tcp_input.c revision 1.379
1 /* $NetBSD: tcp_input.c,v 1.379 2018/02/12 08:22:26 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 1.379 2018/02/12 08:22:26 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) 1010 { 1011 int tiflags; 1012 int todrop; 1013 uint32_t t_flags = 0; 1014 uint64_t *tcps; 1015 1016 tiflags = th->th_flags; 1017 todrop = vp->rcv_nxt - th->th_seq; 1018 1019 if (todrop > 0) { 1020 if (tiflags & TH_SYN) { 1021 tiflags &= ~TH_SYN; 1022 ++th->th_seq; 1023 if (th->th_urp > 1) 1024 --th->th_urp; 1025 else { 1026 tiflags &= ~TH_URG; 1027 th->th_urp = 0; 1028 } 1029 --todrop; 1030 } 1031 if (todrop > tlen || 1032 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1033 /* 1034 * Any valid FIN or RST must be to the left of the 1035 * window. At this point the FIN or RST must be a 1036 * duplicate or out of sequence; drop it. 1037 */ 1038 if (tiflags & TH_RST) 1039 goto drop; 1040 tiflags &= ~(TH_FIN|TH_RST); 1041 /* 1042 * Send an ACK to resynchronize and drop any data. 1043 * But keep on processing for RST or ACK. 1044 */ 1045 t_flags |= TF_ACKNOW; 1046 todrop = tlen; 1047 tcps = TCP_STAT_GETREF(); 1048 tcps[TCP_STAT_RCVDUPPACK] += 1; 1049 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 1050 TCP_STAT_PUTREF(); 1051 } else if ((tiflags & TH_RST) 1052 && th->th_seq != vp->rcv_nxt) { 1053 /* 1054 * Test for reset before adjusting the sequence 1055 * number for overlapping data. 1056 */ 1057 goto dropafterack_ratelim; 1058 } else { 1059 tcps = TCP_STAT_GETREF(); 1060 tcps[TCP_STAT_RCVPARTDUPPACK] += 1; 1061 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 1062 TCP_STAT_PUTREF(); 1063 } 1064 1065 // tcp_new_dsack(tp, th->th_seq, todrop); 1066 // hdroptlen += todrop; /*drop from head afterwards*/ 1067 1068 th->th_seq += todrop; 1069 tlen -= todrop; 1070 1071 if (th->th_urp > todrop) 1072 th->th_urp -= todrop; 1073 else { 1074 tiflags &= ~TH_URG; 1075 th->th_urp = 0; 1076 } 1077 } 1078 1079 /* 1080 * If new data are received on a connection after the 1081 * user processes are gone, then RST the other end. 1082 */ 1083 if (tlen) { 1084 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 1085 goto dropwithreset; 1086 } 1087 1088 /* 1089 * If segment ends after window, drop trailing data 1090 * (and PUSH and FIN); if nothing left, just ACK. 1091 */ 1092 todrop = (th->th_seq + tlen) - (vp->rcv_nxt+vp->rcv_wnd); 1093 1094 if (todrop > 0) { 1095 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 1096 if (todrop >= tlen) { 1097 /* 1098 * The segment actually starts after the window. 1099 * th->th_seq + tlen - vp->rcv_nxt - vp->rcv_wnd >= tlen 1100 * th->th_seq - vp->rcv_nxt - vp->rcv_wnd >= 0 1101 * th->th_seq >= vp->rcv_nxt + vp->rcv_wnd 1102 */ 1103 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 1104 /* 1105 * If a new connection request is received 1106 * while in TIME_WAIT, drop the old connection 1107 * and start over if the sequence numbers 1108 * are above the previous ones. 1109 */ 1110 if ((tiflags & TH_SYN) 1111 && SEQ_GT(th->th_seq, vp->rcv_nxt)) { 1112 /* We only support this in the !NOFDREF case, which 1113 * is to say: not here. 1114 */ 1115 goto dropwithreset; 1116 } 1117 /* 1118 * If window is closed can only take segments at 1119 * window edge, and have to drop data and PUSH from 1120 * incoming segments. Continue processing, but 1121 * remember to ack. Otherwise, drop segment 1122 * and (if not RST) ack. 1123 */ 1124 if (vp->rcv_wnd == 0 && th->th_seq == vp->rcv_nxt) { 1125 t_flags |= TF_ACKNOW; 1126 TCP_STATINC(TCP_STAT_RCVWINPROBE); 1127 } else 1128 goto dropafterack; 1129 } else 1130 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 1131 m_adj(m, -todrop); 1132 tlen -= todrop; 1133 tiflags &= ~(TH_PUSH|TH_FIN); 1134 } 1135 1136 if (tiflags & TH_RST) { 1137 if (th->th_seq != vp->rcv_nxt) 1138 goto dropafterack_ratelim; 1139 1140 vtw_del(vp->ctl, vp->vtw); 1141 goto drop; 1142 } 1143 1144 /* 1145 * If the ACK bit is off we drop the segment and return. 1146 */ 1147 if ((tiflags & TH_ACK) == 0) { 1148 if (t_flags & TF_ACKNOW) 1149 goto dropafterack; 1150 else 1151 goto drop; 1152 } 1153 1154 /* 1155 * In TIME_WAIT state the only thing that should arrive 1156 * is a retransmission of the remote FIN. Acknowledge 1157 * it and restart the finack timer. 1158 */ 1159 vtw_restart(vp); 1160 goto dropafterack; 1161 1162 dropafterack: 1163 /* 1164 * Generate an ACK dropping incoming segment if it occupies 1165 * sequence space, where the ACK reflects our state. 1166 */ 1167 if (tiflags & TH_RST) 1168 goto drop; 1169 goto dropafterack2; 1170 1171 dropafterack_ratelim: 1172 /* 1173 * We may want to rate-limit ACKs against SYN/RST attack. 1174 */ 1175 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 1176 tcp_ackdrop_ppslim) == 0) { 1177 /* XXX stat */ 1178 goto drop; 1179 } 1180 /* ...fall into dropafterack2... */ 1181 1182 dropafterack2: 1183 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, th->th_ack, 1184 TH_ACK); 1185 return; 1186 1187 dropwithreset: 1188 /* 1189 * Generate a RST, dropping incoming segment. 1190 * Make ACK acceptable to originator of segment. 1191 */ 1192 if (tiflags & TH_RST) 1193 goto drop; 1194 1195 if (tiflags & TH_ACK) 1196 tcp_respond(0, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 1197 else { 1198 if (tiflags & TH_SYN) 1199 ++tlen; 1200 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, (tcp_seq)0, 1201 TH_RST|TH_ACK); 1202 } 1203 return; 1204 drop: 1205 m_freem(m); 1206 } 1207 1208 /* 1209 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 1210 */ 1211 void 1212 tcp_input(struct mbuf *m, ...) 1213 { 1214 struct tcphdr *th; 1215 struct ip *ip; 1216 struct inpcb *inp; 1217 #ifdef INET6 1218 struct ip6_hdr *ip6; 1219 struct in6pcb *in6p; 1220 #endif 1221 u_int8_t *optp = NULL; 1222 int optlen = 0; 1223 int len, tlen, toff, hdroptlen = 0; 1224 struct tcpcb *tp = NULL; 1225 int tiflags; 1226 struct socket *so = NULL; 1227 int todrop, acked, ourfinisacked, needoutput = 0; 1228 bool dupseg; 1229 #ifdef TCP_DEBUG 1230 short ostate = 0; 1231 #endif 1232 u_long tiwin; 1233 struct tcp_opt_info opti; 1234 int off, iphlen; 1235 va_list ap; 1236 int af; /* af on the wire */ 1237 struct mbuf *tcp_saveti = NULL; 1238 uint32_t ts_rtt; 1239 uint8_t iptos; 1240 uint64_t *tcps; 1241 vestigial_inpcb_t vestige; 1242 1243 vestige.valid = 0; 1244 1245 MCLAIM(m, &tcp_rx_mowner); 1246 va_start(ap, m); 1247 toff = va_arg(ap, int); 1248 (void)va_arg(ap, int); /* ignore value, advance ap */ 1249 va_end(ap); 1250 1251 TCP_STATINC(TCP_STAT_RCVTOTAL); 1252 1253 memset(&opti, 0, sizeof(opti)); 1254 opti.ts_present = 0; 1255 opti.maxseg = 0; 1256 1257 /* 1258 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 1259 * 1260 * TCP is, by definition, unicast, so we reject all 1261 * multicast outright. 1262 * 1263 * Note, there are additional src/dst address checks in 1264 * the AF-specific code below. 1265 */ 1266 if (m->m_flags & (M_BCAST|M_MCAST)) { 1267 /* XXX stat */ 1268 goto drop; 1269 } 1270 #ifdef INET6 1271 if (m->m_flags & M_ANYCAST6) { 1272 /* XXX stat */ 1273 goto drop; 1274 } 1275 #endif 1276 1277 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, sizeof(struct tcphdr)); 1278 if (th == NULL) { 1279 TCP_STATINC(TCP_STAT_RCVSHORT); 1280 return; 1281 } 1282 1283 /* 1284 * Get IP and TCP header. 1285 * Note: IP leaves IP header in first mbuf. 1286 */ 1287 ip = mtod(m, struct ip *); 1288 switch (ip->ip_v) { 1289 #ifdef INET 1290 case 4: 1291 #ifdef INET6 1292 ip6 = NULL; 1293 #endif 1294 af = AF_INET; 1295 iphlen = sizeof(struct ip); 1296 1297 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1298 in_broadcast(ip->ip_dst, m_get_rcvif_NOMPSAFE(m))) 1299 goto drop; 1300 1301 /* We do the checksum after PCB lookup... */ 1302 len = ntohs(ip->ip_len); 1303 tlen = len - toff; 1304 iptos = ip->ip_tos; 1305 break; 1306 #endif 1307 #ifdef INET6 1308 case 6: 1309 ip = NULL; 1310 iphlen = sizeof(struct ip6_hdr); 1311 af = AF_INET6; 1312 ip6 = mtod(m, struct ip6_hdr *); 1313 1314 /* 1315 * Be proactive about unspecified IPv6 address in source. 1316 * As we use all-zero to indicate unbounded/unconnected pcb, 1317 * unspecified IPv6 address can be used to confuse us. 1318 * 1319 * Note that packets with unspecified IPv6 destination is 1320 * already dropped in ip6_input. 1321 */ 1322 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1323 /* XXX stat */ 1324 goto drop; 1325 } 1326 1327 /* 1328 * Make sure destination address is not multicast. 1329 * Source address checked in ip6_input(). 1330 */ 1331 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1332 /* XXX stat */ 1333 goto drop; 1334 } 1335 1336 /* We do the checksum after PCB lookup... */ 1337 len = m->m_pkthdr.len; 1338 tlen = len - toff; 1339 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1340 break; 1341 #endif 1342 default: 1343 m_freem(m); 1344 return; 1345 } 1346 1347 /* 1348 * Enforce alignment requirements that are violated in 1349 * some cases, see kern/50766 for details. 1350 */ 1351 if (TCP_HDR_ALIGNED_P(th) == 0) { 1352 m = m_copyup(m, toff + sizeof(struct tcphdr), 0); 1353 if (m == NULL) { 1354 TCP_STATINC(TCP_STAT_RCVSHORT); 1355 return; 1356 } 1357 ip = mtod(m, struct ip *); 1358 #ifdef INET6 1359 ip6 = mtod(m, struct ip6_hdr *); 1360 #endif 1361 th = (struct tcphdr *)(mtod(m, char *) + toff); 1362 } 1363 KASSERT(TCP_HDR_ALIGNED_P(th)); 1364 1365 /* 1366 * Check that TCP offset makes sense, pull out TCP options and 1367 * adjust length. 1368 */ 1369 off = th->th_off << 2; 1370 if (off < sizeof(struct tcphdr) || off > tlen) { 1371 TCP_STATINC(TCP_STAT_RCVBADOFF); 1372 goto drop; 1373 } 1374 tlen -= off; 1375 1376 if (off > sizeof(struct tcphdr)) { 1377 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1378 if (th == NULL) { 1379 TCP_STATINC(TCP_STAT_RCVSHORT); 1380 return; 1381 } 1382 KASSERT(TCP_HDR_ALIGNED_P(th)); 1383 optlen = off - sizeof(struct tcphdr); 1384 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1385 /* 1386 * Do quick retrieval of timestamp options ("options 1387 * prediction?"). If timestamp is the only option and it's 1388 * formatted as recommended in RFC 1323 appendix A, we 1389 * quickly get the values now and not bother calling 1390 * tcp_dooptions(), etc. 1391 */ 1392 if ((optlen == TCPOLEN_TSTAMP_APPA || 1393 (optlen > TCPOLEN_TSTAMP_APPA && 1394 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1395 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1396 (th->th_flags & TH_SYN) == 0) { 1397 opti.ts_present = 1; 1398 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1399 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1400 optp = NULL; /* we've parsed the options */ 1401 } 1402 } 1403 tiflags = th->th_flags; 1404 1405 /* 1406 * Checksum extended TCP header and data 1407 */ 1408 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1409 goto badcsum; 1410 1411 /* 1412 * Locate pcb for segment. 1413 */ 1414 findpcb: 1415 inp = NULL; 1416 #ifdef INET6 1417 in6p = NULL; 1418 #endif 1419 switch (af) { 1420 #ifdef INET 1421 case AF_INET: 1422 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1423 ip->ip_dst, th->th_dport, &vestige); 1424 if (inp == NULL && !vestige.valid) { 1425 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1426 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, 1427 th->th_dport); 1428 } 1429 #ifdef INET6 1430 if (inp == NULL && !vestige.valid) { 1431 struct in6_addr s, d; 1432 1433 /* mapped addr case */ 1434 in6_in_2_v4mapin6(&ip->ip_src, &s); 1435 in6_in_2_v4mapin6(&ip->ip_dst, &d); 1436 in6p = in6_pcblookup_connect(&tcbtable, &s, 1437 th->th_sport, &d, th->th_dport, 0, &vestige); 1438 if (in6p == 0 && !vestige.valid) { 1439 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1440 in6p = in6_pcblookup_bind(&tcbtable, &d, 1441 th->th_dport, 0); 1442 } 1443 } 1444 #endif 1445 #ifndef INET6 1446 if (inp == NULL && !vestige.valid) 1447 #else 1448 if (inp == NULL && in6p == NULL && !vestige.valid) 1449 #endif 1450 { 1451 TCP_STATINC(TCP_STAT_NOPORT); 1452 if (tcp_log_refused && 1453 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1454 tcp4_log_refused(ip, th); 1455 } 1456 tcp_fields_to_host(th); 1457 goto dropwithreset_ratelim; 1458 } 1459 #if defined(IPSEC) 1460 if (ipsec_used) { 1461 if (inp && 1462 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 1463 && ipsec4_in_reject(m, inp)) { 1464 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1465 goto drop; 1466 } 1467 #ifdef INET6 1468 else if (in6p && 1469 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 1470 && ipsec6_in_reject(m, in6p)) { 1471 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1472 goto drop; 1473 } 1474 #endif 1475 } 1476 #endif /*IPSEC*/ 1477 break; 1478 #endif /*INET*/ 1479 #ifdef INET6 1480 case AF_INET6: 1481 { 1482 int faith; 1483 1484 #if defined(NFAITH) && NFAITH > 0 1485 faith = faithprefix(&ip6->ip6_dst); 1486 #else 1487 faith = 0; 1488 #endif 1489 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1490 th->th_sport, &ip6->ip6_dst, th->th_dport, faith, &vestige); 1491 if (!in6p && !vestige.valid) { 1492 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1493 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1494 th->th_dport, faith); 1495 } 1496 if (!in6p && !vestige.valid) { 1497 TCP_STATINC(TCP_STAT_NOPORT); 1498 if (tcp_log_refused && 1499 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1500 tcp6_log_refused(ip6, th); 1501 } 1502 tcp_fields_to_host(th); 1503 goto dropwithreset_ratelim; 1504 } 1505 #if defined(IPSEC) 1506 if (ipsec_used && in6p && 1507 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1508 ipsec6_in_reject(m, in6p)) { 1509 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1510 goto drop; 1511 } 1512 #endif /*IPSEC*/ 1513 break; 1514 } 1515 #endif 1516 } 1517 1518 /* 1519 * If the state is CLOSED (i.e., TCB does not exist) then 1520 * all data in the incoming segment is discarded. 1521 * If the TCB exists but is in CLOSED state, it is embryonic, 1522 * but should either do a listen or a connect soon. 1523 */ 1524 tp = NULL; 1525 so = NULL; 1526 if (inp) { 1527 /* Check the minimum TTL for socket. */ 1528 if (ip->ip_ttl < inp->inp_ip_minttl) 1529 goto drop; 1530 1531 tp = intotcpcb(inp); 1532 so = inp->inp_socket; 1533 } 1534 #ifdef INET6 1535 else if (in6p) { 1536 tp = in6totcpcb(in6p); 1537 so = in6p->in6p_socket; 1538 } 1539 #endif 1540 else if (vestige.valid) { 1541 /* We do not support the resurrection of vtw tcpcps. */ 1542 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1543 goto badcsum; 1544 1545 tcp_fields_to_host(th); 1546 tcp_vtw_input(th, &vestige, m, tlen); 1547 m = NULL; 1548 goto drop; 1549 } 1550 1551 if (tp == NULL) { 1552 tcp_fields_to_host(th); 1553 goto dropwithreset_ratelim; 1554 } 1555 if (tp->t_state == TCPS_CLOSED) 1556 goto drop; 1557 1558 KASSERT(so->so_lock == softnet_lock); 1559 KASSERT(solocked(so)); 1560 1561 tcp_fields_to_host(th); 1562 1563 /* Unscale the window into a 32-bit value. */ 1564 if ((tiflags & TH_SYN) == 0) 1565 tiwin = th->th_win << tp->snd_scale; 1566 else 1567 tiwin = th->th_win; 1568 1569 #ifdef INET6 1570 /* save packet options if user wanted */ 1571 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1572 if (in6p->in6p_options) { 1573 m_freem(in6p->in6p_options); 1574 in6p->in6p_options = NULL; 1575 } 1576 KASSERT(ip6 != NULL); 1577 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1578 } 1579 #endif 1580 1581 if (so->so_options & SO_DEBUG) { 1582 #ifdef TCP_DEBUG 1583 ostate = tp->t_state; 1584 #endif 1585 1586 tcp_saveti = NULL; 1587 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1588 goto nosave; 1589 1590 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1591 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1592 if (tcp_saveti == NULL) 1593 goto nosave; 1594 } else { 1595 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1596 if (tcp_saveti == NULL) 1597 goto nosave; 1598 MCLAIM(m, &tcp_mowner); 1599 tcp_saveti->m_len = iphlen; 1600 m_copydata(m, 0, iphlen, 1601 mtod(tcp_saveti, void *)); 1602 } 1603 1604 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1605 m_freem(tcp_saveti); 1606 tcp_saveti = NULL; 1607 } else { 1608 tcp_saveti->m_len += sizeof(struct tcphdr); 1609 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1610 sizeof(struct tcphdr)); 1611 } 1612 nosave:; 1613 } 1614 1615 if (so->so_options & SO_ACCEPTCONN) { 1616 union syn_cache_sa src; 1617 union syn_cache_sa dst; 1618 1619 KASSERT(tp->t_state == TCPS_LISTEN); 1620 1621 memset(&src, 0, sizeof(src)); 1622 memset(&dst, 0, sizeof(dst)); 1623 switch (af) { 1624 #ifdef INET 1625 case AF_INET: 1626 src.sin.sin_len = sizeof(struct sockaddr_in); 1627 src.sin.sin_family = AF_INET; 1628 src.sin.sin_addr = ip->ip_src; 1629 src.sin.sin_port = th->th_sport; 1630 1631 dst.sin.sin_len = sizeof(struct sockaddr_in); 1632 dst.sin.sin_family = AF_INET; 1633 dst.sin.sin_addr = ip->ip_dst; 1634 dst.sin.sin_port = th->th_dport; 1635 break; 1636 #endif 1637 #ifdef INET6 1638 case AF_INET6: 1639 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1640 src.sin6.sin6_family = AF_INET6; 1641 src.sin6.sin6_addr = ip6->ip6_src; 1642 src.sin6.sin6_port = th->th_sport; 1643 1644 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1645 dst.sin6.sin6_family = AF_INET6; 1646 dst.sin6.sin6_addr = ip6->ip6_dst; 1647 dst.sin6.sin6_port = th->th_dport; 1648 break; 1649 #endif /* INET6 */ 1650 } 1651 1652 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1653 if (tiflags & TH_RST) { 1654 syn_cache_reset(&src.sa, &dst.sa, th); 1655 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1656 (TH_ACK|TH_SYN)) { 1657 /* 1658 * Received a SYN,ACK. This should never 1659 * happen while we are in LISTEN. Send an RST. 1660 */ 1661 goto badsyn; 1662 } else if (tiflags & TH_ACK) { 1663 so = syn_cache_get(&src.sa, &dst.sa, 1664 th, toff, tlen, so, m); 1665 if (so == NULL) { 1666 /* 1667 * We don't have a SYN for this ACK; 1668 * send an RST. 1669 */ 1670 goto badsyn; 1671 } else if (so == (struct socket *)(-1)) { 1672 /* 1673 * We were unable to create the 1674 * connection. If the 3-way handshake 1675 * was completed, and RST has been 1676 * sent to the peer. Since the mbuf 1677 * might be in use for the reply, do 1678 * not free it. 1679 */ 1680 m = NULL; 1681 } else { 1682 /* 1683 * We have created a full-blown 1684 * connection. 1685 */ 1686 tp = NULL; 1687 inp = NULL; 1688 #ifdef INET6 1689 in6p = NULL; 1690 #endif 1691 switch (so->so_proto->pr_domain->dom_family) { 1692 #ifdef INET 1693 case AF_INET: 1694 inp = sotoinpcb(so); 1695 tp = intotcpcb(inp); 1696 break; 1697 #endif 1698 #ifdef INET6 1699 case AF_INET6: 1700 in6p = sotoin6pcb(so); 1701 tp = in6totcpcb(in6p); 1702 break; 1703 #endif 1704 } 1705 if (tp == NULL) 1706 goto badsyn; /*XXX*/ 1707 tiwin <<= tp->snd_scale; 1708 goto after_listen; 1709 } 1710 } else { 1711 /* 1712 * None of RST, SYN or ACK was set. 1713 * This is an invalid packet for a 1714 * TCB in LISTEN state. Send a RST. 1715 */ 1716 goto badsyn; 1717 } 1718 } else { 1719 /* 1720 * Received a SYN. 1721 */ 1722 1723 #ifdef INET6 1724 /* 1725 * If deprecated address is forbidden, we do 1726 * not accept SYN to deprecated interface 1727 * address to prevent any new inbound 1728 * connection from getting established. 1729 * When we do not accept SYN, we send a TCP 1730 * RST, with deprecated source address (instead 1731 * of dropping it). We compromise it as it is 1732 * much better for peer to send a RST, and 1733 * RST will be the final packet for the 1734 * exchange. 1735 * 1736 * If we do not forbid deprecated addresses, we 1737 * accept the SYN packet. RFC2462 does not 1738 * suggest dropping SYN in this case. 1739 * If we decipher RFC2462 5.5.4, it says like 1740 * this: 1741 * 1. use of deprecated addr with existing 1742 * communication is okay - "SHOULD continue 1743 * to be used" 1744 * 2. use of it with new communication: 1745 * (2a) "SHOULD NOT be used if alternate 1746 * address with sufficient scope is 1747 * available" 1748 * (2b) nothing mentioned otherwise. 1749 * Here we fall into (2b) case as we have no 1750 * choice in our source address selection - we 1751 * must obey the peer. 1752 * 1753 * The wording in RFC2462 is confusing, and 1754 * there are multiple description text for 1755 * deprecated address handling - worse, they 1756 * are not exactly the same. I believe 5.5.4 1757 * is the best one, so we follow 5.5.4. 1758 */ 1759 if (af == AF_INET6 && !ip6_use_deprecated) { 1760 struct in6_ifaddr *ia6; 1761 int s; 1762 struct ifnet *rcvif = m_get_rcvif(m, &s); 1763 if (rcvif == NULL) 1764 goto dropwithreset; /* XXX */ 1765 if ((ia6 = in6ifa_ifpwithaddr(rcvif, 1766 &ip6->ip6_dst)) && 1767 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1768 tp = NULL; 1769 m_put_rcvif(rcvif, &s); 1770 goto dropwithreset; 1771 } 1772 m_put_rcvif(rcvif, &s); 1773 } 1774 #endif 1775 1776 #if defined(IPSEC) 1777 if (ipsec_used) { 1778 switch (af) { 1779 #ifdef INET 1780 case AF_INET: 1781 /* 1782 * inp can be NULL when receiving an 1783 * IPv4 packet on an IPv4-mapped IPv6 1784 * address. 1785 */ 1786 KASSERT(inp == NULL || 1787 sotoinpcb(so) == inp); 1788 if (!ipsec4_in_reject(m, inp)) 1789 break; 1790 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1791 tp = NULL; 1792 goto dropwithreset; 1793 #endif 1794 #ifdef INET6 1795 case AF_INET6: 1796 KASSERT(sotoin6pcb(so) == in6p); 1797 if (!ipsec6_in_reject(m, in6p)) 1798 break; 1799 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1800 tp = NULL; 1801 goto dropwithreset; 1802 #endif /*INET6*/ 1803 } 1804 } 1805 #endif /*IPSEC*/ 1806 1807 /* 1808 * LISTEN socket received a SYN from itself? This 1809 * can't possibly be valid; drop the packet. 1810 */ 1811 if (th->th_sport == th->th_dport) { 1812 int eq = 0; 1813 1814 switch (af) { 1815 #ifdef INET 1816 case AF_INET: 1817 eq = in_hosteq(ip->ip_src, ip->ip_dst); 1818 break; 1819 #endif 1820 #ifdef INET6 1821 case AF_INET6: 1822 eq = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, 1823 &ip6->ip6_dst); 1824 break; 1825 #endif 1826 } 1827 if (eq) { 1828 TCP_STATINC(TCP_STAT_BADSYN); 1829 goto drop; 1830 } 1831 } 1832 1833 /* 1834 * SYN looks ok; create compressed TCP 1835 * state for it. 1836 */ 1837 if (so->so_qlen <= so->so_qlimit && 1838 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1839 so, m, optp, optlen, &opti)) 1840 m = NULL; 1841 } 1842 1843 goto drop; 1844 } 1845 1846 after_listen: 1847 /* 1848 * Should not happen now that all embryonic connections 1849 * are handled with compressed state. 1850 */ 1851 KASSERT(tp->t_state != TCPS_LISTEN); 1852 1853 /* 1854 * Segment received on connection. 1855 * Reset idle time and keep-alive timer. 1856 */ 1857 tp->t_rcvtime = tcp_now; 1858 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1859 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1860 1861 /* 1862 * Process options. 1863 */ 1864 #ifdef TCP_SIGNATURE 1865 if (optp || (tp->t_flags & TF_SIGNATURE)) 1866 #else 1867 if (optp) 1868 #endif 1869 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1870 goto drop; 1871 1872 if (TCP_SACK_ENABLED(tp)) { 1873 tcp_del_sackholes(tp, th); 1874 } 1875 1876 if (TCP_ECN_ALLOWED(tp)) { 1877 if (tiflags & TH_CWR) { 1878 tp->t_flags &= ~TF_ECN_SND_ECE; 1879 } 1880 switch (iptos & IPTOS_ECN_MASK) { 1881 case IPTOS_ECN_CE: 1882 tp->t_flags |= TF_ECN_SND_ECE; 1883 TCP_STATINC(TCP_STAT_ECN_CE); 1884 break; 1885 case IPTOS_ECN_ECT0: 1886 TCP_STATINC(TCP_STAT_ECN_ECT); 1887 break; 1888 case IPTOS_ECN_ECT1: 1889 /* XXX: ignore for now -- rpaulo */ 1890 break; 1891 } 1892 /* 1893 * Congestion experienced. 1894 * Ignore if we are already trying to recover. 1895 */ 1896 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1897 tp->t_congctl->cong_exp(tp); 1898 } 1899 1900 if (opti.ts_present && opti.ts_ecr) { 1901 /* 1902 * Calculate the RTT from the returned time stamp and the 1903 * connection's time base. If the time stamp is later than 1904 * the current time, or is extremely old, fall back to non-1323 1905 * RTT calculation. Since ts_rtt is unsigned, we can test both 1906 * at the same time. 1907 * 1908 * Note that ts_rtt is in units of slow ticks (500 1909 * ms). Since most earthbound RTTs are < 500 ms, 1910 * observed values will have large quantization noise. 1911 * Our smoothed RTT is then the fraction of observed 1912 * samples that are 1 tick instead of 0 (times 500 1913 * ms). 1914 * 1915 * ts_rtt is increased by 1 to denote a valid sample, 1916 * with 0 indicating an invalid measurement. This 1917 * extra 1 must be removed when ts_rtt is used, or 1918 * else an an erroneous extra 500 ms will result. 1919 */ 1920 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1921 if (ts_rtt > TCP_PAWS_IDLE) 1922 ts_rtt = 0; 1923 } else { 1924 ts_rtt = 0; 1925 } 1926 1927 /* 1928 * Header prediction: check for the two common cases 1929 * of a uni-directional data xfer. If the packet has 1930 * no control flags, is in-sequence, the window didn't 1931 * change and we're not retransmitting, it's a 1932 * candidate. If the length is zero and the ack moved 1933 * forward, we're the sender side of the xfer. Just 1934 * free the data acked & wake any higher level process 1935 * that was blocked waiting for space. If the length 1936 * is non-zero and the ack didn't move, we're the 1937 * receiver side. If we're getting packets in-order 1938 * (the reassembly queue is empty), add the data to 1939 * the socket buffer and note that we need a delayed ack. 1940 */ 1941 if (tp->t_state == TCPS_ESTABLISHED && 1942 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1943 == TH_ACK && 1944 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1945 th->th_seq == tp->rcv_nxt && 1946 tiwin && tiwin == tp->snd_wnd && 1947 tp->snd_nxt == tp->snd_max) { 1948 1949 /* 1950 * If last ACK falls within this segment's sequence numbers, 1951 * record the timestamp. 1952 * NOTE that the test is modified according to the latest 1953 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 1954 * 1955 * note that we already know 1956 * TSTMP_GEQ(opti.ts_val, tp->ts_recent) 1957 */ 1958 if (opti.ts_present && 1959 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1960 tp->ts_recent_age = tcp_now; 1961 tp->ts_recent = opti.ts_val; 1962 } 1963 1964 if (tlen == 0) { 1965 /* Ack prediction. */ 1966 if (SEQ_GT(th->th_ack, tp->snd_una) && 1967 SEQ_LEQ(th->th_ack, tp->snd_max) && 1968 tp->snd_cwnd >= tp->snd_wnd && 1969 tp->t_partialacks < 0) { 1970 /* 1971 * this is a pure ack for outstanding data. 1972 */ 1973 if (ts_rtt) 1974 tcp_xmit_timer(tp, ts_rtt - 1); 1975 else if (tp->t_rtttime && 1976 SEQ_GT(th->th_ack, tp->t_rtseq)) 1977 tcp_xmit_timer(tp, 1978 tcp_now - tp->t_rtttime); 1979 acked = th->th_ack - tp->snd_una; 1980 tcps = TCP_STAT_GETREF(); 1981 tcps[TCP_STAT_PREDACK]++; 1982 tcps[TCP_STAT_RCVACKPACK]++; 1983 tcps[TCP_STAT_RCVACKBYTE] += acked; 1984 TCP_STAT_PUTREF(); 1985 nd6_hint(tp); 1986 1987 if (acked > (tp->t_lastoff - tp->t_inoff)) 1988 tp->t_lastm = NULL; 1989 sbdrop(&so->so_snd, acked); 1990 tp->t_lastoff -= acked; 1991 1992 icmp_check(tp, th, acked); 1993 1994 tp->snd_una = th->th_ack; 1995 tp->snd_fack = tp->snd_una; 1996 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1997 tp->snd_high = tp->snd_una; 1998 m_freem(m); 1999 2000 /* 2001 * If all outstanding data are acked, stop 2002 * retransmit timer, otherwise restart timer 2003 * using current (possibly backed-off) value. 2004 * If process is waiting for space, 2005 * wakeup/selnotify/signal. If data 2006 * are ready to send, let tcp_output 2007 * decide between more output or persist. 2008 */ 2009 if (tp->snd_una == tp->snd_max) 2010 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2011 else if (TCP_TIMER_ISARMED(tp, 2012 TCPT_PERSIST) == 0) 2013 TCP_TIMER_ARM(tp, TCPT_REXMT, 2014 tp->t_rxtcur); 2015 2016 sowwakeup(so); 2017 if (so->so_snd.sb_cc) { 2018 KERNEL_LOCK(1, NULL); 2019 (void) tcp_output(tp); 2020 KERNEL_UNLOCK_ONE(NULL); 2021 } 2022 if (tcp_saveti) 2023 m_freem(tcp_saveti); 2024 return; 2025 } 2026 } else if (th->th_ack == tp->snd_una && 2027 TAILQ_FIRST(&tp->segq) == NULL && 2028 tlen <= sbspace(&so->so_rcv)) { 2029 int newsize = 0; /* automatic sockbuf scaling */ 2030 2031 /* 2032 * this is a pure, in-sequence data packet 2033 * with nothing on the reassembly queue and 2034 * we have enough buffer space to take it. 2035 */ 2036 tp->rcv_nxt += tlen; 2037 tcps = TCP_STAT_GETREF(); 2038 tcps[TCP_STAT_PREDDAT]++; 2039 tcps[TCP_STAT_RCVPACK]++; 2040 tcps[TCP_STAT_RCVBYTE] += tlen; 2041 TCP_STAT_PUTREF(); 2042 nd6_hint(tp); 2043 2044 /* 2045 * Automatic sizing enables the performance of large buffers 2046 * and most of the efficiency of small ones by only allocating 2047 * space when it is needed. 2048 * 2049 * On the receive side the socket buffer memory is only rarely 2050 * used to any significant extent. This allows us to be much 2051 * more aggressive in scaling the receive socket buffer. For 2052 * the case that the buffer space is actually used to a large 2053 * extent and we run out of kernel memory we can simply drop 2054 * the new segments; TCP on the sender will just retransmit it 2055 * later. Setting the buffer size too big may only consume too 2056 * much kernel memory if the application doesn't read() from 2057 * the socket or packet loss or reordering makes use of the 2058 * reassembly queue. 2059 * 2060 * The criteria to step up the receive buffer one notch are: 2061 * 1. the number of bytes received during the time it takes 2062 * one timestamp to be reflected back to us (the RTT); 2063 * 2. received bytes per RTT is within seven eighth of the 2064 * current socket buffer size; 2065 * 3. receive buffer size has not hit maximal automatic size; 2066 * 2067 * This algorithm does one step per RTT at most and only if 2068 * we receive a bulk stream w/o packet losses or reorderings. 2069 * Shrinking the buffer during idle times is not necessary as 2070 * it doesn't consume any memory when idle. 2071 * 2072 * TODO: Only step up if the application is actually serving 2073 * the buffer to better manage the socket buffer resources. 2074 */ 2075 if (tcp_do_autorcvbuf && 2076 opti.ts_ecr && 2077 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 2078 if (opti.ts_ecr > tp->rfbuf_ts && 2079 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 2080 if (tp->rfbuf_cnt > 2081 (so->so_rcv.sb_hiwat / 8 * 7) && 2082 so->so_rcv.sb_hiwat < 2083 tcp_autorcvbuf_max) { 2084 newsize = 2085 min(so->so_rcv.sb_hiwat + 2086 tcp_autorcvbuf_inc, 2087 tcp_autorcvbuf_max); 2088 } 2089 /* Start over with next RTT. */ 2090 tp->rfbuf_ts = 0; 2091 tp->rfbuf_cnt = 0; 2092 } else 2093 tp->rfbuf_cnt += tlen; /* add up */ 2094 } 2095 2096 /* 2097 * Drop TCP, IP headers and TCP options then add data 2098 * to socket buffer. 2099 */ 2100 if (so->so_state & SS_CANTRCVMORE) 2101 m_freem(m); 2102 else { 2103 /* 2104 * Set new socket buffer size. 2105 * Give up when limit is reached. 2106 */ 2107 if (newsize) 2108 if (!sbreserve(&so->so_rcv, 2109 newsize, so)) 2110 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 2111 m_adj(m, toff + off); 2112 sbappendstream(&so->so_rcv, m); 2113 } 2114 sorwakeup(so); 2115 tcp_setup_ack(tp, th); 2116 if (tp->t_flags & TF_ACKNOW) { 2117 KERNEL_LOCK(1, NULL); 2118 (void) tcp_output(tp); 2119 KERNEL_UNLOCK_ONE(NULL); 2120 } 2121 if (tcp_saveti) 2122 m_freem(tcp_saveti); 2123 return; 2124 } 2125 } 2126 2127 /* 2128 * Compute mbuf offset to TCP data segment. 2129 */ 2130 hdroptlen = toff + off; 2131 2132 /* 2133 * Calculate amount of space in receive window, 2134 * and then do TCP input processing. 2135 * Receive window is amount of space in rcv queue, 2136 * but not less than advertised window. 2137 */ 2138 { int win; 2139 2140 win = sbspace(&so->so_rcv); 2141 if (win < 0) 2142 win = 0; 2143 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 2144 } 2145 2146 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 2147 tp->rfbuf_ts = 0; 2148 tp->rfbuf_cnt = 0; 2149 2150 switch (tp->t_state) { 2151 /* 2152 * If the state is SYN_SENT: 2153 * if seg contains an ACK, but not for our SYN, drop the input. 2154 * if seg contains a RST, then drop the connection. 2155 * if seg does not contain SYN, then drop it. 2156 * Otherwise this is an acceptable SYN segment 2157 * initialize tp->rcv_nxt and tp->irs 2158 * if seg contains ack then advance tp->snd_una 2159 * if seg contains a ECE and ECN support is enabled, the stream 2160 * is ECN capable. 2161 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2162 * arrange for segment to be acked (eventually) 2163 * continue processing rest of data/controls, beginning with URG 2164 */ 2165 case TCPS_SYN_SENT: 2166 if ((tiflags & TH_ACK) && 2167 (SEQ_LEQ(th->th_ack, tp->iss) || 2168 SEQ_GT(th->th_ack, tp->snd_max))) 2169 goto dropwithreset; 2170 if (tiflags & TH_RST) { 2171 if (tiflags & TH_ACK) 2172 tp = tcp_drop(tp, ECONNREFUSED); 2173 goto drop; 2174 } 2175 if ((tiflags & TH_SYN) == 0) 2176 goto drop; 2177 if (tiflags & TH_ACK) { 2178 tp->snd_una = th->th_ack; 2179 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2180 tp->snd_nxt = tp->snd_una; 2181 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2182 tp->snd_high = tp->snd_una; 2183 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2184 2185 if ((tiflags & TH_ECE) && tcp_do_ecn) { 2186 tp->t_flags |= TF_ECN_PERMIT; 2187 TCP_STATINC(TCP_STAT_ECN_SHS); 2188 } 2189 2190 } 2191 tp->irs = th->th_seq; 2192 tcp_rcvseqinit(tp); 2193 tp->t_flags |= TF_ACKNOW; 2194 tcp_mss_from_peer(tp, opti.maxseg); 2195 2196 /* 2197 * Initialize the initial congestion window. If we 2198 * had to retransmit the SYN, we must initialize cwnd 2199 * to 1 segment (i.e. the Loss Window). 2200 */ 2201 if (tp->t_flags & TF_SYN_REXMT) 2202 tp->snd_cwnd = tp->t_peermss; 2203 else { 2204 int ss = tcp_init_win; 2205 #ifdef INET 2206 if (inp != NULL && in_localaddr(inp->inp_faddr)) 2207 ss = tcp_init_win_local; 2208 #endif 2209 #ifdef INET6 2210 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 2211 ss = tcp_init_win_local; 2212 #endif 2213 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 2214 } 2215 2216 tcp_rmx_rtt(tp); 2217 if (tiflags & TH_ACK) { 2218 TCP_STATINC(TCP_STAT_CONNECTS); 2219 /* 2220 * move tcp_established before soisconnected 2221 * because upcall handler can drive tcp_output 2222 * functionality. 2223 * XXX we might call soisconnected at the end of 2224 * all processing 2225 */ 2226 tcp_established(tp); 2227 soisconnected(so); 2228 /* Do window scaling on this connection? */ 2229 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2230 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2231 tp->snd_scale = tp->requested_s_scale; 2232 tp->rcv_scale = tp->request_r_scale; 2233 } 2234 TCP_REASS_LOCK(tp); 2235 (void) tcp_reass(tp, NULL, NULL, &tlen); 2236 /* 2237 * if we didn't have to retransmit the SYN, 2238 * use its rtt as our initial srtt & rtt var. 2239 */ 2240 if (tp->t_rtttime) 2241 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2242 } else 2243 tp->t_state = TCPS_SYN_RECEIVED; 2244 2245 /* 2246 * Advance th->th_seq to correspond to first data byte. 2247 * If data, trim to stay within window, 2248 * dropping FIN if necessary. 2249 */ 2250 th->th_seq++; 2251 if (tlen > tp->rcv_wnd) { 2252 todrop = tlen - tp->rcv_wnd; 2253 m_adj(m, -todrop); 2254 tlen = tp->rcv_wnd; 2255 tiflags &= ~TH_FIN; 2256 tcps = TCP_STAT_GETREF(); 2257 tcps[TCP_STAT_RCVPACKAFTERWIN]++; 2258 tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop; 2259 TCP_STAT_PUTREF(); 2260 } 2261 tp->snd_wl1 = th->th_seq - 1; 2262 tp->rcv_up = th->th_seq; 2263 goto step6; 2264 2265 /* 2266 * If the state is SYN_RECEIVED: 2267 * If seg contains an ACK, but not for our SYN, drop the input 2268 * and generate an RST. See page 36, rfc793 2269 */ 2270 case TCPS_SYN_RECEIVED: 2271 if ((tiflags & TH_ACK) && 2272 (SEQ_LEQ(th->th_ack, tp->iss) || 2273 SEQ_GT(th->th_ack, tp->snd_max))) 2274 goto dropwithreset; 2275 break; 2276 } 2277 2278 /* 2279 * States other than LISTEN or SYN_SENT. 2280 * First check timestamp, if present. 2281 * Then check that at least some bytes of segment are within 2282 * receive window. If segment begins before rcv_nxt, 2283 * drop leading data (and SYN); if nothing left, just ack. 2284 * 2285 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2286 * and it's less than ts_recent, drop it. 2287 */ 2288 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2289 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2290 2291 /* Check to see if ts_recent is over 24 days old. */ 2292 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2293 /* 2294 * Invalidate ts_recent. If this segment updates 2295 * ts_recent, the age will be reset later and ts_recent 2296 * will get a valid value. If it does not, setting 2297 * ts_recent to zero will at least satisfy the 2298 * requirement that zero be placed in the timestamp 2299 * echo reply when ts_recent isn't valid. The 2300 * age isn't reset until we get a valid ts_recent 2301 * because we don't want out-of-order segments to be 2302 * dropped when ts_recent is old. 2303 */ 2304 tp->ts_recent = 0; 2305 } else { 2306 tcps = TCP_STAT_GETREF(); 2307 tcps[TCP_STAT_RCVDUPPACK]++; 2308 tcps[TCP_STAT_RCVDUPBYTE] += tlen; 2309 tcps[TCP_STAT_PAWSDROP]++; 2310 TCP_STAT_PUTREF(); 2311 tcp_new_dsack(tp, th->th_seq, tlen); 2312 goto dropafterack; 2313 } 2314 } 2315 2316 todrop = tp->rcv_nxt - th->th_seq; 2317 dupseg = false; 2318 if (todrop > 0) { 2319 if (tiflags & TH_SYN) { 2320 tiflags &= ~TH_SYN; 2321 th->th_seq++; 2322 if (th->th_urp > 1) 2323 th->th_urp--; 2324 else { 2325 tiflags &= ~TH_URG; 2326 th->th_urp = 0; 2327 } 2328 todrop--; 2329 } 2330 if (todrop > tlen || 2331 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2332 /* 2333 * Any valid FIN or RST must be to the left of the 2334 * window. At this point the FIN or RST must be a 2335 * duplicate or out of sequence; drop it. 2336 */ 2337 if (tiflags & TH_RST) 2338 goto drop; 2339 tiflags &= ~(TH_FIN|TH_RST); 2340 /* 2341 * Send an ACK to resynchronize and drop any data. 2342 * But keep on processing for RST or ACK. 2343 */ 2344 tp->t_flags |= TF_ACKNOW; 2345 todrop = tlen; 2346 dupseg = true; 2347 tcps = TCP_STAT_GETREF(); 2348 tcps[TCP_STAT_RCVDUPPACK]++; 2349 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 2350 TCP_STAT_PUTREF(); 2351 } else if ((tiflags & TH_RST) && 2352 th->th_seq != tp->rcv_nxt) { 2353 /* 2354 * Test for reset before adjusting the sequence 2355 * number for overlapping data. 2356 */ 2357 goto dropafterack_ratelim; 2358 } else { 2359 tcps = TCP_STAT_GETREF(); 2360 tcps[TCP_STAT_RCVPARTDUPPACK]++; 2361 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 2362 TCP_STAT_PUTREF(); 2363 } 2364 tcp_new_dsack(tp, th->th_seq, todrop); 2365 hdroptlen += todrop; /*drop from head afterwards*/ 2366 th->th_seq += todrop; 2367 tlen -= todrop; 2368 if (th->th_urp > todrop) 2369 th->th_urp -= todrop; 2370 else { 2371 tiflags &= ~TH_URG; 2372 th->th_urp = 0; 2373 } 2374 } 2375 2376 /* 2377 * If new data are received on a connection after the 2378 * user processes are gone, then RST the other end. 2379 */ 2380 if ((so->so_state & SS_NOFDREF) && 2381 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2382 tp = tcp_close(tp); 2383 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 2384 goto dropwithreset; 2385 } 2386 2387 /* 2388 * If segment ends after window, drop trailing data 2389 * (and PUSH and FIN); if nothing left, just ACK. 2390 */ 2391 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 2392 if (todrop > 0) { 2393 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 2394 if (todrop >= tlen) { 2395 /* 2396 * The segment actually starts after the window. 2397 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2398 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2399 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2400 */ 2401 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 2402 /* 2403 * If a new connection request is received 2404 * while in TIME_WAIT, drop the old connection 2405 * and start over if the sequence numbers 2406 * are above the previous ones. 2407 * 2408 * NOTE: We will checksum the packet again, and 2409 * so we need to put the header fields back into 2410 * network order! 2411 * XXX This kind of sucks, but we don't expect 2412 * XXX this to happen very often, so maybe it 2413 * XXX doesn't matter so much. 2414 */ 2415 if (tiflags & TH_SYN && 2416 tp->t_state == TCPS_TIME_WAIT && 2417 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2418 tp = tcp_close(tp); 2419 tcp_fields_to_net(th); 2420 goto findpcb; 2421 } 2422 /* 2423 * If window is closed can only take segments at 2424 * window edge, and have to drop data and PUSH from 2425 * incoming segments. Continue processing, but 2426 * remember to ack. Otherwise, drop segment 2427 * and (if not RST) ack. 2428 */ 2429 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2430 tp->t_flags |= TF_ACKNOW; 2431 TCP_STATINC(TCP_STAT_RCVWINPROBE); 2432 } else 2433 goto dropafterack; 2434 } else 2435 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 2436 m_adj(m, -todrop); 2437 tlen -= todrop; 2438 tiflags &= ~(TH_PUSH|TH_FIN); 2439 } 2440 2441 /* 2442 * If last ACK falls within this segment's sequence numbers, 2443 * record the timestamp. 2444 * NOTE: 2445 * 1) That the test incorporates suggestions from the latest 2446 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 2447 * 2) That updating only on newer timestamps interferes with 2448 * our earlier PAWS tests, so this check should be solely 2449 * predicated on the sequence space of this segment. 2450 * 3) That we modify the segment boundary check to be 2451 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2452 * instead of RFC1323's 2453 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2454 * This modified check allows us to overcome RFC1323's 2455 * limitations as described in Stevens TCP/IP Illustrated 2456 * Vol. 2 p.869. In such cases, we can still calculate the 2457 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2458 */ 2459 if (opti.ts_present && 2460 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2461 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2462 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2463 tp->ts_recent_age = tcp_now; 2464 tp->ts_recent = opti.ts_val; 2465 } 2466 2467 /* 2468 * If the RST bit is set examine the state: 2469 * SYN_RECEIVED STATE: 2470 * If passive open, return to LISTEN state. 2471 * If active open, inform user that connection was refused. 2472 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 2473 * Inform user that connection was reset, and close tcb. 2474 * CLOSING, LAST_ACK, TIME_WAIT STATES 2475 * Close the tcb. 2476 */ 2477 if (tiflags & TH_RST) { 2478 if (th->th_seq != tp->rcv_nxt) 2479 goto dropafterack_ratelim; 2480 2481 switch (tp->t_state) { 2482 case TCPS_SYN_RECEIVED: 2483 so->so_error = ECONNREFUSED; 2484 goto close; 2485 2486 case TCPS_ESTABLISHED: 2487 case TCPS_FIN_WAIT_1: 2488 case TCPS_FIN_WAIT_2: 2489 case TCPS_CLOSE_WAIT: 2490 so->so_error = ECONNRESET; 2491 close: 2492 tp->t_state = TCPS_CLOSED; 2493 TCP_STATINC(TCP_STAT_DROPS); 2494 tp = tcp_close(tp); 2495 goto drop; 2496 2497 case TCPS_CLOSING: 2498 case TCPS_LAST_ACK: 2499 case TCPS_TIME_WAIT: 2500 tp = tcp_close(tp); 2501 goto drop; 2502 } 2503 } 2504 2505 /* 2506 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2507 * we must be in a synchronized state. RFC791 states (under RST 2508 * generation) that any unacceptable segment (an out-of-order SYN 2509 * qualifies) received in a synchronized state must elicit only an 2510 * empty acknowledgment segment ... and the connection remains in 2511 * the same state. 2512 */ 2513 if (tiflags & TH_SYN) { 2514 if (tp->rcv_nxt == th->th_seq) { 2515 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2516 TH_ACK); 2517 if (tcp_saveti) 2518 m_freem(tcp_saveti); 2519 return; 2520 } 2521 2522 goto dropafterack_ratelim; 2523 } 2524 2525 /* 2526 * If the ACK bit is off we drop the segment and return. 2527 */ 2528 if ((tiflags & TH_ACK) == 0) { 2529 if (tp->t_flags & TF_ACKNOW) 2530 goto dropafterack; 2531 else 2532 goto drop; 2533 } 2534 2535 /* 2536 * Ack processing. 2537 */ 2538 switch (tp->t_state) { 2539 2540 /* 2541 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2542 * ESTABLISHED state and continue processing, otherwise 2543 * send an RST. 2544 */ 2545 case TCPS_SYN_RECEIVED: 2546 if (SEQ_GT(tp->snd_una, th->th_ack) || 2547 SEQ_GT(th->th_ack, tp->snd_max)) 2548 goto dropwithreset; 2549 TCP_STATINC(TCP_STAT_CONNECTS); 2550 soisconnected(so); 2551 tcp_established(tp); 2552 /* Do window scaling? */ 2553 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2554 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2555 tp->snd_scale = tp->requested_s_scale; 2556 tp->rcv_scale = tp->request_r_scale; 2557 } 2558 TCP_REASS_LOCK(tp); 2559 (void) tcp_reass(tp, NULL, NULL, &tlen); 2560 tp->snd_wl1 = th->th_seq - 1; 2561 /* fall into ... */ 2562 2563 /* 2564 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2565 * ACKs. If the ack is in the range 2566 * tp->snd_una < th->th_ack <= tp->snd_max 2567 * then advance tp->snd_una to th->th_ack and drop 2568 * data from the retransmission queue. If this ACK reflects 2569 * more up to date window information we update our window information. 2570 */ 2571 case TCPS_ESTABLISHED: 2572 case TCPS_FIN_WAIT_1: 2573 case TCPS_FIN_WAIT_2: 2574 case TCPS_CLOSE_WAIT: 2575 case TCPS_CLOSING: 2576 case TCPS_LAST_ACK: 2577 case TCPS_TIME_WAIT: 2578 2579 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2580 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2581 TCP_STATINC(TCP_STAT_RCVDUPACK); 2582 /* 2583 * If we have outstanding data (other than 2584 * a window probe), this is a completely 2585 * duplicate ack (ie, window info didn't 2586 * change), the ack is the biggest we've 2587 * seen and we've seen exactly our rexmt 2588 * threshhold of them, assume a packet 2589 * has been dropped and retransmit it. 2590 * Kludge snd_nxt & the congestion 2591 * window so we send only this one 2592 * packet. 2593 */ 2594 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2595 th->th_ack != tp->snd_una) 2596 tp->t_dupacks = 0; 2597 else if (tp->t_partialacks < 0 && 2598 (++tp->t_dupacks == tcprexmtthresh || 2599 TCP_FACK_FASTRECOV(tp))) { 2600 /* 2601 * Do the fast retransmit, and adjust 2602 * congestion control paramenters. 2603 */ 2604 if (tp->t_congctl->fast_retransmit(tp, th)) { 2605 /* False fast retransmit */ 2606 break; 2607 } else 2608 goto drop; 2609 } else if (tp->t_dupacks > tcprexmtthresh) { 2610 tp->snd_cwnd += tp->t_segsz; 2611 KERNEL_LOCK(1, NULL); 2612 (void) tcp_output(tp); 2613 KERNEL_UNLOCK_ONE(NULL); 2614 goto drop; 2615 } 2616 } else { 2617 /* 2618 * If the ack appears to be very old, only 2619 * allow data that is in-sequence. This 2620 * makes it somewhat more difficult to insert 2621 * forged data by guessing sequence numbers. 2622 * Sent an ack to try to update the send 2623 * sequence number on the other side. 2624 */ 2625 if (tlen && th->th_seq != tp->rcv_nxt && 2626 SEQ_LT(th->th_ack, 2627 tp->snd_una - tp->max_sndwnd)) 2628 goto dropafterack; 2629 } 2630 break; 2631 } 2632 /* 2633 * If the congestion window was inflated to account 2634 * for the other side's cached packets, retract it. 2635 */ 2636 tp->t_congctl->fast_retransmit_newack(tp, th); 2637 2638 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2639 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2640 goto dropafterack; 2641 } 2642 acked = th->th_ack - tp->snd_una; 2643 tcps = TCP_STAT_GETREF(); 2644 tcps[TCP_STAT_RCVACKPACK]++; 2645 tcps[TCP_STAT_RCVACKBYTE] += acked; 2646 TCP_STAT_PUTREF(); 2647 2648 /* 2649 * If we have a timestamp reply, update smoothed 2650 * round trip time. If no timestamp is present but 2651 * transmit timer is running and timed sequence 2652 * number was acked, update smoothed round trip time. 2653 * Since we now have an rtt measurement, cancel the 2654 * timer backoff (cf., Phil Karn's retransmit alg.). 2655 * Recompute the initial retransmit timer. 2656 */ 2657 if (ts_rtt) 2658 tcp_xmit_timer(tp, ts_rtt - 1); 2659 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2660 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2661 2662 /* 2663 * If all outstanding data is acked, stop retransmit 2664 * timer and remember to restart (more output or persist). 2665 * If there is more data to be acked, restart retransmit 2666 * timer, using current (possibly backed-off) value. 2667 */ 2668 if (th->th_ack == tp->snd_max) { 2669 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2670 needoutput = 1; 2671 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2672 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2673 2674 /* 2675 * New data has been acked, adjust the congestion window. 2676 */ 2677 tp->t_congctl->newack(tp, th); 2678 2679 nd6_hint(tp); 2680 if (acked > so->so_snd.sb_cc) { 2681 tp->snd_wnd -= so->so_snd.sb_cc; 2682 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2683 ourfinisacked = 1; 2684 } else { 2685 if (acked > (tp->t_lastoff - tp->t_inoff)) 2686 tp->t_lastm = NULL; 2687 sbdrop(&so->so_snd, acked); 2688 tp->t_lastoff -= acked; 2689 if (tp->snd_wnd > acked) 2690 tp->snd_wnd -= acked; 2691 else 2692 tp->snd_wnd = 0; 2693 ourfinisacked = 0; 2694 } 2695 sowwakeup(so); 2696 2697 icmp_check(tp, th, acked); 2698 2699 tp->snd_una = th->th_ack; 2700 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2701 tp->snd_fack = tp->snd_una; 2702 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2703 tp->snd_nxt = tp->snd_una; 2704 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2705 tp->snd_high = tp->snd_una; 2706 2707 switch (tp->t_state) { 2708 2709 /* 2710 * In FIN_WAIT_1 STATE in addition to the processing 2711 * for the ESTABLISHED state if our FIN is now acknowledged 2712 * then enter FIN_WAIT_2. 2713 */ 2714 case TCPS_FIN_WAIT_1: 2715 if (ourfinisacked) { 2716 /* 2717 * If we can't receive any more 2718 * data, then closing user can proceed. 2719 * Starting the timer is contrary to the 2720 * specification, but if we don't get a FIN 2721 * we'll hang forever. 2722 */ 2723 if (so->so_state & SS_CANTRCVMORE) { 2724 soisdisconnected(so); 2725 if (tp->t_maxidle > 0) 2726 TCP_TIMER_ARM(tp, TCPT_2MSL, 2727 tp->t_maxidle); 2728 } 2729 tp->t_state = TCPS_FIN_WAIT_2; 2730 } 2731 break; 2732 2733 /* 2734 * In CLOSING STATE in addition to the processing for 2735 * the ESTABLISHED state if the ACK acknowledges our FIN 2736 * then enter the TIME-WAIT state, otherwise ignore 2737 * the segment. 2738 */ 2739 case TCPS_CLOSING: 2740 if (ourfinisacked) { 2741 tp->t_state = TCPS_TIME_WAIT; 2742 tcp_canceltimers(tp); 2743 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2744 soisdisconnected(so); 2745 } 2746 break; 2747 2748 /* 2749 * In LAST_ACK, we may still be waiting for data to drain 2750 * and/or to be acked, as well as for the ack of our FIN. 2751 * If our FIN is now acknowledged, delete the TCB, 2752 * enter the closed state and return. 2753 */ 2754 case TCPS_LAST_ACK: 2755 if (ourfinisacked) { 2756 tp = tcp_close(tp); 2757 goto drop; 2758 } 2759 break; 2760 2761 /* 2762 * In TIME_WAIT state the only thing that should arrive 2763 * is a retransmission of the remote FIN. Acknowledge 2764 * it and restart the finack timer. 2765 */ 2766 case TCPS_TIME_WAIT: 2767 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2768 goto dropafterack; 2769 } 2770 } 2771 2772 step6: 2773 /* 2774 * Update window information. 2775 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2776 */ 2777 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2778 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2779 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2780 /* keep track of pure window updates */ 2781 if (tlen == 0 && 2782 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2783 TCP_STATINC(TCP_STAT_RCVWINUPD); 2784 tp->snd_wnd = tiwin; 2785 tp->snd_wl1 = th->th_seq; 2786 tp->snd_wl2 = th->th_ack; 2787 if (tp->snd_wnd > tp->max_sndwnd) 2788 tp->max_sndwnd = tp->snd_wnd; 2789 needoutput = 1; 2790 } 2791 2792 /* 2793 * Process segments with URG. 2794 */ 2795 if ((tiflags & TH_URG) && th->th_urp && 2796 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2797 /* 2798 * This is a kludge, but if we receive and accept 2799 * random urgent pointers, we'll crash in 2800 * soreceive. It's hard to imagine someone 2801 * actually wanting to send this much urgent data. 2802 */ 2803 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2804 th->th_urp = 0; /* XXX */ 2805 tiflags &= ~TH_URG; /* XXX */ 2806 goto dodata; /* XXX */ 2807 } 2808 /* 2809 * If this segment advances the known urgent pointer, 2810 * then mark the data stream. This should not happen 2811 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2812 * a FIN has been received from the remote side. 2813 * In these states we ignore the URG. 2814 * 2815 * According to RFC961 (Assigned Protocols), 2816 * the urgent pointer points to the last octet 2817 * of urgent data. We continue, however, 2818 * to consider it to indicate the first octet 2819 * of data past the urgent section as the original 2820 * spec states (in one of two places). 2821 */ 2822 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2823 tp->rcv_up = th->th_seq + th->th_urp; 2824 so->so_oobmark = so->so_rcv.sb_cc + 2825 (tp->rcv_up - tp->rcv_nxt) - 1; 2826 if (so->so_oobmark == 0) 2827 so->so_state |= SS_RCVATMARK; 2828 sohasoutofband(so); 2829 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2830 } 2831 /* 2832 * Remove out of band data so doesn't get presented to user. 2833 * This can happen independent of advancing the URG pointer, 2834 * but if two URG's are pending at once, some out-of-band 2835 * data may creep in... ick. 2836 */ 2837 if (th->th_urp <= (u_int16_t) tlen 2838 #ifdef SO_OOBINLINE 2839 && (so->so_options & SO_OOBINLINE) == 0 2840 #endif 2841 ) 2842 tcp_pulloutofband(so, th, m, hdroptlen); 2843 } else 2844 /* 2845 * If no out of band data is expected, 2846 * pull receive urgent pointer along 2847 * with the receive window. 2848 */ 2849 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2850 tp->rcv_up = tp->rcv_nxt; 2851 dodata: /* XXX */ 2852 2853 /* 2854 * Process the segment text, merging it into the TCP sequencing queue, 2855 * and arranging for acknowledgement of receipt if necessary. 2856 * This process logically involves adjusting tp->rcv_wnd as data 2857 * is presented to the user (this happens in tcp_usrreq.c, 2858 * tcp_rcvd()). If a FIN has already been received on this 2859 * connection then we just ignore the text. 2860 */ 2861 if ((tlen || (tiflags & TH_FIN)) && 2862 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2863 /* 2864 * Insert segment ti into reassembly queue of tcp with 2865 * control block tp. Return TH_FIN if reassembly now includes 2866 * a segment with FIN. The macro form does the common case 2867 * inline (segment is the next to be received on an 2868 * established connection, and the queue is empty), 2869 * avoiding linkage into and removal from the queue and 2870 * repetition of various conversions. 2871 * Set DELACK for segments received in order, but ack 2872 * immediately when segments are out of order 2873 * (so fast retransmit can work). 2874 */ 2875 /* NOTE: this was TCP_REASS() macro, but used only once */ 2876 TCP_REASS_LOCK(tp); 2877 if (th->th_seq == tp->rcv_nxt && 2878 TAILQ_FIRST(&tp->segq) == NULL && 2879 tp->t_state == TCPS_ESTABLISHED) { 2880 tcp_setup_ack(tp, th); 2881 tp->rcv_nxt += tlen; 2882 tiflags = th->th_flags & TH_FIN; 2883 tcps = TCP_STAT_GETREF(); 2884 tcps[TCP_STAT_RCVPACK]++; 2885 tcps[TCP_STAT_RCVBYTE] += tlen; 2886 TCP_STAT_PUTREF(); 2887 nd6_hint(tp); 2888 if (so->so_state & SS_CANTRCVMORE) 2889 m_freem(m); 2890 else { 2891 m_adj(m, hdroptlen); 2892 sbappendstream(&(so)->so_rcv, m); 2893 } 2894 TCP_REASS_UNLOCK(tp); 2895 sorwakeup(so); 2896 } else { 2897 m_adj(m, hdroptlen); 2898 tiflags = tcp_reass(tp, th, m, &tlen); 2899 tp->t_flags |= TF_ACKNOW; 2900 } 2901 2902 /* 2903 * Note the amount of data that peer has sent into 2904 * our window, in order to estimate the sender's 2905 * buffer size. 2906 */ 2907 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2908 } else { 2909 m_freem(m); 2910 m = NULL; 2911 tiflags &= ~TH_FIN; 2912 } 2913 2914 /* 2915 * If FIN is received ACK the FIN and let the user know 2916 * that the connection is closing. Ignore a FIN received before 2917 * the connection is fully established. 2918 */ 2919 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2920 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2921 socantrcvmore(so); 2922 tp->t_flags |= TF_ACKNOW; 2923 tp->rcv_nxt++; 2924 } 2925 switch (tp->t_state) { 2926 2927 /* 2928 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2929 */ 2930 case TCPS_ESTABLISHED: 2931 tp->t_state = TCPS_CLOSE_WAIT; 2932 break; 2933 2934 /* 2935 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2936 * enter the CLOSING state. 2937 */ 2938 case TCPS_FIN_WAIT_1: 2939 tp->t_state = TCPS_CLOSING; 2940 break; 2941 2942 /* 2943 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2944 * starting the time-wait timer, turning off the other 2945 * standard timers. 2946 */ 2947 case TCPS_FIN_WAIT_2: 2948 tp->t_state = TCPS_TIME_WAIT; 2949 tcp_canceltimers(tp); 2950 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2951 soisdisconnected(so); 2952 break; 2953 2954 /* 2955 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2956 */ 2957 case TCPS_TIME_WAIT: 2958 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2959 break; 2960 } 2961 } 2962 #ifdef TCP_DEBUG 2963 if (so->so_options & SO_DEBUG) 2964 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2965 #endif 2966 2967 /* 2968 * Return any desired output. 2969 */ 2970 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2971 KERNEL_LOCK(1, NULL); 2972 (void) tcp_output(tp); 2973 KERNEL_UNLOCK_ONE(NULL); 2974 } 2975 if (tcp_saveti) 2976 m_freem(tcp_saveti); 2977 2978 if (tp->t_state == TCPS_TIME_WAIT 2979 && (so->so_state & SS_NOFDREF) 2980 && (tp->t_inpcb || af != AF_INET) 2981 && (tp->t_in6pcb || af != AF_INET6) 2982 && ((af == AF_INET ? tcp4_vtw_enable : tcp6_vtw_enable) & 1) != 0 2983 && TAILQ_EMPTY(&tp->segq) 2984 && vtw_add(af, tp)) { 2985 ; 2986 } 2987 return; 2988 2989 badsyn: 2990 /* 2991 * Received a bad SYN. Increment counters and dropwithreset. 2992 */ 2993 TCP_STATINC(TCP_STAT_BADSYN); 2994 tp = NULL; 2995 goto dropwithreset; 2996 2997 dropafterack: 2998 /* 2999 * Generate an ACK dropping incoming segment if it occupies 3000 * sequence space, where the ACK reflects our state. 3001 */ 3002 if (tiflags & TH_RST) 3003 goto drop; 3004 goto dropafterack2; 3005 3006 dropafterack_ratelim: 3007 /* 3008 * We may want to rate-limit ACKs against SYN/RST attack. 3009 */ 3010 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 3011 tcp_ackdrop_ppslim) == 0) { 3012 /* XXX stat */ 3013 goto drop; 3014 } 3015 /* ...fall into dropafterack2... */ 3016 3017 dropafterack2: 3018 m_freem(m); 3019 tp->t_flags |= TF_ACKNOW; 3020 KERNEL_LOCK(1, NULL); 3021 (void) tcp_output(tp); 3022 KERNEL_UNLOCK_ONE(NULL); 3023 if (tcp_saveti) 3024 m_freem(tcp_saveti); 3025 return; 3026 3027 dropwithreset_ratelim: 3028 /* 3029 * We may want to rate-limit RSTs in certain situations, 3030 * particularly if we are sending an RST in response to 3031 * an attempt to connect to or otherwise communicate with 3032 * a port for which we have no socket. 3033 */ 3034 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 3035 tcp_rst_ppslim) == 0) { 3036 /* XXX stat */ 3037 goto drop; 3038 } 3039 /* ...fall into dropwithreset... */ 3040 3041 dropwithreset: 3042 /* 3043 * Generate a RST, dropping incoming segment. 3044 * Make ACK acceptable to originator of segment. 3045 */ 3046 if (tiflags & TH_RST) 3047 goto drop; 3048 if (tiflags & TH_ACK) { 3049 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3050 } else { 3051 if (tiflags & TH_SYN) 3052 tlen++; 3053 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 3054 TH_RST|TH_ACK); 3055 } 3056 if (tcp_saveti) 3057 m_freem(tcp_saveti); 3058 return; 3059 3060 badcsum: 3061 drop: 3062 /* 3063 * Drop space held by incoming segment and return. 3064 */ 3065 if (tp) { 3066 if (tp->t_inpcb) 3067 so = tp->t_inpcb->inp_socket; 3068 #ifdef INET6 3069 else if (tp->t_in6pcb) 3070 so = tp->t_in6pcb->in6p_socket; 3071 #endif 3072 else 3073 so = NULL; 3074 #ifdef TCP_DEBUG 3075 if (so && (so->so_options & SO_DEBUG) != 0) 3076 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 3077 #endif 3078 } 3079 if (tcp_saveti) 3080 m_freem(tcp_saveti); 3081 m_freem(m); 3082 return; 3083 } 3084 3085 #ifdef TCP_SIGNATURE 3086 int 3087 tcp_signature_apply(void *fstate, void *data, u_int len) 3088 { 3089 3090 MD5Update(fstate, (u_char *)data, len); 3091 return (0); 3092 } 3093 3094 struct secasvar * 3095 tcp_signature_getsav(struct mbuf *m) 3096 { 3097 struct ip *ip; 3098 struct ip6_hdr *ip6; 3099 3100 ip = mtod(m, struct ip *); 3101 switch (ip->ip_v) { 3102 case 4: 3103 ip = mtod(m, struct ip *); 3104 ip6 = NULL; 3105 break; 3106 case 6: 3107 ip = NULL; 3108 ip6 = mtod(m, struct ip6_hdr *); 3109 break; 3110 default: 3111 return (NULL); 3112 } 3113 3114 #ifdef IPSEC 3115 union sockaddr_union dst; 3116 3117 /* Extract the destination from the IP header in the mbuf. */ 3118 memset(&dst, 0, sizeof(union sockaddr_union)); 3119 if (ip != NULL) { 3120 dst.sa.sa_len = sizeof(struct sockaddr_in); 3121 dst.sa.sa_family = AF_INET; 3122 dst.sin.sin_addr = ip->ip_dst; 3123 } else { 3124 dst.sa.sa_len = sizeof(struct sockaddr_in6); 3125 dst.sa.sa_family = AF_INET6; 3126 dst.sin6.sin6_addr = ip6->ip6_dst; 3127 } 3128 3129 /* 3130 * Look up an SADB entry which matches the address of the peer. 3131 */ 3132 return KEY_LOOKUP_SA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI), 0, 0); 3133 #else 3134 return NULL; 3135 #endif 3136 } 3137 3138 int 3139 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 3140 struct secasvar *sav, char *sig) 3141 { 3142 MD5_CTX ctx; 3143 struct ip *ip; 3144 struct ipovly *ipovly; 3145 #ifdef INET6 3146 struct ip6_hdr *ip6; 3147 struct ip6_hdr_pseudo ip6pseudo; 3148 #endif /* INET6 */ 3149 struct ippseudo ippseudo; 3150 struct tcphdr th0; 3151 int l, tcphdrlen; 3152 3153 if (sav == NULL) 3154 return (-1); 3155 3156 tcphdrlen = th->th_off * 4; 3157 3158 switch (mtod(m, struct ip *)->ip_v) { 3159 case 4: 3160 MD5Init(&ctx); 3161 ip = mtod(m, struct ip *); 3162 memset(&ippseudo, 0, sizeof(ippseudo)); 3163 ipovly = (struct ipovly *)ip; 3164 ippseudo.ippseudo_src = ipovly->ih_src; 3165 ippseudo.ippseudo_dst = ipovly->ih_dst; 3166 ippseudo.ippseudo_pad = 0; 3167 ippseudo.ippseudo_p = IPPROTO_TCP; 3168 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 3169 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 3170 break; 3171 #if INET6 3172 case 6: 3173 MD5Init(&ctx); 3174 ip6 = mtod(m, struct ip6_hdr *); 3175 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 3176 ip6pseudo.ip6ph_src = ip6->ip6_src; 3177 in6_clearscope(&ip6pseudo.ip6ph_src); 3178 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 3179 in6_clearscope(&ip6pseudo.ip6ph_dst); 3180 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 3181 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 3182 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 3183 break; 3184 #endif /* INET6 */ 3185 default: 3186 return (-1); 3187 } 3188 3189 th0 = *th; 3190 th0.th_sum = 0; 3191 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 3192 3193 l = m->m_pkthdr.len - thoff - tcphdrlen; 3194 if (l > 0) 3195 m_apply(m, thoff + tcphdrlen, 3196 m->m_pkthdr.len - thoff - tcphdrlen, 3197 tcp_signature_apply, &ctx); 3198 3199 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 3200 MD5Final(sig, &ctx); 3201 3202 return (0); 3203 } 3204 #endif 3205 3206 /* 3207 * tcp_dooptions: parse and process tcp options. 3208 * 3209 * returns -1 if this segment should be dropped. (eg. wrong signature) 3210 * otherwise returns 0. 3211 */ 3212 3213 static int 3214 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, struct tcphdr *th, 3215 struct mbuf *m, int toff, struct tcp_opt_info *oi) 3216 { 3217 u_int16_t mss; 3218 int opt, optlen = 0; 3219 #ifdef TCP_SIGNATURE 3220 void *sigp = NULL; 3221 char sigbuf[TCP_SIGLEN]; 3222 struct secasvar *sav = NULL; 3223 #endif 3224 3225 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 3226 opt = cp[0]; 3227 if (opt == TCPOPT_EOL) 3228 break; 3229 if (opt == TCPOPT_NOP) 3230 optlen = 1; 3231 else { 3232 if (cnt < 2) 3233 break; 3234 optlen = cp[1]; 3235 if (optlen < 2 || optlen > cnt) 3236 break; 3237 } 3238 switch (opt) { 3239 3240 default: 3241 continue; 3242 3243 case TCPOPT_MAXSEG: 3244 if (optlen != TCPOLEN_MAXSEG) 3245 continue; 3246 if (!(th->th_flags & TH_SYN)) 3247 continue; 3248 if (TCPS_HAVERCVDSYN(tp->t_state)) 3249 continue; 3250 bcopy(cp + 2, &mss, sizeof(mss)); 3251 oi->maxseg = ntohs(mss); 3252 break; 3253 3254 case TCPOPT_WINDOW: 3255 if (optlen != TCPOLEN_WINDOW) 3256 continue; 3257 if (!(th->th_flags & TH_SYN)) 3258 continue; 3259 if (TCPS_HAVERCVDSYN(tp->t_state)) 3260 continue; 3261 tp->t_flags |= TF_RCVD_SCALE; 3262 tp->requested_s_scale = cp[2]; 3263 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3264 char buf[INET6_ADDRSTRLEN]; 3265 struct ip *ip = mtod(m, struct ip *); 3266 #ifdef INET6 3267 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 3268 #endif 3269 if (ip) 3270 in_print(buf, sizeof(buf), 3271 &ip->ip_src); 3272 #ifdef INET6 3273 else if (ip6) 3274 in6_print(buf, sizeof(buf), 3275 &ip6->ip6_src); 3276 #endif 3277 else 3278 strlcpy(buf, "(unknown)", sizeof(buf)); 3279 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3280 "assuming %d\n", 3281 tp->requested_s_scale, buf, 3282 TCP_MAX_WINSHIFT); 3283 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3284 } 3285 break; 3286 3287 case TCPOPT_TIMESTAMP: 3288 if (optlen != TCPOLEN_TIMESTAMP) 3289 continue; 3290 oi->ts_present = 1; 3291 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 3292 NTOHL(oi->ts_val); 3293 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 3294 NTOHL(oi->ts_ecr); 3295 3296 if (!(th->th_flags & TH_SYN)) 3297 continue; 3298 if (TCPS_HAVERCVDSYN(tp->t_state)) 3299 continue; 3300 /* 3301 * A timestamp received in a SYN makes 3302 * it ok to send timestamp requests and replies. 3303 */ 3304 tp->t_flags |= TF_RCVD_TSTMP; 3305 tp->ts_recent = oi->ts_val; 3306 tp->ts_recent_age = tcp_now; 3307 break; 3308 3309 case TCPOPT_SACK_PERMITTED: 3310 if (optlen != TCPOLEN_SACK_PERMITTED) 3311 continue; 3312 if (!(th->th_flags & TH_SYN)) 3313 continue; 3314 if (TCPS_HAVERCVDSYN(tp->t_state)) 3315 continue; 3316 if (tcp_do_sack) { 3317 tp->t_flags |= TF_SACK_PERMIT; 3318 tp->t_flags |= TF_WILL_SACK; 3319 } 3320 break; 3321 3322 case TCPOPT_SACK: 3323 tcp_sack_option(tp, th, cp, optlen); 3324 break; 3325 #ifdef TCP_SIGNATURE 3326 case TCPOPT_SIGNATURE: 3327 if (optlen != TCPOLEN_SIGNATURE) 3328 continue; 3329 if (sigp && memcmp(sigp, cp + 2, TCP_SIGLEN)) 3330 return (-1); 3331 3332 sigp = sigbuf; 3333 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3334 tp->t_flags |= TF_SIGNATURE; 3335 break; 3336 #endif 3337 } 3338 } 3339 3340 #ifndef TCP_SIGNATURE 3341 return 0; 3342 #else 3343 if (tp->t_flags & TF_SIGNATURE) { 3344 sav = tcp_signature_getsav(m); 3345 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3346 return (-1); 3347 } 3348 3349 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) 3350 goto out; 3351 3352 if (sigp) { 3353 char sig[TCP_SIGLEN]; 3354 3355 tcp_fields_to_net(th); 3356 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3357 tcp_fields_to_host(th); 3358 goto out; 3359 } 3360 tcp_fields_to_host(th); 3361 3362 if (memcmp(sig, sigp, TCP_SIGLEN)) { 3363 TCP_STATINC(TCP_STAT_BADSIG); 3364 goto out; 3365 } else 3366 TCP_STATINC(TCP_STAT_GOODSIG); 3367 3368 key_sa_recordxfer(sav, m); 3369 KEY_SA_UNREF(&sav); 3370 } 3371 return 0; 3372 out: 3373 if (sav != NULL) 3374 KEY_SA_UNREF(&sav); 3375 return -1; 3376 #endif 3377 } 3378 3379 /* 3380 * Pull out of band byte out of a segment so 3381 * it doesn't appear in the user's data queue. 3382 * It is still reflected in the segment length for 3383 * sequencing purposes. 3384 */ 3385 void 3386 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3387 struct mbuf *m, int off) 3388 { 3389 int cnt = off + th->th_urp - 1; 3390 3391 while (cnt >= 0) { 3392 if (m->m_len > cnt) { 3393 char *cp = mtod(m, char *) + cnt; 3394 struct tcpcb *tp = sototcpcb(so); 3395 3396 tp->t_iobc = *cp; 3397 tp->t_oobflags |= TCPOOB_HAVEDATA; 3398 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3399 m->m_len--; 3400 return; 3401 } 3402 cnt -= m->m_len; 3403 m = m->m_next; 3404 if (m == 0) 3405 break; 3406 } 3407 panic("tcp_pulloutofband"); 3408 } 3409 3410 /* 3411 * Collect new round-trip time estimate 3412 * and update averages and current timeout. 3413 * 3414 * rtt is in units of slow ticks (typically 500 ms) -- essentially the 3415 * difference of two timestamps. 3416 */ 3417 void 3418 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3419 { 3420 int32_t delta; 3421 3422 TCP_STATINC(TCP_STAT_RTTUPDATED); 3423 if (tp->t_srtt != 0) { 3424 /* 3425 * Compute the amount to add to srtt for smoothing, 3426 * *alpha, or 2^(-TCP_RTT_SHIFT). Because 3427 * srtt is stored in 1/32 slow ticks, we conceptually 3428 * shift left 5 bits, subtract srtt to get the 3429 * diference, and then shift right by TCP_RTT_SHIFT 3430 * (3) to obtain 1/8 of the difference. 3431 */ 3432 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3433 /* 3434 * This can never happen, because delta's lowest 3435 * possible value is 1/8 of t_srtt. But if it does, 3436 * set srtt to some reasonable value, here chosen 3437 * as 1/8 tick. 3438 */ 3439 if ((tp->t_srtt += delta) <= 0) 3440 tp->t_srtt = 1 << 2; 3441 /* 3442 * RFC2988 requires that rttvar be updated first. 3443 * This code is compliant because "delta" is the old 3444 * srtt minus the new observation (scaled). 3445 * 3446 * RFC2988 says: 3447 * rttvar = (1-beta) * rttvar + beta * |srtt-observed| 3448 * 3449 * delta is in units of 1/32 ticks, and has then been 3450 * divided by 8. This is equivalent to being in 1/16s 3451 * units and divided by 4. Subtract from it 1/4 of 3452 * the existing rttvar to form the (signed) amount to 3453 * adjust. 3454 */ 3455 if (delta < 0) 3456 delta = -delta; 3457 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3458 /* 3459 * As with srtt, this should never happen. There is 3460 * no support in RFC2988 for this operation. But 1/4s 3461 * as rttvar when faced with something arguably wrong 3462 * is ok. 3463 */ 3464 if ((tp->t_rttvar += delta) <= 0) 3465 tp->t_rttvar = 1 << 2; 3466 3467 /* 3468 * If srtt exceeds .01 second, ensure we use the 'remote' MSL 3469 * Problem is: it doesn't work. Disabled by defaulting 3470 * tcp_rttlocal to 0; see corresponding code in 3471 * tcp_subr that selects local vs remote in a different way. 3472 * 3473 * The static branch prediction hint here should be removed 3474 * when the rtt estimator is fixed and the rtt_enable code 3475 * is turned back on. 3476 */ 3477 if (__predict_false(tcp_rttlocal) && tcp_msl_enable 3478 && tp->t_srtt > tcp_msl_remote_threshold 3479 && tp->t_msl < tcp_msl_remote) { 3480 tp->t_msl = tcp_msl_remote; 3481 } 3482 } else { 3483 /* 3484 * This is the first measurement. Per RFC2988, 2.2, 3485 * set rtt=R and srtt=R/2. 3486 * For srtt, storage representation is 1/32 ticks, 3487 * so shift left by 5. 3488 * For rttvar, storage representation is 1/16 ticks, 3489 * So shift left by 4, but then right by 1 to halve. 3490 */ 3491 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3492 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3493 } 3494 tp->t_rtttime = 0; 3495 tp->t_rxtshift = 0; 3496 3497 /* 3498 * the retransmit should happen at rtt + 4 * rttvar. 3499 * Because of the way we do the smoothing, srtt and rttvar 3500 * will each average +1/2 tick of bias. When we compute 3501 * the retransmit timer, we want 1/2 tick of rounding and 3502 * 1 extra tick because of +-1/2 tick uncertainty in the 3503 * firing of the timer. The bias will give us exactly the 3504 * 1.5 tick we need. But, because the bias is 3505 * statistical, we have to test that we don't drop below 3506 * the minimum feasible timer (which is 2 ticks). 3507 */ 3508 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3509 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3510 3511 /* 3512 * We received an ack for a packet that wasn't retransmitted; 3513 * it is probably safe to discard any error indications we've 3514 * received recently. This isn't quite right, but close enough 3515 * for now (a route might have failed after we sent a segment, 3516 * and the return path might not be symmetrical). 3517 */ 3518 tp->t_softerror = 0; 3519 } 3520 3521 3522 /* 3523 * TCP compressed state engine. Currently used to hold compressed 3524 * state for SYN_RECEIVED. 3525 */ 3526 3527 u_long syn_cache_count; 3528 u_int32_t syn_hash1, syn_hash2; 3529 3530 #define SYN_HASH(sa, sp, dp) \ 3531 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3532 ((u_int32_t)(sp)))^syn_hash2))) 3533 #ifndef INET6 3534 #define SYN_HASHALL(hash, src, dst) \ 3535 do { \ 3536 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3537 ((const struct sockaddr_in *)(src))->sin_port, \ 3538 ((const struct sockaddr_in *)(dst))->sin_port); \ 3539 } while (/*CONSTCOND*/ 0) 3540 #else 3541 #define SYN_HASH6(sa, sp, dp) \ 3542 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3543 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3544 & 0x7fffffff) 3545 3546 #define SYN_HASHALL(hash, src, dst) \ 3547 do { \ 3548 switch ((src)->sa_family) { \ 3549 case AF_INET: \ 3550 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3551 ((const struct sockaddr_in *)(src))->sin_port, \ 3552 ((const struct sockaddr_in *)(dst))->sin_port); \ 3553 break; \ 3554 case AF_INET6: \ 3555 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3556 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3557 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3558 break; \ 3559 default: \ 3560 hash = 0; \ 3561 } \ 3562 } while (/*CONSTCOND*/0) 3563 #endif /* INET6 */ 3564 3565 static struct pool syn_cache_pool; 3566 3567 /* 3568 * We don't estimate RTT with SYNs, so each packet starts with the default 3569 * RTT and each timer step has a fixed timeout value. 3570 */ 3571 static inline void 3572 syn_cache_timer_arm(struct syn_cache *sc) 3573 { 3574 3575 TCPT_RANGESET(sc->sc_rxtcur, 3576 TCPTV_SRTTDFLT * tcp_backoff[sc->sc_rxtshift], TCPTV_MIN, 3577 TCPTV_REXMTMAX); 3578 callout_reset(&sc->sc_timer, 3579 sc->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, sc); 3580 } 3581 3582 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3583 3584 static inline void 3585 syn_cache_rm(struct syn_cache *sc) 3586 { 3587 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3588 sc, sc_bucketq); 3589 sc->sc_tp = NULL; 3590 LIST_REMOVE(sc, sc_tpq); 3591 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3592 callout_stop(&sc->sc_timer); 3593 syn_cache_count--; 3594 } 3595 3596 static inline void 3597 syn_cache_put(struct syn_cache *sc) 3598 { 3599 if (sc->sc_ipopts) 3600 (void) m_free(sc->sc_ipopts); 3601 rtcache_free(&sc->sc_route); 3602 sc->sc_flags |= SCF_DEAD; 3603 if (!callout_invoking(&sc->sc_timer)) 3604 callout_schedule(&(sc)->sc_timer, 1); 3605 } 3606 3607 void 3608 syn_cache_init(void) 3609 { 3610 int i; 3611 3612 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 3613 "synpl", NULL, IPL_SOFTNET); 3614 3615 /* Initialize the hash buckets. */ 3616 for (i = 0; i < tcp_syn_cache_size; i++) 3617 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3618 } 3619 3620 void 3621 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3622 { 3623 struct syn_cache_head *scp; 3624 struct syn_cache *sc2; 3625 int s; 3626 3627 /* 3628 * If there are no entries in the hash table, reinitialize 3629 * the hash secrets. 3630 */ 3631 if (syn_cache_count == 0) { 3632 syn_hash1 = cprng_fast32(); 3633 syn_hash2 = cprng_fast32(); 3634 } 3635 3636 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3637 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3638 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3639 3640 /* 3641 * Make sure that we don't overflow the per-bucket 3642 * limit or the total cache size limit. 3643 */ 3644 s = splsoftnet(); 3645 if (scp->sch_length >= tcp_syn_bucket_limit) { 3646 TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW); 3647 /* 3648 * The bucket is full. Toss the oldest element in the 3649 * bucket. This will be the first entry in the bucket. 3650 */ 3651 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3652 #ifdef DIAGNOSTIC 3653 /* 3654 * This should never happen; we should always find an 3655 * entry in our bucket. 3656 */ 3657 if (sc2 == NULL) 3658 panic("syn_cache_insert: bucketoverflow: impossible"); 3659 #endif 3660 syn_cache_rm(sc2); 3661 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3662 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3663 struct syn_cache_head *scp2, *sce; 3664 3665 TCP_STATINC(TCP_STAT_SC_OVERFLOWED); 3666 /* 3667 * The cache is full. Toss the oldest entry in the 3668 * first non-empty bucket we can find. 3669 * 3670 * XXX We would really like to toss the oldest 3671 * entry in the cache, but we hope that this 3672 * condition doesn't happen very often. 3673 */ 3674 scp2 = scp; 3675 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3676 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3677 for (++scp2; scp2 != scp; scp2++) { 3678 if (scp2 >= sce) 3679 scp2 = &tcp_syn_cache[0]; 3680 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3681 break; 3682 } 3683 #ifdef DIAGNOSTIC 3684 /* 3685 * This should never happen; we should always find a 3686 * non-empty bucket. 3687 */ 3688 if (scp2 == scp) 3689 panic("syn_cache_insert: cacheoverflow: " 3690 "impossible"); 3691 #endif 3692 } 3693 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3694 syn_cache_rm(sc2); 3695 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3696 } 3697 3698 /* 3699 * Initialize the entry's timer. 3700 */ 3701 sc->sc_rxttot = 0; 3702 sc->sc_rxtshift = 0; 3703 syn_cache_timer_arm(sc); 3704 3705 /* Link it from tcpcb entry */ 3706 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3707 3708 /* Put it into the bucket. */ 3709 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3710 scp->sch_length++; 3711 syn_cache_count++; 3712 3713 TCP_STATINC(TCP_STAT_SC_ADDED); 3714 splx(s); 3715 } 3716 3717 /* 3718 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3719 * If we have retransmitted an entry the maximum number of times, expire 3720 * that entry. 3721 */ 3722 static void 3723 syn_cache_timer(void *arg) 3724 { 3725 struct syn_cache *sc = arg; 3726 3727 mutex_enter(softnet_lock); 3728 KERNEL_LOCK(1, NULL); 3729 3730 callout_ack(&sc->sc_timer); 3731 3732 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3733 TCP_STATINC(TCP_STAT_SC_DELAYED_FREE); 3734 goto free; 3735 } 3736 3737 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3738 /* Drop it -- too many retransmissions. */ 3739 goto dropit; 3740 } 3741 3742 /* 3743 * Compute the total amount of time this entry has 3744 * been on a queue. If this entry has been on longer 3745 * than the keep alive timer would allow, expire it. 3746 */ 3747 sc->sc_rxttot += sc->sc_rxtcur; 3748 if (sc->sc_rxttot >= tcp_keepinit) 3749 goto dropit; 3750 3751 TCP_STATINC(TCP_STAT_SC_RETRANSMITTED); 3752 (void)syn_cache_respond(sc); 3753 3754 /* Advance the timer back-off. */ 3755 sc->sc_rxtshift++; 3756 syn_cache_timer_arm(sc); 3757 3758 goto out; 3759 3760 dropit: 3761 TCP_STATINC(TCP_STAT_SC_TIMED_OUT); 3762 syn_cache_rm(sc); 3763 if (sc->sc_ipopts) 3764 (void) m_free(sc->sc_ipopts); 3765 rtcache_free(&sc->sc_route); 3766 3767 free: 3768 callout_destroy(&sc->sc_timer); 3769 pool_put(&syn_cache_pool, sc); 3770 3771 out: 3772 KERNEL_UNLOCK_ONE(NULL); 3773 mutex_exit(softnet_lock); 3774 } 3775 3776 /* 3777 * Remove syn cache created by the specified tcb entry, 3778 * because this does not make sense to keep them 3779 * (if there's no tcb entry, syn cache entry will never be used) 3780 */ 3781 void 3782 syn_cache_cleanup(struct tcpcb *tp) 3783 { 3784 struct syn_cache *sc, *nsc; 3785 int s; 3786 3787 s = splsoftnet(); 3788 3789 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3790 nsc = LIST_NEXT(sc, sc_tpq); 3791 3792 #ifdef DIAGNOSTIC 3793 if (sc->sc_tp != tp) 3794 panic("invalid sc_tp in syn_cache_cleanup"); 3795 #endif 3796 syn_cache_rm(sc); 3797 syn_cache_put(sc); /* calls pool_put but see spl above */ 3798 } 3799 /* just for safety */ 3800 LIST_INIT(&tp->t_sc); 3801 3802 splx(s); 3803 } 3804 3805 /* 3806 * Find an entry in the syn cache. 3807 */ 3808 struct syn_cache * 3809 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3810 struct syn_cache_head **headp) 3811 { 3812 struct syn_cache *sc; 3813 struct syn_cache_head *scp; 3814 u_int32_t hash; 3815 int s; 3816 3817 SYN_HASHALL(hash, src, dst); 3818 3819 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3820 *headp = scp; 3821 s = splsoftnet(); 3822 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3823 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3824 if (sc->sc_hash != hash) 3825 continue; 3826 if (!memcmp(&sc->sc_src, src, src->sa_len) && 3827 !memcmp(&sc->sc_dst, dst, dst->sa_len)) { 3828 splx(s); 3829 return (sc); 3830 } 3831 } 3832 splx(s); 3833 return (NULL); 3834 } 3835 3836 /* 3837 * This function gets called when we receive an ACK for a 3838 * socket in the LISTEN state. We look up the connection 3839 * in the syn cache, and if its there, we pull it out of 3840 * the cache and turn it into a full-blown connection in 3841 * the SYN-RECEIVED state. 3842 * 3843 * The return values may not be immediately obvious, and their effects 3844 * can be subtle, so here they are: 3845 * 3846 * NULL SYN was not found in cache; caller should drop the 3847 * packet and send an RST. 3848 * 3849 * -1 We were unable to create the new connection, and are 3850 * aborting it. An ACK,RST is being sent to the peer 3851 * (unless we got screwey sequence numbners; see below), 3852 * because the 3-way handshake has been completed. Caller 3853 * should not free the mbuf, since we may be using it. If 3854 * we are not, we will free it. 3855 * 3856 * Otherwise, the return value is a pointer to the new socket 3857 * associated with the connection. 3858 */ 3859 struct socket * 3860 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3861 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3862 struct socket *so, struct mbuf *m) 3863 { 3864 struct syn_cache *sc; 3865 struct syn_cache_head *scp; 3866 struct inpcb *inp = NULL; 3867 #ifdef INET6 3868 struct in6pcb *in6p = NULL; 3869 #endif 3870 struct tcpcb *tp = 0; 3871 int s; 3872 struct socket *oso; 3873 3874 s = splsoftnet(); 3875 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3876 splx(s); 3877 return (NULL); 3878 } 3879 3880 /* 3881 * Verify the sequence and ack numbers. Try getting the correct 3882 * response again. 3883 */ 3884 if ((th->th_ack != sc->sc_iss + 1) || 3885 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3886 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3887 m_freem(m); 3888 (void)syn_cache_respond(sc); 3889 splx(s); 3890 return ((struct socket *)(-1)); 3891 } 3892 3893 /* Remove this cache entry */ 3894 syn_cache_rm(sc); 3895 splx(s); 3896 3897 /* 3898 * Ok, create the full blown connection, and set things up 3899 * as they would have been set up if we had created the 3900 * connection when the SYN arrived. If we can't create 3901 * the connection, abort it. 3902 */ 3903 /* 3904 * inp still has the OLD in_pcb stuff, set the 3905 * v6-related flags on the new guy, too. This is 3906 * done particularly for the case where an AF_INET6 3907 * socket is bound only to a port, and a v4 connection 3908 * comes in on that port. 3909 * we also copy the flowinfo from the original pcb 3910 * to the new one. 3911 */ 3912 oso = so; 3913 so = sonewconn(so, true); 3914 if (so == NULL) 3915 goto resetandabort; 3916 3917 switch (so->so_proto->pr_domain->dom_family) { 3918 #ifdef INET 3919 case AF_INET: 3920 inp = sotoinpcb(so); 3921 break; 3922 #endif 3923 #ifdef INET6 3924 case AF_INET6: 3925 in6p = sotoin6pcb(so); 3926 break; 3927 #endif 3928 } 3929 switch (src->sa_family) { 3930 #ifdef INET 3931 case AF_INET: 3932 if (inp) { 3933 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3934 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3935 inp->inp_options = ip_srcroute(m); 3936 in_pcbstate(inp, INP_BOUND); 3937 if (inp->inp_options == NULL) { 3938 inp->inp_options = sc->sc_ipopts; 3939 sc->sc_ipopts = NULL; 3940 } 3941 } 3942 #ifdef INET6 3943 else if (in6p) { 3944 /* IPv4 packet to AF_INET6 socket */ 3945 memset(&in6p->in6p_laddr, 0, sizeof(in6p->in6p_laddr)); 3946 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3947 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3948 &in6p->in6p_laddr.s6_addr32[3], 3949 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3950 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3951 in6totcpcb(in6p)->t_family = AF_INET; 3952 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3953 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3954 else 3955 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3956 in6_pcbstate(in6p, IN6P_BOUND); 3957 } 3958 #endif 3959 break; 3960 #endif 3961 #ifdef INET6 3962 case AF_INET6: 3963 if (in6p) { 3964 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3965 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3966 in6_pcbstate(in6p, IN6P_BOUND); 3967 } 3968 break; 3969 #endif 3970 } 3971 #ifdef INET6 3972 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3973 struct in6pcb *oin6p = sotoin6pcb(oso); 3974 /* inherit socket options from the listening socket */ 3975 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3976 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3977 m_freem(in6p->in6p_options); 3978 in6p->in6p_options = 0; 3979 } 3980 ip6_savecontrol(in6p, &in6p->in6p_options, 3981 mtod(m, struct ip6_hdr *), m); 3982 } 3983 #endif 3984 3985 #if defined(IPSEC) 3986 if (ipsec_used) { 3987 /* 3988 * we make a copy of policy, instead of sharing the policy, for 3989 * better behavior in terms of SA lookup and dead SA removal. 3990 */ 3991 if (inp) { 3992 /* copy old policy into new socket's */ 3993 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, 3994 inp->inp_sp)) 3995 printf("tcp_input: could not copy policy\n"); 3996 } 3997 #ifdef INET6 3998 else if (in6p) { 3999 /* copy old policy into new socket's */ 4000 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 4001 in6p->in6p_sp)) 4002 printf("tcp_input: could not copy policy\n"); 4003 } 4004 #endif 4005 } 4006 #endif 4007 4008 /* 4009 * Give the new socket our cached route reference. 4010 */ 4011 if (inp) { 4012 rtcache_copy(&inp->inp_route, &sc->sc_route); 4013 rtcache_free(&sc->sc_route); 4014 } 4015 #ifdef INET6 4016 else { 4017 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 4018 rtcache_free(&sc->sc_route); 4019 } 4020 #endif 4021 4022 if (inp) { 4023 struct sockaddr_in sin; 4024 memcpy(&sin, src, src->sa_len); 4025 if (in_pcbconnect(inp, &sin, &lwp0)) { 4026 goto resetandabort; 4027 } 4028 } 4029 #ifdef INET6 4030 else if (in6p) { 4031 struct sockaddr_in6 sin6; 4032 memcpy(&sin6, src, src->sa_len); 4033 if (src->sa_family == AF_INET) { 4034 /* IPv4 packet to AF_INET6 socket */ 4035 in6_sin_2_v4mapsin6((struct sockaddr_in *)src, &sin6); 4036 } 4037 if (in6_pcbconnect(in6p, &sin6, NULL)) { 4038 goto resetandabort; 4039 } 4040 } 4041 #endif 4042 else { 4043 goto resetandabort; 4044 } 4045 4046 if (inp) 4047 tp = intotcpcb(inp); 4048 #ifdef INET6 4049 else if (in6p) 4050 tp = in6totcpcb(in6p); 4051 #endif 4052 else 4053 tp = NULL; 4054 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 4055 if (sc->sc_request_r_scale != 15) { 4056 tp->requested_s_scale = sc->sc_requested_s_scale; 4057 tp->request_r_scale = sc->sc_request_r_scale; 4058 tp->snd_scale = sc->sc_requested_s_scale; 4059 tp->rcv_scale = sc->sc_request_r_scale; 4060 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 4061 } 4062 if (sc->sc_flags & SCF_TIMESTAMP) 4063 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 4064 tp->ts_timebase = sc->sc_timebase; 4065 4066 tp->t_template = tcp_template(tp); 4067 if (tp->t_template == 0) { 4068 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 4069 so = NULL; 4070 m_freem(m); 4071 goto abort; 4072 } 4073 4074 tp->iss = sc->sc_iss; 4075 tp->irs = sc->sc_irs; 4076 tcp_sendseqinit(tp); 4077 tcp_rcvseqinit(tp); 4078 tp->t_state = TCPS_SYN_RECEIVED; 4079 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 4080 TCP_STATINC(TCP_STAT_ACCEPTS); 4081 4082 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 4083 tp->t_flags |= TF_WILL_SACK; 4084 4085 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 4086 tp->t_flags |= TF_ECN_PERMIT; 4087 4088 #ifdef TCP_SIGNATURE 4089 if (sc->sc_flags & SCF_SIGNATURE) 4090 tp->t_flags |= TF_SIGNATURE; 4091 #endif 4092 4093 /* Initialize tp->t_ourmss before we deal with the peer's! */ 4094 tp->t_ourmss = sc->sc_ourmaxseg; 4095 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 4096 4097 /* 4098 * Initialize the initial congestion window. If we 4099 * had to retransmit the SYN,ACK, we must initialize cwnd 4100 * to 1 segment (i.e. the Loss Window). 4101 */ 4102 if (sc->sc_rxtshift) 4103 tp->snd_cwnd = tp->t_peermss; 4104 else { 4105 int ss = tcp_init_win; 4106 #ifdef INET 4107 if (inp != NULL && in_localaddr(inp->inp_faddr)) 4108 ss = tcp_init_win_local; 4109 #endif 4110 #ifdef INET6 4111 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 4112 ss = tcp_init_win_local; 4113 #endif 4114 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 4115 } 4116 4117 tcp_rmx_rtt(tp); 4118 tp->snd_wl1 = sc->sc_irs; 4119 tp->rcv_up = sc->sc_irs + 1; 4120 4121 /* 4122 * This is what whould have happened in tcp_output() when 4123 * the SYN,ACK was sent. 4124 */ 4125 tp->snd_up = tp->snd_una; 4126 tp->snd_max = tp->snd_nxt = tp->iss+1; 4127 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 4128 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 4129 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 4130 tp->last_ack_sent = tp->rcv_nxt; 4131 tp->t_partialacks = -1; 4132 tp->t_dupacks = 0; 4133 4134 TCP_STATINC(TCP_STAT_SC_COMPLETED); 4135 s = splsoftnet(); 4136 syn_cache_put(sc); 4137 splx(s); 4138 return (so); 4139 4140 resetandabort: 4141 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 4142 abort: 4143 if (so != NULL) { 4144 (void) soqremque(so, 1); 4145 (void) soabort(so); 4146 mutex_enter(softnet_lock); 4147 } 4148 s = splsoftnet(); 4149 syn_cache_put(sc); 4150 splx(s); 4151 TCP_STATINC(TCP_STAT_SC_ABORTED); 4152 return ((struct socket *)(-1)); 4153 } 4154 4155 /* 4156 * This function is called when we get a RST for a 4157 * non-existent connection, so that we can see if the 4158 * connection is in the syn cache. If it is, zap it. 4159 */ 4160 4161 void 4162 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 4163 { 4164 struct syn_cache *sc; 4165 struct syn_cache_head *scp; 4166 int s = splsoftnet(); 4167 4168 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4169 splx(s); 4170 return; 4171 } 4172 if (SEQ_LT(th->th_seq, sc->sc_irs) || 4173 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 4174 splx(s); 4175 return; 4176 } 4177 syn_cache_rm(sc); 4178 TCP_STATINC(TCP_STAT_SC_RESET); 4179 syn_cache_put(sc); /* calls pool_put but see spl above */ 4180 splx(s); 4181 } 4182 4183 void 4184 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 4185 struct tcphdr *th) 4186 { 4187 struct syn_cache *sc; 4188 struct syn_cache_head *scp; 4189 int s; 4190 4191 s = splsoftnet(); 4192 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4193 splx(s); 4194 return; 4195 } 4196 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 4197 if (ntohl (th->th_seq) != sc->sc_iss) { 4198 splx(s); 4199 return; 4200 } 4201 4202 /* 4203 * If we've retransmitted 3 times and this is our second error, 4204 * we remove the entry. Otherwise, we allow it to continue on. 4205 * This prevents us from incorrectly nuking an entry during a 4206 * spurious network outage. 4207 * 4208 * See tcp_notify(). 4209 */ 4210 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 4211 sc->sc_flags |= SCF_UNREACH; 4212 splx(s); 4213 return; 4214 } 4215 4216 syn_cache_rm(sc); 4217 TCP_STATINC(TCP_STAT_SC_UNREACH); 4218 syn_cache_put(sc); /* calls pool_put but see spl above */ 4219 splx(s); 4220 } 4221 4222 /* 4223 * Given a LISTEN socket and an inbound SYN request, add 4224 * this to the syn cache, and send back a segment: 4225 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 4226 * to the source. 4227 * 4228 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 4229 * Doing so would require that we hold onto the data and deliver it 4230 * to the application. However, if we are the target of a SYN-flood 4231 * DoS attack, an attacker could send data which would eventually 4232 * consume all available buffer space if it were ACKed. By not ACKing 4233 * the data, we avoid this DoS scenario. 4234 */ 4235 int 4236 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 4237 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 4238 int optlen, struct tcp_opt_info *oi) 4239 { 4240 struct tcpcb tb, *tp; 4241 long win; 4242 struct syn_cache *sc; 4243 struct syn_cache_head *scp; 4244 struct mbuf *ipopts; 4245 struct tcp_opt_info opti; 4246 int s; 4247 4248 tp = sototcpcb(so); 4249 4250 memset(&opti, 0, sizeof(opti)); 4251 4252 /* 4253 * Initialize some local state. 4254 */ 4255 win = sbspace(&so->so_rcv); 4256 if (win > TCP_MAXWIN) 4257 win = TCP_MAXWIN; 4258 4259 #ifdef TCP_SIGNATURE 4260 if (optp || (tp->t_flags & TF_SIGNATURE)) 4261 #else 4262 if (optp) 4263 #endif 4264 { 4265 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4266 #ifdef TCP_SIGNATURE 4267 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4268 #endif 4269 tb.t_state = TCPS_LISTEN; 4270 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 4271 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 4272 return 0; 4273 } else 4274 tb.t_flags = 0; 4275 4276 switch (src->sa_family) { 4277 #ifdef INET 4278 case AF_INET: 4279 /* 4280 * Remember the IP options, if any. 4281 */ 4282 ipopts = ip_srcroute(m); 4283 break; 4284 #endif 4285 default: 4286 ipopts = NULL; 4287 } 4288 4289 /* 4290 * See if we already have an entry for this connection. 4291 * If we do, resend the SYN,ACK. We do not count this 4292 * as a retransmission (XXX though maybe we should). 4293 */ 4294 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4295 TCP_STATINC(TCP_STAT_SC_DUPESYN); 4296 if (ipopts) { 4297 /* 4298 * If we were remembering a previous source route, 4299 * forget it and use the new one we've been given. 4300 */ 4301 if (sc->sc_ipopts) 4302 (void)m_free(sc->sc_ipopts); 4303 sc->sc_ipopts = ipopts; 4304 } 4305 sc->sc_timestamp = tb.ts_recent; 4306 m_freem(m); 4307 if (syn_cache_respond(sc) == 0) { 4308 uint64_t *tcps = TCP_STAT_GETREF(); 4309 tcps[TCP_STAT_SNDACKS]++; 4310 tcps[TCP_STAT_SNDTOTAL]++; 4311 TCP_STAT_PUTREF(); 4312 } 4313 return 1; 4314 } 4315 4316 s = splsoftnet(); 4317 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4318 splx(s); 4319 if (sc == NULL) { 4320 if (ipopts) 4321 (void)m_free(ipopts); 4322 return 0; 4323 } 4324 4325 /* 4326 * Fill in the cache, and put the necessary IP and TCP 4327 * options into the reply. 4328 */ 4329 memset(sc, 0, sizeof(struct syn_cache)); 4330 callout_init(&sc->sc_timer, CALLOUT_MPSAFE); 4331 bcopy(src, &sc->sc_src, src->sa_len); 4332 bcopy(dst, &sc->sc_dst, dst->sa_len); 4333 sc->sc_flags = 0; 4334 sc->sc_ipopts = ipopts; 4335 sc->sc_irs = th->th_seq; 4336 switch (src->sa_family) { 4337 #ifdef INET 4338 case AF_INET: 4339 { 4340 struct sockaddr_in *srcin = (void *)src; 4341 struct sockaddr_in *dstin = (void *)dst; 4342 4343 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4344 &srcin->sin_addr, dstin->sin_port, 4345 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4346 break; 4347 } 4348 #endif /* INET */ 4349 #ifdef INET6 4350 case AF_INET6: 4351 { 4352 struct sockaddr_in6 *srcin6 = (void *)src; 4353 struct sockaddr_in6 *dstin6 = (void *)dst; 4354 4355 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4356 &srcin6->sin6_addr, dstin6->sin6_port, 4357 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4358 break; 4359 } 4360 #endif /* INET6 */ 4361 } 4362 sc->sc_peermaxseg = oi->maxseg; 4363 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4364 m_get_rcvif_NOMPSAFE(m) : NULL, sc->sc_src.sa.sa_family); 4365 sc->sc_win = win; 4366 sc->sc_timebase = tcp_now - 1; /* see tcp_newtcpcb() */ 4367 sc->sc_timestamp = tb.ts_recent; 4368 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4369 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4370 sc->sc_flags |= SCF_TIMESTAMP; 4371 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4372 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4373 sc->sc_requested_s_scale = tb.requested_s_scale; 4374 sc->sc_request_r_scale = 0; 4375 /* 4376 * Pick the smallest possible scaling factor that 4377 * will still allow us to scale up to sb_max. 4378 * 4379 * We do this because there are broken firewalls that 4380 * will corrupt the window scale option, leading to 4381 * the other endpoint believing that our advertised 4382 * window is unscaled. At scale factors larger than 4383 * 5 the unscaled window will drop below 1500 bytes, 4384 * leading to serious problems when traversing these 4385 * broken firewalls. 4386 * 4387 * With the default sbmax of 256K, a scale factor 4388 * of 3 will be chosen by this algorithm. Those who 4389 * choose a larger sbmax should watch out 4390 * for the compatiblity problems mentioned above. 4391 * 4392 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4393 * or <SYN,ACK>) segment itself is never scaled. 4394 */ 4395 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4396 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4397 sc->sc_request_r_scale++; 4398 } else { 4399 sc->sc_requested_s_scale = 15; 4400 sc->sc_request_r_scale = 15; 4401 } 4402 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4403 sc->sc_flags |= SCF_SACK_PERMIT; 4404 4405 /* 4406 * ECN setup packet received. 4407 */ 4408 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4409 sc->sc_flags |= SCF_ECN_PERMIT; 4410 4411 #ifdef TCP_SIGNATURE 4412 if (tb.t_flags & TF_SIGNATURE) 4413 sc->sc_flags |= SCF_SIGNATURE; 4414 #endif 4415 sc->sc_tp = tp; 4416 m_freem(m); 4417 if (syn_cache_respond(sc) == 0) { 4418 uint64_t *tcps = TCP_STAT_GETREF(); 4419 tcps[TCP_STAT_SNDACKS]++; 4420 tcps[TCP_STAT_SNDTOTAL]++; 4421 TCP_STAT_PUTREF(); 4422 syn_cache_insert(sc, tp); 4423 } else { 4424 s = splsoftnet(); 4425 /* 4426 * syn_cache_put() will try to schedule the timer, so 4427 * we need to initialize it 4428 */ 4429 syn_cache_timer_arm(sc); 4430 syn_cache_put(sc); 4431 splx(s); 4432 TCP_STATINC(TCP_STAT_SC_DROPPED); 4433 } 4434 return 1; 4435 } 4436 4437 /* 4438 * syn_cache_respond: (re)send SYN+ACK. 4439 * 4440 * returns 0 on success. otherwise returns an errno, typically ENOBUFS. 4441 */ 4442 4443 int 4444 syn_cache_respond(struct syn_cache *sc) 4445 { 4446 #ifdef INET6 4447 struct rtentry *rt = NULL; 4448 #endif 4449 struct route *ro; 4450 u_int8_t *optp; 4451 int optlen, error; 4452 u_int16_t tlen; 4453 struct ip *ip = NULL; 4454 #ifdef INET6 4455 struct ip6_hdr *ip6 = NULL; 4456 #endif 4457 struct tcpcb *tp = NULL; 4458 struct tcphdr *th; 4459 struct mbuf *m; 4460 u_int hlen; 4461 #ifdef TCP_SIGNATURE 4462 struct secasvar *sav = NULL; 4463 u_int8_t *sigp = NULL; 4464 #endif 4465 4466 ro = &sc->sc_route; 4467 switch (sc->sc_src.sa.sa_family) { 4468 case AF_INET: 4469 hlen = sizeof(struct ip); 4470 break; 4471 #ifdef INET6 4472 case AF_INET6: 4473 hlen = sizeof(struct ip6_hdr); 4474 break; 4475 #endif 4476 default: 4477 return (EAFNOSUPPORT); 4478 } 4479 4480 /* worst case scanario, since we don't know the option size yet */ 4481 tlen = hlen + sizeof(struct tcphdr) + MAX_TCPOPTLEN; 4482 4483 /* 4484 * Create the IP+TCP header from scratch. 4485 */ 4486 #ifdef DIAGNOSTIC 4487 if (max_linkhdr + tlen > MCLBYTES) 4488 return ENOBUFS; 4489 #endif 4490 4491 MGETHDR(m, M_DONTWAIT, MT_DATA); 4492 if (m && (max_linkhdr + tlen) > MHLEN) { 4493 MCLGET(m, M_DONTWAIT); 4494 if ((m->m_flags & M_EXT) == 0) { 4495 m_freem(m); 4496 m = NULL; 4497 } 4498 } 4499 if (m == NULL) 4500 return ENOBUFS; 4501 MCLAIM(m, &tcp_tx_mowner); 4502 4503 /* Fixup the mbuf. */ 4504 m->m_data += max_linkhdr; 4505 if (sc->sc_tp) 4506 tp = sc->sc_tp; 4507 m_reset_rcvif(m); 4508 memset(mtod(m, u_char *), 0, tlen); 4509 4510 switch (sc->sc_src.sa.sa_family) { 4511 case AF_INET: 4512 ip = mtod(m, struct ip *); 4513 ip->ip_v = 4; 4514 ip->ip_dst = sc->sc_src.sin.sin_addr; 4515 ip->ip_src = sc->sc_dst.sin.sin_addr; 4516 ip->ip_p = IPPROTO_TCP; 4517 th = (struct tcphdr *)(ip + 1); 4518 th->th_dport = sc->sc_src.sin.sin_port; 4519 th->th_sport = sc->sc_dst.sin.sin_port; 4520 break; 4521 #ifdef INET6 4522 case AF_INET6: 4523 ip6 = mtod(m, struct ip6_hdr *); 4524 ip6->ip6_vfc = IPV6_VERSION; 4525 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4526 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4527 ip6->ip6_nxt = IPPROTO_TCP; 4528 /* ip6_plen will be updated in ip6_output() */ 4529 th = (struct tcphdr *)(ip6 + 1); 4530 th->th_dport = sc->sc_src.sin6.sin6_port; 4531 th->th_sport = sc->sc_dst.sin6.sin6_port; 4532 break; 4533 #endif 4534 default: 4535 return ENOBUFS; 4536 } 4537 4538 th->th_seq = htonl(sc->sc_iss); 4539 th->th_ack = htonl(sc->sc_irs + 1); 4540 th->th_flags = TH_SYN|TH_ACK; 4541 th->th_win = htons(sc->sc_win); 4542 /* th_x2, th_sum, th_urp already 0 from memset */ 4543 4544 /* Tack on the TCP options. */ 4545 optp = (u_int8_t *)(th + 1); 4546 optlen = 0; 4547 *optp++ = TCPOPT_MAXSEG; 4548 *optp++ = TCPOLEN_MAXSEG; 4549 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4550 *optp++ = sc->sc_ourmaxseg & 0xff; 4551 optlen += TCPOLEN_MAXSEG; 4552 4553 if (sc->sc_request_r_scale != 15) { 4554 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4555 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4556 sc->sc_request_r_scale); 4557 optp += TCPOLEN_WINDOW + TCPOLEN_NOP; 4558 optlen += TCPOLEN_WINDOW + TCPOLEN_NOP; 4559 } 4560 4561 if (sc->sc_flags & SCF_SACK_PERMIT) { 4562 /* Let the peer know that we will SACK. */ 4563 *optp++ = TCPOPT_SACK_PERMITTED; 4564 *optp++ = TCPOLEN_SACK_PERMITTED; 4565 optlen += TCPOLEN_SACK_PERMITTED; 4566 } 4567 4568 if (sc->sc_flags & SCF_TIMESTAMP) { 4569 while (optlen % 4 != 2) { 4570 optlen += TCPOLEN_NOP; 4571 *optp++ = TCPOPT_NOP; 4572 } 4573 *optp++ = TCPOPT_TIMESTAMP; 4574 *optp++ = TCPOLEN_TIMESTAMP; 4575 u_int32_t *lp = (u_int32_t *)(optp); 4576 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4577 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4578 *lp = htonl(sc->sc_timestamp); 4579 optp += TCPOLEN_TIMESTAMP - 2; 4580 optlen += TCPOLEN_TIMESTAMP; 4581 } 4582 4583 #ifdef TCP_SIGNATURE 4584 if (sc->sc_flags & SCF_SIGNATURE) { 4585 sav = tcp_signature_getsav(m); 4586 if (sav == NULL) { 4587 if (m) 4588 m_freem(m); 4589 return (EPERM); 4590 } 4591 4592 *optp++ = TCPOPT_SIGNATURE; 4593 *optp++ = TCPOLEN_SIGNATURE; 4594 sigp = optp; 4595 memset(optp, 0, TCP_SIGLEN); 4596 optp += TCP_SIGLEN; 4597 optlen += TCPOLEN_SIGNATURE; 4598 4599 } 4600 #endif 4601 /* Terminate and pad TCP options to a 4 byte boundary. */ 4602 if (optlen % 4) { 4603 optlen += TCPOLEN_EOL; 4604 *optp++ = TCPOPT_EOL; 4605 } 4606 /* 4607 * According to RFC 793 (STD0007): 4608 * "The content of the header beyond the End-of-Option option 4609 * must be header padding (i.e., zero)." 4610 * and later: "The padding is composed of zeros." 4611 */ 4612 while (optlen % 4) { 4613 optlen += TCPOLEN_PAD; 4614 *optp++ = TCPOPT_PAD; 4615 } 4616 4617 /* compute the actual values now that we've added the options */ 4618 tlen = hlen + sizeof(struct tcphdr) + optlen; 4619 m->m_len = m->m_pkthdr.len = tlen; 4620 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4621 4622 #ifdef TCP_SIGNATURE 4623 if (sav) { 4624 (void)tcp_signature(m, th, hlen, sav, sigp); 4625 key_sa_recordxfer(sav, m); 4626 KEY_SA_UNREF(&sav); 4627 } 4628 #endif 4629 4630 /* 4631 * Send ECN SYN-ACK setup packet. 4632 * Routes can be asymetric, so, even if we receive a packet 4633 * with ECE and CWR set, we must not assume no one will block 4634 * the ECE packet we are about to send. 4635 */ 4636 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4637 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4638 th->th_flags |= TH_ECE; 4639 TCP_STATINC(TCP_STAT_ECN_SHS); 4640 4641 /* 4642 * draft-ietf-tcpm-ecnsyn-00.txt 4643 * 4644 * "[...] a TCP node MAY respond to an ECN-setup 4645 * SYN packet by setting ECT in the responding 4646 * ECN-setup SYN/ACK packet, indicating to routers 4647 * that the SYN/ACK packet is ECN-Capable. 4648 * This allows a congested router along the path 4649 * to mark the packet instead of dropping the 4650 * packet as an indication of congestion." 4651 * 4652 * "[...] There can be a great benefit in setting 4653 * an ECN-capable codepoint in SYN/ACK packets [...] 4654 * Congestion is most likely to occur in 4655 * the server-to-client direction. As a result, 4656 * setting an ECN-capable codepoint in SYN/ACK 4657 * packets can reduce the occurence of three-second 4658 * retransmit timeouts resulting from the drop 4659 * of SYN/ACK packets." 4660 * 4661 * Page 4 and 6, January 2006. 4662 */ 4663 4664 switch (sc->sc_src.sa.sa_family) { 4665 #ifdef INET 4666 case AF_INET: 4667 ip->ip_tos |= IPTOS_ECN_ECT0; 4668 break; 4669 #endif 4670 #ifdef INET6 4671 case AF_INET6: 4672 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4673 break; 4674 #endif 4675 } 4676 TCP_STATINC(TCP_STAT_ECN_ECT); 4677 } 4678 4679 4680 /* Compute the packet's checksum. */ 4681 switch (sc->sc_src.sa.sa_family) { 4682 case AF_INET: 4683 ip->ip_len = htons(tlen - hlen); 4684 th->th_sum = 0; 4685 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4686 break; 4687 #ifdef INET6 4688 case AF_INET6: 4689 ip6->ip6_plen = htons(tlen - hlen); 4690 th->th_sum = 0; 4691 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4692 break; 4693 #endif 4694 } 4695 4696 /* 4697 * Fill in some straggling IP bits. Note the stack expects 4698 * ip_len to be in host order, for convenience. 4699 */ 4700 switch (sc->sc_src.sa.sa_family) { 4701 #ifdef INET 4702 case AF_INET: 4703 ip->ip_len = htons(tlen); 4704 ip->ip_ttl = ip_defttl; 4705 /* XXX tos? */ 4706 break; 4707 #endif 4708 #ifdef INET6 4709 case AF_INET6: 4710 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4711 ip6->ip6_vfc |= IPV6_VERSION; 4712 ip6->ip6_plen = htons(tlen - hlen); 4713 /* ip6_hlim will be initialized afterwards */ 4714 /* XXX flowlabel? */ 4715 break; 4716 #endif 4717 } 4718 4719 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4720 tp = sc->sc_tp; 4721 4722 switch (sc->sc_src.sa.sa_family) { 4723 #ifdef INET 4724 case AF_INET: 4725 error = ip_output(m, sc->sc_ipopts, ro, 4726 (ip_mtudisc ? IP_MTUDISC : 0), 4727 NULL, tp ? tp->t_inpcb : NULL); 4728 break; 4729 #endif 4730 #ifdef INET6 4731 case AF_INET6: 4732 ip6->ip6_hlim = in6_selecthlim(NULL, 4733 (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp : NULL); 4734 rtcache_unref(rt, ro); 4735 4736 error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, 4737 tp ? tp->t_in6pcb : NULL, NULL); 4738 break; 4739 #endif 4740 default: 4741 error = EAFNOSUPPORT; 4742 break; 4743 } 4744 return (error); 4745 } 4746
Indexes created Sun Sep 21 20:09:37 GMT 2025