1 /* $NetBSD: tcp_input.c,v 1.337 2015/03/14 02:08:16 rtr 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 1.337 2015/03/14 02:08:16 rtr Exp $"); 152 153 #include "opt_inet.h" 154 #include "opt_ipsec.h" 155 #include "opt_inet_csum.h" 156 #include "opt_tcp_debug.h" 157 158 #include <sys/param.h> 159 #include <sys/systm.h> 160 #include <sys/malloc.h> 161 #include <sys/mbuf.h> 162 #include <sys/protosw.h> 163 #include <sys/socket.h> 164 #include <sys/socketvar.h> 165 #include <sys/errno.h> 166 #include <sys/syslog.h> 167 #include <sys/pool.h> 168 #include <sys/domain.h> 169 #include <sys/kernel.h> 170 #ifdef TCP_SIGNATURE 171 #include <sys/md5.h> 172 #endif 173 #include <sys/lwp.h> /* for lwp0 */ 174 #include <sys/cprng.h> 175 176 #include <net/if.h> 177 #include <net/route.h> 178 #include <net/if_types.h> 179 180 #include <netinet/in.h> 181 #include <netinet/in_systm.h> 182 #include <netinet/ip.h> 183 #include <netinet/in_pcb.h> 184 #include <netinet/in_var.h> 185 #include <netinet/ip_var.h> 186 #include <netinet/in_offload.h> 187 188 #ifdef INET6 189 #ifndef INET 190 #include <netinet/in.h> 191 #endif 192 #include <netinet/ip6.h> 193 #include <netinet6/ip6_var.h> 194 #include <netinet6/in6_pcb.h> 195 #include <netinet6/ip6_var.h> 196 #include <netinet6/in6_var.h> 197 #include <netinet/icmp6.h> 198 #include <netinet6/nd6.h> 199 #ifdef TCP_SIGNATURE 200 #include <netinet6/scope6_var.h> 201 #endif 202 #endif 203 204 #ifndef INET6 205 /* always need ip6.h for IP6_EXTHDR_GET */ 206 #include <netinet/ip6.h> 207 #endif 208 209 #include <netinet/tcp.h> 210 #include <netinet/tcp_fsm.h> 211 #include <netinet/tcp_seq.h> 212 #include <netinet/tcp_timer.h> 213 #include <netinet/tcp_var.h> 214 #include <netinet/tcp_private.h> 215 #include <netinet/tcpip.h> 216 #include <netinet/tcp_congctl.h> 217 #include <netinet/tcp_debug.h> 218 219 #ifdef INET6 220 #include "faith.h" 221 #if defined(NFAITH) && NFAITH > 0 222 #include <net/if_faith.h> 223 #endif 224 #endif /* INET6 */ 225 226 #ifdef IPSEC 227 #include <netipsec/ipsec.h> 228 #include <netipsec/ipsec_var.h> 229 #include <netipsec/ipsec_private.h> 230 #include <netipsec/key.h> 231 #ifdef INET6 232 #include <netipsec/ipsec6.h> 233 #endif 234 #endif /* IPSEC*/ 235 236 #include <netinet/tcp_vtw.h> 237 238 int tcprexmtthresh = 3; 239 int tcp_log_refused; 240 241 int tcp_do_autorcvbuf = 1; 242 int tcp_autorcvbuf_inc = 16 * 1024; 243 int tcp_autorcvbuf_max = 256 * 1024; 244 int tcp_msl = (TCPTV_MSL / PR_SLOWHZ); 245 246 static int tcp_rst_ppslim_count = 0; 247 static struct timeval tcp_rst_ppslim_last; 248 static int tcp_ackdrop_ppslim_count = 0; 249 static struct timeval tcp_ackdrop_ppslim_last; 250 251 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 252 253 /* for modulo comparisons of timestamps */ 254 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 255 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 256 257 /* 258 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 259 */ 260 #ifdef INET6 261 static inline void 262 nd6_hint(struct tcpcb *tp) 263 { 264 struct rtentry *rt; 265 266 if (tp != NULL && tp->t_in6pcb != NULL && tp->t_family == AF_INET6 && 267 (rt = rtcache_validate(&tp->t_in6pcb->in6p_route)) != NULL) 268 nd6_nud_hint(rt, NULL, 0); 269 } 270 #else 271 static inline void 272 nd6_hint(struct tcpcb *tp) 273 { 274 } 275 #endif 276 277 /* 278 * Compute ACK transmission behavior. Delay the ACK unless 279 * we have already delayed an ACK (must send an ACK every two segments). 280 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 281 * option is enabled. 282 */ 283 static void 284 tcp_setup_ack(struct tcpcb *tp, const struct tcphdr *th) 285 { 286 287 if (tp->t_flags & TF_DELACK || 288 (tcp_ack_on_push && th->th_flags & TH_PUSH)) 289 tp->t_flags |= TF_ACKNOW; 290 else 291 TCP_SET_DELACK(tp); 292 } 293 294 static void 295 icmp_check(struct tcpcb *tp, const struct tcphdr *th, int acked) 296 { 297 298 /* 299 * If we had a pending ICMP message that refers to data that have 300 * just been acknowledged, disregard the recorded ICMP message. 301 */ 302 if ((tp->t_flags & TF_PMTUD_PEND) && 303 SEQ_GT(th->th_ack, tp->t_pmtud_th_seq)) 304 tp->t_flags &= ~TF_PMTUD_PEND; 305 306 /* 307 * Keep track of the largest chunk of data 308 * acknowledged since last PMTU update 309 */ 310 if (tp->t_pmtud_mss_acked < acked) 311 tp->t_pmtud_mss_acked = acked; 312 } 313 314 /* 315 * Convert TCP protocol fields to host order for easier processing. 316 */ 317 static void 318 tcp_fields_to_host(struct tcphdr *th) 319 { 320 321 NTOHL(th->th_seq); 322 NTOHL(th->th_ack); 323 NTOHS(th->th_win); 324 NTOHS(th->th_urp); 325 } 326 327 /* 328 * ... and reverse the above. 329 */ 330 static void 331 tcp_fields_to_net(struct tcphdr *th) 332 { 333 334 HTONL(th->th_seq); 335 HTONL(th->th_ack); 336 HTONS(th->th_win); 337 HTONS(th->th_urp); 338 } 339 340 #ifdef TCP_CSUM_COUNTERS 341 #include <sys/device.h> 342 343 #if defined(INET) 344 extern struct evcnt tcp_hwcsum_ok; 345 extern struct evcnt tcp_hwcsum_bad; 346 extern struct evcnt tcp_hwcsum_data; 347 extern struct evcnt tcp_swcsum; 348 #endif /* defined(INET) */ 349 #if defined(INET6) 350 extern struct evcnt tcp6_hwcsum_ok; 351 extern struct evcnt tcp6_hwcsum_bad; 352 extern struct evcnt tcp6_hwcsum_data; 353 extern struct evcnt tcp6_swcsum; 354 #endif /* defined(INET6) */ 355 356 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 357 358 #else 359 360 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 361 362 #endif /* TCP_CSUM_COUNTERS */ 363 364 #ifdef TCP_REASS_COUNTERS 365 #include <sys/device.h> 366 367 extern struct evcnt tcp_reass_; 368 extern struct evcnt tcp_reass_empty; 369 extern struct evcnt tcp_reass_iteration[8]; 370 extern struct evcnt tcp_reass_prependfirst; 371 extern struct evcnt tcp_reass_prepend; 372 extern struct evcnt tcp_reass_insert; 373 extern struct evcnt tcp_reass_inserttail; 374 extern struct evcnt tcp_reass_append; 375 extern struct evcnt tcp_reass_appendtail; 376 extern struct evcnt tcp_reass_overlaptail; 377 extern struct evcnt tcp_reass_overlapfront; 378 extern struct evcnt tcp_reass_segdup; 379 extern struct evcnt tcp_reass_fragdup; 380 381 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 382 383 #else 384 385 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 386 387 #endif /* TCP_REASS_COUNTERS */ 388 389 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *, 390 int *); 391 static int tcp_dooptions(struct tcpcb *, const u_char *, int, 392 struct tcphdr *, struct mbuf *, int, struct tcp_opt_info *); 393 394 #ifdef INET 395 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 396 #endif 397 #ifdef INET6 398 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 399 #endif 400 401 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 402 403 #if defined(MBUFTRACE) 404 struct mowner tcp_reass_mowner = MOWNER_INIT("tcp", "reass"); 405 #endif /* defined(MBUFTRACE) */ 406 407 static struct pool tcpipqent_pool; 408 409 void 410 tcpipqent_init(void) 411 { 412 413 pool_init(&tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", 414 NULL, IPL_VM); 415 } 416 417 struct ipqent * 418 tcpipqent_alloc(void) 419 { 420 struct ipqent *ipqe; 421 int s; 422 423 s = splvm(); 424 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 425 splx(s); 426 427 return ipqe; 428 } 429 430 void 431 tcpipqent_free(struct ipqent *ipqe) 432 { 433 int s; 434 435 s = splvm(); 436 pool_put(&tcpipqent_pool, ipqe); 437 splx(s); 438 } 439 440 static int 441 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int *tlen) 442 { 443 struct ipqent *p, *q, *nq, *tiqe = NULL; 444 struct socket *so = NULL; 445 int pkt_flags; 446 tcp_seq pkt_seq; 447 unsigned pkt_len; 448 u_long rcvpartdupbyte = 0; 449 u_long rcvoobyte; 450 #ifdef TCP_REASS_COUNTERS 451 u_int count = 0; 452 #endif 453 uint64_t *tcps; 454 455 if (tp->t_inpcb) 456 so = tp->t_inpcb->inp_socket; 457 #ifdef INET6 458 else if (tp->t_in6pcb) 459 so = tp->t_in6pcb->in6p_socket; 460 #endif 461 462 TCP_REASS_LOCK_CHECK(tp); 463 464 /* 465 * Call with th==0 after become established to 466 * force pre-ESTABLISHED data up to user socket. 467 */ 468 if (th == 0) 469 goto present; 470 471 m_claimm(m, &tcp_reass_mowner); 472 473 rcvoobyte = *tlen; 474 /* 475 * Copy these to local variables because the tcpiphdr 476 * gets munged while we are collapsing mbufs. 477 */ 478 pkt_seq = th->th_seq; 479 pkt_len = *tlen; 480 pkt_flags = th->th_flags; 481 482 TCP_REASS_COUNTER_INCR(&tcp_reass_); 483 484 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 485 /* 486 * When we miss a packet, the vast majority of time we get 487 * packets that follow it in order. So optimize for that. 488 */ 489 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 490 p->ipqe_len += pkt_len; 491 p->ipqe_flags |= pkt_flags; 492 m_cat(p->ipre_mlast, m); 493 TRAVERSE(p->ipre_mlast); 494 m = NULL; 495 tiqe = p; 496 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 497 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 498 goto skip_replacement; 499 } 500 /* 501 * While we're here, if the pkt is completely beyond 502 * anything we have, just insert it at the tail. 503 */ 504 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 505 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 506 goto insert_it; 507 } 508 } 509 510 q = TAILQ_FIRST(&tp->segq); 511 512 if (q != NULL) { 513 /* 514 * If this segment immediately precedes the first out-of-order 515 * block, simply slap the segment in front of it and (mostly) 516 * skip the complicated logic. 517 */ 518 if (pkt_seq + pkt_len == q->ipqe_seq) { 519 q->ipqe_seq = pkt_seq; 520 q->ipqe_len += pkt_len; 521 q->ipqe_flags |= pkt_flags; 522 m_cat(m, q->ipqe_m); 523 q->ipqe_m = m; 524 q->ipre_mlast = m; /* last mbuf may have changed */ 525 TRAVERSE(q->ipre_mlast); 526 tiqe = q; 527 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 528 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 529 goto skip_replacement; 530 } 531 } else { 532 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 533 } 534 535 /* 536 * Find a segment which begins after this one does. 537 */ 538 for (p = NULL; q != NULL; q = nq) { 539 nq = TAILQ_NEXT(q, ipqe_q); 540 #ifdef TCP_REASS_COUNTERS 541 count++; 542 #endif 543 /* 544 * If the received segment is just right after this 545 * fragment, merge the two together and then check 546 * for further overlaps. 547 */ 548 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 549 #ifdef TCPREASS_DEBUG 550 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 551 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 552 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 553 #endif 554 pkt_len += q->ipqe_len; 555 pkt_flags |= q->ipqe_flags; 556 pkt_seq = q->ipqe_seq; 557 m_cat(q->ipre_mlast, m); 558 TRAVERSE(q->ipre_mlast); 559 m = q->ipqe_m; 560 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 561 goto free_ipqe; 562 } 563 /* 564 * If the received segment is completely past this 565 * fragment, we need to go the next fragment. 566 */ 567 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 568 p = q; 569 continue; 570 } 571 /* 572 * If the fragment is past the received segment, 573 * it (or any following) can't be concatenated. 574 */ 575 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 576 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 577 break; 578 } 579 580 /* 581 * We've received all the data in this segment before. 582 * mark it as a duplicate and return. 583 */ 584 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 585 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 586 tcps = TCP_STAT_GETREF(); 587 tcps[TCP_STAT_RCVDUPPACK]++; 588 tcps[TCP_STAT_RCVDUPBYTE] += pkt_len; 589 TCP_STAT_PUTREF(); 590 tcp_new_dsack(tp, pkt_seq, pkt_len); 591 m_freem(m); 592 if (tiqe != NULL) { 593 tcpipqent_free(tiqe); 594 } 595 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 596 goto out; 597 } 598 /* 599 * Received segment completely overlaps this fragment 600 * so we drop the fragment (this keeps the temporal 601 * ordering of segments correct). 602 */ 603 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 604 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 605 rcvpartdupbyte += q->ipqe_len; 606 m_freem(q->ipqe_m); 607 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 608 goto free_ipqe; 609 } 610 /* 611 * RX'ed segment extends past the end of the 612 * fragment. Drop the overlapping bytes. Then 613 * merge the fragment and segment then treat as 614 * a longer received packet. 615 */ 616 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 617 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 618 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 619 #ifdef TCPREASS_DEBUG 620 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 621 tp, overlap, 622 pkt_seq, pkt_seq + pkt_len, pkt_len); 623 #endif 624 m_adj(m, overlap); 625 rcvpartdupbyte += overlap; 626 m_cat(q->ipre_mlast, m); 627 TRAVERSE(q->ipre_mlast); 628 m = q->ipqe_m; 629 pkt_seq = q->ipqe_seq; 630 pkt_len += q->ipqe_len - overlap; 631 rcvoobyte -= overlap; 632 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 633 goto free_ipqe; 634 } 635 /* 636 * RX'ed segment extends past the front of the 637 * fragment. Drop the overlapping bytes on the 638 * received packet. The packet will then be 639 * contatentated with this fragment a bit later. 640 */ 641 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 642 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 643 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 644 #ifdef TCPREASS_DEBUG 645 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 646 tp, overlap, 647 pkt_seq, pkt_seq + pkt_len, pkt_len); 648 #endif 649 m_adj(m, -overlap); 650 pkt_len -= overlap; 651 rcvpartdupbyte += overlap; 652 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 653 rcvoobyte -= overlap; 654 } 655 /* 656 * If the received segment immediates precedes this 657 * fragment then tack the fragment onto this segment 658 * and reinsert the data. 659 */ 660 if (q->ipqe_seq == pkt_seq + pkt_len) { 661 #ifdef TCPREASS_DEBUG 662 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 663 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 664 pkt_seq, pkt_seq + pkt_len, pkt_len); 665 #endif 666 pkt_len += q->ipqe_len; 667 pkt_flags |= q->ipqe_flags; 668 m_cat(m, q->ipqe_m); 669 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 670 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 671 tp->t_segqlen--; 672 KASSERT(tp->t_segqlen >= 0); 673 KASSERT(tp->t_segqlen != 0 || 674 (TAILQ_EMPTY(&tp->segq) && 675 TAILQ_EMPTY(&tp->timeq))); 676 if (tiqe == NULL) { 677 tiqe = q; 678 } else { 679 tcpipqent_free(q); 680 } 681 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 682 break; 683 } 684 /* 685 * If the fragment is before the segment, remember it. 686 * When this loop is terminated, p will contain the 687 * pointer to fragment that is right before the received 688 * segment. 689 */ 690 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 691 p = q; 692 693 continue; 694 695 /* 696 * This is a common operation. It also will allow 697 * to save doing a malloc/free in most instances. 698 */ 699 free_ipqe: 700 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 701 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 702 tp->t_segqlen--; 703 KASSERT(tp->t_segqlen >= 0); 704 KASSERT(tp->t_segqlen != 0 || 705 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 706 if (tiqe == NULL) { 707 tiqe = q; 708 } else { 709 tcpipqent_free(q); 710 } 711 } 712 713 #ifdef TCP_REASS_COUNTERS 714 if (count > 7) 715 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 716 else if (count > 0) 717 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 718 #endif 719 720 insert_it: 721 722 /* 723 * Allocate a new queue entry since the received segment did not 724 * collapse onto any other out-of-order block; thus we are allocating 725 * a new block. If it had collapsed, tiqe would not be NULL and 726 * we would be reusing it. 727 * XXX If we can't, just drop the packet. XXX 728 */ 729 if (tiqe == NULL) { 730 tiqe = tcpipqent_alloc(); 731 if (tiqe == NULL) { 732 TCP_STATINC(TCP_STAT_RCVMEMDROP); 733 m_freem(m); 734 goto out; 735 } 736 } 737 738 /* 739 * Update the counters. 740 */ 741 tp->t_rcvoopack++; 742 tcps = TCP_STAT_GETREF(); 743 tcps[TCP_STAT_RCVOOPACK]++; 744 tcps[TCP_STAT_RCVOOBYTE] += rcvoobyte; 745 if (rcvpartdupbyte) { 746 tcps[TCP_STAT_RCVPARTDUPPACK]++; 747 tcps[TCP_STAT_RCVPARTDUPBYTE] += rcvpartdupbyte; 748 } 749 TCP_STAT_PUTREF(); 750 751 /* 752 * Insert the new fragment queue entry into both queues. 753 */ 754 tiqe->ipqe_m = m; 755 tiqe->ipre_mlast = m; 756 tiqe->ipqe_seq = pkt_seq; 757 tiqe->ipqe_len = pkt_len; 758 tiqe->ipqe_flags = pkt_flags; 759 if (p == NULL) { 760 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 761 #ifdef TCPREASS_DEBUG 762 if (tiqe->ipqe_seq != tp->rcv_nxt) 763 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 764 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 765 #endif 766 } else { 767 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 768 #ifdef TCPREASS_DEBUG 769 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 770 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 771 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 772 #endif 773 } 774 tp->t_segqlen++; 775 776 skip_replacement: 777 778 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 779 780 present: 781 /* 782 * Present data to user, advancing rcv_nxt through 783 * completed sequence space. 784 */ 785 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 786 goto out; 787 q = TAILQ_FIRST(&tp->segq); 788 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 789 goto out; 790 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 791 goto out; 792 793 tp->rcv_nxt += q->ipqe_len; 794 pkt_flags = q->ipqe_flags & TH_FIN; 795 nd6_hint(tp); 796 797 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 798 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 799 tp->t_segqlen--; 800 KASSERT(tp->t_segqlen >= 0); 801 KASSERT(tp->t_segqlen != 0 || 802 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 803 if (so->so_state & SS_CANTRCVMORE) 804 m_freem(q->ipqe_m); 805 else 806 sbappendstream(&so->so_rcv, q->ipqe_m); 807 tcpipqent_free(q); 808 TCP_REASS_UNLOCK(tp); 809 sorwakeup(so); 810 return (pkt_flags); 811 out: 812 TCP_REASS_UNLOCK(tp); 813 return (0); 814 } 815 816 #ifdef INET6 817 int 818 tcp6_input(struct mbuf **mp, int *offp, int proto) 819 { 820 struct mbuf *m = *mp; 821 822 /* 823 * draft-itojun-ipv6-tcp-to-anycast 824 * better place to put this in? 825 */ 826 if (m->m_flags & M_ANYCAST6) { 827 struct ip6_hdr *ip6; 828 if (m->m_len < sizeof(struct ip6_hdr)) { 829 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 830 TCP_STATINC(TCP_STAT_RCVSHORT); 831 return IPPROTO_DONE; 832 } 833 } 834 ip6 = mtod(m, struct ip6_hdr *); 835 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 836 (char *)&ip6->ip6_dst - (char *)ip6); 837 return IPPROTO_DONE; 838 } 839 840 tcp_input(m, *offp, proto); 841 return IPPROTO_DONE; 842 } 843 #endif 844 845 #ifdef INET 846 static void 847 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 848 { 849 char src[INET_ADDRSTRLEN]; 850 char dst[INET_ADDRSTRLEN]; 851 852 if (ip) { 853 in_print(src, sizeof(src), &ip->ip_src); 854 in_print(dst, sizeof(dst), &ip->ip_dst); 855 } 856 else { 857 strlcpy(src, "(unknown)", sizeof(src)); 858 strlcpy(dst, "(unknown)", sizeof(dst)); 859 } 860 log(LOG_INFO, 861 "Connection attempt to TCP %s:%d from %s:%d\n", 862 dst, ntohs(th->th_dport), 863 src, ntohs(th->th_sport)); 864 } 865 #endif 866 867 #ifdef INET6 868 static void 869 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 870 { 871 char src[INET6_ADDRSTRLEN]; 872 char dst[INET6_ADDRSTRLEN]; 873 874 if (ip6) { 875 in6_print(src, sizeof(src), &ip6->ip6_src); 876 in6_print(dst, sizeof(dst), &ip6->ip6_dst); 877 } 878 else { 879 strlcpy(src, "(unknown v6)", sizeof(src)); 880 strlcpy(dst, "(unknown v6)", sizeof(dst)); 881 } 882 log(LOG_INFO, 883 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 884 dst, ntohs(th->th_dport), 885 src, ntohs(th->th_sport)); 886 } 887 #endif 888 889 /* 890 * Checksum extended TCP header and data. 891 */ 892 int 893 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, 894 int toff, int off, int tlen) 895 { 896 897 /* 898 * XXX it's better to record and check if this mbuf is 899 * already checked. 900 */ 901 902 switch (af) { 903 #ifdef INET 904 case AF_INET: 905 switch (m->m_pkthdr.csum_flags & 906 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 907 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 908 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 909 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 910 goto badcsum; 911 912 case M_CSUM_TCPv4|M_CSUM_DATA: { 913 u_int32_t hw_csum = m->m_pkthdr.csum_data; 914 915 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 916 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 917 const struct ip *ip = 918 mtod(m, const struct ip *); 919 920 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 921 ip->ip_dst.s_addr, 922 htons(hw_csum + tlen + off + IPPROTO_TCP)); 923 } 924 if ((hw_csum ^ 0xffff) != 0) 925 goto badcsum; 926 break; 927 } 928 929 case M_CSUM_TCPv4: 930 /* Checksum was okay. */ 931 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 932 break; 933 934 default: 935 /* 936 * Must compute it ourselves. Maybe skip checksum 937 * on loopback interfaces. 938 */ 939 if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & 940 IFF_LOOPBACK) || 941 tcp_do_loopback_cksum)) { 942 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 943 if (in4_cksum(m, IPPROTO_TCP, toff, 944 tlen + off) != 0) 945 goto badcsum; 946 } 947 break; 948 } 949 break; 950 #endif /* INET4 */ 951 952 #ifdef INET6 953 case AF_INET6: 954 switch (m->m_pkthdr.csum_flags & 955 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 956 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 957 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 958 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 959 goto badcsum; 960 961 #if 0 /* notyet */ 962 case M_CSUM_TCPv6|M_CSUM_DATA: 963 #endif 964 965 case M_CSUM_TCPv6: 966 /* Checksum was okay. */ 967 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 968 break; 969 970 default: 971 /* 972 * Must compute it ourselves. Maybe skip checksum 973 * on loopback interfaces. 974 */ 975 if (__predict_true((m->m_flags & M_LOOP) == 0 || 976 tcp_do_loopback_cksum)) { 977 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 978 if (in6_cksum(m, IPPROTO_TCP, toff, 979 tlen + off) != 0) 980 goto badcsum; 981 } 982 } 983 break; 984 #endif /* INET6 */ 985 } 986 987 return 0; 988 989 badcsum: 990 TCP_STATINC(TCP_STAT_RCVBADSUM); 991 return -1; 992 } 993 994 /* When a packet arrives addressed to a vestigial tcpbp, we 995 * nevertheless have to respond to it per the spec. 996 */ 997 static void tcp_vtw_input(struct tcphdr *th, vestigial_inpcb_t *vp, 998 struct mbuf *m, int tlen, int multicast) 999 { 1000 int tiflags; 1001 int todrop; 1002 uint32_t t_flags = 0; 1003 uint64_t *tcps; 1004 1005 tiflags = th->th_flags; 1006 todrop = vp->rcv_nxt - th->th_seq; 1007 1008 if (todrop > 0) { 1009 if (tiflags & TH_SYN) { 1010 tiflags &= ~TH_SYN; 1011 ++th->th_seq; 1012 if (th->th_urp > 1) 1013 --th->th_urp; 1014 else { 1015 tiflags &= ~TH_URG; 1016 th->th_urp = 0; 1017 } 1018 --todrop; 1019 } 1020 if (todrop > tlen || 1021 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1022 /* 1023 * Any valid FIN or RST must be to the left of the 1024 * window. At this point the FIN or RST must be a 1025 * duplicate or out of sequence; drop it. 1026 */ 1027 if (tiflags & TH_RST) 1028 goto drop; 1029 tiflags &= ~(TH_FIN|TH_RST); 1030 /* 1031 * Send an ACK to resynchronize and drop any data. 1032 * But keep on processing for RST or ACK. 1033 */ 1034 t_flags |= TF_ACKNOW; 1035 todrop = tlen; 1036 tcps = TCP_STAT_GETREF(); 1037 tcps[TCP_STAT_RCVDUPPACK] += 1; 1038 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 1039 TCP_STAT_PUTREF(); 1040 } else if ((tiflags & TH_RST) 1041 && th->th_seq != vp->rcv_nxt) { 1042 /* 1043 * Test for reset before adjusting the sequence 1044 * number for overlapping data. 1045 */ 1046 goto dropafterack_ratelim; 1047 } else { 1048 tcps = TCP_STAT_GETREF(); 1049 tcps[TCP_STAT_RCVPARTDUPPACK] += 1; 1050 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 1051 TCP_STAT_PUTREF(); 1052 } 1053 1054 // tcp_new_dsack(tp, th->th_seq, todrop); 1055 // hdroptlen += todrop; /*drop from head afterwards*/ 1056 1057 th->th_seq += todrop; 1058 tlen -= todrop; 1059 1060 if (th->th_urp > todrop) 1061 th->th_urp -= todrop; 1062 else { 1063 tiflags &= ~TH_URG; 1064 th->th_urp = 0; 1065 } 1066 } 1067 1068 /* 1069 * If new data are received on a connection after the 1070 * user processes are gone, then RST the other end. 1071 */ 1072 if (tlen) { 1073 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 1074 goto dropwithreset; 1075 } 1076 1077 /* 1078 * If segment ends after window, drop trailing data 1079 * (and PUSH and FIN); if nothing left, just ACK. 1080 */ 1081 todrop = (th->th_seq + tlen) - (vp->rcv_nxt+vp->rcv_wnd); 1082 1083 if (todrop > 0) { 1084 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 1085 if (todrop >= tlen) { 1086 /* 1087 * The segment actually starts after the window. 1088 * th->th_seq + tlen - vp->rcv_nxt - vp->rcv_wnd >= tlen 1089 * th->th_seq - vp->rcv_nxt - vp->rcv_wnd >= 0 1090 * th->th_seq >= vp->rcv_nxt + vp->rcv_wnd 1091 */ 1092 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 1093 /* 1094 * If a new connection request is received 1095 * while in TIME_WAIT, drop the old connection 1096 * and start over if the sequence numbers 1097 * are above the previous ones. 1098 */ 1099 if ((tiflags & TH_SYN) 1100 && SEQ_GT(th->th_seq, vp->rcv_nxt)) { 1101 /* We only support this in the !NOFDREF case, which 1102 * is to say: not here. 1103 */ 1104 goto dropwithreset; 1105 } 1106 /* 1107 * If window is closed can only take segments at 1108 * window edge, and have to drop data and PUSH from 1109 * incoming segments. Continue processing, but 1110 * remember to ack. Otherwise, drop segment 1111 * and (if not RST) ack. 1112 */ 1113 if (vp->rcv_wnd == 0 && th->th_seq == vp->rcv_nxt) { 1114 t_flags |= TF_ACKNOW; 1115 TCP_STATINC(TCP_STAT_RCVWINPROBE); 1116 } else 1117 goto dropafterack; 1118 } else 1119 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 1120 m_adj(m, -todrop); 1121 tlen -= todrop; 1122 tiflags &= ~(TH_PUSH|TH_FIN); 1123 } 1124 1125 if (tiflags & TH_RST) { 1126 if (th->th_seq != vp->rcv_nxt) 1127 goto dropafterack_ratelim; 1128 1129 vtw_del(vp->ctl, vp->vtw); 1130 goto drop; 1131 } 1132 1133 /* 1134 * If the ACK bit is off we drop the segment and return. 1135 */ 1136 if ((tiflags & TH_ACK) == 0) { 1137 if (t_flags & TF_ACKNOW) 1138 goto dropafterack; 1139 else 1140 goto drop; 1141 } 1142 1143 /* 1144 * In TIME_WAIT state the only thing that should arrive 1145 * is a retransmission of the remote FIN. Acknowledge 1146 * it and restart the finack timer. 1147 */ 1148 vtw_restart(vp); 1149 goto dropafterack; 1150 1151 dropafterack: 1152 /* 1153 * Generate an ACK dropping incoming segment if it occupies 1154 * sequence space, where the ACK reflects our state. 1155 */ 1156 if (tiflags & TH_RST) 1157 goto drop; 1158 goto dropafterack2; 1159 1160 dropafterack_ratelim: 1161 /* 1162 * We may want to rate-limit ACKs against SYN/RST attack. 1163 */ 1164 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 1165 tcp_ackdrop_ppslim) == 0) { 1166 /* XXX stat */ 1167 goto drop; 1168 } 1169 /* ...fall into dropafterack2... */ 1170 1171 dropafterack2: 1172 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, th->th_ack, 1173 TH_ACK); 1174 return; 1175 1176 dropwithreset: 1177 /* 1178 * Generate a RST, dropping incoming segment. 1179 * Make ACK acceptable to originator of segment. 1180 */ 1181 if (tiflags & TH_RST) 1182 goto drop; 1183 1184 if (tiflags & TH_ACK) 1185 tcp_respond(0, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 1186 else { 1187 if (tiflags & TH_SYN) 1188 ++tlen; 1189 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, (tcp_seq)0, 1190 TH_RST|TH_ACK); 1191 } 1192 return; 1193 drop: 1194 m_freem(m); 1195 } 1196 1197 /* 1198 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 1199 */ 1200 void 1201 tcp_input(struct mbuf *m, ...) 1202 { 1203 struct tcphdr *th; 1204 struct ip *ip; 1205 struct inpcb *inp; 1206 #ifdef INET6 1207 struct ip6_hdr *ip6; 1208 struct in6pcb *in6p; 1209 #endif 1210 u_int8_t *optp = NULL; 1211 int optlen = 0; 1212 int len, tlen, toff, hdroptlen = 0; 1213 struct tcpcb *tp = 0; 1214 int tiflags; 1215 struct socket *so = NULL; 1216 int todrop, acked, ourfinisacked, needoutput = 0; 1217 bool dupseg; 1218 #ifdef TCP_DEBUG 1219 short ostate = 0; 1220 #endif 1221 u_long tiwin; 1222 struct tcp_opt_info opti; 1223 int off, iphlen; 1224 va_list ap; 1225 int af; /* af on the wire */ 1226 struct mbuf *tcp_saveti = NULL; 1227 uint32_t ts_rtt; 1228 uint8_t iptos; 1229 uint64_t *tcps; 1230 vestigial_inpcb_t vestige; 1231 1232 vestige.valid = 0; 1233 1234 MCLAIM(m, &tcp_rx_mowner); 1235 va_start(ap, m); 1236 toff = va_arg(ap, int); 1237 (void)va_arg(ap, int); /* ignore value, advance ap */ 1238 va_end(ap); 1239 1240 TCP_STATINC(TCP_STAT_RCVTOTAL); 1241 1242 memset(&opti, 0, sizeof(opti)); 1243 opti.ts_present = 0; 1244 opti.maxseg = 0; 1245 1246 /* 1247 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 1248 * 1249 * TCP is, by definition, unicast, so we reject all 1250 * multicast outright. 1251 * 1252 * Note, there are additional src/dst address checks in 1253 * the AF-specific code below. 1254 */ 1255 if (m->m_flags & (M_BCAST|M_MCAST)) { 1256 /* XXX stat */ 1257 goto drop; 1258 } 1259 #ifdef INET6 1260 if (m->m_flags & M_ANYCAST6) { 1261 /* XXX stat */ 1262 goto drop; 1263 } 1264 #endif 1265 1266 /* 1267 * Get IP and TCP header. 1268 * Note: IP leaves IP header in first mbuf. 1269 */ 1270 ip = mtod(m, struct ip *); 1271 switch (ip->ip_v) { 1272 #ifdef INET 1273 case 4: 1274 #ifdef INET6 1275 ip6 = NULL; 1276 #endif 1277 af = AF_INET; 1278 iphlen = sizeof(struct ip); 1279 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1280 sizeof(struct tcphdr)); 1281 if (th == NULL) { 1282 TCP_STATINC(TCP_STAT_RCVSHORT); 1283 return; 1284 } 1285 /* We do the checksum after PCB lookup... */ 1286 len = ntohs(ip->ip_len); 1287 tlen = len - toff; 1288 iptos = ip->ip_tos; 1289 break; 1290 #endif 1291 #ifdef INET6 1292 case 6: 1293 ip = NULL; 1294 iphlen = sizeof(struct ip6_hdr); 1295 af = AF_INET6; 1296 ip6 = mtod(m, struct ip6_hdr *); 1297 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1298 sizeof(struct tcphdr)); 1299 if (th == NULL) { 1300 TCP_STATINC(TCP_STAT_RCVSHORT); 1301 return; 1302 } 1303 1304 /* Be proactive about malicious use of IPv4 mapped address */ 1305 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 1306 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 1307 /* XXX stat */ 1308 goto drop; 1309 } 1310 1311 /* 1312 * Be proactive about unspecified IPv6 address in source. 1313 * As we use all-zero to indicate unbounded/unconnected pcb, 1314 * unspecified IPv6 address can be used to confuse us. 1315 * 1316 * Note that packets with unspecified IPv6 destination is 1317 * already dropped in ip6_input. 1318 */ 1319 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1320 /* XXX stat */ 1321 goto drop; 1322 } 1323 1324 /* 1325 * Make sure destination address is not multicast. 1326 * Source address checked in ip6_input(). 1327 */ 1328 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1329 /* XXX stat */ 1330 goto drop; 1331 } 1332 1333 /* We do the checksum after PCB lookup... */ 1334 len = m->m_pkthdr.len; 1335 tlen = len - toff; 1336 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1337 break; 1338 #endif 1339 default: 1340 m_freem(m); 1341 return; 1342 } 1343 1344 KASSERT(TCP_HDR_ALIGNED_P(th)); 1345 1346 /* 1347 * Check that TCP offset makes sense, 1348 * pull out TCP options and adjust length. XXX 1349 */ 1350 off = th->th_off << 2; 1351 if (off < sizeof (struct tcphdr) || off > tlen) { 1352 TCP_STATINC(TCP_STAT_RCVBADOFF); 1353 goto drop; 1354 } 1355 tlen -= off; 1356 1357 /* 1358 * tcp_input() has been modified to use tlen to mean the TCP data 1359 * length throughout the function. Other functions can use 1360 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1361 * rja 1362 */ 1363 1364 if (off > sizeof (struct tcphdr)) { 1365 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1366 if (th == NULL) { 1367 TCP_STATINC(TCP_STAT_RCVSHORT); 1368 return; 1369 } 1370 /* 1371 * NOTE: ip/ip6 will not be affected by m_pulldown() 1372 * (as they're before toff) and we don't need to update those. 1373 */ 1374 KASSERT(TCP_HDR_ALIGNED_P(th)); 1375 optlen = off - sizeof (struct tcphdr); 1376 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1377 /* 1378 * Do quick retrieval of timestamp options ("options 1379 * prediction?"). If timestamp is the only option and it's 1380 * formatted as recommended in RFC 1323 appendix A, we 1381 * quickly get the values now and not bother calling 1382 * tcp_dooptions(), etc. 1383 */ 1384 if ((optlen == TCPOLEN_TSTAMP_APPA || 1385 (optlen > TCPOLEN_TSTAMP_APPA && 1386 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1387 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1388 (th->th_flags & TH_SYN) == 0) { 1389 opti.ts_present = 1; 1390 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1391 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1392 optp = NULL; /* we've parsed the options */ 1393 } 1394 } 1395 tiflags = th->th_flags; 1396 1397 /* 1398 * Locate pcb for segment. 1399 */ 1400 findpcb: 1401 inp = NULL; 1402 #ifdef INET6 1403 in6p = NULL; 1404 #endif 1405 switch (af) { 1406 #ifdef INET 1407 case AF_INET: 1408 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1409 ip->ip_dst, th->th_dport, 1410 &vestige); 1411 if (inp == 0 && !vestige.valid) { 1412 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1413 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1414 } 1415 #ifdef INET6 1416 if (inp == 0 && !vestige.valid) { 1417 struct in6_addr s, d; 1418 1419 /* mapped addr case */ 1420 memset(&s, 0, sizeof(s)); 1421 s.s6_addr16[5] = htons(0xffff); 1422 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1423 memset(&d, 0, sizeof(d)); 1424 d.s6_addr16[5] = htons(0xffff); 1425 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1426 in6p = in6_pcblookup_connect(&tcbtable, &s, 1427 th->th_sport, &d, th->th_dport, 1428 0, &vestige); 1429 if (in6p == 0 && !vestige.valid) { 1430 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1431 in6p = in6_pcblookup_bind(&tcbtable, &d, 1432 th->th_dport, 0); 1433 } 1434 } 1435 #endif 1436 #ifndef INET6 1437 if (inp == 0 && !vestige.valid) 1438 #else 1439 if (inp == 0 && in6p == 0 && !vestige.valid) 1440 #endif 1441 { 1442 TCP_STATINC(TCP_STAT_NOPORT); 1443 if (tcp_log_refused && 1444 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1445 tcp4_log_refused(ip, th); 1446 } 1447 tcp_fields_to_host(th); 1448 goto dropwithreset_ratelim; 1449 } 1450 #if defined(IPSEC) 1451 if (ipsec_used) { 1452 if (inp && 1453 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 1454 && ipsec4_in_reject(m, inp)) { 1455 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1456 goto drop; 1457 } 1458 #ifdef INET6 1459 else if (in6p && 1460 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 1461 && ipsec6_in_reject_so(m, in6p->in6p_socket)) { 1462 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1463 goto drop; 1464 } 1465 #endif 1466 } 1467 #endif /*IPSEC*/ 1468 break; 1469 #endif /*INET*/ 1470 #ifdef INET6 1471 case AF_INET6: 1472 { 1473 int faith; 1474 1475 #if defined(NFAITH) && NFAITH > 0 1476 faith = faithprefix(&ip6->ip6_dst); 1477 #else 1478 faith = 0; 1479 #endif 1480 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1481 th->th_sport, &ip6->ip6_dst, th->th_dport, faith, &vestige); 1482 if (!in6p && !vestige.valid) { 1483 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1484 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1485 th->th_dport, faith); 1486 } 1487 if (!in6p && !vestige.valid) { 1488 TCP_STATINC(TCP_STAT_NOPORT); 1489 if (tcp_log_refused && 1490 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1491 tcp6_log_refused(ip6, th); 1492 } 1493 tcp_fields_to_host(th); 1494 goto dropwithreset_ratelim; 1495 } 1496 #if defined(IPSEC) 1497 if (ipsec_used && in6p 1498 && (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 1499 && ipsec6_in_reject(m, in6p)) { 1500 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1501 goto drop; 1502 } 1503 #endif /*IPSEC*/ 1504 break; 1505 } 1506 #endif 1507 } 1508 1509 /* 1510 * If the state is CLOSED (i.e., TCB does not exist) then 1511 * all data in the incoming segment is discarded. 1512 * If the TCB exists but is in CLOSED state, it is embryonic, 1513 * but should either do a listen or a connect soon. 1514 */ 1515 tp = NULL; 1516 so = NULL; 1517 if (inp) { 1518 /* Check the minimum TTL for socket. */ 1519 if (ip->ip_ttl < inp->inp_ip_minttl) 1520 goto drop; 1521 1522 tp = intotcpcb(inp); 1523 so = inp->inp_socket; 1524 } 1525 #ifdef INET6 1526 else if (in6p) { 1527 tp = in6totcpcb(in6p); 1528 so = in6p->in6p_socket; 1529 } 1530 #endif 1531 else if (vestige.valid) { 1532 int mc = 0; 1533 1534 /* We do not support the resurrection of vtw tcpcps. 1535 */ 1536 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1537 goto badcsum; 1538 1539 switch (af) { 1540 #ifdef INET6 1541 case AF_INET6: 1542 mc = IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst); 1543 break; 1544 #endif 1545 1546 case AF_INET: 1547 mc = (IN_MULTICAST(ip->ip_dst.s_addr) 1548 || in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)); 1549 break; 1550 } 1551 1552 tcp_fields_to_host(th); 1553 tcp_vtw_input(th, &vestige, m, tlen, mc); 1554 m = 0; 1555 goto drop; 1556 } 1557 1558 if (tp == 0) { 1559 tcp_fields_to_host(th); 1560 goto dropwithreset_ratelim; 1561 } 1562 if (tp->t_state == TCPS_CLOSED) 1563 goto drop; 1564 1565 KASSERT(so->so_lock == softnet_lock); 1566 KASSERT(solocked(so)); 1567 1568 /* 1569 * Checksum extended TCP header and data. 1570 */ 1571 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1572 goto badcsum; 1573 1574 tcp_fields_to_host(th); 1575 1576 /* Unscale the window into a 32-bit value. */ 1577 if ((tiflags & TH_SYN) == 0) 1578 tiwin = th->th_win << tp->snd_scale; 1579 else 1580 tiwin = th->th_win; 1581 1582 #ifdef INET6 1583 /* save packet options if user wanted */ 1584 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1585 if (in6p->in6p_options) { 1586 m_freem(in6p->in6p_options); 1587 in6p->in6p_options = 0; 1588 } 1589 KASSERT(ip6 != NULL); 1590 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1591 } 1592 #endif 1593 1594 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1595 union syn_cache_sa src; 1596 union syn_cache_sa dst; 1597 1598 memset(&src, 0, sizeof(src)); 1599 memset(&dst, 0, sizeof(dst)); 1600 switch (af) { 1601 #ifdef INET 1602 case AF_INET: 1603 src.sin.sin_len = sizeof(struct sockaddr_in); 1604 src.sin.sin_family = AF_INET; 1605 src.sin.sin_addr = ip->ip_src; 1606 src.sin.sin_port = th->th_sport; 1607 1608 dst.sin.sin_len = sizeof(struct sockaddr_in); 1609 dst.sin.sin_family = AF_INET; 1610 dst.sin.sin_addr = ip->ip_dst; 1611 dst.sin.sin_port = th->th_dport; 1612 break; 1613 #endif 1614 #ifdef INET6 1615 case AF_INET6: 1616 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1617 src.sin6.sin6_family = AF_INET6; 1618 src.sin6.sin6_addr = ip6->ip6_src; 1619 src.sin6.sin6_port = th->th_sport; 1620 1621 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1622 dst.sin6.sin6_family = AF_INET6; 1623 dst.sin6.sin6_addr = ip6->ip6_dst; 1624 dst.sin6.sin6_port = th->th_dport; 1625 break; 1626 #endif /* INET6 */ 1627 default: 1628 goto badsyn; /*sanity*/ 1629 } 1630 1631 if (so->so_options & SO_DEBUG) { 1632 #ifdef TCP_DEBUG 1633 ostate = tp->t_state; 1634 #endif 1635 1636 tcp_saveti = NULL; 1637 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1638 goto nosave; 1639 1640 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1641 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1642 if (!tcp_saveti) 1643 goto nosave; 1644 } else { 1645 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1646 if (!tcp_saveti) 1647 goto nosave; 1648 MCLAIM(m, &tcp_mowner); 1649 tcp_saveti->m_len = iphlen; 1650 m_copydata(m, 0, iphlen, 1651 mtod(tcp_saveti, void *)); 1652 } 1653 1654 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1655 m_freem(tcp_saveti); 1656 tcp_saveti = NULL; 1657 } else { 1658 tcp_saveti->m_len += sizeof(struct tcphdr); 1659 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1660 sizeof(struct tcphdr)); 1661 } 1662 nosave:; 1663 } 1664 if (so->so_options & SO_ACCEPTCONN) { 1665 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1666 if (tiflags & TH_RST) { 1667 syn_cache_reset(&src.sa, &dst.sa, th); 1668 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1669 (TH_ACK|TH_SYN)) { 1670 /* 1671 * Received a SYN,ACK. This should 1672 * never happen while we are in 1673 * LISTEN. Send an RST. 1674 */ 1675 goto badsyn; 1676 } else if (tiflags & TH_ACK) { 1677 so = syn_cache_get(&src.sa, &dst.sa, 1678 th, toff, tlen, so, m); 1679 if (so == NULL) { 1680 /* 1681 * We don't have a SYN for 1682 * this ACK; send an RST. 1683 */ 1684 goto badsyn; 1685 } else if (so == 1686 (struct socket *)(-1)) { 1687 /* 1688 * We were unable to create 1689 * the connection. If the 1690 * 3-way handshake was 1691 * completed, and RST has 1692 * been sent to the peer. 1693 * Since the mbuf might be 1694 * in use for the reply, 1695 * do not free it. 1696 */ 1697 m = NULL; 1698 } else { 1699 /* 1700 * We have created a 1701 * full-blown connection. 1702 */ 1703 tp = NULL; 1704 inp = NULL; 1705 #ifdef INET6 1706 in6p = NULL; 1707 #endif 1708 switch (so->so_proto->pr_domain->dom_family) { 1709 #ifdef INET 1710 case AF_INET: 1711 inp = sotoinpcb(so); 1712 tp = intotcpcb(inp); 1713 break; 1714 #endif 1715 #ifdef INET6 1716 case AF_INET6: 1717 in6p = sotoin6pcb(so); 1718 tp = in6totcpcb(in6p); 1719 break; 1720 #endif 1721 } 1722 if (tp == NULL) 1723 goto badsyn; /*XXX*/ 1724 tiwin <<= tp->snd_scale; 1725 goto after_listen; 1726 } 1727 } else { 1728 /* 1729 * None of RST, SYN or ACK was set. 1730 * This is an invalid packet for a 1731 * TCB in LISTEN state. Send a RST. 1732 */ 1733 goto badsyn; 1734 } 1735 } else { 1736 /* 1737 * Received a SYN. 1738 * 1739 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1740 */ 1741 if (m->m_flags & (M_BCAST|M_MCAST)) 1742 goto drop; 1743 1744 switch (af) { 1745 #ifdef INET6 1746 case AF_INET6: 1747 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1748 goto drop; 1749 break; 1750 #endif /* INET6 */ 1751 case AF_INET: 1752 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1753 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1754 goto drop; 1755 break; 1756 } 1757 1758 #ifdef INET6 1759 /* 1760 * If deprecated address is forbidden, we do 1761 * not accept SYN to deprecated interface 1762 * address to prevent any new inbound 1763 * connection from getting established. 1764 * When we do not accept SYN, we send a TCP 1765 * RST, with deprecated source address (instead 1766 * of dropping it). We compromise it as it is 1767 * much better for peer to send a RST, and 1768 * RST will be the final packet for the 1769 * exchange. 1770 * 1771 * If we do not forbid deprecated addresses, we 1772 * accept the SYN packet. RFC2462 does not 1773 * suggest dropping SYN in this case. 1774 * If we decipher RFC2462 5.5.4, it says like 1775 * this: 1776 * 1. use of deprecated addr with existing 1777 * communication is okay - "SHOULD continue 1778 * to be used" 1779 * 2. use of it with new communication: 1780 * (2a) "SHOULD NOT be used if alternate 1781 * address with sufficient scope is 1782 * available" 1783 * (2b) nothing mentioned otherwise. 1784 * Here we fall into (2b) case as we have no 1785 * choice in our source address selection - we 1786 * must obey the peer. 1787 * 1788 * The wording in RFC2462 is confusing, and 1789 * there are multiple description text for 1790 * deprecated address handling - worse, they 1791 * are not exactly the same. I believe 5.5.4 1792 * is the best one, so we follow 5.5.4. 1793 */ 1794 if (af == AF_INET6 && !ip6_use_deprecated) { 1795 struct in6_ifaddr *ia6; 1796 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1797 &ip6->ip6_dst)) && 1798 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1799 tp = NULL; 1800 goto dropwithreset; 1801 } 1802 } 1803 #endif 1804 1805 #if defined(IPSEC) 1806 if (ipsec_used) { 1807 switch (af) { 1808 #ifdef INET 1809 case AF_INET: 1810 if (!ipsec4_in_reject_so(m, so)) 1811 break; 1812 IPSEC_STATINC( 1813 IPSEC_STAT_IN_POLVIO); 1814 tp = NULL; 1815 goto dropwithreset; 1816 #endif 1817 #ifdef INET6 1818 case AF_INET6: 1819 if (!ipsec6_in_reject_so(m, so)) 1820 break; 1821 IPSEC6_STATINC( 1822 IPSEC_STAT_IN_POLVIO); 1823 tp = NULL; 1824 goto dropwithreset; 1825 #endif /*INET6*/ 1826 } 1827 } 1828 #endif /*IPSEC*/ 1829 1830 /* 1831 * LISTEN socket received a SYN 1832 * from itself? This can't possibly 1833 * be valid; drop the packet. 1834 */ 1835 if (th->th_sport == th->th_dport) { 1836 int i; 1837 1838 switch (af) { 1839 #ifdef INET 1840 case AF_INET: 1841 i = in_hosteq(ip->ip_src, ip->ip_dst); 1842 break; 1843 #endif 1844 #ifdef INET6 1845 case AF_INET6: 1846 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1847 break; 1848 #endif 1849 default: 1850 i = 1; 1851 } 1852 if (i) { 1853 TCP_STATINC(TCP_STAT_BADSYN); 1854 goto drop; 1855 } 1856 } 1857 1858 /* 1859 * SYN looks ok; create compressed TCP 1860 * state for it. 1861 */ 1862 if (so->so_qlen <= so->so_qlimit && 1863 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1864 so, m, optp, optlen, &opti)) 1865 m = NULL; 1866 } 1867 goto drop; 1868 } 1869 } 1870 1871 after_listen: 1872 #ifdef DIAGNOSTIC 1873 /* 1874 * Should not happen now that all embryonic connections 1875 * are handled with compressed state. 1876 */ 1877 if (tp->t_state == TCPS_LISTEN) 1878 panic("tcp_input: TCPS_LISTEN"); 1879 #endif 1880 1881 /* 1882 * Segment received on connection. 1883 * Reset idle time and keep-alive timer. 1884 */ 1885 tp->t_rcvtime = tcp_now; 1886 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1887 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1888 1889 /* 1890 * Process options. 1891 */ 1892 #ifdef TCP_SIGNATURE 1893 if (optp || (tp->t_flags & TF_SIGNATURE)) 1894 #else 1895 if (optp) 1896 #endif 1897 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1898 goto drop; 1899 1900 if (TCP_SACK_ENABLED(tp)) { 1901 tcp_del_sackholes(tp, th); 1902 } 1903 1904 if (TCP_ECN_ALLOWED(tp)) { 1905 if (tiflags & TH_CWR) { 1906 tp->t_flags &= ~TF_ECN_SND_ECE; 1907 } 1908 switch (iptos & IPTOS_ECN_MASK) { 1909 case IPTOS_ECN_CE: 1910 tp->t_flags |= TF_ECN_SND_ECE; 1911 TCP_STATINC(TCP_STAT_ECN_CE); 1912 break; 1913 case IPTOS_ECN_ECT0: 1914 TCP_STATINC(TCP_STAT_ECN_ECT); 1915 break; 1916 case IPTOS_ECN_ECT1: 1917 /* XXX: ignore for now -- rpaulo */ 1918 break; 1919 } 1920 /* 1921 * Congestion experienced. 1922 * Ignore if we are already trying to recover. 1923 */ 1924 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1925 tp->t_congctl->cong_exp(tp); 1926 } 1927 1928 if (opti.ts_present && opti.ts_ecr) { 1929 /* 1930 * Calculate the RTT from the returned time stamp and the 1931 * connection's time base. If the time stamp is later than 1932 * the current time, or is extremely old, fall back to non-1323 1933 * RTT calculation. Since ts_rtt is unsigned, we can test both 1934 * at the same time. 1935 * 1936 * Note that ts_rtt is in units of slow ticks (500 1937 * ms). Since most earthbound RTTs are < 500 ms, 1938 * observed values will have large quantization noise. 1939 * Our smoothed RTT is then the fraction of observed 1940 * samples that are 1 tick instead of 0 (times 500 1941 * ms). 1942 * 1943 * ts_rtt is increased by 1 to denote a valid sample, 1944 * with 0 indicating an invalid measurement. This 1945 * extra 1 must be removed when ts_rtt is used, or 1946 * else an an erroneous extra 500 ms will result. 1947 */ 1948 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1949 if (ts_rtt > TCP_PAWS_IDLE) 1950 ts_rtt = 0; 1951 } else { 1952 ts_rtt = 0; 1953 } 1954 1955 /* 1956 * Header prediction: check for the two common cases 1957 * of a uni-directional data xfer. If the packet has 1958 * no control flags, is in-sequence, the window didn't 1959 * change and we're not retransmitting, it's a 1960 * candidate. If the length is zero and the ack moved 1961 * forward, we're the sender side of the xfer. Just 1962 * free the data acked & wake any higher level process 1963 * that was blocked waiting for space. If the length 1964 * is non-zero and the ack didn't move, we're the 1965 * receiver side. If we're getting packets in-order 1966 * (the reassembly queue is empty), add the data to 1967 * the socket buffer and note that we need a delayed ack. 1968 */ 1969 if (tp->t_state == TCPS_ESTABLISHED && 1970 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1971 == TH_ACK && 1972 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1973 th->th_seq == tp->rcv_nxt && 1974 tiwin && tiwin == tp->snd_wnd && 1975 tp->snd_nxt == tp->snd_max) { 1976 1977 /* 1978 * If last ACK falls within this segment's sequence numbers, 1979 * record the timestamp. 1980 * NOTE that the test is modified according to the latest 1981 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1982 * 1983 * note that we already know 1984 * TSTMP_GEQ(opti.ts_val, tp->ts_recent) 1985 */ 1986 if (opti.ts_present && 1987 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1988 tp->ts_recent_age = tcp_now; 1989 tp->ts_recent = opti.ts_val; 1990 } 1991 1992 if (tlen == 0) { 1993 /* Ack prediction. */ 1994 if (SEQ_GT(th->th_ack, tp->snd_una) && 1995 SEQ_LEQ(th->th_ack, tp->snd_max) && 1996 tp->snd_cwnd >= tp->snd_wnd && 1997 tp->t_partialacks < 0) { 1998 /* 1999 * this is a pure ack for outstanding data. 2000 */ 2001 if (ts_rtt) 2002 tcp_xmit_timer(tp, ts_rtt - 1); 2003 else if (tp->t_rtttime && 2004 SEQ_GT(th->th_ack, tp->t_rtseq)) 2005 tcp_xmit_timer(tp, 2006 tcp_now - tp->t_rtttime); 2007 acked = th->th_ack - tp->snd_una; 2008 tcps = TCP_STAT_GETREF(); 2009 tcps[TCP_STAT_PREDACK]++; 2010 tcps[TCP_STAT_RCVACKPACK]++; 2011 tcps[TCP_STAT_RCVACKBYTE] += acked; 2012 TCP_STAT_PUTREF(); 2013 nd6_hint(tp); 2014 2015 if (acked > (tp->t_lastoff - tp->t_inoff)) 2016 tp->t_lastm = NULL; 2017 sbdrop(&so->so_snd, acked); 2018 tp->t_lastoff -= acked; 2019 2020 icmp_check(tp, th, acked); 2021 2022 tp->snd_una = th->th_ack; 2023 tp->snd_fack = tp->snd_una; 2024 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2025 tp->snd_high = tp->snd_una; 2026 m_freem(m); 2027 2028 /* 2029 * If all outstanding data are acked, stop 2030 * retransmit timer, otherwise restart timer 2031 * using current (possibly backed-off) value. 2032 * If process is waiting for space, 2033 * wakeup/selnotify/signal. If data 2034 * are ready to send, let tcp_output 2035 * decide between more output or persist. 2036 */ 2037 if (tp->snd_una == tp->snd_max) 2038 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2039 else if (TCP_TIMER_ISARMED(tp, 2040 TCPT_PERSIST) == 0) 2041 TCP_TIMER_ARM(tp, TCPT_REXMT, 2042 tp->t_rxtcur); 2043 2044 sowwakeup(so); 2045 if (so->so_snd.sb_cc) { 2046 KERNEL_LOCK(1, NULL); 2047 (void) tcp_output(tp); 2048 KERNEL_UNLOCK_ONE(NULL); 2049 } 2050 if (tcp_saveti) 2051 m_freem(tcp_saveti); 2052 return; 2053 } 2054 } else if (th->th_ack == tp->snd_una && 2055 TAILQ_FIRST(&tp->segq) == NULL && 2056 tlen <= sbspace(&so->so_rcv)) { 2057 int newsize = 0; /* automatic sockbuf scaling */ 2058 2059 /* 2060 * this is a pure, in-sequence data packet 2061 * with nothing on the reassembly queue and 2062 * we have enough buffer space to take it. 2063 */ 2064 tp->rcv_nxt += tlen; 2065 tcps = TCP_STAT_GETREF(); 2066 tcps[TCP_STAT_PREDDAT]++; 2067 tcps[TCP_STAT_RCVPACK]++; 2068 tcps[TCP_STAT_RCVBYTE] += tlen; 2069 TCP_STAT_PUTREF(); 2070 nd6_hint(tp); 2071 2072 /* 2073 * Automatic sizing enables the performance of large buffers 2074 * and most of the efficiency of small ones by only allocating 2075 * space when it is needed. 2076 * 2077 * On the receive side the socket buffer memory is only rarely 2078 * used to any significant extent. This allows us to be much 2079 * more aggressive in scaling the receive socket buffer. For 2080 * the case that the buffer space is actually used to a large 2081 * extent and we run out of kernel memory we can simply drop 2082 * the new segments; TCP on the sender will just retransmit it 2083 * later. Setting the buffer size too big may only consume too 2084 * much kernel memory if the application doesn't read() from 2085 * the socket or packet loss or reordering makes use of the 2086 * reassembly queue. 2087 * 2088 * The criteria to step up the receive buffer one notch are: 2089 * 1. the number of bytes received during the time it takes 2090 * one timestamp to be reflected back to us (the RTT); 2091 * 2. received bytes per RTT is within seven eighth of the 2092 * current socket buffer size; 2093 * 3. receive buffer size has not hit maximal automatic size; 2094 * 2095 * This algorithm does one step per RTT at most and only if 2096 * we receive a bulk stream w/o packet losses or reorderings. 2097 * Shrinking the buffer during idle times is not necessary as 2098 * it doesn't consume any memory when idle. 2099 * 2100 * TODO: Only step up if the application is actually serving 2101 * the buffer to better manage the socket buffer resources. 2102 */ 2103 if (tcp_do_autorcvbuf && 2104 opti.ts_ecr && 2105 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 2106 if (opti.ts_ecr > tp->rfbuf_ts && 2107 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 2108 if (tp->rfbuf_cnt > 2109 (so->so_rcv.sb_hiwat / 8 * 7) && 2110 so->so_rcv.sb_hiwat < 2111 tcp_autorcvbuf_max) { 2112 newsize = 2113 min(so->so_rcv.sb_hiwat + 2114 tcp_autorcvbuf_inc, 2115 tcp_autorcvbuf_max); 2116 } 2117 /* Start over with next RTT. */ 2118 tp->rfbuf_ts = 0; 2119 tp->rfbuf_cnt = 0; 2120 } else 2121 tp->rfbuf_cnt += tlen; /* add up */ 2122 } 2123 2124 /* 2125 * Drop TCP, IP headers and TCP options then add data 2126 * to socket buffer. 2127 */ 2128 if (so->so_state & SS_CANTRCVMORE) 2129 m_freem(m); 2130 else { 2131 /* 2132 * Set new socket buffer size. 2133 * Give up when limit is reached. 2134 */ 2135 if (newsize) 2136 if (!sbreserve(&so->so_rcv, 2137 newsize, so)) 2138 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 2139 m_adj(m, toff + off); 2140 sbappendstream(&so->so_rcv, m); 2141 } 2142 sorwakeup(so); 2143 tcp_setup_ack(tp, th); 2144 if (tp->t_flags & TF_ACKNOW) { 2145 KERNEL_LOCK(1, NULL); 2146 (void) tcp_output(tp); 2147 KERNEL_UNLOCK_ONE(NULL); 2148 } 2149 if (tcp_saveti) 2150 m_freem(tcp_saveti); 2151 return; 2152 } 2153 } 2154 2155 /* 2156 * Compute mbuf offset to TCP data segment. 2157 */ 2158 hdroptlen = toff + off; 2159 2160 /* 2161 * Calculate amount of space in receive window, 2162 * and then do TCP input processing. 2163 * Receive window is amount of space in rcv queue, 2164 * but not less than advertised window. 2165 */ 2166 { int win; 2167 2168 win = sbspace(&so->so_rcv); 2169 if (win < 0) 2170 win = 0; 2171 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 2172 } 2173 2174 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 2175 tp->rfbuf_ts = 0; 2176 tp->rfbuf_cnt = 0; 2177 2178 switch (tp->t_state) { 2179 /* 2180 * If the state is SYN_SENT: 2181 * if seg contains an ACK, but not for our SYN, drop the input. 2182 * if seg contains a RST, then drop the connection. 2183 * if seg does not contain SYN, then drop it. 2184 * Otherwise this is an acceptable SYN segment 2185 * initialize tp->rcv_nxt and tp->irs 2186 * if seg contains ack then advance tp->snd_una 2187 * if seg contains a ECE and ECN support is enabled, the stream 2188 * is ECN capable. 2189 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2190 * arrange for segment to be acked (eventually) 2191 * continue processing rest of data/controls, beginning with URG 2192 */ 2193 case TCPS_SYN_SENT: 2194 if ((tiflags & TH_ACK) && 2195 (SEQ_LEQ(th->th_ack, tp->iss) || 2196 SEQ_GT(th->th_ack, tp->snd_max))) 2197 goto dropwithreset; 2198 if (tiflags & TH_RST) { 2199 if (tiflags & TH_ACK) 2200 tp = tcp_drop(tp, ECONNREFUSED); 2201 goto drop; 2202 } 2203 if ((tiflags & TH_SYN) == 0) 2204 goto drop; 2205 if (tiflags & TH_ACK) { 2206 tp->snd_una = th->th_ack; 2207 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2208 tp->snd_nxt = tp->snd_una; 2209 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2210 tp->snd_high = tp->snd_una; 2211 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2212 2213 if ((tiflags & TH_ECE) && tcp_do_ecn) { 2214 tp->t_flags |= TF_ECN_PERMIT; 2215 TCP_STATINC(TCP_STAT_ECN_SHS); 2216 } 2217 2218 } 2219 tp->irs = th->th_seq; 2220 tcp_rcvseqinit(tp); 2221 tp->t_flags |= TF_ACKNOW; 2222 tcp_mss_from_peer(tp, opti.maxseg); 2223 2224 /* 2225 * Initialize the initial congestion window. If we 2226 * had to retransmit the SYN, we must initialize cwnd 2227 * to 1 segment (i.e. the Loss Window). 2228 */ 2229 if (tp->t_flags & TF_SYN_REXMT) 2230 tp->snd_cwnd = tp->t_peermss; 2231 else { 2232 int ss = tcp_init_win; 2233 #ifdef INET 2234 if (inp != NULL && in_localaddr(inp->inp_faddr)) 2235 ss = tcp_init_win_local; 2236 #endif 2237 #ifdef INET6 2238 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 2239 ss = tcp_init_win_local; 2240 #endif 2241 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 2242 } 2243 2244 tcp_rmx_rtt(tp); 2245 if (tiflags & TH_ACK) { 2246 TCP_STATINC(TCP_STAT_CONNECTS); 2247 /* 2248 * move tcp_established before soisconnected 2249 * because upcall handler can drive tcp_output 2250 * functionality. 2251 * XXX we might call soisconnected at the end of 2252 * all processing 2253 */ 2254 tcp_established(tp); 2255 soisconnected(so); 2256 /* Do window scaling on this connection? */ 2257 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2258 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2259 tp->snd_scale = tp->requested_s_scale; 2260 tp->rcv_scale = tp->request_r_scale; 2261 } 2262 TCP_REASS_LOCK(tp); 2263 (void) tcp_reass(tp, NULL, NULL, &tlen); 2264 /* 2265 * if we didn't have to retransmit the SYN, 2266 * use its rtt as our initial srtt & rtt var. 2267 */ 2268 if (tp->t_rtttime) 2269 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2270 } else 2271 tp->t_state = TCPS_SYN_RECEIVED; 2272 2273 /* 2274 * Advance th->th_seq to correspond to first data byte. 2275 * If data, trim to stay within window, 2276 * dropping FIN if necessary. 2277 */ 2278 th->th_seq++; 2279 if (tlen > tp->rcv_wnd) { 2280 todrop = tlen - tp->rcv_wnd; 2281 m_adj(m, -todrop); 2282 tlen = tp->rcv_wnd; 2283 tiflags &= ~TH_FIN; 2284 tcps = TCP_STAT_GETREF(); 2285 tcps[TCP_STAT_RCVPACKAFTERWIN]++; 2286 tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop; 2287 TCP_STAT_PUTREF(); 2288 } 2289 tp->snd_wl1 = th->th_seq - 1; 2290 tp->rcv_up = th->th_seq; 2291 goto step6; 2292 2293 /* 2294 * If the state is SYN_RECEIVED: 2295 * If seg contains an ACK, but not for our SYN, drop the input 2296 * and generate an RST. See page 36, rfc793 2297 */ 2298 case TCPS_SYN_RECEIVED: 2299 if ((tiflags & TH_ACK) && 2300 (SEQ_LEQ(th->th_ack, tp->iss) || 2301 SEQ_GT(th->th_ack, tp->snd_max))) 2302 goto dropwithreset; 2303 break; 2304 } 2305 2306 /* 2307 * States other than LISTEN or SYN_SENT. 2308 * First check timestamp, if present. 2309 * Then check that at least some bytes of segment are within 2310 * receive window. If segment begins before rcv_nxt, 2311 * drop leading data (and SYN); if nothing left, just ack. 2312 * 2313 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2314 * and it's less than ts_recent, drop it. 2315 */ 2316 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2317 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2318 2319 /* Check to see if ts_recent is over 24 days old. */ 2320 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2321 /* 2322 * Invalidate ts_recent. If this segment updates 2323 * ts_recent, the age will be reset later and ts_recent 2324 * will get a valid value. If it does not, setting 2325 * ts_recent to zero will at least satisfy the 2326 * requirement that zero be placed in the timestamp 2327 * echo reply when ts_recent isn't valid. The 2328 * age isn't reset until we get a valid ts_recent 2329 * because we don't want out-of-order segments to be 2330 * dropped when ts_recent is old. 2331 */ 2332 tp->ts_recent = 0; 2333 } else { 2334 tcps = TCP_STAT_GETREF(); 2335 tcps[TCP_STAT_RCVDUPPACK]++; 2336 tcps[TCP_STAT_RCVDUPBYTE] += tlen; 2337 tcps[TCP_STAT_PAWSDROP]++; 2338 TCP_STAT_PUTREF(); 2339 tcp_new_dsack(tp, th->th_seq, tlen); 2340 goto dropafterack; 2341 } 2342 } 2343 2344 todrop = tp->rcv_nxt - th->th_seq; 2345 dupseg = false; 2346 if (todrop > 0) { 2347 if (tiflags & TH_SYN) { 2348 tiflags &= ~TH_SYN; 2349 th->th_seq++; 2350 if (th->th_urp > 1) 2351 th->th_urp--; 2352 else { 2353 tiflags &= ~TH_URG; 2354 th->th_urp = 0; 2355 } 2356 todrop--; 2357 } 2358 if (todrop > tlen || 2359 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2360 /* 2361 * Any valid FIN or RST must be to the left of the 2362 * window. At this point the FIN or RST must be a 2363 * duplicate or out of sequence; drop it. 2364 */ 2365 if (tiflags & TH_RST) 2366 goto drop; 2367 tiflags &= ~(TH_FIN|TH_RST); 2368 /* 2369 * Send an ACK to resynchronize and drop any data. 2370 * But keep on processing for RST or ACK. 2371 */ 2372 tp->t_flags |= TF_ACKNOW; 2373 todrop = tlen; 2374 dupseg = true; 2375 tcps = TCP_STAT_GETREF(); 2376 tcps[TCP_STAT_RCVDUPPACK]++; 2377 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 2378 TCP_STAT_PUTREF(); 2379 } else if ((tiflags & TH_RST) && 2380 th->th_seq != tp->rcv_nxt) { 2381 /* 2382 * Test for reset before adjusting the sequence 2383 * number for overlapping data. 2384 */ 2385 goto dropafterack_ratelim; 2386 } else { 2387 tcps = TCP_STAT_GETREF(); 2388 tcps[TCP_STAT_RCVPARTDUPPACK]++; 2389 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 2390 TCP_STAT_PUTREF(); 2391 } 2392 tcp_new_dsack(tp, th->th_seq, todrop); 2393 hdroptlen += todrop; /*drop from head afterwards*/ 2394 th->th_seq += todrop; 2395 tlen -= todrop; 2396 if (th->th_urp > todrop) 2397 th->th_urp -= todrop; 2398 else { 2399 tiflags &= ~TH_URG; 2400 th->th_urp = 0; 2401 } 2402 } 2403 2404 /* 2405 * If new data are received on a connection after the 2406 * user processes are gone, then RST the other end. 2407 */ 2408 if ((so->so_state & SS_NOFDREF) && 2409 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2410 tp = tcp_close(tp); 2411 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 2412 goto dropwithreset; 2413 } 2414 2415 /* 2416 * If segment ends after window, drop trailing data 2417 * (and PUSH and FIN); if nothing left, just ACK. 2418 */ 2419 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 2420 if (todrop > 0) { 2421 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 2422 if (todrop >= tlen) { 2423 /* 2424 * The segment actually starts after the window. 2425 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2426 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2427 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2428 */ 2429 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 2430 /* 2431 * If a new connection request is received 2432 * while in TIME_WAIT, drop the old connection 2433 * and start over if the sequence numbers 2434 * are above the previous ones. 2435 * 2436 * NOTE: We will checksum the packet again, and 2437 * so we need to put the header fields back into 2438 * network order! 2439 * XXX This kind of sucks, but we don't expect 2440 * XXX this to happen very often, so maybe it 2441 * XXX doesn't matter so much. 2442 */ 2443 if (tiflags & TH_SYN && 2444 tp->t_state == TCPS_TIME_WAIT && 2445 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2446 tp = tcp_close(tp); 2447 tcp_fields_to_net(th); 2448 goto findpcb; 2449 } 2450 /* 2451 * If window is closed can only take segments at 2452 * window edge, and have to drop data and PUSH from 2453 * incoming segments. Continue processing, but 2454 * remember to ack. Otherwise, drop segment 2455 * and (if not RST) ack. 2456 */ 2457 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2458 tp->t_flags |= TF_ACKNOW; 2459 TCP_STATINC(TCP_STAT_RCVWINPROBE); 2460 } else 2461 goto dropafterack; 2462 } else 2463 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 2464 m_adj(m, -todrop); 2465 tlen -= todrop; 2466 tiflags &= ~(TH_PUSH|TH_FIN); 2467 } 2468 2469 /* 2470 * If last ACK falls within this segment's sequence numbers, 2471 * record the timestamp. 2472 * NOTE: 2473 * 1) That the test incorporates suggestions from the latest 2474 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2475 * 2) That updating only on newer timestamps interferes with 2476 * our earlier PAWS tests, so this check should be solely 2477 * predicated on the sequence space of this segment. 2478 * 3) That we modify the segment boundary check to be 2479 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2480 * instead of RFC1323's 2481 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2482 * This modified check allows us to overcome RFC1323's 2483 * limitations as described in Stevens TCP/IP Illustrated 2484 * Vol. 2 p.869. In such cases, we can still calculate the 2485 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2486 */ 2487 if (opti.ts_present && 2488 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2489 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2490 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2491 tp->ts_recent_age = tcp_now; 2492 tp->ts_recent = opti.ts_val; 2493 } 2494 2495 /* 2496 * If the RST bit is set examine the state: 2497 * SYN_RECEIVED STATE: 2498 * If passive open, return to LISTEN state. 2499 * If active open, inform user that connection was refused. 2500 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 2501 * Inform user that connection was reset, and close tcb. 2502 * CLOSING, LAST_ACK, TIME_WAIT STATES 2503 * Close the tcb. 2504 */ 2505 if (tiflags & TH_RST) { 2506 if (th->th_seq != tp->rcv_nxt) 2507 goto dropafterack_ratelim; 2508 2509 switch (tp->t_state) { 2510 case TCPS_SYN_RECEIVED: 2511 so->so_error = ECONNREFUSED; 2512 goto close; 2513 2514 case TCPS_ESTABLISHED: 2515 case TCPS_FIN_WAIT_1: 2516 case TCPS_FIN_WAIT_2: 2517 case TCPS_CLOSE_WAIT: 2518 so->so_error = ECONNRESET; 2519 close: 2520 tp->t_state = TCPS_CLOSED; 2521 TCP_STATINC(TCP_STAT_DROPS); 2522 tp = tcp_close(tp); 2523 goto drop; 2524 2525 case TCPS_CLOSING: 2526 case TCPS_LAST_ACK: 2527 case TCPS_TIME_WAIT: 2528 tp = tcp_close(tp); 2529 goto drop; 2530 } 2531 } 2532 2533 /* 2534 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2535 * we must be in a synchronized state. RFC791 states (under RST 2536 * generation) that any unacceptable segment (an out-of-order SYN 2537 * qualifies) received in a synchronized state must elicit only an 2538 * empty acknowledgment segment ... and the connection remains in 2539 * the same state. 2540 */ 2541 if (tiflags & TH_SYN) { 2542 if (tp->rcv_nxt == th->th_seq) { 2543 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2544 TH_ACK); 2545 if (tcp_saveti) 2546 m_freem(tcp_saveti); 2547 return; 2548 } 2549 2550 goto dropafterack_ratelim; 2551 } 2552 2553 /* 2554 * If the ACK bit is off we drop the segment and return. 2555 */ 2556 if ((tiflags & TH_ACK) == 0) { 2557 if (tp->t_flags & TF_ACKNOW) 2558 goto dropafterack; 2559 else 2560 goto drop; 2561 } 2562 2563 /* 2564 * Ack processing. 2565 */ 2566 switch (tp->t_state) { 2567 2568 /* 2569 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2570 * ESTABLISHED state and continue processing, otherwise 2571 * send an RST. 2572 */ 2573 case TCPS_SYN_RECEIVED: 2574 if (SEQ_GT(tp->snd_una, th->th_ack) || 2575 SEQ_GT(th->th_ack, tp->snd_max)) 2576 goto dropwithreset; 2577 TCP_STATINC(TCP_STAT_CONNECTS); 2578 soisconnected(so); 2579 tcp_established(tp); 2580 /* Do window scaling? */ 2581 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2582 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2583 tp->snd_scale = tp->requested_s_scale; 2584 tp->rcv_scale = tp->request_r_scale; 2585 } 2586 TCP_REASS_LOCK(tp); 2587 (void) tcp_reass(tp, NULL, NULL, &tlen); 2588 tp->snd_wl1 = th->th_seq - 1; 2589 /* fall into ... */ 2590 2591 /* 2592 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2593 * ACKs. If the ack is in the range 2594 * tp->snd_una < th->th_ack <= tp->snd_max 2595 * then advance tp->snd_una to th->th_ack and drop 2596 * data from the retransmission queue. If this ACK reflects 2597 * more up to date window information we update our window information. 2598 */ 2599 case TCPS_ESTABLISHED: 2600 case TCPS_FIN_WAIT_1: 2601 case TCPS_FIN_WAIT_2: 2602 case TCPS_CLOSE_WAIT: 2603 case TCPS_CLOSING: 2604 case TCPS_LAST_ACK: 2605 case TCPS_TIME_WAIT: 2606 2607 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2608 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2609 TCP_STATINC(TCP_STAT_RCVDUPACK); 2610 /* 2611 * If we have outstanding data (other than 2612 * a window probe), this is a completely 2613 * duplicate ack (ie, window info didn't 2614 * change), the ack is the biggest we've 2615 * seen and we've seen exactly our rexmt 2616 * threshhold of them, assume a packet 2617 * has been dropped and retransmit it. 2618 * Kludge snd_nxt & the congestion 2619 * window so we send only this one 2620 * packet. 2621 */ 2622 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2623 th->th_ack != tp->snd_una) 2624 tp->t_dupacks = 0; 2625 else if (tp->t_partialacks < 0 && 2626 (++tp->t_dupacks == tcprexmtthresh || 2627 TCP_FACK_FASTRECOV(tp))) { 2628 /* 2629 * Do the fast retransmit, and adjust 2630 * congestion control paramenters. 2631 */ 2632 if (tp->t_congctl->fast_retransmit(tp, th)) { 2633 /* False fast retransmit */ 2634 break; 2635 } else 2636 goto drop; 2637 } else if (tp->t_dupacks > tcprexmtthresh) { 2638 tp->snd_cwnd += tp->t_segsz; 2639 KERNEL_LOCK(1, NULL); 2640 (void) tcp_output(tp); 2641 KERNEL_UNLOCK_ONE(NULL); 2642 goto drop; 2643 } 2644 } else { 2645 /* 2646 * If the ack appears to be very old, only 2647 * allow data that is in-sequence. This 2648 * makes it somewhat more difficult to insert 2649 * forged data by guessing sequence numbers. 2650 * Sent an ack to try to update the send 2651 * sequence number on the other side. 2652 */ 2653 if (tlen && th->th_seq != tp->rcv_nxt && 2654 SEQ_LT(th->th_ack, 2655 tp->snd_una - tp->max_sndwnd)) 2656 goto dropafterack; 2657 } 2658 break; 2659 } 2660 /* 2661 * If the congestion window was inflated to account 2662 * for the other side's cached packets, retract it. 2663 */ 2664 tp->t_congctl->fast_retransmit_newack(tp, th); 2665 2666 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2667 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2668 goto dropafterack; 2669 } 2670 acked = th->th_ack - tp->snd_una; 2671 tcps = TCP_STAT_GETREF(); 2672 tcps[TCP_STAT_RCVACKPACK]++; 2673 tcps[TCP_STAT_RCVACKBYTE] += acked; 2674 TCP_STAT_PUTREF(); 2675 2676 /* 2677 * If we have a timestamp reply, update smoothed 2678 * round trip time. If no timestamp is present but 2679 * transmit timer is running and timed sequence 2680 * number was acked, update smoothed round trip time. 2681 * Since we now have an rtt measurement, cancel the 2682 * timer backoff (cf., Phil Karn's retransmit alg.). 2683 * Recompute the initial retransmit timer. 2684 */ 2685 if (ts_rtt) 2686 tcp_xmit_timer(tp, ts_rtt - 1); 2687 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2688 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2689 2690 /* 2691 * If all outstanding data is acked, stop retransmit 2692 * timer and remember to restart (more output or persist). 2693 * If there is more data to be acked, restart retransmit 2694 * timer, using current (possibly backed-off) value. 2695 */ 2696 if (th->th_ack == tp->snd_max) { 2697 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2698 needoutput = 1; 2699 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2700 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2701 2702 /* 2703 * New data has been acked, adjust the congestion window. 2704 */ 2705 tp->t_congctl->newack(tp, th); 2706 2707 nd6_hint(tp); 2708 if (acked > so->so_snd.sb_cc) { 2709 tp->snd_wnd -= so->so_snd.sb_cc; 2710 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2711 ourfinisacked = 1; 2712 } else { 2713 if (acked > (tp->t_lastoff - tp->t_inoff)) 2714 tp->t_lastm = NULL; 2715 sbdrop(&so->so_snd, acked); 2716 tp->t_lastoff -= acked; 2717 tp->snd_wnd -= acked; 2718 ourfinisacked = 0; 2719 } 2720 sowwakeup(so); 2721 2722 icmp_check(tp, th, acked); 2723 2724 tp->snd_una = th->th_ack; 2725 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2726 tp->snd_fack = tp->snd_una; 2727 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2728 tp->snd_nxt = tp->snd_una; 2729 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2730 tp->snd_high = tp->snd_una; 2731 2732 switch (tp->t_state) { 2733 2734 /* 2735 * In FIN_WAIT_1 STATE in addition to the processing 2736 * for the ESTABLISHED state if our FIN is now acknowledged 2737 * then enter FIN_WAIT_2. 2738 */ 2739 case TCPS_FIN_WAIT_1: 2740 if (ourfinisacked) { 2741 /* 2742 * If we can't receive any more 2743 * data, then closing user can proceed. 2744 * Starting the timer is contrary to the 2745 * specification, but if we don't get a FIN 2746 * we'll hang forever. 2747 */ 2748 if (so->so_state & SS_CANTRCVMORE) { 2749 soisdisconnected(so); 2750 if (tp->t_maxidle > 0) 2751 TCP_TIMER_ARM(tp, TCPT_2MSL, 2752 tp->t_maxidle); 2753 } 2754 tp->t_state = TCPS_FIN_WAIT_2; 2755 } 2756 break; 2757 2758 /* 2759 * In CLOSING STATE in addition to the processing for 2760 * the ESTABLISHED state if the ACK acknowledges our FIN 2761 * then enter the TIME-WAIT state, otherwise ignore 2762 * the segment. 2763 */ 2764 case TCPS_CLOSING: 2765 if (ourfinisacked) { 2766 tp->t_state = TCPS_TIME_WAIT; 2767 tcp_canceltimers(tp); 2768 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2769 soisdisconnected(so); 2770 } 2771 break; 2772 2773 /* 2774 * In LAST_ACK, we may still be waiting for data to drain 2775 * and/or to be acked, as well as for the ack of our FIN. 2776 * If our FIN is now acknowledged, delete the TCB, 2777 * enter the closed state and return. 2778 */ 2779 case TCPS_LAST_ACK: 2780 if (ourfinisacked) { 2781 tp = tcp_close(tp); 2782 goto drop; 2783 } 2784 break; 2785 2786 /* 2787 * In TIME_WAIT state the only thing that should arrive 2788 * is a retransmission of the remote FIN. Acknowledge 2789 * it and restart the finack timer. 2790 */ 2791 case TCPS_TIME_WAIT: 2792 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2793 goto dropafterack; 2794 } 2795 } 2796 2797 step6: 2798 /* 2799 * Update window information. 2800 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2801 */ 2802 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2803 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2804 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2805 /* keep track of pure window updates */ 2806 if (tlen == 0 && 2807 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2808 TCP_STATINC(TCP_STAT_RCVWINUPD); 2809 tp->snd_wnd = tiwin; 2810 tp->snd_wl1 = th->th_seq; 2811 tp->snd_wl2 = th->th_ack; 2812 if (tp->snd_wnd > tp->max_sndwnd) 2813 tp->max_sndwnd = tp->snd_wnd; 2814 needoutput = 1; 2815 } 2816 2817 /* 2818 * Process segments with URG. 2819 */ 2820 if ((tiflags & TH_URG) && th->th_urp && 2821 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2822 /* 2823 * This is a kludge, but if we receive and accept 2824 * random urgent pointers, we'll crash in 2825 * soreceive. It's hard to imagine someone 2826 * actually wanting to send this much urgent data. 2827 */ 2828 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2829 th->th_urp = 0; /* XXX */ 2830 tiflags &= ~TH_URG; /* XXX */ 2831 goto dodata; /* XXX */ 2832 } 2833 /* 2834 * If this segment advances the known urgent pointer, 2835 * then mark the data stream. This should not happen 2836 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2837 * a FIN has been received from the remote side. 2838 * In these states we ignore the URG. 2839 * 2840 * According to RFC961 (Assigned Protocols), 2841 * the urgent pointer points to the last octet 2842 * of urgent data. We continue, however, 2843 * to consider it to indicate the first octet 2844 * of data past the urgent section as the original 2845 * spec states (in one of two places). 2846 */ 2847 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2848 tp->rcv_up = th->th_seq + th->th_urp; 2849 so->so_oobmark = so->so_rcv.sb_cc + 2850 (tp->rcv_up - tp->rcv_nxt) - 1; 2851 if (so->so_oobmark == 0) 2852 so->so_state |= SS_RCVATMARK; 2853 sohasoutofband(so); 2854 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2855 } 2856 /* 2857 * Remove out of band data so doesn't get presented to user. 2858 * This can happen independent of advancing the URG pointer, 2859 * but if two URG's are pending at once, some out-of-band 2860 * data may creep in... ick. 2861 */ 2862 if (th->th_urp <= (u_int16_t) tlen 2863 #ifdef SO_OOBINLINE 2864 && (so->so_options & SO_OOBINLINE) == 0 2865 #endif 2866 ) 2867 tcp_pulloutofband(so, th, m, hdroptlen); 2868 } else 2869 /* 2870 * If no out of band data is expected, 2871 * pull receive urgent pointer along 2872 * with the receive window. 2873 */ 2874 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2875 tp->rcv_up = tp->rcv_nxt; 2876 dodata: /* XXX */ 2877 2878 /* 2879 * Process the segment text, merging it into the TCP sequencing queue, 2880 * and arranging for acknowledgement of receipt if necessary. 2881 * This process logically involves adjusting tp->rcv_wnd as data 2882 * is presented to the user (this happens in tcp_usrreq.c, 2883 * tcp_rcvd()). If a FIN has already been received on this 2884 * connection then we just ignore the text. 2885 */ 2886 if ((tlen || (tiflags & TH_FIN)) && 2887 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2888 /* 2889 * Insert segment ti into reassembly queue of tcp with 2890 * control block tp. Return TH_FIN if reassembly now includes 2891 * a segment with FIN. The macro form does the common case 2892 * inline (segment is the next to be received on an 2893 * established connection, and the queue is empty), 2894 * avoiding linkage into and removal from the queue and 2895 * repetition of various conversions. 2896 * Set DELACK for segments received in order, but ack 2897 * immediately when segments are out of order 2898 * (so fast retransmit can work). 2899 */ 2900 /* NOTE: this was TCP_REASS() macro, but used only once */ 2901 TCP_REASS_LOCK(tp); 2902 if (th->th_seq == tp->rcv_nxt && 2903 TAILQ_FIRST(&tp->segq) == NULL && 2904 tp->t_state == TCPS_ESTABLISHED) { 2905 tcp_setup_ack(tp, th); 2906 tp->rcv_nxt += tlen; 2907 tiflags = th->th_flags & TH_FIN; 2908 tcps = TCP_STAT_GETREF(); 2909 tcps[TCP_STAT_RCVPACK]++; 2910 tcps[TCP_STAT_RCVBYTE] += tlen; 2911 TCP_STAT_PUTREF(); 2912 nd6_hint(tp); 2913 if (so->so_state & SS_CANTRCVMORE) 2914 m_freem(m); 2915 else { 2916 m_adj(m, hdroptlen); 2917 sbappendstream(&(so)->so_rcv, m); 2918 } 2919 TCP_REASS_UNLOCK(tp); 2920 sorwakeup(so); 2921 } else { 2922 m_adj(m, hdroptlen); 2923 tiflags = tcp_reass(tp, th, m, &tlen); 2924 tp->t_flags |= TF_ACKNOW; 2925 } 2926 2927 /* 2928 * Note the amount of data that peer has sent into 2929 * our window, in order to estimate the sender's 2930 * buffer size. 2931 */ 2932 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2933 } else { 2934 m_freem(m); 2935 m = NULL; 2936 tiflags &= ~TH_FIN; 2937 } 2938 2939 /* 2940 * If FIN is received ACK the FIN and let the user know 2941 * that the connection is closing. Ignore a FIN received before 2942 * the connection is fully established. 2943 */ 2944 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2945 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2946 socantrcvmore(so); 2947 tp->t_flags |= TF_ACKNOW; 2948 tp->rcv_nxt++; 2949 } 2950 switch (tp->t_state) { 2951 2952 /* 2953 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2954 */ 2955 case TCPS_ESTABLISHED: 2956 tp->t_state = TCPS_CLOSE_WAIT; 2957 break; 2958 2959 /* 2960 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2961 * enter the CLOSING state. 2962 */ 2963 case TCPS_FIN_WAIT_1: 2964 tp->t_state = TCPS_CLOSING; 2965 break; 2966 2967 /* 2968 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2969 * starting the time-wait timer, turning off the other 2970 * standard timers. 2971 */ 2972 case TCPS_FIN_WAIT_2: 2973 tp->t_state = TCPS_TIME_WAIT; 2974 tcp_canceltimers(tp); 2975 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2976 soisdisconnected(so); 2977 break; 2978 2979 /* 2980 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2981 */ 2982 case TCPS_TIME_WAIT: 2983 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2984 break; 2985 } 2986 } 2987 #ifdef TCP_DEBUG 2988 if (so->so_options & SO_DEBUG) 2989 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2990 #endif 2991 2992 /* 2993 * Return any desired output. 2994 */ 2995 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2996 KERNEL_LOCK(1, NULL); 2997 (void) tcp_output(tp); 2998 KERNEL_UNLOCK_ONE(NULL); 2999 } 3000 if (tcp_saveti) 3001 m_freem(tcp_saveti); 3002 3003 if (tp->t_state == TCPS_TIME_WAIT 3004 && (so->so_state & SS_NOFDREF) 3005 && (tp->t_inpcb || af != AF_INET) 3006 && (tp->t_in6pcb || af != AF_INET6) 3007 && ((af == AF_INET ? tcp4_vtw_enable : tcp6_vtw_enable) & 1) != 0 3008 && TAILQ_EMPTY(&tp->segq) 3009 && vtw_add(af, tp)) { 3010 ; 3011 } 3012 return; 3013 3014 badsyn: 3015 /* 3016 * Received a bad SYN. Increment counters and dropwithreset. 3017 */ 3018 TCP_STATINC(TCP_STAT_BADSYN); 3019 tp = NULL; 3020 goto dropwithreset; 3021 3022 dropafterack: 3023 /* 3024 * Generate an ACK dropping incoming segment if it occupies 3025 * sequence space, where the ACK reflects our state. 3026 */ 3027 if (tiflags & TH_RST) 3028 goto drop; 3029 goto dropafterack2; 3030 3031 dropafterack_ratelim: 3032 /* 3033 * We may want to rate-limit ACKs against SYN/RST attack. 3034 */ 3035 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 3036 tcp_ackdrop_ppslim) == 0) { 3037 /* XXX stat */ 3038 goto drop; 3039 } 3040 /* ...fall into dropafterack2... */ 3041 3042 dropafterack2: 3043 m_freem(m); 3044 tp->t_flags |= TF_ACKNOW; 3045 KERNEL_LOCK(1, NULL); 3046 (void) tcp_output(tp); 3047 KERNEL_UNLOCK_ONE(NULL); 3048 if (tcp_saveti) 3049 m_freem(tcp_saveti); 3050 return; 3051 3052 dropwithreset_ratelim: 3053 /* 3054 * We may want to rate-limit RSTs in certain situations, 3055 * particularly if we are sending an RST in response to 3056 * an attempt to connect to or otherwise communicate with 3057 * a port for which we have no socket. 3058 */ 3059 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 3060 tcp_rst_ppslim) == 0) { 3061 /* XXX stat */ 3062 goto drop; 3063 } 3064 /* ...fall into dropwithreset... */ 3065 3066 dropwithreset: 3067 /* 3068 * Generate a RST, dropping incoming segment. 3069 * Make ACK acceptable to originator of segment. 3070 */ 3071 if (tiflags & TH_RST) 3072 goto drop; 3073 3074 switch (af) { 3075 #ifdef INET6 3076 case AF_INET6: 3077 /* For following calls to tcp_respond */ 3078 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 3079 goto drop; 3080 break; 3081 #endif /* INET6 */ 3082 case AF_INET: 3083 if (IN_MULTICAST(ip->ip_dst.s_addr) || 3084 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3085 goto drop; 3086 } 3087 3088 if (tiflags & TH_ACK) 3089 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3090 else { 3091 if (tiflags & TH_SYN) 3092 tlen++; 3093 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 3094 TH_RST|TH_ACK); 3095 } 3096 if (tcp_saveti) 3097 m_freem(tcp_saveti); 3098 return; 3099 3100 badcsum: 3101 drop: 3102 /* 3103 * Drop space held by incoming segment and return. 3104 */ 3105 if (tp) { 3106 if (tp->t_inpcb) 3107 so = tp->t_inpcb->inp_socket; 3108 #ifdef INET6 3109 else if (tp->t_in6pcb) 3110 so = tp->t_in6pcb->in6p_socket; 3111 #endif 3112 else 3113 so = NULL; 3114 #ifdef TCP_DEBUG 3115 if (so && (so->so_options & SO_DEBUG) != 0) 3116 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 3117 #endif 3118 } 3119 if (tcp_saveti) 3120 m_freem(tcp_saveti); 3121 m_freem(m); 3122 return; 3123 } 3124 3125 #ifdef TCP_SIGNATURE 3126 int 3127 tcp_signature_apply(void *fstate, void *data, u_int len) 3128 { 3129 3130 MD5Update(fstate, (u_char *)data, len); 3131 return (0); 3132 } 3133 3134 struct secasvar * 3135 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 3136 { 3137 struct ip *ip; 3138 struct ip6_hdr *ip6; 3139 3140 ip = mtod(m, struct ip *); 3141 switch (ip->ip_v) { 3142 case 4: 3143 ip = mtod(m, struct ip *); 3144 ip6 = NULL; 3145 break; 3146 case 6: 3147 ip = NULL; 3148 ip6 = mtod(m, struct ip6_hdr *); 3149 break; 3150 default: 3151 return (NULL); 3152 } 3153 3154 #ifdef IPSEC 3155 if (ipsec_used) { 3156 union sockaddr_union dst; 3157 /* Extract the destination from the IP header in the mbuf. */ 3158 memset(&dst, 0, sizeof(union sockaddr_union)); 3159 if (ip != NULL) { 3160 dst.sa.sa_len = sizeof(struct sockaddr_in); 3161 dst.sa.sa_family = AF_INET; 3162 dst.sin.sin_addr = ip->ip_dst; 3163 } else { 3164 dst.sa.sa_len = sizeof(struct sockaddr_in6); 3165 dst.sa.sa_family = AF_INET6; 3166 dst.sin6.sin6_addr = ip6->ip6_dst; 3167 } 3168 3169 /* 3170 * Look up an SADB entry which matches the address of the peer. 3171 */ 3172 return KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI), 0, 0); 3173 } 3174 return NULL; 3175 #else 3176 if (ip) 3177 return key_allocsa(AF_INET, (void *)&ip->ip_src, 3178 (void *)&ip->ip_dst, IPPROTO_TCP, 3179 htonl(TCP_SIG_SPI), 0, 0); 3180 else 3181 return key_allocsa(AF_INET6, (void *)&ip6->ip6_src, 3182 (void *)&ip6->ip6_dst, IPPROTO_TCP, 3183 htonl(TCP_SIG_SPI), 0, 0); 3184 #endif 3185 } 3186 3187 int 3188 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 3189 struct secasvar *sav, char *sig) 3190 { 3191 MD5_CTX ctx; 3192 struct ip *ip; 3193 struct ipovly *ipovly; 3194 #ifdef INET6 3195 struct ip6_hdr *ip6; 3196 struct ip6_hdr_pseudo ip6pseudo; 3197 #endif /* INET6 */ 3198 struct ippseudo ippseudo; 3199 struct tcphdr th0; 3200 int l, tcphdrlen; 3201 3202 if (sav == NULL) 3203 return (-1); 3204 3205 tcphdrlen = th->th_off * 4; 3206 3207 switch (mtod(m, struct ip *)->ip_v) { 3208 case 4: 3209 MD5Init(&ctx); 3210 ip = mtod(m, struct ip *); 3211 memset(&ippseudo, 0, sizeof(ippseudo)); 3212 ipovly = (struct ipovly *)ip; 3213 ippseudo.ippseudo_src = ipovly->ih_src; 3214 ippseudo.ippseudo_dst = ipovly->ih_dst; 3215 ippseudo.ippseudo_pad = 0; 3216 ippseudo.ippseudo_p = IPPROTO_TCP; 3217 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 3218 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 3219 break; 3220 #if INET6 3221 case 6: 3222 MD5Init(&ctx); 3223 ip6 = mtod(m, struct ip6_hdr *); 3224 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 3225 ip6pseudo.ip6ph_src = ip6->ip6_src; 3226 in6_clearscope(&ip6pseudo.ip6ph_src); 3227 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 3228 in6_clearscope(&ip6pseudo.ip6ph_dst); 3229 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 3230 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 3231 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 3232 break; 3233 #endif /* INET6 */ 3234 default: 3235 return (-1); 3236 } 3237 3238 th0 = *th; 3239 th0.th_sum = 0; 3240 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 3241 3242 l = m->m_pkthdr.len - thoff - tcphdrlen; 3243 if (l > 0) 3244 m_apply(m, thoff + tcphdrlen, 3245 m->m_pkthdr.len - thoff - tcphdrlen, 3246 tcp_signature_apply, &ctx); 3247 3248 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 3249 MD5Final(sig, &ctx); 3250 3251 return (0); 3252 } 3253 #endif 3254 3255 /* 3256 * tcp_dooptions: parse and process tcp options. 3257 * 3258 * returns -1 if this segment should be dropped. (eg. wrong signature) 3259 * otherwise returns 0. 3260 */ 3261 3262 static int 3263 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, 3264 struct tcphdr *th, 3265 struct mbuf *m, int toff, struct tcp_opt_info *oi) 3266 { 3267 u_int16_t mss; 3268 int opt, optlen = 0; 3269 #ifdef TCP_SIGNATURE 3270 void *sigp = NULL; 3271 char sigbuf[TCP_SIGLEN]; 3272 struct secasvar *sav = NULL; 3273 #endif 3274 3275 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 3276 opt = cp[0]; 3277 if (opt == TCPOPT_EOL) 3278 break; 3279 if (opt == TCPOPT_NOP) 3280 optlen = 1; 3281 else { 3282 if (cnt < 2) 3283 break; 3284 optlen = cp[1]; 3285 if (optlen < 2 || optlen > cnt) 3286 break; 3287 } 3288 switch (opt) { 3289 3290 default: 3291 continue; 3292 3293 case TCPOPT_MAXSEG: 3294 if (optlen != TCPOLEN_MAXSEG) 3295 continue; 3296 if (!(th->th_flags & TH_SYN)) 3297 continue; 3298 if (TCPS_HAVERCVDSYN(tp->t_state)) 3299 continue; 3300 bcopy(cp + 2, &mss, sizeof(mss)); 3301 oi->maxseg = ntohs(mss); 3302 break; 3303 3304 case TCPOPT_WINDOW: 3305 if (optlen != TCPOLEN_WINDOW) 3306 continue; 3307 if (!(th->th_flags & TH_SYN)) 3308 continue; 3309 if (TCPS_HAVERCVDSYN(tp->t_state)) 3310 continue; 3311 tp->t_flags |= TF_RCVD_SCALE; 3312 tp->requested_s_scale = cp[2]; 3313 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3314 char buf[INET6_ADDRSTRLEN]; 3315 struct ip *ip = mtod(m, struct ip *); 3316 #ifdef INET6 3317 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 3318 #endif 3319 if (ip) 3320 in_print(buf, sizeof(buf), 3321 &ip->ip_src); 3322 #ifdef INET6 3323 else if (ip6) 3324 in6_print(buf, sizeof(buf), 3325 &ip6->ip6_src); 3326 #endif 3327 else 3328 strlcpy(buf, "(unknown)", sizeof(buf)); 3329 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3330 "assuming %d\n", 3331 tp->requested_s_scale, buf, 3332 TCP_MAX_WINSHIFT); 3333 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3334 } 3335 break; 3336 3337 case TCPOPT_TIMESTAMP: 3338 if (optlen != TCPOLEN_TIMESTAMP) 3339 continue; 3340 oi->ts_present = 1; 3341 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 3342 NTOHL(oi->ts_val); 3343 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 3344 NTOHL(oi->ts_ecr); 3345 3346 if (!(th->th_flags & TH_SYN)) 3347 continue; 3348 if (TCPS_HAVERCVDSYN(tp->t_state)) 3349 continue; 3350 /* 3351 * A timestamp received in a SYN makes 3352 * it ok to send timestamp requests and replies. 3353 */ 3354 tp->t_flags |= TF_RCVD_TSTMP; 3355 tp->ts_recent = oi->ts_val; 3356 tp->ts_recent_age = tcp_now; 3357 break; 3358 3359 case TCPOPT_SACK_PERMITTED: 3360 if (optlen != TCPOLEN_SACK_PERMITTED) 3361 continue; 3362 if (!(th->th_flags & TH_SYN)) 3363 continue; 3364 if (TCPS_HAVERCVDSYN(tp->t_state)) 3365 continue; 3366 if (tcp_do_sack) { 3367 tp->t_flags |= TF_SACK_PERMIT; 3368 tp->t_flags |= TF_WILL_SACK; 3369 } 3370 break; 3371 3372 case TCPOPT_SACK: 3373 tcp_sack_option(tp, th, cp, optlen); 3374 break; 3375 #ifdef TCP_SIGNATURE 3376 case TCPOPT_SIGNATURE: 3377 if (optlen != TCPOLEN_SIGNATURE) 3378 continue; 3379 if (sigp && memcmp(sigp, cp + 2, TCP_SIGLEN)) 3380 return (-1); 3381 3382 sigp = sigbuf; 3383 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3384 tp->t_flags |= TF_SIGNATURE; 3385 break; 3386 #endif 3387 } 3388 } 3389 3390 #ifndef TCP_SIGNATURE 3391 return 0; 3392 #else 3393 if (tp->t_flags & TF_SIGNATURE) { 3394 3395 sav = tcp_signature_getsav(m, th); 3396 3397 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3398 return (-1); 3399 } 3400 3401 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) 3402 goto out; 3403 3404 if (sigp) { 3405 char sig[TCP_SIGLEN]; 3406 3407 tcp_fields_to_net(th); 3408 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3409 tcp_fields_to_host(th); 3410 goto out; 3411 } 3412 tcp_fields_to_host(th); 3413 3414 if (memcmp(sig, sigp, TCP_SIGLEN)) { 3415 TCP_STATINC(TCP_STAT_BADSIG); 3416 goto out; 3417 } else 3418 TCP_STATINC(TCP_STAT_GOODSIG); 3419 3420 key_sa_recordxfer(sav, m); 3421 KEY_FREESAV(&sav); 3422 } 3423 return 0; 3424 out: 3425 if (sav != NULL) 3426 KEY_FREESAV(&sav); 3427 return -1; 3428 #endif 3429 } 3430 3431 /* 3432 * Pull out of band byte out of a segment so 3433 * it doesn't appear in the user's data queue. 3434 * It is still reflected in the segment length for 3435 * sequencing purposes. 3436 */ 3437 void 3438 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3439 struct mbuf *m, int off) 3440 { 3441 int cnt = off + th->th_urp - 1; 3442 3443 while (cnt >= 0) { 3444 if (m->m_len > cnt) { 3445 char *cp = mtod(m, char *) + cnt; 3446 struct tcpcb *tp = sototcpcb(so); 3447 3448 tp->t_iobc = *cp; 3449 tp->t_oobflags |= TCPOOB_HAVEDATA; 3450 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3451 m->m_len--; 3452 return; 3453 } 3454 cnt -= m->m_len; 3455 m = m->m_next; 3456 if (m == 0) 3457 break; 3458 } 3459 panic("tcp_pulloutofband"); 3460 } 3461 3462 /* 3463 * Collect new round-trip time estimate 3464 * and update averages and current timeout. 3465 * 3466 * rtt is in units of slow ticks (typically 500 ms) -- essentially the 3467 * difference of two timestamps. 3468 */ 3469 void 3470 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3471 { 3472 int32_t delta; 3473 3474 TCP_STATINC(TCP_STAT_RTTUPDATED); 3475 if (tp->t_srtt != 0) { 3476 /* 3477 * Compute the amount to add to srtt for smoothing, 3478 * *alpha, or 2^(-TCP_RTT_SHIFT). Because 3479 * srtt is stored in 1/32 slow ticks, we conceptually 3480 * shift left 5 bits, subtract srtt to get the 3481 * diference, and then shift right by TCP_RTT_SHIFT 3482 * (3) to obtain 1/8 of the difference. 3483 */ 3484 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3485 /* 3486 * This can never happen, because delta's lowest 3487 * possible value is 1/8 of t_srtt. But if it does, 3488 * set srtt to some reasonable value, here chosen 3489 * as 1/8 tick. 3490 */ 3491 if ((tp->t_srtt += delta) <= 0) 3492 tp->t_srtt = 1 << 2; 3493 /* 3494 * RFC2988 requires that rttvar be updated first. 3495 * This code is compliant because "delta" is the old 3496 * srtt minus the new observation (scaled). 3497 * 3498 * RFC2988 says: 3499 * rttvar = (1-beta) * rttvar + beta * |srtt-observed| 3500 * 3501 * delta is in units of 1/32 ticks, and has then been 3502 * divided by 8. This is equivalent to being in 1/16s 3503 * units and divided by 4. Subtract from it 1/4 of 3504 * the existing rttvar to form the (signed) amount to 3505 * adjust. 3506 */ 3507 if (delta < 0) 3508 delta = -delta; 3509 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3510 /* 3511 * As with srtt, this should never happen. There is 3512 * no support in RFC2988 for this operation. But 1/4s 3513 * as rttvar when faced with something arguably wrong 3514 * is ok. 3515 */ 3516 if ((tp->t_rttvar += delta) <= 0) 3517 tp->t_rttvar = 1 << 2; 3518 3519 /* 3520 * If srtt exceeds .01 second, ensure we use the 'remote' MSL 3521 * Problem is: it doesn't work. Disabled by defaulting 3522 * tcp_rttlocal to 0; see corresponding code in 3523 * tcp_subr that selects local vs remote in a different way. 3524 * 3525 * The static branch prediction hint here should be removed 3526 * when the rtt estimator is fixed and the rtt_enable code 3527 * is turned back on. 3528 */ 3529 if (__predict_false(tcp_rttlocal) && tcp_msl_enable 3530 && tp->t_srtt > tcp_msl_remote_threshold 3531 && tp->t_msl < tcp_msl_remote) { 3532 tp->t_msl = tcp_msl_remote; 3533 } 3534 } else { 3535 /* 3536 * This is the first measurement. Per RFC2988, 2.2, 3537 * set rtt=R and srtt=R/2. 3538 * For srtt, storage representation is 1/32 ticks, 3539 * so shift left by 5. 3540 * For rttvar, storage representation is 1/16 ticks, 3541 * So shift left by 4, but then right by 1 to halve. 3542 */ 3543 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3544 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3545 } 3546 tp->t_rtttime = 0; 3547 tp->t_rxtshift = 0; 3548 3549 /* 3550 * the retransmit should happen at rtt + 4 * rttvar. 3551 * Because of the way we do the smoothing, srtt and rttvar 3552 * will each average +1/2 tick of bias. When we compute 3553 * the retransmit timer, we want 1/2 tick of rounding and 3554 * 1 extra tick because of +-1/2 tick uncertainty in the 3555 * firing of the timer. The bias will give us exactly the 3556 * 1.5 tick we need. But, because the bias is 3557 * statistical, we have to test that we don't drop below 3558 * the minimum feasible timer (which is 2 ticks). 3559 */ 3560 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3561 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3562 3563 /* 3564 * We received an ack for a packet that wasn't retransmitted; 3565 * it is probably safe to discard any error indications we've 3566 * received recently. This isn't quite right, but close enough 3567 * for now (a route might have failed after we sent a segment, 3568 * and the return path might not be symmetrical). 3569 */ 3570 tp->t_softerror = 0; 3571 } 3572 3573 3574 /* 3575 * TCP compressed state engine. Currently used to hold compressed 3576 * state for SYN_RECEIVED. 3577 */ 3578 3579 u_long syn_cache_count; 3580 u_int32_t syn_hash1, syn_hash2; 3581 3582 #define SYN_HASH(sa, sp, dp) \ 3583 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3584 ((u_int32_t)(sp)))^syn_hash2))) 3585 #ifndef INET6 3586 #define SYN_HASHALL(hash, src, dst) \ 3587 do { \ 3588 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3589 ((const struct sockaddr_in *)(src))->sin_port, \ 3590 ((const struct sockaddr_in *)(dst))->sin_port); \ 3591 } while (/*CONSTCOND*/ 0) 3592 #else 3593 #define SYN_HASH6(sa, sp, dp) \ 3594 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3595 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3596 & 0x7fffffff) 3597 3598 #define SYN_HASHALL(hash, src, dst) \ 3599 do { \ 3600 switch ((src)->sa_family) { \ 3601 case AF_INET: \ 3602 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3603 ((const struct sockaddr_in *)(src))->sin_port, \ 3604 ((const struct sockaddr_in *)(dst))->sin_port); \ 3605 break; \ 3606 case AF_INET6: \ 3607 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3608 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3609 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3610 break; \ 3611 default: \ 3612 hash = 0; \ 3613 } \ 3614 } while (/*CONSTCOND*/0) 3615 #endif /* INET6 */ 3616 3617 static struct pool syn_cache_pool; 3618 3619 /* 3620 * We don't estimate RTT with SYNs, so each packet starts with the default 3621 * RTT and each timer step has a fixed timeout value. 3622 */ 3623 #define SYN_CACHE_TIMER_ARM(sc) \ 3624 do { \ 3625 TCPT_RANGESET((sc)->sc_rxtcur, \ 3626 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3627 TCPTV_REXMTMAX); \ 3628 callout_reset(&(sc)->sc_timer, \ 3629 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3630 } while (/*CONSTCOND*/0) 3631 3632 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3633 3634 static inline void 3635 syn_cache_rm(struct syn_cache *sc) 3636 { 3637 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3638 sc, sc_bucketq); 3639 sc->sc_tp = NULL; 3640 LIST_REMOVE(sc, sc_tpq); 3641 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3642 callout_stop(&sc->sc_timer); 3643 syn_cache_count--; 3644 } 3645 3646 static inline void 3647 syn_cache_put(struct syn_cache *sc) 3648 { 3649 if (sc->sc_ipopts) 3650 (void) m_free(sc->sc_ipopts); 3651 rtcache_free(&sc->sc_route); 3652 sc->sc_flags |= SCF_DEAD; 3653 if (!callout_invoking(&sc->sc_timer)) 3654 callout_schedule(&(sc)->sc_timer, 1); 3655 } 3656 3657 void 3658 syn_cache_init(void) 3659 { 3660 int i; 3661 3662 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 3663 "synpl", NULL, IPL_SOFTNET); 3664 3665 /* Initialize the hash buckets. */ 3666 for (i = 0; i < tcp_syn_cache_size; i++) 3667 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3668 } 3669 3670 void 3671 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3672 { 3673 struct syn_cache_head *scp; 3674 struct syn_cache *sc2; 3675 int s; 3676 3677 /* 3678 * If there are no entries in the hash table, reinitialize 3679 * the hash secrets. 3680 */ 3681 if (syn_cache_count == 0) { 3682 syn_hash1 = cprng_fast32(); 3683 syn_hash2 = cprng_fast32(); 3684 } 3685 3686 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3687 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3688 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3689 3690 /* 3691 * Make sure that we don't overflow the per-bucket 3692 * limit or the total cache size limit. 3693 */ 3694 s = splsoftnet(); 3695 if (scp->sch_length >= tcp_syn_bucket_limit) { 3696 TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW); 3697 /* 3698 * The bucket is full. Toss the oldest element in the 3699 * bucket. This will be the first entry in the bucket. 3700 */ 3701 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3702 #ifdef DIAGNOSTIC 3703 /* 3704 * This should never happen; we should always find an 3705 * entry in our bucket. 3706 */ 3707 if (sc2 == NULL) 3708 panic("syn_cache_insert: bucketoverflow: impossible"); 3709 #endif 3710 syn_cache_rm(sc2); 3711 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3712 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3713 struct syn_cache_head *scp2, *sce; 3714 3715 TCP_STATINC(TCP_STAT_SC_OVERFLOWED); 3716 /* 3717 * The cache is full. Toss the oldest entry in the 3718 * first non-empty bucket we can find. 3719 * 3720 * XXX We would really like to toss the oldest 3721 * entry in the cache, but we hope that this 3722 * condition doesn't happen very often. 3723 */ 3724 scp2 = scp; 3725 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3726 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3727 for (++scp2; scp2 != scp; scp2++) { 3728 if (scp2 >= sce) 3729 scp2 = &tcp_syn_cache[0]; 3730 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3731 break; 3732 } 3733 #ifdef DIAGNOSTIC 3734 /* 3735 * This should never happen; we should always find a 3736 * non-empty bucket. 3737 */ 3738 if (scp2 == scp) 3739 panic("syn_cache_insert: cacheoverflow: " 3740 "impossible"); 3741 #endif 3742 } 3743 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3744 syn_cache_rm(sc2); 3745 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3746 } 3747 3748 /* 3749 * Initialize the entry's timer. 3750 */ 3751 sc->sc_rxttot = 0; 3752 sc->sc_rxtshift = 0; 3753 SYN_CACHE_TIMER_ARM(sc); 3754 3755 /* Link it from tcpcb entry */ 3756 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3757 3758 /* Put it into the bucket. */ 3759 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3760 scp->sch_length++; 3761 syn_cache_count++; 3762 3763 TCP_STATINC(TCP_STAT_SC_ADDED); 3764 splx(s); 3765 } 3766 3767 /* 3768 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3769 * If we have retransmitted an entry the maximum number of times, expire 3770 * that entry. 3771 */ 3772 void 3773 syn_cache_timer(void *arg) 3774 { 3775 struct syn_cache *sc = arg; 3776 3777 mutex_enter(softnet_lock); 3778 KERNEL_LOCK(1, NULL); 3779 callout_ack(&sc->sc_timer); 3780 3781 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3782 TCP_STATINC(TCP_STAT_SC_DELAYED_FREE); 3783 callout_destroy(&sc->sc_timer); 3784 pool_put(&syn_cache_pool, sc); 3785 KERNEL_UNLOCK_ONE(NULL); 3786 mutex_exit(softnet_lock); 3787 return; 3788 } 3789 3790 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3791 /* Drop it -- too many retransmissions. */ 3792 goto dropit; 3793 } 3794 3795 /* 3796 * Compute the total amount of time this entry has 3797 * been on a queue. If this entry has been on longer 3798 * than the keep alive timer would allow, expire it. 3799 */ 3800 sc->sc_rxttot += sc->sc_rxtcur; 3801 if (sc->sc_rxttot >= tcp_keepinit) 3802 goto dropit; 3803 3804 TCP_STATINC(TCP_STAT_SC_RETRANSMITTED); 3805 (void) syn_cache_respond(sc, NULL); 3806 3807 /* Advance the timer back-off. */ 3808 sc->sc_rxtshift++; 3809 SYN_CACHE_TIMER_ARM(sc); 3810 3811 KERNEL_UNLOCK_ONE(NULL); 3812 mutex_exit(softnet_lock); 3813 return; 3814 3815 dropit: 3816 TCP_STATINC(TCP_STAT_SC_TIMED_OUT); 3817 syn_cache_rm(sc); 3818 if (sc->sc_ipopts) 3819 (void) m_free(sc->sc_ipopts); 3820 rtcache_free(&sc->sc_route); 3821 callout_destroy(&sc->sc_timer); 3822 pool_put(&syn_cache_pool, sc); 3823 KERNEL_UNLOCK_ONE(NULL); 3824 mutex_exit(softnet_lock); 3825 } 3826 3827 /* 3828 * Remove syn cache created by the specified tcb entry, 3829 * because this does not make sense to keep them 3830 * (if there's no tcb entry, syn cache entry will never be used) 3831 */ 3832 void 3833 syn_cache_cleanup(struct tcpcb *tp) 3834 { 3835 struct syn_cache *sc, *nsc; 3836 int s; 3837 3838 s = splsoftnet(); 3839 3840 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3841 nsc = LIST_NEXT(sc, sc_tpq); 3842 3843 #ifdef DIAGNOSTIC 3844 if (sc->sc_tp != tp) 3845 panic("invalid sc_tp in syn_cache_cleanup"); 3846 #endif 3847 syn_cache_rm(sc); 3848 syn_cache_put(sc); /* calls pool_put but see spl above */ 3849 } 3850 /* just for safety */ 3851 LIST_INIT(&tp->t_sc); 3852 3853 splx(s); 3854 } 3855 3856 /* 3857 * Find an entry in the syn cache. 3858 */ 3859 struct syn_cache * 3860 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3861 struct syn_cache_head **headp) 3862 { 3863 struct syn_cache *sc; 3864 struct syn_cache_head *scp; 3865 u_int32_t hash; 3866 int s; 3867 3868 SYN_HASHALL(hash, src, dst); 3869 3870 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3871 *headp = scp; 3872 s = splsoftnet(); 3873 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3874 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3875 if (sc->sc_hash != hash) 3876 continue; 3877 if (!memcmp(&sc->sc_src, src, src->sa_len) && 3878 !memcmp(&sc->sc_dst, dst, dst->sa_len)) { 3879 splx(s); 3880 return (sc); 3881 } 3882 } 3883 splx(s); 3884 return (NULL); 3885 } 3886 3887 /* 3888 * This function gets called when we receive an ACK for a 3889 * socket in the LISTEN state. We look up the connection 3890 * in the syn cache, and if its there, we pull it out of 3891 * the cache and turn it into a full-blown connection in 3892 * the SYN-RECEIVED state. 3893 * 3894 * The return values may not be immediately obvious, and their effects 3895 * can be subtle, so here they are: 3896 * 3897 * NULL SYN was not found in cache; caller should drop the 3898 * packet and send an RST. 3899 * 3900 * -1 We were unable to create the new connection, and are 3901 * aborting it. An ACK,RST is being sent to the peer 3902 * (unless we got screwey sequence numbners; see below), 3903 * because the 3-way handshake has been completed. Caller 3904 * should not free the mbuf, since we may be using it. If 3905 * we are not, we will free it. 3906 * 3907 * Otherwise, the return value is a pointer to the new socket 3908 * associated with the connection. 3909 */ 3910 struct socket * 3911 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3912 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3913 struct socket *so, struct mbuf *m) 3914 { 3915 struct syn_cache *sc; 3916 struct syn_cache_head *scp; 3917 struct inpcb *inp = NULL; 3918 #ifdef INET6 3919 struct in6pcb *in6p = NULL; 3920 #endif 3921 struct tcpcb *tp = 0; 3922 struct mbuf *am; 3923 int s; 3924 struct socket *oso; 3925 3926 s = splsoftnet(); 3927 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3928 splx(s); 3929 return (NULL); 3930 } 3931 3932 /* 3933 * Verify the sequence and ack numbers. Try getting the correct 3934 * response again. 3935 */ 3936 if ((th->th_ack != sc->sc_iss + 1) || 3937 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3938 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3939 (void) syn_cache_respond(sc, m); 3940 splx(s); 3941 return ((struct socket *)(-1)); 3942 } 3943 3944 /* Remove this cache entry */ 3945 syn_cache_rm(sc); 3946 splx(s); 3947 3948 /* 3949 * Ok, create the full blown connection, and set things up 3950 * as they would have been set up if we had created the 3951 * connection when the SYN arrived. If we can't create 3952 * the connection, abort it. 3953 */ 3954 /* 3955 * inp still has the OLD in_pcb stuff, set the 3956 * v6-related flags on the new guy, too. This is 3957 * done particularly for the case where an AF_INET6 3958 * socket is bound only to a port, and a v4 connection 3959 * comes in on that port. 3960 * we also copy the flowinfo from the original pcb 3961 * to the new one. 3962 */ 3963 oso = so; 3964 so = sonewconn(so, true); 3965 if (so == NULL) 3966 goto resetandabort; 3967 3968 switch (so->so_proto->pr_domain->dom_family) { 3969 #ifdef INET 3970 case AF_INET: 3971 inp = sotoinpcb(so); 3972 break; 3973 #endif 3974 #ifdef INET6 3975 case AF_INET6: 3976 in6p = sotoin6pcb(so); 3977 break; 3978 #endif 3979 } 3980 switch (src->sa_family) { 3981 #ifdef INET 3982 case AF_INET: 3983 if (inp) { 3984 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3985 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3986 inp->inp_options = ip_srcroute(); 3987 in_pcbstate(inp, INP_BOUND); 3988 if (inp->inp_options == NULL) { 3989 inp->inp_options = sc->sc_ipopts; 3990 sc->sc_ipopts = NULL; 3991 } 3992 } 3993 #ifdef INET6 3994 else if (in6p) { 3995 /* IPv4 packet to AF_INET6 socket */ 3996 memset(&in6p->in6p_laddr, 0, sizeof(in6p->in6p_laddr)); 3997 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3998 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3999 &in6p->in6p_laddr.s6_addr32[3], 4000 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 4001 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 4002 in6totcpcb(in6p)->t_family = AF_INET; 4003 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 4004 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 4005 else 4006 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 4007 in6_pcbstate(in6p, IN6P_BOUND); 4008 } 4009 #endif 4010 break; 4011 #endif 4012 #ifdef INET6 4013 case AF_INET6: 4014 if (in6p) { 4015 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 4016 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 4017 in6_pcbstate(in6p, IN6P_BOUND); 4018 } 4019 break; 4020 #endif 4021 } 4022 #ifdef INET6 4023 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 4024 struct in6pcb *oin6p = sotoin6pcb(oso); 4025 /* inherit socket options from the listening socket */ 4026 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 4027 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 4028 m_freem(in6p->in6p_options); 4029 in6p->in6p_options = 0; 4030 } 4031 ip6_savecontrol(in6p, &in6p->in6p_options, 4032 mtod(m, struct ip6_hdr *), m); 4033 } 4034 #endif 4035 4036 #if defined(IPSEC) 4037 if (ipsec_used) { 4038 /* 4039 * we make a copy of policy, instead of sharing the policy, for 4040 * better behavior in terms of SA lookup and dead SA removal. 4041 */ 4042 if (inp) { 4043 /* copy old policy into new socket's */ 4044 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, 4045 inp->inp_sp)) 4046 printf("tcp_input: could not copy policy\n"); 4047 } 4048 #ifdef INET6 4049 else if (in6p) { 4050 /* copy old policy into new socket's */ 4051 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 4052 in6p->in6p_sp)) 4053 printf("tcp_input: could not copy policy\n"); 4054 } 4055 #endif 4056 } 4057 #endif 4058 4059 /* 4060 * Give the new socket our cached route reference. 4061 */ 4062 if (inp) { 4063 rtcache_copy(&inp->inp_route, &sc->sc_route); 4064 rtcache_free(&sc->sc_route); 4065 } 4066 #ifdef INET6 4067 else { 4068 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 4069 rtcache_free(&sc->sc_route); 4070 } 4071 #endif 4072 4073 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 4074 if (am == NULL) 4075 goto resetandabort; 4076 MCLAIM(am, &tcp_mowner); 4077 am->m_len = src->sa_len; 4078 bcopy(src, mtod(am, void *), src->sa_len); 4079 if (inp) { 4080 if (in_pcbconnect(inp, am, &lwp0)) { 4081 (void) m_free(am); 4082 goto resetandabort; 4083 } 4084 } 4085 #ifdef INET6 4086 else if (in6p) { 4087 if (src->sa_family == AF_INET) { 4088 /* IPv4 packet to AF_INET6 socket */ 4089 struct sockaddr_in6 *sin6; 4090 sin6 = mtod(am, struct sockaddr_in6 *); 4091 am->m_len = sizeof(*sin6); 4092 memset(sin6, 0, sizeof(*sin6)); 4093 sin6->sin6_family = AF_INET6; 4094 sin6->sin6_len = sizeof(*sin6); 4095 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 4096 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 4097 bcopy(&((struct sockaddr_in *)src)->sin_addr, 4098 &sin6->sin6_addr.s6_addr32[3], 4099 sizeof(sin6->sin6_addr.s6_addr32[3])); 4100 } 4101 if (in6_pcbconnect(in6p, am, NULL)) { 4102 (void) m_free(am); 4103 goto resetandabort; 4104 } 4105 } 4106 #endif 4107 else { 4108 (void) m_free(am); 4109 goto resetandabort; 4110 } 4111 (void) m_free(am); 4112 4113 if (inp) 4114 tp = intotcpcb(inp); 4115 #ifdef INET6 4116 else if (in6p) 4117 tp = in6totcpcb(in6p); 4118 #endif 4119 else 4120 tp = NULL; 4121 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 4122 if (sc->sc_request_r_scale != 15) { 4123 tp->requested_s_scale = sc->sc_requested_s_scale; 4124 tp->request_r_scale = sc->sc_request_r_scale; 4125 tp->snd_scale = sc->sc_requested_s_scale; 4126 tp->rcv_scale = sc->sc_request_r_scale; 4127 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 4128 } 4129 if (sc->sc_flags & SCF_TIMESTAMP) 4130 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 4131 tp->ts_timebase = sc->sc_timebase; 4132 4133 tp->t_template = tcp_template(tp); 4134 if (tp->t_template == 0) { 4135 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 4136 so = NULL; 4137 m_freem(m); 4138 goto abort; 4139 } 4140 4141 tp->iss = sc->sc_iss; 4142 tp->irs = sc->sc_irs; 4143 tcp_sendseqinit(tp); 4144 tcp_rcvseqinit(tp); 4145 tp->t_state = TCPS_SYN_RECEIVED; 4146 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 4147 TCP_STATINC(TCP_STAT_ACCEPTS); 4148 4149 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 4150 tp->t_flags |= TF_WILL_SACK; 4151 4152 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 4153 tp->t_flags |= TF_ECN_PERMIT; 4154 4155 #ifdef TCP_SIGNATURE 4156 if (sc->sc_flags & SCF_SIGNATURE) 4157 tp->t_flags |= TF_SIGNATURE; 4158 #endif 4159 4160 /* Initialize tp->t_ourmss before we deal with the peer's! */ 4161 tp->t_ourmss = sc->sc_ourmaxseg; 4162 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 4163 4164 /* 4165 * Initialize the initial congestion window. If we 4166 * had to retransmit the SYN,ACK, we must initialize cwnd 4167 * to 1 segment (i.e. the Loss Window). 4168 */ 4169 if (sc->sc_rxtshift) 4170 tp->snd_cwnd = tp->t_peermss; 4171 else { 4172 int ss = tcp_init_win; 4173 #ifdef INET 4174 if (inp != NULL && in_localaddr(inp->inp_faddr)) 4175 ss = tcp_init_win_local; 4176 #endif 4177 #ifdef INET6 4178 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 4179 ss = tcp_init_win_local; 4180 #endif 4181 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 4182 } 4183 4184 tcp_rmx_rtt(tp); 4185 tp->snd_wl1 = sc->sc_irs; 4186 tp->rcv_up = sc->sc_irs + 1; 4187 4188 /* 4189 * This is what whould have happened in tcp_output() when 4190 * the SYN,ACK was sent. 4191 */ 4192 tp->snd_up = tp->snd_una; 4193 tp->snd_max = tp->snd_nxt = tp->iss+1; 4194 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 4195 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 4196 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 4197 tp->last_ack_sent = tp->rcv_nxt; 4198 tp->t_partialacks = -1; 4199 tp->t_dupacks = 0; 4200 4201 TCP_STATINC(TCP_STAT_SC_COMPLETED); 4202 s = splsoftnet(); 4203 syn_cache_put(sc); 4204 splx(s); 4205 return (so); 4206 4207 resetandabort: 4208 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 4209 abort: 4210 if (so != NULL) { 4211 (void) soqremque(so, 1); 4212 (void) soabort(so); 4213 mutex_enter(softnet_lock); 4214 } 4215 s = splsoftnet(); 4216 syn_cache_put(sc); 4217 splx(s); 4218 TCP_STATINC(TCP_STAT_SC_ABORTED); 4219 return ((struct socket *)(-1)); 4220 } 4221 4222 /* 4223 * This function is called when we get a RST for a 4224 * non-existent connection, so that we can see if the 4225 * connection is in the syn cache. If it is, zap it. 4226 */ 4227 4228 void 4229 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 4230 { 4231 struct syn_cache *sc; 4232 struct syn_cache_head *scp; 4233 int s = splsoftnet(); 4234 4235 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4236 splx(s); 4237 return; 4238 } 4239 if (SEQ_LT(th->th_seq, sc->sc_irs) || 4240 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 4241 splx(s); 4242 return; 4243 } 4244 syn_cache_rm(sc); 4245 TCP_STATINC(TCP_STAT_SC_RESET); 4246 syn_cache_put(sc); /* calls pool_put but see spl above */ 4247 splx(s); 4248 } 4249 4250 void 4251 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 4252 struct tcphdr *th) 4253 { 4254 struct syn_cache *sc; 4255 struct syn_cache_head *scp; 4256 int s; 4257 4258 s = splsoftnet(); 4259 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4260 splx(s); 4261 return; 4262 } 4263 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 4264 if (ntohl (th->th_seq) != sc->sc_iss) { 4265 splx(s); 4266 return; 4267 } 4268 4269 /* 4270 * If we've retransmitted 3 times and this is our second error, 4271 * we remove the entry. Otherwise, we allow it to continue on. 4272 * This prevents us from incorrectly nuking an entry during a 4273 * spurious network outage. 4274 * 4275 * See tcp_notify(). 4276 */ 4277 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 4278 sc->sc_flags |= SCF_UNREACH; 4279 splx(s); 4280 return; 4281 } 4282 4283 syn_cache_rm(sc); 4284 TCP_STATINC(TCP_STAT_SC_UNREACH); 4285 syn_cache_put(sc); /* calls pool_put but see spl above */ 4286 splx(s); 4287 } 4288 4289 /* 4290 * Given a LISTEN socket and an inbound SYN request, add 4291 * this to the syn cache, and send back a segment: 4292 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 4293 * to the source. 4294 * 4295 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 4296 * Doing so would require that we hold onto the data and deliver it 4297 * to the application. However, if we are the target of a SYN-flood 4298 * DoS attack, an attacker could send data which would eventually 4299 * consume all available buffer space if it were ACKed. By not ACKing 4300 * the data, we avoid this DoS scenario. 4301 */ 4302 4303 int 4304 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 4305 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 4306 int optlen, struct tcp_opt_info *oi) 4307 { 4308 struct tcpcb tb, *tp; 4309 long win; 4310 struct syn_cache *sc; 4311 struct syn_cache_head *scp; 4312 struct mbuf *ipopts; 4313 struct tcp_opt_info opti; 4314 int s; 4315 4316 tp = sototcpcb(so); 4317 4318 memset(&opti, 0, sizeof(opti)); 4319 4320 /* 4321 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 4322 * 4323 * Note this check is performed in tcp_input() very early on. 4324 */ 4325 4326 /* 4327 * Initialize some local state. 4328 */ 4329 win = sbspace(&so->so_rcv); 4330 if (win > TCP_MAXWIN) 4331 win = TCP_MAXWIN; 4332 4333 switch (src->sa_family) { 4334 #ifdef INET 4335 case AF_INET: 4336 /* 4337 * Remember the IP options, if any. 4338 */ 4339 ipopts = ip_srcroute(); 4340 break; 4341 #endif 4342 default: 4343 ipopts = NULL; 4344 } 4345 4346 #ifdef TCP_SIGNATURE 4347 if (optp || (tp->t_flags & TF_SIGNATURE)) 4348 #else 4349 if (optp) 4350 #endif 4351 { 4352 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4353 #ifdef TCP_SIGNATURE 4354 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4355 #endif 4356 tb.t_state = TCPS_LISTEN; 4357 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 4358 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 4359 return (0); 4360 } else 4361 tb.t_flags = 0; 4362 4363 /* 4364 * See if we already have an entry for this connection. 4365 * If we do, resend the SYN,ACK. We do not count this 4366 * as a retransmission (XXX though maybe we should). 4367 */ 4368 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4369 TCP_STATINC(TCP_STAT_SC_DUPESYN); 4370 if (ipopts) { 4371 /* 4372 * If we were remembering a previous source route, 4373 * forget it and use the new one we've been given. 4374 */ 4375 if (sc->sc_ipopts) 4376 (void) m_free(sc->sc_ipopts); 4377 sc->sc_ipopts = ipopts; 4378 } 4379 sc->sc_timestamp = tb.ts_recent; 4380 if (syn_cache_respond(sc, m) == 0) { 4381 uint64_t *tcps = TCP_STAT_GETREF(); 4382 tcps[TCP_STAT_SNDACKS]++; 4383 tcps[TCP_STAT_SNDTOTAL]++; 4384 TCP_STAT_PUTREF(); 4385 } 4386 return (1); 4387 } 4388 4389 s = splsoftnet(); 4390 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4391 splx(s); 4392 if (sc == NULL) { 4393 if (ipopts) 4394 (void) m_free(ipopts); 4395 return (0); 4396 } 4397 4398 /* 4399 * Fill in the cache, and put the necessary IP and TCP 4400 * options into the reply. 4401 */ 4402 memset(sc, 0, sizeof(struct syn_cache)); 4403 callout_init(&sc->sc_timer, CALLOUT_MPSAFE); 4404 bcopy(src, &sc->sc_src, src->sa_len); 4405 bcopy(dst, &sc->sc_dst, dst->sa_len); 4406 sc->sc_flags = 0; 4407 sc->sc_ipopts = ipopts; 4408 sc->sc_irs = th->th_seq; 4409 switch (src->sa_family) { 4410 #ifdef INET 4411 case AF_INET: 4412 { 4413 struct sockaddr_in *srcin = (void *) src; 4414 struct sockaddr_in *dstin = (void *) dst; 4415 4416 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4417 &srcin->sin_addr, dstin->sin_port, 4418 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4419 break; 4420 } 4421 #endif /* INET */ 4422 #ifdef INET6 4423 case AF_INET6: 4424 { 4425 struct sockaddr_in6 *srcin6 = (void *) src; 4426 struct sockaddr_in6 *dstin6 = (void *) dst; 4427 4428 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4429 &srcin6->sin6_addr, dstin6->sin6_port, 4430 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4431 break; 4432 } 4433 #endif /* INET6 */ 4434 } 4435 sc->sc_peermaxseg = oi->maxseg; 4436 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4437 m->m_pkthdr.rcvif : NULL, 4438 sc->sc_src.sa.sa_family); 4439 sc->sc_win = win; 4440 sc->sc_timebase = tcp_now - 1; /* see tcp_newtcpcb() */ 4441 sc->sc_timestamp = tb.ts_recent; 4442 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4443 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4444 sc->sc_flags |= SCF_TIMESTAMP; 4445 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4446 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4447 sc->sc_requested_s_scale = tb.requested_s_scale; 4448 sc->sc_request_r_scale = 0; 4449 /* 4450 * Pick the smallest possible scaling factor that 4451 * will still allow us to scale up to sb_max. 4452 * 4453 * We do this because there are broken firewalls that 4454 * will corrupt the window scale option, leading to 4455 * the other endpoint believing that our advertised 4456 * window is unscaled. At scale factors larger than 4457 * 5 the unscaled window will drop below 1500 bytes, 4458 * leading to serious problems when traversing these 4459 * broken firewalls. 4460 * 4461 * With the default sbmax of 256K, a scale factor 4462 * of 3 will be chosen by this algorithm. Those who 4463 * choose a larger sbmax should watch out 4464 * for the compatiblity problems mentioned above. 4465 * 4466 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4467 * or <SYN,ACK>) segment itself is never scaled. 4468 */ 4469 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4470 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4471 sc->sc_request_r_scale++; 4472 } else { 4473 sc->sc_requested_s_scale = 15; 4474 sc->sc_request_r_scale = 15; 4475 } 4476 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4477 sc->sc_flags |= SCF_SACK_PERMIT; 4478 4479 /* 4480 * ECN setup packet recieved. 4481 */ 4482 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4483 sc->sc_flags |= SCF_ECN_PERMIT; 4484 4485 #ifdef TCP_SIGNATURE 4486 if (tb.t_flags & TF_SIGNATURE) 4487 sc->sc_flags |= SCF_SIGNATURE; 4488 #endif 4489 sc->sc_tp = tp; 4490 if (syn_cache_respond(sc, m) == 0) { 4491 uint64_t *tcps = TCP_STAT_GETREF(); 4492 tcps[TCP_STAT_SNDACKS]++; 4493 tcps[TCP_STAT_SNDTOTAL]++; 4494 TCP_STAT_PUTREF(); 4495 syn_cache_insert(sc, tp); 4496 } else { 4497 s = splsoftnet(); 4498 /* 4499 * syn_cache_put() will try to schedule the timer, so 4500 * we need to initialize it 4501 */ 4502 SYN_CACHE_TIMER_ARM(sc); 4503 syn_cache_put(sc); 4504 splx(s); 4505 TCP_STATINC(TCP_STAT_SC_DROPPED); 4506 } 4507 return (1); 4508 } 4509 4510 /* 4511 * syn_cache_respond: (re)send SYN+ACK. 4512 * 4513 * returns 0 on success. otherwise returns an errno, typically ENOBUFS. 4514 */ 4515 4516 int 4517 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4518 { 4519 #ifdef INET6 4520 struct rtentry *rt; 4521 #endif 4522 struct route *ro; 4523 u_int8_t *optp; 4524 int optlen, error; 4525 u_int16_t tlen; 4526 struct ip *ip = NULL; 4527 #ifdef INET6 4528 struct ip6_hdr *ip6 = NULL; 4529 #endif 4530 struct tcpcb *tp = NULL; 4531 struct tcphdr *th; 4532 u_int hlen; 4533 struct socket *so; 4534 4535 ro = &sc->sc_route; 4536 switch (sc->sc_src.sa.sa_family) { 4537 case AF_INET: 4538 hlen = sizeof(struct ip); 4539 break; 4540 #ifdef INET6 4541 case AF_INET6: 4542 hlen = sizeof(struct ip6_hdr); 4543 break; 4544 #endif 4545 default: 4546 if (m) 4547 m_freem(m); 4548 return (EAFNOSUPPORT); 4549 } 4550 4551 /* Compute the size of the TCP options. */ 4552 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4553 ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) + 4554 #ifdef TCP_SIGNATURE 4555 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4556 #endif 4557 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4558 4559 tlen = hlen + sizeof(struct tcphdr) + optlen; 4560 4561 /* 4562 * Create the IP+TCP header from scratch. 4563 */ 4564 if (m) 4565 m_freem(m); 4566 #ifdef DIAGNOSTIC 4567 if (max_linkhdr + tlen > MCLBYTES) 4568 return (ENOBUFS); 4569 #endif 4570 MGETHDR(m, M_DONTWAIT, MT_DATA); 4571 if (m && (max_linkhdr + tlen) > MHLEN) { 4572 MCLGET(m, M_DONTWAIT); 4573 if ((m->m_flags & M_EXT) == 0) { 4574 m_freem(m); 4575 m = NULL; 4576 } 4577 } 4578 if (m == NULL) 4579 return (ENOBUFS); 4580 MCLAIM(m, &tcp_tx_mowner); 4581 4582 /* Fixup the mbuf. */ 4583 m->m_data += max_linkhdr; 4584 m->m_len = m->m_pkthdr.len = tlen; 4585 if (sc->sc_tp) { 4586 tp = sc->sc_tp; 4587 if (tp->t_inpcb) 4588 so = tp->t_inpcb->inp_socket; 4589 #ifdef INET6 4590 else if (tp->t_in6pcb) 4591 so = tp->t_in6pcb->in6p_socket; 4592 #endif 4593 else 4594 so = NULL; 4595 } else 4596 so = NULL; 4597 m->m_pkthdr.rcvif = NULL; 4598 memset(mtod(m, u_char *), 0, tlen); 4599 4600 switch (sc->sc_src.sa.sa_family) { 4601 case AF_INET: 4602 ip = mtod(m, struct ip *); 4603 ip->ip_v = 4; 4604 ip->ip_dst = sc->sc_src.sin.sin_addr; 4605 ip->ip_src = sc->sc_dst.sin.sin_addr; 4606 ip->ip_p = IPPROTO_TCP; 4607 th = (struct tcphdr *)(ip + 1); 4608 th->th_dport = sc->sc_src.sin.sin_port; 4609 th->th_sport = sc->sc_dst.sin.sin_port; 4610 break; 4611 #ifdef INET6 4612 case AF_INET6: 4613 ip6 = mtod(m, struct ip6_hdr *); 4614 ip6->ip6_vfc = IPV6_VERSION; 4615 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4616 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4617 ip6->ip6_nxt = IPPROTO_TCP; 4618 /* ip6_plen will be updated in ip6_output() */ 4619 th = (struct tcphdr *)(ip6 + 1); 4620 th->th_dport = sc->sc_src.sin6.sin6_port; 4621 th->th_sport = sc->sc_dst.sin6.sin6_port; 4622 break; 4623 #endif 4624 default: 4625 th = NULL; 4626 } 4627 4628 th->th_seq = htonl(sc->sc_iss); 4629 th->th_ack = htonl(sc->sc_irs + 1); 4630 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4631 th->th_flags = TH_SYN|TH_ACK; 4632 th->th_win = htons(sc->sc_win); 4633 /* th_sum already 0 */ 4634 /* th_urp already 0 */ 4635 4636 /* Tack on the TCP options. */ 4637 optp = (u_int8_t *)(th + 1); 4638 *optp++ = TCPOPT_MAXSEG; 4639 *optp++ = 4; 4640 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4641 *optp++ = sc->sc_ourmaxseg & 0xff; 4642 4643 if (sc->sc_request_r_scale != 15) { 4644 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4645 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4646 sc->sc_request_r_scale); 4647 optp += 4; 4648 } 4649 4650 if (sc->sc_flags & SCF_TIMESTAMP) { 4651 u_int32_t *lp = (u_int32_t *)(optp); 4652 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4653 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4654 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4655 *lp = htonl(sc->sc_timestamp); 4656 optp += TCPOLEN_TSTAMP_APPA; 4657 } 4658 4659 if (sc->sc_flags & SCF_SACK_PERMIT) { 4660 u_int8_t *p = optp; 4661 4662 /* Let the peer know that we will SACK. */ 4663 p[0] = TCPOPT_SACK_PERMITTED; 4664 p[1] = 2; 4665 p[2] = TCPOPT_NOP; 4666 p[3] = TCPOPT_NOP; 4667 optp += 4; 4668 } 4669 4670 /* 4671 * Send ECN SYN-ACK setup packet. 4672 * Routes can be asymetric, so, even if we receive a packet 4673 * with ECE and CWR set, we must not assume no one will block 4674 * the ECE packet we are about to send. 4675 */ 4676 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4677 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4678 th->th_flags |= TH_ECE; 4679 TCP_STATINC(TCP_STAT_ECN_SHS); 4680 4681 /* 4682 * draft-ietf-tcpm-ecnsyn-00.txt 4683 * 4684 * "[...] a TCP node MAY respond to an ECN-setup 4685 * SYN packet by setting ECT in the responding 4686 * ECN-setup SYN/ACK packet, indicating to routers 4687 * that the SYN/ACK packet is ECN-Capable. 4688 * This allows a congested router along the path 4689 * to mark the packet instead of dropping the 4690 * packet as an indication of congestion." 4691 * 4692 * "[...] There can be a great benefit in setting 4693 * an ECN-capable codepoint in SYN/ACK packets [...] 4694 * Congestion is most likely to occur in 4695 * the server-to-client direction. As a result, 4696 * setting an ECN-capable codepoint in SYN/ACK 4697 * packets can reduce the occurence of three-second 4698 * retransmit timeouts resulting from the drop 4699 * of SYN/ACK packets." 4700 * 4701 * Page 4 and 6, January 2006. 4702 */ 4703 4704 switch (sc->sc_src.sa.sa_family) { 4705 #ifdef INET 4706 case AF_INET: 4707 ip->ip_tos |= IPTOS_ECN_ECT0; 4708 break; 4709 #endif 4710 #ifdef INET6 4711 case AF_INET6: 4712 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4713 break; 4714 #endif 4715 } 4716 TCP_STATINC(TCP_STAT_ECN_ECT); 4717 } 4718 4719 #ifdef TCP_SIGNATURE 4720 if (sc->sc_flags & SCF_SIGNATURE) { 4721 struct secasvar *sav; 4722 u_int8_t *sigp; 4723 4724 sav = tcp_signature_getsav(m, th); 4725 4726 if (sav == NULL) { 4727 if (m) 4728 m_freem(m); 4729 return (EPERM); 4730 } 4731 4732 *optp++ = TCPOPT_SIGNATURE; 4733 *optp++ = TCPOLEN_SIGNATURE; 4734 sigp = optp; 4735 memset(optp, 0, TCP_SIGLEN); 4736 optp += TCP_SIGLEN; 4737 *optp++ = TCPOPT_NOP; 4738 *optp++ = TCPOPT_EOL; 4739 4740 (void)tcp_signature(m, th, hlen, sav, sigp); 4741 4742 key_sa_recordxfer(sav, m); 4743 KEY_FREESAV(&sav); 4744 } 4745 #endif 4746 4747 /* Compute the packet's checksum. */ 4748 switch (sc->sc_src.sa.sa_family) { 4749 case AF_INET: 4750 ip->ip_len = htons(tlen - hlen); 4751 th->th_sum = 0; 4752 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4753 break; 4754 #ifdef INET6 4755 case AF_INET6: 4756 ip6->ip6_plen = htons(tlen - hlen); 4757 th->th_sum = 0; 4758 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4759 break; 4760 #endif 4761 } 4762 4763 /* 4764 * Fill in some straggling IP bits. Note the stack expects 4765 * ip_len to be in host order, for convenience. 4766 */ 4767 switch (sc->sc_src.sa.sa_family) { 4768 #ifdef INET 4769 case AF_INET: 4770 ip->ip_len = htons(tlen); 4771 ip->ip_ttl = ip_defttl; 4772 /* XXX tos? */ 4773 break; 4774 #endif 4775 #ifdef INET6 4776 case AF_INET6: 4777 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4778 ip6->ip6_vfc |= IPV6_VERSION; 4779 ip6->ip6_plen = htons(tlen - hlen); 4780 /* ip6_hlim will be initialized afterwards */ 4781 /* XXX flowlabel? */ 4782 break; 4783 #endif 4784 } 4785 4786 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4787 tp = sc->sc_tp; 4788 4789 switch (sc->sc_src.sa.sa_family) { 4790 #ifdef INET 4791 case AF_INET: 4792 error = ip_output(m, sc->sc_ipopts, ro, 4793 (ip_mtudisc ? IP_MTUDISC : 0), 4794 NULL, so); 4795 break; 4796 #endif 4797 #ifdef INET6 4798 case AF_INET6: 4799 ip6->ip6_hlim = in6_selecthlim(NULL, 4800 (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp 4801 : NULL); 4802 4803 error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, so, NULL); 4804 break; 4805 #endif 4806 default: 4807 error = EAFNOSUPPORT; 4808 break; 4809 } 4810 return (error); 4811 } 4812