1 /* $NetBSD: tcp_input.c,v 1.314 2011/05/25 23:20:57 gdt 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 1.314 2011/05/25 23:20:57 gdt Exp $"); 152 153 #include "opt_inet.h" 154 #include "opt_ipsec.h" 155 #include "opt_inet_csum.h" 156 #include "opt_tcp_debug.h" 157 158 #include <sys/param.h> 159 #include <sys/systm.h> 160 #include <sys/malloc.h> 161 #include <sys/mbuf.h> 162 #include <sys/protosw.h> 163 #include <sys/socket.h> 164 #include <sys/socketvar.h> 165 #include <sys/errno.h> 166 #include <sys/syslog.h> 167 #include <sys/pool.h> 168 #include <sys/domain.h> 169 #include <sys/kernel.h> 170 #ifdef TCP_SIGNATURE 171 #include <sys/md5.h> 172 #endif 173 #include <sys/lwp.h> /* for lwp0 */ 174 175 #include <net/if.h> 176 #include <net/route.h> 177 #include <net/if_types.h> 178 179 #include <netinet/in.h> 180 #include <netinet/in_systm.h> 181 #include <netinet/ip.h> 182 #include <netinet/in_pcb.h> 183 #include <netinet/in_var.h> 184 #include <netinet/ip_var.h> 185 #include <netinet/in_offload.h> 186 187 #ifdef INET6 188 #ifndef INET 189 #include <netinet/in.h> 190 #endif 191 #include <netinet/ip6.h> 192 #include <netinet6/ip6_var.h> 193 #include <netinet6/in6_pcb.h> 194 #include <netinet6/ip6_var.h> 195 #include <netinet6/in6_var.h> 196 #include <netinet/icmp6.h> 197 #include <netinet6/nd6.h> 198 #ifdef TCP_SIGNATURE 199 #include <netinet6/scope6_var.h> 200 #endif 201 #endif 202 203 #ifndef INET6 204 /* always need ip6.h for IP6_EXTHDR_GET */ 205 #include <netinet/ip6.h> 206 #endif 207 208 #include <netinet/tcp.h> 209 #include <netinet/tcp_fsm.h> 210 #include <netinet/tcp_seq.h> 211 #include <netinet/tcp_timer.h> 212 #include <netinet/tcp_var.h> 213 #include <netinet/tcp_private.h> 214 #include <netinet/tcpip.h> 215 #include <netinet/tcp_congctl.h> 216 #include <netinet/tcp_debug.h> 217 218 #include <machine/stdarg.h> 219 220 #ifdef IPSEC 221 #include <netinet6/ipsec.h> 222 #include <netinet6/ipsec_private.h> 223 #include <netkey/key.h> 224 #endif /*IPSEC*/ 225 #ifdef INET6 226 #include "faith.h" 227 #if defined(NFAITH) && NFAITH > 0 228 #include <net/if_faith.h> 229 #endif 230 #endif /* IPSEC */ 231 232 #ifdef FAST_IPSEC 233 #include <netipsec/ipsec.h> 234 #include <netipsec/ipsec_var.h> 235 #include <netipsec/ipsec_private.h> 236 #include <netipsec/key.h> 237 #ifdef INET6 238 #include <netipsec/ipsec6.h> 239 #endif 240 #endif /* FAST_IPSEC*/ 241 242 #include <netinet/tcp_vtw.h> 243 244 int tcprexmtthresh = 3; 245 int tcp_log_refused; 246 247 int tcp_do_autorcvbuf = 1; 248 int tcp_autorcvbuf_inc = 16 * 1024; 249 int tcp_autorcvbuf_max = 256 * 1024; 250 int tcp_msl = (TCPTV_MSL / PR_SLOWHZ); 251 252 static int tcp_rst_ppslim_count = 0; 253 static struct timeval tcp_rst_ppslim_last; 254 static int tcp_ackdrop_ppslim_count = 0; 255 static struct timeval tcp_ackdrop_ppslim_last; 256 257 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 258 259 /* for modulo comparisons of timestamps */ 260 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 261 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 262 263 /* 264 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 265 */ 266 #ifdef INET6 267 static inline void 268 nd6_hint(struct tcpcb *tp) 269 { 270 struct rtentry *rt; 271 272 if (tp != NULL && tp->t_in6pcb != NULL && tp->t_family == AF_INET6 && 273 (rt = rtcache_validate(&tp->t_in6pcb->in6p_route)) != NULL) 274 nd6_nud_hint(rt, NULL, 0); 275 } 276 #else 277 static inline void 278 nd6_hint(struct tcpcb *tp) 279 { 280 } 281 #endif 282 283 /* 284 * Compute ACK transmission behavior. Delay the ACK unless 285 * we have already delayed an ACK (must send an ACK every two segments). 286 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 287 * option is enabled. 288 */ 289 static void 290 tcp_setup_ack(struct tcpcb *tp, const struct tcphdr *th) 291 { 292 293 if (tp->t_flags & TF_DELACK || 294 (tcp_ack_on_push && th->th_flags & TH_PUSH)) 295 tp->t_flags |= TF_ACKNOW; 296 else 297 TCP_SET_DELACK(tp); 298 } 299 300 static void 301 icmp_check(struct tcpcb *tp, const struct tcphdr *th, int acked) 302 { 303 304 /* 305 * If we had a pending ICMP message that refers to data that have 306 * just been acknowledged, disregard the recorded ICMP message. 307 */ 308 if ((tp->t_flags & TF_PMTUD_PEND) && 309 SEQ_GT(th->th_ack, tp->t_pmtud_th_seq)) 310 tp->t_flags &= ~TF_PMTUD_PEND; 311 312 /* 313 * Keep track of the largest chunk of data 314 * acknowledged since last PMTU update 315 */ 316 if (tp->t_pmtud_mss_acked < acked) 317 tp->t_pmtud_mss_acked = acked; 318 } 319 320 /* 321 * Convert TCP protocol fields to host order for easier processing. 322 */ 323 static void 324 tcp_fields_to_host(struct tcphdr *th) 325 { 326 327 NTOHL(th->th_seq); 328 NTOHL(th->th_ack); 329 NTOHS(th->th_win); 330 NTOHS(th->th_urp); 331 } 332 333 /* 334 * ... and reverse the above. 335 */ 336 static void 337 tcp_fields_to_net(struct tcphdr *th) 338 { 339 340 HTONL(th->th_seq); 341 HTONL(th->th_ack); 342 HTONS(th->th_win); 343 HTONS(th->th_urp); 344 } 345 346 #ifdef TCP_CSUM_COUNTERS 347 #include <sys/device.h> 348 349 #if defined(INET) 350 extern struct evcnt tcp_hwcsum_ok; 351 extern struct evcnt tcp_hwcsum_bad; 352 extern struct evcnt tcp_hwcsum_data; 353 extern struct evcnt tcp_swcsum; 354 #endif /* defined(INET) */ 355 #if defined(INET6) 356 extern struct evcnt tcp6_hwcsum_ok; 357 extern struct evcnt tcp6_hwcsum_bad; 358 extern struct evcnt tcp6_hwcsum_data; 359 extern struct evcnt tcp6_swcsum; 360 #endif /* defined(INET6) */ 361 362 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 363 364 #else 365 366 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 367 368 #endif /* TCP_CSUM_COUNTERS */ 369 370 #ifdef TCP_REASS_COUNTERS 371 #include <sys/device.h> 372 373 extern struct evcnt tcp_reass_; 374 extern struct evcnt tcp_reass_empty; 375 extern struct evcnt tcp_reass_iteration[8]; 376 extern struct evcnt tcp_reass_prependfirst; 377 extern struct evcnt tcp_reass_prepend; 378 extern struct evcnt tcp_reass_insert; 379 extern struct evcnt tcp_reass_inserttail; 380 extern struct evcnt tcp_reass_append; 381 extern struct evcnt tcp_reass_appendtail; 382 extern struct evcnt tcp_reass_overlaptail; 383 extern struct evcnt tcp_reass_overlapfront; 384 extern struct evcnt tcp_reass_segdup; 385 extern struct evcnt tcp_reass_fragdup; 386 387 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 388 389 #else 390 391 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 392 393 #endif /* TCP_REASS_COUNTERS */ 394 395 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *, 396 int *); 397 static int tcp_dooptions(struct tcpcb *, const u_char *, int, 398 struct tcphdr *, struct mbuf *, int, struct tcp_opt_info *); 399 400 #ifdef INET 401 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 402 #endif 403 #ifdef INET6 404 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 405 #endif 406 407 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 408 409 #if defined(MBUFTRACE) 410 struct mowner tcp_reass_mowner = MOWNER_INIT("tcp", "reass"); 411 #endif /* defined(MBUFTRACE) */ 412 413 static struct pool tcpipqent_pool; 414 415 void 416 tcpipqent_init(void) 417 { 418 419 pool_init(&tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", 420 NULL, IPL_VM); 421 } 422 423 struct ipqent * 424 tcpipqent_alloc(void) 425 { 426 struct ipqent *ipqe; 427 int s; 428 429 s = splvm(); 430 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 431 splx(s); 432 433 return ipqe; 434 } 435 436 void 437 tcpipqent_free(struct ipqent *ipqe) 438 { 439 int s; 440 441 s = splvm(); 442 pool_put(&tcpipqent_pool, ipqe); 443 splx(s); 444 } 445 446 static int 447 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int *tlen) 448 { 449 struct ipqent *p, *q, *nq, *tiqe = NULL; 450 struct socket *so = NULL; 451 int pkt_flags; 452 tcp_seq pkt_seq; 453 unsigned pkt_len; 454 u_long rcvpartdupbyte = 0; 455 u_long rcvoobyte; 456 #ifdef TCP_REASS_COUNTERS 457 u_int count = 0; 458 #endif 459 uint64_t *tcps; 460 461 if (tp->t_inpcb) 462 so = tp->t_inpcb->inp_socket; 463 #ifdef INET6 464 else if (tp->t_in6pcb) 465 so = tp->t_in6pcb->in6p_socket; 466 #endif 467 468 TCP_REASS_LOCK_CHECK(tp); 469 470 /* 471 * Call with th==0 after become established to 472 * force pre-ESTABLISHED data up to user socket. 473 */ 474 if (th == 0) 475 goto present; 476 477 m_claimm(m, &tcp_reass_mowner); 478 479 rcvoobyte = *tlen; 480 /* 481 * Copy these to local variables because the tcpiphdr 482 * gets munged while we are collapsing mbufs. 483 */ 484 pkt_seq = th->th_seq; 485 pkt_len = *tlen; 486 pkt_flags = th->th_flags; 487 488 TCP_REASS_COUNTER_INCR(&tcp_reass_); 489 490 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 491 /* 492 * When we miss a packet, the vast majority of time we get 493 * packets that follow it in order. So optimize for that. 494 */ 495 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 496 p->ipqe_len += pkt_len; 497 p->ipqe_flags |= pkt_flags; 498 m_cat(p->ipre_mlast, m); 499 TRAVERSE(p->ipre_mlast); 500 m = NULL; 501 tiqe = p; 502 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 503 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 504 goto skip_replacement; 505 } 506 /* 507 * While we're here, if the pkt is completely beyond 508 * anything we have, just insert it at the tail. 509 */ 510 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 511 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 512 goto insert_it; 513 } 514 } 515 516 q = TAILQ_FIRST(&tp->segq); 517 518 if (q != NULL) { 519 /* 520 * If this segment immediately precedes the first out-of-order 521 * block, simply slap the segment in front of it and (mostly) 522 * skip the complicated logic. 523 */ 524 if (pkt_seq + pkt_len == q->ipqe_seq) { 525 q->ipqe_seq = pkt_seq; 526 q->ipqe_len += pkt_len; 527 q->ipqe_flags |= pkt_flags; 528 m_cat(m, q->ipqe_m); 529 q->ipqe_m = m; 530 q->ipre_mlast = m; /* last mbuf may have changed */ 531 TRAVERSE(q->ipre_mlast); 532 tiqe = q; 533 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 534 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 535 goto skip_replacement; 536 } 537 } else { 538 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 539 } 540 541 /* 542 * Find a segment which begins after this one does. 543 */ 544 for (p = NULL; q != NULL; q = nq) { 545 nq = TAILQ_NEXT(q, ipqe_q); 546 #ifdef TCP_REASS_COUNTERS 547 count++; 548 #endif 549 /* 550 * If the received segment is just right after this 551 * fragment, merge the two together and then check 552 * for further overlaps. 553 */ 554 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 555 #ifdef TCPREASS_DEBUG 556 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 557 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 558 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 559 #endif 560 pkt_len += q->ipqe_len; 561 pkt_flags |= q->ipqe_flags; 562 pkt_seq = q->ipqe_seq; 563 m_cat(q->ipre_mlast, m); 564 TRAVERSE(q->ipre_mlast); 565 m = q->ipqe_m; 566 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 567 goto free_ipqe; 568 } 569 /* 570 * If the received segment is completely past this 571 * fragment, we need to go the next fragment. 572 */ 573 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 574 p = q; 575 continue; 576 } 577 /* 578 * If the fragment is past the received segment, 579 * it (or any following) can't be concatenated. 580 */ 581 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 582 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 583 break; 584 } 585 586 /* 587 * We've received all the data in this segment before. 588 * mark it as a duplicate and return. 589 */ 590 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 591 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 592 tcps = TCP_STAT_GETREF(); 593 tcps[TCP_STAT_RCVDUPPACK]++; 594 tcps[TCP_STAT_RCVDUPBYTE] += pkt_len; 595 TCP_STAT_PUTREF(); 596 tcp_new_dsack(tp, pkt_seq, pkt_len); 597 m_freem(m); 598 if (tiqe != NULL) { 599 tcpipqent_free(tiqe); 600 } 601 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 602 goto out; 603 } 604 /* 605 * Received segment completely overlaps this fragment 606 * so we drop the fragment (this keeps the temporal 607 * ordering of segments correct). 608 */ 609 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 610 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 611 rcvpartdupbyte += q->ipqe_len; 612 m_freem(q->ipqe_m); 613 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 614 goto free_ipqe; 615 } 616 /* 617 * RX'ed segment extends past the end of the 618 * fragment. Drop the overlapping bytes. Then 619 * merge the fragment and segment then treat as 620 * a longer received packet. 621 */ 622 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 623 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 624 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 625 #ifdef TCPREASS_DEBUG 626 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 627 tp, overlap, 628 pkt_seq, pkt_seq + pkt_len, pkt_len); 629 #endif 630 m_adj(m, overlap); 631 rcvpartdupbyte += overlap; 632 m_cat(q->ipre_mlast, m); 633 TRAVERSE(q->ipre_mlast); 634 m = q->ipqe_m; 635 pkt_seq = q->ipqe_seq; 636 pkt_len += q->ipqe_len - overlap; 637 rcvoobyte -= overlap; 638 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 639 goto free_ipqe; 640 } 641 /* 642 * RX'ed segment extends past the front of the 643 * fragment. Drop the overlapping bytes on the 644 * received packet. The packet will then be 645 * contatentated with this fragment a bit later. 646 */ 647 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 648 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 649 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 650 #ifdef TCPREASS_DEBUG 651 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 652 tp, overlap, 653 pkt_seq, pkt_seq + pkt_len, pkt_len); 654 #endif 655 m_adj(m, -overlap); 656 pkt_len -= overlap; 657 rcvpartdupbyte += overlap; 658 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 659 rcvoobyte -= overlap; 660 } 661 /* 662 * If the received segment immediates precedes this 663 * fragment then tack the fragment onto this segment 664 * and reinsert the data. 665 */ 666 if (q->ipqe_seq == pkt_seq + pkt_len) { 667 #ifdef TCPREASS_DEBUG 668 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 669 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 670 pkt_seq, pkt_seq + pkt_len, pkt_len); 671 #endif 672 pkt_len += q->ipqe_len; 673 pkt_flags |= q->ipqe_flags; 674 m_cat(m, q->ipqe_m); 675 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 676 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 677 tp->t_segqlen--; 678 KASSERT(tp->t_segqlen >= 0); 679 KASSERT(tp->t_segqlen != 0 || 680 (TAILQ_EMPTY(&tp->segq) && 681 TAILQ_EMPTY(&tp->timeq))); 682 if (tiqe == NULL) { 683 tiqe = q; 684 } else { 685 tcpipqent_free(q); 686 } 687 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 688 break; 689 } 690 /* 691 * If the fragment is before the segment, remember it. 692 * When this loop is terminated, p will contain the 693 * pointer to fragment that is right before the received 694 * segment. 695 */ 696 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 697 p = q; 698 699 continue; 700 701 /* 702 * This is a common operation. It also will allow 703 * to save doing a malloc/free in most instances. 704 */ 705 free_ipqe: 706 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 707 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 708 tp->t_segqlen--; 709 KASSERT(tp->t_segqlen >= 0); 710 KASSERT(tp->t_segqlen != 0 || 711 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 712 if (tiqe == NULL) { 713 tiqe = q; 714 } else { 715 tcpipqent_free(q); 716 } 717 } 718 719 #ifdef TCP_REASS_COUNTERS 720 if (count > 7) 721 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 722 else if (count > 0) 723 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 724 #endif 725 726 insert_it: 727 728 /* 729 * Allocate a new queue entry since the received segment did not 730 * collapse onto any other out-of-order block; thus we are allocating 731 * a new block. If it had collapsed, tiqe would not be NULL and 732 * we would be reusing it. 733 * XXX If we can't, just drop the packet. XXX 734 */ 735 if (tiqe == NULL) { 736 tiqe = tcpipqent_alloc(); 737 if (tiqe == NULL) { 738 TCP_STATINC(TCP_STAT_RCVMEMDROP); 739 m_freem(m); 740 goto out; 741 } 742 } 743 744 /* 745 * Update the counters. 746 */ 747 tcps = TCP_STAT_GETREF(); 748 tcps[TCP_STAT_RCVOOPACK]++; 749 tcps[TCP_STAT_RCVOOBYTE] += rcvoobyte; 750 if (rcvpartdupbyte) { 751 tcps[TCP_STAT_RCVPARTDUPPACK]++; 752 tcps[TCP_STAT_RCVPARTDUPBYTE] += rcvpartdupbyte; 753 } 754 TCP_STAT_PUTREF(); 755 756 /* 757 * Insert the new fragment queue entry into both queues. 758 */ 759 tiqe->ipqe_m = m; 760 tiqe->ipre_mlast = m; 761 tiqe->ipqe_seq = pkt_seq; 762 tiqe->ipqe_len = pkt_len; 763 tiqe->ipqe_flags = pkt_flags; 764 if (p == NULL) { 765 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 766 #ifdef TCPREASS_DEBUG 767 if (tiqe->ipqe_seq != tp->rcv_nxt) 768 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 769 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 770 #endif 771 } else { 772 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 773 #ifdef TCPREASS_DEBUG 774 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 775 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 776 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 777 #endif 778 } 779 tp->t_segqlen++; 780 781 skip_replacement: 782 783 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 784 785 present: 786 /* 787 * Present data to user, advancing rcv_nxt through 788 * completed sequence space. 789 */ 790 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 791 goto out; 792 q = TAILQ_FIRST(&tp->segq); 793 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 794 goto out; 795 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 796 goto out; 797 798 tp->rcv_nxt += q->ipqe_len; 799 pkt_flags = q->ipqe_flags & TH_FIN; 800 nd6_hint(tp); 801 802 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 803 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 804 tp->t_segqlen--; 805 KASSERT(tp->t_segqlen >= 0); 806 KASSERT(tp->t_segqlen != 0 || 807 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 808 if (so->so_state & SS_CANTRCVMORE) 809 m_freem(q->ipqe_m); 810 else 811 sbappendstream(&so->so_rcv, q->ipqe_m); 812 tcpipqent_free(q); 813 TCP_REASS_UNLOCK(tp); 814 sorwakeup(so); 815 return (pkt_flags); 816 out: 817 TCP_REASS_UNLOCK(tp); 818 return (0); 819 } 820 821 #ifdef INET6 822 int 823 tcp6_input(struct mbuf **mp, int *offp, int proto) 824 { 825 struct mbuf *m = *mp; 826 827 /* 828 * draft-itojun-ipv6-tcp-to-anycast 829 * better place to put this in? 830 */ 831 if (m->m_flags & M_ANYCAST6) { 832 struct ip6_hdr *ip6; 833 if (m->m_len < sizeof(struct ip6_hdr)) { 834 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 835 TCP_STATINC(TCP_STAT_RCVSHORT); 836 return IPPROTO_DONE; 837 } 838 } 839 ip6 = mtod(m, struct ip6_hdr *); 840 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 841 (char *)&ip6->ip6_dst - (char *)ip6); 842 return IPPROTO_DONE; 843 } 844 845 tcp_input(m, *offp, proto); 846 return IPPROTO_DONE; 847 } 848 #endif 849 850 #ifdef INET 851 static void 852 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 853 { 854 char src[4*sizeof "123"]; 855 char dst[4*sizeof "123"]; 856 857 if (ip) { 858 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 859 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 860 } 861 else { 862 strlcpy(src, "(unknown)", sizeof(src)); 863 strlcpy(dst, "(unknown)", sizeof(dst)); 864 } 865 log(LOG_INFO, 866 "Connection attempt to TCP %s:%d from %s:%d\n", 867 dst, ntohs(th->th_dport), 868 src, ntohs(th->th_sport)); 869 } 870 #endif 871 872 #ifdef INET6 873 static void 874 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 875 { 876 char src[INET6_ADDRSTRLEN]; 877 char dst[INET6_ADDRSTRLEN]; 878 879 if (ip6) { 880 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 881 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 882 } 883 else { 884 strlcpy(src, "(unknown v6)", sizeof(src)); 885 strlcpy(dst, "(unknown v6)", sizeof(dst)); 886 } 887 log(LOG_INFO, 888 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 889 dst, ntohs(th->th_dport), 890 src, ntohs(th->th_sport)); 891 } 892 #endif 893 894 /* 895 * Checksum extended TCP header and data. 896 */ 897 int 898 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, 899 int toff, int off, int tlen) 900 { 901 902 /* 903 * XXX it's better to record and check if this mbuf is 904 * already checked. 905 */ 906 907 switch (af) { 908 #ifdef INET 909 case AF_INET: 910 switch (m->m_pkthdr.csum_flags & 911 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 912 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 913 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 914 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 915 goto badcsum; 916 917 case M_CSUM_TCPv4|M_CSUM_DATA: { 918 u_int32_t hw_csum = m->m_pkthdr.csum_data; 919 920 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 921 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 922 const struct ip *ip = 923 mtod(m, const struct ip *); 924 925 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 926 ip->ip_dst.s_addr, 927 htons(hw_csum + tlen + off + IPPROTO_TCP)); 928 } 929 if ((hw_csum ^ 0xffff) != 0) 930 goto badcsum; 931 break; 932 } 933 934 case M_CSUM_TCPv4: 935 /* Checksum was okay. */ 936 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 937 break; 938 939 default: 940 /* 941 * Must compute it ourselves. Maybe skip checksum 942 * on loopback interfaces. 943 */ 944 if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & 945 IFF_LOOPBACK) || 946 tcp_do_loopback_cksum)) { 947 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 948 if (in4_cksum(m, IPPROTO_TCP, toff, 949 tlen + off) != 0) 950 goto badcsum; 951 } 952 break; 953 } 954 break; 955 #endif /* INET4 */ 956 957 #ifdef INET6 958 case AF_INET6: 959 switch (m->m_pkthdr.csum_flags & 960 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 961 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 962 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 963 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 964 goto badcsum; 965 966 #if 0 /* notyet */ 967 case M_CSUM_TCPv6|M_CSUM_DATA: 968 #endif 969 970 case M_CSUM_TCPv6: 971 /* Checksum was okay. */ 972 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 973 break; 974 975 default: 976 /* 977 * Must compute it ourselves. Maybe skip checksum 978 * on loopback interfaces. 979 */ 980 if (__predict_true((m->m_flags & M_LOOP) == 0 || 981 tcp_do_loopback_cksum)) { 982 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 983 if (in6_cksum(m, IPPROTO_TCP, toff, 984 tlen + off) != 0) 985 goto badcsum; 986 } 987 } 988 break; 989 #endif /* INET6 */ 990 } 991 992 return 0; 993 994 badcsum: 995 TCP_STATINC(TCP_STAT_RCVBADSUM); 996 return -1; 997 } 998 999 /* When a packet arrives addressed to a vestigial tcpbp, we 1000 * nevertheless have to respond to it per the spec. 1001 */ 1002 static void tcp_vtw_input(struct tcphdr *th, vestigial_inpcb_t *vp, 1003 struct mbuf *m, int tlen, int multicast) 1004 { 1005 int tiflags; 1006 int todrop, dupseg; 1007 uint32_t t_flags = 0; 1008 uint64_t *tcps; 1009 1010 tiflags = th->th_flags; 1011 todrop = vp->rcv_nxt - th->th_seq; 1012 dupseg = false; 1013 1014 if (todrop > 0) { 1015 if (tiflags & TH_SYN) { 1016 tiflags &= ~TH_SYN; 1017 ++th->th_seq; 1018 if (th->th_urp > 1) 1019 --th->th_urp; 1020 else { 1021 tiflags &= ~TH_URG; 1022 th->th_urp = 0; 1023 } 1024 --todrop; 1025 } 1026 if (todrop > tlen || 1027 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1028 /* 1029 * Any valid FIN or RST must be to the left of the 1030 * window. At this point the FIN or RST must be a 1031 * duplicate or out of sequence; drop it. 1032 */ 1033 if (tiflags & TH_RST) 1034 goto drop; 1035 tiflags &= ~(TH_FIN|TH_RST); 1036 /* 1037 * Send an ACK to resynchronize and drop any data. 1038 * But keep on processing for RST or ACK. 1039 */ 1040 t_flags |= TF_ACKNOW; 1041 todrop = tlen; 1042 dupseg = true; 1043 tcps = TCP_STAT_GETREF(); 1044 tcps[TCP_STAT_RCVDUPPACK] += 1; 1045 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 1046 TCP_STAT_PUTREF(); 1047 } else if ((tiflags & TH_RST) 1048 && th->th_seq != vp->rcv_nxt) { 1049 /* 1050 * Test for reset before adjusting the sequence 1051 * number for overlapping data. 1052 */ 1053 goto dropafterack_ratelim; 1054 } else { 1055 tcps = TCP_STAT_GETREF(); 1056 tcps[TCP_STAT_RCVPARTDUPPACK] += 1; 1057 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 1058 TCP_STAT_PUTREF(); 1059 } 1060 1061 // tcp_new_dsack(tp, th->th_seq, todrop); 1062 // hdroptlen += todrop; /*drop from head afterwards*/ 1063 1064 th->th_seq += todrop; 1065 tlen -= todrop; 1066 1067 if (th->th_urp > todrop) 1068 th->th_urp -= todrop; 1069 else { 1070 tiflags &= ~TH_URG; 1071 th->th_urp = 0; 1072 } 1073 } 1074 1075 /* 1076 * If new data are received on a connection after the 1077 * user processes are gone, then RST the other end. 1078 */ 1079 if (tlen) { 1080 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 1081 goto dropwithreset; 1082 } 1083 1084 /* 1085 * If segment ends after window, drop trailing data 1086 * (and PUSH and FIN); if nothing left, just ACK. 1087 */ 1088 todrop = (th->th_seq + tlen) - (vp->rcv_nxt+vp->rcv_wnd); 1089 1090 if (todrop > 0) { 1091 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 1092 if (todrop >= tlen) { 1093 /* 1094 * The segment actually starts after the window. 1095 * th->th_seq + tlen - vp->rcv_nxt - vp->rcv_wnd >= tlen 1096 * th->th_seq - vp->rcv_nxt - vp->rcv_wnd >= 0 1097 * th->th_seq >= vp->rcv_nxt + vp->rcv_wnd 1098 */ 1099 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 1100 /* 1101 * If a new connection request is received 1102 * while in TIME_WAIT, drop the old connection 1103 * and start over if the sequence numbers 1104 * are above the previous ones. 1105 */ 1106 if ((tiflags & TH_SYN) 1107 && SEQ_GT(th->th_seq, vp->rcv_nxt)) { 1108 /* We only support this in the !NOFDREF case, which 1109 * is to say: not here. 1110 */ 1111 goto dropwithreset;; 1112 } 1113 /* 1114 * If window is closed can only take segments at 1115 * window edge, and have to drop data and PUSH from 1116 * incoming segments. Continue processing, but 1117 * remember to ack. Otherwise, drop segment 1118 * and (if not RST) ack. 1119 */ 1120 if (vp->rcv_wnd == 0 && th->th_seq == vp->rcv_nxt) { 1121 t_flags |= TF_ACKNOW; 1122 TCP_STATINC(TCP_STAT_RCVWINPROBE); 1123 } else 1124 goto dropafterack; 1125 } else 1126 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 1127 m_adj(m, -todrop); 1128 tlen -= todrop; 1129 tiflags &= ~(TH_PUSH|TH_FIN); 1130 } 1131 1132 if (tiflags & TH_RST) { 1133 if (th->th_seq != vp->rcv_nxt) 1134 goto dropafterack_ratelim; 1135 1136 vtw_del(vp->ctl, vp->vtw); 1137 goto drop; 1138 } 1139 1140 /* 1141 * If the ACK bit is off we drop the segment and return. 1142 */ 1143 if ((tiflags & TH_ACK) == 0) { 1144 if (t_flags & TF_ACKNOW) 1145 goto dropafterack; 1146 else 1147 goto drop; 1148 } 1149 1150 /* 1151 * In TIME_WAIT state the only thing that should arrive 1152 * is a retransmission of the remote FIN. Acknowledge 1153 * it and restart the finack timer. 1154 */ 1155 vtw_restart(vp); 1156 goto dropafterack; 1157 1158 dropafterack: 1159 /* 1160 * Generate an ACK dropping incoming segment if it occupies 1161 * sequence space, where the ACK reflects our state. 1162 */ 1163 if (tiflags & TH_RST) 1164 goto drop; 1165 goto dropafterack2; 1166 1167 dropafterack_ratelim: 1168 /* 1169 * We may want to rate-limit ACKs against SYN/RST attack. 1170 */ 1171 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 1172 tcp_ackdrop_ppslim) == 0) { 1173 /* XXX stat */ 1174 goto drop; 1175 } 1176 /* ...fall into dropafterack2... */ 1177 1178 dropafterack2: 1179 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, th->th_ack, 1180 TH_ACK); 1181 return; 1182 1183 dropwithreset: 1184 /* 1185 * Generate a RST, dropping incoming segment. 1186 * Make ACK acceptable to originator of segment. 1187 */ 1188 if (tiflags & TH_RST) 1189 goto drop; 1190 1191 if (tiflags & TH_ACK) 1192 tcp_respond(0, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 1193 else { 1194 if (tiflags & TH_SYN) 1195 ++tlen; 1196 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, (tcp_seq)0, 1197 TH_RST|TH_ACK); 1198 } 1199 return; 1200 drop: 1201 m_freem(m); 1202 } 1203 1204 /* 1205 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 1206 */ 1207 void 1208 tcp_input(struct mbuf *m, ...) 1209 { 1210 struct tcphdr *th; 1211 struct ip *ip; 1212 struct inpcb *inp; 1213 #ifdef INET6 1214 struct ip6_hdr *ip6; 1215 struct in6pcb *in6p; 1216 #endif 1217 u_int8_t *optp = NULL; 1218 int optlen = 0; 1219 int len, tlen, toff, hdroptlen = 0; 1220 struct tcpcb *tp = 0; 1221 int tiflags; 1222 struct socket *so = NULL; 1223 int todrop, acked, ourfinisacked, needoutput = 0; 1224 bool dupseg; 1225 #ifdef TCP_DEBUG 1226 short ostate = 0; 1227 #endif 1228 u_long tiwin; 1229 struct tcp_opt_info opti; 1230 int off, iphlen; 1231 va_list ap; 1232 int af; /* af on the wire */ 1233 struct mbuf *tcp_saveti = NULL; 1234 uint32_t ts_rtt; 1235 uint8_t iptos; 1236 uint64_t *tcps; 1237 vestigial_inpcb_t vestige; 1238 1239 vestige.valid = 0; 1240 1241 MCLAIM(m, &tcp_rx_mowner); 1242 va_start(ap, m); 1243 toff = va_arg(ap, int); 1244 (void)va_arg(ap, int); /* ignore value, advance ap */ 1245 va_end(ap); 1246 1247 TCP_STATINC(TCP_STAT_RCVTOTAL); 1248 1249 memset(&opti, 0, sizeof(opti)); 1250 opti.ts_present = 0; 1251 opti.maxseg = 0; 1252 1253 /* 1254 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 1255 * 1256 * TCP is, by definition, unicast, so we reject all 1257 * multicast outright. 1258 * 1259 * Note, there are additional src/dst address checks in 1260 * the AF-specific code below. 1261 */ 1262 if (m->m_flags & (M_BCAST|M_MCAST)) { 1263 /* XXX stat */ 1264 goto drop; 1265 } 1266 #ifdef INET6 1267 if (m->m_flags & M_ANYCAST6) { 1268 /* XXX stat */ 1269 goto drop; 1270 } 1271 #endif 1272 1273 /* 1274 * Get IP and TCP header. 1275 * Note: IP leaves IP header in first mbuf. 1276 */ 1277 ip = mtod(m, struct ip *); 1278 switch (ip->ip_v) { 1279 #ifdef INET 1280 case 4: 1281 #ifdef INET6 1282 ip6 = NULL; 1283 #endif 1284 af = AF_INET; 1285 iphlen = sizeof(struct ip); 1286 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1287 sizeof(struct tcphdr)); 1288 if (th == NULL) { 1289 TCP_STATINC(TCP_STAT_RCVSHORT); 1290 return; 1291 } 1292 /* We do the checksum after PCB lookup... */ 1293 len = ntohs(ip->ip_len); 1294 tlen = len - toff; 1295 iptos = ip->ip_tos; 1296 break; 1297 #endif 1298 #ifdef INET6 1299 case 6: 1300 ip = NULL; 1301 iphlen = sizeof(struct ip6_hdr); 1302 af = AF_INET6; 1303 ip6 = mtod(m, struct ip6_hdr *); 1304 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1305 sizeof(struct tcphdr)); 1306 if (th == NULL) { 1307 TCP_STATINC(TCP_STAT_RCVSHORT); 1308 return; 1309 } 1310 1311 /* Be proactive about malicious use of IPv4 mapped address */ 1312 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 1313 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 1314 /* XXX stat */ 1315 goto drop; 1316 } 1317 1318 /* 1319 * Be proactive about unspecified IPv6 address in source. 1320 * As we use all-zero to indicate unbounded/unconnected pcb, 1321 * unspecified IPv6 address can be used to confuse us. 1322 * 1323 * Note that packets with unspecified IPv6 destination is 1324 * already dropped in ip6_input. 1325 */ 1326 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1327 /* XXX stat */ 1328 goto drop; 1329 } 1330 1331 /* 1332 * Make sure destination address is not multicast. 1333 * Source address checked in ip6_input(). 1334 */ 1335 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1336 /* XXX stat */ 1337 goto drop; 1338 } 1339 1340 /* We do the checksum after PCB lookup... */ 1341 len = m->m_pkthdr.len; 1342 tlen = len - toff; 1343 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1344 break; 1345 #endif 1346 default: 1347 m_freem(m); 1348 return; 1349 } 1350 1351 KASSERT(TCP_HDR_ALIGNED_P(th)); 1352 1353 /* 1354 * Check that TCP offset makes sense, 1355 * pull out TCP options and adjust length. XXX 1356 */ 1357 off = th->th_off << 2; 1358 if (off < sizeof (struct tcphdr) || off > tlen) { 1359 TCP_STATINC(TCP_STAT_RCVBADOFF); 1360 goto drop; 1361 } 1362 tlen -= off; 1363 1364 /* 1365 * tcp_input() has been modified to use tlen to mean the TCP data 1366 * length throughout the function. Other functions can use 1367 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1368 * rja 1369 */ 1370 1371 if (off > sizeof (struct tcphdr)) { 1372 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1373 if (th == NULL) { 1374 TCP_STATINC(TCP_STAT_RCVSHORT); 1375 return; 1376 } 1377 /* 1378 * NOTE: ip/ip6 will not be affected by m_pulldown() 1379 * (as they're before toff) and we don't need to update those. 1380 */ 1381 KASSERT(TCP_HDR_ALIGNED_P(th)); 1382 optlen = off - sizeof (struct tcphdr); 1383 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1384 /* 1385 * Do quick retrieval of timestamp options ("options 1386 * prediction?"). If timestamp is the only option and it's 1387 * formatted as recommended in RFC 1323 appendix A, we 1388 * quickly get the values now and not bother calling 1389 * tcp_dooptions(), etc. 1390 */ 1391 if ((optlen == TCPOLEN_TSTAMP_APPA || 1392 (optlen > TCPOLEN_TSTAMP_APPA && 1393 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1394 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1395 (th->th_flags & TH_SYN) == 0) { 1396 opti.ts_present = 1; 1397 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1398 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1399 optp = NULL; /* we've parsed the options */ 1400 } 1401 } 1402 tiflags = th->th_flags; 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) || defined(FAST_IPSEC) 1458 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1459 ipsec4_in_reject(m, inp)) { 1460 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1461 goto drop; 1462 } 1463 #ifdef INET6 1464 else if (in6p && 1465 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1466 ipsec6_in_reject_so(m, in6p->in6p_socket)) { 1467 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1468 goto drop; 1469 } 1470 #endif 1471 #endif /*IPSEC*/ 1472 break; 1473 #endif /*INET*/ 1474 #ifdef INET6 1475 case AF_INET6: 1476 { 1477 int faith; 1478 1479 #if defined(NFAITH) && NFAITH > 0 1480 faith = faithprefix(&ip6->ip6_dst); 1481 #else 1482 faith = 0; 1483 #endif 1484 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1485 th->th_sport, &ip6->ip6_dst, th->th_dport, faith, &vestige); 1486 if (!in6p && !vestige.valid) { 1487 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1488 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1489 th->th_dport, faith); 1490 } 1491 if (!in6p && !vestige.valid) { 1492 TCP_STATINC(TCP_STAT_NOPORT); 1493 if (tcp_log_refused && 1494 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1495 tcp6_log_refused(ip6, th); 1496 } 1497 tcp_fields_to_host(th); 1498 goto dropwithreset_ratelim; 1499 } 1500 #if defined(IPSEC) || defined(FAST_IPSEC) 1501 if (in6p 1502 && (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 1503 && ipsec6_in_reject(m, in6p)) { 1504 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1505 goto drop; 1506 } 1507 #endif /*IPSEC*/ 1508 break; 1509 } 1510 #endif 1511 } 1512 1513 /* 1514 * If the state is CLOSED (i.e., TCB does not exist) then 1515 * all data in the incoming segment is discarded. 1516 * If the TCB exists but is in CLOSED state, it is embryonic, 1517 * but should either do a listen or a connect soon. 1518 */ 1519 tp = NULL; 1520 so = NULL; 1521 if (inp) { 1522 /* Check the minimum TTL for socket. */ 1523 if (ip->ip_ttl < inp->inp_ip_minttl) 1524 goto drop; 1525 1526 tp = intotcpcb(inp); 1527 so = inp->inp_socket; 1528 } 1529 #ifdef INET6 1530 else if (in6p) { 1531 tp = in6totcpcb(in6p); 1532 so = in6p->in6p_socket; 1533 } 1534 #endif 1535 else if (vestige.valid) { 1536 int mc = 0; 1537 1538 /* We do not support the resurrection of vtw tcpcps. 1539 */ 1540 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1541 goto badcsum; 1542 1543 switch (af) { 1544 #ifdef INET6 1545 case AF_INET6: 1546 mc = IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst); 1547 break; 1548 #endif 1549 1550 case AF_INET: 1551 mc = (IN_MULTICAST(ip->ip_dst.s_addr) 1552 || in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)); 1553 break; 1554 } 1555 1556 tcp_fields_to_host(th); 1557 tcp_vtw_input(th, &vestige, m, tlen, mc); 1558 m = 0; 1559 goto drop; 1560 } 1561 1562 if (tp == 0) { 1563 tcp_fields_to_host(th); 1564 goto dropwithreset_ratelim; 1565 } 1566 if (tp->t_state == TCPS_CLOSED) 1567 goto drop; 1568 1569 KASSERT(so->so_lock == softnet_lock); 1570 KASSERT(solocked(so)); 1571 1572 /* 1573 * Checksum extended TCP header and data. 1574 */ 1575 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1576 goto badcsum; 1577 1578 tcp_fields_to_host(th); 1579 1580 /* Unscale the window into a 32-bit value. */ 1581 if ((tiflags & TH_SYN) == 0) 1582 tiwin = th->th_win << tp->snd_scale; 1583 else 1584 tiwin = th->th_win; 1585 1586 #ifdef INET6 1587 /* save packet options if user wanted */ 1588 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1589 if (in6p->in6p_options) { 1590 m_freem(in6p->in6p_options); 1591 in6p->in6p_options = 0; 1592 } 1593 KASSERT(ip6 != NULL); 1594 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1595 } 1596 #endif 1597 1598 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1599 union syn_cache_sa src; 1600 union syn_cache_sa dst; 1601 1602 memset(&src, 0, sizeof(src)); 1603 memset(&dst, 0, sizeof(dst)); 1604 switch (af) { 1605 #ifdef INET 1606 case AF_INET: 1607 src.sin.sin_len = sizeof(struct sockaddr_in); 1608 src.sin.sin_family = AF_INET; 1609 src.sin.sin_addr = ip->ip_src; 1610 src.sin.sin_port = th->th_sport; 1611 1612 dst.sin.sin_len = sizeof(struct sockaddr_in); 1613 dst.sin.sin_family = AF_INET; 1614 dst.sin.sin_addr = ip->ip_dst; 1615 dst.sin.sin_port = th->th_dport; 1616 break; 1617 #endif 1618 #ifdef INET6 1619 case AF_INET6: 1620 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1621 src.sin6.sin6_family = AF_INET6; 1622 src.sin6.sin6_addr = ip6->ip6_src; 1623 src.sin6.sin6_port = th->th_sport; 1624 1625 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1626 dst.sin6.sin6_family = AF_INET6; 1627 dst.sin6.sin6_addr = ip6->ip6_dst; 1628 dst.sin6.sin6_port = th->th_dport; 1629 break; 1630 #endif /* INET6 */ 1631 default: 1632 goto badsyn; /*sanity*/ 1633 } 1634 1635 if (so->so_options & SO_DEBUG) { 1636 #ifdef TCP_DEBUG 1637 ostate = tp->t_state; 1638 #endif 1639 1640 tcp_saveti = NULL; 1641 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1642 goto nosave; 1643 1644 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1645 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1646 if (!tcp_saveti) 1647 goto nosave; 1648 } else { 1649 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1650 if (!tcp_saveti) 1651 goto nosave; 1652 MCLAIM(m, &tcp_mowner); 1653 tcp_saveti->m_len = iphlen; 1654 m_copydata(m, 0, iphlen, 1655 mtod(tcp_saveti, void *)); 1656 } 1657 1658 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1659 m_freem(tcp_saveti); 1660 tcp_saveti = NULL; 1661 } else { 1662 tcp_saveti->m_len += sizeof(struct tcphdr); 1663 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1664 sizeof(struct tcphdr)); 1665 } 1666 nosave:; 1667 } 1668 if (so->so_options & SO_ACCEPTCONN) { 1669 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1670 if (tiflags & TH_RST) { 1671 syn_cache_reset(&src.sa, &dst.sa, th); 1672 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1673 (TH_ACK|TH_SYN)) { 1674 /* 1675 * Received a SYN,ACK. This should 1676 * never happen while we are in 1677 * LISTEN. Send an RST. 1678 */ 1679 goto badsyn; 1680 } else if (tiflags & TH_ACK) { 1681 so = syn_cache_get(&src.sa, &dst.sa, 1682 th, toff, tlen, so, m); 1683 if (so == NULL) { 1684 /* 1685 * We don't have a SYN for 1686 * this ACK; send an RST. 1687 */ 1688 goto badsyn; 1689 } else if (so == 1690 (struct socket *)(-1)) { 1691 /* 1692 * We were unable to create 1693 * the connection. If the 1694 * 3-way handshake was 1695 * completed, and RST has 1696 * been sent to the peer. 1697 * Since the mbuf might be 1698 * in use for the reply, 1699 * do not free it. 1700 */ 1701 m = NULL; 1702 } else { 1703 /* 1704 * We have created a 1705 * full-blown connection. 1706 */ 1707 tp = NULL; 1708 inp = NULL; 1709 #ifdef INET6 1710 in6p = NULL; 1711 #endif 1712 switch (so->so_proto->pr_domain->dom_family) { 1713 #ifdef INET 1714 case AF_INET: 1715 inp = sotoinpcb(so); 1716 tp = intotcpcb(inp); 1717 break; 1718 #endif 1719 #ifdef INET6 1720 case AF_INET6: 1721 in6p = sotoin6pcb(so); 1722 tp = in6totcpcb(in6p); 1723 break; 1724 #endif 1725 } 1726 if (tp == NULL) 1727 goto badsyn; /*XXX*/ 1728 tiwin <<= tp->snd_scale; 1729 goto after_listen; 1730 } 1731 } else { 1732 /* 1733 * None of RST, SYN or ACK was set. 1734 * This is an invalid packet for a 1735 * TCB in LISTEN state. Send a RST. 1736 */ 1737 goto badsyn; 1738 } 1739 } else { 1740 /* 1741 * Received a SYN. 1742 * 1743 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1744 */ 1745 if (m->m_flags & (M_BCAST|M_MCAST)) 1746 goto drop; 1747 1748 switch (af) { 1749 #ifdef INET6 1750 case AF_INET6: 1751 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1752 goto drop; 1753 break; 1754 #endif /* INET6 */ 1755 case AF_INET: 1756 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1757 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1758 goto drop; 1759 break; 1760 } 1761 1762 #ifdef INET6 1763 /* 1764 * If deprecated address is forbidden, we do 1765 * not accept SYN to deprecated interface 1766 * address to prevent any new inbound 1767 * connection from getting established. 1768 * When we do not accept SYN, we send a TCP 1769 * RST, with deprecated source address (instead 1770 * of dropping it). We compromise it as it is 1771 * much better for peer to send a RST, and 1772 * RST will be the final packet for the 1773 * exchange. 1774 * 1775 * If we do not forbid deprecated addresses, we 1776 * accept the SYN packet. RFC2462 does not 1777 * suggest dropping SYN in this case. 1778 * If we decipher RFC2462 5.5.4, it says like 1779 * this: 1780 * 1. use of deprecated addr with existing 1781 * communication is okay - "SHOULD continue 1782 * to be used" 1783 * 2. use of it with new communication: 1784 * (2a) "SHOULD NOT be used if alternate 1785 * address with sufficient scope is 1786 * available" 1787 * (2b) nothing mentioned otherwise. 1788 * Here we fall into (2b) case as we have no 1789 * choice in our source address selection - we 1790 * must obey the peer. 1791 * 1792 * The wording in RFC2462 is confusing, and 1793 * there are multiple description text for 1794 * deprecated address handling - worse, they 1795 * are not exactly the same. I believe 5.5.4 1796 * is the best one, so we follow 5.5.4. 1797 */ 1798 if (af == AF_INET6 && !ip6_use_deprecated) { 1799 struct in6_ifaddr *ia6; 1800 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1801 &ip6->ip6_dst)) && 1802 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1803 tp = NULL; 1804 goto dropwithreset; 1805 } 1806 } 1807 #endif 1808 1809 #if defined(IPSEC) || defined(FAST_IPSEC) 1810 switch (af) { 1811 #ifdef INET 1812 case AF_INET: 1813 if (ipsec4_in_reject_so(m, so)) { 1814 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1815 tp = NULL; 1816 goto dropwithreset; 1817 } 1818 break; 1819 #endif 1820 #ifdef INET6 1821 case AF_INET6: 1822 if (ipsec6_in_reject_so(m, so)) { 1823 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1824 tp = NULL; 1825 goto dropwithreset; 1826 } 1827 break; 1828 #endif /*INET6*/ 1829 } 1830 #endif /*IPSEC*/ 1831 1832 /* 1833 * LISTEN socket received a SYN 1834 * from itself? This can't possibly 1835 * be valid; drop the packet. 1836 */ 1837 if (th->th_sport == th->th_dport) { 1838 int i; 1839 1840 switch (af) { 1841 #ifdef INET 1842 case AF_INET: 1843 i = in_hosteq(ip->ip_src, ip->ip_dst); 1844 break; 1845 #endif 1846 #ifdef INET6 1847 case AF_INET6: 1848 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1849 break; 1850 #endif 1851 default: 1852 i = 1; 1853 } 1854 if (i) { 1855 TCP_STATINC(TCP_STAT_BADSYN); 1856 goto drop; 1857 } 1858 } 1859 1860 /* 1861 * SYN looks ok; create compressed TCP 1862 * state for it. 1863 */ 1864 if (so->so_qlen <= so->so_qlimit && 1865 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1866 so, m, optp, optlen, &opti)) 1867 m = NULL; 1868 } 1869 goto drop; 1870 } 1871 } 1872 1873 after_listen: 1874 #ifdef DIAGNOSTIC 1875 /* 1876 * Should not happen now that all embryonic connections 1877 * are handled with compressed state. 1878 */ 1879 if (tp->t_state == TCPS_LISTEN) 1880 panic("tcp_input: TCPS_LISTEN"); 1881 #endif 1882 1883 /* 1884 * Segment received on connection. 1885 * Reset idle time and keep-alive timer. 1886 */ 1887 tp->t_rcvtime = tcp_now; 1888 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1889 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1890 1891 /* 1892 * Process options. 1893 */ 1894 #ifdef TCP_SIGNATURE 1895 if (optp || (tp->t_flags & TF_SIGNATURE)) 1896 #else 1897 if (optp) 1898 #endif 1899 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1900 goto drop; 1901 1902 if (TCP_SACK_ENABLED(tp)) { 1903 tcp_del_sackholes(tp, th); 1904 } 1905 1906 if (TCP_ECN_ALLOWED(tp)) { 1907 if (tiflags & TH_CWR) { 1908 tp->t_flags &= ~TF_ECN_SND_ECE; 1909 } 1910 switch (iptos & IPTOS_ECN_MASK) { 1911 case IPTOS_ECN_CE: 1912 tp->t_flags |= TF_ECN_SND_ECE; 1913 TCP_STATINC(TCP_STAT_ECN_CE); 1914 break; 1915 case IPTOS_ECN_ECT0: 1916 TCP_STATINC(TCP_STAT_ECN_ECT); 1917 break; 1918 case IPTOS_ECN_ECT1: 1919 /* XXX: ignore for now -- rpaulo */ 1920 break; 1921 } 1922 /* 1923 * Congestion experienced. 1924 * Ignore if we are already trying to recover. 1925 */ 1926 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1927 tp->t_congctl->cong_exp(tp); 1928 } 1929 1930 if (opti.ts_present && opti.ts_ecr) { 1931 /* 1932 * Calculate the RTT from the returned time stamp and the 1933 * connection's time base. If the time stamp is later than 1934 * the current time, or is extremely old, fall back to non-1323 1935 * RTT calculation. Since ts_rtt is unsigned, we can test both 1936 * at the same time. 1937 * 1938 * Note that ts_rtt is in units of slow ticks (500 1939 * ms). Since most earthbound RTTs are < 500 ms, 1940 * observed values will have large quantization noise. 1941 * Our smoothed RTT is then the fraction of observed 1942 * samples that are 1 tick instead of 0 (times 500 1943 * ms). 1944 * 1945 * ts_rtt is increased by 1 to denote a valid sample, 1946 * with 0 indicating an invalid measurement. This 1947 * extra 1 must be removed when ts_rtt is used, or 1948 * else an an erroneous extra 500 ms will result. 1949 */ 1950 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1951 if (ts_rtt > TCP_PAWS_IDLE) 1952 ts_rtt = 0; 1953 } else { 1954 ts_rtt = 0; 1955 } 1956 1957 /* 1958 * Header prediction: check for the two common cases 1959 * of a uni-directional data xfer. If the packet has 1960 * no control flags, is in-sequence, the window didn't 1961 * change and we're not retransmitting, it's a 1962 * candidate. If the length is zero and the ack moved 1963 * forward, we're the sender side of the xfer. Just 1964 * free the data acked & wake any higher level process 1965 * that was blocked waiting for space. If the length 1966 * is non-zero and the ack didn't move, we're the 1967 * receiver side. If we're getting packets in-order 1968 * (the reassembly queue is empty), add the data to 1969 * the socket buffer and note that we need a delayed ack. 1970 */ 1971 if (tp->t_state == TCPS_ESTABLISHED && 1972 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1973 == TH_ACK && 1974 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1975 th->th_seq == tp->rcv_nxt && 1976 tiwin && tiwin == tp->snd_wnd && 1977 tp->snd_nxt == tp->snd_max) { 1978 1979 /* 1980 * If last ACK falls within this segment's sequence numbers, 1981 * record the timestamp. 1982 * NOTE that the test is modified according to the latest 1983 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1984 * 1985 * note that we already know 1986 * TSTMP_GEQ(opti.ts_val, tp->ts_recent) 1987 */ 1988 if (opti.ts_present && 1989 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1990 tp->ts_recent_age = tcp_now; 1991 tp->ts_recent = opti.ts_val; 1992 } 1993 1994 if (tlen == 0) { 1995 /* Ack prediction. */ 1996 if (SEQ_GT(th->th_ack, tp->snd_una) && 1997 SEQ_LEQ(th->th_ack, tp->snd_max) && 1998 tp->snd_cwnd >= tp->snd_wnd && 1999 tp->t_partialacks < 0) { 2000 /* 2001 * this is a pure ack for outstanding data. 2002 */ 2003 if (ts_rtt) 2004 tcp_xmit_timer(tp, ts_rtt - 1); 2005 else if (tp->t_rtttime && 2006 SEQ_GT(th->th_ack, tp->t_rtseq)) 2007 tcp_xmit_timer(tp, 2008 tcp_now - tp->t_rtttime); 2009 acked = th->th_ack - tp->snd_una; 2010 tcps = TCP_STAT_GETREF(); 2011 tcps[TCP_STAT_PREDACK]++; 2012 tcps[TCP_STAT_RCVACKPACK]++; 2013 tcps[TCP_STAT_RCVACKBYTE] += acked; 2014 TCP_STAT_PUTREF(); 2015 nd6_hint(tp); 2016 2017 if (acked > (tp->t_lastoff - tp->t_inoff)) 2018 tp->t_lastm = NULL; 2019 sbdrop(&so->so_snd, acked); 2020 tp->t_lastoff -= acked; 2021 2022 icmp_check(tp, th, acked); 2023 2024 tp->snd_una = th->th_ack; 2025 tp->snd_fack = tp->snd_una; 2026 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2027 tp->snd_high = tp->snd_una; 2028 m_freem(m); 2029 2030 /* 2031 * If all outstanding data are acked, stop 2032 * retransmit timer, otherwise restart timer 2033 * using current (possibly backed-off) value. 2034 * If process is waiting for space, 2035 * wakeup/selnotify/signal. If data 2036 * are ready to send, let tcp_output 2037 * decide between more output or persist. 2038 */ 2039 if (tp->snd_una == tp->snd_max) 2040 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2041 else if (TCP_TIMER_ISARMED(tp, 2042 TCPT_PERSIST) == 0) 2043 TCP_TIMER_ARM(tp, TCPT_REXMT, 2044 tp->t_rxtcur); 2045 2046 sowwakeup(so); 2047 if (so->so_snd.sb_cc) { 2048 KERNEL_LOCK(1, NULL); 2049 (void) tcp_output(tp); 2050 KERNEL_UNLOCK_ONE(NULL); 2051 } 2052 if (tcp_saveti) 2053 m_freem(tcp_saveti); 2054 return; 2055 } 2056 } else if (th->th_ack == tp->snd_una && 2057 TAILQ_FIRST(&tp->segq) == NULL && 2058 tlen <= sbspace(&so->so_rcv)) { 2059 int newsize = 0; /* automatic sockbuf scaling */ 2060 2061 /* 2062 * this is a pure, in-sequence data packet 2063 * with nothing on the reassembly queue and 2064 * we have enough buffer space to take it. 2065 */ 2066 tp->rcv_nxt += tlen; 2067 tcps = TCP_STAT_GETREF(); 2068 tcps[TCP_STAT_PREDDAT]++; 2069 tcps[TCP_STAT_RCVPACK]++; 2070 tcps[TCP_STAT_RCVBYTE] += tlen; 2071 TCP_STAT_PUTREF(); 2072 nd6_hint(tp); 2073 2074 /* 2075 * Automatic sizing enables the performance of large buffers 2076 * and most of the efficiency of small ones by only allocating 2077 * space when it is needed. 2078 * 2079 * On the receive side the socket buffer memory is only rarely 2080 * used to any significant extent. This allows us to be much 2081 * more aggressive in scaling the receive socket buffer. For 2082 * the case that the buffer space is actually used to a large 2083 * extent and we run out of kernel memory we can simply drop 2084 * the new segments; TCP on the sender will just retransmit it 2085 * later. Setting the buffer size too big may only consume too 2086 * much kernel memory if the application doesn't read() from 2087 * the socket or packet loss or reordering makes use of the 2088 * reassembly queue. 2089 * 2090 * The criteria to step up the receive buffer one notch are: 2091 * 1. the number of bytes received during the time it takes 2092 * one timestamp to be reflected back to us (the RTT); 2093 * 2. received bytes per RTT is within seven eighth of the 2094 * current socket buffer size; 2095 * 3. receive buffer size has not hit maximal automatic size; 2096 * 2097 * This algorithm does one step per RTT at most and only if 2098 * we receive a bulk stream w/o packet losses or reorderings. 2099 * Shrinking the buffer during idle times is not necessary as 2100 * it doesn't consume any memory when idle. 2101 * 2102 * TODO: Only step up if the application is actually serving 2103 * the buffer to better manage the socket buffer resources. 2104 */ 2105 if (tcp_do_autorcvbuf && 2106 opti.ts_ecr && 2107 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 2108 if (opti.ts_ecr > tp->rfbuf_ts && 2109 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 2110 if (tp->rfbuf_cnt > 2111 (so->so_rcv.sb_hiwat / 8 * 7) && 2112 so->so_rcv.sb_hiwat < 2113 tcp_autorcvbuf_max) { 2114 newsize = 2115 min(so->so_rcv.sb_hiwat + 2116 tcp_autorcvbuf_inc, 2117 tcp_autorcvbuf_max); 2118 } 2119 /* Start over with next RTT. */ 2120 tp->rfbuf_ts = 0; 2121 tp->rfbuf_cnt = 0; 2122 } else 2123 tp->rfbuf_cnt += tlen; /* add up */ 2124 } 2125 2126 /* 2127 * Drop TCP, IP headers and TCP options then add data 2128 * to socket buffer. 2129 */ 2130 if (so->so_state & SS_CANTRCVMORE) 2131 m_freem(m); 2132 else { 2133 /* 2134 * Set new socket buffer size. 2135 * Give up when limit is reached. 2136 */ 2137 if (newsize) 2138 if (!sbreserve(&so->so_rcv, 2139 newsize, so)) 2140 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 2141 m_adj(m, toff + off); 2142 sbappendstream(&so->so_rcv, m); 2143 } 2144 sorwakeup(so); 2145 tcp_setup_ack(tp, th); 2146 if (tp->t_flags & TF_ACKNOW) { 2147 KERNEL_LOCK(1, NULL); 2148 (void) tcp_output(tp); 2149 KERNEL_UNLOCK_ONE(NULL); 2150 } 2151 if (tcp_saveti) 2152 m_freem(tcp_saveti); 2153 return; 2154 } 2155 } 2156 2157 /* 2158 * Compute mbuf offset to TCP data segment. 2159 */ 2160 hdroptlen = toff + off; 2161 2162 /* 2163 * Calculate amount of space in receive window, 2164 * and then do TCP input processing. 2165 * Receive window is amount of space in rcv queue, 2166 * but not less than advertised window. 2167 */ 2168 { int win; 2169 2170 win = sbspace(&so->so_rcv); 2171 if (win < 0) 2172 win = 0; 2173 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 2174 } 2175 2176 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 2177 tp->rfbuf_ts = 0; 2178 tp->rfbuf_cnt = 0; 2179 2180 switch (tp->t_state) { 2181 /* 2182 * If the state is SYN_SENT: 2183 * if seg contains an ACK, but not for our SYN, drop the input. 2184 * if seg contains a RST, then drop the connection. 2185 * if seg does not contain SYN, then drop it. 2186 * Otherwise this is an acceptable SYN segment 2187 * initialize tp->rcv_nxt and tp->irs 2188 * if seg contains ack then advance tp->snd_una 2189 * if seg contains a ECE and ECN support is enabled, the stream 2190 * is ECN capable. 2191 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2192 * arrange for segment to be acked (eventually) 2193 * continue processing rest of data/controls, beginning with URG 2194 */ 2195 case TCPS_SYN_SENT: 2196 if ((tiflags & TH_ACK) && 2197 (SEQ_LEQ(th->th_ack, tp->iss) || 2198 SEQ_GT(th->th_ack, tp->snd_max))) 2199 goto dropwithreset; 2200 if (tiflags & TH_RST) { 2201 if (tiflags & TH_ACK) 2202 tp = tcp_drop(tp, ECONNREFUSED); 2203 goto drop; 2204 } 2205 if ((tiflags & TH_SYN) == 0) 2206 goto drop; 2207 if (tiflags & TH_ACK) { 2208 tp->snd_una = th->th_ack; 2209 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2210 tp->snd_nxt = tp->snd_una; 2211 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2212 tp->snd_high = tp->snd_una; 2213 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2214 2215 if ((tiflags & TH_ECE) && tcp_do_ecn) { 2216 tp->t_flags |= TF_ECN_PERMIT; 2217 TCP_STATINC(TCP_STAT_ECN_SHS); 2218 } 2219 2220 } 2221 tp->irs = th->th_seq; 2222 tcp_rcvseqinit(tp); 2223 tp->t_flags |= TF_ACKNOW; 2224 tcp_mss_from_peer(tp, opti.maxseg); 2225 2226 /* 2227 * Initialize the initial congestion window. If we 2228 * had to retransmit the SYN, we must initialize cwnd 2229 * to 1 segment (i.e. the Loss Window). 2230 */ 2231 if (tp->t_flags & TF_SYN_REXMT) 2232 tp->snd_cwnd = tp->t_peermss; 2233 else { 2234 int ss = tcp_init_win; 2235 #ifdef INET 2236 if (inp != NULL && in_localaddr(inp->inp_faddr)) 2237 ss = tcp_init_win_local; 2238 #endif 2239 #ifdef INET6 2240 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 2241 ss = tcp_init_win_local; 2242 #endif 2243 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 2244 } 2245 2246 tcp_rmx_rtt(tp); 2247 if (tiflags & TH_ACK) { 2248 TCP_STATINC(TCP_STAT_CONNECTS); 2249 /* 2250 * move tcp_established before soisconnected 2251 * because upcall handler can drive tcp_output 2252 * functionality. 2253 * XXX we might call soisconnected at the end of 2254 * all processing 2255 */ 2256 tcp_established(tp); 2257 soisconnected(so); 2258 /* Do window scaling on this connection? */ 2259 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2260 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2261 tp->snd_scale = tp->requested_s_scale; 2262 tp->rcv_scale = tp->request_r_scale; 2263 } 2264 TCP_REASS_LOCK(tp); 2265 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2266 /* 2267 * if we didn't have to retransmit the SYN, 2268 * use its rtt as our initial srtt & rtt var. 2269 */ 2270 if (tp->t_rtttime) 2271 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2272 } else 2273 tp->t_state = TCPS_SYN_RECEIVED; 2274 2275 /* 2276 * Advance th->th_seq to correspond to first data byte. 2277 * If data, trim to stay within window, 2278 * dropping FIN if necessary. 2279 */ 2280 th->th_seq++; 2281 if (tlen > tp->rcv_wnd) { 2282 todrop = tlen - tp->rcv_wnd; 2283 m_adj(m, -todrop); 2284 tlen = tp->rcv_wnd; 2285 tiflags &= ~TH_FIN; 2286 tcps = TCP_STAT_GETREF(); 2287 tcps[TCP_STAT_RCVPACKAFTERWIN]++; 2288 tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop; 2289 TCP_STAT_PUTREF(); 2290 } 2291 tp->snd_wl1 = th->th_seq - 1; 2292 tp->rcv_up = th->th_seq; 2293 goto step6; 2294 2295 /* 2296 * If the state is SYN_RECEIVED: 2297 * If seg contains an ACK, but not for our SYN, drop the input 2298 * and generate an RST. See page 36, rfc793 2299 */ 2300 case TCPS_SYN_RECEIVED: 2301 if ((tiflags & TH_ACK) && 2302 (SEQ_LEQ(th->th_ack, tp->iss) || 2303 SEQ_GT(th->th_ack, tp->snd_max))) 2304 goto dropwithreset; 2305 break; 2306 } 2307 2308 /* 2309 * States other than LISTEN or SYN_SENT. 2310 * First check timestamp, if present. 2311 * Then check that at least some bytes of segment are within 2312 * receive window. If segment begins before rcv_nxt, 2313 * drop leading data (and SYN); if nothing left, just ack. 2314 * 2315 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2316 * and it's less than ts_recent, drop it. 2317 */ 2318 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2319 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2320 2321 /* Check to see if ts_recent is over 24 days old. */ 2322 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2323 /* 2324 * Invalidate ts_recent. If this segment updates 2325 * ts_recent, the age will be reset later and ts_recent 2326 * will get a valid value. If it does not, setting 2327 * ts_recent to zero will at least satisfy the 2328 * requirement that zero be placed in the timestamp 2329 * echo reply when ts_recent isn't valid. The 2330 * age isn't reset until we get a valid ts_recent 2331 * because we don't want out-of-order segments to be 2332 * dropped when ts_recent is old. 2333 */ 2334 tp->ts_recent = 0; 2335 } else { 2336 tcps = TCP_STAT_GETREF(); 2337 tcps[TCP_STAT_RCVDUPPACK]++; 2338 tcps[TCP_STAT_RCVDUPBYTE] += tlen; 2339 tcps[TCP_STAT_PAWSDROP]++; 2340 TCP_STAT_PUTREF(); 2341 tcp_new_dsack(tp, th->th_seq, tlen); 2342 goto dropafterack; 2343 } 2344 } 2345 2346 todrop = tp->rcv_nxt - th->th_seq; 2347 dupseg = false; 2348 if (todrop > 0) { 2349 if (tiflags & TH_SYN) { 2350 tiflags &= ~TH_SYN; 2351 th->th_seq++; 2352 if (th->th_urp > 1) 2353 th->th_urp--; 2354 else { 2355 tiflags &= ~TH_URG; 2356 th->th_urp = 0; 2357 } 2358 todrop--; 2359 } 2360 if (todrop > tlen || 2361 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2362 /* 2363 * Any valid FIN or RST must be to the left of the 2364 * window. At this point the FIN or RST must be a 2365 * duplicate or out of sequence; drop it. 2366 */ 2367 if (tiflags & TH_RST) 2368 goto drop; 2369 tiflags &= ~(TH_FIN|TH_RST); 2370 /* 2371 * Send an ACK to resynchronize and drop any data. 2372 * But keep on processing for RST or ACK. 2373 */ 2374 tp->t_flags |= TF_ACKNOW; 2375 todrop = tlen; 2376 dupseg = true; 2377 tcps = TCP_STAT_GETREF(); 2378 tcps[TCP_STAT_RCVDUPPACK]++; 2379 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 2380 TCP_STAT_PUTREF(); 2381 } else if ((tiflags & TH_RST) && 2382 th->th_seq != tp->rcv_nxt) { 2383 /* 2384 * Test for reset before adjusting the sequence 2385 * number for overlapping data. 2386 */ 2387 goto dropafterack_ratelim; 2388 } else { 2389 tcps = TCP_STAT_GETREF(); 2390 tcps[TCP_STAT_RCVPARTDUPPACK]++; 2391 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 2392 TCP_STAT_PUTREF(); 2393 } 2394 tcp_new_dsack(tp, th->th_seq, todrop); 2395 hdroptlen += todrop; /*drop from head afterwards*/ 2396 th->th_seq += todrop; 2397 tlen -= todrop; 2398 if (th->th_urp > todrop) 2399 th->th_urp -= todrop; 2400 else { 2401 tiflags &= ~TH_URG; 2402 th->th_urp = 0; 2403 } 2404 } 2405 2406 /* 2407 * If new data are received on a connection after the 2408 * user processes are gone, then RST the other end. 2409 */ 2410 if ((so->so_state & SS_NOFDREF) && 2411 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2412 tp = tcp_close(tp); 2413 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 2414 goto dropwithreset; 2415 } 2416 2417 /* 2418 * If segment ends after window, drop trailing data 2419 * (and PUSH and FIN); if nothing left, just ACK. 2420 */ 2421 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 2422 if (todrop > 0) { 2423 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 2424 if (todrop >= tlen) { 2425 /* 2426 * The segment actually starts after the window. 2427 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2428 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2429 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2430 */ 2431 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 2432 /* 2433 * If a new connection request is received 2434 * while in TIME_WAIT, drop the old connection 2435 * and start over if the sequence numbers 2436 * are above the previous ones. 2437 * 2438 * NOTE: We will checksum the packet again, and 2439 * so we need to put the header fields back into 2440 * network order! 2441 * XXX This kind of sucks, but we don't expect 2442 * XXX this to happen very often, so maybe it 2443 * XXX doesn't matter so much. 2444 */ 2445 if (tiflags & TH_SYN && 2446 tp->t_state == TCPS_TIME_WAIT && 2447 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2448 tp = tcp_close(tp); 2449 tcp_fields_to_net(th); 2450 goto findpcb; 2451 } 2452 /* 2453 * If window is closed can only take segments at 2454 * window edge, and have to drop data and PUSH from 2455 * incoming segments. Continue processing, but 2456 * remember to ack. Otherwise, drop segment 2457 * and (if not RST) ack. 2458 */ 2459 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2460 tp->t_flags |= TF_ACKNOW; 2461 TCP_STATINC(TCP_STAT_RCVWINPROBE); 2462 } else 2463 goto dropafterack; 2464 } else 2465 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 2466 m_adj(m, -todrop); 2467 tlen -= todrop; 2468 tiflags &= ~(TH_PUSH|TH_FIN); 2469 } 2470 2471 /* 2472 * If last ACK falls within this segment's sequence numbers, 2473 * record the timestamp. 2474 * NOTE: 2475 * 1) That the test incorporates suggestions from the latest 2476 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2477 * 2) That updating only on newer timestamps interferes with 2478 * our earlier PAWS tests, so this check should be solely 2479 * predicated on the sequence space of this segment. 2480 * 3) That we modify the segment boundary check to be 2481 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2482 * instead of RFC1323's 2483 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2484 * This modified check allows us to overcome RFC1323's 2485 * limitations as described in Stevens TCP/IP Illustrated 2486 * Vol. 2 p.869. In such cases, we can still calculate the 2487 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2488 */ 2489 if (opti.ts_present && 2490 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2491 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2492 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2493 tp->ts_recent_age = tcp_now; 2494 tp->ts_recent = opti.ts_val; 2495 } 2496 2497 /* 2498 * If the RST bit is set examine the state: 2499 * SYN_RECEIVED STATE: 2500 * If passive open, return to LISTEN state. 2501 * If active open, inform user that connection was refused. 2502 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 2503 * Inform user that connection was reset, and close tcb. 2504 * CLOSING, LAST_ACK, TIME_WAIT STATES 2505 * Close the tcb. 2506 */ 2507 if (tiflags & TH_RST) { 2508 if (th->th_seq != tp->rcv_nxt) 2509 goto dropafterack_ratelim; 2510 2511 switch (tp->t_state) { 2512 case TCPS_SYN_RECEIVED: 2513 so->so_error = ECONNREFUSED; 2514 goto close; 2515 2516 case TCPS_ESTABLISHED: 2517 case TCPS_FIN_WAIT_1: 2518 case TCPS_FIN_WAIT_2: 2519 case TCPS_CLOSE_WAIT: 2520 so->so_error = ECONNRESET; 2521 close: 2522 tp->t_state = TCPS_CLOSED; 2523 TCP_STATINC(TCP_STAT_DROPS); 2524 tp = tcp_close(tp); 2525 goto drop; 2526 2527 case TCPS_CLOSING: 2528 case TCPS_LAST_ACK: 2529 case TCPS_TIME_WAIT: 2530 tp = tcp_close(tp); 2531 goto drop; 2532 } 2533 } 2534 2535 /* 2536 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2537 * we must be in a synchronized state. RFC791 states (under RST 2538 * generation) that any unacceptable segment (an out-of-order SYN 2539 * qualifies) received in a synchronized state must elicit only an 2540 * empty acknowledgment segment ... and the connection remains in 2541 * the same state. 2542 */ 2543 if (tiflags & TH_SYN) { 2544 if (tp->rcv_nxt == th->th_seq) { 2545 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2546 TH_ACK); 2547 if (tcp_saveti) 2548 m_freem(tcp_saveti); 2549 return; 2550 } 2551 2552 goto dropafterack_ratelim; 2553 } 2554 2555 /* 2556 * If the ACK bit is off we drop the segment and return. 2557 */ 2558 if ((tiflags & TH_ACK) == 0) { 2559 if (tp->t_flags & TF_ACKNOW) 2560 goto dropafterack; 2561 else 2562 goto drop; 2563 } 2564 2565 /* 2566 * Ack processing. 2567 */ 2568 switch (tp->t_state) { 2569 2570 /* 2571 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2572 * ESTABLISHED state and continue processing, otherwise 2573 * send an RST. 2574 */ 2575 case TCPS_SYN_RECEIVED: 2576 if (SEQ_GT(tp->snd_una, th->th_ack) || 2577 SEQ_GT(th->th_ack, tp->snd_max)) 2578 goto dropwithreset; 2579 TCP_STATINC(TCP_STAT_CONNECTS); 2580 soisconnected(so); 2581 tcp_established(tp); 2582 /* Do window scaling? */ 2583 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2584 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2585 tp->snd_scale = tp->requested_s_scale; 2586 tp->rcv_scale = tp->request_r_scale; 2587 } 2588 TCP_REASS_LOCK(tp); 2589 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2590 tp->snd_wl1 = th->th_seq - 1; 2591 /* fall into ... */ 2592 2593 /* 2594 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2595 * ACKs. If the ack is in the range 2596 * tp->snd_una < th->th_ack <= tp->snd_max 2597 * then advance tp->snd_una to th->th_ack and drop 2598 * data from the retransmission queue. If this ACK reflects 2599 * more up to date window information we update our window information. 2600 */ 2601 case TCPS_ESTABLISHED: 2602 case TCPS_FIN_WAIT_1: 2603 case TCPS_FIN_WAIT_2: 2604 case TCPS_CLOSE_WAIT: 2605 case TCPS_CLOSING: 2606 case TCPS_LAST_ACK: 2607 case TCPS_TIME_WAIT: 2608 2609 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2610 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2611 TCP_STATINC(TCP_STAT_RCVDUPACK); 2612 /* 2613 * If we have outstanding data (other than 2614 * a window probe), this is a completely 2615 * duplicate ack (ie, window info didn't 2616 * change), the ack is the biggest we've 2617 * seen and we've seen exactly our rexmt 2618 * threshhold of them, assume a packet 2619 * has been dropped and retransmit it. 2620 * Kludge snd_nxt & the congestion 2621 * window so we send only this one 2622 * packet. 2623 */ 2624 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2625 th->th_ack != tp->snd_una) 2626 tp->t_dupacks = 0; 2627 else if (tp->t_partialacks < 0 && 2628 (++tp->t_dupacks == tcprexmtthresh || 2629 TCP_FACK_FASTRECOV(tp))) { 2630 /* 2631 * Do the fast retransmit, and adjust 2632 * congestion control paramenters. 2633 */ 2634 if (tp->t_congctl->fast_retransmit(tp, th)) { 2635 /* False fast retransmit */ 2636 break; 2637 } else 2638 goto drop; 2639 } else if (tp->t_dupacks > tcprexmtthresh) { 2640 tp->snd_cwnd += tp->t_segsz; 2641 KERNEL_LOCK(1, NULL); 2642 (void) tcp_output(tp); 2643 KERNEL_UNLOCK_ONE(NULL); 2644 goto drop; 2645 } 2646 } else { 2647 /* 2648 * If the ack appears to be very old, only 2649 * allow data that is in-sequence. This 2650 * makes it somewhat more difficult to insert 2651 * forged data by guessing sequence numbers. 2652 * Sent an ack to try to update the send 2653 * sequence number on the other side. 2654 */ 2655 if (tlen && th->th_seq != tp->rcv_nxt && 2656 SEQ_LT(th->th_ack, 2657 tp->snd_una - tp->max_sndwnd)) 2658 goto dropafterack; 2659 } 2660 break; 2661 } 2662 /* 2663 * If the congestion window was inflated to account 2664 * for the other side's cached packets, retract it. 2665 */ 2666 /* XXX: make SACK have his own congestion control 2667 * struct -- rpaulo */ 2668 if (TCP_SACK_ENABLED(tp)) 2669 tcp_sack_newack(tp, th); 2670 else 2671 tp->t_congctl->fast_retransmit_newack(tp, th); 2672 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2673 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2674 goto dropafterack; 2675 } 2676 acked = th->th_ack - tp->snd_una; 2677 tcps = TCP_STAT_GETREF(); 2678 tcps[TCP_STAT_RCVACKPACK]++; 2679 tcps[TCP_STAT_RCVACKBYTE] += acked; 2680 TCP_STAT_PUTREF(); 2681 2682 /* 2683 * If we have a timestamp reply, update smoothed 2684 * round trip time. If no timestamp is present but 2685 * transmit timer is running and timed sequence 2686 * number was acked, update smoothed round trip time. 2687 * Since we now have an rtt measurement, cancel the 2688 * timer backoff (cf., Phil Karn's retransmit alg.). 2689 * Recompute the initial retransmit timer. 2690 */ 2691 if (ts_rtt) 2692 tcp_xmit_timer(tp, ts_rtt - 1); 2693 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2694 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2695 2696 /* 2697 * If all outstanding data is acked, stop retransmit 2698 * timer and remember to restart (more output or persist). 2699 * If there is more data to be acked, restart retransmit 2700 * timer, using current (possibly backed-off) value. 2701 */ 2702 if (th->th_ack == tp->snd_max) { 2703 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2704 needoutput = 1; 2705 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2706 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2707 2708 /* 2709 * New data has been acked, adjust the congestion window. 2710 */ 2711 tp->t_congctl->newack(tp, th); 2712 2713 nd6_hint(tp); 2714 if (acked > so->so_snd.sb_cc) { 2715 tp->snd_wnd -= so->so_snd.sb_cc; 2716 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2717 ourfinisacked = 1; 2718 } else { 2719 if (acked > (tp->t_lastoff - tp->t_inoff)) 2720 tp->t_lastm = NULL; 2721 sbdrop(&so->so_snd, acked); 2722 tp->t_lastoff -= acked; 2723 tp->snd_wnd -= acked; 2724 ourfinisacked = 0; 2725 } 2726 sowwakeup(so); 2727 2728 icmp_check(tp, th, acked); 2729 2730 tp->snd_una = th->th_ack; 2731 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2732 tp->snd_fack = tp->snd_una; 2733 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2734 tp->snd_nxt = tp->snd_una; 2735 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2736 tp->snd_high = tp->snd_una; 2737 2738 switch (tp->t_state) { 2739 2740 /* 2741 * In FIN_WAIT_1 STATE in addition to the processing 2742 * for the ESTABLISHED state if our FIN is now acknowledged 2743 * then enter FIN_WAIT_2. 2744 */ 2745 case TCPS_FIN_WAIT_1: 2746 if (ourfinisacked) { 2747 /* 2748 * If we can't receive any more 2749 * data, then closing user can proceed. 2750 * Starting the timer is contrary to the 2751 * specification, but if we don't get a FIN 2752 * we'll hang forever. 2753 */ 2754 if (so->so_state & SS_CANTRCVMORE) { 2755 soisdisconnected(so); 2756 if (tp->t_maxidle > 0) 2757 TCP_TIMER_ARM(tp, TCPT_2MSL, 2758 tp->t_maxidle); 2759 } 2760 tp->t_state = TCPS_FIN_WAIT_2; 2761 } 2762 break; 2763 2764 /* 2765 * In CLOSING STATE in addition to the processing for 2766 * the ESTABLISHED state if the ACK acknowledges our FIN 2767 * then enter the TIME-WAIT state, otherwise ignore 2768 * the segment. 2769 */ 2770 case TCPS_CLOSING: 2771 if (ourfinisacked) { 2772 tp->t_state = TCPS_TIME_WAIT; 2773 tcp_canceltimers(tp); 2774 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2775 soisdisconnected(so); 2776 } 2777 break; 2778 2779 /* 2780 * In LAST_ACK, we may still be waiting for data to drain 2781 * and/or to be acked, as well as for the ack of our FIN. 2782 * If our FIN is now acknowledged, delete the TCB, 2783 * enter the closed state and return. 2784 */ 2785 case TCPS_LAST_ACK: 2786 if (ourfinisacked) { 2787 tp = tcp_close(tp); 2788 goto drop; 2789 } 2790 break; 2791 2792 /* 2793 * In TIME_WAIT state the only thing that should arrive 2794 * is a retransmission of the remote FIN. Acknowledge 2795 * it and restart the finack timer. 2796 */ 2797 case TCPS_TIME_WAIT: 2798 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2799 goto dropafterack; 2800 } 2801 } 2802 2803 step6: 2804 /* 2805 * Update window information. 2806 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2807 */ 2808 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2809 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2810 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2811 /* keep track of pure window updates */ 2812 if (tlen == 0 && 2813 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2814 TCP_STATINC(TCP_STAT_RCVWINUPD); 2815 tp->snd_wnd = tiwin; 2816 tp->snd_wl1 = th->th_seq; 2817 tp->snd_wl2 = th->th_ack; 2818 if (tp->snd_wnd > tp->max_sndwnd) 2819 tp->max_sndwnd = tp->snd_wnd; 2820 needoutput = 1; 2821 } 2822 2823 /* 2824 * Process segments with URG. 2825 */ 2826 if ((tiflags & TH_URG) && th->th_urp && 2827 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2828 /* 2829 * This is a kludge, but if we receive and accept 2830 * random urgent pointers, we'll crash in 2831 * soreceive. It's hard to imagine someone 2832 * actually wanting to send this much urgent data. 2833 */ 2834 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2835 th->th_urp = 0; /* XXX */ 2836 tiflags &= ~TH_URG; /* XXX */ 2837 goto dodata; /* XXX */ 2838 } 2839 /* 2840 * If this segment advances the known urgent pointer, 2841 * then mark the data stream. This should not happen 2842 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2843 * a FIN has been received from the remote side. 2844 * In these states we ignore the URG. 2845 * 2846 * According to RFC961 (Assigned Protocols), 2847 * the urgent pointer points to the last octet 2848 * of urgent data. We continue, however, 2849 * to consider it to indicate the first octet 2850 * of data past the urgent section as the original 2851 * spec states (in one of two places). 2852 */ 2853 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2854 tp->rcv_up = th->th_seq + th->th_urp; 2855 so->so_oobmark = so->so_rcv.sb_cc + 2856 (tp->rcv_up - tp->rcv_nxt) - 1; 2857 if (so->so_oobmark == 0) 2858 so->so_state |= SS_RCVATMARK; 2859 sohasoutofband(so); 2860 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2861 } 2862 /* 2863 * Remove out of band data so doesn't get presented to user. 2864 * This can happen independent of advancing the URG pointer, 2865 * but if two URG's are pending at once, some out-of-band 2866 * data may creep in... ick. 2867 */ 2868 if (th->th_urp <= (u_int16_t) tlen 2869 #ifdef SO_OOBINLINE 2870 && (so->so_options & SO_OOBINLINE) == 0 2871 #endif 2872 ) 2873 tcp_pulloutofband(so, th, m, hdroptlen); 2874 } else 2875 /* 2876 * If no out of band data is expected, 2877 * pull receive urgent pointer along 2878 * with the receive window. 2879 */ 2880 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2881 tp->rcv_up = tp->rcv_nxt; 2882 dodata: /* XXX */ 2883 2884 /* 2885 * Process the segment text, merging it into the TCP sequencing queue, 2886 * and arranging for acknowledgement of receipt if necessary. 2887 * This process logically involves adjusting tp->rcv_wnd as data 2888 * is presented to the user (this happens in tcp_usrreq.c, 2889 * case PRU_RCVD). If a FIN has already been received on this 2890 * connection then we just ignore the text. 2891 */ 2892 if ((tlen || (tiflags & TH_FIN)) && 2893 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2894 /* 2895 * Insert segment ti into reassembly queue of tcp with 2896 * control block tp. Return TH_FIN if reassembly now includes 2897 * a segment with FIN. The macro form does the common case 2898 * inline (segment is the next to be received on an 2899 * established connection, and the queue is empty), 2900 * avoiding linkage into and removal from the queue and 2901 * repetition of various conversions. 2902 * Set DELACK for segments received in order, but ack 2903 * immediately when segments are out of order 2904 * (so fast retransmit can work). 2905 */ 2906 /* NOTE: this was TCP_REASS() macro, but used only once */ 2907 TCP_REASS_LOCK(tp); 2908 if (th->th_seq == tp->rcv_nxt && 2909 TAILQ_FIRST(&tp->segq) == NULL && 2910 tp->t_state == TCPS_ESTABLISHED) { 2911 tcp_setup_ack(tp, th); 2912 tp->rcv_nxt += tlen; 2913 tiflags = th->th_flags & TH_FIN; 2914 tcps = TCP_STAT_GETREF(); 2915 tcps[TCP_STAT_RCVPACK]++; 2916 tcps[TCP_STAT_RCVBYTE] += tlen; 2917 TCP_STAT_PUTREF(); 2918 nd6_hint(tp); 2919 if (so->so_state & SS_CANTRCVMORE) 2920 m_freem(m); 2921 else { 2922 m_adj(m, hdroptlen); 2923 sbappendstream(&(so)->so_rcv, m); 2924 } 2925 TCP_REASS_UNLOCK(tp); 2926 sorwakeup(so); 2927 } else { 2928 m_adj(m, hdroptlen); 2929 tiflags = tcp_reass(tp, th, m, &tlen); 2930 tp->t_flags |= TF_ACKNOW; 2931 TCP_REASS_UNLOCK(tp); 2932 } 2933 2934 /* 2935 * Note the amount of data that peer has sent into 2936 * our window, in order to estimate the sender's 2937 * buffer size. 2938 */ 2939 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2940 } else { 2941 m_freem(m); 2942 m = NULL; 2943 tiflags &= ~TH_FIN; 2944 } 2945 2946 /* 2947 * If FIN is received ACK the FIN and let the user know 2948 * that the connection is closing. Ignore a FIN received before 2949 * the connection is fully established. 2950 */ 2951 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2952 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2953 socantrcvmore(so); 2954 tp->t_flags |= TF_ACKNOW; 2955 tp->rcv_nxt++; 2956 } 2957 switch (tp->t_state) { 2958 2959 /* 2960 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2961 */ 2962 case TCPS_ESTABLISHED: 2963 tp->t_state = TCPS_CLOSE_WAIT; 2964 break; 2965 2966 /* 2967 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2968 * enter the CLOSING state. 2969 */ 2970 case TCPS_FIN_WAIT_1: 2971 tp->t_state = TCPS_CLOSING; 2972 break; 2973 2974 /* 2975 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2976 * starting the time-wait timer, turning off the other 2977 * standard timers. 2978 */ 2979 case TCPS_FIN_WAIT_2: 2980 tp->t_state = TCPS_TIME_WAIT; 2981 tcp_canceltimers(tp); 2982 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2983 soisdisconnected(so); 2984 break; 2985 2986 /* 2987 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2988 */ 2989 case TCPS_TIME_WAIT: 2990 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2991 break; 2992 } 2993 } 2994 #ifdef TCP_DEBUG 2995 if (so->so_options & SO_DEBUG) 2996 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2997 #endif 2998 2999 /* 3000 * Return any desired output. 3001 */ 3002 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 3003 KERNEL_LOCK(1, NULL); 3004 (void) tcp_output(tp); 3005 KERNEL_UNLOCK_ONE(NULL); 3006 } 3007 if (tcp_saveti) 3008 m_freem(tcp_saveti); 3009 3010 if (tp->t_state == TCPS_TIME_WAIT 3011 && (so->so_state & SS_NOFDREF) 3012 && (tp->t_inpcb || af != AF_INET) 3013 && (tp->t_in6pcb || af != AF_INET6) 3014 && ((af == AF_INET ? tcp4_vtw_enable : tcp6_vtw_enable) & 1) != 0 3015 && TAILQ_EMPTY(&tp->segq) 3016 && vtw_add(af, tp)) { 3017 ; 3018 } 3019 return; 3020 3021 badsyn: 3022 /* 3023 * Received a bad SYN. Increment counters and dropwithreset. 3024 */ 3025 TCP_STATINC(TCP_STAT_BADSYN); 3026 tp = NULL; 3027 goto dropwithreset; 3028 3029 dropafterack: 3030 /* 3031 * Generate an ACK dropping incoming segment if it occupies 3032 * sequence space, where the ACK reflects our state. 3033 */ 3034 if (tiflags & TH_RST) 3035 goto drop; 3036 goto dropafterack2; 3037 3038 dropafterack_ratelim: 3039 /* 3040 * We may want to rate-limit ACKs against SYN/RST attack. 3041 */ 3042 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 3043 tcp_ackdrop_ppslim) == 0) { 3044 /* XXX stat */ 3045 goto drop; 3046 } 3047 /* ...fall into dropafterack2... */ 3048 3049 dropafterack2: 3050 m_freem(m); 3051 tp->t_flags |= TF_ACKNOW; 3052 KERNEL_LOCK(1, NULL); 3053 (void) tcp_output(tp); 3054 KERNEL_UNLOCK_ONE(NULL); 3055 if (tcp_saveti) 3056 m_freem(tcp_saveti); 3057 return; 3058 3059 dropwithreset_ratelim: 3060 /* 3061 * We may want to rate-limit RSTs in certain situations, 3062 * particularly if we are sending an RST in response to 3063 * an attempt to connect to or otherwise communicate with 3064 * a port for which we have no socket. 3065 */ 3066 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 3067 tcp_rst_ppslim) == 0) { 3068 /* XXX stat */ 3069 goto drop; 3070 } 3071 /* ...fall into dropwithreset... */ 3072 3073 dropwithreset: 3074 /* 3075 * Generate a RST, dropping incoming segment. 3076 * Make ACK acceptable to originator of segment. 3077 */ 3078 if (tiflags & TH_RST) 3079 goto drop; 3080 3081 switch (af) { 3082 #ifdef INET6 3083 case AF_INET6: 3084 /* For following calls to tcp_respond */ 3085 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 3086 goto drop; 3087 break; 3088 #endif /* INET6 */ 3089 case AF_INET: 3090 if (IN_MULTICAST(ip->ip_dst.s_addr) || 3091 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3092 goto drop; 3093 } 3094 3095 if (tiflags & TH_ACK) 3096 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3097 else { 3098 if (tiflags & TH_SYN) 3099 tlen++; 3100 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 3101 TH_RST|TH_ACK); 3102 } 3103 if (tcp_saveti) 3104 m_freem(tcp_saveti); 3105 return; 3106 3107 badcsum: 3108 drop: 3109 /* 3110 * Drop space held by incoming segment and return. 3111 */ 3112 if (tp) { 3113 if (tp->t_inpcb) 3114 so = tp->t_inpcb->inp_socket; 3115 #ifdef INET6 3116 else if (tp->t_in6pcb) 3117 so = tp->t_in6pcb->in6p_socket; 3118 #endif 3119 else 3120 so = NULL; 3121 #ifdef TCP_DEBUG 3122 if (so && (so->so_options & SO_DEBUG) != 0) 3123 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 3124 #endif 3125 } 3126 if (tcp_saveti) 3127 m_freem(tcp_saveti); 3128 m_freem(m); 3129 return; 3130 } 3131 3132 #ifdef TCP_SIGNATURE 3133 int 3134 tcp_signature_apply(void *fstate, void *data, u_int len) 3135 { 3136 3137 MD5Update(fstate, (u_char *)data, len); 3138 return (0); 3139 } 3140 3141 struct secasvar * 3142 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 3143 { 3144 struct secasvar *sav; 3145 #ifdef FAST_IPSEC 3146 union sockaddr_union dst; 3147 #endif 3148 struct ip *ip; 3149 struct ip6_hdr *ip6; 3150 3151 ip = mtod(m, struct ip *); 3152 switch (ip->ip_v) { 3153 case 4: 3154 ip = mtod(m, struct ip *); 3155 ip6 = NULL; 3156 break; 3157 case 6: 3158 ip = NULL; 3159 ip6 = mtod(m, struct ip6_hdr *); 3160 break; 3161 default: 3162 return (NULL); 3163 } 3164 3165 #ifdef FAST_IPSEC 3166 /* Extract the destination from the IP header in the mbuf. */ 3167 memset(&dst, 0, sizeof(union sockaddr_union)); 3168 if (ip !=NULL) { 3169 dst.sa.sa_len = sizeof(struct sockaddr_in); 3170 dst.sa.sa_family = AF_INET; 3171 dst.sin.sin_addr = ip->ip_dst; 3172 } else { 3173 dst.sa.sa_len = sizeof(struct sockaddr_in6); 3174 dst.sa.sa_family = AF_INET6; 3175 dst.sin6.sin6_addr = ip6->ip6_dst; 3176 } 3177 3178 /* 3179 * Look up an SADB entry which matches the address of the peer. 3180 */ 3181 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 3182 #else 3183 if (ip) 3184 sav = key_allocsa(AF_INET, (void *)&ip->ip_src, 3185 (void *)&ip->ip_dst, IPPROTO_TCP, 3186 htonl(TCP_SIG_SPI), 0, 0); 3187 else 3188 sav = key_allocsa(AF_INET6, (void *)&ip6->ip6_src, 3189 (void *)&ip6->ip6_dst, IPPROTO_TCP, 3190 htonl(TCP_SIG_SPI), 0, 0); 3191 #endif 3192 3193 return (sav); /* freesav must be performed by caller */ 3194 } 3195 3196 int 3197 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 3198 struct secasvar *sav, char *sig) 3199 { 3200 MD5_CTX ctx; 3201 struct ip *ip; 3202 struct ipovly *ipovly; 3203 struct ip6_hdr *ip6; 3204 struct ippseudo ippseudo; 3205 struct ip6_hdr_pseudo ip6pseudo; 3206 struct tcphdr th0; 3207 int l, tcphdrlen; 3208 3209 if (sav == NULL) 3210 return (-1); 3211 3212 tcphdrlen = th->th_off * 4; 3213 3214 switch (mtod(m, struct ip *)->ip_v) { 3215 case 4: 3216 ip = mtod(m, struct ip *); 3217 ip6 = NULL; 3218 break; 3219 case 6: 3220 ip = NULL; 3221 ip6 = mtod(m, struct ip6_hdr *); 3222 break; 3223 default: 3224 return (-1); 3225 } 3226 3227 MD5Init(&ctx); 3228 3229 if (ip) { 3230 memset(&ippseudo, 0, sizeof(ippseudo)); 3231 ipovly = (struct ipovly *)ip; 3232 ippseudo.ippseudo_src = ipovly->ih_src; 3233 ippseudo.ippseudo_dst = ipovly->ih_dst; 3234 ippseudo.ippseudo_pad = 0; 3235 ippseudo.ippseudo_p = IPPROTO_TCP; 3236 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 3237 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 3238 } else { 3239 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 3240 ip6pseudo.ip6ph_src = ip6->ip6_src; 3241 in6_clearscope(&ip6pseudo.ip6ph_src); 3242 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 3243 in6_clearscope(&ip6pseudo.ip6ph_dst); 3244 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 3245 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 3246 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 3247 } 3248 3249 th0 = *th; 3250 th0.th_sum = 0; 3251 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 3252 3253 l = m->m_pkthdr.len - thoff - tcphdrlen; 3254 if (l > 0) 3255 m_apply(m, thoff + tcphdrlen, 3256 m->m_pkthdr.len - thoff - tcphdrlen, 3257 tcp_signature_apply, &ctx); 3258 3259 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 3260 MD5Final(sig, &ctx); 3261 3262 return (0); 3263 } 3264 #endif 3265 3266 /* 3267 * tcp_dooptions: parse and process tcp options. 3268 * 3269 * returns -1 if this segment should be dropped. (eg. wrong signature) 3270 * otherwise returns 0. 3271 */ 3272 3273 static int 3274 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, 3275 struct tcphdr *th, 3276 struct mbuf *m, int toff, struct tcp_opt_info *oi) 3277 { 3278 u_int16_t mss; 3279 int opt, optlen = 0; 3280 #ifdef TCP_SIGNATURE 3281 void *sigp = NULL; 3282 char sigbuf[TCP_SIGLEN]; 3283 struct secasvar *sav = NULL; 3284 #endif 3285 3286 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 3287 opt = cp[0]; 3288 if (opt == TCPOPT_EOL) 3289 break; 3290 if (opt == TCPOPT_NOP) 3291 optlen = 1; 3292 else { 3293 if (cnt < 2) 3294 break; 3295 optlen = cp[1]; 3296 if (optlen < 2 || optlen > cnt) 3297 break; 3298 } 3299 switch (opt) { 3300 3301 default: 3302 continue; 3303 3304 case TCPOPT_MAXSEG: 3305 if (optlen != TCPOLEN_MAXSEG) 3306 continue; 3307 if (!(th->th_flags & TH_SYN)) 3308 continue; 3309 if (TCPS_HAVERCVDSYN(tp->t_state)) 3310 continue; 3311 bcopy(cp + 2, &mss, sizeof(mss)); 3312 oi->maxseg = ntohs(mss); 3313 break; 3314 3315 case TCPOPT_WINDOW: 3316 if (optlen != TCPOLEN_WINDOW) 3317 continue; 3318 if (!(th->th_flags & TH_SYN)) 3319 continue; 3320 if (TCPS_HAVERCVDSYN(tp->t_state)) 3321 continue; 3322 tp->t_flags |= TF_RCVD_SCALE; 3323 tp->requested_s_scale = cp[2]; 3324 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3325 #if 0 /*XXX*/ 3326 char *p; 3327 3328 if (ip) 3329 p = ntohl(ip->ip_src); 3330 #ifdef INET6 3331 else if (ip6) 3332 p = ip6_sprintf(&ip6->ip6_src); 3333 #endif 3334 else 3335 p = "(unknown)"; 3336 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3337 "assuming %d\n", 3338 tp->requested_s_scale, p, 3339 TCP_MAX_WINSHIFT); 3340 #else 3341 log(LOG_ERR, "TCP: invalid wscale %d, " 3342 "assuming %d\n", 3343 tp->requested_s_scale, 3344 TCP_MAX_WINSHIFT); 3345 #endif 3346 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3347 } 3348 break; 3349 3350 case TCPOPT_TIMESTAMP: 3351 if (optlen != TCPOLEN_TIMESTAMP) 3352 continue; 3353 oi->ts_present = 1; 3354 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 3355 NTOHL(oi->ts_val); 3356 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 3357 NTOHL(oi->ts_ecr); 3358 3359 if (!(th->th_flags & TH_SYN)) 3360 continue; 3361 if (TCPS_HAVERCVDSYN(tp->t_state)) 3362 continue; 3363 /* 3364 * A timestamp received in a SYN makes 3365 * it ok to send timestamp requests and replies. 3366 */ 3367 tp->t_flags |= TF_RCVD_TSTMP; 3368 tp->ts_recent = oi->ts_val; 3369 tp->ts_recent_age = tcp_now; 3370 break; 3371 3372 case TCPOPT_SACK_PERMITTED: 3373 if (optlen != TCPOLEN_SACK_PERMITTED) 3374 continue; 3375 if (!(th->th_flags & TH_SYN)) 3376 continue; 3377 if (TCPS_HAVERCVDSYN(tp->t_state)) 3378 continue; 3379 if (tcp_do_sack) { 3380 tp->t_flags |= TF_SACK_PERMIT; 3381 tp->t_flags |= TF_WILL_SACK; 3382 } 3383 break; 3384 3385 case TCPOPT_SACK: 3386 tcp_sack_option(tp, th, cp, optlen); 3387 break; 3388 #ifdef TCP_SIGNATURE 3389 case TCPOPT_SIGNATURE: 3390 if (optlen != TCPOLEN_SIGNATURE) 3391 continue; 3392 if (sigp && memcmp(sigp, cp + 2, TCP_SIGLEN)) 3393 return (-1); 3394 3395 sigp = sigbuf; 3396 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3397 tp->t_flags |= TF_SIGNATURE; 3398 break; 3399 #endif 3400 } 3401 } 3402 3403 #ifdef TCP_SIGNATURE 3404 if (tp->t_flags & TF_SIGNATURE) { 3405 3406 sav = tcp_signature_getsav(m, th); 3407 3408 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3409 return (-1); 3410 } 3411 3412 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 3413 if (sav == NULL) 3414 return (-1); 3415 #ifdef FAST_IPSEC 3416 KEY_FREESAV(&sav); 3417 #else 3418 key_freesav(sav); 3419 #endif 3420 return (-1); 3421 } 3422 3423 if (sigp) { 3424 char sig[TCP_SIGLEN]; 3425 3426 tcp_fields_to_net(th); 3427 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3428 tcp_fields_to_host(th); 3429 if (sav == NULL) 3430 return (-1); 3431 #ifdef FAST_IPSEC 3432 KEY_FREESAV(&sav); 3433 #else 3434 key_freesav(sav); 3435 #endif 3436 return (-1); 3437 } 3438 tcp_fields_to_host(th); 3439 3440 if (memcmp(sig, sigp, TCP_SIGLEN)) { 3441 TCP_STATINC(TCP_STAT_BADSIG); 3442 if (sav == NULL) 3443 return (-1); 3444 #ifdef FAST_IPSEC 3445 KEY_FREESAV(&sav); 3446 #else 3447 key_freesav(sav); 3448 #endif 3449 return (-1); 3450 } else 3451 TCP_STATINC(TCP_STAT_GOODSIG); 3452 3453 key_sa_recordxfer(sav, m); 3454 #ifdef FAST_IPSEC 3455 KEY_FREESAV(&sav); 3456 #else 3457 key_freesav(sav); 3458 #endif 3459 } 3460 #endif 3461 3462 return (0); 3463 } 3464 3465 /* 3466 * Pull out of band byte out of a segment so 3467 * it doesn't appear in the user's data queue. 3468 * It is still reflected in the segment length for 3469 * sequencing purposes. 3470 */ 3471 void 3472 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3473 struct mbuf *m, int off) 3474 { 3475 int cnt = off + th->th_urp - 1; 3476 3477 while (cnt >= 0) { 3478 if (m->m_len > cnt) { 3479 char *cp = mtod(m, char *) + cnt; 3480 struct tcpcb *tp = sototcpcb(so); 3481 3482 tp->t_iobc = *cp; 3483 tp->t_oobflags |= TCPOOB_HAVEDATA; 3484 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3485 m->m_len--; 3486 return; 3487 } 3488 cnt -= m->m_len; 3489 m = m->m_next; 3490 if (m == 0) 3491 break; 3492 } 3493 panic("tcp_pulloutofband"); 3494 } 3495 3496 /* 3497 * Collect new round-trip time estimate 3498 * and update averages and current timeout. 3499 * 3500 * rtt is in units of slow ticks (typically 500 ms) -- essentially the 3501 * difference of two timestamps. 3502 */ 3503 void 3504 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3505 { 3506 int32_t delta; 3507 3508 TCP_STATINC(TCP_STAT_RTTUPDATED); 3509 if (tp->t_srtt != 0) { 3510 /* 3511 * Compute the amount to add to srtt for smoothing, 3512 * *alpha, or 2^(-TCP_RTT_SHIFT). Because 3513 * srtt is stored in 1/32 slow ticks, we conceptually 3514 * shift left 5 bits, subtract srtt to get the 3515 * diference, and then shift right by TCP_RTT_SHIFT 3516 * (3) to obtain 1/8 of the difference. 3517 */ 3518 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3519 /* 3520 * This can never happen, because delta's lowest 3521 * possible value is 1/8 of t_srtt. But if it does, 3522 * set srtt to some reasonable value, here chosen 3523 * as 1/8 tick. 3524 */ 3525 if ((tp->t_srtt += delta) <= 0) 3526 tp->t_srtt = 1 << 2; 3527 /* 3528 * RFC2988 requires that rttvar be updated first. 3529 * This code is compliant because "delta" is the old 3530 * srtt minus the new observation (scaled). 3531 * 3532 * RFC2988 says: 3533 * rttvar = (1-beta) * rttvar + beta * |srtt-observed| 3534 * 3535 * delta is in units of 1/32 ticks, and has then been 3536 * divided by 8. This is equivalent to being in 1/16s 3537 * units and divided by 4. Subtract from it 1/4 of 3538 * the existing rttvar to form the (signed) amount to 3539 * adjust. 3540 */ 3541 if (delta < 0) 3542 delta = -delta; 3543 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3544 /* 3545 * As with srtt, this should never happen. There is 3546 * no support in RFC2988 for this operation. But 1/4s 3547 * as rttvar when faced with something arguably wrong 3548 * is ok. 3549 */ 3550 if ((tp->t_rttvar += delta) <= 0) 3551 tp->t_rttvar = 1 << 2; 3552 3553 /* 3554 * If srtt exceeds .01 second, ensure we use the 'remote' MSL 3555 * Problem is: it doesn't work. Disabled by defaulting 3556 * tcp_rttlocal to 0; see corresponding code in 3557 * tcp_subr that selects local vs remote in a different way. 3558 * 3559 * The static branch prediction hint here should be removed 3560 * when the rtt estimator is fixed and the rtt_enable code 3561 * is turned back on. 3562 */ 3563 if (__predict_false(tcp_rttlocal) && tcp_msl_enable 3564 && tp->t_srtt > tcp_msl_remote_threshold 3565 && tp->t_msl < tcp_msl_remote) { 3566 tp->t_msl = tcp_msl_remote; 3567 } 3568 } else { 3569 /* 3570 * This is the first measurement. Per RFC2988, 2.2, 3571 * set rtt=R and srtt=R/2. 3572 * For srtt, storage representation is 1/32 ticks, 3573 * so shift left by 5. 3574 * For rttvar, storage representation is 1/16 ticks, 3575 * So shift left by 4, but then right by 1 to halve. 3576 */ 3577 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3578 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3579 } 3580 tp->t_rtttime = 0; 3581 tp->t_rxtshift = 0; 3582 3583 /* 3584 * the retransmit should happen at rtt + 4 * rttvar. 3585 * Because of the way we do the smoothing, srtt and rttvar 3586 * will each average +1/2 tick of bias. When we compute 3587 * the retransmit timer, we want 1/2 tick of rounding and 3588 * 1 extra tick because of +-1/2 tick uncertainty in the 3589 * firing of the timer. The bias will give us exactly the 3590 * 1.5 tick we need. But, because the bias is 3591 * statistical, we have to test that we don't drop below 3592 * the minimum feasible timer (which is 2 ticks). 3593 */ 3594 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3595 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3596 3597 /* 3598 * We received an ack for a packet that wasn't retransmitted; 3599 * it is probably safe to discard any error indications we've 3600 * received recently. This isn't quite right, but close enough 3601 * for now (a route might have failed after we sent a segment, 3602 * and the return path might not be symmetrical). 3603 */ 3604 tp->t_softerror = 0; 3605 } 3606 3607 3608 /* 3609 * TCP compressed state engine. Currently used to hold compressed 3610 * state for SYN_RECEIVED. 3611 */ 3612 3613 u_long syn_cache_count; 3614 u_int32_t syn_hash1, syn_hash2; 3615 3616 #define SYN_HASH(sa, sp, dp) \ 3617 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3618 ((u_int32_t)(sp)))^syn_hash2))) 3619 #ifndef INET6 3620 #define SYN_HASHALL(hash, src, dst) \ 3621 do { \ 3622 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3623 ((const struct sockaddr_in *)(src))->sin_port, \ 3624 ((const struct sockaddr_in *)(dst))->sin_port); \ 3625 } while (/*CONSTCOND*/ 0) 3626 #else 3627 #define SYN_HASH6(sa, sp, dp) \ 3628 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3629 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3630 & 0x7fffffff) 3631 3632 #define SYN_HASHALL(hash, src, dst) \ 3633 do { \ 3634 switch ((src)->sa_family) { \ 3635 case AF_INET: \ 3636 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3637 ((const struct sockaddr_in *)(src))->sin_port, \ 3638 ((const struct sockaddr_in *)(dst))->sin_port); \ 3639 break; \ 3640 case AF_INET6: \ 3641 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3642 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3643 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3644 break; \ 3645 default: \ 3646 hash = 0; \ 3647 } \ 3648 } while (/*CONSTCOND*/0) 3649 #endif /* INET6 */ 3650 3651 static struct pool syn_cache_pool; 3652 3653 /* 3654 * We don't estimate RTT with SYNs, so each packet starts with the default 3655 * RTT and each timer step has a fixed timeout value. 3656 */ 3657 #define SYN_CACHE_TIMER_ARM(sc) \ 3658 do { \ 3659 TCPT_RANGESET((sc)->sc_rxtcur, \ 3660 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3661 TCPTV_REXMTMAX); \ 3662 callout_reset(&(sc)->sc_timer, \ 3663 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3664 } while (/*CONSTCOND*/0) 3665 3666 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3667 3668 static inline void 3669 syn_cache_rm(struct syn_cache *sc) 3670 { 3671 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3672 sc, sc_bucketq); 3673 sc->sc_tp = NULL; 3674 LIST_REMOVE(sc, sc_tpq); 3675 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3676 callout_stop(&sc->sc_timer); 3677 syn_cache_count--; 3678 } 3679 3680 static inline void 3681 syn_cache_put(struct syn_cache *sc) 3682 { 3683 if (sc->sc_ipopts) 3684 (void) m_free(sc->sc_ipopts); 3685 rtcache_free(&sc->sc_route); 3686 sc->sc_flags |= SCF_DEAD; 3687 if (!callout_invoking(&sc->sc_timer)) 3688 callout_schedule(&(sc)->sc_timer, 1); 3689 } 3690 3691 void 3692 syn_cache_init(void) 3693 { 3694 int i; 3695 3696 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 3697 "synpl", NULL, IPL_SOFTNET); 3698 3699 /* Initialize the hash buckets. */ 3700 for (i = 0; i < tcp_syn_cache_size; i++) 3701 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3702 } 3703 3704 void 3705 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3706 { 3707 struct syn_cache_head *scp; 3708 struct syn_cache *sc2; 3709 int s; 3710 3711 /* 3712 * If there are no entries in the hash table, reinitialize 3713 * the hash secrets. 3714 */ 3715 if (syn_cache_count == 0) { 3716 syn_hash1 = arc4random(); 3717 syn_hash2 = arc4random(); 3718 } 3719 3720 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3721 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3722 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3723 3724 /* 3725 * Make sure that we don't overflow the per-bucket 3726 * limit or the total cache size limit. 3727 */ 3728 s = splsoftnet(); 3729 if (scp->sch_length >= tcp_syn_bucket_limit) { 3730 TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW); 3731 /* 3732 * The bucket is full. Toss the oldest element in the 3733 * bucket. This will be the first entry in the bucket. 3734 */ 3735 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3736 #ifdef DIAGNOSTIC 3737 /* 3738 * This should never happen; we should always find an 3739 * entry in our bucket. 3740 */ 3741 if (sc2 == NULL) 3742 panic("syn_cache_insert: bucketoverflow: impossible"); 3743 #endif 3744 syn_cache_rm(sc2); 3745 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3746 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3747 struct syn_cache_head *scp2, *sce; 3748 3749 TCP_STATINC(TCP_STAT_SC_OVERFLOWED); 3750 /* 3751 * The cache is full. Toss the oldest entry in the 3752 * first non-empty bucket we can find. 3753 * 3754 * XXX We would really like to toss the oldest 3755 * entry in the cache, but we hope that this 3756 * condition doesn't happen very often. 3757 */ 3758 scp2 = scp; 3759 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3760 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3761 for (++scp2; scp2 != scp; scp2++) { 3762 if (scp2 >= sce) 3763 scp2 = &tcp_syn_cache[0]; 3764 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3765 break; 3766 } 3767 #ifdef DIAGNOSTIC 3768 /* 3769 * This should never happen; we should always find a 3770 * non-empty bucket. 3771 */ 3772 if (scp2 == scp) 3773 panic("syn_cache_insert: cacheoverflow: " 3774 "impossible"); 3775 #endif 3776 } 3777 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3778 syn_cache_rm(sc2); 3779 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3780 } 3781 3782 /* 3783 * Initialize the entry's timer. 3784 */ 3785 sc->sc_rxttot = 0; 3786 sc->sc_rxtshift = 0; 3787 SYN_CACHE_TIMER_ARM(sc); 3788 3789 /* Link it from tcpcb entry */ 3790 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3791 3792 /* Put it into the bucket. */ 3793 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3794 scp->sch_length++; 3795 syn_cache_count++; 3796 3797 TCP_STATINC(TCP_STAT_SC_ADDED); 3798 splx(s); 3799 } 3800 3801 /* 3802 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3803 * If we have retransmitted an entry the maximum number of times, expire 3804 * that entry. 3805 */ 3806 void 3807 syn_cache_timer(void *arg) 3808 { 3809 struct syn_cache *sc = arg; 3810 3811 mutex_enter(softnet_lock); 3812 KERNEL_LOCK(1, NULL); 3813 callout_ack(&sc->sc_timer); 3814 3815 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3816 TCP_STATINC(TCP_STAT_SC_DELAYED_FREE); 3817 callout_destroy(&sc->sc_timer); 3818 pool_put(&syn_cache_pool, sc); 3819 KERNEL_UNLOCK_ONE(NULL); 3820 mutex_exit(softnet_lock); 3821 return; 3822 } 3823 3824 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3825 /* Drop it -- too many retransmissions. */ 3826 goto dropit; 3827 } 3828 3829 /* 3830 * Compute the total amount of time this entry has 3831 * been on a queue. If this entry has been on longer 3832 * than the keep alive timer would allow, expire it. 3833 */ 3834 sc->sc_rxttot += sc->sc_rxtcur; 3835 if (sc->sc_rxttot >= tcp_keepinit) 3836 goto dropit; 3837 3838 TCP_STATINC(TCP_STAT_SC_RETRANSMITTED); 3839 (void) syn_cache_respond(sc, NULL); 3840 3841 /* Advance the timer back-off. */ 3842 sc->sc_rxtshift++; 3843 SYN_CACHE_TIMER_ARM(sc); 3844 3845 KERNEL_UNLOCK_ONE(NULL); 3846 mutex_exit(softnet_lock); 3847 return; 3848 3849 dropit: 3850 TCP_STATINC(TCP_STAT_SC_TIMED_OUT); 3851 syn_cache_rm(sc); 3852 if (sc->sc_ipopts) 3853 (void) m_free(sc->sc_ipopts); 3854 rtcache_free(&sc->sc_route); 3855 callout_destroy(&sc->sc_timer); 3856 pool_put(&syn_cache_pool, sc); 3857 KERNEL_UNLOCK_ONE(NULL); 3858 mutex_exit(softnet_lock); 3859 } 3860 3861 /* 3862 * Remove syn cache created by the specified tcb entry, 3863 * because this does not make sense to keep them 3864 * (if there's no tcb entry, syn cache entry will never be used) 3865 */ 3866 void 3867 syn_cache_cleanup(struct tcpcb *tp) 3868 { 3869 struct syn_cache *sc, *nsc; 3870 int s; 3871 3872 s = splsoftnet(); 3873 3874 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3875 nsc = LIST_NEXT(sc, sc_tpq); 3876 3877 #ifdef DIAGNOSTIC 3878 if (sc->sc_tp != tp) 3879 panic("invalid sc_tp in syn_cache_cleanup"); 3880 #endif 3881 syn_cache_rm(sc); 3882 syn_cache_put(sc); /* calls pool_put but see spl above */ 3883 } 3884 /* just for safety */ 3885 LIST_INIT(&tp->t_sc); 3886 3887 splx(s); 3888 } 3889 3890 /* 3891 * Find an entry in the syn cache. 3892 */ 3893 struct syn_cache * 3894 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3895 struct syn_cache_head **headp) 3896 { 3897 struct syn_cache *sc; 3898 struct syn_cache_head *scp; 3899 u_int32_t hash; 3900 int s; 3901 3902 SYN_HASHALL(hash, src, dst); 3903 3904 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3905 *headp = scp; 3906 s = splsoftnet(); 3907 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3908 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3909 if (sc->sc_hash != hash) 3910 continue; 3911 if (!memcmp(&sc->sc_src, src, src->sa_len) && 3912 !memcmp(&sc->sc_dst, dst, dst->sa_len)) { 3913 splx(s); 3914 return (sc); 3915 } 3916 } 3917 splx(s); 3918 return (NULL); 3919 } 3920 3921 /* 3922 * This function gets called when we receive an ACK for a 3923 * socket in the LISTEN state. We look up the connection 3924 * in the syn cache, and if its there, we pull it out of 3925 * the cache and turn it into a full-blown connection in 3926 * the SYN-RECEIVED state. 3927 * 3928 * The return values may not be immediately obvious, and their effects 3929 * can be subtle, so here they are: 3930 * 3931 * NULL SYN was not found in cache; caller should drop the 3932 * packet and send an RST. 3933 * 3934 * -1 We were unable to create the new connection, and are 3935 * aborting it. An ACK,RST is being sent to the peer 3936 * (unless we got screwey sequence numbners; see below), 3937 * because the 3-way handshake has been completed. Caller 3938 * should not free the mbuf, since we may be using it. If 3939 * we are not, we will free it. 3940 * 3941 * Otherwise, the return value is a pointer to the new socket 3942 * associated with the connection. 3943 */ 3944 struct socket * 3945 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3946 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3947 struct socket *so, struct mbuf *m) 3948 { 3949 struct syn_cache *sc; 3950 struct syn_cache_head *scp; 3951 struct inpcb *inp = NULL; 3952 #ifdef INET6 3953 struct in6pcb *in6p = NULL; 3954 #endif 3955 struct tcpcb *tp = 0; 3956 struct mbuf *am; 3957 int s; 3958 struct socket *oso; 3959 3960 s = splsoftnet(); 3961 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3962 splx(s); 3963 return (NULL); 3964 } 3965 3966 /* 3967 * Verify the sequence and ack numbers. Try getting the correct 3968 * response again. 3969 */ 3970 if ((th->th_ack != sc->sc_iss + 1) || 3971 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3972 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3973 (void) syn_cache_respond(sc, m); 3974 splx(s); 3975 return ((struct socket *)(-1)); 3976 } 3977 3978 /* Remove this cache entry */ 3979 syn_cache_rm(sc); 3980 splx(s); 3981 3982 /* 3983 * Ok, create the full blown connection, and set things up 3984 * as they would have been set up if we had created the 3985 * connection when the SYN arrived. If we can't create 3986 * the connection, abort it. 3987 */ 3988 /* 3989 * inp still has the OLD in_pcb stuff, set the 3990 * v6-related flags on the new guy, too. This is 3991 * done particularly for the case where an AF_INET6 3992 * socket is bound only to a port, and a v4 connection 3993 * comes in on that port. 3994 * we also copy the flowinfo from the original pcb 3995 * to the new one. 3996 */ 3997 oso = so; 3998 so = sonewconn(so, SS_ISCONNECTED); 3999 if (so == NULL) 4000 goto resetandabort; 4001 4002 switch (so->so_proto->pr_domain->dom_family) { 4003 #ifdef INET 4004 case AF_INET: 4005 inp = sotoinpcb(so); 4006 break; 4007 #endif 4008 #ifdef INET6 4009 case AF_INET6: 4010 in6p = sotoin6pcb(so); 4011 break; 4012 #endif 4013 } 4014 switch (src->sa_family) { 4015 #ifdef INET 4016 case AF_INET: 4017 if (inp) { 4018 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 4019 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 4020 inp->inp_options = ip_srcroute(); 4021 in_pcbstate(inp, INP_BOUND); 4022 if (inp->inp_options == NULL) { 4023 inp->inp_options = sc->sc_ipopts; 4024 sc->sc_ipopts = NULL; 4025 } 4026 } 4027 #ifdef INET6 4028 else if (in6p) { 4029 /* IPv4 packet to AF_INET6 socket */ 4030 memset(&in6p->in6p_laddr, 0, sizeof(in6p->in6p_laddr)); 4031 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 4032 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 4033 &in6p->in6p_laddr.s6_addr32[3], 4034 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 4035 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 4036 in6totcpcb(in6p)->t_family = AF_INET; 4037 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 4038 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 4039 else 4040 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 4041 in6_pcbstate(in6p, IN6P_BOUND); 4042 } 4043 #endif 4044 break; 4045 #endif 4046 #ifdef INET6 4047 case AF_INET6: 4048 if (in6p) { 4049 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 4050 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 4051 in6_pcbstate(in6p, IN6P_BOUND); 4052 } 4053 break; 4054 #endif 4055 } 4056 #ifdef INET6 4057 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 4058 struct in6pcb *oin6p = sotoin6pcb(oso); 4059 /* inherit socket options from the listening socket */ 4060 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 4061 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 4062 m_freem(in6p->in6p_options); 4063 in6p->in6p_options = 0; 4064 } 4065 ip6_savecontrol(in6p, &in6p->in6p_options, 4066 mtod(m, struct ip6_hdr *), m); 4067 } 4068 #endif 4069 4070 #if defined(IPSEC) || defined(FAST_IPSEC) 4071 /* 4072 * we make a copy of policy, instead of sharing the policy, 4073 * for better behavior in terms of SA lookup and dead SA removal. 4074 */ 4075 if (inp) { 4076 /* copy old policy into new socket's */ 4077 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 4078 printf("tcp_input: could not copy policy\n"); 4079 } 4080 #ifdef INET6 4081 else if (in6p) { 4082 /* copy old policy into new socket's */ 4083 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 4084 in6p->in6p_sp)) 4085 printf("tcp_input: could not copy policy\n"); 4086 } 4087 #endif 4088 #endif 4089 4090 /* 4091 * Give the new socket our cached route reference. 4092 */ 4093 if (inp) { 4094 rtcache_copy(&inp->inp_route, &sc->sc_route); 4095 rtcache_free(&sc->sc_route); 4096 } 4097 #ifdef INET6 4098 else { 4099 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 4100 rtcache_free(&sc->sc_route); 4101 } 4102 #endif 4103 4104 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 4105 if (am == NULL) 4106 goto resetandabort; 4107 MCLAIM(am, &tcp_mowner); 4108 am->m_len = src->sa_len; 4109 bcopy(src, mtod(am, void *), src->sa_len); 4110 if (inp) { 4111 if (in_pcbconnect(inp, am, &lwp0)) { 4112 (void) m_free(am); 4113 goto resetandabort; 4114 } 4115 } 4116 #ifdef INET6 4117 else if (in6p) { 4118 if (src->sa_family == AF_INET) { 4119 /* IPv4 packet to AF_INET6 socket */ 4120 struct sockaddr_in6 *sin6; 4121 sin6 = mtod(am, struct sockaddr_in6 *); 4122 am->m_len = sizeof(*sin6); 4123 memset(sin6, 0, sizeof(*sin6)); 4124 sin6->sin6_family = AF_INET6; 4125 sin6->sin6_len = sizeof(*sin6); 4126 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 4127 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 4128 bcopy(&((struct sockaddr_in *)src)->sin_addr, 4129 &sin6->sin6_addr.s6_addr32[3], 4130 sizeof(sin6->sin6_addr.s6_addr32[3])); 4131 } 4132 if (in6_pcbconnect(in6p, am, NULL)) { 4133 (void) m_free(am); 4134 goto resetandabort; 4135 } 4136 } 4137 #endif 4138 else { 4139 (void) m_free(am); 4140 goto resetandabort; 4141 } 4142 (void) m_free(am); 4143 4144 if (inp) 4145 tp = intotcpcb(inp); 4146 #ifdef INET6 4147 else if (in6p) 4148 tp = in6totcpcb(in6p); 4149 #endif 4150 else 4151 tp = NULL; 4152 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 4153 if (sc->sc_request_r_scale != 15) { 4154 tp->requested_s_scale = sc->sc_requested_s_scale; 4155 tp->request_r_scale = sc->sc_request_r_scale; 4156 tp->snd_scale = sc->sc_requested_s_scale; 4157 tp->rcv_scale = sc->sc_request_r_scale; 4158 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 4159 } 4160 if (sc->sc_flags & SCF_TIMESTAMP) 4161 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 4162 tp->ts_timebase = sc->sc_timebase; 4163 4164 tp->t_template = tcp_template(tp); 4165 if (tp->t_template == 0) { 4166 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 4167 so = NULL; 4168 m_freem(m); 4169 goto abort; 4170 } 4171 4172 tp->iss = sc->sc_iss; 4173 tp->irs = sc->sc_irs; 4174 tcp_sendseqinit(tp); 4175 tcp_rcvseqinit(tp); 4176 tp->t_state = TCPS_SYN_RECEIVED; 4177 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 4178 TCP_STATINC(TCP_STAT_ACCEPTS); 4179 4180 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 4181 tp->t_flags |= TF_WILL_SACK; 4182 4183 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 4184 tp->t_flags |= TF_ECN_PERMIT; 4185 4186 #ifdef TCP_SIGNATURE 4187 if (sc->sc_flags & SCF_SIGNATURE) 4188 tp->t_flags |= TF_SIGNATURE; 4189 #endif 4190 4191 /* Initialize tp->t_ourmss before we deal with the peer's! */ 4192 tp->t_ourmss = sc->sc_ourmaxseg; 4193 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 4194 4195 /* 4196 * Initialize the initial congestion window. If we 4197 * had to retransmit the SYN,ACK, we must initialize cwnd 4198 * to 1 segment (i.e. the Loss Window). 4199 */ 4200 if (sc->sc_rxtshift) 4201 tp->snd_cwnd = tp->t_peermss; 4202 else { 4203 int ss = tcp_init_win; 4204 #ifdef INET 4205 if (inp != NULL && in_localaddr(inp->inp_faddr)) 4206 ss = tcp_init_win_local; 4207 #endif 4208 #ifdef INET6 4209 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 4210 ss = tcp_init_win_local; 4211 #endif 4212 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 4213 } 4214 4215 tcp_rmx_rtt(tp); 4216 tp->snd_wl1 = sc->sc_irs; 4217 tp->rcv_up = sc->sc_irs + 1; 4218 4219 /* 4220 * This is what whould have happened in tcp_output() when 4221 * the SYN,ACK was sent. 4222 */ 4223 tp->snd_up = tp->snd_una; 4224 tp->snd_max = tp->snd_nxt = tp->iss+1; 4225 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 4226 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 4227 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 4228 tp->last_ack_sent = tp->rcv_nxt; 4229 tp->t_partialacks = -1; 4230 tp->t_dupacks = 0; 4231 4232 TCP_STATINC(TCP_STAT_SC_COMPLETED); 4233 s = splsoftnet(); 4234 syn_cache_put(sc); 4235 splx(s); 4236 return (so); 4237 4238 resetandabort: 4239 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 4240 abort: 4241 if (so != NULL) { 4242 (void) soqremque(so, 1); 4243 (void) soabort(so); 4244 mutex_enter(softnet_lock); 4245 } 4246 s = splsoftnet(); 4247 syn_cache_put(sc); 4248 splx(s); 4249 TCP_STATINC(TCP_STAT_SC_ABORTED); 4250 return ((struct socket *)(-1)); 4251 } 4252 4253 /* 4254 * This function is called when we get a RST for a 4255 * non-existent connection, so that we can see if the 4256 * connection is in the syn cache. If it is, zap it. 4257 */ 4258 4259 void 4260 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 4261 { 4262 struct syn_cache *sc; 4263 struct syn_cache_head *scp; 4264 int s = splsoftnet(); 4265 4266 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4267 splx(s); 4268 return; 4269 } 4270 if (SEQ_LT(th->th_seq, sc->sc_irs) || 4271 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 4272 splx(s); 4273 return; 4274 } 4275 syn_cache_rm(sc); 4276 TCP_STATINC(TCP_STAT_SC_RESET); 4277 syn_cache_put(sc); /* calls pool_put but see spl above */ 4278 splx(s); 4279 } 4280 4281 void 4282 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 4283 struct tcphdr *th) 4284 { 4285 struct syn_cache *sc; 4286 struct syn_cache_head *scp; 4287 int s; 4288 4289 s = splsoftnet(); 4290 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4291 splx(s); 4292 return; 4293 } 4294 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 4295 if (ntohl (th->th_seq) != sc->sc_iss) { 4296 splx(s); 4297 return; 4298 } 4299 4300 /* 4301 * If we've retransmitted 3 times and this is our second error, 4302 * we remove the entry. Otherwise, we allow it to continue on. 4303 * This prevents us from incorrectly nuking an entry during a 4304 * spurious network outage. 4305 * 4306 * See tcp_notify(). 4307 */ 4308 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 4309 sc->sc_flags |= SCF_UNREACH; 4310 splx(s); 4311 return; 4312 } 4313 4314 syn_cache_rm(sc); 4315 TCP_STATINC(TCP_STAT_SC_UNREACH); 4316 syn_cache_put(sc); /* calls pool_put but see spl above */ 4317 splx(s); 4318 } 4319 4320 /* 4321 * Given a LISTEN socket and an inbound SYN request, add 4322 * this to the syn cache, and send back a segment: 4323 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 4324 * to the source. 4325 * 4326 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 4327 * Doing so would require that we hold onto the data and deliver it 4328 * to the application. However, if we are the target of a SYN-flood 4329 * DoS attack, an attacker could send data which would eventually 4330 * consume all available buffer space if it were ACKed. By not ACKing 4331 * the data, we avoid this DoS scenario. 4332 */ 4333 4334 int 4335 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 4336 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 4337 int optlen, struct tcp_opt_info *oi) 4338 { 4339 struct tcpcb tb, *tp; 4340 long win; 4341 struct syn_cache *sc; 4342 struct syn_cache_head *scp; 4343 struct mbuf *ipopts; 4344 struct tcp_opt_info opti; 4345 int s; 4346 4347 tp = sototcpcb(so); 4348 4349 memset(&opti, 0, sizeof(opti)); 4350 4351 /* 4352 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 4353 * 4354 * Note this check is performed in tcp_input() very early on. 4355 */ 4356 4357 /* 4358 * Initialize some local state. 4359 */ 4360 win = sbspace(&so->so_rcv); 4361 if (win > TCP_MAXWIN) 4362 win = TCP_MAXWIN; 4363 4364 switch (src->sa_family) { 4365 #ifdef INET 4366 case AF_INET: 4367 /* 4368 * Remember the IP options, if any. 4369 */ 4370 ipopts = ip_srcroute(); 4371 break; 4372 #endif 4373 default: 4374 ipopts = NULL; 4375 } 4376 4377 #ifdef TCP_SIGNATURE 4378 if (optp || (tp->t_flags & TF_SIGNATURE)) 4379 #else 4380 if (optp) 4381 #endif 4382 { 4383 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4384 #ifdef TCP_SIGNATURE 4385 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4386 #endif 4387 tb.t_state = TCPS_LISTEN; 4388 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 4389 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 4390 return (0); 4391 } else 4392 tb.t_flags = 0; 4393 4394 /* 4395 * See if we already have an entry for this connection. 4396 * If we do, resend the SYN,ACK. We do not count this 4397 * as a retransmission (XXX though maybe we should). 4398 */ 4399 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4400 TCP_STATINC(TCP_STAT_SC_DUPESYN); 4401 if (ipopts) { 4402 /* 4403 * If we were remembering a previous source route, 4404 * forget it and use the new one we've been given. 4405 */ 4406 if (sc->sc_ipopts) 4407 (void) m_free(sc->sc_ipopts); 4408 sc->sc_ipopts = ipopts; 4409 } 4410 sc->sc_timestamp = tb.ts_recent; 4411 if (syn_cache_respond(sc, m) == 0) { 4412 uint64_t *tcps = TCP_STAT_GETREF(); 4413 tcps[TCP_STAT_SNDACKS]++; 4414 tcps[TCP_STAT_SNDTOTAL]++; 4415 TCP_STAT_PUTREF(); 4416 } 4417 return (1); 4418 } 4419 4420 s = splsoftnet(); 4421 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4422 splx(s); 4423 if (sc == NULL) { 4424 if (ipopts) 4425 (void) m_free(ipopts); 4426 return (0); 4427 } 4428 4429 /* 4430 * Fill in the cache, and put the necessary IP and TCP 4431 * options into the reply. 4432 */ 4433 memset(sc, 0, sizeof(struct syn_cache)); 4434 callout_init(&sc->sc_timer, CALLOUT_MPSAFE); 4435 bcopy(src, &sc->sc_src, src->sa_len); 4436 bcopy(dst, &sc->sc_dst, dst->sa_len); 4437 sc->sc_flags = 0; 4438 sc->sc_ipopts = ipopts; 4439 sc->sc_irs = th->th_seq; 4440 switch (src->sa_family) { 4441 #ifdef INET 4442 case AF_INET: 4443 { 4444 struct sockaddr_in *srcin = (void *) src; 4445 struct sockaddr_in *dstin = (void *) dst; 4446 4447 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4448 &srcin->sin_addr, dstin->sin_port, 4449 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4450 break; 4451 } 4452 #endif /* INET */ 4453 #ifdef INET6 4454 case AF_INET6: 4455 { 4456 struct sockaddr_in6 *srcin6 = (void *) src; 4457 struct sockaddr_in6 *dstin6 = (void *) dst; 4458 4459 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4460 &srcin6->sin6_addr, dstin6->sin6_port, 4461 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4462 break; 4463 } 4464 #endif /* INET6 */ 4465 } 4466 sc->sc_peermaxseg = oi->maxseg; 4467 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4468 m->m_pkthdr.rcvif : NULL, 4469 sc->sc_src.sa.sa_family); 4470 sc->sc_win = win; 4471 sc->sc_timebase = tcp_now - 1; /* see tcp_newtcpcb() */ 4472 sc->sc_timestamp = tb.ts_recent; 4473 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4474 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4475 sc->sc_flags |= SCF_TIMESTAMP; 4476 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4477 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4478 sc->sc_requested_s_scale = tb.requested_s_scale; 4479 sc->sc_request_r_scale = 0; 4480 /* 4481 * Pick the smallest possible scaling factor that 4482 * will still allow us to scale up to sb_max. 4483 * 4484 * We do this because there are broken firewalls that 4485 * will corrupt the window scale option, leading to 4486 * the other endpoint believing that our advertised 4487 * window is unscaled. At scale factors larger than 4488 * 5 the unscaled window will drop below 1500 bytes, 4489 * leading to serious problems when traversing these 4490 * broken firewalls. 4491 * 4492 * With the default sbmax of 256K, a scale factor 4493 * of 3 will be chosen by this algorithm. Those who 4494 * choose a larger sbmax should watch out 4495 * for the compatiblity problems mentioned above. 4496 * 4497 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4498 * or <SYN,ACK>) segment itself is never scaled. 4499 */ 4500 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4501 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4502 sc->sc_request_r_scale++; 4503 } else { 4504 sc->sc_requested_s_scale = 15; 4505 sc->sc_request_r_scale = 15; 4506 } 4507 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4508 sc->sc_flags |= SCF_SACK_PERMIT; 4509 4510 /* 4511 * ECN setup packet recieved. 4512 */ 4513 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4514 sc->sc_flags |= SCF_ECN_PERMIT; 4515 4516 #ifdef TCP_SIGNATURE 4517 if (tb.t_flags & TF_SIGNATURE) 4518 sc->sc_flags |= SCF_SIGNATURE; 4519 #endif 4520 sc->sc_tp = tp; 4521 if (syn_cache_respond(sc, m) == 0) { 4522 uint64_t *tcps = TCP_STAT_GETREF(); 4523 tcps[TCP_STAT_SNDACKS]++; 4524 tcps[TCP_STAT_SNDTOTAL]++; 4525 TCP_STAT_PUTREF(); 4526 syn_cache_insert(sc, tp); 4527 } else { 4528 s = splsoftnet(); 4529 /* 4530 * syn_cache_put() will try to schedule the timer, so 4531 * we need to initialize it 4532 */ 4533 SYN_CACHE_TIMER_ARM(sc); 4534 syn_cache_put(sc); 4535 splx(s); 4536 TCP_STATINC(TCP_STAT_SC_DROPPED); 4537 } 4538 return (1); 4539 } 4540 4541 int 4542 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4543 { 4544 #ifdef INET6 4545 struct rtentry *rt; 4546 #endif 4547 struct route *ro; 4548 u_int8_t *optp; 4549 int optlen, error; 4550 u_int16_t tlen; 4551 struct ip *ip = NULL; 4552 #ifdef INET6 4553 struct ip6_hdr *ip6 = NULL; 4554 #endif 4555 struct tcpcb *tp = NULL; 4556 struct tcphdr *th; 4557 u_int hlen; 4558 struct socket *so; 4559 4560 ro = &sc->sc_route; 4561 switch (sc->sc_src.sa.sa_family) { 4562 case AF_INET: 4563 hlen = sizeof(struct ip); 4564 break; 4565 #ifdef INET6 4566 case AF_INET6: 4567 hlen = sizeof(struct ip6_hdr); 4568 break; 4569 #endif 4570 default: 4571 if (m) 4572 m_freem(m); 4573 return (EAFNOSUPPORT); 4574 } 4575 4576 /* Compute the size of the TCP options. */ 4577 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4578 ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) + 4579 #ifdef TCP_SIGNATURE 4580 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4581 #endif 4582 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4583 4584 tlen = hlen + sizeof(struct tcphdr) + optlen; 4585 4586 /* 4587 * Create the IP+TCP header from scratch. 4588 */ 4589 if (m) 4590 m_freem(m); 4591 #ifdef DIAGNOSTIC 4592 if (max_linkhdr + tlen > MCLBYTES) 4593 return (ENOBUFS); 4594 #endif 4595 MGETHDR(m, M_DONTWAIT, MT_DATA); 4596 if (m && (max_linkhdr + tlen) > MHLEN) { 4597 MCLGET(m, M_DONTWAIT); 4598 if ((m->m_flags & M_EXT) == 0) { 4599 m_freem(m); 4600 m = NULL; 4601 } 4602 } 4603 if (m == NULL) 4604 return (ENOBUFS); 4605 MCLAIM(m, &tcp_tx_mowner); 4606 4607 /* Fixup the mbuf. */ 4608 m->m_data += max_linkhdr; 4609 m->m_len = m->m_pkthdr.len = tlen; 4610 if (sc->sc_tp) { 4611 tp = sc->sc_tp; 4612 if (tp->t_inpcb) 4613 so = tp->t_inpcb->inp_socket; 4614 #ifdef INET6 4615 else if (tp->t_in6pcb) 4616 so = tp->t_in6pcb->in6p_socket; 4617 #endif 4618 else 4619 so = NULL; 4620 } else 4621 so = NULL; 4622 m->m_pkthdr.rcvif = NULL; 4623 memset(mtod(m, u_char *), 0, tlen); 4624 4625 switch (sc->sc_src.sa.sa_family) { 4626 case AF_INET: 4627 ip = mtod(m, struct ip *); 4628 ip->ip_v = 4; 4629 ip->ip_dst = sc->sc_src.sin.sin_addr; 4630 ip->ip_src = sc->sc_dst.sin.sin_addr; 4631 ip->ip_p = IPPROTO_TCP; 4632 th = (struct tcphdr *)(ip + 1); 4633 th->th_dport = sc->sc_src.sin.sin_port; 4634 th->th_sport = sc->sc_dst.sin.sin_port; 4635 break; 4636 #ifdef INET6 4637 case AF_INET6: 4638 ip6 = mtod(m, struct ip6_hdr *); 4639 ip6->ip6_vfc = IPV6_VERSION; 4640 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4641 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4642 ip6->ip6_nxt = IPPROTO_TCP; 4643 /* ip6_plen will be updated in ip6_output() */ 4644 th = (struct tcphdr *)(ip6 + 1); 4645 th->th_dport = sc->sc_src.sin6.sin6_port; 4646 th->th_sport = sc->sc_dst.sin6.sin6_port; 4647 break; 4648 #endif 4649 default: 4650 th = NULL; 4651 } 4652 4653 th->th_seq = htonl(sc->sc_iss); 4654 th->th_ack = htonl(sc->sc_irs + 1); 4655 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4656 th->th_flags = TH_SYN|TH_ACK; 4657 th->th_win = htons(sc->sc_win); 4658 /* th_sum already 0 */ 4659 /* th_urp already 0 */ 4660 4661 /* Tack on the TCP options. */ 4662 optp = (u_int8_t *)(th + 1); 4663 *optp++ = TCPOPT_MAXSEG; 4664 *optp++ = 4; 4665 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4666 *optp++ = sc->sc_ourmaxseg & 0xff; 4667 4668 if (sc->sc_request_r_scale != 15) { 4669 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4670 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4671 sc->sc_request_r_scale); 4672 optp += 4; 4673 } 4674 4675 if (sc->sc_flags & SCF_TIMESTAMP) { 4676 u_int32_t *lp = (u_int32_t *)(optp); 4677 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4678 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4679 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4680 *lp = htonl(sc->sc_timestamp); 4681 optp += TCPOLEN_TSTAMP_APPA; 4682 } 4683 4684 if (sc->sc_flags & SCF_SACK_PERMIT) { 4685 u_int8_t *p = optp; 4686 4687 /* Let the peer know that we will SACK. */ 4688 p[0] = TCPOPT_SACK_PERMITTED; 4689 p[1] = 2; 4690 p[2] = TCPOPT_NOP; 4691 p[3] = TCPOPT_NOP; 4692 optp += 4; 4693 } 4694 4695 /* 4696 * Send ECN SYN-ACK setup packet. 4697 * Routes can be asymetric, so, even if we receive a packet 4698 * with ECE and CWR set, we must not assume no one will block 4699 * the ECE packet we are about to send. 4700 */ 4701 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4702 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4703 th->th_flags |= TH_ECE; 4704 TCP_STATINC(TCP_STAT_ECN_SHS); 4705 4706 /* 4707 * draft-ietf-tcpm-ecnsyn-00.txt 4708 * 4709 * "[...] a TCP node MAY respond to an ECN-setup 4710 * SYN packet by setting ECT in the responding 4711 * ECN-setup SYN/ACK packet, indicating to routers 4712 * that the SYN/ACK packet is ECN-Capable. 4713 * This allows a congested router along the path 4714 * to mark the packet instead of dropping the 4715 * packet as an indication of congestion." 4716 * 4717 * "[...] There can be a great benefit in setting 4718 * an ECN-capable codepoint in SYN/ACK packets [...] 4719 * Congestion is most likely to occur in 4720 * the server-to-client direction. As a result, 4721 * setting an ECN-capable codepoint in SYN/ACK 4722 * packets can reduce the occurence of three-second 4723 * retransmit timeouts resulting from the drop 4724 * of SYN/ACK packets." 4725 * 4726 * Page 4 and 6, January 2006. 4727 */ 4728 4729 switch (sc->sc_src.sa.sa_family) { 4730 #ifdef INET 4731 case AF_INET: 4732 ip->ip_tos |= IPTOS_ECN_ECT0; 4733 break; 4734 #endif 4735 #ifdef INET6 4736 case AF_INET6: 4737 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4738 break; 4739 #endif 4740 } 4741 TCP_STATINC(TCP_STAT_ECN_ECT); 4742 } 4743 4744 #ifdef TCP_SIGNATURE 4745 if (sc->sc_flags & SCF_SIGNATURE) { 4746 struct secasvar *sav; 4747 u_int8_t *sigp; 4748 4749 sav = tcp_signature_getsav(m, th); 4750 4751 if (sav == NULL) { 4752 if (m) 4753 m_freem(m); 4754 return (EPERM); 4755 } 4756 4757 *optp++ = TCPOPT_SIGNATURE; 4758 *optp++ = TCPOLEN_SIGNATURE; 4759 sigp = optp; 4760 memset(optp, 0, TCP_SIGLEN); 4761 optp += TCP_SIGLEN; 4762 *optp++ = TCPOPT_NOP; 4763 *optp++ = TCPOPT_EOL; 4764 4765 (void)tcp_signature(m, th, hlen, sav, sigp); 4766 4767 key_sa_recordxfer(sav, m); 4768 #ifdef FAST_IPSEC 4769 KEY_FREESAV(&sav); 4770 #else 4771 key_freesav(sav); 4772 #endif 4773 } 4774 #endif 4775 4776 /* Compute the packet's checksum. */ 4777 switch (sc->sc_src.sa.sa_family) { 4778 case AF_INET: 4779 ip->ip_len = htons(tlen - hlen); 4780 th->th_sum = 0; 4781 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4782 break; 4783 #ifdef INET6 4784 case AF_INET6: 4785 ip6->ip6_plen = htons(tlen - hlen); 4786 th->th_sum = 0; 4787 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4788 break; 4789 #endif 4790 } 4791 4792 /* 4793 * Fill in some straggling IP bits. Note the stack expects 4794 * ip_len to be in host order, for convenience. 4795 */ 4796 switch (sc->sc_src.sa.sa_family) { 4797 #ifdef INET 4798 case AF_INET: 4799 ip->ip_len = htons(tlen); 4800 ip->ip_ttl = ip_defttl; 4801 /* XXX tos? */ 4802 break; 4803 #endif 4804 #ifdef INET6 4805 case AF_INET6: 4806 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4807 ip6->ip6_vfc |= IPV6_VERSION; 4808 ip6->ip6_plen = htons(tlen - hlen); 4809 /* ip6_hlim will be initialized afterwards */ 4810 /* XXX flowlabel? */ 4811 break; 4812 #endif 4813 } 4814 4815 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4816 tp = sc->sc_tp; 4817 4818 switch (sc->sc_src.sa.sa_family) { 4819 #ifdef INET 4820 case AF_INET: 4821 error = ip_output(m, sc->sc_ipopts, ro, 4822 (ip_mtudisc ? IP_MTUDISC : 0), 4823 (struct ip_moptions *)NULL, so); 4824 break; 4825 #endif 4826 #ifdef INET6 4827 case AF_INET6: 4828 ip6->ip6_hlim = in6_selecthlim(NULL, 4829 (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp 4830 : NULL); 4831 4832 error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, so, NULL); 4833 break; 4834 #endif 4835 default: 4836 error = EAFNOSUPPORT; 4837 break; 4838 } 4839 return (error); 4840 } 4841