1 /* $NetBSD: tcp_input.c,v 1.344 2015/08/24 22:21:26 pooka 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 1.344 2015/08/24 22:21:26 pooka Exp $"); 152 153 #ifdef _KERNEL_OPT 154 #include "opt_inet.h" 155 #include "opt_ipsec.h" 156 #include "opt_inet_csum.h" 157 #include "opt_tcp_debug.h" 158 #endif 159 160 #include <sys/param.h> 161 #include <sys/systm.h> 162 #include <sys/malloc.h> 163 #include <sys/mbuf.h> 164 #include <sys/protosw.h> 165 #include <sys/socket.h> 166 #include <sys/socketvar.h> 167 #include <sys/errno.h> 168 #include <sys/syslog.h> 169 #include <sys/pool.h> 170 #include <sys/domain.h> 171 #include <sys/kernel.h> 172 #ifdef TCP_SIGNATURE 173 #include <sys/md5.h> 174 #endif 175 #include <sys/lwp.h> /* for lwp0 */ 176 #include <sys/cprng.h> 177 178 #include <net/if.h> 179 #include <net/if_types.h> 180 181 #include <netinet/in.h> 182 #include <netinet/in_systm.h> 183 #include <netinet/ip.h> 184 #include <netinet/in_pcb.h> 185 #include <netinet/in_var.h> 186 #include <netinet/ip_var.h> 187 #include <netinet/in_offload.h> 188 189 #ifdef INET6 190 #ifndef INET 191 #include <netinet/in.h> 192 #endif 193 #include <netinet/ip6.h> 194 #include <netinet6/ip6_var.h> 195 #include <netinet6/in6_pcb.h> 196 #include <netinet6/ip6_var.h> 197 #include <netinet6/in6_var.h> 198 #include <netinet/icmp6.h> 199 #include <netinet6/nd6.h> 200 #ifdef TCP_SIGNATURE 201 #include <netinet6/scope6_var.h> 202 #endif 203 #endif 204 205 #ifndef INET6 206 /* always need ip6.h for IP6_EXTHDR_GET */ 207 #include <netinet/ip6.h> 208 #endif 209 210 #include <netinet/tcp.h> 211 #include <netinet/tcp_fsm.h> 212 #include <netinet/tcp_seq.h> 213 #include <netinet/tcp_timer.h> 214 #include <netinet/tcp_var.h> 215 #include <netinet/tcp_private.h> 216 #include <netinet/tcpip.h> 217 #include <netinet/tcp_congctl.h> 218 #include <netinet/tcp_debug.h> 219 220 #ifdef INET6 221 #include "faith.h" 222 #if defined(NFAITH) && NFAITH > 0 223 #include <net/if_faith.h> 224 #endif 225 #endif /* INET6 */ 226 227 #ifdef IPSEC 228 #include <netipsec/ipsec.h> 229 #include <netipsec/ipsec_var.h> 230 #include <netipsec/ipsec_private.h> 231 #include <netipsec/key.h> 232 #ifdef INET6 233 #include <netipsec/ipsec6.h> 234 #endif 235 #endif /* IPSEC*/ 236 237 #include <netinet/tcp_vtw.h> 238 239 int tcprexmtthresh = 3; 240 int tcp_log_refused; 241 242 int tcp_do_autorcvbuf = 1; 243 int tcp_autorcvbuf_inc = 16 * 1024; 244 int tcp_autorcvbuf_max = 256 * 1024; 245 int tcp_msl = (TCPTV_MSL / PR_SLOWHZ); 246 247 static int tcp_rst_ppslim_count = 0; 248 static struct timeval tcp_rst_ppslim_last; 249 static int tcp_ackdrop_ppslim_count = 0; 250 static struct timeval tcp_ackdrop_ppslim_last; 251 252 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 253 254 /* for modulo comparisons of timestamps */ 255 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 256 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 257 258 /* 259 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 260 */ 261 #ifdef INET6 262 static inline void 263 nd6_hint(struct tcpcb *tp) 264 { 265 struct rtentry *rt; 266 267 if (tp != NULL && tp->t_in6pcb != NULL && tp->t_family == AF_INET6 && 268 (rt = rtcache_validate(&tp->t_in6pcb->in6p_route)) != NULL) 269 nd6_nud_hint(rt); 270 } 271 #else 272 static inline void 273 nd6_hint(struct tcpcb *tp) 274 { 275 } 276 #endif 277 278 /* 279 * Compute ACK transmission behavior. Delay the ACK unless 280 * we have already delayed an ACK (must send an ACK every two segments). 281 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 282 * option is enabled. 283 */ 284 static void 285 tcp_setup_ack(struct tcpcb *tp, const struct tcphdr *th) 286 { 287 288 if (tp->t_flags & TF_DELACK || 289 (tcp_ack_on_push && th->th_flags & TH_PUSH)) 290 tp->t_flags |= TF_ACKNOW; 291 else 292 TCP_SET_DELACK(tp); 293 } 294 295 static void 296 icmp_check(struct tcpcb *tp, const struct tcphdr *th, int acked) 297 { 298 299 /* 300 * If we had a pending ICMP message that refers to data that have 301 * just been acknowledged, disregard the recorded ICMP message. 302 */ 303 if ((tp->t_flags & TF_PMTUD_PEND) && 304 SEQ_GT(th->th_ack, tp->t_pmtud_th_seq)) 305 tp->t_flags &= ~TF_PMTUD_PEND; 306 307 /* 308 * Keep track of the largest chunk of data 309 * acknowledged since last PMTU update 310 */ 311 if (tp->t_pmtud_mss_acked < acked) 312 tp->t_pmtud_mss_acked = acked; 313 } 314 315 /* 316 * Convert TCP protocol fields to host order for easier processing. 317 */ 318 static void 319 tcp_fields_to_host(struct tcphdr *th) 320 { 321 322 NTOHL(th->th_seq); 323 NTOHL(th->th_ack); 324 NTOHS(th->th_win); 325 NTOHS(th->th_urp); 326 } 327 328 /* 329 * ... and reverse the above. 330 */ 331 static void 332 tcp_fields_to_net(struct tcphdr *th) 333 { 334 335 HTONL(th->th_seq); 336 HTONL(th->th_ack); 337 HTONS(th->th_win); 338 HTONS(th->th_urp); 339 } 340 341 #ifdef TCP_CSUM_COUNTERS 342 #include <sys/device.h> 343 344 #if defined(INET) 345 extern struct evcnt tcp_hwcsum_ok; 346 extern struct evcnt tcp_hwcsum_bad; 347 extern struct evcnt tcp_hwcsum_data; 348 extern struct evcnt tcp_swcsum; 349 #endif /* defined(INET) */ 350 #if defined(INET6) 351 extern struct evcnt tcp6_hwcsum_ok; 352 extern struct evcnt tcp6_hwcsum_bad; 353 extern struct evcnt tcp6_hwcsum_data; 354 extern struct evcnt tcp6_swcsum; 355 #endif /* defined(INET6) */ 356 357 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 358 359 #else 360 361 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 362 363 #endif /* TCP_CSUM_COUNTERS */ 364 365 #ifdef TCP_REASS_COUNTERS 366 #include <sys/device.h> 367 368 extern struct evcnt tcp_reass_; 369 extern struct evcnt tcp_reass_empty; 370 extern struct evcnt tcp_reass_iteration[8]; 371 extern struct evcnt tcp_reass_prependfirst; 372 extern struct evcnt tcp_reass_prepend; 373 extern struct evcnt tcp_reass_insert; 374 extern struct evcnt tcp_reass_inserttail; 375 extern struct evcnt tcp_reass_append; 376 extern struct evcnt tcp_reass_appendtail; 377 extern struct evcnt tcp_reass_overlaptail; 378 extern struct evcnt tcp_reass_overlapfront; 379 extern struct evcnt tcp_reass_segdup; 380 extern struct evcnt tcp_reass_fragdup; 381 382 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 383 384 #else 385 386 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 387 388 #endif /* TCP_REASS_COUNTERS */ 389 390 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *, 391 int *); 392 static int tcp_dooptions(struct tcpcb *, const u_char *, int, 393 struct tcphdr *, struct mbuf *, int, struct tcp_opt_info *); 394 395 #ifdef INET 396 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 397 #endif 398 #ifdef INET6 399 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 400 #endif 401 402 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 403 404 #if defined(MBUFTRACE) 405 struct mowner tcp_reass_mowner = MOWNER_INIT("tcp", "reass"); 406 #endif /* defined(MBUFTRACE) */ 407 408 static struct pool tcpipqent_pool; 409 410 void 411 tcpipqent_init(void) 412 { 413 414 pool_init(&tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", 415 NULL, IPL_VM); 416 } 417 418 struct ipqent * 419 tcpipqent_alloc(void) 420 { 421 struct ipqent *ipqe; 422 int s; 423 424 s = splvm(); 425 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 426 splx(s); 427 428 return ipqe; 429 } 430 431 void 432 tcpipqent_free(struct ipqent *ipqe) 433 { 434 int s; 435 436 s = splvm(); 437 pool_put(&tcpipqent_pool, ipqe); 438 splx(s); 439 } 440 441 static int 442 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int *tlen) 443 { 444 struct ipqent *p, *q, *nq, *tiqe = NULL; 445 struct socket *so = NULL; 446 int pkt_flags; 447 tcp_seq pkt_seq; 448 unsigned pkt_len; 449 u_long rcvpartdupbyte = 0; 450 u_long rcvoobyte; 451 #ifdef TCP_REASS_COUNTERS 452 u_int count = 0; 453 #endif 454 uint64_t *tcps; 455 456 if (tp->t_inpcb) 457 so = tp->t_inpcb->inp_socket; 458 #ifdef INET6 459 else if (tp->t_in6pcb) 460 so = tp->t_in6pcb->in6p_socket; 461 #endif 462 463 TCP_REASS_LOCK_CHECK(tp); 464 465 /* 466 * Call with th==0 after become established to 467 * force pre-ESTABLISHED data up to user socket. 468 */ 469 if (th == 0) 470 goto present; 471 472 m_claimm(m, &tcp_reass_mowner); 473 474 rcvoobyte = *tlen; 475 /* 476 * Copy these to local variables because the tcpiphdr 477 * gets munged while we are collapsing mbufs. 478 */ 479 pkt_seq = th->th_seq; 480 pkt_len = *tlen; 481 pkt_flags = th->th_flags; 482 483 TCP_REASS_COUNTER_INCR(&tcp_reass_); 484 485 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 486 /* 487 * When we miss a packet, the vast majority of time we get 488 * packets that follow it in order. So optimize for that. 489 */ 490 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 491 p->ipqe_len += pkt_len; 492 p->ipqe_flags |= pkt_flags; 493 m_cat(p->ipre_mlast, m); 494 TRAVERSE(p->ipre_mlast); 495 m = NULL; 496 tiqe = p; 497 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 498 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 499 goto skip_replacement; 500 } 501 /* 502 * While we're here, if the pkt is completely beyond 503 * anything we have, just insert it at the tail. 504 */ 505 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 506 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 507 goto insert_it; 508 } 509 } 510 511 q = TAILQ_FIRST(&tp->segq); 512 513 if (q != NULL) { 514 /* 515 * If this segment immediately precedes the first out-of-order 516 * block, simply slap the segment in front of it and (mostly) 517 * skip the complicated logic. 518 */ 519 if (pkt_seq + pkt_len == q->ipqe_seq) { 520 q->ipqe_seq = pkt_seq; 521 q->ipqe_len += pkt_len; 522 q->ipqe_flags |= pkt_flags; 523 m_cat(m, q->ipqe_m); 524 q->ipqe_m = m; 525 q->ipre_mlast = m; /* last mbuf may have changed */ 526 TRAVERSE(q->ipre_mlast); 527 tiqe = q; 528 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 529 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 530 goto skip_replacement; 531 } 532 } else { 533 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 534 } 535 536 /* 537 * Find a segment which begins after this one does. 538 */ 539 for (p = NULL; q != NULL; q = nq) { 540 nq = TAILQ_NEXT(q, ipqe_q); 541 #ifdef TCP_REASS_COUNTERS 542 count++; 543 #endif 544 /* 545 * If the received segment is just right after this 546 * fragment, merge the two together and then check 547 * for further overlaps. 548 */ 549 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 550 #ifdef TCPREASS_DEBUG 551 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 552 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 553 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 554 #endif 555 pkt_len += q->ipqe_len; 556 pkt_flags |= q->ipqe_flags; 557 pkt_seq = q->ipqe_seq; 558 m_cat(q->ipre_mlast, m); 559 TRAVERSE(q->ipre_mlast); 560 m = q->ipqe_m; 561 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 562 goto free_ipqe; 563 } 564 /* 565 * If the received segment is completely past this 566 * fragment, we need to go the next fragment. 567 */ 568 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 569 p = q; 570 continue; 571 } 572 /* 573 * If the fragment is past the received segment, 574 * it (or any following) can't be concatenated. 575 */ 576 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 577 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 578 break; 579 } 580 581 /* 582 * We've received all the data in this segment before. 583 * mark it as a duplicate and return. 584 */ 585 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 586 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 587 tcps = TCP_STAT_GETREF(); 588 tcps[TCP_STAT_RCVDUPPACK]++; 589 tcps[TCP_STAT_RCVDUPBYTE] += pkt_len; 590 TCP_STAT_PUTREF(); 591 tcp_new_dsack(tp, pkt_seq, pkt_len); 592 m_freem(m); 593 if (tiqe != NULL) { 594 tcpipqent_free(tiqe); 595 } 596 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 597 goto out; 598 } 599 /* 600 * Received segment completely overlaps this fragment 601 * so we drop the fragment (this keeps the temporal 602 * ordering of segments correct). 603 */ 604 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 605 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 606 rcvpartdupbyte += q->ipqe_len; 607 m_freem(q->ipqe_m); 608 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 609 goto free_ipqe; 610 } 611 /* 612 * RX'ed segment extends past the end of the 613 * fragment. Drop the overlapping bytes. Then 614 * merge the fragment and segment then treat as 615 * a longer received packet. 616 */ 617 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 618 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 619 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 620 #ifdef TCPREASS_DEBUG 621 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 622 tp, overlap, 623 pkt_seq, pkt_seq + pkt_len, pkt_len); 624 #endif 625 m_adj(m, overlap); 626 rcvpartdupbyte += overlap; 627 m_cat(q->ipre_mlast, m); 628 TRAVERSE(q->ipre_mlast); 629 m = q->ipqe_m; 630 pkt_seq = q->ipqe_seq; 631 pkt_len += q->ipqe_len - overlap; 632 rcvoobyte -= overlap; 633 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 634 goto free_ipqe; 635 } 636 /* 637 * RX'ed segment extends past the front of the 638 * fragment. Drop the overlapping bytes on the 639 * received packet. The packet will then be 640 * contatentated with this fragment a bit later. 641 */ 642 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 643 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 644 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 645 #ifdef TCPREASS_DEBUG 646 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 647 tp, overlap, 648 pkt_seq, pkt_seq + pkt_len, pkt_len); 649 #endif 650 m_adj(m, -overlap); 651 pkt_len -= overlap; 652 rcvpartdupbyte += overlap; 653 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 654 rcvoobyte -= overlap; 655 } 656 /* 657 * If the received segment immediates precedes this 658 * fragment then tack the fragment onto this segment 659 * and reinsert the data. 660 */ 661 if (q->ipqe_seq == pkt_seq + pkt_len) { 662 #ifdef TCPREASS_DEBUG 663 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 664 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 665 pkt_seq, pkt_seq + pkt_len, pkt_len); 666 #endif 667 pkt_len += q->ipqe_len; 668 pkt_flags |= q->ipqe_flags; 669 m_cat(m, q->ipqe_m); 670 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 671 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 672 tp->t_segqlen--; 673 KASSERT(tp->t_segqlen >= 0); 674 KASSERT(tp->t_segqlen != 0 || 675 (TAILQ_EMPTY(&tp->segq) && 676 TAILQ_EMPTY(&tp->timeq))); 677 if (tiqe == NULL) { 678 tiqe = q; 679 } else { 680 tcpipqent_free(q); 681 } 682 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 683 break; 684 } 685 /* 686 * If the fragment is before the segment, remember it. 687 * When this loop is terminated, p will contain the 688 * pointer to fragment that is right before the received 689 * segment. 690 */ 691 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 692 p = q; 693 694 continue; 695 696 /* 697 * This is a common operation. It also will allow 698 * to save doing a malloc/free in most instances. 699 */ 700 free_ipqe: 701 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 702 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 703 tp->t_segqlen--; 704 KASSERT(tp->t_segqlen >= 0); 705 KASSERT(tp->t_segqlen != 0 || 706 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 707 if (tiqe == NULL) { 708 tiqe = q; 709 } else { 710 tcpipqent_free(q); 711 } 712 } 713 714 #ifdef TCP_REASS_COUNTERS 715 if (count > 7) 716 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 717 else if (count > 0) 718 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 719 #endif 720 721 insert_it: 722 723 /* 724 * Allocate a new queue entry since the received segment did not 725 * collapse onto any other out-of-order block; thus we are allocating 726 * a new block. If it had collapsed, tiqe would not be NULL and 727 * we would be reusing it. 728 * XXX If we can't, just drop the packet. XXX 729 */ 730 if (tiqe == NULL) { 731 tiqe = tcpipqent_alloc(); 732 if (tiqe == NULL) { 733 TCP_STATINC(TCP_STAT_RCVMEMDROP); 734 m_freem(m); 735 goto out; 736 } 737 } 738 739 /* 740 * Update the counters. 741 */ 742 tp->t_rcvoopack++; 743 tcps = TCP_STAT_GETREF(); 744 tcps[TCP_STAT_RCVOOPACK]++; 745 tcps[TCP_STAT_RCVOOBYTE] += rcvoobyte; 746 if (rcvpartdupbyte) { 747 tcps[TCP_STAT_RCVPARTDUPPACK]++; 748 tcps[TCP_STAT_RCVPARTDUPBYTE] += rcvpartdupbyte; 749 } 750 TCP_STAT_PUTREF(); 751 752 /* 753 * Insert the new fragment queue entry into both queues. 754 */ 755 tiqe->ipqe_m = m; 756 tiqe->ipre_mlast = m; 757 tiqe->ipqe_seq = pkt_seq; 758 tiqe->ipqe_len = pkt_len; 759 tiqe->ipqe_flags = pkt_flags; 760 if (p == NULL) { 761 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 762 #ifdef TCPREASS_DEBUG 763 if (tiqe->ipqe_seq != tp->rcv_nxt) 764 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 765 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 766 #endif 767 } else { 768 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 769 #ifdef TCPREASS_DEBUG 770 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 771 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 772 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 773 #endif 774 } 775 tp->t_segqlen++; 776 777 skip_replacement: 778 779 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 780 781 present: 782 /* 783 * Present data to user, advancing rcv_nxt through 784 * completed sequence space. 785 */ 786 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 787 goto out; 788 q = TAILQ_FIRST(&tp->segq); 789 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 790 goto out; 791 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 792 goto out; 793 794 tp->rcv_nxt += q->ipqe_len; 795 pkt_flags = q->ipqe_flags & TH_FIN; 796 nd6_hint(tp); 797 798 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 799 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 800 tp->t_segqlen--; 801 KASSERT(tp->t_segqlen >= 0); 802 KASSERT(tp->t_segqlen != 0 || 803 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 804 if (so->so_state & SS_CANTRCVMORE) 805 m_freem(q->ipqe_m); 806 else 807 sbappendstream(&so->so_rcv, q->ipqe_m); 808 tcpipqent_free(q); 809 TCP_REASS_UNLOCK(tp); 810 sorwakeup(so); 811 return (pkt_flags); 812 out: 813 TCP_REASS_UNLOCK(tp); 814 return (0); 815 } 816 817 #ifdef INET6 818 int 819 tcp6_input(struct mbuf **mp, int *offp, int proto) 820 { 821 struct mbuf *m = *mp; 822 823 /* 824 * draft-itojun-ipv6-tcp-to-anycast 825 * better place to put this in? 826 */ 827 if (m->m_flags & M_ANYCAST6) { 828 struct ip6_hdr *ip6; 829 if (m->m_len < sizeof(struct ip6_hdr)) { 830 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 831 TCP_STATINC(TCP_STAT_RCVSHORT); 832 return IPPROTO_DONE; 833 } 834 } 835 ip6 = mtod(m, struct ip6_hdr *); 836 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 837 (char *)&ip6->ip6_dst - (char *)ip6); 838 return IPPROTO_DONE; 839 } 840 841 tcp_input(m, *offp, proto); 842 return IPPROTO_DONE; 843 } 844 #endif 845 846 #ifdef INET 847 static void 848 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 849 { 850 char src[INET_ADDRSTRLEN]; 851 char dst[INET_ADDRSTRLEN]; 852 853 if (ip) { 854 in_print(src, sizeof(src), &ip->ip_src); 855 in_print(dst, sizeof(dst), &ip->ip_dst); 856 } 857 else { 858 strlcpy(src, "(unknown)", sizeof(src)); 859 strlcpy(dst, "(unknown)", sizeof(dst)); 860 } 861 log(LOG_INFO, 862 "Connection attempt to TCP %s:%d from %s:%d\n", 863 dst, ntohs(th->th_dport), 864 src, ntohs(th->th_sport)); 865 } 866 #endif 867 868 #ifdef INET6 869 static void 870 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 871 { 872 char src[INET6_ADDRSTRLEN]; 873 char dst[INET6_ADDRSTRLEN]; 874 875 if (ip6) { 876 in6_print(src, sizeof(src), &ip6->ip6_src); 877 in6_print(dst, sizeof(dst), &ip6->ip6_dst); 878 } 879 else { 880 strlcpy(src, "(unknown v6)", sizeof(src)); 881 strlcpy(dst, "(unknown v6)", sizeof(dst)); 882 } 883 log(LOG_INFO, 884 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 885 dst, ntohs(th->th_dport), 886 src, ntohs(th->th_sport)); 887 } 888 #endif 889 890 /* 891 * Checksum extended TCP header and data. 892 */ 893 int 894 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, 895 int toff, int off, int tlen) 896 { 897 898 /* 899 * XXX it's better to record and check if this mbuf is 900 * already checked. 901 */ 902 903 switch (af) { 904 #ifdef INET 905 case AF_INET: 906 switch (m->m_pkthdr.csum_flags & 907 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 908 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 909 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 910 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 911 goto badcsum; 912 913 case M_CSUM_TCPv4|M_CSUM_DATA: { 914 u_int32_t hw_csum = m->m_pkthdr.csum_data; 915 916 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 917 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 918 const struct ip *ip = 919 mtod(m, const struct ip *); 920 921 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 922 ip->ip_dst.s_addr, 923 htons(hw_csum + tlen + off + IPPROTO_TCP)); 924 } 925 if ((hw_csum ^ 0xffff) != 0) 926 goto badcsum; 927 break; 928 } 929 930 case M_CSUM_TCPv4: 931 /* Checksum was okay. */ 932 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 933 break; 934 935 default: 936 /* 937 * Must compute it ourselves. Maybe skip checksum 938 * on loopback interfaces. 939 */ 940 if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & 941 IFF_LOOPBACK) || 942 tcp_do_loopback_cksum)) { 943 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 944 if (in4_cksum(m, IPPROTO_TCP, toff, 945 tlen + off) != 0) 946 goto badcsum; 947 } 948 break; 949 } 950 break; 951 #endif /* INET4 */ 952 953 #ifdef INET6 954 case AF_INET6: 955 switch (m->m_pkthdr.csum_flags & 956 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 957 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 958 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 959 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 960 goto badcsum; 961 962 #if 0 /* notyet */ 963 case M_CSUM_TCPv6|M_CSUM_DATA: 964 #endif 965 966 case M_CSUM_TCPv6: 967 /* Checksum was okay. */ 968 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 969 break; 970 971 default: 972 /* 973 * Must compute it ourselves. Maybe skip checksum 974 * on loopback interfaces. 975 */ 976 if (__predict_true((m->m_flags & M_LOOP) == 0 || 977 tcp_do_loopback_cksum)) { 978 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 979 if (in6_cksum(m, IPPROTO_TCP, toff, 980 tlen + off) != 0) 981 goto badcsum; 982 } 983 } 984 break; 985 #endif /* INET6 */ 986 } 987 988 return 0; 989 990 badcsum: 991 TCP_STATINC(TCP_STAT_RCVBADSUM); 992 return -1; 993 } 994 995 /* When a packet arrives addressed to a vestigial tcpbp, we 996 * nevertheless have to respond to it per the spec. 997 */ 998 static void tcp_vtw_input(struct tcphdr *th, vestigial_inpcb_t *vp, 999 struct mbuf *m, int tlen, int multicast) 1000 { 1001 int tiflags; 1002 int todrop; 1003 uint32_t t_flags = 0; 1004 uint64_t *tcps; 1005 1006 tiflags = th->th_flags; 1007 todrop = vp->rcv_nxt - th->th_seq; 1008 1009 if (todrop > 0) { 1010 if (tiflags & TH_SYN) { 1011 tiflags &= ~TH_SYN; 1012 ++th->th_seq; 1013 if (th->th_urp > 1) 1014 --th->th_urp; 1015 else { 1016 tiflags &= ~TH_URG; 1017 th->th_urp = 0; 1018 } 1019 --todrop; 1020 } 1021 if (todrop > tlen || 1022 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1023 /* 1024 * Any valid FIN or RST must be to the left of the 1025 * window. At this point the FIN or RST must be a 1026 * duplicate or out of sequence; drop it. 1027 */ 1028 if (tiflags & TH_RST) 1029 goto drop; 1030 tiflags &= ~(TH_FIN|TH_RST); 1031 /* 1032 * Send an ACK to resynchronize and drop any data. 1033 * But keep on processing for RST or ACK. 1034 */ 1035 t_flags |= TF_ACKNOW; 1036 todrop = tlen; 1037 tcps = TCP_STAT_GETREF(); 1038 tcps[TCP_STAT_RCVDUPPACK] += 1; 1039 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 1040 TCP_STAT_PUTREF(); 1041 } else if ((tiflags & TH_RST) 1042 && th->th_seq != vp->rcv_nxt) { 1043 /* 1044 * Test for reset before adjusting the sequence 1045 * number for overlapping data. 1046 */ 1047 goto dropafterack_ratelim; 1048 } else { 1049 tcps = TCP_STAT_GETREF(); 1050 tcps[TCP_STAT_RCVPARTDUPPACK] += 1; 1051 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 1052 TCP_STAT_PUTREF(); 1053 } 1054 1055 // tcp_new_dsack(tp, th->th_seq, todrop); 1056 // hdroptlen += todrop; /*drop from head afterwards*/ 1057 1058 th->th_seq += todrop; 1059 tlen -= todrop; 1060 1061 if (th->th_urp > todrop) 1062 th->th_urp -= todrop; 1063 else { 1064 tiflags &= ~TH_URG; 1065 th->th_urp = 0; 1066 } 1067 } 1068 1069 /* 1070 * If new data are received on a connection after the 1071 * user processes are gone, then RST the other end. 1072 */ 1073 if (tlen) { 1074 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 1075 goto dropwithreset; 1076 } 1077 1078 /* 1079 * If segment ends after window, drop trailing data 1080 * (and PUSH and FIN); if nothing left, just ACK. 1081 */ 1082 todrop = (th->th_seq + tlen) - (vp->rcv_nxt+vp->rcv_wnd); 1083 1084 if (todrop > 0) { 1085 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 1086 if (todrop >= tlen) { 1087 /* 1088 * The segment actually starts after the window. 1089 * th->th_seq + tlen - vp->rcv_nxt - vp->rcv_wnd >= tlen 1090 * th->th_seq - vp->rcv_nxt - vp->rcv_wnd >= 0 1091 * th->th_seq >= vp->rcv_nxt + vp->rcv_wnd 1092 */ 1093 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 1094 /* 1095 * If a new connection request is received 1096 * while in TIME_WAIT, drop the old connection 1097 * and start over if the sequence numbers 1098 * are above the previous ones. 1099 */ 1100 if ((tiflags & TH_SYN) 1101 && SEQ_GT(th->th_seq, vp->rcv_nxt)) { 1102 /* We only support this in the !NOFDREF case, which 1103 * is to say: not here. 1104 */ 1105 goto dropwithreset; 1106 } 1107 /* 1108 * If window is closed can only take segments at 1109 * window edge, and have to drop data and PUSH from 1110 * incoming segments. Continue processing, but 1111 * remember to ack. Otherwise, drop segment 1112 * and (if not RST) ack. 1113 */ 1114 if (vp->rcv_wnd == 0 && th->th_seq == vp->rcv_nxt) { 1115 t_flags |= TF_ACKNOW; 1116 TCP_STATINC(TCP_STAT_RCVWINPROBE); 1117 } else 1118 goto dropafterack; 1119 } else 1120 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 1121 m_adj(m, -todrop); 1122 tlen -= todrop; 1123 tiflags &= ~(TH_PUSH|TH_FIN); 1124 } 1125 1126 if (tiflags & TH_RST) { 1127 if (th->th_seq != vp->rcv_nxt) 1128 goto dropafterack_ratelim; 1129 1130 vtw_del(vp->ctl, vp->vtw); 1131 goto drop; 1132 } 1133 1134 /* 1135 * If the ACK bit is off we drop the segment and return. 1136 */ 1137 if ((tiflags & TH_ACK) == 0) { 1138 if (t_flags & TF_ACKNOW) 1139 goto dropafterack; 1140 else 1141 goto drop; 1142 } 1143 1144 /* 1145 * In TIME_WAIT state the only thing that should arrive 1146 * is a retransmission of the remote FIN. Acknowledge 1147 * it and restart the finack timer. 1148 */ 1149 vtw_restart(vp); 1150 goto dropafterack; 1151 1152 dropafterack: 1153 /* 1154 * Generate an ACK dropping incoming segment if it occupies 1155 * sequence space, where the ACK reflects our state. 1156 */ 1157 if (tiflags & TH_RST) 1158 goto drop; 1159 goto dropafterack2; 1160 1161 dropafterack_ratelim: 1162 /* 1163 * We may want to rate-limit ACKs against SYN/RST attack. 1164 */ 1165 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 1166 tcp_ackdrop_ppslim) == 0) { 1167 /* XXX stat */ 1168 goto drop; 1169 } 1170 /* ...fall into dropafterack2... */ 1171 1172 dropafterack2: 1173 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, th->th_ack, 1174 TH_ACK); 1175 return; 1176 1177 dropwithreset: 1178 /* 1179 * Generate a RST, dropping incoming segment. 1180 * Make ACK acceptable to originator of segment. 1181 */ 1182 if (tiflags & TH_RST) 1183 goto drop; 1184 1185 if (tiflags & TH_ACK) 1186 tcp_respond(0, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 1187 else { 1188 if (tiflags & TH_SYN) 1189 ++tlen; 1190 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, (tcp_seq)0, 1191 TH_RST|TH_ACK); 1192 } 1193 return; 1194 drop: 1195 m_freem(m); 1196 } 1197 1198 /* 1199 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 1200 */ 1201 void 1202 tcp_input(struct mbuf *m, ...) 1203 { 1204 struct tcphdr *th; 1205 struct ip *ip; 1206 struct inpcb *inp; 1207 #ifdef INET6 1208 struct ip6_hdr *ip6; 1209 struct in6pcb *in6p; 1210 #endif 1211 u_int8_t *optp = NULL; 1212 int optlen = 0; 1213 int len, tlen, toff, hdroptlen = 0; 1214 struct tcpcb *tp = 0; 1215 int tiflags; 1216 struct socket *so = NULL; 1217 int todrop, acked, ourfinisacked, needoutput = 0; 1218 bool dupseg; 1219 #ifdef TCP_DEBUG 1220 short ostate = 0; 1221 #endif 1222 u_long tiwin; 1223 struct tcp_opt_info opti; 1224 int off, iphlen; 1225 va_list ap; 1226 int af; /* af on the wire */ 1227 struct mbuf *tcp_saveti = NULL; 1228 uint32_t ts_rtt; 1229 uint8_t iptos; 1230 uint64_t *tcps; 1231 vestigial_inpcb_t vestige; 1232 1233 vestige.valid = 0; 1234 1235 MCLAIM(m, &tcp_rx_mowner); 1236 va_start(ap, m); 1237 toff = va_arg(ap, int); 1238 (void)va_arg(ap, int); /* ignore value, advance ap */ 1239 va_end(ap); 1240 1241 TCP_STATINC(TCP_STAT_RCVTOTAL); 1242 1243 memset(&opti, 0, sizeof(opti)); 1244 opti.ts_present = 0; 1245 opti.maxseg = 0; 1246 1247 /* 1248 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 1249 * 1250 * TCP is, by definition, unicast, so we reject all 1251 * multicast outright. 1252 * 1253 * Note, there are additional src/dst address checks in 1254 * the AF-specific code below. 1255 */ 1256 if (m->m_flags & (M_BCAST|M_MCAST)) { 1257 /* XXX stat */ 1258 goto drop; 1259 } 1260 #ifdef INET6 1261 if (m->m_flags & M_ANYCAST6) { 1262 /* XXX stat */ 1263 goto drop; 1264 } 1265 #endif 1266 1267 /* 1268 * Get IP and TCP header. 1269 * Note: IP leaves IP header in first mbuf. 1270 */ 1271 ip = mtod(m, struct ip *); 1272 switch (ip->ip_v) { 1273 #ifdef INET 1274 case 4: 1275 #ifdef INET6 1276 ip6 = NULL; 1277 #endif 1278 af = AF_INET; 1279 iphlen = sizeof(struct ip); 1280 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1281 sizeof(struct tcphdr)); 1282 if (th == NULL) { 1283 TCP_STATINC(TCP_STAT_RCVSHORT); 1284 return; 1285 } 1286 /* We do the checksum after PCB lookup... */ 1287 len = ntohs(ip->ip_len); 1288 tlen = len - toff; 1289 iptos = ip->ip_tos; 1290 break; 1291 #endif 1292 #ifdef INET6 1293 case 6: 1294 ip = NULL; 1295 iphlen = sizeof(struct ip6_hdr); 1296 af = AF_INET6; 1297 ip6 = mtod(m, struct ip6_hdr *); 1298 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1299 sizeof(struct tcphdr)); 1300 if (th == NULL) { 1301 TCP_STATINC(TCP_STAT_RCVSHORT); 1302 return; 1303 } 1304 1305 /* Be proactive about malicious use of IPv4 mapped address */ 1306 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 1307 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 1308 /* XXX stat */ 1309 goto drop; 1310 } 1311 1312 /* 1313 * Be proactive about unspecified IPv6 address in source. 1314 * As we use all-zero to indicate unbounded/unconnected pcb, 1315 * unspecified IPv6 address can be used to confuse us. 1316 * 1317 * Note that packets with unspecified IPv6 destination is 1318 * already dropped in ip6_input. 1319 */ 1320 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1321 /* XXX stat */ 1322 goto drop; 1323 } 1324 1325 /* 1326 * Make sure destination address is not multicast. 1327 * Source address checked in ip6_input(). 1328 */ 1329 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1330 /* XXX stat */ 1331 goto drop; 1332 } 1333 1334 /* We do the checksum after PCB lookup... */ 1335 len = m->m_pkthdr.len; 1336 tlen = len - toff; 1337 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1338 break; 1339 #endif 1340 default: 1341 m_freem(m); 1342 return; 1343 } 1344 1345 KASSERT(TCP_HDR_ALIGNED_P(th)); 1346 1347 /* 1348 * Check that TCP offset makes sense, 1349 * pull out TCP options and adjust length. XXX 1350 */ 1351 off = th->th_off << 2; 1352 if (off < sizeof (struct tcphdr) || off > tlen) { 1353 TCP_STATINC(TCP_STAT_RCVBADOFF); 1354 goto drop; 1355 } 1356 tlen -= off; 1357 1358 /* 1359 * tcp_input() has been modified to use tlen to mean the TCP data 1360 * length throughout the function. Other functions can use 1361 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1362 * rja 1363 */ 1364 1365 if (off > sizeof (struct tcphdr)) { 1366 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1367 if (th == NULL) { 1368 TCP_STATINC(TCP_STAT_RCVSHORT); 1369 return; 1370 } 1371 /* 1372 * NOTE: ip/ip6 will not be affected by m_pulldown() 1373 * (as they're before toff) and we don't need to update those. 1374 */ 1375 KASSERT(TCP_HDR_ALIGNED_P(th)); 1376 optlen = off - sizeof (struct tcphdr); 1377 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1378 /* 1379 * Do quick retrieval of timestamp options ("options 1380 * prediction?"). If timestamp is the only option and it's 1381 * formatted as recommended in RFC 1323 appendix A, we 1382 * quickly get the values now and not bother calling 1383 * tcp_dooptions(), etc. 1384 */ 1385 if ((optlen == TCPOLEN_TSTAMP_APPA || 1386 (optlen > TCPOLEN_TSTAMP_APPA && 1387 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1388 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1389 (th->th_flags & TH_SYN) == 0) { 1390 opti.ts_present = 1; 1391 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1392 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1393 optp = NULL; /* we've parsed the options */ 1394 } 1395 } 1396 tiflags = th->th_flags; 1397 1398 /* 1399 * Checksum extended TCP header and data 1400 */ 1401 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1402 goto badcsum; 1403 1404 /* 1405 * Locate pcb for segment. 1406 */ 1407 findpcb: 1408 inp = NULL; 1409 #ifdef INET6 1410 in6p = NULL; 1411 #endif 1412 switch (af) { 1413 #ifdef INET 1414 case AF_INET: 1415 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1416 ip->ip_dst, th->th_dport, 1417 &vestige); 1418 if (inp == 0 && !vestige.valid) { 1419 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1420 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1421 } 1422 #ifdef INET6 1423 if (inp == 0 && !vestige.valid) { 1424 struct in6_addr s, d; 1425 1426 /* mapped addr case */ 1427 memset(&s, 0, sizeof(s)); 1428 s.s6_addr16[5] = htons(0xffff); 1429 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1430 memset(&d, 0, sizeof(d)); 1431 d.s6_addr16[5] = htons(0xffff); 1432 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1433 in6p = in6_pcblookup_connect(&tcbtable, &s, 1434 th->th_sport, &d, th->th_dport, 1435 0, &vestige); 1436 if (in6p == 0 && !vestige.valid) { 1437 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1438 in6p = in6_pcblookup_bind(&tcbtable, &d, 1439 th->th_dport, 0); 1440 } 1441 } 1442 #endif 1443 #ifndef INET6 1444 if (inp == 0 && !vestige.valid) 1445 #else 1446 if (inp == 0 && in6p == 0 && !vestige.valid) 1447 #endif 1448 { 1449 TCP_STATINC(TCP_STAT_NOPORT); 1450 if (tcp_log_refused && 1451 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1452 tcp4_log_refused(ip, th); 1453 } 1454 tcp_fields_to_host(th); 1455 goto dropwithreset_ratelim; 1456 } 1457 #if defined(IPSEC) 1458 if (ipsec_used) { 1459 if (inp && 1460 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 1461 && ipsec4_in_reject(m, inp)) { 1462 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1463 goto drop; 1464 } 1465 #ifdef INET6 1466 else if (in6p && 1467 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 1468 && ipsec6_in_reject_so(m, in6p->in6p_socket)) { 1469 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1470 goto drop; 1471 } 1472 #endif 1473 } 1474 #endif /*IPSEC*/ 1475 break; 1476 #endif /*INET*/ 1477 #ifdef INET6 1478 case AF_INET6: 1479 { 1480 int faith; 1481 1482 #if defined(NFAITH) && NFAITH > 0 1483 faith = faithprefix(&ip6->ip6_dst); 1484 #else 1485 faith = 0; 1486 #endif 1487 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1488 th->th_sport, &ip6->ip6_dst, th->th_dport, faith, &vestige); 1489 if (!in6p && !vestige.valid) { 1490 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1491 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1492 th->th_dport, faith); 1493 } 1494 if (!in6p && !vestige.valid) { 1495 TCP_STATINC(TCP_STAT_NOPORT); 1496 if (tcp_log_refused && 1497 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1498 tcp6_log_refused(ip6, th); 1499 } 1500 tcp_fields_to_host(th); 1501 goto dropwithreset_ratelim; 1502 } 1503 #if defined(IPSEC) 1504 if (ipsec_used && in6p 1505 && (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 1506 && ipsec6_in_reject(m, in6p)) { 1507 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1508 goto drop; 1509 } 1510 #endif /*IPSEC*/ 1511 break; 1512 } 1513 #endif 1514 } 1515 1516 /* 1517 * If the state is CLOSED (i.e., TCB does not exist) then 1518 * all data in the incoming segment is discarded. 1519 * If the TCB exists but is in CLOSED state, it is embryonic, 1520 * but should either do a listen or a connect soon. 1521 */ 1522 tp = NULL; 1523 so = NULL; 1524 if (inp) { 1525 /* Check the minimum TTL for socket. */ 1526 if (ip->ip_ttl < inp->inp_ip_minttl) 1527 goto drop; 1528 1529 tp = intotcpcb(inp); 1530 so = inp->inp_socket; 1531 } 1532 #ifdef INET6 1533 else if (in6p) { 1534 tp = in6totcpcb(in6p); 1535 so = in6p->in6p_socket; 1536 } 1537 #endif 1538 else if (vestige.valid) { 1539 int mc = 0; 1540 1541 /* We do not support the resurrection of vtw tcpcps. 1542 */ 1543 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1544 goto badcsum; 1545 1546 switch (af) { 1547 #ifdef INET6 1548 case AF_INET6: 1549 mc = IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst); 1550 break; 1551 #endif 1552 1553 case AF_INET: 1554 mc = (IN_MULTICAST(ip->ip_dst.s_addr) 1555 || in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)); 1556 break; 1557 } 1558 1559 tcp_fields_to_host(th); 1560 tcp_vtw_input(th, &vestige, m, tlen, mc); 1561 m = 0; 1562 goto drop; 1563 } 1564 1565 if (tp == 0) { 1566 tcp_fields_to_host(th); 1567 goto dropwithreset_ratelim; 1568 } 1569 if (tp->t_state == TCPS_CLOSED) 1570 goto drop; 1571 1572 KASSERT(so->so_lock == softnet_lock); 1573 KASSERT(solocked(so)); 1574 1575 tcp_fields_to_host(th); 1576 1577 /* Unscale the window into a 32-bit value. */ 1578 if ((tiflags & TH_SYN) == 0) 1579 tiwin = th->th_win << tp->snd_scale; 1580 else 1581 tiwin = th->th_win; 1582 1583 #ifdef INET6 1584 /* save packet options if user wanted */ 1585 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1586 if (in6p->in6p_options) { 1587 m_freem(in6p->in6p_options); 1588 in6p->in6p_options = 0; 1589 } 1590 KASSERT(ip6 != NULL); 1591 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1592 } 1593 #endif 1594 1595 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1596 union syn_cache_sa src; 1597 union syn_cache_sa dst; 1598 1599 memset(&src, 0, sizeof(src)); 1600 memset(&dst, 0, sizeof(dst)); 1601 switch (af) { 1602 #ifdef INET 1603 case AF_INET: 1604 src.sin.sin_len = sizeof(struct sockaddr_in); 1605 src.sin.sin_family = AF_INET; 1606 src.sin.sin_addr = ip->ip_src; 1607 src.sin.sin_port = th->th_sport; 1608 1609 dst.sin.sin_len = sizeof(struct sockaddr_in); 1610 dst.sin.sin_family = AF_INET; 1611 dst.sin.sin_addr = ip->ip_dst; 1612 dst.sin.sin_port = th->th_dport; 1613 break; 1614 #endif 1615 #ifdef INET6 1616 case AF_INET6: 1617 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1618 src.sin6.sin6_family = AF_INET6; 1619 src.sin6.sin6_addr = ip6->ip6_src; 1620 src.sin6.sin6_port = th->th_sport; 1621 1622 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1623 dst.sin6.sin6_family = AF_INET6; 1624 dst.sin6.sin6_addr = ip6->ip6_dst; 1625 dst.sin6.sin6_port = th->th_dport; 1626 break; 1627 #endif /* INET6 */ 1628 default: 1629 goto badsyn; /*sanity*/ 1630 } 1631 1632 if (so->so_options & SO_DEBUG) { 1633 #ifdef TCP_DEBUG 1634 ostate = tp->t_state; 1635 #endif 1636 1637 tcp_saveti = NULL; 1638 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1639 goto nosave; 1640 1641 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1642 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1643 if (!tcp_saveti) 1644 goto nosave; 1645 } else { 1646 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1647 if (!tcp_saveti) 1648 goto nosave; 1649 MCLAIM(m, &tcp_mowner); 1650 tcp_saveti->m_len = iphlen; 1651 m_copydata(m, 0, iphlen, 1652 mtod(tcp_saveti, void *)); 1653 } 1654 1655 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1656 m_freem(tcp_saveti); 1657 tcp_saveti = NULL; 1658 } else { 1659 tcp_saveti->m_len += sizeof(struct tcphdr); 1660 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1661 sizeof(struct tcphdr)); 1662 } 1663 nosave:; 1664 } 1665 if (so->so_options & SO_ACCEPTCONN) { 1666 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1667 if (tiflags & TH_RST) { 1668 syn_cache_reset(&src.sa, &dst.sa, th); 1669 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1670 (TH_ACK|TH_SYN)) { 1671 /* 1672 * Received a SYN,ACK. This should 1673 * never happen while we are in 1674 * LISTEN. Send an RST. 1675 */ 1676 goto badsyn; 1677 } else if (tiflags & TH_ACK) { 1678 so = syn_cache_get(&src.sa, &dst.sa, 1679 th, toff, tlen, so, m); 1680 if (so == NULL) { 1681 /* 1682 * We don't have a SYN for 1683 * this ACK; send an RST. 1684 */ 1685 goto badsyn; 1686 } else if (so == 1687 (struct socket *)(-1)) { 1688 /* 1689 * We were unable to create 1690 * the connection. If the 1691 * 3-way handshake was 1692 * completed, and RST has 1693 * been sent to the peer. 1694 * Since the mbuf might be 1695 * in use for the reply, 1696 * do not free it. 1697 */ 1698 m = NULL; 1699 } else { 1700 /* 1701 * We have created a 1702 * full-blown connection. 1703 */ 1704 tp = NULL; 1705 inp = NULL; 1706 #ifdef INET6 1707 in6p = NULL; 1708 #endif 1709 switch (so->so_proto->pr_domain->dom_family) { 1710 #ifdef INET 1711 case AF_INET: 1712 inp = sotoinpcb(so); 1713 tp = intotcpcb(inp); 1714 break; 1715 #endif 1716 #ifdef INET6 1717 case AF_INET6: 1718 in6p = sotoin6pcb(so); 1719 tp = in6totcpcb(in6p); 1720 break; 1721 #endif 1722 } 1723 if (tp == NULL) 1724 goto badsyn; /*XXX*/ 1725 tiwin <<= tp->snd_scale; 1726 goto after_listen; 1727 } 1728 } else { 1729 /* 1730 * None of RST, SYN or ACK was set. 1731 * This is an invalid packet for a 1732 * TCB in LISTEN state. Send a RST. 1733 */ 1734 goto badsyn; 1735 } 1736 } else { 1737 /* 1738 * Received a SYN. 1739 * 1740 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1741 */ 1742 if (m->m_flags & (M_BCAST|M_MCAST)) 1743 goto drop; 1744 1745 switch (af) { 1746 #ifdef INET6 1747 case AF_INET6: 1748 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1749 goto drop; 1750 break; 1751 #endif /* INET6 */ 1752 case AF_INET: 1753 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1754 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1755 goto drop; 1756 break; 1757 } 1758 1759 #ifdef INET6 1760 /* 1761 * If deprecated address is forbidden, we do 1762 * not accept SYN to deprecated interface 1763 * address to prevent any new inbound 1764 * connection from getting established. 1765 * When we do not accept SYN, we send a TCP 1766 * RST, with deprecated source address (instead 1767 * of dropping it). We compromise it as it is 1768 * much better for peer to send a RST, and 1769 * RST will be the final packet for the 1770 * exchange. 1771 * 1772 * If we do not forbid deprecated addresses, we 1773 * accept the SYN packet. RFC2462 does not 1774 * suggest dropping SYN in this case. 1775 * If we decipher RFC2462 5.5.4, it says like 1776 * this: 1777 * 1. use of deprecated addr with existing 1778 * communication is okay - "SHOULD continue 1779 * to be used" 1780 * 2. use of it with new communication: 1781 * (2a) "SHOULD NOT be used if alternate 1782 * address with sufficient scope is 1783 * available" 1784 * (2b) nothing mentioned otherwise. 1785 * Here we fall into (2b) case as we have no 1786 * choice in our source address selection - we 1787 * must obey the peer. 1788 * 1789 * The wording in RFC2462 is confusing, and 1790 * there are multiple description text for 1791 * deprecated address handling - worse, they 1792 * are not exactly the same. I believe 5.5.4 1793 * is the best one, so we follow 5.5.4. 1794 */ 1795 if (af == AF_INET6 && !ip6_use_deprecated) { 1796 struct in6_ifaddr *ia6; 1797 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1798 &ip6->ip6_dst)) && 1799 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1800 tp = NULL; 1801 goto dropwithreset; 1802 } 1803 } 1804 #endif 1805 1806 #if defined(IPSEC) 1807 if (ipsec_used) { 1808 switch (af) { 1809 #ifdef INET 1810 case AF_INET: 1811 if (!ipsec4_in_reject_so(m, so)) 1812 break; 1813 IPSEC_STATINC( 1814 IPSEC_STAT_IN_POLVIO); 1815 tp = NULL; 1816 goto dropwithreset; 1817 #endif 1818 #ifdef INET6 1819 case AF_INET6: 1820 if (!ipsec6_in_reject_so(m, so)) 1821 break; 1822 IPSEC6_STATINC( 1823 IPSEC_STAT_IN_POLVIO); 1824 tp = NULL; 1825 goto dropwithreset; 1826 #endif /*INET6*/ 1827 } 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 tp->t_congctl->fast_retransmit_newack(tp, th); 2666 2667 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2668 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2669 goto dropafterack; 2670 } 2671 acked = th->th_ack - tp->snd_una; 2672 tcps = TCP_STAT_GETREF(); 2673 tcps[TCP_STAT_RCVACKPACK]++; 2674 tcps[TCP_STAT_RCVACKBYTE] += acked; 2675 TCP_STAT_PUTREF(); 2676 2677 /* 2678 * If we have a timestamp reply, update smoothed 2679 * round trip time. If no timestamp is present but 2680 * transmit timer is running and timed sequence 2681 * number was acked, update smoothed round trip time. 2682 * Since we now have an rtt measurement, cancel the 2683 * timer backoff (cf., Phil Karn's retransmit alg.). 2684 * Recompute the initial retransmit timer. 2685 */ 2686 if (ts_rtt) 2687 tcp_xmit_timer(tp, ts_rtt - 1); 2688 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2689 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2690 2691 /* 2692 * If all outstanding data is acked, stop retransmit 2693 * timer and remember to restart (more output or persist). 2694 * If there is more data to be acked, restart retransmit 2695 * timer, using current (possibly backed-off) value. 2696 */ 2697 if (th->th_ack == tp->snd_max) { 2698 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2699 needoutput = 1; 2700 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2701 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2702 2703 /* 2704 * New data has been acked, adjust the congestion window. 2705 */ 2706 tp->t_congctl->newack(tp, th); 2707 2708 nd6_hint(tp); 2709 if (acked > so->so_snd.sb_cc) { 2710 tp->snd_wnd -= so->so_snd.sb_cc; 2711 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2712 ourfinisacked = 1; 2713 } else { 2714 if (acked > (tp->t_lastoff - tp->t_inoff)) 2715 tp->t_lastm = NULL; 2716 sbdrop(&so->so_snd, acked); 2717 tp->t_lastoff -= acked; 2718 if (tp->snd_wnd > acked) 2719 tp->snd_wnd -= acked; 2720 else 2721 tp->snd_wnd = 0; 2722 ourfinisacked = 0; 2723 } 2724 sowwakeup(so); 2725 2726 icmp_check(tp, th, acked); 2727 2728 tp->snd_una = th->th_ack; 2729 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2730 tp->snd_fack = tp->snd_una; 2731 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2732 tp->snd_nxt = tp->snd_una; 2733 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2734 tp->snd_high = tp->snd_una; 2735 2736 switch (tp->t_state) { 2737 2738 /* 2739 * In FIN_WAIT_1 STATE in addition to the processing 2740 * for the ESTABLISHED state if our FIN is now acknowledged 2741 * then enter FIN_WAIT_2. 2742 */ 2743 case TCPS_FIN_WAIT_1: 2744 if (ourfinisacked) { 2745 /* 2746 * If we can't receive any more 2747 * data, then closing user can proceed. 2748 * Starting the timer is contrary to the 2749 * specification, but if we don't get a FIN 2750 * we'll hang forever. 2751 */ 2752 if (so->so_state & SS_CANTRCVMORE) { 2753 soisdisconnected(so); 2754 if (tp->t_maxidle > 0) 2755 TCP_TIMER_ARM(tp, TCPT_2MSL, 2756 tp->t_maxidle); 2757 } 2758 tp->t_state = TCPS_FIN_WAIT_2; 2759 } 2760 break; 2761 2762 /* 2763 * In CLOSING STATE in addition to the processing for 2764 * the ESTABLISHED state if the ACK acknowledges our FIN 2765 * then enter the TIME-WAIT state, otherwise ignore 2766 * the segment. 2767 */ 2768 case TCPS_CLOSING: 2769 if (ourfinisacked) { 2770 tp->t_state = TCPS_TIME_WAIT; 2771 tcp_canceltimers(tp); 2772 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2773 soisdisconnected(so); 2774 } 2775 break; 2776 2777 /* 2778 * In LAST_ACK, we may still be waiting for data to drain 2779 * and/or to be acked, as well as for the ack of our FIN. 2780 * If our FIN is now acknowledged, delete the TCB, 2781 * enter the closed state and return. 2782 */ 2783 case TCPS_LAST_ACK: 2784 if (ourfinisacked) { 2785 tp = tcp_close(tp); 2786 goto drop; 2787 } 2788 break; 2789 2790 /* 2791 * In TIME_WAIT state the only thing that should arrive 2792 * is a retransmission of the remote FIN. Acknowledge 2793 * it and restart the finack timer. 2794 */ 2795 case TCPS_TIME_WAIT: 2796 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2797 goto dropafterack; 2798 } 2799 } 2800 2801 step6: 2802 /* 2803 * Update window information. 2804 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2805 */ 2806 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2807 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2808 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2809 /* keep track of pure window updates */ 2810 if (tlen == 0 && 2811 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2812 TCP_STATINC(TCP_STAT_RCVWINUPD); 2813 tp->snd_wnd = tiwin; 2814 tp->snd_wl1 = th->th_seq; 2815 tp->snd_wl2 = th->th_ack; 2816 if (tp->snd_wnd > tp->max_sndwnd) 2817 tp->max_sndwnd = tp->snd_wnd; 2818 needoutput = 1; 2819 } 2820 2821 /* 2822 * Process segments with URG. 2823 */ 2824 if ((tiflags & TH_URG) && th->th_urp && 2825 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2826 /* 2827 * This is a kludge, but if we receive and accept 2828 * random urgent pointers, we'll crash in 2829 * soreceive. It's hard to imagine someone 2830 * actually wanting to send this much urgent data. 2831 */ 2832 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2833 th->th_urp = 0; /* XXX */ 2834 tiflags &= ~TH_URG; /* XXX */ 2835 goto dodata; /* XXX */ 2836 } 2837 /* 2838 * If this segment advances the known urgent pointer, 2839 * then mark the data stream. This should not happen 2840 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2841 * a FIN has been received from the remote side. 2842 * In these states we ignore the URG. 2843 * 2844 * According to RFC961 (Assigned Protocols), 2845 * the urgent pointer points to the last octet 2846 * of urgent data. We continue, however, 2847 * to consider it to indicate the first octet 2848 * of data past the urgent section as the original 2849 * spec states (in one of two places). 2850 */ 2851 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2852 tp->rcv_up = th->th_seq + th->th_urp; 2853 so->so_oobmark = so->so_rcv.sb_cc + 2854 (tp->rcv_up - tp->rcv_nxt) - 1; 2855 if (so->so_oobmark == 0) 2856 so->so_state |= SS_RCVATMARK; 2857 sohasoutofband(so); 2858 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2859 } 2860 /* 2861 * Remove out of band data so doesn't get presented to user. 2862 * This can happen independent of advancing the URG pointer, 2863 * but if two URG's are pending at once, some out-of-band 2864 * data may creep in... ick. 2865 */ 2866 if (th->th_urp <= (u_int16_t) tlen 2867 #ifdef SO_OOBINLINE 2868 && (so->so_options & SO_OOBINLINE) == 0 2869 #endif 2870 ) 2871 tcp_pulloutofband(so, th, m, hdroptlen); 2872 } else 2873 /* 2874 * If no out of band data is expected, 2875 * pull receive urgent pointer along 2876 * with the receive window. 2877 */ 2878 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2879 tp->rcv_up = tp->rcv_nxt; 2880 dodata: /* XXX */ 2881 2882 /* 2883 * Process the segment text, merging it into the TCP sequencing queue, 2884 * and arranging for acknowledgement of receipt if necessary. 2885 * This process logically involves adjusting tp->rcv_wnd as data 2886 * is presented to the user (this happens in tcp_usrreq.c, 2887 * tcp_rcvd()). If a FIN has already been received on this 2888 * connection then we just ignore the text. 2889 */ 2890 if ((tlen || (tiflags & TH_FIN)) && 2891 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2892 /* 2893 * Insert segment ti into reassembly queue of tcp with 2894 * control block tp. Return TH_FIN if reassembly now includes 2895 * a segment with FIN. The macro form does the common case 2896 * inline (segment is the next to be received on an 2897 * established connection, and the queue is empty), 2898 * avoiding linkage into and removal from the queue and 2899 * repetition of various conversions. 2900 * Set DELACK for segments received in order, but ack 2901 * immediately when segments are out of order 2902 * (so fast retransmit can work). 2903 */ 2904 /* NOTE: this was TCP_REASS() macro, but used only once */ 2905 TCP_REASS_LOCK(tp); 2906 if (th->th_seq == tp->rcv_nxt && 2907 TAILQ_FIRST(&tp->segq) == NULL && 2908 tp->t_state == TCPS_ESTABLISHED) { 2909 tcp_setup_ack(tp, th); 2910 tp->rcv_nxt += tlen; 2911 tiflags = th->th_flags & TH_FIN; 2912 tcps = TCP_STAT_GETREF(); 2913 tcps[TCP_STAT_RCVPACK]++; 2914 tcps[TCP_STAT_RCVBYTE] += tlen; 2915 TCP_STAT_PUTREF(); 2916 nd6_hint(tp); 2917 if (so->so_state & SS_CANTRCVMORE) 2918 m_freem(m); 2919 else { 2920 m_adj(m, hdroptlen); 2921 sbappendstream(&(so)->so_rcv, m); 2922 } 2923 TCP_REASS_UNLOCK(tp); 2924 sorwakeup(so); 2925 } else { 2926 m_adj(m, hdroptlen); 2927 tiflags = tcp_reass(tp, th, m, &tlen); 2928 tp->t_flags |= TF_ACKNOW; 2929 } 2930 2931 /* 2932 * Note the amount of data that peer has sent into 2933 * our window, in order to estimate the sender's 2934 * buffer size. 2935 */ 2936 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2937 } else { 2938 m_freem(m); 2939 m = NULL; 2940 tiflags &= ~TH_FIN; 2941 } 2942 2943 /* 2944 * If FIN is received ACK the FIN and let the user know 2945 * that the connection is closing. Ignore a FIN received before 2946 * the connection is fully established. 2947 */ 2948 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2949 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2950 socantrcvmore(so); 2951 tp->t_flags |= TF_ACKNOW; 2952 tp->rcv_nxt++; 2953 } 2954 switch (tp->t_state) { 2955 2956 /* 2957 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2958 */ 2959 case TCPS_ESTABLISHED: 2960 tp->t_state = TCPS_CLOSE_WAIT; 2961 break; 2962 2963 /* 2964 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2965 * enter the CLOSING state. 2966 */ 2967 case TCPS_FIN_WAIT_1: 2968 tp->t_state = TCPS_CLOSING; 2969 break; 2970 2971 /* 2972 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2973 * starting the time-wait timer, turning off the other 2974 * standard timers. 2975 */ 2976 case TCPS_FIN_WAIT_2: 2977 tp->t_state = TCPS_TIME_WAIT; 2978 tcp_canceltimers(tp); 2979 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2980 soisdisconnected(so); 2981 break; 2982 2983 /* 2984 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2985 */ 2986 case TCPS_TIME_WAIT: 2987 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2988 break; 2989 } 2990 } 2991 #ifdef TCP_DEBUG 2992 if (so->so_options & SO_DEBUG) 2993 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2994 #endif 2995 2996 /* 2997 * Return any desired output. 2998 */ 2999 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 3000 KERNEL_LOCK(1, NULL); 3001 (void) tcp_output(tp); 3002 KERNEL_UNLOCK_ONE(NULL); 3003 } 3004 if (tcp_saveti) 3005 m_freem(tcp_saveti); 3006 3007 if (tp->t_state == TCPS_TIME_WAIT 3008 && (so->so_state & SS_NOFDREF) 3009 && (tp->t_inpcb || af != AF_INET) 3010 && (tp->t_in6pcb || af != AF_INET6) 3011 && ((af == AF_INET ? tcp4_vtw_enable : tcp6_vtw_enable) & 1) != 0 3012 && TAILQ_EMPTY(&tp->segq) 3013 && vtw_add(af, tp)) { 3014 ; 3015 } 3016 return; 3017 3018 badsyn: 3019 /* 3020 * Received a bad SYN. Increment counters and dropwithreset. 3021 */ 3022 TCP_STATINC(TCP_STAT_BADSYN); 3023 tp = NULL; 3024 goto dropwithreset; 3025 3026 dropafterack: 3027 /* 3028 * Generate an ACK dropping incoming segment if it occupies 3029 * sequence space, where the ACK reflects our state. 3030 */ 3031 if (tiflags & TH_RST) 3032 goto drop; 3033 goto dropafterack2; 3034 3035 dropafterack_ratelim: 3036 /* 3037 * We may want to rate-limit ACKs against SYN/RST attack. 3038 */ 3039 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 3040 tcp_ackdrop_ppslim) == 0) { 3041 /* XXX stat */ 3042 goto drop; 3043 } 3044 /* ...fall into dropafterack2... */ 3045 3046 dropafterack2: 3047 m_freem(m); 3048 tp->t_flags |= TF_ACKNOW; 3049 KERNEL_LOCK(1, NULL); 3050 (void) tcp_output(tp); 3051 KERNEL_UNLOCK_ONE(NULL); 3052 if (tcp_saveti) 3053 m_freem(tcp_saveti); 3054 return; 3055 3056 dropwithreset_ratelim: 3057 /* 3058 * We may want to rate-limit RSTs in certain situations, 3059 * particularly if we are sending an RST in response to 3060 * an attempt to connect to or otherwise communicate with 3061 * a port for which we have no socket. 3062 */ 3063 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 3064 tcp_rst_ppslim) == 0) { 3065 /* XXX stat */ 3066 goto drop; 3067 } 3068 /* ...fall into dropwithreset... */ 3069 3070 dropwithreset: 3071 /* 3072 * Generate a RST, dropping incoming segment. 3073 * Make ACK acceptable to originator of segment. 3074 */ 3075 if (tiflags & TH_RST) 3076 goto drop; 3077 3078 switch (af) { 3079 #ifdef INET6 3080 case AF_INET6: 3081 /* For following calls to tcp_respond */ 3082 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 3083 goto drop; 3084 break; 3085 #endif /* INET6 */ 3086 case AF_INET: 3087 if (IN_MULTICAST(ip->ip_dst.s_addr) || 3088 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3089 goto drop; 3090 } 3091 3092 if (tiflags & TH_ACK) 3093 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3094 else { 3095 if (tiflags & TH_SYN) 3096 tlen++; 3097 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 3098 TH_RST|TH_ACK); 3099 } 3100 if (tcp_saveti) 3101 m_freem(tcp_saveti); 3102 return; 3103 3104 badcsum: 3105 drop: 3106 /* 3107 * Drop space held by incoming segment and return. 3108 */ 3109 if (tp) { 3110 if (tp->t_inpcb) 3111 so = tp->t_inpcb->inp_socket; 3112 #ifdef INET6 3113 else if (tp->t_in6pcb) 3114 so = tp->t_in6pcb->in6p_socket; 3115 #endif 3116 else 3117 so = NULL; 3118 #ifdef TCP_DEBUG 3119 if (so && (so->so_options & SO_DEBUG) != 0) 3120 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 3121 #endif 3122 } 3123 if (tcp_saveti) 3124 m_freem(tcp_saveti); 3125 m_freem(m); 3126 return; 3127 } 3128 3129 #ifdef TCP_SIGNATURE 3130 int 3131 tcp_signature_apply(void *fstate, void *data, u_int len) 3132 { 3133 3134 MD5Update(fstate, (u_char *)data, len); 3135 return (0); 3136 } 3137 3138 struct secasvar * 3139 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 3140 { 3141 struct ip *ip; 3142 struct ip6_hdr *ip6; 3143 3144 ip = mtod(m, struct ip *); 3145 switch (ip->ip_v) { 3146 case 4: 3147 ip = mtod(m, struct ip *); 3148 ip6 = NULL; 3149 break; 3150 case 6: 3151 ip = NULL; 3152 ip6 = mtod(m, struct ip6_hdr *); 3153 break; 3154 default: 3155 return (NULL); 3156 } 3157 3158 #ifdef IPSEC 3159 if (ipsec_used) { 3160 union sockaddr_union dst; 3161 /* Extract the destination from the IP header in the mbuf. */ 3162 memset(&dst, 0, sizeof(union sockaddr_union)); 3163 if (ip != NULL) { 3164 dst.sa.sa_len = sizeof(struct sockaddr_in); 3165 dst.sa.sa_family = AF_INET; 3166 dst.sin.sin_addr = ip->ip_dst; 3167 } else { 3168 dst.sa.sa_len = sizeof(struct sockaddr_in6); 3169 dst.sa.sa_family = AF_INET6; 3170 dst.sin6.sin6_addr = ip6->ip6_dst; 3171 } 3172 3173 /* 3174 * Look up an SADB entry which matches the address of the peer. 3175 */ 3176 return KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI), 0, 0); 3177 } 3178 return NULL; 3179 #else 3180 if (ip) 3181 return key_allocsa(AF_INET, (void *)&ip->ip_src, 3182 (void *)&ip->ip_dst, IPPROTO_TCP, 3183 htonl(TCP_SIG_SPI), 0, 0); 3184 else 3185 return key_allocsa(AF_INET6, (void *)&ip6->ip6_src, 3186 (void *)&ip6->ip6_dst, IPPROTO_TCP, 3187 htonl(TCP_SIG_SPI), 0, 0); 3188 #endif 3189 } 3190 3191 int 3192 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 3193 struct secasvar *sav, char *sig) 3194 { 3195 MD5_CTX ctx; 3196 struct ip *ip; 3197 struct ipovly *ipovly; 3198 #ifdef INET6 3199 struct ip6_hdr *ip6; 3200 struct ip6_hdr_pseudo ip6pseudo; 3201 #endif /* INET6 */ 3202 struct ippseudo ippseudo; 3203 struct tcphdr th0; 3204 int l, tcphdrlen; 3205 3206 if (sav == NULL) 3207 return (-1); 3208 3209 tcphdrlen = th->th_off * 4; 3210 3211 switch (mtod(m, struct ip *)->ip_v) { 3212 case 4: 3213 MD5Init(&ctx); 3214 ip = mtod(m, struct ip *); 3215 memset(&ippseudo, 0, sizeof(ippseudo)); 3216 ipovly = (struct ipovly *)ip; 3217 ippseudo.ippseudo_src = ipovly->ih_src; 3218 ippseudo.ippseudo_dst = ipovly->ih_dst; 3219 ippseudo.ippseudo_pad = 0; 3220 ippseudo.ippseudo_p = IPPROTO_TCP; 3221 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 3222 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 3223 break; 3224 #if INET6 3225 case 6: 3226 MD5Init(&ctx); 3227 ip6 = mtod(m, struct ip6_hdr *); 3228 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 3229 ip6pseudo.ip6ph_src = ip6->ip6_src; 3230 in6_clearscope(&ip6pseudo.ip6ph_src); 3231 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 3232 in6_clearscope(&ip6pseudo.ip6ph_dst); 3233 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 3234 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 3235 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 3236 break; 3237 #endif /* INET6 */ 3238 default: 3239 return (-1); 3240 } 3241 3242 th0 = *th; 3243 th0.th_sum = 0; 3244 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 3245 3246 l = m->m_pkthdr.len - thoff - tcphdrlen; 3247 if (l > 0) 3248 m_apply(m, thoff + tcphdrlen, 3249 m->m_pkthdr.len - thoff - tcphdrlen, 3250 tcp_signature_apply, &ctx); 3251 3252 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 3253 MD5Final(sig, &ctx); 3254 3255 return (0); 3256 } 3257 #endif 3258 3259 /* 3260 * tcp_dooptions: parse and process tcp options. 3261 * 3262 * returns -1 if this segment should be dropped. (eg. wrong signature) 3263 * otherwise returns 0. 3264 */ 3265 3266 static int 3267 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, 3268 struct tcphdr *th, 3269 struct mbuf *m, int toff, struct tcp_opt_info *oi) 3270 { 3271 u_int16_t mss; 3272 int opt, optlen = 0; 3273 #ifdef TCP_SIGNATURE 3274 void *sigp = NULL; 3275 char sigbuf[TCP_SIGLEN]; 3276 struct secasvar *sav = NULL; 3277 #endif 3278 3279 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 3280 opt = cp[0]; 3281 if (opt == TCPOPT_EOL) 3282 break; 3283 if (opt == TCPOPT_NOP) 3284 optlen = 1; 3285 else { 3286 if (cnt < 2) 3287 break; 3288 optlen = cp[1]; 3289 if (optlen < 2 || optlen > cnt) 3290 break; 3291 } 3292 switch (opt) { 3293 3294 default: 3295 continue; 3296 3297 case TCPOPT_MAXSEG: 3298 if (optlen != TCPOLEN_MAXSEG) 3299 continue; 3300 if (!(th->th_flags & TH_SYN)) 3301 continue; 3302 if (TCPS_HAVERCVDSYN(tp->t_state)) 3303 continue; 3304 bcopy(cp + 2, &mss, sizeof(mss)); 3305 oi->maxseg = ntohs(mss); 3306 break; 3307 3308 case TCPOPT_WINDOW: 3309 if (optlen != TCPOLEN_WINDOW) 3310 continue; 3311 if (!(th->th_flags & TH_SYN)) 3312 continue; 3313 if (TCPS_HAVERCVDSYN(tp->t_state)) 3314 continue; 3315 tp->t_flags |= TF_RCVD_SCALE; 3316 tp->requested_s_scale = cp[2]; 3317 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3318 char buf[INET6_ADDRSTRLEN]; 3319 struct ip *ip = mtod(m, struct ip *); 3320 #ifdef INET6 3321 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 3322 #endif 3323 if (ip) 3324 in_print(buf, sizeof(buf), 3325 &ip->ip_src); 3326 #ifdef INET6 3327 else if (ip6) 3328 in6_print(buf, sizeof(buf), 3329 &ip6->ip6_src); 3330 #endif 3331 else 3332 strlcpy(buf, "(unknown)", sizeof(buf)); 3333 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3334 "assuming %d\n", 3335 tp->requested_s_scale, buf, 3336 TCP_MAX_WINSHIFT); 3337 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3338 } 3339 break; 3340 3341 case TCPOPT_TIMESTAMP: 3342 if (optlen != TCPOLEN_TIMESTAMP) 3343 continue; 3344 oi->ts_present = 1; 3345 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 3346 NTOHL(oi->ts_val); 3347 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 3348 NTOHL(oi->ts_ecr); 3349 3350 if (!(th->th_flags & TH_SYN)) 3351 continue; 3352 if (TCPS_HAVERCVDSYN(tp->t_state)) 3353 continue; 3354 /* 3355 * A timestamp received in a SYN makes 3356 * it ok to send timestamp requests and replies. 3357 */ 3358 tp->t_flags |= TF_RCVD_TSTMP; 3359 tp->ts_recent = oi->ts_val; 3360 tp->ts_recent_age = tcp_now; 3361 break; 3362 3363 case TCPOPT_SACK_PERMITTED: 3364 if (optlen != TCPOLEN_SACK_PERMITTED) 3365 continue; 3366 if (!(th->th_flags & TH_SYN)) 3367 continue; 3368 if (TCPS_HAVERCVDSYN(tp->t_state)) 3369 continue; 3370 if (tcp_do_sack) { 3371 tp->t_flags |= TF_SACK_PERMIT; 3372 tp->t_flags |= TF_WILL_SACK; 3373 } 3374 break; 3375 3376 case TCPOPT_SACK: 3377 tcp_sack_option(tp, th, cp, optlen); 3378 break; 3379 #ifdef TCP_SIGNATURE 3380 case TCPOPT_SIGNATURE: 3381 if (optlen != TCPOLEN_SIGNATURE) 3382 continue; 3383 if (sigp && memcmp(sigp, cp + 2, TCP_SIGLEN)) 3384 return (-1); 3385 3386 sigp = sigbuf; 3387 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3388 tp->t_flags |= TF_SIGNATURE; 3389 break; 3390 #endif 3391 } 3392 } 3393 3394 #ifndef TCP_SIGNATURE 3395 return 0; 3396 #else 3397 if (tp->t_flags & TF_SIGNATURE) { 3398 3399 sav = tcp_signature_getsav(m, th); 3400 3401 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3402 return (-1); 3403 } 3404 3405 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) 3406 goto out; 3407 3408 if (sigp) { 3409 char sig[TCP_SIGLEN]; 3410 3411 tcp_fields_to_net(th); 3412 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3413 tcp_fields_to_host(th); 3414 goto out; 3415 } 3416 tcp_fields_to_host(th); 3417 3418 if (memcmp(sig, sigp, TCP_SIGLEN)) { 3419 TCP_STATINC(TCP_STAT_BADSIG); 3420 goto out; 3421 } else 3422 TCP_STATINC(TCP_STAT_GOODSIG); 3423 3424 key_sa_recordxfer(sav, m); 3425 KEY_FREESAV(&sav); 3426 } 3427 return 0; 3428 out: 3429 if (sav != NULL) 3430 KEY_FREESAV(&sav); 3431 return -1; 3432 #endif 3433 } 3434 3435 /* 3436 * Pull out of band byte out of a segment so 3437 * it doesn't appear in the user's data queue. 3438 * It is still reflected in the segment length for 3439 * sequencing purposes. 3440 */ 3441 void 3442 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3443 struct mbuf *m, int off) 3444 { 3445 int cnt = off + th->th_urp - 1; 3446 3447 while (cnt >= 0) { 3448 if (m->m_len > cnt) { 3449 char *cp = mtod(m, char *) + cnt; 3450 struct tcpcb *tp = sototcpcb(so); 3451 3452 tp->t_iobc = *cp; 3453 tp->t_oobflags |= TCPOOB_HAVEDATA; 3454 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3455 m->m_len--; 3456 return; 3457 } 3458 cnt -= m->m_len; 3459 m = m->m_next; 3460 if (m == 0) 3461 break; 3462 } 3463 panic("tcp_pulloutofband"); 3464 } 3465 3466 /* 3467 * Collect new round-trip time estimate 3468 * and update averages and current timeout. 3469 * 3470 * rtt is in units of slow ticks (typically 500 ms) -- essentially the 3471 * difference of two timestamps. 3472 */ 3473 void 3474 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3475 { 3476 int32_t delta; 3477 3478 TCP_STATINC(TCP_STAT_RTTUPDATED); 3479 if (tp->t_srtt != 0) { 3480 /* 3481 * Compute the amount to add to srtt for smoothing, 3482 * *alpha, or 2^(-TCP_RTT_SHIFT). Because 3483 * srtt is stored in 1/32 slow ticks, we conceptually 3484 * shift left 5 bits, subtract srtt to get the 3485 * diference, and then shift right by TCP_RTT_SHIFT 3486 * (3) to obtain 1/8 of the difference. 3487 */ 3488 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3489 /* 3490 * This can never happen, because delta's lowest 3491 * possible value is 1/8 of t_srtt. But if it does, 3492 * set srtt to some reasonable value, here chosen 3493 * as 1/8 tick. 3494 */ 3495 if ((tp->t_srtt += delta) <= 0) 3496 tp->t_srtt = 1 << 2; 3497 /* 3498 * RFC2988 requires that rttvar be updated first. 3499 * This code is compliant because "delta" is the old 3500 * srtt minus the new observation (scaled). 3501 * 3502 * RFC2988 says: 3503 * rttvar = (1-beta) * rttvar + beta * |srtt-observed| 3504 * 3505 * delta is in units of 1/32 ticks, and has then been 3506 * divided by 8. This is equivalent to being in 1/16s 3507 * units and divided by 4. Subtract from it 1/4 of 3508 * the existing rttvar to form the (signed) amount to 3509 * adjust. 3510 */ 3511 if (delta < 0) 3512 delta = -delta; 3513 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3514 /* 3515 * As with srtt, this should never happen. There is 3516 * no support in RFC2988 for this operation. But 1/4s 3517 * as rttvar when faced with something arguably wrong 3518 * is ok. 3519 */ 3520 if ((tp->t_rttvar += delta) <= 0) 3521 tp->t_rttvar = 1 << 2; 3522 3523 /* 3524 * If srtt exceeds .01 second, ensure we use the 'remote' MSL 3525 * Problem is: it doesn't work. Disabled by defaulting 3526 * tcp_rttlocal to 0; see corresponding code in 3527 * tcp_subr that selects local vs remote in a different way. 3528 * 3529 * The static branch prediction hint here should be removed 3530 * when the rtt estimator is fixed and the rtt_enable code 3531 * is turned back on. 3532 */ 3533 if (__predict_false(tcp_rttlocal) && tcp_msl_enable 3534 && tp->t_srtt > tcp_msl_remote_threshold 3535 && tp->t_msl < tcp_msl_remote) { 3536 tp->t_msl = tcp_msl_remote; 3537 } 3538 } else { 3539 /* 3540 * This is the first measurement. Per RFC2988, 2.2, 3541 * set rtt=R and srtt=R/2. 3542 * For srtt, storage representation is 1/32 ticks, 3543 * so shift left by 5. 3544 * For rttvar, storage representation is 1/16 ticks, 3545 * So shift left by 4, but then right by 1 to halve. 3546 */ 3547 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3548 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3549 } 3550 tp->t_rtttime = 0; 3551 tp->t_rxtshift = 0; 3552 3553 /* 3554 * the retransmit should happen at rtt + 4 * rttvar. 3555 * Because of the way we do the smoothing, srtt and rttvar 3556 * will each average +1/2 tick of bias. When we compute 3557 * the retransmit timer, we want 1/2 tick of rounding and 3558 * 1 extra tick because of +-1/2 tick uncertainty in the 3559 * firing of the timer. The bias will give us exactly the 3560 * 1.5 tick we need. But, because the bias is 3561 * statistical, we have to test that we don't drop below 3562 * the minimum feasible timer (which is 2 ticks). 3563 */ 3564 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3565 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3566 3567 /* 3568 * We received an ack for a packet that wasn't retransmitted; 3569 * it is probably safe to discard any error indications we've 3570 * received recently. This isn't quite right, but close enough 3571 * for now (a route might have failed after we sent a segment, 3572 * and the return path might not be symmetrical). 3573 */ 3574 tp->t_softerror = 0; 3575 } 3576 3577 3578 /* 3579 * TCP compressed state engine. Currently used to hold compressed 3580 * state for SYN_RECEIVED. 3581 */ 3582 3583 u_long syn_cache_count; 3584 u_int32_t syn_hash1, syn_hash2; 3585 3586 #define SYN_HASH(sa, sp, dp) \ 3587 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3588 ((u_int32_t)(sp)))^syn_hash2))) 3589 #ifndef INET6 3590 #define SYN_HASHALL(hash, src, dst) \ 3591 do { \ 3592 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3593 ((const struct sockaddr_in *)(src))->sin_port, \ 3594 ((const struct sockaddr_in *)(dst))->sin_port); \ 3595 } while (/*CONSTCOND*/ 0) 3596 #else 3597 #define SYN_HASH6(sa, sp, dp) \ 3598 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3599 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3600 & 0x7fffffff) 3601 3602 #define SYN_HASHALL(hash, src, dst) \ 3603 do { \ 3604 switch ((src)->sa_family) { \ 3605 case AF_INET: \ 3606 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3607 ((const struct sockaddr_in *)(src))->sin_port, \ 3608 ((const struct sockaddr_in *)(dst))->sin_port); \ 3609 break; \ 3610 case AF_INET6: \ 3611 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3612 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3613 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3614 break; \ 3615 default: \ 3616 hash = 0; \ 3617 } \ 3618 } while (/*CONSTCOND*/0) 3619 #endif /* INET6 */ 3620 3621 static struct pool syn_cache_pool; 3622 3623 /* 3624 * We don't estimate RTT with SYNs, so each packet starts with the default 3625 * RTT and each timer step has a fixed timeout value. 3626 */ 3627 #define SYN_CACHE_TIMER_ARM(sc) \ 3628 do { \ 3629 TCPT_RANGESET((sc)->sc_rxtcur, \ 3630 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3631 TCPTV_REXMTMAX); \ 3632 callout_reset(&(sc)->sc_timer, \ 3633 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3634 } while (/*CONSTCOND*/0) 3635 3636 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3637 3638 static inline void 3639 syn_cache_rm(struct syn_cache *sc) 3640 { 3641 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3642 sc, sc_bucketq); 3643 sc->sc_tp = NULL; 3644 LIST_REMOVE(sc, sc_tpq); 3645 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3646 callout_stop(&sc->sc_timer); 3647 syn_cache_count--; 3648 } 3649 3650 static inline void 3651 syn_cache_put(struct syn_cache *sc) 3652 { 3653 if (sc->sc_ipopts) 3654 (void) m_free(sc->sc_ipopts); 3655 rtcache_free(&sc->sc_route); 3656 sc->sc_flags |= SCF_DEAD; 3657 if (!callout_invoking(&sc->sc_timer)) 3658 callout_schedule(&(sc)->sc_timer, 1); 3659 } 3660 3661 void 3662 syn_cache_init(void) 3663 { 3664 int i; 3665 3666 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 3667 "synpl", NULL, IPL_SOFTNET); 3668 3669 /* Initialize the hash buckets. */ 3670 for (i = 0; i < tcp_syn_cache_size; i++) 3671 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3672 } 3673 3674 void 3675 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3676 { 3677 struct syn_cache_head *scp; 3678 struct syn_cache *sc2; 3679 int s; 3680 3681 /* 3682 * If there are no entries in the hash table, reinitialize 3683 * the hash secrets. 3684 */ 3685 if (syn_cache_count == 0) { 3686 syn_hash1 = cprng_fast32(); 3687 syn_hash2 = cprng_fast32(); 3688 } 3689 3690 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3691 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3692 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3693 3694 /* 3695 * Make sure that we don't overflow the per-bucket 3696 * limit or the total cache size limit. 3697 */ 3698 s = splsoftnet(); 3699 if (scp->sch_length >= tcp_syn_bucket_limit) { 3700 TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW); 3701 /* 3702 * The bucket is full. Toss the oldest element in the 3703 * bucket. This will be the first entry in the bucket. 3704 */ 3705 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3706 #ifdef DIAGNOSTIC 3707 /* 3708 * This should never happen; we should always find an 3709 * entry in our bucket. 3710 */ 3711 if (sc2 == NULL) 3712 panic("syn_cache_insert: bucketoverflow: impossible"); 3713 #endif 3714 syn_cache_rm(sc2); 3715 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3716 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3717 struct syn_cache_head *scp2, *sce; 3718 3719 TCP_STATINC(TCP_STAT_SC_OVERFLOWED); 3720 /* 3721 * The cache is full. Toss the oldest entry in the 3722 * first non-empty bucket we can find. 3723 * 3724 * XXX We would really like to toss the oldest 3725 * entry in the cache, but we hope that this 3726 * condition doesn't happen very often. 3727 */ 3728 scp2 = scp; 3729 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3730 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3731 for (++scp2; scp2 != scp; scp2++) { 3732 if (scp2 >= sce) 3733 scp2 = &tcp_syn_cache[0]; 3734 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3735 break; 3736 } 3737 #ifdef DIAGNOSTIC 3738 /* 3739 * This should never happen; we should always find a 3740 * non-empty bucket. 3741 */ 3742 if (scp2 == scp) 3743 panic("syn_cache_insert: cacheoverflow: " 3744 "impossible"); 3745 #endif 3746 } 3747 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3748 syn_cache_rm(sc2); 3749 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3750 } 3751 3752 /* 3753 * Initialize the entry's timer. 3754 */ 3755 sc->sc_rxttot = 0; 3756 sc->sc_rxtshift = 0; 3757 SYN_CACHE_TIMER_ARM(sc); 3758 3759 /* Link it from tcpcb entry */ 3760 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3761 3762 /* Put it into the bucket. */ 3763 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3764 scp->sch_length++; 3765 syn_cache_count++; 3766 3767 TCP_STATINC(TCP_STAT_SC_ADDED); 3768 splx(s); 3769 } 3770 3771 /* 3772 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3773 * If we have retransmitted an entry the maximum number of times, expire 3774 * that entry. 3775 */ 3776 void 3777 syn_cache_timer(void *arg) 3778 { 3779 struct syn_cache *sc = arg; 3780 3781 mutex_enter(softnet_lock); 3782 KERNEL_LOCK(1, NULL); 3783 callout_ack(&sc->sc_timer); 3784 3785 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3786 TCP_STATINC(TCP_STAT_SC_DELAYED_FREE); 3787 callout_destroy(&sc->sc_timer); 3788 pool_put(&syn_cache_pool, sc); 3789 KERNEL_UNLOCK_ONE(NULL); 3790 mutex_exit(softnet_lock); 3791 return; 3792 } 3793 3794 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3795 /* Drop it -- too many retransmissions. */ 3796 goto dropit; 3797 } 3798 3799 /* 3800 * Compute the total amount of time this entry has 3801 * been on a queue. If this entry has been on longer 3802 * than the keep alive timer would allow, expire it. 3803 */ 3804 sc->sc_rxttot += sc->sc_rxtcur; 3805 if (sc->sc_rxttot >= tcp_keepinit) 3806 goto dropit; 3807 3808 TCP_STATINC(TCP_STAT_SC_RETRANSMITTED); 3809 (void) syn_cache_respond(sc, NULL); 3810 3811 /* Advance the timer back-off. */ 3812 sc->sc_rxtshift++; 3813 SYN_CACHE_TIMER_ARM(sc); 3814 3815 KERNEL_UNLOCK_ONE(NULL); 3816 mutex_exit(softnet_lock); 3817 return; 3818 3819 dropit: 3820 TCP_STATINC(TCP_STAT_SC_TIMED_OUT); 3821 syn_cache_rm(sc); 3822 if (sc->sc_ipopts) 3823 (void) m_free(sc->sc_ipopts); 3824 rtcache_free(&sc->sc_route); 3825 callout_destroy(&sc->sc_timer); 3826 pool_put(&syn_cache_pool, sc); 3827 KERNEL_UNLOCK_ONE(NULL); 3828 mutex_exit(softnet_lock); 3829 } 3830 3831 /* 3832 * Remove syn cache created by the specified tcb entry, 3833 * because this does not make sense to keep them 3834 * (if there's no tcb entry, syn cache entry will never be used) 3835 */ 3836 void 3837 syn_cache_cleanup(struct tcpcb *tp) 3838 { 3839 struct syn_cache *sc, *nsc; 3840 int s; 3841 3842 s = splsoftnet(); 3843 3844 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3845 nsc = LIST_NEXT(sc, sc_tpq); 3846 3847 #ifdef DIAGNOSTIC 3848 if (sc->sc_tp != tp) 3849 panic("invalid sc_tp in syn_cache_cleanup"); 3850 #endif 3851 syn_cache_rm(sc); 3852 syn_cache_put(sc); /* calls pool_put but see spl above */ 3853 } 3854 /* just for safety */ 3855 LIST_INIT(&tp->t_sc); 3856 3857 splx(s); 3858 } 3859 3860 /* 3861 * Find an entry in the syn cache. 3862 */ 3863 struct syn_cache * 3864 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3865 struct syn_cache_head **headp) 3866 { 3867 struct syn_cache *sc; 3868 struct syn_cache_head *scp; 3869 u_int32_t hash; 3870 int s; 3871 3872 SYN_HASHALL(hash, src, dst); 3873 3874 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3875 *headp = scp; 3876 s = splsoftnet(); 3877 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3878 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3879 if (sc->sc_hash != hash) 3880 continue; 3881 if (!memcmp(&sc->sc_src, src, src->sa_len) && 3882 !memcmp(&sc->sc_dst, dst, dst->sa_len)) { 3883 splx(s); 3884 return (sc); 3885 } 3886 } 3887 splx(s); 3888 return (NULL); 3889 } 3890 3891 /* 3892 * This function gets called when we receive an ACK for a 3893 * socket in the LISTEN state. We look up the connection 3894 * in the syn cache, and if its there, we pull it out of 3895 * the cache and turn it into a full-blown connection in 3896 * the SYN-RECEIVED state. 3897 * 3898 * The return values may not be immediately obvious, and their effects 3899 * can be subtle, so here they are: 3900 * 3901 * NULL SYN was not found in cache; caller should drop the 3902 * packet and send an RST. 3903 * 3904 * -1 We were unable to create the new connection, and are 3905 * aborting it. An ACK,RST is being sent to the peer 3906 * (unless we got screwey sequence numbners; see below), 3907 * because the 3-way handshake has been completed. Caller 3908 * should not free the mbuf, since we may be using it. If 3909 * we are not, we will free it. 3910 * 3911 * Otherwise, the return value is a pointer to the new socket 3912 * associated with the connection. 3913 */ 3914 struct socket * 3915 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3916 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3917 struct socket *so, struct mbuf *m) 3918 { 3919 struct syn_cache *sc; 3920 struct syn_cache_head *scp; 3921 struct inpcb *inp = NULL; 3922 #ifdef INET6 3923 struct in6pcb *in6p = NULL; 3924 #endif 3925 struct tcpcb *tp = 0; 3926 int s; 3927 struct socket *oso; 3928 3929 s = splsoftnet(); 3930 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3931 splx(s); 3932 return (NULL); 3933 } 3934 3935 /* 3936 * Verify the sequence and ack numbers. Try getting the correct 3937 * response again. 3938 */ 3939 if ((th->th_ack != sc->sc_iss + 1) || 3940 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3941 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3942 (void) syn_cache_respond(sc, m); 3943 splx(s); 3944 return ((struct socket *)(-1)); 3945 } 3946 3947 /* Remove this cache entry */ 3948 syn_cache_rm(sc); 3949 splx(s); 3950 3951 /* 3952 * Ok, create the full blown connection, and set things up 3953 * as they would have been set up if we had created the 3954 * connection when the SYN arrived. If we can't create 3955 * the connection, abort it. 3956 */ 3957 /* 3958 * inp still has the OLD in_pcb stuff, set the 3959 * v6-related flags on the new guy, too. This is 3960 * done particularly for the case where an AF_INET6 3961 * socket is bound only to a port, and a v4 connection 3962 * comes in on that port. 3963 * we also copy the flowinfo from the original pcb 3964 * to the new one. 3965 */ 3966 oso = so; 3967 so = sonewconn(so, true); 3968 if (so == NULL) 3969 goto resetandabort; 3970 3971 switch (so->so_proto->pr_domain->dom_family) { 3972 #ifdef INET 3973 case AF_INET: 3974 inp = sotoinpcb(so); 3975 break; 3976 #endif 3977 #ifdef INET6 3978 case AF_INET6: 3979 in6p = sotoin6pcb(so); 3980 break; 3981 #endif 3982 } 3983 switch (src->sa_family) { 3984 #ifdef INET 3985 case AF_INET: 3986 if (inp) { 3987 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3988 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3989 inp->inp_options = ip_srcroute(); 3990 in_pcbstate(inp, INP_BOUND); 3991 if (inp->inp_options == NULL) { 3992 inp->inp_options = sc->sc_ipopts; 3993 sc->sc_ipopts = NULL; 3994 } 3995 } 3996 #ifdef INET6 3997 else if (in6p) { 3998 /* IPv4 packet to AF_INET6 socket */ 3999 memset(&in6p->in6p_laddr, 0, sizeof(in6p->in6p_laddr)); 4000 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 4001 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 4002 &in6p->in6p_laddr.s6_addr32[3], 4003 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 4004 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 4005 in6totcpcb(in6p)->t_family = AF_INET; 4006 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 4007 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 4008 else 4009 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 4010 in6_pcbstate(in6p, IN6P_BOUND); 4011 } 4012 #endif 4013 break; 4014 #endif 4015 #ifdef INET6 4016 case AF_INET6: 4017 if (in6p) { 4018 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 4019 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 4020 in6_pcbstate(in6p, IN6P_BOUND); 4021 } 4022 break; 4023 #endif 4024 } 4025 #ifdef INET6 4026 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 4027 struct in6pcb *oin6p = sotoin6pcb(oso); 4028 /* inherit socket options from the listening socket */ 4029 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 4030 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 4031 m_freem(in6p->in6p_options); 4032 in6p->in6p_options = 0; 4033 } 4034 ip6_savecontrol(in6p, &in6p->in6p_options, 4035 mtod(m, struct ip6_hdr *), m); 4036 } 4037 #endif 4038 4039 #if defined(IPSEC) 4040 if (ipsec_used) { 4041 /* 4042 * we make a copy of policy, instead of sharing the policy, for 4043 * better behavior in terms of SA lookup and dead SA removal. 4044 */ 4045 if (inp) { 4046 /* copy old policy into new socket's */ 4047 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, 4048 inp->inp_sp)) 4049 printf("tcp_input: could not copy policy\n"); 4050 } 4051 #ifdef INET6 4052 else if (in6p) { 4053 /* copy old policy into new socket's */ 4054 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 4055 in6p->in6p_sp)) 4056 printf("tcp_input: could not copy policy\n"); 4057 } 4058 #endif 4059 } 4060 #endif 4061 4062 /* 4063 * Give the new socket our cached route reference. 4064 */ 4065 if (inp) { 4066 rtcache_copy(&inp->inp_route, &sc->sc_route); 4067 rtcache_free(&sc->sc_route); 4068 } 4069 #ifdef INET6 4070 else { 4071 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 4072 rtcache_free(&sc->sc_route); 4073 } 4074 #endif 4075 4076 if (inp) { 4077 struct sockaddr_in sin; 4078 memcpy(&sin, src, src->sa_len); 4079 if (in_pcbconnect(inp, &sin, &lwp0)) { 4080 goto resetandabort; 4081 } 4082 } 4083 #ifdef INET6 4084 else if (in6p) { 4085 struct sockaddr_in6 sin6; 4086 memcpy(&sin6, src, src->sa_len); 4087 if (src->sa_family == AF_INET) { 4088 /* IPv4 packet to AF_INET6 socket */ 4089 memset(&sin6, 0, sizeof(sin6)); 4090 sin6.sin6_family = AF_INET6; 4091 sin6.sin6_len = sizeof(sin6); 4092 sin6.sin6_port = ((struct sockaddr_in *)src)->sin_port; 4093 sin6.sin6_addr.s6_addr16[5] = htons(0xffff); 4094 bcopy(&((struct sockaddr_in *)src)->sin_addr, 4095 &sin6.sin6_addr.s6_addr32[3], 4096 sizeof(sin6.sin6_addr.s6_addr32[3])); 4097 } 4098 if (in6_pcbconnect(in6p, &sin6, NULL)) { 4099 goto resetandabort; 4100 } 4101 } 4102 #endif 4103 else { 4104 goto resetandabort; 4105 } 4106 4107 if (inp) 4108 tp = intotcpcb(inp); 4109 #ifdef INET6 4110 else if (in6p) 4111 tp = in6totcpcb(in6p); 4112 #endif 4113 else 4114 tp = NULL; 4115 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 4116 if (sc->sc_request_r_scale != 15) { 4117 tp->requested_s_scale = sc->sc_requested_s_scale; 4118 tp->request_r_scale = sc->sc_request_r_scale; 4119 tp->snd_scale = sc->sc_requested_s_scale; 4120 tp->rcv_scale = sc->sc_request_r_scale; 4121 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 4122 } 4123 if (sc->sc_flags & SCF_TIMESTAMP) 4124 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 4125 tp->ts_timebase = sc->sc_timebase; 4126 4127 tp->t_template = tcp_template(tp); 4128 if (tp->t_template == 0) { 4129 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 4130 so = NULL; 4131 m_freem(m); 4132 goto abort; 4133 } 4134 4135 tp->iss = sc->sc_iss; 4136 tp->irs = sc->sc_irs; 4137 tcp_sendseqinit(tp); 4138 tcp_rcvseqinit(tp); 4139 tp->t_state = TCPS_SYN_RECEIVED; 4140 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 4141 TCP_STATINC(TCP_STAT_ACCEPTS); 4142 4143 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 4144 tp->t_flags |= TF_WILL_SACK; 4145 4146 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 4147 tp->t_flags |= TF_ECN_PERMIT; 4148 4149 #ifdef TCP_SIGNATURE 4150 if (sc->sc_flags & SCF_SIGNATURE) 4151 tp->t_flags |= TF_SIGNATURE; 4152 #endif 4153 4154 /* Initialize tp->t_ourmss before we deal with the peer's! */ 4155 tp->t_ourmss = sc->sc_ourmaxseg; 4156 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 4157 4158 /* 4159 * Initialize the initial congestion window. If we 4160 * had to retransmit the SYN,ACK, we must initialize cwnd 4161 * to 1 segment (i.e. the Loss Window). 4162 */ 4163 if (sc->sc_rxtshift) 4164 tp->snd_cwnd = tp->t_peermss; 4165 else { 4166 int ss = tcp_init_win; 4167 #ifdef INET 4168 if (inp != NULL && in_localaddr(inp->inp_faddr)) 4169 ss = tcp_init_win_local; 4170 #endif 4171 #ifdef INET6 4172 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 4173 ss = tcp_init_win_local; 4174 #endif 4175 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 4176 } 4177 4178 tcp_rmx_rtt(tp); 4179 tp->snd_wl1 = sc->sc_irs; 4180 tp->rcv_up = sc->sc_irs + 1; 4181 4182 /* 4183 * This is what whould have happened in tcp_output() when 4184 * the SYN,ACK was sent. 4185 */ 4186 tp->snd_up = tp->snd_una; 4187 tp->snd_max = tp->snd_nxt = tp->iss+1; 4188 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 4189 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 4190 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 4191 tp->last_ack_sent = tp->rcv_nxt; 4192 tp->t_partialacks = -1; 4193 tp->t_dupacks = 0; 4194 4195 TCP_STATINC(TCP_STAT_SC_COMPLETED); 4196 s = splsoftnet(); 4197 syn_cache_put(sc); 4198 splx(s); 4199 return (so); 4200 4201 resetandabort: 4202 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 4203 abort: 4204 if (so != NULL) { 4205 (void) soqremque(so, 1); 4206 (void) soabort(so); 4207 mutex_enter(softnet_lock); 4208 } 4209 s = splsoftnet(); 4210 syn_cache_put(sc); 4211 splx(s); 4212 TCP_STATINC(TCP_STAT_SC_ABORTED); 4213 return ((struct socket *)(-1)); 4214 } 4215 4216 /* 4217 * This function is called when we get a RST for a 4218 * non-existent connection, so that we can see if the 4219 * connection is in the syn cache. If it is, zap it. 4220 */ 4221 4222 void 4223 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 4224 { 4225 struct syn_cache *sc; 4226 struct syn_cache_head *scp; 4227 int s = splsoftnet(); 4228 4229 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4230 splx(s); 4231 return; 4232 } 4233 if (SEQ_LT(th->th_seq, sc->sc_irs) || 4234 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 4235 splx(s); 4236 return; 4237 } 4238 syn_cache_rm(sc); 4239 TCP_STATINC(TCP_STAT_SC_RESET); 4240 syn_cache_put(sc); /* calls pool_put but see spl above */ 4241 splx(s); 4242 } 4243 4244 void 4245 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 4246 struct tcphdr *th) 4247 { 4248 struct syn_cache *sc; 4249 struct syn_cache_head *scp; 4250 int s; 4251 4252 s = splsoftnet(); 4253 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4254 splx(s); 4255 return; 4256 } 4257 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 4258 if (ntohl (th->th_seq) != sc->sc_iss) { 4259 splx(s); 4260 return; 4261 } 4262 4263 /* 4264 * If we've retransmitted 3 times and this is our second error, 4265 * we remove the entry. Otherwise, we allow it to continue on. 4266 * This prevents us from incorrectly nuking an entry during a 4267 * spurious network outage. 4268 * 4269 * See tcp_notify(). 4270 */ 4271 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 4272 sc->sc_flags |= SCF_UNREACH; 4273 splx(s); 4274 return; 4275 } 4276 4277 syn_cache_rm(sc); 4278 TCP_STATINC(TCP_STAT_SC_UNREACH); 4279 syn_cache_put(sc); /* calls pool_put but see spl above */ 4280 splx(s); 4281 } 4282 4283 /* 4284 * Given a LISTEN socket and an inbound SYN request, add 4285 * this to the syn cache, and send back a segment: 4286 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 4287 * to the source. 4288 * 4289 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 4290 * Doing so would require that we hold onto the data and deliver it 4291 * to the application. However, if we are the target of a SYN-flood 4292 * DoS attack, an attacker could send data which would eventually 4293 * consume all available buffer space if it were ACKed. By not ACKing 4294 * the data, we avoid this DoS scenario. 4295 */ 4296 4297 int 4298 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 4299 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 4300 int optlen, struct tcp_opt_info *oi) 4301 { 4302 struct tcpcb tb, *tp; 4303 long win; 4304 struct syn_cache *sc; 4305 struct syn_cache_head *scp; 4306 struct mbuf *ipopts; 4307 struct tcp_opt_info opti; 4308 int s; 4309 4310 tp = sototcpcb(so); 4311 4312 memset(&opti, 0, sizeof(opti)); 4313 4314 /* 4315 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 4316 * 4317 * Note this check is performed in tcp_input() very early on. 4318 */ 4319 4320 /* 4321 * Initialize some local state. 4322 */ 4323 win = sbspace(&so->so_rcv); 4324 if (win > TCP_MAXWIN) 4325 win = TCP_MAXWIN; 4326 4327 switch (src->sa_family) { 4328 #ifdef INET 4329 case AF_INET: 4330 /* 4331 * Remember the IP options, if any. 4332 */ 4333 ipopts = ip_srcroute(); 4334 break; 4335 #endif 4336 default: 4337 ipopts = NULL; 4338 } 4339 4340 #ifdef TCP_SIGNATURE 4341 if (optp || (tp->t_flags & TF_SIGNATURE)) 4342 #else 4343 if (optp) 4344 #endif 4345 { 4346 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4347 #ifdef TCP_SIGNATURE 4348 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4349 #endif 4350 tb.t_state = TCPS_LISTEN; 4351 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 4352 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 4353 return (0); 4354 } else 4355 tb.t_flags = 0; 4356 4357 /* 4358 * See if we already have an entry for this connection. 4359 * If we do, resend the SYN,ACK. We do not count this 4360 * as a retransmission (XXX though maybe we should). 4361 */ 4362 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4363 TCP_STATINC(TCP_STAT_SC_DUPESYN); 4364 if (ipopts) { 4365 /* 4366 * If we were remembering a previous source route, 4367 * forget it and use the new one we've been given. 4368 */ 4369 if (sc->sc_ipopts) 4370 (void) m_free(sc->sc_ipopts); 4371 sc->sc_ipopts = ipopts; 4372 } 4373 sc->sc_timestamp = tb.ts_recent; 4374 if (syn_cache_respond(sc, m) == 0) { 4375 uint64_t *tcps = TCP_STAT_GETREF(); 4376 tcps[TCP_STAT_SNDACKS]++; 4377 tcps[TCP_STAT_SNDTOTAL]++; 4378 TCP_STAT_PUTREF(); 4379 } 4380 return (1); 4381 } 4382 4383 s = splsoftnet(); 4384 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4385 splx(s); 4386 if (sc == NULL) { 4387 if (ipopts) 4388 (void) m_free(ipopts); 4389 return (0); 4390 } 4391 4392 /* 4393 * Fill in the cache, and put the necessary IP and TCP 4394 * options into the reply. 4395 */ 4396 memset(sc, 0, sizeof(struct syn_cache)); 4397 callout_init(&sc->sc_timer, CALLOUT_MPSAFE); 4398 bcopy(src, &sc->sc_src, src->sa_len); 4399 bcopy(dst, &sc->sc_dst, dst->sa_len); 4400 sc->sc_flags = 0; 4401 sc->sc_ipopts = ipopts; 4402 sc->sc_irs = th->th_seq; 4403 switch (src->sa_family) { 4404 #ifdef INET 4405 case AF_INET: 4406 { 4407 struct sockaddr_in *srcin = (void *) src; 4408 struct sockaddr_in *dstin = (void *) dst; 4409 4410 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4411 &srcin->sin_addr, dstin->sin_port, 4412 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4413 break; 4414 } 4415 #endif /* INET */ 4416 #ifdef INET6 4417 case AF_INET6: 4418 { 4419 struct sockaddr_in6 *srcin6 = (void *) src; 4420 struct sockaddr_in6 *dstin6 = (void *) dst; 4421 4422 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4423 &srcin6->sin6_addr, dstin6->sin6_port, 4424 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4425 break; 4426 } 4427 #endif /* INET6 */ 4428 } 4429 sc->sc_peermaxseg = oi->maxseg; 4430 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4431 m->m_pkthdr.rcvif : NULL, 4432 sc->sc_src.sa.sa_family); 4433 sc->sc_win = win; 4434 sc->sc_timebase = tcp_now - 1; /* see tcp_newtcpcb() */ 4435 sc->sc_timestamp = tb.ts_recent; 4436 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4437 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4438 sc->sc_flags |= SCF_TIMESTAMP; 4439 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4440 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4441 sc->sc_requested_s_scale = tb.requested_s_scale; 4442 sc->sc_request_r_scale = 0; 4443 /* 4444 * Pick the smallest possible scaling factor that 4445 * will still allow us to scale up to sb_max. 4446 * 4447 * We do this because there are broken firewalls that 4448 * will corrupt the window scale option, leading to 4449 * the other endpoint believing that our advertised 4450 * window is unscaled. At scale factors larger than 4451 * 5 the unscaled window will drop below 1500 bytes, 4452 * leading to serious problems when traversing these 4453 * broken firewalls. 4454 * 4455 * With the default sbmax of 256K, a scale factor 4456 * of 3 will be chosen by this algorithm. Those who 4457 * choose a larger sbmax should watch out 4458 * for the compatiblity problems mentioned above. 4459 * 4460 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4461 * or <SYN,ACK>) segment itself is never scaled. 4462 */ 4463 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4464 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4465 sc->sc_request_r_scale++; 4466 } else { 4467 sc->sc_requested_s_scale = 15; 4468 sc->sc_request_r_scale = 15; 4469 } 4470 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4471 sc->sc_flags |= SCF_SACK_PERMIT; 4472 4473 /* 4474 * ECN setup packet recieved. 4475 */ 4476 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4477 sc->sc_flags |= SCF_ECN_PERMIT; 4478 4479 #ifdef TCP_SIGNATURE 4480 if (tb.t_flags & TF_SIGNATURE) 4481 sc->sc_flags |= SCF_SIGNATURE; 4482 #endif 4483 sc->sc_tp = tp; 4484 if (syn_cache_respond(sc, m) == 0) { 4485 uint64_t *tcps = TCP_STAT_GETREF(); 4486 tcps[TCP_STAT_SNDACKS]++; 4487 tcps[TCP_STAT_SNDTOTAL]++; 4488 TCP_STAT_PUTREF(); 4489 syn_cache_insert(sc, tp); 4490 } else { 4491 s = splsoftnet(); 4492 /* 4493 * syn_cache_put() will try to schedule the timer, so 4494 * we need to initialize it 4495 */ 4496 SYN_CACHE_TIMER_ARM(sc); 4497 syn_cache_put(sc); 4498 splx(s); 4499 TCP_STATINC(TCP_STAT_SC_DROPPED); 4500 } 4501 return (1); 4502 } 4503 4504 /* 4505 * syn_cache_respond: (re)send SYN+ACK. 4506 * 4507 * returns 0 on success. otherwise returns an errno, typically ENOBUFS. 4508 */ 4509 4510 int 4511 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4512 { 4513 #ifdef INET6 4514 struct rtentry *rt; 4515 #endif 4516 struct route *ro; 4517 u_int8_t *optp; 4518 int optlen, error; 4519 u_int16_t tlen; 4520 struct ip *ip = NULL; 4521 #ifdef INET6 4522 struct ip6_hdr *ip6 = NULL; 4523 #endif 4524 struct tcpcb *tp = NULL; 4525 struct tcphdr *th; 4526 u_int hlen; 4527 struct socket *so; 4528 4529 ro = &sc->sc_route; 4530 switch (sc->sc_src.sa.sa_family) { 4531 case AF_INET: 4532 hlen = sizeof(struct ip); 4533 break; 4534 #ifdef INET6 4535 case AF_INET6: 4536 hlen = sizeof(struct ip6_hdr); 4537 break; 4538 #endif 4539 default: 4540 if (m) 4541 m_freem(m); 4542 return (EAFNOSUPPORT); 4543 } 4544 4545 /* Compute the size of the TCP options. */ 4546 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4547 ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) + 4548 #ifdef TCP_SIGNATURE 4549 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4550 #endif 4551 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4552 4553 tlen = hlen + sizeof(struct tcphdr) + optlen; 4554 4555 /* 4556 * Create the IP+TCP header from scratch. 4557 */ 4558 if (m) 4559 m_freem(m); 4560 #ifdef DIAGNOSTIC 4561 if (max_linkhdr + tlen > MCLBYTES) 4562 return (ENOBUFS); 4563 #endif 4564 MGETHDR(m, M_DONTWAIT, MT_DATA); 4565 if (m && (max_linkhdr + tlen) > MHLEN) { 4566 MCLGET(m, M_DONTWAIT); 4567 if ((m->m_flags & M_EXT) == 0) { 4568 m_freem(m); 4569 m = NULL; 4570 } 4571 } 4572 if (m == NULL) 4573 return (ENOBUFS); 4574 MCLAIM(m, &tcp_tx_mowner); 4575 4576 /* Fixup the mbuf. */ 4577 m->m_data += max_linkhdr; 4578 m->m_len = m->m_pkthdr.len = tlen; 4579 if (sc->sc_tp) { 4580 tp = sc->sc_tp; 4581 if (tp->t_inpcb) 4582 so = tp->t_inpcb->inp_socket; 4583 #ifdef INET6 4584 else if (tp->t_in6pcb) 4585 so = tp->t_in6pcb->in6p_socket; 4586 #endif 4587 else 4588 so = NULL; 4589 } else 4590 so = NULL; 4591 m->m_pkthdr.rcvif = NULL; 4592 memset(mtod(m, u_char *), 0, tlen); 4593 4594 switch (sc->sc_src.sa.sa_family) { 4595 case AF_INET: 4596 ip = mtod(m, struct ip *); 4597 ip->ip_v = 4; 4598 ip->ip_dst = sc->sc_src.sin.sin_addr; 4599 ip->ip_src = sc->sc_dst.sin.sin_addr; 4600 ip->ip_p = IPPROTO_TCP; 4601 th = (struct tcphdr *)(ip + 1); 4602 th->th_dport = sc->sc_src.sin.sin_port; 4603 th->th_sport = sc->sc_dst.sin.sin_port; 4604 break; 4605 #ifdef INET6 4606 case AF_INET6: 4607 ip6 = mtod(m, struct ip6_hdr *); 4608 ip6->ip6_vfc = IPV6_VERSION; 4609 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4610 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4611 ip6->ip6_nxt = IPPROTO_TCP; 4612 /* ip6_plen will be updated in ip6_output() */ 4613 th = (struct tcphdr *)(ip6 + 1); 4614 th->th_dport = sc->sc_src.sin6.sin6_port; 4615 th->th_sport = sc->sc_dst.sin6.sin6_port; 4616 break; 4617 #endif 4618 default: 4619 th = NULL; 4620 } 4621 4622 th->th_seq = htonl(sc->sc_iss); 4623 th->th_ack = htonl(sc->sc_irs + 1); 4624 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4625 th->th_flags = TH_SYN|TH_ACK; 4626 th->th_win = htons(sc->sc_win); 4627 /* th_sum already 0 */ 4628 /* th_urp already 0 */ 4629 4630 /* Tack on the TCP options. */ 4631 optp = (u_int8_t *)(th + 1); 4632 *optp++ = TCPOPT_MAXSEG; 4633 *optp++ = 4; 4634 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4635 *optp++ = sc->sc_ourmaxseg & 0xff; 4636 4637 if (sc->sc_request_r_scale != 15) { 4638 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4639 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4640 sc->sc_request_r_scale); 4641 optp += 4; 4642 } 4643 4644 if (sc->sc_flags & SCF_TIMESTAMP) { 4645 u_int32_t *lp = (u_int32_t *)(optp); 4646 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4647 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4648 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4649 *lp = htonl(sc->sc_timestamp); 4650 optp += TCPOLEN_TSTAMP_APPA; 4651 } 4652 4653 if (sc->sc_flags & SCF_SACK_PERMIT) { 4654 u_int8_t *p = optp; 4655 4656 /* Let the peer know that we will SACK. */ 4657 p[0] = TCPOPT_SACK_PERMITTED; 4658 p[1] = 2; 4659 p[2] = TCPOPT_NOP; 4660 p[3] = TCPOPT_NOP; 4661 optp += 4; 4662 } 4663 4664 /* 4665 * Send ECN SYN-ACK setup packet. 4666 * Routes can be asymetric, so, even if we receive a packet 4667 * with ECE and CWR set, we must not assume no one will block 4668 * the ECE packet we are about to send. 4669 */ 4670 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4671 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4672 th->th_flags |= TH_ECE; 4673 TCP_STATINC(TCP_STAT_ECN_SHS); 4674 4675 /* 4676 * draft-ietf-tcpm-ecnsyn-00.txt 4677 * 4678 * "[...] a TCP node MAY respond to an ECN-setup 4679 * SYN packet by setting ECT in the responding 4680 * ECN-setup SYN/ACK packet, indicating to routers 4681 * that the SYN/ACK packet is ECN-Capable. 4682 * This allows a congested router along the path 4683 * to mark the packet instead of dropping the 4684 * packet as an indication of congestion." 4685 * 4686 * "[...] There can be a great benefit in setting 4687 * an ECN-capable codepoint in SYN/ACK packets [...] 4688 * Congestion is most likely to occur in 4689 * the server-to-client direction. As a result, 4690 * setting an ECN-capable codepoint in SYN/ACK 4691 * packets can reduce the occurence of three-second 4692 * retransmit timeouts resulting from the drop 4693 * of SYN/ACK packets." 4694 * 4695 * Page 4 and 6, January 2006. 4696 */ 4697 4698 switch (sc->sc_src.sa.sa_family) { 4699 #ifdef INET 4700 case AF_INET: 4701 ip->ip_tos |= IPTOS_ECN_ECT0; 4702 break; 4703 #endif 4704 #ifdef INET6 4705 case AF_INET6: 4706 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4707 break; 4708 #endif 4709 } 4710 TCP_STATINC(TCP_STAT_ECN_ECT); 4711 } 4712 4713 #ifdef TCP_SIGNATURE 4714 if (sc->sc_flags & SCF_SIGNATURE) { 4715 struct secasvar *sav; 4716 u_int8_t *sigp; 4717 4718 sav = tcp_signature_getsav(m, th); 4719 4720 if (sav == NULL) { 4721 if (m) 4722 m_freem(m); 4723 return (EPERM); 4724 } 4725 4726 *optp++ = TCPOPT_SIGNATURE; 4727 *optp++ = TCPOLEN_SIGNATURE; 4728 sigp = optp; 4729 memset(optp, 0, TCP_SIGLEN); 4730 optp += TCP_SIGLEN; 4731 *optp++ = TCPOPT_NOP; 4732 *optp++ = TCPOPT_EOL; 4733 4734 (void)tcp_signature(m, th, hlen, sav, sigp); 4735 4736 key_sa_recordxfer(sav, m); 4737 KEY_FREESAV(&sav); 4738 } 4739 #endif 4740 4741 /* Compute the packet's checksum. */ 4742 switch (sc->sc_src.sa.sa_family) { 4743 case AF_INET: 4744 ip->ip_len = htons(tlen - hlen); 4745 th->th_sum = 0; 4746 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4747 break; 4748 #ifdef INET6 4749 case AF_INET6: 4750 ip6->ip6_plen = htons(tlen - hlen); 4751 th->th_sum = 0; 4752 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4753 break; 4754 #endif 4755 } 4756 4757 /* 4758 * Fill in some straggling IP bits. Note the stack expects 4759 * ip_len to be in host order, for convenience. 4760 */ 4761 switch (sc->sc_src.sa.sa_family) { 4762 #ifdef INET 4763 case AF_INET: 4764 ip->ip_len = htons(tlen); 4765 ip->ip_ttl = ip_defttl; 4766 /* XXX tos? */ 4767 break; 4768 #endif 4769 #ifdef INET6 4770 case AF_INET6: 4771 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4772 ip6->ip6_vfc |= IPV6_VERSION; 4773 ip6->ip6_plen = htons(tlen - hlen); 4774 /* ip6_hlim will be initialized afterwards */ 4775 /* XXX flowlabel? */ 4776 break; 4777 #endif 4778 } 4779 4780 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4781 tp = sc->sc_tp; 4782 4783 switch (sc->sc_src.sa.sa_family) { 4784 #ifdef INET 4785 case AF_INET: 4786 error = ip_output(m, sc->sc_ipopts, ro, 4787 (ip_mtudisc ? IP_MTUDISC : 0), 4788 NULL, so); 4789 break; 4790 #endif 4791 #ifdef INET6 4792 case AF_INET6: 4793 ip6->ip6_hlim = in6_selecthlim(NULL, 4794 (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp : NULL); 4795 4796 error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, so, NULL); 4797 break; 4798 #endif 4799 default: 4800 error = EAFNOSUPPORT; 4801 break; 4802 } 4803 return (error); 4804 } 4805