777 dst.sin.sin_addr = ip->ip_dst; 2778 2779 /* 2780 * Look up an SADB entry which matches the address of the peer. 2781 */ 2782 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2783 #else 2784 if (ip) 2785 sav = key_allocsa(AF_INET, (caddr_t)&ip->ip_src, 2786 (caddr_t)&ip->ip_dst, IPPROTO_TCP, 2787 htonl(TCP_SIG_SPI), 0, 0); 2788 else 2789 sav = key_allocsa(AF_INET6, (caddr_t)&ip6->ip6_src, 2790 (caddr_t)&ip6->ip6_dst, IPPROTO_TCP, 2791 htonl(TCP_SIG_SPI), 0, 0); 2792 #endif 2793 2794 return (sav); /* freesav must be performed by caller */ 2795 } 2796 2797 int 2798 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 2799 struct secasvar *sav, char *sig) 2800 { 2801 MD5_CTX ctx; 2802 struct ip *ip; 2803 struct ipovly *ipovly; 2804 struct ip6_hdr *ip6; 2805 struct ippseudo ippseudo; 2806 struct ip6_hdr_pseudo ip6pseudo; 2807 struct tcphdr th0; 2808 int l, tcphdrlen; 2809 2810 if (sav == NULL) 2811 return (-1); 2812 2813 tcphdrlen = th->th_off * 4; 2814 2815 switch (mtod(m, struct ip *)->ip_v) { 2816 case 4: 2817 ip = mtod(m, struct ip *); 2818 ip6 = NULL; 2819 break; 2820 case 6: 2821 ip = NULL; 2822 ip6 = mtod(m, struct ip6_hdr *); 2823 break; 2824 default: 2825 return (-1); 2826 } 2827 2828 MD5Init(&ctx); 2829 2830 if (ip) { 2831 memset(&ippseudo, 0, sizeof(ippseudo)); 2832 ipovly = (struct ipovly *)ip; 2833 ippseudo.ippseudo_src = ipovly->ih_src; 2834 ippseudo.ippseudo_dst = ipovly->ih_dst; 2835 ippseudo.ippseudo_pad = 0; 2836 ippseudo.ippseudo_p = IPPROTO_TCP; 2837 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 2838 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 2839 } else { 2840 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 2841 ip6pseudo.ip6ph_src = ip6->ip6_src; 2842 in6_clearscope(&ip6pseudo.ip6ph_src); 2843 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 2844 in6_clearscope(&ip6pseudo.ip6ph_dst); 2845 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 2846 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 2847 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 2848 } 2849 2850 th0 = *th; 2851 th0.th_sum = 0; 2852 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 2853 2854 l = m->m_pkthdr.len - thoff - tcphdrlen; 2855 if (l > 0) 2856 m_apply(m, thoff + tcphdrlen, 2857 m->m_pkthdr.len - thoff - tcphdrlen, 2858 tcp_signature_apply, &ctx); 2859 2860 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2861 MD5Final(sig, &ctx); 2862 2863 return (0); 2864 } 2865 #endif 2866 2867 int 2868 tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcphdr *th, 2869 struct mbuf *m __unused, int toff __unused, struct tcp_opt_info *oi) 2870 { 2871 u_int16_t mss; 2872 int opt, optlen = 0; 2873 #ifdef TCP_SIGNATURE 2874 caddr_t sigp = NULL; 2875 char sigbuf[TCP_SIGLEN]; 2876 struct secasvar *sav = NULL; 2877 #endif 2878 2879 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 2880 opt = cp[0]; 2881 if (opt == TCPOPT_EOL) 2882 break; 2883 if (opt == TCPOPT_NOP) 2884 optlen = 1; 2885 else { 2886 if (cnt < 2) 2887 break; 2888 optlen = cp[1]; 2889 if (optlen < 2 || optlen > cnt) 2890 break; 2891 } 2892 switch (opt) { 2893 2894 default: 2895 continue; 2896 2897 case TCPOPT_MAXSEG: 2898 if (optlen != TCPOLEN_MAXSEG) 2899 continue; 2900 if (!(th->th_flags & TH_SYN)) 2901 continue; 2902 if (TCPS_HAVERCVDSYN(tp->t_state)) 2903 continue; 2904 bcopy(cp + 2, &mss, sizeof(mss)); 2905 oi->maxseg = ntohs(mss); 2906 break; 2907 2908 case TCPOPT_WINDOW: 2909 if (optlen != TCPOLEN_WINDOW) 2910 continue; 2911 if (!(th->th_flags & TH_SYN)) 2912 continue; 2913 if (TCPS_HAVERCVDSYN(tp->t_state)) 2914 continue; 2915 tp->t_flags |= TF_RCVD_SCALE; 2916 tp->requested_s_scale = cp[2]; 2917 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2918 #if 0 /*XXX*/ 2919 char *p; 2920 2921 if (ip) 2922 p = ntohl(ip->ip_src); 2923 #ifdef INET6 2924 else if (ip6) 2925 p = ip6_sprintf(&ip6->ip6_src); 2926 #endif 2927 else 2928 p = "(unknown)"; 2929 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2930 "assuming %d\n", 2931 tp->requested_s_scale, p, 2932 TCP_MAX_WINSHIFT); 2933 #else 2934 log(LOG_ERR, "TCP: invalid wscale %d, " 2935 "assuming %d\n", 2936 tp->requested_s_scale, 2937 TCP_MAX_WINSHIFT); 2938 #endif 2939 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2940 } 2941 break; 2942 2943 case TCPOPT_TIMESTAMP: 2944 if (optlen != TCPOLEN_TIMESTAMP) 2945 continue; 2946 oi->ts_present = 1; 2947 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2948 NTOHL(oi->ts_val); 2949 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2950 NTOHL(oi->ts_ecr); 2951 2952 if (!(th->th_flags & TH_SYN)) 2953 continue; 2954 if (TCPS_HAVERCVDSYN(tp->t_state)) 2955 continue; 2956 /* 2957 * A timestamp received in a SYN makes 2958 * it ok to send timestamp requests and replies. 2959 */ 2960 tp->t_flags |= TF_RCVD_TSTMP; 2961 tp->ts_recent = oi->ts_val; 2962 tp->ts_recent_age = tcp_now; 2963 break; 2964 2965 case TCPOPT_SACK_PERMITTED: 2966 if (optlen != TCPOLEN_SACK_PERMITTED) 2967 continue; 2968 if (!(th->th_flags & TH_SYN)) 2969 continue; 2970 if (TCPS_HAVERCVDSYN(tp->t_state)) 2971 continue; 2972 if (tcp_do_sack) { 2973 tp->t_flags |= TF_SACK_PERMIT; 2974 tp->t_flags |= TF_WILL_SACK; 2975 } 2976 break; 2977 2978 case TCPOPT_SACK: 2979 tcp_sack_option(tp, th, cp, optlen); 2980 break; 2981 #ifdef TCP_SIGNATURE 2982 case TCPOPT_SIGNATURE: 2983 if (optlen != TCPOLEN_SIGNATURE) 2984 continue; 2985 if (sigp && bcmp(sigp, cp + 2, TCP_SIGLEN)) 2986 return (-1); 2987 2988 sigp = sigbuf; 2989 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 2990 memset(cp + 2, 0, TCP_SIGLEN); 2991 tp->t_flags |= TF_SIGNATURE; 2992 break; 2993 #endif 2994 } 2995 } 2996 2997 #ifdef TCP_SIGNATURE 2998 if (tp->t_flags & TF_SIGNATURE) { 2999 3000 sav = tcp_signature_getsav(m, th); 3001 3002 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3003 return (-1); 3004 } 3005 3006 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 3007 if (sav == NULL) 3008 return (-1); 3009 #ifdef FAST_IPSEC 3010 KEY_FREESAV(&sav); 3011 #else 3012 key_freesav(sav); 3013 #endif 3014 return (-1); 3015 } 3016 3017 if (sigp) { 3018 char sig[TCP_SIGLEN]; 3019 3020 TCP_FIELDS_TO_NET(th); 3021 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3022 TCP_FIELDS_TO_HOST(th); 3023 if (sav == NULL) 3024 return (-1); 3025 #ifdef FAST_IPSEC 3026 KEY_FREESAV(&sav); 3027 #else 3028 key_freesav(sav); 3029 #endif 3030 return (-1); 3031 } 3032 TCP_FIELDS_TO_HOST(th); 3033 3034 if (bcmp(sig, sigp, TCP_SIGLEN)) { 3035 tcpstat.tcps_badsig++; 3036 if (sav == NULL) 3037 return (-1); 3038 #ifdef FAST_IPSEC 3039 KEY_FREESAV(&sav); 3040 #else 3041 key_freesav(sav); 3042 #endif 3043 return (-1); 3044 } else 3045 tcpstat.tcps_goodsig++; 3046 3047 key_sa_recordxfer(sav, m); 3048 #ifdef FAST_IPSEC 3049 KEY_FREESAV(&sav); 3050 #else 3051 key_freesav(sav); 3052 #endif 3053 } 3054 #endif 3055 3056 return (0); 3057 } 3058 3059 /* 3060 * Pull out of band byte out of a segment so 3061 * it doesn't appear in the user's data queue. 3062 * It is still reflected in the segment length for 3063 * sequencing purposes. 3064 */ 3065 void 3066 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3067 struct mbuf *m, int off) 3068 { 3069 int cnt = off + th->th_urp - 1; 3070 3071 while (cnt >= 0) { 3072 if (m->m_len > cnt) { 3073 char *cp = mtod(m, caddr_t) + cnt; 3074 struct tcpcb *tp = sototcpcb(so); 3075 3076 tp->t_iobc = *cp; 3077 tp->t_oobflags |= TCPOOB_HAVEDATA; 3078 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3079 m->m_len--; 3080 return; 3081 } 3082 cnt -= m->m_len; 3083 m = m->m_next; 3084 if (m == 0) 3085 break; 3086 } 3087 panic("tcp_pulloutofband"); 3088 } 3089 3090 /* 3091 * Collect new round-trip time estimate 3092 * and update averages and current timeout. 3093 */ 3094 void 3095 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3096 { 3097 int32_t delta; 3098 3099 tcpstat.tcps_rttupdated++; 3100 if (tp->t_srtt != 0) { 3101 /* 3102 * srtt is stored as fixed point with 3 bits after the 3103 * binary point (i.e., scaled by 8). The following magic 3104 * is equivalent to the smoothing algorithm in rfc793 with 3105 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3106 * point). Adjust rtt to origin 0. 3107 */ 3108 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3109 if ((tp->t_srtt += delta) <= 0) 3110 tp->t_srtt = 1 << 2; 3111 /* 3112 * We accumulate a smoothed rtt variance (actually, a 3113 * smoothed mean difference), then set the retransmit 3114 * timer to smoothed rtt + 4 times the smoothed variance. 3115 * rttvar is stored as fixed point with 2 bits after the 3116 * binary point (scaled by 4). The following is 3117 * equivalent to rfc793 smoothing with an alpha of .75 3118 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3119 * rfc793's wired-in beta. 3120 */ 3121 if (delta < 0) 3122 delta = -delta; 3123 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3124 if ((tp->t_rttvar += delta) <= 0) 3125 tp->t_rttvar = 1 << 2; 3126 } else { 3127 /* 3128 * No rtt measurement yet - use the unsmoothed rtt. 3129 * Set the variance to half the rtt (so our first 3130 * retransmit happens at 3*rtt). 3131 */ 3132 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3133 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3134 } 3135 tp->t_rtttime = 0; 3136 tp->t_rxtshift = 0; 3137 3138 /* 3139 * the retransmit should happen at rtt + 4 * rttvar. 3140 * Because of the way we do the smoothing, srtt and rttvar 3141 * will each average +1/2 tick of bias. When we compute 3142 * the retransmit timer, we want 1/2 tick of rounding and 3143 * 1 extra tick because of +-1/2 tick uncertainty in the 3144 * firing of the timer. The bias will give us exactly the 3145 * 1.5 tick we need. But, because the bias is 3146 * statistical, we have to test that we don't drop below 3147 * the minimum feasible timer (which is 2 ticks). 3148 */ 3149 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3150 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3151 3152 /* 3153 * We received an ack for a packet that wasn't retransmitted; 3154 * it is probably safe to discard any error indications we've 3155 * received recently. This isn't quite right, but close enough 3156 * for now (a route might have failed after we sent a segment, 3157 * and the return path might not be symmetrical). 3158 */ 3159 tp->t_softerror = 0; 3160 } 3161 3162 3163 /* 3164 * TCP compressed state engine. Currently used to hold compressed 3165 * state for SYN_RECEIVED. 3166 */ 3167 3168 u_long syn_cache_count; 3169 u_int32_t syn_hash1, syn_hash2; 3170 3171 #define SYN_HASH(sa, sp, dp) \ 3172 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3173 ((u_int32_t)(sp)))^syn_hash2))) 3174 #ifndef INET6 3175 #define SYN_HASHALL(hash, src, dst) \ 3176 do { \ 3177 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3178 ((const struct sockaddr_in *)(src))->sin_port, \ 3179 ((const struct sockaddr_in *)(dst))->sin_port); \ 3180 } while (/*CONSTCOND*/ 0) 3181 #else 3182 #define SYN_HASH6(sa, sp, dp) \ 3183 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3184 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3185 & 0x7fffffff) 3186 3187 #define SYN_HASHALL(hash, src, dst) \ 3188 do { \ 3189 switch ((src)->sa_family) { \ 3190 case AF_INET: \ 3191 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3192 ((const struct sockaddr_in *)(src))->sin_port, \ 3193 ((const struct sockaddr_in *)(dst))->sin_port); \ 3194 break; \ 3195 case AF_INET6: \ 3196 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3197 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3198 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3199 break; \ 3200 default: \ 3201 hash = 0; \ 3202 } \ 3203 } while (/*CONSTCOND*/0) 3204 #endif /* INET6 */ 3205 3206 #define SYN_CACHE_RM(sc) \ 3207 do { \ 3208 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 3209 (sc), sc_bucketq); \ 3210 (sc)->sc_tp = NULL; \ 3211 LIST_REMOVE((sc), sc_tpq); \ 3212 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 3213 callout_stop(&(sc)->sc_timer); \ 3214 syn_cache_count--; \ 3215 } while (/*CONSTCOND*/0) 3216 3217 #define SYN_CACHE_PUT(sc) \ 3218 do { \ 3219 if ((sc)->sc_ipopts) \ 3220 (void) m_free((sc)->sc_ipopts); \ 3221 if ((sc)->sc_route4.ro_rt != NULL) \ 3222 RTFREE((sc)->sc_route4.ro_rt); \ 3223 if (callout_invoking(&(sc)->sc_timer)) \ 3224 (sc)->sc_flags |= SCF_DEAD; \ 3225 else \ 3226 pool_put(&syn_cache_pool, (sc)); \ 3227 } while (/*CONSTCOND*/0) 3228 3229 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL); 3230 3231 /* 3232 * We don't estimate RTT with SYNs, so each packet starts with the default 3233 * RTT and each timer step has a fixed timeout value. 3234 */ 3235 #define SYN_CACHE_TIMER_ARM(sc) \ 3236 do { \ 3237 TCPT_RANGESET((sc)->sc_rxtcur, \ 3238 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3239 TCPTV_REXMTMAX); \ 3240 callout_reset(&(sc)->sc_timer, \ 3241 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3242 } while (/*CONSTCOND*/0) 3243 3244 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3245 3246 void 3247 syn_cache_init(void) 3248 { 3249 int i; 3250 3251 /* Initialize the hash buckets. */ 3252 for (i = 0; i < tcp_syn_cache_size; i++) 3253 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3254 } 3255 3256 void 3257 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3258 { 3259 struct syn_cache_head *scp; 3260 struct syn_cache *sc2; 3261 int s; 3262 3263 /* 3264 * If there are no entries in the hash table, reinitialize 3265 * the hash secrets. 3266 */ 3267 if (syn_cache_count == 0) { 3268 syn_hash1 = arc4random(); 3269 syn_hash2 = arc4random(); 3270 } 3271 3272 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3273 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3274 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3275 3276 /* 3277 * Make sure that we don't overflow the per-bucket 3278 * limit or the total cache size limit. 3279 */ 3280 s = splsoftnet(); 3281 if (scp->sch_length >= tcp_syn_bucket_limit) { 3282 tcpstat.tcps_sc_bucketoverflow++; 3283 /* 3284 * The bucket is full. Toss the oldest element in the 3285 * bucket. This will be the first entry in the bucket. 3286 */ 3287 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3288 #ifdef DIAGNOSTIC 3289 /* 3290 * This should never happen; we should always find an 3291 * entry in our bucket. 3292 */ 3293 if (sc2 == NULL) 3294 panic("syn_cache_insert: bucketoverflow: impossible"); 3295 #endif 3296 SYN_CACHE_RM(sc2); 3297 SYN_CACHE_PUT(sc2); /* calls pool_put but see spl above */ 3298 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3299 struct syn_cache_head *scp2, *sce; 3300 3301 tcpstat.tcps_sc_overflowed++; 3302 /* 3303 * The cache is full. Toss the oldest entry in the 3304 * first non-empty bucket we can find. 3305 * 3306 * XXX We would really like to toss the oldest 3307 * entry in the cache, but we hope that this 3308 * condition doesn't happen very often. 3309 */ 3310 scp2 = scp; 3311 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3312 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3313 for (++scp2; scp2 != scp; scp2++) { 3314 if (scp2 >= sce) 3315 scp2 = &tcp_syn_cache[0]; 3316 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3317 break; 3318 } 3319 #ifdef DIAGNOSTIC 3320 /* 3321 * This should never happen; we should always find a 3322 * non-empty bucket. 3323 */ 3324 if (scp2 == scp) 3325 panic("syn_cache_insert: cacheoverflow: " 3326 "impossible"); 3327 #endif 3328 } 3329 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3330 SYN_CACHE_RM(sc2); 3331 SYN_CACHE_PUT(sc2); /* calls pool_put but see spl above */ 3332 } 3333 3334 /* 3335 * Initialize the entry's timer. 3336 */ 3337 sc->sc_rxttot = 0; 3338 sc->sc_rxtshift = 0; 3339 SYN_CACHE_TIMER_ARM(sc); 3340 3341 /* Link it from tcpcb entry */ 3342 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3343 3344 /* Put it into the bucket. */ 3345 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3346 scp->sch_length++; 3347 syn_cache_count++; 3348 3349 tcpstat.tcps_sc_added++; 3350 splx(s); 3351 } 3352 3353 /* 3354 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3355 * If we have retransmitted an entry the maximum number of times, expire 3356 * that entry. 3357 */ 3358 void 3359 syn_cache_timer(void *arg) 3360 { 3361 struct syn_cache *sc = arg; 3362 int s; 3363 3364 s = splsoftnet(); 3365 callout_ack(&sc->sc_timer); 3366 3367 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3368 tcpstat.tcps_sc_delayed_free++; 3369 pool_put(&syn_cache_pool, sc); 3370 splx(s); 3371 return; 3372 } 3373 3374 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3375 /* Drop it -- too many retransmissions. */ 3376 goto dropit; 3377 } 3378 3379 /* 3380 * Compute the total amount of time this entry has 3381 * been on a queue. If this entry has been on longer 3382 * than the keep alive timer would allow, expire it. 3383 */ 3384 sc->sc_rxttot += sc->sc_rxtcur; 3385 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) 3386 goto dropit; 3387 3388 tcpstat.tcps_sc_retransmitted++; 3389 (void) syn_cache_respond(sc, NULL); 3390 3391 /* Advance the timer back-off. */ 3392 sc->sc_rxtshift++; 3393 SYN_CACHE_TIMER_ARM(sc); 3394 3395 splx(s); 3396 return; 3397 3398 dropit: 3399 tcpstat.tcps_sc_timed_out++; 3400 SYN_CACHE_RM(sc); 3401 SYN_CACHE_PUT(sc); /* calls pool_put but see spl above */ 3402 splx(s); 3403 } 3404 3405 /* 3406 * Remove syn cache created by the specified tcb entry, 3407 * because this does not make sense to keep them 3408 * (if there's no tcb entry, syn cache entry will never be used) 3409 */ 3410 void 3411 syn_cache_cleanup(struct tcpcb *tp) 3412 { 3413 struct syn_cache *sc, *nsc; 3414 int s; 3415 3416 s = splsoftnet(); 3417 3418 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3419 nsc = LIST_NEXT(sc, sc_tpq); 3420 3421 #ifdef DIAGNOSTIC 3422 if (sc->sc_tp != tp) 3423 panic("invalid sc_tp in syn_cache_cleanup"); 3424 #endif 3425 SYN_CACHE_RM(sc); 3426 SYN_CACHE_PUT(sc); /* calls pool_put but see spl above */ 3427 } 3428 /* just for safety */ 3429 LIST_INIT(&tp->t_sc); 3430 3431 splx(s); 3432 } 3433 3434 /* 3435 * Find an entry in the syn cache. 3436 */ 3437 struct syn_cache * 3438 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3439 struct syn_cache_head **headp) 3440 { 3441 struct syn_cache *sc; 3442 struct syn_cache_head *scp; 3443 u_int32_t hash; 3444 int s; 3445 3446 SYN_HASHALL(hash, src, dst); 3447 3448 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3449 *headp = scp; 3450 s = splsoftnet(); 3451 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3452 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3453 if (sc->sc_hash != hash) 3454 continue; 3455 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3456 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3457 splx(s); 3458 return (sc); 3459 } 3460 } 3461 splx(s); 3462 return (NULL); 3463 } 3464 3465 /* 3466 * This function gets called when we receive an ACK for a 3467 * socket in the LISTEN state. We look up the connection 3468 * in the syn cache, and if its there, we pull it out of 3469 * the cache and turn it into a full-blown connection in 3470 * the SYN-RECEIVED state. 3471 * 3472 * The return values may not be immediately obvious, and their effects 3473 * can be subtle, so here they are: 3474 * 3475 * NULL SYN was not found in cache; caller should drop the 3476 * packet and send an RST. 3477 * 3478 * -1 We were unable to create the new connection, and are 3479 * aborting it. An ACK,RST is being sent to the peer 3480 * (unless we got screwey sequence numbners; see below), 3481 * because the 3-way handshake has been completed. Caller 3482 * should not free the mbuf, since we may be using it. If 3483 * we are not, we will free it. 3484 * 3485 * Otherwise, the return value is a pointer to the new socket 3486 * associated with the connection. 3487 */ 3488 struct socket * 3489 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3490 struct tcphdr *th, unsigned int hlen __unused, unsigned int tlen __unused, 3491 struct socket *so, struct mbuf *m) 3492 { 3493 struct syn_cache *sc; 3494 struct syn_cache_head *scp; 3495 struct inpcb *inp = NULL; 3496 #ifdef INET6 3497 struct in6pcb *in6p = NULL; 3498 #endif 3499 struct tcpcb *tp = 0; 3500 struct mbuf *am; 3501 int s; 3502 struct socket *oso; 3503 3504 s = splsoftnet(); 3505 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3506 splx(s); 3507 return (NULL); 3508 } 3509 3510 /* 3511 * Verify the sequence and ack numbers. Try getting the correct 3512 * response again. 3513 */ 3514 if ((th->th_ack != sc->sc_iss + 1) || 3515 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3516 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3517 (void) syn_cache_respond(sc, m); 3518 splx(s); 3519 return ((struct socket *)(-1)); 3520 } 3521 3522 /* Remove this cache entry */ 3523 SYN_CACHE_RM(sc); 3524 splx(s); 3525 3526 /* 3527 * Ok, create the full blown connection, and set things up 3528 * as they would have been set up if we had created the 3529 * connection when the SYN arrived. If we can't create 3530 * the connection, abort it. 3531 */ 3532 /* 3533 * inp still has the OLD in_pcb stuff, set the 3534 * v6-related flags on the new guy, too. This is 3535 * done particularly for the case where an AF_INET6 3536 * socket is bound only to a port, and a v4 connection 3537 * comes in on that port. 3538 * we also copy the flowinfo from the original pcb 3539 * to the new one. 3540 */ 3541 oso = so; 3542 so = sonewconn(so, SS_ISCONNECTED); 3543 if (so == NULL) 3544 goto resetandabort; 3545 3546 switch (so->so_proto->pr_domain->dom_family) { 3547 #ifdef INET 3548 case AF_INET: 3549 inp = sotoinpcb(so); 3550 break; 3551 #endif 3552 #ifdef INET6 3553 case AF_INET6: 3554 in6p = sotoin6pcb(so); 3555 break; 3556 #endif 3557 } 3558 switch (src->sa_family) { 3559 #ifdef INET 3560 case AF_INET: 3561 if (inp) { 3562 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3563 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3564 inp->inp_options = ip_srcroute(); 3565 in_pcbstate(inp, INP_BOUND); 3566 if (inp->inp_options == NULL) { 3567 inp->inp_options = sc->sc_ipopts; 3568 sc->sc_ipopts = NULL; 3569 } 3570 } 3571 #ifdef INET6 3572 else if (in6p) { 3573 /* IPv4 packet to AF_INET6 socket */ 3574 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3575 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3576 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3577 &in6p->in6p_laddr.s6_addr32[3], 3578 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3579 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3580 in6totcpcb(in6p)->t_family = AF_INET; 3581 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3582 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3583 else 3584 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3585 in6_pcbstate(in6p, IN6P_BOUND); 3586 } 3587 #endif 3588 break; 3589 #endif 3590 #ifdef INET6 3591 case AF_INET6: 3592 if (in6p) { 3593 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3594 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3595 in6_pcbstate(in6p, IN6P_BOUND); 3596 } 3597 break; 3598 #endif 3599 } 3600 #ifdef INET6 3601 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3602 struct in6pcb *oin6p = sotoin6pcb(oso); 3603 /* inherit socket options from the listening socket */ 3604 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3605 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3606 m_freem(in6p->in6p_options); 3607 in6p->in6p_options = 0; 3608 } 3609 ip6_savecontrol(in6p, &in6p->in6p_options, 3610 mtod(m, struct ip6_hdr *), m); 3611 } 3612 #endif 3613 3614 #if defined(IPSEC) || defined(FAST_IPSEC) 3615 /* 3616 * we make a copy of policy, instead of sharing the policy, 3617 * for better behavior in terms of SA lookup and dead SA removal. 3618 */ 3619 if (inp) { 3620 /* copy old policy into new socket's */ 3621 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3622 printf("tcp_input: could not copy policy\n"); 3623 } 3624 #ifdef INET6 3625 else if (in6p) { 3626 /* copy old policy into new socket's */ 3627 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3628 in6p->in6p_sp)) 3629 printf("tcp_input: could not copy policy\n"); 3630 } 3631 #endif 3632 #endif 3633 3634 /* 3635 * Give the new socket our cached route reference. 3636 */ 3637 if (inp) 3638 inp->inp_route = sc->sc_route4; /* struct assignment */ 3639 #ifdef INET6 3640 else 3641 in6p->in6p_route = sc->sc_route6; 3642 #endif 3643 sc->sc_route4.ro_rt = NULL; 3644 3645 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3646 if (am == NULL) 3647 goto resetandabort; 3648 MCLAIM(am, &tcp_mowner); 3649 am->m_len = src->sa_len; 3650 bcopy(src, mtod(am, caddr_t), src->sa_len); 3651 if (inp) { 3652 if (in_pcbconnect(inp, am, NULL)) { 3653 (void) m_free(am); 3654 goto resetandabort; 3655 } 3656 } 3657 #ifdef INET6 3658 else if (in6p) { 3659 if (src->sa_family == AF_INET) { 3660 /* IPv4 packet to AF_INET6 socket */ 3661 struct sockaddr_in6 *sin6; 3662 sin6 = mtod(am, struct sockaddr_in6 *); 3663 am->m_len = sizeof(*sin6); 3664 bzero(sin6, sizeof(*sin6)); 3665 sin6->sin6_family = AF_INET6; 3666 sin6->sin6_len = sizeof(*sin6); 3667 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3668 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3669 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3670 &sin6->sin6_addr.s6_addr32[3], 3671 sizeof(sin6->sin6_addr.s6_addr32[3])); 3672 } 3673 if (in6_pcbconnect(in6p, am, NULL)) { 3674 (void) m_free(am); 3675 goto resetandabort; 3676 } 3677 } 3678 #endif 3679 else { 3680 (void) m_free(am); 3681 goto resetandabort; 3682 } 3683 (void) m_free(am); 3684 3685 if (inp) 3686 tp = intotcpcb(inp); 3687 #ifdef INET6 3688 else if (in6p) 3689 tp = in6totcpcb(in6p); 3690 #endif 3691 else 3692 tp = NULL; 3693 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3694 if (sc->sc_request_r_scale != 15) { 3695 tp->requested_s_scale = sc->sc_requested_s_scale; 3696 tp->request_r_scale = sc->sc_request_r_scale; 3697 tp->snd_scale = sc->sc_requested_s_scale; 3698 tp->rcv_scale = sc->sc_request_r_scale; 3699 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3700 } 3701 if (sc->sc_flags & SCF_TIMESTAMP) 3702 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3703 tp->ts_timebase = sc->sc_timebase; 3704 3705 tp->t_template = tcp_template(tp); 3706 if (tp->t_template == 0) { 3707 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3708 so = NULL; 3709 m_freem(m); 3710 goto abort; 3711 } 3712 3713 tp->iss = sc->sc_iss; 3714 tp->irs = sc->sc_irs; 3715 tcp_sendseqinit(tp); 3716 tcp_rcvseqinit(tp); 3717 tp->t_state = TCPS_SYN_RECEIVED; 3718 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3719 tcpstat.tcps_accepts++; 3720 3721 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 3722 tp->t_flags |= TF_WILL_SACK; 3723 3724 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 3725 tp->t_flags |= TF_ECN_PERMIT; 3726 3727 #ifdef TCP_SIGNATURE 3728 if (sc->sc_flags & SCF_SIGNATURE) 3729 tp->t_flags |= TF_SIGNATURE; 3730 #endif 3731 3732 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3733 tp->t_ourmss = sc->sc_ourmaxseg; 3734 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3735 3736 /* 3737 * Initialize the initial congestion window. If we 3738 * had to retransmit the SYN,ACK, we must initialize cwnd 3739 * to 1 segment (i.e. the Loss Window). 3740 */ 3741 if (sc->sc_rxtshift) 3742 tp->snd_cwnd = tp->t_peermss; 3743 else { 3744 int ss = tcp_init_win; 3745 #ifdef INET 3746 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3747 ss = tcp_init_win_local; 3748 #endif 3749 #ifdef INET6 3750 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3751 ss = tcp_init_win_local; 3752 #endif 3753 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3754 } 3755 3756 tcp_rmx_rtt(tp); 3757 tp->snd_wl1 = sc->sc_irs; 3758 tp->rcv_up = sc->sc_irs + 1; 3759 3760 /* 3761 * This is what whould have happened in tcp_output() when 3762 * the SYN,ACK was sent. 3763 */ 3764 tp->snd_up = tp->snd_una; 3765 tp->snd_max = tp->snd_nxt = tp->iss+1; 3766 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3767 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3768 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3769 tp->last_ack_sent = tp->rcv_nxt; 3770 tp->t_partialacks = -1; 3771 tp->t_dupacks = 0; 3772 3773 tcpstat.tcps_sc_completed++; 3774 s = splsoftnet(); 3775 SYN_CACHE_PUT(sc); 3776 splx(s); 3777 return (so); 3778 3779 resetandabort: 3780 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3781 abort: 3782 if (so != NULL) 3783 (void) soabort(so); 3784 s = splsoftnet(); 3785 SYN_CACHE_PUT(sc); 3786 splx(s); 3787 tcpstat.tcps_sc_aborted++; 3788 return ((struct socket *)(-1)); 3789 } 3790 3791 /* 3792 * This function is called when we get a RST for a 3793 * non-existent connection, so that we can see if the 3794 * connection is in the syn cache. If it is, zap it. 3795 */ 3796 3797 void 3798 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 3799 { 3800 struct syn_cache *sc; 3801 struct syn_cache_head *scp; 3802 int s = splsoftnet(); 3803 3804 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3805 splx(s); 3806 return; 3807 } 3808 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3809 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3810 splx(s); 3811 return; 3812 } 3813 SYN_CACHE_RM(sc); 3814 tcpstat.tcps_sc_reset++; 3815 SYN_CACHE_PUT(sc); /* calls pool_put but see spl above */ 3816 splx(s); 3817 } 3818 3819 void 3820 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 3821 struct tcphdr *th) 3822 { 3823 struct syn_cache *sc; 3824 struct syn_cache_head *scp; 3825 int s; 3826 3827 s = splsoftnet(); 3828 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3829 splx(s); 3830 return; 3831 } 3832 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3833 if (ntohl (th->th_seq) != sc->sc_iss) { 3834 splx(s); 3835 return; 3836 } 3837 3838 /* 3839 * If we've retransmitted 3 times and this is our second error, 3840 * we remove the entry. Otherwise, we allow it to continue on. 3841 * This prevents us from incorrectly nuking an entry during a 3842 * spurious network outage. 3843 * 3844 * See tcp_notify(). 3845 */ 3846 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3847 sc->sc_flags |= SCF_UNREACH; 3848 splx(s); 3849 return; 3850 } 3851 3852 SYN_CACHE_RM(sc); 3853 tcpstat.tcps_sc_unreach++; 3854 SYN_CACHE_PUT(sc); /* calls pool_put but see spl above */ 3855 splx(s); 3856 } 3857 3858 /* 3859 * Given a LISTEN socket and an inbound SYN request, add 3860 * this to the syn cache, and send back a segment: 3861 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3862 * to the source. 3863 * 3864 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3865 * Doing so would require that we hold onto the data and deliver it 3866 * to the application. However, if we are the target of a SYN-flood 3867 * DoS attack, an attacker could send data which would eventually 3868 * consume all available buffer space if it were ACKed. By not ACKing 3869 * the data, we avoid this DoS scenario. 3870 */ 3871 3872 int 3873 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3874 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 3875 int optlen, struct tcp_opt_info *oi) 3876 { 3877 struct tcpcb tb, *tp; 3878 long win; 3879 struct syn_cache *sc; 3880 struct syn_cache_head *scp; 3881 struct mbuf *ipopts; 3882 struct tcp_opt_info opti; 3883 int s; 3884 3885 tp = sototcpcb(so); 3886 3887 bzero(&opti, sizeof(opti)); 3888 3889 /* 3890 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3891 * 3892 * Note this check is performed in tcp_input() very early on. 3893 */ 3894 3895 /* 3896 * Initialize some local state. 3897 */ 3898 win = sbspace(&so->so_rcv); 3899 if (win > TCP_MAXWIN) 3900 win = TCP_MAXWIN; 3901 3902 switch (src->sa_family) { 3903 #ifdef INET 3904 case AF_INET: 3905 /* 3906 * Remember the IP options, if any. 3907 */ 3908 ipopts = ip_srcroute(); 3909 break; 3910 #endif 3911 default: 3912 ipopts = NULL; 3913 } 3914 3915 #ifdef TCP_SIGNATURE 3916 if (optp || (tp->t_flags & TF_SIGNATURE)) 3917 #else 3918 if (optp) 3919 #endif 3920 { 3921 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3922 #ifdef TCP_SIGNATURE 3923 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 3924 #endif 3925 tb.t_state = TCPS_LISTEN; 3926 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 3927 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 3928 return (0); 3929 } else 3930 tb.t_flags = 0; 3931 3932 /* 3933 * See if we already have an entry for this connection. 3934 * If we do, resend the SYN,ACK. We do not count this 3935 * as a retransmission (XXX though maybe we should). 3936 */ 3937 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3938 tcpstat.tcps_sc_dupesyn++; 3939 if (ipopts) { 3940 /* 3941 * If we were remembering a previous source route, 3942 * forget it and use the new one we've been given. 3943 */ 3944 if (sc->sc_ipopts) 3945 (void) m_free(sc->sc_ipopts); 3946 sc->sc_ipopts = ipopts; 3947 } 3948 sc->sc_timestamp = tb.ts_recent; 3949 if (syn_cache_respond(sc, m) == 0) { 3950 tcpstat.tcps_sndacks++; 3951 tcpstat.tcps_sndtotal++; 3952 } 3953 return (1); 3954 } 3955 3956 s = splsoftnet(); 3957 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3958 splx(s); 3959 if (sc == NULL) { 3960 if (ipopts) 3961 (void) m_free(ipopts); 3962 return (0); 3963 } 3964 3965 /* 3966 * Fill in the cache, and put the necessary IP and TCP 3967 * options into the reply. 3968 */ 3969 bzero(sc, sizeof(struct syn_cache)); 3970 callout_init(&sc->sc_timer); 3971 bcopy(src, &sc->sc_src, src->sa_len); 3972 bcopy(dst, &sc->sc_dst, dst->sa_len); 3973 sc->sc_flags = 0; 3974 sc->sc_ipopts = ipopts; 3975 sc->sc_irs = th->th_seq; 3976 switch (src->sa_family) { 3977 #ifdef INET 3978 case AF_INET: 3979 { 3980 struct sockaddr_in *srcin = (void *) src; 3981 struct sockaddr_in *dstin = (void *) dst; 3982 3983 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 3984 &srcin->sin_addr, dstin->sin_port, 3985 srcin->sin_port, sizeof(dstin->sin_addr), 0); 3986 break; 3987 } 3988 #endif /* INET */ 3989 #ifdef INET6 3990 case AF_INET6: 3991 { 3992 struct sockaddr_in6 *srcin6 = (void *) src; 3993 struct sockaddr_in6 *dstin6 = (void *) dst; 3994 3995 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 3996 &srcin6->sin6_addr, dstin6->sin6_port, 3997 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 3998 break; 3999 } 4000 #endif /* INET6 */ 4001 } 4002 sc->sc_peermaxseg = oi->maxseg; 4003 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4004 m->m_pkthdr.rcvif : NULL, 4005 sc->sc_src.sa.sa_family); 4006 sc->sc_win = win; 4007 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 4008 sc->sc_timestamp = tb.ts_recent; 4009 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4010 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4011 sc->sc_flags |= SCF_TIMESTAMP; 4012 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4013 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4014 sc->sc_requested_s_scale = tb.requested_s_scale; 4015 sc->sc_request_r_scale = 0; 4016 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4017 TCP_MAXWIN << sc->sc_request_r_scale < 4018 so->so_rcv.sb_hiwat) 4019 sc->sc_request_r_scale++; 4020 } else { 4021 sc->sc_requested_s_scale = 15; 4022 sc->sc_request_r_scale = 15; 4023 } 4024 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4025 sc->sc_flags |= SCF_SACK_PERMIT; 4026 4027 /* 4028 * ECN setup packet recieved. 4029 */ 4030 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4031 sc->sc_flags |= SCF_ECN_PERMIT; 4032 4033 #ifdef TCP_SIGNATURE 4034 if (tb.t_flags & TF_SIGNATURE) 4035 sc->sc_flags |= SCF_SIGNATURE; 4036 #endif 4037 sc->sc_tp = tp; 4038 if (syn_cache_respond(sc, m) == 0) { 4039 syn_cache_insert(sc, tp); 4040 tcpstat.tcps_sndacks++; 4041 tcpstat.tcps_sndtotal++; 4042 } else { 4043 s = splsoftnet(); 4044 SYN_CACHE_PUT(sc); 4045 splx(s); 4046 tcpstat.tcps_sc_dropped++; 4047 } 4048 return (1); 4049 } 4050 4051 int 4052 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4053 { 4054 struct route *ro; 4055 u_int8_t *optp; 4056 int optlen, error; 4057 u_int16_t tlen; 4058 struct ip *ip = NULL; 4059 #ifdef INET6 4060 struct ip6_hdr *ip6 = NULL; 4061 #endif 4062 struct tcpcb *tp = NULL; 4063 struct tcphdr *th; 4064 u_int hlen; 4065 struct socket *so; 4066 4067 switch (sc->sc_src.sa.sa_family) { 4068 case AF_INET: 4069 hlen = sizeof(struct ip); 4070 ro = &sc->sc_route4; 4071 break; 4072 #ifdef INET6 4073 case AF_INET6: 4074 hlen = sizeof(struct ip6_hdr); 4075 ro = (struct route *)&sc->sc_route6; 4076 break; 4077 #endif 4078 default: 4079 if (m) 4080 m_freem(m); 4081 return (EAFNOSUPPORT); 4082 } 4083 4084 /* Compute the size of the TCP options. */ 4085 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4086 ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) + 4087 #ifdef TCP_SIGNATURE 4088 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4089 #endif 4090 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4091 4092 tlen = hlen + sizeof(struct tcphdr) + optlen; 4093 4094 /* 4095 * Create the IP+TCP header from scratch. 4096 */ 4097 if (m) 4098 m_freem(m); 4099 #ifdef DIAGNOSTIC 4100 if (max_linkhdr + tlen > MCLBYTES) 4101 return (ENOBUFS); 4102 #endif 4103 MGETHDR(m, M_DONTWAIT, MT_DATA); 4104 if (m && tlen > MHLEN) { 4105 MCLGET(m, M_DONTWAIT); 4106 if ((m->m_flags & M_EXT) == 0) { 4107 m_freem(m); 4108 m = NULL; 4109 } 4110 } 4111 if (m == NULL) 4112 return (ENOBUFS); 4113 MCLAIM(m, &tcp_tx_mowner); 4114 4115 /* Fixup the mbuf. */ 4116 m->m_data += max_linkhdr; 4117 m->m_len = m->m_pkthdr.len = tlen; 4118 if (sc->sc_tp) { 4119 tp = sc->sc_tp; 4120 if (tp->t_inpcb) 4121 so = tp->t_inpcb->inp_socket; 4122 #ifdef INET6 4123 else if (tp->t_in6pcb) 4124 so = tp->t_in6pcb->in6p_socket; 4125 #endif 4126 else 4127 so = NULL; 4128 } else 4129 so = NULL; 4130 m->m_pkthdr.rcvif = NULL; 4131 memset(mtod(m, u_char *), 0, tlen); 4132 4133 switch (sc->sc_src.sa.sa_family) { 4134 case AF_INET: 4135 ip = mtod(m, struct ip *); 4136 ip->ip_v = 4; 4137 ip->ip_dst = sc->sc_src.sin.sin_addr; 4138 ip->ip_src = sc->sc_dst.sin.sin_addr; 4139 ip->ip_p = IPPROTO_TCP; 4140 th = (struct tcphdr *)(ip + 1); 4141 th->th_dport = sc->sc_src.sin.sin_port; 4142 th->th_sport = sc->sc_dst.sin.sin_port; 4143 break; 4144 #ifdef INET6 4145 case AF_INET6: 4146 ip6 = mtod(m, struct ip6_hdr *); 4147 ip6->ip6_vfc = IPV6_VERSION; 4148 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4149 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4150 ip6->ip6_nxt = IPPROTO_TCP; 4151 /* ip6_plen will be updated in ip6_output() */ 4152 th = (struct tcphdr *)(ip6 + 1); 4153 th->th_dport = sc->sc_src.sin6.sin6_port; 4154 th->th_sport = sc->sc_dst.sin6.sin6_port; 4155 break; 4156 #endif 4157 default: 4158 th = NULL; 4159 } 4160 4161 th->th_seq = htonl(sc->sc_iss); 4162 th->th_ack = htonl(sc->sc_irs + 1); 4163 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4164 th->th_flags = TH_SYN|TH_ACK; 4165 th->th_win = htons(sc->sc_win); 4166 /* th_sum already 0 */ 4167 /* th_urp already 0 */ 4168 4169 /* Tack on the TCP options. */ 4170 optp = (u_int8_t *)(th + 1); 4171 *optp++ = TCPOPT_MAXSEG; 4172 *optp++ = 4; 4173 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4174 *optp++ = sc->sc_ourmaxseg & 0xff; 4175 4176 if (sc->sc_request_r_scale != 15) { 4177 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4178 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4179 sc->sc_request_r_scale); 4180 optp += 4; 4181 } 4182 4183 if (sc->sc_flags & SCF_TIMESTAMP) { 4184 u_int32_t *lp = (u_int32_t *)(optp); 4185 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4186 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4187 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4188 *lp = htonl(sc->sc_timestamp); 4189 optp += TCPOLEN_TSTAMP_APPA; 4190 } 4191 4192 if (sc->sc_flags & SCF_SACK_PERMIT) { 4193 u_int8_t *p = optp; 4194 4195 /* Let the peer know that we will SACK. */ 4196 p[0] = TCPOPT_SACK_PERMITTED; 4197 p[1] = 2; 4198 p[2] = TCPOPT_NOP; 4199 p[3] = TCPOPT_NOP; 4200 optp += 4; 4201 } 4202 4203 /* 4204 * Send ECN SYN-ACK setup packet. 4205 * Routes can be asymetric, so, even if we receive a packet 4206 * with ECE and CWR set, we must not assume no one will block 4207 * the ECE packet we are about to send. 4208 */ 4209 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4210 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4211 th->th_flags |= TH_ECE; 4212 tcpstat.tcps_ecn_shs++; 4213 4214 /* 4215 * draft-ietf-tcpm-ecnsyn-00.txt 4216 * 4217 * "[...] a TCP node MAY respond to an ECN-setup 4218 * SYN packet by setting ECT in the responding 4219 * ECN-setup SYN/ACK packet, indicating to routers 4220 * that the SYN/ACK packet is ECN-Capable. 4221 * This allows a congested router along the path 4222 * to mark the packet instead of dropping the 4223 * packet as an indication of congestion." 4224 * 4225 * "[...] There can be a great benefit in setting 4226 * an ECN-capable codepoint in SYN/ACK packets [...] 4227 * Congestion is most likely to occur in 4228 * the server-to-client direction. As a result, 4229 * setting an ECN-capable codepoint in SYN/ACK 4230 * packets can reduce the occurence of three-second 4231 * retransmit timeouts resulting from the drop 4232 * of SYN/ACK packets." 4233 * 4234 * Page 4 and 6, January 2006. 4235 */ 4236 4237 switch (sc->sc_src.sa.sa_family) { 4238 #ifdef INET 4239 case AF_INET: 4240 ip->ip_tos |= IPTOS_ECN_ECT0; 4241 break; 4242 #endif 4243 #ifdef INET6 4244 case AF_INET6: 4245 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4246 break; 4247 #endif 4248 } 4249 tcpstat.tcps_ecn_ect++; 4250 } 4251 4252 #ifdef TCP_SIGNATURE 4253 if (sc->sc_flags & SCF_SIGNATURE) { 4254 struct secasvar *sav; 4255 u_int8_t *sigp; 4256 4257 sav = tcp_signature_getsav(m, th); 4258 4259 if (sav == NULL) { 4260 if (m) 4261 m_freem(m); 4262 return (EPERM); 4263 } 4264 4265 *optp++ = TCPOPT_SIGNATURE; 4266 *optp++ = TCPOLEN_SIGNATURE; 4267 sigp = optp; 4268 bzero(optp, TCP_SIGLEN); 4269 optp += TCP_SIGLEN; 4270 *optp++ = TCPOPT_NOP; 4271 *optp++ = TCPOPT_EOL; 4272 4273 (void)tcp_signature(m, th, hlen, sav, sigp); 4274 4275 key_sa_recordxfer(sav, m); 4276 #ifdef FAST_IPSEC 4277 KEY_FREESAV(&sav); 4278 #else 4279 key_freesav(sav); 4280 #endif 4281 } 4282 #endif 4283 4284 /* Compute the packet's checksum. */ 4285 switch (sc->sc_src.sa.sa_family) { 4286 case AF_INET: 4287 ip->ip_len = htons(tlen - hlen); 4288 th->th_sum = 0; 4289 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4290 break; 4291 #ifdef INET6 4292 case AF_INET6: 4293 ip6->ip6_plen = htons(tlen - hlen); 4294 th->th_sum = 0; 4295 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4296 break; 4297 #endif 4298 } 4299 4300 /* 4301 * Fill in some straggling IP bits. Note the stack expects 4302 * ip_len to be in host order, for convenience. 4303 */ 4304 switch (sc->sc_src.sa.sa_family) { 4305 #ifdef INET 4306 case AF_INET: 4307 ip->ip_len = htons(tlen); 4308 ip->ip_ttl = ip_defttl; 4309 /* XXX tos? */ 4310 break; 4311 #endif 4312 #ifdef INET6 4313 case AF_INET6: 4314 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4315 ip6->ip6_vfc |= IPV6_VERSION; 4316 ip6->ip6_plen = htons(tlen - hlen); 4317 /* ip6_hlim will be initialized afterwards */ 4318 /* XXX flowlabel? */ 4319 break; 4320 #endif 4321 } 4322 4323 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4324 tp = sc->sc_tp; 4325 4326 switch (sc->sc_src.sa.sa_family) { 4327 #ifdef INET 4328 case AF_INET: 4329 error = ip_output(m, sc->sc_ipopts, ro, 4330 (ip_mtudisc ? IP_MTUDISC : 0), 4331 (struct ip_moptions *)NULL, so); 4332 break; 4333 #endif 4334 #ifdef INET6 4335 case AF_INET6: 4336 ip6->ip6_hlim = in6_selecthlim(NULL, 4337 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 4338 4339 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0, 4340 (struct ip6_moptions *)0, so, NULL); 4341 break; 4342 #endif 4343 default: 4344 error = EAFNOSUPPORT; 4345 break; 4346 } 4347 return (error); 4348 } 4349
Indexes created Sun Sep 21 20:09:37 GMT 2025