tcp_input.c revision 1.370
1 /* $NetBSD: tcp_input.c,v 1.370 2018/02/08 20:06:21 maxv 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 1.370 2018/02/08 20:06:21 maxv 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 static void syn_cache_timer(void *); 253 254 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 255 256 /* for modulo comparisons of timestamps */ 257 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 258 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 259 260 /* 261 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 262 */ 263 #ifdef INET6 264 static inline void 265 nd6_hint(struct tcpcb *tp) 266 { 267 struct rtentry *rt = NULL; 268 269 if (tp != NULL && tp->t_in6pcb != NULL && tp->t_family == AF_INET6 && 270 (rt = rtcache_validate(&tp->t_in6pcb->in6p_route)) != NULL) 271 nd6_nud_hint(rt); 272 rtcache_unref(rt, &tp->t_in6pcb->in6p_route); 273 } 274 #else 275 static inline void 276 nd6_hint(struct tcpcb *tp) 277 { 278 } 279 #endif 280 281 /* 282 * Compute ACK transmission behavior. Delay the ACK unless 283 * we have already delayed an ACK (must send an ACK every two segments). 284 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 285 * option is enabled. 286 */ 287 static void 288 tcp_setup_ack(struct tcpcb *tp, const struct tcphdr *th) 289 { 290 291 if (tp->t_flags & TF_DELACK || 292 (tcp_ack_on_push && th->th_flags & TH_PUSH)) 293 tp->t_flags |= TF_ACKNOW; 294 else 295 TCP_SET_DELACK(tp); 296 } 297 298 static void 299 icmp_check(struct tcpcb *tp, const struct tcphdr *th, int acked) 300 { 301 302 /* 303 * If we had a pending ICMP message that refers to data that have 304 * just been acknowledged, disregard the recorded ICMP message. 305 */ 306 if ((tp->t_flags & TF_PMTUD_PEND) && 307 SEQ_GT(th->th_ack, tp->t_pmtud_th_seq)) 308 tp->t_flags &= ~TF_PMTUD_PEND; 309 310 /* 311 * Keep track of the largest chunk of data 312 * acknowledged since last PMTU update 313 */ 314 if (tp->t_pmtud_mss_acked < acked) 315 tp->t_pmtud_mss_acked = acked; 316 } 317 318 /* 319 * Convert TCP protocol fields to host order for easier processing. 320 */ 321 static void 322 tcp_fields_to_host(struct tcphdr *th) 323 { 324 325 NTOHL(th->th_seq); 326 NTOHL(th->th_ack); 327 NTOHS(th->th_win); 328 NTOHS(th->th_urp); 329 } 330 331 /* 332 * ... and reverse the above. 333 */ 334 static void 335 tcp_fields_to_net(struct tcphdr *th) 336 { 337 338 HTONL(th->th_seq); 339 HTONL(th->th_ack); 340 HTONS(th->th_win); 341 HTONS(th->th_urp); 342 } 343 344 #ifdef TCP_CSUM_COUNTERS 345 #include <sys/device.h> 346 347 #if defined(INET) 348 extern struct evcnt tcp_hwcsum_ok; 349 extern struct evcnt tcp_hwcsum_bad; 350 extern struct evcnt tcp_hwcsum_data; 351 extern struct evcnt tcp_swcsum; 352 #endif /* defined(INET) */ 353 #if defined(INET6) 354 extern struct evcnt tcp6_hwcsum_ok; 355 extern struct evcnt tcp6_hwcsum_bad; 356 extern struct evcnt tcp6_hwcsum_data; 357 extern struct evcnt tcp6_swcsum; 358 #endif /* defined(INET6) */ 359 360 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 361 362 #else 363 364 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 365 366 #endif /* TCP_CSUM_COUNTERS */ 367 368 #ifdef TCP_REASS_COUNTERS 369 #include <sys/device.h> 370 371 extern struct evcnt tcp_reass_; 372 extern struct evcnt tcp_reass_empty; 373 extern struct evcnt tcp_reass_iteration[8]; 374 extern struct evcnt tcp_reass_prependfirst; 375 extern struct evcnt tcp_reass_prepend; 376 extern struct evcnt tcp_reass_insert; 377 extern struct evcnt tcp_reass_inserttail; 378 extern struct evcnt tcp_reass_append; 379 extern struct evcnt tcp_reass_appendtail; 380 extern struct evcnt tcp_reass_overlaptail; 381 extern struct evcnt tcp_reass_overlapfront; 382 extern struct evcnt tcp_reass_segdup; 383 extern struct evcnt tcp_reass_fragdup; 384 385 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 386 387 #else 388 389 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 390 391 #endif /* TCP_REASS_COUNTERS */ 392 393 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *, 394 int *); 395 static int tcp_dooptions(struct tcpcb *, const u_char *, int, 396 struct tcphdr *, struct mbuf *, int, struct tcp_opt_info *); 397 398 #ifdef INET 399 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 400 #endif 401 #ifdef INET6 402 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 403 #endif 404 405 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 406 407 #if defined(MBUFTRACE) 408 struct mowner tcp_reass_mowner = MOWNER_INIT("tcp", "reass"); 409 #endif /* defined(MBUFTRACE) */ 410 411 static struct pool tcpipqent_pool; 412 413 void 414 tcpipqent_init(void) 415 { 416 417 pool_init(&tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", 418 NULL, IPL_VM); 419 } 420 421 struct ipqent * 422 tcpipqent_alloc(void) 423 { 424 struct ipqent *ipqe; 425 int s; 426 427 s = splvm(); 428 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 429 splx(s); 430 431 return ipqe; 432 } 433 434 void 435 tcpipqent_free(struct ipqent *ipqe) 436 { 437 int s; 438 439 s = splvm(); 440 pool_put(&tcpipqent_pool, ipqe); 441 splx(s); 442 } 443 444 static int 445 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int *tlen) 446 { 447 struct ipqent *p, *q, *nq, *tiqe = NULL; 448 struct socket *so = NULL; 449 int pkt_flags; 450 tcp_seq pkt_seq; 451 unsigned pkt_len; 452 u_long rcvpartdupbyte = 0; 453 u_long rcvoobyte; 454 #ifdef TCP_REASS_COUNTERS 455 u_int count = 0; 456 #endif 457 uint64_t *tcps; 458 459 if (tp->t_inpcb) 460 so = tp->t_inpcb->inp_socket; 461 #ifdef INET6 462 else if (tp->t_in6pcb) 463 so = tp->t_in6pcb->in6p_socket; 464 #endif 465 466 TCP_REASS_LOCK_CHECK(tp); 467 468 /* 469 * Call with th==0 after become established to 470 * force pre-ESTABLISHED data up to user socket. 471 */ 472 if (th == 0) 473 goto present; 474 475 m_claimm(m, &tcp_reass_mowner); 476 477 rcvoobyte = *tlen; 478 /* 479 * Copy these to local variables because the tcpiphdr 480 * gets munged while we are collapsing mbufs. 481 */ 482 pkt_seq = th->th_seq; 483 pkt_len = *tlen; 484 pkt_flags = th->th_flags; 485 486 TCP_REASS_COUNTER_INCR(&tcp_reass_); 487 488 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 489 /* 490 * When we miss a packet, the vast majority of time we get 491 * packets that follow it in order. So optimize for that. 492 */ 493 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 494 p->ipqe_len += pkt_len; 495 p->ipqe_flags |= pkt_flags; 496 m_cat(p->ipre_mlast, m); 497 TRAVERSE(p->ipre_mlast); 498 m = NULL; 499 tiqe = p; 500 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 501 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 502 goto skip_replacement; 503 } 504 /* 505 * While we're here, if the pkt is completely beyond 506 * anything we have, just insert it at the tail. 507 */ 508 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 509 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 510 goto insert_it; 511 } 512 } 513 514 q = TAILQ_FIRST(&tp->segq); 515 516 if (q != NULL) { 517 /* 518 * If this segment immediately precedes the first out-of-order 519 * block, simply slap the segment in front of it and (mostly) 520 * skip the complicated logic. 521 */ 522 if (pkt_seq + pkt_len == q->ipqe_seq) { 523 q->ipqe_seq = pkt_seq; 524 q->ipqe_len += pkt_len; 525 q->ipqe_flags |= pkt_flags; 526 m_cat(m, q->ipqe_m); 527 q->ipqe_m = m; 528 q->ipre_mlast = m; /* last mbuf may have changed */ 529 TRAVERSE(q->ipre_mlast); 530 tiqe = q; 531 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 532 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 533 goto skip_replacement; 534 } 535 } else { 536 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 537 } 538 539 /* 540 * Find a segment which begins after this one does. 541 */ 542 for (p = NULL; q != NULL; q = nq) { 543 nq = TAILQ_NEXT(q, ipqe_q); 544 #ifdef TCP_REASS_COUNTERS 545 count++; 546 #endif 547 /* 548 * If the received segment is just right after this 549 * fragment, merge the two together and then check 550 * for further overlaps. 551 */ 552 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 553 #ifdef TCPREASS_DEBUG 554 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 555 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 556 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 557 #endif 558 pkt_len += q->ipqe_len; 559 pkt_flags |= q->ipqe_flags; 560 pkt_seq = q->ipqe_seq; 561 m_cat(q->ipre_mlast, m); 562 TRAVERSE(q->ipre_mlast); 563 m = q->ipqe_m; 564 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 565 goto free_ipqe; 566 } 567 /* 568 * If the received segment is completely past this 569 * fragment, we need to go the next fragment. 570 */ 571 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 572 p = q; 573 continue; 574 } 575 /* 576 * If the fragment is past the received segment, 577 * it (or any following) can't be concatenated. 578 */ 579 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 580 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 581 break; 582 } 583 584 /* 585 * We've received all the data in this segment before. 586 * mark it as a duplicate and return. 587 */ 588 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 589 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 590 tcps = TCP_STAT_GETREF(); 591 tcps[TCP_STAT_RCVDUPPACK]++; 592 tcps[TCP_STAT_RCVDUPBYTE] += pkt_len; 593 TCP_STAT_PUTREF(); 594 tcp_new_dsack(tp, pkt_seq, pkt_len); 595 m_freem(m); 596 if (tiqe != NULL) { 597 tcpipqent_free(tiqe); 598 } 599 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 600 goto out; 601 } 602 /* 603 * Received segment completely overlaps this fragment 604 * so we drop the fragment (this keeps the temporal 605 * ordering of segments correct). 606 */ 607 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 608 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 609 rcvpartdupbyte += q->ipqe_len; 610 m_freem(q->ipqe_m); 611 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 612 goto free_ipqe; 613 } 614 /* 615 * RX'ed segment extends past the end of the 616 * fragment. Drop the overlapping bytes. Then 617 * merge the fragment and segment then treat as 618 * a longer received packet. 619 */ 620 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 621 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 622 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 623 #ifdef TCPREASS_DEBUG 624 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 625 tp, overlap, 626 pkt_seq, pkt_seq + pkt_len, pkt_len); 627 #endif 628 m_adj(m, overlap); 629 rcvpartdupbyte += overlap; 630 m_cat(q->ipre_mlast, m); 631 TRAVERSE(q->ipre_mlast); 632 m = q->ipqe_m; 633 pkt_seq = q->ipqe_seq; 634 pkt_len += q->ipqe_len - overlap; 635 rcvoobyte -= overlap; 636 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 637 goto free_ipqe; 638 } 639 /* 640 * RX'ed segment extends past the front of the 641 * fragment. Drop the overlapping bytes on the 642 * received packet. The packet will then be 643 * contatentated with this fragment a bit later. 644 */ 645 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 646 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 647 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 648 #ifdef TCPREASS_DEBUG 649 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 650 tp, overlap, 651 pkt_seq, pkt_seq + pkt_len, pkt_len); 652 #endif 653 m_adj(m, -overlap); 654 pkt_len -= overlap; 655 rcvpartdupbyte += overlap; 656 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 657 rcvoobyte -= overlap; 658 } 659 /* 660 * If the received segment immediates precedes this 661 * fragment then tack the fragment onto this segment 662 * and reinsert the data. 663 */ 664 if (q->ipqe_seq == pkt_seq + pkt_len) { 665 #ifdef TCPREASS_DEBUG 666 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 667 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 668 pkt_seq, pkt_seq + pkt_len, pkt_len); 669 #endif 670 pkt_len += q->ipqe_len; 671 pkt_flags |= q->ipqe_flags; 672 m_cat(m, q->ipqe_m); 673 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 674 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 675 tp->t_segqlen--; 676 KASSERT(tp->t_segqlen >= 0); 677 KASSERT(tp->t_segqlen != 0 || 678 (TAILQ_EMPTY(&tp->segq) && 679 TAILQ_EMPTY(&tp->timeq))); 680 if (tiqe == NULL) { 681 tiqe = q; 682 } else { 683 tcpipqent_free(q); 684 } 685 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 686 break; 687 } 688 /* 689 * If the fragment is before the segment, remember it. 690 * When this loop is terminated, p will contain the 691 * pointer to fragment that is right before the received 692 * segment. 693 */ 694 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 695 p = q; 696 697 continue; 698 699 /* 700 * This is a common operation. It also will allow 701 * to save doing a malloc/free in most instances. 702 */ 703 free_ipqe: 704 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 705 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 706 tp->t_segqlen--; 707 KASSERT(tp->t_segqlen >= 0); 708 KASSERT(tp->t_segqlen != 0 || 709 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 710 if (tiqe == NULL) { 711 tiqe = q; 712 } else { 713 tcpipqent_free(q); 714 } 715 } 716 717 #ifdef TCP_REASS_COUNTERS 718 if (count > 7) 719 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 720 else if (count > 0) 721 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 722 #endif 723 724 insert_it: 725 726 /* 727 * Allocate a new queue entry since the received segment did not 728 * collapse onto any other out-of-order block; thus we are allocating 729 * a new block. If it had collapsed, tiqe would not be NULL and 730 * we would be reusing it. 731 * XXX If we can't, just drop the packet. XXX 732 */ 733 if (tiqe == NULL) { 734 tiqe = tcpipqent_alloc(); 735 if (tiqe == NULL) { 736 TCP_STATINC(TCP_STAT_RCVMEMDROP); 737 m_freem(m); 738 goto out; 739 } 740 } 741 742 /* 743 * Update the counters. 744 */ 745 tp->t_rcvoopack++; 746 tcps = TCP_STAT_GETREF(); 747 tcps[TCP_STAT_RCVOOPACK]++; 748 tcps[TCP_STAT_RCVOOBYTE] += rcvoobyte; 749 if (rcvpartdupbyte) { 750 tcps[TCP_STAT_RCVPARTDUPPACK]++; 751 tcps[TCP_STAT_RCVPARTDUPBYTE] += rcvpartdupbyte; 752 } 753 TCP_STAT_PUTREF(); 754 755 /* 756 * Insert the new fragment queue entry into both queues. 757 */ 758 tiqe->ipqe_m = m; 759 tiqe->ipre_mlast = m; 760 tiqe->ipqe_seq = pkt_seq; 761 tiqe->ipqe_len = pkt_len; 762 tiqe->ipqe_flags = pkt_flags; 763 if (p == NULL) { 764 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 765 #ifdef TCPREASS_DEBUG 766 if (tiqe->ipqe_seq != tp->rcv_nxt) 767 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 768 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 769 #endif 770 } else { 771 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 772 #ifdef TCPREASS_DEBUG 773 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 774 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 775 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 776 #endif 777 } 778 tp->t_segqlen++; 779 780 skip_replacement: 781 782 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 783 784 present: 785 /* 786 * Present data to user, advancing rcv_nxt through 787 * completed sequence space. 788 */ 789 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 790 goto out; 791 q = TAILQ_FIRST(&tp->segq); 792 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 793 goto out; 794 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 795 goto out; 796 797 tp->rcv_nxt += q->ipqe_len; 798 pkt_flags = q->ipqe_flags & TH_FIN; 799 nd6_hint(tp); 800 801 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 802 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 803 tp->t_segqlen--; 804 KASSERT(tp->t_segqlen >= 0); 805 KASSERT(tp->t_segqlen != 0 || 806 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 807 if (so->so_state & SS_CANTRCVMORE) 808 m_freem(q->ipqe_m); 809 else 810 sbappendstream(&so->so_rcv, q->ipqe_m); 811 tcpipqent_free(q); 812 TCP_REASS_UNLOCK(tp); 813 sorwakeup(so); 814 return (pkt_flags); 815 out: 816 TCP_REASS_UNLOCK(tp); 817 return (0); 818 } 819 820 #ifdef INET6 821 int 822 tcp6_input(struct mbuf **mp, int *offp, int proto) 823 { 824 struct mbuf *m = *mp; 825 826 /* 827 * draft-itojun-ipv6-tcp-to-anycast 828 * better place to put this in? 829 */ 830 if (m->m_flags & M_ANYCAST6) { 831 struct ip6_hdr *ip6; 832 if (m->m_len < sizeof(struct ip6_hdr)) { 833 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 834 TCP_STATINC(TCP_STAT_RCVSHORT); 835 return IPPROTO_DONE; 836 } 837 } 838 ip6 = mtod(m, struct ip6_hdr *); 839 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 840 (char *)&ip6->ip6_dst - (char *)ip6); 841 return IPPROTO_DONE; 842 } 843 844 tcp_input(m, *offp, proto); 845 return IPPROTO_DONE; 846 } 847 #endif 848 849 #ifdef INET 850 static void 851 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 852 { 853 char src[INET_ADDRSTRLEN]; 854 char dst[INET_ADDRSTRLEN]; 855 856 if (ip) { 857 in_print(src, sizeof(src), &ip->ip_src); 858 in_print(dst, sizeof(dst), &ip->ip_dst); 859 } 860 else { 861 strlcpy(src, "(unknown)", sizeof(src)); 862 strlcpy(dst, "(unknown)", sizeof(dst)); 863 } 864 log(LOG_INFO, 865 "Connection attempt to TCP %s:%d from %s:%d\n", 866 dst, ntohs(th->th_dport), 867 src, ntohs(th->th_sport)); 868 } 869 #endif 870 871 #ifdef INET6 872 static void 873 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 874 { 875 char src[INET6_ADDRSTRLEN]; 876 char dst[INET6_ADDRSTRLEN]; 877 878 if (ip6) { 879 in6_print(src, sizeof(src), &ip6->ip6_src); 880 in6_print(dst, sizeof(dst), &ip6->ip6_dst); 881 } 882 else { 883 strlcpy(src, "(unknown v6)", sizeof(src)); 884 strlcpy(dst, "(unknown v6)", sizeof(dst)); 885 } 886 log(LOG_INFO, 887 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 888 dst, ntohs(th->th_dport), 889 src, ntohs(th->th_sport)); 890 } 891 #endif 892 893 /* 894 * Checksum extended TCP header and data. 895 */ 896 int 897 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, 898 int toff, int off, int tlen) 899 { 900 struct ifnet *rcvif; 901 int s; 902 903 /* 904 * XXX it's better to record and check if this mbuf is 905 * already checked. 906 */ 907 908 rcvif = m_get_rcvif(m, &s); 909 if (__predict_false(rcvif == NULL)) 910 goto badcsum; /* XXX */ 911 912 switch (af) { 913 #ifdef INET 914 case AF_INET: 915 switch (m->m_pkthdr.csum_flags & 916 ((rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 917 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 918 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 919 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 920 goto badcsum; 921 922 case M_CSUM_TCPv4|M_CSUM_DATA: { 923 u_int32_t hw_csum = m->m_pkthdr.csum_data; 924 925 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 926 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 927 const struct ip *ip = 928 mtod(m, const struct ip *); 929 930 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 931 ip->ip_dst.s_addr, 932 htons(hw_csum + tlen + off + IPPROTO_TCP)); 933 } 934 if ((hw_csum ^ 0xffff) != 0) 935 goto badcsum; 936 break; 937 } 938 939 case M_CSUM_TCPv4: 940 /* Checksum was okay. */ 941 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 942 break; 943 944 default: 945 /* 946 * Must compute it ourselves. Maybe skip checksum 947 * on loopback interfaces. 948 */ 949 if (__predict_true(!(rcvif->if_flags & IFF_LOOPBACK) || 950 tcp_do_loopback_cksum)) { 951 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 952 if (in4_cksum(m, IPPROTO_TCP, toff, 953 tlen + off) != 0) 954 goto badcsum; 955 } 956 break; 957 } 958 break; 959 #endif /* INET4 */ 960 961 #ifdef INET6 962 case AF_INET6: 963 switch (m->m_pkthdr.csum_flags & 964 ((rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 965 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 966 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 967 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 968 goto badcsum; 969 970 #if 0 /* notyet */ 971 case M_CSUM_TCPv6|M_CSUM_DATA: 972 #endif 973 974 case M_CSUM_TCPv6: 975 /* Checksum was okay. */ 976 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 977 break; 978 979 default: 980 /* 981 * Must compute it ourselves. Maybe skip checksum 982 * on loopback interfaces. 983 */ 984 if (__predict_true((m->m_flags & M_LOOP) == 0 || 985 tcp_do_loopback_cksum)) { 986 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 987 if (in6_cksum(m, IPPROTO_TCP, toff, 988 tlen + off) != 0) 989 goto badcsum; 990 } 991 } 992 break; 993 #endif /* INET6 */ 994 } 995 m_put_rcvif(rcvif, &s); 996 997 return 0; 998 999 badcsum: 1000 m_put_rcvif(rcvif, &s); 1001 TCP_STATINC(TCP_STAT_RCVBADSUM); 1002 return -1; 1003 } 1004 1005 /* When a packet arrives addressed to a vestigial tcpbp, we 1006 * nevertheless have to respond to it per the spec. 1007 */ 1008 static void tcp_vtw_input(struct tcphdr *th, vestigial_inpcb_t *vp, 1009 struct mbuf *m, int tlen, int multicast) 1010 { 1011 int tiflags; 1012 int todrop; 1013 uint32_t t_flags = 0; 1014 uint64_t *tcps; 1015 1016 tiflags = th->th_flags; 1017 todrop = vp->rcv_nxt - th->th_seq; 1018 1019 if (todrop > 0) { 1020 if (tiflags & TH_SYN) { 1021 tiflags &= ~TH_SYN; 1022 ++th->th_seq; 1023 if (th->th_urp > 1) 1024 --th->th_urp; 1025 else { 1026 tiflags &= ~TH_URG; 1027 th->th_urp = 0; 1028 } 1029 --todrop; 1030 } 1031 if (todrop > tlen || 1032 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1033 /* 1034 * Any valid FIN or RST must be to the left of the 1035 * window. At this point the FIN or RST must be a 1036 * duplicate or out of sequence; drop it. 1037 */ 1038 if (tiflags & TH_RST) 1039 goto drop; 1040 tiflags &= ~(TH_FIN|TH_RST); 1041 /* 1042 * Send an ACK to resynchronize and drop any data. 1043 * But keep on processing for RST or ACK. 1044 */ 1045 t_flags |= TF_ACKNOW; 1046 todrop = tlen; 1047 tcps = TCP_STAT_GETREF(); 1048 tcps[TCP_STAT_RCVDUPPACK] += 1; 1049 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 1050 TCP_STAT_PUTREF(); 1051 } else if ((tiflags & TH_RST) 1052 && th->th_seq != vp->rcv_nxt) { 1053 /* 1054 * Test for reset before adjusting the sequence 1055 * number for overlapping data. 1056 */ 1057 goto dropafterack_ratelim; 1058 } else { 1059 tcps = TCP_STAT_GETREF(); 1060 tcps[TCP_STAT_RCVPARTDUPPACK] += 1; 1061 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 1062 TCP_STAT_PUTREF(); 1063 } 1064 1065 // tcp_new_dsack(tp, th->th_seq, todrop); 1066 // hdroptlen += todrop; /*drop from head afterwards*/ 1067 1068 th->th_seq += todrop; 1069 tlen -= todrop; 1070 1071 if (th->th_urp > todrop) 1072 th->th_urp -= todrop; 1073 else { 1074 tiflags &= ~TH_URG; 1075 th->th_urp = 0; 1076 } 1077 } 1078 1079 /* 1080 * If new data are received on a connection after the 1081 * user processes are gone, then RST the other end. 1082 */ 1083 if (tlen) { 1084 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 1085 goto dropwithreset; 1086 } 1087 1088 /* 1089 * If segment ends after window, drop trailing data 1090 * (and PUSH and FIN); if nothing left, just ACK. 1091 */ 1092 todrop = (th->th_seq + tlen) - (vp->rcv_nxt+vp->rcv_wnd); 1093 1094 if (todrop > 0) { 1095 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 1096 if (todrop >= tlen) { 1097 /* 1098 * The segment actually starts after the window. 1099 * th->th_seq + tlen - vp->rcv_nxt - vp->rcv_wnd >= tlen 1100 * th->th_seq - vp->rcv_nxt - vp->rcv_wnd >= 0 1101 * th->th_seq >= vp->rcv_nxt + vp->rcv_wnd 1102 */ 1103 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 1104 /* 1105 * If a new connection request is received 1106 * while in TIME_WAIT, drop the old connection 1107 * and start over if the sequence numbers 1108 * are above the previous ones. 1109 */ 1110 if ((tiflags & TH_SYN) 1111 && SEQ_GT(th->th_seq, vp->rcv_nxt)) { 1112 /* We only support this in the !NOFDREF case, which 1113 * is to say: not here. 1114 */ 1115 goto dropwithreset; 1116 } 1117 /* 1118 * If window is closed can only take segments at 1119 * window edge, and have to drop data and PUSH from 1120 * incoming segments. Continue processing, but 1121 * remember to ack. Otherwise, drop segment 1122 * and (if not RST) ack. 1123 */ 1124 if (vp->rcv_wnd == 0 && th->th_seq == vp->rcv_nxt) { 1125 t_flags |= TF_ACKNOW; 1126 TCP_STATINC(TCP_STAT_RCVWINPROBE); 1127 } else 1128 goto dropafterack; 1129 } else 1130 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 1131 m_adj(m, -todrop); 1132 tlen -= todrop; 1133 tiflags &= ~(TH_PUSH|TH_FIN); 1134 } 1135 1136 if (tiflags & TH_RST) { 1137 if (th->th_seq != vp->rcv_nxt) 1138 goto dropafterack_ratelim; 1139 1140 vtw_del(vp->ctl, vp->vtw); 1141 goto drop; 1142 } 1143 1144 /* 1145 * If the ACK bit is off we drop the segment and return. 1146 */ 1147 if ((tiflags & TH_ACK) == 0) { 1148 if (t_flags & TF_ACKNOW) 1149 goto dropafterack; 1150 else 1151 goto drop; 1152 } 1153 1154 /* 1155 * In TIME_WAIT state the only thing that should arrive 1156 * is a retransmission of the remote FIN. Acknowledge 1157 * it and restart the finack timer. 1158 */ 1159 vtw_restart(vp); 1160 goto dropafterack; 1161 1162 dropafterack: 1163 /* 1164 * Generate an ACK dropping incoming segment if it occupies 1165 * sequence space, where the ACK reflects our state. 1166 */ 1167 if (tiflags & TH_RST) 1168 goto drop; 1169 goto dropafterack2; 1170 1171 dropafterack_ratelim: 1172 /* 1173 * We may want to rate-limit ACKs against SYN/RST attack. 1174 */ 1175 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 1176 tcp_ackdrop_ppslim) == 0) { 1177 /* XXX stat */ 1178 goto drop; 1179 } 1180 /* ...fall into dropafterack2... */ 1181 1182 dropafterack2: 1183 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, th->th_ack, 1184 TH_ACK); 1185 return; 1186 1187 dropwithreset: 1188 /* 1189 * Generate a RST, dropping incoming segment. 1190 * Make ACK acceptable to originator of segment. 1191 */ 1192 if (tiflags & TH_RST) 1193 goto drop; 1194 1195 if (tiflags & TH_ACK) 1196 tcp_respond(0, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 1197 else { 1198 if (tiflags & TH_SYN) 1199 ++tlen; 1200 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, (tcp_seq)0, 1201 TH_RST|TH_ACK); 1202 } 1203 return; 1204 drop: 1205 m_freem(m); 1206 } 1207 1208 /* 1209 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 1210 */ 1211 void 1212 tcp_input(struct mbuf *m, ...) 1213 { 1214 struct tcphdr *th; 1215 struct ip *ip; 1216 struct inpcb *inp; 1217 #ifdef INET6 1218 struct ip6_hdr *ip6; 1219 struct in6pcb *in6p; 1220 #endif 1221 u_int8_t *optp = NULL; 1222 int optlen = 0; 1223 int len, tlen, toff, hdroptlen = 0; 1224 struct tcpcb *tp = NULL; 1225 int tiflags; 1226 struct socket *so = NULL; 1227 int todrop, acked, ourfinisacked, needoutput = 0; 1228 bool dupseg; 1229 #ifdef TCP_DEBUG 1230 short ostate = 0; 1231 #endif 1232 u_long tiwin; 1233 struct tcp_opt_info opti; 1234 int off, iphlen; 1235 va_list ap; 1236 int af; /* af on the wire */ 1237 struct mbuf *tcp_saveti = NULL; 1238 uint32_t ts_rtt; 1239 uint8_t iptos; 1240 uint64_t *tcps; 1241 vestigial_inpcb_t vestige; 1242 1243 vestige.valid = 0; 1244 1245 MCLAIM(m, &tcp_rx_mowner); 1246 va_start(ap, m); 1247 toff = va_arg(ap, int); 1248 (void)va_arg(ap, int); /* ignore value, advance ap */ 1249 va_end(ap); 1250 1251 TCP_STATINC(TCP_STAT_RCVTOTAL); 1252 1253 memset(&opti, 0, sizeof(opti)); 1254 opti.ts_present = 0; 1255 opti.maxseg = 0; 1256 1257 /* 1258 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 1259 * 1260 * TCP is, by definition, unicast, so we reject all 1261 * multicast outright. 1262 * 1263 * Note, there are additional src/dst address checks in 1264 * the AF-specific code below. 1265 */ 1266 if (m->m_flags & (M_BCAST|M_MCAST)) { 1267 /* XXX stat */ 1268 goto drop; 1269 } 1270 #ifdef INET6 1271 if (m->m_flags & M_ANYCAST6) { 1272 /* XXX stat */ 1273 goto drop; 1274 } 1275 #endif 1276 1277 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, sizeof(struct tcphdr)); 1278 if (th == NULL) { 1279 TCP_STATINC(TCP_STAT_RCVSHORT); 1280 return; 1281 } 1282 1283 /* 1284 * Get IP and TCP header. 1285 * Note: IP leaves IP header in first mbuf. 1286 */ 1287 ip = mtod(m, struct ip *); 1288 switch (ip->ip_v) { 1289 #ifdef INET 1290 case 4: 1291 #ifdef INET6 1292 ip6 = NULL; 1293 #endif 1294 af = AF_INET; 1295 iphlen = sizeof(struct ip); 1296 1297 /* We do the checksum after PCB lookup... */ 1298 len = ntohs(ip->ip_len); 1299 tlen = len - toff; 1300 iptos = ip->ip_tos; 1301 break; 1302 #endif 1303 #ifdef INET6 1304 case 6: 1305 ip = NULL; 1306 iphlen = sizeof(struct ip6_hdr); 1307 af = AF_INET6; 1308 ip6 = mtod(m, struct ip6_hdr *); 1309 1310 /* 1311 * Be proactive about unspecified IPv6 address in source. 1312 * As we use all-zero to indicate unbounded/unconnected pcb, 1313 * unspecified IPv6 address can be used to confuse us. 1314 * 1315 * Note that packets with unspecified IPv6 destination is 1316 * already dropped in ip6_input. 1317 */ 1318 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1319 /* XXX stat */ 1320 goto drop; 1321 } 1322 1323 /* 1324 * Make sure destination address is not multicast. 1325 * Source address checked in ip6_input(). 1326 */ 1327 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1328 /* XXX stat */ 1329 goto drop; 1330 } 1331 1332 /* We do the checksum after PCB lookup... */ 1333 len = m->m_pkthdr.len; 1334 tlen = len - toff; 1335 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1336 break; 1337 #endif 1338 default: 1339 m_freem(m); 1340 return; 1341 } 1342 1343 /* 1344 * Enforce alignment requirements that are violated in 1345 * some cases, see kern/50766 for details. 1346 */ 1347 if (TCP_HDR_ALIGNED_P(th) == 0) { 1348 m = m_copyup(m, toff + sizeof(struct tcphdr), 0); 1349 if (m == NULL) { 1350 TCP_STATINC(TCP_STAT_RCVSHORT); 1351 return; 1352 } 1353 ip = mtod(m, struct ip *); 1354 #ifdef INET6 1355 ip6 = mtod(m, struct ip6_hdr *); 1356 #endif 1357 th = (struct tcphdr *)(mtod(m, char *) + toff); 1358 } 1359 KASSERT(TCP_HDR_ALIGNED_P(th)); 1360 1361 /* 1362 * Check that TCP offset makes sense, pull out TCP options and 1363 * adjust length. 1364 */ 1365 off = th->th_off << 2; 1366 if (off < sizeof(struct tcphdr) || off > tlen) { 1367 TCP_STATINC(TCP_STAT_RCVBADOFF); 1368 goto drop; 1369 } 1370 tlen -= off; 1371 1372 if (off > sizeof(struct tcphdr)) { 1373 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1374 if (th == NULL) { 1375 TCP_STATINC(TCP_STAT_RCVSHORT); 1376 return; 1377 } 1378 KASSERT(TCP_HDR_ALIGNED_P(th)); 1379 optlen = off - sizeof (struct tcphdr); 1380 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1381 /* 1382 * Do quick retrieval of timestamp options ("options 1383 * prediction?"). If timestamp is the only option and it's 1384 * formatted as recommended in RFC 1323 appendix A, we 1385 * quickly get the values now and not bother calling 1386 * tcp_dooptions(), etc. 1387 */ 1388 if ((optlen == TCPOLEN_TSTAMP_APPA || 1389 (optlen > TCPOLEN_TSTAMP_APPA && 1390 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1391 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1392 (th->th_flags & TH_SYN) == 0) { 1393 opti.ts_present = 1; 1394 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1395 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1396 optp = NULL; /* we've parsed the options */ 1397 } 1398 } 1399 tiflags = th->th_flags; 1400 1401 /* 1402 * Checksum extended TCP header and data 1403 */ 1404 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1405 goto badcsum; 1406 1407 /* 1408 * Locate pcb for segment. 1409 */ 1410 findpcb: 1411 inp = NULL; 1412 #ifdef INET6 1413 in6p = NULL; 1414 #endif 1415 switch (af) { 1416 #ifdef INET 1417 case AF_INET: 1418 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1419 ip->ip_dst, th->th_dport, &vestige); 1420 if (inp == 0 && !vestige.valid) { 1421 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1422 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, 1423 th->th_dport); 1424 } 1425 #ifdef INET6 1426 if (inp == 0 && !vestige.valid) { 1427 struct in6_addr s, d; 1428 1429 /* mapped addr case */ 1430 in6_in_2_v4mapin6(&ip->ip_src, &s); 1431 in6_in_2_v4mapin6(&ip->ip_dst, &d); 1432 in6p = in6_pcblookup_connect(&tcbtable, &s, 1433 th->th_sport, &d, th->th_dport, 0, &vestige); 1434 if (in6p == 0 && !vestige.valid) { 1435 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1436 in6p = in6_pcblookup_bind(&tcbtable, &d, 1437 th->th_dport, 0); 1438 } 1439 } 1440 #endif 1441 #ifndef INET6 1442 if (inp == 0 && !vestige.valid) 1443 #else 1444 if (inp == 0 && in6p == 0 && !vestige.valid) 1445 #endif 1446 { 1447 TCP_STATINC(TCP_STAT_NOPORT); 1448 if (tcp_log_refused && 1449 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1450 tcp4_log_refused(ip, th); 1451 } 1452 tcp_fields_to_host(th); 1453 goto dropwithreset_ratelim; 1454 } 1455 #if defined(IPSEC) 1456 if (ipsec_used) { 1457 if (inp && 1458 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 1459 && ipsec4_in_reject(m, inp)) { 1460 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1461 goto drop; 1462 } 1463 #ifdef INET6 1464 else if (in6p && 1465 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 1466 && ipsec6_in_reject(m, in6p)) { 1467 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1468 goto drop; 1469 } 1470 #endif 1471 } 1472 #endif /*IPSEC*/ 1473 break; 1474 #endif /*INET*/ 1475 #ifdef INET6 1476 case AF_INET6: 1477 { 1478 int faith; 1479 1480 #if defined(NFAITH) && NFAITH > 0 1481 faith = faithprefix(&ip6->ip6_dst); 1482 #else 1483 faith = 0; 1484 #endif 1485 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1486 th->th_sport, &ip6->ip6_dst, th->th_dport, faith, &vestige); 1487 if (!in6p && !vestige.valid) { 1488 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1489 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1490 th->th_dport, faith); 1491 } 1492 if (!in6p && !vestige.valid) { 1493 TCP_STATINC(TCP_STAT_NOPORT); 1494 if (tcp_log_refused && 1495 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1496 tcp6_log_refused(ip6, th); 1497 } 1498 tcp_fields_to_host(th); 1499 goto dropwithreset_ratelim; 1500 } 1501 #if defined(IPSEC) 1502 if (ipsec_used && in6p 1503 && (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 1504 && ipsec6_in_reject(m, in6p)) { 1505 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1506 goto drop; 1507 } 1508 #endif /*IPSEC*/ 1509 break; 1510 } 1511 #endif 1512 } 1513 1514 /* 1515 * If the state is CLOSED (i.e., TCB does not exist) then 1516 * all data in the incoming segment is discarded. 1517 * If the TCB exists but is in CLOSED state, it is embryonic, 1518 * but should either do a listen or a connect soon. 1519 */ 1520 tp = NULL; 1521 so = NULL; 1522 if (inp) { 1523 /* Check the minimum TTL for socket. */ 1524 if (ip->ip_ttl < inp->inp_ip_minttl) 1525 goto drop; 1526 1527 tp = intotcpcb(inp); 1528 so = inp->inp_socket; 1529 } 1530 #ifdef INET6 1531 else if (in6p) { 1532 tp = in6totcpcb(in6p); 1533 so = in6p->in6p_socket; 1534 } 1535 #endif 1536 else if (vestige.valid) { 1537 int mc = 0; 1538 1539 /* We do not support the resurrection of vtw tcpcps. */ 1540 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1541 goto badcsum; 1542 1543 switch (af) { 1544 #ifdef INET6 1545 case AF_INET6: 1546 mc = IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst); 1547 break; 1548 #endif 1549 1550 case AF_INET: 1551 mc = (IN_MULTICAST(ip->ip_dst.s_addr) || 1552 in_broadcast(ip->ip_dst, m_get_rcvif_NOMPSAFE(m))); 1553 break; 1554 } 1555 1556 tcp_fields_to_host(th); 1557 tcp_vtw_input(th, &vestige, m, tlen, mc); 1558 m = NULL; 1559 goto drop; 1560 } 1561 1562 if (tp == NULL) { 1563 tcp_fields_to_host(th); 1564 goto dropwithreset_ratelim; 1565 } 1566 if (tp->t_state == TCPS_CLOSED) 1567 goto drop; 1568 1569 KASSERT(so->so_lock == softnet_lock); 1570 KASSERT(solocked(so)); 1571 1572 tcp_fields_to_host(th); 1573 1574 /* Unscale the window into a 32-bit value. */ 1575 if ((tiflags & TH_SYN) == 0) 1576 tiwin = th->th_win << tp->snd_scale; 1577 else 1578 tiwin = th->th_win; 1579 1580 #ifdef INET6 1581 /* save packet options if user wanted */ 1582 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1583 if (in6p->in6p_options) { 1584 m_freem(in6p->in6p_options); 1585 in6p->in6p_options = 0; 1586 } 1587 KASSERT(ip6 != NULL); 1588 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1589 } 1590 #endif 1591 1592 if (so->so_options & SO_DEBUG) { 1593 #ifdef TCP_DEBUG 1594 ostate = tp->t_state; 1595 #endif 1596 1597 tcp_saveti = NULL; 1598 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1599 goto nosave; 1600 1601 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1602 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1603 if (!tcp_saveti) 1604 goto nosave; 1605 } else { 1606 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1607 if (!tcp_saveti) 1608 goto nosave; 1609 MCLAIM(m, &tcp_mowner); 1610 tcp_saveti->m_len = iphlen; 1611 m_copydata(m, 0, iphlen, 1612 mtod(tcp_saveti, void *)); 1613 } 1614 1615 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1616 m_freem(tcp_saveti); 1617 tcp_saveti = NULL; 1618 } else { 1619 tcp_saveti->m_len += sizeof(struct tcphdr); 1620 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1621 sizeof(struct tcphdr)); 1622 } 1623 nosave:; 1624 } 1625 1626 if (so->so_options & SO_ACCEPTCONN) { 1627 union syn_cache_sa src; 1628 union syn_cache_sa dst; 1629 1630 memset(&src, 0, sizeof(src)); 1631 memset(&dst, 0, sizeof(dst)); 1632 switch (af) { 1633 #ifdef INET 1634 case AF_INET: 1635 src.sin.sin_len = sizeof(struct sockaddr_in); 1636 src.sin.sin_family = AF_INET; 1637 src.sin.sin_addr = ip->ip_src; 1638 src.sin.sin_port = th->th_sport; 1639 1640 dst.sin.sin_len = sizeof(struct sockaddr_in); 1641 dst.sin.sin_family = AF_INET; 1642 dst.sin.sin_addr = ip->ip_dst; 1643 dst.sin.sin_port = th->th_dport; 1644 break; 1645 #endif 1646 #ifdef INET6 1647 case AF_INET6: 1648 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1649 src.sin6.sin6_family = AF_INET6; 1650 src.sin6.sin6_addr = ip6->ip6_src; 1651 src.sin6.sin6_port = th->th_sport; 1652 1653 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1654 dst.sin6.sin6_family = AF_INET6; 1655 dst.sin6.sin6_addr = ip6->ip6_dst; 1656 dst.sin6.sin6_port = th->th_dport; 1657 break; 1658 #endif /* INET6 */ 1659 default: 1660 goto badsyn; /*sanity*/ 1661 } 1662 1663 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1664 if (tiflags & TH_RST) { 1665 syn_cache_reset(&src.sa, &dst.sa, th); 1666 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1667 (TH_ACK|TH_SYN)) { 1668 /* 1669 * Received a SYN,ACK. This should 1670 * never happen while we are in 1671 * LISTEN. Send an RST. 1672 */ 1673 goto badsyn; 1674 } else if (tiflags & TH_ACK) { 1675 so = syn_cache_get(&src.sa, &dst.sa, 1676 th, toff, tlen, so, m); 1677 if (so == NULL) { 1678 /* 1679 * We don't have a SYN for this ACK; 1680 * send an RST. 1681 */ 1682 goto badsyn; 1683 } else if (so == (struct socket *)(-1)) { 1684 /* 1685 * We were unable to create the 1686 * connection. If the 3-way handshake 1687 * was completed, and RST has been 1688 * sent to the peer. Since the mbuf 1689 * might be in use for the reply, do 1690 * not free it. 1691 */ 1692 m = NULL; 1693 } else { 1694 /* 1695 * We have created a full-blown 1696 * connection. 1697 */ 1698 tp = NULL; 1699 inp = NULL; 1700 #ifdef INET6 1701 in6p = NULL; 1702 #endif 1703 switch (so->so_proto->pr_domain->dom_family) { 1704 #ifdef INET 1705 case AF_INET: 1706 inp = sotoinpcb(so); 1707 tp = intotcpcb(inp); 1708 break; 1709 #endif 1710 #ifdef INET6 1711 case AF_INET6: 1712 in6p = sotoin6pcb(so); 1713 tp = in6totcpcb(in6p); 1714 break; 1715 #endif 1716 } 1717 if (tp == NULL) 1718 goto badsyn; /*XXX*/ 1719 tiwin <<= tp->snd_scale; 1720 goto after_listen; 1721 } 1722 } else { 1723 /* 1724 * None of RST, SYN or ACK was set. 1725 * This is an invalid packet for a 1726 * TCB in LISTEN state. Send a RST. 1727 */ 1728 goto badsyn; 1729 } 1730 } else { 1731 /* 1732 * Received a SYN. 1733 */ 1734 1735 switch (af) { 1736 #ifdef INET6 1737 case AF_INET6: 1738 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1739 goto drop; 1740 break; 1741 #endif /* INET6 */ 1742 case AF_INET: 1743 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1744 in_broadcast(ip->ip_dst, 1745 m_get_rcvif_NOMPSAFE(m))) 1746 goto drop; 1747 break; 1748 } 1749 1750 #ifdef INET6 1751 /* 1752 * If deprecated address is forbidden, we do 1753 * not accept SYN to deprecated interface 1754 * address to prevent any new inbound 1755 * connection from getting established. 1756 * When we do not accept SYN, we send a TCP 1757 * RST, with deprecated source address (instead 1758 * of dropping it). We compromise it as it is 1759 * much better for peer to send a RST, and 1760 * RST will be the final packet for the 1761 * exchange. 1762 * 1763 * If we do not forbid deprecated addresses, we 1764 * accept the SYN packet. RFC2462 does not 1765 * suggest dropping SYN in this case. 1766 * If we decipher RFC2462 5.5.4, it says like 1767 * this: 1768 * 1. use of deprecated addr with existing 1769 * communication is okay - "SHOULD continue 1770 * to be used" 1771 * 2. use of it with new communication: 1772 * (2a) "SHOULD NOT be used if alternate 1773 * address with sufficient scope is 1774 * available" 1775 * (2b) nothing mentioned otherwise. 1776 * Here we fall into (2b) case as we have no 1777 * choice in our source address selection - we 1778 * must obey the peer. 1779 * 1780 * The wording in RFC2462 is confusing, and 1781 * there are multiple description text for 1782 * deprecated address handling - worse, they 1783 * are not exactly the same. I believe 5.5.4 1784 * is the best one, so we follow 5.5.4. 1785 */ 1786 if (af == AF_INET6 && !ip6_use_deprecated) { 1787 struct in6_ifaddr *ia6; 1788 int s; 1789 struct ifnet *rcvif = m_get_rcvif(m, &s); 1790 if (rcvif == NULL) 1791 goto dropwithreset; /* XXX */ 1792 if ((ia6 = in6ifa_ifpwithaddr(rcvif, 1793 &ip6->ip6_dst)) && 1794 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1795 tp = NULL; 1796 m_put_rcvif(rcvif, &s); 1797 goto dropwithreset; 1798 } 1799 m_put_rcvif(rcvif, &s); 1800 } 1801 #endif 1802 1803 #if defined(IPSEC) 1804 if (ipsec_used) { 1805 switch (af) { 1806 #ifdef INET 1807 case AF_INET: 1808 /* 1809 * inp can be NULL when receiving an 1810 * IPv4 packet on an IPv4-mapped IPv6 1811 * address. 1812 */ 1813 KASSERT(inp == NULL || 1814 sotoinpcb(so) == inp); 1815 if (!ipsec4_in_reject(m, inp)) 1816 break; 1817 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1818 tp = NULL; 1819 goto dropwithreset; 1820 #endif 1821 #ifdef INET6 1822 case AF_INET6: 1823 KASSERT(sotoin6pcb(so) == in6p); 1824 if (!ipsec6_in_reject(m, in6p)) 1825 break; 1826 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1827 tp = NULL; 1828 goto dropwithreset; 1829 #endif /*INET6*/ 1830 } 1831 } 1832 #endif /*IPSEC*/ 1833 1834 /* 1835 * LISTEN socket received a SYN from itself? This 1836 * can't possibly be valid; drop the packet. 1837 */ 1838 if (th->th_sport == th->th_dport) { 1839 int i; 1840 1841 switch (af) { 1842 #ifdef INET 1843 case AF_INET: 1844 i = in_hosteq(ip->ip_src, ip->ip_dst); 1845 break; 1846 #endif 1847 #ifdef INET6 1848 case AF_INET6: 1849 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, 1850 &ip6->ip6_dst); 1851 break; 1852 #endif 1853 default: 1854 i = 1; 1855 } 1856 if (i) { 1857 TCP_STATINC(TCP_STAT_BADSYN); 1858 goto drop; 1859 } 1860 } 1861 1862 /* 1863 * SYN looks ok; create compressed TCP 1864 * state for it. 1865 */ 1866 if (so->so_qlen <= so->so_qlimit && 1867 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1868 so, m, optp, optlen, &opti)) 1869 m = NULL; 1870 } 1871 1872 goto drop; 1873 } 1874 1875 after_listen: 1876 /* 1877 * Should not happen now that all embryonic connections 1878 * are handled with compressed state. 1879 */ 1880 KASSERT(tp->t_state != TCPS_LISTEN); 1881 1882 /* 1883 * Segment received on connection. 1884 * Reset idle time and keep-alive timer. 1885 */ 1886 tp->t_rcvtime = tcp_now; 1887 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1888 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1889 1890 /* 1891 * Process options. 1892 */ 1893 #ifdef TCP_SIGNATURE 1894 if (optp || (tp->t_flags & TF_SIGNATURE)) 1895 #else 1896 if (optp) 1897 #endif 1898 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1899 goto drop; 1900 1901 if (TCP_SACK_ENABLED(tp)) { 1902 tcp_del_sackholes(tp, th); 1903 } 1904 1905 if (TCP_ECN_ALLOWED(tp)) { 1906 if (tiflags & TH_CWR) { 1907 tp->t_flags &= ~TF_ECN_SND_ECE; 1908 } 1909 switch (iptos & IPTOS_ECN_MASK) { 1910 case IPTOS_ECN_CE: 1911 tp->t_flags |= TF_ECN_SND_ECE; 1912 TCP_STATINC(TCP_STAT_ECN_CE); 1913 break; 1914 case IPTOS_ECN_ECT0: 1915 TCP_STATINC(TCP_STAT_ECN_ECT); 1916 break; 1917 case IPTOS_ECN_ECT1: 1918 /* XXX: ignore for now -- rpaulo */ 1919 break; 1920 } 1921 /* 1922 * Congestion experienced. 1923 * Ignore if we are already trying to recover. 1924 */ 1925 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1926 tp->t_congctl->cong_exp(tp); 1927 } 1928 1929 if (opti.ts_present && opti.ts_ecr) { 1930 /* 1931 * Calculate the RTT from the returned time stamp and the 1932 * connection's time base. If the time stamp is later than 1933 * the current time, or is extremely old, fall back to non-1323 1934 * RTT calculation. Since ts_rtt is unsigned, we can test both 1935 * at the same time. 1936 * 1937 * Note that ts_rtt is in units of slow ticks (500 1938 * ms). Since most earthbound RTTs are < 500 ms, 1939 * observed values will have large quantization noise. 1940 * Our smoothed RTT is then the fraction of observed 1941 * samples that are 1 tick instead of 0 (times 500 1942 * ms). 1943 * 1944 * ts_rtt is increased by 1 to denote a valid sample, 1945 * with 0 indicating an invalid measurement. This 1946 * extra 1 must be removed when ts_rtt is used, or 1947 * else an an erroneous extra 500 ms will result. 1948 */ 1949 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1950 if (ts_rtt > TCP_PAWS_IDLE) 1951 ts_rtt = 0; 1952 } else { 1953 ts_rtt = 0; 1954 } 1955 1956 /* 1957 * Header prediction: check for the two common cases 1958 * of a uni-directional data xfer. If the packet has 1959 * no control flags, is in-sequence, the window didn't 1960 * change and we're not retransmitting, it's a 1961 * candidate. If the length is zero and the ack moved 1962 * forward, we're the sender side of the xfer. Just 1963 * free the data acked & wake any higher level process 1964 * that was blocked waiting for space. If the length 1965 * is non-zero and the ack didn't move, we're the 1966 * receiver side. If we're getting packets in-order 1967 * (the reassembly queue is empty), add the data to 1968 * the socket buffer and note that we need a delayed ack. 1969 */ 1970 if (tp->t_state == TCPS_ESTABLISHED && 1971 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1972 == TH_ACK && 1973 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1974 th->th_seq == tp->rcv_nxt && 1975 tiwin && tiwin == tp->snd_wnd && 1976 tp->snd_nxt == tp->snd_max) { 1977 1978 /* 1979 * If last ACK falls within this segment's sequence numbers, 1980 * record the timestamp. 1981 * NOTE that the test is modified according to the latest 1982 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 1983 * 1984 * note that we already know 1985 * TSTMP_GEQ(opti.ts_val, tp->ts_recent) 1986 */ 1987 if (opti.ts_present && 1988 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1989 tp->ts_recent_age = tcp_now; 1990 tp->ts_recent = opti.ts_val; 1991 } 1992 1993 if (tlen == 0) { 1994 /* Ack prediction. */ 1995 if (SEQ_GT(th->th_ack, tp->snd_una) && 1996 SEQ_LEQ(th->th_ack, tp->snd_max) && 1997 tp->snd_cwnd >= tp->snd_wnd && 1998 tp->t_partialacks < 0) { 1999 /* 2000 * this is a pure ack for outstanding data. 2001 */ 2002 if (ts_rtt) 2003 tcp_xmit_timer(tp, ts_rtt - 1); 2004 else if (tp->t_rtttime && 2005 SEQ_GT(th->th_ack, tp->t_rtseq)) 2006 tcp_xmit_timer(tp, 2007 tcp_now - tp->t_rtttime); 2008 acked = th->th_ack - tp->snd_una; 2009 tcps = TCP_STAT_GETREF(); 2010 tcps[TCP_STAT_PREDACK]++; 2011 tcps[TCP_STAT_RCVACKPACK]++; 2012 tcps[TCP_STAT_RCVACKBYTE] += acked; 2013 TCP_STAT_PUTREF(); 2014 nd6_hint(tp); 2015 2016 if (acked > (tp->t_lastoff - tp->t_inoff)) 2017 tp->t_lastm = NULL; 2018 sbdrop(&so->so_snd, acked); 2019 tp->t_lastoff -= acked; 2020 2021 icmp_check(tp, th, acked); 2022 2023 tp->snd_una = th->th_ack; 2024 tp->snd_fack = tp->snd_una; 2025 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2026 tp->snd_high = tp->snd_una; 2027 m_freem(m); 2028 2029 /* 2030 * If all outstanding data are acked, stop 2031 * retransmit timer, otherwise restart timer 2032 * using current (possibly backed-off) value. 2033 * If process is waiting for space, 2034 * wakeup/selnotify/signal. If data 2035 * are ready to send, let tcp_output 2036 * decide between more output or persist. 2037 */ 2038 if (tp->snd_una == tp->snd_max) 2039 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2040 else if (TCP_TIMER_ISARMED(tp, 2041 TCPT_PERSIST) == 0) 2042 TCP_TIMER_ARM(tp, TCPT_REXMT, 2043 tp->t_rxtcur); 2044 2045 sowwakeup(so); 2046 if (so->so_snd.sb_cc) { 2047 KERNEL_LOCK(1, NULL); 2048 (void) tcp_output(tp); 2049 KERNEL_UNLOCK_ONE(NULL); 2050 } 2051 if (tcp_saveti) 2052 m_freem(tcp_saveti); 2053 return; 2054 } 2055 } else if (th->th_ack == tp->snd_una && 2056 TAILQ_FIRST(&tp->segq) == NULL && 2057 tlen <= sbspace(&so->so_rcv)) { 2058 int newsize = 0; /* automatic sockbuf scaling */ 2059 2060 /* 2061 * this is a pure, in-sequence data packet 2062 * with nothing on the reassembly queue and 2063 * we have enough buffer space to take it. 2064 */ 2065 tp->rcv_nxt += tlen; 2066 tcps = TCP_STAT_GETREF(); 2067 tcps[TCP_STAT_PREDDAT]++; 2068 tcps[TCP_STAT_RCVPACK]++; 2069 tcps[TCP_STAT_RCVBYTE] += tlen; 2070 TCP_STAT_PUTREF(); 2071 nd6_hint(tp); 2072 2073 /* 2074 * Automatic sizing enables the performance of large buffers 2075 * and most of the efficiency of small ones by only allocating 2076 * space when it is needed. 2077 * 2078 * On the receive side the socket buffer memory is only rarely 2079 * used to any significant extent. This allows us to be much 2080 * more aggressive in scaling the receive socket buffer. For 2081 * the case that the buffer space is actually used to a large 2082 * extent and we run out of kernel memory we can simply drop 2083 * the new segments; TCP on the sender will just retransmit it 2084 * later. Setting the buffer size too big may only consume too 2085 * much kernel memory if the application doesn't read() from 2086 * the socket or packet loss or reordering makes use of the 2087 * reassembly queue. 2088 * 2089 * The criteria to step up the receive buffer one notch are: 2090 * 1. the number of bytes received during the time it takes 2091 * one timestamp to be reflected back to us (the RTT); 2092 * 2. received bytes per RTT is within seven eighth of the 2093 * current socket buffer size; 2094 * 3. receive buffer size has not hit maximal automatic size; 2095 * 2096 * This algorithm does one step per RTT at most and only if 2097 * we receive a bulk stream w/o packet losses or reorderings. 2098 * Shrinking the buffer during idle times is not necessary as 2099 * it doesn't consume any memory when idle. 2100 * 2101 * TODO: Only step up if the application is actually serving 2102 * the buffer to better manage the socket buffer resources. 2103 */ 2104 if (tcp_do_autorcvbuf && 2105 opti.ts_ecr && 2106 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 2107 if (opti.ts_ecr > tp->rfbuf_ts && 2108 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 2109 if (tp->rfbuf_cnt > 2110 (so->so_rcv.sb_hiwat / 8 * 7) && 2111 so->so_rcv.sb_hiwat < 2112 tcp_autorcvbuf_max) { 2113 newsize = 2114 min(so->so_rcv.sb_hiwat + 2115 tcp_autorcvbuf_inc, 2116 tcp_autorcvbuf_max); 2117 } 2118 /* Start over with next RTT. */ 2119 tp->rfbuf_ts = 0; 2120 tp->rfbuf_cnt = 0; 2121 } else 2122 tp->rfbuf_cnt += tlen; /* add up */ 2123 } 2124 2125 /* 2126 * Drop TCP, IP headers and TCP options then add data 2127 * to socket buffer. 2128 */ 2129 if (so->so_state & SS_CANTRCVMORE) 2130 m_freem(m); 2131 else { 2132 /* 2133 * Set new socket buffer size. 2134 * Give up when limit is reached. 2135 */ 2136 if (newsize) 2137 if (!sbreserve(&so->so_rcv, 2138 newsize, so)) 2139 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 2140 m_adj(m, toff + off); 2141 sbappendstream(&so->so_rcv, m); 2142 } 2143 sorwakeup(so); 2144 tcp_setup_ack(tp, th); 2145 if (tp->t_flags & TF_ACKNOW) { 2146 KERNEL_LOCK(1, NULL); 2147 (void) tcp_output(tp); 2148 KERNEL_UNLOCK_ONE(NULL); 2149 } 2150 if (tcp_saveti) 2151 m_freem(tcp_saveti); 2152 return; 2153 } 2154 } 2155 2156 /* 2157 * Compute mbuf offset to TCP data segment. 2158 */ 2159 hdroptlen = toff + off; 2160 2161 /* 2162 * Calculate amount of space in receive window, 2163 * and then do TCP input processing. 2164 * Receive window is amount of space in rcv queue, 2165 * but not less than advertised window. 2166 */ 2167 { int win; 2168 2169 win = sbspace(&so->so_rcv); 2170 if (win < 0) 2171 win = 0; 2172 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 2173 } 2174 2175 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 2176 tp->rfbuf_ts = 0; 2177 tp->rfbuf_cnt = 0; 2178 2179 switch (tp->t_state) { 2180 /* 2181 * If the state is SYN_SENT: 2182 * if seg contains an ACK, but not for our SYN, drop the input. 2183 * if seg contains a RST, then drop the connection. 2184 * if seg does not contain SYN, then drop it. 2185 * Otherwise this is an acceptable SYN segment 2186 * initialize tp->rcv_nxt and tp->irs 2187 * if seg contains ack then advance tp->snd_una 2188 * if seg contains a ECE and ECN support is enabled, the stream 2189 * is ECN capable. 2190 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2191 * arrange for segment to be acked (eventually) 2192 * continue processing rest of data/controls, beginning with URG 2193 */ 2194 case TCPS_SYN_SENT: 2195 if ((tiflags & TH_ACK) && 2196 (SEQ_LEQ(th->th_ack, tp->iss) || 2197 SEQ_GT(th->th_ack, tp->snd_max))) 2198 goto dropwithreset; 2199 if (tiflags & TH_RST) { 2200 if (tiflags & TH_ACK) 2201 tp = tcp_drop(tp, ECONNREFUSED); 2202 goto drop; 2203 } 2204 if ((tiflags & TH_SYN) == 0) 2205 goto drop; 2206 if (tiflags & TH_ACK) { 2207 tp->snd_una = th->th_ack; 2208 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2209 tp->snd_nxt = tp->snd_una; 2210 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2211 tp->snd_high = tp->snd_una; 2212 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2213 2214 if ((tiflags & TH_ECE) && tcp_do_ecn) { 2215 tp->t_flags |= TF_ECN_PERMIT; 2216 TCP_STATINC(TCP_STAT_ECN_SHS); 2217 } 2218 2219 } 2220 tp->irs = th->th_seq; 2221 tcp_rcvseqinit(tp); 2222 tp->t_flags |= TF_ACKNOW; 2223 tcp_mss_from_peer(tp, opti.maxseg); 2224 2225 /* 2226 * Initialize the initial congestion window. If we 2227 * had to retransmit the SYN, we must initialize cwnd 2228 * to 1 segment (i.e. the Loss Window). 2229 */ 2230 if (tp->t_flags & TF_SYN_REXMT) 2231 tp->snd_cwnd = tp->t_peermss; 2232 else { 2233 int ss = tcp_init_win; 2234 #ifdef INET 2235 if (inp != NULL && in_localaddr(inp->inp_faddr)) 2236 ss = tcp_init_win_local; 2237 #endif 2238 #ifdef INET6 2239 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 2240 ss = tcp_init_win_local; 2241 #endif 2242 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 2243 } 2244 2245 tcp_rmx_rtt(tp); 2246 if (tiflags & TH_ACK) { 2247 TCP_STATINC(TCP_STAT_CONNECTS); 2248 /* 2249 * move tcp_established before soisconnected 2250 * because upcall handler can drive tcp_output 2251 * functionality. 2252 * XXX we might call soisconnected at the end of 2253 * all processing 2254 */ 2255 tcp_established(tp); 2256 soisconnected(so); 2257 /* Do window scaling on this connection? */ 2258 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2259 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2260 tp->snd_scale = tp->requested_s_scale; 2261 tp->rcv_scale = tp->request_r_scale; 2262 } 2263 TCP_REASS_LOCK(tp); 2264 (void) tcp_reass(tp, NULL, NULL, &tlen); 2265 /* 2266 * if we didn't have to retransmit the SYN, 2267 * use its rtt as our initial srtt & rtt var. 2268 */ 2269 if (tp->t_rtttime) 2270 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2271 } else 2272 tp->t_state = TCPS_SYN_RECEIVED; 2273 2274 /* 2275 * Advance th->th_seq to correspond to first data byte. 2276 * If data, trim to stay within window, 2277 * dropping FIN if necessary. 2278 */ 2279 th->th_seq++; 2280 if (tlen > tp->rcv_wnd) { 2281 todrop = tlen - tp->rcv_wnd; 2282 m_adj(m, -todrop); 2283 tlen = tp->rcv_wnd; 2284 tiflags &= ~TH_FIN; 2285 tcps = TCP_STAT_GETREF(); 2286 tcps[TCP_STAT_RCVPACKAFTERWIN]++; 2287 tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop; 2288 TCP_STAT_PUTREF(); 2289 } 2290 tp->snd_wl1 = th->th_seq - 1; 2291 tp->rcv_up = th->th_seq; 2292 goto step6; 2293 2294 /* 2295 * If the state is SYN_RECEIVED: 2296 * If seg contains an ACK, but not for our SYN, drop the input 2297 * and generate an RST. See page 36, rfc793 2298 */ 2299 case TCPS_SYN_RECEIVED: 2300 if ((tiflags & TH_ACK) && 2301 (SEQ_LEQ(th->th_ack, tp->iss) || 2302 SEQ_GT(th->th_ack, tp->snd_max))) 2303 goto dropwithreset; 2304 break; 2305 } 2306 2307 /* 2308 * States other than LISTEN or SYN_SENT. 2309 * First check timestamp, if present. 2310 * Then check that at least some bytes of segment are within 2311 * receive window. If segment begins before rcv_nxt, 2312 * drop leading data (and SYN); if nothing left, just ack. 2313 * 2314 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2315 * and it's less than ts_recent, drop it. 2316 */ 2317 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2318 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2319 2320 /* Check to see if ts_recent is over 24 days old. */ 2321 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2322 /* 2323 * Invalidate ts_recent. If this segment updates 2324 * ts_recent, the age will be reset later and ts_recent 2325 * will get a valid value. If it does not, setting 2326 * ts_recent to zero will at least satisfy the 2327 * requirement that zero be placed in the timestamp 2328 * echo reply when ts_recent isn't valid. The 2329 * age isn't reset until we get a valid ts_recent 2330 * because we don't want out-of-order segments to be 2331 * dropped when ts_recent is old. 2332 */ 2333 tp->ts_recent = 0; 2334 } else { 2335 tcps = TCP_STAT_GETREF(); 2336 tcps[TCP_STAT_RCVDUPPACK]++; 2337 tcps[TCP_STAT_RCVDUPBYTE] += tlen; 2338 tcps[TCP_STAT_PAWSDROP]++; 2339 TCP_STAT_PUTREF(); 2340 tcp_new_dsack(tp, th->th_seq, tlen); 2341 goto dropafterack; 2342 } 2343 } 2344 2345 todrop = tp->rcv_nxt - th->th_seq; 2346 dupseg = false; 2347 if (todrop > 0) { 2348 if (tiflags & TH_SYN) { 2349 tiflags &= ~TH_SYN; 2350 th->th_seq++; 2351 if (th->th_urp > 1) 2352 th->th_urp--; 2353 else { 2354 tiflags &= ~TH_URG; 2355 th->th_urp = 0; 2356 } 2357 todrop--; 2358 } 2359 if (todrop > tlen || 2360 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2361 /* 2362 * Any valid FIN or RST must be to the left of the 2363 * window. At this point the FIN or RST must be a 2364 * duplicate or out of sequence; drop it. 2365 */ 2366 if (tiflags & TH_RST) 2367 goto drop; 2368 tiflags &= ~(TH_FIN|TH_RST); 2369 /* 2370 * Send an ACK to resynchronize and drop any data. 2371 * But keep on processing for RST or ACK. 2372 */ 2373 tp->t_flags |= TF_ACKNOW; 2374 todrop = tlen; 2375 dupseg = true; 2376 tcps = TCP_STAT_GETREF(); 2377 tcps[TCP_STAT_RCVDUPPACK]++; 2378 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 2379 TCP_STAT_PUTREF(); 2380 } else if ((tiflags & TH_RST) && 2381 th->th_seq != tp->rcv_nxt) { 2382 /* 2383 * Test for reset before adjusting the sequence 2384 * number for overlapping data. 2385 */ 2386 goto dropafterack_ratelim; 2387 } else { 2388 tcps = TCP_STAT_GETREF(); 2389 tcps[TCP_STAT_RCVPARTDUPPACK]++; 2390 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 2391 TCP_STAT_PUTREF(); 2392 } 2393 tcp_new_dsack(tp, th->th_seq, todrop); 2394 hdroptlen += todrop; /*drop from head afterwards*/ 2395 th->th_seq += todrop; 2396 tlen -= todrop; 2397 if (th->th_urp > todrop) 2398 th->th_urp -= todrop; 2399 else { 2400 tiflags &= ~TH_URG; 2401 th->th_urp = 0; 2402 } 2403 } 2404 2405 /* 2406 * If new data are received on a connection after the 2407 * user processes are gone, then RST the other end. 2408 */ 2409 if ((so->so_state & SS_NOFDREF) && 2410 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2411 tp = tcp_close(tp); 2412 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 2413 goto dropwithreset; 2414 } 2415 2416 /* 2417 * If segment ends after window, drop trailing data 2418 * (and PUSH and FIN); if nothing left, just ACK. 2419 */ 2420 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 2421 if (todrop > 0) { 2422 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 2423 if (todrop >= tlen) { 2424 /* 2425 * The segment actually starts after the window. 2426 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2427 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2428 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2429 */ 2430 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 2431 /* 2432 * If a new connection request is received 2433 * while in TIME_WAIT, drop the old connection 2434 * and start over if the sequence numbers 2435 * are above the previous ones. 2436 * 2437 * NOTE: We will checksum the packet again, and 2438 * so we need to put the header fields back into 2439 * network order! 2440 * XXX This kind of sucks, but we don't expect 2441 * XXX this to happen very often, so maybe it 2442 * XXX doesn't matter so much. 2443 */ 2444 if (tiflags & TH_SYN && 2445 tp->t_state == TCPS_TIME_WAIT && 2446 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2447 tp = tcp_close(tp); 2448 tcp_fields_to_net(th); 2449 goto findpcb; 2450 } 2451 /* 2452 * If window is closed can only take segments at 2453 * window edge, and have to drop data and PUSH from 2454 * incoming segments. Continue processing, but 2455 * remember to ack. Otherwise, drop segment 2456 * and (if not RST) ack. 2457 */ 2458 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2459 tp->t_flags |= TF_ACKNOW; 2460 TCP_STATINC(TCP_STAT_RCVWINPROBE); 2461 } else 2462 goto dropafterack; 2463 } else 2464 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 2465 m_adj(m, -todrop); 2466 tlen -= todrop; 2467 tiflags &= ~(TH_PUSH|TH_FIN); 2468 } 2469 2470 /* 2471 * If last ACK falls within this segment's sequence numbers, 2472 * record the timestamp. 2473 * NOTE: 2474 * 1) That the test incorporates suggestions from the latest 2475 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 2476 * 2) That updating only on newer timestamps interferes with 2477 * our earlier PAWS tests, so this check should be solely 2478 * predicated on the sequence space of this segment. 2479 * 3) That we modify the segment boundary check to be 2480 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2481 * instead of RFC1323's 2482 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2483 * This modified check allows us to overcome RFC1323's 2484 * limitations as described in Stevens TCP/IP Illustrated 2485 * Vol. 2 p.869. In such cases, we can still calculate the 2486 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2487 */ 2488 if (opti.ts_present && 2489 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2490 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2491 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2492 tp->ts_recent_age = tcp_now; 2493 tp->ts_recent = opti.ts_val; 2494 } 2495 2496 /* 2497 * If the RST bit is set examine the state: 2498 * SYN_RECEIVED STATE: 2499 * If passive open, return to LISTEN state. 2500 * If active open, inform user that connection was refused. 2501 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 2502 * Inform user that connection was reset, and close tcb. 2503 * CLOSING, LAST_ACK, TIME_WAIT STATES 2504 * Close the tcb. 2505 */ 2506 if (tiflags & TH_RST) { 2507 if (th->th_seq != tp->rcv_nxt) 2508 goto dropafterack_ratelim; 2509 2510 switch (tp->t_state) { 2511 case TCPS_SYN_RECEIVED: 2512 so->so_error = ECONNREFUSED; 2513 goto close; 2514 2515 case TCPS_ESTABLISHED: 2516 case TCPS_FIN_WAIT_1: 2517 case TCPS_FIN_WAIT_2: 2518 case TCPS_CLOSE_WAIT: 2519 so->so_error = ECONNRESET; 2520 close: 2521 tp->t_state = TCPS_CLOSED; 2522 TCP_STATINC(TCP_STAT_DROPS); 2523 tp = tcp_close(tp); 2524 goto drop; 2525 2526 case TCPS_CLOSING: 2527 case TCPS_LAST_ACK: 2528 case TCPS_TIME_WAIT: 2529 tp = tcp_close(tp); 2530 goto drop; 2531 } 2532 } 2533 2534 /* 2535 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2536 * we must be in a synchronized state. RFC791 states (under RST 2537 * generation) that any unacceptable segment (an out-of-order SYN 2538 * qualifies) received in a synchronized state must elicit only an 2539 * empty acknowledgment segment ... and the connection remains in 2540 * the same state. 2541 */ 2542 if (tiflags & TH_SYN) { 2543 if (tp->rcv_nxt == th->th_seq) { 2544 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2545 TH_ACK); 2546 if (tcp_saveti) 2547 m_freem(tcp_saveti); 2548 return; 2549 } 2550 2551 goto dropafterack_ratelim; 2552 } 2553 2554 /* 2555 * If the ACK bit is off we drop the segment and return. 2556 */ 2557 if ((tiflags & TH_ACK) == 0) { 2558 if (tp->t_flags & TF_ACKNOW) 2559 goto dropafterack; 2560 else 2561 goto drop; 2562 } 2563 2564 /* 2565 * Ack processing. 2566 */ 2567 switch (tp->t_state) { 2568 2569 /* 2570 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2571 * ESTABLISHED state and continue processing, otherwise 2572 * send an RST. 2573 */ 2574 case TCPS_SYN_RECEIVED: 2575 if (SEQ_GT(tp->snd_una, th->th_ack) || 2576 SEQ_GT(th->th_ack, tp->snd_max)) 2577 goto dropwithreset; 2578 TCP_STATINC(TCP_STAT_CONNECTS); 2579 soisconnected(so); 2580 tcp_established(tp); 2581 /* Do window scaling? */ 2582 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2583 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2584 tp->snd_scale = tp->requested_s_scale; 2585 tp->rcv_scale = tp->request_r_scale; 2586 } 2587 TCP_REASS_LOCK(tp); 2588 (void) tcp_reass(tp, NULL, NULL, &tlen); 2589 tp->snd_wl1 = th->th_seq - 1; 2590 /* fall into ... */ 2591 2592 /* 2593 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2594 * ACKs. If the ack is in the range 2595 * tp->snd_una < th->th_ack <= tp->snd_max 2596 * then advance tp->snd_una to th->th_ack and drop 2597 * data from the retransmission queue. If this ACK reflects 2598 * more up to date window information we update our window information. 2599 */ 2600 case TCPS_ESTABLISHED: 2601 case TCPS_FIN_WAIT_1: 2602 case TCPS_FIN_WAIT_2: 2603 case TCPS_CLOSE_WAIT: 2604 case TCPS_CLOSING: 2605 case TCPS_LAST_ACK: 2606 case TCPS_TIME_WAIT: 2607 2608 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2609 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2610 TCP_STATINC(TCP_STAT_RCVDUPACK); 2611 /* 2612 * If we have outstanding data (other than 2613 * a window probe), this is a completely 2614 * duplicate ack (ie, window info didn't 2615 * change), the ack is the biggest we've 2616 * seen and we've seen exactly our rexmt 2617 * threshhold of them, assume a packet 2618 * has been dropped and retransmit it. 2619 * Kludge snd_nxt & the congestion 2620 * window so we send only this one 2621 * packet. 2622 */ 2623 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2624 th->th_ack != tp->snd_una) 2625 tp->t_dupacks = 0; 2626 else if (tp->t_partialacks < 0 && 2627 (++tp->t_dupacks == tcprexmtthresh || 2628 TCP_FACK_FASTRECOV(tp))) { 2629 /* 2630 * Do the fast retransmit, and adjust 2631 * congestion control paramenters. 2632 */ 2633 if (tp->t_congctl->fast_retransmit(tp, th)) { 2634 /* False fast retransmit */ 2635 break; 2636 } else 2637 goto drop; 2638 } else if (tp->t_dupacks > tcprexmtthresh) { 2639 tp->snd_cwnd += tp->t_segsz; 2640 KERNEL_LOCK(1, NULL); 2641 (void) tcp_output(tp); 2642 KERNEL_UNLOCK_ONE(NULL); 2643 goto drop; 2644 } 2645 } else { 2646 /* 2647 * If the ack appears to be very old, only 2648 * allow data that is in-sequence. This 2649 * makes it somewhat more difficult to insert 2650 * forged data by guessing sequence numbers. 2651 * Sent an ack to try to update the send 2652 * sequence number on the other side. 2653 */ 2654 if (tlen && th->th_seq != tp->rcv_nxt && 2655 SEQ_LT(th->th_ack, 2656 tp->snd_una - tp->max_sndwnd)) 2657 goto dropafterack; 2658 } 2659 break; 2660 } 2661 /* 2662 * If the congestion window was inflated to account 2663 * for the other side's cached packets, retract it. 2664 */ 2665 tp->t_congctl->fast_retransmit_newack(tp, th); 2666 2667 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2668 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2669 goto dropafterack; 2670 } 2671 acked = th->th_ack - tp->snd_una; 2672 tcps = TCP_STAT_GETREF(); 2673 tcps[TCP_STAT_RCVACKPACK]++; 2674 tcps[TCP_STAT_RCVACKBYTE] += acked; 2675 TCP_STAT_PUTREF(); 2676 2677 /* 2678 * If we have a timestamp reply, update smoothed 2679 * round trip time. If no timestamp is present but 2680 * transmit timer is running and timed sequence 2681 * number was acked, update smoothed round trip time. 2682 * Since we now have an rtt measurement, cancel the 2683 * timer backoff (cf., Phil Karn's retransmit alg.). 2684 * Recompute the initial retransmit timer. 2685 */ 2686 if (ts_rtt) 2687 tcp_xmit_timer(tp, ts_rtt - 1); 2688 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2689 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2690 2691 /* 2692 * If all outstanding data is acked, stop retransmit 2693 * timer and remember to restart (more output or persist). 2694 * If there is more data to be acked, restart retransmit 2695 * timer, using current (possibly backed-off) value. 2696 */ 2697 if (th->th_ack == tp->snd_max) { 2698 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2699 needoutput = 1; 2700 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2701 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2702 2703 /* 2704 * New data has been acked, adjust the congestion window. 2705 */ 2706 tp->t_congctl->newack(tp, th); 2707 2708 nd6_hint(tp); 2709 if (acked > so->so_snd.sb_cc) { 2710 tp->snd_wnd -= so->so_snd.sb_cc; 2711 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2712 ourfinisacked = 1; 2713 } else { 2714 if (acked > (tp->t_lastoff - tp->t_inoff)) 2715 tp->t_lastm = NULL; 2716 sbdrop(&so->so_snd, acked); 2717 tp->t_lastoff -= acked; 2718 if (tp->snd_wnd > acked) 2719 tp->snd_wnd -= acked; 2720 else 2721 tp->snd_wnd = 0; 2722 ourfinisacked = 0; 2723 } 2724 sowwakeup(so); 2725 2726 icmp_check(tp, th, acked); 2727 2728 tp->snd_una = th->th_ack; 2729 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2730 tp->snd_fack = tp->snd_una; 2731 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2732 tp->snd_nxt = tp->snd_una; 2733 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2734 tp->snd_high = tp->snd_una; 2735 2736 switch (tp->t_state) { 2737 2738 /* 2739 * In FIN_WAIT_1 STATE in addition to the processing 2740 * for the ESTABLISHED state if our FIN is now acknowledged 2741 * then enter FIN_WAIT_2. 2742 */ 2743 case TCPS_FIN_WAIT_1: 2744 if (ourfinisacked) { 2745 /* 2746 * If we can't receive any more 2747 * data, then closing user can proceed. 2748 * Starting the timer is contrary to the 2749 * specification, but if we don't get a FIN 2750 * we'll hang forever. 2751 */ 2752 if (so->so_state & SS_CANTRCVMORE) { 2753 soisdisconnected(so); 2754 if (tp->t_maxidle > 0) 2755 TCP_TIMER_ARM(tp, TCPT_2MSL, 2756 tp->t_maxidle); 2757 } 2758 tp->t_state = TCPS_FIN_WAIT_2; 2759 } 2760 break; 2761 2762 /* 2763 * In CLOSING STATE in addition to the processing for 2764 * the ESTABLISHED state if the ACK acknowledges our FIN 2765 * then enter the TIME-WAIT state, otherwise ignore 2766 * the segment. 2767 */ 2768 case TCPS_CLOSING: 2769 if (ourfinisacked) { 2770 tp->t_state = TCPS_TIME_WAIT; 2771 tcp_canceltimers(tp); 2772 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2773 soisdisconnected(so); 2774 } 2775 break; 2776 2777 /* 2778 * In LAST_ACK, we may still be waiting for data to drain 2779 * and/or to be acked, as well as for the ack of our FIN. 2780 * If our FIN is now acknowledged, delete the TCB, 2781 * enter the closed state and return. 2782 */ 2783 case TCPS_LAST_ACK: 2784 if (ourfinisacked) { 2785 tp = tcp_close(tp); 2786 goto drop; 2787 } 2788 break; 2789 2790 /* 2791 * In TIME_WAIT state the only thing that should arrive 2792 * is a retransmission of the remote FIN. Acknowledge 2793 * it and restart the finack timer. 2794 */ 2795 case TCPS_TIME_WAIT: 2796 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2797 goto dropafterack; 2798 } 2799 } 2800 2801 step6: 2802 /* 2803 * Update window information. 2804 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2805 */ 2806 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2807 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2808 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2809 /* keep track of pure window updates */ 2810 if (tlen == 0 && 2811 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2812 TCP_STATINC(TCP_STAT_RCVWINUPD); 2813 tp->snd_wnd = tiwin; 2814 tp->snd_wl1 = th->th_seq; 2815 tp->snd_wl2 = th->th_ack; 2816 if (tp->snd_wnd > tp->max_sndwnd) 2817 tp->max_sndwnd = tp->snd_wnd; 2818 needoutput = 1; 2819 } 2820 2821 /* 2822 * Process segments with URG. 2823 */ 2824 if ((tiflags & TH_URG) && th->th_urp && 2825 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2826 /* 2827 * This is a kludge, but if we receive and accept 2828 * random urgent pointers, we'll crash in 2829 * soreceive. It's hard to imagine someone 2830 * actually wanting to send this much urgent data. 2831 */ 2832 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2833 th->th_urp = 0; /* XXX */ 2834 tiflags &= ~TH_URG; /* XXX */ 2835 goto dodata; /* XXX */ 2836 } 2837 /* 2838 * If this segment advances the known urgent pointer, 2839 * then mark the data stream. This should not happen 2840 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2841 * a FIN has been received from the remote side. 2842 * In these states we ignore the URG. 2843 * 2844 * According to RFC961 (Assigned Protocols), 2845 * the urgent pointer points to the last octet 2846 * of urgent data. We continue, however, 2847 * to consider it to indicate the first octet 2848 * of data past the urgent section as the original 2849 * spec states (in one of two places). 2850 */ 2851 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2852 tp->rcv_up = th->th_seq + th->th_urp; 2853 so->so_oobmark = so->so_rcv.sb_cc + 2854 (tp->rcv_up - tp->rcv_nxt) - 1; 2855 if (so->so_oobmark == 0) 2856 so->so_state |= SS_RCVATMARK; 2857 sohasoutofband(so); 2858 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2859 } 2860 /* 2861 * Remove out of band data so doesn't get presented to user. 2862 * This can happen independent of advancing the URG pointer, 2863 * but if two URG's are pending at once, some out-of-band 2864 * data may creep in... ick. 2865 */ 2866 if (th->th_urp <= (u_int16_t) tlen 2867 #ifdef SO_OOBINLINE 2868 && (so->so_options & SO_OOBINLINE) == 0 2869 #endif 2870 ) 2871 tcp_pulloutofband(so, th, m, hdroptlen); 2872 } else 2873 /* 2874 * If no out of band data is expected, 2875 * pull receive urgent pointer along 2876 * with the receive window. 2877 */ 2878 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2879 tp->rcv_up = tp->rcv_nxt; 2880 dodata: /* XXX */ 2881 2882 /* 2883 * Process the segment text, merging it into the TCP sequencing queue, 2884 * and arranging for acknowledgement of receipt if necessary. 2885 * This process logically involves adjusting tp->rcv_wnd as data 2886 * is presented to the user (this happens in tcp_usrreq.c, 2887 * tcp_rcvd()). If a FIN has already been received on this 2888 * connection then we just ignore the text. 2889 */ 2890 if ((tlen || (tiflags & TH_FIN)) && 2891 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2892 /* 2893 * Insert segment ti into reassembly queue of tcp with 2894 * control block tp. Return TH_FIN if reassembly now includes 2895 * a segment with FIN. The macro form does the common case 2896 * inline (segment is the next to be received on an 2897 * established connection, and the queue is empty), 2898 * avoiding linkage into and removal from the queue and 2899 * repetition of various conversions. 2900 * Set DELACK for segments received in order, but ack 2901 * immediately when segments are out of order 2902 * (so fast retransmit can work). 2903 */ 2904 /* NOTE: this was TCP_REASS() macro, but used only once */ 2905 TCP_REASS_LOCK(tp); 2906 if (th->th_seq == tp->rcv_nxt && 2907 TAILQ_FIRST(&tp->segq) == NULL && 2908 tp->t_state == TCPS_ESTABLISHED) { 2909 tcp_setup_ack(tp, th); 2910 tp->rcv_nxt += tlen; 2911 tiflags = th->th_flags & TH_FIN; 2912 tcps = TCP_STAT_GETREF(); 2913 tcps[TCP_STAT_RCVPACK]++; 2914 tcps[TCP_STAT_RCVBYTE] += tlen; 2915 TCP_STAT_PUTREF(); 2916 nd6_hint(tp); 2917 if (so->so_state & SS_CANTRCVMORE) 2918 m_freem(m); 2919 else { 2920 m_adj(m, hdroptlen); 2921 sbappendstream(&(so)->so_rcv, m); 2922 } 2923 TCP_REASS_UNLOCK(tp); 2924 sorwakeup(so); 2925 } else { 2926 m_adj(m, hdroptlen); 2927 tiflags = tcp_reass(tp, th, m, &tlen); 2928 tp->t_flags |= TF_ACKNOW; 2929 } 2930 2931 /* 2932 * Note the amount of data that peer has sent into 2933 * our window, in order to estimate the sender's 2934 * buffer size. 2935 */ 2936 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2937 } else { 2938 m_freem(m); 2939 m = NULL; 2940 tiflags &= ~TH_FIN; 2941 } 2942 2943 /* 2944 * If FIN is received ACK the FIN and let the user know 2945 * that the connection is closing. Ignore a FIN received before 2946 * the connection is fully established. 2947 */ 2948 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2949 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2950 socantrcvmore(so); 2951 tp->t_flags |= TF_ACKNOW; 2952 tp->rcv_nxt++; 2953 } 2954 switch (tp->t_state) { 2955 2956 /* 2957 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2958 */ 2959 case TCPS_ESTABLISHED: 2960 tp->t_state = TCPS_CLOSE_WAIT; 2961 break; 2962 2963 /* 2964 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2965 * enter the CLOSING state. 2966 */ 2967 case TCPS_FIN_WAIT_1: 2968 tp->t_state = TCPS_CLOSING; 2969 break; 2970 2971 /* 2972 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2973 * starting the time-wait timer, turning off the other 2974 * standard timers. 2975 */ 2976 case TCPS_FIN_WAIT_2: 2977 tp->t_state = TCPS_TIME_WAIT; 2978 tcp_canceltimers(tp); 2979 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2980 soisdisconnected(so); 2981 break; 2982 2983 /* 2984 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2985 */ 2986 case TCPS_TIME_WAIT: 2987 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2988 break; 2989 } 2990 } 2991 #ifdef TCP_DEBUG 2992 if (so->so_options & SO_DEBUG) 2993 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2994 #endif 2995 2996 /* 2997 * Return any desired output. 2998 */ 2999 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 3000 KERNEL_LOCK(1, NULL); 3001 (void) tcp_output(tp); 3002 KERNEL_UNLOCK_ONE(NULL); 3003 } 3004 if (tcp_saveti) 3005 m_freem(tcp_saveti); 3006 3007 if (tp->t_state == TCPS_TIME_WAIT 3008 && (so->so_state & SS_NOFDREF) 3009 && (tp->t_inpcb || af != AF_INET) 3010 && (tp->t_in6pcb || af != AF_INET6) 3011 && ((af == AF_INET ? tcp4_vtw_enable : tcp6_vtw_enable) & 1) != 0 3012 && TAILQ_EMPTY(&tp->segq) 3013 && vtw_add(af, tp)) { 3014 ; 3015 } 3016 return; 3017 3018 badsyn: 3019 /* 3020 * Received a bad SYN. Increment counters and dropwithreset. 3021 */ 3022 TCP_STATINC(TCP_STAT_BADSYN); 3023 tp = NULL; 3024 goto dropwithreset; 3025 3026 dropafterack: 3027 /* 3028 * Generate an ACK dropping incoming segment if it occupies 3029 * sequence space, where the ACK reflects our state. 3030 */ 3031 if (tiflags & TH_RST) 3032 goto drop; 3033 goto dropafterack2; 3034 3035 dropafterack_ratelim: 3036 /* 3037 * We may want to rate-limit ACKs against SYN/RST attack. 3038 */ 3039 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 3040 tcp_ackdrop_ppslim) == 0) { 3041 /* XXX stat */ 3042 goto drop; 3043 } 3044 /* ...fall into dropafterack2... */ 3045 3046 dropafterack2: 3047 m_freem(m); 3048 tp->t_flags |= TF_ACKNOW; 3049 KERNEL_LOCK(1, NULL); 3050 (void) tcp_output(tp); 3051 KERNEL_UNLOCK_ONE(NULL); 3052 if (tcp_saveti) 3053 m_freem(tcp_saveti); 3054 return; 3055 3056 dropwithreset_ratelim: 3057 /* 3058 * We may want to rate-limit RSTs in certain situations, 3059 * particularly if we are sending an RST in response to 3060 * an attempt to connect to or otherwise communicate with 3061 * a port for which we have no socket. 3062 */ 3063 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 3064 tcp_rst_ppslim) == 0) { 3065 /* XXX stat */ 3066 goto drop; 3067 } 3068 /* ...fall into dropwithreset... */ 3069 3070 dropwithreset: 3071 /* 3072 * Generate a RST, dropping incoming segment. 3073 * Make ACK acceptable to originator of segment. 3074 */ 3075 if (tiflags & TH_RST) 3076 goto drop; 3077 3078 switch (af) { 3079 #ifdef INET6 3080 case AF_INET6: 3081 /* For following calls to tcp_respond */ 3082 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 3083 goto drop; 3084 break; 3085 #endif /* INET6 */ 3086 case AF_INET: 3087 if (IN_MULTICAST(ip->ip_dst.s_addr) || 3088 in_broadcast(ip->ip_dst, m_get_rcvif_NOMPSAFE(m))) 3089 goto drop; 3090 } 3091 3092 if (tiflags & TH_ACK) 3093 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3094 else { 3095 if (tiflags & TH_SYN) 3096 tlen++; 3097 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 3098 TH_RST|TH_ACK); 3099 } 3100 if (tcp_saveti) 3101 m_freem(tcp_saveti); 3102 return; 3103 3104 badcsum: 3105 drop: 3106 /* 3107 * Drop space held by incoming segment and return. 3108 */ 3109 if (tp) { 3110 if (tp->t_inpcb) 3111 so = tp->t_inpcb->inp_socket; 3112 #ifdef INET6 3113 else if (tp->t_in6pcb) 3114 so = tp->t_in6pcb->in6p_socket; 3115 #endif 3116 else 3117 so = NULL; 3118 #ifdef TCP_DEBUG 3119 if (so && (so->so_options & SO_DEBUG) != 0) 3120 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 3121 #endif 3122 } 3123 if (tcp_saveti) 3124 m_freem(tcp_saveti); 3125 m_freem(m); 3126 return; 3127 } 3128 3129 #ifdef TCP_SIGNATURE 3130 int 3131 tcp_signature_apply(void *fstate, void *data, u_int len) 3132 { 3133 3134 MD5Update(fstate, (u_char *)data, len); 3135 return (0); 3136 } 3137 3138 struct secasvar * 3139 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 3140 { 3141 struct ip *ip; 3142 struct ip6_hdr *ip6; 3143 3144 ip = mtod(m, struct ip *); 3145 switch (ip->ip_v) { 3146 case 4: 3147 ip = mtod(m, struct ip *); 3148 ip6 = NULL; 3149 break; 3150 case 6: 3151 ip = NULL; 3152 ip6 = mtod(m, struct ip6_hdr *); 3153 break; 3154 default: 3155 return (NULL); 3156 } 3157 3158 #ifdef IPSEC 3159 union sockaddr_union dst; 3160 3161 /* Extract the destination from the IP header in the mbuf. */ 3162 memset(&dst, 0, sizeof(union sockaddr_union)); 3163 if (ip != NULL) { 3164 dst.sa.sa_len = sizeof(struct sockaddr_in); 3165 dst.sa.sa_family = AF_INET; 3166 dst.sin.sin_addr = ip->ip_dst; 3167 } else { 3168 dst.sa.sa_len = sizeof(struct sockaddr_in6); 3169 dst.sa.sa_family = AF_INET6; 3170 dst.sin6.sin6_addr = ip6->ip6_dst; 3171 } 3172 3173 /* 3174 * Look up an SADB entry which matches the address of the peer. 3175 */ 3176 return KEY_LOOKUP_SA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI), 0, 0); 3177 #else 3178 return NULL; 3179 #endif 3180 } 3181 3182 int 3183 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 3184 struct secasvar *sav, char *sig) 3185 { 3186 MD5_CTX ctx; 3187 struct ip *ip; 3188 struct ipovly *ipovly; 3189 #ifdef INET6 3190 struct ip6_hdr *ip6; 3191 struct ip6_hdr_pseudo ip6pseudo; 3192 #endif /* INET6 */ 3193 struct ippseudo ippseudo; 3194 struct tcphdr th0; 3195 int l, tcphdrlen; 3196 3197 if (sav == NULL) 3198 return (-1); 3199 3200 tcphdrlen = th->th_off * 4; 3201 3202 switch (mtod(m, struct ip *)->ip_v) { 3203 case 4: 3204 MD5Init(&ctx); 3205 ip = mtod(m, struct ip *); 3206 memset(&ippseudo, 0, sizeof(ippseudo)); 3207 ipovly = (struct ipovly *)ip; 3208 ippseudo.ippseudo_src = ipovly->ih_src; 3209 ippseudo.ippseudo_dst = ipovly->ih_dst; 3210 ippseudo.ippseudo_pad = 0; 3211 ippseudo.ippseudo_p = IPPROTO_TCP; 3212 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 3213 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 3214 break; 3215 #if INET6 3216 case 6: 3217 MD5Init(&ctx); 3218 ip6 = mtod(m, struct ip6_hdr *); 3219 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 3220 ip6pseudo.ip6ph_src = ip6->ip6_src; 3221 in6_clearscope(&ip6pseudo.ip6ph_src); 3222 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 3223 in6_clearscope(&ip6pseudo.ip6ph_dst); 3224 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 3225 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 3226 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 3227 break; 3228 #endif /* INET6 */ 3229 default: 3230 return (-1); 3231 } 3232 3233 th0 = *th; 3234 th0.th_sum = 0; 3235 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 3236 3237 l = m->m_pkthdr.len - thoff - tcphdrlen; 3238 if (l > 0) 3239 m_apply(m, thoff + tcphdrlen, 3240 m->m_pkthdr.len - thoff - tcphdrlen, 3241 tcp_signature_apply, &ctx); 3242 3243 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 3244 MD5Final(sig, &ctx); 3245 3246 return (0); 3247 } 3248 #endif 3249 3250 /* 3251 * tcp_dooptions: parse and process tcp options. 3252 * 3253 * returns -1 if this segment should be dropped. (eg. wrong signature) 3254 * otherwise returns 0. 3255 */ 3256 3257 static int 3258 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, 3259 struct tcphdr *th, 3260 struct mbuf *m, int toff, struct tcp_opt_info *oi) 3261 { 3262 u_int16_t mss; 3263 int opt, optlen = 0; 3264 #ifdef TCP_SIGNATURE 3265 void *sigp = NULL; 3266 char sigbuf[TCP_SIGLEN]; 3267 struct secasvar *sav = NULL; 3268 #endif 3269 3270 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 3271 opt = cp[0]; 3272 if (opt == TCPOPT_EOL) 3273 break; 3274 if (opt == TCPOPT_NOP) 3275 optlen = 1; 3276 else { 3277 if (cnt < 2) 3278 break; 3279 optlen = cp[1]; 3280 if (optlen < 2 || optlen > cnt) 3281 break; 3282 } 3283 switch (opt) { 3284 3285 default: 3286 continue; 3287 3288 case TCPOPT_MAXSEG: 3289 if (optlen != TCPOLEN_MAXSEG) 3290 continue; 3291 if (!(th->th_flags & TH_SYN)) 3292 continue; 3293 if (TCPS_HAVERCVDSYN(tp->t_state)) 3294 continue; 3295 bcopy(cp + 2, &mss, sizeof(mss)); 3296 oi->maxseg = ntohs(mss); 3297 break; 3298 3299 case TCPOPT_WINDOW: 3300 if (optlen != TCPOLEN_WINDOW) 3301 continue; 3302 if (!(th->th_flags & TH_SYN)) 3303 continue; 3304 if (TCPS_HAVERCVDSYN(tp->t_state)) 3305 continue; 3306 tp->t_flags |= TF_RCVD_SCALE; 3307 tp->requested_s_scale = cp[2]; 3308 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3309 char buf[INET6_ADDRSTRLEN]; 3310 struct ip *ip = mtod(m, struct ip *); 3311 #ifdef INET6 3312 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 3313 #endif 3314 if (ip) 3315 in_print(buf, sizeof(buf), 3316 &ip->ip_src); 3317 #ifdef INET6 3318 else if (ip6) 3319 in6_print(buf, sizeof(buf), 3320 &ip6->ip6_src); 3321 #endif 3322 else 3323 strlcpy(buf, "(unknown)", sizeof(buf)); 3324 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3325 "assuming %d\n", 3326 tp->requested_s_scale, buf, 3327 TCP_MAX_WINSHIFT); 3328 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3329 } 3330 break; 3331 3332 case TCPOPT_TIMESTAMP: 3333 if (optlen != TCPOLEN_TIMESTAMP) 3334 continue; 3335 oi->ts_present = 1; 3336 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 3337 NTOHL(oi->ts_val); 3338 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 3339 NTOHL(oi->ts_ecr); 3340 3341 if (!(th->th_flags & TH_SYN)) 3342 continue; 3343 if (TCPS_HAVERCVDSYN(tp->t_state)) 3344 continue; 3345 /* 3346 * A timestamp received in a SYN makes 3347 * it ok to send timestamp requests and replies. 3348 */ 3349 tp->t_flags |= TF_RCVD_TSTMP; 3350 tp->ts_recent = oi->ts_val; 3351 tp->ts_recent_age = tcp_now; 3352 break; 3353 3354 case TCPOPT_SACK_PERMITTED: 3355 if (optlen != TCPOLEN_SACK_PERMITTED) 3356 continue; 3357 if (!(th->th_flags & TH_SYN)) 3358 continue; 3359 if (TCPS_HAVERCVDSYN(tp->t_state)) 3360 continue; 3361 if (tcp_do_sack) { 3362 tp->t_flags |= TF_SACK_PERMIT; 3363 tp->t_flags |= TF_WILL_SACK; 3364 } 3365 break; 3366 3367 case TCPOPT_SACK: 3368 tcp_sack_option(tp, th, cp, optlen); 3369 break; 3370 #ifdef TCP_SIGNATURE 3371 case TCPOPT_SIGNATURE: 3372 if (optlen != TCPOLEN_SIGNATURE) 3373 continue; 3374 if (sigp && memcmp(sigp, cp + 2, TCP_SIGLEN)) 3375 return (-1); 3376 3377 sigp = sigbuf; 3378 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3379 tp->t_flags |= TF_SIGNATURE; 3380 break; 3381 #endif 3382 } 3383 } 3384 3385 #ifndef TCP_SIGNATURE 3386 return 0; 3387 #else 3388 if (tp->t_flags & TF_SIGNATURE) { 3389 3390 sav = tcp_signature_getsav(m, th); 3391 3392 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3393 return (-1); 3394 } 3395 3396 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) 3397 goto out; 3398 3399 if (sigp) { 3400 char sig[TCP_SIGLEN]; 3401 3402 tcp_fields_to_net(th); 3403 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3404 tcp_fields_to_host(th); 3405 goto out; 3406 } 3407 tcp_fields_to_host(th); 3408 3409 if (memcmp(sig, sigp, TCP_SIGLEN)) { 3410 TCP_STATINC(TCP_STAT_BADSIG); 3411 goto out; 3412 } else 3413 TCP_STATINC(TCP_STAT_GOODSIG); 3414 3415 key_sa_recordxfer(sav, m); 3416 KEY_SA_UNREF(&sav); 3417 } 3418 return 0; 3419 out: 3420 if (sav != NULL) 3421 KEY_SA_UNREF(&sav); 3422 return -1; 3423 #endif 3424 } 3425 3426 /* 3427 * Pull out of band byte out of a segment so 3428 * it doesn't appear in the user's data queue. 3429 * It is still reflected in the segment length for 3430 * sequencing purposes. 3431 */ 3432 void 3433 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3434 struct mbuf *m, int off) 3435 { 3436 int cnt = off + th->th_urp - 1; 3437 3438 while (cnt >= 0) { 3439 if (m->m_len > cnt) { 3440 char *cp = mtod(m, char *) + cnt; 3441 struct tcpcb *tp = sototcpcb(so); 3442 3443 tp->t_iobc = *cp; 3444 tp->t_oobflags |= TCPOOB_HAVEDATA; 3445 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3446 m->m_len--; 3447 return; 3448 } 3449 cnt -= m->m_len; 3450 m = m->m_next; 3451 if (m == 0) 3452 break; 3453 } 3454 panic("tcp_pulloutofband"); 3455 } 3456 3457 /* 3458 * Collect new round-trip time estimate 3459 * and update averages and current timeout. 3460 * 3461 * rtt is in units of slow ticks (typically 500 ms) -- essentially the 3462 * difference of two timestamps. 3463 */ 3464 void 3465 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3466 { 3467 int32_t delta; 3468 3469 TCP_STATINC(TCP_STAT_RTTUPDATED); 3470 if (tp->t_srtt != 0) { 3471 /* 3472 * Compute the amount to add to srtt for smoothing, 3473 * *alpha, or 2^(-TCP_RTT_SHIFT). Because 3474 * srtt is stored in 1/32 slow ticks, we conceptually 3475 * shift left 5 bits, subtract srtt to get the 3476 * diference, and then shift right by TCP_RTT_SHIFT 3477 * (3) to obtain 1/8 of the difference. 3478 */ 3479 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3480 /* 3481 * This can never happen, because delta's lowest 3482 * possible value is 1/8 of t_srtt. But if it does, 3483 * set srtt to some reasonable value, here chosen 3484 * as 1/8 tick. 3485 */ 3486 if ((tp->t_srtt += delta) <= 0) 3487 tp->t_srtt = 1 << 2; 3488 /* 3489 * RFC2988 requires that rttvar be updated first. 3490 * This code is compliant because "delta" is the old 3491 * srtt minus the new observation (scaled). 3492 * 3493 * RFC2988 says: 3494 * rttvar = (1-beta) * rttvar + beta * |srtt-observed| 3495 * 3496 * delta is in units of 1/32 ticks, and has then been 3497 * divided by 8. This is equivalent to being in 1/16s 3498 * units and divided by 4. Subtract from it 1/4 of 3499 * the existing rttvar to form the (signed) amount to 3500 * adjust. 3501 */ 3502 if (delta < 0) 3503 delta = -delta; 3504 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3505 /* 3506 * As with srtt, this should never happen. There is 3507 * no support in RFC2988 for this operation. But 1/4s 3508 * as rttvar when faced with something arguably wrong 3509 * is ok. 3510 */ 3511 if ((tp->t_rttvar += delta) <= 0) 3512 tp->t_rttvar = 1 << 2; 3513 3514 /* 3515 * If srtt exceeds .01 second, ensure we use the 'remote' MSL 3516 * Problem is: it doesn't work. Disabled by defaulting 3517 * tcp_rttlocal to 0; see corresponding code in 3518 * tcp_subr that selects local vs remote in a different way. 3519 * 3520 * The static branch prediction hint here should be removed 3521 * when the rtt estimator is fixed and the rtt_enable code 3522 * is turned back on. 3523 */ 3524 if (__predict_false(tcp_rttlocal) && tcp_msl_enable 3525 && tp->t_srtt > tcp_msl_remote_threshold 3526 && tp->t_msl < tcp_msl_remote) { 3527 tp->t_msl = tcp_msl_remote; 3528 } 3529 } else { 3530 /* 3531 * This is the first measurement. Per RFC2988, 2.2, 3532 * set rtt=R and srtt=R/2. 3533 * For srtt, storage representation is 1/32 ticks, 3534 * so shift left by 5. 3535 * For rttvar, storage representation is 1/16 ticks, 3536 * So shift left by 4, but then right by 1 to halve. 3537 */ 3538 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3539 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3540 } 3541 tp->t_rtttime = 0; 3542 tp->t_rxtshift = 0; 3543 3544 /* 3545 * the retransmit should happen at rtt + 4 * rttvar. 3546 * Because of the way we do the smoothing, srtt and rttvar 3547 * will each average +1/2 tick of bias. When we compute 3548 * the retransmit timer, we want 1/2 tick of rounding and 3549 * 1 extra tick because of +-1/2 tick uncertainty in the 3550 * firing of the timer. The bias will give us exactly the 3551 * 1.5 tick we need. But, because the bias is 3552 * statistical, we have to test that we don't drop below 3553 * the minimum feasible timer (which is 2 ticks). 3554 */ 3555 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3556 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3557 3558 /* 3559 * We received an ack for a packet that wasn't retransmitted; 3560 * it is probably safe to discard any error indications we've 3561 * received recently. This isn't quite right, but close enough 3562 * for now (a route might have failed after we sent a segment, 3563 * and the return path might not be symmetrical). 3564 */ 3565 tp->t_softerror = 0; 3566 } 3567 3568 3569 /* 3570 * TCP compressed state engine. Currently used to hold compressed 3571 * state for SYN_RECEIVED. 3572 */ 3573 3574 u_long syn_cache_count; 3575 u_int32_t syn_hash1, syn_hash2; 3576 3577 #define SYN_HASH(sa, sp, dp) \ 3578 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3579 ((u_int32_t)(sp)))^syn_hash2))) 3580 #ifndef INET6 3581 #define SYN_HASHALL(hash, src, dst) \ 3582 do { \ 3583 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3584 ((const struct sockaddr_in *)(src))->sin_port, \ 3585 ((const struct sockaddr_in *)(dst))->sin_port); \ 3586 } while (/*CONSTCOND*/ 0) 3587 #else 3588 #define SYN_HASH6(sa, sp, dp) \ 3589 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3590 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3591 & 0x7fffffff) 3592 3593 #define SYN_HASHALL(hash, src, dst) \ 3594 do { \ 3595 switch ((src)->sa_family) { \ 3596 case AF_INET: \ 3597 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3598 ((const struct sockaddr_in *)(src))->sin_port, \ 3599 ((const struct sockaddr_in *)(dst))->sin_port); \ 3600 break; \ 3601 case AF_INET6: \ 3602 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3603 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3604 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3605 break; \ 3606 default: \ 3607 hash = 0; \ 3608 } \ 3609 } while (/*CONSTCOND*/0) 3610 #endif /* INET6 */ 3611 3612 static struct pool syn_cache_pool; 3613 3614 /* 3615 * We don't estimate RTT with SYNs, so each packet starts with the default 3616 * RTT and each timer step has a fixed timeout value. 3617 */ 3618 static inline void 3619 syn_cache_timer_arm(struct syn_cache *sc) 3620 { 3621 3622 TCPT_RANGESET(sc->sc_rxtcur, 3623 TCPTV_SRTTDFLT * tcp_backoff[sc->sc_rxtshift], TCPTV_MIN, 3624 TCPTV_REXMTMAX); 3625 callout_reset(&sc->sc_timer, 3626 sc->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, sc); 3627 } 3628 3629 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3630 3631 static inline void 3632 syn_cache_rm(struct syn_cache *sc) 3633 { 3634 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3635 sc, sc_bucketq); 3636 sc->sc_tp = NULL; 3637 LIST_REMOVE(sc, sc_tpq); 3638 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3639 callout_stop(&sc->sc_timer); 3640 syn_cache_count--; 3641 } 3642 3643 static inline void 3644 syn_cache_put(struct syn_cache *sc) 3645 { 3646 if (sc->sc_ipopts) 3647 (void) m_free(sc->sc_ipopts); 3648 rtcache_free(&sc->sc_route); 3649 sc->sc_flags |= SCF_DEAD; 3650 if (!callout_invoking(&sc->sc_timer)) 3651 callout_schedule(&(sc)->sc_timer, 1); 3652 } 3653 3654 void 3655 syn_cache_init(void) 3656 { 3657 int i; 3658 3659 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 3660 "synpl", NULL, IPL_SOFTNET); 3661 3662 /* Initialize the hash buckets. */ 3663 for (i = 0; i < tcp_syn_cache_size; i++) 3664 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3665 } 3666 3667 void 3668 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3669 { 3670 struct syn_cache_head *scp; 3671 struct syn_cache *sc2; 3672 int s; 3673 3674 /* 3675 * If there are no entries in the hash table, reinitialize 3676 * the hash secrets. 3677 */ 3678 if (syn_cache_count == 0) { 3679 syn_hash1 = cprng_fast32(); 3680 syn_hash2 = cprng_fast32(); 3681 } 3682 3683 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3684 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3685 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3686 3687 /* 3688 * Make sure that we don't overflow the per-bucket 3689 * limit or the total cache size limit. 3690 */ 3691 s = splsoftnet(); 3692 if (scp->sch_length >= tcp_syn_bucket_limit) { 3693 TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW); 3694 /* 3695 * The bucket is full. Toss the oldest element in the 3696 * bucket. This will be the first entry in the bucket. 3697 */ 3698 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3699 #ifdef DIAGNOSTIC 3700 /* 3701 * This should never happen; we should always find an 3702 * entry in our bucket. 3703 */ 3704 if (sc2 == NULL) 3705 panic("syn_cache_insert: bucketoverflow: impossible"); 3706 #endif 3707 syn_cache_rm(sc2); 3708 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3709 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3710 struct syn_cache_head *scp2, *sce; 3711 3712 TCP_STATINC(TCP_STAT_SC_OVERFLOWED); 3713 /* 3714 * The cache is full. Toss the oldest entry in the 3715 * first non-empty bucket we can find. 3716 * 3717 * XXX We would really like to toss the oldest 3718 * entry in the cache, but we hope that this 3719 * condition doesn't happen very often. 3720 */ 3721 scp2 = scp; 3722 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3723 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3724 for (++scp2; scp2 != scp; scp2++) { 3725 if (scp2 >= sce) 3726 scp2 = &tcp_syn_cache[0]; 3727 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3728 break; 3729 } 3730 #ifdef DIAGNOSTIC 3731 /* 3732 * This should never happen; we should always find a 3733 * non-empty bucket. 3734 */ 3735 if (scp2 == scp) 3736 panic("syn_cache_insert: cacheoverflow: " 3737 "impossible"); 3738 #endif 3739 } 3740 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3741 syn_cache_rm(sc2); 3742 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3743 } 3744 3745 /* 3746 * Initialize the entry's timer. 3747 */ 3748 sc->sc_rxttot = 0; 3749 sc->sc_rxtshift = 0; 3750 syn_cache_timer_arm(sc); 3751 3752 /* Link it from tcpcb entry */ 3753 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3754 3755 /* Put it into the bucket. */ 3756 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3757 scp->sch_length++; 3758 syn_cache_count++; 3759 3760 TCP_STATINC(TCP_STAT_SC_ADDED); 3761 splx(s); 3762 } 3763 3764 /* 3765 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3766 * If we have retransmitted an entry the maximum number of times, expire 3767 * that entry. 3768 */ 3769 static void 3770 syn_cache_timer(void *arg) 3771 { 3772 struct syn_cache *sc = arg; 3773 3774 mutex_enter(softnet_lock); 3775 KERNEL_LOCK(1, NULL); 3776 3777 callout_ack(&sc->sc_timer); 3778 3779 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3780 TCP_STATINC(TCP_STAT_SC_DELAYED_FREE); 3781 goto free; 3782 } 3783 3784 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3785 /* Drop it -- too many retransmissions. */ 3786 goto dropit; 3787 } 3788 3789 /* 3790 * Compute the total amount of time this entry has 3791 * been on a queue. If this entry has been on longer 3792 * than the keep alive timer would allow, expire it. 3793 */ 3794 sc->sc_rxttot += sc->sc_rxtcur; 3795 if (sc->sc_rxttot >= tcp_keepinit) 3796 goto dropit; 3797 3798 TCP_STATINC(TCP_STAT_SC_RETRANSMITTED); 3799 (void) syn_cache_respond(sc, NULL); 3800 3801 /* Advance the timer back-off. */ 3802 sc->sc_rxtshift++; 3803 syn_cache_timer_arm(sc); 3804 3805 goto out; 3806 3807 dropit: 3808 TCP_STATINC(TCP_STAT_SC_TIMED_OUT); 3809 syn_cache_rm(sc); 3810 if (sc->sc_ipopts) 3811 (void) m_free(sc->sc_ipopts); 3812 rtcache_free(&sc->sc_route); 3813 3814 free: 3815 callout_destroy(&sc->sc_timer); 3816 pool_put(&syn_cache_pool, sc); 3817 3818 out: 3819 KERNEL_UNLOCK_ONE(NULL); 3820 mutex_exit(softnet_lock); 3821 } 3822 3823 /* 3824 * Remove syn cache created by the specified tcb entry, 3825 * because this does not make sense to keep them 3826 * (if there's no tcb entry, syn cache entry will never be used) 3827 */ 3828 void 3829 syn_cache_cleanup(struct tcpcb *tp) 3830 { 3831 struct syn_cache *sc, *nsc; 3832 int s; 3833 3834 s = splsoftnet(); 3835 3836 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3837 nsc = LIST_NEXT(sc, sc_tpq); 3838 3839 #ifdef DIAGNOSTIC 3840 if (sc->sc_tp != tp) 3841 panic("invalid sc_tp in syn_cache_cleanup"); 3842 #endif 3843 syn_cache_rm(sc); 3844 syn_cache_put(sc); /* calls pool_put but see spl above */ 3845 } 3846 /* just for safety */ 3847 LIST_INIT(&tp->t_sc); 3848 3849 splx(s); 3850 } 3851 3852 /* 3853 * Find an entry in the syn cache. 3854 */ 3855 struct syn_cache * 3856 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3857 struct syn_cache_head **headp) 3858 { 3859 struct syn_cache *sc; 3860 struct syn_cache_head *scp; 3861 u_int32_t hash; 3862 int s; 3863 3864 SYN_HASHALL(hash, src, dst); 3865 3866 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3867 *headp = scp; 3868 s = splsoftnet(); 3869 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3870 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3871 if (sc->sc_hash != hash) 3872 continue; 3873 if (!memcmp(&sc->sc_src, src, src->sa_len) && 3874 !memcmp(&sc->sc_dst, dst, dst->sa_len)) { 3875 splx(s); 3876 return (sc); 3877 } 3878 } 3879 splx(s); 3880 return (NULL); 3881 } 3882 3883 /* 3884 * This function gets called when we receive an ACK for a 3885 * socket in the LISTEN state. We look up the connection 3886 * in the syn cache, and if its there, we pull it out of 3887 * the cache and turn it into a full-blown connection in 3888 * the SYN-RECEIVED state. 3889 * 3890 * The return values may not be immediately obvious, and their effects 3891 * can be subtle, so here they are: 3892 * 3893 * NULL SYN was not found in cache; caller should drop the 3894 * packet and send an RST. 3895 * 3896 * -1 We were unable to create the new connection, and are 3897 * aborting it. An ACK,RST is being sent to the peer 3898 * (unless we got screwey sequence numbners; see below), 3899 * because the 3-way handshake has been completed. Caller 3900 * should not free the mbuf, since we may be using it. If 3901 * we are not, we will free it. 3902 * 3903 * Otherwise, the return value is a pointer to the new socket 3904 * associated with the connection. 3905 */ 3906 struct socket * 3907 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3908 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3909 struct socket *so, struct mbuf *m) 3910 { 3911 struct syn_cache *sc; 3912 struct syn_cache_head *scp; 3913 struct inpcb *inp = NULL; 3914 #ifdef INET6 3915 struct in6pcb *in6p = NULL; 3916 #endif 3917 struct tcpcb *tp = 0; 3918 int s; 3919 struct socket *oso; 3920 3921 s = splsoftnet(); 3922 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3923 splx(s); 3924 return (NULL); 3925 } 3926 3927 /* 3928 * Verify the sequence and ack numbers. Try getting the correct 3929 * response again. 3930 */ 3931 if ((th->th_ack != sc->sc_iss + 1) || 3932 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3933 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3934 (void) syn_cache_respond(sc, m); 3935 splx(s); 3936 return ((struct socket *)(-1)); 3937 } 3938 3939 /* Remove this cache entry */ 3940 syn_cache_rm(sc); 3941 splx(s); 3942 3943 /* 3944 * Ok, create the full blown connection, and set things up 3945 * as they would have been set up if we had created the 3946 * connection when the SYN arrived. If we can't create 3947 * the connection, abort it. 3948 */ 3949 /* 3950 * inp still has the OLD in_pcb stuff, set the 3951 * v6-related flags on the new guy, too. This is 3952 * done particularly for the case where an AF_INET6 3953 * socket is bound only to a port, and a v4 connection 3954 * comes in on that port. 3955 * we also copy the flowinfo from the original pcb 3956 * to the new one. 3957 */ 3958 oso = so; 3959 so = sonewconn(so, true); 3960 if (so == NULL) 3961 goto resetandabort; 3962 3963 switch (so->so_proto->pr_domain->dom_family) { 3964 #ifdef INET 3965 case AF_INET: 3966 inp = sotoinpcb(so); 3967 break; 3968 #endif 3969 #ifdef INET6 3970 case AF_INET6: 3971 in6p = sotoin6pcb(so); 3972 break; 3973 #endif 3974 } 3975 switch (src->sa_family) { 3976 #ifdef INET 3977 case AF_INET: 3978 if (inp) { 3979 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3980 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3981 inp->inp_options = ip_srcroute(m); 3982 in_pcbstate(inp, INP_BOUND); 3983 if (inp->inp_options == NULL) { 3984 inp->inp_options = sc->sc_ipopts; 3985 sc->sc_ipopts = NULL; 3986 } 3987 } 3988 #ifdef INET6 3989 else if (in6p) { 3990 /* IPv4 packet to AF_INET6 socket */ 3991 memset(&in6p->in6p_laddr, 0, sizeof(in6p->in6p_laddr)); 3992 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3993 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3994 &in6p->in6p_laddr.s6_addr32[3], 3995 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3996 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3997 in6totcpcb(in6p)->t_family = AF_INET; 3998 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3999 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 4000 else 4001 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 4002 in6_pcbstate(in6p, IN6P_BOUND); 4003 } 4004 #endif 4005 break; 4006 #endif 4007 #ifdef INET6 4008 case AF_INET6: 4009 if (in6p) { 4010 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 4011 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 4012 in6_pcbstate(in6p, IN6P_BOUND); 4013 } 4014 break; 4015 #endif 4016 } 4017 #ifdef INET6 4018 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 4019 struct in6pcb *oin6p = sotoin6pcb(oso); 4020 /* inherit socket options from the listening socket */ 4021 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 4022 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 4023 m_freem(in6p->in6p_options); 4024 in6p->in6p_options = 0; 4025 } 4026 ip6_savecontrol(in6p, &in6p->in6p_options, 4027 mtod(m, struct ip6_hdr *), m); 4028 } 4029 #endif 4030 4031 #if defined(IPSEC) 4032 if (ipsec_used) { 4033 /* 4034 * we make a copy of policy, instead of sharing the policy, for 4035 * better behavior in terms of SA lookup and dead SA removal. 4036 */ 4037 if (inp) { 4038 /* copy old policy into new socket's */ 4039 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, 4040 inp->inp_sp)) 4041 printf("tcp_input: could not copy policy\n"); 4042 } 4043 #ifdef INET6 4044 else if (in6p) { 4045 /* copy old policy into new socket's */ 4046 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 4047 in6p->in6p_sp)) 4048 printf("tcp_input: could not copy policy\n"); 4049 } 4050 #endif 4051 } 4052 #endif 4053 4054 /* 4055 * Give the new socket our cached route reference. 4056 */ 4057 if (inp) { 4058 rtcache_copy(&inp->inp_route, &sc->sc_route); 4059 rtcache_free(&sc->sc_route); 4060 } 4061 #ifdef INET6 4062 else { 4063 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 4064 rtcache_free(&sc->sc_route); 4065 } 4066 #endif 4067 4068 if (inp) { 4069 struct sockaddr_in sin; 4070 memcpy(&sin, src, src->sa_len); 4071 if (in_pcbconnect(inp, &sin, &lwp0)) { 4072 goto resetandabort; 4073 } 4074 } 4075 #ifdef INET6 4076 else if (in6p) { 4077 struct sockaddr_in6 sin6; 4078 memcpy(&sin6, src, src->sa_len); 4079 if (src->sa_family == AF_INET) { 4080 /* IPv4 packet to AF_INET6 socket */ 4081 in6_sin_2_v4mapsin6((struct sockaddr_in *)src, &sin6); 4082 } 4083 if (in6_pcbconnect(in6p, &sin6, NULL)) { 4084 goto resetandabort; 4085 } 4086 } 4087 #endif 4088 else { 4089 goto resetandabort; 4090 } 4091 4092 if (inp) 4093 tp = intotcpcb(inp); 4094 #ifdef INET6 4095 else if (in6p) 4096 tp = in6totcpcb(in6p); 4097 #endif 4098 else 4099 tp = NULL; 4100 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 4101 if (sc->sc_request_r_scale != 15) { 4102 tp->requested_s_scale = sc->sc_requested_s_scale; 4103 tp->request_r_scale = sc->sc_request_r_scale; 4104 tp->snd_scale = sc->sc_requested_s_scale; 4105 tp->rcv_scale = sc->sc_request_r_scale; 4106 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 4107 } 4108 if (sc->sc_flags & SCF_TIMESTAMP) 4109 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 4110 tp->ts_timebase = sc->sc_timebase; 4111 4112 tp->t_template = tcp_template(tp); 4113 if (tp->t_template == 0) { 4114 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 4115 so = NULL; 4116 m_freem(m); 4117 goto abort; 4118 } 4119 4120 tp->iss = sc->sc_iss; 4121 tp->irs = sc->sc_irs; 4122 tcp_sendseqinit(tp); 4123 tcp_rcvseqinit(tp); 4124 tp->t_state = TCPS_SYN_RECEIVED; 4125 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 4126 TCP_STATINC(TCP_STAT_ACCEPTS); 4127 4128 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 4129 tp->t_flags |= TF_WILL_SACK; 4130 4131 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 4132 tp->t_flags |= TF_ECN_PERMIT; 4133 4134 #ifdef TCP_SIGNATURE 4135 if (sc->sc_flags & SCF_SIGNATURE) 4136 tp->t_flags |= TF_SIGNATURE; 4137 #endif 4138 4139 /* Initialize tp->t_ourmss before we deal with the peer's! */ 4140 tp->t_ourmss = sc->sc_ourmaxseg; 4141 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 4142 4143 /* 4144 * Initialize the initial congestion window. If we 4145 * had to retransmit the SYN,ACK, we must initialize cwnd 4146 * to 1 segment (i.e. the Loss Window). 4147 */ 4148 if (sc->sc_rxtshift) 4149 tp->snd_cwnd = tp->t_peermss; 4150 else { 4151 int ss = tcp_init_win; 4152 #ifdef INET 4153 if (inp != NULL && in_localaddr(inp->inp_faddr)) 4154 ss = tcp_init_win_local; 4155 #endif 4156 #ifdef INET6 4157 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 4158 ss = tcp_init_win_local; 4159 #endif 4160 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 4161 } 4162 4163 tcp_rmx_rtt(tp); 4164 tp->snd_wl1 = sc->sc_irs; 4165 tp->rcv_up = sc->sc_irs + 1; 4166 4167 /* 4168 * This is what whould have happened in tcp_output() when 4169 * the SYN,ACK was sent. 4170 */ 4171 tp->snd_up = tp->snd_una; 4172 tp->snd_max = tp->snd_nxt = tp->iss+1; 4173 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 4174 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 4175 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 4176 tp->last_ack_sent = tp->rcv_nxt; 4177 tp->t_partialacks = -1; 4178 tp->t_dupacks = 0; 4179 4180 TCP_STATINC(TCP_STAT_SC_COMPLETED); 4181 s = splsoftnet(); 4182 syn_cache_put(sc); 4183 splx(s); 4184 return (so); 4185 4186 resetandabort: 4187 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 4188 abort: 4189 if (so != NULL) { 4190 (void) soqremque(so, 1); 4191 (void) soabort(so); 4192 mutex_enter(softnet_lock); 4193 } 4194 s = splsoftnet(); 4195 syn_cache_put(sc); 4196 splx(s); 4197 TCP_STATINC(TCP_STAT_SC_ABORTED); 4198 return ((struct socket *)(-1)); 4199 } 4200 4201 /* 4202 * This function is called when we get a RST for a 4203 * non-existent connection, so that we can see if the 4204 * connection is in the syn cache. If it is, zap it. 4205 */ 4206 4207 void 4208 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 4209 { 4210 struct syn_cache *sc; 4211 struct syn_cache_head *scp; 4212 int s = splsoftnet(); 4213 4214 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4215 splx(s); 4216 return; 4217 } 4218 if (SEQ_LT(th->th_seq, sc->sc_irs) || 4219 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 4220 splx(s); 4221 return; 4222 } 4223 syn_cache_rm(sc); 4224 TCP_STATINC(TCP_STAT_SC_RESET); 4225 syn_cache_put(sc); /* calls pool_put but see spl above */ 4226 splx(s); 4227 } 4228 4229 void 4230 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 4231 struct tcphdr *th) 4232 { 4233 struct syn_cache *sc; 4234 struct syn_cache_head *scp; 4235 int s; 4236 4237 s = splsoftnet(); 4238 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4239 splx(s); 4240 return; 4241 } 4242 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 4243 if (ntohl (th->th_seq) != sc->sc_iss) { 4244 splx(s); 4245 return; 4246 } 4247 4248 /* 4249 * If we've retransmitted 3 times and this is our second error, 4250 * we remove the entry. Otherwise, we allow it to continue on. 4251 * This prevents us from incorrectly nuking an entry during a 4252 * spurious network outage. 4253 * 4254 * See tcp_notify(). 4255 */ 4256 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 4257 sc->sc_flags |= SCF_UNREACH; 4258 splx(s); 4259 return; 4260 } 4261 4262 syn_cache_rm(sc); 4263 TCP_STATINC(TCP_STAT_SC_UNREACH); 4264 syn_cache_put(sc); /* calls pool_put but see spl above */ 4265 splx(s); 4266 } 4267 4268 /* 4269 * Given a LISTEN socket and an inbound SYN request, add 4270 * this to the syn cache, and send back a segment: 4271 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 4272 * to the source. 4273 * 4274 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 4275 * Doing so would require that we hold onto the data and deliver it 4276 * to the application. However, if we are the target of a SYN-flood 4277 * DoS attack, an attacker could send data which would eventually 4278 * consume all available buffer space if it were ACKed. By not ACKing 4279 * the data, we avoid this DoS scenario. 4280 */ 4281 4282 int 4283 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 4284 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 4285 int optlen, struct tcp_opt_info *oi) 4286 { 4287 struct tcpcb tb, *tp; 4288 long win; 4289 struct syn_cache *sc; 4290 struct syn_cache_head *scp; 4291 struct mbuf *ipopts; 4292 struct tcp_opt_info opti; 4293 int s; 4294 4295 tp = sototcpcb(so); 4296 4297 memset(&opti, 0, sizeof(opti)); 4298 4299 /* 4300 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 4301 * 4302 * Note this check is performed in tcp_input() very early on. 4303 */ 4304 4305 /* 4306 * Initialize some local state. 4307 */ 4308 win = sbspace(&so->so_rcv); 4309 if (win > TCP_MAXWIN) 4310 win = TCP_MAXWIN; 4311 4312 switch (src->sa_family) { 4313 #ifdef INET 4314 case AF_INET: 4315 /* 4316 * Remember the IP options, if any. 4317 */ 4318 ipopts = ip_srcroute(m); 4319 break; 4320 #endif 4321 default: 4322 ipopts = NULL; 4323 } 4324 4325 #ifdef TCP_SIGNATURE 4326 if (optp || (tp->t_flags & TF_SIGNATURE)) 4327 #else 4328 if (optp) 4329 #endif 4330 { 4331 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4332 #ifdef TCP_SIGNATURE 4333 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4334 #endif 4335 tb.t_state = TCPS_LISTEN; 4336 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 4337 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 4338 return (0); 4339 } else 4340 tb.t_flags = 0; 4341 4342 /* 4343 * See if we already have an entry for this connection. 4344 * If we do, resend the SYN,ACK. We do not count this 4345 * as a retransmission (XXX though maybe we should). 4346 */ 4347 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4348 TCP_STATINC(TCP_STAT_SC_DUPESYN); 4349 if (ipopts) { 4350 /* 4351 * If we were remembering a previous source route, 4352 * forget it and use the new one we've been given. 4353 */ 4354 if (sc->sc_ipopts) 4355 (void) m_free(sc->sc_ipopts); 4356 sc->sc_ipopts = ipopts; 4357 } 4358 sc->sc_timestamp = tb.ts_recent; 4359 if (syn_cache_respond(sc, m) == 0) { 4360 uint64_t *tcps = TCP_STAT_GETREF(); 4361 tcps[TCP_STAT_SNDACKS]++; 4362 tcps[TCP_STAT_SNDTOTAL]++; 4363 TCP_STAT_PUTREF(); 4364 } 4365 return (1); 4366 } 4367 4368 s = splsoftnet(); 4369 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4370 splx(s); 4371 if (sc == NULL) { 4372 if (ipopts) 4373 (void) m_free(ipopts); 4374 return (0); 4375 } 4376 4377 /* 4378 * Fill in the cache, and put the necessary IP and TCP 4379 * options into the reply. 4380 */ 4381 memset(sc, 0, sizeof(struct syn_cache)); 4382 callout_init(&sc->sc_timer, CALLOUT_MPSAFE); 4383 bcopy(src, &sc->sc_src, src->sa_len); 4384 bcopy(dst, &sc->sc_dst, dst->sa_len); 4385 sc->sc_flags = 0; 4386 sc->sc_ipopts = ipopts; 4387 sc->sc_irs = th->th_seq; 4388 switch (src->sa_family) { 4389 #ifdef INET 4390 case AF_INET: 4391 { 4392 struct sockaddr_in *srcin = (void *) src; 4393 struct sockaddr_in *dstin = (void *) dst; 4394 4395 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4396 &srcin->sin_addr, dstin->sin_port, 4397 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4398 break; 4399 } 4400 #endif /* INET */ 4401 #ifdef INET6 4402 case AF_INET6: 4403 { 4404 struct sockaddr_in6 *srcin6 = (void *) src; 4405 struct sockaddr_in6 *dstin6 = (void *) dst; 4406 4407 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4408 &srcin6->sin6_addr, dstin6->sin6_port, 4409 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4410 break; 4411 } 4412 #endif /* INET6 */ 4413 } 4414 sc->sc_peermaxseg = oi->maxseg; 4415 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4416 m_get_rcvif_NOMPSAFE(m) : NULL, 4417 sc->sc_src.sa.sa_family); 4418 sc->sc_win = win; 4419 sc->sc_timebase = tcp_now - 1; /* see tcp_newtcpcb() */ 4420 sc->sc_timestamp = tb.ts_recent; 4421 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4422 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4423 sc->sc_flags |= SCF_TIMESTAMP; 4424 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4425 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4426 sc->sc_requested_s_scale = tb.requested_s_scale; 4427 sc->sc_request_r_scale = 0; 4428 /* 4429 * Pick the smallest possible scaling factor that 4430 * will still allow us to scale up to sb_max. 4431 * 4432 * We do this because there are broken firewalls that 4433 * will corrupt the window scale option, leading to 4434 * the other endpoint believing that our advertised 4435 * window is unscaled. At scale factors larger than 4436 * 5 the unscaled window will drop below 1500 bytes, 4437 * leading to serious problems when traversing these 4438 * broken firewalls. 4439 * 4440 * With the default sbmax of 256K, a scale factor 4441 * of 3 will be chosen by this algorithm. Those who 4442 * choose a larger sbmax should watch out 4443 * for the compatiblity problems mentioned above. 4444 * 4445 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4446 * or <SYN,ACK>) segment itself is never scaled. 4447 */ 4448 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4449 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4450 sc->sc_request_r_scale++; 4451 } else { 4452 sc->sc_requested_s_scale = 15; 4453 sc->sc_request_r_scale = 15; 4454 } 4455 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4456 sc->sc_flags |= SCF_SACK_PERMIT; 4457 4458 /* 4459 * ECN setup packet received. 4460 */ 4461 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4462 sc->sc_flags |= SCF_ECN_PERMIT; 4463 4464 #ifdef TCP_SIGNATURE 4465 if (tb.t_flags & TF_SIGNATURE) 4466 sc->sc_flags |= SCF_SIGNATURE; 4467 #endif 4468 sc->sc_tp = tp; 4469 if (syn_cache_respond(sc, m) == 0) { 4470 uint64_t *tcps = TCP_STAT_GETREF(); 4471 tcps[TCP_STAT_SNDACKS]++; 4472 tcps[TCP_STAT_SNDTOTAL]++; 4473 TCP_STAT_PUTREF(); 4474 syn_cache_insert(sc, tp); 4475 } else { 4476 s = splsoftnet(); 4477 /* 4478 * syn_cache_put() will try to schedule the timer, so 4479 * we need to initialize it 4480 */ 4481 syn_cache_timer_arm(sc); 4482 syn_cache_put(sc); 4483 splx(s); 4484 TCP_STATINC(TCP_STAT_SC_DROPPED); 4485 } 4486 return (1); 4487 } 4488 4489 /* 4490 * syn_cache_respond: (re)send SYN+ACK. 4491 * 4492 * returns 0 on success. otherwise returns an errno, typically ENOBUFS. 4493 */ 4494 4495 int 4496 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4497 { 4498 #ifdef INET6 4499 struct rtentry *rt = NULL; 4500 #endif 4501 struct route *ro; 4502 u_int8_t *optp; 4503 int optlen, error; 4504 u_int16_t tlen; 4505 struct ip *ip = NULL; 4506 #ifdef INET6 4507 struct ip6_hdr *ip6 = NULL; 4508 #endif 4509 struct tcpcb *tp = NULL; 4510 struct tcphdr *th; 4511 u_int hlen; 4512 #ifdef TCP_SIGNATURE 4513 struct secasvar *sav = NULL; 4514 u_int8_t *sigp = NULL; 4515 #endif 4516 4517 ro = &sc->sc_route; 4518 switch (sc->sc_src.sa.sa_family) { 4519 case AF_INET: 4520 hlen = sizeof(struct ip); 4521 break; 4522 #ifdef INET6 4523 case AF_INET6: 4524 hlen = sizeof(struct ip6_hdr); 4525 break; 4526 #endif 4527 default: 4528 if (m) 4529 m_freem(m); 4530 return (EAFNOSUPPORT); 4531 } 4532 4533 /* worst case scanario, since we don't know the option size yet */ 4534 tlen = hlen + sizeof(struct tcphdr) + MAX_TCPOPTLEN; 4535 4536 /* 4537 * Create the IP+TCP header from scratch. 4538 */ 4539 if (m) 4540 m_freem(m); 4541 #ifdef DIAGNOSTIC 4542 if (max_linkhdr + tlen > MCLBYTES) 4543 return ENOBUFS; 4544 #endif 4545 4546 MGETHDR(m, M_DONTWAIT, MT_DATA); 4547 if (m && (max_linkhdr + tlen) > MHLEN) { 4548 MCLGET(m, M_DONTWAIT); 4549 if ((m->m_flags & M_EXT) == 0) { 4550 m_freem(m); 4551 m = NULL; 4552 } 4553 } 4554 if (m == NULL) 4555 return ENOBUFS; 4556 MCLAIM(m, &tcp_tx_mowner); 4557 4558 /* Fixup the mbuf. */ 4559 m->m_data += max_linkhdr; 4560 if (sc->sc_tp) 4561 tp = sc->sc_tp; 4562 m_reset_rcvif(m); 4563 memset(mtod(m, u_char *), 0, tlen); 4564 4565 switch (sc->sc_src.sa.sa_family) { 4566 case AF_INET: 4567 ip = mtod(m, struct ip *); 4568 ip->ip_v = 4; 4569 ip->ip_dst = sc->sc_src.sin.sin_addr; 4570 ip->ip_src = sc->sc_dst.sin.sin_addr; 4571 ip->ip_p = IPPROTO_TCP; 4572 th = (struct tcphdr *)(ip + 1); 4573 th->th_dport = sc->sc_src.sin.sin_port; 4574 th->th_sport = sc->sc_dst.sin.sin_port; 4575 break; 4576 #ifdef INET6 4577 case AF_INET6: 4578 ip6 = mtod(m, struct ip6_hdr *); 4579 ip6->ip6_vfc = IPV6_VERSION; 4580 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4581 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4582 ip6->ip6_nxt = IPPROTO_TCP; 4583 /* ip6_plen will be updated in ip6_output() */ 4584 th = (struct tcphdr *)(ip6 + 1); 4585 th->th_dport = sc->sc_src.sin6.sin6_port; 4586 th->th_sport = sc->sc_dst.sin6.sin6_port; 4587 break; 4588 #endif 4589 default: 4590 return ENOBUFS; 4591 } 4592 4593 th->th_seq = htonl(sc->sc_iss); 4594 th->th_ack = htonl(sc->sc_irs + 1); 4595 th->th_flags = TH_SYN|TH_ACK; 4596 th->th_win = htons(sc->sc_win); 4597 /* th_x2, th_sum, th_urp already 0 from memset */ 4598 4599 /* Tack on the TCP options. */ 4600 optp = (u_int8_t *)(th + 1); 4601 optlen = 0; 4602 *optp++ = TCPOPT_MAXSEG; 4603 *optp++ = TCPOLEN_MAXSEG; 4604 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4605 *optp++ = sc->sc_ourmaxseg & 0xff; 4606 optlen += TCPOLEN_MAXSEG; 4607 4608 if (sc->sc_request_r_scale != 15) { 4609 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4610 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4611 sc->sc_request_r_scale); 4612 optp += TCPOLEN_WINDOW + TCPOLEN_NOP; 4613 optlen += TCPOLEN_WINDOW + TCPOLEN_NOP; 4614 } 4615 4616 if (sc->sc_flags & SCF_SACK_PERMIT) { 4617 /* Let the peer know that we will SACK. */ 4618 *optp++ = TCPOPT_SACK_PERMITTED; 4619 *optp++ = TCPOLEN_SACK_PERMITTED; 4620 optlen += TCPOLEN_SACK_PERMITTED; 4621 } 4622 4623 if (sc->sc_flags & SCF_TIMESTAMP) { 4624 while (optlen % 4 != 2) { 4625 optlen += TCPOLEN_NOP; 4626 *optp++ = TCPOPT_NOP; 4627 } 4628 *optp++ = TCPOPT_TIMESTAMP; 4629 *optp++ = TCPOLEN_TIMESTAMP; 4630 u_int32_t *lp = (u_int32_t *)(optp); 4631 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4632 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4633 *lp = htonl(sc->sc_timestamp); 4634 optp += TCPOLEN_TIMESTAMP - 2; 4635 optlen += TCPOLEN_TIMESTAMP; 4636 } 4637 4638 #ifdef TCP_SIGNATURE 4639 if (sc->sc_flags & SCF_SIGNATURE) { 4640 4641 sav = tcp_signature_getsav(m, th); 4642 4643 if (sav == NULL) { 4644 if (m) 4645 m_freem(m); 4646 return (EPERM); 4647 } 4648 4649 *optp++ = TCPOPT_SIGNATURE; 4650 *optp++ = TCPOLEN_SIGNATURE; 4651 sigp = optp; 4652 memset(optp, 0, TCP_SIGLEN); 4653 optp += TCP_SIGLEN; 4654 optlen += TCPOLEN_SIGNATURE; 4655 4656 } 4657 #endif 4658 /* Terminate and pad TCP options to a 4 byte boundary. */ 4659 if (optlen % 4) { 4660 optlen += TCPOLEN_EOL; 4661 *optp++ = TCPOPT_EOL; 4662 } 4663 /* 4664 * According to RFC 793 (STD0007): 4665 * "The content of the header beyond the End-of-Option option 4666 * must be header padding (i.e., zero)." 4667 * and later: "The padding is composed of zeros." 4668 */ 4669 while (optlen % 4) { 4670 optlen += TCPOLEN_PAD; 4671 *optp++ = TCPOPT_PAD; 4672 } 4673 4674 /* compute the actual values now that we've added the options */ 4675 tlen = hlen + sizeof(struct tcphdr) + optlen; 4676 m->m_len = m->m_pkthdr.len = tlen; 4677 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4678 4679 #ifdef TCP_SIGNATURE 4680 if (sav) { 4681 (void)tcp_signature(m, th, hlen, sav, sigp); 4682 key_sa_recordxfer(sav, m); 4683 KEY_SA_UNREF(&sav); 4684 } 4685 #endif 4686 4687 /* 4688 * Send ECN SYN-ACK setup packet. 4689 * Routes can be asymetric, so, even if we receive a packet 4690 * with ECE and CWR set, we must not assume no one will block 4691 * the ECE packet we are about to send. 4692 */ 4693 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4694 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4695 th->th_flags |= TH_ECE; 4696 TCP_STATINC(TCP_STAT_ECN_SHS); 4697 4698 /* 4699 * draft-ietf-tcpm-ecnsyn-00.txt 4700 * 4701 * "[...] a TCP node MAY respond to an ECN-setup 4702 * SYN packet by setting ECT in the responding 4703 * ECN-setup SYN/ACK packet, indicating to routers 4704 * that the SYN/ACK packet is ECN-Capable. 4705 * This allows a congested router along the path 4706 * to mark the packet instead of dropping the 4707 * packet as an indication of congestion." 4708 * 4709 * "[...] There can be a great benefit in setting 4710 * an ECN-capable codepoint in SYN/ACK packets [...] 4711 * Congestion is most likely to occur in 4712 * the server-to-client direction. As a result, 4713 * setting an ECN-capable codepoint in SYN/ACK 4714 * packets can reduce the occurence of three-second 4715 * retransmit timeouts resulting from the drop 4716 * of SYN/ACK packets." 4717 * 4718 * Page 4 and 6, January 2006. 4719 */ 4720 4721 switch (sc->sc_src.sa.sa_family) { 4722 #ifdef INET 4723 case AF_INET: 4724 ip->ip_tos |= IPTOS_ECN_ECT0; 4725 break; 4726 #endif 4727 #ifdef INET6 4728 case AF_INET6: 4729 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4730 break; 4731 #endif 4732 } 4733 TCP_STATINC(TCP_STAT_ECN_ECT); 4734 } 4735 4736 4737 /* Compute the packet's checksum. */ 4738 switch (sc->sc_src.sa.sa_family) { 4739 case AF_INET: 4740 ip->ip_len = htons(tlen - hlen); 4741 th->th_sum = 0; 4742 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4743 break; 4744 #ifdef INET6 4745 case AF_INET6: 4746 ip6->ip6_plen = htons(tlen - hlen); 4747 th->th_sum = 0; 4748 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4749 break; 4750 #endif 4751 } 4752 4753 /* 4754 * Fill in some straggling IP bits. Note the stack expects 4755 * ip_len to be in host order, for convenience. 4756 */ 4757 switch (sc->sc_src.sa.sa_family) { 4758 #ifdef INET 4759 case AF_INET: 4760 ip->ip_len = htons(tlen); 4761 ip->ip_ttl = ip_defttl; 4762 /* XXX tos? */ 4763 break; 4764 #endif 4765 #ifdef INET6 4766 case AF_INET6: 4767 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4768 ip6->ip6_vfc |= IPV6_VERSION; 4769 ip6->ip6_plen = htons(tlen - hlen); 4770 /* ip6_hlim will be initialized afterwards */ 4771 /* XXX flowlabel? */ 4772 break; 4773 #endif 4774 } 4775 4776 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4777 tp = sc->sc_tp; 4778 4779 switch (sc->sc_src.sa.sa_family) { 4780 #ifdef INET 4781 case AF_INET: 4782 error = ip_output(m, sc->sc_ipopts, ro, 4783 (ip_mtudisc ? IP_MTUDISC : 0), 4784 NULL, tp ? tp->t_inpcb : NULL); 4785 break; 4786 #endif 4787 #ifdef INET6 4788 case AF_INET6: 4789 ip6->ip6_hlim = in6_selecthlim(NULL, 4790 (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp : NULL); 4791 rtcache_unref(rt, ro); 4792 4793 error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, 4794 tp ? tp->t_in6pcb : NULL, NULL); 4795 break; 4796 #endif 4797 default: 4798 error = EAFNOSUPPORT; 4799 break; 4800 } 4801 return (error); 4802 } 4803
Indexes created Tue Sep 23 07:09:52 GMT 2025