tcp_input.c revision 1.368
1 /* $NetBSD: tcp_input.c,v 1.368 2018/02/08 19:38:21 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 1.368 2018/02/08 19:38:21 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 /* 1278 * Get IP and TCP header. 1279 * Note: IP leaves IP header in first mbuf. 1280 */ 1281 ip = mtod(m, struct ip *); 1282 switch (ip->ip_v) { 1283 #ifdef INET 1284 case 4: 1285 #ifdef INET6 1286 ip6 = NULL; 1287 #endif 1288 af = AF_INET; 1289 iphlen = sizeof(struct ip); 1290 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1291 sizeof(struct tcphdr)); 1292 if (th == NULL) { 1293 TCP_STATINC(TCP_STAT_RCVSHORT); 1294 return; 1295 } 1296 /* We do the checksum after PCB lookup... */ 1297 len = ntohs(ip->ip_len); 1298 tlen = len - toff; 1299 iptos = ip->ip_tos; 1300 break; 1301 #endif 1302 #ifdef INET6 1303 case 6: 1304 ip = NULL; 1305 iphlen = sizeof(struct ip6_hdr); 1306 af = AF_INET6; 1307 ip6 = mtod(m, struct ip6_hdr *); 1308 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1309 sizeof(struct tcphdr)); 1310 if (th == NULL) { 1311 TCP_STATINC(TCP_STAT_RCVSHORT); 1312 return; 1313 } 1314 1315 /* Be proactive about malicious use of IPv4 mapped address */ 1316 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 1317 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 1318 /* XXX stat */ 1319 goto drop; 1320 } 1321 1322 /* 1323 * Be proactive about unspecified IPv6 address in source. 1324 * As we use all-zero to indicate unbounded/unconnected pcb, 1325 * unspecified IPv6 address can be used to confuse us. 1326 * 1327 * Note that packets with unspecified IPv6 destination is 1328 * already dropped in ip6_input. 1329 */ 1330 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1331 /* XXX stat */ 1332 goto drop; 1333 } 1334 1335 /* 1336 * Make sure destination address is not multicast. 1337 * Source address checked in ip6_input(). 1338 */ 1339 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1340 /* XXX stat */ 1341 goto drop; 1342 } 1343 1344 /* We do the checksum after PCB lookup... */ 1345 len = m->m_pkthdr.len; 1346 tlen = len - toff; 1347 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1348 break; 1349 #endif 1350 default: 1351 m_freem(m); 1352 return; 1353 } 1354 1355 /* 1356 * Enforce alignment requirements that are violated in 1357 * some cases, see kern/50766 for details. 1358 */ 1359 if (TCP_HDR_ALIGNED_P(th) == 0) { 1360 m = m_copyup(m, toff + sizeof(struct tcphdr), 0); 1361 if (m == NULL) { 1362 TCP_STATINC(TCP_STAT_RCVSHORT); 1363 return; 1364 } 1365 ip = mtod(m, struct ip *); 1366 #ifdef INET6 1367 ip6 = mtod(m, struct ip6_hdr *); 1368 #endif 1369 th = (struct tcphdr *)(mtod(m, char *) + toff); 1370 } 1371 KASSERT(TCP_HDR_ALIGNED_P(th)); 1372 1373 /* 1374 * Check that TCP offset makes sense, pull out TCP options and 1375 * adjust length. 1376 */ 1377 off = th->th_off << 2; 1378 if (off < sizeof(struct tcphdr) || off > tlen) { 1379 TCP_STATINC(TCP_STAT_RCVBADOFF); 1380 goto drop; 1381 } 1382 tlen -= off; 1383 1384 if (off > sizeof(struct tcphdr)) { 1385 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1386 if (th == NULL) { 1387 TCP_STATINC(TCP_STAT_RCVSHORT); 1388 return; 1389 } 1390 KASSERT(TCP_HDR_ALIGNED_P(th)); 1391 optlen = off - sizeof (struct tcphdr); 1392 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1393 /* 1394 * Do quick retrieval of timestamp options ("options 1395 * prediction?"). If timestamp is the only option and it's 1396 * formatted as recommended in RFC 1323 appendix A, we 1397 * quickly get the values now and not bother calling 1398 * tcp_dooptions(), etc. 1399 */ 1400 if ((optlen == TCPOLEN_TSTAMP_APPA || 1401 (optlen > TCPOLEN_TSTAMP_APPA && 1402 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1403 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1404 (th->th_flags & TH_SYN) == 0) { 1405 opti.ts_present = 1; 1406 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1407 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1408 optp = NULL; /* we've parsed the options */ 1409 } 1410 } 1411 tiflags = th->th_flags; 1412 1413 /* 1414 * Checksum extended TCP header and data 1415 */ 1416 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1417 goto badcsum; 1418 1419 /* 1420 * Locate pcb for segment. 1421 */ 1422 findpcb: 1423 inp = NULL; 1424 #ifdef INET6 1425 in6p = NULL; 1426 #endif 1427 switch (af) { 1428 #ifdef INET 1429 case AF_INET: 1430 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1431 ip->ip_dst, th->th_dport, &vestige); 1432 if (inp == 0 && !vestige.valid) { 1433 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1434 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, 1435 th->th_dport); 1436 } 1437 #ifdef INET6 1438 if (inp == 0 && !vestige.valid) { 1439 struct in6_addr s, d; 1440 1441 /* mapped addr case */ 1442 in6_in_2_v4mapin6(&ip->ip_src, &s); 1443 in6_in_2_v4mapin6(&ip->ip_dst, &d); 1444 in6p = in6_pcblookup_connect(&tcbtable, &s, 1445 th->th_sport, &d, th->th_dport, 0, &vestige); 1446 if (in6p == 0 && !vestige.valid) { 1447 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1448 in6p = in6_pcblookup_bind(&tcbtable, &d, 1449 th->th_dport, 0); 1450 } 1451 } 1452 #endif 1453 #ifndef INET6 1454 if (inp == 0 && !vestige.valid) 1455 #else 1456 if (inp == 0 && in6p == 0 && !vestige.valid) 1457 #endif 1458 { 1459 TCP_STATINC(TCP_STAT_NOPORT); 1460 if (tcp_log_refused && 1461 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1462 tcp4_log_refused(ip, th); 1463 } 1464 tcp_fields_to_host(th); 1465 goto dropwithreset_ratelim; 1466 } 1467 #if defined(IPSEC) 1468 if (ipsec_used) { 1469 if (inp && 1470 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 1471 && ipsec4_in_reject(m, inp)) { 1472 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1473 goto drop; 1474 } 1475 #ifdef INET6 1476 else if (in6p && 1477 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 1478 && ipsec6_in_reject(m, in6p)) { 1479 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1480 goto drop; 1481 } 1482 #endif 1483 } 1484 #endif /*IPSEC*/ 1485 break; 1486 #endif /*INET*/ 1487 #ifdef INET6 1488 case AF_INET6: 1489 { 1490 int faith; 1491 1492 #if defined(NFAITH) && NFAITH > 0 1493 faith = faithprefix(&ip6->ip6_dst); 1494 #else 1495 faith = 0; 1496 #endif 1497 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1498 th->th_sport, &ip6->ip6_dst, th->th_dport, faith, &vestige); 1499 if (!in6p && !vestige.valid) { 1500 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1501 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1502 th->th_dport, faith); 1503 } 1504 if (!in6p && !vestige.valid) { 1505 TCP_STATINC(TCP_STAT_NOPORT); 1506 if (tcp_log_refused && 1507 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1508 tcp6_log_refused(ip6, th); 1509 } 1510 tcp_fields_to_host(th); 1511 goto dropwithreset_ratelim; 1512 } 1513 #if defined(IPSEC) 1514 if (ipsec_used && in6p 1515 && (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 1516 && ipsec6_in_reject(m, in6p)) { 1517 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1518 goto drop; 1519 } 1520 #endif /*IPSEC*/ 1521 break; 1522 } 1523 #endif 1524 } 1525 1526 /* 1527 * If the state is CLOSED (i.e., TCB does not exist) then 1528 * all data in the incoming segment is discarded. 1529 * If the TCB exists but is in CLOSED state, it is embryonic, 1530 * but should either do a listen or a connect soon. 1531 */ 1532 tp = NULL; 1533 so = NULL; 1534 if (inp) { 1535 /* Check the minimum TTL for socket. */ 1536 if (ip->ip_ttl < inp->inp_ip_minttl) 1537 goto drop; 1538 1539 tp = intotcpcb(inp); 1540 so = inp->inp_socket; 1541 } 1542 #ifdef INET6 1543 else if (in6p) { 1544 tp = in6totcpcb(in6p); 1545 so = in6p->in6p_socket; 1546 } 1547 #endif 1548 else if (vestige.valid) { 1549 int mc = 0; 1550 1551 /* We do not support the resurrection of vtw tcpcps. */ 1552 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1553 goto badcsum; 1554 1555 switch (af) { 1556 #ifdef INET6 1557 case AF_INET6: 1558 mc = IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst); 1559 break; 1560 #endif 1561 1562 case AF_INET: 1563 mc = (IN_MULTICAST(ip->ip_dst.s_addr) || 1564 in_broadcast(ip->ip_dst, m_get_rcvif_NOMPSAFE(m))); 1565 break; 1566 } 1567 1568 tcp_fields_to_host(th); 1569 tcp_vtw_input(th, &vestige, m, tlen, mc); 1570 m = NULL; 1571 goto drop; 1572 } 1573 1574 if (tp == NULL) { 1575 tcp_fields_to_host(th); 1576 goto dropwithreset_ratelim; 1577 } 1578 if (tp->t_state == TCPS_CLOSED) 1579 goto drop; 1580 1581 KASSERT(so->so_lock == softnet_lock); 1582 KASSERT(solocked(so)); 1583 1584 tcp_fields_to_host(th); 1585 1586 /* Unscale the window into a 32-bit value. */ 1587 if ((tiflags & TH_SYN) == 0) 1588 tiwin = th->th_win << tp->snd_scale; 1589 else 1590 tiwin = th->th_win; 1591 1592 #ifdef INET6 1593 /* save packet options if user wanted */ 1594 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1595 if (in6p->in6p_options) { 1596 m_freem(in6p->in6p_options); 1597 in6p->in6p_options = 0; 1598 } 1599 KASSERT(ip6 != NULL); 1600 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1601 } 1602 #endif 1603 1604 if (so->so_options & SO_DEBUG) { 1605 #ifdef TCP_DEBUG 1606 ostate = tp->t_state; 1607 #endif 1608 1609 tcp_saveti = NULL; 1610 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1611 goto nosave; 1612 1613 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1614 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1615 if (!tcp_saveti) 1616 goto nosave; 1617 } else { 1618 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1619 if (!tcp_saveti) 1620 goto nosave; 1621 MCLAIM(m, &tcp_mowner); 1622 tcp_saveti->m_len = iphlen; 1623 m_copydata(m, 0, iphlen, 1624 mtod(tcp_saveti, void *)); 1625 } 1626 1627 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1628 m_freem(tcp_saveti); 1629 tcp_saveti = NULL; 1630 } else { 1631 tcp_saveti->m_len += sizeof(struct tcphdr); 1632 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1633 sizeof(struct tcphdr)); 1634 } 1635 nosave:; 1636 } 1637 1638 if (so->so_options & SO_ACCEPTCONN) { 1639 union syn_cache_sa src; 1640 union syn_cache_sa dst; 1641 1642 memset(&src, 0, sizeof(src)); 1643 memset(&dst, 0, sizeof(dst)); 1644 switch (af) { 1645 #ifdef INET 1646 case AF_INET: 1647 src.sin.sin_len = sizeof(struct sockaddr_in); 1648 src.sin.sin_family = AF_INET; 1649 src.sin.sin_addr = ip->ip_src; 1650 src.sin.sin_port = th->th_sport; 1651 1652 dst.sin.sin_len = sizeof(struct sockaddr_in); 1653 dst.sin.sin_family = AF_INET; 1654 dst.sin.sin_addr = ip->ip_dst; 1655 dst.sin.sin_port = th->th_dport; 1656 break; 1657 #endif 1658 #ifdef INET6 1659 case AF_INET6: 1660 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1661 src.sin6.sin6_family = AF_INET6; 1662 src.sin6.sin6_addr = ip6->ip6_src; 1663 src.sin6.sin6_port = th->th_sport; 1664 1665 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1666 dst.sin6.sin6_family = AF_INET6; 1667 dst.sin6.sin6_addr = ip6->ip6_dst; 1668 dst.sin6.sin6_port = th->th_dport; 1669 break; 1670 #endif /* INET6 */ 1671 default: 1672 goto badsyn; /*sanity*/ 1673 } 1674 1675 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1676 if (tiflags & TH_RST) { 1677 syn_cache_reset(&src.sa, &dst.sa, th); 1678 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1679 (TH_ACK|TH_SYN)) { 1680 /* 1681 * Received a SYN,ACK. This should 1682 * never happen while we are in 1683 * LISTEN. Send an RST. 1684 */ 1685 goto badsyn; 1686 } else if (tiflags & TH_ACK) { 1687 so = syn_cache_get(&src.sa, &dst.sa, 1688 th, toff, tlen, so, m); 1689 if (so == NULL) { 1690 /* 1691 * We don't have a SYN for this ACK; 1692 * send an RST. 1693 */ 1694 goto badsyn; 1695 } else if (so == (struct socket *)(-1)) { 1696 /* 1697 * We were unable to create the 1698 * connection. If the 3-way handshake 1699 * was completed, and RST has been 1700 * sent to the peer. Since the mbuf 1701 * might be in use for the reply, do 1702 * not free it. 1703 */ 1704 m = NULL; 1705 } else { 1706 /* 1707 * We have created a full-blown 1708 * connection. 1709 */ 1710 tp = NULL; 1711 inp = NULL; 1712 #ifdef INET6 1713 in6p = NULL; 1714 #endif 1715 switch (so->so_proto->pr_domain->dom_family) { 1716 #ifdef INET 1717 case AF_INET: 1718 inp = sotoinpcb(so); 1719 tp = intotcpcb(inp); 1720 break; 1721 #endif 1722 #ifdef INET6 1723 case AF_INET6: 1724 in6p = sotoin6pcb(so); 1725 tp = in6totcpcb(in6p); 1726 break; 1727 #endif 1728 } 1729 if (tp == NULL) 1730 goto badsyn; /*XXX*/ 1731 tiwin <<= tp->snd_scale; 1732 goto after_listen; 1733 } 1734 } else { 1735 /* 1736 * None of RST, SYN or ACK was set. 1737 * This is an invalid packet for a 1738 * TCB in LISTEN state. Send a RST. 1739 */ 1740 goto badsyn; 1741 } 1742 } else { 1743 /* 1744 * Received a SYN. 1745 */ 1746 1747 switch (af) { 1748 #ifdef INET6 1749 case AF_INET6: 1750 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1751 goto drop; 1752 break; 1753 #endif /* INET6 */ 1754 case AF_INET: 1755 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1756 in_broadcast(ip->ip_dst, 1757 m_get_rcvif_NOMPSAFE(m))) 1758 goto drop; 1759 break; 1760 } 1761 1762 #ifdef INET6 1763 /* 1764 * If deprecated address is forbidden, we do 1765 * not accept SYN to deprecated interface 1766 * address to prevent any new inbound 1767 * connection from getting established. 1768 * When we do not accept SYN, we send a TCP 1769 * RST, with deprecated source address (instead 1770 * of dropping it). We compromise it as it is 1771 * much better for peer to send a RST, and 1772 * RST will be the final packet for the 1773 * exchange. 1774 * 1775 * If we do not forbid deprecated addresses, we 1776 * accept the SYN packet. RFC2462 does not 1777 * suggest dropping SYN in this case. 1778 * If we decipher RFC2462 5.5.4, it says like 1779 * this: 1780 * 1. use of deprecated addr with existing 1781 * communication is okay - "SHOULD continue 1782 * to be used" 1783 * 2. use of it with new communication: 1784 * (2a) "SHOULD NOT be used if alternate 1785 * address with sufficient scope is 1786 * available" 1787 * (2b) nothing mentioned otherwise. 1788 * Here we fall into (2b) case as we have no 1789 * choice in our source address selection - we 1790 * must obey the peer. 1791 * 1792 * The wording in RFC2462 is confusing, and 1793 * there are multiple description text for 1794 * deprecated address handling - worse, they 1795 * are not exactly the same. I believe 5.5.4 1796 * is the best one, so we follow 5.5.4. 1797 */ 1798 if (af == AF_INET6 && !ip6_use_deprecated) { 1799 struct in6_ifaddr *ia6; 1800 int s; 1801 struct ifnet *rcvif = m_get_rcvif(m, &s); 1802 if (rcvif == NULL) 1803 goto dropwithreset; /* XXX */ 1804 if ((ia6 = in6ifa_ifpwithaddr(rcvif, 1805 &ip6->ip6_dst)) && 1806 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1807 tp = NULL; 1808 m_put_rcvif(rcvif, &s); 1809 goto dropwithreset; 1810 } 1811 m_put_rcvif(rcvif, &s); 1812 } 1813 #endif 1814 1815 #if defined(IPSEC) 1816 if (ipsec_used) { 1817 switch (af) { 1818 #ifdef INET 1819 case AF_INET: 1820 /* 1821 * inp can be NULL when receiving an 1822 * IPv4 packet on an IPv4-mapped IPv6 1823 * address. 1824 */ 1825 KASSERT(inp == NULL || 1826 sotoinpcb(so) == inp); 1827 if (!ipsec4_in_reject(m, inp)) 1828 break; 1829 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1830 tp = NULL; 1831 goto dropwithreset; 1832 #endif 1833 #ifdef INET6 1834 case AF_INET6: 1835 KASSERT(sotoin6pcb(so) == in6p); 1836 if (!ipsec6_in_reject(m, in6p)) 1837 break; 1838 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1839 tp = NULL; 1840 goto dropwithreset; 1841 #endif /*INET6*/ 1842 } 1843 } 1844 #endif /*IPSEC*/ 1845 1846 /* 1847 * LISTEN socket received a SYN from itself? This 1848 * can't possibly be valid; drop the packet. 1849 */ 1850 if (th->th_sport == th->th_dport) { 1851 int i; 1852 1853 switch (af) { 1854 #ifdef INET 1855 case AF_INET: 1856 i = in_hosteq(ip->ip_src, ip->ip_dst); 1857 break; 1858 #endif 1859 #ifdef INET6 1860 case AF_INET6: 1861 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, 1862 &ip6->ip6_dst); 1863 break; 1864 #endif 1865 default: 1866 i = 1; 1867 } 1868 if (i) { 1869 TCP_STATINC(TCP_STAT_BADSYN); 1870 goto drop; 1871 } 1872 } 1873 1874 /* 1875 * SYN looks ok; create compressed TCP 1876 * state for it. 1877 */ 1878 if (so->so_qlen <= so->so_qlimit && 1879 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1880 so, m, optp, optlen, &opti)) 1881 m = NULL; 1882 } 1883 1884 goto drop; 1885 } 1886 1887 after_listen: 1888 /* 1889 * Should not happen now that all embryonic connections 1890 * are handled with compressed state. 1891 */ 1892 KASSERT(tp->t_state != TCPS_LISTEN); 1893 1894 /* 1895 * Segment received on connection. 1896 * Reset idle time and keep-alive timer. 1897 */ 1898 tp->t_rcvtime = tcp_now; 1899 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1900 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1901 1902 /* 1903 * Process options. 1904 */ 1905 #ifdef TCP_SIGNATURE 1906 if (optp || (tp->t_flags & TF_SIGNATURE)) 1907 #else 1908 if (optp) 1909 #endif 1910 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1911 goto drop; 1912 1913 if (TCP_SACK_ENABLED(tp)) { 1914 tcp_del_sackholes(tp, th); 1915 } 1916 1917 if (TCP_ECN_ALLOWED(tp)) { 1918 if (tiflags & TH_CWR) { 1919 tp->t_flags &= ~TF_ECN_SND_ECE; 1920 } 1921 switch (iptos & IPTOS_ECN_MASK) { 1922 case IPTOS_ECN_CE: 1923 tp->t_flags |= TF_ECN_SND_ECE; 1924 TCP_STATINC(TCP_STAT_ECN_CE); 1925 break; 1926 case IPTOS_ECN_ECT0: 1927 TCP_STATINC(TCP_STAT_ECN_ECT); 1928 break; 1929 case IPTOS_ECN_ECT1: 1930 /* XXX: ignore for now -- rpaulo */ 1931 break; 1932 } 1933 /* 1934 * Congestion experienced. 1935 * Ignore if we are already trying to recover. 1936 */ 1937 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1938 tp->t_congctl->cong_exp(tp); 1939 } 1940 1941 if (opti.ts_present && opti.ts_ecr) { 1942 /* 1943 * Calculate the RTT from the returned time stamp and the 1944 * connection's time base. If the time stamp is later than 1945 * the current time, or is extremely old, fall back to non-1323 1946 * RTT calculation. Since ts_rtt is unsigned, we can test both 1947 * at the same time. 1948 * 1949 * Note that ts_rtt is in units of slow ticks (500 1950 * ms). Since most earthbound RTTs are < 500 ms, 1951 * observed values will have large quantization noise. 1952 * Our smoothed RTT is then the fraction of observed 1953 * samples that are 1 tick instead of 0 (times 500 1954 * ms). 1955 * 1956 * ts_rtt is increased by 1 to denote a valid sample, 1957 * with 0 indicating an invalid measurement. This 1958 * extra 1 must be removed when ts_rtt is used, or 1959 * else an an erroneous extra 500 ms will result. 1960 */ 1961 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1962 if (ts_rtt > TCP_PAWS_IDLE) 1963 ts_rtt = 0; 1964 } else { 1965 ts_rtt = 0; 1966 } 1967 1968 /* 1969 * Header prediction: check for the two common cases 1970 * of a uni-directional data xfer. If the packet has 1971 * no control flags, is in-sequence, the window didn't 1972 * change and we're not retransmitting, it's a 1973 * candidate. If the length is zero and the ack moved 1974 * forward, we're the sender side of the xfer. Just 1975 * free the data acked & wake any higher level process 1976 * that was blocked waiting for space. If the length 1977 * is non-zero and the ack didn't move, we're the 1978 * receiver side. If we're getting packets in-order 1979 * (the reassembly queue is empty), add the data to 1980 * the socket buffer and note that we need a delayed ack. 1981 */ 1982 if (tp->t_state == TCPS_ESTABLISHED && 1983 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1984 == TH_ACK && 1985 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1986 th->th_seq == tp->rcv_nxt && 1987 tiwin && tiwin == tp->snd_wnd && 1988 tp->snd_nxt == tp->snd_max) { 1989 1990 /* 1991 * If last ACK falls within this segment's sequence numbers, 1992 * record the timestamp. 1993 * NOTE that the test is modified according to the latest 1994 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 1995 * 1996 * note that we already know 1997 * TSTMP_GEQ(opti.ts_val, tp->ts_recent) 1998 */ 1999 if (opti.ts_present && 2000 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 2001 tp->ts_recent_age = tcp_now; 2002 tp->ts_recent = opti.ts_val; 2003 } 2004 2005 if (tlen == 0) { 2006 /* Ack prediction. */ 2007 if (SEQ_GT(th->th_ack, tp->snd_una) && 2008 SEQ_LEQ(th->th_ack, tp->snd_max) && 2009 tp->snd_cwnd >= tp->snd_wnd && 2010 tp->t_partialacks < 0) { 2011 /* 2012 * this is a pure ack for outstanding data. 2013 */ 2014 if (ts_rtt) 2015 tcp_xmit_timer(tp, ts_rtt - 1); 2016 else if (tp->t_rtttime && 2017 SEQ_GT(th->th_ack, tp->t_rtseq)) 2018 tcp_xmit_timer(tp, 2019 tcp_now - tp->t_rtttime); 2020 acked = th->th_ack - tp->snd_una; 2021 tcps = TCP_STAT_GETREF(); 2022 tcps[TCP_STAT_PREDACK]++; 2023 tcps[TCP_STAT_RCVACKPACK]++; 2024 tcps[TCP_STAT_RCVACKBYTE] += acked; 2025 TCP_STAT_PUTREF(); 2026 nd6_hint(tp); 2027 2028 if (acked > (tp->t_lastoff - tp->t_inoff)) 2029 tp->t_lastm = NULL; 2030 sbdrop(&so->so_snd, acked); 2031 tp->t_lastoff -= acked; 2032 2033 icmp_check(tp, th, acked); 2034 2035 tp->snd_una = th->th_ack; 2036 tp->snd_fack = tp->snd_una; 2037 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2038 tp->snd_high = tp->snd_una; 2039 m_freem(m); 2040 2041 /* 2042 * If all outstanding data are acked, stop 2043 * retransmit timer, otherwise restart timer 2044 * using current (possibly backed-off) value. 2045 * If process is waiting for space, 2046 * wakeup/selnotify/signal. If data 2047 * are ready to send, let tcp_output 2048 * decide between more output or persist. 2049 */ 2050 if (tp->snd_una == tp->snd_max) 2051 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2052 else if (TCP_TIMER_ISARMED(tp, 2053 TCPT_PERSIST) == 0) 2054 TCP_TIMER_ARM(tp, TCPT_REXMT, 2055 tp->t_rxtcur); 2056 2057 sowwakeup(so); 2058 if (so->so_snd.sb_cc) { 2059 KERNEL_LOCK(1, NULL); 2060 (void) tcp_output(tp); 2061 KERNEL_UNLOCK_ONE(NULL); 2062 } 2063 if (tcp_saveti) 2064 m_freem(tcp_saveti); 2065 return; 2066 } 2067 } else if (th->th_ack == tp->snd_una && 2068 TAILQ_FIRST(&tp->segq) == NULL && 2069 tlen <= sbspace(&so->so_rcv)) { 2070 int newsize = 0; /* automatic sockbuf scaling */ 2071 2072 /* 2073 * this is a pure, in-sequence data packet 2074 * with nothing on the reassembly queue and 2075 * we have enough buffer space to take it. 2076 */ 2077 tp->rcv_nxt += tlen; 2078 tcps = TCP_STAT_GETREF(); 2079 tcps[TCP_STAT_PREDDAT]++; 2080 tcps[TCP_STAT_RCVPACK]++; 2081 tcps[TCP_STAT_RCVBYTE] += tlen; 2082 TCP_STAT_PUTREF(); 2083 nd6_hint(tp); 2084 2085 /* 2086 * Automatic sizing enables the performance of large buffers 2087 * and most of the efficiency of small ones by only allocating 2088 * space when it is needed. 2089 * 2090 * On the receive side the socket buffer memory is only rarely 2091 * used to any significant extent. This allows us to be much 2092 * more aggressive in scaling the receive socket buffer. For 2093 * the case that the buffer space is actually used to a large 2094 * extent and we run out of kernel memory we can simply drop 2095 * the new segments; TCP on the sender will just retransmit it 2096 * later. Setting the buffer size too big may only consume too 2097 * much kernel memory if the application doesn't read() from 2098 * the socket or packet loss or reordering makes use of the 2099 * reassembly queue. 2100 * 2101 * The criteria to step up the receive buffer one notch are: 2102 * 1. the number of bytes received during the time it takes 2103 * one timestamp to be reflected back to us (the RTT); 2104 * 2. received bytes per RTT is within seven eighth of the 2105 * current socket buffer size; 2106 * 3. receive buffer size has not hit maximal automatic size; 2107 * 2108 * This algorithm does one step per RTT at most and only if 2109 * we receive a bulk stream w/o packet losses or reorderings. 2110 * Shrinking the buffer during idle times is not necessary as 2111 * it doesn't consume any memory when idle. 2112 * 2113 * TODO: Only step up if the application is actually serving 2114 * the buffer to better manage the socket buffer resources. 2115 */ 2116 if (tcp_do_autorcvbuf && 2117 opti.ts_ecr && 2118 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 2119 if (opti.ts_ecr > tp->rfbuf_ts && 2120 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 2121 if (tp->rfbuf_cnt > 2122 (so->so_rcv.sb_hiwat / 8 * 7) && 2123 so->so_rcv.sb_hiwat < 2124 tcp_autorcvbuf_max) { 2125 newsize = 2126 min(so->so_rcv.sb_hiwat + 2127 tcp_autorcvbuf_inc, 2128 tcp_autorcvbuf_max); 2129 } 2130 /* Start over with next RTT. */ 2131 tp->rfbuf_ts = 0; 2132 tp->rfbuf_cnt = 0; 2133 } else 2134 tp->rfbuf_cnt += tlen; /* add up */ 2135 } 2136 2137 /* 2138 * Drop TCP, IP headers and TCP options then add data 2139 * to socket buffer. 2140 */ 2141 if (so->so_state & SS_CANTRCVMORE) 2142 m_freem(m); 2143 else { 2144 /* 2145 * Set new socket buffer size. 2146 * Give up when limit is reached. 2147 */ 2148 if (newsize) 2149 if (!sbreserve(&so->so_rcv, 2150 newsize, so)) 2151 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 2152 m_adj(m, toff + off); 2153 sbappendstream(&so->so_rcv, m); 2154 } 2155 sorwakeup(so); 2156 tcp_setup_ack(tp, th); 2157 if (tp->t_flags & TF_ACKNOW) { 2158 KERNEL_LOCK(1, NULL); 2159 (void) tcp_output(tp); 2160 KERNEL_UNLOCK_ONE(NULL); 2161 } 2162 if (tcp_saveti) 2163 m_freem(tcp_saveti); 2164 return; 2165 } 2166 } 2167 2168 /* 2169 * Compute mbuf offset to TCP data segment. 2170 */ 2171 hdroptlen = toff + off; 2172 2173 /* 2174 * Calculate amount of space in receive window, 2175 * and then do TCP input processing. 2176 * Receive window is amount of space in rcv queue, 2177 * but not less than advertised window. 2178 */ 2179 { int win; 2180 2181 win = sbspace(&so->so_rcv); 2182 if (win < 0) 2183 win = 0; 2184 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 2185 } 2186 2187 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 2188 tp->rfbuf_ts = 0; 2189 tp->rfbuf_cnt = 0; 2190 2191 switch (tp->t_state) { 2192 /* 2193 * If the state is SYN_SENT: 2194 * if seg contains an ACK, but not for our SYN, drop the input. 2195 * if seg contains a RST, then drop the connection. 2196 * if seg does not contain SYN, then drop it. 2197 * Otherwise this is an acceptable SYN segment 2198 * initialize tp->rcv_nxt and tp->irs 2199 * if seg contains ack then advance tp->snd_una 2200 * if seg contains a ECE and ECN support is enabled, the stream 2201 * is ECN capable. 2202 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2203 * arrange for segment to be acked (eventually) 2204 * continue processing rest of data/controls, beginning with URG 2205 */ 2206 case TCPS_SYN_SENT: 2207 if ((tiflags & TH_ACK) && 2208 (SEQ_LEQ(th->th_ack, tp->iss) || 2209 SEQ_GT(th->th_ack, tp->snd_max))) 2210 goto dropwithreset; 2211 if (tiflags & TH_RST) { 2212 if (tiflags & TH_ACK) 2213 tp = tcp_drop(tp, ECONNREFUSED); 2214 goto drop; 2215 } 2216 if ((tiflags & TH_SYN) == 0) 2217 goto drop; 2218 if (tiflags & TH_ACK) { 2219 tp->snd_una = th->th_ack; 2220 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2221 tp->snd_nxt = tp->snd_una; 2222 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2223 tp->snd_high = tp->snd_una; 2224 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2225 2226 if ((tiflags & TH_ECE) && tcp_do_ecn) { 2227 tp->t_flags |= TF_ECN_PERMIT; 2228 TCP_STATINC(TCP_STAT_ECN_SHS); 2229 } 2230 2231 } 2232 tp->irs = th->th_seq; 2233 tcp_rcvseqinit(tp); 2234 tp->t_flags |= TF_ACKNOW; 2235 tcp_mss_from_peer(tp, opti.maxseg); 2236 2237 /* 2238 * Initialize the initial congestion window. If we 2239 * had to retransmit the SYN, we must initialize cwnd 2240 * to 1 segment (i.e. the Loss Window). 2241 */ 2242 if (tp->t_flags & TF_SYN_REXMT) 2243 tp->snd_cwnd = tp->t_peermss; 2244 else { 2245 int ss = tcp_init_win; 2246 #ifdef INET 2247 if (inp != NULL && in_localaddr(inp->inp_faddr)) 2248 ss = tcp_init_win_local; 2249 #endif 2250 #ifdef INET6 2251 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 2252 ss = tcp_init_win_local; 2253 #endif 2254 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 2255 } 2256 2257 tcp_rmx_rtt(tp); 2258 if (tiflags & TH_ACK) { 2259 TCP_STATINC(TCP_STAT_CONNECTS); 2260 /* 2261 * move tcp_established before soisconnected 2262 * because upcall handler can drive tcp_output 2263 * functionality. 2264 * XXX we might call soisconnected at the end of 2265 * all processing 2266 */ 2267 tcp_established(tp); 2268 soisconnected(so); 2269 /* Do window scaling on this connection? */ 2270 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2271 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2272 tp->snd_scale = tp->requested_s_scale; 2273 tp->rcv_scale = tp->request_r_scale; 2274 } 2275 TCP_REASS_LOCK(tp); 2276 (void) tcp_reass(tp, NULL, NULL, &tlen); 2277 /* 2278 * if we didn't have to retransmit the SYN, 2279 * use its rtt as our initial srtt & rtt var. 2280 */ 2281 if (tp->t_rtttime) 2282 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2283 } else 2284 tp->t_state = TCPS_SYN_RECEIVED; 2285 2286 /* 2287 * Advance th->th_seq to correspond to first data byte. 2288 * If data, trim to stay within window, 2289 * dropping FIN if necessary. 2290 */ 2291 th->th_seq++; 2292 if (tlen > tp->rcv_wnd) { 2293 todrop = tlen - tp->rcv_wnd; 2294 m_adj(m, -todrop); 2295 tlen = tp->rcv_wnd; 2296 tiflags &= ~TH_FIN; 2297 tcps = TCP_STAT_GETREF(); 2298 tcps[TCP_STAT_RCVPACKAFTERWIN]++; 2299 tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop; 2300 TCP_STAT_PUTREF(); 2301 } 2302 tp->snd_wl1 = th->th_seq - 1; 2303 tp->rcv_up = th->th_seq; 2304 goto step6; 2305 2306 /* 2307 * If the state is SYN_RECEIVED: 2308 * If seg contains an ACK, but not for our SYN, drop the input 2309 * and generate an RST. See page 36, rfc793 2310 */ 2311 case TCPS_SYN_RECEIVED: 2312 if ((tiflags & TH_ACK) && 2313 (SEQ_LEQ(th->th_ack, tp->iss) || 2314 SEQ_GT(th->th_ack, tp->snd_max))) 2315 goto dropwithreset; 2316 break; 2317 } 2318 2319 /* 2320 * States other than LISTEN or SYN_SENT. 2321 * First check timestamp, if present. 2322 * Then check that at least some bytes of segment are within 2323 * receive window. If segment begins before rcv_nxt, 2324 * drop leading data (and SYN); if nothing left, just ack. 2325 * 2326 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2327 * and it's less than ts_recent, drop it. 2328 */ 2329 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2330 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2331 2332 /* Check to see if ts_recent is over 24 days old. */ 2333 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2334 /* 2335 * Invalidate ts_recent. If this segment updates 2336 * ts_recent, the age will be reset later and ts_recent 2337 * will get a valid value. If it does not, setting 2338 * ts_recent to zero will at least satisfy the 2339 * requirement that zero be placed in the timestamp 2340 * echo reply when ts_recent isn't valid. The 2341 * age isn't reset until we get a valid ts_recent 2342 * because we don't want out-of-order segments to be 2343 * dropped when ts_recent is old. 2344 */ 2345 tp->ts_recent = 0; 2346 } else { 2347 tcps = TCP_STAT_GETREF(); 2348 tcps[TCP_STAT_RCVDUPPACK]++; 2349 tcps[TCP_STAT_RCVDUPBYTE] += tlen; 2350 tcps[TCP_STAT_PAWSDROP]++; 2351 TCP_STAT_PUTREF(); 2352 tcp_new_dsack(tp, th->th_seq, tlen); 2353 goto dropafterack; 2354 } 2355 } 2356 2357 todrop = tp->rcv_nxt - th->th_seq; 2358 dupseg = false; 2359 if (todrop > 0) { 2360 if (tiflags & TH_SYN) { 2361 tiflags &= ~TH_SYN; 2362 th->th_seq++; 2363 if (th->th_urp > 1) 2364 th->th_urp--; 2365 else { 2366 tiflags &= ~TH_URG; 2367 th->th_urp = 0; 2368 } 2369 todrop--; 2370 } 2371 if (todrop > tlen || 2372 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2373 /* 2374 * Any valid FIN or RST must be to the left of the 2375 * window. At this point the FIN or RST must be a 2376 * duplicate or out of sequence; drop it. 2377 */ 2378 if (tiflags & TH_RST) 2379 goto drop; 2380 tiflags &= ~(TH_FIN|TH_RST); 2381 /* 2382 * Send an ACK to resynchronize and drop any data. 2383 * But keep on processing for RST or ACK. 2384 */ 2385 tp->t_flags |= TF_ACKNOW; 2386 todrop = tlen; 2387 dupseg = true; 2388 tcps = TCP_STAT_GETREF(); 2389 tcps[TCP_STAT_RCVDUPPACK]++; 2390 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 2391 TCP_STAT_PUTREF(); 2392 } else if ((tiflags & TH_RST) && 2393 th->th_seq != tp->rcv_nxt) { 2394 /* 2395 * Test for reset before adjusting the sequence 2396 * number for overlapping data. 2397 */ 2398 goto dropafterack_ratelim; 2399 } else { 2400 tcps = TCP_STAT_GETREF(); 2401 tcps[TCP_STAT_RCVPARTDUPPACK]++; 2402 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 2403 TCP_STAT_PUTREF(); 2404 } 2405 tcp_new_dsack(tp, th->th_seq, todrop); 2406 hdroptlen += todrop; /*drop from head afterwards*/ 2407 th->th_seq += todrop; 2408 tlen -= todrop; 2409 if (th->th_urp > todrop) 2410 th->th_urp -= todrop; 2411 else { 2412 tiflags &= ~TH_URG; 2413 th->th_urp = 0; 2414 } 2415 } 2416 2417 /* 2418 * If new data are received on a connection after the 2419 * user processes are gone, then RST the other end. 2420 */ 2421 if ((so->so_state & SS_NOFDREF) && 2422 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2423 tp = tcp_close(tp); 2424 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 2425 goto dropwithreset; 2426 } 2427 2428 /* 2429 * If segment ends after window, drop trailing data 2430 * (and PUSH and FIN); if nothing left, just ACK. 2431 */ 2432 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 2433 if (todrop > 0) { 2434 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 2435 if (todrop >= tlen) { 2436 /* 2437 * The segment actually starts after the window. 2438 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2439 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2440 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2441 */ 2442 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 2443 /* 2444 * If a new connection request is received 2445 * while in TIME_WAIT, drop the old connection 2446 * and start over if the sequence numbers 2447 * are above the previous ones. 2448 * 2449 * NOTE: We will checksum the packet again, and 2450 * so we need to put the header fields back into 2451 * network order! 2452 * XXX This kind of sucks, but we don't expect 2453 * XXX this to happen very often, so maybe it 2454 * XXX doesn't matter so much. 2455 */ 2456 if (tiflags & TH_SYN && 2457 tp->t_state == TCPS_TIME_WAIT && 2458 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2459 tp = tcp_close(tp); 2460 tcp_fields_to_net(th); 2461 goto findpcb; 2462 } 2463 /* 2464 * If window is closed can only take segments at 2465 * window edge, and have to drop data and PUSH from 2466 * incoming segments. Continue processing, but 2467 * remember to ack. Otherwise, drop segment 2468 * and (if not RST) ack. 2469 */ 2470 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2471 tp->t_flags |= TF_ACKNOW; 2472 TCP_STATINC(TCP_STAT_RCVWINPROBE); 2473 } else 2474 goto dropafterack; 2475 } else 2476 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 2477 m_adj(m, -todrop); 2478 tlen -= todrop; 2479 tiflags &= ~(TH_PUSH|TH_FIN); 2480 } 2481 2482 /* 2483 * If last ACK falls within this segment's sequence numbers, 2484 * record the timestamp. 2485 * NOTE: 2486 * 1) That the test incorporates suggestions from the latest 2487 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 2488 * 2) That updating only on newer timestamps interferes with 2489 * our earlier PAWS tests, so this check should be solely 2490 * predicated on the sequence space of this segment. 2491 * 3) That we modify the segment boundary check to be 2492 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2493 * instead of RFC1323's 2494 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2495 * This modified check allows us to overcome RFC1323's 2496 * limitations as described in Stevens TCP/IP Illustrated 2497 * Vol. 2 p.869. In such cases, we can still calculate the 2498 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2499 */ 2500 if (opti.ts_present && 2501 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2502 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2503 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2504 tp->ts_recent_age = tcp_now; 2505 tp->ts_recent = opti.ts_val; 2506 } 2507 2508 /* 2509 * If the RST bit is set examine the state: 2510 * SYN_RECEIVED STATE: 2511 * If passive open, return to LISTEN state. 2512 * If active open, inform user that connection was refused. 2513 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 2514 * Inform user that connection was reset, and close tcb. 2515 * CLOSING, LAST_ACK, TIME_WAIT STATES 2516 * Close the tcb. 2517 */ 2518 if (tiflags & TH_RST) { 2519 if (th->th_seq != tp->rcv_nxt) 2520 goto dropafterack_ratelim; 2521 2522 switch (tp->t_state) { 2523 case TCPS_SYN_RECEIVED: 2524 so->so_error = ECONNREFUSED; 2525 goto close; 2526 2527 case TCPS_ESTABLISHED: 2528 case TCPS_FIN_WAIT_1: 2529 case TCPS_FIN_WAIT_2: 2530 case TCPS_CLOSE_WAIT: 2531 so->so_error = ECONNRESET; 2532 close: 2533 tp->t_state = TCPS_CLOSED; 2534 TCP_STATINC(TCP_STAT_DROPS); 2535 tp = tcp_close(tp); 2536 goto drop; 2537 2538 case TCPS_CLOSING: 2539 case TCPS_LAST_ACK: 2540 case TCPS_TIME_WAIT: 2541 tp = tcp_close(tp); 2542 goto drop; 2543 } 2544 } 2545 2546 /* 2547 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2548 * we must be in a synchronized state. RFC791 states (under RST 2549 * generation) that any unacceptable segment (an out-of-order SYN 2550 * qualifies) received in a synchronized state must elicit only an 2551 * empty acknowledgment segment ... and the connection remains in 2552 * the same state. 2553 */ 2554 if (tiflags & TH_SYN) { 2555 if (tp->rcv_nxt == th->th_seq) { 2556 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2557 TH_ACK); 2558 if (tcp_saveti) 2559 m_freem(tcp_saveti); 2560 return; 2561 } 2562 2563 goto dropafterack_ratelim; 2564 } 2565 2566 /* 2567 * If the ACK bit is off we drop the segment and return. 2568 */ 2569 if ((tiflags & TH_ACK) == 0) { 2570 if (tp->t_flags & TF_ACKNOW) 2571 goto dropafterack; 2572 else 2573 goto drop; 2574 } 2575 2576 /* 2577 * Ack processing. 2578 */ 2579 switch (tp->t_state) { 2580 2581 /* 2582 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2583 * ESTABLISHED state and continue processing, otherwise 2584 * send an RST. 2585 */ 2586 case TCPS_SYN_RECEIVED: 2587 if (SEQ_GT(tp->snd_una, th->th_ack) || 2588 SEQ_GT(th->th_ack, tp->snd_max)) 2589 goto dropwithreset; 2590 TCP_STATINC(TCP_STAT_CONNECTS); 2591 soisconnected(so); 2592 tcp_established(tp); 2593 /* Do window scaling? */ 2594 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2595 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2596 tp->snd_scale = tp->requested_s_scale; 2597 tp->rcv_scale = tp->request_r_scale; 2598 } 2599 TCP_REASS_LOCK(tp); 2600 (void) tcp_reass(tp, NULL, NULL, &tlen); 2601 tp->snd_wl1 = th->th_seq - 1; 2602 /* fall into ... */ 2603 2604 /* 2605 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2606 * ACKs. If the ack is in the range 2607 * tp->snd_una < th->th_ack <= tp->snd_max 2608 * then advance tp->snd_una to th->th_ack and drop 2609 * data from the retransmission queue. If this ACK reflects 2610 * more up to date window information we update our window information. 2611 */ 2612 case TCPS_ESTABLISHED: 2613 case TCPS_FIN_WAIT_1: 2614 case TCPS_FIN_WAIT_2: 2615 case TCPS_CLOSE_WAIT: 2616 case TCPS_CLOSING: 2617 case TCPS_LAST_ACK: 2618 case TCPS_TIME_WAIT: 2619 2620 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2621 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2622 TCP_STATINC(TCP_STAT_RCVDUPACK); 2623 /* 2624 * If we have outstanding data (other than 2625 * a window probe), this is a completely 2626 * duplicate ack (ie, window info didn't 2627 * change), the ack is the biggest we've 2628 * seen and we've seen exactly our rexmt 2629 * threshhold of them, assume a packet 2630 * has been dropped and retransmit it. 2631 * Kludge snd_nxt & the congestion 2632 * window so we send only this one 2633 * packet. 2634 */ 2635 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2636 th->th_ack != tp->snd_una) 2637 tp->t_dupacks = 0; 2638 else if (tp->t_partialacks < 0 && 2639 (++tp->t_dupacks == tcprexmtthresh || 2640 TCP_FACK_FASTRECOV(tp))) { 2641 /* 2642 * Do the fast retransmit, and adjust 2643 * congestion control paramenters. 2644 */ 2645 if (tp->t_congctl->fast_retransmit(tp, th)) { 2646 /* False fast retransmit */ 2647 break; 2648 } else 2649 goto drop; 2650 } else if (tp->t_dupacks > tcprexmtthresh) { 2651 tp->snd_cwnd += tp->t_segsz; 2652 KERNEL_LOCK(1, NULL); 2653 (void) tcp_output(tp); 2654 KERNEL_UNLOCK_ONE(NULL); 2655 goto drop; 2656 } 2657 } else { 2658 /* 2659 * If the ack appears to be very old, only 2660 * allow data that is in-sequence. This 2661 * makes it somewhat more difficult to insert 2662 * forged data by guessing sequence numbers. 2663 * Sent an ack to try to update the send 2664 * sequence number on the other side. 2665 */ 2666 if (tlen && th->th_seq != tp->rcv_nxt && 2667 SEQ_LT(th->th_ack, 2668 tp->snd_una - tp->max_sndwnd)) 2669 goto dropafterack; 2670 } 2671 break; 2672 } 2673 /* 2674 * If the congestion window was inflated to account 2675 * for the other side's cached packets, retract it. 2676 */ 2677 tp->t_congctl->fast_retransmit_newack(tp, th); 2678 2679 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2680 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2681 goto dropafterack; 2682 } 2683 acked = th->th_ack - tp->snd_una; 2684 tcps = TCP_STAT_GETREF(); 2685 tcps[TCP_STAT_RCVACKPACK]++; 2686 tcps[TCP_STAT_RCVACKBYTE] += acked; 2687 TCP_STAT_PUTREF(); 2688 2689 /* 2690 * If we have a timestamp reply, update smoothed 2691 * round trip time. If no timestamp is present but 2692 * transmit timer is running and timed sequence 2693 * number was acked, update smoothed round trip time. 2694 * Since we now have an rtt measurement, cancel the 2695 * timer backoff (cf., Phil Karn's retransmit alg.). 2696 * Recompute the initial retransmit timer. 2697 */ 2698 if (ts_rtt) 2699 tcp_xmit_timer(tp, ts_rtt - 1); 2700 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2701 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2702 2703 /* 2704 * If all outstanding data is acked, stop retransmit 2705 * timer and remember to restart (more output or persist). 2706 * If there is more data to be acked, restart retransmit 2707 * timer, using current (possibly backed-off) value. 2708 */ 2709 if (th->th_ack == tp->snd_max) { 2710 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2711 needoutput = 1; 2712 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2713 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2714 2715 /* 2716 * New data has been acked, adjust the congestion window. 2717 */ 2718 tp->t_congctl->newack(tp, th); 2719 2720 nd6_hint(tp); 2721 if (acked > so->so_snd.sb_cc) { 2722 tp->snd_wnd -= so->so_snd.sb_cc; 2723 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2724 ourfinisacked = 1; 2725 } else { 2726 if (acked > (tp->t_lastoff - tp->t_inoff)) 2727 tp->t_lastm = NULL; 2728 sbdrop(&so->so_snd, acked); 2729 tp->t_lastoff -= acked; 2730 if (tp->snd_wnd > acked) 2731 tp->snd_wnd -= acked; 2732 else 2733 tp->snd_wnd = 0; 2734 ourfinisacked = 0; 2735 } 2736 sowwakeup(so); 2737 2738 icmp_check(tp, th, acked); 2739 2740 tp->snd_una = th->th_ack; 2741 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2742 tp->snd_fack = tp->snd_una; 2743 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2744 tp->snd_nxt = tp->snd_una; 2745 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2746 tp->snd_high = tp->snd_una; 2747 2748 switch (tp->t_state) { 2749 2750 /* 2751 * In FIN_WAIT_1 STATE in addition to the processing 2752 * for the ESTABLISHED state if our FIN is now acknowledged 2753 * then enter FIN_WAIT_2. 2754 */ 2755 case TCPS_FIN_WAIT_1: 2756 if (ourfinisacked) { 2757 /* 2758 * If we can't receive any more 2759 * data, then closing user can proceed. 2760 * Starting the timer is contrary to the 2761 * specification, but if we don't get a FIN 2762 * we'll hang forever. 2763 */ 2764 if (so->so_state & SS_CANTRCVMORE) { 2765 soisdisconnected(so); 2766 if (tp->t_maxidle > 0) 2767 TCP_TIMER_ARM(tp, TCPT_2MSL, 2768 tp->t_maxidle); 2769 } 2770 tp->t_state = TCPS_FIN_WAIT_2; 2771 } 2772 break; 2773 2774 /* 2775 * In CLOSING STATE in addition to the processing for 2776 * the ESTABLISHED state if the ACK acknowledges our FIN 2777 * then enter the TIME-WAIT state, otherwise ignore 2778 * the segment. 2779 */ 2780 case TCPS_CLOSING: 2781 if (ourfinisacked) { 2782 tp->t_state = TCPS_TIME_WAIT; 2783 tcp_canceltimers(tp); 2784 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2785 soisdisconnected(so); 2786 } 2787 break; 2788 2789 /* 2790 * In LAST_ACK, we may still be waiting for data to drain 2791 * and/or to be acked, as well as for the ack of our FIN. 2792 * If our FIN is now acknowledged, delete the TCB, 2793 * enter the closed state and return. 2794 */ 2795 case TCPS_LAST_ACK: 2796 if (ourfinisacked) { 2797 tp = tcp_close(tp); 2798 goto drop; 2799 } 2800 break; 2801 2802 /* 2803 * In TIME_WAIT state the only thing that should arrive 2804 * is a retransmission of the remote FIN. Acknowledge 2805 * it and restart the finack timer. 2806 */ 2807 case TCPS_TIME_WAIT: 2808 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2809 goto dropafterack; 2810 } 2811 } 2812 2813 step6: 2814 /* 2815 * Update window information. 2816 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2817 */ 2818 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2819 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2820 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2821 /* keep track of pure window updates */ 2822 if (tlen == 0 && 2823 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2824 TCP_STATINC(TCP_STAT_RCVWINUPD); 2825 tp->snd_wnd = tiwin; 2826 tp->snd_wl1 = th->th_seq; 2827 tp->snd_wl2 = th->th_ack; 2828 if (tp->snd_wnd > tp->max_sndwnd) 2829 tp->max_sndwnd = tp->snd_wnd; 2830 needoutput = 1; 2831 } 2832 2833 /* 2834 * Process segments with URG. 2835 */ 2836 if ((tiflags & TH_URG) && th->th_urp && 2837 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2838 /* 2839 * This is a kludge, but if we receive and accept 2840 * random urgent pointers, we'll crash in 2841 * soreceive. It's hard to imagine someone 2842 * actually wanting to send this much urgent data. 2843 */ 2844 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2845 th->th_urp = 0; /* XXX */ 2846 tiflags &= ~TH_URG; /* XXX */ 2847 goto dodata; /* XXX */ 2848 } 2849 /* 2850 * If this segment advances the known urgent pointer, 2851 * then mark the data stream. This should not happen 2852 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2853 * a FIN has been received from the remote side. 2854 * In these states we ignore the URG. 2855 * 2856 * According to RFC961 (Assigned Protocols), 2857 * the urgent pointer points to the last octet 2858 * of urgent data. We continue, however, 2859 * to consider it to indicate the first octet 2860 * of data past the urgent section as the original 2861 * spec states (in one of two places). 2862 */ 2863 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2864 tp->rcv_up = th->th_seq + th->th_urp; 2865 so->so_oobmark = so->so_rcv.sb_cc + 2866 (tp->rcv_up - tp->rcv_nxt) - 1; 2867 if (so->so_oobmark == 0) 2868 so->so_state |= SS_RCVATMARK; 2869 sohasoutofband(so); 2870 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2871 } 2872 /* 2873 * Remove out of band data so doesn't get presented to user. 2874 * This can happen independent of advancing the URG pointer, 2875 * but if two URG's are pending at once, some out-of-band 2876 * data may creep in... ick. 2877 */ 2878 if (th->th_urp <= (u_int16_t) tlen 2879 #ifdef SO_OOBINLINE 2880 && (so->so_options & SO_OOBINLINE) == 0 2881 #endif 2882 ) 2883 tcp_pulloutofband(so, th, m, hdroptlen); 2884 } else 2885 /* 2886 * If no out of band data is expected, 2887 * pull receive urgent pointer along 2888 * with the receive window. 2889 */ 2890 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2891 tp->rcv_up = tp->rcv_nxt; 2892 dodata: /* XXX */ 2893 2894 /* 2895 * Process the segment text, merging it into the TCP sequencing queue, 2896 * and arranging for acknowledgement of receipt if necessary. 2897 * This process logically involves adjusting tp->rcv_wnd as data 2898 * is presented to the user (this happens in tcp_usrreq.c, 2899 * tcp_rcvd()). If a FIN has already been received on this 2900 * connection then we just ignore the text. 2901 */ 2902 if ((tlen || (tiflags & TH_FIN)) && 2903 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2904 /* 2905 * Insert segment ti into reassembly queue of tcp with 2906 * control block tp. Return TH_FIN if reassembly now includes 2907 * a segment with FIN. The macro form does the common case 2908 * inline (segment is the next to be received on an 2909 * established connection, and the queue is empty), 2910 * avoiding linkage into and removal from the queue and 2911 * repetition of various conversions. 2912 * Set DELACK for segments received in order, but ack 2913 * immediately when segments are out of order 2914 * (so fast retransmit can work). 2915 */ 2916 /* NOTE: this was TCP_REASS() macro, but used only once */ 2917 TCP_REASS_LOCK(tp); 2918 if (th->th_seq == tp->rcv_nxt && 2919 TAILQ_FIRST(&tp->segq) == NULL && 2920 tp->t_state == TCPS_ESTABLISHED) { 2921 tcp_setup_ack(tp, th); 2922 tp->rcv_nxt += tlen; 2923 tiflags = th->th_flags & TH_FIN; 2924 tcps = TCP_STAT_GETREF(); 2925 tcps[TCP_STAT_RCVPACK]++; 2926 tcps[TCP_STAT_RCVBYTE] += tlen; 2927 TCP_STAT_PUTREF(); 2928 nd6_hint(tp); 2929 if (so->so_state & SS_CANTRCVMORE) 2930 m_freem(m); 2931 else { 2932 m_adj(m, hdroptlen); 2933 sbappendstream(&(so)->so_rcv, m); 2934 } 2935 TCP_REASS_UNLOCK(tp); 2936 sorwakeup(so); 2937 } else { 2938 m_adj(m, hdroptlen); 2939 tiflags = tcp_reass(tp, th, m, &tlen); 2940 tp->t_flags |= TF_ACKNOW; 2941 } 2942 2943 /* 2944 * Note the amount of data that peer has sent into 2945 * our window, in order to estimate the sender's 2946 * buffer size. 2947 */ 2948 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2949 } else { 2950 m_freem(m); 2951 m = NULL; 2952 tiflags &= ~TH_FIN; 2953 } 2954 2955 /* 2956 * If FIN is received ACK the FIN and let the user know 2957 * that the connection is closing. Ignore a FIN received before 2958 * the connection is fully established. 2959 */ 2960 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2961 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2962 socantrcvmore(so); 2963 tp->t_flags |= TF_ACKNOW; 2964 tp->rcv_nxt++; 2965 } 2966 switch (tp->t_state) { 2967 2968 /* 2969 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2970 */ 2971 case TCPS_ESTABLISHED: 2972 tp->t_state = TCPS_CLOSE_WAIT; 2973 break; 2974 2975 /* 2976 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2977 * enter the CLOSING state. 2978 */ 2979 case TCPS_FIN_WAIT_1: 2980 tp->t_state = TCPS_CLOSING; 2981 break; 2982 2983 /* 2984 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2985 * starting the time-wait timer, turning off the other 2986 * standard timers. 2987 */ 2988 case TCPS_FIN_WAIT_2: 2989 tp->t_state = TCPS_TIME_WAIT; 2990 tcp_canceltimers(tp); 2991 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2992 soisdisconnected(so); 2993 break; 2994 2995 /* 2996 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2997 */ 2998 case TCPS_TIME_WAIT: 2999 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 3000 break; 3001 } 3002 } 3003 #ifdef TCP_DEBUG 3004 if (so->so_options & SO_DEBUG) 3005 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 3006 #endif 3007 3008 /* 3009 * Return any desired output. 3010 */ 3011 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 3012 KERNEL_LOCK(1, NULL); 3013 (void) tcp_output(tp); 3014 KERNEL_UNLOCK_ONE(NULL); 3015 } 3016 if (tcp_saveti) 3017 m_freem(tcp_saveti); 3018 3019 if (tp->t_state == TCPS_TIME_WAIT 3020 && (so->so_state & SS_NOFDREF) 3021 && (tp->t_inpcb || af != AF_INET) 3022 && (tp->t_in6pcb || af != AF_INET6) 3023 && ((af == AF_INET ? tcp4_vtw_enable : tcp6_vtw_enable) & 1) != 0 3024 && TAILQ_EMPTY(&tp->segq) 3025 && vtw_add(af, tp)) { 3026 ; 3027 } 3028 return; 3029 3030 badsyn: 3031 /* 3032 * Received a bad SYN. Increment counters and dropwithreset. 3033 */ 3034 TCP_STATINC(TCP_STAT_BADSYN); 3035 tp = NULL; 3036 goto dropwithreset; 3037 3038 dropafterack: 3039 /* 3040 * Generate an ACK dropping incoming segment if it occupies 3041 * sequence space, where the ACK reflects our state. 3042 */ 3043 if (tiflags & TH_RST) 3044 goto drop; 3045 goto dropafterack2; 3046 3047 dropafterack_ratelim: 3048 /* 3049 * We may want to rate-limit ACKs against SYN/RST attack. 3050 */ 3051 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 3052 tcp_ackdrop_ppslim) == 0) { 3053 /* XXX stat */ 3054 goto drop; 3055 } 3056 /* ...fall into dropafterack2... */ 3057 3058 dropafterack2: 3059 m_freem(m); 3060 tp->t_flags |= TF_ACKNOW; 3061 KERNEL_LOCK(1, NULL); 3062 (void) tcp_output(tp); 3063 KERNEL_UNLOCK_ONE(NULL); 3064 if (tcp_saveti) 3065 m_freem(tcp_saveti); 3066 return; 3067 3068 dropwithreset_ratelim: 3069 /* 3070 * We may want to rate-limit RSTs in certain situations, 3071 * particularly if we are sending an RST in response to 3072 * an attempt to connect to or otherwise communicate with 3073 * a port for which we have no socket. 3074 */ 3075 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 3076 tcp_rst_ppslim) == 0) { 3077 /* XXX stat */ 3078 goto drop; 3079 } 3080 /* ...fall into dropwithreset... */ 3081 3082 dropwithreset: 3083 /* 3084 * Generate a RST, dropping incoming segment. 3085 * Make ACK acceptable to originator of segment. 3086 */ 3087 if (tiflags & TH_RST) 3088 goto drop; 3089 3090 switch (af) { 3091 #ifdef INET6 3092 case AF_INET6: 3093 /* For following calls to tcp_respond */ 3094 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 3095 goto drop; 3096 break; 3097 #endif /* INET6 */ 3098 case AF_INET: 3099 if (IN_MULTICAST(ip->ip_dst.s_addr) || 3100 in_broadcast(ip->ip_dst, m_get_rcvif_NOMPSAFE(m))) 3101 goto drop; 3102 } 3103 3104 if (tiflags & TH_ACK) 3105 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3106 else { 3107 if (tiflags & TH_SYN) 3108 tlen++; 3109 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 3110 TH_RST|TH_ACK); 3111 } 3112 if (tcp_saveti) 3113 m_freem(tcp_saveti); 3114 return; 3115 3116 badcsum: 3117 drop: 3118 /* 3119 * Drop space held by incoming segment and return. 3120 */ 3121 if (tp) { 3122 if (tp->t_inpcb) 3123 so = tp->t_inpcb->inp_socket; 3124 #ifdef INET6 3125 else if (tp->t_in6pcb) 3126 so = tp->t_in6pcb->in6p_socket; 3127 #endif 3128 else 3129 so = NULL; 3130 #ifdef TCP_DEBUG 3131 if (so && (so->so_options & SO_DEBUG) != 0) 3132 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 3133 #endif 3134 } 3135 if (tcp_saveti) 3136 m_freem(tcp_saveti); 3137 m_freem(m); 3138 return; 3139 } 3140 3141 #ifdef TCP_SIGNATURE 3142 int 3143 tcp_signature_apply(void *fstate, void *data, u_int len) 3144 { 3145 3146 MD5Update(fstate, (u_char *)data, len); 3147 return (0); 3148 } 3149 3150 struct secasvar * 3151 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 3152 { 3153 struct ip *ip; 3154 struct ip6_hdr *ip6; 3155 3156 ip = mtod(m, struct ip *); 3157 switch (ip->ip_v) { 3158 case 4: 3159 ip = mtod(m, struct ip *); 3160 ip6 = NULL; 3161 break; 3162 case 6: 3163 ip = NULL; 3164 ip6 = mtod(m, struct ip6_hdr *); 3165 break; 3166 default: 3167 return (NULL); 3168 } 3169 3170 #ifdef IPSEC 3171 union sockaddr_union dst; 3172 3173 /* Extract the destination from the IP header in the mbuf. */ 3174 memset(&dst, 0, sizeof(union sockaddr_union)); 3175 if (ip != NULL) { 3176 dst.sa.sa_len = sizeof(struct sockaddr_in); 3177 dst.sa.sa_family = AF_INET; 3178 dst.sin.sin_addr = ip->ip_dst; 3179 } else { 3180 dst.sa.sa_len = sizeof(struct sockaddr_in6); 3181 dst.sa.sa_family = AF_INET6; 3182 dst.sin6.sin6_addr = ip6->ip6_dst; 3183 } 3184 3185 /* 3186 * Look up an SADB entry which matches the address of the peer. 3187 */ 3188 return KEY_LOOKUP_SA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI), 0, 0); 3189 #else 3190 return NULL; 3191 #endif 3192 } 3193 3194 int 3195 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 3196 struct secasvar *sav, char *sig) 3197 { 3198 MD5_CTX ctx; 3199 struct ip *ip; 3200 struct ipovly *ipovly; 3201 #ifdef INET6 3202 struct ip6_hdr *ip6; 3203 struct ip6_hdr_pseudo ip6pseudo; 3204 #endif /* INET6 */ 3205 struct ippseudo ippseudo; 3206 struct tcphdr th0; 3207 int l, tcphdrlen; 3208 3209 if (sav == NULL) 3210 return (-1); 3211 3212 tcphdrlen = th->th_off * 4; 3213 3214 switch (mtod(m, struct ip *)->ip_v) { 3215 case 4: 3216 MD5Init(&ctx); 3217 ip = mtod(m, struct ip *); 3218 memset(&ippseudo, 0, sizeof(ippseudo)); 3219 ipovly = (struct ipovly *)ip; 3220 ippseudo.ippseudo_src = ipovly->ih_src; 3221 ippseudo.ippseudo_dst = ipovly->ih_dst; 3222 ippseudo.ippseudo_pad = 0; 3223 ippseudo.ippseudo_p = IPPROTO_TCP; 3224 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 3225 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 3226 break; 3227 #if INET6 3228 case 6: 3229 MD5Init(&ctx); 3230 ip6 = mtod(m, struct ip6_hdr *); 3231 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 3232 ip6pseudo.ip6ph_src = ip6->ip6_src; 3233 in6_clearscope(&ip6pseudo.ip6ph_src); 3234 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 3235 in6_clearscope(&ip6pseudo.ip6ph_dst); 3236 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 3237 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 3238 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 3239 break; 3240 #endif /* INET6 */ 3241 default: 3242 return (-1); 3243 } 3244 3245 th0 = *th; 3246 th0.th_sum = 0; 3247 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 3248 3249 l = m->m_pkthdr.len - thoff - tcphdrlen; 3250 if (l > 0) 3251 m_apply(m, thoff + tcphdrlen, 3252 m->m_pkthdr.len - thoff - tcphdrlen, 3253 tcp_signature_apply, &ctx); 3254 3255 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 3256 MD5Final(sig, &ctx); 3257 3258 return (0); 3259 } 3260 #endif 3261 3262 /* 3263 * tcp_dooptions: parse and process tcp options. 3264 * 3265 * returns -1 if this segment should be dropped. (eg. wrong signature) 3266 * otherwise returns 0. 3267 */ 3268 3269 static int 3270 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, 3271 struct tcphdr *th, 3272 struct mbuf *m, int toff, struct tcp_opt_info *oi) 3273 { 3274 u_int16_t mss; 3275 int opt, optlen = 0; 3276 #ifdef TCP_SIGNATURE 3277 void *sigp = NULL; 3278 char sigbuf[TCP_SIGLEN]; 3279 struct secasvar *sav = NULL; 3280 #endif 3281 3282 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 3283 opt = cp[0]; 3284 if (opt == TCPOPT_EOL) 3285 break; 3286 if (opt == TCPOPT_NOP) 3287 optlen = 1; 3288 else { 3289 if (cnt < 2) 3290 break; 3291 optlen = cp[1]; 3292 if (optlen < 2 || optlen > cnt) 3293 break; 3294 } 3295 switch (opt) { 3296 3297 default: 3298 continue; 3299 3300 case TCPOPT_MAXSEG: 3301 if (optlen != TCPOLEN_MAXSEG) 3302 continue; 3303 if (!(th->th_flags & TH_SYN)) 3304 continue; 3305 if (TCPS_HAVERCVDSYN(tp->t_state)) 3306 continue; 3307 bcopy(cp + 2, &mss, sizeof(mss)); 3308 oi->maxseg = ntohs(mss); 3309 break; 3310 3311 case TCPOPT_WINDOW: 3312 if (optlen != TCPOLEN_WINDOW) 3313 continue; 3314 if (!(th->th_flags & TH_SYN)) 3315 continue; 3316 if (TCPS_HAVERCVDSYN(tp->t_state)) 3317 continue; 3318 tp->t_flags |= TF_RCVD_SCALE; 3319 tp->requested_s_scale = cp[2]; 3320 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3321 char buf[INET6_ADDRSTRLEN]; 3322 struct ip *ip = mtod(m, struct ip *); 3323 #ifdef INET6 3324 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 3325 #endif 3326 if (ip) 3327 in_print(buf, sizeof(buf), 3328 &ip->ip_src); 3329 #ifdef INET6 3330 else if (ip6) 3331 in6_print(buf, sizeof(buf), 3332 &ip6->ip6_src); 3333 #endif 3334 else 3335 strlcpy(buf, "(unknown)", sizeof(buf)); 3336 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3337 "assuming %d\n", 3338 tp->requested_s_scale, buf, 3339 TCP_MAX_WINSHIFT); 3340 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3341 } 3342 break; 3343 3344 case TCPOPT_TIMESTAMP: 3345 if (optlen != TCPOLEN_TIMESTAMP) 3346 continue; 3347 oi->ts_present = 1; 3348 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 3349 NTOHL(oi->ts_val); 3350 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 3351 NTOHL(oi->ts_ecr); 3352 3353 if (!(th->th_flags & TH_SYN)) 3354 continue; 3355 if (TCPS_HAVERCVDSYN(tp->t_state)) 3356 continue; 3357 /* 3358 * A timestamp received in a SYN makes 3359 * it ok to send timestamp requests and replies. 3360 */ 3361 tp->t_flags |= TF_RCVD_TSTMP; 3362 tp->ts_recent = oi->ts_val; 3363 tp->ts_recent_age = tcp_now; 3364 break; 3365 3366 case TCPOPT_SACK_PERMITTED: 3367 if (optlen != TCPOLEN_SACK_PERMITTED) 3368 continue; 3369 if (!(th->th_flags & TH_SYN)) 3370 continue; 3371 if (TCPS_HAVERCVDSYN(tp->t_state)) 3372 continue; 3373 if (tcp_do_sack) { 3374 tp->t_flags |= TF_SACK_PERMIT; 3375 tp->t_flags |= TF_WILL_SACK; 3376 } 3377 break; 3378 3379 case TCPOPT_SACK: 3380 tcp_sack_option(tp, th, cp, optlen); 3381 break; 3382 #ifdef TCP_SIGNATURE 3383 case TCPOPT_SIGNATURE: 3384 if (optlen != TCPOLEN_SIGNATURE) 3385 continue; 3386 if (sigp && memcmp(sigp, cp + 2, TCP_SIGLEN)) 3387 return (-1); 3388 3389 sigp = sigbuf; 3390 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3391 tp->t_flags |= TF_SIGNATURE; 3392 break; 3393 #endif 3394 } 3395 } 3396 3397 #ifndef TCP_SIGNATURE 3398 return 0; 3399 #else 3400 if (tp->t_flags & TF_SIGNATURE) { 3401 3402 sav = tcp_signature_getsav(m, th); 3403 3404 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3405 return (-1); 3406 } 3407 3408 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) 3409 goto out; 3410 3411 if (sigp) { 3412 char sig[TCP_SIGLEN]; 3413 3414 tcp_fields_to_net(th); 3415 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3416 tcp_fields_to_host(th); 3417 goto out; 3418 } 3419 tcp_fields_to_host(th); 3420 3421 if (memcmp(sig, sigp, TCP_SIGLEN)) { 3422 TCP_STATINC(TCP_STAT_BADSIG); 3423 goto out; 3424 } else 3425 TCP_STATINC(TCP_STAT_GOODSIG); 3426 3427 key_sa_recordxfer(sav, m); 3428 KEY_SA_UNREF(&sav); 3429 } 3430 return 0; 3431 out: 3432 if (sav != NULL) 3433 KEY_SA_UNREF(&sav); 3434 return -1; 3435 #endif 3436 } 3437 3438 /* 3439 * Pull out of band byte out of a segment so 3440 * it doesn't appear in the user's data queue. 3441 * It is still reflected in the segment length for 3442 * sequencing purposes. 3443 */ 3444 void 3445 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3446 struct mbuf *m, int off) 3447 { 3448 int cnt = off + th->th_urp - 1; 3449 3450 while (cnt >= 0) { 3451 if (m->m_len > cnt) { 3452 char *cp = mtod(m, char *) + cnt; 3453 struct tcpcb *tp = sototcpcb(so); 3454 3455 tp->t_iobc = *cp; 3456 tp->t_oobflags |= TCPOOB_HAVEDATA; 3457 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3458 m->m_len--; 3459 return; 3460 } 3461 cnt -= m->m_len; 3462 m = m->m_next; 3463 if (m == 0) 3464 break; 3465 } 3466 panic("tcp_pulloutofband"); 3467 } 3468 3469 /* 3470 * Collect new round-trip time estimate 3471 * and update averages and current timeout. 3472 * 3473 * rtt is in units of slow ticks (typically 500 ms) -- essentially the 3474 * difference of two timestamps. 3475 */ 3476 void 3477 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3478 { 3479 int32_t delta; 3480 3481 TCP_STATINC(TCP_STAT_RTTUPDATED); 3482 if (tp->t_srtt != 0) { 3483 /* 3484 * Compute the amount to add to srtt for smoothing, 3485 * *alpha, or 2^(-TCP_RTT_SHIFT). Because 3486 * srtt is stored in 1/32 slow ticks, we conceptually 3487 * shift left 5 bits, subtract srtt to get the 3488 * diference, and then shift right by TCP_RTT_SHIFT 3489 * (3) to obtain 1/8 of the difference. 3490 */ 3491 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3492 /* 3493 * This can never happen, because delta's lowest 3494 * possible value is 1/8 of t_srtt. But if it does, 3495 * set srtt to some reasonable value, here chosen 3496 * as 1/8 tick. 3497 */ 3498 if ((tp->t_srtt += delta) <= 0) 3499 tp->t_srtt = 1 << 2; 3500 /* 3501 * RFC2988 requires that rttvar be updated first. 3502 * This code is compliant because "delta" is the old 3503 * srtt minus the new observation (scaled). 3504 * 3505 * RFC2988 says: 3506 * rttvar = (1-beta) * rttvar + beta * |srtt-observed| 3507 * 3508 * delta is in units of 1/32 ticks, and has then been 3509 * divided by 8. This is equivalent to being in 1/16s 3510 * units and divided by 4. Subtract from it 1/4 of 3511 * the existing rttvar to form the (signed) amount to 3512 * adjust. 3513 */ 3514 if (delta < 0) 3515 delta = -delta; 3516 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3517 /* 3518 * As with srtt, this should never happen. There is 3519 * no support in RFC2988 for this operation. But 1/4s 3520 * as rttvar when faced with something arguably wrong 3521 * is ok. 3522 */ 3523 if ((tp->t_rttvar += delta) <= 0) 3524 tp->t_rttvar = 1 << 2; 3525 3526 /* 3527 * If srtt exceeds .01 second, ensure we use the 'remote' MSL 3528 * Problem is: it doesn't work. Disabled by defaulting 3529 * tcp_rttlocal to 0; see corresponding code in 3530 * tcp_subr that selects local vs remote in a different way. 3531 * 3532 * The static branch prediction hint here should be removed 3533 * when the rtt estimator is fixed and the rtt_enable code 3534 * is turned back on. 3535 */ 3536 if (__predict_false(tcp_rttlocal) && tcp_msl_enable 3537 && tp->t_srtt > tcp_msl_remote_threshold 3538 && tp->t_msl < tcp_msl_remote) { 3539 tp->t_msl = tcp_msl_remote; 3540 } 3541 } else { 3542 /* 3543 * This is the first measurement. Per RFC2988, 2.2, 3544 * set rtt=R and srtt=R/2. 3545 * For srtt, storage representation is 1/32 ticks, 3546 * so shift left by 5. 3547 * For rttvar, storage representation is 1/16 ticks, 3548 * So shift left by 4, but then right by 1 to halve. 3549 */ 3550 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3551 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3552 } 3553 tp->t_rtttime = 0; 3554 tp->t_rxtshift = 0; 3555 3556 /* 3557 * the retransmit should happen at rtt + 4 * rttvar. 3558 * Because of the way we do the smoothing, srtt and rttvar 3559 * will each average +1/2 tick of bias. When we compute 3560 * the retransmit timer, we want 1/2 tick of rounding and 3561 * 1 extra tick because of +-1/2 tick uncertainty in the 3562 * firing of the timer. The bias will give us exactly the 3563 * 1.5 tick we need. But, because the bias is 3564 * statistical, we have to test that we don't drop below 3565 * the minimum feasible timer (which is 2 ticks). 3566 */ 3567 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3568 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3569 3570 /* 3571 * We received an ack for a packet that wasn't retransmitted; 3572 * it is probably safe to discard any error indications we've 3573 * received recently. This isn't quite right, but close enough 3574 * for now (a route might have failed after we sent a segment, 3575 * and the return path might not be symmetrical). 3576 */ 3577 tp->t_softerror = 0; 3578 } 3579 3580 3581 /* 3582 * TCP compressed state engine. Currently used to hold compressed 3583 * state for SYN_RECEIVED. 3584 */ 3585 3586 u_long syn_cache_count; 3587 u_int32_t syn_hash1, syn_hash2; 3588 3589 #define SYN_HASH(sa, sp, dp) \ 3590 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3591 ((u_int32_t)(sp)))^syn_hash2))) 3592 #ifndef INET6 3593 #define SYN_HASHALL(hash, src, dst) \ 3594 do { \ 3595 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3596 ((const struct sockaddr_in *)(src))->sin_port, \ 3597 ((const struct sockaddr_in *)(dst))->sin_port); \ 3598 } while (/*CONSTCOND*/ 0) 3599 #else 3600 #define SYN_HASH6(sa, sp, dp) \ 3601 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3602 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3603 & 0x7fffffff) 3604 3605 #define SYN_HASHALL(hash, src, dst) \ 3606 do { \ 3607 switch ((src)->sa_family) { \ 3608 case AF_INET: \ 3609 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3610 ((const struct sockaddr_in *)(src))->sin_port, \ 3611 ((const struct sockaddr_in *)(dst))->sin_port); \ 3612 break; \ 3613 case AF_INET6: \ 3614 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3615 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3616 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3617 break; \ 3618 default: \ 3619 hash = 0; \ 3620 } \ 3621 } while (/*CONSTCOND*/0) 3622 #endif /* INET6 */ 3623 3624 static struct pool syn_cache_pool; 3625 3626 /* 3627 * We don't estimate RTT with SYNs, so each packet starts with the default 3628 * RTT and each timer step has a fixed timeout value. 3629 */ 3630 static inline void 3631 syn_cache_timer_arm(struct syn_cache *sc) 3632 { 3633 3634 TCPT_RANGESET(sc->sc_rxtcur, 3635 TCPTV_SRTTDFLT * tcp_backoff[sc->sc_rxtshift], TCPTV_MIN, 3636 TCPTV_REXMTMAX); 3637 callout_reset(&sc->sc_timer, 3638 sc->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, sc); 3639 } 3640 3641 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3642 3643 static inline void 3644 syn_cache_rm(struct syn_cache *sc) 3645 { 3646 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3647 sc, sc_bucketq); 3648 sc->sc_tp = NULL; 3649 LIST_REMOVE(sc, sc_tpq); 3650 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3651 callout_stop(&sc->sc_timer); 3652 syn_cache_count--; 3653 } 3654 3655 static inline void 3656 syn_cache_put(struct syn_cache *sc) 3657 { 3658 if (sc->sc_ipopts) 3659 (void) m_free(sc->sc_ipopts); 3660 rtcache_free(&sc->sc_route); 3661 sc->sc_flags |= SCF_DEAD; 3662 if (!callout_invoking(&sc->sc_timer)) 3663 callout_schedule(&(sc)->sc_timer, 1); 3664 } 3665 3666 void 3667 syn_cache_init(void) 3668 { 3669 int i; 3670 3671 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 3672 "synpl", NULL, IPL_SOFTNET); 3673 3674 /* Initialize the hash buckets. */ 3675 for (i = 0; i < tcp_syn_cache_size; i++) 3676 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3677 } 3678 3679 void 3680 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3681 { 3682 struct syn_cache_head *scp; 3683 struct syn_cache *sc2; 3684 int s; 3685 3686 /* 3687 * If there are no entries in the hash table, reinitialize 3688 * the hash secrets. 3689 */ 3690 if (syn_cache_count == 0) { 3691 syn_hash1 = cprng_fast32(); 3692 syn_hash2 = cprng_fast32(); 3693 } 3694 3695 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3696 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3697 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3698 3699 /* 3700 * Make sure that we don't overflow the per-bucket 3701 * limit or the total cache size limit. 3702 */ 3703 s = splsoftnet(); 3704 if (scp->sch_length >= tcp_syn_bucket_limit) { 3705 TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW); 3706 /* 3707 * The bucket is full. Toss the oldest element in the 3708 * bucket. This will be the first entry in the bucket. 3709 */ 3710 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3711 #ifdef DIAGNOSTIC 3712 /* 3713 * This should never happen; we should always find an 3714 * entry in our bucket. 3715 */ 3716 if (sc2 == NULL) 3717 panic("syn_cache_insert: bucketoverflow: impossible"); 3718 #endif 3719 syn_cache_rm(sc2); 3720 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3721 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3722 struct syn_cache_head *scp2, *sce; 3723 3724 TCP_STATINC(TCP_STAT_SC_OVERFLOWED); 3725 /* 3726 * The cache is full. Toss the oldest entry in the 3727 * first non-empty bucket we can find. 3728 * 3729 * XXX We would really like to toss the oldest 3730 * entry in the cache, but we hope that this 3731 * condition doesn't happen very often. 3732 */ 3733 scp2 = scp; 3734 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3735 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3736 for (++scp2; scp2 != scp; scp2++) { 3737 if (scp2 >= sce) 3738 scp2 = &tcp_syn_cache[0]; 3739 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3740 break; 3741 } 3742 #ifdef DIAGNOSTIC 3743 /* 3744 * This should never happen; we should always find a 3745 * non-empty bucket. 3746 */ 3747 if (scp2 == scp) 3748 panic("syn_cache_insert: cacheoverflow: " 3749 "impossible"); 3750 #endif 3751 } 3752 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3753 syn_cache_rm(sc2); 3754 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3755 } 3756 3757 /* 3758 * Initialize the entry's timer. 3759 */ 3760 sc->sc_rxttot = 0; 3761 sc->sc_rxtshift = 0; 3762 syn_cache_timer_arm(sc); 3763 3764 /* Link it from tcpcb entry */ 3765 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3766 3767 /* Put it into the bucket. */ 3768 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3769 scp->sch_length++; 3770 syn_cache_count++; 3771 3772 TCP_STATINC(TCP_STAT_SC_ADDED); 3773 splx(s); 3774 } 3775 3776 /* 3777 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3778 * If we have retransmitted an entry the maximum number of times, expire 3779 * that entry. 3780 */ 3781 static void 3782 syn_cache_timer(void *arg) 3783 { 3784 struct syn_cache *sc = arg; 3785 3786 mutex_enter(softnet_lock); 3787 KERNEL_LOCK(1, NULL); 3788 3789 callout_ack(&sc->sc_timer); 3790 3791 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3792 TCP_STATINC(TCP_STAT_SC_DELAYED_FREE); 3793 goto free; 3794 } 3795 3796 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3797 /* Drop it -- too many retransmissions. */ 3798 goto dropit; 3799 } 3800 3801 /* 3802 * Compute the total amount of time this entry has 3803 * been on a queue. If this entry has been on longer 3804 * than the keep alive timer would allow, expire it. 3805 */ 3806 sc->sc_rxttot += sc->sc_rxtcur; 3807 if (sc->sc_rxttot >= tcp_keepinit) 3808 goto dropit; 3809 3810 TCP_STATINC(TCP_STAT_SC_RETRANSMITTED); 3811 (void) syn_cache_respond(sc, NULL); 3812 3813 /* Advance the timer back-off. */ 3814 sc->sc_rxtshift++; 3815 syn_cache_timer_arm(sc); 3816 3817 goto out; 3818 3819 dropit: 3820 TCP_STATINC(TCP_STAT_SC_TIMED_OUT); 3821 syn_cache_rm(sc); 3822 if (sc->sc_ipopts) 3823 (void) m_free(sc->sc_ipopts); 3824 rtcache_free(&sc->sc_route); 3825 3826 free: 3827 callout_destroy(&sc->sc_timer); 3828 pool_put(&syn_cache_pool, sc); 3829 3830 out: 3831 KERNEL_UNLOCK_ONE(NULL); 3832 mutex_exit(softnet_lock); 3833 } 3834 3835 /* 3836 * Remove syn cache created by the specified tcb entry, 3837 * because this does not make sense to keep them 3838 * (if there's no tcb entry, syn cache entry will never be used) 3839 */ 3840 void 3841 syn_cache_cleanup(struct tcpcb *tp) 3842 { 3843 struct syn_cache *sc, *nsc; 3844 int s; 3845 3846 s = splsoftnet(); 3847 3848 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3849 nsc = LIST_NEXT(sc, sc_tpq); 3850 3851 #ifdef DIAGNOSTIC 3852 if (sc->sc_tp != tp) 3853 panic("invalid sc_tp in syn_cache_cleanup"); 3854 #endif 3855 syn_cache_rm(sc); 3856 syn_cache_put(sc); /* calls pool_put but see spl above */ 3857 } 3858 /* just for safety */ 3859 LIST_INIT(&tp->t_sc); 3860 3861 splx(s); 3862 } 3863 3864 /* 3865 * Find an entry in the syn cache. 3866 */ 3867 struct syn_cache * 3868 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3869 struct syn_cache_head **headp) 3870 { 3871 struct syn_cache *sc; 3872 struct syn_cache_head *scp; 3873 u_int32_t hash; 3874 int s; 3875 3876 SYN_HASHALL(hash, src, dst); 3877 3878 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3879 *headp = scp; 3880 s = splsoftnet(); 3881 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3882 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3883 if (sc->sc_hash != hash) 3884 continue; 3885 if (!memcmp(&sc->sc_src, src, src->sa_len) && 3886 !memcmp(&sc->sc_dst, dst, dst->sa_len)) { 3887 splx(s); 3888 return (sc); 3889 } 3890 } 3891 splx(s); 3892 return (NULL); 3893 } 3894 3895 /* 3896 * This function gets called when we receive an ACK for a 3897 * socket in the LISTEN state. We look up the connection 3898 * in the syn cache, and if its there, we pull it out of 3899 * the cache and turn it into a full-blown connection in 3900 * the SYN-RECEIVED state. 3901 * 3902 * The return values may not be immediately obvious, and their effects 3903 * can be subtle, so here they are: 3904 * 3905 * NULL SYN was not found in cache; caller should drop the 3906 * packet and send an RST. 3907 * 3908 * -1 We were unable to create the new connection, and are 3909 * aborting it. An ACK,RST is being sent to the peer 3910 * (unless we got screwey sequence numbners; see below), 3911 * because the 3-way handshake has been completed. Caller 3912 * should not free the mbuf, since we may be using it. If 3913 * we are not, we will free it. 3914 * 3915 * Otherwise, the return value is a pointer to the new socket 3916 * associated with the connection. 3917 */ 3918 struct socket * 3919 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3920 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3921 struct socket *so, struct mbuf *m) 3922 { 3923 struct syn_cache *sc; 3924 struct syn_cache_head *scp; 3925 struct inpcb *inp = NULL; 3926 #ifdef INET6 3927 struct in6pcb *in6p = NULL; 3928 #endif 3929 struct tcpcb *tp = 0; 3930 int s; 3931 struct socket *oso; 3932 3933 s = splsoftnet(); 3934 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3935 splx(s); 3936 return (NULL); 3937 } 3938 3939 /* 3940 * Verify the sequence and ack numbers. Try getting the correct 3941 * response again. 3942 */ 3943 if ((th->th_ack != sc->sc_iss + 1) || 3944 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3945 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3946 (void) syn_cache_respond(sc, m); 3947 splx(s); 3948 return ((struct socket *)(-1)); 3949 } 3950 3951 /* Remove this cache entry */ 3952 syn_cache_rm(sc); 3953 splx(s); 3954 3955 /* 3956 * Ok, create the full blown connection, and set things up 3957 * as they would have been set up if we had created the 3958 * connection when the SYN arrived. If we can't create 3959 * the connection, abort it. 3960 */ 3961 /* 3962 * inp still has the OLD in_pcb stuff, set the 3963 * v6-related flags on the new guy, too. This is 3964 * done particularly for the case where an AF_INET6 3965 * socket is bound only to a port, and a v4 connection 3966 * comes in on that port. 3967 * we also copy the flowinfo from the original pcb 3968 * to the new one. 3969 */ 3970 oso = so; 3971 so = sonewconn(so, true); 3972 if (so == NULL) 3973 goto resetandabort; 3974 3975 switch (so->so_proto->pr_domain->dom_family) { 3976 #ifdef INET 3977 case AF_INET: 3978 inp = sotoinpcb(so); 3979 break; 3980 #endif 3981 #ifdef INET6 3982 case AF_INET6: 3983 in6p = sotoin6pcb(so); 3984 break; 3985 #endif 3986 } 3987 switch (src->sa_family) { 3988 #ifdef INET 3989 case AF_INET: 3990 if (inp) { 3991 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3992 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3993 inp->inp_options = ip_srcroute(m); 3994 in_pcbstate(inp, INP_BOUND); 3995 if (inp->inp_options == NULL) { 3996 inp->inp_options = sc->sc_ipopts; 3997 sc->sc_ipopts = NULL; 3998 } 3999 } 4000 #ifdef INET6 4001 else if (in6p) { 4002 /* IPv4 packet to AF_INET6 socket */ 4003 memset(&in6p->in6p_laddr, 0, sizeof(in6p->in6p_laddr)); 4004 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 4005 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 4006 &in6p->in6p_laddr.s6_addr32[3], 4007 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 4008 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 4009 in6totcpcb(in6p)->t_family = AF_INET; 4010 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 4011 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 4012 else 4013 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 4014 in6_pcbstate(in6p, IN6P_BOUND); 4015 } 4016 #endif 4017 break; 4018 #endif 4019 #ifdef INET6 4020 case AF_INET6: 4021 if (in6p) { 4022 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 4023 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 4024 in6_pcbstate(in6p, IN6P_BOUND); 4025 } 4026 break; 4027 #endif 4028 } 4029 #ifdef INET6 4030 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 4031 struct in6pcb *oin6p = sotoin6pcb(oso); 4032 /* inherit socket options from the listening socket */ 4033 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 4034 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 4035 m_freem(in6p->in6p_options); 4036 in6p->in6p_options = 0; 4037 } 4038 ip6_savecontrol(in6p, &in6p->in6p_options, 4039 mtod(m, struct ip6_hdr *), m); 4040 } 4041 #endif 4042 4043 #if defined(IPSEC) 4044 if (ipsec_used) { 4045 /* 4046 * we make a copy of policy, instead of sharing the policy, for 4047 * better behavior in terms of SA lookup and dead SA removal. 4048 */ 4049 if (inp) { 4050 /* copy old policy into new socket's */ 4051 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, 4052 inp->inp_sp)) 4053 printf("tcp_input: could not copy policy\n"); 4054 } 4055 #ifdef INET6 4056 else if (in6p) { 4057 /* copy old policy into new socket's */ 4058 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 4059 in6p->in6p_sp)) 4060 printf("tcp_input: could not copy policy\n"); 4061 } 4062 #endif 4063 } 4064 #endif 4065 4066 /* 4067 * Give the new socket our cached route reference. 4068 */ 4069 if (inp) { 4070 rtcache_copy(&inp->inp_route, &sc->sc_route); 4071 rtcache_free(&sc->sc_route); 4072 } 4073 #ifdef INET6 4074 else { 4075 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 4076 rtcache_free(&sc->sc_route); 4077 } 4078 #endif 4079 4080 if (inp) { 4081 struct sockaddr_in sin; 4082 memcpy(&sin, src, src->sa_len); 4083 if (in_pcbconnect(inp, &sin, &lwp0)) { 4084 goto resetandabort; 4085 } 4086 } 4087 #ifdef INET6 4088 else if (in6p) { 4089 struct sockaddr_in6 sin6; 4090 memcpy(&sin6, src, src->sa_len); 4091 if (src->sa_family == AF_INET) { 4092 /* IPv4 packet to AF_INET6 socket */ 4093 in6_sin_2_v4mapsin6((struct sockaddr_in *)src, &sin6); 4094 } 4095 if (in6_pcbconnect(in6p, &sin6, NULL)) { 4096 goto resetandabort; 4097 } 4098 } 4099 #endif 4100 else { 4101 goto resetandabort; 4102 } 4103 4104 if (inp) 4105 tp = intotcpcb(inp); 4106 #ifdef INET6 4107 else if (in6p) 4108 tp = in6totcpcb(in6p); 4109 #endif 4110 else 4111 tp = NULL; 4112 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 4113 if (sc->sc_request_r_scale != 15) { 4114 tp->requested_s_scale = sc->sc_requested_s_scale; 4115 tp->request_r_scale = sc->sc_request_r_scale; 4116 tp->snd_scale = sc->sc_requested_s_scale; 4117 tp->rcv_scale = sc->sc_request_r_scale; 4118 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 4119 } 4120 if (sc->sc_flags & SCF_TIMESTAMP) 4121 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 4122 tp->ts_timebase = sc->sc_timebase; 4123 4124 tp->t_template = tcp_template(tp); 4125 if (tp->t_template == 0) { 4126 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 4127 so = NULL; 4128 m_freem(m); 4129 goto abort; 4130 } 4131 4132 tp->iss = sc->sc_iss; 4133 tp->irs = sc->sc_irs; 4134 tcp_sendseqinit(tp); 4135 tcp_rcvseqinit(tp); 4136 tp->t_state = TCPS_SYN_RECEIVED; 4137 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 4138 TCP_STATINC(TCP_STAT_ACCEPTS); 4139 4140 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 4141 tp->t_flags |= TF_WILL_SACK; 4142 4143 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 4144 tp->t_flags |= TF_ECN_PERMIT; 4145 4146 #ifdef TCP_SIGNATURE 4147 if (sc->sc_flags & SCF_SIGNATURE) 4148 tp->t_flags |= TF_SIGNATURE; 4149 #endif 4150 4151 /* Initialize tp->t_ourmss before we deal with the peer's! */ 4152 tp->t_ourmss = sc->sc_ourmaxseg; 4153 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 4154 4155 /* 4156 * Initialize the initial congestion window. If we 4157 * had to retransmit the SYN,ACK, we must initialize cwnd 4158 * to 1 segment (i.e. the Loss Window). 4159 */ 4160 if (sc->sc_rxtshift) 4161 tp->snd_cwnd = tp->t_peermss; 4162 else { 4163 int ss = tcp_init_win; 4164 #ifdef INET 4165 if (inp != NULL && in_localaddr(inp->inp_faddr)) 4166 ss = tcp_init_win_local; 4167 #endif 4168 #ifdef INET6 4169 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 4170 ss = tcp_init_win_local; 4171 #endif 4172 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 4173 } 4174 4175 tcp_rmx_rtt(tp); 4176 tp->snd_wl1 = sc->sc_irs; 4177 tp->rcv_up = sc->sc_irs + 1; 4178 4179 /* 4180 * This is what whould have happened in tcp_output() when 4181 * the SYN,ACK was sent. 4182 */ 4183 tp->snd_up = tp->snd_una; 4184 tp->snd_max = tp->snd_nxt = tp->iss+1; 4185 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 4186 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 4187 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 4188 tp->last_ack_sent = tp->rcv_nxt; 4189 tp->t_partialacks = -1; 4190 tp->t_dupacks = 0; 4191 4192 TCP_STATINC(TCP_STAT_SC_COMPLETED); 4193 s = splsoftnet(); 4194 syn_cache_put(sc); 4195 splx(s); 4196 return (so); 4197 4198 resetandabort: 4199 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 4200 abort: 4201 if (so != NULL) { 4202 (void) soqremque(so, 1); 4203 (void) soabort(so); 4204 mutex_enter(softnet_lock); 4205 } 4206 s = splsoftnet(); 4207 syn_cache_put(sc); 4208 splx(s); 4209 TCP_STATINC(TCP_STAT_SC_ABORTED); 4210 return ((struct socket *)(-1)); 4211 } 4212 4213 /* 4214 * This function is called when we get a RST for a 4215 * non-existent connection, so that we can see if the 4216 * connection is in the syn cache. If it is, zap it. 4217 */ 4218 4219 void 4220 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 4221 { 4222 struct syn_cache *sc; 4223 struct syn_cache_head *scp; 4224 int s = splsoftnet(); 4225 4226 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4227 splx(s); 4228 return; 4229 } 4230 if (SEQ_LT(th->th_seq, sc->sc_irs) || 4231 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 4232 splx(s); 4233 return; 4234 } 4235 syn_cache_rm(sc); 4236 TCP_STATINC(TCP_STAT_SC_RESET); 4237 syn_cache_put(sc); /* calls pool_put but see spl above */ 4238 splx(s); 4239 } 4240 4241 void 4242 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 4243 struct tcphdr *th) 4244 { 4245 struct syn_cache *sc; 4246 struct syn_cache_head *scp; 4247 int s; 4248 4249 s = splsoftnet(); 4250 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4251 splx(s); 4252 return; 4253 } 4254 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 4255 if (ntohl (th->th_seq) != sc->sc_iss) { 4256 splx(s); 4257 return; 4258 } 4259 4260 /* 4261 * If we've retransmitted 3 times and this is our second error, 4262 * we remove the entry. Otherwise, we allow it to continue on. 4263 * This prevents us from incorrectly nuking an entry during a 4264 * spurious network outage. 4265 * 4266 * See tcp_notify(). 4267 */ 4268 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 4269 sc->sc_flags |= SCF_UNREACH; 4270 splx(s); 4271 return; 4272 } 4273 4274 syn_cache_rm(sc); 4275 TCP_STATINC(TCP_STAT_SC_UNREACH); 4276 syn_cache_put(sc); /* calls pool_put but see spl above */ 4277 splx(s); 4278 } 4279 4280 /* 4281 * Given a LISTEN socket and an inbound SYN request, add 4282 * this to the syn cache, and send back a segment: 4283 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 4284 * to the source. 4285 * 4286 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 4287 * Doing so would require that we hold onto the data and deliver it 4288 * to the application. However, if we are the target of a SYN-flood 4289 * DoS attack, an attacker could send data which would eventually 4290 * consume all available buffer space if it were ACKed. By not ACKing 4291 * the data, we avoid this DoS scenario. 4292 */ 4293 4294 int 4295 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 4296 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 4297 int optlen, struct tcp_opt_info *oi) 4298 { 4299 struct tcpcb tb, *tp; 4300 long win; 4301 struct syn_cache *sc; 4302 struct syn_cache_head *scp; 4303 struct mbuf *ipopts; 4304 struct tcp_opt_info opti; 4305 int s; 4306 4307 tp = sototcpcb(so); 4308 4309 memset(&opti, 0, sizeof(opti)); 4310 4311 /* 4312 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 4313 * 4314 * Note this check is performed in tcp_input() very early on. 4315 */ 4316 4317 /* 4318 * Initialize some local state. 4319 */ 4320 win = sbspace(&so->so_rcv); 4321 if (win > TCP_MAXWIN) 4322 win = TCP_MAXWIN; 4323 4324 switch (src->sa_family) { 4325 #ifdef INET 4326 case AF_INET: 4327 /* 4328 * Remember the IP options, if any. 4329 */ 4330 ipopts = ip_srcroute(m); 4331 break; 4332 #endif 4333 default: 4334 ipopts = NULL; 4335 } 4336 4337 #ifdef TCP_SIGNATURE 4338 if (optp || (tp->t_flags & TF_SIGNATURE)) 4339 #else 4340 if (optp) 4341 #endif 4342 { 4343 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4344 #ifdef TCP_SIGNATURE 4345 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4346 #endif 4347 tb.t_state = TCPS_LISTEN; 4348 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 4349 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 4350 return (0); 4351 } else 4352 tb.t_flags = 0; 4353 4354 /* 4355 * See if we already have an entry for this connection. 4356 * If we do, resend the SYN,ACK. We do not count this 4357 * as a retransmission (XXX though maybe we should). 4358 */ 4359 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4360 TCP_STATINC(TCP_STAT_SC_DUPESYN); 4361 if (ipopts) { 4362 /* 4363 * If we were remembering a previous source route, 4364 * forget it and use the new one we've been given. 4365 */ 4366 if (sc->sc_ipopts) 4367 (void) m_free(sc->sc_ipopts); 4368 sc->sc_ipopts = ipopts; 4369 } 4370 sc->sc_timestamp = tb.ts_recent; 4371 if (syn_cache_respond(sc, m) == 0) { 4372 uint64_t *tcps = TCP_STAT_GETREF(); 4373 tcps[TCP_STAT_SNDACKS]++; 4374 tcps[TCP_STAT_SNDTOTAL]++; 4375 TCP_STAT_PUTREF(); 4376 } 4377 return (1); 4378 } 4379 4380 s = splsoftnet(); 4381 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4382 splx(s); 4383 if (sc == NULL) { 4384 if (ipopts) 4385 (void) m_free(ipopts); 4386 return (0); 4387 } 4388 4389 /* 4390 * Fill in the cache, and put the necessary IP and TCP 4391 * options into the reply. 4392 */ 4393 memset(sc, 0, sizeof(struct syn_cache)); 4394 callout_init(&sc->sc_timer, CALLOUT_MPSAFE); 4395 bcopy(src, &sc->sc_src, src->sa_len); 4396 bcopy(dst, &sc->sc_dst, dst->sa_len); 4397 sc->sc_flags = 0; 4398 sc->sc_ipopts = ipopts; 4399 sc->sc_irs = th->th_seq; 4400 switch (src->sa_family) { 4401 #ifdef INET 4402 case AF_INET: 4403 { 4404 struct sockaddr_in *srcin = (void *) src; 4405 struct sockaddr_in *dstin = (void *) dst; 4406 4407 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4408 &srcin->sin_addr, dstin->sin_port, 4409 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4410 break; 4411 } 4412 #endif /* INET */ 4413 #ifdef INET6 4414 case AF_INET6: 4415 { 4416 struct sockaddr_in6 *srcin6 = (void *) src; 4417 struct sockaddr_in6 *dstin6 = (void *) dst; 4418 4419 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4420 &srcin6->sin6_addr, dstin6->sin6_port, 4421 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4422 break; 4423 } 4424 #endif /* INET6 */ 4425 } 4426 sc->sc_peermaxseg = oi->maxseg; 4427 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4428 m_get_rcvif_NOMPSAFE(m) : NULL, 4429 sc->sc_src.sa.sa_family); 4430 sc->sc_win = win; 4431 sc->sc_timebase = tcp_now - 1; /* see tcp_newtcpcb() */ 4432 sc->sc_timestamp = tb.ts_recent; 4433 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4434 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4435 sc->sc_flags |= SCF_TIMESTAMP; 4436 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4437 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4438 sc->sc_requested_s_scale = tb.requested_s_scale; 4439 sc->sc_request_r_scale = 0; 4440 /* 4441 * Pick the smallest possible scaling factor that 4442 * will still allow us to scale up to sb_max. 4443 * 4444 * We do this because there are broken firewalls that 4445 * will corrupt the window scale option, leading to 4446 * the other endpoint believing that our advertised 4447 * window is unscaled. At scale factors larger than 4448 * 5 the unscaled window will drop below 1500 bytes, 4449 * leading to serious problems when traversing these 4450 * broken firewalls. 4451 * 4452 * With the default sbmax of 256K, a scale factor 4453 * of 3 will be chosen by this algorithm. Those who 4454 * choose a larger sbmax should watch out 4455 * for the compatiblity problems mentioned above. 4456 * 4457 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4458 * or <SYN,ACK>) segment itself is never scaled. 4459 */ 4460 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4461 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4462 sc->sc_request_r_scale++; 4463 } else { 4464 sc->sc_requested_s_scale = 15; 4465 sc->sc_request_r_scale = 15; 4466 } 4467 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4468 sc->sc_flags |= SCF_SACK_PERMIT; 4469 4470 /* 4471 * ECN setup packet received. 4472 */ 4473 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4474 sc->sc_flags |= SCF_ECN_PERMIT; 4475 4476 #ifdef TCP_SIGNATURE 4477 if (tb.t_flags & TF_SIGNATURE) 4478 sc->sc_flags |= SCF_SIGNATURE; 4479 #endif 4480 sc->sc_tp = tp; 4481 if (syn_cache_respond(sc, m) == 0) { 4482 uint64_t *tcps = TCP_STAT_GETREF(); 4483 tcps[TCP_STAT_SNDACKS]++; 4484 tcps[TCP_STAT_SNDTOTAL]++; 4485 TCP_STAT_PUTREF(); 4486 syn_cache_insert(sc, tp); 4487 } else { 4488 s = splsoftnet(); 4489 /* 4490 * syn_cache_put() will try to schedule the timer, so 4491 * we need to initialize it 4492 */ 4493 syn_cache_timer_arm(sc); 4494 syn_cache_put(sc); 4495 splx(s); 4496 TCP_STATINC(TCP_STAT_SC_DROPPED); 4497 } 4498 return (1); 4499 } 4500 4501 /* 4502 * syn_cache_respond: (re)send SYN+ACK. 4503 * 4504 * returns 0 on success. otherwise returns an errno, typically ENOBUFS. 4505 */ 4506 4507 int 4508 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4509 { 4510 #ifdef INET6 4511 struct rtentry *rt = NULL; 4512 #endif 4513 struct route *ro; 4514 u_int8_t *optp; 4515 int optlen, error; 4516 u_int16_t tlen; 4517 struct ip *ip = NULL; 4518 #ifdef INET6 4519 struct ip6_hdr *ip6 = NULL; 4520 #endif 4521 struct tcpcb *tp = NULL; 4522 struct tcphdr *th; 4523 u_int hlen; 4524 #ifdef TCP_SIGNATURE 4525 struct secasvar *sav = NULL; 4526 u_int8_t *sigp = NULL; 4527 #endif 4528 4529 ro = &sc->sc_route; 4530 switch (sc->sc_src.sa.sa_family) { 4531 case AF_INET: 4532 hlen = sizeof(struct ip); 4533 break; 4534 #ifdef INET6 4535 case AF_INET6: 4536 hlen = sizeof(struct ip6_hdr); 4537 break; 4538 #endif 4539 default: 4540 if (m) 4541 m_freem(m); 4542 return (EAFNOSUPPORT); 4543 } 4544 4545 /* worst case scanario, since we don't know the option size yet */ 4546 tlen = hlen + sizeof(struct tcphdr) + MAX_TCPOPTLEN; 4547 4548 /* 4549 * Create the IP+TCP header from scratch. 4550 */ 4551 if (m) 4552 m_freem(m); 4553 #ifdef DIAGNOSTIC 4554 if (max_linkhdr + tlen > MCLBYTES) 4555 return ENOBUFS; 4556 #endif 4557 4558 MGETHDR(m, M_DONTWAIT, MT_DATA); 4559 if (m && (max_linkhdr + tlen) > MHLEN) { 4560 MCLGET(m, M_DONTWAIT); 4561 if ((m->m_flags & M_EXT) == 0) { 4562 m_freem(m); 4563 m = NULL; 4564 } 4565 } 4566 if (m == NULL) 4567 return ENOBUFS; 4568 MCLAIM(m, &tcp_tx_mowner); 4569 4570 /* Fixup the mbuf. */ 4571 m->m_data += max_linkhdr; 4572 if (sc->sc_tp) 4573 tp = sc->sc_tp; 4574 m_reset_rcvif(m); 4575 memset(mtod(m, u_char *), 0, tlen); 4576 4577 switch (sc->sc_src.sa.sa_family) { 4578 case AF_INET: 4579 ip = mtod(m, struct ip *); 4580 ip->ip_v = 4; 4581 ip->ip_dst = sc->sc_src.sin.sin_addr; 4582 ip->ip_src = sc->sc_dst.sin.sin_addr; 4583 ip->ip_p = IPPROTO_TCP; 4584 th = (struct tcphdr *)(ip + 1); 4585 th->th_dport = sc->sc_src.sin.sin_port; 4586 th->th_sport = sc->sc_dst.sin.sin_port; 4587 break; 4588 #ifdef INET6 4589 case AF_INET6: 4590 ip6 = mtod(m, struct ip6_hdr *); 4591 ip6->ip6_vfc = IPV6_VERSION; 4592 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4593 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4594 ip6->ip6_nxt = IPPROTO_TCP; 4595 /* ip6_plen will be updated in ip6_output() */ 4596 th = (struct tcphdr *)(ip6 + 1); 4597 th->th_dport = sc->sc_src.sin6.sin6_port; 4598 th->th_sport = sc->sc_dst.sin6.sin6_port; 4599 break; 4600 #endif 4601 default: 4602 return ENOBUFS; 4603 } 4604 4605 th->th_seq = htonl(sc->sc_iss); 4606 th->th_ack = htonl(sc->sc_irs + 1); 4607 th->th_flags = TH_SYN|TH_ACK; 4608 th->th_win = htons(sc->sc_win); 4609 /* th_x2, th_sum, th_urp already 0 from memset */ 4610 4611 /* Tack on the TCP options. */ 4612 optp = (u_int8_t *)(th + 1); 4613 optlen = 0; 4614 *optp++ = TCPOPT_MAXSEG; 4615 *optp++ = TCPOLEN_MAXSEG; 4616 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4617 *optp++ = sc->sc_ourmaxseg & 0xff; 4618 optlen += TCPOLEN_MAXSEG; 4619 4620 if (sc->sc_request_r_scale != 15) { 4621 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4622 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4623 sc->sc_request_r_scale); 4624 optp += TCPOLEN_WINDOW + TCPOLEN_NOP; 4625 optlen += TCPOLEN_WINDOW + TCPOLEN_NOP; 4626 } 4627 4628 if (sc->sc_flags & SCF_SACK_PERMIT) { 4629 /* Let the peer know that we will SACK. */ 4630 *optp++ = TCPOPT_SACK_PERMITTED; 4631 *optp++ = TCPOLEN_SACK_PERMITTED; 4632 optlen += TCPOLEN_SACK_PERMITTED; 4633 } 4634 4635 if (sc->sc_flags & SCF_TIMESTAMP) { 4636 while (optlen % 4 != 2) { 4637 optlen += TCPOLEN_NOP; 4638 *optp++ = TCPOPT_NOP; 4639 } 4640 *optp++ = TCPOPT_TIMESTAMP; 4641 *optp++ = TCPOLEN_TIMESTAMP; 4642 u_int32_t *lp = (u_int32_t *)(optp); 4643 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4644 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4645 *lp = htonl(sc->sc_timestamp); 4646 optp += TCPOLEN_TIMESTAMP - 2; 4647 optlen += TCPOLEN_TIMESTAMP; 4648 } 4649 4650 #ifdef TCP_SIGNATURE 4651 if (sc->sc_flags & SCF_SIGNATURE) { 4652 4653 sav = tcp_signature_getsav(m, th); 4654 4655 if (sav == NULL) { 4656 if (m) 4657 m_freem(m); 4658 return (EPERM); 4659 } 4660 4661 *optp++ = TCPOPT_SIGNATURE; 4662 *optp++ = TCPOLEN_SIGNATURE; 4663 sigp = optp; 4664 memset(optp, 0, TCP_SIGLEN); 4665 optp += TCP_SIGLEN; 4666 optlen += TCPOLEN_SIGNATURE; 4667 4668 } 4669 #endif 4670 /* Terminate and pad TCP options to a 4 byte boundary. */ 4671 if (optlen % 4) { 4672 optlen += TCPOLEN_EOL; 4673 *optp++ = TCPOPT_EOL; 4674 } 4675 /* 4676 * According to RFC 793 (STD0007): 4677 * "The content of the header beyond the End-of-Option option 4678 * must be header padding (i.e., zero)." 4679 * and later: "The padding is composed of zeros." 4680 */ 4681 while (optlen % 4) { 4682 optlen += TCPOLEN_PAD; 4683 *optp++ = TCPOPT_PAD; 4684 } 4685 4686 /* compute the actual values now that we've added the options */ 4687 tlen = hlen + sizeof(struct tcphdr) + optlen; 4688 m->m_len = m->m_pkthdr.len = tlen; 4689 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4690 4691 #ifdef TCP_SIGNATURE 4692 if (sav) { 4693 (void)tcp_signature(m, th, hlen, sav, sigp); 4694 key_sa_recordxfer(sav, m); 4695 KEY_SA_UNREF(&sav); 4696 } 4697 #endif 4698 4699 /* 4700 * Send ECN SYN-ACK setup packet. 4701 * Routes can be asymetric, so, even if we receive a packet 4702 * with ECE and CWR set, we must not assume no one will block 4703 * the ECE packet we are about to send. 4704 */ 4705 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4706 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4707 th->th_flags |= TH_ECE; 4708 TCP_STATINC(TCP_STAT_ECN_SHS); 4709 4710 /* 4711 * draft-ietf-tcpm-ecnsyn-00.txt 4712 * 4713 * "[...] a TCP node MAY respond to an ECN-setup 4714 * SYN packet by setting ECT in the responding 4715 * ECN-setup SYN/ACK packet, indicating to routers 4716 * that the SYN/ACK packet is ECN-Capable. 4717 * This allows a congested router along the path 4718 * to mark the packet instead of dropping the 4719 * packet as an indication of congestion." 4720 * 4721 * "[...] There can be a great benefit in setting 4722 * an ECN-capable codepoint in SYN/ACK packets [...] 4723 * Congestion is most likely to occur in 4724 * the server-to-client direction. As a result, 4725 * setting an ECN-capable codepoint in SYN/ACK 4726 * packets can reduce the occurence of three-second 4727 * retransmit timeouts resulting from the drop 4728 * of SYN/ACK packets." 4729 * 4730 * Page 4 and 6, January 2006. 4731 */ 4732 4733 switch (sc->sc_src.sa.sa_family) { 4734 #ifdef INET 4735 case AF_INET: 4736 ip->ip_tos |= IPTOS_ECN_ECT0; 4737 break; 4738 #endif 4739 #ifdef INET6 4740 case AF_INET6: 4741 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4742 break; 4743 #endif 4744 } 4745 TCP_STATINC(TCP_STAT_ECN_ECT); 4746 } 4747 4748 4749 /* Compute the packet's checksum. */ 4750 switch (sc->sc_src.sa.sa_family) { 4751 case AF_INET: 4752 ip->ip_len = htons(tlen - hlen); 4753 th->th_sum = 0; 4754 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4755 break; 4756 #ifdef INET6 4757 case AF_INET6: 4758 ip6->ip6_plen = htons(tlen - hlen); 4759 th->th_sum = 0; 4760 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4761 break; 4762 #endif 4763 } 4764 4765 /* 4766 * Fill in some straggling IP bits. Note the stack expects 4767 * ip_len to be in host order, for convenience. 4768 */ 4769 switch (sc->sc_src.sa.sa_family) { 4770 #ifdef INET 4771 case AF_INET: 4772 ip->ip_len = htons(tlen); 4773 ip->ip_ttl = ip_defttl; 4774 /* XXX tos? */ 4775 break; 4776 #endif 4777 #ifdef INET6 4778 case AF_INET6: 4779 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4780 ip6->ip6_vfc |= IPV6_VERSION; 4781 ip6->ip6_plen = htons(tlen - hlen); 4782 /* ip6_hlim will be initialized afterwards */ 4783 /* XXX flowlabel? */ 4784 break; 4785 #endif 4786 } 4787 4788 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4789 tp = sc->sc_tp; 4790 4791 switch (sc->sc_src.sa.sa_family) { 4792 #ifdef INET 4793 case AF_INET: 4794 error = ip_output(m, sc->sc_ipopts, ro, 4795 (ip_mtudisc ? IP_MTUDISC : 0), 4796 NULL, tp ? tp->t_inpcb : NULL); 4797 break; 4798 #endif 4799 #ifdef INET6 4800 case AF_INET6: 4801 ip6->ip6_hlim = in6_selecthlim(NULL, 4802 (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp : NULL); 4803 rtcache_unref(rt, ro); 4804 4805 error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, 4806 tp ? tp->t_in6pcb : NULL, NULL); 4807 break; 4808 #endif 4809 default: 4810 error = EAFNOSUPPORT; 4811 break; 4812 } 4813 return (error); 4814 } 4815
Indexes created Thu Sep 25 16:09:42 GMT 2025