tcp_input.c revision 1.234
1 /* $NetBSD: tcp_input.c,v 1.234 2005/08/11 22:25:18 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 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 * 82 * Redistribution and use in source and binary forms, with or without 83 * modification, are permitted provided that the following conditions 84 * are met: 85 * 1. Redistributions of source code must retain the above copyright 86 * notice, this list of conditions and the following disclaimer. 87 * 2. Redistributions in binary form must reproduce the above copyright 88 * notice, this list of conditions and the following disclaimer in the 89 * documentation and/or other materials provided with the distribution. 90 * 3. All advertising materials mentioning features or use of this software 91 * must display the following acknowledgement: 92 * This product includes software developed by the NetBSD 93 * Foundation, Inc. and its contributors. 94 * 4. Neither the name of The NetBSD Foundation nor the names of its 95 * contributors may be used to endorse or promote products derived 96 * from this software without specific prior written permission. 97 * 98 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 99 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 100 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 101 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 102 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 103 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 104 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 105 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 106 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 107 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 108 * POSSIBILITY OF SUCH DAMAGE. 109 */ 110 111 /* 112 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 113 * The Regents of the University of California. All rights reserved. 114 * 115 * Redistribution and use in source and binary forms, with or without 116 * modification, are permitted provided that the following conditions 117 * are met: 118 * 1. Redistributions of source code must retain the above copyright 119 * notice, this list of conditions and the following disclaimer. 120 * 2. Redistributions in binary form must reproduce the above copyright 121 * notice, this list of conditions and the following disclaimer in the 122 * documentation and/or other materials provided with the distribution. 123 * 3. Neither the name of the University nor the names of its contributors 124 * may be used to endorse or promote products derived from this software 125 * without specific prior written permission. 126 * 127 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 128 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 129 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 130 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 131 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 132 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 133 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 134 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 135 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 136 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 137 * SUCH DAMAGE. 138 * 139 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 140 */ 141 142 /* 143 * TODO list for SYN cache stuff: 144 * 145 * Find room for a "state" field, which is needed to keep a 146 * compressed state for TIME_WAIT TCBs. It's been noted already 147 * that this is fairly important for very high-volume web and 148 * mail servers, which use a large number of short-lived 149 * connections. 150 */ 151 152 #include <sys/cdefs.h> 153 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.234 2005/08/11 22:25:18 christos Exp $"); 154 155 #include "opt_inet.h" 156 #include "opt_ipsec.h" 157 #include "opt_inet_csum.h" 158 #include "opt_tcp_debug.h" 159 160 #include <sys/param.h> 161 #include <sys/systm.h> 162 #include <sys/malloc.h> 163 #include <sys/mbuf.h> 164 #include <sys/protosw.h> 165 #include <sys/socket.h> 166 #include <sys/socketvar.h> 167 #include <sys/errno.h> 168 #include <sys/syslog.h> 169 #include <sys/pool.h> 170 #include <sys/domain.h> 171 #include <sys/kernel.h> 172 #ifdef TCP_SIGNATURE 173 #include <sys/md5.h> 174 #endif 175 176 #include <net/if.h> 177 #include <net/route.h> 178 #include <net/if_types.h> 179 180 #include <netinet/in.h> 181 #include <netinet/in_systm.h> 182 #include <netinet/ip.h> 183 #include <netinet/in_pcb.h> 184 #include <netinet/in_var.h> 185 #include <netinet/ip_var.h> 186 #include <netinet/in_offload.h> 187 188 #ifdef INET6 189 #ifndef INET 190 #include <netinet/in.h> 191 #endif 192 #include <netinet/ip6.h> 193 #include <netinet6/ip6_var.h> 194 #include <netinet6/in6_pcb.h> 195 #include <netinet6/ip6_var.h> 196 #include <netinet6/in6_var.h> 197 #include <netinet/icmp6.h> 198 #include <netinet6/nd6.h> 199 #endif 200 201 #ifndef INET6 202 /* always need ip6.h for IP6_EXTHDR_GET */ 203 #include <netinet/ip6.h> 204 #endif 205 206 #include <netinet/tcp.h> 207 #include <netinet/tcp_fsm.h> 208 #include <netinet/tcp_seq.h> 209 #include <netinet/tcp_timer.h> 210 #include <netinet/tcp_var.h> 211 #include <netinet/tcpip.h> 212 #include <netinet/tcp_debug.h> 213 214 #include <machine/stdarg.h> 215 216 #ifdef IPSEC 217 #include <netinet6/ipsec.h> 218 #include <netkey/key.h> 219 #endif /*IPSEC*/ 220 #ifdef INET6 221 #include "faith.h" 222 #if defined(NFAITH) && NFAITH > 0 223 #include <net/if_faith.h> 224 #endif 225 #endif /* IPSEC */ 226 227 #ifdef FAST_IPSEC 228 #include <netipsec/ipsec.h> 229 #include <netipsec/ipsec_var.h> /* XXX ipsecstat namespace */ 230 #include <netipsec/key.h> 231 #ifdef INET6 232 #include <netipsec/ipsec6.h> 233 #endif 234 #endif /* FAST_IPSEC*/ 235 236 int tcprexmtthresh = 3; 237 int tcp_log_refused; 238 239 static int tcp_rst_ppslim_count = 0; 240 static struct timeval tcp_rst_ppslim_last; 241 static int tcp_ackdrop_ppslim_count = 0; 242 static struct timeval tcp_ackdrop_ppslim_last; 243 244 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 245 246 /* for modulo comparisons of timestamps */ 247 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 248 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 249 250 /* 251 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 252 */ 253 #ifdef INET6 254 #define ND6_HINT(tp) \ 255 do { \ 256 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 && \ 257 tp->t_in6pcb->in6p_route.ro_rt) { \ 258 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \ 259 } \ 260 } while (/*CONSTCOND*/ 0) 261 #else 262 #define ND6_HINT(tp) 263 #endif 264 265 /* 266 * Macro to compute ACK transmission behavior. Delay the ACK unless 267 * we have already delayed an ACK (must send an ACK every two segments). 268 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 269 * option is enabled. 270 */ 271 #define TCP_SETUP_ACK(tp, th) \ 272 do { \ 273 if ((tp)->t_flags & TF_DELACK || \ 274 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 275 tp->t_flags |= TF_ACKNOW; \ 276 else \ 277 TCP_SET_DELACK(tp); \ 278 } while (/*CONSTCOND*/ 0) 279 280 #define ICMP_CHECK(tp, th, acked) \ 281 do { \ 282 /* \ 283 * If we had a pending ICMP message that \ 284 * refers to data that have just been \ 285 * acknowledged, disregard the recorded ICMP \ 286 * message. \ 287 */ \ 288 if (((tp)->t_flags & TF_PMTUD_PEND) && \ 289 SEQ_GT((th)->th_ack, (tp)->t_pmtud_th_seq)) \ 290 (tp)->t_flags &= ~TF_PMTUD_PEND; \ 291 \ 292 /* \ 293 * Keep track of the largest chunk of data \ 294 * acknowledged since last PMTU update \ 295 */ \ 296 if ((tp)->t_pmtud_mss_acked < (acked)) \ 297 (tp)->t_pmtud_mss_acked = (acked); \ 298 } while (/*CONSTCOND*/ 0) 299 300 /* 301 * Convert TCP protocol fields to host order for easier processing. 302 */ 303 #define TCP_FIELDS_TO_HOST(th) \ 304 do { \ 305 NTOHL((th)->th_seq); \ 306 NTOHL((th)->th_ack); \ 307 NTOHS((th)->th_win); \ 308 NTOHS((th)->th_urp); \ 309 } while (/*CONSTCOND*/ 0) 310 311 /* 312 * ... and reverse the above. 313 */ 314 #define TCP_FIELDS_TO_NET(th) \ 315 do { \ 316 HTONL((th)->th_seq); \ 317 HTONL((th)->th_ack); \ 318 HTONS((th)->th_win); \ 319 HTONS((th)->th_urp); \ 320 } while (/*CONSTCOND*/ 0) 321 322 #ifdef TCP_CSUM_COUNTERS 323 #include <sys/device.h> 324 325 #if defined(INET) 326 extern struct evcnt tcp_hwcsum_ok; 327 extern struct evcnt tcp_hwcsum_bad; 328 extern struct evcnt tcp_hwcsum_data; 329 extern struct evcnt tcp_swcsum; 330 #endif /* defined(INET) */ 331 #if defined(INET6) 332 extern struct evcnt tcp6_hwcsum_ok; 333 extern struct evcnt tcp6_hwcsum_bad; 334 extern struct evcnt tcp6_hwcsum_data; 335 extern struct evcnt tcp6_swcsum; 336 #endif /* defined(INET6) */ 337 338 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 339 340 #else 341 342 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 343 344 #endif /* TCP_CSUM_COUNTERS */ 345 346 #ifdef TCP_REASS_COUNTERS 347 #include <sys/device.h> 348 349 extern struct evcnt tcp_reass_; 350 extern struct evcnt tcp_reass_empty; 351 extern struct evcnt tcp_reass_iteration[8]; 352 extern struct evcnt tcp_reass_prependfirst; 353 extern struct evcnt tcp_reass_prepend; 354 extern struct evcnt tcp_reass_insert; 355 extern struct evcnt tcp_reass_inserttail; 356 extern struct evcnt tcp_reass_append; 357 extern struct evcnt tcp_reass_appendtail; 358 extern struct evcnt tcp_reass_overlaptail; 359 extern struct evcnt tcp_reass_overlapfront; 360 extern struct evcnt tcp_reass_segdup; 361 extern struct evcnt tcp_reass_fragdup; 362 363 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 364 365 #else 366 367 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 368 369 #endif /* TCP_REASS_COUNTERS */ 370 371 #ifdef INET 372 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 373 #endif 374 #ifdef INET6 375 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 376 #endif 377 378 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 379 380 POOL_INIT(tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", NULL); 381 382 struct ipqent * 383 tcpipqent_alloc() 384 { 385 struct ipqent *ipqe; 386 int s; 387 388 s = splvm(); 389 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 390 splx(s); 391 392 return ipqe; 393 } 394 395 void 396 tcpipqent_free(struct ipqent *ipqe) 397 { 398 int s; 399 400 s = splvm(); 401 pool_put(&tcpipqent_pool, ipqe); 402 splx(s); 403 } 404 405 int 406 tcp_reass(struct tcpcb *tp, struct tcphdr *th, struct mbuf *m, int *tlen) 407 { 408 struct ipqent *p, *q, *nq, *tiqe = NULL; 409 struct socket *so = NULL; 410 int pkt_flags; 411 tcp_seq pkt_seq; 412 unsigned pkt_len; 413 u_long rcvpartdupbyte = 0; 414 u_long rcvoobyte; 415 #ifdef TCP_REASS_COUNTERS 416 u_int count = 0; 417 #endif 418 419 if (tp->t_inpcb) 420 so = tp->t_inpcb->inp_socket; 421 #ifdef INET6 422 else if (tp->t_in6pcb) 423 so = tp->t_in6pcb->in6p_socket; 424 #endif 425 426 TCP_REASS_LOCK_CHECK(tp); 427 428 /* 429 * Call with th==0 after become established to 430 * force pre-ESTABLISHED data up to user socket. 431 */ 432 if (th == 0) 433 goto present; 434 435 rcvoobyte = *tlen; 436 /* 437 * Copy these to local variables because the tcpiphdr 438 * gets munged while we are collapsing mbufs. 439 */ 440 pkt_seq = th->th_seq; 441 pkt_len = *tlen; 442 pkt_flags = th->th_flags; 443 444 TCP_REASS_COUNTER_INCR(&tcp_reass_); 445 446 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 447 /* 448 * When we miss a packet, the vast majority of time we get 449 * packets that follow it in order. So optimize for that. 450 */ 451 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 452 p->ipqe_len += pkt_len; 453 p->ipqe_flags |= pkt_flags; 454 m_cat(p->ipre_mlast, m); 455 TRAVERSE(p->ipre_mlast); 456 m = NULL; 457 tiqe = p; 458 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 459 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 460 goto skip_replacement; 461 } 462 /* 463 * While we're here, if the pkt is completely beyond 464 * anything we have, just insert it at the tail. 465 */ 466 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 467 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 468 goto insert_it; 469 } 470 } 471 472 q = TAILQ_FIRST(&tp->segq); 473 474 if (q != NULL) { 475 /* 476 * If this segment immediately precedes the first out-of-order 477 * block, simply slap the segment in front of it and (mostly) 478 * skip the complicated logic. 479 */ 480 if (pkt_seq + pkt_len == q->ipqe_seq) { 481 q->ipqe_seq = pkt_seq; 482 q->ipqe_len += pkt_len; 483 q->ipqe_flags |= pkt_flags; 484 m_cat(m, q->ipqe_m); 485 q->ipqe_m = m; 486 q->ipre_mlast = m; /* last mbuf may have changed */ 487 TRAVERSE(q->ipre_mlast); 488 tiqe = q; 489 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 490 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 491 goto skip_replacement; 492 } 493 } else { 494 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 495 } 496 497 /* 498 * Find a segment which begins after this one does. 499 */ 500 for (p = NULL; q != NULL; q = nq) { 501 nq = TAILQ_NEXT(q, ipqe_q); 502 #ifdef TCP_REASS_COUNTERS 503 count++; 504 #endif 505 /* 506 * If the received segment is just right after this 507 * fragment, merge the two together and then check 508 * for further overlaps. 509 */ 510 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 511 #ifdef TCPREASS_DEBUG 512 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 513 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 514 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 515 #endif 516 pkt_len += q->ipqe_len; 517 pkt_flags |= q->ipqe_flags; 518 pkt_seq = q->ipqe_seq; 519 m_cat(q->ipre_mlast, m); 520 TRAVERSE(q->ipre_mlast); 521 m = q->ipqe_m; 522 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 523 goto free_ipqe; 524 } 525 /* 526 * If the received segment is completely past this 527 * fragment, we need to go the next fragment. 528 */ 529 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 530 p = q; 531 continue; 532 } 533 /* 534 * If the fragment is past the received segment, 535 * it (or any following) can't be concatenated. 536 */ 537 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 538 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 539 break; 540 } 541 542 /* 543 * We've received all the data in this segment before. 544 * mark it as a duplicate and return. 545 */ 546 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 547 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 548 tcpstat.tcps_rcvduppack++; 549 tcpstat.tcps_rcvdupbyte += pkt_len; 550 tcp_new_dsack(tp, pkt_seq, pkt_len); 551 m_freem(m); 552 if (tiqe != NULL) { 553 tcpipqent_free(tiqe); 554 } 555 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 556 return (0); 557 } 558 /* 559 * Received segment completely overlaps this fragment 560 * so we drop the fragment (this keeps the temporal 561 * ordering of segments correct). 562 */ 563 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 564 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 565 rcvpartdupbyte += q->ipqe_len; 566 m_freem(q->ipqe_m); 567 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 568 goto free_ipqe; 569 } 570 /* 571 * RX'ed segment extends past the end of the 572 * fragment. Drop the overlapping bytes. Then 573 * merge the fragment and segment then treat as 574 * a longer received packet. 575 */ 576 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 577 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 578 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 579 #ifdef TCPREASS_DEBUG 580 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 581 tp, overlap, 582 pkt_seq, pkt_seq + pkt_len, pkt_len); 583 #endif 584 m_adj(m, overlap); 585 rcvpartdupbyte += overlap; 586 m_cat(q->ipre_mlast, m); 587 TRAVERSE(q->ipre_mlast); 588 m = q->ipqe_m; 589 pkt_seq = q->ipqe_seq; 590 pkt_len += q->ipqe_len - overlap; 591 rcvoobyte -= overlap; 592 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 593 goto free_ipqe; 594 } 595 /* 596 * RX'ed segment extends past the front of the 597 * fragment. Drop the overlapping bytes on the 598 * received packet. The packet will then be 599 * contatentated with this fragment a bit later. 600 */ 601 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 602 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 603 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 604 #ifdef TCPREASS_DEBUG 605 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 606 tp, overlap, 607 pkt_seq, pkt_seq + pkt_len, pkt_len); 608 #endif 609 m_adj(m, -overlap); 610 pkt_len -= overlap; 611 rcvpartdupbyte += overlap; 612 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 613 rcvoobyte -= overlap; 614 } 615 /* 616 * If the received segment immediates precedes this 617 * fragment then tack the fragment onto this segment 618 * and reinsert the data. 619 */ 620 if (q->ipqe_seq == pkt_seq + pkt_len) { 621 #ifdef TCPREASS_DEBUG 622 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 623 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 624 pkt_seq, pkt_seq + pkt_len, pkt_len); 625 #endif 626 pkt_len += q->ipqe_len; 627 pkt_flags |= q->ipqe_flags; 628 m_cat(m, q->ipqe_m); 629 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 630 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 631 tp->t_segqlen--; 632 KASSERT(tp->t_segqlen >= 0); 633 KASSERT(tp->t_segqlen != 0 || 634 (TAILQ_EMPTY(&tp->segq) && 635 TAILQ_EMPTY(&tp->timeq))); 636 if (tiqe == NULL) { 637 tiqe = q; 638 } else { 639 tcpipqent_free(q); 640 } 641 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 642 break; 643 } 644 /* 645 * If the fragment is before the segment, remember it. 646 * When this loop is terminated, p will contain the 647 * pointer to fragment that is right before the received 648 * segment. 649 */ 650 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 651 p = q; 652 653 continue; 654 655 /* 656 * This is a common operation. It also will allow 657 * to save doing a malloc/free in most instances. 658 */ 659 free_ipqe: 660 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 661 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 662 tp->t_segqlen--; 663 KASSERT(tp->t_segqlen >= 0); 664 KASSERT(tp->t_segqlen != 0 || 665 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 666 if (tiqe == NULL) { 667 tiqe = q; 668 } else { 669 tcpipqent_free(q); 670 } 671 } 672 673 #ifdef TCP_REASS_COUNTERS 674 if (count > 7) 675 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 676 else if (count > 0) 677 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 678 #endif 679 680 insert_it: 681 682 /* 683 * Allocate a new queue entry since the received segment did not 684 * collapse onto any other out-of-order block; thus we are allocating 685 * a new block. If it had collapsed, tiqe would not be NULL and 686 * we would be reusing it. 687 * XXX If we can't, just drop the packet. XXX 688 */ 689 if (tiqe == NULL) { 690 tiqe = tcpipqent_alloc(); 691 if (tiqe == NULL) { 692 tcpstat.tcps_rcvmemdrop++; 693 m_freem(m); 694 return (0); 695 } 696 } 697 698 /* 699 * Update the counters. 700 */ 701 tcpstat.tcps_rcvoopack++; 702 tcpstat.tcps_rcvoobyte += rcvoobyte; 703 if (rcvpartdupbyte) { 704 tcpstat.tcps_rcvpartduppack++; 705 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 706 } 707 708 /* 709 * Insert the new fragment queue entry into both queues. 710 */ 711 tiqe->ipqe_m = m; 712 tiqe->ipre_mlast = m; 713 tiqe->ipqe_seq = pkt_seq; 714 tiqe->ipqe_len = pkt_len; 715 tiqe->ipqe_flags = pkt_flags; 716 if (p == NULL) { 717 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 718 #ifdef TCPREASS_DEBUG 719 if (tiqe->ipqe_seq != tp->rcv_nxt) 720 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 721 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 722 #endif 723 } else { 724 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 725 #ifdef TCPREASS_DEBUG 726 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 727 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 728 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 729 #endif 730 } 731 tp->t_segqlen++; 732 733 skip_replacement: 734 735 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 736 737 present: 738 /* 739 * Present data to user, advancing rcv_nxt through 740 * completed sequence space. 741 */ 742 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 743 return (0); 744 q = TAILQ_FIRST(&tp->segq); 745 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 746 return (0); 747 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 748 return (0); 749 750 tp->rcv_nxt += q->ipqe_len; 751 pkt_flags = q->ipqe_flags & TH_FIN; 752 ND6_HINT(tp); 753 754 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 755 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 756 tp->t_segqlen--; 757 KASSERT(tp->t_segqlen >= 0); 758 KASSERT(tp->t_segqlen != 0 || 759 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 760 if (so->so_state & SS_CANTRCVMORE) 761 m_freem(q->ipqe_m); 762 else 763 sbappendstream(&so->so_rcv, q->ipqe_m); 764 tcpipqent_free(q); 765 sorwakeup(so); 766 return (pkt_flags); 767 } 768 769 #ifdef INET6 770 int 771 tcp6_input(struct mbuf **mp, int *offp, int proto) 772 { 773 struct mbuf *m = *mp; 774 775 /* 776 * draft-itojun-ipv6-tcp-to-anycast 777 * better place to put this in? 778 */ 779 if (m->m_flags & M_ANYCAST6) { 780 struct ip6_hdr *ip6; 781 if (m->m_len < sizeof(struct ip6_hdr)) { 782 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 783 tcpstat.tcps_rcvshort++; 784 return IPPROTO_DONE; 785 } 786 } 787 ip6 = mtod(m, struct ip6_hdr *); 788 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 789 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 790 return IPPROTO_DONE; 791 } 792 793 tcp_input(m, *offp, proto); 794 return IPPROTO_DONE; 795 } 796 #endif 797 798 #ifdef INET 799 static void 800 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 801 { 802 char src[4*sizeof "123"]; 803 char dst[4*sizeof "123"]; 804 805 if (ip) { 806 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 807 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 808 } 809 else { 810 strlcpy(src, "(unknown)", sizeof(src)); 811 strlcpy(dst, "(unknown)", sizeof(dst)); 812 } 813 log(LOG_INFO, 814 "Connection attempt to TCP %s:%d from %s:%d\n", 815 dst, ntohs(th->th_dport), 816 src, ntohs(th->th_sport)); 817 } 818 #endif 819 820 #ifdef INET6 821 static void 822 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 823 { 824 char src[INET6_ADDRSTRLEN]; 825 char dst[INET6_ADDRSTRLEN]; 826 827 if (ip6) { 828 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 829 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 830 } 831 else { 832 strlcpy(src, "(unknown v6)", sizeof(src)); 833 strlcpy(dst, "(unknown v6)", sizeof(dst)); 834 } 835 log(LOG_INFO, 836 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 837 dst, ntohs(th->th_dport), 838 src, ntohs(th->th_sport)); 839 } 840 #endif 841 842 /* 843 * Checksum extended TCP header and data. 844 */ 845 int 846 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, int toff, 847 int off, int tlen) 848 { 849 850 /* 851 * XXX it's better to record and check if this mbuf is 852 * already checked. 853 */ 854 855 switch (af) { 856 #ifdef INET 857 case AF_INET: 858 switch (m->m_pkthdr.csum_flags & 859 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 860 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 861 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 862 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 863 goto badcsum; 864 865 case M_CSUM_TCPv4|M_CSUM_DATA: { 866 u_int32_t hw_csum = m->m_pkthdr.csum_data; 867 868 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 869 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 870 const struct ip *ip = 871 mtod(m, const struct ip *); 872 873 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 874 ip->ip_dst.s_addr, 875 htons(hw_csum + tlen + off + IPPROTO_TCP)); 876 } 877 if ((hw_csum ^ 0xffff) != 0) 878 goto badcsum; 879 break; 880 } 881 882 case M_CSUM_TCPv4: 883 /* Checksum was okay. */ 884 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 885 break; 886 887 default: 888 /* 889 * Must compute it ourselves. Maybe skip checksum 890 * on loopback interfaces. 891 */ 892 if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & 893 IFF_LOOPBACK) || 894 tcp_do_loopback_cksum)) { 895 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 896 if (in4_cksum(m, IPPROTO_TCP, toff, 897 tlen + off) != 0) 898 goto badcsum; 899 } 900 break; 901 } 902 break; 903 #endif /* INET4 */ 904 905 #ifdef INET6 906 case AF_INET6: 907 switch (m->m_pkthdr.csum_flags & 908 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 909 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 910 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 911 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 912 goto badcsum; 913 914 #if 0 /* notyet */ 915 case M_CSUM_TCPv6|M_CSUM_DATA: 916 #endif 917 918 case M_CSUM_TCPv6: 919 /* Checksum was okay. */ 920 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 921 break; 922 923 default: 924 /* 925 * Must compute it ourselves. Maybe skip checksum 926 * on loopback interfaces. 927 */ 928 if (__predict_true((m->m_flags & M_LOOP) == 0 || 929 tcp_do_loopback_cksum)) { 930 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 931 if (in6_cksum(m, IPPROTO_TCP, toff, 932 tlen + off) != 0) 933 goto badcsum; 934 } 935 } 936 break; 937 #endif /* INET6 */ 938 } 939 940 return 0; 941 942 badcsum: 943 tcpstat.tcps_rcvbadsum++; 944 return -1; 945 } 946 947 /* 948 * TCP input routine, follows pages 65-76 of the 949 * protocol specification dated September, 1981 very closely. 950 */ 951 void 952 tcp_input(struct mbuf *m, ...) 953 { 954 struct tcphdr *th; 955 struct ip *ip; 956 struct inpcb *inp; 957 #ifdef INET6 958 struct ip6_hdr *ip6; 959 struct in6pcb *in6p; 960 #endif 961 u_int8_t *optp = NULL; 962 int optlen = 0; 963 int len, tlen, toff, hdroptlen = 0; 964 struct tcpcb *tp = 0; 965 int tiflags; 966 struct socket *so = NULL; 967 int todrop, dupseg, acked, ourfinisacked, needoutput = 0; 968 #ifdef TCP_DEBUG 969 short ostate = 0; 970 #endif 971 int iss = 0; 972 u_long tiwin; 973 struct tcp_opt_info opti; 974 int off, iphlen; 975 va_list ap; 976 int af; /* af on the wire */ 977 struct mbuf *tcp_saveti = NULL; 978 uint32_t ts_rtt; 979 980 MCLAIM(m, &tcp_rx_mowner); 981 va_start(ap, m); 982 toff = va_arg(ap, int); 983 (void)va_arg(ap, int); /* ignore value, advance ap */ 984 va_end(ap); 985 986 tcpstat.tcps_rcvtotal++; 987 988 bzero(&opti, sizeof(opti)); 989 opti.ts_present = 0; 990 opti.maxseg = 0; 991 992 /* 993 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 994 * 995 * TCP is, by definition, unicast, so we reject all 996 * multicast outright. 997 * 998 * Note, there are additional src/dst address checks in 999 * the AF-specific code below. 1000 */ 1001 if (m->m_flags & (M_BCAST|M_MCAST)) { 1002 /* XXX stat */ 1003 goto drop; 1004 } 1005 #ifdef INET6 1006 if (m->m_flags & M_ANYCAST6) { 1007 /* XXX stat */ 1008 goto drop; 1009 } 1010 #endif 1011 1012 /* 1013 * Get IP and TCP header. 1014 * Note: IP leaves IP header in first mbuf. 1015 */ 1016 ip = mtod(m, struct ip *); 1017 #ifdef INET6 1018 ip6 = NULL; 1019 #endif 1020 switch (ip->ip_v) { 1021 #ifdef INET 1022 case 4: 1023 af = AF_INET; 1024 iphlen = sizeof(struct ip); 1025 ip = mtod(m, struct ip *); 1026 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1027 sizeof(struct tcphdr)); 1028 if (th == NULL) { 1029 tcpstat.tcps_rcvshort++; 1030 return; 1031 } 1032 /* We do the checksum after PCB lookup... */ 1033 len = ntohs(ip->ip_len); 1034 tlen = len - toff; 1035 break; 1036 #endif 1037 #ifdef INET6 1038 case 6: 1039 ip = NULL; 1040 iphlen = sizeof(struct ip6_hdr); 1041 af = AF_INET6; 1042 ip6 = mtod(m, struct ip6_hdr *); 1043 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1044 sizeof(struct tcphdr)); 1045 if (th == NULL) { 1046 tcpstat.tcps_rcvshort++; 1047 return; 1048 } 1049 1050 /* Be proactive about malicious use of IPv4 mapped address */ 1051 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 1052 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 1053 /* XXX stat */ 1054 goto drop; 1055 } 1056 1057 /* 1058 * Be proactive about unspecified IPv6 address in source. 1059 * As we use all-zero to indicate unbounded/unconnected pcb, 1060 * unspecified IPv6 address can be used to confuse us. 1061 * 1062 * Note that packets with unspecified IPv6 destination is 1063 * already dropped in ip6_input. 1064 */ 1065 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1066 /* XXX stat */ 1067 goto drop; 1068 } 1069 1070 /* 1071 * Make sure destination address is not multicast. 1072 * Source address checked in ip6_input(). 1073 */ 1074 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1075 /* XXX stat */ 1076 goto drop; 1077 } 1078 1079 /* We do the checksum after PCB lookup... */ 1080 len = m->m_pkthdr.len; 1081 tlen = len - toff; 1082 break; 1083 #endif 1084 default: 1085 m_freem(m); 1086 return; 1087 } 1088 1089 KASSERT(TCP_HDR_ALIGNED_P(th)); 1090 1091 /* 1092 * Check that TCP offset makes sense, 1093 * pull out TCP options and adjust length. XXX 1094 */ 1095 off = th->th_off << 2; 1096 if (off < sizeof (struct tcphdr) || off > tlen) { 1097 tcpstat.tcps_rcvbadoff++; 1098 goto drop; 1099 } 1100 tlen -= off; 1101 1102 /* 1103 * tcp_input() has been modified to use tlen to mean the TCP data 1104 * length throughout the function. Other functions can use 1105 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1106 * rja 1107 */ 1108 1109 if (off > sizeof (struct tcphdr)) { 1110 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1111 if (th == NULL) { 1112 tcpstat.tcps_rcvshort++; 1113 return; 1114 } 1115 /* 1116 * NOTE: ip/ip6 will not be affected by m_pulldown() 1117 * (as they're before toff) and we don't need to update those. 1118 */ 1119 KASSERT(TCP_HDR_ALIGNED_P(th)); 1120 optlen = off - sizeof (struct tcphdr); 1121 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1122 /* 1123 * Do quick retrieval of timestamp options ("options 1124 * prediction?"). If timestamp is the only option and it's 1125 * formatted as recommended in RFC 1323 appendix A, we 1126 * quickly get the values now and not bother calling 1127 * tcp_dooptions(), etc. 1128 */ 1129 if ((optlen == TCPOLEN_TSTAMP_APPA || 1130 (optlen > TCPOLEN_TSTAMP_APPA && 1131 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1132 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1133 (th->th_flags & TH_SYN) == 0) { 1134 opti.ts_present = 1; 1135 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1136 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1137 optp = NULL; /* we've parsed the options */ 1138 } 1139 } 1140 tiflags = th->th_flags; 1141 1142 /* 1143 * Locate pcb for segment. 1144 */ 1145 findpcb: 1146 inp = NULL; 1147 #ifdef INET6 1148 in6p = NULL; 1149 #endif 1150 switch (af) { 1151 #ifdef INET 1152 case AF_INET: 1153 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1154 ip->ip_dst, th->th_dport); 1155 if (inp == 0) { 1156 ++tcpstat.tcps_pcbhashmiss; 1157 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1158 } 1159 #ifdef INET6 1160 if (inp == 0) { 1161 struct in6_addr s, d; 1162 1163 /* mapped addr case */ 1164 bzero(&s, sizeof(s)); 1165 s.s6_addr16[5] = htons(0xffff); 1166 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1167 bzero(&d, sizeof(d)); 1168 d.s6_addr16[5] = htons(0xffff); 1169 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1170 in6p = in6_pcblookup_connect(&tcbtable, &s, 1171 th->th_sport, &d, th->th_dport, 0); 1172 if (in6p == 0) { 1173 ++tcpstat.tcps_pcbhashmiss; 1174 in6p = in6_pcblookup_bind(&tcbtable, &d, 1175 th->th_dport, 0); 1176 } 1177 } 1178 #endif 1179 #ifndef INET6 1180 if (inp == 0) 1181 #else 1182 if (inp == 0 && in6p == 0) 1183 #endif 1184 { 1185 ++tcpstat.tcps_noport; 1186 if (tcp_log_refused && 1187 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1188 tcp4_log_refused(ip, th); 1189 } 1190 TCP_FIELDS_TO_HOST(th); 1191 goto dropwithreset_ratelim; 1192 } 1193 #if defined(IPSEC) || defined(FAST_IPSEC) 1194 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1195 ipsec4_in_reject(m, inp)) { 1196 ipsecstat.in_polvio++; 1197 goto drop; 1198 } 1199 #ifdef INET6 1200 else if (in6p && 1201 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1202 ipsec4_in_reject_so(m, in6p->in6p_socket)) { 1203 ipsecstat.in_polvio++; 1204 goto drop; 1205 } 1206 #endif 1207 #endif /*IPSEC*/ 1208 break; 1209 #endif /*INET*/ 1210 #ifdef INET6 1211 case AF_INET6: 1212 { 1213 int faith; 1214 1215 #if defined(NFAITH) && NFAITH > 0 1216 faith = faithprefix(&ip6->ip6_dst); 1217 #else 1218 faith = 0; 1219 #endif 1220 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1221 th->th_sport, &ip6->ip6_dst, th->th_dport, faith); 1222 if (in6p == NULL) { 1223 ++tcpstat.tcps_pcbhashmiss; 1224 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1225 th->th_dport, faith); 1226 } 1227 if (in6p == NULL) { 1228 ++tcpstat.tcps_noport; 1229 if (tcp_log_refused && 1230 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1231 tcp6_log_refused(ip6, th); 1232 } 1233 TCP_FIELDS_TO_HOST(th); 1234 goto dropwithreset_ratelim; 1235 } 1236 #if defined(IPSEC) || defined(FAST_IPSEC) 1237 if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1238 ipsec6_in_reject(m, in6p)) { 1239 ipsec6stat.in_polvio++; 1240 goto drop; 1241 } 1242 #endif /*IPSEC*/ 1243 break; 1244 } 1245 #endif 1246 } 1247 1248 /* 1249 * If the state is CLOSED (i.e., TCB does not exist) then 1250 * all data in the incoming segment is discarded. 1251 * If the TCB exists but is in CLOSED state, it is embryonic, 1252 * but should either do a listen or a connect soon. 1253 */ 1254 tp = NULL; 1255 so = NULL; 1256 if (inp) { 1257 tp = intotcpcb(inp); 1258 so = inp->inp_socket; 1259 } 1260 #ifdef INET6 1261 else if (in6p) { 1262 tp = in6totcpcb(in6p); 1263 so = in6p->in6p_socket; 1264 } 1265 #endif 1266 if (tp == 0) { 1267 TCP_FIELDS_TO_HOST(th); 1268 goto dropwithreset_ratelim; 1269 } 1270 if (tp->t_state == TCPS_CLOSED) 1271 goto drop; 1272 1273 /* 1274 * Checksum extended TCP header and data. 1275 */ 1276 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1277 goto badcsum; 1278 1279 TCP_FIELDS_TO_HOST(th); 1280 1281 /* Unscale the window into a 32-bit value. */ 1282 if ((tiflags & TH_SYN) == 0) 1283 tiwin = th->th_win << tp->snd_scale; 1284 else 1285 tiwin = th->th_win; 1286 1287 #ifdef INET6 1288 /* save packet options if user wanted */ 1289 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1290 if (in6p->in6p_options) { 1291 m_freem(in6p->in6p_options); 1292 in6p->in6p_options = 0; 1293 } 1294 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1295 } 1296 #endif 1297 1298 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1299 union syn_cache_sa src; 1300 union syn_cache_sa dst; 1301 1302 bzero(&src, sizeof(src)); 1303 bzero(&dst, sizeof(dst)); 1304 switch (af) { 1305 #ifdef INET 1306 case AF_INET: 1307 src.sin.sin_len = sizeof(struct sockaddr_in); 1308 src.sin.sin_family = AF_INET; 1309 src.sin.sin_addr = ip->ip_src; 1310 src.sin.sin_port = th->th_sport; 1311 1312 dst.sin.sin_len = sizeof(struct sockaddr_in); 1313 dst.sin.sin_family = AF_INET; 1314 dst.sin.sin_addr = ip->ip_dst; 1315 dst.sin.sin_port = th->th_dport; 1316 break; 1317 #endif 1318 #ifdef INET6 1319 case AF_INET6: 1320 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1321 src.sin6.sin6_family = AF_INET6; 1322 src.sin6.sin6_addr = ip6->ip6_src; 1323 src.sin6.sin6_port = th->th_sport; 1324 1325 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1326 dst.sin6.sin6_family = AF_INET6; 1327 dst.sin6.sin6_addr = ip6->ip6_dst; 1328 dst.sin6.sin6_port = th->th_dport; 1329 break; 1330 #endif /* INET6 */ 1331 default: 1332 goto badsyn; /*sanity*/ 1333 } 1334 1335 if (so->so_options & SO_DEBUG) { 1336 #ifdef TCP_DEBUG 1337 ostate = tp->t_state; 1338 #endif 1339 1340 tcp_saveti = NULL; 1341 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1342 goto nosave; 1343 1344 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1345 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1346 if (!tcp_saveti) 1347 goto nosave; 1348 } else { 1349 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1350 if (!tcp_saveti) 1351 goto nosave; 1352 MCLAIM(m, &tcp_mowner); 1353 tcp_saveti->m_len = iphlen; 1354 m_copydata(m, 0, iphlen, 1355 mtod(tcp_saveti, caddr_t)); 1356 } 1357 1358 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1359 m_freem(tcp_saveti); 1360 tcp_saveti = NULL; 1361 } else { 1362 tcp_saveti->m_len += sizeof(struct tcphdr); 1363 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1364 sizeof(struct tcphdr)); 1365 } 1366 nosave:; 1367 } 1368 if (so->so_options & SO_ACCEPTCONN) { 1369 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1370 if (tiflags & TH_RST) { 1371 syn_cache_reset(&src.sa, &dst.sa, th); 1372 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1373 (TH_ACK|TH_SYN)) { 1374 /* 1375 * Received a SYN,ACK. This should 1376 * never happen while we are in 1377 * LISTEN. Send an RST. 1378 */ 1379 goto badsyn; 1380 } else if (tiflags & TH_ACK) { 1381 so = syn_cache_get(&src.sa, &dst.sa, 1382 th, toff, tlen, so, m); 1383 if (so == NULL) { 1384 /* 1385 * We don't have a SYN for 1386 * this ACK; send an RST. 1387 */ 1388 goto badsyn; 1389 } else if (so == 1390 (struct socket *)(-1)) { 1391 /* 1392 * We were unable to create 1393 * the connection. If the 1394 * 3-way handshake was 1395 * completed, and RST has 1396 * been sent to the peer. 1397 * Since the mbuf might be 1398 * in use for the reply, 1399 * do not free it. 1400 */ 1401 m = NULL; 1402 } else { 1403 /* 1404 * We have created a 1405 * full-blown connection. 1406 */ 1407 tp = NULL; 1408 inp = NULL; 1409 #ifdef INET6 1410 in6p = NULL; 1411 #endif 1412 switch (so->so_proto->pr_domain->dom_family) { 1413 #ifdef INET 1414 case AF_INET: 1415 inp = sotoinpcb(so); 1416 tp = intotcpcb(inp); 1417 break; 1418 #endif 1419 #ifdef INET6 1420 case AF_INET6: 1421 in6p = sotoin6pcb(so); 1422 tp = in6totcpcb(in6p); 1423 break; 1424 #endif 1425 } 1426 if (tp == NULL) 1427 goto badsyn; /*XXX*/ 1428 tiwin <<= tp->snd_scale; 1429 goto after_listen; 1430 } 1431 } else { 1432 /* 1433 * None of RST, SYN or ACK was set. 1434 * This is an invalid packet for a 1435 * TCB in LISTEN state. Send a RST. 1436 */ 1437 goto badsyn; 1438 } 1439 } else { 1440 /* 1441 * Received a SYN. 1442 */ 1443 1444 #ifdef INET6 1445 /* 1446 * If deprecated address is forbidden, we do 1447 * not accept SYN to deprecated interface 1448 * address to prevent any new inbound 1449 * connection from getting established. 1450 * When we do not accept SYN, we send a TCP 1451 * RST, with deprecated source address (instead 1452 * of dropping it). We compromise it as it is 1453 * much better for peer to send a RST, and 1454 * RST will be the final packet for the 1455 * exchange. 1456 * 1457 * If we do not forbid deprecated addresses, we 1458 * accept the SYN packet. RFC2462 does not 1459 * suggest dropping SYN in this case. 1460 * If we decipher RFC2462 5.5.4, it says like 1461 * this: 1462 * 1. use of deprecated addr with existing 1463 * communication is okay - "SHOULD continue 1464 * to be used" 1465 * 2. use of it with new communication: 1466 * (2a) "SHOULD NOT be used if alternate 1467 * address with sufficient scope is 1468 * available" 1469 * (2b) nothing mentioned otherwise. 1470 * Here we fall into (2b) case as we have no 1471 * choice in our source address selection - we 1472 * must obey the peer. 1473 * 1474 * The wording in RFC2462 is confusing, and 1475 * there are multiple description text for 1476 * deprecated address handling - worse, they 1477 * are not exactly the same. I believe 5.5.4 1478 * is the best one, so we follow 5.5.4. 1479 */ 1480 if (af == AF_INET6 && !ip6_use_deprecated) { 1481 struct in6_ifaddr *ia6; 1482 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1483 &ip6->ip6_dst)) && 1484 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1485 tp = NULL; 1486 goto dropwithreset; 1487 } 1488 } 1489 #endif 1490 1491 #ifdef IPSEC 1492 switch (af) { 1493 #ifdef INET 1494 case AF_INET: 1495 if (ipsec4_in_reject_so(m, so)) { 1496 ipsecstat.in_polvio++; 1497 tp = NULL; 1498 goto dropwithreset; 1499 } 1500 break; 1501 #endif 1502 #ifdef INET6 1503 case AF_INET6: 1504 if (ipsec6_in_reject_so(m, so)) { 1505 ipsec6stat.in_polvio++; 1506 tp = NULL; 1507 goto dropwithreset; 1508 } 1509 break; 1510 #endif 1511 } 1512 #endif 1513 1514 /* 1515 * LISTEN socket received a SYN 1516 * from itself? This can't possibly 1517 * be valid; drop the packet. 1518 */ 1519 if (th->th_sport == th->th_dport) { 1520 int i; 1521 1522 switch (af) { 1523 #ifdef INET 1524 case AF_INET: 1525 i = in_hosteq(ip->ip_src, ip->ip_dst); 1526 break; 1527 #endif 1528 #ifdef INET6 1529 case AF_INET6: 1530 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1531 break; 1532 #endif 1533 default: 1534 i = 1; 1535 } 1536 if (i) { 1537 tcpstat.tcps_badsyn++; 1538 goto drop; 1539 } 1540 } 1541 1542 /* 1543 * SYN looks ok; create compressed TCP 1544 * state for it. 1545 */ 1546 if (so->so_qlen <= so->so_qlimit && 1547 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1548 so, m, optp, optlen, &opti)) 1549 m = NULL; 1550 } 1551 goto drop; 1552 } 1553 } 1554 1555 after_listen: 1556 #ifdef DIAGNOSTIC 1557 /* 1558 * Should not happen now that all embryonic connections 1559 * are handled with compressed state. 1560 */ 1561 if (tp->t_state == TCPS_LISTEN) 1562 panic("tcp_input: TCPS_LISTEN"); 1563 #endif 1564 1565 /* 1566 * Segment received on connection. 1567 * Reset idle time and keep-alive timer. 1568 */ 1569 tp->t_rcvtime = tcp_now; 1570 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1571 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1572 1573 /* 1574 * Process options. 1575 */ 1576 #ifdef TCP_SIGNATURE 1577 if (optp || (tp->t_flags & TF_SIGNATURE)) 1578 #else 1579 if (optp) 1580 #endif 1581 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1582 goto drop; 1583 1584 if (TCP_SACK_ENABLED(tp)) { 1585 tcp_del_sackholes(tp, th); 1586 } 1587 1588 if (opti.ts_present && opti.ts_ecr) { 1589 /* 1590 * Calculate the RTT from the returned time stamp and the 1591 * connection's time base. If the time stamp is later than 1592 * the current time, or is extremely old, fall back to non-1323 1593 * RTT calculation. Since ts_ecr is unsigned, we can test both 1594 * at the same time. 1595 */ 1596 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1597 if (ts_rtt > TCP_PAWS_IDLE) 1598 ts_rtt = 0; 1599 } else { 1600 ts_rtt = 0; 1601 } 1602 1603 /* 1604 * Header prediction: check for the two common cases 1605 * of a uni-directional data xfer. If the packet has 1606 * no control flags, is in-sequence, the window didn't 1607 * change and we're not retransmitting, it's a 1608 * candidate. If the length is zero and the ack moved 1609 * forward, we're the sender side of the xfer. Just 1610 * free the data acked & wake any higher level process 1611 * that was blocked waiting for space. If the length 1612 * is non-zero and the ack didn't move, we're the 1613 * receiver side. If we're getting packets in-order 1614 * (the reassembly queue is empty), add the data to 1615 * the socket buffer and note that we need a delayed ack. 1616 */ 1617 if (tp->t_state == TCPS_ESTABLISHED && 1618 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1619 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1620 th->th_seq == tp->rcv_nxt && 1621 tiwin && tiwin == tp->snd_wnd && 1622 tp->snd_nxt == tp->snd_max) { 1623 1624 /* 1625 * If last ACK falls within this segment's sequence numbers, 1626 * record the timestamp. 1627 * NOTE: 1628 * 1) That the test incorporates suggestions from the latest 1629 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 1630 * 2) That updating only on newer timestamps interferes with 1631 * our earlier PAWS tests, so this check should be solely 1632 * predicated on the sequence space of this segment. 1633 * 3) That we modify the segment boundary check to be 1634 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1635 * instead of RFC1323's 1636 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1637 * This modified check allows us to overcome RFC1323's 1638 * limitations as described in Stevens TCP/IP Illustrated 1639 * Vol. 2 p.869. In such cases, we can still calculate the 1640 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1641 */ 1642 if (opti.ts_present && 1643 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1644 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1645 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1646 tp->ts_recent_age = tcp_now; 1647 tp->ts_recent = opti.ts_val; 1648 } 1649 1650 if (tlen == 0) { 1651 /* Ack prediction. */ 1652 if (SEQ_GT(th->th_ack, tp->snd_una) && 1653 SEQ_LEQ(th->th_ack, tp->snd_max) && 1654 tp->snd_cwnd >= tp->snd_wnd && 1655 tp->t_partialacks < 0) { 1656 /* 1657 * this is a pure ack for outstanding data. 1658 */ 1659 ++tcpstat.tcps_predack; 1660 if (ts_rtt) 1661 tcp_xmit_timer(tp, ts_rtt); 1662 else if (tp->t_rtttime && 1663 SEQ_GT(th->th_ack, tp->t_rtseq)) 1664 tcp_xmit_timer(tp, 1665 tcp_now - tp->t_rtttime); 1666 acked = th->th_ack - tp->snd_una; 1667 tcpstat.tcps_rcvackpack++; 1668 tcpstat.tcps_rcvackbyte += acked; 1669 ND6_HINT(tp); 1670 1671 if (acked > (tp->t_lastoff - tp->t_inoff)) 1672 tp->t_lastm = NULL; 1673 sbdrop(&so->so_snd, acked); 1674 tp->t_lastoff -= acked; 1675 1676 ICMP_CHECK(tp, th, acked); 1677 1678 tp->snd_una = th->th_ack; 1679 tp->snd_fack = tp->snd_una; 1680 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1681 tp->snd_high = tp->snd_una; 1682 m_freem(m); 1683 1684 /* 1685 * If all outstanding data are acked, stop 1686 * retransmit timer, otherwise restart timer 1687 * using current (possibly backed-off) value. 1688 * If process is waiting for space, 1689 * wakeup/selwakeup/signal. If data 1690 * are ready to send, let tcp_output 1691 * decide between more output or persist. 1692 */ 1693 if (tp->snd_una == tp->snd_max) 1694 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1695 else if (TCP_TIMER_ISARMED(tp, 1696 TCPT_PERSIST) == 0) 1697 TCP_TIMER_ARM(tp, TCPT_REXMT, 1698 tp->t_rxtcur); 1699 1700 sowwakeup(so); 1701 if (so->so_snd.sb_cc) 1702 (void) tcp_output(tp); 1703 if (tcp_saveti) 1704 m_freem(tcp_saveti); 1705 return; 1706 } 1707 } else if (th->th_ack == tp->snd_una && 1708 TAILQ_FIRST(&tp->segq) == NULL && 1709 tlen <= sbspace(&so->so_rcv)) { 1710 /* 1711 * this is a pure, in-sequence data packet 1712 * with nothing on the reassembly queue and 1713 * we have enough buffer space to take it. 1714 */ 1715 ++tcpstat.tcps_preddat; 1716 tp->rcv_nxt += tlen; 1717 tcpstat.tcps_rcvpack++; 1718 tcpstat.tcps_rcvbyte += tlen; 1719 ND6_HINT(tp); 1720 /* 1721 * Drop TCP, IP headers and TCP options then add data 1722 * to socket buffer. 1723 */ 1724 if (so->so_state & SS_CANTRCVMORE) 1725 m_freem(m); 1726 else { 1727 m_adj(m, toff + off); 1728 sbappendstream(&so->so_rcv, m); 1729 } 1730 sorwakeup(so); 1731 TCP_SETUP_ACK(tp, th); 1732 if (tp->t_flags & TF_ACKNOW) 1733 (void) tcp_output(tp); 1734 if (tcp_saveti) 1735 m_freem(tcp_saveti); 1736 return; 1737 } 1738 } 1739 1740 /* 1741 * Compute mbuf offset to TCP data segment. 1742 */ 1743 hdroptlen = toff + off; 1744 1745 /* 1746 * Calculate amount of space in receive window, 1747 * and then do TCP input processing. 1748 * Receive window is amount of space in rcv queue, 1749 * but not less than advertised window. 1750 */ 1751 { int win; 1752 1753 win = sbspace(&so->so_rcv); 1754 if (win < 0) 1755 win = 0; 1756 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1757 } 1758 1759 switch (tp->t_state) { 1760 case TCPS_LISTEN: 1761 /* 1762 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1763 */ 1764 if (m->m_flags & (M_BCAST|M_MCAST)) 1765 goto drop; 1766 switch (af) { 1767 #ifdef INET6 1768 case AF_INET6: 1769 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1770 goto drop; 1771 break; 1772 #endif /* INET6 */ 1773 case AF_INET: 1774 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1775 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1776 goto drop; 1777 break; 1778 } 1779 break; 1780 1781 /* 1782 * If the state is SYN_SENT: 1783 * if seg contains an ACK, but not for our SYN, drop the input. 1784 * if seg contains a RST, then drop the connection. 1785 * if seg does not contain SYN, then drop it. 1786 * Otherwise this is an acceptable SYN segment 1787 * initialize tp->rcv_nxt and tp->irs 1788 * if seg contains ack then advance tp->snd_una 1789 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1790 * arrange for segment to be acked (eventually) 1791 * continue processing rest of data/controls, beginning with URG 1792 */ 1793 case TCPS_SYN_SENT: 1794 if ((tiflags & TH_ACK) && 1795 (SEQ_LEQ(th->th_ack, tp->iss) || 1796 SEQ_GT(th->th_ack, tp->snd_max))) 1797 goto dropwithreset; 1798 if (tiflags & TH_RST) { 1799 if (tiflags & TH_ACK) 1800 tp = tcp_drop(tp, ECONNREFUSED); 1801 goto drop; 1802 } 1803 if ((tiflags & TH_SYN) == 0) 1804 goto drop; 1805 if (tiflags & TH_ACK) { 1806 tp->snd_una = th->th_ack; 1807 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1808 tp->snd_nxt = tp->snd_una; 1809 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1810 tp->snd_high = tp->snd_una; 1811 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1812 } 1813 tp->irs = th->th_seq; 1814 tcp_rcvseqinit(tp); 1815 tp->t_flags |= TF_ACKNOW; 1816 tcp_mss_from_peer(tp, opti.maxseg); 1817 1818 /* 1819 * Initialize the initial congestion window. If we 1820 * had to retransmit the SYN, we must initialize cwnd 1821 * to 1 segment (i.e. the Loss Window). 1822 */ 1823 if (tp->t_flags & TF_SYN_REXMT) 1824 tp->snd_cwnd = tp->t_peermss; 1825 else { 1826 int ss = tcp_init_win; 1827 #ifdef INET 1828 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1829 ss = tcp_init_win_local; 1830 #endif 1831 #ifdef INET6 1832 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1833 ss = tcp_init_win_local; 1834 #endif 1835 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1836 } 1837 1838 tcp_rmx_rtt(tp); 1839 if (tiflags & TH_ACK) { 1840 tcpstat.tcps_connects++; 1841 soisconnected(so); 1842 tcp_established(tp); 1843 /* Do window scaling on this connection? */ 1844 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1845 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1846 tp->snd_scale = tp->requested_s_scale; 1847 tp->rcv_scale = tp->request_r_scale; 1848 } 1849 TCP_REASS_LOCK(tp); 1850 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1851 TCP_REASS_UNLOCK(tp); 1852 /* 1853 * if we didn't have to retransmit the SYN, 1854 * use its rtt as our initial srtt & rtt var. 1855 */ 1856 if (tp->t_rtttime) 1857 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1858 } else 1859 tp->t_state = TCPS_SYN_RECEIVED; 1860 1861 /* 1862 * Advance th->th_seq to correspond to first data byte. 1863 * If data, trim to stay within window, 1864 * dropping FIN if necessary. 1865 */ 1866 th->th_seq++; 1867 if (tlen > tp->rcv_wnd) { 1868 todrop = tlen - tp->rcv_wnd; 1869 m_adj(m, -todrop); 1870 tlen = tp->rcv_wnd; 1871 tiflags &= ~TH_FIN; 1872 tcpstat.tcps_rcvpackafterwin++; 1873 tcpstat.tcps_rcvbyteafterwin += todrop; 1874 } 1875 tp->snd_wl1 = th->th_seq - 1; 1876 tp->rcv_up = th->th_seq; 1877 goto step6; 1878 1879 /* 1880 * If the state is SYN_RECEIVED: 1881 * If seg contains an ACK, but not for our SYN, drop the input 1882 * and generate an RST. See page 36, rfc793 1883 */ 1884 case TCPS_SYN_RECEIVED: 1885 if ((tiflags & TH_ACK) && 1886 (SEQ_LEQ(th->th_ack, tp->iss) || 1887 SEQ_GT(th->th_ack, tp->snd_max))) 1888 goto dropwithreset; 1889 break; 1890 } 1891 1892 /* 1893 * States other than LISTEN or SYN_SENT. 1894 * First check timestamp, if present. 1895 * Then check that at least some bytes of segment are within 1896 * receive window. If segment begins before rcv_nxt, 1897 * drop leading data (and SYN); if nothing left, just ack. 1898 * 1899 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1900 * and it's less than ts_recent, drop it. 1901 */ 1902 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1903 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1904 1905 /* Check to see if ts_recent is over 24 days old. */ 1906 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 1907 /* 1908 * Invalidate ts_recent. If this segment updates 1909 * ts_recent, the age will be reset later and ts_recent 1910 * will get a valid value. If it does not, setting 1911 * ts_recent to zero will at least satisfy the 1912 * requirement that zero be placed in the timestamp 1913 * echo reply when ts_recent isn't valid. The 1914 * age isn't reset until we get a valid ts_recent 1915 * because we don't want out-of-order segments to be 1916 * dropped when ts_recent is old. 1917 */ 1918 tp->ts_recent = 0; 1919 } else { 1920 tcpstat.tcps_rcvduppack++; 1921 tcpstat.tcps_rcvdupbyte += tlen; 1922 tcpstat.tcps_pawsdrop++; 1923 tcp_new_dsack(tp, th->th_seq, tlen); 1924 goto dropafterack; 1925 } 1926 } 1927 1928 todrop = tp->rcv_nxt - th->th_seq; 1929 dupseg = FALSE; 1930 if (todrop > 0) { 1931 if (tiflags & TH_SYN) { 1932 tiflags &= ~TH_SYN; 1933 th->th_seq++; 1934 if (th->th_urp > 1) 1935 th->th_urp--; 1936 else { 1937 tiflags &= ~TH_URG; 1938 th->th_urp = 0; 1939 } 1940 todrop--; 1941 } 1942 if (todrop > tlen || 1943 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1944 /* 1945 * Any valid FIN or RST must be to the left of the 1946 * window. At this point the FIN or RST must be a 1947 * duplicate or out of sequence; drop it. 1948 */ 1949 if (tiflags & TH_RST) 1950 goto drop; 1951 tiflags &= ~(TH_FIN|TH_RST); 1952 /* 1953 * Send an ACK to resynchronize and drop any data. 1954 * But keep on processing for RST or ACK. 1955 */ 1956 tp->t_flags |= TF_ACKNOW; 1957 todrop = tlen; 1958 dupseg = TRUE; 1959 tcpstat.tcps_rcvdupbyte += todrop; 1960 tcpstat.tcps_rcvduppack++; 1961 } else if ((tiflags & TH_RST) && 1962 th->th_seq != tp->last_ack_sent) { 1963 /* 1964 * Test for reset before adjusting the sequence 1965 * number for overlapping data. 1966 */ 1967 goto dropafterack_ratelim; 1968 } else { 1969 tcpstat.tcps_rcvpartduppack++; 1970 tcpstat.tcps_rcvpartdupbyte += todrop; 1971 } 1972 tcp_new_dsack(tp, th->th_seq, todrop); 1973 hdroptlen += todrop; /*drop from head afterwards*/ 1974 th->th_seq += todrop; 1975 tlen -= todrop; 1976 if (th->th_urp > todrop) 1977 th->th_urp -= todrop; 1978 else { 1979 tiflags &= ~TH_URG; 1980 th->th_urp = 0; 1981 } 1982 } 1983 1984 /* 1985 * If new data are received on a connection after the 1986 * user processes are gone, then RST the other end. 1987 */ 1988 if ((so->so_state & SS_NOFDREF) && 1989 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1990 tp = tcp_close(tp); 1991 tcpstat.tcps_rcvafterclose++; 1992 goto dropwithreset; 1993 } 1994 1995 /* 1996 * If segment ends after window, drop trailing data 1997 * (and PUSH and FIN); if nothing left, just ACK. 1998 */ 1999 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 2000 if (todrop > 0) { 2001 tcpstat.tcps_rcvpackafterwin++; 2002 if (todrop >= tlen) { 2003 /* 2004 * The segment actually starts after the window. 2005 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2006 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2007 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2008 */ 2009 tcpstat.tcps_rcvbyteafterwin += tlen; 2010 /* 2011 * If a new connection request is received 2012 * while in TIME_WAIT, drop the old connection 2013 * and start over if the sequence numbers 2014 * are above the previous ones. 2015 * 2016 * NOTE: We will checksum the packet again, and 2017 * so we need to put the header fields back into 2018 * network order! 2019 * XXX This kind of sucks, but we don't expect 2020 * XXX this to happen very often, so maybe it 2021 * XXX doesn't matter so much. 2022 */ 2023 if (tiflags & TH_SYN && 2024 tp->t_state == TCPS_TIME_WAIT && 2025 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2026 iss = tcp_new_iss(tp, tp->snd_nxt); 2027 tp = tcp_close(tp); 2028 TCP_FIELDS_TO_NET(th); 2029 goto findpcb; 2030 } 2031 /* 2032 * If window is closed can only take segments at 2033 * window edge, and have to drop data and PUSH from 2034 * incoming segments. Continue processing, but 2035 * remember to ack. Otherwise, drop segment 2036 * and (if not RST) ack. 2037 */ 2038 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2039 tp->t_flags |= TF_ACKNOW; 2040 tcpstat.tcps_rcvwinprobe++; 2041 } else 2042 goto dropafterack; 2043 } else 2044 tcpstat.tcps_rcvbyteafterwin += todrop; 2045 m_adj(m, -todrop); 2046 tlen -= todrop; 2047 tiflags &= ~(TH_PUSH|TH_FIN); 2048 } 2049 2050 /* 2051 * If last ACK falls within this segment's sequence numbers, 2052 * and the timestamp is newer, record it. 2053 */ 2054 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 2055 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2056 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 2057 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2058 tp->ts_recent_age = tcp_now; 2059 tp->ts_recent = opti.ts_val; 2060 } 2061 2062 /* 2063 * If the RST bit is set examine the state: 2064 * SYN_RECEIVED STATE: 2065 * If passive open, return to LISTEN state. 2066 * If active open, inform user that connection was refused. 2067 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 2068 * Inform user that connection was reset, and close tcb. 2069 * CLOSING, LAST_ACK, TIME_WAIT STATES 2070 * Close the tcb. 2071 */ 2072 if (tiflags & TH_RST) { 2073 if (th->th_seq != tp->last_ack_sent) 2074 goto dropafterack_ratelim; 2075 2076 switch (tp->t_state) { 2077 case TCPS_SYN_RECEIVED: 2078 so->so_error = ECONNREFUSED; 2079 goto close; 2080 2081 case TCPS_ESTABLISHED: 2082 case TCPS_FIN_WAIT_1: 2083 case TCPS_FIN_WAIT_2: 2084 case TCPS_CLOSE_WAIT: 2085 so->so_error = ECONNRESET; 2086 close: 2087 tp->t_state = TCPS_CLOSED; 2088 tcpstat.tcps_drops++; 2089 tp = tcp_close(tp); 2090 goto drop; 2091 2092 case TCPS_CLOSING: 2093 case TCPS_LAST_ACK: 2094 case TCPS_TIME_WAIT: 2095 tp = tcp_close(tp); 2096 goto drop; 2097 } 2098 } 2099 2100 /* 2101 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2102 * we must be in a synchronized state. RFC791 states (under RST 2103 * generation) that any unacceptable segment (an out-of-order SYN 2104 * qualifies) received in a synchronized state must elicit only an 2105 * empty acknowledgment segment ... and the connection remains in 2106 * the same state. 2107 */ 2108 if (tiflags & TH_SYN) { 2109 if (tp->rcv_nxt == th->th_seq) { 2110 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2111 TH_ACK); 2112 if (tcp_saveti) 2113 m_freem(tcp_saveti); 2114 return; 2115 } 2116 2117 goto dropafterack_ratelim; 2118 } 2119 2120 /* 2121 * If the ACK bit is off we drop the segment and return. 2122 */ 2123 if ((tiflags & TH_ACK) == 0) { 2124 if (tp->t_flags & TF_ACKNOW) 2125 goto dropafterack; 2126 else 2127 goto drop; 2128 } 2129 2130 /* 2131 * Ack processing. 2132 */ 2133 switch (tp->t_state) { 2134 2135 /* 2136 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2137 * ESTABLISHED state and continue processing, otherwise 2138 * send an RST. 2139 */ 2140 case TCPS_SYN_RECEIVED: 2141 if (SEQ_GT(tp->snd_una, th->th_ack) || 2142 SEQ_GT(th->th_ack, tp->snd_max)) 2143 goto dropwithreset; 2144 tcpstat.tcps_connects++; 2145 soisconnected(so); 2146 tcp_established(tp); 2147 /* Do window scaling? */ 2148 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2149 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2150 tp->snd_scale = tp->requested_s_scale; 2151 tp->rcv_scale = tp->request_r_scale; 2152 } 2153 TCP_REASS_LOCK(tp); 2154 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2155 TCP_REASS_UNLOCK(tp); 2156 tp->snd_wl1 = th->th_seq - 1; 2157 /* fall into ... */ 2158 2159 /* 2160 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2161 * ACKs. If the ack is in the range 2162 * tp->snd_una < th->th_ack <= tp->snd_max 2163 * then advance tp->snd_una to th->th_ack and drop 2164 * data from the retransmission queue. If this ACK reflects 2165 * more up to date window information we update our window information. 2166 */ 2167 case TCPS_ESTABLISHED: 2168 case TCPS_FIN_WAIT_1: 2169 case TCPS_FIN_WAIT_2: 2170 case TCPS_CLOSE_WAIT: 2171 case TCPS_CLOSING: 2172 case TCPS_LAST_ACK: 2173 case TCPS_TIME_WAIT: 2174 2175 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2176 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2177 tcpstat.tcps_rcvdupack++; 2178 /* 2179 * If we have outstanding data (other than 2180 * a window probe), this is a completely 2181 * duplicate ack (ie, window info didn't 2182 * change), the ack is the biggest we've 2183 * seen and we've seen exactly our rexmt 2184 * threshhold of them, assume a packet 2185 * has been dropped and retransmit it. 2186 * Kludge snd_nxt & the congestion 2187 * window so we send only this one 2188 * packet. 2189 * 2190 * We know we're losing at the current 2191 * window size so do congestion avoidance 2192 * (set ssthresh to half the current window 2193 * and pull our congestion window back to 2194 * the new ssthresh). 2195 * 2196 * Dup acks mean that packets have left the 2197 * network (they're now cached at the receiver) 2198 * so bump cwnd by the amount in the receiver 2199 * to keep a constant cwnd packets in the 2200 * network. 2201 * 2202 * If we are using TCP/SACK, then enter 2203 * Fast Recovery if the receiver SACKs 2204 * data that is tcprexmtthresh * MSS 2205 * bytes past the last ACKed segment, 2206 * irrespective of the number of DupAcks. 2207 */ 2208 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2209 th->th_ack != tp->snd_una) 2210 tp->t_dupacks = 0; 2211 else if (tp->t_partialacks < 0 && 2212 (++tp->t_dupacks == tcprexmtthresh || 2213 TCP_FACK_FASTRECOV(tp))) { 2214 tcp_seq onxt; 2215 u_int win; 2216 2217 if (tcp_do_newreno && 2218 SEQ_LT(th->th_ack, tp->snd_high)) { 2219 /* 2220 * False fast retransmit after 2221 * timeout. Do not enter fast 2222 * recovery. 2223 */ 2224 tp->t_dupacks = 0; 2225 break; 2226 } 2227 2228 onxt = tp->snd_nxt; 2229 win = min(tp->snd_wnd, tp->snd_cwnd) / 2230 2 / tp->t_segsz; 2231 if (win < 2) 2232 win = 2; 2233 tp->snd_ssthresh = win * tp->t_segsz; 2234 tp->snd_recover = tp->snd_max; 2235 tp->t_partialacks = 0; 2236 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2237 tp->t_rtttime = 0; 2238 if (TCP_SACK_ENABLED(tp)) { 2239 tp->t_dupacks = tcprexmtthresh; 2240 tp->sack_newdata = tp->snd_nxt; 2241 tp->snd_cwnd = tp->t_segsz; 2242 (void) tcp_output(tp); 2243 goto drop; 2244 } 2245 tp->snd_nxt = th->th_ack; 2246 tp->snd_cwnd = tp->t_segsz; 2247 (void) tcp_output(tp); 2248 tp->snd_cwnd = tp->snd_ssthresh + 2249 tp->t_segsz * tp->t_dupacks; 2250 if (SEQ_GT(onxt, tp->snd_nxt)) 2251 tp->snd_nxt = onxt; 2252 goto drop; 2253 } else if (tp->t_dupacks > tcprexmtthresh) { 2254 tp->snd_cwnd += tp->t_segsz; 2255 (void) tcp_output(tp); 2256 goto drop; 2257 } 2258 } else { 2259 /* 2260 * If the ack appears to be very old, only 2261 * allow data that is in-sequence. This 2262 * makes it somewhat more difficult to insert 2263 * forged data by guessing sequence numbers. 2264 * Sent an ack to try to update the send 2265 * sequence number on the other side. 2266 */ 2267 if (tlen && th->th_seq != tp->rcv_nxt && 2268 SEQ_LT(th->th_ack, 2269 tp->snd_una - tp->max_sndwnd)) 2270 goto dropafterack; 2271 } 2272 break; 2273 } 2274 /* 2275 * If the congestion window was inflated to account 2276 * for the other side's cached packets, retract it. 2277 */ 2278 if (TCP_SACK_ENABLED(tp)) 2279 tcp_sack_newack(tp, th); 2280 else if (tcp_do_newreno) 2281 tcp_newreno_newack(tp, th); 2282 else 2283 tcp_reno_newack(tp, th); 2284 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2285 tcpstat.tcps_rcvacktoomuch++; 2286 goto dropafterack; 2287 } 2288 acked = th->th_ack - tp->snd_una; 2289 tcpstat.tcps_rcvackpack++; 2290 tcpstat.tcps_rcvackbyte += acked; 2291 2292 /* 2293 * If we have a timestamp reply, update smoothed 2294 * round trip time. If no timestamp is present but 2295 * transmit timer is running and timed sequence 2296 * number was acked, update smoothed round trip time. 2297 * Since we now have an rtt measurement, cancel the 2298 * timer backoff (cf., Phil Karn's retransmit alg.). 2299 * Recompute the initial retransmit timer. 2300 */ 2301 if (ts_rtt) 2302 tcp_xmit_timer(tp, ts_rtt); 2303 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2304 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2305 2306 /* 2307 * If all outstanding data is acked, stop retransmit 2308 * timer and remember to restart (more output or persist). 2309 * If there is more data to be acked, restart retransmit 2310 * timer, using current (possibly backed-off) value. 2311 */ 2312 if (th->th_ack == tp->snd_max) { 2313 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2314 needoutput = 1; 2315 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2316 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2317 /* 2318 * When new data is acked, open the congestion window. 2319 * If the window gives us less than ssthresh packets 2320 * in flight, open exponentially (segsz per packet). 2321 * Otherwise open linearly: segsz per window 2322 * (segsz^2 / cwnd per packet). 2323 * 2324 * If we are still in fast recovery (meaning we are using 2325 * NewReno and we have only received partial acks), do not 2326 * inflate the window yet. 2327 */ 2328 if (tp->t_partialacks < 0) { 2329 u_int cw = tp->snd_cwnd; 2330 u_int incr = tp->t_segsz; 2331 2332 if (cw >= tp->snd_ssthresh) 2333 incr = incr * incr / cw; 2334 tp->snd_cwnd = min(cw + incr, 2335 TCP_MAXWIN << tp->snd_scale); 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, int toff, 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 void 3163 tcp_reno_newack(struct tcpcb *tp, struct tcphdr *th) 3164 { 3165 if (tp->t_partialacks < 0) { 3166 /* 3167 * We were not in fast recovery. Reset the duplicate ack 3168 * counter. 3169 */ 3170 tp->t_dupacks = 0; 3171 } else { 3172 /* 3173 * Clamp the congestion window to the crossover point and 3174 * exit fast recovery. 3175 */ 3176 if (tp->snd_cwnd > tp->snd_ssthresh) 3177 tp->snd_cwnd = tp->snd_ssthresh; 3178 tp->t_partialacks = -1; 3179 tp->t_dupacks = 0; 3180 } 3181 } 3182 3183 /* 3184 * Implement the NewReno response to a new ack, checking for partial acks in 3185 * fast recovery. 3186 */ 3187 void 3188 tcp_newreno_newack(struct tcpcb *tp, struct tcphdr *th) 3189 { 3190 if (tp->t_partialacks < 0) { 3191 /* 3192 * We were not in fast recovery. Reset the duplicate ack 3193 * counter. 3194 */ 3195 tp->t_dupacks = 0; 3196 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) { 3197 /* 3198 * This is a partial ack. Retransmit the first unacknowledged 3199 * segment and deflate the congestion window by the amount of 3200 * acknowledged data. Do not exit fast recovery. 3201 */ 3202 tcp_seq onxt = tp->snd_nxt; 3203 u_long ocwnd = tp->snd_cwnd; 3204 3205 /* 3206 * snd_una has not yet been updated and the socket's send 3207 * buffer has not yet drained off the ACK'd data, so we 3208 * have to leave snd_una as it was to get the correct data 3209 * offset in tcp_output(). 3210 */ 3211 if (++tp->t_partialacks == 1) 3212 TCP_TIMER_DISARM(tp, TCPT_REXMT); 3213 tp->t_rtttime = 0; 3214 tp->snd_nxt = th->th_ack; 3215 /* 3216 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 3217 * is not yet updated when we're called. 3218 */ 3219 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 3220 (void) tcp_output(tp); 3221 tp->snd_cwnd = ocwnd; 3222 if (SEQ_GT(onxt, tp->snd_nxt)) 3223 tp->snd_nxt = onxt; 3224 /* 3225 * Partial window deflation. Relies on fact that tp->snd_una 3226 * not updated yet. 3227 */ 3228 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 3229 } else { 3230 /* 3231 * Complete ack. Inflate the congestion window to ssthresh 3232 * and exit fast recovery. 3233 * 3234 * Window inflation should have left us with approx. 3235 * snd_ssthresh outstanding data. But in case we 3236 * would be inclined to send a burst, better to do 3237 * it via the slow start mechanism. 3238 */ 3239 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 3240 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 3241 + tp->t_segsz; 3242 else 3243 tp->snd_cwnd = tp->snd_ssthresh; 3244 tp->t_partialacks = -1; 3245 tp->t_dupacks = 0; 3246 } 3247 } 3248 3249 3250 /* 3251 * TCP compressed state engine. Currently used to hold compressed 3252 * state for SYN_RECEIVED. 3253 */ 3254 3255 u_long syn_cache_count; 3256 u_int32_t syn_hash1, syn_hash2; 3257 3258 #define SYN_HASH(sa, sp, dp) \ 3259 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3260 ((u_int32_t)(sp)))^syn_hash2))) 3261 #ifndef INET6 3262 #define SYN_HASHALL(hash, src, dst) \ 3263 do { \ 3264 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3265 ((const struct sockaddr_in *)(src))->sin_port, \ 3266 ((const struct sockaddr_in *)(dst))->sin_port); \ 3267 } while (/*CONSTCOND*/ 0) 3268 #else 3269 #define SYN_HASH6(sa, sp, dp) \ 3270 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3271 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3272 & 0x7fffffff) 3273 3274 #define SYN_HASHALL(hash, src, dst) \ 3275 do { \ 3276 switch ((src)->sa_family) { \ 3277 case AF_INET: \ 3278 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3279 ((const struct sockaddr_in *)(src))->sin_port, \ 3280 ((const struct sockaddr_in *)(dst))->sin_port); \ 3281 break; \ 3282 case AF_INET6: \ 3283 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3284 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3285 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3286 break; \ 3287 default: \ 3288 hash = 0; \ 3289 } \ 3290 } while (/*CONSTCOND*/0) 3291 #endif /* INET6 */ 3292 3293 #define SYN_CACHE_RM(sc) \ 3294 do { \ 3295 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 3296 (sc), sc_bucketq); \ 3297 (sc)->sc_tp = NULL; \ 3298 LIST_REMOVE((sc), sc_tpq); \ 3299 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 3300 callout_stop(&(sc)->sc_timer); \ 3301 syn_cache_count--; \ 3302 } while (/*CONSTCOND*/0) 3303 3304 #define SYN_CACHE_PUT(sc) \ 3305 do { \ 3306 if ((sc)->sc_ipopts) \ 3307 (void) m_free((sc)->sc_ipopts); \ 3308 if ((sc)->sc_route4.ro_rt != NULL) \ 3309 RTFREE((sc)->sc_route4.ro_rt); \ 3310 if (callout_invoking(&(sc)->sc_timer)) \ 3311 (sc)->sc_flags |= SCF_DEAD; \ 3312 else \ 3313 pool_put(&syn_cache_pool, (sc)); \ 3314 } while (/*CONSTCOND*/0) 3315 3316 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL); 3317 3318 /* 3319 * We don't estimate RTT with SYNs, so each packet starts with the default 3320 * RTT and each timer step has a fixed timeout value. 3321 */ 3322 #define SYN_CACHE_TIMER_ARM(sc) \ 3323 do { \ 3324 TCPT_RANGESET((sc)->sc_rxtcur, \ 3325 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3326 TCPTV_REXMTMAX); \ 3327 callout_reset(&(sc)->sc_timer, \ 3328 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3329 } while (/*CONSTCOND*/0) 3330 3331 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3332 3333 void 3334 syn_cache_init(void) 3335 { 3336 int i; 3337 3338 /* Initialize the hash buckets. */ 3339 for (i = 0; i < tcp_syn_cache_size; i++) 3340 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3341 } 3342 3343 void 3344 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3345 { 3346 struct syn_cache_head *scp; 3347 struct syn_cache *sc2; 3348 int s; 3349 3350 /* 3351 * If there are no entries in the hash table, reinitialize 3352 * the hash secrets. 3353 */ 3354 if (syn_cache_count == 0) { 3355 syn_hash1 = arc4random(); 3356 syn_hash2 = arc4random(); 3357 } 3358 3359 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3360 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3361 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3362 3363 /* 3364 * Make sure that we don't overflow the per-bucket 3365 * limit or the total cache size limit. 3366 */ 3367 s = splsoftnet(); 3368 if (scp->sch_length >= tcp_syn_bucket_limit) { 3369 tcpstat.tcps_sc_bucketoverflow++; 3370 /* 3371 * The bucket is full. Toss the oldest element in the 3372 * bucket. This will be the first entry in the bucket. 3373 */ 3374 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3375 #ifdef DIAGNOSTIC 3376 /* 3377 * This should never happen; we should always find an 3378 * entry in our bucket. 3379 */ 3380 if (sc2 == NULL) 3381 panic("syn_cache_insert: bucketoverflow: impossible"); 3382 #endif 3383 SYN_CACHE_RM(sc2); 3384 SYN_CACHE_PUT(sc2); 3385 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3386 struct syn_cache_head *scp2, *sce; 3387 3388 tcpstat.tcps_sc_overflowed++; 3389 /* 3390 * The cache is full. Toss the oldest entry in the 3391 * first non-empty bucket we can find. 3392 * 3393 * XXX We would really like to toss the oldest 3394 * entry in the cache, but we hope that this 3395 * condition doesn't happen very often. 3396 */ 3397 scp2 = scp; 3398 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3399 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3400 for (++scp2; scp2 != scp; scp2++) { 3401 if (scp2 >= sce) 3402 scp2 = &tcp_syn_cache[0]; 3403 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3404 break; 3405 } 3406 #ifdef DIAGNOSTIC 3407 /* 3408 * This should never happen; we should always find a 3409 * non-empty bucket. 3410 */ 3411 if (scp2 == scp) 3412 panic("syn_cache_insert: cacheoverflow: " 3413 "impossible"); 3414 #endif 3415 } 3416 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3417 SYN_CACHE_RM(sc2); 3418 SYN_CACHE_PUT(sc2); 3419 } 3420 3421 /* 3422 * Initialize the entry's timer. 3423 */ 3424 sc->sc_rxttot = 0; 3425 sc->sc_rxtshift = 0; 3426 SYN_CACHE_TIMER_ARM(sc); 3427 3428 /* Link it from tcpcb entry */ 3429 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3430 3431 /* Put it into the bucket. */ 3432 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3433 scp->sch_length++; 3434 syn_cache_count++; 3435 3436 tcpstat.tcps_sc_added++; 3437 splx(s); 3438 } 3439 3440 /* 3441 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3442 * If we have retransmitted an entry the maximum number of times, expire 3443 * that entry. 3444 */ 3445 void 3446 syn_cache_timer(void *arg) 3447 { 3448 struct syn_cache *sc = arg; 3449 int s; 3450 3451 s = splsoftnet(); 3452 callout_ack(&sc->sc_timer); 3453 3454 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3455 tcpstat.tcps_sc_delayed_free++; 3456 pool_put(&syn_cache_pool, sc); 3457 splx(s); 3458 return; 3459 } 3460 3461 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3462 /* Drop it -- too many retransmissions. */ 3463 goto dropit; 3464 } 3465 3466 /* 3467 * Compute the total amount of time this entry has 3468 * been on a queue. If this entry has been on longer 3469 * than the keep alive timer would allow, expire it. 3470 */ 3471 sc->sc_rxttot += sc->sc_rxtcur; 3472 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) 3473 goto dropit; 3474 3475 tcpstat.tcps_sc_retransmitted++; 3476 (void) syn_cache_respond(sc, NULL); 3477 3478 /* Advance the timer back-off. */ 3479 sc->sc_rxtshift++; 3480 SYN_CACHE_TIMER_ARM(sc); 3481 3482 splx(s); 3483 return; 3484 3485 dropit: 3486 tcpstat.tcps_sc_timed_out++; 3487 SYN_CACHE_RM(sc); 3488 SYN_CACHE_PUT(sc); 3489 splx(s); 3490 } 3491 3492 /* 3493 * Remove syn cache created by the specified tcb entry, 3494 * because this does not make sense to keep them 3495 * (if there's no tcb entry, syn cache entry will never be used) 3496 */ 3497 void 3498 syn_cache_cleanup(struct tcpcb *tp) 3499 { 3500 struct syn_cache *sc, *nsc; 3501 int s; 3502 3503 s = splsoftnet(); 3504 3505 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3506 nsc = LIST_NEXT(sc, sc_tpq); 3507 3508 #ifdef DIAGNOSTIC 3509 if (sc->sc_tp != tp) 3510 panic("invalid sc_tp in syn_cache_cleanup"); 3511 #endif 3512 SYN_CACHE_RM(sc); 3513 SYN_CACHE_PUT(sc); 3514 } 3515 /* just for safety */ 3516 LIST_INIT(&tp->t_sc); 3517 3518 splx(s); 3519 } 3520 3521 /* 3522 * Find an entry in the syn cache. 3523 */ 3524 struct syn_cache * 3525 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3526 struct syn_cache_head **headp) 3527 { 3528 struct syn_cache *sc; 3529 struct syn_cache_head *scp; 3530 u_int32_t hash; 3531 int s; 3532 3533 SYN_HASHALL(hash, src, dst); 3534 3535 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3536 *headp = scp; 3537 s = splsoftnet(); 3538 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3539 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3540 if (sc->sc_hash != hash) 3541 continue; 3542 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3543 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3544 splx(s); 3545 return (sc); 3546 } 3547 } 3548 splx(s); 3549 return (NULL); 3550 } 3551 3552 /* 3553 * This function gets called when we receive an ACK for a 3554 * socket in the LISTEN state. We look up the connection 3555 * in the syn cache, and if its there, we pull it out of 3556 * the cache and turn it into a full-blown connection in 3557 * the SYN-RECEIVED state. 3558 * 3559 * The return values may not be immediately obvious, and their effects 3560 * can be subtle, so here they are: 3561 * 3562 * NULL SYN was not found in cache; caller should drop the 3563 * packet and send an RST. 3564 * 3565 * -1 We were unable to create the new connection, and are 3566 * aborting it. An ACK,RST is being sent to the peer 3567 * (unless we got screwey sequence numbners; see below), 3568 * because the 3-way handshake has been completed. Caller 3569 * should not free the mbuf, since we may be using it. If 3570 * we are not, we will free it. 3571 * 3572 * Otherwise, the return value is a pointer to the new socket 3573 * associated with the connection. 3574 */ 3575 struct socket * 3576 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3577 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3578 struct socket *so, struct mbuf *m) 3579 { 3580 struct syn_cache *sc; 3581 struct syn_cache_head *scp; 3582 struct inpcb *inp = NULL; 3583 #ifdef INET6 3584 struct in6pcb *in6p = NULL; 3585 #endif 3586 struct tcpcb *tp = 0; 3587 struct mbuf *am; 3588 int s; 3589 struct socket *oso; 3590 3591 s = splsoftnet(); 3592 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3593 splx(s); 3594 return (NULL); 3595 } 3596 3597 /* 3598 * Verify the sequence and ack numbers. Try getting the correct 3599 * response again. 3600 */ 3601 if ((th->th_ack != sc->sc_iss + 1) || 3602 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3603 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3604 (void) syn_cache_respond(sc, m); 3605 splx(s); 3606 return ((struct socket *)(-1)); 3607 } 3608 3609 /* Remove this cache entry */ 3610 SYN_CACHE_RM(sc); 3611 splx(s); 3612 3613 /* 3614 * Ok, create the full blown connection, and set things up 3615 * as they would have been set up if we had created the 3616 * connection when the SYN arrived. If we can't create 3617 * the connection, abort it. 3618 */ 3619 /* 3620 * inp still has the OLD in_pcb stuff, set the 3621 * v6-related flags on the new guy, too. This is 3622 * done particularly for the case where an AF_INET6 3623 * socket is bound only to a port, and a v4 connection 3624 * comes in on that port. 3625 * we also copy the flowinfo from the original pcb 3626 * to the new one. 3627 */ 3628 oso = so; 3629 so = sonewconn(so, SS_ISCONNECTED); 3630 if (so == NULL) 3631 goto resetandabort; 3632 3633 switch (so->so_proto->pr_domain->dom_family) { 3634 #ifdef INET 3635 case AF_INET: 3636 inp = sotoinpcb(so); 3637 break; 3638 #endif 3639 #ifdef INET6 3640 case AF_INET6: 3641 in6p = sotoin6pcb(so); 3642 break; 3643 #endif 3644 } 3645 switch (src->sa_family) { 3646 #ifdef INET 3647 case AF_INET: 3648 if (inp) { 3649 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3650 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3651 inp->inp_options = ip_srcroute(); 3652 in_pcbstate(inp, INP_BOUND); 3653 if (inp->inp_options == NULL) { 3654 inp->inp_options = sc->sc_ipopts; 3655 sc->sc_ipopts = NULL; 3656 } 3657 } 3658 #ifdef INET6 3659 else if (in6p) { 3660 /* IPv4 packet to AF_INET6 socket */ 3661 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3662 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3663 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3664 &in6p->in6p_laddr.s6_addr32[3], 3665 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3666 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3667 in6totcpcb(in6p)->t_family = AF_INET; 3668 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3669 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3670 else 3671 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3672 in6_pcbstate(in6p, IN6P_BOUND); 3673 } 3674 #endif 3675 break; 3676 #endif 3677 #ifdef INET6 3678 case AF_INET6: 3679 if (in6p) { 3680 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3681 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3682 in6_pcbstate(in6p, IN6P_BOUND); 3683 } 3684 break; 3685 #endif 3686 } 3687 #ifdef INET6 3688 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3689 struct in6pcb *oin6p = sotoin6pcb(oso); 3690 /* inherit socket options from the listening socket */ 3691 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3692 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3693 m_freem(in6p->in6p_options); 3694 in6p->in6p_options = 0; 3695 } 3696 ip6_savecontrol(in6p, &in6p->in6p_options, 3697 mtod(m, struct ip6_hdr *), m); 3698 } 3699 #endif 3700 3701 #if defined(IPSEC) || defined(FAST_IPSEC) 3702 /* 3703 * we make a copy of policy, instead of sharing the policy, 3704 * for better behavior in terms of SA lookup and dead SA removal. 3705 */ 3706 if (inp) { 3707 /* copy old policy into new socket's */ 3708 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3709 printf("tcp_input: could not copy policy\n"); 3710 } 3711 #ifdef INET6 3712 else if (in6p) { 3713 /* copy old policy into new socket's */ 3714 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3715 in6p->in6p_sp)) 3716 printf("tcp_input: could not copy policy\n"); 3717 } 3718 #endif 3719 #endif 3720 3721 /* 3722 * Give the new socket our cached route reference. 3723 */ 3724 if (inp) 3725 inp->inp_route = sc->sc_route4; /* struct assignment */ 3726 #ifdef INET6 3727 else 3728 in6p->in6p_route = sc->sc_route6; 3729 #endif 3730 sc->sc_route4.ro_rt = NULL; 3731 3732 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3733 if (am == NULL) 3734 goto resetandabort; 3735 MCLAIM(am, &tcp_mowner); 3736 am->m_len = src->sa_len; 3737 bcopy(src, mtod(am, caddr_t), src->sa_len); 3738 if (inp) { 3739 if (in_pcbconnect(inp, am)) { 3740 (void) m_free(am); 3741 goto resetandabort; 3742 } 3743 } 3744 #ifdef INET6 3745 else if (in6p) { 3746 if (src->sa_family == AF_INET) { 3747 /* IPv4 packet to AF_INET6 socket */ 3748 struct sockaddr_in6 *sin6; 3749 sin6 = mtod(am, struct sockaddr_in6 *); 3750 am->m_len = sizeof(*sin6); 3751 bzero(sin6, sizeof(*sin6)); 3752 sin6->sin6_family = AF_INET6; 3753 sin6->sin6_len = sizeof(*sin6); 3754 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3755 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3756 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3757 &sin6->sin6_addr.s6_addr32[3], 3758 sizeof(sin6->sin6_addr.s6_addr32[3])); 3759 } 3760 if (in6_pcbconnect(in6p, am)) { 3761 (void) m_free(am); 3762 goto resetandabort; 3763 } 3764 } 3765 #endif 3766 else { 3767 (void) m_free(am); 3768 goto resetandabort; 3769 } 3770 (void) m_free(am); 3771 3772 if (inp) 3773 tp = intotcpcb(inp); 3774 #ifdef INET6 3775 else if (in6p) 3776 tp = in6totcpcb(in6p); 3777 #endif 3778 else 3779 tp = NULL; 3780 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3781 if (sc->sc_request_r_scale != 15) { 3782 tp->requested_s_scale = sc->sc_requested_s_scale; 3783 tp->request_r_scale = sc->sc_request_r_scale; 3784 tp->snd_scale = sc->sc_requested_s_scale; 3785 tp->rcv_scale = sc->sc_request_r_scale; 3786 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3787 } 3788 if (sc->sc_flags & SCF_TIMESTAMP) 3789 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3790 tp->ts_timebase = sc->sc_timebase; 3791 3792 tp->t_template = tcp_template(tp); 3793 if (tp->t_template == 0) { 3794 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3795 so = NULL; 3796 m_freem(m); 3797 goto abort; 3798 } 3799 3800 tp->iss = sc->sc_iss; 3801 tp->irs = sc->sc_irs; 3802 tcp_sendseqinit(tp); 3803 tcp_rcvseqinit(tp); 3804 tp->t_state = TCPS_SYN_RECEIVED; 3805 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3806 tcpstat.tcps_accepts++; 3807 3808 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 3809 tp->t_flags |= TF_WILL_SACK; 3810 3811 #ifdef TCP_SIGNATURE 3812 if (sc->sc_flags & SCF_SIGNATURE) 3813 tp->t_flags |= TF_SIGNATURE; 3814 #endif 3815 3816 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3817 tp->t_ourmss = sc->sc_ourmaxseg; 3818 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3819 3820 /* 3821 * Initialize the initial congestion window. If we 3822 * had to retransmit the SYN,ACK, we must initialize cwnd 3823 * to 1 segment (i.e. the Loss Window). 3824 */ 3825 if (sc->sc_rxtshift) 3826 tp->snd_cwnd = tp->t_peermss; 3827 else { 3828 int ss = tcp_init_win; 3829 #ifdef INET 3830 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3831 ss = tcp_init_win_local; 3832 #endif 3833 #ifdef INET6 3834 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3835 ss = tcp_init_win_local; 3836 #endif 3837 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3838 } 3839 3840 tcp_rmx_rtt(tp); 3841 tp->snd_wl1 = sc->sc_irs; 3842 tp->rcv_up = sc->sc_irs + 1; 3843 3844 /* 3845 * This is what whould have happened in tcp_output() when 3846 * the SYN,ACK was sent. 3847 */ 3848 tp->snd_up = tp->snd_una; 3849 tp->snd_max = tp->snd_nxt = tp->iss+1; 3850 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3851 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3852 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3853 tp->last_ack_sent = tp->rcv_nxt; 3854 tp->t_partialacks = -1; 3855 tp->t_dupacks = 0; 3856 3857 tcpstat.tcps_sc_completed++; 3858 SYN_CACHE_PUT(sc); 3859 return (so); 3860 3861 resetandabort: 3862 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3863 abort: 3864 if (so != NULL) 3865 (void) soabort(so); 3866 SYN_CACHE_PUT(sc); 3867 tcpstat.tcps_sc_aborted++; 3868 return ((struct socket *)(-1)); 3869 } 3870 3871 /* 3872 * This function is called when we get a RST for a 3873 * non-existent connection, so that we can see if the 3874 * connection is in the syn cache. If it is, zap it. 3875 */ 3876 3877 void 3878 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 3879 { 3880 struct syn_cache *sc; 3881 struct syn_cache_head *scp; 3882 int s = splsoftnet(); 3883 3884 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3885 splx(s); 3886 return; 3887 } 3888 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3889 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3890 splx(s); 3891 return; 3892 } 3893 SYN_CACHE_RM(sc); 3894 splx(s); 3895 tcpstat.tcps_sc_reset++; 3896 SYN_CACHE_PUT(sc); 3897 } 3898 3899 void 3900 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 3901 struct tcphdr *th) 3902 { 3903 struct syn_cache *sc; 3904 struct syn_cache_head *scp; 3905 int s; 3906 3907 s = splsoftnet(); 3908 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3909 splx(s); 3910 return; 3911 } 3912 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3913 if (ntohl (th->th_seq) != sc->sc_iss) { 3914 splx(s); 3915 return; 3916 } 3917 3918 /* 3919 * If we've retransmitted 3 times and this is our second error, 3920 * we remove the entry. Otherwise, we allow it to continue on. 3921 * This prevents us from incorrectly nuking an entry during a 3922 * spurious network outage. 3923 * 3924 * See tcp_notify(). 3925 */ 3926 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3927 sc->sc_flags |= SCF_UNREACH; 3928 splx(s); 3929 return; 3930 } 3931 3932 SYN_CACHE_RM(sc); 3933 splx(s); 3934 tcpstat.tcps_sc_unreach++; 3935 SYN_CACHE_PUT(sc); 3936 } 3937 3938 /* 3939 * Given a LISTEN socket and an inbound SYN request, add 3940 * this to the syn cache, and send back a segment: 3941 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3942 * to the source. 3943 * 3944 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3945 * Doing so would require that we hold onto the data and deliver it 3946 * to the application. However, if we are the target of a SYN-flood 3947 * DoS attack, an attacker could send data which would eventually 3948 * consume all available buffer space if it were ACKed. By not ACKing 3949 * the data, we avoid this DoS scenario. 3950 */ 3951 3952 int 3953 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3954 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 3955 int optlen, struct tcp_opt_info *oi) 3956 { 3957 struct tcpcb tb, *tp; 3958 long win; 3959 struct syn_cache *sc; 3960 struct syn_cache_head *scp; 3961 struct mbuf *ipopts; 3962 struct tcp_opt_info opti; 3963 3964 tp = sototcpcb(so); 3965 3966 bzero(&opti, sizeof(opti)); 3967 3968 /* 3969 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3970 * 3971 * Note this check is performed in tcp_input() very early on. 3972 */ 3973 3974 /* 3975 * Initialize some local state. 3976 */ 3977 win = sbspace(&so->so_rcv); 3978 if (win > TCP_MAXWIN) 3979 win = TCP_MAXWIN; 3980 3981 switch (src->sa_family) { 3982 #ifdef INET 3983 case AF_INET: 3984 /* 3985 * Remember the IP options, if any. 3986 */ 3987 ipopts = ip_srcroute(); 3988 break; 3989 #endif 3990 default: 3991 ipopts = NULL; 3992 } 3993 3994 #ifdef TCP_SIGNATURE 3995 if (optp || (tp->t_flags & TF_SIGNATURE)) 3996 #else 3997 if (optp) 3998 #endif 3999 { 4000 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4001 #ifdef TCP_SIGNATURE 4002 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4003 #endif 4004 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 4005 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 4006 return (0); 4007 } else 4008 tb.t_flags = 0; 4009 4010 /* 4011 * See if we already have an entry for this connection. 4012 * If we do, resend the SYN,ACK. We do not count this 4013 * as a retransmission (XXX though maybe we should). 4014 */ 4015 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4016 tcpstat.tcps_sc_dupesyn++; 4017 if (ipopts) { 4018 /* 4019 * If we were remembering a previous source route, 4020 * forget it and use the new one we've been given. 4021 */ 4022 if (sc->sc_ipopts) 4023 (void) m_free(sc->sc_ipopts); 4024 sc->sc_ipopts = ipopts; 4025 } 4026 sc->sc_timestamp = tb.ts_recent; 4027 if (syn_cache_respond(sc, m) == 0) { 4028 tcpstat.tcps_sndacks++; 4029 tcpstat.tcps_sndtotal++; 4030 } 4031 return (1); 4032 } 4033 4034 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4035 if (sc == NULL) { 4036 if (ipopts) 4037 (void) m_free(ipopts); 4038 return (0); 4039 } 4040 4041 /* 4042 * Fill in the cache, and put the necessary IP and TCP 4043 * options into the reply. 4044 */ 4045 bzero(sc, sizeof(struct syn_cache)); 4046 callout_init(&sc->sc_timer); 4047 bcopy(src, &sc->sc_src, src->sa_len); 4048 bcopy(dst, &sc->sc_dst, dst->sa_len); 4049 sc->sc_flags = 0; 4050 sc->sc_ipopts = ipopts; 4051 sc->sc_irs = th->th_seq; 4052 switch (src->sa_family) { 4053 #ifdef INET 4054 case AF_INET: 4055 { 4056 struct sockaddr_in *srcin = (void *) src; 4057 struct sockaddr_in *dstin = (void *) dst; 4058 4059 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4060 &srcin->sin_addr, dstin->sin_port, 4061 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4062 break; 4063 } 4064 #endif /* INET */ 4065 #ifdef INET6 4066 case AF_INET6: 4067 { 4068 struct sockaddr_in6 *srcin6 = (void *) src; 4069 struct sockaddr_in6 *dstin6 = (void *) dst; 4070 4071 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4072 &srcin6->sin6_addr, dstin6->sin6_port, 4073 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4074 break; 4075 } 4076 #endif /* INET6 */ 4077 } 4078 sc->sc_peermaxseg = oi->maxseg; 4079 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4080 m->m_pkthdr.rcvif : NULL, 4081 sc->sc_src.sa.sa_family); 4082 sc->sc_win = win; 4083 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 4084 sc->sc_timestamp = tb.ts_recent; 4085 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4086 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4087 sc->sc_flags |= SCF_TIMESTAMP; 4088 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4089 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4090 sc->sc_requested_s_scale = tb.requested_s_scale; 4091 sc->sc_request_r_scale = 0; 4092 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4093 TCP_MAXWIN << sc->sc_request_r_scale < 4094 so->so_rcv.sb_hiwat) 4095 sc->sc_request_r_scale++; 4096 } else { 4097 sc->sc_requested_s_scale = 15; 4098 sc->sc_request_r_scale = 15; 4099 } 4100 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4101 sc->sc_flags |= SCF_SACK_PERMIT; 4102 #ifdef TCP_SIGNATURE 4103 if (tb.t_flags & TF_SIGNATURE) 4104 sc->sc_flags |= SCF_SIGNATURE; 4105 #endif 4106 sc->sc_tp = tp; 4107 if (syn_cache_respond(sc, m) == 0) { 4108 syn_cache_insert(sc, tp); 4109 tcpstat.tcps_sndacks++; 4110 tcpstat.tcps_sndtotal++; 4111 } else { 4112 SYN_CACHE_PUT(sc); 4113 tcpstat.tcps_sc_dropped++; 4114 } 4115 return (1); 4116 } 4117 4118 int 4119 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4120 { 4121 struct route *ro; 4122 u_int8_t *optp; 4123 int optlen, error; 4124 u_int16_t tlen; 4125 struct ip *ip = NULL; 4126 #ifdef INET6 4127 struct ip6_hdr *ip6 = NULL; 4128 #endif 4129 struct tcpcb *tp; 4130 struct tcphdr *th; 4131 u_int hlen; 4132 struct socket *so; 4133 4134 switch (sc->sc_src.sa.sa_family) { 4135 case AF_INET: 4136 hlen = sizeof(struct ip); 4137 ro = &sc->sc_route4; 4138 break; 4139 #ifdef INET6 4140 case AF_INET6: 4141 hlen = sizeof(struct ip6_hdr); 4142 ro = (struct route *)&sc->sc_route6; 4143 break; 4144 #endif 4145 default: 4146 if (m) 4147 m_freem(m); 4148 return (EAFNOSUPPORT); 4149 } 4150 4151 /* Compute the size of the TCP options. */ 4152 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4153 ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) + 4154 #ifdef TCP_SIGNATURE 4155 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4156 #endif 4157 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4158 4159 tlen = hlen + sizeof(struct tcphdr) + optlen; 4160 4161 /* 4162 * Create the IP+TCP header from scratch. 4163 */ 4164 if (m) 4165 m_freem(m); 4166 #ifdef DIAGNOSTIC 4167 if (max_linkhdr + tlen > MCLBYTES) 4168 return (ENOBUFS); 4169 #endif 4170 MGETHDR(m, M_DONTWAIT, MT_DATA); 4171 if (m && tlen > MHLEN) { 4172 MCLGET(m, M_DONTWAIT); 4173 if ((m->m_flags & M_EXT) == 0) { 4174 m_freem(m); 4175 m = NULL; 4176 } 4177 } 4178 if (m == NULL) 4179 return (ENOBUFS); 4180 MCLAIM(m, &tcp_tx_mowner); 4181 4182 /* Fixup the mbuf. */ 4183 m->m_data += max_linkhdr; 4184 m->m_len = m->m_pkthdr.len = tlen; 4185 if (sc->sc_tp) { 4186 tp = sc->sc_tp; 4187 if (tp->t_inpcb) 4188 so = tp->t_inpcb->inp_socket; 4189 #ifdef INET6 4190 else if (tp->t_in6pcb) 4191 so = tp->t_in6pcb->in6p_socket; 4192 #endif 4193 else 4194 so = NULL; 4195 } else 4196 so = NULL; 4197 m->m_pkthdr.rcvif = NULL; 4198 memset(mtod(m, u_char *), 0, tlen); 4199 4200 switch (sc->sc_src.sa.sa_family) { 4201 case AF_INET: 4202 ip = mtod(m, struct ip *); 4203 ip->ip_v = 4; 4204 ip->ip_dst = sc->sc_src.sin.sin_addr; 4205 ip->ip_src = sc->sc_dst.sin.sin_addr; 4206 ip->ip_p = IPPROTO_TCP; 4207 th = (struct tcphdr *)(ip + 1); 4208 th->th_dport = sc->sc_src.sin.sin_port; 4209 th->th_sport = sc->sc_dst.sin.sin_port; 4210 break; 4211 #ifdef INET6 4212 case AF_INET6: 4213 ip6 = mtod(m, struct ip6_hdr *); 4214 ip6->ip6_vfc = IPV6_VERSION; 4215 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4216 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4217 ip6->ip6_nxt = IPPROTO_TCP; 4218 /* ip6_plen will be updated in ip6_output() */ 4219 th = (struct tcphdr *)(ip6 + 1); 4220 th->th_dport = sc->sc_src.sin6.sin6_port; 4221 th->th_sport = sc->sc_dst.sin6.sin6_port; 4222 break; 4223 #endif 4224 default: 4225 th = NULL; 4226 } 4227 4228 th->th_seq = htonl(sc->sc_iss); 4229 th->th_ack = htonl(sc->sc_irs + 1); 4230 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4231 th->th_flags = TH_SYN|TH_ACK; 4232 th->th_win = htons(sc->sc_win); 4233 /* th_sum already 0 */ 4234 /* th_urp already 0 */ 4235 4236 /* Tack on the TCP options. */ 4237 optp = (u_int8_t *)(th + 1); 4238 *optp++ = TCPOPT_MAXSEG; 4239 *optp++ = 4; 4240 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4241 *optp++ = sc->sc_ourmaxseg & 0xff; 4242 4243 if (sc->sc_request_r_scale != 15) { 4244 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4245 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4246 sc->sc_request_r_scale); 4247 optp += 4; 4248 } 4249 4250 if (sc->sc_flags & SCF_TIMESTAMP) { 4251 u_int32_t *lp = (u_int32_t *)(optp); 4252 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4253 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4254 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4255 *lp = htonl(sc->sc_timestamp); 4256 optp += TCPOLEN_TSTAMP_APPA; 4257 } 4258 4259 if (sc->sc_flags & SCF_SACK_PERMIT) { 4260 u_int8_t *p = optp; 4261 4262 /* Let the peer know that we will SACK. */ 4263 p[0] = TCPOPT_SACK_PERMITTED; 4264 p[1] = 2; 4265 p[2] = TCPOPT_NOP; 4266 p[3] = TCPOPT_NOP; 4267 optp += 4; 4268 } 4269 4270 #ifdef TCP_SIGNATURE 4271 if (sc->sc_flags & SCF_SIGNATURE) { 4272 struct secasvar *sav; 4273 u_int8_t *sigp; 4274 4275 sav = tcp_signature_getsav(m, th); 4276 4277 if (sav == NULL) { 4278 if (m) 4279 m_freem(m); 4280 return (EPERM); 4281 } 4282 4283 *optp++ = TCPOPT_SIGNATURE; 4284 *optp++ = TCPOLEN_SIGNATURE; 4285 sigp = optp; 4286 bzero(optp, TCP_SIGLEN); 4287 optp += TCP_SIGLEN; 4288 *optp++ = TCPOPT_NOP; 4289 *optp++ = TCPOPT_EOL; 4290 4291 (void)tcp_signature(m, th, hlen, sav, sigp); 4292 4293 key_sa_recordxfer(sav, m); 4294 #ifdef FAST_IPSEC 4295 KEY_FREESAV(&sav); 4296 #else 4297 key_freesav(sav); 4298 #endif 4299 } 4300 #endif 4301 4302 /* Compute the packet's checksum. */ 4303 switch (sc->sc_src.sa.sa_family) { 4304 case AF_INET: 4305 ip->ip_len = htons(tlen - hlen); 4306 th->th_sum = 0; 4307 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4308 break; 4309 #ifdef INET6 4310 case AF_INET6: 4311 ip6->ip6_plen = htons(tlen - hlen); 4312 th->th_sum = 0; 4313 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4314 break; 4315 #endif 4316 } 4317 4318 /* 4319 * Fill in some straggling IP bits. Note the stack expects 4320 * ip_len to be in host order, for convenience. 4321 */ 4322 switch (sc->sc_src.sa.sa_family) { 4323 #ifdef INET 4324 case AF_INET: 4325 ip->ip_len = htons(tlen); 4326 ip->ip_ttl = ip_defttl; 4327 /* XXX tos? */ 4328 break; 4329 #endif 4330 #ifdef INET6 4331 case AF_INET6: 4332 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4333 ip6->ip6_vfc |= IPV6_VERSION; 4334 ip6->ip6_plen = htons(tlen - hlen); 4335 /* ip6_hlim will be initialized afterwards */ 4336 /* XXX flowlabel? */ 4337 break; 4338 #endif 4339 } 4340 4341 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4342 tp = sc->sc_tp; 4343 4344 switch (sc->sc_src.sa.sa_family) { 4345 #ifdef INET 4346 case AF_INET: 4347 error = ip_output(m, sc->sc_ipopts, ro, 4348 (ip_mtudisc ? IP_MTUDISC : 0), 4349 (struct ip_moptions *)NULL, so); 4350 break; 4351 #endif 4352 #ifdef INET6 4353 case AF_INET6: 4354 ip6->ip6_hlim = in6_selecthlim(NULL, 4355 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 4356 4357 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0, 4358 (struct ip6_moptions *)0, so, NULL); 4359 break; 4360 #endif 4361 default: 4362 error = EAFNOSUPPORT; 4363 break; 4364 } 4365 return (error); 4366 } 4367
Indexes created Tue Sep 23 07:09:52 GMT 2025