1 /* $NetBSD: tcp_input.c,v 1.359 2017/07/07 01:37:34 ozaki-r 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 1.359 2017/07/07 01:37:34 ozaki-r Exp $"); 152 153 #ifdef _KERNEL_OPT 154 #include "opt_inet.h" 155 #include "opt_ipsec.h" 156 #include "opt_inet_csum.h" 157 #include "opt_tcp_debug.h" 158 #endif 159 160 #include <sys/param.h> 161 #include <sys/systm.h> 162 #include <sys/malloc.h> 163 #include <sys/mbuf.h> 164 #include <sys/protosw.h> 165 #include <sys/socket.h> 166 #include <sys/socketvar.h> 167 #include <sys/errno.h> 168 #include <sys/syslog.h> 169 #include <sys/pool.h> 170 #include <sys/domain.h> 171 #include <sys/kernel.h> 172 #ifdef TCP_SIGNATURE 173 #include <sys/md5.h> 174 #endif 175 #include <sys/lwp.h> /* for lwp0 */ 176 #include <sys/cprng.h> 177 178 #include <net/if.h> 179 #include <net/if_types.h> 180 181 #include <netinet/in.h> 182 #include <netinet/in_systm.h> 183 #include <netinet/ip.h> 184 #include <netinet/in_pcb.h> 185 #include <netinet/in_var.h> 186 #include <netinet/ip_var.h> 187 #include <netinet/in_offload.h> 188 189 #ifdef INET6 190 #ifndef INET 191 #include <netinet/in.h> 192 #endif 193 #include <netinet/ip6.h> 194 #include <netinet6/ip6_var.h> 195 #include <netinet6/in6_pcb.h> 196 #include <netinet6/ip6_var.h> 197 #include <netinet6/in6_var.h> 198 #include <netinet/icmp6.h> 199 #include <netinet6/nd6.h> 200 #ifdef TCP_SIGNATURE 201 #include <netinet6/scope6_var.h> 202 #endif 203 #endif 204 205 #ifndef INET6 206 /* always need ip6.h for IP6_EXTHDR_GET */ 207 #include <netinet/ip6.h> 208 #endif 209 210 #include <netinet/tcp.h> 211 #include <netinet/tcp_fsm.h> 212 #include <netinet/tcp_seq.h> 213 #include <netinet/tcp_timer.h> 214 #include <netinet/tcp_var.h> 215 #include <netinet/tcp_private.h> 216 #include <netinet/tcpip.h> 217 #include <netinet/tcp_congctl.h> 218 #include <netinet/tcp_debug.h> 219 220 #ifdef INET6 221 #include "faith.h" 222 #if defined(NFAITH) && NFAITH > 0 223 #include <net/if_faith.h> 224 #endif 225 #endif /* INET6 */ 226 227 #ifdef IPSEC 228 #include <netipsec/ipsec.h> 229 #include <netipsec/ipsec_var.h> 230 #include <netipsec/ipsec_private.h> 231 #include <netipsec/key.h> 232 #ifdef INET6 233 #include <netipsec/ipsec6.h> 234 #endif 235 #endif /* IPSEC*/ 236 237 #include <netinet/tcp_vtw.h> 238 239 int tcprexmtthresh = 3; 240 int tcp_log_refused; 241 242 int tcp_do_autorcvbuf = 1; 243 int tcp_autorcvbuf_inc = 16 * 1024; 244 int tcp_autorcvbuf_max = 256 * 1024; 245 int tcp_msl = (TCPTV_MSL / PR_SLOWHZ); 246 247 static int tcp_rst_ppslim_count = 0; 248 static struct timeval tcp_rst_ppslim_last; 249 static int tcp_ackdrop_ppslim_count = 0; 250 static struct timeval tcp_ackdrop_ppslim_last; 251 252 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 253 254 /* for modulo comparisons of timestamps */ 255 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 256 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 257 258 /* 259 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 260 */ 261 #ifdef INET6 262 static inline void 263 nd6_hint(struct tcpcb *tp) 264 { 265 struct rtentry *rt = NULL; 266 267 if (tp != NULL && tp->t_in6pcb != NULL && tp->t_family == AF_INET6 && 268 (rt = rtcache_validate(&tp->t_in6pcb->in6p_route)) != NULL) 269 nd6_nud_hint(rt); 270 rtcache_unref(rt, &tp->t_in6pcb->in6p_route); 271 } 272 #else 273 static inline void 274 nd6_hint(struct tcpcb *tp) 275 { 276 } 277 #endif 278 279 /* 280 * Compute ACK transmission behavior. Delay the ACK unless 281 * we have already delayed an ACK (must send an ACK every two segments). 282 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 283 * option is enabled. 284 */ 285 static void 286 tcp_setup_ack(struct tcpcb *tp, const struct tcphdr *th) 287 { 288 289 if (tp->t_flags & TF_DELACK || 290 (tcp_ack_on_push && th->th_flags & TH_PUSH)) 291 tp->t_flags |= TF_ACKNOW; 292 else 293 TCP_SET_DELACK(tp); 294 } 295 296 static void 297 icmp_check(struct tcpcb *tp, const struct tcphdr *th, int acked) 298 { 299 300 /* 301 * If we had a pending ICMP message that refers to data that have 302 * just been acknowledged, disregard the recorded ICMP message. 303 */ 304 if ((tp->t_flags & TF_PMTUD_PEND) && 305 SEQ_GT(th->th_ack, tp->t_pmtud_th_seq)) 306 tp->t_flags &= ~TF_PMTUD_PEND; 307 308 /* 309 * Keep track of the largest chunk of data 310 * acknowledged since last PMTU update 311 */ 312 if (tp->t_pmtud_mss_acked < acked) 313 tp->t_pmtud_mss_acked = acked; 314 } 315 316 /* 317 * Convert TCP protocol fields to host order for easier processing. 318 */ 319 static void 320 tcp_fields_to_host(struct tcphdr *th) 321 { 322 323 NTOHL(th->th_seq); 324 NTOHL(th->th_ack); 325 NTOHS(th->th_win); 326 NTOHS(th->th_urp); 327 } 328 329 /* 330 * ... and reverse the above. 331 */ 332 static void 333 tcp_fields_to_net(struct tcphdr *th) 334 { 335 336 HTONL(th->th_seq); 337 HTONL(th->th_ack); 338 HTONS(th->th_win); 339 HTONS(th->th_urp); 340 } 341 342 #ifdef TCP_CSUM_COUNTERS 343 #include <sys/device.h> 344 345 #if defined(INET) 346 extern struct evcnt tcp_hwcsum_ok; 347 extern struct evcnt tcp_hwcsum_bad; 348 extern struct evcnt tcp_hwcsum_data; 349 extern struct evcnt tcp_swcsum; 350 #endif /* defined(INET) */ 351 #if defined(INET6) 352 extern struct evcnt tcp6_hwcsum_ok; 353 extern struct evcnt tcp6_hwcsum_bad; 354 extern struct evcnt tcp6_hwcsum_data; 355 extern struct evcnt tcp6_swcsum; 356 #endif /* defined(INET6) */ 357 358 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 359 360 #else 361 362 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 363 364 #endif /* TCP_CSUM_COUNTERS */ 365 366 #ifdef TCP_REASS_COUNTERS 367 #include <sys/device.h> 368 369 extern struct evcnt tcp_reass_; 370 extern struct evcnt tcp_reass_empty; 371 extern struct evcnt tcp_reass_iteration[8]; 372 extern struct evcnt tcp_reass_prependfirst; 373 extern struct evcnt tcp_reass_prepend; 374 extern struct evcnt tcp_reass_insert; 375 extern struct evcnt tcp_reass_inserttail; 376 extern struct evcnt tcp_reass_append; 377 extern struct evcnt tcp_reass_appendtail; 378 extern struct evcnt tcp_reass_overlaptail; 379 extern struct evcnt tcp_reass_overlapfront; 380 extern struct evcnt tcp_reass_segdup; 381 extern struct evcnt tcp_reass_fragdup; 382 383 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 384 385 #else 386 387 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 388 389 #endif /* TCP_REASS_COUNTERS */ 390 391 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *, 392 int *); 393 static int tcp_dooptions(struct tcpcb *, const u_char *, int, 394 struct tcphdr *, struct mbuf *, int, struct tcp_opt_info *); 395 396 #ifdef INET 397 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 398 #endif 399 #ifdef INET6 400 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 401 #endif 402 403 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 404 405 #if defined(MBUFTRACE) 406 struct mowner tcp_reass_mowner = MOWNER_INIT("tcp", "reass"); 407 #endif /* defined(MBUFTRACE) */ 408 409 static struct pool tcpipqent_pool; 410 411 void 412 tcpipqent_init(void) 413 { 414 415 pool_init(&tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", 416 NULL, IPL_VM); 417 } 418 419 struct ipqent * 420 tcpipqent_alloc(void) 421 { 422 struct ipqent *ipqe; 423 int s; 424 425 s = splvm(); 426 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 427 splx(s); 428 429 return ipqe; 430 } 431 432 void 433 tcpipqent_free(struct ipqent *ipqe) 434 { 435 int s; 436 437 s = splvm(); 438 pool_put(&tcpipqent_pool, ipqe); 439 splx(s); 440 } 441 442 static int 443 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int *tlen) 444 { 445 struct ipqent *p, *q, *nq, *tiqe = NULL; 446 struct socket *so = NULL; 447 int pkt_flags; 448 tcp_seq pkt_seq; 449 unsigned pkt_len; 450 u_long rcvpartdupbyte = 0; 451 u_long rcvoobyte; 452 #ifdef TCP_REASS_COUNTERS 453 u_int count = 0; 454 #endif 455 uint64_t *tcps; 456 457 if (tp->t_inpcb) 458 so = tp->t_inpcb->inp_socket; 459 #ifdef INET6 460 else if (tp->t_in6pcb) 461 so = tp->t_in6pcb->in6p_socket; 462 #endif 463 464 TCP_REASS_LOCK_CHECK(tp); 465 466 /* 467 * Call with th==0 after become established to 468 * force pre-ESTABLISHED data up to user socket. 469 */ 470 if (th == 0) 471 goto present; 472 473 m_claimm(m, &tcp_reass_mowner); 474 475 rcvoobyte = *tlen; 476 /* 477 * Copy these to local variables because the tcpiphdr 478 * gets munged while we are collapsing mbufs. 479 */ 480 pkt_seq = th->th_seq; 481 pkt_len = *tlen; 482 pkt_flags = th->th_flags; 483 484 TCP_REASS_COUNTER_INCR(&tcp_reass_); 485 486 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 487 /* 488 * When we miss a packet, the vast majority of time we get 489 * packets that follow it in order. So optimize for that. 490 */ 491 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 492 p->ipqe_len += pkt_len; 493 p->ipqe_flags |= pkt_flags; 494 m_cat(p->ipre_mlast, m); 495 TRAVERSE(p->ipre_mlast); 496 m = NULL; 497 tiqe = p; 498 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 499 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 500 goto skip_replacement; 501 } 502 /* 503 * While we're here, if the pkt is completely beyond 504 * anything we have, just insert it at the tail. 505 */ 506 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 507 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 508 goto insert_it; 509 } 510 } 511 512 q = TAILQ_FIRST(&tp->segq); 513 514 if (q != NULL) { 515 /* 516 * If this segment immediately precedes the first out-of-order 517 * block, simply slap the segment in front of it and (mostly) 518 * skip the complicated logic. 519 */ 520 if (pkt_seq + pkt_len == q->ipqe_seq) { 521 q->ipqe_seq = pkt_seq; 522 q->ipqe_len += pkt_len; 523 q->ipqe_flags |= pkt_flags; 524 m_cat(m, q->ipqe_m); 525 q->ipqe_m = m; 526 q->ipre_mlast = m; /* last mbuf may have changed */ 527 TRAVERSE(q->ipre_mlast); 528 tiqe = q; 529 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 530 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 531 goto skip_replacement; 532 } 533 } else { 534 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 535 } 536 537 /* 538 * Find a segment which begins after this one does. 539 */ 540 for (p = NULL; q != NULL; q = nq) { 541 nq = TAILQ_NEXT(q, ipqe_q); 542 #ifdef TCP_REASS_COUNTERS 543 count++; 544 #endif 545 /* 546 * If the received segment is just right after this 547 * fragment, merge the two together and then check 548 * for further overlaps. 549 */ 550 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 551 #ifdef TCPREASS_DEBUG 552 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 553 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 554 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 555 #endif 556 pkt_len += q->ipqe_len; 557 pkt_flags |= q->ipqe_flags; 558 pkt_seq = q->ipqe_seq; 559 m_cat(q->ipre_mlast, m); 560 TRAVERSE(q->ipre_mlast); 561 m = q->ipqe_m; 562 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 563 goto free_ipqe; 564 } 565 /* 566 * If the received segment is completely past this 567 * fragment, we need to go the next fragment. 568 */ 569 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 570 p = q; 571 continue; 572 } 573 /* 574 * If the fragment is past the received segment, 575 * it (or any following) can't be concatenated. 576 */ 577 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 578 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 579 break; 580 } 581 582 /* 583 * We've received all the data in this segment before. 584 * mark it as a duplicate and return. 585 */ 586 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 587 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 588 tcps = TCP_STAT_GETREF(); 589 tcps[TCP_STAT_RCVDUPPACK]++; 590 tcps[TCP_STAT_RCVDUPBYTE] += pkt_len; 591 TCP_STAT_PUTREF(); 592 tcp_new_dsack(tp, pkt_seq, pkt_len); 593 m_freem(m); 594 if (tiqe != NULL) { 595 tcpipqent_free(tiqe); 596 } 597 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 598 goto out; 599 } 600 /* 601 * Received segment completely overlaps this fragment 602 * so we drop the fragment (this keeps the temporal 603 * ordering of segments correct). 604 */ 605 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 606 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 607 rcvpartdupbyte += q->ipqe_len; 608 m_freem(q->ipqe_m); 609 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 610 goto free_ipqe; 611 } 612 /* 613 * RX'ed segment extends past the end of the 614 * fragment. Drop the overlapping bytes. Then 615 * merge the fragment and segment then treat as 616 * a longer received packet. 617 */ 618 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 619 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 620 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 621 #ifdef TCPREASS_DEBUG 622 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 623 tp, overlap, 624 pkt_seq, pkt_seq + pkt_len, pkt_len); 625 #endif 626 m_adj(m, overlap); 627 rcvpartdupbyte += overlap; 628 m_cat(q->ipre_mlast, m); 629 TRAVERSE(q->ipre_mlast); 630 m = q->ipqe_m; 631 pkt_seq = q->ipqe_seq; 632 pkt_len += q->ipqe_len - overlap; 633 rcvoobyte -= overlap; 634 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 635 goto free_ipqe; 636 } 637 /* 638 * RX'ed segment extends past the front of the 639 * fragment. Drop the overlapping bytes on the 640 * received packet. The packet will then be 641 * contatentated with this fragment a bit later. 642 */ 643 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 644 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 645 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 646 #ifdef TCPREASS_DEBUG 647 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 648 tp, overlap, 649 pkt_seq, pkt_seq + pkt_len, pkt_len); 650 #endif 651 m_adj(m, -overlap); 652 pkt_len -= overlap; 653 rcvpartdupbyte += overlap; 654 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 655 rcvoobyte -= overlap; 656 } 657 /* 658 * If the received segment immediates precedes this 659 * fragment then tack the fragment onto this segment 660 * and reinsert the data. 661 */ 662 if (q->ipqe_seq == pkt_seq + pkt_len) { 663 #ifdef TCPREASS_DEBUG 664 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 665 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 666 pkt_seq, pkt_seq + pkt_len, pkt_len); 667 #endif 668 pkt_len += q->ipqe_len; 669 pkt_flags |= q->ipqe_flags; 670 m_cat(m, q->ipqe_m); 671 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 672 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 673 tp->t_segqlen--; 674 KASSERT(tp->t_segqlen >= 0); 675 KASSERT(tp->t_segqlen != 0 || 676 (TAILQ_EMPTY(&tp->segq) && 677 TAILQ_EMPTY(&tp->timeq))); 678 if (tiqe == NULL) { 679 tiqe = q; 680 } else { 681 tcpipqent_free(q); 682 } 683 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 684 break; 685 } 686 /* 687 * If the fragment is before the segment, remember it. 688 * When this loop is terminated, p will contain the 689 * pointer to fragment that is right before the received 690 * segment. 691 */ 692 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 693 p = q; 694 695 continue; 696 697 /* 698 * This is a common operation. It also will allow 699 * to save doing a malloc/free in most instances. 700 */ 701 free_ipqe: 702 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 703 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 704 tp->t_segqlen--; 705 KASSERT(tp->t_segqlen >= 0); 706 KASSERT(tp->t_segqlen != 0 || 707 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 708 if (tiqe == NULL) { 709 tiqe = q; 710 } else { 711 tcpipqent_free(q); 712 } 713 } 714 715 #ifdef TCP_REASS_COUNTERS 716 if (count > 7) 717 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 718 else if (count > 0) 719 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 720 #endif 721 722 insert_it: 723 724 /* 725 * Allocate a new queue entry since the received segment did not 726 * collapse onto any other out-of-order block; thus we are allocating 727 * a new block. If it had collapsed, tiqe would not be NULL and 728 * we would be reusing it. 729 * XXX If we can't, just drop the packet. XXX 730 */ 731 if (tiqe == NULL) { 732 tiqe = tcpipqent_alloc(); 733 if (tiqe == NULL) { 734 TCP_STATINC(TCP_STAT_RCVMEMDROP); 735 m_freem(m); 736 goto out; 737 } 738 } 739 740 /* 741 * Update the counters. 742 */ 743 tp->t_rcvoopack++; 744 tcps = TCP_STAT_GETREF(); 745 tcps[TCP_STAT_RCVOOPACK]++; 746 tcps[TCP_STAT_RCVOOBYTE] += rcvoobyte; 747 if (rcvpartdupbyte) { 748 tcps[TCP_STAT_RCVPARTDUPPACK]++; 749 tcps[TCP_STAT_RCVPARTDUPBYTE] += rcvpartdupbyte; 750 } 751 TCP_STAT_PUTREF(); 752 753 /* 754 * Insert the new fragment queue entry into both queues. 755 */ 756 tiqe->ipqe_m = m; 757 tiqe->ipre_mlast = m; 758 tiqe->ipqe_seq = pkt_seq; 759 tiqe->ipqe_len = pkt_len; 760 tiqe->ipqe_flags = pkt_flags; 761 if (p == NULL) { 762 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 763 #ifdef TCPREASS_DEBUG 764 if (tiqe->ipqe_seq != tp->rcv_nxt) 765 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 766 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 767 #endif 768 } else { 769 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 770 #ifdef TCPREASS_DEBUG 771 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 772 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 773 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 774 #endif 775 } 776 tp->t_segqlen++; 777 778 skip_replacement: 779 780 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 781 782 present: 783 /* 784 * Present data to user, advancing rcv_nxt through 785 * completed sequence space. 786 */ 787 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 788 goto out; 789 q = TAILQ_FIRST(&tp->segq); 790 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 791 goto out; 792 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 793 goto out; 794 795 tp->rcv_nxt += q->ipqe_len; 796 pkt_flags = q->ipqe_flags & TH_FIN; 797 nd6_hint(tp); 798 799 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 800 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 801 tp->t_segqlen--; 802 KASSERT(tp->t_segqlen >= 0); 803 KASSERT(tp->t_segqlen != 0 || 804 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 805 if (so->so_state & SS_CANTRCVMORE) 806 m_freem(q->ipqe_m); 807 else 808 sbappendstream(&so->so_rcv, q->ipqe_m); 809 tcpipqent_free(q); 810 TCP_REASS_UNLOCK(tp); 811 sorwakeup(so); 812 return (pkt_flags); 813 out: 814 TCP_REASS_UNLOCK(tp); 815 return (0); 816 } 817 818 #ifdef INET6 819 int 820 tcp6_input(struct mbuf **mp, int *offp, int proto) 821 { 822 struct mbuf *m = *mp; 823 824 /* 825 * draft-itojun-ipv6-tcp-to-anycast 826 * better place to put this in? 827 */ 828 if (m->m_flags & M_ANYCAST6) { 829 struct ip6_hdr *ip6; 830 if (m->m_len < sizeof(struct ip6_hdr)) { 831 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 832 TCP_STATINC(TCP_STAT_RCVSHORT); 833 return IPPROTO_DONE; 834 } 835 } 836 ip6 = mtod(m, struct ip6_hdr *); 837 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 838 (char *)&ip6->ip6_dst - (char *)ip6); 839 return IPPROTO_DONE; 840 } 841 842 tcp_input(m, *offp, proto); 843 return IPPROTO_DONE; 844 } 845 #endif 846 847 #ifdef INET 848 static void 849 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 850 { 851 char src[INET_ADDRSTRLEN]; 852 char dst[INET_ADDRSTRLEN]; 853 854 if (ip) { 855 in_print(src, sizeof(src), &ip->ip_src); 856 in_print(dst, sizeof(dst), &ip->ip_dst); 857 } 858 else { 859 strlcpy(src, "(unknown)", sizeof(src)); 860 strlcpy(dst, "(unknown)", sizeof(dst)); 861 } 862 log(LOG_INFO, 863 "Connection attempt to TCP %s:%d from %s:%d\n", 864 dst, ntohs(th->th_dport), 865 src, ntohs(th->th_sport)); 866 } 867 #endif 868 869 #ifdef INET6 870 static void 871 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 872 { 873 char src[INET6_ADDRSTRLEN]; 874 char dst[INET6_ADDRSTRLEN]; 875 876 if (ip6) { 877 in6_print(src, sizeof(src), &ip6->ip6_src); 878 in6_print(dst, sizeof(dst), &ip6->ip6_dst); 879 } 880 else { 881 strlcpy(src, "(unknown v6)", sizeof(src)); 882 strlcpy(dst, "(unknown v6)", sizeof(dst)); 883 } 884 log(LOG_INFO, 885 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 886 dst, ntohs(th->th_dport), 887 src, ntohs(th->th_sport)); 888 } 889 #endif 890 891 /* 892 * Checksum extended TCP header and data. 893 */ 894 int 895 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, 896 int toff, int off, int tlen) 897 { 898 struct ifnet *rcvif; 899 int s; 900 901 /* 902 * XXX it's better to record and check if this mbuf is 903 * already checked. 904 */ 905 906 rcvif = m_get_rcvif(m, &s); 907 if (__predict_false(rcvif == NULL)) 908 goto badcsum; /* XXX */ 909 910 switch (af) { 911 #ifdef INET 912 case AF_INET: 913 switch (m->m_pkthdr.csum_flags & 914 ((rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 915 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 916 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 917 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 918 goto badcsum; 919 920 case M_CSUM_TCPv4|M_CSUM_DATA: { 921 u_int32_t hw_csum = m->m_pkthdr.csum_data; 922 923 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 924 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 925 const struct ip *ip = 926 mtod(m, const struct ip *); 927 928 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 929 ip->ip_dst.s_addr, 930 htons(hw_csum + tlen + off + IPPROTO_TCP)); 931 } 932 if ((hw_csum ^ 0xffff) != 0) 933 goto badcsum; 934 break; 935 } 936 937 case M_CSUM_TCPv4: 938 /* Checksum was okay. */ 939 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 940 break; 941 942 default: 943 /* 944 * Must compute it ourselves. Maybe skip checksum 945 * on loopback interfaces. 946 */ 947 if (__predict_true(!(rcvif->if_flags & IFF_LOOPBACK) || 948 tcp_do_loopback_cksum)) { 949 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 950 if (in4_cksum(m, IPPROTO_TCP, toff, 951 tlen + off) != 0) 952 goto badcsum; 953 } 954 break; 955 } 956 break; 957 #endif /* INET4 */ 958 959 #ifdef INET6 960 case AF_INET6: 961 switch (m->m_pkthdr.csum_flags & 962 ((rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 963 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 964 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 965 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 966 goto badcsum; 967 968 #if 0 /* notyet */ 969 case M_CSUM_TCPv6|M_CSUM_DATA: 970 #endif 971 972 case M_CSUM_TCPv6: 973 /* Checksum was okay. */ 974 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 975 break; 976 977 default: 978 /* 979 * Must compute it ourselves. Maybe skip checksum 980 * on loopback interfaces. 981 */ 982 if (__predict_true((m->m_flags & M_LOOP) == 0 || 983 tcp_do_loopback_cksum)) { 984 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 985 if (in6_cksum(m, IPPROTO_TCP, toff, 986 tlen + off) != 0) 987 goto badcsum; 988 } 989 } 990 break; 991 #endif /* INET6 */ 992 } 993 m_put_rcvif(rcvif, &s); 994 995 return 0; 996 997 badcsum: 998 m_put_rcvif(rcvif, &s); 999 TCP_STATINC(TCP_STAT_RCVBADSUM); 1000 return -1; 1001 } 1002 1003 /* When a packet arrives addressed to a vestigial tcpbp, we 1004 * nevertheless have to respond to it per the spec. 1005 */ 1006 static void tcp_vtw_input(struct tcphdr *th, vestigial_inpcb_t *vp, 1007 struct mbuf *m, int tlen, int multicast) 1008 { 1009 int tiflags; 1010 int todrop; 1011 uint32_t t_flags = 0; 1012 uint64_t *tcps; 1013 1014 tiflags = th->th_flags; 1015 todrop = vp->rcv_nxt - th->th_seq; 1016 1017 if (todrop > 0) { 1018 if (tiflags & TH_SYN) { 1019 tiflags &= ~TH_SYN; 1020 ++th->th_seq; 1021 if (th->th_urp > 1) 1022 --th->th_urp; 1023 else { 1024 tiflags &= ~TH_URG; 1025 th->th_urp = 0; 1026 } 1027 --todrop; 1028 } 1029 if (todrop > tlen || 1030 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1031 /* 1032 * Any valid FIN or RST must be to the left of the 1033 * window. At this point the FIN or RST must be a 1034 * duplicate or out of sequence; drop it. 1035 */ 1036 if (tiflags & TH_RST) 1037 goto drop; 1038 tiflags &= ~(TH_FIN|TH_RST); 1039 /* 1040 * Send an ACK to resynchronize and drop any data. 1041 * But keep on processing for RST or ACK. 1042 */ 1043 t_flags |= TF_ACKNOW; 1044 todrop = tlen; 1045 tcps = TCP_STAT_GETREF(); 1046 tcps[TCP_STAT_RCVDUPPACK] += 1; 1047 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 1048 TCP_STAT_PUTREF(); 1049 } else if ((tiflags & TH_RST) 1050 && th->th_seq != vp->rcv_nxt) { 1051 /* 1052 * Test for reset before adjusting the sequence 1053 * number for overlapping data. 1054 */ 1055 goto dropafterack_ratelim; 1056 } else { 1057 tcps = TCP_STAT_GETREF(); 1058 tcps[TCP_STAT_RCVPARTDUPPACK] += 1; 1059 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 1060 TCP_STAT_PUTREF(); 1061 } 1062 1063 // tcp_new_dsack(tp, th->th_seq, todrop); 1064 // hdroptlen += todrop; /*drop from head afterwards*/ 1065 1066 th->th_seq += todrop; 1067 tlen -= todrop; 1068 1069 if (th->th_urp > todrop) 1070 th->th_urp -= todrop; 1071 else { 1072 tiflags &= ~TH_URG; 1073 th->th_urp = 0; 1074 } 1075 } 1076 1077 /* 1078 * If new data are received on a connection after the 1079 * user processes are gone, then RST the other end. 1080 */ 1081 if (tlen) { 1082 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 1083 goto dropwithreset; 1084 } 1085 1086 /* 1087 * If segment ends after window, drop trailing data 1088 * (and PUSH and FIN); if nothing left, just ACK. 1089 */ 1090 todrop = (th->th_seq + tlen) - (vp->rcv_nxt+vp->rcv_wnd); 1091 1092 if (todrop > 0) { 1093 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 1094 if (todrop >= tlen) { 1095 /* 1096 * The segment actually starts after the window. 1097 * th->th_seq + tlen - vp->rcv_nxt - vp->rcv_wnd >= tlen 1098 * th->th_seq - vp->rcv_nxt - vp->rcv_wnd >= 0 1099 * th->th_seq >= vp->rcv_nxt + vp->rcv_wnd 1100 */ 1101 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 1102 /* 1103 * If a new connection request is received 1104 * while in TIME_WAIT, drop the old connection 1105 * and start over if the sequence numbers 1106 * are above the previous ones. 1107 */ 1108 if ((tiflags & TH_SYN) 1109 && SEQ_GT(th->th_seq, vp->rcv_nxt)) { 1110 /* We only support this in the !NOFDREF case, which 1111 * is to say: not here. 1112 */ 1113 goto dropwithreset; 1114 } 1115 /* 1116 * If window is closed can only take segments at 1117 * window edge, and have to drop data and PUSH from 1118 * incoming segments. Continue processing, but 1119 * remember to ack. Otherwise, drop segment 1120 * and (if not RST) ack. 1121 */ 1122 if (vp->rcv_wnd == 0 && th->th_seq == vp->rcv_nxt) { 1123 t_flags |= TF_ACKNOW; 1124 TCP_STATINC(TCP_STAT_RCVWINPROBE); 1125 } else 1126 goto dropafterack; 1127 } else 1128 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 1129 m_adj(m, -todrop); 1130 tlen -= todrop; 1131 tiflags &= ~(TH_PUSH|TH_FIN); 1132 } 1133 1134 if (tiflags & TH_RST) { 1135 if (th->th_seq != vp->rcv_nxt) 1136 goto dropafterack_ratelim; 1137 1138 vtw_del(vp->ctl, vp->vtw); 1139 goto drop; 1140 } 1141 1142 /* 1143 * If the ACK bit is off we drop the segment and return. 1144 */ 1145 if ((tiflags & TH_ACK) == 0) { 1146 if (t_flags & TF_ACKNOW) 1147 goto dropafterack; 1148 else 1149 goto drop; 1150 } 1151 1152 /* 1153 * In TIME_WAIT state the only thing that should arrive 1154 * is a retransmission of the remote FIN. Acknowledge 1155 * it and restart the finack timer. 1156 */ 1157 vtw_restart(vp); 1158 goto dropafterack; 1159 1160 dropafterack: 1161 /* 1162 * Generate an ACK dropping incoming segment if it occupies 1163 * sequence space, where the ACK reflects our state. 1164 */ 1165 if (tiflags & TH_RST) 1166 goto drop; 1167 goto dropafterack2; 1168 1169 dropafterack_ratelim: 1170 /* 1171 * We may want to rate-limit ACKs against SYN/RST attack. 1172 */ 1173 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 1174 tcp_ackdrop_ppslim) == 0) { 1175 /* XXX stat */ 1176 goto drop; 1177 } 1178 /* ...fall into dropafterack2... */ 1179 1180 dropafterack2: 1181 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, th->th_ack, 1182 TH_ACK); 1183 return; 1184 1185 dropwithreset: 1186 /* 1187 * Generate a RST, dropping incoming segment. 1188 * Make ACK acceptable to originator of segment. 1189 */ 1190 if (tiflags & TH_RST) 1191 goto drop; 1192 1193 if (tiflags & TH_ACK) 1194 tcp_respond(0, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 1195 else { 1196 if (tiflags & TH_SYN) 1197 ++tlen; 1198 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, (tcp_seq)0, 1199 TH_RST|TH_ACK); 1200 } 1201 return; 1202 drop: 1203 m_freem(m); 1204 } 1205 1206 /* 1207 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 1208 */ 1209 void 1210 tcp_input(struct mbuf *m, ...) 1211 { 1212 struct tcphdr *th; 1213 struct ip *ip; 1214 struct inpcb *inp; 1215 #ifdef INET6 1216 struct ip6_hdr *ip6; 1217 struct in6pcb *in6p; 1218 #endif 1219 u_int8_t *optp = NULL; 1220 int optlen = 0; 1221 int len, tlen, toff, hdroptlen = 0; 1222 struct tcpcb *tp = 0; 1223 int tiflags; 1224 struct socket *so = NULL; 1225 int todrop, acked, ourfinisacked, needoutput = 0; 1226 bool dupseg; 1227 #ifdef TCP_DEBUG 1228 short ostate = 0; 1229 #endif 1230 u_long tiwin; 1231 struct tcp_opt_info opti; 1232 int off, iphlen; 1233 va_list ap; 1234 int af; /* af on the wire */ 1235 struct mbuf *tcp_saveti = NULL; 1236 uint32_t ts_rtt; 1237 uint8_t iptos; 1238 uint64_t *tcps; 1239 vestigial_inpcb_t vestige; 1240 1241 vestige.valid = 0; 1242 1243 MCLAIM(m, &tcp_rx_mowner); 1244 va_start(ap, m); 1245 toff = va_arg(ap, int); 1246 (void)va_arg(ap, int); /* ignore value, advance ap */ 1247 va_end(ap); 1248 1249 TCP_STATINC(TCP_STAT_RCVTOTAL); 1250 1251 memset(&opti, 0, sizeof(opti)); 1252 opti.ts_present = 0; 1253 opti.maxseg = 0; 1254 1255 /* 1256 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 1257 * 1258 * TCP is, by definition, unicast, so we reject all 1259 * multicast outright. 1260 * 1261 * Note, there are additional src/dst address checks in 1262 * the AF-specific code below. 1263 */ 1264 if (m->m_flags & (M_BCAST|M_MCAST)) { 1265 /* XXX stat */ 1266 goto drop; 1267 } 1268 #ifdef INET6 1269 if (m->m_flags & M_ANYCAST6) { 1270 /* XXX stat */ 1271 goto drop; 1272 } 1273 #endif 1274 1275 /* 1276 * Get IP and TCP header. 1277 * Note: IP leaves IP header in first mbuf. 1278 */ 1279 ip = mtod(m, struct ip *); 1280 switch (ip->ip_v) { 1281 #ifdef INET 1282 case 4: 1283 #ifdef INET6 1284 ip6 = NULL; 1285 #endif 1286 af = AF_INET; 1287 iphlen = sizeof(struct ip); 1288 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1289 sizeof(struct tcphdr)); 1290 if (th == NULL) { 1291 TCP_STATINC(TCP_STAT_RCVSHORT); 1292 return; 1293 } 1294 /* We do the checksum after PCB lookup... */ 1295 len = ntohs(ip->ip_len); 1296 tlen = len - toff; 1297 iptos = ip->ip_tos; 1298 break; 1299 #endif 1300 #ifdef INET6 1301 case 6: 1302 ip = NULL; 1303 iphlen = sizeof(struct ip6_hdr); 1304 af = AF_INET6; 1305 ip6 = mtod(m, struct ip6_hdr *); 1306 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1307 sizeof(struct tcphdr)); 1308 if (th == NULL) { 1309 TCP_STATINC(TCP_STAT_RCVSHORT); 1310 return; 1311 } 1312 1313 /* Be proactive about malicious use of IPv4 mapped address */ 1314 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 1315 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 1316 /* XXX stat */ 1317 goto drop; 1318 } 1319 1320 /* 1321 * Be proactive about unspecified IPv6 address in source. 1322 * As we use all-zero to indicate unbounded/unconnected pcb, 1323 * unspecified IPv6 address can be used to confuse us. 1324 * 1325 * Note that packets with unspecified IPv6 destination is 1326 * already dropped in ip6_input. 1327 */ 1328 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1329 /* XXX stat */ 1330 goto drop; 1331 } 1332 1333 /* 1334 * Make sure destination address is not multicast. 1335 * Source address checked in ip6_input(). 1336 */ 1337 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1338 /* XXX stat */ 1339 goto drop; 1340 } 1341 1342 /* We do the checksum after PCB lookup... */ 1343 len = m->m_pkthdr.len; 1344 tlen = len - toff; 1345 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1346 break; 1347 #endif 1348 default: 1349 m_freem(m); 1350 return; 1351 } 1352 /* 1353 * Enforce alignment requirements that are violated in 1354 * some cases, see kern/50766 for details. 1355 */ 1356 if (TCP_HDR_ALIGNED_P(th) == 0) { 1357 m = m_copyup(m, toff + sizeof(struct tcphdr), 0); 1358 if (m == NULL) { 1359 TCP_STATINC(TCP_STAT_RCVSHORT); 1360 return; 1361 } 1362 ip = mtod(m, struct ip *); 1363 #ifdef INET6 1364 ip6 = mtod(m, struct ip6_hdr *); 1365 #endif 1366 th = (struct tcphdr *)(mtod(m, char *) + toff); 1367 } 1368 KASSERT(TCP_HDR_ALIGNED_P(th)); 1369 1370 /* 1371 * Check that TCP offset makes sense, 1372 * pull out TCP options and adjust length. XXX 1373 */ 1374 off = th->th_off << 2; 1375 if (off < sizeof (struct tcphdr) || off > tlen) { 1376 TCP_STATINC(TCP_STAT_RCVBADOFF); 1377 goto drop; 1378 } 1379 tlen -= off; 1380 1381 /* 1382 * tcp_input() has been modified to use tlen to mean the TCP data 1383 * length throughout the function. Other functions can use 1384 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1385 * rja 1386 */ 1387 1388 if (off > sizeof (struct tcphdr)) { 1389 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1390 if (th == NULL) { 1391 TCP_STATINC(TCP_STAT_RCVSHORT); 1392 return; 1393 } 1394 /* 1395 * NOTE: ip/ip6 will not be affected by m_pulldown() 1396 * (as they're before toff) and we don't need to update those. 1397 */ 1398 KASSERT(TCP_HDR_ALIGNED_P(th)); 1399 optlen = off - sizeof (struct tcphdr); 1400 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1401 /* 1402 * Do quick retrieval of timestamp options ("options 1403 * prediction?"). If timestamp is the only option and it's 1404 * formatted as recommended in RFC 1323 appendix A, we 1405 * quickly get the values now and not bother calling 1406 * tcp_dooptions(), etc. 1407 */ 1408 if ((optlen == TCPOLEN_TSTAMP_APPA || 1409 (optlen > TCPOLEN_TSTAMP_APPA && 1410 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1411 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1412 (th->th_flags & TH_SYN) == 0) { 1413 opti.ts_present = 1; 1414 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1415 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1416 optp = NULL; /* we've parsed the options */ 1417 } 1418 } 1419 tiflags = th->th_flags; 1420 1421 /* 1422 * Checksum extended TCP header and data 1423 */ 1424 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1425 goto badcsum; 1426 1427 /* 1428 * Locate pcb for segment. 1429 */ 1430 findpcb: 1431 inp = NULL; 1432 #ifdef INET6 1433 in6p = NULL; 1434 #endif 1435 switch (af) { 1436 #ifdef INET 1437 case AF_INET: 1438 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1439 ip->ip_dst, th->th_dport, 1440 &vestige); 1441 if (inp == 0 && !vestige.valid) { 1442 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1443 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1444 } 1445 #ifdef INET6 1446 if (inp == 0 && !vestige.valid) { 1447 struct in6_addr s, d; 1448 1449 /* mapped addr case */ 1450 in6_in_2_v4mapin6(&ip->ip_src, &s); 1451 in6_in_2_v4mapin6(&ip->ip_dst, &d); 1452 in6p = in6_pcblookup_connect(&tcbtable, &s, 1453 th->th_sport, &d, th->th_dport, 1454 0, &vestige); 1455 if (in6p == 0 && !vestige.valid) { 1456 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1457 in6p = in6_pcblookup_bind(&tcbtable, &d, 1458 th->th_dport, 0); 1459 } 1460 } 1461 #endif 1462 #ifndef INET6 1463 if (inp == 0 && !vestige.valid) 1464 #else 1465 if (inp == 0 && in6p == 0 && !vestige.valid) 1466 #endif 1467 { 1468 TCP_STATINC(TCP_STAT_NOPORT); 1469 if (tcp_log_refused && 1470 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1471 tcp4_log_refused(ip, th); 1472 } 1473 tcp_fields_to_host(th); 1474 goto dropwithreset_ratelim; 1475 } 1476 #if defined(IPSEC) 1477 if (ipsec_used) { 1478 if (inp && 1479 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 1480 && ipsec4_in_reject(m, inp)) { 1481 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1482 goto drop; 1483 } 1484 #ifdef INET6 1485 else if (in6p && 1486 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 1487 && ipsec6_in_reject(m, in6p)) { 1488 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1489 goto drop; 1490 } 1491 #endif 1492 } 1493 #endif /*IPSEC*/ 1494 break; 1495 #endif /*INET*/ 1496 #ifdef INET6 1497 case AF_INET6: 1498 { 1499 int faith; 1500 1501 #if defined(NFAITH) && NFAITH > 0 1502 faith = faithprefix(&ip6->ip6_dst); 1503 #else 1504 faith = 0; 1505 #endif 1506 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1507 th->th_sport, &ip6->ip6_dst, th->th_dport, faith, &vestige); 1508 if (!in6p && !vestige.valid) { 1509 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1510 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1511 th->th_dport, faith); 1512 } 1513 if (!in6p && !vestige.valid) { 1514 TCP_STATINC(TCP_STAT_NOPORT); 1515 if (tcp_log_refused && 1516 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1517 tcp6_log_refused(ip6, th); 1518 } 1519 tcp_fields_to_host(th); 1520 goto dropwithreset_ratelim; 1521 } 1522 #if defined(IPSEC) 1523 if (ipsec_used && in6p 1524 && (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 1525 && ipsec6_in_reject(m, in6p)) { 1526 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1527 goto drop; 1528 } 1529 #endif /*IPSEC*/ 1530 break; 1531 } 1532 #endif 1533 } 1534 1535 /* 1536 * If the state is CLOSED (i.e., TCB does not exist) then 1537 * all data in the incoming segment is discarded. 1538 * If the TCB exists but is in CLOSED state, it is embryonic, 1539 * but should either do a listen or a connect soon. 1540 */ 1541 tp = NULL; 1542 so = NULL; 1543 if (inp) { 1544 /* Check the minimum TTL for socket. */ 1545 if (ip->ip_ttl < inp->inp_ip_minttl) 1546 goto drop; 1547 1548 tp = intotcpcb(inp); 1549 so = inp->inp_socket; 1550 } 1551 #ifdef INET6 1552 else if (in6p) { 1553 tp = in6totcpcb(in6p); 1554 so = in6p->in6p_socket; 1555 } 1556 #endif 1557 else if (vestige.valid) { 1558 int mc = 0; 1559 1560 /* We do not support the resurrection of vtw tcpcps. 1561 */ 1562 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1563 goto badcsum; 1564 1565 switch (af) { 1566 #ifdef INET6 1567 case AF_INET6: 1568 mc = IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst); 1569 break; 1570 #endif 1571 1572 case AF_INET: 1573 mc = (IN_MULTICAST(ip->ip_dst.s_addr) 1574 || in_broadcast(ip->ip_dst, 1575 m_get_rcvif_NOMPSAFE(m))); 1576 break; 1577 } 1578 1579 tcp_fields_to_host(th); 1580 tcp_vtw_input(th, &vestige, m, tlen, mc); 1581 m = 0; 1582 goto drop; 1583 } 1584 1585 if (tp == 0) { 1586 tcp_fields_to_host(th); 1587 goto dropwithreset_ratelim; 1588 } 1589 if (tp->t_state == TCPS_CLOSED) 1590 goto drop; 1591 1592 KASSERT(so->so_lock == softnet_lock); 1593 KASSERT(solocked(so)); 1594 1595 tcp_fields_to_host(th); 1596 1597 /* Unscale the window into a 32-bit value. */ 1598 if ((tiflags & TH_SYN) == 0) 1599 tiwin = th->th_win << tp->snd_scale; 1600 else 1601 tiwin = th->th_win; 1602 1603 #ifdef INET6 1604 /* save packet options if user wanted */ 1605 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1606 if (in6p->in6p_options) { 1607 m_freem(in6p->in6p_options); 1608 in6p->in6p_options = 0; 1609 } 1610 KASSERT(ip6 != NULL); 1611 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1612 } 1613 #endif 1614 1615 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1616 union syn_cache_sa src; 1617 union syn_cache_sa dst; 1618 1619 memset(&src, 0, sizeof(src)); 1620 memset(&dst, 0, sizeof(dst)); 1621 switch (af) { 1622 #ifdef INET 1623 case AF_INET: 1624 src.sin.sin_len = sizeof(struct sockaddr_in); 1625 src.sin.sin_family = AF_INET; 1626 src.sin.sin_addr = ip->ip_src; 1627 src.sin.sin_port = th->th_sport; 1628 1629 dst.sin.sin_len = sizeof(struct sockaddr_in); 1630 dst.sin.sin_family = AF_INET; 1631 dst.sin.sin_addr = ip->ip_dst; 1632 dst.sin.sin_port = th->th_dport; 1633 break; 1634 #endif 1635 #ifdef INET6 1636 case AF_INET6: 1637 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1638 src.sin6.sin6_family = AF_INET6; 1639 src.sin6.sin6_addr = ip6->ip6_src; 1640 src.sin6.sin6_port = th->th_sport; 1641 1642 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1643 dst.sin6.sin6_family = AF_INET6; 1644 dst.sin6.sin6_addr = ip6->ip6_dst; 1645 dst.sin6.sin6_port = th->th_dport; 1646 break; 1647 #endif /* INET6 */ 1648 default: 1649 goto badsyn; /*sanity*/ 1650 } 1651 1652 if (so->so_options & SO_DEBUG) { 1653 #ifdef TCP_DEBUG 1654 ostate = tp->t_state; 1655 #endif 1656 1657 tcp_saveti = NULL; 1658 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1659 goto nosave; 1660 1661 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1662 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1663 if (!tcp_saveti) 1664 goto nosave; 1665 } else { 1666 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1667 if (!tcp_saveti) 1668 goto nosave; 1669 MCLAIM(m, &tcp_mowner); 1670 tcp_saveti->m_len = iphlen; 1671 m_copydata(m, 0, iphlen, 1672 mtod(tcp_saveti, void *)); 1673 } 1674 1675 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1676 m_freem(tcp_saveti); 1677 tcp_saveti = NULL; 1678 } else { 1679 tcp_saveti->m_len += sizeof(struct tcphdr); 1680 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1681 sizeof(struct tcphdr)); 1682 } 1683 nosave:; 1684 } 1685 if (so->so_options & SO_ACCEPTCONN) { 1686 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1687 if (tiflags & TH_RST) { 1688 syn_cache_reset(&src.sa, &dst.sa, th); 1689 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1690 (TH_ACK|TH_SYN)) { 1691 /* 1692 * Received a SYN,ACK. This should 1693 * never happen while we are in 1694 * LISTEN. Send an RST. 1695 */ 1696 goto badsyn; 1697 } else if (tiflags & TH_ACK) { 1698 so = syn_cache_get(&src.sa, &dst.sa, 1699 th, toff, tlen, so, m); 1700 if (so == NULL) { 1701 /* 1702 * We don't have a SYN for 1703 * this ACK; send an RST. 1704 */ 1705 goto badsyn; 1706 } else if (so == 1707 (struct socket *)(-1)) { 1708 /* 1709 * We were unable to create 1710 * the connection. If the 1711 * 3-way handshake was 1712 * completed, and RST has 1713 * been sent to the peer. 1714 * Since the mbuf might be 1715 * in use for the reply, 1716 * do not free it. 1717 */ 1718 m = NULL; 1719 } else { 1720 /* 1721 * We have created a 1722 * full-blown connection. 1723 */ 1724 tp = NULL; 1725 inp = NULL; 1726 #ifdef INET6 1727 in6p = NULL; 1728 #endif 1729 switch (so->so_proto->pr_domain->dom_family) { 1730 #ifdef INET 1731 case AF_INET: 1732 inp = sotoinpcb(so); 1733 tp = intotcpcb(inp); 1734 break; 1735 #endif 1736 #ifdef INET6 1737 case AF_INET6: 1738 in6p = sotoin6pcb(so); 1739 tp = in6totcpcb(in6p); 1740 break; 1741 #endif 1742 } 1743 if (tp == NULL) 1744 goto badsyn; /*XXX*/ 1745 tiwin <<= tp->snd_scale; 1746 goto after_listen; 1747 } 1748 } else { 1749 /* 1750 * None of RST, SYN or ACK was set. 1751 * This is an invalid packet for a 1752 * TCB in LISTEN state. Send a RST. 1753 */ 1754 goto badsyn; 1755 } 1756 } else { 1757 /* 1758 * Received a SYN. 1759 * 1760 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1761 */ 1762 if (m->m_flags & (M_BCAST|M_MCAST)) 1763 goto drop; 1764 1765 switch (af) { 1766 #ifdef INET6 1767 case AF_INET6: 1768 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1769 goto drop; 1770 break; 1771 #endif /* INET6 */ 1772 case AF_INET: 1773 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1774 in_broadcast(ip->ip_dst, 1775 m_get_rcvif_NOMPSAFE(m))) 1776 goto drop; 1777 break; 1778 } 1779 1780 #ifdef INET6 1781 /* 1782 * If deprecated address is forbidden, we do 1783 * not accept SYN to deprecated interface 1784 * address to prevent any new inbound 1785 * connection from getting established. 1786 * When we do not accept SYN, we send a TCP 1787 * RST, with deprecated source address (instead 1788 * of dropping it). We compromise it as it is 1789 * much better for peer to send a RST, and 1790 * RST will be the final packet for the 1791 * exchange. 1792 * 1793 * If we do not forbid deprecated addresses, we 1794 * accept the SYN packet. RFC2462 does not 1795 * suggest dropping SYN in this case. 1796 * If we decipher RFC2462 5.5.4, it says like 1797 * this: 1798 * 1. use of deprecated addr with existing 1799 * communication is okay - "SHOULD continue 1800 * to be used" 1801 * 2. use of it with new communication: 1802 * (2a) "SHOULD NOT be used if alternate 1803 * address with sufficient scope is 1804 * available" 1805 * (2b) nothing mentioned otherwise. 1806 * Here we fall into (2b) case as we have no 1807 * choice in our source address selection - we 1808 * must obey the peer. 1809 * 1810 * The wording in RFC2462 is confusing, and 1811 * there are multiple description text for 1812 * deprecated address handling - worse, they 1813 * are not exactly the same. I believe 5.5.4 1814 * is the best one, so we follow 5.5.4. 1815 */ 1816 if (af == AF_INET6 && !ip6_use_deprecated) { 1817 struct in6_ifaddr *ia6; 1818 int s; 1819 struct ifnet *rcvif = m_get_rcvif(m, &s); 1820 if (rcvif == NULL) 1821 goto dropwithreset; /* XXX */ 1822 if ((ia6 = in6ifa_ifpwithaddr(rcvif, 1823 &ip6->ip6_dst)) && 1824 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1825 tp = NULL; 1826 m_put_rcvif(rcvif, &s); 1827 goto dropwithreset; 1828 } 1829 m_put_rcvif(rcvif, &s); 1830 } 1831 #endif 1832 1833 #if defined(IPSEC) 1834 if (ipsec_used) { 1835 switch (af) { 1836 #ifdef INET 1837 case AF_INET: 1838 /* 1839 * inp can be NULL when 1840 * receiving an IPv4 packet on 1841 * an IPv4-mapped IPv6 address. 1842 */ 1843 KASSERT(inp == NULL || 1844 sotoinpcb(so) == inp); 1845 if (!ipsec4_in_reject(m, inp)) 1846 break; 1847 IPSEC_STATINC( 1848 IPSEC_STAT_IN_POLVIO); 1849 tp = NULL; 1850 goto dropwithreset; 1851 #endif 1852 #ifdef INET6 1853 case AF_INET6: 1854 KASSERT(sotoin6pcb(so) == in6p); 1855 if (!ipsec6_in_reject(m, in6p)) 1856 break; 1857 IPSEC6_STATINC( 1858 IPSEC_STAT_IN_POLVIO); 1859 tp = NULL; 1860 goto dropwithreset; 1861 #endif /*INET6*/ 1862 } 1863 } 1864 #endif /*IPSEC*/ 1865 1866 /* 1867 * LISTEN socket received a SYN 1868 * from itself? This can't possibly 1869 * be valid; drop the packet. 1870 */ 1871 if (th->th_sport == th->th_dport) { 1872 int i; 1873 1874 switch (af) { 1875 #ifdef INET 1876 case AF_INET: 1877 i = in_hosteq(ip->ip_src, ip->ip_dst); 1878 break; 1879 #endif 1880 #ifdef INET6 1881 case AF_INET6: 1882 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1883 break; 1884 #endif 1885 default: 1886 i = 1; 1887 } 1888 if (i) { 1889 TCP_STATINC(TCP_STAT_BADSYN); 1890 goto drop; 1891 } 1892 } 1893 1894 /* 1895 * SYN looks ok; create compressed TCP 1896 * state for it. 1897 */ 1898 if (so->so_qlen <= so->so_qlimit && 1899 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1900 so, m, optp, optlen, &opti)) 1901 m = NULL; 1902 } 1903 goto drop; 1904 } 1905 } 1906 1907 after_listen: 1908 #ifdef DIAGNOSTIC 1909 /* 1910 * Should not happen now that all embryonic connections 1911 * are handled with compressed state. 1912 */ 1913 if (tp->t_state == TCPS_LISTEN) 1914 panic("tcp_input: TCPS_LISTEN"); 1915 #endif 1916 1917 /* 1918 * Segment received on connection. 1919 * Reset idle time and keep-alive timer. 1920 */ 1921 tp->t_rcvtime = tcp_now; 1922 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1923 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1924 1925 /* 1926 * Process options. 1927 */ 1928 #ifdef TCP_SIGNATURE 1929 if (optp || (tp->t_flags & TF_SIGNATURE)) 1930 #else 1931 if (optp) 1932 #endif 1933 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1934 goto drop; 1935 1936 if (TCP_SACK_ENABLED(tp)) { 1937 tcp_del_sackholes(tp, th); 1938 } 1939 1940 if (TCP_ECN_ALLOWED(tp)) { 1941 if (tiflags & TH_CWR) { 1942 tp->t_flags &= ~TF_ECN_SND_ECE; 1943 } 1944 switch (iptos & IPTOS_ECN_MASK) { 1945 case IPTOS_ECN_CE: 1946 tp->t_flags |= TF_ECN_SND_ECE; 1947 TCP_STATINC(TCP_STAT_ECN_CE); 1948 break; 1949 case IPTOS_ECN_ECT0: 1950 TCP_STATINC(TCP_STAT_ECN_ECT); 1951 break; 1952 case IPTOS_ECN_ECT1: 1953 /* XXX: ignore for now -- rpaulo */ 1954 break; 1955 } 1956 /* 1957 * Congestion experienced. 1958 * Ignore if we are already trying to recover. 1959 */ 1960 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1961 tp->t_congctl->cong_exp(tp); 1962 } 1963 1964 if (opti.ts_present && opti.ts_ecr) { 1965 /* 1966 * Calculate the RTT from the returned time stamp and the 1967 * connection's time base. If the time stamp is later than 1968 * the current time, or is extremely old, fall back to non-1323 1969 * RTT calculation. Since ts_rtt is unsigned, we can test both 1970 * at the same time. 1971 * 1972 * Note that ts_rtt is in units of slow ticks (500 1973 * ms). Since most earthbound RTTs are < 500 ms, 1974 * observed values will have large quantization noise. 1975 * Our smoothed RTT is then the fraction of observed 1976 * samples that are 1 tick instead of 0 (times 500 1977 * ms). 1978 * 1979 * ts_rtt is increased by 1 to denote a valid sample, 1980 * with 0 indicating an invalid measurement. This 1981 * extra 1 must be removed when ts_rtt is used, or 1982 * else an an erroneous extra 500 ms will result. 1983 */ 1984 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1985 if (ts_rtt > TCP_PAWS_IDLE) 1986 ts_rtt = 0; 1987 } else { 1988 ts_rtt = 0; 1989 } 1990 1991 /* 1992 * Header prediction: check for the two common cases 1993 * of a uni-directional data xfer. If the packet has 1994 * no control flags, is in-sequence, the window didn't 1995 * change and we're not retransmitting, it's a 1996 * candidate. If the length is zero and the ack moved 1997 * forward, we're the sender side of the xfer. Just 1998 * free the data acked & wake any higher level process 1999 * that was blocked waiting for space. If the length 2000 * is non-zero and the ack didn't move, we're the 2001 * receiver side. If we're getting packets in-order 2002 * (the reassembly queue is empty), add the data to 2003 * the socket buffer and note that we need a delayed ack. 2004 */ 2005 if (tp->t_state == TCPS_ESTABLISHED && 2006 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 2007 == TH_ACK && 2008 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 2009 th->th_seq == tp->rcv_nxt && 2010 tiwin && tiwin == tp->snd_wnd && 2011 tp->snd_nxt == tp->snd_max) { 2012 2013 /* 2014 * If last ACK falls within this segment's sequence numbers, 2015 * record the timestamp. 2016 * NOTE that the test is modified according to the latest 2017 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2018 * 2019 * note that we already know 2020 * TSTMP_GEQ(opti.ts_val, tp->ts_recent) 2021 */ 2022 if (opti.ts_present && 2023 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 2024 tp->ts_recent_age = tcp_now; 2025 tp->ts_recent = opti.ts_val; 2026 } 2027 2028 if (tlen == 0) { 2029 /* Ack prediction. */ 2030 if (SEQ_GT(th->th_ack, tp->snd_una) && 2031 SEQ_LEQ(th->th_ack, tp->snd_max) && 2032 tp->snd_cwnd >= tp->snd_wnd && 2033 tp->t_partialacks < 0) { 2034 /* 2035 * this is a pure ack for outstanding data. 2036 */ 2037 if (ts_rtt) 2038 tcp_xmit_timer(tp, ts_rtt - 1); 2039 else if (tp->t_rtttime && 2040 SEQ_GT(th->th_ack, tp->t_rtseq)) 2041 tcp_xmit_timer(tp, 2042 tcp_now - tp->t_rtttime); 2043 acked = th->th_ack - tp->snd_una; 2044 tcps = TCP_STAT_GETREF(); 2045 tcps[TCP_STAT_PREDACK]++; 2046 tcps[TCP_STAT_RCVACKPACK]++; 2047 tcps[TCP_STAT_RCVACKBYTE] += acked; 2048 TCP_STAT_PUTREF(); 2049 nd6_hint(tp); 2050 2051 if (acked > (tp->t_lastoff - tp->t_inoff)) 2052 tp->t_lastm = NULL; 2053 sbdrop(&so->so_snd, acked); 2054 tp->t_lastoff -= acked; 2055 2056 icmp_check(tp, th, acked); 2057 2058 tp->snd_una = th->th_ack; 2059 tp->snd_fack = tp->snd_una; 2060 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2061 tp->snd_high = tp->snd_una; 2062 m_freem(m); 2063 2064 /* 2065 * If all outstanding data are acked, stop 2066 * retransmit timer, otherwise restart timer 2067 * using current (possibly backed-off) value. 2068 * If process is waiting for space, 2069 * wakeup/selnotify/signal. If data 2070 * are ready to send, let tcp_output 2071 * decide between more output or persist. 2072 */ 2073 if (tp->snd_una == tp->snd_max) 2074 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2075 else if (TCP_TIMER_ISARMED(tp, 2076 TCPT_PERSIST) == 0) 2077 TCP_TIMER_ARM(tp, TCPT_REXMT, 2078 tp->t_rxtcur); 2079 2080 sowwakeup(so); 2081 if (so->so_snd.sb_cc) { 2082 KERNEL_LOCK(1, NULL); 2083 (void) tcp_output(tp); 2084 KERNEL_UNLOCK_ONE(NULL); 2085 } 2086 if (tcp_saveti) 2087 m_freem(tcp_saveti); 2088 return; 2089 } 2090 } else if (th->th_ack == tp->snd_una && 2091 TAILQ_FIRST(&tp->segq) == NULL && 2092 tlen <= sbspace(&so->so_rcv)) { 2093 int newsize = 0; /* automatic sockbuf scaling */ 2094 2095 /* 2096 * this is a pure, in-sequence data packet 2097 * with nothing on the reassembly queue and 2098 * we have enough buffer space to take it. 2099 */ 2100 tp->rcv_nxt += tlen; 2101 tcps = TCP_STAT_GETREF(); 2102 tcps[TCP_STAT_PREDDAT]++; 2103 tcps[TCP_STAT_RCVPACK]++; 2104 tcps[TCP_STAT_RCVBYTE] += tlen; 2105 TCP_STAT_PUTREF(); 2106 nd6_hint(tp); 2107 2108 /* 2109 * Automatic sizing enables the performance of large buffers 2110 * and most of the efficiency of small ones by only allocating 2111 * space when it is needed. 2112 * 2113 * On the receive side the socket buffer memory is only rarely 2114 * used to any significant extent. This allows us to be much 2115 * more aggressive in scaling the receive socket buffer. For 2116 * the case that the buffer space is actually used to a large 2117 * extent and we run out of kernel memory we can simply drop 2118 * the new segments; TCP on the sender will just retransmit it 2119 * later. Setting the buffer size too big may only consume too 2120 * much kernel memory if the application doesn't read() from 2121 * the socket or packet loss or reordering makes use of the 2122 * reassembly queue. 2123 * 2124 * The criteria to step up the receive buffer one notch are: 2125 * 1. the number of bytes received during the time it takes 2126 * one timestamp to be reflected back to us (the RTT); 2127 * 2. received bytes per RTT is within seven eighth of the 2128 * current socket buffer size; 2129 * 3. receive buffer size has not hit maximal automatic size; 2130 * 2131 * This algorithm does one step per RTT at most and only if 2132 * we receive a bulk stream w/o packet losses or reorderings. 2133 * Shrinking the buffer during idle times is not necessary as 2134 * it doesn't consume any memory when idle. 2135 * 2136 * TODO: Only step up if the application is actually serving 2137 * the buffer to better manage the socket buffer resources. 2138 */ 2139 if (tcp_do_autorcvbuf && 2140 opti.ts_ecr && 2141 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 2142 if (opti.ts_ecr > tp->rfbuf_ts && 2143 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 2144 if (tp->rfbuf_cnt > 2145 (so->so_rcv.sb_hiwat / 8 * 7) && 2146 so->so_rcv.sb_hiwat < 2147 tcp_autorcvbuf_max) { 2148 newsize = 2149 min(so->so_rcv.sb_hiwat + 2150 tcp_autorcvbuf_inc, 2151 tcp_autorcvbuf_max); 2152 } 2153 /* Start over with next RTT. */ 2154 tp->rfbuf_ts = 0; 2155 tp->rfbuf_cnt = 0; 2156 } else 2157 tp->rfbuf_cnt += tlen; /* add up */ 2158 } 2159 2160 /* 2161 * Drop TCP, IP headers and TCP options then add data 2162 * to socket buffer. 2163 */ 2164 if (so->so_state & SS_CANTRCVMORE) 2165 m_freem(m); 2166 else { 2167 /* 2168 * Set new socket buffer size. 2169 * Give up when limit is reached. 2170 */ 2171 if (newsize) 2172 if (!sbreserve(&so->so_rcv, 2173 newsize, so)) 2174 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 2175 m_adj(m, toff + off); 2176 sbappendstream(&so->so_rcv, m); 2177 } 2178 sorwakeup(so); 2179 tcp_setup_ack(tp, th); 2180 if (tp->t_flags & TF_ACKNOW) { 2181 KERNEL_LOCK(1, NULL); 2182 (void) tcp_output(tp); 2183 KERNEL_UNLOCK_ONE(NULL); 2184 } 2185 if (tcp_saveti) 2186 m_freem(tcp_saveti); 2187 return; 2188 } 2189 } 2190 2191 /* 2192 * Compute mbuf offset to TCP data segment. 2193 */ 2194 hdroptlen = toff + off; 2195 2196 /* 2197 * Calculate amount of space in receive window, 2198 * and then do TCP input processing. 2199 * Receive window is amount of space in rcv queue, 2200 * but not less than advertised window. 2201 */ 2202 { int win; 2203 2204 win = sbspace(&so->so_rcv); 2205 if (win < 0) 2206 win = 0; 2207 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 2208 } 2209 2210 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 2211 tp->rfbuf_ts = 0; 2212 tp->rfbuf_cnt = 0; 2213 2214 switch (tp->t_state) { 2215 /* 2216 * If the state is SYN_SENT: 2217 * if seg contains an ACK, but not for our SYN, drop the input. 2218 * if seg contains a RST, then drop the connection. 2219 * if seg does not contain SYN, then drop it. 2220 * Otherwise this is an acceptable SYN segment 2221 * initialize tp->rcv_nxt and tp->irs 2222 * if seg contains ack then advance tp->snd_una 2223 * if seg contains a ECE and ECN support is enabled, the stream 2224 * is ECN capable. 2225 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2226 * arrange for segment to be acked (eventually) 2227 * continue processing rest of data/controls, beginning with URG 2228 */ 2229 case TCPS_SYN_SENT: 2230 if ((tiflags & TH_ACK) && 2231 (SEQ_LEQ(th->th_ack, tp->iss) || 2232 SEQ_GT(th->th_ack, tp->snd_max))) 2233 goto dropwithreset; 2234 if (tiflags & TH_RST) { 2235 if (tiflags & TH_ACK) 2236 tp = tcp_drop(tp, ECONNREFUSED); 2237 goto drop; 2238 } 2239 if ((tiflags & TH_SYN) == 0) 2240 goto drop; 2241 if (tiflags & TH_ACK) { 2242 tp->snd_una = th->th_ack; 2243 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2244 tp->snd_nxt = tp->snd_una; 2245 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2246 tp->snd_high = tp->snd_una; 2247 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2248 2249 if ((tiflags & TH_ECE) && tcp_do_ecn) { 2250 tp->t_flags |= TF_ECN_PERMIT; 2251 TCP_STATINC(TCP_STAT_ECN_SHS); 2252 } 2253 2254 } 2255 tp->irs = th->th_seq; 2256 tcp_rcvseqinit(tp); 2257 tp->t_flags |= TF_ACKNOW; 2258 tcp_mss_from_peer(tp, opti.maxseg); 2259 2260 /* 2261 * Initialize the initial congestion window. If we 2262 * had to retransmit the SYN, we must initialize cwnd 2263 * to 1 segment (i.e. the Loss Window). 2264 */ 2265 if (tp->t_flags & TF_SYN_REXMT) 2266 tp->snd_cwnd = tp->t_peermss; 2267 else { 2268 int ss = tcp_init_win; 2269 #ifdef INET 2270 if (inp != NULL && in_localaddr(inp->inp_faddr)) 2271 ss = tcp_init_win_local; 2272 #endif 2273 #ifdef INET6 2274 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 2275 ss = tcp_init_win_local; 2276 #endif 2277 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 2278 } 2279 2280 tcp_rmx_rtt(tp); 2281 if (tiflags & TH_ACK) { 2282 TCP_STATINC(TCP_STAT_CONNECTS); 2283 /* 2284 * move tcp_established before soisconnected 2285 * because upcall handler can drive tcp_output 2286 * functionality. 2287 * XXX we might call soisconnected at the end of 2288 * all processing 2289 */ 2290 tcp_established(tp); 2291 soisconnected(so); 2292 /* Do window scaling on this connection? */ 2293 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2294 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2295 tp->snd_scale = tp->requested_s_scale; 2296 tp->rcv_scale = tp->request_r_scale; 2297 } 2298 TCP_REASS_LOCK(tp); 2299 (void) tcp_reass(tp, NULL, NULL, &tlen); 2300 /* 2301 * if we didn't have to retransmit the SYN, 2302 * use its rtt as our initial srtt & rtt var. 2303 */ 2304 if (tp->t_rtttime) 2305 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2306 } else 2307 tp->t_state = TCPS_SYN_RECEIVED; 2308 2309 /* 2310 * Advance th->th_seq to correspond to first data byte. 2311 * If data, trim to stay within window, 2312 * dropping FIN if necessary. 2313 */ 2314 th->th_seq++; 2315 if (tlen > tp->rcv_wnd) { 2316 todrop = tlen - tp->rcv_wnd; 2317 m_adj(m, -todrop); 2318 tlen = tp->rcv_wnd; 2319 tiflags &= ~TH_FIN; 2320 tcps = TCP_STAT_GETREF(); 2321 tcps[TCP_STAT_RCVPACKAFTERWIN]++; 2322 tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop; 2323 TCP_STAT_PUTREF(); 2324 } 2325 tp->snd_wl1 = th->th_seq - 1; 2326 tp->rcv_up = th->th_seq; 2327 goto step6; 2328 2329 /* 2330 * If the state is SYN_RECEIVED: 2331 * If seg contains an ACK, but not for our SYN, drop the input 2332 * and generate an RST. See page 36, rfc793 2333 */ 2334 case TCPS_SYN_RECEIVED: 2335 if ((tiflags & TH_ACK) && 2336 (SEQ_LEQ(th->th_ack, tp->iss) || 2337 SEQ_GT(th->th_ack, tp->snd_max))) 2338 goto dropwithreset; 2339 break; 2340 } 2341 2342 /* 2343 * States other than LISTEN or SYN_SENT. 2344 * First check timestamp, if present. 2345 * Then check that at least some bytes of segment are within 2346 * receive window. If segment begins before rcv_nxt, 2347 * drop leading data (and SYN); if nothing left, just ack. 2348 * 2349 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2350 * and it's less than ts_recent, drop it. 2351 */ 2352 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2353 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2354 2355 /* Check to see if ts_recent is over 24 days old. */ 2356 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2357 /* 2358 * Invalidate ts_recent. If this segment updates 2359 * ts_recent, the age will be reset later and ts_recent 2360 * will get a valid value. If it does not, setting 2361 * ts_recent to zero will at least satisfy the 2362 * requirement that zero be placed in the timestamp 2363 * echo reply when ts_recent isn't valid. The 2364 * age isn't reset until we get a valid ts_recent 2365 * because we don't want out-of-order segments to be 2366 * dropped when ts_recent is old. 2367 */ 2368 tp->ts_recent = 0; 2369 } else { 2370 tcps = TCP_STAT_GETREF(); 2371 tcps[TCP_STAT_RCVDUPPACK]++; 2372 tcps[TCP_STAT_RCVDUPBYTE] += tlen; 2373 tcps[TCP_STAT_PAWSDROP]++; 2374 TCP_STAT_PUTREF(); 2375 tcp_new_dsack(tp, th->th_seq, tlen); 2376 goto dropafterack; 2377 } 2378 } 2379 2380 todrop = tp->rcv_nxt - th->th_seq; 2381 dupseg = false; 2382 if (todrop > 0) { 2383 if (tiflags & TH_SYN) { 2384 tiflags &= ~TH_SYN; 2385 th->th_seq++; 2386 if (th->th_urp > 1) 2387 th->th_urp--; 2388 else { 2389 tiflags &= ~TH_URG; 2390 th->th_urp = 0; 2391 } 2392 todrop--; 2393 } 2394 if (todrop > tlen || 2395 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2396 /* 2397 * Any valid FIN or RST must be to the left of the 2398 * window. At this point the FIN or RST must be a 2399 * duplicate or out of sequence; drop it. 2400 */ 2401 if (tiflags & TH_RST) 2402 goto drop; 2403 tiflags &= ~(TH_FIN|TH_RST); 2404 /* 2405 * Send an ACK to resynchronize and drop any data. 2406 * But keep on processing for RST or ACK. 2407 */ 2408 tp->t_flags |= TF_ACKNOW; 2409 todrop = tlen; 2410 dupseg = true; 2411 tcps = TCP_STAT_GETREF(); 2412 tcps[TCP_STAT_RCVDUPPACK]++; 2413 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 2414 TCP_STAT_PUTREF(); 2415 } else if ((tiflags & TH_RST) && 2416 th->th_seq != tp->rcv_nxt) { 2417 /* 2418 * Test for reset before adjusting the sequence 2419 * number for overlapping data. 2420 */ 2421 goto dropafterack_ratelim; 2422 } else { 2423 tcps = TCP_STAT_GETREF(); 2424 tcps[TCP_STAT_RCVPARTDUPPACK]++; 2425 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 2426 TCP_STAT_PUTREF(); 2427 } 2428 tcp_new_dsack(tp, th->th_seq, todrop); 2429 hdroptlen += todrop; /*drop from head afterwards*/ 2430 th->th_seq += todrop; 2431 tlen -= todrop; 2432 if (th->th_urp > todrop) 2433 th->th_urp -= todrop; 2434 else { 2435 tiflags &= ~TH_URG; 2436 th->th_urp = 0; 2437 } 2438 } 2439 2440 /* 2441 * If new data are received on a connection after the 2442 * user processes are gone, then RST the other end. 2443 */ 2444 if ((so->so_state & SS_NOFDREF) && 2445 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2446 tp = tcp_close(tp); 2447 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 2448 goto dropwithreset; 2449 } 2450 2451 /* 2452 * If segment ends after window, drop trailing data 2453 * (and PUSH and FIN); if nothing left, just ACK. 2454 */ 2455 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 2456 if (todrop > 0) { 2457 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 2458 if (todrop >= tlen) { 2459 /* 2460 * The segment actually starts after the window. 2461 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2462 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2463 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2464 */ 2465 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 2466 /* 2467 * If a new connection request is received 2468 * while in TIME_WAIT, drop the old connection 2469 * and start over if the sequence numbers 2470 * are above the previous ones. 2471 * 2472 * NOTE: We will checksum the packet again, and 2473 * so we need to put the header fields back into 2474 * network order! 2475 * XXX This kind of sucks, but we don't expect 2476 * XXX this to happen very often, so maybe it 2477 * XXX doesn't matter so much. 2478 */ 2479 if (tiflags & TH_SYN && 2480 tp->t_state == TCPS_TIME_WAIT && 2481 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2482 tp = tcp_close(tp); 2483 tcp_fields_to_net(th); 2484 goto findpcb; 2485 } 2486 /* 2487 * If window is closed can only take segments at 2488 * window edge, and have to drop data and PUSH from 2489 * incoming segments. Continue processing, but 2490 * remember to ack. Otherwise, drop segment 2491 * and (if not RST) ack. 2492 */ 2493 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2494 tp->t_flags |= TF_ACKNOW; 2495 TCP_STATINC(TCP_STAT_RCVWINPROBE); 2496 } else 2497 goto dropafterack; 2498 } else 2499 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 2500 m_adj(m, -todrop); 2501 tlen -= todrop; 2502 tiflags &= ~(TH_PUSH|TH_FIN); 2503 } 2504 2505 /* 2506 * If last ACK falls within this segment's sequence numbers, 2507 * record the timestamp. 2508 * NOTE: 2509 * 1) That the test incorporates suggestions from the latest 2510 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2511 * 2) That updating only on newer timestamps interferes with 2512 * our earlier PAWS tests, so this check should be solely 2513 * predicated on the sequence space of this segment. 2514 * 3) That we modify the segment boundary check to be 2515 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2516 * instead of RFC1323's 2517 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2518 * This modified check allows us to overcome RFC1323's 2519 * limitations as described in Stevens TCP/IP Illustrated 2520 * Vol. 2 p.869. In such cases, we can still calculate the 2521 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2522 */ 2523 if (opti.ts_present && 2524 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2525 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2526 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2527 tp->ts_recent_age = tcp_now; 2528 tp->ts_recent = opti.ts_val; 2529 } 2530 2531 /* 2532 * If the RST bit is set examine the state: 2533 * SYN_RECEIVED STATE: 2534 * If passive open, return to LISTEN state. 2535 * If active open, inform user that connection was refused. 2536 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 2537 * Inform user that connection was reset, and close tcb. 2538 * CLOSING, LAST_ACK, TIME_WAIT STATES 2539 * Close the tcb. 2540 */ 2541 if (tiflags & TH_RST) { 2542 if (th->th_seq != tp->rcv_nxt) 2543 goto dropafterack_ratelim; 2544 2545 switch (tp->t_state) { 2546 case TCPS_SYN_RECEIVED: 2547 so->so_error = ECONNREFUSED; 2548 goto close; 2549 2550 case TCPS_ESTABLISHED: 2551 case TCPS_FIN_WAIT_1: 2552 case TCPS_FIN_WAIT_2: 2553 case TCPS_CLOSE_WAIT: 2554 so->so_error = ECONNRESET; 2555 close: 2556 tp->t_state = TCPS_CLOSED; 2557 TCP_STATINC(TCP_STAT_DROPS); 2558 tp = tcp_close(tp); 2559 goto drop; 2560 2561 case TCPS_CLOSING: 2562 case TCPS_LAST_ACK: 2563 case TCPS_TIME_WAIT: 2564 tp = tcp_close(tp); 2565 goto drop; 2566 } 2567 } 2568 2569 /* 2570 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2571 * we must be in a synchronized state. RFC791 states (under RST 2572 * generation) that any unacceptable segment (an out-of-order SYN 2573 * qualifies) received in a synchronized state must elicit only an 2574 * empty acknowledgment segment ... and the connection remains in 2575 * the same state. 2576 */ 2577 if (tiflags & TH_SYN) { 2578 if (tp->rcv_nxt == th->th_seq) { 2579 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2580 TH_ACK); 2581 if (tcp_saveti) 2582 m_freem(tcp_saveti); 2583 return; 2584 } 2585 2586 goto dropafterack_ratelim; 2587 } 2588 2589 /* 2590 * If the ACK bit is off we drop the segment and return. 2591 */ 2592 if ((tiflags & TH_ACK) == 0) { 2593 if (tp->t_flags & TF_ACKNOW) 2594 goto dropafterack; 2595 else 2596 goto drop; 2597 } 2598 2599 /* 2600 * Ack processing. 2601 */ 2602 switch (tp->t_state) { 2603 2604 /* 2605 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2606 * ESTABLISHED state and continue processing, otherwise 2607 * send an RST. 2608 */ 2609 case TCPS_SYN_RECEIVED: 2610 if (SEQ_GT(tp->snd_una, th->th_ack) || 2611 SEQ_GT(th->th_ack, tp->snd_max)) 2612 goto dropwithreset; 2613 TCP_STATINC(TCP_STAT_CONNECTS); 2614 soisconnected(so); 2615 tcp_established(tp); 2616 /* Do window scaling? */ 2617 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2618 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2619 tp->snd_scale = tp->requested_s_scale; 2620 tp->rcv_scale = tp->request_r_scale; 2621 } 2622 TCP_REASS_LOCK(tp); 2623 (void) tcp_reass(tp, NULL, NULL, &tlen); 2624 tp->snd_wl1 = th->th_seq - 1; 2625 /* fall into ... */ 2626 2627 /* 2628 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2629 * ACKs. If the ack is in the range 2630 * tp->snd_una < th->th_ack <= tp->snd_max 2631 * then advance tp->snd_una to th->th_ack and drop 2632 * data from the retransmission queue. If this ACK reflects 2633 * more up to date window information we update our window information. 2634 */ 2635 case TCPS_ESTABLISHED: 2636 case TCPS_FIN_WAIT_1: 2637 case TCPS_FIN_WAIT_2: 2638 case TCPS_CLOSE_WAIT: 2639 case TCPS_CLOSING: 2640 case TCPS_LAST_ACK: 2641 case TCPS_TIME_WAIT: 2642 2643 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2644 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2645 TCP_STATINC(TCP_STAT_RCVDUPACK); 2646 /* 2647 * If we have outstanding data (other than 2648 * a window probe), this is a completely 2649 * duplicate ack (ie, window info didn't 2650 * change), the ack is the biggest we've 2651 * seen and we've seen exactly our rexmt 2652 * threshhold of them, assume a packet 2653 * has been dropped and retransmit it. 2654 * Kludge snd_nxt & the congestion 2655 * window so we send only this one 2656 * packet. 2657 */ 2658 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2659 th->th_ack != tp->snd_una) 2660 tp->t_dupacks = 0; 2661 else if (tp->t_partialacks < 0 && 2662 (++tp->t_dupacks == tcprexmtthresh || 2663 TCP_FACK_FASTRECOV(tp))) { 2664 /* 2665 * Do the fast retransmit, and adjust 2666 * congestion control paramenters. 2667 */ 2668 if (tp->t_congctl->fast_retransmit(tp, th)) { 2669 /* False fast retransmit */ 2670 break; 2671 } else 2672 goto drop; 2673 } else if (tp->t_dupacks > tcprexmtthresh) { 2674 tp->snd_cwnd += tp->t_segsz; 2675 KERNEL_LOCK(1, NULL); 2676 (void) tcp_output(tp); 2677 KERNEL_UNLOCK_ONE(NULL); 2678 goto drop; 2679 } 2680 } else { 2681 /* 2682 * If the ack appears to be very old, only 2683 * allow data that is in-sequence. This 2684 * makes it somewhat more difficult to insert 2685 * forged data by guessing sequence numbers. 2686 * Sent an ack to try to update the send 2687 * sequence number on the other side. 2688 */ 2689 if (tlen && th->th_seq != tp->rcv_nxt && 2690 SEQ_LT(th->th_ack, 2691 tp->snd_una - tp->max_sndwnd)) 2692 goto dropafterack; 2693 } 2694 break; 2695 } 2696 /* 2697 * If the congestion window was inflated to account 2698 * for the other side's cached packets, retract it. 2699 */ 2700 tp->t_congctl->fast_retransmit_newack(tp, th); 2701 2702 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2703 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2704 goto dropafterack; 2705 } 2706 acked = th->th_ack - tp->snd_una; 2707 tcps = TCP_STAT_GETREF(); 2708 tcps[TCP_STAT_RCVACKPACK]++; 2709 tcps[TCP_STAT_RCVACKBYTE] += acked; 2710 TCP_STAT_PUTREF(); 2711 2712 /* 2713 * If we have a timestamp reply, update smoothed 2714 * round trip time. If no timestamp is present but 2715 * transmit timer is running and timed sequence 2716 * number was acked, update smoothed round trip time. 2717 * Since we now have an rtt measurement, cancel the 2718 * timer backoff (cf., Phil Karn's retransmit alg.). 2719 * Recompute the initial retransmit timer. 2720 */ 2721 if (ts_rtt) 2722 tcp_xmit_timer(tp, ts_rtt - 1); 2723 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2724 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2725 2726 /* 2727 * If all outstanding data is acked, stop retransmit 2728 * timer and remember to restart (more output or persist). 2729 * If there is more data to be acked, restart retransmit 2730 * timer, using current (possibly backed-off) value. 2731 */ 2732 if (th->th_ack == tp->snd_max) { 2733 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2734 needoutput = 1; 2735 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2736 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2737 2738 /* 2739 * New data has been acked, adjust the congestion window. 2740 */ 2741 tp->t_congctl->newack(tp, th); 2742 2743 nd6_hint(tp); 2744 if (acked > so->so_snd.sb_cc) { 2745 tp->snd_wnd -= so->so_snd.sb_cc; 2746 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2747 ourfinisacked = 1; 2748 } else { 2749 if (acked > (tp->t_lastoff - tp->t_inoff)) 2750 tp->t_lastm = NULL; 2751 sbdrop(&so->so_snd, acked); 2752 tp->t_lastoff -= acked; 2753 if (tp->snd_wnd > acked) 2754 tp->snd_wnd -= acked; 2755 else 2756 tp->snd_wnd = 0; 2757 ourfinisacked = 0; 2758 } 2759 sowwakeup(so); 2760 2761 icmp_check(tp, th, acked); 2762 2763 tp->snd_una = th->th_ack; 2764 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2765 tp->snd_fack = tp->snd_una; 2766 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2767 tp->snd_nxt = tp->snd_una; 2768 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2769 tp->snd_high = tp->snd_una; 2770 2771 switch (tp->t_state) { 2772 2773 /* 2774 * In FIN_WAIT_1 STATE in addition to the processing 2775 * for the ESTABLISHED state if our FIN is now acknowledged 2776 * then enter FIN_WAIT_2. 2777 */ 2778 case TCPS_FIN_WAIT_1: 2779 if (ourfinisacked) { 2780 /* 2781 * If we can't receive any more 2782 * data, then closing user can proceed. 2783 * Starting the timer is contrary to the 2784 * specification, but if we don't get a FIN 2785 * we'll hang forever. 2786 */ 2787 if (so->so_state & SS_CANTRCVMORE) { 2788 soisdisconnected(so); 2789 if (tp->t_maxidle > 0) 2790 TCP_TIMER_ARM(tp, TCPT_2MSL, 2791 tp->t_maxidle); 2792 } 2793 tp->t_state = TCPS_FIN_WAIT_2; 2794 } 2795 break; 2796 2797 /* 2798 * In CLOSING STATE in addition to the processing for 2799 * the ESTABLISHED state if the ACK acknowledges our FIN 2800 * then enter the TIME-WAIT state, otherwise ignore 2801 * the segment. 2802 */ 2803 case TCPS_CLOSING: 2804 if (ourfinisacked) { 2805 tp->t_state = TCPS_TIME_WAIT; 2806 tcp_canceltimers(tp); 2807 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2808 soisdisconnected(so); 2809 } 2810 break; 2811 2812 /* 2813 * In LAST_ACK, we may still be waiting for data to drain 2814 * and/or to be acked, as well as for the ack of our FIN. 2815 * If our FIN is now acknowledged, delete the TCB, 2816 * enter the closed state and return. 2817 */ 2818 case TCPS_LAST_ACK: 2819 if (ourfinisacked) { 2820 tp = tcp_close(tp); 2821 goto drop; 2822 } 2823 break; 2824 2825 /* 2826 * In TIME_WAIT state the only thing that should arrive 2827 * is a retransmission of the remote FIN. Acknowledge 2828 * it and restart the finack timer. 2829 */ 2830 case TCPS_TIME_WAIT: 2831 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2832 goto dropafterack; 2833 } 2834 } 2835 2836 step6: 2837 /* 2838 * Update window information. 2839 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2840 */ 2841 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2842 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2843 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2844 /* keep track of pure window updates */ 2845 if (tlen == 0 && 2846 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2847 TCP_STATINC(TCP_STAT_RCVWINUPD); 2848 tp->snd_wnd = tiwin; 2849 tp->snd_wl1 = th->th_seq; 2850 tp->snd_wl2 = th->th_ack; 2851 if (tp->snd_wnd > tp->max_sndwnd) 2852 tp->max_sndwnd = tp->snd_wnd; 2853 needoutput = 1; 2854 } 2855 2856 /* 2857 * Process segments with URG. 2858 */ 2859 if ((tiflags & TH_URG) && th->th_urp && 2860 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2861 /* 2862 * This is a kludge, but if we receive and accept 2863 * random urgent pointers, we'll crash in 2864 * soreceive. It's hard to imagine someone 2865 * actually wanting to send this much urgent data. 2866 */ 2867 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2868 th->th_urp = 0; /* XXX */ 2869 tiflags &= ~TH_URG; /* XXX */ 2870 goto dodata; /* XXX */ 2871 } 2872 /* 2873 * If this segment advances the known urgent pointer, 2874 * then mark the data stream. This should not happen 2875 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2876 * a FIN has been received from the remote side. 2877 * In these states we ignore the URG. 2878 * 2879 * According to RFC961 (Assigned Protocols), 2880 * the urgent pointer points to the last octet 2881 * of urgent data. We continue, however, 2882 * to consider it to indicate the first octet 2883 * of data past the urgent section as the original 2884 * spec states (in one of two places). 2885 */ 2886 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2887 tp->rcv_up = th->th_seq + th->th_urp; 2888 so->so_oobmark = so->so_rcv.sb_cc + 2889 (tp->rcv_up - tp->rcv_nxt) - 1; 2890 if (so->so_oobmark == 0) 2891 so->so_state |= SS_RCVATMARK; 2892 sohasoutofband(so); 2893 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2894 } 2895 /* 2896 * Remove out of band data so doesn't get presented to user. 2897 * This can happen independent of advancing the URG pointer, 2898 * but if two URG's are pending at once, some out-of-band 2899 * data may creep in... ick. 2900 */ 2901 if (th->th_urp <= (u_int16_t) tlen 2902 #ifdef SO_OOBINLINE 2903 && (so->so_options & SO_OOBINLINE) == 0 2904 #endif 2905 ) 2906 tcp_pulloutofband(so, th, m, hdroptlen); 2907 } else 2908 /* 2909 * If no out of band data is expected, 2910 * pull receive urgent pointer along 2911 * with the receive window. 2912 */ 2913 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2914 tp->rcv_up = tp->rcv_nxt; 2915 dodata: /* XXX */ 2916 2917 /* 2918 * Process the segment text, merging it into the TCP sequencing queue, 2919 * and arranging for acknowledgement of receipt if necessary. 2920 * This process logically involves adjusting tp->rcv_wnd as data 2921 * is presented to the user (this happens in tcp_usrreq.c, 2922 * tcp_rcvd()). If a FIN has already been received on this 2923 * connection then we just ignore the text. 2924 */ 2925 if ((tlen || (tiflags & TH_FIN)) && 2926 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2927 /* 2928 * Insert segment ti into reassembly queue of tcp with 2929 * control block tp. Return TH_FIN if reassembly now includes 2930 * a segment with FIN. The macro form does the common case 2931 * inline (segment is the next to be received on an 2932 * established connection, and the queue is empty), 2933 * avoiding linkage into and removal from the queue and 2934 * repetition of various conversions. 2935 * Set DELACK for segments received in order, but ack 2936 * immediately when segments are out of order 2937 * (so fast retransmit can work). 2938 */ 2939 /* NOTE: this was TCP_REASS() macro, but used only once */ 2940 TCP_REASS_LOCK(tp); 2941 if (th->th_seq == tp->rcv_nxt && 2942 TAILQ_FIRST(&tp->segq) == NULL && 2943 tp->t_state == TCPS_ESTABLISHED) { 2944 tcp_setup_ack(tp, th); 2945 tp->rcv_nxt += tlen; 2946 tiflags = th->th_flags & TH_FIN; 2947 tcps = TCP_STAT_GETREF(); 2948 tcps[TCP_STAT_RCVPACK]++; 2949 tcps[TCP_STAT_RCVBYTE] += tlen; 2950 TCP_STAT_PUTREF(); 2951 nd6_hint(tp); 2952 if (so->so_state & SS_CANTRCVMORE) 2953 m_freem(m); 2954 else { 2955 m_adj(m, hdroptlen); 2956 sbappendstream(&(so)->so_rcv, m); 2957 } 2958 TCP_REASS_UNLOCK(tp); 2959 sorwakeup(so); 2960 } else { 2961 m_adj(m, hdroptlen); 2962 tiflags = tcp_reass(tp, th, m, &tlen); 2963 tp->t_flags |= TF_ACKNOW; 2964 } 2965 2966 /* 2967 * Note the amount of data that peer has sent into 2968 * our window, in order to estimate the sender's 2969 * buffer size. 2970 */ 2971 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2972 } else { 2973 m_freem(m); 2974 m = NULL; 2975 tiflags &= ~TH_FIN; 2976 } 2977 2978 /* 2979 * If FIN is received ACK the FIN and let the user know 2980 * that the connection is closing. Ignore a FIN received before 2981 * the connection is fully established. 2982 */ 2983 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2984 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2985 socantrcvmore(so); 2986 tp->t_flags |= TF_ACKNOW; 2987 tp->rcv_nxt++; 2988 } 2989 switch (tp->t_state) { 2990 2991 /* 2992 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2993 */ 2994 case TCPS_ESTABLISHED: 2995 tp->t_state = TCPS_CLOSE_WAIT; 2996 break; 2997 2998 /* 2999 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3000 * enter the CLOSING state. 3001 */ 3002 case TCPS_FIN_WAIT_1: 3003 tp->t_state = TCPS_CLOSING; 3004 break; 3005 3006 /* 3007 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3008 * starting the time-wait timer, turning off the other 3009 * standard timers. 3010 */ 3011 case TCPS_FIN_WAIT_2: 3012 tp->t_state = TCPS_TIME_WAIT; 3013 tcp_canceltimers(tp); 3014 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 3015 soisdisconnected(so); 3016 break; 3017 3018 /* 3019 * In TIME_WAIT state restart the 2 MSL time_wait timer. 3020 */ 3021 case TCPS_TIME_WAIT: 3022 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 3023 break; 3024 } 3025 } 3026 #ifdef TCP_DEBUG 3027 if (so->so_options & SO_DEBUG) 3028 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 3029 #endif 3030 3031 /* 3032 * Return any desired output. 3033 */ 3034 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 3035 KERNEL_LOCK(1, NULL); 3036 (void) tcp_output(tp); 3037 KERNEL_UNLOCK_ONE(NULL); 3038 } 3039 if (tcp_saveti) 3040 m_freem(tcp_saveti); 3041 3042 if (tp->t_state == TCPS_TIME_WAIT 3043 && (so->so_state & SS_NOFDREF) 3044 && (tp->t_inpcb || af != AF_INET) 3045 && (tp->t_in6pcb || af != AF_INET6) 3046 && ((af == AF_INET ? tcp4_vtw_enable : tcp6_vtw_enable) & 1) != 0 3047 && TAILQ_EMPTY(&tp->segq) 3048 && vtw_add(af, tp)) { 3049 ; 3050 } 3051 return; 3052 3053 badsyn: 3054 /* 3055 * Received a bad SYN. Increment counters and dropwithreset. 3056 */ 3057 TCP_STATINC(TCP_STAT_BADSYN); 3058 tp = NULL; 3059 goto dropwithreset; 3060 3061 dropafterack: 3062 /* 3063 * Generate an ACK dropping incoming segment if it occupies 3064 * sequence space, where the ACK reflects our state. 3065 */ 3066 if (tiflags & TH_RST) 3067 goto drop; 3068 goto dropafterack2; 3069 3070 dropafterack_ratelim: 3071 /* 3072 * We may want to rate-limit ACKs against SYN/RST attack. 3073 */ 3074 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 3075 tcp_ackdrop_ppslim) == 0) { 3076 /* XXX stat */ 3077 goto drop; 3078 } 3079 /* ...fall into dropafterack2... */ 3080 3081 dropafterack2: 3082 m_freem(m); 3083 tp->t_flags |= TF_ACKNOW; 3084 KERNEL_LOCK(1, NULL); 3085 (void) tcp_output(tp); 3086 KERNEL_UNLOCK_ONE(NULL); 3087 if (tcp_saveti) 3088 m_freem(tcp_saveti); 3089 return; 3090 3091 dropwithreset_ratelim: 3092 /* 3093 * We may want to rate-limit RSTs in certain situations, 3094 * particularly if we are sending an RST in response to 3095 * an attempt to connect to or otherwise communicate with 3096 * a port for which we have no socket. 3097 */ 3098 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 3099 tcp_rst_ppslim) == 0) { 3100 /* XXX stat */ 3101 goto drop; 3102 } 3103 /* ...fall into dropwithreset... */ 3104 3105 dropwithreset: 3106 /* 3107 * Generate a RST, dropping incoming segment. 3108 * Make ACK acceptable to originator of segment. 3109 */ 3110 if (tiflags & TH_RST) 3111 goto drop; 3112 3113 switch (af) { 3114 #ifdef INET6 3115 case AF_INET6: 3116 /* For following calls to tcp_respond */ 3117 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 3118 goto drop; 3119 break; 3120 #endif /* INET6 */ 3121 case AF_INET: 3122 if (IN_MULTICAST(ip->ip_dst.s_addr) || 3123 in_broadcast(ip->ip_dst, m_get_rcvif_NOMPSAFE(m))) 3124 goto drop; 3125 } 3126 3127 if (tiflags & TH_ACK) 3128 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3129 else { 3130 if (tiflags & TH_SYN) 3131 tlen++; 3132 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 3133 TH_RST|TH_ACK); 3134 } 3135 if (tcp_saveti) 3136 m_freem(tcp_saveti); 3137 return; 3138 3139 badcsum: 3140 drop: 3141 /* 3142 * Drop space held by incoming segment and return. 3143 */ 3144 if (tp) { 3145 if (tp->t_inpcb) 3146 so = tp->t_inpcb->inp_socket; 3147 #ifdef INET6 3148 else if (tp->t_in6pcb) 3149 so = tp->t_in6pcb->in6p_socket; 3150 #endif 3151 else 3152 so = NULL; 3153 #ifdef TCP_DEBUG 3154 if (so && (so->so_options & SO_DEBUG) != 0) 3155 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 3156 #endif 3157 } 3158 if (tcp_saveti) 3159 m_freem(tcp_saveti); 3160 m_freem(m); 3161 return; 3162 } 3163 3164 #ifdef TCP_SIGNATURE 3165 int 3166 tcp_signature_apply(void *fstate, void *data, u_int len) 3167 { 3168 3169 MD5Update(fstate, (u_char *)data, len); 3170 return (0); 3171 } 3172 3173 struct secasvar * 3174 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 3175 { 3176 struct ip *ip; 3177 struct ip6_hdr *ip6; 3178 3179 ip = mtod(m, struct ip *); 3180 switch (ip->ip_v) { 3181 case 4: 3182 ip = mtod(m, struct ip *); 3183 ip6 = NULL; 3184 break; 3185 case 6: 3186 ip = NULL; 3187 ip6 = mtod(m, struct ip6_hdr *); 3188 break; 3189 default: 3190 return (NULL); 3191 } 3192 3193 #ifdef IPSEC 3194 union sockaddr_union dst; 3195 3196 /* Extract the destination from the IP header in the mbuf. */ 3197 memset(&dst, 0, sizeof(union sockaddr_union)); 3198 if (ip != NULL) { 3199 dst.sa.sa_len = sizeof(struct sockaddr_in); 3200 dst.sa.sa_family = AF_INET; 3201 dst.sin.sin_addr = ip->ip_dst; 3202 } else { 3203 dst.sa.sa_len = sizeof(struct sockaddr_in6); 3204 dst.sa.sa_family = AF_INET6; 3205 dst.sin6.sin6_addr = ip6->ip6_dst; 3206 } 3207 3208 /* 3209 * Look up an SADB entry which matches the address of the peer. 3210 */ 3211 return KEY_LOOKUP_SA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI), 0, 0); 3212 #else 3213 return NULL; 3214 #endif 3215 } 3216 3217 int 3218 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 3219 struct secasvar *sav, char *sig) 3220 { 3221 MD5_CTX ctx; 3222 struct ip *ip; 3223 struct ipovly *ipovly; 3224 #ifdef INET6 3225 struct ip6_hdr *ip6; 3226 struct ip6_hdr_pseudo ip6pseudo; 3227 #endif /* INET6 */ 3228 struct ippseudo ippseudo; 3229 struct tcphdr th0; 3230 int l, tcphdrlen; 3231 3232 if (sav == NULL) 3233 return (-1); 3234 3235 tcphdrlen = th->th_off * 4; 3236 3237 switch (mtod(m, struct ip *)->ip_v) { 3238 case 4: 3239 MD5Init(&ctx); 3240 ip = mtod(m, struct ip *); 3241 memset(&ippseudo, 0, sizeof(ippseudo)); 3242 ipovly = (struct ipovly *)ip; 3243 ippseudo.ippseudo_src = ipovly->ih_src; 3244 ippseudo.ippseudo_dst = ipovly->ih_dst; 3245 ippseudo.ippseudo_pad = 0; 3246 ippseudo.ippseudo_p = IPPROTO_TCP; 3247 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 3248 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 3249 break; 3250 #if INET6 3251 case 6: 3252 MD5Init(&ctx); 3253 ip6 = mtod(m, struct ip6_hdr *); 3254 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 3255 ip6pseudo.ip6ph_src = ip6->ip6_src; 3256 in6_clearscope(&ip6pseudo.ip6ph_src); 3257 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 3258 in6_clearscope(&ip6pseudo.ip6ph_dst); 3259 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 3260 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 3261 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 3262 break; 3263 #endif /* INET6 */ 3264 default: 3265 return (-1); 3266 } 3267 3268 th0 = *th; 3269 th0.th_sum = 0; 3270 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 3271 3272 l = m->m_pkthdr.len - thoff - tcphdrlen; 3273 if (l > 0) 3274 m_apply(m, thoff + tcphdrlen, 3275 m->m_pkthdr.len - thoff - tcphdrlen, 3276 tcp_signature_apply, &ctx); 3277 3278 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 3279 MD5Final(sig, &ctx); 3280 3281 return (0); 3282 } 3283 #endif 3284 3285 /* 3286 * tcp_dooptions: parse and process tcp options. 3287 * 3288 * returns -1 if this segment should be dropped. (eg. wrong signature) 3289 * otherwise returns 0. 3290 */ 3291 3292 static int 3293 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, 3294 struct tcphdr *th, 3295 struct mbuf *m, int toff, struct tcp_opt_info *oi) 3296 { 3297 u_int16_t mss; 3298 int opt, optlen = 0; 3299 #ifdef TCP_SIGNATURE 3300 void *sigp = NULL; 3301 char sigbuf[TCP_SIGLEN]; 3302 struct secasvar *sav = NULL; 3303 #endif 3304 3305 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 3306 opt = cp[0]; 3307 if (opt == TCPOPT_EOL) 3308 break; 3309 if (opt == TCPOPT_NOP) 3310 optlen = 1; 3311 else { 3312 if (cnt < 2) 3313 break; 3314 optlen = cp[1]; 3315 if (optlen < 2 || optlen > cnt) 3316 break; 3317 } 3318 switch (opt) { 3319 3320 default: 3321 continue; 3322 3323 case TCPOPT_MAXSEG: 3324 if (optlen != TCPOLEN_MAXSEG) 3325 continue; 3326 if (!(th->th_flags & TH_SYN)) 3327 continue; 3328 if (TCPS_HAVERCVDSYN(tp->t_state)) 3329 continue; 3330 bcopy(cp + 2, &mss, sizeof(mss)); 3331 oi->maxseg = ntohs(mss); 3332 break; 3333 3334 case TCPOPT_WINDOW: 3335 if (optlen != TCPOLEN_WINDOW) 3336 continue; 3337 if (!(th->th_flags & TH_SYN)) 3338 continue; 3339 if (TCPS_HAVERCVDSYN(tp->t_state)) 3340 continue; 3341 tp->t_flags |= TF_RCVD_SCALE; 3342 tp->requested_s_scale = cp[2]; 3343 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3344 char buf[INET6_ADDRSTRLEN]; 3345 struct ip *ip = mtod(m, struct ip *); 3346 #ifdef INET6 3347 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 3348 #endif 3349 if (ip) 3350 in_print(buf, sizeof(buf), 3351 &ip->ip_src); 3352 #ifdef INET6 3353 else if (ip6) 3354 in6_print(buf, sizeof(buf), 3355 &ip6->ip6_src); 3356 #endif 3357 else 3358 strlcpy(buf, "(unknown)", sizeof(buf)); 3359 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3360 "assuming %d\n", 3361 tp->requested_s_scale, buf, 3362 TCP_MAX_WINSHIFT); 3363 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3364 } 3365 break; 3366 3367 case TCPOPT_TIMESTAMP: 3368 if (optlen != TCPOLEN_TIMESTAMP) 3369 continue; 3370 oi->ts_present = 1; 3371 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 3372 NTOHL(oi->ts_val); 3373 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 3374 NTOHL(oi->ts_ecr); 3375 3376 if (!(th->th_flags & TH_SYN)) 3377 continue; 3378 if (TCPS_HAVERCVDSYN(tp->t_state)) 3379 continue; 3380 /* 3381 * A timestamp received in a SYN makes 3382 * it ok to send timestamp requests and replies. 3383 */ 3384 tp->t_flags |= TF_RCVD_TSTMP; 3385 tp->ts_recent = oi->ts_val; 3386 tp->ts_recent_age = tcp_now; 3387 break; 3388 3389 case TCPOPT_SACK_PERMITTED: 3390 if (optlen != TCPOLEN_SACK_PERMITTED) 3391 continue; 3392 if (!(th->th_flags & TH_SYN)) 3393 continue; 3394 if (TCPS_HAVERCVDSYN(tp->t_state)) 3395 continue; 3396 if (tcp_do_sack) { 3397 tp->t_flags |= TF_SACK_PERMIT; 3398 tp->t_flags |= TF_WILL_SACK; 3399 } 3400 break; 3401 3402 case TCPOPT_SACK: 3403 tcp_sack_option(tp, th, cp, optlen); 3404 break; 3405 #ifdef TCP_SIGNATURE 3406 case TCPOPT_SIGNATURE: 3407 if (optlen != TCPOLEN_SIGNATURE) 3408 continue; 3409 if (sigp && memcmp(sigp, cp + 2, TCP_SIGLEN)) 3410 return (-1); 3411 3412 sigp = sigbuf; 3413 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3414 tp->t_flags |= TF_SIGNATURE; 3415 break; 3416 #endif 3417 } 3418 } 3419 3420 #ifndef TCP_SIGNATURE 3421 return 0; 3422 #else 3423 if (tp->t_flags & TF_SIGNATURE) { 3424 3425 sav = tcp_signature_getsav(m, th); 3426 3427 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3428 return (-1); 3429 } 3430 3431 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) 3432 goto out; 3433 3434 if (sigp) { 3435 char sig[TCP_SIGLEN]; 3436 3437 tcp_fields_to_net(th); 3438 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3439 tcp_fields_to_host(th); 3440 goto out; 3441 } 3442 tcp_fields_to_host(th); 3443 3444 if (memcmp(sig, sigp, TCP_SIGLEN)) { 3445 TCP_STATINC(TCP_STAT_BADSIG); 3446 goto out; 3447 } else 3448 TCP_STATINC(TCP_STAT_GOODSIG); 3449 3450 key_sa_recordxfer(sav, m); 3451 KEY_FREESAV(&sav); 3452 } 3453 return 0; 3454 out: 3455 if (sav != NULL) 3456 KEY_FREESAV(&sav); 3457 return -1; 3458 #endif 3459 } 3460 3461 /* 3462 * Pull out of band byte out of a segment so 3463 * it doesn't appear in the user's data queue. 3464 * It is still reflected in the segment length for 3465 * sequencing purposes. 3466 */ 3467 void 3468 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3469 struct mbuf *m, int off) 3470 { 3471 int cnt = off + th->th_urp - 1; 3472 3473 while (cnt >= 0) { 3474 if (m->m_len > cnt) { 3475 char *cp = mtod(m, char *) + cnt; 3476 struct tcpcb *tp = sototcpcb(so); 3477 3478 tp->t_iobc = *cp; 3479 tp->t_oobflags |= TCPOOB_HAVEDATA; 3480 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3481 m->m_len--; 3482 return; 3483 } 3484 cnt -= m->m_len; 3485 m = m->m_next; 3486 if (m == 0) 3487 break; 3488 } 3489 panic("tcp_pulloutofband"); 3490 } 3491 3492 /* 3493 * Collect new round-trip time estimate 3494 * and update averages and current timeout. 3495 * 3496 * rtt is in units of slow ticks (typically 500 ms) -- essentially the 3497 * difference of two timestamps. 3498 */ 3499 void 3500 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3501 { 3502 int32_t delta; 3503 3504 TCP_STATINC(TCP_STAT_RTTUPDATED); 3505 if (tp->t_srtt != 0) { 3506 /* 3507 * Compute the amount to add to srtt for smoothing, 3508 * *alpha, or 2^(-TCP_RTT_SHIFT). Because 3509 * srtt is stored in 1/32 slow ticks, we conceptually 3510 * shift left 5 bits, subtract srtt to get the 3511 * diference, and then shift right by TCP_RTT_SHIFT 3512 * (3) to obtain 1/8 of the difference. 3513 */ 3514 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3515 /* 3516 * This can never happen, because delta's lowest 3517 * possible value is 1/8 of t_srtt. But if it does, 3518 * set srtt to some reasonable value, here chosen 3519 * as 1/8 tick. 3520 */ 3521 if ((tp->t_srtt += delta) <= 0) 3522 tp->t_srtt = 1 << 2; 3523 /* 3524 * RFC2988 requires that rttvar be updated first. 3525 * This code is compliant because "delta" is the old 3526 * srtt minus the new observation (scaled). 3527 * 3528 * RFC2988 says: 3529 * rttvar = (1-beta) * rttvar + beta * |srtt-observed| 3530 * 3531 * delta is in units of 1/32 ticks, and has then been 3532 * divided by 8. This is equivalent to being in 1/16s 3533 * units and divided by 4. Subtract from it 1/4 of 3534 * the existing rttvar to form the (signed) amount to 3535 * adjust. 3536 */ 3537 if (delta < 0) 3538 delta = -delta; 3539 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3540 /* 3541 * As with srtt, this should never happen. There is 3542 * no support in RFC2988 for this operation. But 1/4s 3543 * as rttvar when faced with something arguably wrong 3544 * is ok. 3545 */ 3546 if ((tp->t_rttvar += delta) <= 0) 3547 tp->t_rttvar = 1 << 2; 3548 3549 /* 3550 * If srtt exceeds .01 second, ensure we use the 'remote' MSL 3551 * Problem is: it doesn't work. Disabled by defaulting 3552 * tcp_rttlocal to 0; see corresponding code in 3553 * tcp_subr that selects local vs remote in a different way. 3554 * 3555 * The static branch prediction hint here should be removed 3556 * when the rtt estimator is fixed and the rtt_enable code 3557 * is turned back on. 3558 */ 3559 if (__predict_false(tcp_rttlocal) && tcp_msl_enable 3560 && tp->t_srtt > tcp_msl_remote_threshold 3561 && tp->t_msl < tcp_msl_remote) { 3562 tp->t_msl = tcp_msl_remote; 3563 } 3564 } else { 3565 /* 3566 * This is the first measurement. Per RFC2988, 2.2, 3567 * set rtt=R and srtt=R/2. 3568 * For srtt, storage representation is 1/32 ticks, 3569 * so shift left by 5. 3570 * For rttvar, storage representation is 1/16 ticks, 3571 * So shift left by 4, but then right by 1 to halve. 3572 */ 3573 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3574 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3575 } 3576 tp->t_rtttime = 0; 3577 tp->t_rxtshift = 0; 3578 3579 /* 3580 * the retransmit should happen at rtt + 4 * rttvar. 3581 * Because of the way we do the smoothing, srtt and rttvar 3582 * will each average +1/2 tick of bias. When we compute 3583 * the retransmit timer, we want 1/2 tick of rounding and 3584 * 1 extra tick because of +-1/2 tick uncertainty in the 3585 * firing of the timer. The bias will give us exactly the 3586 * 1.5 tick we need. But, because the bias is 3587 * statistical, we have to test that we don't drop below 3588 * the minimum feasible timer (which is 2 ticks). 3589 */ 3590 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3591 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3592 3593 /* 3594 * We received an ack for a packet that wasn't retransmitted; 3595 * it is probably safe to discard any error indications we've 3596 * received recently. This isn't quite right, but close enough 3597 * for now (a route might have failed after we sent a segment, 3598 * and the return path might not be symmetrical). 3599 */ 3600 tp->t_softerror = 0; 3601 } 3602 3603 3604 /* 3605 * TCP compressed state engine. Currently used to hold compressed 3606 * state for SYN_RECEIVED. 3607 */ 3608 3609 u_long syn_cache_count; 3610 u_int32_t syn_hash1, syn_hash2; 3611 3612 #define SYN_HASH(sa, sp, dp) \ 3613 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3614 ((u_int32_t)(sp)))^syn_hash2))) 3615 #ifndef INET6 3616 #define SYN_HASHALL(hash, src, dst) \ 3617 do { \ 3618 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3619 ((const struct sockaddr_in *)(src))->sin_port, \ 3620 ((const struct sockaddr_in *)(dst))->sin_port); \ 3621 } while (/*CONSTCOND*/ 0) 3622 #else 3623 #define SYN_HASH6(sa, sp, dp) \ 3624 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3625 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3626 & 0x7fffffff) 3627 3628 #define SYN_HASHALL(hash, src, dst) \ 3629 do { \ 3630 switch ((src)->sa_family) { \ 3631 case AF_INET: \ 3632 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3633 ((const struct sockaddr_in *)(src))->sin_port, \ 3634 ((const struct sockaddr_in *)(dst))->sin_port); \ 3635 break; \ 3636 case AF_INET6: \ 3637 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3638 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3639 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3640 break; \ 3641 default: \ 3642 hash = 0; \ 3643 } \ 3644 } while (/*CONSTCOND*/0) 3645 #endif /* INET6 */ 3646 3647 static struct pool syn_cache_pool; 3648 3649 /* 3650 * We don't estimate RTT with SYNs, so each packet starts with the default 3651 * RTT and each timer step has a fixed timeout value. 3652 */ 3653 #define SYN_CACHE_TIMER_ARM(sc) \ 3654 do { \ 3655 TCPT_RANGESET((sc)->sc_rxtcur, \ 3656 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3657 TCPTV_REXMTMAX); \ 3658 callout_reset(&(sc)->sc_timer, \ 3659 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3660 } while (/*CONSTCOND*/0) 3661 3662 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3663 3664 static inline void 3665 syn_cache_rm(struct syn_cache *sc) 3666 { 3667 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3668 sc, sc_bucketq); 3669 sc->sc_tp = NULL; 3670 LIST_REMOVE(sc, sc_tpq); 3671 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3672 callout_stop(&sc->sc_timer); 3673 syn_cache_count--; 3674 } 3675 3676 static inline void 3677 syn_cache_put(struct syn_cache *sc) 3678 { 3679 if (sc->sc_ipopts) 3680 (void) m_free(sc->sc_ipopts); 3681 rtcache_free(&sc->sc_route); 3682 sc->sc_flags |= SCF_DEAD; 3683 if (!callout_invoking(&sc->sc_timer)) 3684 callout_schedule(&(sc)->sc_timer, 1); 3685 } 3686 3687 void 3688 syn_cache_init(void) 3689 { 3690 int i; 3691 3692 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 3693 "synpl", NULL, IPL_SOFTNET); 3694 3695 /* Initialize the hash buckets. */ 3696 for (i = 0; i < tcp_syn_cache_size; i++) 3697 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3698 } 3699 3700 void 3701 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3702 { 3703 struct syn_cache_head *scp; 3704 struct syn_cache *sc2; 3705 int s; 3706 3707 /* 3708 * If there are no entries in the hash table, reinitialize 3709 * the hash secrets. 3710 */ 3711 if (syn_cache_count == 0) { 3712 syn_hash1 = cprng_fast32(); 3713 syn_hash2 = cprng_fast32(); 3714 } 3715 3716 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3717 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3718 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3719 3720 /* 3721 * Make sure that we don't overflow the per-bucket 3722 * limit or the total cache size limit. 3723 */ 3724 s = splsoftnet(); 3725 if (scp->sch_length >= tcp_syn_bucket_limit) { 3726 TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW); 3727 /* 3728 * The bucket is full. Toss the oldest element in the 3729 * bucket. This will be the first entry in the bucket. 3730 */ 3731 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3732 #ifdef DIAGNOSTIC 3733 /* 3734 * This should never happen; we should always find an 3735 * entry in our bucket. 3736 */ 3737 if (sc2 == NULL) 3738 panic("syn_cache_insert: bucketoverflow: impossible"); 3739 #endif 3740 syn_cache_rm(sc2); 3741 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3742 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3743 struct syn_cache_head *scp2, *sce; 3744 3745 TCP_STATINC(TCP_STAT_SC_OVERFLOWED); 3746 /* 3747 * The cache is full. Toss the oldest entry in the 3748 * first non-empty bucket we can find. 3749 * 3750 * XXX We would really like to toss the oldest 3751 * entry in the cache, but we hope that this 3752 * condition doesn't happen very often. 3753 */ 3754 scp2 = scp; 3755 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3756 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3757 for (++scp2; scp2 != scp; scp2++) { 3758 if (scp2 >= sce) 3759 scp2 = &tcp_syn_cache[0]; 3760 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3761 break; 3762 } 3763 #ifdef DIAGNOSTIC 3764 /* 3765 * This should never happen; we should always find a 3766 * non-empty bucket. 3767 */ 3768 if (scp2 == scp) 3769 panic("syn_cache_insert: cacheoverflow: " 3770 "impossible"); 3771 #endif 3772 } 3773 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3774 syn_cache_rm(sc2); 3775 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3776 } 3777 3778 /* 3779 * Initialize the entry's timer. 3780 */ 3781 sc->sc_rxttot = 0; 3782 sc->sc_rxtshift = 0; 3783 SYN_CACHE_TIMER_ARM(sc); 3784 3785 /* Link it from tcpcb entry */ 3786 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3787 3788 /* Put it into the bucket. */ 3789 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3790 scp->sch_length++; 3791 syn_cache_count++; 3792 3793 TCP_STATINC(TCP_STAT_SC_ADDED); 3794 splx(s); 3795 } 3796 3797 /* 3798 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3799 * If we have retransmitted an entry the maximum number of times, expire 3800 * that entry. 3801 */ 3802 void 3803 syn_cache_timer(void *arg) 3804 { 3805 struct syn_cache *sc = arg; 3806 3807 mutex_enter(softnet_lock); 3808 KERNEL_LOCK(1, NULL); 3809 callout_ack(&sc->sc_timer); 3810 3811 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3812 TCP_STATINC(TCP_STAT_SC_DELAYED_FREE); 3813 callout_destroy(&sc->sc_timer); 3814 pool_put(&syn_cache_pool, sc); 3815 KERNEL_UNLOCK_ONE(NULL); 3816 mutex_exit(softnet_lock); 3817 return; 3818 } 3819 3820 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3821 /* Drop it -- too many retransmissions. */ 3822 goto dropit; 3823 } 3824 3825 /* 3826 * Compute the total amount of time this entry has 3827 * been on a queue. If this entry has been on longer 3828 * than the keep alive timer would allow, expire it. 3829 */ 3830 sc->sc_rxttot += sc->sc_rxtcur; 3831 if (sc->sc_rxttot >= tcp_keepinit) 3832 goto dropit; 3833 3834 TCP_STATINC(TCP_STAT_SC_RETRANSMITTED); 3835 (void) syn_cache_respond(sc, NULL); 3836 3837 /* Advance the timer back-off. */ 3838 sc->sc_rxtshift++; 3839 SYN_CACHE_TIMER_ARM(sc); 3840 3841 KERNEL_UNLOCK_ONE(NULL); 3842 mutex_exit(softnet_lock); 3843 return; 3844 3845 dropit: 3846 TCP_STATINC(TCP_STAT_SC_TIMED_OUT); 3847 syn_cache_rm(sc); 3848 if (sc->sc_ipopts) 3849 (void) m_free(sc->sc_ipopts); 3850 rtcache_free(&sc->sc_route); 3851 callout_destroy(&sc->sc_timer); 3852 pool_put(&syn_cache_pool, sc); 3853 KERNEL_UNLOCK_ONE(NULL); 3854 mutex_exit(softnet_lock); 3855 } 3856 3857 /* 3858 * Remove syn cache created by the specified tcb entry, 3859 * because this does not make sense to keep them 3860 * (if there's no tcb entry, syn cache entry will never be used) 3861 */ 3862 void 3863 syn_cache_cleanup(struct tcpcb *tp) 3864 { 3865 struct syn_cache *sc, *nsc; 3866 int s; 3867 3868 s = splsoftnet(); 3869 3870 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3871 nsc = LIST_NEXT(sc, sc_tpq); 3872 3873 #ifdef DIAGNOSTIC 3874 if (sc->sc_tp != tp) 3875 panic("invalid sc_tp in syn_cache_cleanup"); 3876 #endif 3877 syn_cache_rm(sc); 3878 syn_cache_put(sc); /* calls pool_put but see spl above */ 3879 } 3880 /* just for safety */ 3881 LIST_INIT(&tp->t_sc); 3882 3883 splx(s); 3884 } 3885 3886 /* 3887 * Find an entry in the syn cache. 3888 */ 3889 struct syn_cache * 3890 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3891 struct syn_cache_head **headp) 3892 { 3893 struct syn_cache *sc; 3894 struct syn_cache_head *scp; 3895 u_int32_t hash; 3896 int s; 3897 3898 SYN_HASHALL(hash, src, dst); 3899 3900 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3901 *headp = scp; 3902 s = splsoftnet(); 3903 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3904 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3905 if (sc->sc_hash != hash) 3906 continue; 3907 if (!memcmp(&sc->sc_src, src, src->sa_len) && 3908 !memcmp(&sc->sc_dst, dst, dst->sa_len)) { 3909 splx(s); 3910 return (sc); 3911 } 3912 } 3913 splx(s); 3914 return (NULL); 3915 } 3916 3917 /* 3918 * This function gets called when we receive an ACK for a 3919 * socket in the LISTEN state. We look up the connection 3920 * in the syn cache, and if its there, we pull it out of 3921 * the cache and turn it into a full-blown connection in 3922 * the SYN-RECEIVED state. 3923 * 3924 * The return values may not be immediately obvious, and their effects 3925 * can be subtle, so here they are: 3926 * 3927 * NULL SYN was not found in cache; caller should drop the 3928 * packet and send an RST. 3929 * 3930 * -1 We were unable to create the new connection, and are 3931 * aborting it. An ACK,RST is being sent to the peer 3932 * (unless we got screwey sequence numbners; see below), 3933 * because the 3-way handshake has been completed. Caller 3934 * should not free the mbuf, since we may be using it. If 3935 * we are not, we will free it. 3936 * 3937 * Otherwise, the return value is a pointer to the new socket 3938 * associated with the connection. 3939 */ 3940 struct socket * 3941 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3942 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3943 struct socket *so, struct mbuf *m) 3944 { 3945 struct syn_cache *sc; 3946 struct syn_cache_head *scp; 3947 struct inpcb *inp = NULL; 3948 #ifdef INET6 3949 struct in6pcb *in6p = NULL; 3950 #endif 3951 struct tcpcb *tp = 0; 3952 int s; 3953 struct socket *oso; 3954 3955 s = splsoftnet(); 3956 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3957 splx(s); 3958 return (NULL); 3959 } 3960 3961 /* 3962 * Verify the sequence and ack numbers. Try getting the correct 3963 * response again. 3964 */ 3965 if ((th->th_ack != sc->sc_iss + 1) || 3966 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3967 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3968 (void) syn_cache_respond(sc, m); 3969 splx(s); 3970 return ((struct socket *)(-1)); 3971 } 3972 3973 /* Remove this cache entry */ 3974 syn_cache_rm(sc); 3975 splx(s); 3976 3977 /* 3978 * Ok, create the full blown connection, and set things up 3979 * as they would have been set up if we had created the 3980 * connection when the SYN arrived. If we can't create 3981 * the connection, abort it. 3982 */ 3983 /* 3984 * inp still has the OLD in_pcb stuff, set the 3985 * v6-related flags on the new guy, too. This is 3986 * done particularly for the case where an AF_INET6 3987 * socket is bound only to a port, and a v4 connection 3988 * comes in on that port. 3989 * we also copy the flowinfo from the original pcb 3990 * to the new one. 3991 */ 3992 oso = so; 3993 so = sonewconn(so, true); 3994 if (so == NULL) 3995 goto resetandabort; 3996 3997 switch (so->so_proto->pr_domain->dom_family) { 3998 #ifdef INET 3999 case AF_INET: 4000 inp = sotoinpcb(so); 4001 break; 4002 #endif 4003 #ifdef INET6 4004 case AF_INET6: 4005 in6p = sotoin6pcb(so); 4006 break; 4007 #endif 4008 } 4009 switch (src->sa_family) { 4010 #ifdef INET 4011 case AF_INET: 4012 if (inp) { 4013 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 4014 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 4015 inp->inp_options = ip_srcroute(m); 4016 in_pcbstate(inp, INP_BOUND); 4017 if (inp->inp_options == NULL) { 4018 inp->inp_options = sc->sc_ipopts; 4019 sc->sc_ipopts = NULL; 4020 } 4021 } 4022 #ifdef INET6 4023 else if (in6p) { 4024 /* IPv4 packet to AF_INET6 socket */ 4025 memset(&in6p->in6p_laddr, 0, sizeof(in6p->in6p_laddr)); 4026 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 4027 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 4028 &in6p->in6p_laddr.s6_addr32[3], 4029 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 4030 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 4031 in6totcpcb(in6p)->t_family = AF_INET; 4032 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 4033 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 4034 else 4035 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 4036 in6_pcbstate(in6p, IN6P_BOUND); 4037 } 4038 #endif 4039 break; 4040 #endif 4041 #ifdef INET6 4042 case AF_INET6: 4043 if (in6p) { 4044 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 4045 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 4046 in6_pcbstate(in6p, IN6P_BOUND); 4047 } 4048 break; 4049 #endif 4050 } 4051 #ifdef INET6 4052 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 4053 struct in6pcb *oin6p = sotoin6pcb(oso); 4054 /* inherit socket options from the listening socket */ 4055 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 4056 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 4057 m_freem(in6p->in6p_options); 4058 in6p->in6p_options = 0; 4059 } 4060 ip6_savecontrol(in6p, &in6p->in6p_options, 4061 mtod(m, struct ip6_hdr *), m); 4062 } 4063 #endif 4064 4065 #if defined(IPSEC) 4066 if (ipsec_used) { 4067 /* 4068 * we make a copy of policy, instead of sharing the policy, for 4069 * better behavior in terms of SA lookup and dead SA removal. 4070 */ 4071 if (inp) { 4072 /* copy old policy into new socket's */ 4073 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, 4074 inp->inp_sp)) 4075 printf("tcp_input: could not copy policy\n"); 4076 } 4077 #ifdef INET6 4078 else if (in6p) { 4079 /* copy old policy into new socket's */ 4080 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 4081 in6p->in6p_sp)) 4082 printf("tcp_input: could not copy policy\n"); 4083 } 4084 #endif 4085 } 4086 #endif 4087 4088 /* 4089 * Give the new socket our cached route reference. 4090 */ 4091 if (inp) { 4092 rtcache_copy(&inp->inp_route, &sc->sc_route); 4093 rtcache_free(&sc->sc_route); 4094 } 4095 #ifdef INET6 4096 else { 4097 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 4098 rtcache_free(&sc->sc_route); 4099 } 4100 #endif 4101 4102 if (inp) { 4103 struct sockaddr_in sin; 4104 memcpy(&sin, src, src->sa_len); 4105 if (in_pcbconnect(inp, &sin, &lwp0)) { 4106 goto resetandabort; 4107 } 4108 } 4109 #ifdef INET6 4110 else if (in6p) { 4111 struct sockaddr_in6 sin6; 4112 memcpy(&sin6, src, src->sa_len); 4113 if (src->sa_family == AF_INET) { 4114 /* IPv4 packet to AF_INET6 socket */ 4115 in6_sin_2_v4mapsin6((struct sockaddr_in *)src, &sin6); 4116 } 4117 if (in6_pcbconnect(in6p, &sin6, NULL)) { 4118 goto resetandabort; 4119 } 4120 } 4121 #endif 4122 else { 4123 goto resetandabort; 4124 } 4125 4126 if (inp) 4127 tp = intotcpcb(inp); 4128 #ifdef INET6 4129 else if (in6p) 4130 tp = in6totcpcb(in6p); 4131 #endif 4132 else 4133 tp = NULL; 4134 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 4135 if (sc->sc_request_r_scale != 15) { 4136 tp->requested_s_scale = sc->sc_requested_s_scale; 4137 tp->request_r_scale = sc->sc_request_r_scale; 4138 tp->snd_scale = sc->sc_requested_s_scale; 4139 tp->rcv_scale = sc->sc_request_r_scale; 4140 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 4141 } 4142 if (sc->sc_flags & SCF_TIMESTAMP) 4143 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 4144 tp->ts_timebase = sc->sc_timebase; 4145 4146 tp->t_template = tcp_template(tp); 4147 if (tp->t_template == 0) { 4148 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 4149 so = NULL; 4150 m_freem(m); 4151 goto abort; 4152 } 4153 4154 tp->iss = sc->sc_iss; 4155 tp->irs = sc->sc_irs; 4156 tcp_sendseqinit(tp); 4157 tcp_rcvseqinit(tp); 4158 tp->t_state = TCPS_SYN_RECEIVED; 4159 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 4160 TCP_STATINC(TCP_STAT_ACCEPTS); 4161 4162 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 4163 tp->t_flags |= TF_WILL_SACK; 4164 4165 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 4166 tp->t_flags |= TF_ECN_PERMIT; 4167 4168 #ifdef TCP_SIGNATURE 4169 if (sc->sc_flags & SCF_SIGNATURE) 4170 tp->t_flags |= TF_SIGNATURE; 4171 #endif 4172 4173 /* Initialize tp->t_ourmss before we deal with the peer's! */ 4174 tp->t_ourmss = sc->sc_ourmaxseg; 4175 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 4176 4177 /* 4178 * Initialize the initial congestion window. If we 4179 * had to retransmit the SYN,ACK, we must initialize cwnd 4180 * to 1 segment (i.e. the Loss Window). 4181 */ 4182 if (sc->sc_rxtshift) 4183 tp->snd_cwnd = tp->t_peermss; 4184 else { 4185 int ss = tcp_init_win; 4186 #ifdef INET 4187 if (inp != NULL && in_localaddr(inp->inp_faddr)) 4188 ss = tcp_init_win_local; 4189 #endif 4190 #ifdef INET6 4191 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 4192 ss = tcp_init_win_local; 4193 #endif 4194 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 4195 } 4196 4197 tcp_rmx_rtt(tp); 4198 tp->snd_wl1 = sc->sc_irs; 4199 tp->rcv_up = sc->sc_irs + 1; 4200 4201 /* 4202 * This is what whould have happened in tcp_output() when 4203 * the SYN,ACK was sent. 4204 */ 4205 tp->snd_up = tp->snd_una; 4206 tp->snd_max = tp->snd_nxt = tp->iss+1; 4207 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 4208 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 4209 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 4210 tp->last_ack_sent = tp->rcv_nxt; 4211 tp->t_partialacks = -1; 4212 tp->t_dupacks = 0; 4213 4214 TCP_STATINC(TCP_STAT_SC_COMPLETED); 4215 s = splsoftnet(); 4216 syn_cache_put(sc); 4217 splx(s); 4218 return (so); 4219 4220 resetandabort: 4221 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 4222 abort: 4223 if (so != NULL) { 4224 (void) soqremque(so, 1); 4225 (void) soabort(so); 4226 mutex_enter(softnet_lock); 4227 } 4228 s = splsoftnet(); 4229 syn_cache_put(sc); 4230 splx(s); 4231 TCP_STATINC(TCP_STAT_SC_ABORTED); 4232 return ((struct socket *)(-1)); 4233 } 4234 4235 /* 4236 * This function is called when we get a RST for a 4237 * non-existent connection, so that we can see if the 4238 * connection is in the syn cache. If it is, zap it. 4239 */ 4240 4241 void 4242 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 4243 { 4244 struct syn_cache *sc; 4245 struct syn_cache_head *scp; 4246 int s = splsoftnet(); 4247 4248 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4249 splx(s); 4250 return; 4251 } 4252 if (SEQ_LT(th->th_seq, sc->sc_irs) || 4253 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 4254 splx(s); 4255 return; 4256 } 4257 syn_cache_rm(sc); 4258 TCP_STATINC(TCP_STAT_SC_RESET); 4259 syn_cache_put(sc); /* calls pool_put but see spl above */ 4260 splx(s); 4261 } 4262 4263 void 4264 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 4265 struct tcphdr *th) 4266 { 4267 struct syn_cache *sc; 4268 struct syn_cache_head *scp; 4269 int s; 4270 4271 s = splsoftnet(); 4272 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4273 splx(s); 4274 return; 4275 } 4276 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 4277 if (ntohl (th->th_seq) != sc->sc_iss) { 4278 splx(s); 4279 return; 4280 } 4281 4282 /* 4283 * If we've retransmitted 3 times and this is our second error, 4284 * we remove the entry. Otherwise, we allow it to continue on. 4285 * This prevents us from incorrectly nuking an entry during a 4286 * spurious network outage. 4287 * 4288 * See tcp_notify(). 4289 */ 4290 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 4291 sc->sc_flags |= SCF_UNREACH; 4292 splx(s); 4293 return; 4294 } 4295 4296 syn_cache_rm(sc); 4297 TCP_STATINC(TCP_STAT_SC_UNREACH); 4298 syn_cache_put(sc); /* calls pool_put but see spl above */ 4299 splx(s); 4300 } 4301 4302 /* 4303 * Given a LISTEN socket and an inbound SYN request, add 4304 * this to the syn cache, and send back a segment: 4305 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 4306 * to the source. 4307 * 4308 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 4309 * Doing so would require that we hold onto the data and deliver it 4310 * to the application. However, if we are the target of a SYN-flood 4311 * DoS attack, an attacker could send data which would eventually 4312 * consume all available buffer space if it were ACKed. By not ACKing 4313 * the data, we avoid this DoS scenario. 4314 */ 4315 4316 int 4317 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 4318 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 4319 int optlen, struct tcp_opt_info *oi) 4320 { 4321 struct tcpcb tb, *tp; 4322 long win; 4323 struct syn_cache *sc; 4324 struct syn_cache_head *scp; 4325 struct mbuf *ipopts; 4326 struct tcp_opt_info opti; 4327 int s; 4328 4329 tp = sototcpcb(so); 4330 4331 memset(&opti, 0, sizeof(opti)); 4332 4333 /* 4334 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 4335 * 4336 * Note this check is performed in tcp_input() very early on. 4337 */ 4338 4339 /* 4340 * Initialize some local state. 4341 */ 4342 win = sbspace(&so->so_rcv); 4343 if (win > TCP_MAXWIN) 4344 win = TCP_MAXWIN; 4345 4346 switch (src->sa_family) { 4347 #ifdef INET 4348 case AF_INET: 4349 /* 4350 * Remember the IP options, if any. 4351 */ 4352 ipopts = ip_srcroute(m); 4353 break; 4354 #endif 4355 default: 4356 ipopts = NULL; 4357 } 4358 4359 #ifdef TCP_SIGNATURE 4360 if (optp || (tp->t_flags & TF_SIGNATURE)) 4361 #else 4362 if (optp) 4363 #endif 4364 { 4365 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4366 #ifdef TCP_SIGNATURE 4367 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4368 #endif 4369 tb.t_state = TCPS_LISTEN; 4370 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 4371 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 4372 return (0); 4373 } else 4374 tb.t_flags = 0; 4375 4376 /* 4377 * See if we already have an entry for this connection. 4378 * If we do, resend the SYN,ACK. We do not count this 4379 * as a retransmission (XXX though maybe we should). 4380 */ 4381 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4382 TCP_STATINC(TCP_STAT_SC_DUPESYN); 4383 if (ipopts) { 4384 /* 4385 * If we were remembering a previous source route, 4386 * forget it and use the new one we've been given. 4387 */ 4388 if (sc->sc_ipopts) 4389 (void) m_free(sc->sc_ipopts); 4390 sc->sc_ipopts = ipopts; 4391 } 4392 sc->sc_timestamp = tb.ts_recent; 4393 if (syn_cache_respond(sc, m) == 0) { 4394 uint64_t *tcps = TCP_STAT_GETREF(); 4395 tcps[TCP_STAT_SNDACKS]++; 4396 tcps[TCP_STAT_SNDTOTAL]++; 4397 TCP_STAT_PUTREF(); 4398 } 4399 return (1); 4400 } 4401 4402 s = splsoftnet(); 4403 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4404 splx(s); 4405 if (sc == NULL) { 4406 if (ipopts) 4407 (void) m_free(ipopts); 4408 return (0); 4409 } 4410 4411 /* 4412 * Fill in the cache, and put the necessary IP and TCP 4413 * options into the reply. 4414 */ 4415 memset(sc, 0, sizeof(struct syn_cache)); 4416 callout_init(&sc->sc_timer, CALLOUT_MPSAFE); 4417 bcopy(src, &sc->sc_src, src->sa_len); 4418 bcopy(dst, &sc->sc_dst, dst->sa_len); 4419 sc->sc_flags = 0; 4420 sc->sc_ipopts = ipopts; 4421 sc->sc_irs = th->th_seq; 4422 switch (src->sa_family) { 4423 #ifdef INET 4424 case AF_INET: 4425 { 4426 struct sockaddr_in *srcin = (void *) src; 4427 struct sockaddr_in *dstin = (void *) dst; 4428 4429 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4430 &srcin->sin_addr, dstin->sin_port, 4431 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4432 break; 4433 } 4434 #endif /* INET */ 4435 #ifdef INET6 4436 case AF_INET6: 4437 { 4438 struct sockaddr_in6 *srcin6 = (void *) src; 4439 struct sockaddr_in6 *dstin6 = (void *) dst; 4440 4441 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4442 &srcin6->sin6_addr, dstin6->sin6_port, 4443 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4444 break; 4445 } 4446 #endif /* INET6 */ 4447 } 4448 sc->sc_peermaxseg = oi->maxseg; 4449 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4450 m_get_rcvif_NOMPSAFE(m) : NULL, 4451 sc->sc_src.sa.sa_family); 4452 sc->sc_win = win; 4453 sc->sc_timebase = tcp_now - 1; /* see tcp_newtcpcb() */ 4454 sc->sc_timestamp = tb.ts_recent; 4455 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4456 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4457 sc->sc_flags |= SCF_TIMESTAMP; 4458 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4459 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4460 sc->sc_requested_s_scale = tb.requested_s_scale; 4461 sc->sc_request_r_scale = 0; 4462 /* 4463 * Pick the smallest possible scaling factor that 4464 * will still allow us to scale up to sb_max. 4465 * 4466 * We do this because there are broken firewalls that 4467 * will corrupt the window scale option, leading to 4468 * the other endpoint believing that our advertised 4469 * window is unscaled. At scale factors larger than 4470 * 5 the unscaled window will drop below 1500 bytes, 4471 * leading to serious problems when traversing these 4472 * broken firewalls. 4473 * 4474 * With the default sbmax of 256K, a scale factor 4475 * of 3 will be chosen by this algorithm. Those who 4476 * choose a larger sbmax should watch out 4477 * for the compatiblity problems mentioned above. 4478 * 4479 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4480 * or <SYN,ACK>) segment itself is never scaled. 4481 */ 4482 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4483 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4484 sc->sc_request_r_scale++; 4485 } else { 4486 sc->sc_requested_s_scale = 15; 4487 sc->sc_request_r_scale = 15; 4488 } 4489 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4490 sc->sc_flags |= SCF_SACK_PERMIT; 4491 4492 /* 4493 * ECN setup packet recieved. 4494 */ 4495 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4496 sc->sc_flags |= SCF_ECN_PERMIT; 4497 4498 #ifdef TCP_SIGNATURE 4499 if (tb.t_flags & TF_SIGNATURE) 4500 sc->sc_flags |= SCF_SIGNATURE; 4501 #endif 4502 sc->sc_tp = tp; 4503 if (syn_cache_respond(sc, m) == 0) { 4504 uint64_t *tcps = TCP_STAT_GETREF(); 4505 tcps[TCP_STAT_SNDACKS]++; 4506 tcps[TCP_STAT_SNDTOTAL]++; 4507 TCP_STAT_PUTREF(); 4508 syn_cache_insert(sc, tp); 4509 } else { 4510 s = splsoftnet(); 4511 /* 4512 * syn_cache_put() will try to schedule the timer, so 4513 * we need to initialize it 4514 */ 4515 SYN_CACHE_TIMER_ARM(sc); 4516 syn_cache_put(sc); 4517 splx(s); 4518 TCP_STATINC(TCP_STAT_SC_DROPPED); 4519 } 4520 return (1); 4521 } 4522 4523 /* 4524 * syn_cache_respond: (re)send SYN+ACK. 4525 * 4526 * returns 0 on success. otherwise returns an errno, typically ENOBUFS. 4527 */ 4528 4529 int 4530 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4531 { 4532 #ifdef INET6 4533 struct rtentry *rt = NULL; 4534 #endif 4535 struct route *ro; 4536 u_int8_t *optp; 4537 int optlen, error; 4538 u_int16_t tlen; 4539 struct ip *ip = NULL; 4540 #ifdef INET6 4541 struct ip6_hdr *ip6 = NULL; 4542 #endif 4543 struct tcpcb *tp = NULL; 4544 struct tcphdr *th; 4545 u_int hlen; 4546 #ifdef TCP_SIGNATURE 4547 struct secasvar *sav = NULL; 4548 u_int8_t *sigp = NULL; 4549 #endif 4550 4551 ro = &sc->sc_route; 4552 switch (sc->sc_src.sa.sa_family) { 4553 case AF_INET: 4554 hlen = sizeof(struct ip); 4555 break; 4556 #ifdef INET6 4557 case AF_INET6: 4558 hlen = sizeof(struct ip6_hdr); 4559 break; 4560 #endif 4561 default: 4562 if (m) 4563 m_freem(m); 4564 return (EAFNOSUPPORT); 4565 } 4566 4567 /* worst case scanario, since we don't know the option size yet */ 4568 tlen = hlen + sizeof(struct tcphdr) + MAX_TCPOPTLEN; 4569 4570 /* 4571 * Create the IP+TCP header from scratch. 4572 */ 4573 if (m) 4574 m_freem(m); 4575 #ifdef DIAGNOSTIC 4576 if (max_linkhdr + tlen > MCLBYTES) 4577 return ENOBUFS; 4578 #endif 4579 4580 MGETHDR(m, M_DONTWAIT, MT_DATA); 4581 if (m && (max_linkhdr + tlen) > MHLEN) { 4582 MCLGET(m, M_DONTWAIT); 4583 if ((m->m_flags & M_EXT) == 0) { 4584 m_freem(m); 4585 m = NULL; 4586 } 4587 } 4588 if (m == NULL) 4589 return ENOBUFS; 4590 MCLAIM(m, &tcp_tx_mowner); 4591 4592 /* Fixup the mbuf. */ 4593 m->m_data += max_linkhdr; 4594 if (sc->sc_tp) 4595 tp = sc->sc_tp; 4596 m_reset_rcvif(m); 4597 memset(mtod(m, u_char *), 0, tlen); 4598 4599 switch (sc->sc_src.sa.sa_family) { 4600 case AF_INET: 4601 ip = mtod(m, struct ip *); 4602 ip->ip_v = 4; 4603 ip->ip_dst = sc->sc_src.sin.sin_addr; 4604 ip->ip_src = sc->sc_dst.sin.sin_addr; 4605 ip->ip_p = IPPROTO_TCP; 4606 th = (struct tcphdr *)(ip + 1); 4607 th->th_dport = sc->sc_src.sin.sin_port; 4608 th->th_sport = sc->sc_dst.sin.sin_port; 4609 break; 4610 #ifdef INET6 4611 case AF_INET6: 4612 ip6 = mtod(m, struct ip6_hdr *); 4613 ip6->ip6_vfc = IPV6_VERSION; 4614 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4615 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4616 ip6->ip6_nxt = IPPROTO_TCP; 4617 /* ip6_plen will be updated in ip6_output() */ 4618 th = (struct tcphdr *)(ip6 + 1); 4619 th->th_dport = sc->sc_src.sin6.sin6_port; 4620 th->th_sport = sc->sc_dst.sin6.sin6_port; 4621 break; 4622 #endif 4623 default: 4624 return ENOBUFS; 4625 } 4626 4627 th->th_seq = htonl(sc->sc_iss); 4628 th->th_ack = htonl(sc->sc_irs + 1); 4629 th->th_flags = TH_SYN|TH_ACK; 4630 th->th_win = htons(sc->sc_win); 4631 /* th_x2, th_sum, th_urp already 0 from memset */ 4632 4633 /* Tack on the TCP options. */ 4634 optp = (u_int8_t *)(th + 1); 4635 optlen = 0; 4636 *optp++ = TCPOPT_MAXSEG; 4637 *optp++ = TCPOLEN_MAXSEG; 4638 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4639 *optp++ = sc->sc_ourmaxseg & 0xff; 4640 optlen += TCPOLEN_MAXSEG; 4641 4642 if (sc->sc_request_r_scale != 15) { 4643 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4644 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4645 sc->sc_request_r_scale); 4646 optp += TCPOLEN_WINDOW + TCPOLEN_NOP; 4647 optlen += TCPOLEN_WINDOW + TCPOLEN_NOP; 4648 } 4649 4650 if (sc->sc_flags & SCF_SACK_PERMIT) { 4651 /* Let the peer know that we will SACK. */ 4652 *optp++ = TCPOPT_SACK_PERMITTED; 4653 *optp++ = TCPOLEN_SACK_PERMITTED; 4654 optlen += TCPOLEN_SACK_PERMITTED; 4655 } 4656 4657 if (sc->sc_flags & SCF_TIMESTAMP) { 4658 while (optlen % 4 != 2) { 4659 optlen += TCPOLEN_NOP; 4660 *optp++ = TCPOPT_NOP; 4661 } 4662 *optp++ = TCPOPT_TIMESTAMP; 4663 *optp++ = TCPOLEN_TIMESTAMP; 4664 u_int32_t *lp = (u_int32_t *)(optp); 4665 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4666 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4667 *lp = htonl(sc->sc_timestamp); 4668 optp += TCPOLEN_TIMESTAMP - 2; 4669 optlen += TCPOLEN_TIMESTAMP; 4670 } 4671 4672 #ifdef TCP_SIGNATURE 4673 if (sc->sc_flags & SCF_SIGNATURE) { 4674 4675 sav = tcp_signature_getsav(m, th); 4676 4677 if (sav == NULL) { 4678 if (m) 4679 m_freem(m); 4680 return (EPERM); 4681 } 4682 4683 *optp++ = TCPOPT_SIGNATURE; 4684 *optp++ = TCPOLEN_SIGNATURE; 4685 sigp = optp; 4686 memset(optp, 0, TCP_SIGLEN); 4687 optp += TCP_SIGLEN; 4688 optlen += TCPOLEN_SIGNATURE; 4689 4690 } 4691 #endif 4692 /* Terminate and pad TCP options to a 4 byte boundary. */ 4693 if (optlen % 4) { 4694 optlen += TCPOLEN_EOL; 4695 *optp++ = TCPOPT_EOL; 4696 } 4697 /* 4698 * According to RFC 793 (STD0007): 4699 * "The content of the header beyond the End-of-Option option 4700 * must be header padding (i.e., zero)." 4701 * and later: "The padding is composed of zeros." 4702 */ 4703 while (optlen % 4) { 4704 optlen += TCPOLEN_PAD; 4705 *optp++ = TCPOPT_PAD; 4706 } 4707 4708 /* compute the actual values now that we've added the options */ 4709 tlen = hlen + sizeof(struct tcphdr) + optlen; 4710 m->m_len = m->m_pkthdr.len = tlen; 4711 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4712 4713 #ifdef TCP_SIGNATURE 4714 if (sav) { 4715 (void)tcp_signature(m, th, hlen, sav, sigp); 4716 key_sa_recordxfer(sav, m); 4717 KEY_FREESAV(&sav); 4718 } 4719 #endif 4720 4721 /* 4722 * Send ECN SYN-ACK setup packet. 4723 * Routes can be asymetric, so, even if we receive a packet 4724 * with ECE and CWR set, we must not assume no one will block 4725 * the ECE packet we are about to send. 4726 */ 4727 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4728 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4729 th->th_flags |= TH_ECE; 4730 TCP_STATINC(TCP_STAT_ECN_SHS); 4731 4732 /* 4733 * draft-ietf-tcpm-ecnsyn-00.txt 4734 * 4735 * "[...] a TCP node MAY respond to an ECN-setup 4736 * SYN packet by setting ECT in the responding 4737 * ECN-setup SYN/ACK packet, indicating to routers 4738 * that the SYN/ACK packet is ECN-Capable. 4739 * This allows a congested router along the path 4740 * to mark the packet instead of dropping the 4741 * packet as an indication of congestion." 4742 * 4743 * "[...] There can be a great benefit in setting 4744 * an ECN-capable codepoint in SYN/ACK packets [...] 4745 * Congestion is most likely to occur in 4746 * the server-to-client direction. As a result, 4747 * setting an ECN-capable codepoint in SYN/ACK 4748 * packets can reduce the occurence of three-second 4749 * retransmit timeouts resulting from the drop 4750 * of SYN/ACK packets." 4751 * 4752 * Page 4 and 6, January 2006. 4753 */ 4754 4755 switch (sc->sc_src.sa.sa_family) { 4756 #ifdef INET 4757 case AF_INET: 4758 ip->ip_tos |= IPTOS_ECN_ECT0; 4759 break; 4760 #endif 4761 #ifdef INET6 4762 case AF_INET6: 4763 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4764 break; 4765 #endif 4766 } 4767 TCP_STATINC(TCP_STAT_ECN_ECT); 4768 } 4769 4770 4771 /* Compute the packet's checksum. */ 4772 switch (sc->sc_src.sa.sa_family) { 4773 case AF_INET: 4774 ip->ip_len = htons(tlen - hlen); 4775 th->th_sum = 0; 4776 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4777 break; 4778 #ifdef INET6 4779 case AF_INET6: 4780 ip6->ip6_plen = htons(tlen - hlen); 4781 th->th_sum = 0; 4782 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4783 break; 4784 #endif 4785 } 4786 4787 /* 4788 * Fill in some straggling IP bits. Note the stack expects 4789 * ip_len to be in host order, for convenience. 4790 */ 4791 switch (sc->sc_src.sa.sa_family) { 4792 #ifdef INET 4793 case AF_INET: 4794 ip->ip_len = htons(tlen); 4795 ip->ip_ttl = ip_defttl; 4796 /* XXX tos? */ 4797 break; 4798 #endif 4799 #ifdef INET6 4800 case AF_INET6: 4801 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4802 ip6->ip6_vfc |= IPV6_VERSION; 4803 ip6->ip6_plen = htons(tlen - hlen); 4804 /* ip6_hlim will be initialized afterwards */ 4805 /* XXX flowlabel? */ 4806 break; 4807 #endif 4808 } 4809 4810 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4811 tp = sc->sc_tp; 4812 4813 switch (sc->sc_src.sa.sa_family) { 4814 #ifdef INET 4815 case AF_INET: 4816 error = ip_output(m, sc->sc_ipopts, ro, 4817 (ip_mtudisc ? IP_MTUDISC : 0), 4818 NULL, tp ? tp->t_inpcb : NULL); 4819 break; 4820 #endif 4821 #ifdef INET6 4822 case AF_INET6: 4823 ip6->ip6_hlim = in6_selecthlim(NULL, 4824 (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp : NULL); 4825 rtcache_unref(rt, ro); 4826 4827 error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, 4828 tp ? tp->t_in6pcb : NULL, NULL); 4829 break; 4830 #endif 4831 default: 4832 error = EAFNOSUPPORT; 4833 break; 4834 } 4835 return (error); 4836 } 4837