1 /* $NetBSD: tcp_input.c,v 1.347 2016/06/10 13:31:44 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 1.347 2016/06/10 13:31:44 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; 266 267 if (tp != NULL && tp->t_in6pcb != NULL && tp->t_family == AF_INET6 && 268 (rt = rtcache_validate(&tp->t_in6pcb->in6p_route)) != NULL) 269 nd6_nud_hint(rt); 270 } 271 #else 272 static inline void 273 nd6_hint(struct tcpcb *tp) 274 { 275 } 276 #endif 277 278 /* 279 * Compute ACK transmission behavior. Delay the ACK unless 280 * we have already delayed an ACK (must send an ACK every two segments). 281 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 282 * option is enabled. 283 */ 284 static void 285 tcp_setup_ack(struct tcpcb *tp, const struct tcphdr *th) 286 { 287 288 if (tp->t_flags & TF_DELACK || 289 (tcp_ack_on_push && th->th_flags & TH_PUSH)) 290 tp->t_flags |= TF_ACKNOW; 291 else 292 TCP_SET_DELACK(tp); 293 } 294 295 static void 296 icmp_check(struct tcpcb *tp, const struct tcphdr *th, int acked) 297 { 298 299 /* 300 * If we had a pending ICMP message that refers to data that have 301 * just been acknowledged, disregard the recorded ICMP message. 302 */ 303 if ((tp->t_flags & TF_PMTUD_PEND) && 304 SEQ_GT(th->th_ack, tp->t_pmtud_th_seq)) 305 tp->t_flags &= ~TF_PMTUD_PEND; 306 307 /* 308 * Keep track of the largest chunk of data 309 * acknowledged since last PMTU update 310 */ 311 if (tp->t_pmtud_mss_acked < acked) 312 tp->t_pmtud_mss_acked = acked; 313 } 314 315 /* 316 * Convert TCP protocol fields to host order for easier processing. 317 */ 318 static void 319 tcp_fields_to_host(struct tcphdr *th) 320 { 321 322 NTOHL(th->th_seq); 323 NTOHL(th->th_ack); 324 NTOHS(th->th_win); 325 NTOHS(th->th_urp); 326 } 327 328 /* 329 * ... and reverse the above. 330 */ 331 static void 332 tcp_fields_to_net(struct tcphdr *th) 333 { 334 335 HTONL(th->th_seq); 336 HTONL(th->th_ack); 337 HTONS(th->th_win); 338 HTONS(th->th_urp); 339 } 340 341 #ifdef TCP_CSUM_COUNTERS 342 #include <sys/device.h> 343 344 #if defined(INET) 345 extern struct evcnt tcp_hwcsum_ok; 346 extern struct evcnt tcp_hwcsum_bad; 347 extern struct evcnt tcp_hwcsum_data; 348 extern struct evcnt tcp_swcsum; 349 #endif /* defined(INET) */ 350 #if defined(INET6) 351 extern struct evcnt tcp6_hwcsum_ok; 352 extern struct evcnt tcp6_hwcsum_bad; 353 extern struct evcnt tcp6_hwcsum_data; 354 extern struct evcnt tcp6_swcsum; 355 #endif /* defined(INET6) */ 356 357 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 358 359 #else 360 361 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 362 363 #endif /* TCP_CSUM_COUNTERS */ 364 365 #ifdef TCP_REASS_COUNTERS 366 #include <sys/device.h> 367 368 extern struct evcnt tcp_reass_; 369 extern struct evcnt tcp_reass_empty; 370 extern struct evcnt tcp_reass_iteration[8]; 371 extern struct evcnt tcp_reass_prependfirst; 372 extern struct evcnt tcp_reass_prepend; 373 extern struct evcnt tcp_reass_insert; 374 extern struct evcnt tcp_reass_inserttail; 375 extern struct evcnt tcp_reass_append; 376 extern struct evcnt tcp_reass_appendtail; 377 extern struct evcnt tcp_reass_overlaptail; 378 extern struct evcnt tcp_reass_overlapfront; 379 extern struct evcnt tcp_reass_segdup; 380 extern struct evcnt tcp_reass_fragdup; 381 382 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 383 384 #else 385 386 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 387 388 #endif /* TCP_REASS_COUNTERS */ 389 390 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *, 391 int *); 392 static int tcp_dooptions(struct tcpcb *, const u_char *, int, 393 struct tcphdr *, struct mbuf *, int, struct tcp_opt_info *); 394 395 #ifdef INET 396 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 397 #endif 398 #ifdef INET6 399 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 400 #endif 401 402 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 403 404 #if defined(MBUFTRACE) 405 struct mowner tcp_reass_mowner = MOWNER_INIT("tcp", "reass"); 406 #endif /* defined(MBUFTRACE) */ 407 408 static struct pool tcpipqent_pool; 409 410 void 411 tcpipqent_init(void) 412 { 413 414 pool_init(&tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", 415 NULL, IPL_VM); 416 } 417 418 struct ipqent * 419 tcpipqent_alloc(void) 420 { 421 struct ipqent *ipqe; 422 int s; 423 424 s = splvm(); 425 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 426 splx(s); 427 428 return ipqe; 429 } 430 431 void 432 tcpipqent_free(struct ipqent *ipqe) 433 { 434 int s; 435 436 s = splvm(); 437 pool_put(&tcpipqent_pool, ipqe); 438 splx(s); 439 } 440 441 static int 442 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int *tlen) 443 { 444 struct ipqent *p, *q, *nq, *tiqe = NULL; 445 struct socket *so = NULL; 446 int pkt_flags; 447 tcp_seq pkt_seq; 448 unsigned pkt_len; 449 u_long rcvpartdupbyte = 0; 450 u_long rcvoobyte; 451 #ifdef TCP_REASS_COUNTERS 452 u_int count = 0; 453 #endif 454 uint64_t *tcps; 455 456 if (tp->t_inpcb) 457 so = tp->t_inpcb->inp_socket; 458 #ifdef INET6 459 else if (tp->t_in6pcb) 460 so = tp->t_in6pcb->in6p_socket; 461 #endif 462 463 TCP_REASS_LOCK_CHECK(tp); 464 465 /* 466 * Call with th==0 after become established to 467 * force pre-ESTABLISHED data up to user socket. 468 */ 469 if (th == 0) 470 goto present; 471 472 m_claimm(m, &tcp_reass_mowner); 473 474 rcvoobyte = *tlen; 475 /* 476 * Copy these to local variables because the tcpiphdr 477 * gets munged while we are collapsing mbufs. 478 */ 479 pkt_seq = th->th_seq; 480 pkt_len = *tlen; 481 pkt_flags = th->th_flags; 482 483 TCP_REASS_COUNTER_INCR(&tcp_reass_); 484 485 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 486 /* 487 * When we miss a packet, the vast majority of time we get 488 * packets that follow it in order. So optimize for that. 489 */ 490 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 491 p->ipqe_len += pkt_len; 492 p->ipqe_flags |= pkt_flags; 493 m_cat(p->ipre_mlast, m); 494 TRAVERSE(p->ipre_mlast); 495 m = NULL; 496 tiqe = p; 497 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 498 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 499 goto skip_replacement; 500 } 501 /* 502 * While we're here, if the pkt is completely beyond 503 * anything we have, just insert it at the tail. 504 */ 505 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 506 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 507 goto insert_it; 508 } 509 } 510 511 q = TAILQ_FIRST(&tp->segq); 512 513 if (q != NULL) { 514 /* 515 * If this segment immediately precedes the first out-of-order 516 * block, simply slap the segment in front of it and (mostly) 517 * skip the complicated logic. 518 */ 519 if (pkt_seq + pkt_len == q->ipqe_seq) { 520 q->ipqe_seq = pkt_seq; 521 q->ipqe_len += pkt_len; 522 q->ipqe_flags |= pkt_flags; 523 m_cat(m, q->ipqe_m); 524 q->ipqe_m = m; 525 q->ipre_mlast = m; /* last mbuf may have changed */ 526 TRAVERSE(q->ipre_mlast); 527 tiqe = q; 528 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 529 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 530 goto skip_replacement; 531 } 532 } else { 533 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 534 } 535 536 /* 537 * Find a segment which begins after this one does. 538 */ 539 for (p = NULL; q != NULL; q = nq) { 540 nq = TAILQ_NEXT(q, ipqe_q); 541 #ifdef TCP_REASS_COUNTERS 542 count++; 543 #endif 544 /* 545 * If the received segment is just right after this 546 * fragment, merge the two together and then check 547 * for further overlaps. 548 */ 549 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 550 #ifdef TCPREASS_DEBUG 551 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 552 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 553 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 554 #endif 555 pkt_len += q->ipqe_len; 556 pkt_flags |= q->ipqe_flags; 557 pkt_seq = q->ipqe_seq; 558 m_cat(q->ipre_mlast, m); 559 TRAVERSE(q->ipre_mlast); 560 m = q->ipqe_m; 561 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 562 goto free_ipqe; 563 } 564 /* 565 * If the received segment is completely past this 566 * fragment, we need to go the next fragment. 567 */ 568 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 569 p = q; 570 continue; 571 } 572 /* 573 * If the fragment is past the received segment, 574 * it (or any following) can't be concatenated. 575 */ 576 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 577 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 578 break; 579 } 580 581 /* 582 * We've received all the data in this segment before. 583 * mark it as a duplicate and return. 584 */ 585 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 586 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 587 tcps = TCP_STAT_GETREF(); 588 tcps[TCP_STAT_RCVDUPPACK]++; 589 tcps[TCP_STAT_RCVDUPBYTE] += pkt_len; 590 TCP_STAT_PUTREF(); 591 tcp_new_dsack(tp, pkt_seq, pkt_len); 592 m_freem(m); 593 if (tiqe != NULL) { 594 tcpipqent_free(tiqe); 595 } 596 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 597 goto out; 598 } 599 /* 600 * Received segment completely overlaps this fragment 601 * so we drop the fragment (this keeps the temporal 602 * ordering of segments correct). 603 */ 604 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 605 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 606 rcvpartdupbyte += q->ipqe_len; 607 m_freem(q->ipqe_m); 608 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 609 goto free_ipqe; 610 } 611 /* 612 * RX'ed segment extends past the end of the 613 * fragment. Drop the overlapping bytes. Then 614 * merge the fragment and segment then treat as 615 * a longer received packet. 616 */ 617 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 618 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 619 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 620 #ifdef TCPREASS_DEBUG 621 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 622 tp, overlap, 623 pkt_seq, pkt_seq + pkt_len, pkt_len); 624 #endif 625 m_adj(m, overlap); 626 rcvpartdupbyte += overlap; 627 m_cat(q->ipre_mlast, m); 628 TRAVERSE(q->ipre_mlast); 629 m = q->ipqe_m; 630 pkt_seq = q->ipqe_seq; 631 pkt_len += q->ipqe_len - overlap; 632 rcvoobyte -= overlap; 633 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 634 goto free_ipqe; 635 } 636 /* 637 * RX'ed segment extends past the front of the 638 * fragment. Drop the overlapping bytes on the 639 * received packet. The packet will then be 640 * contatentated with this fragment a bit later. 641 */ 642 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 643 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 644 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 645 #ifdef TCPREASS_DEBUG 646 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 647 tp, overlap, 648 pkt_seq, pkt_seq + pkt_len, pkt_len); 649 #endif 650 m_adj(m, -overlap); 651 pkt_len -= overlap; 652 rcvpartdupbyte += overlap; 653 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 654 rcvoobyte -= overlap; 655 } 656 /* 657 * If the received segment immediates precedes this 658 * fragment then tack the fragment onto this segment 659 * and reinsert the data. 660 */ 661 if (q->ipqe_seq == pkt_seq + pkt_len) { 662 #ifdef TCPREASS_DEBUG 663 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 664 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 665 pkt_seq, pkt_seq + pkt_len, pkt_len); 666 #endif 667 pkt_len += q->ipqe_len; 668 pkt_flags |= q->ipqe_flags; 669 m_cat(m, q->ipqe_m); 670 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 671 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 672 tp->t_segqlen--; 673 KASSERT(tp->t_segqlen >= 0); 674 KASSERT(tp->t_segqlen != 0 || 675 (TAILQ_EMPTY(&tp->segq) && 676 TAILQ_EMPTY(&tp->timeq))); 677 if (tiqe == NULL) { 678 tiqe = q; 679 } else { 680 tcpipqent_free(q); 681 } 682 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 683 break; 684 } 685 /* 686 * If the fragment is before the segment, remember it. 687 * When this loop is terminated, p will contain the 688 * pointer to fragment that is right before the received 689 * segment. 690 */ 691 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 692 p = q; 693 694 continue; 695 696 /* 697 * This is a common operation. It also will allow 698 * to save doing a malloc/free in most instances. 699 */ 700 free_ipqe: 701 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 702 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 703 tp->t_segqlen--; 704 KASSERT(tp->t_segqlen >= 0); 705 KASSERT(tp->t_segqlen != 0 || 706 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 707 if (tiqe == NULL) { 708 tiqe = q; 709 } else { 710 tcpipqent_free(q); 711 } 712 } 713 714 #ifdef TCP_REASS_COUNTERS 715 if (count > 7) 716 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 717 else if (count > 0) 718 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 719 #endif 720 721 insert_it: 722 723 /* 724 * Allocate a new queue entry since the received segment did not 725 * collapse onto any other out-of-order block; thus we are allocating 726 * a new block. If it had collapsed, tiqe would not be NULL and 727 * we would be reusing it. 728 * XXX If we can't, just drop the packet. XXX 729 */ 730 if (tiqe == NULL) { 731 tiqe = tcpipqent_alloc(); 732 if (tiqe == NULL) { 733 TCP_STATINC(TCP_STAT_RCVMEMDROP); 734 m_freem(m); 735 goto out; 736 } 737 } 738 739 /* 740 * Update the counters. 741 */ 742 tp->t_rcvoopack++; 743 tcps = TCP_STAT_GETREF(); 744 tcps[TCP_STAT_RCVOOPACK]++; 745 tcps[TCP_STAT_RCVOOBYTE] += rcvoobyte; 746 if (rcvpartdupbyte) { 747 tcps[TCP_STAT_RCVPARTDUPPACK]++; 748 tcps[TCP_STAT_RCVPARTDUPBYTE] += rcvpartdupbyte; 749 } 750 TCP_STAT_PUTREF(); 751 752 /* 753 * Insert the new fragment queue entry into both queues. 754 */ 755 tiqe->ipqe_m = m; 756 tiqe->ipre_mlast = m; 757 tiqe->ipqe_seq = pkt_seq; 758 tiqe->ipqe_len = pkt_len; 759 tiqe->ipqe_flags = pkt_flags; 760 if (p == NULL) { 761 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 762 #ifdef TCPREASS_DEBUG 763 if (tiqe->ipqe_seq != tp->rcv_nxt) 764 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 765 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 766 #endif 767 } else { 768 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 769 #ifdef TCPREASS_DEBUG 770 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 771 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 772 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 773 #endif 774 } 775 tp->t_segqlen++; 776 777 skip_replacement: 778 779 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 780 781 present: 782 /* 783 * Present data to user, advancing rcv_nxt through 784 * completed sequence space. 785 */ 786 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 787 goto out; 788 q = TAILQ_FIRST(&tp->segq); 789 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 790 goto out; 791 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 792 goto out; 793 794 tp->rcv_nxt += q->ipqe_len; 795 pkt_flags = q->ipqe_flags & TH_FIN; 796 nd6_hint(tp); 797 798 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 799 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 800 tp->t_segqlen--; 801 KASSERT(tp->t_segqlen >= 0); 802 KASSERT(tp->t_segqlen != 0 || 803 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 804 if (so->so_state & SS_CANTRCVMORE) 805 m_freem(q->ipqe_m); 806 else 807 sbappendstream(&so->so_rcv, q->ipqe_m); 808 tcpipqent_free(q); 809 TCP_REASS_UNLOCK(tp); 810 sorwakeup(so); 811 return (pkt_flags); 812 out: 813 TCP_REASS_UNLOCK(tp); 814 return (0); 815 } 816 817 #ifdef INET6 818 int 819 tcp6_input(struct mbuf **mp, int *offp, int proto) 820 { 821 struct mbuf *m = *mp; 822 823 /* 824 * draft-itojun-ipv6-tcp-to-anycast 825 * better place to put this in? 826 */ 827 if (m->m_flags & M_ANYCAST6) { 828 struct ip6_hdr *ip6; 829 if (m->m_len < sizeof(struct ip6_hdr)) { 830 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 831 TCP_STATINC(TCP_STAT_RCVSHORT); 832 return IPPROTO_DONE; 833 } 834 } 835 ip6 = mtod(m, struct ip6_hdr *); 836 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 837 (char *)&ip6->ip6_dst - (char *)ip6); 838 return IPPROTO_DONE; 839 } 840 841 tcp_input(m, *offp, proto); 842 return IPPROTO_DONE; 843 } 844 #endif 845 846 #ifdef INET 847 static void 848 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 849 { 850 char src[INET_ADDRSTRLEN]; 851 char dst[INET_ADDRSTRLEN]; 852 853 if (ip) { 854 in_print(src, sizeof(src), &ip->ip_src); 855 in_print(dst, sizeof(dst), &ip->ip_dst); 856 } 857 else { 858 strlcpy(src, "(unknown)", sizeof(src)); 859 strlcpy(dst, "(unknown)", sizeof(dst)); 860 } 861 log(LOG_INFO, 862 "Connection attempt to TCP %s:%d from %s:%d\n", 863 dst, ntohs(th->th_dport), 864 src, ntohs(th->th_sport)); 865 } 866 #endif 867 868 #ifdef INET6 869 static void 870 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 871 { 872 char src[INET6_ADDRSTRLEN]; 873 char dst[INET6_ADDRSTRLEN]; 874 875 if (ip6) { 876 in6_print(src, sizeof(src), &ip6->ip6_src); 877 in6_print(dst, sizeof(dst), &ip6->ip6_dst); 878 } 879 else { 880 strlcpy(src, "(unknown v6)", sizeof(src)); 881 strlcpy(dst, "(unknown v6)", sizeof(dst)); 882 } 883 log(LOG_INFO, 884 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 885 dst, ntohs(th->th_dport), 886 src, ntohs(th->th_sport)); 887 } 888 #endif 889 890 /* 891 * Checksum extended TCP header and data. 892 */ 893 int 894 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, 895 int toff, int off, int tlen) 896 { 897 struct ifnet *rcvif; 898 int s; 899 900 /* 901 * XXX it's better to record and check if this mbuf is 902 * already checked. 903 */ 904 905 rcvif = m_get_rcvif(m, &s); 906 907 switch (af) { 908 #ifdef INET 909 case AF_INET: 910 switch (m->m_pkthdr.csum_flags & 911 ((rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 912 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 913 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 914 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 915 goto badcsum; 916 917 case M_CSUM_TCPv4|M_CSUM_DATA: { 918 u_int32_t hw_csum = m->m_pkthdr.csum_data; 919 920 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 921 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 922 const struct ip *ip = 923 mtod(m, const struct ip *); 924 925 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 926 ip->ip_dst.s_addr, 927 htons(hw_csum + tlen + off + IPPROTO_TCP)); 928 } 929 if ((hw_csum ^ 0xffff) != 0) 930 goto badcsum; 931 break; 932 } 933 934 case M_CSUM_TCPv4: 935 /* Checksum was okay. */ 936 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 937 break; 938 939 default: 940 /* 941 * Must compute it ourselves. Maybe skip checksum 942 * on loopback interfaces. 943 */ 944 if (__predict_true(!(rcvif->if_flags & IFF_LOOPBACK) || 945 tcp_do_loopback_cksum)) { 946 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 947 if (in4_cksum(m, IPPROTO_TCP, toff, 948 tlen + off) != 0) 949 goto badcsum; 950 } 951 break; 952 } 953 break; 954 #endif /* INET4 */ 955 956 #ifdef INET6 957 case AF_INET6: 958 switch (m->m_pkthdr.csum_flags & 959 ((rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 960 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 961 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 962 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 963 goto badcsum; 964 965 #if 0 /* notyet */ 966 case M_CSUM_TCPv6|M_CSUM_DATA: 967 #endif 968 969 case M_CSUM_TCPv6: 970 /* Checksum was okay. */ 971 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 972 break; 973 974 default: 975 /* 976 * Must compute it ourselves. Maybe skip checksum 977 * on loopback interfaces. 978 */ 979 if (__predict_true((m->m_flags & M_LOOP) == 0 || 980 tcp_do_loopback_cksum)) { 981 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 982 if (in6_cksum(m, IPPROTO_TCP, toff, 983 tlen + off) != 0) 984 goto badcsum; 985 } 986 } 987 break; 988 #endif /* INET6 */ 989 } 990 m_put_rcvif(rcvif, &s); 991 992 return 0; 993 994 badcsum: 995 m_put_rcvif(rcvif, &s); 996 TCP_STATINC(TCP_STAT_RCVBADSUM); 997 return -1; 998 } 999 1000 /* When a packet arrives addressed to a vestigial tcpbp, we 1001 * nevertheless have to respond to it per the spec. 1002 */ 1003 static void tcp_vtw_input(struct tcphdr *th, vestigial_inpcb_t *vp, 1004 struct mbuf *m, int tlen, int multicast) 1005 { 1006 int tiflags; 1007 int todrop; 1008 uint32_t t_flags = 0; 1009 uint64_t *tcps; 1010 1011 tiflags = th->th_flags; 1012 todrop = vp->rcv_nxt - th->th_seq; 1013 1014 if (todrop > 0) { 1015 if (tiflags & TH_SYN) { 1016 tiflags &= ~TH_SYN; 1017 ++th->th_seq; 1018 if (th->th_urp > 1) 1019 --th->th_urp; 1020 else { 1021 tiflags &= ~TH_URG; 1022 th->th_urp = 0; 1023 } 1024 --todrop; 1025 } 1026 if (todrop > tlen || 1027 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1028 /* 1029 * Any valid FIN or RST must be to the left of the 1030 * window. At this point the FIN or RST must be a 1031 * duplicate or out of sequence; drop it. 1032 */ 1033 if (tiflags & TH_RST) 1034 goto drop; 1035 tiflags &= ~(TH_FIN|TH_RST); 1036 /* 1037 * Send an ACK to resynchronize and drop any data. 1038 * But keep on processing for RST or ACK. 1039 */ 1040 t_flags |= TF_ACKNOW; 1041 todrop = tlen; 1042 tcps = TCP_STAT_GETREF(); 1043 tcps[TCP_STAT_RCVDUPPACK] += 1; 1044 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 1045 TCP_STAT_PUTREF(); 1046 } else if ((tiflags & TH_RST) 1047 && th->th_seq != vp->rcv_nxt) { 1048 /* 1049 * Test for reset before adjusting the sequence 1050 * number for overlapping data. 1051 */ 1052 goto dropafterack_ratelim; 1053 } else { 1054 tcps = TCP_STAT_GETREF(); 1055 tcps[TCP_STAT_RCVPARTDUPPACK] += 1; 1056 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 1057 TCP_STAT_PUTREF(); 1058 } 1059 1060 // tcp_new_dsack(tp, th->th_seq, todrop); 1061 // hdroptlen += todrop; /*drop from head afterwards*/ 1062 1063 th->th_seq += todrop; 1064 tlen -= todrop; 1065 1066 if (th->th_urp > todrop) 1067 th->th_urp -= todrop; 1068 else { 1069 tiflags &= ~TH_URG; 1070 th->th_urp = 0; 1071 } 1072 } 1073 1074 /* 1075 * If new data are received on a connection after the 1076 * user processes are gone, then RST the other end. 1077 */ 1078 if (tlen) { 1079 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 1080 goto dropwithreset; 1081 } 1082 1083 /* 1084 * If segment ends after window, drop trailing data 1085 * (and PUSH and FIN); if nothing left, just ACK. 1086 */ 1087 todrop = (th->th_seq + tlen) - (vp->rcv_nxt+vp->rcv_wnd); 1088 1089 if (todrop > 0) { 1090 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 1091 if (todrop >= tlen) { 1092 /* 1093 * The segment actually starts after the window. 1094 * th->th_seq + tlen - vp->rcv_nxt - vp->rcv_wnd >= tlen 1095 * th->th_seq - vp->rcv_nxt - vp->rcv_wnd >= 0 1096 * th->th_seq >= vp->rcv_nxt + vp->rcv_wnd 1097 */ 1098 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 1099 /* 1100 * If a new connection request is received 1101 * while in TIME_WAIT, drop the old connection 1102 * and start over if the sequence numbers 1103 * are above the previous ones. 1104 */ 1105 if ((tiflags & TH_SYN) 1106 && SEQ_GT(th->th_seq, vp->rcv_nxt)) { 1107 /* We only support this in the !NOFDREF case, which 1108 * is to say: not here. 1109 */ 1110 goto dropwithreset; 1111 } 1112 /* 1113 * If window is closed can only take segments at 1114 * window edge, and have to drop data and PUSH from 1115 * incoming segments. Continue processing, but 1116 * remember to ack. Otherwise, drop segment 1117 * and (if not RST) ack. 1118 */ 1119 if (vp->rcv_wnd == 0 && th->th_seq == vp->rcv_nxt) { 1120 t_flags |= TF_ACKNOW; 1121 TCP_STATINC(TCP_STAT_RCVWINPROBE); 1122 } else 1123 goto dropafterack; 1124 } else 1125 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 1126 m_adj(m, -todrop); 1127 tlen -= todrop; 1128 tiflags &= ~(TH_PUSH|TH_FIN); 1129 } 1130 1131 if (tiflags & TH_RST) { 1132 if (th->th_seq != vp->rcv_nxt) 1133 goto dropafterack_ratelim; 1134 1135 vtw_del(vp->ctl, vp->vtw); 1136 goto drop; 1137 } 1138 1139 /* 1140 * If the ACK bit is off we drop the segment and return. 1141 */ 1142 if ((tiflags & TH_ACK) == 0) { 1143 if (t_flags & TF_ACKNOW) 1144 goto dropafterack; 1145 else 1146 goto drop; 1147 } 1148 1149 /* 1150 * In TIME_WAIT state the only thing that should arrive 1151 * is a retransmission of the remote FIN. Acknowledge 1152 * it and restart the finack timer. 1153 */ 1154 vtw_restart(vp); 1155 goto dropafterack; 1156 1157 dropafterack: 1158 /* 1159 * Generate an ACK dropping incoming segment if it occupies 1160 * sequence space, where the ACK reflects our state. 1161 */ 1162 if (tiflags & TH_RST) 1163 goto drop; 1164 goto dropafterack2; 1165 1166 dropafterack_ratelim: 1167 /* 1168 * We may want to rate-limit ACKs against SYN/RST attack. 1169 */ 1170 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 1171 tcp_ackdrop_ppslim) == 0) { 1172 /* XXX stat */ 1173 goto drop; 1174 } 1175 /* ...fall into dropafterack2... */ 1176 1177 dropafterack2: 1178 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, th->th_ack, 1179 TH_ACK); 1180 return; 1181 1182 dropwithreset: 1183 /* 1184 * Generate a RST, dropping incoming segment. 1185 * Make ACK acceptable to originator of segment. 1186 */ 1187 if (tiflags & TH_RST) 1188 goto drop; 1189 1190 if (tiflags & TH_ACK) 1191 tcp_respond(0, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 1192 else { 1193 if (tiflags & TH_SYN) 1194 ++tlen; 1195 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, (tcp_seq)0, 1196 TH_RST|TH_ACK); 1197 } 1198 return; 1199 drop: 1200 m_freem(m); 1201 } 1202 1203 /* 1204 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 1205 */ 1206 void 1207 tcp_input(struct mbuf *m, ...) 1208 { 1209 struct tcphdr *th; 1210 struct ip *ip; 1211 struct inpcb *inp; 1212 #ifdef INET6 1213 struct ip6_hdr *ip6; 1214 struct in6pcb *in6p; 1215 #endif 1216 u_int8_t *optp = NULL; 1217 int optlen = 0; 1218 int len, tlen, toff, hdroptlen = 0; 1219 struct tcpcb *tp = 0; 1220 int tiflags; 1221 struct socket *so = NULL; 1222 int todrop, acked, ourfinisacked, needoutput = 0; 1223 bool dupseg; 1224 #ifdef TCP_DEBUG 1225 short ostate = 0; 1226 #endif 1227 u_long tiwin; 1228 struct tcp_opt_info opti; 1229 int off, iphlen; 1230 va_list ap; 1231 int af; /* af on the wire */ 1232 struct mbuf *tcp_saveti = NULL; 1233 uint32_t ts_rtt; 1234 uint8_t iptos; 1235 uint64_t *tcps; 1236 vestigial_inpcb_t vestige; 1237 1238 vestige.valid = 0; 1239 1240 MCLAIM(m, &tcp_rx_mowner); 1241 va_start(ap, m); 1242 toff = va_arg(ap, int); 1243 (void)va_arg(ap, int); /* ignore value, advance ap */ 1244 va_end(ap); 1245 1246 TCP_STATINC(TCP_STAT_RCVTOTAL); 1247 1248 memset(&opti, 0, sizeof(opti)); 1249 opti.ts_present = 0; 1250 opti.maxseg = 0; 1251 1252 /* 1253 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 1254 * 1255 * TCP is, by definition, unicast, so we reject all 1256 * multicast outright. 1257 * 1258 * Note, there are additional src/dst address checks in 1259 * the AF-specific code below. 1260 */ 1261 if (m->m_flags & (M_BCAST|M_MCAST)) { 1262 /* XXX stat */ 1263 goto drop; 1264 } 1265 #ifdef INET6 1266 if (m->m_flags & M_ANYCAST6) { 1267 /* XXX stat */ 1268 goto drop; 1269 } 1270 #endif 1271 1272 /* 1273 * Get IP and TCP header. 1274 * Note: IP leaves IP header in first mbuf. 1275 */ 1276 ip = mtod(m, struct ip *); 1277 switch (ip->ip_v) { 1278 #ifdef INET 1279 case 4: 1280 #ifdef INET6 1281 ip6 = NULL; 1282 #endif 1283 af = AF_INET; 1284 iphlen = sizeof(struct ip); 1285 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1286 sizeof(struct tcphdr)); 1287 if (th == NULL) { 1288 TCP_STATINC(TCP_STAT_RCVSHORT); 1289 return; 1290 } 1291 /* We do the checksum after PCB lookup... */ 1292 len = ntohs(ip->ip_len); 1293 tlen = len - toff; 1294 iptos = ip->ip_tos; 1295 break; 1296 #endif 1297 #ifdef INET6 1298 case 6: 1299 ip = NULL; 1300 iphlen = sizeof(struct ip6_hdr); 1301 af = AF_INET6; 1302 ip6 = mtod(m, struct ip6_hdr *); 1303 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1304 sizeof(struct tcphdr)); 1305 if (th == NULL) { 1306 TCP_STATINC(TCP_STAT_RCVSHORT); 1307 return; 1308 } 1309 1310 /* Be proactive about malicious use of IPv4 mapped address */ 1311 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 1312 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 1313 /* XXX stat */ 1314 goto drop; 1315 } 1316 1317 /* 1318 * Be proactive about unspecified IPv6 address in source. 1319 * As we use all-zero to indicate unbounded/unconnected pcb, 1320 * unspecified IPv6 address can be used to confuse us. 1321 * 1322 * Note that packets with unspecified IPv6 destination is 1323 * already dropped in ip6_input. 1324 */ 1325 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1326 /* XXX stat */ 1327 goto drop; 1328 } 1329 1330 /* 1331 * Make sure destination address is not multicast. 1332 * Source address checked in ip6_input(). 1333 */ 1334 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1335 /* XXX stat */ 1336 goto drop; 1337 } 1338 1339 /* We do the checksum after PCB lookup... */ 1340 len = m->m_pkthdr.len; 1341 tlen = len - toff; 1342 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1343 break; 1344 #endif 1345 default: 1346 m_freem(m); 1347 return; 1348 } 1349 1350 KASSERT(TCP_HDR_ALIGNED_P(th)); 1351 1352 /* 1353 * Check that TCP offset makes sense, 1354 * pull out TCP options and adjust length. XXX 1355 */ 1356 off = th->th_off << 2; 1357 if (off < sizeof (struct tcphdr) || off > tlen) { 1358 TCP_STATINC(TCP_STAT_RCVBADOFF); 1359 goto drop; 1360 } 1361 tlen -= off; 1362 1363 /* 1364 * tcp_input() has been modified to use tlen to mean the TCP data 1365 * length throughout the function. Other functions can use 1366 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1367 * rja 1368 */ 1369 1370 if (off > sizeof (struct tcphdr)) { 1371 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1372 if (th == NULL) { 1373 TCP_STATINC(TCP_STAT_RCVSHORT); 1374 return; 1375 } 1376 /* 1377 * NOTE: ip/ip6 will not be affected by m_pulldown() 1378 * (as they're before toff) and we don't need to update those. 1379 */ 1380 KASSERT(TCP_HDR_ALIGNED_P(th)); 1381 optlen = off - sizeof (struct tcphdr); 1382 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1383 /* 1384 * Do quick retrieval of timestamp options ("options 1385 * prediction?"). If timestamp is the only option and it's 1386 * formatted as recommended in RFC 1323 appendix A, we 1387 * quickly get the values now and not bother calling 1388 * tcp_dooptions(), etc. 1389 */ 1390 if ((optlen == TCPOLEN_TSTAMP_APPA || 1391 (optlen > TCPOLEN_TSTAMP_APPA && 1392 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1393 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1394 (th->th_flags & TH_SYN) == 0) { 1395 opti.ts_present = 1; 1396 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1397 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1398 optp = NULL; /* we've parsed the options */ 1399 } 1400 } 1401 tiflags = th->th_flags; 1402 1403 /* 1404 * Checksum extended TCP header and data 1405 */ 1406 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1407 goto badcsum; 1408 1409 /* 1410 * Locate pcb for segment. 1411 */ 1412 findpcb: 1413 inp = NULL; 1414 #ifdef INET6 1415 in6p = NULL; 1416 #endif 1417 switch (af) { 1418 #ifdef INET 1419 case AF_INET: 1420 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1421 ip->ip_dst, th->th_dport, 1422 &vestige); 1423 if (inp == 0 && !vestige.valid) { 1424 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1425 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1426 } 1427 #ifdef INET6 1428 if (inp == 0 && !vestige.valid) { 1429 struct in6_addr s, d; 1430 1431 /* mapped addr case */ 1432 in6_in_2_v4mapin6(&ip->ip_src, &s); 1433 in6_in_2_v4mapin6(&ip->ip_dst, &d); 1434 in6p = in6_pcblookup_connect(&tcbtable, &s, 1435 th->th_sport, &d, th->th_dport, 1436 0, &vestige); 1437 if (in6p == 0 && !vestige.valid) { 1438 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1439 in6p = in6_pcblookup_bind(&tcbtable, &d, 1440 th->th_dport, 0); 1441 } 1442 } 1443 #endif 1444 #ifndef INET6 1445 if (inp == 0 && !vestige.valid) 1446 #else 1447 if (inp == 0 && in6p == 0 && !vestige.valid) 1448 #endif 1449 { 1450 TCP_STATINC(TCP_STAT_NOPORT); 1451 if (tcp_log_refused && 1452 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1453 tcp4_log_refused(ip, th); 1454 } 1455 tcp_fields_to_host(th); 1456 goto dropwithreset_ratelim; 1457 } 1458 #if defined(IPSEC) 1459 if (ipsec_used) { 1460 if (inp && 1461 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 1462 && ipsec4_in_reject(m, inp)) { 1463 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1464 goto drop; 1465 } 1466 #ifdef INET6 1467 else if (in6p && 1468 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 1469 && ipsec6_in_reject_so(m, in6p->in6p_socket)) { 1470 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1471 goto drop; 1472 } 1473 #endif 1474 } 1475 #endif /*IPSEC*/ 1476 break; 1477 #endif /*INET*/ 1478 #ifdef INET6 1479 case AF_INET6: 1480 { 1481 int faith; 1482 1483 #if defined(NFAITH) && NFAITH > 0 1484 faith = faithprefix(&ip6->ip6_dst); 1485 #else 1486 faith = 0; 1487 #endif 1488 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1489 th->th_sport, &ip6->ip6_dst, th->th_dport, faith, &vestige); 1490 if (!in6p && !vestige.valid) { 1491 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1492 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1493 th->th_dport, faith); 1494 } 1495 if (!in6p && !vestige.valid) { 1496 TCP_STATINC(TCP_STAT_NOPORT); 1497 if (tcp_log_refused && 1498 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1499 tcp6_log_refused(ip6, th); 1500 } 1501 tcp_fields_to_host(th); 1502 goto dropwithreset_ratelim; 1503 } 1504 #if defined(IPSEC) 1505 if (ipsec_used && in6p 1506 && (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 1507 && ipsec6_in_reject(m, in6p)) { 1508 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1509 goto drop; 1510 } 1511 #endif /*IPSEC*/ 1512 break; 1513 } 1514 #endif 1515 } 1516 1517 /* 1518 * If the state is CLOSED (i.e., TCB does not exist) then 1519 * all data in the incoming segment is discarded. 1520 * If the TCB exists but is in CLOSED state, it is embryonic, 1521 * but should either do a listen or a connect soon. 1522 */ 1523 tp = NULL; 1524 so = NULL; 1525 if (inp) { 1526 /* Check the minimum TTL for socket. */ 1527 if (ip->ip_ttl < inp->inp_ip_minttl) 1528 goto drop; 1529 1530 tp = intotcpcb(inp); 1531 so = inp->inp_socket; 1532 } 1533 #ifdef INET6 1534 else if (in6p) { 1535 tp = in6totcpcb(in6p); 1536 so = in6p->in6p_socket; 1537 } 1538 #endif 1539 else if (vestige.valid) { 1540 int mc = 0; 1541 1542 /* We do not support the resurrection of vtw tcpcps. 1543 */ 1544 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1545 goto badcsum; 1546 1547 switch (af) { 1548 #ifdef INET6 1549 case AF_INET6: 1550 mc = IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst); 1551 break; 1552 #endif 1553 1554 case AF_INET: 1555 mc = (IN_MULTICAST(ip->ip_dst.s_addr) 1556 || in_broadcast(ip->ip_dst, 1557 m_get_rcvif_NOMPSAFE(m))); 1558 break; 1559 } 1560 1561 tcp_fields_to_host(th); 1562 tcp_vtw_input(th, &vestige, m, tlen, mc); 1563 m = 0; 1564 goto drop; 1565 } 1566 1567 if (tp == 0) { 1568 tcp_fields_to_host(th); 1569 goto dropwithreset_ratelim; 1570 } 1571 if (tp->t_state == TCPS_CLOSED) 1572 goto drop; 1573 1574 KASSERT(so->so_lock == softnet_lock); 1575 KASSERT(solocked(so)); 1576 1577 tcp_fields_to_host(th); 1578 1579 /* Unscale the window into a 32-bit value. */ 1580 if ((tiflags & TH_SYN) == 0) 1581 tiwin = th->th_win << tp->snd_scale; 1582 else 1583 tiwin = th->th_win; 1584 1585 #ifdef INET6 1586 /* save packet options if user wanted */ 1587 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1588 if (in6p->in6p_options) { 1589 m_freem(in6p->in6p_options); 1590 in6p->in6p_options = 0; 1591 } 1592 KASSERT(ip6 != NULL); 1593 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1594 } 1595 #endif 1596 1597 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1598 union syn_cache_sa src; 1599 union syn_cache_sa dst; 1600 1601 memset(&src, 0, sizeof(src)); 1602 memset(&dst, 0, sizeof(dst)); 1603 switch (af) { 1604 #ifdef INET 1605 case AF_INET: 1606 src.sin.sin_len = sizeof(struct sockaddr_in); 1607 src.sin.sin_family = AF_INET; 1608 src.sin.sin_addr = ip->ip_src; 1609 src.sin.sin_port = th->th_sport; 1610 1611 dst.sin.sin_len = sizeof(struct sockaddr_in); 1612 dst.sin.sin_family = AF_INET; 1613 dst.sin.sin_addr = ip->ip_dst; 1614 dst.sin.sin_port = th->th_dport; 1615 break; 1616 #endif 1617 #ifdef INET6 1618 case AF_INET6: 1619 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1620 src.sin6.sin6_family = AF_INET6; 1621 src.sin6.sin6_addr = ip6->ip6_src; 1622 src.sin6.sin6_port = th->th_sport; 1623 1624 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1625 dst.sin6.sin6_family = AF_INET6; 1626 dst.sin6.sin6_addr = ip6->ip6_dst; 1627 dst.sin6.sin6_port = th->th_dport; 1628 break; 1629 #endif /* INET6 */ 1630 default: 1631 goto badsyn; /*sanity*/ 1632 } 1633 1634 if (so->so_options & SO_DEBUG) { 1635 #ifdef TCP_DEBUG 1636 ostate = tp->t_state; 1637 #endif 1638 1639 tcp_saveti = NULL; 1640 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1641 goto nosave; 1642 1643 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1644 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1645 if (!tcp_saveti) 1646 goto nosave; 1647 } else { 1648 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1649 if (!tcp_saveti) 1650 goto nosave; 1651 MCLAIM(m, &tcp_mowner); 1652 tcp_saveti->m_len = iphlen; 1653 m_copydata(m, 0, iphlen, 1654 mtod(tcp_saveti, void *)); 1655 } 1656 1657 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1658 m_freem(tcp_saveti); 1659 tcp_saveti = NULL; 1660 } else { 1661 tcp_saveti->m_len += sizeof(struct tcphdr); 1662 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1663 sizeof(struct tcphdr)); 1664 } 1665 nosave:; 1666 } 1667 if (so->so_options & SO_ACCEPTCONN) { 1668 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1669 if (tiflags & TH_RST) { 1670 syn_cache_reset(&src.sa, &dst.sa, th); 1671 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1672 (TH_ACK|TH_SYN)) { 1673 /* 1674 * Received a SYN,ACK. This should 1675 * never happen while we are in 1676 * LISTEN. Send an RST. 1677 */ 1678 goto badsyn; 1679 } else if (tiflags & TH_ACK) { 1680 so = syn_cache_get(&src.sa, &dst.sa, 1681 th, toff, tlen, so, m); 1682 if (so == NULL) { 1683 /* 1684 * We don't have a SYN for 1685 * this ACK; send an RST. 1686 */ 1687 goto badsyn; 1688 } else if (so == 1689 (struct socket *)(-1)) { 1690 /* 1691 * We were unable to create 1692 * the connection. If the 1693 * 3-way handshake was 1694 * completed, and RST has 1695 * been sent to the peer. 1696 * Since the mbuf might be 1697 * in use for the reply, 1698 * do not free it. 1699 */ 1700 m = NULL; 1701 } else { 1702 /* 1703 * We have created a 1704 * full-blown connection. 1705 */ 1706 tp = NULL; 1707 inp = NULL; 1708 #ifdef INET6 1709 in6p = NULL; 1710 #endif 1711 switch (so->so_proto->pr_domain->dom_family) { 1712 #ifdef INET 1713 case AF_INET: 1714 inp = sotoinpcb(so); 1715 tp = intotcpcb(inp); 1716 break; 1717 #endif 1718 #ifdef INET6 1719 case AF_INET6: 1720 in6p = sotoin6pcb(so); 1721 tp = in6totcpcb(in6p); 1722 break; 1723 #endif 1724 } 1725 if (tp == NULL) 1726 goto badsyn; /*XXX*/ 1727 tiwin <<= tp->snd_scale; 1728 goto after_listen; 1729 } 1730 } else { 1731 /* 1732 * None of RST, SYN or ACK was set. 1733 * This is an invalid packet for a 1734 * TCB in LISTEN state. Send a RST. 1735 */ 1736 goto badsyn; 1737 } 1738 } else { 1739 /* 1740 * Received a SYN. 1741 * 1742 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1743 */ 1744 if (m->m_flags & (M_BCAST|M_MCAST)) 1745 goto drop; 1746 1747 switch (af) { 1748 #ifdef INET6 1749 case AF_INET6: 1750 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1751 goto drop; 1752 break; 1753 #endif /* INET6 */ 1754 case AF_INET: 1755 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1756 in_broadcast(ip->ip_dst, 1757 m_get_rcvif_NOMPSAFE(m))) 1758 goto drop; 1759 break; 1760 } 1761 1762 #ifdef INET6 1763 /* 1764 * If deprecated address is forbidden, we do 1765 * not accept SYN to deprecated interface 1766 * address to prevent any new inbound 1767 * connection from getting established. 1768 * When we do not accept SYN, we send a TCP 1769 * RST, with deprecated source address (instead 1770 * of dropping it). We compromise it as it is 1771 * much better for peer to send a RST, and 1772 * RST will be the final packet for the 1773 * exchange. 1774 * 1775 * If we do not forbid deprecated addresses, we 1776 * accept the SYN packet. RFC2462 does not 1777 * suggest dropping SYN in this case. 1778 * If we decipher RFC2462 5.5.4, it says like 1779 * this: 1780 * 1. use of deprecated addr with existing 1781 * communication is okay - "SHOULD continue 1782 * to be used" 1783 * 2. use of it with new communication: 1784 * (2a) "SHOULD NOT be used if alternate 1785 * address with sufficient scope is 1786 * available" 1787 * (2b) nothing mentioned otherwise. 1788 * Here we fall into (2b) case as we have no 1789 * choice in our source address selection - we 1790 * must obey the peer. 1791 * 1792 * The wording in RFC2462 is confusing, and 1793 * there are multiple description text for 1794 * deprecated address handling - worse, they 1795 * are not exactly the same. I believe 5.5.4 1796 * is the best one, so we follow 5.5.4. 1797 */ 1798 if (af == AF_INET6 && !ip6_use_deprecated) { 1799 struct in6_ifaddr *ia6; 1800 int s; 1801 struct ifnet *rcvif = m_get_rcvif(m, &s); 1802 if (rcvif == NULL) 1803 goto dropwithreset; /* XXX */ 1804 if ((ia6 = in6ifa_ifpwithaddr(rcvif, 1805 &ip6->ip6_dst)) && 1806 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1807 tp = NULL; 1808 m_put_rcvif(rcvif, &s); 1809 goto dropwithreset; 1810 } 1811 m_put_rcvif(rcvif, &s); 1812 } 1813 #endif 1814 1815 #if defined(IPSEC) 1816 if (ipsec_used) { 1817 switch (af) { 1818 #ifdef INET 1819 case AF_INET: 1820 if (!ipsec4_in_reject_so(m, so)) 1821 break; 1822 IPSEC_STATINC( 1823 IPSEC_STAT_IN_POLVIO); 1824 tp = NULL; 1825 goto dropwithreset; 1826 #endif 1827 #ifdef INET6 1828 case AF_INET6: 1829 if (!ipsec6_in_reject_so(m, so)) 1830 break; 1831 IPSEC6_STATINC( 1832 IPSEC_STAT_IN_POLVIO); 1833 tp = NULL; 1834 goto dropwithreset; 1835 #endif /*INET6*/ 1836 } 1837 } 1838 #endif /*IPSEC*/ 1839 1840 /* 1841 * LISTEN socket received a SYN 1842 * from itself? This can't possibly 1843 * be valid; drop the packet. 1844 */ 1845 if (th->th_sport == th->th_dport) { 1846 int i; 1847 1848 switch (af) { 1849 #ifdef INET 1850 case AF_INET: 1851 i = in_hosteq(ip->ip_src, ip->ip_dst); 1852 break; 1853 #endif 1854 #ifdef INET6 1855 case AF_INET6: 1856 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1857 break; 1858 #endif 1859 default: 1860 i = 1; 1861 } 1862 if (i) { 1863 TCP_STATINC(TCP_STAT_BADSYN); 1864 goto drop; 1865 } 1866 } 1867 1868 /* 1869 * SYN looks ok; create compressed TCP 1870 * state for it. 1871 */ 1872 if (so->so_qlen <= so->so_qlimit && 1873 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1874 so, m, optp, optlen, &opti)) 1875 m = NULL; 1876 } 1877 goto drop; 1878 } 1879 } 1880 1881 after_listen: 1882 #ifdef DIAGNOSTIC 1883 /* 1884 * Should not happen now that all embryonic connections 1885 * are handled with compressed state. 1886 */ 1887 if (tp->t_state == TCPS_LISTEN) 1888 panic("tcp_input: TCPS_LISTEN"); 1889 #endif 1890 1891 /* 1892 * Segment received on connection. 1893 * Reset idle time and keep-alive timer. 1894 */ 1895 tp->t_rcvtime = tcp_now; 1896 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1897 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1898 1899 /* 1900 * Process options. 1901 */ 1902 #ifdef TCP_SIGNATURE 1903 if (optp || (tp->t_flags & TF_SIGNATURE)) 1904 #else 1905 if (optp) 1906 #endif 1907 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1908 goto drop; 1909 1910 if (TCP_SACK_ENABLED(tp)) { 1911 tcp_del_sackholes(tp, th); 1912 } 1913 1914 if (TCP_ECN_ALLOWED(tp)) { 1915 if (tiflags & TH_CWR) { 1916 tp->t_flags &= ~TF_ECN_SND_ECE; 1917 } 1918 switch (iptos & IPTOS_ECN_MASK) { 1919 case IPTOS_ECN_CE: 1920 tp->t_flags |= TF_ECN_SND_ECE; 1921 TCP_STATINC(TCP_STAT_ECN_CE); 1922 break; 1923 case IPTOS_ECN_ECT0: 1924 TCP_STATINC(TCP_STAT_ECN_ECT); 1925 break; 1926 case IPTOS_ECN_ECT1: 1927 /* XXX: ignore for now -- rpaulo */ 1928 break; 1929 } 1930 /* 1931 * Congestion experienced. 1932 * Ignore if we are already trying to recover. 1933 */ 1934 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1935 tp->t_congctl->cong_exp(tp); 1936 } 1937 1938 if (opti.ts_present && opti.ts_ecr) { 1939 /* 1940 * Calculate the RTT from the returned time stamp and the 1941 * connection's time base. If the time stamp is later than 1942 * the current time, or is extremely old, fall back to non-1323 1943 * RTT calculation. Since ts_rtt is unsigned, we can test both 1944 * at the same time. 1945 * 1946 * Note that ts_rtt is in units of slow ticks (500 1947 * ms). Since most earthbound RTTs are < 500 ms, 1948 * observed values will have large quantization noise. 1949 * Our smoothed RTT is then the fraction of observed 1950 * samples that are 1 tick instead of 0 (times 500 1951 * ms). 1952 * 1953 * ts_rtt is increased by 1 to denote a valid sample, 1954 * with 0 indicating an invalid measurement. This 1955 * extra 1 must be removed when ts_rtt is used, or 1956 * else an an erroneous extra 500 ms will result. 1957 */ 1958 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1959 if (ts_rtt > TCP_PAWS_IDLE) 1960 ts_rtt = 0; 1961 } else { 1962 ts_rtt = 0; 1963 } 1964 1965 /* 1966 * Header prediction: check for the two common cases 1967 * of a uni-directional data xfer. If the packet has 1968 * no control flags, is in-sequence, the window didn't 1969 * change and we're not retransmitting, it's a 1970 * candidate. If the length is zero and the ack moved 1971 * forward, we're the sender side of the xfer. Just 1972 * free the data acked & wake any higher level process 1973 * that was blocked waiting for space. If the length 1974 * is non-zero and the ack didn't move, we're the 1975 * receiver side. If we're getting packets in-order 1976 * (the reassembly queue is empty), add the data to 1977 * the socket buffer and note that we need a delayed ack. 1978 */ 1979 if (tp->t_state == TCPS_ESTABLISHED && 1980 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1981 == TH_ACK && 1982 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1983 th->th_seq == tp->rcv_nxt && 1984 tiwin && tiwin == tp->snd_wnd && 1985 tp->snd_nxt == tp->snd_max) { 1986 1987 /* 1988 * If last ACK falls within this segment's sequence numbers, 1989 * record the timestamp. 1990 * NOTE that the test is modified according to the latest 1991 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1992 * 1993 * note that we already know 1994 * TSTMP_GEQ(opti.ts_val, tp->ts_recent) 1995 */ 1996 if (opti.ts_present && 1997 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1998 tp->ts_recent_age = tcp_now; 1999 tp->ts_recent = opti.ts_val; 2000 } 2001 2002 if (tlen == 0) { 2003 /* Ack prediction. */ 2004 if (SEQ_GT(th->th_ack, tp->snd_una) && 2005 SEQ_LEQ(th->th_ack, tp->snd_max) && 2006 tp->snd_cwnd >= tp->snd_wnd && 2007 tp->t_partialacks < 0) { 2008 /* 2009 * this is a pure ack for outstanding data. 2010 */ 2011 if (ts_rtt) 2012 tcp_xmit_timer(tp, ts_rtt - 1); 2013 else if (tp->t_rtttime && 2014 SEQ_GT(th->th_ack, tp->t_rtseq)) 2015 tcp_xmit_timer(tp, 2016 tcp_now - tp->t_rtttime); 2017 acked = th->th_ack - tp->snd_una; 2018 tcps = TCP_STAT_GETREF(); 2019 tcps[TCP_STAT_PREDACK]++; 2020 tcps[TCP_STAT_RCVACKPACK]++; 2021 tcps[TCP_STAT_RCVACKBYTE] += acked; 2022 TCP_STAT_PUTREF(); 2023 nd6_hint(tp); 2024 2025 if (acked > (tp->t_lastoff - tp->t_inoff)) 2026 tp->t_lastm = NULL; 2027 sbdrop(&so->so_snd, acked); 2028 tp->t_lastoff -= acked; 2029 2030 icmp_check(tp, th, acked); 2031 2032 tp->snd_una = th->th_ack; 2033 tp->snd_fack = tp->snd_una; 2034 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2035 tp->snd_high = tp->snd_una; 2036 m_freem(m); 2037 2038 /* 2039 * If all outstanding data are acked, stop 2040 * retransmit timer, otherwise restart timer 2041 * using current (possibly backed-off) value. 2042 * If process is waiting for space, 2043 * wakeup/selnotify/signal. If data 2044 * are ready to send, let tcp_output 2045 * decide between more output or persist. 2046 */ 2047 if (tp->snd_una == tp->snd_max) 2048 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2049 else if (TCP_TIMER_ISARMED(tp, 2050 TCPT_PERSIST) == 0) 2051 TCP_TIMER_ARM(tp, TCPT_REXMT, 2052 tp->t_rxtcur); 2053 2054 sowwakeup(so); 2055 if (so->so_snd.sb_cc) { 2056 KERNEL_LOCK(1, NULL); 2057 (void) tcp_output(tp); 2058 KERNEL_UNLOCK_ONE(NULL); 2059 } 2060 if (tcp_saveti) 2061 m_freem(tcp_saveti); 2062 return; 2063 } 2064 } else if (th->th_ack == tp->snd_una && 2065 TAILQ_FIRST(&tp->segq) == NULL && 2066 tlen <= sbspace(&so->so_rcv)) { 2067 int newsize = 0; /* automatic sockbuf scaling */ 2068 2069 /* 2070 * this is a pure, in-sequence data packet 2071 * with nothing on the reassembly queue and 2072 * we have enough buffer space to take it. 2073 */ 2074 tp->rcv_nxt += tlen; 2075 tcps = TCP_STAT_GETREF(); 2076 tcps[TCP_STAT_PREDDAT]++; 2077 tcps[TCP_STAT_RCVPACK]++; 2078 tcps[TCP_STAT_RCVBYTE] += tlen; 2079 TCP_STAT_PUTREF(); 2080 nd6_hint(tp); 2081 2082 /* 2083 * Automatic sizing enables the performance of large buffers 2084 * and most of the efficiency of small ones by only allocating 2085 * space when it is needed. 2086 * 2087 * On the receive side the socket buffer memory is only rarely 2088 * used to any significant extent. This allows us to be much 2089 * more aggressive in scaling the receive socket buffer. For 2090 * the case that the buffer space is actually used to a large 2091 * extent and we run out of kernel memory we can simply drop 2092 * the new segments; TCP on the sender will just retransmit it 2093 * later. Setting the buffer size too big may only consume too 2094 * much kernel memory if the application doesn't read() from 2095 * the socket or packet loss or reordering makes use of the 2096 * reassembly queue. 2097 * 2098 * The criteria to step up the receive buffer one notch are: 2099 * 1. the number of bytes received during the time it takes 2100 * one timestamp to be reflected back to us (the RTT); 2101 * 2. received bytes per RTT is within seven eighth of the 2102 * current socket buffer size; 2103 * 3. receive buffer size has not hit maximal automatic size; 2104 * 2105 * This algorithm does one step per RTT at most and only if 2106 * we receive a bulk stream w/o packet losses or reorderings. 2107 * Shrinking the buffer during idle times is not necessary as 2108 * it doesn't consume any memory when idle. 2109 * 2110 * TODO: Only step up if the application is actually serving 2111 * the buffer to better manage the socket buffer resources. 2112 */ 2113 if (tcp_do_autorcvbuf && 2114 opti.ts_ecr && 2115 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 2116 if (opti.ts_ecr > tp->rfbuf_ts && 2117 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 2118 if (tp->rfbuf_cnt > 2119 (so->so_rcv.sb_hiwat / 8 * 7) && 2120 so->so_rcv.sb_hiwat < 2121 tcp_autorcvbuf_max) { 2122 newsize = 2123 min(so->so_rcv.sb_hiwat + 2124 tcp_autorcvbuf_inc, 2125 tcp_autorcvbuf_max); 2126 } 2127 /* Start over with next RTT. */ 2128 tp->rfbuf_ts = 0; 2129 tp->rfbuf_cnt = 0; 2130 } else 2131 tp->rfbuf_cnt += tlen; /* add up */ 2132 } 2133 2134 /* 2135 * Drop TCP, IP headers and TCP options then add data 2136 * to socket buffer. 2137 */ 2138 if (so->so_state & SS_CANTRCVMORE) 2139 m_freem(m); 2140 else { 2141 /* 2142 * Set new socket buffer size. 2143 * Give up when limit is reached. 2144 */ 2145 if (newsize) 2146 if (!sbreserve(&so->so_rcv, 2147 newsize, so)) 2148 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 2149 m_adj(m, toff + off); 2150 sbappendstream(&so->so_rcv, m); 2151 } 2152 sorwakeup(so); 2153 tcp_setup_ack(tp, th); 2154 if (tp->t_flags & TF_ACKNOW) { 2155 KERNEL_LOCK(1, NULL); 2156 (void) tcp_output(tp); 2157 KERNEL_UNLOCK_ONE(NULL); 2158 } 2159 if (tcp_saveti) 2160 m_freem(tcp_saveti); 2161 return; 2162 } 2163 } 2164 2165 /* 2166 * Compute mbuf offset to TCP data segment. 2167 */ 2168 hdroptlen = toff + off; 2169 2170 /* 2171 * Calculate amount of space in receive window, 2172 * and then do TCP input processing. 2173 * Receive window is amount of space in rcv queue, 2174 * but not less than advertised window. 2175 */ 2176 { int win; 2177 2178 win = sbspace(&so->so_rcv); 2179 if (win < 0) 2180 win = 0; 2181 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 2182 } 2183 2184 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 2185 tp->rfbuf_ts = 0; 2186 tp->rfbuf_cnt = 0; 2187 2188 switch (tp->t_state) { 2189 /* 2190 * If the state is SYN_SENT: 2191 * if seg contains an ACK, but not for our SYN, drop the input. 2192 * if seg contains a RST, then drop the connection. 2193 * if seg does not contain SYN, then drop it. 2194 * Otherwise this is an acceptable SYN segment 2195 * initialize tp->rcv_nxt and tp->irs 2196 * if seg contains ack then advance tp->snd_una 2197 * if seg contains a ECE and ECN support is enabled, the stream 2198 * is ECN capable. 2199 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2200 * arrange for segment to be acked (eventually) 2201 * continue processing rest of data/controls, beginning with URG 2202 */ 2203 case TCPS_SYN_SENT: 2204 if ((tiflags & TH_ACK) && 2205 (SEQ_LEQ(th->th_ack, tp->iss) || 2206 SEQ_GT(th->th_ack, tp->snd_max))) 2207 goto dropwithreset; 2208 if (tiflags & TH_RST) { 2209 if (tiflags & TH_ACK) 2210 tp = tcp_drop(tp, ECONNREFUSED); 2211 goto drop; 2212 } 2213 if ((tiflags & TH_SYN) == 0) 2214 goto drop; 2215 if (tiflags & TH_ACK) { 2216 tp->snd_una = th->th_ack; 2217 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2218 tp->snd_nxt = tp->snd_una; 2219 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2220 tp->snd_high = tp->snd_una; 2221 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2222 2223 if ((tiflags & TH_ECE) && tcp_do_ecn) { 2224 tp->t_flags |= TF_ECN_PERMIT; 2225 TCP_STATINC(TCP_STAT_ECN_SHS); 2226 } 2227 2228 } 2229 tp->irs = th->th_seq; 2230 tcp_rcvseqinit(tp); 2231 tp->t_flags |= TF_ACKNOW; 2232 tcp_mss_from_peer(tp, opti.maxseg); 2233 2234 /* 2235 * Initialize the initial congestion window. If we 2236 * had to retransmit the SYN, we must initialize cwnd 2237 * to 1 segment (i.e. the Loss Window). 2238 */ 2239 if (tp->t_flags & TF_SYN_REXMT) 2240 tp->snd_cwnd = tp->t_peermss; 2241 else { 2242 int ss = tcp_init_win; 2243 #ifdef INET 2244 if (inp != NULL && in_localaddr(inp->inp_faddr)) 2245 ss = tcp_init_win_local; 2246 #endif 2247 #ifdef INET6 2248 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 2249 ss = tcp_init_win_local; 2250 #endif 2251 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 2252 } 2253 2254 tcp_rmx_rtt(tp); 2255 if (tiflags & TH_ACK) { 2256 TCP_STATINC(TCP_STAT_CONNECTS); 2257 /* 2258 * move tcp_established before soisconnected 2259 * because upcall handler can drive tcp_output 2260 * functionality. 2261 * XXX we might call soisconnected at the end of 2262 * all processing 2263 */ 2264 tcp_established(tp); 2265 soisconnected(so); 2266 /* Do window scaling on this connection? */ 2267 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2268 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2269 tp->snd_scale = tp->requested_s_scale; 2270 tp->rcv_scale = tp->request_r_scale; 2271 } 2272 TCP_REASS_LOCK(tp); 2273 (void) tcp_reass(tp, NULL, NULL, &tlen); 2274 /* 2275 * if we didn't have to retransmit the SYN, 2276 * use its rtt as our initial srtt & rtt var. 2277 */ 2278 if (tp->t_rtttime) 2279 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2280 } else 2281 tp->t_state = TCPS_SYN_RECEIVED; 2282 2283 /* 2284 * Advance th->th_seq to correspond to first data byte. 2285 * If data, trim to stay within window, 2286 * dropping FIN if necessary. 2287 */ 2288 th->th_seq++; 2289 if (tlen > tp->rcv_wnd) { 2290 todrop = tlen - tp->rcv_wnd; 2291 m_adj(m, -todrop); 2292 tlen = tp->rcv_wnd; 2293 tiflags &= ~TH_FIN; 2294 tcps = TCP_STAT_GETREF(); 2295 tcps[TCP_STAT_RCVPACKAFTERWIN]++; 2296 tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop; 2297 TCP_STAT_PUTREF(); 2298 } 2299 tp->snd_wl1 = th->th_seq - 1; 2300 tp->rcv_up = th->th_seq; 2301 goto step6; 2302 2303 /* 2304 * If the state is SYN_RECEIVED: 2305 * If seg contains an ACK, but not for our SYN, drop the input 2306 * and generate an RST. See page 36, rfc793 2307 */ 2308 case TCPS_SYN_RECEIVED: 2309 if ((tiflags & TH_ACK) && 2310 (SEQ_LEQ(th->th_ack, tp->iss) || 2311 SEQ_GT(th->th_ack, tp->snd_max))) 2312 goto dropwithreset; 2313 break; 2314 } 2315 2316 /* 2317 * States other than LISTEN or SYN_SENT. 2318 * First check timestamp, if present. 2319 * Then check that at least some bytes of segment are within 2320 * receive window. If segment begins before rcv_nxt, 2321 * drop leading data (and SYN); if nothing left, just ack. 2322 * 2323 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2324 * and it's less than ts_recent, drop it. 2325 */ 2326 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2327 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2328 2329 /* Check to see if ts_recent is over 24 days old. */ 2330 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2331 /* 2332 * Invalidate ts_recent. If this segment updates 2333 * ts_recent, the age will be reset later and ts_recent 2334 * will get a valid value. If it does not, setting 2335 * ts_recent to zero will at least satisfy the 2336 * requirement that zero be placed in the timestamp 2337 * echo reply when ts_recent isn't valid. The 2338 * age isn't reset until we get a valid ts_recent 2339 * because we don't want out-of-order segments to be 2340 * dropped when ts_recent is old. 2341 */ 2342 tp->ts_recent = 0; 2343 } else { 2344 tcps = TCP_STAT_GETREF(); 2345 tcps[TCP_STAT_RCVDUPPACK]++; 2346 tcps[TCP_STAT_RCVDUPBYTE] += tlen; 2347 tcps[TCP_STAT_PAWSDROP]++; 2348 TCP_STAT_PUTREF(); 2349 tcp_new_dsack(tp, th->th_seq, tlen); 2350 goto dropafterack; 2351 } 2352 } 2353 2354 todrop = tp->rcv_nxt - th->th_seq; 2355 dupseg = false; 2356 if (todrop > 0) { 2357 if (tiflags & TH_SYN) { 2358 tiflags &= ~TH_SYN; 2359 th->th_seq++; 2360 if (th->th_urp > 1) 2361 th->th_urp--; 2362 else { 2363 tiflags &= ~TH_URG; 2364 th->th_urp = 0; 2365 } 2366 todrop--; 2367 } 2368 if (todrop > tlen || 2369 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2370 /* 2371 * Any valid FIN or RST must be to the left of the 2372 * window. At this point the FIN or RST must be a 2373 * duplicate or out of sequence; drop it. 2374 */ 2375 if (tiflags & TH_RST) 2376 goto drop; 2377 tiflags &= ~(TH_FIN|TH_RST); 2378 /* 2379 * Send an ACK to resynchronize and drop any data. 2380 * But keep on processing for RST or ACK. 2381 */ 2382 tp->t_flags |= TF_ACKNOW; 2383 todrop = tlen; 2384 dupseg = true; 2385 tcps = TCP_STAT_GETREF(); 2386 tcps[TCP_STAT_RCVDUPPACK]++; 2387 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 2388 TCP_STAT_PUTREF(); 2389 } else if ((tiflags & TH_RST) && 2390 th->th_seq != tp->rcv_nxt) { 2391 /* 2392 * Test for reset before adjusting the sequence 2393 * number for overlapping data. 2394 */ 2395 goto dropafterack_ratelim; 2396 } else { 2397 tcps = TCP_STAT_GETREF(); 2398 tcps[TCP_STAT_RCVPARTDUPPACK]++; 2399 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 2400 TCP_STAT_PUTREF(); 2401 } 2402 tcp_new_dsack(tp, th->th_seq, todrop); 2403 hdroptlen += todrop; /*drop from head afterwards*/ 2404 th->th_seq += todrop; 2405 tlen -= todrop; 2406 if (th->th_urp > todrop) 2407 th->th_urp -= todrop; 2408 else { 2409 tiflags &= ~TH_URG; 2410 th->th_urp = 0; 2411 } 2412 } 2413 2414 /* 2415 * If new data are received on a connection after the 2416 * user processes are gone, then RST the other end. 2417 */ 2418 if ((so->so_state & SS_NOFDREF) && 2419 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2420 tp = tcp_close(tp); 2421 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 2422 goto dropwithreset; 2423 } 2424 2425 /* 2426 * If segment ends after window, drop trailing data 2427 * (and PUSH and FIN); if nothing left, just ACK. 2428 */ 2429 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 2430 if (todrop > 0) { 2431 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 2432 if (todrop >= tlen) { 2433 /* 2434 * The segment actually starts after the window. 2435 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2436 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2437 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2438 */ 2439 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 2440 /* 2441 * If a new connection request is received 2442 * while in TIME_WAIT, drop the old connection 2443 * and start over if the sequence numbers 2444 * are above the previous ones. 2445 * 2446 * NOTE: We will checksum the packet again, and 2447 * so we need to put the header fields back into 2448 * network order! 2449 * XXX This kind of sucks, but we don't expect 2450 * XXX this to happen very often, so maybe it 2451 * XXX doesn't matter so much. 2452 */ 2453 if (tiflags & TH_SYN && 2454 tp->t_state == TCPS_TIME_WAIT && 2455 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2456 tp = tcp_close(tp); 2457 tcp_fields_to_net(th); 2458 goto findpcb; 2459 } 2460 /* 2461 * If window is closed can only take segments at 2462 * window edge, and have to drop data and PUSH from 2463 * incoming segments. Continue processing, but 2464 * remember to ack. Otherwise, drop segment 2465 * and (if not RST) ack. 2466 */ 2467 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2468 tp->t_flags |= TF_ACKNOW; 2469 TCP_STATINC(TCP_STAT_RCVWINPROBE); 2470 } else 2471 goto dropafterack; 2472 } else 2473 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 2474 m_adj(m, -todrop); 2475 tlen -= todrop; 2476 tiflags &= ~(TH_PUSH|TH_FIN); 2477 } 2478 2479 /* 2480 * If last ACK falls within this segment's sequence numbers, 2481 * record the timestamp. 2482 * NOTE: 2483 * 1) That the test incorporates suggestions from the latest 2484 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2485 * 2) That updating only on newer timestamps interferes with 2486 * our earlier PAWS tests, so this check should be solely 2487 * predicated on the sequence space of this segment. 2488 * 3) That we modify the segment boundary check to be 2489 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2490 * instead of RFC1323's 2491 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2492 * This modified check allows us to overcome RFC1323's 2493 * limitations as described in Stevens TCP/IP Illustrated 2494 * Vol. 2 p.869. In such cases, we can still calculate the 2495 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2496 */ 2497 if (opti.ts_present && 2498 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2499 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2500 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2501 tp->ts_recent_age = tcp_now; 2502 tp->ts_recent = opti.ts_val; 2503 } 2504 2505 /* 2506 * If the RST bit is set examine the state: 2507 * SYN_RECEIVED STATE: 2508 * If passive open, return to LISTEN state. 2509 * If active open, inform user that connection was refused. 2510 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 2511 * Inform user that connection was reset, and close tcb. 2512 * CLOSING, LAST_ACK, TIME_WAIT STATES 2513 * Close the tcb. 2514 */ 2515 if (tiflags & TH_RST) { 2516 if (th->th_seq != tp->rcv_nxt) 2517 goto dropafterack_ratelim; 2518 2519 switch (tp->t_state) { 2520 case TCPS_SYN_RECEIVED: 2521 so->so_error = ECONNREFUSED; 2522 goto close; 2523 2524 case TCPS_ESTABLISHED: 2525 case TCPS_FIN_WAIT_1: 2526 case TCPS_FIN_WAIT_2: 2527 case TCPS_CLOSE_WAIT: 2528 so->so_error = ECONNRESET; 2529 close: 2530 tp->t_state = TCPS_CLOSED; 2531 TCP_STATINC(TCP_STAT_DROPS); 2532 tp = tcp_close(tp); 2533 goto drop; 2534 2535 case TCPS_CLOSING: 2536 case TCPS_LAST_ACK: 2537 case TCPS_TIME_WAIT: 2538 tp = tcp_close(tp); 2539 goto drop; 2540 } 2541 } 2542 2543 /* 2544 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2545 * we must be in a synchronized state. RFC791 states (under RST 2546 * generation) that any unacceptable segment (an out-of-order SYN 2547 * qualifies) received in a synchronized state must elicit only an 2548 * empty acknowledgment segment ... and the connection remains in 2549 * the same state. 2550 */ 2551 if (tiflags & TH_SYN) { 2552 if (tp->rcv_nxt == th->th_seq) { 2553 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2554 TH_ACK); 2555 if (tcp_saveti) 2556 m_freem(tcp_saveti); 2557 return; 2558 } 2559 2560 goto dropafterack_ratelim; 2561 } 2562 2563 /* 2564 * If the ACK bit is off we drop the segment and return. 2565 */ 2566 if ((tiflags & TH_ACK) == 0) { 2567 if (tp->t_flags & TF_ACKNOW) 2568 goto dropafterack; 2569 else 2570 goto drop; 2571 } 2572 2573 /* 2574 * Ack processing. 2575 */ 2576 switch (tp->t_state) { 2577 2578 /* 2579 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2580 * ESTABLISHED state and continue processing, otherwise 2581 * send an RST. 2582 */ 2583 case TCPS_SYN_RECEIVED: 2584 if (SEQ_GT(tp->snd_una, th->th_ack) || 2585 SEQ_GT(th->th_ack, tp->snd_max)) 2586 goto dropwithreset; 2587 TCP_STATINC(TCP_STAT_CONNECTS); 2588 soisconnected(so); 2589 tcp_established(tp); 2590 /* Do window scaling? */ 2591 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2592 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2593 tp->snd_scale = tp->requested_s_scale; 2594 tp->rcv_scale = tp->request_r_scale; 2595 } 2596 TCP_REASS_LOCK(tp); 2597 (void) tcp_reass(tp, NULL, NULL, &tlen); 2598 tp->snd_wl1 = th->th_seq - 1; 2599 /* fall into ... */ 2600 2601 /* 2602 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2603 * ACKs. If the ack is in the range 2604 * tp->snd_una < th->th_ack <= tp->snd_max 2605 * then advance tp->snd_una to th->th_ack and drop 2606 * data from the retransmission queue. If this ACK reflects 2607 * more up to date window information we update our window information. 2608 */ 2609 case TCPS_ESTABLISHED: 2610 case TCPS_FIN_WAIT_1: 2611 case TCPS_FIN_WAIT_2: 2612 case TCPS_CLOSE_WAIT: 2613 case TCPS_CLOSING: 2614 case TCPS_LAST_ACK: 2615 case TCPS_TIME_WAIT: 2616 2617 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2618 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2619 TCP_STATINC(TCP_STAT_RCVDUPACK); 2620 /* 2621 * If we have outstanding data (other than 2622 * a window probe), this is a completely 2623 * duplicate ack (ie, window info didn't 2624 * change), the ack is the biggest we've 2625 * seen and we've seen exactly our rexmt 2626 * threshhold of them, assume a packet 2627 * has been dropped and retransmit it. 2628 * Kludge snd_nxt & the congestion 2629 * window so we send only this one 2630 * packet. 2631 */ 2632 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2633 th->th_ack != tp->snd_una) 2634 tp->t_dupacks = 0; 2635 else if (tp->t_partialacks < 0 && 2636 (++tp->t_dupacks == tcprexmtthresh || 2637 TCP_FACK_FASTRECOV(tp))) { 2638 /* 2639 * Do the fast retransmit, and adjust 2640 * congestion control paramenters. 2641 */ 2642 if (tp->t_congctl->fast_retransmit(tp, th)) { 2643 /* False fast retransmit */ 2644 break; 2645 } else 2646 goto drop; 2647 } else if (tp->t_dupacks > tcprexmtthresh) { 2648 tp->snd_cwnd += tp->t_segsz; 2649 KERNEL_LOCK(1, NULL); 2650 (void) tcp_output(tp); 2651 KERNEL_UNLOCK_ONE(NULL); 2652 goto drop; 2653 } 2654 } else { 2655 /* 2656 * If the ack appears to be very old, only 2657 * allow data that is in-sequence. This 2658 * makes it somewhat more difficult to insert 2659 * forged data by guessing sequence numbers. 2660 * Sent an ack to try to update the send 2661 * sequence number on the other side. 2662 */ 2663 if (tlen && th->th_seq != tp->rcv_nxt && 2664 SEQ_LT(th->th_ack, 2665 tp->snd_una - tp->max_sndwnd)) 2666 goto dropafterack; 2667 } 2668 break; 2669 } 2670 /* 2671 * If the congestion window was inflated to account 2672 * for the other side's cached packets, retract it. 2673 */ 2674 tp->t_congctl->fast_retransmit_newack(tp, th); 2675 2676 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2677 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2678 goto dropafterack; 2679 } 2680 acked = th->th_ack - tp->snd_una; 2681 tcps = TCP_STAT_GETREF(); 2682 tcps[TCP_STAT_RCVACKPACK]++; 2683 tcps[TCP_STAT_RCVACKBYTE] += acked; 2684 TCP_STAT_PUTREF(); 2685 2686 /* 2687 * If we have a timestamp reply, update smoothed 2688 * round trip time. If no timestamp is present but 2689 * transmit timer is running and timed sequence 2690 * number was acked, update smoothed round trip time. 2691 * Since we now have an rtt measurement, cancel the 2692 * timer backoff (cf., Phil Karn's retransmit alg.). 2693 * Recompute the initial retransmit timer. 2694 */ 2695 if (ts_rtt) 2696 tcp_xmit_timer(tp, ts_rtt - 1); 2697 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2698 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2699 2700 /* 2701 * If all outstanding data is acked, stop retransmit 2702 * timer and remember to restart (more output or persist). 2703 * If there is more data to be acked, restart retransmit 2704 * timer, using current (possibly backed-off) value. 2705 */ 2706 if (th->th_ack == tp->snd_max) { 2707 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2708 needoutput = 1; 2709 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2710 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2711 2712 /* 2713 * New data has been acked, adjust the congestion window. 2714 */ 2715 tp->t_congctl->newack(tp, th); 2716 2717 nd6_hint(tp); 2718 if (acked > so->so_snd.sb_cc) { 2719 tp->snd_wnd -= so->so_snd.sb_cc; 2720 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2721 ourfinisacked = 1; 2722 } else { 2723 if (acked > (tp->t_lastoff - tp->t_inoff)) 2724 tp->t_lastm = NULL; 2725 sbdrop(&so->so_snd, acked); 2726 tp->t_lastoff -= acked; 2727 if (tp->snd_wnd > acked) 2728 tp->snd_wnd -= acked; 2729 else 2730 tp->snd_wnd = 0; 2731 ourfinisacked = 0; 2732 } 2733 sowwakeup(so); 2734 2735 icmp_check(tp, th, acked); 2736 2737 tp->snd_una = th->th_ack; 2738 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2739 tp->snd_fack = tp->snd_una; 2740 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2741 tp->snd_nxt = tp->snd_una; 2742 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2743 tp->snd_high = tp->snd_una; 2744 2745 switch (tp->t_state) { 2746 2747 /* 2748 * In FIN_WAIT_1 STATE in addition to the processing 2749 * for the ESTABLISHED state if our FIN is now acknowledged 2750 * then enter FIN_WAIT_2. 2751 */ 2752 case TCPS_FIN_WAIT_1: 2753 if (ourfinisacked) { 2754 /* 2755 * If we can't receive any more 2756 * data, then closing user can proceed. 2757 * Starting the timer is contrary to the 2758 * specification, but if we don't get a FIN 2759 * we'll hang forever. 2760 */ 2761 if (so->so_state & SS_CANTRCVMORE) { 2762 soisdisconnected(so); 2763 if (tp->t_maxidle > 0) 2764 TCP_TIMER_ARM(tp, TCPT_2MSL, 2765 tp->t_maxidle); 2766 } 2767 tp->t_state = TCPS_FIN_WAIT_2; 2768 } 2769 break; 2770 2771 /* 2772 * In CLOSING STATE in addition to the processing for 2773 * the ESTABLISHED state if the ACK acknowledges our FIN 2774 * then enter the TIME-WAIT state, otherwise ignore 2775 * the segment. 2776 */ 2777 case TCPS_CLOSING: 2778 if (ourfinisacked) { 2779 tp->t_state = TCPS_TIME_WAIT; 2780 tcp_canceltimers(tp); 2781 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2782 soisdisconnected(so); 2783 } 2784 break; 2785 2786 /* 2787 * In LAST_ACK, we may still be waiting for data to drain 2788 * and/or to be acked, as well as for the ack of our FIN. 2789 * If our FIN is now acknowledged, delete the TCB, 2790 * enter the closed state and return. 2791 */ 2792 case TCPS_LAST_ACK: 2793 if (ourfinisacked) { 2794 tp = tcp_close(tp); 2795 goto drop; 2796 } 2797 break; 2798 2799 /* 2800 * In TIME_WAIT state the only thing that should arrive 2801 * is a retransmission of the remote FIN. Acknowledge 2802 * it and restart the finack timer. 2803 */ 2804 case TCPS_TIME_WAIT: 2805 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2806 goto dropafterack; 2807 } 2808 } 2809 2810 step6: 2811 /* 2812 * Update window information. 2813 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2814 */ 2815 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2816 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2817 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2818 /* keep track of pure window updates */ 2819 if (tlen == 0 && 2820 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2821 TCP_STATINC(TCP_STAT_RCVWINUPD); 2822 tp->snd_wnd = tiwin; 2823 tp->snd_wl1 = th->th_seq; 2824 tp->snd_wl2 = th->th_ack; 2825 if (tp->snd_wnd > tp->max_sndwnd) 2826 tp->max_sndwnd = tp->snd_wnd; 2827 needoutput = 1; 2828 } 2829 2830 /* 2831 * Process segments with URG. 2832 */ 2833 if ((tiflags & TH_URG) && th->th_urp && 2834 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2835 /* 2836 * This is a kludge, but if we receive and accept 2837 * random urgent pointers, we'll crash in 2838 * soreceive. It's hard to imagine someone 2839 * actually wanting to send this much urgent data. 2840 */ 2841 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2842 th->th_urp = 0; /* XXX */ 2843 tiflags &= ~TH_URG; /* XXX */ 2844 goto dodata; /* XXX */ 2845 } 2846 /* 2847 * If this segment advances the known urgent pointer, 2848 * then mark the data stream. This should not happen 2849 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2850 * a FIN has been received from the remote side. 2851 * In these states we ignore the URG. 2852 * 2853 * According to RFC961 (Assigned Protocols), 2854 * the urgent pointer points to the last octet 2855 * of urgent data. We continue, however, 2856 * to consider it to indicate the first octet 2857 * of data past the urgent section as the original 2858 * spec states (in one of two places). 2859 */ 2860 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2861 tp->rcv_up = th->th_seq + th->th_urp; 2862 so->so_oobmark = so->so_rcv.sb_cc + 2863 (tp->rcv_up - tp->rcv_nxt) - 1; 2864 if (so->so_oobmark == 0) 2865 so->so_state |= SS_RCVATMARK; 2866 sohasoutofband(so); 2867 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2868 } 2869 /* 2870 * Remove out of band data so doesn't get presented to user. 2871 * This can happen independent of advancing the URG pointer, 2872 * but if two URG's are pending at once, some out-of-band 2873 * data may creep in... ick. 2874 */ 2875 if (th->th_urp <= (u_int16_t) tlen 2876 #ifdef SO_OOBINLINE 2877 && (so->so_options & SO_OOBINLINE) == 0 2878 #endif 2879 ) 2880 tcp_pulloutofband(so, th, m, hdroptlen); 2881 } else 2882 /* 2883 * If no out of band data is expected, 2884 * pull receive urgent pointer along 2885 * with the receive window. 2886 */ 2887 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2888 tp->rcv_up = tp->rcv_nxt; 2889 dodata: /* XXX */ 2890 2891 /* 2892 * Process the segment text, merging it into the TCP sequencing queue, 2893 * and arranging for acknowledgement of receipt if necessary. 2894 * This process logically involves adjusting tp->rcv_wnd as data 2895 * is presented to the user (this happens in tcp_usrreq.c, 2896 * tcp_rcvd()). If a FIN has already been received on this 2897 * connection then we just ignore the text. 2898 */ 2899 if ((tlen || (tiflags & TH_FIN)) && 2900 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2901 /* 2902 * Insert segment ti into reassembly queue of tcp with 2903 * control block tp. Return TH_FIN if reassembly now includes 2904 * a segment with FIN. The macro form does the common case 2905 * inline (segment is the next to be received on an 2906 * established connection, and the queue is empty), 2907 * avoiding linkage into and removal from the queue and 2908 * repetition of various conversions. 2909 * Set DELACK for segments received in order, but ack 2910 * immediately when segments are out of order 2911 * (so fast retransmit can work). 2912 */ 2913 /* NOTE: this was TCP_REASS() macro, but used only once */ 2914 TCP_REASS_LOCK(tp); 2915 if (th->th_seq == tp->rcv_nxt && 2916 TAILQ_FIRST(&tp->segq) == NULL && 2917 tp->t_state == TCPS_ESTABLISHED) { 2918 tcp_setup_ack(tp, th); 2919 tp->rcv_nxt += tlen; 2920 tiflags = th->th_flags & TH_FIN; 2921 tcps = TCP_STAT_GETREF(); 2922 tcps[TCP_STAT_RCVPACK]++; 2923 tcps[TCP_STAT_RCVBYTE] += tlen; 2924 TCP_STAT_PUTREF(); 2925 nd6_hint(tp); 2926 if (so->so_state & SS_CANTRCVMORE) 2927 m_freem(m); 2928 else { 2929 m_adj(m, hdroptlen); 2930 sbappendstream(&(so)->so_rcv, m); 2931 } 2932 TCP_REASS_UNLOCK(tp); 2933 sorwakeup(so); 2934 } else { 2935 m_adj(m, hdroptlen); 2936 tiflags = tcp_reass(tp, th, m, &tlen); 2937 tp->t_flags |= TF_ACKNOW; 2938 } 2939 2940 /* 2941 * Note the amount of data that peer has sent into 2942 * our window, in order to estimate the sender's 2943 * buffer size. 2944 */ 2945 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2946 } else { 2947 m_freem(m); 2948 m = NULL; 2949 tiflags &= ~TH_FIN; 2950 } 2951 2952 /* 2953 * If FIN is received ACK the FIN and let the user know 2954 * that the connection is closing. Ignore a FIN received before 2955 * the connection is fully established. 2956 */ 2957 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2958 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2959 socantrcvmore(so); 2960 tp->t_flags |= TF_ACKNOW; 2961 tp->rcv_nxt++; 2962 } 2963 switch (tp->t_state) { 2964 2965 /* 2966 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2967 */ 2968 case TCPS_ESTABLISHED: 2969 tp->t_state = TCPS_CLOSE_WAIT; 2970 break; 2971 2972 /* 2973 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2974 * enter the CLOSING state. 2975 */ 2976 case TCPS_FIN_WAIT_1: 2977 tp->t_state = TCPS_CLOSING; 2978 break; 2979 2980 /* 2981 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2982 * starting the time-wait timer, turning off the other 2983 * standard timers. 2984 */ 2985 case TCPS_FIN_WAIT_2: 2986 tp->t_state = TCPS_TIME_WAIT; 2987 tcp_canceltimers(tp); 2988 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2989 soisdisconnected(so); 2990 break; 2991 2992 /* 2993 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2994 */ 2995 case TCPS_TIME_WAIT: 2996 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2997 break; 2998 } 2999 } 3000 #ifdef TCP_DEBUG 3001 if (so->so_options & SO_DEBUG) 3002 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 3003 #endif 3004 3005 /* 3006 * Return any desired output. 3007 */ 3008 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 3009 KERNEL_LOCK(1, NULL); 3010 (void) tcp_output(tp); 3011 KERNEL_UNLOCK_ONE(NULL); 3012 } 3013 if (tcp_saveti) 3014 m_freem(tcp_saveti); 3015 3016 if (tp->t_state == TCPS_TIME_WAIT 3017 && (so->so_state & SS_NOFDREF) 3018 && (tp->t_inpcb || af != AF_INET) 3019 && (tp->t_in6pcb || af != AF_INET6) 3020 && ((af == AF_INET ? tcp4_vtw_enable : tcp6_vtw_enable) & 1) != 0 3021 && TAILQ_EMPTY(&tp->segq) 3022 && vtw_add(af, tp)) { 3023 ; 3024 } 3025 return; 3026 3027 badsyn: 3028 /* 3029 * Received a bad SYN. Increment counters and dropwithreset. 3030 */ 3031 TCP_STATINC(TCP_STAT_BADSYN); 3032 tp = NULL; 3033 goto dropwithreset; 3034 3035 dropafterack: 3036 /* 3037 * Generate an ACK dropping incoming segment if it occupies 3038 * sequence space, where the ACK reflects our state. 3039 */ 3040 if (tiflags & TH_RST) 3041 goto drop; 3042 goto dropafterack2; 3043 3044 dropafterack_ratelim: 3045 /* 3046 * We may want to rate-limit ACKs against SYN/RST attack. 3047 */ 3048 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 3049 tcp_ackdrop_ppslim) == 0) { 3050 /* XXX stat */ 3051 goto drop; 3052 } 3053 /* ...fall into dropafterack2... */ 3054 3055 dropafterack2: 3056 m_freem(m); 3057 tp->t_flags |= TF_ACKNOW; 3058 KERNEL_LOCK(1, NULL); 3059 (void) tcp_output(tp); 3060 KERNEL_UNLOCK_ONE(NULL); 3061 if (tcp_saveti) 3062 m_freem(tcp_saveti); 3063 return; 3064 3065 dropwithreset_ratelim: 3066 /* 3067 * We may want to rate-limit RSTs in certain situations, 3068 * particularly if we are sending an RST in response to 3069 * an attempt to connect to or otherwise communicate with 3070 * a port for which we have no socket. 3071 */ 3072 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 3073 tcp_rst_ppslim) == 0) { 3074 /* XXX stat */ 3075 goto drop; 3076 } 3077 /* ...fall into dropwithreset... */ 3078 3079 dropwithreset: 3080 /* 3081 * Generate a RST, dropping incoming segment. 3082 * Make ACK acceptable to originator of segment. 3083 */ 3084 if (tiflags & TH_RST) 3085 goto drop; 3086 3087 switch (af) { 3088 #ifdef INET6 3089 case AF_INET6: 3090 /* For following calls to tcp_respond */ 3091 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 3092 goto drop; 3093 break; 3094 #endif /* INET6 */ 3095 case AF_INET: 3096 if (IN_MULTICAST(ip->ip_dst.s_addr) || 3097 in_broadcast(ip->ip_dst, m_get_rcvif_NOMPSAFE(m))) 3098 goto drop; 3099 } 3100 3101 if (tiflags & TH_ACK) 3102 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3103 else { 3104 if (tiflags & TH_SYN) 3105 tlen++; 3106 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 3107 TH_RST|TH_ACK); 3108 } 3109 if (tcp_saveti) 3110 m_freem(tcp_saveti); 3111 return; 3112 3113 badcsum: 3114 drop: 3115 /* 3116 * Drop space held by incoming segment and return. 3117 */ 3118 if (tp) { 3119 if (tp->t_inpcb) 3120 so = tp->t_inpcb->inp_socket; 3121 #ifdef INET6 3122 else if (tp->t_in6pcb) 3123 so = tp->t_in6pcb->in6p_socket; 3124 #endif 3125 else 3126 so = NULL; 3127 #ifdef TCP_DEBUG 3128 if (so && (so->so_options & SO_DEBUG) != 0) 3129 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 3130 #endif 3131 } 3132 if (tcp_saveti) 3133 m_freem(tcp_saveti); 3134 m_freem(m); 3135 return; 3136 } 3137 3138 #ifdef TCP_SIGNATURE 3139 int 3140 tcp_signature_apply(void *fstate, void *data, u_int len) 3141 { 3142 3143 MD5Update(fstate, (u_char *)data, len); 3144 return (0); 3145 } 3146 3147 struct secasvar * 3148 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 3149 { 3150 struct ip *ip; 3151 struct ip6_hdr *ip6; 3152 3153 ip = mtod(m, struct ip *); 3154 switch (ip->ip_v) { 3155 case 4: 3156 ip = mtod(m, struct ip *); 3157 ip6 = NULL; 3158 break; 3159 case 6: 3160 ip = NULL; 3161 ip6 = mtod(m, struct ip6_hdr *); 3162 break; 3163 default: 3164 return (NULL); 3165 } 3166 3167 #ifdef IPSEC 3168 if (ipsec_used) { 3169 union sockaddr_union dst; 3170 /* Extract the destination from the IP header in the mbuf. */ 3171 memset(&dst, 0, sizeof(union sockaddr_union)); 3172 if (ip != NULL) { 3173 dst.sa.sa_len = sizeof(struct sockaddr_in); 3174 dst.sa.sa_family = AF_INET; 3175 dst.sin.sin_addr = ip->ip_dst; 3176 } else { 3177 dst.sa.sa_len = sizeof(struct sockaddr_in6); 3178 dst.sa.sa_family = AF_INET6; 3179 dst.sin6.sin6_addr = ip6->ip6_dst; 3180 } 3181 3182 /* 3183 * Look up an SADB entry which matches the address of the peer. 3184 */ 3185 return KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI), 0, 0); 3186 } 3187 return NULL; 3188 #else 3189 if (ip) 3190 return key_allocsa(AF_INET, (void *)&ip->ip_src, 3191 (void *)&ip->ip_dst, IPPROTO_TCP, 3192 htonl(TCP_SIG_SPI), 0, 0); 3193 else 3194 return key_allocsa(AF_INET6, (void *)&ip6->ip6_src, 3195 (void *)&ip6->ip6_dst, IPPROTO_TCP, 3196 htonl(TCP_SIG_SPI), 0, 0); 3197 #endif 3198 } 3199 3200 int 3201 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 3202 struct secasvar *sav, char *sig) 3203 { 3204 MD5_CTX ctx; 3205 struct ip *ip; 3206 struct ipovly *ipovly; 3207 #ifdef INET6 3208 struct ip6_hdr *ip6; 3209 struct ip6_hdr_pseudo ip6pseudo; 3210 #endif /* INET6 */ 3211 struct ippseudo ippseudo; 3212 struct tcphdr th0; 3213 int l, tcphdrlen; 3214 3215 if (sav == NULL) 3216 return (-1); 3217 3218 tcphdrlen = th->th_off * 4; 3219 3220 switch (mtod(m, struct ip *)->ip_v) { 3221 case 4: 3222 MD5Init(&ctx); 3223 ip = mtod(m, struct ip *); 3224 memset(&ippseudo, 0, sizeof(ippseudo)); 3225 ipovly = (struct ipovly *)ip; 3226 ippseudo.ippseudo_src = ipovly->ih_src; 3227 ippseudo.ippseudo_dst = ipovly->ih_dst; 3228 ippseudo.ippseudo_pad = 0; 3229 ippseudo.ippseudo_p = IPPROTO_TCP; 3230 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 3231 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 3232 break; 3233 #if INET6 3234 case 6: 3235 MD5Init(&ctx); 3236 ip6 = mtod(m, struct ip6_hdr *); 3237 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 3238 ip6pseudo.ip6ph_src = ip6->ip6_src; 3239 in6_clearscope(&ip6pseudo.ip6ph_src); 3240 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 3241 in6_clearscope(&ip6pseudo.ip6ph_dst); 3242 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 3243 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 3244 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 3245 break; 3246 #endif /* INET6 */ 3247 default: 3248 return (-1); 3249 } 3250 3251 th0 = *th; 3252 th0.th_sum = 0; 3253 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 3254 3255 l = m->m_pkthdr.len - thoff - tcphdrlen; 3256 if (l > 0) 3257 m_apply(m, thoff + tcphdrlen, 3258 m->m_pkthdr.len - thoff - tcphdrlen, 3259 tcp_signature_apply, &ctx); 3260 3261 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 3262 MD5Final(sig, &ctx); 3263 3264 return (0); 3265 } 3266 #endif 3267 3268 /* 3269 * tcp_dooptions: parse and process tcp options. 3270 * 3271 * returns -1 if this segment should be dropped. (eg. wrong signature) 3272 * otherwise returns 0. 3273 */ 3274 3275 static int 3276 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, 3277 struct tcphdr *th, 3278 struct mbuf *m, int toff, struct tcp_opt_info *oi) 3279 { 3280 u_int16_t mss; 3281 int opt, optlen = 0; 3282 #ifdef TCP_SIGNATURE 3283 void *sigp = NULL; 3284 char sigbuf[TCP_SIGLEN]; 3285 struct secasvar *sav = NULL; 3286 #endif 3287 3288 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 3289 opt = cp[0]; 3290 if (opt == TCPOPT_EOL) 3291 break; 3292 if (opt == TCPOPT_NOP) 3293 optlen = 1; 3294 else { 3295 if (cnt < 2) 3296 break; 3297 optlen = cp[1]; 3298 if (optlen < 2 || optlen > cnt) 3299 break; 3300 } 3301 switch (opt) { 3302 3303 default: 3304 continue; 3305 3306 case TCPOPT_MAXSEG: 3307 if (optlen != TCPOLEN_MAXSEG) 3308 continue; 3309 if (!(th->th_flags & TH_SYN)) 3310 continue; 3311 if (TCPS_HAVERCVDSYN(tp->t_state)) 3312 continue; 3313 bcopy(cp + 2, &mss, sizeof(mss)); 3314 oi->maxseg = ntohs(mss); 3315 break; 3316 3317 case TCPOPT_WINDOW: 3318 if (optlen != TCPOLEN_WINDOW) 3319 continue; 3320 if (!(th->th_flags & TH_SYN)) 3321 continue; 3322 if (TCPS_HAVERCVDSYN(tp->t_state)) 3323 continue; 3324 tp->t_flags |= TF_RCVD_SCALE; 3325 tp->requested_s_scale = cp[2]; 3326 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3327 char buf[INET6_ADDRSTRLEN]; 3328 struct ip *ip = mtod(m, struct ip *); 3329 #ifdef INET6 3330 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 3331 #endif 3332 if (ip) 3333 in_print(buf, sizeof(buf), 3334 &ip->ip_src); 3335 #ifdef INET6 3336 else if (ip6) 3337 in6_print(buf, sizeof(buf), 3338 &ip6->ip6_src); 3339 #endif 3340 else 3341 strlcpy(buf, "(unknown)", sizeof(buf)); 3342 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3343 "assuming %d\n", 3344 tp->requested_s_scale, buf, 3345 TCP_MAX_WINSHIFT); 3346 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3347 } 3348 break; 3349 3350 case TCPOPT_TIMESTAMP: 3351 if (optlen != TCPOLEN_TIMESTAMP) 3352 continue; 3353 oi->ts_present = 1; 3354 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 3355 NTOHL(oi->ts_val); 3356 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 3357 NTOHL(oi->ts_ecr); 3358 3359 if (!(th->th_flags & TH_SYN)) 3360 continue; 3361 if (TCPS_HAVERCVDSYN(tp->t_state)) 3362 continue; 3363 /* 3364 * A timestamp received in a SYN makes 3365 * it ok to send timestamp requests and replies. 3366 */ 3367 tp->t_flags |= TF_RCVD_TSTMP; 3368 tp->ts_recent = oi->ts_val; 3369 tp->ts_recent_age = tcp_now; 3370 break; 3371 3372 case TCPOPT_SACK_PERMITTED: 3373 if (optlen != TCPOLEN_SACK_PERMITTED) 3374 continue; 3375 if (!(th->th_flags & TH_SYN)) 3376 continue; 3377 if (TCPS_HAVERCVDSYN(tp->t_state)) 3378 continue; 3379 if (tcp_do_sack) { 3380 tp->t_flags |= TF_SACK_PERMIT; 3381 tp->t_flags |= TF_WILL_SACK; 3382 } 3383 break; 3384 3385 case TCPOPT_SACK: 3386 tcp_sack_option(tp, th, cp, optlen); 3387 break; 3388 #ifdef TCP_SIGNATURE 3389 case TCPOPT_SIGNATURE: 3390 if (optlen != TCPOLEN_SIGNATURE) 3391 continue; 3392 if (sigp && memcmp(sigp, cp + 2, TCP_SIGLEN)) 3393 return (-1); 3394 3395 sigp = sigbuf; 3396 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3397 tp->t_flags |= TF_SIGNATURE; 3398 break; 3399 #endif 3400 } 3401 } 3402 3403 #ifndef TCP_SIGNATURE 3404 return 0; 3405 #else 3406 if (tp->t_flags & TF_SIGNATURE) { 3407 3408 sav = tcp_signature_getsav(m, th); 3409 3410 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3411 return (-1); 3412 } 3413 3414 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) 3415 goto out; 3416 3417 if (sigp) { 3418 char sig[TCP_SIGLEN]; 3419 3420 tcp_fields_to_net(th); 3421 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3422 tcp_fields_to_host(th); 3423 goto out; 3424 } 3425 tcp_fields_to_host(th); 3426 3427 if (memcmp(sig, sigp, TCP_SIGLEN)) { 3428 TCP_STATINC(TCP_STAT_BADSIG); 3429 goto out; 3430 } else 3431 TCP_STATINC(TCP_STAT_GOODSIG); 3432 3433 key_sa_recordxfer(sav, m); 3434 KEY_FREESAV(&sav); 3435 } 3436 return 0; 3437 out: 3438 if (sav != NULL) 3439 KEY_FREESAV(&sav); 3440 return -1; 3441 #endif 3442 } 3443 3444 /* 3445 * Pull out of band byte out of a segment so 3446 * it doesn't appear in the user's data queue. 3447 * It is still reflected in the segment length for 3448 * sequencing purposes. 3449 */ 3450 void 3451 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3452 struct mbuf *m, int off) 3453 { 3454 int cnt = off + th->th_urp - 1; 3455 3456 while (cnt >= 0) { 3457 if (m->m_len > cnt) { 3458 char *cp = mtod(m, char *) + cnt; 3459 struct tcpcb *tp = sototcpcb(so); 3460 3461 tp->t_iobc = *cp; 3462 tp->t_oobflags |= TCPOOB_HAVEDATA; 3463 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3464 m->m_len--; 3465 return; 3466 } 3467 cnt -= m->m_len; 3468 m = m->m_next; 3469 if (m == 0) 3470 break; 3471 } 3472 panic("tcp_pulloutofband"); 3473 } 3474 3475 /* 3476 * Collect new round-trip time estimate 3477 * and update averages and current timeout. 3478 * 3479 * rtt is in units of slow ticks (typically 500 ms) -- essentially the 3480 * difference of two timestamps. 3481 */ 3482 void 3483 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3484 { 3485 int32_t delta; 3486 3487 TCP_STATINC(TCP_STAT_RTTUPDATED); 3488 if (tp->t_srtt != 0) { 3489 /* 3490 * Compute the amount to add to srtt for smoothing, 3491 * *alpha, or 2^(-TCP_RTT_SHIFT). Because 3492 * srtt is stored in 1/32 slow ticks, we conceptually 3493 * shift left 5 bits, subtract srtt to get the 3494 * diference, and then shift right by TCP_RTT_SHIFT 3495 * (3) to obtain 1/8 of the difference. 3496 */ 3497 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3498 /* 3499 * This can never happen, because delta's lowest 3500 * possible value is 1/8 of t_srtt. But if it does, 3501 * set srtt to some reasonable value, here chosen 3502 * as 1/8 tick. 3503 */ 3504 if ((tp->t_srtt += delta) <= 0) 3505 tp->t_srtt = 1 << 2; 3506 /* 3507 * RFC2988 requires that rttvar be updated first. 3508 * This code is compliant because "delta" is the old 3509 * srtt minus the new observation (scaled). 3510 * 3511 * RFC2988 says: 3512 * rttvar = (1-beta) * rttvar + beta * |srtt-observed| 3513 * 3514 * delta is in units of 1/32 ticks, and has then been 3515 * divided by 8. This is equivalent to being in 1/16s 3516 * units and divided by 4. Subtract from it 1/4 of 3517 * the existing rttvar to form the (signed) amount to 3518 * adjust. 3519 */ 3520 if (delta < 0) 3521 delta = -delta; 3522 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3523 /* 3524 * As with srtt, this should never happen. There is 3525 * no support in RFC2988 for this operation. But 1/4s 3526 * as rttvar when faced with something arguably wrong 3527 * is ok. 3528 */ 3529 if ((tp->t_rttvar += delta) <= 0) 3530 tp->t_rttvar = 1 << 2; 3531 3532 /* 3533 * If srtt exceeds .01 second, ensure we use the 'remote' MSL 3534 * Problem is: it doesn't work. Disabled by defaulting 3535 * tcp_rttlocal to 0; see corresponding code in 3536 * tcp_subr that selects local vs remote in a different way. 3537 * 3538 * The static branch prediction hint here should be removed 3539 * when the rtt estimator is fixed and the rtt_enable code 3540 * is turned back on. 3541 */ 3542 if (__predict_false(tcp_rttlocal) && tcp_msl_enable 3543 && tp->t_srtt > tcp_msl_remote_threshold 3544 && tp->t_msl < tcp_msl_remote) { 3545 tp->t_msl = tcp_msl_remote; 3546 } 3547 } else { 3548 /* 3549 * This is the first measurement. Per RFC2988, 2.2, 3550 * set rtt=R and srtt=R/2. 3551 * For srtt, storage representation is 1/32 ticks, 3552 * so shift left by 5. 3553 * For rttvar, storage representation is 1/16 ticks, 3554 * So shift left by 4, but then right by 1 to halve. 3555 */ 3556 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3557 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3558 } 3559 tp->t_rtttime = 0; 3560 tp->t_rxtshift = 0; 3561 3562 /* 3563 * the retransmit should happen at rtt + 4 * rttvar. 3564 * Because of the way we do the smoothing, srtt and rttvar 3565 * will each average +1/2 tick of bias. When we compute 3566 * the retransmit timer, we want 1/2 tick of rounding and 3567 * 1 extra tick because of +-1/2 tick uncertainty in the 3568 * firing of the timer. The bias will give us exactly the 3569 * 1.5 tick we need. But, because the bias is 3570 * statistical, we have to test that we don't drop below 3571 * the minimum feasible timer (which is 2 ticks). 3572 */ 3573 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3574 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3575 3576 /* 3577 * We received an ack for a packet that wasn't retransmitted; 3578 * it is probably safe to discard any error indications we've 3579 * received recently. This isn't quite right, but close enough 3580 * for now (a route might have failed after we sent a segment, 3581 * and the return path might not be symmetrical). 3582 */ 3583 tp->t_softerror = 0; 3584 } 3585 3586 3587 /* 3588 * TCP compressed state engine. Currently used to hold compressed 3589 * state for SYN_RECEIVED. 3590 */ 3591 3592 u_long syn_cache_count; 3593 u_int32_t syn_hash1, syn_hash2; 3594 3595 #define SYN_HASH(sa, sp, dp) \ 3596 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3597 ((u_int32_t)(sp)))^syn_hash2))) 3598 #ifndef INET6 3599 #define SYN_HASHALL(hash, src, dst) \ 3600 do { \ 3601 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3602 ((const struct sockaddr_in *)(src))->sin_port, \ 3603 ((const struct sockaddr_in *)(dst))->sin_port); \ 3604 } while (/*CONSTCOND*/ 0) 3605 #else 3606 #define SYN_HASH6(sa, sp, dp) \ 3607 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3608 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3609 & 0x7fffffff) 3610 3611 #define SYN_HASHALL(hash, src, dst) \ 3612 do { \ 3613 switch ((src)->sa_family) { \ 3614 case AF_INET: \ 3615 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3616 ((const struct sockaddr_in *)(src))->sin_port, \ 3617 ((const struct sockaddr_in *)(dst))->sin_port); \ 3618 break; \ 3619 case AF_INET6: \ 3620 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3621 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3622 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3623 break; \ 3624 default: \ 3625 hash = 0; \ 3626 } \ 3627 } while (/*CONSTCOND*/0) 3628 #endif /* INET6 */ 3629 3630 static struct pool syn_cache_pool; 3631 3632 /* 3633 * We don't estimate RTT with SYNs, so each packet starts with the default 3634 * RTT and each timer step has a fixed timeout value. 3635 */ 3636 #define SYN_CACHE_TIMER_ARM(sc) \ 3637 do { \ 3638 TCPT_RANGESET((sc)->sc_rxtcur, \ 3639 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3640 TCPTV_REXMTMAX); \ 3641 callout_reset(&(sc)->sc_timer, \ 3642 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3643 } while (/*CONSTCOND*/0) 3644 3645 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3646 3647 static inline void 3648 syn_cache_rm(struct syn_cache *sc) 3649 { 3650 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3651 sc, sc_bucketq); 3652 sc->sc_tp = NULL; 3653 LIST_REMOVE(sc, sc_tpq); 3654 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3655 callout_stop(&sc->sc_timer); 3656 syn_cache_count--; 3657 } 3658 3659 static inline void 3660 syn_cache_put(struct syn_cache *sc) 3661 { 3662 if (sc->sc_ipopts) 3663 (void) m_free(sc->sc_ipopts); 3664 rtcache_free(&sc->sc_route); 3665 sc->sc_flags |= SCF_DEAD; 3666 if (!callout_invoking(&sc->sc_timer)) 3667 callout_schedule(&(sc)->sc_timer, 1); 3668 } 3669 3670 void 3671 syn_cache_init(void) 3672 { 3673 int i; 3674 3675 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 3676 "synpl", NULL, IPL_SOFTNET); 3677 3678 /* Initialize the hash buckets. */ 3679 for (i = 0; i < tcp_syn_cache_size; i++) 3680 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3681 } 3682 3683 void 3684 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3685 { 3686 struct syn_cache_head *scp; 3687 struct syn_cache *sc2; 3688 int s; 3689 3690 /* 3691 * If there are no entries in the hash table, reinitialize 3692 * the hash secrets. 3693 */ 3694 if (syn_cache_count == 0) { 3695 syn_hash1 = cprng_fast32(); 3696 syn_hash2 = cprng_fast32(); 3697 } 3698 3699 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3700 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3701 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3702 3703 /* 3704 * Make sure that we don't overflow the per-bucket 3705 * limit or the total cache size limit. 3706 */ 3707 s = splsoftnet(); 3708 if (scp->sch_length >= tcp_syn_bucket_limit) { 3709 TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW); 3710 /* 3711 * The bucket is full. Toss the oldest element in the 3712 * bucket. This will be the first entry in the bucket. 3713 */ 3714 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3715 #ifdef DIAGNOSTIC 3716 /* 3717 * This should never happen; we should always find an 3718 * entry in our bucket. 3719 */ 3720 if (sc2 == NULL) 3721 panic("syn_cache_insert: bucketoverflow: impossible"); 3722 #endif 3723 syn_cache_rm(sc2); 3724 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3725 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3726 struct syn_cache_head *scp2, *sce; 3727 3728 TCP_STATINC(TCP_STAT_SC_OVERFLOWED); 3729 /* 3730 * The cache is full. Toss the oldest entry in the 3731 * first non-empty bucket we can find. 3732 * 3733 * XXX We would really like to toss the oldest 3734 * entry in the cache, but we hope that this 3735 * condition doesn't happen very often. 3736 */ 3737 scp2 = scp; 3738 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3739 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3740 for (++scp2; scp2 != scp; scp2++) { 3741 if (scp2 >= sce) 3742 scp2 = &tcp_syn_cache[0]; 3743 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3744 break; 3745 } 3746 #ifdef DIAGNOSTIC 3747 /* 3748 * This should never happen; we should always find a 3749 * non-empty bucket. 3750 */ 3751 if (scp2 == scp) 3752 panic("syn_cache_insert: cacheoverflow: " 3753 "impossible"); 3754 #endif 3755 } 3756 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3757 syn_cache_rm(sc2); 3758 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3759 } 3760 3761 /* 3762 * Initialize the entry's timer. 3763 */ 3764 sc->sc_rxttot = 0; 3765 sc->sc_rxtshift = 0; 3766 SYN_CACHE_TIMER_ARM(sc); 3767 3768 /* Link it from tcpcb entry */ 3769 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3770 3771 /* Put it into the bucket. */ 3772 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3773 scp->sch_length++; 3774 syn_cache_count++; 3775 3776 TCP_STATINC(TCP_STAT_SC_ADDED); 3777 splx(s); 3778 } 3779 3780 /* 3781 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3782 * If we have retransmitted an entry the maximum number of times, expire 3783 * that entry. 3784 */ 3785 void 3786 syn_cache_timer(void *arg) 3787 { 3788 struct syn_cache *sc = arg; 3789 3790 mutex_enter(softnet_lock); 3791 KERNEL_LOCK(1, NULL); 3792 callout_ack(&sc->sc_timer); 3793 3794 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3795 TCP_STATINC(TCP_STAT_SC_DELAYED_FREE); 3796 callout_destroy(&sc->sc_timer); 3797 pool_put(&syn_cache_pool, sc); 3798 KERNEL_UNLOCK_ONE(NULL); 3799 mutex_exit(softnet_lock); 3800 return; 3801 } 3802 3803 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3804 /* Drop it -- too many retransmissions. */ 3805 goto dropit; 3806 } 3807 3808 /* 3809 * Compute the total amount of time this entry has 3810 * been on a queue. If this entry has been on longer 3811 * than the keep alive timer would allow, expire it. 3812 */ 3813 sc->sc_rxttot += sc->sc_rxtcur; 3814 if (sc->sc_rxttot >= tcp_keepinit) 3815 goto dropit; 3816 3817 TCP_STATINC(TCP_STAT_SC_RETRANSMITTED); 3818 (void) syn_cache_respond(sc, NULL); 3819 3820 /* Advance the timer back-off. */ 3821 sc->sc_rxtshift++; 3822 SYN_CACHE_TIMER_ARM(sc); 3823 3824 KERNEL_UNLOCK_ONE(NULL); 3825 mutex_exit(softnet_lock); 3826 return; 3827 3828 dropit: 3829 TCP_STATINC(TCP_STAT_SC_TIMED_OUT); 3830 syn_cache_rm(sc); 3831 if (sc->sc_ipopts) 3832 (void) m_free(sc->sc_ipopts); 3833 rtcache_free(&sc->sc_route); 3834 callout_destroy(&sc->sc_timer); 3835 pool_put(&syn_cache_pool, sc); 3836 KERNEL_UNLOCK_ONE(NULL); 3837 mutex_exit(softnet_lock); 3838 } 3839 3840 /* 3841 * Remove syn cache created by the specified tcb entry, 3842 * because this does not make sense to keep them 3843 * (if there's no tcb entry, syn cache entry will never be used) 3844 */ 3845 void 3846 syn_cache_cleanup(struct tcpcb *tp) 3847 { 3848 struct syn_cache *sc, *nsc; 3849 int s; 3850 3851 s = splsoftnet(); 3852 3853 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3854 nsc = LIST_NEXT(sc, sc_tpq); 3855 3856 #ifdef DIAGNOSTIC 3857 if (sc->sc_tp != tp) 3858 panic("invalid sc_tp in syn_cache_cleanup"); 3859 #endif 3860 syn_cache_rm(sc); 3861 syn_cache_put(sc); /* calls pool_put but see spl above */ 3862 } 3863 /* just for safety */ 3864 LIST_INIT(&tp->t_sc); 3865 3866 splx(s); 3867 } 3868 3869 /* 3870 * Find an entry in the syn cache. 3871 */ 3872 struct syn_cache * 3873 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3874 struct syn_cache_head **headp) 3875 { 3876 struct syn_cache *sc; 3877 struct syn_cache_head *scp; 3878 u_int32_t hash; 3879 int s; 3880 3881 SYN_HASHALL(hash, src, dst); 3882 3883 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3884 *headp = scp; 3885 s = splsoftnet(); 3886 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3887 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3888 if (sc->sc_hash != hash) 3889 continue; 3890 if (!memcmp(&sc->sc_src, src, src->sa_len) && 3891 !memcmp(&sc->sc_dst, dst, dst->sa_len)) { 3892 splx(s); 3893 return (sc); 3894 } 3895 } 3896 splx(s); 3897 return (NULL); 3898 } 3899 3900 /* 3901 * This function gets called when we receive an ACK for a 3902 * socket in the LISTEN state. We look up the connection 3903 * in the syn cache, and if its there, we pull it out of 3904 * the cache and turn it into a full-blown connection in 3905 * the SYN-RECEIVED state. 3906 * 3907 * The return values may not be immediately obvious, and their effects 3908 * can be subtle, so here they are: 3909 * 3910 * NULL SYN was not found in cache; caller should drop the 3911 * packet and send an RST. 3912 * 3913 * -1 We were unable to create the new connection, and are 3914 * aborting it. An ACK,RST is being sent to the peer 3915 * (unless we got screwey sequence numbners; see below), 3916 * because the 3-way handshake has been completed. Caller 3917 * should not free the mbuf, since we may be using it. If 3918 * we are not, we will free it. 3919 * 3920 * Otherwise, the return value is a pointer to the new socket 3921 * associated with the connection. 3922 */ 3923 struct socket * 3924 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3925 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3926 struct socket *so, struct mbuf *m) 3927 { 3928 struct syn_cache *sc; 3929 struct syn_cache_head *scp; 3930 struct inpcb *inp = NULL; 3931 #ifdef INET6 3932 struct in6pcb *in6p = NULL; 3933 #endif 3934 struct tcpcb *tp = 0; 3935 int s; 3936 struct socket *oso; 3937 3938 s = splsoftnet(); 3939 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3940 splx(s); 3941 return (NULL); 3942 } 3943 3944 /* 3945 * Verify the sequence and ack numbers. Try getting the correct 3946 * response again. 3947 */ 3948 if ((th->th_ack != sc->sc_iss + 1) || 3949 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3950 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3951 (void) syn_cache_respond(sc, m); 3952 splx(s); 3953 return ((struct socket *)(-1)); 3954 } 3955 3956 /* Remove this cache entry */ 3957 syn_cache_rm(sc); 3958 splx(s); 3959 3960 /* 3961 * Ok, create the full blown connection, and set things up 3962 * as they would have been set up if we had created the 3963 * connection when the SYN arrived. If we can't create 3964 * the connection, abort it. 3965 */ 3966 /* 3967 * inp still has the OLD in_pcb stuff, set the 3968 * v6-related flags on the new guy, too. This is 3969 * done particularly for the case where an AF_INET6 3970 * socket is bound only to a port, and a v4 connection 3971 * comes in on that port. 3972 * we also copy the flowinfo from the original pcb 3973 * to the new one. 3974 */ 3975 oso = so; 3976 so = sonewconn(so, true); 3977 if (so == NULL) 3978 goto resetandabort; 3979 3980 switch (so->so_proto->pr_domain->dom_family) { 3981 #ifdef INET 3982 case AF_INET: 3983 inp = sotoinpcb(so); 3984 break; 3985 #endif 3986 #ifdef INET6 3987 case AF_INET6: 3988 in6p = sotoin6pcb(so); 3989 break; 3990 #endif 3991 } 3992 switch (src->sa_family) { 3993 #ifdef INET 3994 case AF_INET: 3995 if (inp) { 3996 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3997 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3998 inp->inp_options = ip_srcroute(); 3999 in_pcbstate(inp, INP_BOUND); 4000 if (inp->inp_options == NULL) { 4001 inp->inp_options = sc->sc_ipopts; 4002 sc->sc_ipopts = NULL; 4003 } 4004 } 4005 #ifdef INET6 4006 else if (in6p) { 4007 /* IPv4 packet to AF_INET6 socket */ 4008 memset(&in6p->in6p_laddr, 0, sizeof(in6p->in6p_laddr)); 4009 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 4010 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 4011 &in6p->in6p_laddr.s6_addr32[3], 4012 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 4013 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 4014 in6totcpcb(in6p)->t_family = AF_INET; 4015 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 4016 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 4017 else 4018 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 4019 in6_pcbstate(in6p, IN6P_BOUND); 4020 } 4021 #endif 4022 break; 4023 #endif 4024 #ifdef INET6 4025 case AF_INET6: 4026 if (in6p) { 4027 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 4028 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 4029 in6_pcbstate(in6p, IN6P_BOUND); 4030 } 4031 break; 4032 #endif 4033 } 4034 #ifdef INET6 4035 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 4036 struct in6pcb *oin6p = sotoin6pcb(oso); 4037 /* inherit socket options from the listening socket */ 4038 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 4039 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 4040 m_freem(in6p->in6p_options); 4041 in6p->in6p_options = 0; 4042 } 4043 ip6_savecontrol(in6p, &in6p->in6p_options, 4044 mtod(m, struct ip6_hdr *), m); 4045 } 4046 #endif 4047 4048 #if defined(IPSEC) 4049 if (ipsec_used) { 4050 /* 4051 * we make a copy of policy, instead of sharing the policy, for 4052 * better behavior in terms of SA lookup and dead SA removal. 4053 */ 4054 if (inp) { 4055 /* copy old policy into new socket's */ 4056 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, 4057 inp->inp_sp)) 4058 printf("tcp_input: could not copy policy\n"); 4059 } 4060 #ifdef INET6 4061 else if (in6p) { 4062 /* copy old policy into new socket's */ 4063 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 4064 in6p->in6p_sp)) 4065 printf("tcp_input: could not copy policy\n"); 4066 } 4067 #endif 4068 } 4069 #endif 4070 4071 /* 4072 * Give the new socket our cached route reference. 4073 */ 4074 if (inp) { 4075 rtcache_copy(&inp->inp_route, &sc->sc_route); 4076 rtcache_free(&sc->sc_route); 4077 } 4078 #ifdef INET6 4079 else { 4080 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 4081 rtcache_free(&sc->sc_route); 4082 } 4083 #endif 4084 4085 if (inp) { 4086 struct sockaddr_in sin; 4087 memcpy(&sin, src, src->sa_len); 4088 if (in_pcbconnect(inp, &sin, &lwp0)) { 4089 goto resetandabort; 4090 } 4091 } 4092 #ifdef INET6 4093 else if (in6p) { 4094 struct sockaddr_in6 sin6; 4095 memcpy(&sin6, src, src->sa_len); 4096 if (src->sa_family == AF_INET) { 4097 /* IPv4 packet to AF_INET6 socket */ 4098 in6_sin_2_v4mapsin6((struct sockaddr_in *)src, &sin6); 4099 } 4100 if (in6_pcbconnect(in6p, &sin6, NULL)) { 4101 goto resetandabort; 4102 } 4103 } 4104 #endif 4105 else { 4106 goto resetandabort; 4107 } 4108 4109 if (inp) 4110 tp = intotcpcb(inp); 4111 #ifdef INET6 4112 else if (in6p) 4113 tp = in6totcpcb(in6p); 4114 #endif 4115 else 4116 tp = NULL; 4117 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 4118 if (sc->sc_request_r_scale != 15) { 4119 tp->requested_s_scale = sc->sc_requested_s_scale; 4120 tp->request_r_scale = sc->sc_request_r_scale; 4121 tp->snd_scale = sc->sc_requested_s_scale; 4122 tp->rcv_scale = sc->sc_request_r_scale; 4123 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 4124 } 4125 if (sc->sc_flags & SCF_TIMESTAMP) 4126 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 4127 tp->ts_timebase = sc->sc_timebase; 4128 4129 tp->t_template = tcp_template(tp); 4130 if (tp->t_template == 0) { 4131 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 4132 so = NULL; 4133 m_freem(m); 4134 goto abort; 4135 } 4136 4137 tp->iss = sc->sc_iss; 4138 tp->irs = sc->sc_irs; 4139 tcp_sendseqinit(tp); 4140 tcp_rcvseqinit(tp); 4141 tp->t_state = TCPS_SYN_RECEIVED; 4142 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 4143 TCP_STATINC(TCP_STAT_ACCEPTS); 4144 4145 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 4146 tp->t_flags |= TF_WILL_SACK; 4147 4148 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 4149 tp->t_flags |= TF_ECN_PERMIT; 4150 4151 #ifdef TCP_SIGNATURE 4152 if (sc->sc_flags & SCF_SIGNATURE) 4153 tp->t_flags |= TF_SIGNATURE; 4154 #endif 4155 4156 /* Initialize tp->t_ourmss before we deal with the peer's! */ 4157 tp->t_ourmss = sc->sc_ourmaxseg; 4158 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 4159 4160 /* 4161 * Initialize the initial congestion window. If we 4162 * had to retransmit the SYN,ACK, we must initialize cwnd 4163 * to 1 segment (i.e. the Loss Window). 4164 */ 4165 if (sc->sc_rxtshift) 4166 tp->snd_cwnd = tp->t_peermss; 4167 else { 4168 int ss = tcp_init_win; 4169 #ifdef INET 4170 if (inp != NULL && in_localaddr(inp->inp_faddr)) 4171 ss = tcp_init_win_local; 4172 #endif 4173 #ifdef INET6 4174 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 4175 ss = tcp_init_win_local; 4176 #endif 4177 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 4178 } 4179 4180 tcp_rmx_rtt(tp); 4181 tp->snd_wl1 = sc->sc_irs; 4182 tp->rcv_up = sc->sc_irs + 1; 4183 4184 /* 4185 * This is what whould have happened in tcp_output() when 4186 * the SYN,ACK was sent. 4187 */ 4188 tp->snd_up = tp->snd_una; 4189 tp->snd_max = tp->snd_nxt = tp->iss+1; 4190 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 4191 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 4192 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 4193 tp->last_ack_sent = tp->rcv_nxt; 4194 tp->t_partialacks = -1; 4195 tp->t_dupacks = 0; 4196 4197 TCP_STATINC(TCP_STAT_SC_COMPLETED); 4198 s = splsoftnet(); 4199 syn_cache_put(sc); 4200 splx(s); 4201 return (so); 4202 4203 resetandabort: 4204 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 4205 abort: 4206 if (so != NULL) { 4207 (void) soqremque(so, 1); 4208 (void) soabort(so); 4209 mutex_enter(softnet_lock); 4210 } 4211 s = splsoftnet(); 4212 syn_cache_put(sc); 4213 splx(s); 4214 TCP_STATINC(TCP_STAT_SC_ABORTED); 4215 return ((struct socket *)(-1)); 4216 } 4217 4218 /* 4219 * This function is called when we get a RST for a 4220 * non-existent connection, so that we can see if the 4221 * connection is in the syn cache. If it is, zap it. 4222 */ 4223 4224 void 4225 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 4226 { 4227 struct syn_cache *sc; 4228 struct syn_cache_head *scp; 4229 int s = splsoftnet(); 4230 4231 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4232 splx(s); 4233 return; 4234 } 4235 if (SEQ_LT(th->th_seq, sc->sc_irs) || 4236 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 4237 splx(s); 4238 return; 4239 } 4240 syn_cache_rm(sc); 4241 TCP_STATINC(TCP_STAT_SC_RESET); 4242 syn_cache_put(sc); /* calls pool_put but see spl above */ 4243 splx(s); 4244 } 4245 4246 void 4247 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 4248 struct tcphdr *th) 4249 { 4250 struct syn_cache *sc; 4251 struct syn_cache_head *scp; 4252 int s; 4253 4254 s = splsoftnet(); 4255 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4256 splx(s); 4257 return; 4258 } 4259 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 4260 if (ntohl (th->th_seq) != sc->sc_iss) { 4261 splx(s); 4262 return; 4263 } 4264 4265 /* 4266 * If we've retransmitted 3 times and this is our second error, 4267 * we remove the entry. Otherwise, we allow it to continue on. 4268 * This prevents us from incorrectly nuking an entry during a 4269 * spurious network outage. 4270 * 4271 * See tcp_notify(). 4272 */ 4273 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 4274 sc->sc_flags |= SCF_UNREACH; 4275 splx(s); 4276 return; 4277 } 4278 4279 syn_cache_rm(sc); 4280 TCP_STATINC(TCP_STAT_SC_UNREACH); 4281 syn_cache_put(sc); /* calls pool_put but see spl above */ 4282 splx(s); 4283 } 4284 4285 /* 4286 * Given a LISTEN socket and an inbound SYN request, add 4287 * this to the syn cache, and send back a segment: 4288 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 4289 * to the source. 4290 * 4291 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 4292 * Doing so would require that we hold onto the data and deliver it 4293 * to the application. However, if we are the target of a SYN-flood 4294 * DoS attack, an attacker could send data which would eventually 4295 * consume all available buffer space if it were ACKed. By not ACKing 4296 * the data, we avoid this DoS scenario. 4297 */ 4298 4299 int 4300 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 4301 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 4302 int optlen, struct tcp_opt_info *oi) 4303 { 4304 struct tcpcb tb, *tp; 4305 long win; 4306 struct syn_cache *sc; 4307 struct syn_cache_head *scp; 4308 struct mbuf *ipopts; 4309 struct tcp_opt_info opti; 4310 int s; 4311 4312 tp = sototcpcb(so); 4313 4314 memset(&opti, 0, sizeof(opti)); 4315 4316 /* 4317 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 4318 * 4319 * Note this check is performed in tcp_input() very early on. 4320 */ 4321 4322 /* 4323 * Initialize some local state. 4324 */ 4325 win = sbspace(&so->so_rcv); 4326 if (win > TCP_MAXWIN) 4327 win = TCP_MAXWIN; 4328 4329 switch (src->sa_family) { 4330 #ifdef INET 4331 case AF_INET: 4332 /* 4333 * Remember the IP options, if any. 4334 */ 4335 ipopts = ip_srcroute(); 4336 break; 4337 #endif 4338 default: 4339 ipopts = NULL; 4340 } 4341 4342 #ifdef TCP_SIGNATURE 4343 if (optp || (tp->t_flags & TF_SIGNATURE)) 4344 #else 4345 if (optp) 4346 #endif 4347 { 4348 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4349 #ifdef TCP_SIGNATURE 4350 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4351 #endif 4352 tb.t_state = TCPS_LISTEN; 4353 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 4354 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 4355 return (0); 4356 } else 4357 tb.t_flags = 0; 4358 4359 /* 4360 * See if we already have an entry for this connection. 4361 * If we do, resend the SYN,ACK. We do not count this 4362 * as a retransmission (XXX though maybe we should). 4363 */ 4364 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4365 TCP_STATINC(TCP_STAT_SC_DUPESYN); 4366 if (ipopts) { 4367 /* 4368 * If we were remembering a previous source route, 4369 * forget it and use the new one we've been given. 4370 */ 4371 if (sc->sc_ipopts) 4372 (void) m_free(sc->sc_ipopts); 4373 sc->sc_ipopts = ipopts; 4374 } 4375 sc->sc_timestamp = tb.ts_recent; 4376 if (syn_cache_respond(sc, m) == 0) { 4377 uint64_t *tcps = TCP_STAT_GETREF(); 4378 tcps[TCP_STAT_SNDACKS]++; 4379 tcps[TCP_STAT_SNDTOTAL]++; 4380 TCP_STAT_PUTREF(); 4381 } 4382 return (1); 4383 } 4384 4385 s = splsoftnet(); 4386 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4387 splx(s); 4388 if (sc == NULL) { 4389 if (ipopts) 4390 (void) m_free(ipopts); 4391 return (0); 4392 } 4393 4394 /* 4395 * Fill in the cache, and put the necessary IP and TCP 4396 * options into the reply. 4397 */ 4398 memset(sc, 0, sizeof(struct syn_cache)); 4399 callout_init(&sc->sc_timer, CALLOUT_MPSAFE); 4400 bcopy(src, &sc->sc_src, src->sa_len); 4401 bcopy(dst, &sc->sc_dst, dst->sa_len); 4402 sc->sc_flags = 0; 4403 sc->sc_ipopts = ipopts; 4404 sc->sc_irs = th->th_seq; 4405 switch (src->sa_family) { 4406 #ifdef INET 4407 case AF_INET: 4408 { 4409 struct sockaddr_in *srcin = (void *) src; 4410 struct sockaddr_in *dstin = (void *) dst; 4411 4412 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4413 &srcin->sin_addr, dstin->sin_port, 4414 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4415 break; 4416 } 4417 #endif /* INET */ 4418 #ifdef INET6 4419 case AF_INET6: 4420 { 4421 struct sockaddr_in6 *srcin6 = (void *) src; 4422 struct sockaddr_in6 *dstin6 = (void *) dst; 4423 4424 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4425 &srcin6->sin6_addr, dstin6->sin6_port, 4426 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4427 break; 4428 } 4429 #endif /* INET6 */ 4430 } 4431 sc->sc_peermaxseg = oi->maxseg; 4432 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4433 m_get_rcvif_NOMPSAFE(m) : NULL, 4434 sc->sc_src.sa.sa_family); 4435 sc->sc_win = win; 4436 sc->sc_timebase = tcp_now - 1; /* see tcp_newtcpcb() */ 4437 sc->sc_timestamp = tb.ts_recent; 4438 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4439 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4440 sc->sc_flags |= SCF_TIMESTAMP; 4441 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4442 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4443 sc->sc_requested_s_scale = tb.requested_s_scale; 4444 sc->sc_request_r_scale = 0; 4445 /* 4446 * Pick the smallest possible scaling factor that 4447 * will still allow us to scale up to sb_max. 4448 * 4449 * We do this because there are broken firewalls that 4450 * will corrupt the window scale option, leading to 4451 * the other endpoint believing that our advertised 4452 * window is unscaled. At scale factors larger than 4453 * 5 the unscaled window will drop below 1500 bytes, 4454 * leading to serious problems when traversing these 4455 * broken firewalls. 4456 * 4457 * With the default sbmax of 256K, a scale factor 4458 * of 3 will be chosen by this algorithm. Those who 4459 * choose a larger sbmax should watch out 4460 * for the compatiblity problems mentioned above. 4461 * 4462 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4463 * or <SYN,ACK>) segment itself is never scaled. 4464 */ 4465 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4466 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4467 sc->sc_request_r_scale++; 4468 } else { 4469 sc->sc_requested_s_scale = 15; 4470 sc->sc_request_r_scale = 15; 4471 } 4472 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4473 sc->sc_flags |= SCF_SACK_PERMIT; 4474 4475 /* 4476 * ECN setup packet recieved. 4477 */ 4478 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4479 sc->sc_flags |= SCF_ECN_PERMIT; 4480 4481 #ifdef TCP_SIGNATURE 4482 if (tb.t_flags & TF_SIGNATURE) 4483 sc->sc_flags |= SCF_SIGNATURE; 4484 #endif 4485 sc->sc_tp = tp; 4486 if (syn_cache_respond(sc, m) == 0) { 4487 uint64_t *tcps = TCP_STAT_GETREF(); 4488 tcps[TCP_STAT_SNDACKS]++; 4489 tcps[TCP_STAT_SNDTOTAL]++; 4490 TCP_STAT_PUTREF(); 4491 syn_cache_insert(sc, tp); 4492 } else { 4493 s = splsoftnet(); 4494 /* 4495 * syn_cache_put() will try to schedule the timer, so 4496 * we need to initialize it 4497 */ 4498 SYN_CACHE_TIMER_ARM(sc); 4499 syn_cache_put(sc); 4500 splx(s); 4501 TCP_STATINC(TCP_STAT_SC_DROPPED); 4502 } 4503 return (1); 4504 } 4505 4506 /* 4507 * syn_cache_respond: (re)send SYN+ACK. 4508 * 4509 * returns 0 on success. otherwise returns an errno, typically ENOBUFS. 4510 */ 4511 4512 int 4513 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4514 { 4515 #ifdef INET6 4516 struct rtentry *rt; 4517 #endif 4518 struct route *ro; 4519 u_int8_t *optp; 4520 int optlen, error; 4521 u_int16_t tlen; 4522 struct ip *ip = NULL; 4523 #ifdef INET6 4524 struct ip6_hdr *ip6 = NULL; 4525 #endif 4526 struct tcpcb *tp = NULL; 4527 struct tcphdr *th; 4528 u_int hlen; 4529 struct socket *so; 4530 4531 ro = &sc->sc_route; 4532 switch (sc->sc_src.sa.sa_family) { 4533 case AF_INET: 4534 hlen = sizeof(struct ip); 4535 break; 4536 #ifdef INET6 4537 case AF_INET6: 4538 hlen = sizeof(struct ip6_hdr); 4539 break; 4540 #endif 4541 default: 4542 if (m) 4543 m_freem(m); 4544 return (EAFNOSUPPORT); 4545 } 4546 4547 /* Compute the size of the TCP options. */ 4548 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4549 ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) + 4550 #ifdef TCP_SIGNATURE 4551 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4552 #endif 4553 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4554 4555 tlen = hlen + sizeof(struct tcphdr) + optlen; 4556 4557 /* 4558 * Create the IP+TCP header from scratch. 4559 */ 4560 if (m) 4561 m_freem(m); 4562 #ifdef DIAGNOSTIC 4563 if (max_linkhdr + tlen > MCLBYTES) 4564 return (ENOBUFS); 4565 #endif 4566 MGETHDR(m, M_DONTWAIT, MT_DATA); 4567 if (m && (max_linkhdr + tlen) > MHLEN) { 4568 MCLGET(m, M_DONTWAIT); 4569 if ((m->m_flags & M_EXT) == 0) { 4570 m_freem(m); 4571 m = NULL; 4572 } 4573 } 4574 if (m == NULL) 4575 return (ENOBUFS); 4576 MCLAIM(m, &tcp_tx_mowner); 4577 4578 /* Fixup the mbuf. */ 4579 m->m_data += max_linkhdr; 4580 m->m_len = m->m_pkthdr.len = tlen; 4581 if (sc->sc_tp) { 4582 tp = sc->sc_tp; 4583 if (tp->t_inpcb) 4584 so = tp->t_inpcb->inp_socket; 4585 #ifdef INET6 4586 else if (tp->t_in6pcb) 4587 so = tp->t_in6pcb->in6p_socket; 4588 #endif 4589 else 4590 so = NULL; 4591 } else 4592 so = NULL; 4593 m_reset_rcvif(m); 4594 memset(mtod(m, u_char *), 0, tlen); 4595 4596 switch (sc->sc_src.sa.sa_family) { 4597 case AF_INET: 4598 ip = mtod(m, struct ip *); 4599 ip->ip_v = 4; 4600 ip->ip_dst = sc->sc_src.sin.sin_addr; 4601 ip->ip_src = sc->sc_dst.sin.sin_addr; 4602 ip->ip_p = IPPROTO_TCP; 4603 th = (struct tcphdr *)(ip + 1); 4604 th->th_dport = sc->sc_src.sin.sin_port; 4605 th->th_sport = sc->sc_dst.sin.sin_port; 4606 break; 4607 #ifdef INET6 4608 case AF_INET6: 4609 ip6 = mtod(m, struct ip6_hdr *); 4610 ip6->ip6_vfc = IPV6_VERSION; 4611 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4612 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4613 ip6->ip6_nxt = IPPROTO_TCP; 4614 /* ip6_plen will be updated in ip6_output() */ 4615 th = (struct tcphdr *)(ip6 + 1); 4616 th->th_dport = sc->sc_src.sin6.sin6_port; 4617 th->th_sport = sc->sc_dst.sin6.sin6_port; 4618 break; 4619 #endif 4620 default: 4621 th = NULL; 4622 } 4623 4624 th->th_seq = htonl(sc->sc_iss); 4625 th->th_ack = htonl(sc->sc_irs + 1); 4626 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4627 th->th_flags = TH_SYN|TH_ACK; 4628 th->th_win = htons(sc->sc_win); 4629 /* th_sum already 0 */ 4630 /* th_urp already 0 */ 4631 4632 /* Tack on the TCP options. */ 4633 optp = (u_int8_t *)(th + 1); 4634 *optp++ = TCPOPT_MAXSEG; 4635 *optp++ = 4; 4636 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4637 *optp++ = sc->sc_ourmaxseg & 0xff; 4638 4639 if (sc->sc_request_r_scale != 15) { 4640 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4641 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4642 sc->sc_request_r_scale); 4643 optp += 4; 4644 } 4645 4646 if (sc->sc_flags & SCF_TIMESTAMP) { 4647 u_int32_t *lp = (u_int32_t *)(optp); 4648 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4649 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4650 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4651 *lp = htonl(sc->sc_timestamp); 4652 optp += TCPOLEN_TSTAMP_APPA; 4653 } 4654 4655 if (sc->sc_flags & SCF_SACK_PERMIT) { 4656 u_int8_t *p = optp; 4657 4658 /* Let the peer know that we will SACK. */ 4659 p[0] = TCPOPT_SACK_PERMITTED; 4660 p[1] = 2; 4661 p[2] = TCPOPT_NOP; 4662 p[3] = TCPOPT_NOP; 4663 optp += 4; 4664 } 4665 4666 /* 4667 * Send ECN SYN-ACK setup packet. 4668 * Routes can be asymetric, so, even if we receive a packet 4669 * with ECE and CWR set, we must not assume no one will block 4670 * the ECE packet we are about to send. 4671 */ 4672 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4673 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4674 th->th_flags |= TH_ECE; 4675 TCP_STATINC(TCP_STAT_ECN_SHS); 4676 4677 /* 4678 * draft-ietf-tcpm-ecnsyn-00.txt 4679 * 4680 * "[...] a TCP node MAY respond to an ECN-setup 4681 * SYN packet by setting ECT in the responding 4682 * ECN-setup SYN/ACK packet, indicating to routers 4683 * that the SYN/ACK packet is ECN-Capable. 4684 * This allows a congested router along the path 4685 * to mark the packet instead of dropping the 4686 * packet as an indication of congestion." 4687 * 4688 * "[...] There can be a great benefit in setting 4689 * an ECN-capable codepoint in SYN/ACK packets [...] 4690 * Congestion is most likely to occur in 4691 * the server-to-client direction. As a result, 4692 * setting an ECN-capable codepoint in SYN/ACK 4693 * packets can reduce the occurence of three-second 4694 * retransmit timeouts resulting from the drop 4695 * of SYN/ACK packets." 4696 * 4697 * Page 4 and 6, January 2006. 4698 */ 4699 4700 switch (sc->sc_src.sa.sa_family) { 4701 #ifdef INET 4702 case AF_INET: 4703 ip->ip_tos |= IPTOS_ECN_ECT0; 4704 break; 4705 #endif 4706 #ifdef INET6 4707 case AF_INET6: 4708 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4709 break; 4710 #endif 4711 } 4712 TCP_STATINC(TCP_STAT_ECN_ECT); 4713 } 4714 4715 #ifdef TCP_SIGNATURE 4716 if (sc->sc_flags & SCF_SIGNATURE) { 4717 struct secasvar *sav; 4718 u_int8_t *sigp; 4719 4720 sav = tcp_signature_getsav(m, th); 4721 4722 if (sav == NULL) { 4723 if (m) 4724 m_freem(m); 4725 return (EPERM); 4726 } 4727 4728 *optp++ = TCPOPT_SIGNATURE; 4729 *optp++ = TCPOLEN_SIGNATURE; 4730 sigp = optp; 4731 memset(optp, 0, TCP_SIGLEN); 4732 optp += TCP_SIGLEN; 4733 *optp++ = TCPOPT_NOP; 4734 *optp++ = TCPOPT_EOL; 4735 4736 (void)tcp_signature(m, th, hlen, sav, sigp); 4737 4738 key_sa_recordxfer(sav, m); 4739 KEY_FREESAV(&sav); 4740 } 4741 #endif 4742 4743 /* Compute the packet's checksum. */ 4744 switch (sc->sc_src.sa.sa_family) { 4745 case AF_INET: 4746 ip->ip_len = htons(tlen - hlen); 4747 th->th_sum = 0; 4748 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4749 break; 4750 #ifdef INET6 4751 case AF_INET6: 4752 ip6->ip6_plen = htons(tlen - hlen); 4753 th->th_sum = 0; 4754 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4755 break; 4756 #endif 4757 } 4758 4759 /* 4760 * Fill in some straggling IP bits. Note the stack expects 4761 * ip_len to be in host order, for convenience. 4762 */ 4763 switch (sc->sc_src.sa.sa_family) { 4764 #ifdef INET 4765 case AF_INET: 4766 ip->ip_len = htons(tlen); 4767 ip->ip_ttl = ip_defttl; 4768 /* XXX tos? */ 4769 break; 4770 #endif 4771 #ifdef INET6 4772 case AF_INET6: 4773 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4774 ip6->ip6_vfc |= IPV6_VERSION; 4775 ip6->ip6_plen = htons(tlen - hlen); 4776 /* ip6_hlim will be initialized afterwards */ 4777 /* XXX flowlabel? */ 4778 break; 4779 #endif 4780 } 4781 4782 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4783 tp = sc->sc_tp; 4784 4785 switch (sc->sc_src.sa.sa_family) { 4786 #ifdef INET 4787 case AF_INET: 4788 error = ip_output(m, sc->sc_ipopts, ro, 4789 (ip_mtudisc ? IP_MTUDISC : 0), 4790 NULL, so); 4791 break; 4792 #endif 4793 #ifdef INET6 4794 case AF_INET6: 4795 ip6->ip6_hlim = in6_selecthlim(NULL, 4796 (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp : NULL); 4797 4798 error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, so, NULL); 4799 break; 4800 #endif 4801 default: 4802 error = EAFNOSUPPORT; 4803 break; 4804 } 4805 return (error); 4806 } 4807