1 /* $NetBSD: tcp_input.c,v 1.321 2012/01/11 14:39:08 drochner Exp $ */ 2 3 /* 4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the project nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 /* 33 * @(#)COPYRIGHT 1.1 (NRL) 17 January 1995 34 * 35 * NRL grants permission for redistribution and use in source and binary 36 * forms, with or without modification, of the software and documentation 37 * created at NRL provided that the following conditions are met: 38 * 39 * 1. Redistributions of source code must retain the above copyright 40 * notice, this list of conditions and the following disclaimer. 41 * 2. Redistributions in binary form must reproduce the above copyright 42 * notice, this list of conditions and the following disclaimer in the 43 * documentation and/or other materials provided with the distribution. 44 * 3. All advertising materials mentioning features or use of this software 45 * must display the following acknowledgements: 46 * This product includes software developed by the University of 47 * California, Berkeley and its contributors. 48 * This product includes software developed at the Information 49 * Technology Division, US Naval Research Laboratory. 50 * 4. Neither the name of the NRL nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS 55 * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 56 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A 57 * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NRL OR 58 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 59 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 60 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 61 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 62 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 63 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 64 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 65 * 66 * The views and conclusions contained in the software and documentation 67 * are those of the authors and should not be interpreted as representing 68 * official policies, either expressed or implied, of the US Naval 69 * Research Laboratory (NRL). 70 */ 71 72 /*- 73 * Copyright (c) 1997, 1998, 1999, 2001, 2005, 2006, 74 * 2011 The NetBSD Foundation, Inc. 75 * All rights reserved. 76 * 77 * This code is derived from software contributed to The NetBSD Foundation 78 * by Coyote Point Systems, Inc. 79 * This code is derived from software contributed to The NetBSD Foundation 80 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 81 * Facility, NASA Ames Research Center. 82 * This code is derived from software contributed to The NetBSD Foundation 83 * by Charles M. Hannum. 84 * This code is derived from software contributed to The NetBSD Foundation 85 * by Rui Paulo. 86 * 87 * Redistribution and use in source and binary forms, with or without 88 * modification, are permitted provided that the following conditions 89 * are met: 90 * 1. Redistributions of source code must retain the above copyright 91 * notice, this list of conditions and the following disclaimer. 92 * 2. Redistributions in binary form must reproduce the above copyright 93 * notice, this list of conditions and the following disclaimer in the 94 * documentation and/or other materials provided with the distribution. 95 * 96 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 97 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 98 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 99 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 100 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 101 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 102 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 103 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 104 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 105 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 106 * POSSIBILITY OF SUCH DAMAGE. 107 */ 108 109 /* 110 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 111 * The Regents of the University of California. All rights reserved. 112 * 113 * Redistribution and use in source and binary forms, with or without 114 * modification, are permitted provided that the following conditions 115 * are met: 116 * 1. Redistributions of source code must retain the above copyright 117 * notice, this list of conditions and the following disclaimer. 118 * 2. Redistributions in binary form must reproduce the above copyright 119 * notice, this list of conditions and the following disclaimer in the 120 * documentation and/or other materials provided with the distribution. 121 * 3. Neither the name of the University nor the names of its contributors 122 * may be used to endorse or promote products derived from this software 123 * without specific prior written permission. 124 * 125 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 126 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 127 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 128 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 129 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 130 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 131 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 132 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 133 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 134 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 135 * SUCH DAMAGE. 136 * 137 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 138 */ 139 140 /* 141 * TODO list for SYN cache stuff: 142 * 143 * Find room for a "state" field, which is needed to keep a 144 * compressed state for TIME_WAIT TCBs. It's been noted already 145 * that this is fairly important for very high-volume web and 146 * mail servers, which use a large number of short-lived 147 * connections. 148 */ 149 150 #include <sys/cdefs.h> 151 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.321 2012/01/11 14:39:08 drochner Exp $"); 152 153 #include "opt_inet.h" 154 #include "opt_ipsec.h" 155 #include "opt_inet_csum.h" 156 #include "opt_tcp_debug.h" 157 158 #include <sys/param.h> 159 #include <sys/systm.h> 160 #include <sys/malloc.h> 161 #include <sys/mbuf.h> 162 #include <sys/protosw.h> 163 #include <sys/socket.h> 164 #include <sys/socketvar.h> 165 #include <sys/errno.h> 166 #include <sys/syslog.h> 167 #include <sys/pool.h> 168 #include <sys/domain.h> 169 #include <sys/kernel.h> 170 #ifdef TCP_SIGNATURE 171 #include <sys/md5.h> 172 #endif 173 #include <sys/lwp.h> /* for lwp0 */ 174 #include <sys/cprng.h> 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 #ifdef TCP_SIGNATURE 200 #include <netinet6/scope6_var.h> 201 #endif 202 #endif 203 204 #ifndef INET6 205 /* always need ip6.h for IP6_EXTHDR_GET */ 206 #include <netinet/ip6.h> 207 #endif 208 209 #include <netinet/tcp.h> 210 #include <netinet/tcp_fsm.h> 211 #include <netinet/tcp_seq.h> 212 #include <netinet/tcp_timer.h> 213 #include <netinet/tcp_var.h> 214 #include <netinet/tcp_private.h> 215 #include <netinet/tcpip.h> 216 #include <netinet/tcp_congctl.h> 217 #include <netinet/tcp_debug.h> 218 219 #ifdef KAME_IPSEC 220 #include <netinet6/ipsec.h> 221 #include <netinet6/ipsec_private.h> 222 #include <netkey/key.h> 223 #endif /*KAME_IPSEC*/ 224 #ifdef INET6 225 #include "faith.h" 226 #if defined(NFAITH) && NFAITH > 0 227 #include <net/if_faith.h> 228 #endif 229 #endif /* INET6 */ 230 231 #ifdef FAST_IPSEC 232 #include <netipsec/ipsec.h> 233 #include <netipsec/ipsec_var.h> 234 #include <netipsec/ipsec_private.h> 235 #include <netipsec/key.h> 236 #ifdef INET6 237 #include <netipsec/ipsec6.h> 238 #endif 239 #endif /* FAST_IPSEC*/ 240 241 #include <netinet/tcp_vtw.h> 242 243 int tcprexmtthresh = 3; 244 int tcp_log_refused; 245 246 int tcp_do_autorcvbuf = 1; 247 int tcp_autorcvbuf_inc = 16 * 1024; 248 int tcp_autorcvbuf_max = 256 * 1024; 249 int tcp_msl = (TCPTV_MSL / PR_SLOWHZ); 250 251 static int tcp_rst_ppslim_count = 0; 252 static struct timeval tcp_rst_ppslim_last; 253 static int tcp_ackdrop_ppslim_count = 0; 254 static struct timeval tcp_ackdrop_ppslim_last; 255 256 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 257 258 /* for modulo comparisons of timestamps */ 259 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 260 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 261 262 /* 263 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 264 */ 265 #ifdef INET6 266 static inline void 267 nd6_hint(struct tcpcb *tp) 268 { 269 struct rtentry *rt; 270 271 if (tp != NULL && tp->t_in6pcb != NULL && tp->t_family == AF_INET6 && 272 (rt = rtcache_validate(&tp->t_in6pcb->in6p_route)) != NULL) 273 nd6_nud_hint(rt, NULL, 0); 274 } 275 #else 276 static inline void 277 nd6_hint(struct tcpcb *tp) 278 { 279 } 280 #endif 281 282 /* 283 * Compute ACK transmission behavior. Delay the ACK unless 284 * we have already delayed an ACK (must send an ACK every two segments). 285 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 286 * option is enabled. 287 */ 288 static void 289 tcp_setup_ack(struct tcpcb *tp, const struct tcphdr *th) 290 { 291 292 if (tp->t_flags & TF_DELACK || 293 (tcp_ack_on_push && th->th_flags & TH_PUSH)) 294 tp->t_flags |= TF_ACKNOW; 295 else 296 TCP_SET_DELACK(tp); 297 } 298 299 static void 300 icmp_check(struct tcpcb *tp, const struct tcphdr *th, int acked) 301 { 302 303 /* 304 * If we had a pending ICMP message that refers to data that have 305 * just been acknowledged, disregard the recorded ICMP message. 306 */ 307 if ((tp->t_flags & TF_PMTUD_PEND) && 308 SEQ_GT(th->th_ack, tp->t_pmtud_th_seq)) 309 tp->t_flags &= ~TF_PMTUD_PEND; 310 311 /* 312 * Keep track of the largest chunk of data 313 * acknowledged since last PMTU update 314 */ 315 if (tp->t_pmtud_mss_acked < acked) 316 tp->t_pmtud_mss_acked = acked; 317 } 318 319 /* 320 * Convert TCP protocol fields to host order for easier processing. 321 */ 322 static void 323 tcp_fields_to_host(struct tcphdr *th) 324 { 325 326 NTOHL(th->th_seq); 327 NTOHL(th->th_ack); 328 NTOHS(th->th_win); 329 NTOHS(th->th_urp); 330 } 331 332 /* 333 * ... and reverse the above. 334 */ 335 static void 336 tcp_fields_to_net(struct tcphdr *th) 337 { 338 339 HTONL(th->th_seq); 340 HTONL(th->th_ack); 341 HTONS(th->th_win); 342 HTONS(th->th_urp); 343 } 344 345 #ifdef TCP_CSUM_COUNTERS 346 #include <sys/device.h> 347 348 #if defined(INET) 349 extern struct evcnt tcp_hwcsum_ok; 350 extern struct evcnt tcp_hwcsum_bad; 351 extern struct evcnt tcp_hwcsum_data; 352 extern struct evcnt tcp_swcsum; 353 #endif /* defined(INET) */ 354 #if defined(INET6) 355 extern struct evcnt tcp6_hwcsum_ok; 356 extern struct evcnt tcp6_hwcsum_bad; 357 extern struct evcnt tcp6_hwcsum_data; 358 extern struct evcnt tcp6_swcsum; 359 #endif /* defined(INET6) */ 360 361 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 362 363 #else 364 365 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 366 367 #endif /* TCP_CSUM_COUNTERS */ 368 369 #ifdef TCP_REASS_COUNTERS 370 #include <sys/device.h> 371 372 extern struct evcnt tcp_reass_; 373 extern struct evcnt tcp_reass_empty; 374 extern struct evcnt tcp_reass_iteration[8]; 375 extern struct evcnt tcp_reass_prependfirst; 376 extern struct evcnt tcp_reass_prepend; 377 extern struct evcnt tcp_reass_insert; 378 extern struct evcnt tcp_reass_inserttail; 379 extern struct evcnt tcp_reass_append; 380 extern struct evcnt tcp_reass_appendtail; 381 extern struct evcnt tcp_reass_overlaptail; 382 extern struct evcnt tcp_reass_overlapfront; 383 extern struct evcnt tcp_reass_segdup; 384 extern struct evcnt tcp_reass_fragdup; 385 386 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 387 388 #else 389 390 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 391 392 #endif /* TCP_REASS_COUNTERS */ 393 394 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *, 395 int *); 396 static int tcp_dooptions(struct tcpcb *, const u_char *, int, 397 struct tcphdr *, struct mbuf *, int, struct tcp_opt_info *); 398 399 #ifdef INET 400 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 401 #endif 402 #ifdef INET6 403 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 404 #endif 405 406 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 407 408 #if defined(MBUFTRACE) 409 struct mowner tcp_reass_mowner = MOWNER_INIT("tcp", "reass"); 410 #endif /* defined(MBUFTRACE) */ 411 412 static struct pool tcpipqent_pool; 413 414 void 415 tcpipqent_init(void) 416 { 417 418 pool_init(&tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", 419 NULL, IPL_VM); 420 } 421 422 struct ipqent * 423 tcpipqent_alloc(void) 424 { 425 struct ipqent *ipqe; 426 int s; 427 428 s = splvm(); 429 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 430 splx(s); 431 432 return ipqe; 433 } 434 435 void 436 tcpipqent_free(struct ipqent *ipqe) 437 { 438 int s; 439 440 s = splvm(); 441 pool_put(&tcpipqent_pool, ipqe); 442 splx(s); 443 } 444 445 static int 446 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int *tlen) 447 { 448 struct ipqent *p, *q, *nq, *tiqe = NULL; 449 struct socket *so = NULL; 450 int pkt_flags; 451 tcp_seq pkt_seq; 452 unsigned pkt_len; 453 u_long rcvpartdupbyte = 0; 454 u_long rcvoobyte; 455 #ifdef TCP_REASS_COUNTERS 456 u_int count = 0; 457 #endif 458 uint64_t *tcps; 459 460 if (tp->t_inpcb) 461 so = tp->t_inpcb->inp_socket; 462 #ifdef INET6 463 else if (tp->t_in6pcb) 464 so = tp->t_in6pcb->in6p_socket; 465 #endif 466 467 TCP_REASS_LOCK_CHECK(tp); 468 469 /* 470 * Call with th==0 after become established to 471 * force pre-ESTABLISHED data up to user socket. 472 */ 473 if (th == 0) 474 goto present; 475 476 m_claimm(m, &tcp_reass_mowner); 477 478 rcvoobyte = *tlen; 479 /* 480 * Copy these to local variables because the tcpiphdr 481 * gets munged while we are collapsing mbufs. 482 */ 483 pkt_seq = th->th_seq; 484 pkt_len = *tlen; 485 pkt_flags = th->th_flags; 486 487 TCP_REASS_COUNTER_INCR(&tcp_reass_); 488 489 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 490 /* 491 * When we miss a packet, the vast majority of time we get 492 * packets that follow it in order. So optimize for that. 493 */ 494 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 495 p->ipqe_len += pkt_len; 496 p->ipqe_flags |= pkt_flags; 497 m_cat(p->ipre_mlast, m); 498 TRAVERSE(p->ipre_mlast); 499 m = NULL; 500 tiqe = p; 501 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 502 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 503 goto skip_replacement; 504 } 505 /* 506 * While we're here, if the pkt is completely beyond 507 * anything we have, just insert it at the tail. 508 */ 509 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 510 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 511 goto insert_it; 512 } 513 } 514 515 q = TAILQ_FIRST(&tp->segq); 516 517 if (q != NULL) { 518 /* 519 * If this segment immediately precedes the first out-of-order 520 * block, simply slap the segment in front of it and (mostly) 521 * skip the complicated logic. 522 */ 523 if (pkt_seq + pkt_len == q->ipqe_seq) { 524 q->ipqe_seq = pkt_seq; 525 q->ipqe_len += pkt_len; 526 q->ipqe_flags |= pkt_flags; 527 m_cat(m, q->ipqe_m); 528 q->ipqe_m = m; 529 q->ipre_mlast = m; /* last mbuf may have changed */ 530 TRAVERSE(q->ipre_mlast); 531 tiqe = q; 532 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 533 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 534 goto skip_replacement; 535 } 536 } else { 537 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 538 } 539 540 /* 541 * Find a segment which begins after this one does. 542 */ 543 for (p = NULL; q != NULL; q = nq) { 544 nq = TAILQ_NEXT(q, ipqe_q); 545 #ifdef TCP_REASS_COUNTERS 546 count++; 547 #endif 548 /* 549 * If the received segment is just right after this 550 * fragment, merge the two together and then check 551 * for further overlaps. 552 */ 553 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 554 #ifdef TCPREASS_DEBUG 555 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 556 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 557 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 558 #endif 559 pkt_len += q->ipqe_len; 560 pkt_flags |= q->ipqe_flags; 561 pkt_seq = q->ipqe_seq; 562 m_cat(q->ipre_mlast, m); 563 TRAVERSE(q->ipre_mlast); 564 m = q->ipqe_m; 565 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 566 goto free_ipqe; 567 } 568 /* 569 * If the received segment is completely past this 570 * fragment, we need to go the next fragment. 571 */ 572 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 573 p = q; 574 continue; 575 } 576 /* 577 * If the fragment is past the received segment, 578 * it (or any following) can't be concatenated. 579 */ 580 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 581 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 582 break; 583 } 584 585 /* 586 * We've received all the data in this segment before. 587 * mark it as a duplicate and return. 588 */ 589 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 590 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 591 tcps = TCP_STAT_GETREF(); 592 tcps[TCP_STAT_RCVDUPPACK]++; 593 tcps[TCP_STAT_RCVDUPBYTE] += pkt_len; 594 TCP_STAT_PUTREF(); 595 tcp_new_dsack(tp, pkt_seq, pkt_len); 596 m_freem(m); 597 if (tiqe != NULL) { 598 tcpipqent_free(tiqe); 599 } 600 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 601 goto out; 602 } 603 /* 604 * Received segment completely overlaps this fragment 605 * so we drop the fragment (this keeps the temporal 606 * ordering of segments correct). 607 */ 608 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 609 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 610 rcvpartdupbyte += q->ipqe_len; 611 m_freem(q->ipqe_m); 612 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 613 goto free_ipqe; 614 } 615 /* 616 * RX'ed segment extends past the end of the 617 * fragment. Drop the overlapping bytes. Then 618 * merge the fragment and segment then treat as 619 * a longer received packet. 620 */ 621 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 622 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 623 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 624 #ifdef TCPREASS_DEBUG 625 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 626 tp, overlap, 627 pkt_seq, pkt_seq + pkt_len, pkt_len); 628 #endif 629 m_adj(m, overlap); 630 rcvpartdupbyte += overlap; 631 m_cat(q->ipre_mlast, m); 632 TRAVERSE(q->ipre_mlast); 633 m = q->ipqe_m; 634 pkt_seq = q->ipqe_seq; 635 pkt_len += q->ipqe_len - overlap; 636 rcvoobyte -= overlap; 637 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 638 goto free_ipqe; 639 } 640 /* 641 * RX'ed segment extends past the front of the 642 * fragment. Drop the overlapping bytes on the 643 * received packet. The packet will then be 644 * contatentated with this fragment a bit later. 645 */ 646 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 647 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 648 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 649 #ifdef TCPREASS_DEBUG 650 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 651 tp, overlap, 652 pkt_seq, pkt_seq + pkt_len, pkt_len); 653 #endif 654 m_adj(m, -overlap); 655 pkt_len -= overlap; 656 rcvpartdupbyte += overlap; 657 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 658 rcvoobyte -= overlap; 659 } 660 /* 661 * If the received segment immediates precedes this 662 * fragment then tack the fragment onto this segment 663 * and reinsert the data. 664 */ 665 if (q->ipqe_seq == pkt_seq + pkt_len) { 666 #ifdef TCPREASS_DEBUG 667 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 668 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 669 pkt_seq, pkt_seq + pkt_len, pkt_len); 670 #endif 671 pkt_len += q->ipqe_len; 672 pkt_flags |= q->ipqe_flags; 673 m_cat(m, q->ipqe_m); 674 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 675 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 676 tp->t_segqlen--; 677 KASSERT(tp->t_segqlen >= 0); 678 KASSERT(tp->t_segqlen != 0 || 679 (TAILQ_EMPTY(&tp->segq) && 680 TAILQ_EMPTY(&tp->timeq))); 681 if (tiqe == NULL) { 682 tiqe = q; 683 } else { 684 tcpipqent_free(q); 685 } 686 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 687 break; 688 } 689 /* 690 * If the fragment is before the segment, remember it. 691 * When this loop is terminated, p will contain the 692 * pointer to fragment that is right before the received 693 * segment. 694 */ 695 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 696 p = q; 697 698 continue; 699 700 /* 701 * This is a common operation. It also will allow 702 * to save doing a malloc/free in most instances. 703 */ 704 free_ipqe: 705 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 706 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 707 tp->t_segqlen--; 708 KASSERT(tp->t_segqlen >= 0); 709 KASSERT(tp->t_segqlen != 0 || 710 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 711 if (tiqe == NULL) { 712 tiqe = q; 713 } else { 714 tcpipqent_free(q); 715 } 716 } 717 718 #ifdef TCP_REASS_COUNTERS 719 if (count > 7) 720 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 721 else if (count > 0) 722 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 723 #endif 724 725 insert_it: 726 727 /* 728 * Allocate a new queue entry since the received segment did not 729 * collapse onto any other out-of-order block; thus we are allocating 730 * a new block. If it had collapsed, tiqe would not be NULL and 731 * we would be reusing it. 732 * XXX If we can't, just drop the packet. XXX 733 */ 734 if (tiqe == NULL) { 735 tiqe = tcpipqent_alloc(); 736 if (tiqe == NULL) { 737 TCP_STATINC(TCP_STAT_RCVMEMDROP); 738 m_freem(m); 739 goto out; 740 } 741 } 742 743 /* 744 * Update the counters. 745 */ 746 tcps = TCP_STAT_GETREF(); 747 tcps[TCP_STAT_RCVOOPACK]++; 748 tcps[TCP_STAT_RCVOOBYTE] += rcvoobyte; 749 if (rcvpartdupbyte) { 750 tcps[TCP_STAT_RCVPARTDUPPACK]++; 751 tcps[TCP_STAT_RCVPARTDUPBYTE] += rcvpartdupbyte; 752 } 753 TCP_STAT_PUTREF(); 754 755 /* 756 * Insert the new fragment queue entry into both queues. 757 */ 758 tiqe->ipqe_m = m; 759 tiqe->ipre_mlast = m; 760 tiqe->ipqe_seq = pkt_seq; 761 tiqe->ipqe_len = pkt_len; 762 tiqe->ipqe_flags = pkt_flags; 763 if (p == NULL) { 764 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 765 #ifdef TCPREASS_DEBUG 766 if (tiqe->ipqe_seq != tp->rcv_nxt) 767 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 768 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 769 #endif 770 } else { 771 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 772 #ifdef TCPREASS_DEBUG 773 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 774 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 775 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 776 #endif 777 } 778 tp->t_segqlen++; 779 780 skip_replacement: 781 782 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 783 784 present: 785 /* 786 * Present data to user, advancing rcv_nxt through 787 * completed sequence space. 788 */ 789 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 790 goto out; 791 q = TAILQ_FIRST(&tp->segq); 792 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 793 goto out; 794 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 795 goto out; 796 797 tp->rcv_nxt += q->ipqe_len; 798 pkt_flags = q->ipqe_flags & TH_FIN; 799 nd6_hint(tp); 800 801 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 802 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 803 tp->t_segqlen--; 804 KASSERT(tp->t_segqlen >= 0); 805 KASSERT(tp->t_segqlen != 0 || 806 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 807 if (so->so_state & SS_CANTRCVMORE) 808 m_freem(q->ipqe_m); 809 else 810 sbappendstream(&so->so_rcv, q->ipqe_m); 811 tcpipqent_free(q); 812 TCP_REASS_UNLOCK(tp); 813 sorwakeup(so); 814 return (pkt_flags); 815 out: 816 TCP_REASS_UNLOCK(tp); 817 return (0); 818 } 819 820 #ifdef INET6 821 int 822 tcp6_input(struct mbuf **mp, int *offp, int proto) 823 { 824 struct mbuf *m = *mp; 825 826 /* 827 * draft-itojun-ipv6-tcp-to-anycast 828 * better place to put this in? 829 */ 830 if (m->m_flags & M_ANYCAST6) { 831 struct ip6_hdr *ip6; 832 if (m->m_len < sizeof(struct ip6_hdr)) { 833 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 834 TCP_STATINC(TCP_STAT_RCVSHORT); 835 return IPPROTO_DONE; 836 } 837 } 838 ip6 = mtod(m, struct ip6_hdr *); 839 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 840 (char *)&ip6->ip6_dst - (char *)ip6); 841 return IPPROTO_DONE; 842 } 843 844 tcp_input(m, *offp, proto); 845 return IPPROTO_DONE; 846 } 847 #endif 848 849 #ifdef INET 850 static void 851 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 852 { 853 char src[4*sizeof "123"]; 854 char dst[4*sizeof "123"]; 855 856 if (ip) { 857 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 858 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 859 } 860 else { 861 strlcpy(src, "(unknown)", sizeof(src)); 862 strlcpy(dst, "(unknown)", sizeof(dst)); 863 } 864 log(LOG_INFO, 865 "Connection attempt to TCP %s:%d from %s:%d\n", 866 dst, ntohs(th->th_dport), 867 src, ntohs(th->th_sport)); 868 } 869 #endif 870 871 #ifdef INET6 872 static void 873 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 874 { 875 char src[INET6_ADDRSTRLEN]; 876 char dst[INET6_ADDRSTRLEN]; 877 878 if (ip6) { 879 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 880 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 881 } 882 else { 883 strlcpy(src, "(unknown v6)", sizeof(src)); 884 strlcpy(dst, "(unknown v6)", sizeof(dst)); 885 } 886 log(LOG_INFO, 887 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 888 dst, ntohs(th->th_dport), 889 src, ntohs(th->th_sport)); 890 } 891 #endif 892 893 /* 894 * Checksum extended TCP header and data. 895 */ 896 int 897 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, 898 int toff, int off, int tlen) 899 { 900 901 /* 902 * XXX it's better to record and check if this mbuf is 903 * already checked. 904 */ 905 906 switch (af) { 907 #ifdef INET 908 case AF_INET: 909 switch (m->m_pkthdr.csum_flags & 910 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 911 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 912 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 913 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 914 goto badcsum; 915 916 case M_CSUM_TCPv4|M_CSUM_DATA: { 917 u_int32_t hw_csum = m->m_pkthdr.csum_data; 918 919 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 920 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 921 const struct ip *ip = 922 mtod(m, const struct ip *); 923 924 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 925 ip->ip_dst.s_addr, 926 htons(hw_csum + tlen + off + IPPROTO_TCP)); 927 } 928 if ((hw_csum ^ 0xffff) != 0) 929 goto badcsum; 930 break; 931 } 932 933 case M_CSUM_TCPv4: 934 /* Checksum was okay. */ 935 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 936 break; 937 938 default: 939 /* 940 * Must compute it ourselves. Maybe skip checksum 941 * on loopback interfaces. 942 */ 943 if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & 944 IFF_LOOPBACK) || 945 tcp_do_loopback_cksum)) { 946 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 947 if (in4_cksum(m, IPPROTO_TCP, toff, 948 tlen + off) != 0) 949 goto badcsum; 950 } 951 break; 952 } 953 break; 954 #endif /* INET4 */ 955 956 #ifdef INET6 957 case AF_INET6: 958 switch (m->m_pkthdr.csum_flags & 959 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 960 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 961 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 962 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 963 goto badcsum; 964 965 #if 0 /* notyet */ 966 case M_CSUM_TCPv6|M_CSUM_DATA: 967 #endif 968 969 case M_CSUM_TCPv6: 970 /* Checksum was okay. */ 971 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 972 break; 973 974 default: 975 /* 976 * Must compute it ourselves. Maybe skip checksum 977 * on loopback interfaces. 978 */ 979 if (__predict_true((m->m_flags & M_LOOP) == 0 || 980 tcp_do_loopback_cksum)) { 981 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 982 if (in6_cksum(m, IPPROTO_TCP, toff, 983 tlen + off) != 0) 984 goto badcsum; 985 } 986 } 987 break; 988 #endif /* INET6 */ 989 } 990 991 return 0; 992 993 badcsum: 994 TCP_STATINC(TCP_STAT_RCVBADSUM); 995 return -1; 996 } 997 998 /* When a packet arrives addressed to a vestigial tcpbp, we 999 * nevertheless have to respond to it per the spec. 1000 */ 1001 static void tcp_vtw_input(struct tcphdr *th, vestigial_inpcb_t *vp, 1002 struct mbuf *m, int tlen, int multicast) 1003 { 1004 int tiflags; 1005 int todrop, dupseg; 1006 uint32_t t_flags = 0; 1007 uint64_t *tcps; 1008 1009 tiflags = th->th_flags; 1010 todrop = vp->rcv_nxt - th->th_seq; 1011 dupseg = false; 1012 1013 if (todrop > 0) { 1014 if (tiflags & TH_SYN) { 1015 tiflags &= ~TH_SYN; 1016 ++th->th_seq; 1017 if (th->th_urp > 1) 1018 --th->th_urp; 1019 else { 1020 tiflags &= ~TH_URG; 1021 th->th_urp = 0; 1022 } 1023 --todrop; 1024 } 1025 if (todrop > tlen || 1026 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1027 /* 1028 * Any valid FIN or RST must be to the left of the 1029 * window. At this point the FIN or RST must be a 1030 * duplicate or out of sequence; drop it. 1031 */ 1032 if (tiflags & TH_RST) 1033 goto drop; 1034 tiflags &= ~(TH_FIN|TH_RST); 1035 /* 1036 * Send an ACK to resynchronize and drop any data. 1037 * But keep on processing for RST or ACK. 1038 */ 1039 t_flags |= TF_ACKNOW; 1040 todrop = tlen; 1041 dupseg = true; 1042 tcps = TCP_STAT_GETREF(); 1043 tcps[TCP_STAT_RCVDUPPACK] += 1; 1044 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 1045 TCP_STAT_PUTREF(); 1046 } else if ((tiflags & TH_RST) 1047 && th->th_seq != vp->rcv_nxt) { 1048 /* 1049 * Test for reset before adjusting the sequence 1050 * number for overlapping data. 1051 */ 1052 goto dropafterack_ratelim; 1053 } else { 1054 tcps = TCP_STAT_GETREF(); 1055 tcps[TCP_STAT_RCVPARTDUPPACK] += 1; 1056 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 1057 TCP_STAT_PUTREF(); 1058 } 1059 1060 // tcp_new_dsack(tp, th->th_seq, todrop); 1061 // hdroptlen += todrop; /*drop from head afterwards*/ 1062 1063 th->th_seq += todrop; 1064 tlen -= todrop; 1065 1066 if (th->th_urp > todrop) 1067 th->th_urp -= todrop; 1068 else { 1069 tiflags &= ~TH_URG; 1070 th->th_urp = 0; 1071 } 1072 } 1073 1074 /* 1075 * If new data are received on a connection after the 1076 * user processes are gone, then RST the other end. 1077 */ 1078 if (tlen) { 1079 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 1080 goto dropwithreset; 1081 } 1082 1083 /* 1084 * If segment ends after window, drop trailing data 1085 * (and PUSH and FIN); if nothing left, just ACK. 1086 */ 1087 todrop = (th->th_seq + tlen) - (vp->rcv_nxt+vp->rcv_wnd); 1088 1089 if (todrop > 0) { 1090 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 1091 if (todrop >= tlen) { 1092 /* 1093 * The segment actually starts after the window. 1094 * th->th_seq + tlen - vp->rcv_nxt - vp->rcv_wnd >= tlen 1095 * th->th_seq - vp->rcv_nxt - vp->rcv_wnd >= 0 1096 * th->th_seq >= vp->rcv_nxt + vp->rcv_wnd 1097 */ 1098 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 1099 /* 1100 * If a new connection request is received 1101 * while in TIME_WAIT, drop the old connection 1102 * and start over if the sequence numbers 1103 * are above the previous ones. 1104 */ 1105 if ((tiflags & TH_SYN) 1106 && SEQ_GT(th->th_seq, vp->rcv_nxt)) { 1107 /* We only support this in the !NOFDREF case, which 1108 * is to say: not here. 1109 */ 1110 goto dropwithreset;; 1111 } 1112 /* 1113 * If window is closed can only take segments at 1114 * window edge, and have to drop data and PUSH from 1115 * incoming segments. Continue processing, but 1116 * remember to ack. Otherwise, drop segment 1117 * and (if not RST) ack. 1118 */ 1119 if (vp->rcv_wnd == 0 && th->th_seq == vp->rcv_nxt) { 1120 t_flags |= TF_ACKNOW; 1121 TCP_STATINC(TCP_STAT_RCVWINPROBE); 1122 } else 1123 goto dropafterack; 1124 } else 1125 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 1126 m_adj(m, -todrop); 1127 tlen -= todrop; 1128 tiflags &= ~(TH_PUSH|TH_FIN); 1129 } 1130 1131 if (tiflags & TH_RST) { 1132 if (th->th_seq != vp->rcv_nxt) 1133 goto dropafterack_ratelim; 1134 1135 vtw_del(vp->ctl, vp->vtw); 1136 goto drop; 1137 } 1138 1139 /* 1140 * If the ACK bit is off we drop the segment and return. 1141 */ 1142 if ((tiflags & TH_ACK) == 0) { 1143 if (t_flags & TF_ACKNOW) 1144 goto dropafterack; 1145 else 1146 goto drop; 1147 } 1148 1149 /* 1150 * In TIME_WAIT state the only thing that should arrive 1151 * is a retransmission of the remote FIN. Acknowledge 1152 * it and restart the finack timer. 1153 */ 1154 vtw_restart(vp); 1155 goto dropafterack; 1156 1157 dropafterack: 1158 /* 1159 * Generate an ACK dropping incoming segment if it occupies 1160 * sequence space, where the ACK reflects our state. 1161 */ 1162 if (tiflags & TH_RST) 1163 goto drop; 1164 goto dropafterack2; 1165 1166 dropafterack_ratelim: 1167 /* 1168 * We may want to rate-limit ACKs against SYN/RST attack. 1169 */ 1170 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 1171 tcp_ackdrop_ppslim) == 0) { 1172 /* XXX stat */ 1173 goto drop; 1174 } 1175 /* ...fall into dropafterack2... */ 1176 1177 dropafterack2: 1178 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, th->th_ack, 1179 TH_ACK); 1180 return; 1181 1182 dropwithreset: 1183 /* 1184 * Generate a RST, dropping incoming segment. 1185 * Make ACK acceptable to originator of segment. 1186 */ 1187 if (tiflags & TH_RST) 1188 goto drop; 1189 1190 if (tiflags & TH_ACK) 1191 tcp_respond(0, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 1192 else { 1193 if (tiflags & TH_SYN) 1194 ++tlen; 1195 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, (tcp_seq)0, 1196 TH_RST|TH_ACK); 1197 } 1198 return; 1199 drop: 1200 m_freem(m); 1201 } 1202 1203 /* 1204 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 1205 */ 1206 void 1207 tcp_input(struct mbuf *m, ...) 1208 { 1209 struct tcphdr *th; 1210 struct ip *ip; 1211 struct inpcb *inp; 1212 #ifdef INET6 1213 struct ip6_hdr *ip6; 1214 struct in6pcb *in6p; 1215 #endif 1216 u_int8_t *optp = NULL; 1217 int optlen = 0; 1218 int len, tlen, toff, hdroptlen = 0; 1219 struct tcpcb *tp = 0; 1220 int tiflags; 1221 struct socket *so = NULL; 1222 int todrop, acked, ourfinisacked, needoutput = 0; 1223 bool dupseg; 1224 #ifdef TCP_DEBUG 1225 short ostate = 0; 1226 #endif 1227 u_long tiwin; 1228 struct tcp_opt_info opti; 1229 int off, iphlen; 1230 va_list ap; 1231 int af; /* af on the wire */ 1232 struct mbuf *tcp_saveti = NULL; 1233 uint32_t ts_rtt; 1234 uint8_t iptos; 1235 uint64_t *tcps; 1236 vestigial_inpcb_t vestige; 1237 1238 vestige.valid = 0; 1239 1240 MCLAIM(m, &tcp_rx_mowner); 1241 va_start(ap, m); 1242 toff = va_arg(ap, int); 1243 (void)va_arg(ap, int); /* ignore value, advance ap */ 1244 va_end(ap); 1245 1246 TCP_STATINC(TCP_STAT_RCVTOTAL); 1247 1248 memset(&opti, 0, sizeof(opti)); 1249 opti.ts_present = 0; 1250 opti.maxseg = 0; 1251 1252 /* 1253 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 1254 * 1255 * TCP is, by definition, unicast, so we reject all 1256 * multicast outright. 1257 * 1258 * Note, there are additional src/dst address checks in 1259 * the AF-specific code below. 1260 */ 1261 if (m->m_flags & (M_BCAST|M_MCAST)) { 1262 /* XXX stat */ 1263 goto drop; 1264 } 1265 #ifdef INET6 1266 if (m->m_flags & M_ANYCAST6) { 1267 /* XXX stat */ 1268 goto drop; 1269 } 1270 #endif 1271 1272 /* 1273 * Get IP and TCP header. 1274 * Note: IP leaves IP header in first mbuf. 1275 */ 1276 ip = mtod(m, struct ip *); 1277 switch (ip->ip_v) { 1278 #ifdef INET 1279 case 4: 1280 #ifdef INET6 1281 ip6 = NULL; 1282 #endif 1283 af = AF_INET; 1284 iphlen = sizeof(struct ip); 1285 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1286 sizeof(struct tcphdr)); 1287 if (th == NULL) { 1288 TCP_STATINC(TCP_STAT_RCVSHORT); 1289 return; 1290 } 1291 /* We do the checksum after PCB lookup... */ 1292 len = ntohs(ip->ip_len); 1293 tlen = len - toff; 1294 iptos = ip->ip_tos; 1295 break; 1296 #endif 1297 #ifdef INET6 1298 case 6: 1299 ip = NULL; 1300 iphlen = sizeof(struct ip6_hdr); 1301 af = AF_INET6; 1302 ip6 = mtod(m, struct ip6_hdr *); 1303 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1304 sizeof(struct tcphdr)); 1305 if (th == NULL) { 1306 TCP_STATINC(TCP_STAT_RCVSHORT); 1307 return; 1308 } 1309 1310 /* Be proactive about malicious use of IPv4 mapped address */ 1311 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 1312 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 1313 /* XXX stat */ 1314 goto drop; 1315 } 1316 1317 /* 1318 * Be proactive about unspecified IPv6 address in source. 1319 * As we use all-zero to indicate unbounded/unconnected pcb, 1320 * unspecified IPv6 address can be used to confuse us. 1321 * 1322 * Note that packets with unspecified IPv6 destination is 1323 * already dropped in ip6_input. 1324 */ 1325 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1326 /* XXX stat */ 1327 goto drop; 1328 } 1329 1330 /* 1331 * Make sure destination address is not multicast. 1332 * Source address checked in ip6_input(). 1333 */ 1334 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1335 /* XXX stat */ 1336 goto drop; 1337 } 1338 1339 /* We do the checksum after PCB lookup... */ 1340 len = m->m_pkthdr.len; 1341 tlen = len - toff; 1342 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1343 break; 1344 #endif 1345 default: 1346 m_freem(m); 1347 return; 1348 } 1349 1350 KASSERT(TCP_HDR_ALIGNED_P(th)); 1351 1352 /* 1353 * Check that TCP offset makes sense, 1354 * pull out TCP options and adjust length. XXX 1355 */ 1356 off = th->th_off << 2; 1357 if (off < sizeof (struct tcphdr) || off > tlen) { 1358 TCP_STATINC(TCP_STAT_RCVBADOFF); 1359 goto drop; 1360 } 1361 tlen -= off; 1362 1363 /* 1364 * tcp_input() has been modified to use tlen to mean the TCP data 1365 * length throughout the function. Other functions can use 1366 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1367 * rja 1368 */ 1369 1370 if (off > sizeof (struct tcphdr)) { 1371 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1372 if (th == NULL) { 1373 TCP_STATINC(TCP_STAT_RCVSHORT); 1374 return; 1375 } 1376 /* 1377 * NOTE: ip/ip6 will not be affected by m_pulldown() 1378 * (as they're before toff) and we don't need to update those. 1379 */ 1380 KASSERT(TCP_HDR_ALIGNED_P(th)); 1381 optlen = off - sizeof (struct tcphdr); 1382 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1383 /* 1384 * Do quick retrieval of timestamp options ("options 1385 * prediction?"). If timestamp is the only option and it's 1386 * formatted as recommended in RFC 1323 appendix A, we 1387 * quickly get the values now and not bother calling 1388 * tcp_dooptions(), etc. 1389 */ 1390 if ((optlen == TCPOLEN_TSTAMP_APPA || 1391 (optlen > TCPOLEN_TSTAMP_APPA && 1392 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1393 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1394 (th->th_flags & TH_SYN) == 0) { 1395 opti.ts_present = 1; 1396 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1397 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1398 optp = NULL; /* we've parsed the options */ 1399 } 1400 } 1401 tiflags = th->th_flags; 1402 1403 /* 1404 * Locate pcb for segment. 1405 */ 1406 findpcb: 1407 inp = NULL; 1408 #ifdef INET6 1409 in6p = NULL; 1410 #endif 1411 switch (af) { 1412 #ifdef INET 1413 case AF_INET: 1414 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1415 ip->ip_dst, th->th_dport, 1416 &vestige); 1417 if (inp == 0 && !vestige.valid) { 1418 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1419 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1420 } 1421 #ifdef INET6 1422 if (inp == 0 && !vestige.valid) { 1423 struct in6_addr s, d; 1424 1425 /* mapped addr case */ 1426 memset(&s, 0, sizeof(s)); 1427 s.s6_addr16[5] = htons(0xffff); 1428 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1429 memset(&d, 0, sizeof(d)); 1430 d.s6_addr16[5] = htons(0xffff); 1431 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1432 in6p = in6_pcblookup_connect(&tcbtable, &s, 1433 th->th_sport, &d, th->th_dport, 1434 0, &vestige); 1435 if (in6p == 0 && !vestige.valid) { 1436 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1437 in6p = in6_pcblookup_bind(&tcbtable, &d, 1438 th->th_dport, 0); 1439 } 1440 } 1441 #endif 1442 #ifndef INET6 1443 if (inp == 0 && !vestige.valid) 1444 #else 1445 if (inp == 0 && in6p == 0 && !vestige.valid) 1446 #endif 1447 { 1448 TCP_STATINC(TCP_STAT_NOPORT); 1449 if (tcp_log_refused && 1450 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1451 tcp4_log_refused(ip, th); 1452 } 1453 tcp_fields_to_host(th); 1454 goto dropwithreset_ratelim; 1455 } 1456 #if defined(KAME_IPSEC) || defined(FAST_IPSEC) 1457 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1458 ipsec4_in_reject(m, inp)) { 1459 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1460 goto drop; 1461 } 1462 #ifdef INET6 1463 else if (in6p && 1464 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1465 ipsec6_in_reject_so(m, in6p->in6p_socket)) { 1466 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1467 goto drop; 1468 } 1469 #endif 1470 #endif /*IPSEC*/ 1471 break; 1472 #endif /*INET*/ 1473 #ifdef INET6 1474 case AF_INET6: 1475 { 1476 int faith; 1477 1478 #if defined(NFAITH) && NFAITH > 0 1479 faith = faithprefix(&ip6->ip6_dst); 1480 #else 1481 faith = 0; 1482 #endif 1483 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1484 th->th_sport, &ip6->ip6_dst, th->th_dport, faith, &vestige); 1485 if (!in6p && !vestige.valid) { 1486 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1487 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1488 th->th_dport, faith); 1489 } 1490 if (!in6p && !vestige.valid) { 1491 TCP_STATINC(TCP_STAT_NOPORT); 1492 if (tcp_log_refused && 1493 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1494 tcp6_log_refused(ip6, th); 1495 } 1496 tcp_fields_to_host(th); 1497 goto dropwithreset_ratelim; 1498 } 1499 #if defined(KAME_IPSEC) || defined(FAST_IPSEC) 1500 if (in6p 1501 && (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 1502 && ipsec6_in_reject(m, in6p)) { 1503 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1504 goto drop; 1505 } 1506 #endif /*IPSEC*/ 1507 break; 1508 } 1509 #endif 1510 } 1511 1512 /* 1513 * If the state is CLOSED (i.e., TCB does not exist) then 1514 * all data in the incoming segment is discarded. 1515 * If the TCB exists but is in CLOSED state, it is embryonic, 1516 * but should either do a listen or a connect soon. 1517 */ 1518 tp = NULL; 1519 so = NULL; 1520 if (inp) { 1521 /* Check the minimum TTL for socket. */ 1522 if (ip->ip_ttl < inp->inp_ip_minttl) 1523 goto drop; 1524 1525 tp = intotcpcb(inp); 1526 so = inp->inp_socket; 1527 } 1528 #ifdef INET6 1529 else if (in6p) { 1530 tp = in6totcpcb(in6p); 1531 so = in6p->in6p_socket; 1532 } 1533 #endif 1534 else if (vestige.valid) { 1535 int mc = 0; 1536 1537 /* We do not support the resurrection of vtw tcpcps. 1538 */ 1539 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1540 goto badcsum; 1541 1542 switch (af) { 1543 #ifdef INET6 1544 case AF_INET6: 1545 mc = IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst); 1546 break; 1547 #endif 1548 1549 case AF_INET: 1550 mc = (IN_MULTICAST(ip->ip_dst.s_addr) 1551 || in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)); 1552 break; 1553 } 1554 1555 tcp_fields_to_host(th); 1556 tcp_vtw_input(th, &vestige, m, tlen, mc); 1557 m = 0; 1558 goto drop; 1559 } 1560 1561 if (tp == 0) { 1562 tcp_fields_to_host(th); 1563 goto dropwithreset_ratelim; 1564 } 1565 if (tp->t_state == TCPS_CLOSED) 1566 goto drop; 1567 1568 KASSERT(so->so_lock == softnet_lock); 1569 KASSERT(solocked(so)); 1570 1571 /* 1572 * Checksum extended TCP header and data. 1573 */ 1574 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1575 goto badcsum; 1576 1577 tcp_fields_to_host(th); 1578 1579 /* Unscale the window into a 32-bit value. */ 1580 if ((tiflags & TH_SYN) == 0) 1581 tiwin = th->th_win << tp->snd_scale; 1582 else 1583 tiwin = th->th_win; 1584 1585 #ifdef INET6 1586 /* save packet options if user wanted */ 1587 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1588 if (in6p->in6p_options) { 1589 m_freem(in6p->in6p_options); 1590 in6p->in6p_options = 0; 1591 } 1592 KASSERT(ip6 != NULL); 1593 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1594 } 1595 #endif 1596 1597 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1598 union syn_cache_sa src; 1599 union syn_cache_sa dst; 1600 1601 memset(&src, 0, sizeof(src)); 1602 memset(&dst, 0, sizeof(dst)); 1603 switch (af) { 1604 #ifdef INET 1605 case AF_INET: 1606 src.sin.sin_len = sizeof(struct sockaddr_in); 1607 src.sin.sin_family = AF_INET; 1608 src.sin.sin_addr = ip->ip_src; 1609 src.sin.sin_port = th->th_sport; 1610 1611 dst.sin.sin_len = sizeof(struct sockaddr_in); 1612 dst.sin.sin_family = AF_INET; 1613 dst.sin.sin_addr = ip->ip_dst; 1614 dst.sin.sin_port = th->th_dport; 1615 break; 1616 #endif 1617 #ifdef INET6 1618 case AF_INET6: 1619 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1620 src.sin6.sin6_family = AF_INET6; 1621 src.sin6.sin6_addr = ip6->ip6_src; 1622 src.sin6.sin6_port = th->th_sport; 1623 1624 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1625 dst.sin6.sin6_family = AF_INET6; 1626 dst.sin6.sin6_addr = ip6->ip6_dst; 1627 dst.sin6.sin6_port = th->th_dport; 1628 break; 1629 #endif /* INET6 */ 1630 default: 1631 goto badsyn; /*sanity*/ 1632 } 1633 1634 if (so->so_options & SO_DEBUG) { 1635 #ifdef TCP_DEBUG 1636 ostate = tp->t_state; 1637 #endif 1638 1639 tcp_saveti = NULL; 1640 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1641 goto nosave; 1642 1643 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1644 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1645 if (!tcp_saveti) 1646 goto nosave; 1647 } else { 1648 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1649 if (!tcp_saveti) 1650 goto nosave; 1651 MCLAIM(m, &tcp_mowner); 1652 tcp_saveti->m_len = iphlen; 1653 m_copydata(m, 0, iphlen, 1654 mtod(tcp_saveti, void *)); 1655 } 1656 1657 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1658 m_freem(tcp_saveti); 1659 tcp_saveti = NULL; 1660 } else { 1661 tcp_saveti->m_len += sizeof(struct tcphdr); 1662 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1663 sizeof(struct tcphdr)); 1664 } 1665 nosave:; 1666 } 1667 if (so->so_options & SO_ACCEPTCONN) { 1668 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1669 if (tiflags & TH_RST) { 1670 syn_cache_reset(&src.sa, &dst.sa, th); 1671 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1672 (TH_ACK|TH_SYN)) { 1673 /* 1674 * Received a SYN,ACK. This should 1675 * never happen while we are in 1676 * LISTEN. Send an RST. 1677 */ 1678 goto badsyn; 1679 } else if (tiflags & TH_ACK) { 1680 so = syn_cache_get(&src.sa, &dst.sa, 1681 th, toff, tlen, so, m); 1682 if (so == NULL) { 1683 /* 1684 * We don't have a SYN for 1685 * this ACK; send an RST. 1686 */ 1687 goto badsyn; 1688 } else if (so == 1689 (struct socket *)(-1)) { 1690 /* 1691 * We were unable to create 1692 * the connection. If the 1693 * 3-way handshake was 1694 * completed, and RST has 1695 * been sent to the peer. 1696 * Since the mbuf might be 1697 * in use for the reply, 1698 * do not free it. 1699 */ 1700 m = NULL; 1701 } else { 1702 /* 1703 * We have created a 1704 * full-blown connection. 1705 */ 1706 tp = NULL; 1707 inp = NULL; 1708 #ifdef INET6 1709 in6p = NULL; 1710 #endif 1711 switch (so->so_proto->pr_domain->dom_family) { 1712 #ifdef INET 1713 case AF_INET: 1714 inp = sotoinpcb(so); 1715 tp = intotcpcb(inp); 1716 break; 1717 #endif 1718 #ifdef INET6 1719 case AF_INET6: 1720 in6p = sotoin6pcb(so); 1721 tp = in6totcpcb(in6p); 1722 break; 1723 #endif 1724 } 1725 if (tp == NULL) 1726 goto badsyn; /*XXX*/ 1727 tiwin <<= tp->snd_scale; 1728 goto after_listen; 1729 } 1730 } else { 1731 /* 1732 * None of RST, SYN or ACK was set. 1733 * This is an invalid packet for a 1734 * TCB in LISTEN state. Send a RST. 1735 */ 1736 goto badsyn; 1737 } 1738 } else { 1739 /* 1740 * Received a SYN. 1741 * 1742 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1743 */ 1744 if (m->m_flags & (M_BCAST|M_MCAST)) 1745 goto drop; 1746 1747 switch (af) { 1748 #ifdef INET6 1749 case AF_INET6: 1750 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1751 goto drop; 1752 break; 1753 #endif /* INET6 */ 1754 case AF_INET: 1755 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1756 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1757 goto drop; 1758 break; 1759 } 1760 1761 #ifdef INET6 1762 /* 1763 * If deprecated address is forbidden, we do 1764 * not accept SYN to deprecated interface 1765 * address to prevent any new inbound 1766 * connection from getting established. 1767 * When we do not accept SYN, we send a TCP 1768 * RST, with deprecated source address (instead 1769 * of dropping it). We compromise it as it is 1770 * much better for peer to send a RST, and 1771 * RST will be the final packet for the 1772 * exchange. 1773 * 1774 * If we do not forbid deprecated addresses, we 1775 * accept the SYN packet. RFC2462 does not 1776 * suggest dropping SYN in this case. 1777 * If we decipher RFC2462 5.5.4, it says like 1778 * this: 1779 * 1. use of deprecated addr with existing 1780 * communication is okay - "SHOULD continue 1781 * to be used" 1782 * 2. use of it with new communication: 1783 * (2a) "SHOULD NOT be used if alternate 1784 * address with sufficient scope is 1785 * available" 1786 * (2b) nothing mentioned otherwise. 1787 * Here we fall into (2b) case as we have no 1788 * choice in our source address selection - we 1789 * must obey the peer. 1790 * 1791 * The wording in RFC2462 is confusing, and 1792 * there are multiple description text for 1793 * deprecated address handling - worse, they 1794 * are not exactly the same. I believe 5.5.4 1795 * is the best one, so we follow 5.5.4. 1796 */ 1797 if (af == AF_INET6 && !ip6_use_deprecated) { 1798 struct in6_ifaddr *ia6; 1799 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1800 &ip6->ip6_dst)) && 1801 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1802 tp = NULL; 1803 goto dropwithreset; 1804 } 1805 } 1806 #endif 1807 1808 #if defined(KAME_IPSEC) || defined(FAST_IPSEC) 1809 switch (af) { 1810 #ifdef INET 1811 case AF_INET: 1812 if (ipsec4_in_reject_so(m, so)) { 1813 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1814 tp = NULL; 1815 goto dropwithreset; 1816 } 1817 break; 1818 #endif 1819 #ifdef INET6 1820 case AF_INET6: 1821 if (ipsec6_in_reject_so(m, so)) { 1822 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1823 tp = NULL; 1824 goto dropwithreset; 1825 } 1826 break; 1827 #endif /*INET6*/ 1828 } 1829 #endif /*IPSEC*/ 1830 1831 /* 1832 * LISTEN socket received a SYN 1833 * from itself? This can't possibly 1834 * be valid; drop the packet. 1835 */ 1836 if (th->th_sport == th->th_dport) { 1837 int i; 1838 1839 switch (af) { 1840 #ifdef INET 1841 case AF_INET: 1842 i = in_hosteq(ip->ip_src, ip->ip_dst); 1843 break; 1844 #endif 1845 #ifdef INET6 1846 case AF_INET6: 1847 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1848 break; 1849 #endif 1850 default: 1851 i = 1; 1852 } 1853 if (i) { 1854 TCP_STATINC(TCP_STAT_BADSYN); 1855 goto drop; 1856 } 1857 } 1858 1859 /* 1860 * SYN looks ok; create compressed TCP 1861 * state for it. 1862 */ 1863 if (so->so_qlen <= so->so_qlimit && 1864 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1865 so, m, optp, optlen, &opti)) 1866 m = NULL; 1867 } 1868 goto drop; 1869 } 1870 } 1871 1872 after_listen: 1873 #ifdef DIAGNOSTIC 1874 /* 1875 * Should not happen now that all embryonic connections 1876 * are handled with compressed state. 1877 */ 1878 if (tp->t_state == TCPS_LISTEN) 1879 panic("tcp_input: TCPS_LISTEN"); 1880 #endif 1881 1882 /* 1883 * Segment received on connection. 1884 * Reset idle time and keep-alive timer. 1885 */ 1886 tp->t_rcvtime = tcp_now; 1887 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1888 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1889 1890 /* 1891 * Process options. 1892 */ 1893 #ifdef TCP_SIGNATURE 1894 if (optp || (tp->t_flags & TF_SIGNATURE)) 1895 #else 1896 if (optp) 1897 #endif 1898 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1899 goto drop; 1900 1901 if (TCP_SACK_ENABLED(tp)) { 1902 tcp_del_sackholes(tp, th); 1903 } 1904 1905 if (TCP_ECN_ALLOWED(tp)) { 1906 if (tiflags & TH_CWR) { 1907 tp->t_flags &= ~TF_ECN_SND_ECE; 1908 } 1909 switch (iptos & IPTOS_ECN_MASK) { 1910 case IPTOS_ECN_CE: 1911 tp->t_flags |= TF_ECN_SND_ECE; 1912 TCP_STATINC(TCP_STAT_ECN_CE); 1913 break; 1914 case IPTOS_ECN_ECT0: 1915 TCP_STATINC(TCP_STAT_ECN_ECT); 1916 break; 1917 case IPTOS_ECN_ECT1: 1918 /* XXX: ignore for now -- rpaulo */ 1919 break; 1920 } 1921 /* 1922 * Congestion experienced. 1923 * Ignore if we are already trying to recover. 1924 */ 1925 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1926 tp->t_congctl->cong_exp(tp); 1927 } 1928 1929 if (opti.ts_present && opti.ts_ecr) { 1930 /* 1931 * Calculate the RTT from the returned time stamp and the 1932 * connection's time base. If the time stamp is later than 1933 * the current time, or is extremely old, fall back to non-1323 1934 * RTT calculation. Since ts_rtt is unsigned, we can test both 1935 * at the same time. 1936 * 1937 * Note that ts_rtt is in units of slow ticks (500 1938 * ms). Since most earthbound RTTs are < 500 ms, 1939 * observed values will have large quantization noise. 1940 * Our smoothed RTT is then the fraction of observed 1941 * samples that are 1 tick instead of 0 (times 500 1942 * ms). 1943 * 1944 * ts_rtt is increased by 1 to denote a valid sample, 1945 * with 0 indicating an invalid measurement. This 1946 * extra 1 must be removed when ts_rtt is used, or 1947 * else an an erroneous extra 500 ms will result. 1948 */ 1949 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1950 if (ts_rtt > TCP_PAWS_IDLE) 1951 ts_rtt = 0; 1952 } else { 1953 ts_rtt = 0; 1954 } 1955 1956 /* 1957 * Header prediction: check for the two common cases 1958 * of a uni-directional data xfer. If the packet has 1959 * no control flags, is in-sequence, the window didn't 1960 * change and we're not retransmitting, it's a 1961 * candidate. If the length is zero and the ack moved 1962 * forward, we're the sender side of the xfer. Just 1963 * free the data acked & wake any higher level process 1964 * that was blocked waiting for space. If the length 1965 * is non-zero and the ack didn't move, we're the 1966 * receiver side. If we're getting packets in-order 1967 * (the reassembly queue is empty), add the data to 1968 * the socket buffer and note that we need a delayed ack. 1969 */ 1970 if (tp->t_state == TCPS_ESTABLISHED && 1971 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1972 == TH_ACK && 1973 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1974 th->th_seq == tp->rcv_nxt && 1975 tiwin && tiwin == tp->snd_wnd && 1976 tp->snd_nxt == tp->snd_max) { 1977 1978 /* 1979 * If last ACK falls within this segment's sequence numbers, 1980 * record the timestamp. 1981 * NOTE that the test is modified according to the latest 1982 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1983 * 1984 * note that we already know 1985 * TSTMP_GEQ(opti.ts_val, tp->ts_recent) 1986 */ 1987 if (opti.ts_present && 1988 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1989 tp->ts_recent_age = tcp_now; 1990 tp->ts_recent = opti.ts_val; 1991 } 1992 1993 if (tlen == 0) { 1994 /* Ack prediction. */ 1995 if (SEQ_GT(th->th_ack, tp->snd_una) && 1996 SEQ_LEQ(th->th_ack, tp->snd_max) && 1997 tp->snd_cwnd >= tp->snd_wnd && 1998 tp->t_partialacks < 0) { 1999 /* 2000 * this is a pure ack for outstanding data. 2001 */ 2002 if (ts_rtt) 2003 tcp_xmit_timer(tp, ts_rtt - 1); 2004 else if (tp->t_rtttime && 2005 SEQ_GT(th->th_ack, tp->t_rtseq)) 2006 tcp_xmit_timer(tp, 2007 tcp_now - tp->t_rtttime); 2008 acked = th->th_ack - tp->snd_una; 2009 tcps = TCP_STAT_GETREF(); 2010 tcps[TCP_STAT_PREDACK]++; 2011 tcps[TCP_STAT_RCVACKPACK]++; 2012 tcps[TCP_STAT_RCVACKBYTE] += acked; 2013 TCP_STAT_PUTREF(); 2014 nd6_hint(tp); 2015 2016 if (acked > (tp->t_lastoff - tp->t_inoff)) 2017 tp->t_lastm = NULL; 2018 sbdrop(&so->so_snd, acked); 2019 tp->t_lastoff -= acked; 2020 2021 icmp_check(tp, th, acked); 2022 2023 tp->snd_una = th->th_ack; 2024 tp->snd_fack = tp->snd_una; 2025 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2026 tp->snd_high = tp->snd_una; 2027 m_freem(m); 2028 2029 /* 2030 * If all outstanding data are acked, stop 2031 * retransmit timer, otherwise restart timer 2032 * using current (possibly backed-off) value. 2033 * If process is waiting for space, 2034 * wakeup/selnotify/signal. If data 2035 * are ready to send, let tcp_output 2036 * decide between more output or persist. 2037 */ 2038 if (tp->snd_una == tp->snd_max) 2039 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2040 else if (TCP_TIMER_ISARMED(tp, 2041 TCPT_PERSIST) == 0) 2042 TCP_TIMER_ARM(tp, TCPT_REXMT, 2043 tp->t_rxtcur); 2044 2045 sowwakeup(so); 2046 if (so->so_snd.sb_cc) { 2047 KERNEL_LOCK(1, NULL); 2048 (void) tcp_output(tp); 2049 KERNEL_UNLOCK_ONE(NULL); 2050 } 2051 if (tcp_saveti) 2052 m_freem(tcp_saveti); 2053 return; 2054 } 2055 } else if (th->th_ack == tp->snd_una && 2056 TAILQ_FIRST(&tp->segq) == NULL && 2057 tlen <= sbspace(&so->so_rcv)) { 2058 int newsize = 0; /* automatic sockbuf scaling */ 2059 2060 /* 2061 * this is a pure, in-sequence data packet 2062 * with nothing on the reassembly queue and 2063 * we have enough buffer space to take it. 2064 */ 2065 tp->rcv_nxt += tlen; 2066 tcps = TCP_STAT_GETREF(); 2067 tcps[TCP_STAT_PREDDAT]++; 2068 tcps[TCP_STAT_RCVPACK]++; 2069 tcps[TCP_STAT_RCVBYTE] += tlen; 2070 TCP_STAT_PUTREF(); 2071 nd6_hint(tp); 2072 2073 /* 2074 * Automatic sizing enables the performance of large buffers 2075 * and most of the efficiency of small ones by only allocating 2076 * space when it is needed. 2077 * 2078 * On the receive side the socket buffer memory is only rarely 2079 * used to any significant extent. This allows us to be much 2080 * more aggressive in scaling the receive socket buffer. For 2081 * the case that the buffer space is actually used to a large 2082 * extent and we run out of kernel memory we can simply drop 2083 * the new segments; TCP on the sender will just retransmit it 2084 * later. Setting the buffer size too big may only consume too 2085 * much kernel memory if the application doesn't read() from 2086 * the socket or packet loss or reordering makes use of the 2087 * reassembly queue. 2088 * 2089 * The criteria to step up the receive buffer one notch are: 2090 * 1. the number of bytes received during the time it takes 2091 * one timestamp to be reflected back to us (the RTT); 2092 * 2. received bytes per RTT is within seven eighth of the 2093 * current socket buffer size; 2094 * 3. receive buffer size has not hit maximal automatic size; 2095 * 2096 * This algorithm does one step per RTT at most and only if 2097 * we receive a bulk stream w/o packet losses or reorderings. 2098 * Shrinking the buffer during idle times is not necessary as 2099 * it doesn't consume any memory when idle. 2100 * 2101 * TODO: Only step up if the application is actually serving 2102 * the buffer to better manage the socket buffer resources. 2103 */ 2104 if (tcp_do_autorcvbuf && 2105 opti.ts_ecr && 2106 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 2107 if (opti.ts_ecr > tp->rfbuf_ts && 2108 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 2109 if (tp->rfbuf_cnt > 2110 (so->so_rcv.sb_hiwat / 8 * 7) && 2111 so->so_rcv.sb_hiwat < 2112 tcp_autorcvbuf_max) { 2113 newsize = 2114 min(so->so_rcv.sb_hiwat + 2115 tcp_autorcvbuf_inc, 2116 tcp_autorcvbuf_max); 2117 } 2118 /* Start over with next RTT. */ 2119 tp->rfbuf_ts = 0; 2120 tp->rfbuf_cnt = 0; 2121 } else 2122 tp->rfbuf_cnt += tlen; /* add up */ 2123 } 2124 2125 /* 2126 * Drop TCP, IP headers and TCP options then add data 2127 * to socket buffer. 2128 */ 2129 if (so->so_state & SS_CANTRCVMORE) 2130 m_freem(m); 2131 else { 2132 /* 2133 * Set new socket buffer size. 2134 * Give up when limit is reached. 2135 */ 2136 if (newsize) 2137 if (!sbreserve(&so->so_rcv, 2138 newsize, so)) 2139 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 2140 m_adj(m, toff + off); 2141 sbappendstream(&so->so_rcv, m); 2142 } 2143 sorwakeup(so); 2144 tcp_setup_ack(tp, th); 2145 if (tp->t_flags & TF_ACKNOW) { 2146 KERNEL_LOCK(1, NULL); 2147 (void) tcp_output(tp); 2148 KERNEL_UNLOCK_ONE(NULL); 2149 } 2150 if (tcp_saveti) 2151 m_freem(tcp_saveti); 2152 return; 2153 } 2154 } 2155 2156 /* 2157 * Compute mbuf offset to TCP data segment. 2158 */ 2159 hdroptlen = toff + off; 2160 2161 /* 2162 * Calculate amount of space in receive window, 2163 * and then do TCP input processing. 2164 * Receive window is amount of space in rcv queue, 2165 * but not less than advertised window. 2166 */ 2167 { int win; 2168 2169 win = sbspace(&so->so_rcv); 2170 if (win < 0) 2171 win = 0; 2172 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 2173 } 2174 2175 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 2176 tp->rfbuf_ts = 0; 2177 tp->rfbuf_cnt = 0; 2178 2179 switch (tp->t_state) { 2180 /* 2181 * If the state is SYN_SENT: 2182 * if seg contains an ACK, but not for our SYN, drop the input. 2183 * if seg contains a RST, then drop the connection. 2184 * if seg does not contain SYN, then drop it. 2185 * Otherwise this is an acceptable SYN segment 2186 * initialize tp->rcv_nxt and tp->irs 2187 * if seg contains ack then advance tp->snd_una 2188 * if seg contains a ECE and ECN support is enabled, the stream 2189 * is ECN capable. 2190 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2191 * arrange for segment to be acked (eventually) 2192 * continue processing rest of data/controls, beginning with URG 2193 */ 2194 case TCPS_SYN_SENT: 2195 if ((tiflags & TH_ACK) && 2196 (SEQ_LEQ(th->th_ack, tp->iss) || 2197 SEQ_GT(th->th_ack, tp->snd_max))) 2198 goto dropwithreset; 2199 if (tiflags & TH_RST) { 2200 if (tiflags & TH_ACK) 2201 tp = tcp_drop(tp, ECONNREFUSED); 2202 goto drop; 2203 } 2204 if ((tiflags & TH_SYN) == 0) 2205 goto drop; 2206 if (tiflags & TH_ACK) { 2207 tp->snd_una = th->th_ack; 2208 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2209 tp->snd_nxt = tp->snd_una; 2210 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2211 tp->snd_high = tp->snd_una; 2212 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2213 2214 if ((tiflags & TH_ECE) && tcp_do_ecn) { 2215 tp->t_flags |= TF_ECN_PERMIT; 2216 TCP_STATINC(TCP_STAT_ECN_SHS); 2217 } 2218 2219 } 2220 tp->irs = th->th_seq; 2221 tcp_rcvseqinit(tp); 2222 tp->t_flags |= TF_ACKNOW; 2223 tcp_mss_from_peer(tp, opti.maxseg); 2224 2225 /* 2226 * Initialize the initial congestion window. If we 2227 * had to retransmit the SYN, we must initialize cwnd 2228 * to 1 segment (i.e. the Loss Window). 2229 */ 2230 if (tp->t_flags & TF_SYN_REXMT) 2231 tp->snd_cwnd = tp->t_peermss; 2232 else { 2233 int ss = tcp_init_win; 2234 #ifdef INET 2235 if (inp != NULL && in_localaddr(inp->inp_faddr)) 2236 ss = tcp_init_win_local; 2237 #endif 2238 #ifdef INET6 2239 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 2240 ss = tcp_init_win_local; 2241 #endif 2242 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 2243 } 2244 2245 tcp_rmx_rtt(tp); 2246 if (tiflags & TH_ACK) { 2247 TCP_STATINC(TCP_STAT_CONNECTS); 2248 /* 2249 * move tcp_established before soisconnected 2250 * because upcall handler can drive tcp_output 2251 * functionality. 2252 * XXX we might call soisconnected at the end of 2253 * all processing 2254 */ 2255 tcp_established(tp); 2256 soisconnected(so); 2257 /* Do window scaling on this connection? */ 2258 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2259 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2260 tp->snd_scale = tp->requested_s_scale; 2261 tp->rcv_scale = tp->request_r_scale; 2262 } 2263 TCP_REASS_LOCK(tp); 2264 (void) tcp_reass(tp, NULL, NULL, &tlen); 2265 /* 2266 * if we didn't have to retransmit the SYN, 2267 * use its rtt as our initial srtt & rtt var. 2268 */ 2269 if (tp->t_rtttime) 2270 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2271 } else 2272 tp->t_state = TCPS_SYN_RECEIVED; 2273 2274 /* 2275 * Advance th->th_seq to correspond to first data byte. 2276 * If data, trim to stay within window, 2277 * dropping FIN if necessary. 2278 */ 2279 th->th_seq++; 2280 if (tlen > tp->rcv_wnd) { 2281 todrop = tlen - tp->rcv_wnd; 2282 m_adj(m, -todrop); 2283 tlen = tp->rcv_wnd; 2284 tiflags &= ~TH_FIN; 2285 tcps = TCP_STAT_GETREF(); 2286 tcps[TCP_STAT_RCVPACKAFTERWIN]++; 2287 tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop; 2288 TCP_STAT_PUTREF(); 2289 } 2290 tp->snd_wl1 = th->th_seq - 1; 2291 tp->rcv_up = th->th_seq; 2292 goto step6; 2293 2294 /* 2295 * If the state is SYN_RECEIVED: 2296 * If seg contains an ACK, but not for our SYN, drop the input 2297 * and generate an RST. See page 36, rfc793 2298 */ 2299 case TCPS_SYN_RECEIVED: 2300 if ((tiflags & TH_ACK) && 2301 (SEQ_LEQ(th->th_ack, tp->iss) || 2302 SEQ_GT(th->th_ack, tp->snd_max))) 2303 goto dropwithreset; 2304 break; 2305 } 2306 2307 /* 2308 * States other than LISTEN or SYN_SENT. 2309 * First check timestamp, if present. 2310 * Then check that at least some bytes of segment are within 2311 * receive window. If segment begins before rcv_nxt, 2312 * drop leading data (and SYN); if nothing left, just ack. 2313 * 2314 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2315 * and it's less than ts_recent, drop it. 2316 */ 2317 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2318 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2319 2320 /* Check to see if ts_recent is over 24 days old. */ 2321 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2322 /* 2323 * Invalidate ts_recent. If this segment updates 2324 * ts_recent, the age will be reset later and ts_recent 2325 * will get a valid value. If it does not, setting 2326 * ts_recent to zero will at least satisfy the 2327 * requirement that zero be placed in the timestamp 2328 * echo reply when ts_recent isn't valid. The 2329 * age isn't reset until we get a valid ts_recent 2330 * because we don't want out-of-order segments to be 2331 * dropped when ts_recent is old. 2332 */ 2333 tp->ts_recent = 0; 2334 } else { 2335 tcps = TCP_STAT_GETREF(); 2336 tcps[TCP_STAT_RCVDUPPACK]++; 2337 tcps[TCP_STAT_RCVDUPBYTE] += tlen; 2338 tcps[TCP_STAT_PAWSDROP]++; 2339 TCP_STAT_PUTREF(); 2340 tcp_new_dsack(tp, th->th_seq, tlen); 2341 goto dropafterack; 2342 } 2343 } 2344 2345 todrop = tp->rcv_nxt - th->th_seq; 2346 dupseg = false; 2347 if (todrop > 0) { 2348 if (tiflags & TH_SYN) { 2349 tiflags &= ~TH_SYN; 2350 th->th_seq++; 2351 if (th->th_urp > 1) 2352 th->th_urp--; 2353 else { 2354 tiflags &= ~TH_URG; 2355 th->th_urp = 0; 2356 } 2357 todrop--; 2358 } 2359 if (todrop > tlen || 2360 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2361 /* 2362 * Any valid FIN or RST must be to the left of the 2363 * window. At this point the FIN or RST must be a 2364 * duplicate or out of sequence; drop it. 2365 */ 2366 if (tiflags & TH_RST) 2367 goto drop; 2368 tiflags &= ~(TH_FIN|TH_RST); 2369 /* 2370 * Send an ACK to resynchronize and drop any data. 2371 * But keep on processing for RST or ACK. 2372 */ 2373 tp->t_flags |= TF_ACKNOW; 2374 todrop = tlen; 2375 dupseg = true; 2376 tcps = TCP_STAT_GETREF(); 2377 tcps[TCP_STAT_RCVDUPPACK]++; 2378 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 2379 TCP_STAT_PUTREF(); 2380 } else if ((tiflags & TH_RST) && 2381 th->th_seq != tp->rcv_nxt) { 2382 /* 2383 * Test for reset before adjusting the sequence 2384 * number for overlapping data. 2385 */ 2386 goto dropafterack_ratelim; 2387 } else { 2388 tcps = TCP_STAT_GETREF(); 2389 tcps[TCP_STAT_RCVPARTDUPPACK]++; 2390 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 2391 TCP_STAT_PUTREF(); 2392 } 2393 tcp_new_dsack(tp, th->th_seq, todrop); 2394 hdroptlen += todrop; /*drop from head afterwards*/ 2395 th->th_seq += todrop; 2396 tlen -= todrop; 2397 if (th->th_urp > todrop) 2398 th->th_urp -= todrop; 2399 else { 2400 tiflags &= ~TH_URG; 2401 th->th_urp = 0; 2402 } 2403 } 2404 2405 /* 2406 * If new data are received on a connection after the 2407 * user processes are gone, then RST the other end. 2408 */ 2409 if ((so->so_state & SS_NOFDREF) && 2410 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2411 tp = tcp_close(tp); 2412 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 2413 goto dropwithreset; 2414 } 2415 2416 /* 2417 * If segment ends after window, drop trailing data 2418 * (and PUSH and FIN); if nothing left, just ACK. 2419 */ 2420 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 2421 if (todrop > 0) { 2422 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 2423 if (todrop >= tlen) { 2424 /* 2425 * The segment actually starts after the window. 2426 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2427 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2428 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2429 */ 2430 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 2431 /* 2432 * If a new connection request is received 2433 * while in TIME_WAIT, drop the old connection 2434 * and start over if the sequence numbers 2435 * are above the previous ones. 2436 * 2437 * NOTE: We will checksum the packet again, and 2438 * so we need to put the header fields back into 2439 * network order! 2440 * XXX This kind of sucks, but we don't expect 2441 * XXX this to happen very often, so maybe it 2442 * XXX doesn't matter so much. 2443 */ 2444 if (tiflags & TH_SYN && 2445 tp->t_state == TCPS_TIME_WAIT && 2446 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2447 tp = tcp_close(tp); 2448 tcp_fields_to_net(th); 2449 goto findpcb; 2450 } 2451 /* 2452 * If window is closed can only take segments at 2453 * window edge, and have to drop data and PUSH from 2454 * incoming segments. Continue processing, but 2455 * remember to ack. Otherwise, drop segment 2456 * and (if not RST) ack. 2457 */ 2458 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2459 tp->t_flags |= TF_ACKNOW; 2460 TCP_STATINC(TCP_STAT_RCVWINPROBE); 2461 } else 2462 goto dropafterack; 2463 } else 2464 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 2465 m_adj(m, -todrop); 2466 tlen -= todrop; 2467 tiflags &= ~(TH_PUSH|TH_FIN); 2468 } 2469 2470 /* 2471 * If last ACK falls within this segment's sequence numbers, 2472 * record the timestamp. 2473 * NOTE: 2474 * 1) That the test incorporates suggestions from the latest 2475 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2476 * 2) That updating only on newer timestamps interferes with 2477 * our earlier PAWS tests, so this check should be solely 2478 * predicated on the sequence space of this segment. 2479 * 3) That we modify the segment boundary check to be 2480 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2481 * instead of RFC1323's 2482 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2483 * This modified check allows us to overcome RFC1323's 2484 * limitations as described in Stevens TCP/IP Illustrated 2485 * Vol. 2 p.869. In such cases, we can still calculate the 2486 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2487 */ 2488 if (opti.ts_present && 2489 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2490 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2491 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2492 tp->ts_recent_age = tcp_now; 2493 tp->ts_recent = opti.ts_val; 2494 } 2495 2496 /* 2497 * If the RST bit is set examine the state: 2498 * SYN_RECEIVED STATE: 2499 * If passive open, return to LISTEN state. 2500 * If active open, inform user that connection was refused. 2501 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 2502 * Inform user that connection was reset, and close tcb. 2503 * CLOSING, LAST_ACK, TIME_WAIT STATES 2504 * Close the tcb. 2505 */ 2506 if (tiflags & TH_RST) { 2507 if (th->th_seq != tp->rcv_nxt) 2508 goto dropafterack_ratelim; 2509 2510 switch (tp->t_state) { 2511 case TCPS_SYN_RECEIVED: 2512 so->so_error = ECONNREFUSED; 2513 goto close; 2514 2515 case TCPS_ESTABLISHED: 2516 case TCPS_FIN_WAIT_1: 2517 case TCPS_FIN_WAIT_2: 2518 case TCPS_CLOSE_WAIT: 2519 so->so_error = ECONNRESET; 2520 close: 2521 tp->t_state = TCPS_CLOSED; 2522 TCP_STATINC(TCP_STAT_DROPS); 2523 tp = tcp_close(tp); 2524 goto drop; 2525 2526 case TCPS_CLOSING: 2527 case TCPS_LAST_ACK: 2528 case TCPS_TIME_WAIT: 2529 tp = tcp_close(tp); 2530 goto drop; 2531 } 2532 } 2533 2534 /* 2535 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2536 * we must be in a synchronized state. RFC791 states (under RST 2537 * generation) that any unacceptable segment (an out-of-order SYN 2538 * qualifies) received in a synchronized state must elicit only an 2539 * empty acknowledgment segment ... and the connection remains in 2540 * the same state. 2541 */ 2542 if (tiflags & TH_SYN) { 2543 if (tp->rcv_nxt == th->th_seq) { 2544 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2545 TH_ACK); 2546 if (tcp_saveti) 2547 m_freem(tcp_saveti); 2548 return; 2549 } 2550 2551 goto dropafterack_ratelim; 2552 } 2553 2554 /* 2555 * If the ACK bit is off we drop the segment and return. 2556 */ 2557 if ((tiflags & TH_ACK) == 0) { 2558 if (tp->t_flags & TF_ACKNOW) 2559 goto dropafterack; 2560 else 2561 goto drop; 2562 } 2563 2564 /* 2565 * Ack processing. 2566 */ 2567 switch (tp->t_state) { 2568 2569 /* 2570 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2571 * ESTABLISHED state and continue processing, otherwise 2572 * send an RST. 2573 */ 2574 case TCPS_SYN_RECEIVED: 2575 if (SEQ_GT(tp->snd_una, th->th_ack) || 2576 SEQ_GT(th->th_ack, tp->snd_max)) 2577 goto dropwithreset; 2578 TCP_STATINC(TCP_STAT_CONNECTS); 2579 soisconnected(so); 2580 tcp_established(tp); 2581 /* Do window scaling? */ 2582 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2583 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2584 tp->snd_scale = tp->requested_s_scale; 2585 tp->rcv_scale = tp->request_r_scale; 2586 } 2587 TCP_REASS_LOCK(tp); 2588 (void) tcp_reass(tp, NULL, NULL, &tlen); 2589 tp->snd_wl1 = th->th_seq - 1; 2590 /* fall into ... */ 2591 2592 /* 2593 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2594 * ACKs. If the ack is in the range 2595 * tp->snd_una < th->th_ack <= tp->snd_max 2596 * then advance tp->snd_una to th->th_ack and drop 2597 * data from the retransmission queue. If this ACK reflects 2598 * more up to date window information we update our window information. 2599 */ 2600 case TCPS_ESTABLISHED: 2601 case TCPS_FIN_WAIT_1: 2602 case TCPS_FIN_WAIT_2: 2603 case TCPS_CLOSE_WAIT: 2604 case TCPS_CLOSING: 2605 case TCPS_LAST_ACK: 2606 case TCPS_TIME_WAIT: 2607 2608 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2609 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2610 TCP_STATINC(TCP_STAT_RCVDUPACK); 2611 /* 2612 * If we have outstanding data (other than 2613 * a window probe), this is a completely 2614 * duplicate ack (ie, window info didn't 2615 * change), the ack is the biggest we've 2616 * seen and we've seen exactly our rexmt 2617 * threshhold of them, assume a packet 2618 * has been dropped and retransmit it. 2619 * Kludge snd_nxt & the congestion 2620 * window so we send only this one 2621 * packet. 2622 */ 2623 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2624 th->th_ack != tp->snd_una) 2625 tp->t_dupacks = 0; 2626 else if (tp->t_partialacks < 0 && 2627 (++tp->t_dupacks == tcprexmtthresh || 2628 TCP_FACK_FASTRECOV(tp))) { 2629 /* 2630 * Do the fast retransmit, and adjust 2631 * congestion control paramenters. 2632 */ 2633 if (tp->t_congctl->fast_retransmit(tp, th)) { 2634 /* False fast retransmit */ 2635 break; 2636 } else 2637 goto drop; 2638 } else if (tp->t_dupacks > tcprexmtthresh) { 2639 tp->snd_cwnd += tp->t_segsz; 2640 KERNEL_LOCK(1, NULL); 2641 (void) tcp_output(tp); 2642 KERNEL_UNLOCK_ONE(NULL); 2643 goto drop; 2644 } 2645 } else { 2646 /* 2647 * If the ack appears to be very old, only 2648 * allow data that is in-sequence. This 2649 * makes it somewhat more difficult to insert 2650 * forged data by guessing sequence numbers. 2651 * Sent an ack to try to update the send 2652 * sequence number on the other side. 2653 */ 2654 if (tlen && th->th_seq != tp->rcv_nxt && 2655 SEQ_LT(th->th_ack, 2656 tp->snd_una - tp->max_sndwnd)) 2657 goto dropafterack; 2658 } 2659 break; 2660 } 2661 /* 2662 * If the congestion window was inflated to account 2663 * for the other side's cached packets, retract it. 2664 */ 2665 /* XXX: make SACK have his own congestion control 2666 * struct -- rpaulo */ 2667 if (TCP_SACK_ENABLED(tp)) 2668 tcp_sack_newack(tp, th); 2669 else 2670 tp->t_congctl->fast_retransmit_newack(tp, th); 2671 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2672 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2673 goto dropafterack; 2674 } 2675 acked = th->th_ack - tp->snd_una; 2676 tcps = TCP_STAT_GETREF(); 2677 tcps[TCP_STAT_RCVACKPACK]++; 2678 tcps[TCP_STAT_RCVACKBYTE] += acked; 2679 TCP_STAT_PUTREF(); 2680 2681 /* 2682 * If we have a timestamp reply, update smoothed 2683 * round trip time. If no timestamp is present but 2684 * transmit timer is running and timed sequence 2685 * number was acked, update smoothed round trip time. 2686 * Since we now have an rtt measurement, cancel the 2687 * timer backoff (cf., Phil Karn's retransmit alg.). 2688 * Recompute the initial retransmit timer. 2689 */ 2690 if (ts_rtt) 2691 tcp_xmit_timer(tp, ts_rtt - 1); 2692 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2693 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2694 2695 /* 2696 * If all outstanding data is acked, stop retransmit 2697 * timer and remember to restart (more output or persist). 2698 * If there is more data to be acked, restart retransmit 2699 * timer, using current (possibly backed-off) value. 2700 */ 2701 if (th->th_ack == tp->snd_max) { 2702 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2703 needoutput = 1; 2704 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2705 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2706 2707 /* 2708 * New data has been acked, adjust the congestion window. 2709 */ 2710 tp->t_congctl->newack(tp, th); 2711 2712 nd6_hint(tp); 2713 if (acked > so->so_snd.sb_cc) { 2714 tp->snd_wnd -= so->so_snd.sb_cc; 2715 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2716 ourfinisacked = 1; 2717 } else { 2718 if (acked > (tp->t_lastoff - tp->t_inoff)) 2719 tp->t_lastm = NULL; 2720 sbdrop(&so->so_snd, acked); 2721 tp->t_lastoff -= acked; 2722 tp->snd_wnd -= acked; 2723 ourfinisacked = 0; 2724 } 2725 sowwakeup(so); 2726 2727 icmp_check(tp, th, acked); 2728 2729 tp->snd_una = th->th_ack; 2730 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2731 tp->snd_fack = tp->snd_una; 2732 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2733 tp->snd_nxt = tp->snd_una; 2734 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2735 tp->snd_high = tp->snd_una; 2736 2737 switch (tp->t_state) { 2738 2739 /* 2740 * In FIN_WAIT_1 STATE in addition to the processing 2741 * for the ESTABLISHED state if our FIN is now acknowledged 2742 * then enter FIN_WAIT_2. 2743 */ 2744 case TCPS_FIN_WAIT_1: 2745 if (ourfinisacked) { 2746 /* 2747 * If we can't receive any more 2748 * data, then closing user can proceed. 2749 * Starting the timer is contrary to the 2750 * specification, but if we don't get a FIN 2751 * we'll hang forever. 2752 */ 2753 if (so->so_state & SS_CANTRCVMORE) { 2754 soisdisconnected(so); 2755 if (tp->t_maxidle > 0) 2756 TCP_TIMER_ARM(tp, TCPT_2MSL, 2757 tp->t_maxidle); 2758 } 2759 tp->t_state = TCPS_FIN_WAIT_2; 2760 } 2761 break; 2762 2763 /* 2764 * In CLOSING STATE in addition to the processing for 2765 * the ESTABLISHED state if the ACK acknowledges our FIN 2766 * then enter the TIME-WAIT state, otherwise ignore 2767 * the segment. 2768 */ 2769 case TCPS_CLOSING: 2770 if (ourfinisacked) { 2771 tp->t_state = TCPS_TIME_WAIT; 2772 tcp_canceltimers(tp); 2773 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2774 soisdisconnected(so); 2775 } 2776 break; 2777 2778 /* 2779 * In LAST_ACK, we may still be waiting for data to drain 2780 * and/or to be acked, as well as for the ack of our FIN. 2781 * If our FIN is now acknowledged, delete the TCB, 2782 * enter the closed state and return. 2783 */ 2784 case TCPS_LAST_ACK: 2785 if (ourfinisacked) { 2786 tp = tcp_close(tp); 2787 goto drop; 2788 } 2789 break; 2790 2791 /* 2792 * In TIME_WAIT state the only thing that should arrive 2793 * is a retransmission of the remote FIN. Acknowledge 2794 * it and restart the finack timer. 2795 */ 2796 case TCPS_TIME_WAIT: 2797 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2798 goto dropafterack; 2799 } 2800 } 2801 2802 step6: 2803 /* 2804 * Update window information. 2805 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2806 */ 2807 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2808 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2809 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2810 /* keep track of pure window updates */ 2811 if (tlen == 0 && 2812 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2813 TCP_STATINC(TCP_STAT_RCVWINUPD); 2814 tp->snd_wnd = tiwin; 2815 tp->snd_wl1 = th->th_seq; 2816 tp->snd_wl2 = th->th_ack; 2817 if (tp->snd_wnd > tp->max_sndwnd) 2818 tp->max_sndwnd = tp->snd_wnd; 2819 needoutput = 1; 2820 } 2821 2822 /* 2823 * Process segments with URG. 2824 */ 2825 if ((tiflags & TH_URG) && th->th_urp && 2826 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2827 /* 2828 * This is a kludge, but if we receive and accept 2829 * random urgent pointers, we'll crash in 2830 * soreceive. It's hard to imagine someone 2831 * actually wanting to send this much urgent data. 2832 */ 2833 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2834 th->th_urp = 0; /* XXX */ 2835 tiflags &= ~TH_URG; /* XXX */ 2836 goto dodata; /* XXX */ 2837 } 2838 /* 2839 * If this segment advances the known urgent pointer, 2840 * then mark the data stream. This should not happen 2841 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2842 * a FIN has been received from the remote side. 2843 * In these states we ignore the URG. 2844 * 2845 * According to RFC961 (Assigned Protocols), 2846 * the urgent pointer points to the last octet 2847 * of urgent data. We continue, however, 2848 * to consider it to indicate the first octet 2849 * of data past the urgent section as the original 2850 * spec states (in one of two places). 2851 */ 2852 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2853 tp->rcv_up = th->th_seq + th->th_urp; 2854 so->so_oobmark = so->so_rcv.sb_cc + 2855 (tp->rcv_up - tp->rcv_nxt) - 1; 2856 if (so->so_oobmark == 0) 2857 so->so_state |= SS_RCVATMARK; 2858 sohasoutofband(so); 2859 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2860 } 2861 /* 2862 * Remove out of band data so doesn't get presented to user. 2863 * This can happen independent of advancing the URG pointer, 2864 * but if two URG's are pending at once, some out-of-band 2865 * data may creep in... ick. 2866 */ 2867 if (th->th_urp <= (u_int16_t) tlen 2868 #ifdef SO_OOBINLINE 2869 && (so->so_options & SO_OOBINLINE) == 0 2870 #endif 2871 ) 2872 tcp_pulloutofband(so, th, m, hdroptlen); 2873 } else 2874 /* 2875 * If no out of band data is expected, 2876 * pull receive urgent pointer along 2877 * with the receive window. 2878 */ 2879 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2880 tp->rcv_up = tp->rcv_nxt; 2881 dodata: /* XXX */ 2882 2883 /* 2884 * Process the segment text, merging it into the TCP sequencing queue, 2885 * and arranging for acknowledgement of receipt if necessary. 2886 * This process logically involves adjusting tp->rcv_wnd as data 2887 * is presented to the user (this happens in tcp_usrreq.c, 2888 * case PRU_RCVD). If a FIN has already been received on this 2889 * connection then we just ignore the text. 2890 */ 2891 if ((tlen || (tiflags & TH_FIN)) && 2892 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2893 /* 2894 * Insert segment ti into reassembly queue of tcp with 2895 * control block tp. Return TH_FIN if reassembly now includes 2896 * a segment with FIN. The macro form does the common case 2897 * inline (segment is the next to be received on an 2898 * established connection, and the queue is empty), 2899 * avoiding linkage into and removal from the queue and 2900 * repetition of various conversions. 2901 * Set DELACK for segments received in order, but ack 2902 * immediately when segments are out of order 2903 * (so fast retransmit can work). 2904 */ 2905 /* NOTE: this was TCP_REASS() macro, but used only once */ 2906 TCP_REASS_LOCK(tp); 2907 if (th->th_seq == tp->rcv_nxt && 2908 TAILQ_FIRST(&tp->segq) == NULL && 2909 tp->t_state == TCPS_ESTABLISHED) { 2910 tcp_setup_ack(tp, th); 2911 tp->rcv_nxt += tlen; 2912 tiflags = th->th_flags & TH_FIN; 2913 tcps = TCP_STAT_GETREF(); 2914 tcps[TCP_STAT_RCVPACK]++; 2915 tcps[TCP_STAT_RCVBYTE] += tlen; 2916 TCP_STAT_PUTREF(); 2917 nd6_hint(tp); 2918 if (so->so_state & SS_CANTRCVMORE) 2919 m_freem(m); 2920 else { 2921 m_adj(m, hdroptlen); 2922 sbappendstream(&(so)->so_rcv, m); 2923 } 2924 TCP_REASS_UNLOCK(tp); 2925 sorwakeup(so); 2926 } else { 2927 m_adj(m, hdroptlen); 2928 tiflags = tcp_reass(tp, th, m, &tlen); 2929 tp->t_flags |= TF_ACKNOW; 2930 } 2931 2932 /* 2933 * Note the amount of data that peer has sent into 2934 * our window, in order to estimate the sender's 2935 * buffer size. 2936 */ 2937 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2938 } else { 2939 m_freem(m); 2940 m = NULL; 2941 tiflags &= ~TH_FIN; 2942 } 2943 2944 /* 2945 * If FIN is received ACK the FIN and let the user know 2946 * that the connection is closing. Ignore a FIN received before 2947 * the connection is fully established. 2948 */ 2949 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2950 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2951 socantrcvmore(so); 2952 tp->t_flags |= TF_ACKNOW; 2953 tp->rcv_nxt++; 2954 } 2955 switch (tp->t_state) { 2956 2957 /* 2958 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2959 */ 2960 case TCPS_ESTABLISHED: 2961 tp->t_state = TCPS_CLOSE_WAIT; 2962 break; 2963 2964 /* 2965 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2966 * enter the CLOSING state. 2967 */ 2968 case TCPS_FIN_WAIT_1: 2969 tp->t_state = TCPS_CLOSING; 2970 break; 2971 2972 /* 2973 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2974 * starting the time-wait timer, turning off the other 2975 * standard timers. 2976 */ 2977 case TCPS_FIN_WAIT_2: 2978 tp->t_state = TCPS_TIME_WAIT; 2979 tcp_canceltimers(tp); 2980 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2981 soisdisconnected(so); 2982 break; 2983 2984 /* 2985 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2986 */ 2987 case TCPS_TIME_WAIT: 2988 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2989 break; 2990 } 2991 } 2992 #ifdef TCP_DEBUG 2993 if (so->so_options & SO_DEBUG) 2994 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2995 #endif 2996 2997 /* 2998 * Return any desired output. 2999 */ 3000 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 3001 KERNEL_LOCK(1, NULL); 3002 (void) tcp_output(tp); 3003 KERNEL_UNLOCK_ONE(NULL); 3004 } 3005 if (tcp_saveti) 3006 m_freem(tcp_saveti); 3007 3008 if (tp->t_state == TCPS_TIME_WAIT 3009 && (so->so_state & SS_NOFDREF) 3010 && (tp->t_inpcb || af != AF_INET) 3011 && (tp->t_in6pcb || af != AF_INET6) 3012 && ((af == AF_INET ? tcp4_vtw_enable : tcp6_vtw_enable) & 1) != 0 3013 && TAILQ_EMPTY(&tp->segq) 3014 && vtw_add(af, tp)) { 3015 ; 3016 } 3017 return; 3018 3019 badsyn: 3020 /* 3021 * Received a bad SYN. Increment counters and dropwithreset. 3022 */ 3023 TCP_STATINC(TCP_STAT_BADSYN); 3024 tp = NULL; 3025 goto dropwithreset; 3026 3027 dropafterack: 3028 /* 3029 * Generate an ACK dropping incoming segment if it occupies 3030 * sequence space, where the ACK reflects our state. 3031 */ 3032 if (tiflags & TH_RST) 3033 goto drop; 3034 goto dropafterack2; 3035 3036 dropafterack_ratelim: 3037 /* 3038 * We may want to rate-limit ACKs against SYN/RST attack. 3039 */ 3040 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 3041 tcp_ackdrop_ppslim) == 0) { 3042 /* XXX stat */ 3043 goto drop; 3044 } 3045 /* ...fall into dropafterack2... */ 3046 3047 dropafterack2: 3048 m_freem(m); 3049 tp->t_flags |= TF_ACKNOW; 3050 KERNEL_LOCK(1, NULL); 3051 (void) tcp_output(tp); 3052 KERNEL_UNLOCK_ONE(NULL); 3053 if (tcp_saveti) 3054 m_freem(tcp_saveti); 3055 return; 3056 3057 dropwithreset_ratelim: 3058 /* 3059 * We may want to rate-limit RSTs in certain situations, 3060 * particularly if we are sending an RST in response to 3061 * an attempt to connect to or otherwise communicate with 3062 * a port for which we have no socket. 3063 */ 3064 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 3065 tcp_rst_ppslim) == 0) { 3066 /* XXX stat */ 3067 goto drop; 3068 } 3069 /* ...fall into dropwithreset... */ 3070 3071 dropwithreset: 3072 /* 3073 * Generate a RST, dropping incoming segment. 3074 * Make ACK acceptable to originator of segment. 3075 */ 3076 if (tiflags & TH_RST) 3077 goto drop; 3078 3079 switch (af) { 3080 #ifdef INET6 3081 case AF_INET6: 3082 /* For following calls to tcp_respond */ 3083 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 3084 goto drop; 3085 break; 3086 #endif /* INET6 */ 3087 case AF_INET: 3088 if (IN_MULTICAST(ip->ip_dst.s_addr) || 3089 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3090 goto drop; 3091 } 3092 3093 if (tiflags & TH_ACK) 3094 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3095 else { 3096 if (tiflags & TH_SYN) 3097 tlen++; 3098 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 3099 TH_RST|TH_ACK); 3100 } 3101 if (tcp_saveti) 3102 m_freem(tcp_saveti); 3103 return; 3104 3105 badcsum: 3106 drop: 3107 /* 3108 * Drop space held by incoming segment and return. 3109 */ 3110 if (tp) { 3111 if (tp->t_inpcb) 3112 so = tp->t_inpcb->inp_socket; 3113 #ifdef INET6 3114 else if (tp->t_in6pcb) 3115 so = tp->t_in6pcb->in6p_socket; 3116 #endif 3117 else 3118 so = NULL; 3119 #ifdef TCP_DEBUG 3120 if (so && (so->so_options & SO_DEBUG) != 0) 3121 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 3122 #endif 3123 } 3124 if (tcp_saveti) 3125 m_freem(tcp_saveti); 3126 m_freem(m); 3127 return; 3128 } 3129 3130 #ifdef TCP_SIGNATURE 3131 int 3132 tcp_signature_apply(void *fstate, void *data, u_int len) 3133 { 3134 3135 MD5Update(fstate, (u_char *)data, len); 3136 return (0); 3137 } 3138 3139 struct secasvar * 3140 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 3141 { 3142 struct secasvar *sav; 3143 #ifdef FAST_IPSEC 3144 union sockaddr_union dst; 3145 #endif 3146 struct ip *ip; 3147 struct ip6_hdr *ip6; 3148 3149 ip = mtod(m, struct ip *); 3150 switch (ip->ip_v) { 3151 case 4: 3152 ip = mtod(m, struct ip *); 3153 ip6 = NULL; 3154 break; 3155 case 6: 3156 ip = NULL; 3157 ip6 = mtod(m, struct ip6_hdr *); 3158 break; 3159 default: 3160 return (NULL); 3161 } 3162 3163 #ifdef FAST_IPSEC 3164 /* Extract the destination from the IP header in the mbuf. */ 3165 memset(&dst, 0, sizeof(union sockaddr_union)); 3166 if (ip !=NULL) { 3167 dst.sa.sa_len = sizeof(struct sockaddr_in); 3168 dst.sa.sa_family = AF_INET; 3169 dst.sin.sin_addr = ip->ip_dst; 3170 } else { 3171 dst.sa.sa_len = sizeof(struct sockaddr_in6); 3172 dst.sa.sa_family = AF_INET6; 3173 dst.sin6.sin6_addr = ip6->ip6_dst; 3174 } 3175 3176 /* 3177 * Look up an SADB entry which matches the address of the peer. 3178 */ 3179 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI), 0, 0); 3180 #else 3181 if (ip) 3182 sav = key_allocsa(AF_INET, (void *)&ip->ip_src, 3183 (void *)&ip->ip_dst, IPPROTO_TCP, 3184 htonl(TCP_SIG_SPI), 0, 0); 3185 else 3186 sav = key_allocsa(AF_INET6, (void *)&ip6->ip6_src, 3187 (void *)&ip6->ip6_dst, IPPROTO_TCP, 3188 htonl(TCP_SIG_SPI), 0, 0); 3189 #endif 3190 3191 return (sav); /* freesav must be performed by caller */ 3192 } 3193 3194 int 3195 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 3196 struct secasvar *sav, char *sig) 3197 { 3198 MD5_CTX ctx; 3199 struct ip *ip; 3200 struct ipovly *ipovly; 3201 struct ip6_hdr *ip6; 3202 struct ippseudo ippseudo; 3203 struct ip6_hdr_pseudo ip6pseudo; 3204 struct tcphdr th0; 3205 int l, tcphdrlen; 3206 3207 if (sav == NULL) 3208 return (-1); 3209 3210 tcphdrlen = th->th_off * 4; 3211 3212 switch (mtod(m, struct ip *)->ip_v) { 3213 case 4: 3214 ip = mtod(m, struct ip *); 3215 ip6 = NULL; 3216 break; 3217 case 6: 3218 ip = NULL; 3219 ip6 = mtod(m, struct ip6_hdr *); 3220 break; 3221 default: 3222 return (-1); 3223 } 3224 3225 MD5Init(&ctx); 3226 3227 if (ip) { 3228 memset(&ippseudo, 0, sizeof(ippseudo)); 3229 ipovly = (struct ipovly *)ip; 3230 ippseudo.ippseudo_src = ipovly->ih_src; 3231 ippseudo.ippseudo_dst = ipovly->ih_dst; 3232 ippseudo.ippseudo_pad = 0; 3233 ippseudo.ippseudo_p = IPPROTO_TCP; 3234 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 3235 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 3236 } else { 3237 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 3238 ip6pseudo.ip6ph_src = ip6->ip6_src; 3239 in6_clearscope(&ip6pseudo.ip6ph_src); 3240 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 3241 in6_clearscope(&ip6pseudo.ip6ph_dst); 3242 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 3243 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 3244 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 3245 } 3246 3247 th0 = *th; 3248 th0.th_sum = 0; 3249 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 3250 3251 l = m->m_pkthdr.len - thoff - tcphdrlen; 3252 if (l > 0) 3253 m_apply(m, thoff + tcphdrlen, 3254 m->m_pkthdr.len - thoff - tcphdrlen, 3255 tcp_signature_apply, &ctx); 3256 3257 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 3258 MD5Final(sig, &ctx); 3259 3260 return (0); 3261 } 3262 #endif 3263 3264 /* 3265 * tcp_dooptions: parse and process tcp options. 3266 * 3267 * returns -1 if this segment should be dropped. (eg. wrong signature) 3268 * otherwise returns 0. 3269 */ 3270 3271 static int 3272 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, 3273 struct tcphdr *th, 3274 struct mbuf *m, int toff, struct tcp_opt_info *oi) 3275 { 3276 u_int16_t mss; 3277 int opt, optlen = 0; 3278 #ifdef TCP_SIGNATURE 3279 void *sigp = NULL; 3280 char sigbuf[TCP_SIGLEN]; 3281 struct secasvar *sav = NULL; 3282 #endif 3283 3284 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 3285 opt = cp[0]; 3286 if (opt == TCPOPT_EOL) 3287 break; 3288 if (opt == TCPOPT_NOP) 3289 optlen = 1; 3290 else { 3291 if (cnt < 2) 3292 break; 3293 optlen = cp[1]; 3294 if (optlen < 2 || optlen > cnt) 3295 break; 3296 } 3297 switch (opt) { 3298 3299 default: 3300 continue; 3301 3302 case TCPOPT_MAXSEG: 3303 if (optlen != TCPOLEN_MAXSEG) 3304 continue; 3305 if (!(th->th_flags & TH_SYN)) 3306 continue; 3307 if (TCPS_HAVERCVDSYN(tp->t_state)) 3308 continue; 3309 bcopy(cp + 2, &mss, sizeof(mss)); 3310 oi->maxseg = ntohs(mss); 3311 break; 3312 3313 case TCPOPT_WINDOW: 3314 if (optlen != TCPOLEN_WINDOW) 3315 continue; 3316 if (!(th->th_flags & TH_SYN)) 3317 continue; 3318 if (TCPS_HAVERCVDSYN(tp->t_state)) 3319 continue; 3320 tp->t_flags |= TF_RCVD_SCALE; 3321 tp->requested_s_scale = cp[2]; 3322 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3323 #if 0 /*XXX*/ 3324 char *p; 3325 3326 if (ip) 3327 p = ntohl(ip->ip_src); 3328 #ifdef INET6 3329 else if (ip6) 3330 p = ip6_sprintf(&ip6->ip6_src); 3331 #endif 3332 else 3333 p = "(unknown)"; 3334 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3335 "assuming %d\n", 3336 tp->requested_s_scale, p, 3337 TCP_MAX_WINSHIFT); 3338 #else 3339 log(LOG_ERR, "TCP: invalid wscale %d, " 3340 "assuming %d\n", 3341 tp->requested_s_scale, 3342 TCP_MAX_WINSHIFT); 3343 #endif 3344 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3345 } 3346 break; 3347 3348 case TCPOPT_TIMESTAMP: 3349 if (optlen != TCPOLEN_TIMESTAMP) 3350 continue; 3351 oi->ts_present = 1; 3352 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 3353 NTOHL(oi->ts_val); 3354 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 3355 NTOHL(oi->ts_ecr); 3356 3357 if (!(th->th_flags & TH_SYN)) 3358 continue; 3359 if (TCPS_HAVERCVDSYN(tp->t_state)) 3360 continue; 3361 /* 3362 * A timestamp received in a SYN makes 3363 * it ok to send timestamp requests and replies. 3364 */ 3365 tp->t_flags |= TF_RCVD_TSTMP; 3366 tp->ts_recent = oi->ts_val; 3367 tp->ts_recent_age = tcp_now; 3368 break; 3369 3370 case TCPOPT_SACK_PERMITTED: 3371 if (optlen != TCPOLEN_SACK_PERMITTED) 3372 continue; 3373 if (!(th->th_flags & TH_SYN)) 3374 continue; 3375 if (TCPS_HAVERCVDSYN(tp->t_state)) 3376 continue; 3377 if (tcp_do_sack) { 3378 tp->t_flags |= TF_SACK_PERMIT; 3379 tp->t_flags |= TF_WILL_SACK; 3380 } 3381 break; 3382 3383 case TCPOPT_SACK: 3384 tcp_sack_option(tp, th, cp, optlen); 3385 break; 3386 #ifdef TCP_SIGNATURE 3387 case TCPOPT_SIGNATURE: 3388 if (optlen != TCPOLEN_SIGNATURE) 3389 continue; 3390 if (sigp && memcmp(sigp, cp + 2, TCP_SIGLEN)) 3391 return (-1); 3392 3393 sigp = sigbuf; 3394 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3395 tp->t_flags |= TF_SIGNATURE; 3396 break; 3397 #endif 3398 } 3399 } 3400 3401 #ifdef TCP_SIGNATURE 3402 if (tp->t_flags & TF_SIGNATURE) { 3403 3404 sav = tcp_signature_getsav(m, th); 3405 3406 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3407 return (-1); 3408 } 3409 3410 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 3411 if (sav == NULL) 3412 return (-1); 3413 #ifdef FAST_IPSEC 3414 KEY_FREESAV(&sav); 3415 #else 3416 key_freesav(sav); 3417 #endif 3418 return (-1); 3419 } 3420 3421 if (sigp) { 3422 char sig[TCP_SIGLEN]; 3423 3424 tcp_fields_to_net(th); 3425 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3426 tcp_fields_to_host(th); 3427 if (sav == NULL) 3428 return (-1); 3429 #ifdef FAST_IPSEC 3430 KEY_FREESAV(&sav); 3431 #else 3432 key_freesav(sav); 3433 #endif 3434 return (-1); 3435 } 3436 tcp_fields_to_host(th); 3437 3438 if (memcmp(sig, sigp, TCP_SIGLEN)) { 3439 TCP_STATINC(TCP_STAT_BADSIG); 3440 if (sav == NULL) 3441 return (-1); 3442 #ifdef FAST_IPSEC 3443 KEY_FREESAV(&sav); 3444 #else 3445 key_freesav(sav); 3446 #endif 3447 return (-1); 3448 } else 3449 TCP_STATINC(TCP_STAT_GOODSIG); 3450 3451 key_sa_recordxfer(sav, m); 3452 #ifdef FAST_IPSEC 3453 KEY_FREESAV(&sav); 3454 #else 3455 key_freesav(sav); 3456 #endif 3457 } 3458 #endif 3459 3460 return (0); 3461 } 3462 3463 /* 3464 * Pull out of band byte out of a segment so 3465 * it doesn't appear in the user's data queue. 3466 * It is still reflected in the segment length for 3467 * sequencing purposes. 3468 */ 3469 void 3470 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3471 struct mbuf *m, int off) 3472 { 3473 int cnt = off + th->th_urp - 1; 3474 3475 while (cnt >= 0) { 3476 if (m->m_len > cnt) { 3477 char *cp = mtod(m, char *) + cnt; 3478 struct tcpcb *tp = sototcpcb(so); 3479 3480 tp->t_iobc = *cp; 3481 tp->t_oobflags |= TCPOOB_HAVEDATA; 3482 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3483 m->m_len--; 3484 return; 3485 } 3486 cnt -= m->m_len; 3487 m = m->m_next; 3488 if (m == 0) 3489 break; 3490 } 3491 panic("tcp_pulloutofband"); 3492 } 3493 3494 /* 3495 * Collect new round-trip time estimate 3496 * and update averages and current timeout. 3497 * 3498 * rtt is in units of slow ticks (typically 500 ms) -- essentially the 3499 * difference of two timestamps. 3500 */ 3501 void 3502 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3503 { 3504 int32_t delta; 3505 3506 TCP_STATINC(TCP_STAT_RTTUPDATED); 3507 if (tp->t_srtt != 0) { 3508 /* 3509 * Compute the amount to add to srtt for smoothing, 3510 * *alpha, or 2^(-TCP_RTT_SHIFT). Because 3511 * srtt is stored in 1/32 slow ticks, we conceptually 3512 * shift left 5 bits, subtract srtt to get the 3513 * diference, and then shift right by TCP_RTT_SHIFT 3514 * (3) to obtain 1/8 of the difference. 3515 */ 3516 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3517 /* 3518 * This can never happen, because delta's lowest 3519 * possible value is 1/8 of t_srtt. But if it does, 3520 * set srtt to some reasonable value, here chosen 3521 * as 1/8 tick. 3522 */ 3523 if ((tp->t_srtt += delta) <= 0) 3524 tp->t_srtt = 1 << 2; 3525 /* 3526 * RFC2988 requires that rttvar be updated first. 3527 * This code is compliant because "delta" is the old 3528 * srtt minus the new observation (scaled). 3529 * 3530 * RFC2988 says: 3531 * rttvar = (1-beta) * rttvar + beta * |srtt-observed| 3532 * 3533 * delta is in units of 1/32 ticks, and has then been 3534 * divided by 8. This is equivalent to being in 1/16s 3535 * units and divided by 4. Subtract from it 1/4 of 3536 * the existing rttvar to form the (signed) amount to 3537 * adjust. 3538 */ 3539 if (delta < 0) 3540 delta = -delta; 3541 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3542 /* 3543 * As with srtt, this should never happen. There is 3544 * no support in RFC2988 for this operation. But 1/4s 3545 * as rttvar when faced with something arguably wrong 3546 * is ok. 3547 */ 3548 if ((tp->t_rttvar += delta) <= 0) 3549 tp->t_rttvar = 1 << 2; 3550 3551 /* 3552 * If srtt exceeds .01 second, ensure we use the 'remote' MSL 3553 * Problem is: it doesn't work. Disabled by defaulting 3554 * tcp_rttlocal to 0; see corresponding code in 3555 * tcp_subr that selects local vs remote in a different way. 3556 * 3557 * The static branch prediction hint here should be removed 3558 * when the rtt estimator is fixed and the rtt_enable code 3559 * is turned back on. 3560 */ 3561 if (__predict_false(tcp_rttlocal) && tcp_msl_enable 3562 && tp->t_srtt > tcp_msl_remote_threshold 3563 && tp->t_msl < tcp_msl_remote) { 3564 tp->t_msl = tcp_msl_remote; 3565 } 3566 } else { 3567 /* 3568 * This is the first measurement. Per RFC2988, 2.2, 3569 * set rtt=R and srtt=R/2. 3570 * For srtt, storage representation is 1/32 ticks, 3571 * so shift left by 5. 3572 * For rttvar, storage representation is 1/16 ticks, 3573 * So shift left by 4, but then right by 1 to halve. 3574 */ 3575 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3576 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3577 } 3578 tp->t_rtttime = 0; 3579 tp->t_rxtshift = 0; 3580 3581 /* 3582 * the retransmit should happen at rtt + 4 * rttvar. 3583 * Because of the way we do the smoothing, srtt and rttvar 3584 * will each average +1/2 tick of bias. When we compute 3585 * the retransmit timer, we want 1/2 tick of rounding and 3586 * 1 extra tick because of +-1/2 tick uncertainty in the 3587 * firing of the timer. The bias will give us exactly the 3588 * 1.5 tick we need. But, because the bias is 3589 * statistical, we have to test that we don't drop below 3590 * the minimum feasible timer (which is 2 ticks). 3591 */ 3592 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3593 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3594 3595 /* 3596 * We received an ack for a packet that wasn't retransmitted; 3597 * it is probably safe to discard any error indications we've 3598 * received recently. This isn't quite right, but close enough 3599 * for now (a route might have failed after we sent a segment, 3600 * and the return path might not be symmetrical). 3601 */ 3602 tp->t_softerror = 0; 3603 } 3604 3605 3606 /* 3607 * TCP compressed state engine. Currently used to hold compressed 3608 * state for SYN_RECEIVED. 3609 */ 3610 3611 u_long syn_cache_count; 3612 u_int32_t syn_hash1, syn_hash2; 3613 3614 #define SYN_HASH(sa, sp, dp) \ 3615 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3616 ((u_int32_t)(sp)))^syn_hash2))) 3617 #ifndef INET6 3618 #define SYN_HASHALL(hash, src, dst) \ 3619 do { \ 3620 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3621 ((const struct sockaddr_in *)(src))->sin_port, \ 3622 ((const struct sockaddr_in *)(dst))->sin_port); \ 3623 } while (/*CONSTCOND*/ 0) 3624 #else 3625 #define SYN_HASH6(sa, sp, dp) \ 3626 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3627 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3628 & 0x7fffffff) 3629 3630 #define SYN_HASHALL(hash, src, dst) \ 3631 do { \ 3632 switch ((src)->sa_family) { \ 3633 case AF_INET: \ 3634 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3635 ((const struct sockaddr_in *)(src))->sin_port, \ 3636 ((const struct sockaddr_in *)(dst))->sin_port); \ 3637 break; \ 3638 case AF_INET6: \ 3639 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3640 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3641 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3642 break; \ 3643 default: \ 3644 hash = 0; \ 3645 } \ 3646 } while (/*CONSTCOND*/0) 3647 #endif /* INET6 */ 3648 3649 static struct pool syn_cache_pool; 3650 3651 /* 3652 * We don't estimate RTT with SYNs, so each packet starts with the default 3653 * RTT and each timer step has a fixed timeout value. 3654 */ 3655 #define SYN_CACHE_TIMER_ARM(sc) \ 3656 do { \ 3657 TCPT_RANGESET((sc)->sc_rxtcur, \ 3658 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3659 TCPTV_REXMTMAX); \ 3660 callout_reset(&(sc)->sc_timer, \ 3661 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3662 } while (/*CONSTCOND*/0) 3663 3664 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3665 3666 static inline void 3667 syn_cache_rm(struct syn_cache *sc) 3668 { 3669 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3670 sc, sc_bucketq); 3671 sc->sc_tp = NULL; 3672 LIST_REMOVE(sc, sc_tpq); 3673 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3674 callout_stop(&sc->sc_timer); 3675 syn_cache_count--; 3676 } 3677 3678 static inline void 3679 syn_cache_put(struct syn_cache *sc) 3680 { 3681 if (sc->sc_ipopts) 3682 (void) m_free(sc->sc_ipopts); 3683 rtcache_free(&sc->sc_route); 3684 sc->sc_flags |= SCF_DEAD; 3685 if (!callout_invoking(&sc->sc_timer)) 3686 callout_schedule(&(sc)->sc_timer, 1); 3687 } 3688 3689 void 3690 syn_cache_init(void) 3691 { 3692 int i; 3693 3694 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 3695 "synpl", NULL, IPL_SOFTNET); 3696 3697 /* Initialize the hash buckets. */ 3698 for (i = 0; i < tcp_syn_cache_size; i++) 3699 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3700 } 3701 3702 void 3703 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3704 { 3705 struct syn_cache_head *scp; 3706 struct syn_cache *sc2; 3707 int s; 3708 3709 /* 3710 * If there are no entries in the hash table, reinitialize 3711 * the hash secrets. 3712 */ 3713 if (syn_cache_count == 0) { 3714 syn_hash1 = cprng_fast32(); 3715 syn_hash2 = cprng_fast32(); 3716 } 3717 3718 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3719 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3720 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3721 3722 /* 3723 * Make sure that we don't overflow the per-bucket 3724 * limit or the total cache size limit. 3725 */ 3726 s = splsoftnet(); 3727 if (scp->sch_length >= tcp_syn_bucket_limit) { 3728 TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW); 3729 /* 3730 * The bucket is full. Toss the oldest element in the 3731 * bucket. This will be the first entry in the bucket. 3732 */ 3733 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3734 #ifdef DIAGNOSTIC 3735 /* 3736 * This should never happen; we should always find an 3737 * entry in our bucket. 3738 */ 3739 if (sc2 == NULL) 3740 panic("syn_cache_insert: bucketoverflow: impossible"); 3741 #endif 3742 syn_cache_rm(sc2); 3743 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3744 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3745 struct syn_cache_head *scp2, *sce; 3746 3747 TCP_STATINC(TCP_STAT_SC_OVERFLOWED); 3748 /* 3749 * The cache is full. Toss the oldest entry in the 3750 * first non-empty bucket we can find. 3751 * 3752 * XXX We would really like to toss the oldest 3753 * entry in the cache, but we hope that this 3754 * condition doesn't happen very often. 3755 */ 3756 scp2 = scp; 3757 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3758 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3759 for (++scp2; scp2 != scp; scp2++) { 3760 if (scp2 >= sce) 3761 scp2 = &tcp_syn_cache[0]; 3762 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3763 break; 3764 } 3765 #ifdef DIAGNOSTIC 3766 /* 3767 * This should never happen; we should always find a 3768 * non-empty bucket. 3769 */ 3770 if (scp2 == scp) 3771 panic("syn_cache_insert: cacheoverflow: " 3772 "impossible"); 3773 #endif 3774 } 3775 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3776 syn_cache_rm(sc2); 3777 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3778 } 3779 3780 /* 3781 * Initialize the entry's timer. 3782 */ 3783 sc->sc_rxttot = 0; 3784 sc->sc_rxtshift = 0; 3785 SYN_CACHE_TIMER_ARM(sc); 3786 3787 /* Link it from tcpcb entry */ 3788 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3789 3790 /* Put it into the bucket. */ 3791 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3792 scp->sch_length++; 3793 syn_cache_count++; 3794 3795 TCP_STATINC(TCP_STAT_SC_ADDED); 3796 splx(s); 3797 } 3798 3799 /* 3800 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3801 * If we have retransmitted an entry the maximum number of times, expire 3802 * that entry. 3803 */ 3804 void 3805 syn_cache_timer(void *arg) 3806 { 3807 struct syn_cache *sc = arg; 3808 3809 mutex_enter(softnet_lock); 3810 KERNEL_LOCK(1, NULL); 3811 callout_ack(&sc->sc_timer); 3812 3813 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3814 TCP_STATINC(TCP_STAT_SC_DELAYED_FREE); 3815 callout_destroy(&sc->sc_timer); 3816 pool_put(&syn_cache_pool, sc); 3817 KERNEL_UNLOCK_ONE(NULL); 3818 mutex_exit(softnet_lock); 3819 return; 3820 } 3821 3822 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3823 /* Drop it -- too many retransmissions. */ 3824 goto dropit; 3825 } 3826 3827 /* 3828 * Compute the total amount of time this entry has 3829 * been on a queue. If this entry has been on longer 3830 * than the keep alive timer would allow, expire it. 3831 */ 3832 sc->sc_rxttot += sc->sc_rxtcur; 3833 if (sc->sc_rxttot >= tcp_keepinit) 3834 goto dropit; 3835 3836 TCP_STATINC(TCP_STAT_SC_RETRANSMITTED); 3837 (void) syn_cache_respond(sc, NULL); 3838 3839 /* Advance the timer back-off. */ 3840 sc->sc_rxtshift++; 3841 SYN_CACHE_TIMER_ARM(sc); 3842 3843 KERNEL_UNLOCK_ONE(NULL); 3844 mutex_exit(softnet_lock); 3845 return; 3846 3847 dropit: 3848 TCP_STATINC(TCP_STAT_SC_TIMED_OUT); 3849 syn_cache_rm(sc); 3850 if (sc->sc_ipopts) 3851 (void) m_free(sc->sc_ipopts); 3852 rtcache_free(&sc->sc_route); 3853 callout_destroy(&sc->sc_timer); 3854 pool_put(&syn_cache_pool, sc); 3855 KERNEL_UNLOCK_ONE(NULL); 3856 mutex_exit(softnet_lock); 3857 } 3858 3859 /* 3860 * Remove syn cache created by the specified tcb entry, 3861 * because this does not make sense to keep them 3862 * (if there's no tcb entry, syn cache entry will never be used) 3863 */ 3864 void 3865 syn_cache_cleanup(struct tcpcb *tp) 3866 { 3867 struct syn_cache *sc, *nsc; 3868 int s; 3869 3870 s = splsoftnet(); 3871 3872 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3873 nsc = LIST_NEXT(sc, sc_tpq); 3874 3875 #ifdef DIAGNOSTIC 3876 if (sc->sc_tp != tp) 3877 panic("invalid sc_tp in syn_cache_cleanup"); 3878 #endif 3879 syn_cache_rm(sc); 3880 syn_cache_put(sc); /* calls pool_put but see spl above */ 3881 } 3882 /* just for safety */ 3883 LIST_INIT(&tp->t_sc); 3884 3885 splx(s); 3886 } 3887 3888 /* 3889 * Find an entry in the syn cache. 3890 */ 3891 struct syn_cache * 3892 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3893 struct syn_cache_head **headp) 3894 { 3895 struct syn_cache *sc; 3896 struct syn_cache_head *scp; 3897 u_int32_t hash; 3898 int s; 3899 3900 SYN_HASHALL(hash, src, dst); 3901 3902 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3903 *headp = scp; 3904 s = splsoftnet(); 3905 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3906 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3907 if (sc->sc_hash != hash) 3908 continue; 3909 if (!memcmp(&sc->sc_src, src, src->sa_len) && 3910 !memcmp(&sc->sc_dst, dst, dst->sa_len)) { 3911 splx(s); 3912 return (sc); 3913 } 3914 } 3915 splx(s); 3916 return (NULL); 3917 } 3918 3919 /* 3920 * This function gets called when we receive an ACK for a 3921 * socket in the LISTEN state. We look up the connection 3922 * in the syn cache, and if its there, we pull it out of 3923 * the cache and turn it into a full-blown connection in 3924 * the SYN-RECEIVED state. 3925 * 3926 * The return values may not be immediately obvious, and their effects 3927 * can be subtle, so here they are: 3928 * 3929 * NULL SYN was not found in cache; caller should drop the 3930 * packet and send an RST. 3931 * 3932 * -1 We were unable to create the new connection, and are 3933 * aborting it. An ACK,RST is being sent to the peer 3934 * (unless we got screwey sequence numbners; see below), 3935 * because the 3-way handshake has been completed. Caller 3936 * should not free the mbuf, since we may be using it. If 3937 * we are not, we will free it. 3938 * 3939 * Otherwise, the return value is a pointer to the new socket 3940 * associated with the connection. 3941 */ 3942 struct socket * 3943 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3944 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3945 struct socket *so, struct mbuf *m) 3946 { 3947 struct syn_cache *sc; 3948 struct syn_cache_head *scp; 3949 struct inpcb *inp = NULL; 3950 #ifdef INET6 3951 struct in6pcb *in6p = NULL; 3952 #endif 3953 struct tcpcb *tp = 0; 3954 struct mbuf *am; 3955 int s; 3956 struct socket *oso; 3957 3958 s = splsoftnet(); 3959 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3960 splx(s); 3961 return (NULL); 3962 } 3963 3964 /* 3965 * Verify the sequence and ack numbers. Try getting the correct 3966 * response again. 3967 */ 3968 if ((th->th_ack != sc->sc_iss + 1) || 3969 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3970 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3971 (void) syn_cache_respond(sc, m); 3972 splx(s); 3973 return ((struct socket *)(-1)); 3974 } 3975 3976 /* Remove this cache entry */ 3977 syn_cache_rm(sc); 3978 splx(s); 3979 3980 /* 3981 * Ok, create the full blown connection, and set things up 3982 * as they would have been set up if we had created the 3983 * connection when the SYN arrived. If we can't create 3984 * the connection, abort it. 3985 */ 3986 /* 3987 * inp still has the OLD in_pcb stuff, set the 3988 * v6-related flags on the new guy, too. This is 3989 * done particularly for the case where an AF_INET6 3990 * socket is bound only to a port, and a v4 connection 3991 * comes in on that port. 3992 * we also copy the flowinfo from the original pcb 3993 * to the new one. 3994 */ 3995 oso = so; 3996 so = sonewconn(so, SS_ISCONNECTED); 3997 if (so == NULL) 3998 goto resetandabort; 3999 4000 switch (so->so_proto->pr_domain->dom_family) { 4001 #ifdef INET 4002 case AF_INET: 4003 inp = sotoinpcb(so); 4004 break; 4005 #endif 4006 #ifdef INET6 4007 case AF_INET6: 4008 in6p = sotoin6pcb(so); 4009 break; 4010 #endif 4011 } 4012 switch (src->sa_family) { 4013 #ifdef INET 4014 case AF_INET: 4015 if (inp) { 4016 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 4017 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 4018 inp->inp_options = ip_srcroute(); 4019 in_pcbstate(inp, INP_BOUND); 4020 if (inp->inp_options == NULL) { 4021 inp->inp_options = sc->sc_ipopts; 4022 sc->sc_ipopts = NULL; 4023 } 4024 } 4025 #ifdef INET6 4026 else if (in6p) { 4027 /* IPv4 packet to AF_INET6 socket */ 4028 memset(&in6p->in6p_laddr, 0, sizeof(in6p->in6p_laddr)); 4029 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 4030 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 4031 &in6p->in6p_laddr.s6_addr32[3], 4032 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 4033 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 4034 in6totcpcb(in6p)->t_family = AF_INET; 4035 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 4036 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 4037 else 4038 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 4039 in6_pcbstate(in6p, IN6P_BOUND); 4040 } 4041 #endif 4042 break; 4043 #endif 4044 #ifdef INET6 4045 case AF_INET6: 4046 if (in6p) { 4047 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 4048 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 4049 in6_pcbstate(in6p, IN6P_BOUND); 4050 } 4051 break; 4052 #endif 4053 } 4054 #ifdef INET6 4055 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 4056 struct in6pcb *oin6p = sotoin6pcb(oso); 4057 /* inherit socket options from the listening socket */ 4058 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 4059 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 4060 m_freem(in6p->in6p_options); 4061 in6p->in6p_options = 0; 4062 } 4063 ip6_savecontrol(in6p, &in6p->in6p_options, 4064 mtod(m, struct ip6_hdr *), m); 4065 } 4066 #endif 4067 4068 #if defined(KAME_IPSEC) || defined(FAST_IPSEC) 4069 /* 4070 * we make a copy of policy, instead of sharing the policy, 4071 * for better behavior in terms of SA lookup and dead SA removal. 4072 */ 4073 if (inp) { 4074 /* copy old policy into new socket's */ 4075 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 4076 printf("tcp_input: could not copy policy\n"); 4077 } 4078 #ifdef INET6 4079 else if (in6p) { 4080 /* copy old policy into new socket's */ 4081 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 4082 in6p->in6p_sp)) 4083 printf("tcp_input: could not copy policy\n"); 4084 } 4085 #endif 4086 #endif 4087 4088 /* 4089 * Give the new socket our cached route reference. 4090 */ 4091 if (inp) { 4092 rtcache_copy(&inp->inp_route, &sc->sc_route); 4093 rtcache_free(&sc->sc_route); 4094 } 4095 #ifdef INET6 4096 else { 4097 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 4098 rtcache_free(&sc->sc_route); 4099 } 4100 #endif 4101 4102 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 4103 if (am == NULL) 4104 goto resetandabort; 4105 MCLAIM(am, &tcp_mowner); 4106 am->m_len = src->sa_len; 4107 bcopy(src, mtod(am, void *), src->sa_len); 4108 if (inp) { 4109 if (in_pcbconnect(inp, am, &lwp0)) { 4110 (void) m_free(am); 4111 goto resetandabort; 4112 } 4113 } 4114 #ifdef INET6 4115 else if (in6p) { 4116 if (src->sa_family == AF_INET) { 4117 /* IPv4 packet to AF_INET6 socket */ 4118 struct sockaddr_in6 *sin6; 4119 sin6 = mtod(am, struct sockaddr_in6 *); 4120 am->m_len = sizeof(*sin6); 4121 memset(sin6, 0, sizeof(*sin6)); 4122 sin6->sin6_family = AF_INET6; 4123 sin6->sin6_len = sizeof(*sin6); 4124 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 4125 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 4126 bcopy(&((struct sockaddr_in *)src)->sin_addr, 4127 &sin6->sin6_addr.s6_addr32[3], 4128 sizeof(sin6->sin6_addr.s6_addr32[3])); 4129 } 4130 if (in6_pcbconnect(in6p, am, NULL)) { 4131 (void) m_free(am); 4132 goto resetandabort; 4133 } 4134 } 4135 #endif 4136 else { 4137 (void) m_free(am); 4138 goto resetandabort; 4139 } 4140 (void) m_free(am); 4141 4142 if (inp) 4143 tp = intotcpcb(inp); 4144 #ifdef INET6 4145 else if (in6p) 4146 tp = in6totcpcb(in6p); 4147 #endif 4148 else 4149 tp = NULL; 4150 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 4151 if (sc->sc_request_r_scale != 15) { 4152 tp->requested_s_scale = sc->sc_requested_s_scale; 4153 tp->request_r_scale = sc->sc_request_r_scale; 4154 tp->snd_scale = sc->sc_requested_s_scale; 4155 tp->rcv_scale = sc->sc_request_r_scale; 4156 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 4157 } 4158 if (sc->sc_flags & SCF_TIMESTAMP) 4159 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 4160 tp->ts_timebase = sc->sc_timebase; 4161 4162 tp->t_template = tcp_template(tp); 4163 if (tp->t_template == 0) { 4164 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 4165 so = NULL; 4166 m_freem(m); 4167 goto abort; 4168 } 4169 4170 tp->iss = sc->sc_iss; 4171 tp->irs = sc->sc_irs; 4172 tcp_sendseqinit(tp); 4173 tcp_rcvseqinit(tp); 4174 tp->t_state = TCPS_SYN_RECEIVED; 4175 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 4176 TCP_STATINC(TCP_STAT_ACCEPTS); 4177 4178 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 4179 tp->t_flags |= TF_WILL_SACK; 4180 4181 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 4182 tp->t_flags |= TF_ECN_PERMIT; 4183 4184 #ifdef TCP_SIGNATURE 4185 if (sc->sc_flags & SCF_SIGNATURE) 4186 tp->t_flags |= TF_SIGNATURE; 4187 #endif 4188 4189 /* Initialize tp->t_ourmss before we deal with the peer's! */ 4190 tp->t_ourmss = sc->sc_ourmaxseg; 4191 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 4192 4193 /* 4194 * Initialize the initial congestion window. If we 4195 * had to retransmit the SYN,ACK, we must initialize cwnd 4196 * to 1 segment (i.e. the Loss Window). 4197 */ 4198 if (sc->sc_rxtshift) 4199 tp->snd_cwnd = tp->t_peermss; 4200 else { 4201 int ss = tcp_init_win; 4202 #ifdef INET 4203 if (inp != NULL && in_localaddr(inp->inp_faddr)) 4204 ss = tcp_init_win_local; 4205 #endif 4206 #ifdef INET6 4207 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 4208 ss = tcp_init_win_local; 4209 #endif 4210 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 4211 } 4212 4213 tcp_rmx_rtt(tp); 4214 tp->snd_wl1 = sc->sc_irs; 4215 tp->rcv_up = sc->sc_irs + 1; 4216 4217 /* 4218 * This is what whould have happened in tcp_output() when 4219 * the SYN,ACK was sent. 4220 */ 4221 tp->snd_up = tp->snd_una; 4222 tp->snd_max = tp->snd_nxt = tp->iss+1; 4223 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 4224 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 4225 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 4226 tp->last_ack_sent = tp->rcv_nxt; 4227 tp->t_partialacks = -1; 4228 tp->t_dupacks = 0; 4229 4230 TCP_STATINC(TCP_STAT_SC_COMPLETED); 4231 s = splsoftnet(); 4232 syn_cache_put(sc); 4233 splx(s); 4234 return (so); 4235 4236 resetandabort: 4237 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 4238 abort: 4239 if (so != NULL) { 4240 (void) soqremque(so, 1); 4241 (void) soabort(so); 4242 mutex_enter(softnet_lock); 4243 } 4244 s = splsoftnet(); 4245 syn_cache_put(sc); 4246 splx(s); 4247 TCP_STATINC(TCP_STAT_SC_ABORTED); 4248 return ((struct socket *)(-1)); 4249 } 4250 4251 /* 4252 * This function is called when we get a RST for a 4253 * non-existent connection, so that we can see if the 4254 * connection is in the syn cache. If it is, zap it. 4255 */ 4256 4257 void 4258 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 4259 { 4260 struct syn_cache *sc; 4261 struct syn_cache_head *scp; 4262 int s = splsoftnet(); 4263 4264 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4265 splx(s); 4266 return; 4267 } 4268 if (SEQ_LT(th->th_seq, sc->sc_irs) || 4269 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 4270 splx(s); 4271 return; 4272 } 4273 syn_cache_rm(sc); 4274 TCP_STATINC(TCP_STAT_SC_RESET); 4275 syn_cache_put(sc); /* calls pool_put but see spl above */ 4276 splx(s); 4277 } 4278 4279 void 4280 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 4281 struct tcphdr *th) 4282 { 4283 struct syn_cache *sc; 4284 struct syn_cache_head *scp; 4285 int s; 4286 4287 s = splsoftnet(); 4288 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4289 splx(s); 4290 return; 4291 } 4292 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 4293 if (ntohl (th->th_seq) != sc->sc_iss) { 4294 splx(s); 4295 return; 4296 } 4297 4298 /* 4299 * If we've retransmitted 3 times and this is our second error, 4300 * we remove the entry. Otherwise, we allow it to continue on. 4301 * This prevents us from incorrectly nuking an entry during a 4302 * spurious network outage. 4303 * 4304 * See tcp_notify(). 4305 */ 4306 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 4307 sc->sc_flags |= SCF_UNREACH; 4308 splx(s); 4309 return; 4310 } 4311 4312 syn_cache_rm(sc); 4313 TCP_STATINC(TCP_STAT_SC_UNREACH); 4314 syn_cache_put(sc); /* calls pool_put but see spl above */ 4315 splx(s); 4316 } 4317 4318 /* 4319 * Given a LISTEN socket and an inbound SYN request, add 4320 * this to the syn cache, and send back a segment: 4321 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 4322 * to the source. 4323 * 4324 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 4325 * Doing so would require that we hold onto the data and deliver it 4326 * to the application. However, if we are the target of a SYN-flood 4327 * DoS attack, an attacker could send data which would eventually 4328 * consume all available buffer space if it were ACKed. By not ACKing 4329 * the data, we avoid this DoS scenario. 4330 */ 4331 4332 int 4333 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 4334 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 4335 int optlen, struct tcp_opt_info *oi) 4336 { 4337 struct tcpcb tb, *tp; 4338 long win; 4339 struct syn_cache *sc; 4340 struct syn_cache_head *scp; 4341 struct mbuf *ipopts; 4342 struct tcp_opt_info opti; 4343 int s; 4344 4345 tp = sototcpcb(so); 4346 4347 memset(&opti, 0, sizeof(opti)); 4348 4349 /* 4350 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 4351 * 4352 * Note this check is performed in tcp_input() very early on. 4353 */ 4354 4355 /* 4356 * Initialize some local state. 4357 */ 4358 win = sbspace(&so->so_rcv); 4359 if (win > TCP_MAXWIN) 4360 win = TCP_MAXWIN; 4361 4362 switch (src->sa_family) { 4363 #ifdef INET 4364 case AF_INET: 4365 /* 4366 * Remember the IP options, if any. 4367 */ 4368 ipopts = ip_srcroute(); 4369 break; 4370 #endif 4371 default: 4372 ipopts = NULL; 4373 } 4374 4375 #ifdef TCP_SIGNATURE 4376 if (optp || (tp->t_flags & TF_SIGNATURE)) 4377 #else 4378 if (optp) 4379 #endif 4380 { 4381 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4382 #ifdef TCP_SIGNATURE 4383 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4384 #endif 4385 tb.t_state = TCPS_LISTEN; 4386 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 4387 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 4388 return (0); 4389 } else 4390 tb.t_flags = 0; 4391 4392 /* 4393 * See if we already have an entry for this connection. 4394 * If we do, resend the SYN,ACK. We do not count this 4395 * as a retransmission (XXX though maybe we should). 4396 */ 4397 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4398 TCP_STATINC(TCP_STAT_SC_DUPESYN); 4399 if (ipopts) { 4400 /* 4401 * If we were remembering a previous source route, 4402 * forget it and use the new one we've been given. 4403 */ 4404 if (sc->sc_ipopts) 4405 (void) m_free(sc->sc_ipopts); 4406 sc->sc_ipopts = ipopts; 4407 } 4408 sc->sc_timestamp = tb.ts_recent; 4409 if (syn_cache_respond(sc, m) == 0) { 4410 uint64_t *tcps = TCP_STAT_GETREF(); 4411 tcps[TCP_STAT_SNDACKS]++; 4412 tcps[TCP_STAT_SNDTOTAL]++; 4413 TCP_STAT_PUTREF(); 4414 } 4415 return (1); 4416 } 4417 4418 s = splsoftnet(); 4419 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4420 splx(s); 4421 if (sc == NULL) { 4422 if (ipopts) 4423 (void) m_free(ipopts); 4424 return (0); 4425 } 4426 4427 /* 4428 * Fill in the cache, and put the necessary IP and TCP 4429 * options into the reply. 4430 */ 4431 memset(sc, 0, sizeof(struct syn_cache)); 4432 callout_init(&sc->sc_timer, CALLOUT_MPSAFE); 4433 bcopy(src, &sc->sc_src, src->sa_len); 4434 bcopy(dst, &sc->sc_dst, dst->sa_len); 4435 sc->sc_flags = 0; 4436 sc->sc_ipopts = ipopts; 4437 sc->sc_irs = th->th_seq; 4438 switch (src->sa_family) { 4439 #ifdef INET 4440 case AF_INET: 4441 { 4442 struct sockaddr_in *srcin = (void *) src; 4443 struct sockaddr_in *dstin = (void *) dst; 4444 4445 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4446 &srcin->sin_addr, dstin->sin_port, 4447 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4448 break; 4449 } 4450 #endif /* INET */ 4451 #ifdef INET6 4452 case AF_INET6: 4453 { 4454 struct sockaddr_in6 *srcin6 = (void *) src; 4455 struct sockaddr_in6 *dstin6 = (void *) dst; 4456 4457 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4458 &srcin6->sin6_addr, dstin6->sin6_port, 4459 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4460 break; 4461 } 4462 #endif /* INET6 */ 4463 } 4464 sc->sc_peermaxseg = oi->maxseg; 4465 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4466 m->m_pkthdr.rcvif : NULL, 4467 sc->sc_src.sa.sa_family); 4468 sc->sc_win = win; 4469 sc->sc_timebase = tcp_now - 1; /* see tcp_newtcpcb() */ 4470 sc->sc_timestamp = tb.ts_recent; 4471 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4472 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4473 sc->sc_flags |= SCF_TIMESTAMP; 4474 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4475 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4476 sc->sc_requested_s_scale = tb.requested_s_scale; 4477 sc->sc_request_r_scale = 0; 4478 /* 4479 * Pick the smallest possible scaling factor that 4480 * will still allow us to scale up to sb_max. 4481 * 4482 * We do this because there are broken firewalls that 4483 * will corrupt the window scale option, leading to 4484 * the other endpoint believing that our advertised 4485 * window is unscaled. At scale factors larger than 4486 * 5 the unscaled window will drop below 1500 bytes, 4487 * leading to serious problems when traversing these 4488 * broken firewalls. 4489 * 4490 * With the default sbmax of 256K, a scale factor 4491 * of 3 will be chosen by this algorithm. Those who 4492 * choose a larger sbmax should watch out 4493 * for the compatiblity problems mentioned above. 4494 * 4495 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4496 * or <SYN,ACK>) segment itself is never scaled. 4497 */ 4498 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4499 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4500 sc->sc_request_r_scale++; 4501 } else { 4502 sc->sc_requested_s_scale = 15; 4503 sc->sc_request_r_scale = 15; 4504 } 4505 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4506 sc->sc_flags |= SCF_SACK_PERMIT; 4507 4508 /* 4509 * ECN setup packet recieved. 4510 */ 4511 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4512 sc->sc_flags |= SCF_ECN_PERMIT; 4513 4514 #ifdef TCP_SIGNATURE 4515 if (tb.t_flags & TF_SIGNATURE) 4516 sc->sc_flags |= SCF_SIGNATURE; 4517 #endif 4518 sc->sc_tp = tp; 4519 if (syn_cache_respond(sc, m) == 0) { 4520 uint64_t *tcps = TCP_STAT_GETREF(); 4521 tcps[TCP_STAT_SNDACKS]++; 4522 tcps[TCP_STAT_SNDTOTAL]++; 4523 TCP_STAT_PUTREF(); 4524 syn_cache_insert(sc, tp); 4525 } else { 4526 s = splsoftnet(); 4527 /* 4528 * syn_cache_put() will try to schedule the timer, so 4529 * we need to initialize it 4530 */ 4531 SYN_CACHE_TIMER_ARM(sc); 4532 syn_cache_put(sc); 4533 splx(s); 4534 TCP_STATINC(TCP_STAT_SC_DROPPED); 4535 } 4536 return (1); 4537 } 4538 4539 int 4540 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4541 { 4542 #ifdef INET6 4543 struct rtentry *rt; 4544 #endif 4545 struct route *ro; 4546 u_int8_t *optp; 4547 int optlen, error; 4548 u_int16_t tlen; 4549 struct ip *ip = NULL; 4550 #ifdef INET6 4551 struct ip6_hdr *ip6 = NULL; 4552 #endif 4553 struct tcpcb *tp = NULL; 4554 struct tcphdr *th; 4555 u_int hlen; 4556 struct socket *so; 4557 4558 ro = &sc->sc_route; 4559 switch (sc->sc_src.sa.sa_family) { 4560 case AF_INET: 4561 hlen = sizeof(struct ip); 4562 break; 4563 #ifdef INET6 4564 case AF_INET6: 4565 hlen = sizeof(struct ip6_hdr); 4566 break; 4567 #endif 4568 default: 4569 if (m) 4570 m_freem(m); 4571 return (EAFNOSUPPORT); 4572 } 4573 4574 /* Compute the size of the TCP options. */ 4575 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4576 ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) + 4577 #ifdef TCP_SIGNATURE 4578 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4579 #endif 4580 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4581 4582 tlen = hlen + sizeof(struct tcphdr) + optlen; 4583 4584 /* 4585 * Create the IP+TCP header from scratch. 4586 */ 4587 if (m) 4588 m_freem(m); 4589 #ifdef DIAGNOSTIC 4590 if (max_linkhdr + tlen > MCLBYTES) 4591 return (ENOBUFS); 4592 #endif 4593 MGETHDR(m, M_DONTWAIT, MT_DATA); 4594 if (m && (max_linkhdr + tlen) > MHLEN) { 4595 MCLGET(m, M_DONTWAIT); 4596 if ((m->m_flags & M_EXT) == 0) { 4597 m_freem(m); 4598 m = NULL; 4599 } 4600 } 4601 if (m == NULL) 4602 return (ENOBUFS); 4603 MCLAIM(m, &tcp_tx_mowner); 4604 4605 /* Fixup the mbuf. */ 4606 m->m_data += max_linkhdr; 4607 m->m_len = m->m_pkthdr.len = tlen; 4608 if (sc->sc_tp) { 4609 tp = sc->sc_tp; 4610 if (tp->t_inpcb) 4611 so = tp->t_inpcb->inp_socket; 4612 #ifdef INET6 4613 else if (tp->t_in6pcb) 4614 so = tp->t_in6pcb->in6p_socket; 4615 #endif 4616 else 4617 so = NULL; 4618 } else 4619 so = NULL; 4620 m->m_pkthdr.rcvif = NULL; 4621 memset(mtod(m, u_char *), 0, tlen); 4622 4623 switch (sc->sc_src.sa.sa_family) { 4624 case AF_INET: 4625 ip = mtod(m, struct ip *); 4626 ip->ip_v = 4; 4627 ip->ip_dst = sc->sc_src.sin.sin_addr; 4628 ip->ip_src = sc->sc_dst.sin.sin_addr; 4629 ip->ip_p = IPPROTO_TCP; 4630 th = (struct tcphdr *)(ip + 1); 4631 th->th_dport = sc->sc_src.sin.sin_port; 4632 th->th_sport = sc->sc_dst.sin.sin_port; 4633 break; 4634 #ifdef INET6 4635 case AF_INET6: 4636 ip6 = mtod(m, struct ip6_hdr *); 4637 ip6->ip6_vfc = IPV6_VERSION; 4638 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4639 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4640 ip6->ip6_nxt = IPPROTO_TCP; 4641 /* ip6_plen will be updated in ip6_output() */ 4642 th = (struct tcphdr *)(ip6 + 1); 4643 th->th_dport = sc->sc_src.sin6.sin6_port; 4644 th->th_sport = sc->sc_dst.sin6.sin6_port; 4645 break; 4646 #endif 4647 default: 4648 th = NULL; 4649 } 4650 4651 th->th_seq = htonl(sc->sc_iss); 4652 th->th_ack = htonl(sc->sc_irs + 1); 4653 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4654 th->th_flags = TH_SYN|TH_ACK; 4655 th->th_win = htons(sc->sc_win); 4656 /* th_sum already 0 */ 4657 /* th_urp already 0 */ 4658 4659 /* Tack on the TCP options. */ 4660 optp = (u_int8_t *)(th + 1); 4661 *optp++ = TCPOPT_MAXSEG; 4662 *optp++ = 4; 4663 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4664 *optp++ = sc->sc_ourmaxseg & 0xff; 4665 4666 if (sc->sc_request_r_scale != 15) { 4667 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4668 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4669 sc->sc_request_r_scale); 4670 optp += 4; 4671 } 4672 4673 if (sc->sc_flags & SCF_TIMESTAMP) { 4674 u_int32_t *lp = (u_int32_t *)(optp); 4675 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4676 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4677 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4678 *lp = htonl(sc->sc_timestamp); 4679 optp += TCPOLEN_TSTAMP_APPA; 4680 } 4681 4682 if (sc->sc_flags & SCF_SACK_PERMIT) { 4683 u_int8_t *p = optp; 4684 4685 /* Let the peer know that we will SACK. */ 4686 p[0] = TCPOPT_SACK_PERMITTED; 4687 p[1] = 2; 4688 p[2] = TCPOPT_NOP; 4689 p[3] = TCPOPT_NOP; 4690 optp += 4; 4691 } 4692 4693 /* 4694 * Send ECN SYN-ACK setup packet. 4695 * Routes can be asymetric, so, even if we receive a packet 4696 * with ECE and CWR set, we must not assume no one will block 4697 * the ECE packet we are about to send. 4698 */ 4699 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4700 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4701 th->th_flags |= TH_ECE; 4702 TCP_STATINC(TCP_STAT_ECN_SHS); 4703 4704 /* 4705 * draft-ietf-tcpm-ecnsyn-00.txt 4706 * 4707 * "[...] a TCP node MAY respond to an ECN-setup 4708 * SYN packet by setting ECT in the responding 4709 * ECN-setup SYN/ACK packet, indicating to routers 4710 * that the SYN/ACK packet is ECN-Capable. 4711 * This allows a congested router along the path 4712 * to mark the packet instead of dropping the 4713 * packet as an indication of congestion." 4714 * 4715 * "[...] There can be a great benefit in setting 4716 * an ECN-capable codepoint in SYN/ACK packets [...] 4717 * Congestion is most likely to occur in 4718 * the server-to-client direction. As a result, 4719 * setting an ECN-capable codepoint in SYN/ACK 4720 * packets can reduce the occurence of three-second 4721 * retransmit timeouts resulting from the drop 4722 * of SYN/ACK packets." 4723 * 4724 * Page 4 and 6, January 2006. 4725 */ 4726 4727 switch (sc->sc_src.sa.sa_family) { 4728 #ifdef INET 4729 case AF_INET: 4730 ip->ip_tos |= IPTOS_ECN_ECT0; 4731 break; 4732 #endif 4733 #ifdef INET6 4734 case AF_INET6: 4735 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4736 break; 4737 #endif 4738 } 4739 TCP_STATINC(TCP_STAT_ECN_ECT); 4740 } 4741 4742 #ifdef TCP_SIGNATURE 4743 if (sc->sc_flags & SCF_SIGNATURE) { 4744 struct secasvar *sav; 4745 u_int8_t *sigp; 4746 4747 sav = tcp_signature_getsav(m, th); 4748 4749 if (sav == NULL) { 4750 if (m) 4751 m_freem(m); 4752 return (EPERM); 4753 } 4754 4755 *optp++ = TCPOPT_SIGNATURE; 4756 *optp++ = TCPOLEN_SIGNATURE; 4757 sigp = optp; 4758 memset(optp, 0, TCP_SIGLEN); 4759 optp += TCP_SIGLEN; 4760 *optp++ = TCPOPT_NOP; 4761 *optp++ = TCPOPT_EOL; 4762 4763 (void)tcp_signature(m, th, hlen, sav, sigp); 4764 4765 key_sa_recordxfer(sav, m); 4766 #ifdef FAST_IPSEC 4767 KEY_FREESAV(&sav); 4768 #else 4769 key_freesav(sav); 4770 #endif 4771 } 4772 #endif 4773 4774 /* Compute the packet's checksum. */ 4775 switch (sc->sc_src.sa.sa_family) { 4776 case AF_INET: 4777 ip->ip_len = htons(tlen - hlen); 4778 th->th_sum = 0; 4779 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4780 break; 4781 #ifdef INET6 4782 case AF_INET6: 4783 ip6->ip6_plen = htons(tlen - hlen); 4784 th->th_sum = 0; 4785 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4786 break; 4787 #endif 4788 } 4789 4790 /* 4791 * Fill in some straggling IP bits. Note the stack expects 4792 * ip_len to be in host order, for convenience. 4793 */ 4794 switch (sc->sc_src.sa.sa_family) { 4795 #ifdef INET 4796 case AF_INET: 4797 ip->ip_len = htons(tlen); 4798 ip->ip_ttl = ip_defttl; 4799 /* XXX tos? */ 4800 break; 4801 #endif 4802 #ifdef INET6 4803 case AF_INET6: 4804 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4805 ip6->ip6_vfc |= IPV6_VERSION; 4806 ip6->ip6_plen = htons(tlen - hlen); 4807 /* ip6_hlim will be initialized afterwards */ 4808 /* XXX flowlabel? */ 4809 break; 4810 #endif 4811 } 4812 4813 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4814 tp = sc->sc_tp; 4815 4816 switch (sc->sc_src.sa.sa_family) { 4817 #ifdef INET 4818 case AF_INET: 4819 error = ip_output(m, sc->sc_ipopts, ro, 4820 (ip_mtudisc ? IP_MTUDISC : 0), 4821 NULL, so); 4822 break; 4823 #endif 4824 #ifdef INET6 4825 case AF_INET6: 4826 ip6->ip6_hlim = in6_selecthlim(NULL, 4827 (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp 4828 : NULL); 4829 4830 error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, so, NULL); 4831 break; 4832 #endif 4833 default: 4834 error = EAFNOSUPPORT; 4835 break; 4836 } 4837 return (error); 4838 } 4839