tcp_input.c revision 1.373
1 /* $NetBSD: tcp_input.c,v 1.373 2018/02/08 20:41:36 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 1.373 2018/02/08 20:41:36 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 == 0 && !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 == 0 && !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 == 0 && !vestige.valid) 1447 #else 1448 if (inp == 0 && in6p == 0 && !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 = 0; 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) 1593 goto nosave; 1594 } else { 1595 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1596 if (!tcp_saveti) 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 memset(&src, 0, sizeof(src)); 1620 memset(&dst, 0, sizeof(dst)); 1621 switch (af) { 1622 #ifdef INET 1623 case AF_INET: 1624 src.sin.sin_len = sizeof(struct sockaddr_in); 1625 src.sin.sin_family = AF_INET; 1626 src.sin.sin_addr = ip->ip_src; 1627 src.sin.sin_port = th->th_sport; 1628 1629 dst.sin.sin_len = sizeof(struct sockaddr_in); 1630 dst.sin.sin_family = AF_INET; 1631 dst.sin.sin_addr = ip->ip_dst; 1632 dst.sin.sin_port = th->th_dport; 1633 break; 1634 #endif 1635 #ifdef INET6 1636 case AF_INET6: 1637 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1638 src.sin6.sin6_family = AF_INET6; 1639 src.sin6.sin6_addr = ip6->ip6_src; 1640 src.sin6.sin6_port = th->th_sport; 1641 1642 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1643 dst.sin6.sin6_family = AF_INET6; 1644 dst.sin6.sin6_addr = ip6->ip6_dst; 1645 dst.sin6.sin6_port = th->th_dport; 1646 break; 1647 #endif /* INET6 */ 1648 } 1649 1650 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1651 if (tiflags & TH_RST) { 1652 syn_cache_reset(&src.sa, &dst.sa, th); 1653 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1654 (TH_ACK|TH_SYN)) { 1655 /* 1656 * Received a SYN,ACK. This should 1657 * never happen while we are in 1658 * LISTEN. Send an RST. 1659 */ 1660 goto badsyn; 1661 } else if (tiflags & TH_ACK) { 1662 so = syn_cache_get(&src.sa, &dst.sa, 1663 th, toff, tlen, so, m); 1664 if (so == NULL) { 1665 /* 1666 * We don't have a SYN for this ACK; 1667 * send an RST. 1668 */ 1669 goto badsyn; 1670 } else if (so == (struct socket *)(-1)) { 1671 /* 1672 * We were unable to create the 1673 * connection. If the 3-way handshake 1674 * was completed, and RST has been 1675 * sent to the peer. Since the mbuf 1676 * might be in use for the reply, do 1677 * not free it. 1678 */ 1679 m = NULL; 1680 } else { 1681 /* 1682 * We have created a full-blown 1683 * connection. 1684 */ 1685 tp = NULL; 1686 inp = NULL; 1687 #ifdef INET6 1688 in6p = NULL; 1689 #endif 1690 switch (so->so_proto->pr_domain->dom_family) { 1691 #ifdef INET 1692 case AF_INET: 1693 inp = sotoinpcb(so); 1694 tp = intotcpcb(inp); 1695 break; 1696 #endif 1697 #ifdef INET6 1698 case AF_INET6: 1699 in6p = sotoin6pcb(so); 1700 tp = in6totcpcb(in6p); 1701 break; 1702 #endif 1703 } 1704 if (tp == NULL) 1705 goto badsyn; /*XXX*/ 1706 tiwin <<= tp->snd_scale; 1707 goto after_listen; 1708 } 1709 } else { 1710 /* 1711 * None of RST, SYN or ACK was set. 1712 * This is an invalid packet for a 1713 * TCB in LISTEN state. Send a RST. 1714 */ 1715 goto badsyn; 1716 } 1717 } else { 1718 /* 1719 * Received a SYN. 1720 */ 1721 1722 #ifdef INET6 1723 /* 1724 * If deprecated address is forbidden, we do 1725 * not accept SYN to deprecated interface 1726 * address to prevent any new inbound 1727 * connection from getting established. 1728 * When we do not accept SYN, we send a TCP 1729 * RST, with deprecated source address (instead 1730 * of dropping it). We compromise it as it is 1731 * much better for peer to send a RST, and 1732 * RST will be the final packet for the 1733 * exchange. 1734 * 1735 * If we do not forbid deprecated addresses, we 1736 * accept the SYN packet. RFC2462 does not 1737 * suggest dropping SYN in this case. 1738 * If we decipher RFC2462 5.5.4, it says like 1739 * this: 1740 * 1. use of deprecated addr with existing 1741 * communication is okay - "SHOULD continue 1742 * to be used" 1743 * 2. use of it with new communication: 1744 * (2a) "SHOULD NOT be used if alternate 1745 * address with sufficient scope is 1746 * available" 1747 * (2b) nothing mentioned otherwise. 1748 * Here we fall into (2b) case as we have no 1749 * choice in our source address selection - we 1750 * must obey the peer. 1751 * 1752 * The wording in RFC2462 is confusing, and 1753 * there are multiple description text for 1754 * deprecated address handling - worse, they 1755 * are not exactly the same. I believe 5.5.4 1756 * is the best one, so we follow 5.5.4. 1757 */ 1758 if (af == AF_INET6 && !ip6_use_deprecated) { 1759 struct in6_ifaddr *ia6; 1760 int s; 1761 struct ifnet *rcvif = m_get_rcvif(m, &s); 1762 if (rcvif == NULL) 1763 goto dropwithreset; /* XXX */ 1764 if ((ia6 = in6ifa_ifpwithaddr(rcvif, 1765 &ip6->ip6_dst)) && 1766 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1767 tp = NULL; 1768 m_put_rcvif(rcvif, &s); 1769 goto dropwithreset; 1770 } 1771 m_put_rcvif(rcvif, &s); 1772 } 1773 #endif 1774 1775 #if defined(IPSEC) 1776 if (ipsec_used) { 1777 switch (af) { 1778 #ifdef INET 1779 case AF_INET: 1780 /* 1781 * inp can be NULL when receiving an 1782 * IPv4 packet on an IPv4-mapped IPv6 1783 * address. 1784 */ 1785 KASSERT(inp == NULL || 1786 sotoinpcb(so) == inp); 1787 if (!ipsec4_in_reject(m, inp)) 1788 break; 1789 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1790 tp = NULL; 1791 goto dropwithreset; 1792 #endif 1793 #ifdef INET6 1794 case AF_INET6: 1795 KASSERT(sotoin6pcb(so) == in6p); 1796 if (!ipsec6_in_reject(m, in6p)) 1797 break; 1798 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1799 tp = NULL; 1800 goto dropwithreset; 1801 #endif /*INET6*/ 1802 } 1803 } 1804 #endif /*IPSEC*/ 1805 1806 /* 1807 * LISTEN socket received a SYN from itself? This 1808 * can't possibly be valid; drop the packet. 1809 */ 1810 if (th->th_sport == th->th_dport) { 1811 int i; 1812 1813 switch (af) { 1814 #ifdef INET 1815 case AF_INET: 1816 i = in_hosteq(ip->ip_src, ip->ip_dst); 1817 break; 1818 #endif 1819 #ifdef INET6 1820 case AF_INET6: 1821 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, 1822 &ip6->ip6_dst); 1823 break; 1824 #endif 1825 default: 1826 i = 1; 1827 } 1828 if (i) { 1829 TCP_STATINC(TCP_STAT_BADSYN); 1830 goto drop; 1831 } 1832 } 1833 1834 /* 1835 * SYN looks ok; create compressed TCP 1836 * state for it. 1837 */ 1838 if (so->so_qlen <= so->so_qlimit && 1839 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1840 so, m, optp, optlen, &opti)) 1841 m = NULL; 1842 } 1843 1844 goto drop; 1845 } 1846 1847 after_listen: 1848 /* 1849 * Should not happen now that all embryonic connections 1850 * are handled with compressed state. 1851 */ 1852 KASSERT(tp->t_state != TCPS_LISTEN); 1853 1854 /* 1855 * Segment received on connection. 1856 * Reset idle time and keep-alive timer. 1857 */ 1858 tp->t_rcvtime = tcp_now; 1859 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1860 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1861 1862 /* 1863 * Process options. 1864 */ 1865 #ifdef TCP_SIGNATURE 1866 if (optp || (tp->t_flags & TF_SIGNATURE)) 1867 #else 1868 if (optp) 1869 #endif 1870 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1871 goto drop; 1872 1873 if (TCP_SACK_ENABLED(tp)) { 1874 tcp_del_sackholes(tp, th); 1875 } 1876 1877 if (TCP_ECN_ALLOWED(tp)) { 1878 if (tiflags & TH_CWR) { 1879 tp->t_flags &= ~TF_ECN_SND_ECE; 1880 } 1881 switch (iptos & IPTOS_ECN_MASK) { 1882 case IPTOS_ECN_CE: 1883 tp->t_flags |= TF_ECN_SND_ECE; 1884 TCP_STATINC(TCP_STAT_ECN_CE); 1885 break; 1886 case IPTOS_ECN_ECT0: 1887 TCP_STATINC(TCP_STAT_ECN_ECT); 1888 break; 1889 case IPTOS_ECN_ECT1: 1890 /* XXX: ignore for now -- rpaulo */ 1891 break; 1892 } 1893 /* 1894 * Congestion experienced. 1895 * Ignore if we are already trying to recover. 1896 */ 1897 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1898 tp->t_congctl->cong_exp(tp); 1899 } 1900 1901 if (opti.ts_present && opti.ts_ecr) { 1902 /* 1903 * Calculate the RTT from the returned time stamp and the 1904 * connection's time base. If the time stamp is later than 1905 * the current time, or is extremely old, fall back to non-1323 1906 * RTT calculation. Since ts_rtt is unsigned, we can test both 1907 * at the same time. 1908 * 1909 * Note that ts_rtt is in units of slow ticks (500 1910 * ms). Since most earthbound RTTs are < 500 ms, 1911 * observed values will have large quantization noise. 1912 * Our smoothed RTT is then the fraction of observed 1913 * samples that are 1 tick instead of 0 (times 500 1914 * ms). 1915 * 1916 * ts_rtt is increased by 1 to denote a valid sample, 1917 * with 0 indicating an invalid measurement. This 1918 * extra 1 must be removed when ts_rtt is used, or 1919 * else an an erroneous extra 500 ms will result. 1920 */ 1921 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1922 if (ts_rtt > TCP_PAWS_IDLE) 1923 ts_rtt = 0; 1924 } else { 1925 ts_rtt = 0; 1926 } 1927 1928 /* 1929 * Header prediction: check for the two common cases 1930 * of a uni-directional data xfer. If the packet has 1931 * no control flags, is in-sequence, the window didn't 1932 * change and we're not retransmitting, it's a 1933 * candidate. If the length is zero and the ack moved 1934 * forward, we're the sender side of the xfer. Just 1935 * free the data acked & wake any higher level process 1936 * that was blocked waiting for space. If the length 1937 * is non-zero and the ack didn't move, we're the 1938 * receiver side. If we're getting packets in-order 1939 * (the reassembly queue is empty), add the data to 1940 * the socket buffer and note that we need a delayed ack. 1941 */ 1942 if (tp->t_state == TCPS_ESTABLISHED && 1943 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1944 == TH_ACK && 1945 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1946 th->th_seq == tp->rcv_nxt && 1947 tiwin && tiwin == tp->snd_wnd && 1948 tp->snd_nxt == tp->snd_max) { 1949 1950 /* 1951 * If last ACK falls within this segment's sequence numbers, 1952 * record the timestamp. 1953 * NOTE that the test is modified according to the latest 1954 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 1955 * 1956 * note that we already know 1957 * TSTMP_GEQ(opti.ts_val, tp->ts_recent) 1958 */ 1959 if (opti.ts_present && 1960 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1961 tp->ts_recent_age = tcp_now; 1962 tp->ts_recent = opti.ts_val; 1963 } 1964 1965 if (tlen == 0) { 1966 /* Ack prediction. */ 1967 if (SEQ_GT(th->th_ack, tp->snd_una) && 1968 SEQ_LEQ(th->th_ack, tp->snd_max) && 1969 tp->snd_cwnd >= tp->snd_wnd && 1970 tp->t_partialacks < 0) { 1971 /* 1972 * this is a pure ack for outstanding data. 1973 */ 1974 if (ts_rtt) 1975 tcp_xmit_timer(tp, ts_rtt - 1); 1976 else if (tp->t_rtttime && 1977 SEQ_GT(th->th_ack, tp->t_rtseq)) 1978 tcp_xmit_timer(tp, 1979 tcp_now - tp->t_rtttime); 1980 acked = th->th_ack - tp->snd_una; 1981 tcps = TCP_STAT_GETREF(); 1982 tcps[TCP_STAT_PREDACK]++; 1983 tcps[TCP_STAT_RCVACKPACK]++; 1984 tcps[TCP_STAT_RCVACKBYTE] += acked; 1985 TCP_STAT_PUTREF(); 1986 nd6_hint(tp); 1987 1988 if (acked > (tp->t_lastoff - tp->t_inoff)) 1989 tp->t_lastm = NULL; 1990 sbdrop(&so->so_snd, acked); 1991 tp->t_lastoff -= acked; 1992 1993 icmp_check(tp, th, acked); 1994 1995 tp->snd_una = th->th_ack; 1996 tp->snd_fack = tp->snd_una; 1997 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1998 tp->snd_high = tp->snd_una; 1999 m_freem(m); 2000 2001 /* 2002 * If all outstanding data are acked, stop 2003 * retransmit timer, otherwise restart timer 2004 * using current (possibly backed-off) value. 2005 * If process is waiting for space, 2006 * wakeup/selnotify/signal. If data 2007 * are ready to send, let tcp_output 2008 * decide between more output or persist. 2009 */ 2010 if (tp->snd_una == tp->snd_max) 2011 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2012 else if (TCP_TIMER_ISARMED(tp, 2013 TCPT_PERSIST) == 0) 2014 TCP_TIMER_ARM(tp, TCPT_REXMT, 2015 tp->t_rxtcur); 2016 2017 sowwakeup(so); 2018 if (so->so_snd.sb_cc) { 2019 KERNEL_LOCK(1, NULL); 2020 (void) tcp_output(tp); 2021 KERNEL_UNLOCK_ONE(NULL); 2022 } 2023 if (tcp_saveti) 2024 m_freem(tcp_saveti); 2025 return; 2026 } 2027 } else if (th->th_ack == tp->snd_una && 2028 TAILQ_FIRST(&tp->segq) == NULL && 2029 tlen <= sbspace(&so->so_rcv)) { 2030 int newsize = 0; /* automatic sockbuf scaling */ 2031 2032 /* 2033 * this is a pure, in-sequence data packet 2034 * with nothing on the reassembly queue and 2035 * we have enough buffer space to take it. 2036 */ 2037 tp->rcv_nxt += tlen; 2038 tcps = TCP_STAT_GETREF(); 2039 tcps[TCP_STAT_PREDDAT]++; 2040 tcps[TCP_STAT_RCVPACK]++; 2041 tcps[TCP_STAT_RCVBYTE] += tlen; 2042 TCP_STAT_PUTREF(); 2043 nd6_hint(tp); 2044 2045 /* 2046 * Automatic sizing enables the performance of large buffers 2047 * and most of the efficiency of small ones by only allocating 2048 * space when it is needed. 2049 * 2050 * On the receive side the socket buffer memory is only rarely 2051 * used to any significant extent. This allows us to be much 2052 * more aggressive in scaling the receive socket buffer. For 2053 * the case that the buffer space is actually used to a large 2054 * extent and we run out of kernel memory we can simply drop 2055 * the new segments; TCP on the sender will just retransmit it 2056 * later. Setting the buffer size too big may only consume too 2057 * much kernel memory if the application doesn't read() from 2058 * the socket or packet loss or reordering makes use of the 2059 * reassembly queue. 2060 * 2061 * The criteria to step up the receive buffer one notch are: 2062 * 1. the number of bytes received during the time it takes 2063 * one timestamp to be reflected back to us (the RTT); 2064 * 2. received bytes per RTT is within seven eighth of the 2065 * current socket buffer size; 2066 * 3. receive buffer size has not hit maximal automatic size; 2067 * 2068 * This algorithm does one step per RTT at most and only if 2069 * we receive a bulk stream w/o packet losses or reorderings. 2070 * Shrinking the buffer during idle times is not necessary as 2071 * it doesn't consume any memory when idle. 2072 * 2073 * TODO: Only step up if the application is actually serving 2074 * the buffer to better manage the socket buffer resources. 2075 */ 2076 if (tcp_do_autorcvbuf && 2077 opti.ts_ecr && 2078 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 2079 if (opti.ts_ecr > tp->rfbuf_ts && 2080 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 2081 if (tp->rfbuf_cnt > 2082 (so->so_rcv.sb_hiwat / 8 * 7) && 2083 so->so_rcv.sb_hiwat < 2084 tcp_autorcvbuf_max) { 2085 newsize = 2086 min(so->so_rcv.sb_hiwat + 2087 tcp_autorcvbuf_inc, 2088 tcp_autorcvbuf_max); 2089 } 2090 /* Start over with next RTT. */ 2091 tp->rfbuf_ts = 0; 2092 tp->rfbuf_cnt = 0; 2093 } else 2094 tp->rfbuf_cnt += tlen; /* add up */ 2095 } 2096 2097 /* 2098 * Drop TCP, IP headers and TCP options then add data 2099 * to socket buffer. 2100 */ 2101 if (so->so_state & SS_CANTRCVMORE) 2102 m_freem(m); 2103 else { 2104 /* 2105 * Set new socket buffer size. 2106 * Give up when limit is reached. 2107 */ 2108 if (newsize) 2109 if (!sbreserve(&so->so_rcv, 2110 newsize, so)) 2111 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 2112 m_adj(m, toff + off); 2113 sbappendstream(&so->so_rcv, m); 2114 } 2115 sorwakeup(so); 2116 tcp_setup_ack(tp, th); 2117 if (tp->t_flags & TF_ACKNOW) { 2118 KERNEL_LOCK(1, NULL); 2119 (void) tcp_output(tp); 2120 KERNEL_UNLOCK_ONE(NULL); 2121 } 2122 if (tcp_saveti) 2123 m_freem(tcp_saveti); 2124 return; 2125 } 2126 } 2127 2128 /* 2129 * Compute mbuf offset to TCP data segment. 2130 */ 2131 hdroptlen = toff + off; 2132 2133 /* 2134 * Calculate amount of space in receive window, 2135 * and then do TCP input processing. 2136 * Receive window is amount of space in rcv queue, 2137 * but not less than advertised window. 2138 */ 2139 { int win; 2140 2141 win = sbspace(&so->so_rcv); 2142 if (win < 0) 2143 win = 0; 2144 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 2145 } 2146 2147 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 2148 tp->rfbuf_ts = 0; 2149 tp->rfbuf_cnt = 0; 2150 2151 switch (tp->t_state) { 2152 /* 2153 * If the state is SYN_SENT: 2154 * if seg contains an ACK, but not for our SYN, drop the input. 2155 * if seg contains a RST, then drop the connection. 2156 * if seg does not contain SYN, then drop it. 2157 * Otherwise this is an acceptable SYN segment 2158 * initialize tp->rcv_nxt and tp->irs 2159 * if seg contains ack then advance tp->snd_una 2160 * if seg contains a ECE and ECN support is enabled, the stream 2161 * is ECN capable. 2162 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2163 * arrange for segment to be acked (eventually) 2164 * continue processing rest of data/controls, beginning with URG 2165 */ 2166 case TCPS_SYN_SENT: 2167 if ((tiflags & TH_ACK) && 2168 (SEQ_LEQ(th->th_ack, tp->iss) || 2169 SEQ_GT(th->th_ack, tp->snd_max))) 2170 goto dropwithreset; 2171 if (tiflags & TH_RST) { 2172 if (tiflags & TH_ACK) 2173 tp = tcp_drop(tp, ECONNREFUSED); 2174 goto drop; 2175 } 2176 if ((tiflags & TH_SYN) == 0) 2177 goto drop; 2178 if (tiflags & TH_ACK) { 2179 tp->snd_una = th->th_ack; 2180 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2181 tp->snd_nxt = tp->snd_una; 2182 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2183 tp->snd_high = tp->snd_una; 2184 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2185 2186 if ((tiflags & TH_ECE) && tcp_do_ecn) { 2187 tp->t_flags |= TF_ECN_PERMIT; 2188 TCP_STATINC(TCP_STAT_ECN_SHS); 2189 } 2190 2191 } 2192 tp->irs = th->th_seq; 2193 tcp_rcvseqinit(tp); 2194 tp->t_flags |= TF_ACKNOW; 2195 tcp_mss_from_peer(tp, opti.maxseg); 2196 2197 /* 2198 * Initialize the initial congestion window. If we 2199 * had to retransmit the SYN, we must initialize cwnd 2200 * to 1 segment (i.e. the Loss Window). 2201 */ 2202 if (tp->t_flags & TF_SYN_REXMT) 2203 tp->snd_cwnd = tp->t_peermss; 2204 else { 2205 int ss = tcp_init_win; 2206 #ifdef INET 2207 if (inp != NULL && in_localaddr(inp->inp_faddr)) 2208 ss = tcp_init_win_local; 2209 #endif 2210 #ifdef INET6 2211 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 2212 ss = tcp_init_win_local; 2213 #endif 2214 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 2215 } 2216 2217 tcp_rmx_rtt(tp); 2218 if (tiflags & TH_ACK) { 2219 TCP_STATINC(TCP_STAT_CONNECTS); 2220 /* 2221 * move tcp_established before soisconnected 2222 * because upcall handler can drive tcp_output 2223 * functionality. 2224 * XXX we might call soisconnected at the end of 2225 * all processing 2226 */ 2227 tcp_established(tp); 2228 soisconnected(so); 2229 /* Do window scaling on this connection? */ 2230 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2231 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2232 tp->snd_scale = tp->requested_s_scale; 2233 tp->rcv_scale = tp->request_r_scale; 2234 } 2235 TCP_REASS_LOCK(tp); 2236 (void) tcp_reass(tp, NULL, NULL, &tlen); 2237 /* 2238 * if we didn't have to retransmit the SYN, 2239 * use its rtt as our initial srtt & rtt var. 2240 */ 2241 if (tp->t_rtttime) 2242 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2243 } else 2244 tp->t_state = TCPS_SYN_RECEIVED; 2245 2246 /* 2247 * Advance th->th_seq to correspond to first data byte. 2248 * If data, trim to stay within window, 2249 * dropping FIN if necessary. 2250 */ 2251 th->th_seq++; 2252 if (tlen > tp->rcv_wnd) { 2253 todrop = tlen - tp->rcv_wnd; 2254 m_adj(m, -todrop); 2255 tlen = tp->rcv_wnd; 2256 tiflags &= ~TH_FIN; 2257 tcps = TCP_STAT_GETREF(); 2258 tcps[TCP_STAT_RCVPACKAFTERWIN]++; 2259 tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop; 2260 TCP_STAT_PUTREF(); 2261 } 2262 tp->snd_wl1 = th->th_seq - 1; 2263 tp->rcv_up = th->th_seq; 2264 goto step6; 2265 2266 /* 2267 * If the state is SYN_RECEIVED: 2268 * If seg contains an ACK, but not for our SYN, drop the input 2269 * and generate an RST. See page 36, rfc793 2270 */ 2271 case TCPS_SYN_RECEIVED: 2272 if ((tiflags & TH_ACK) && 2273 (SEQ_LEQ(th->th_ack, tp->iss) || 2274 SEQ_GT(th->th_ack, tp->snd_max))) 2275 goto dropwithreset; 2276 break; 2277 } 2278 2279 /* 2280 * States other than LISTEN or SYN_SENT. 2281 * First check timestamp, if present. 2282 * Then check that at least some bytes of segment are within 2283 * receive window. If segment begins before rcv_nxt, 2284 * drop leading data (and SYN); if nothing left, just ack. 2285 * 2286 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2287 * and it's less than ts_recent, drop it. 2288 */ 2289 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2290 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2291 2292 /* Check to see if ts_recent is over 24 days old. */ 2293 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2294 /* 2295 * Invalidate ts_recent. If this segment updates 2296 * ts_recent, the age will be reset later and ts_recent 2297 * will get a valid value. If it does not, setting 2298 * ts_recent to zero will at least satisfy the 2299 * requirement that zero be placed in the timestamp 2300 * echo reply when ts_recent isn't valid. The 2301 * age isn't reset until we get a valid ts_recent 2302 * because we don't want out-of-order segments to be 2303 * dropped when ts_recent is old. 2304 */ 2305 tp->ts_recent = 0; 2306 } else { 2307 tcps = TCP_STAT_GETREF(); 2308 tcps[TCP_STAT_RCVDUPPACK]++; 2309 tcps[TCP_STAT_RCVDUPBYTE] += tlen; 2310 tcps[TCP_STAT_PAWSDROP]++; 2311 TCP_STAT_PUTREF(); 2312 tcp_new_dsack(tp, th->th_seq, tlen); 2313 goto dropafterack; 2314 } 2315 } 2316 2317 todrop = tp->rcv_nxt - th->th_seq; 2318 dupseg = false; 2319 if (todrop > 0) { 2320 if (tiflags & TH_SYN) { 2321 tiflags &= ~TH_SYN; 2322 th->th_seq++; 2323 if (th->th_urp > 1) 2324 th->th_urp--; 2325 else { 2326 tiflags &= ~TH_URG; 2327 th->th_urp = 0; 2328 } 2329 todrop--; 2330 } 2331 if (todrop > tlen || 2332 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2333 /* 2334 * Any valid FIN or RST must be to the left of the 2335 * window. At this point the FIN or RST must be a 2336 * duplicate or out of sequence; drop it. 2337 */ 2338 if (tiflags & TH_RST) 2339 goto drop; 2340 tiflags &= ~(TH_FIN|TH_RST); 2341 /* 2342 * Send an ACK to resynchronize and drop any data. 2343 * But keep on processing for RST or ACK. 2344 */ 2345 tp->t_flags |= TF_ACKNOW; 2346 todrop = tlen; 2347 dupseg = true; 2348 tcps = TCP_STAT_GETREF(); 2349 tcps[TCP_STAT_RCVDUPPACK]++; 2350 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 2351 TCP_STAT_PUTREF(); 2352 } else if ((tiflags & TH_RST) && 2353 th->th_seq != tp->rcv_nxt) { 2354 /* 2355 * Test for reset before adjusting the sequence 2356 * number for overlapping data. 2357 */ 2358 goto dropafterack_ratelim; 2359 } else { 2360 tcps = TCP_STAT_GETREF(); 2361 tcps[TCP_STAT_RCVPARTDUPPACK]++; 2362 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 2363 TCP_STAT_PUTREF(); 2364 } 2365 tcp_new_dsack(tp, th->th_seq, todrop); 2366 hdroptlen += todrop; /*drop from head afterwards*/ 2367 th->th_seq += todrop; 2368 tlen -= todrop; 2369 if (th->th_urp > todrop) 2370 th->th_urp -= todrop; 2371 else { 2372 tiflags &= ~TH_URG; 2373 th->th_urp = 0; 2374 } 2375 } 2376 2377 /* 2378 * If new data are received on a connection after the 2379 * user processes are gone, then RST the other end. 2380 */ 2381 if ((so->so_state & SS_NOFDREF) && 2382 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2383 tp = tcp_close(tp); 2384 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 2385 goto dropwithreset; 2386 } 2387 2388 /* 2389 * If segment ends after window, drop trailing data 2390 * (and PUSH and FIN); if nothing left, just ACK. 2391 */ 2392 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 2393 if (todrop > 0) { 2394 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 2395 if (todrop >= tlen) { 2396 /* 2397 * The segment actually starts after the window. 2398 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2399 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2400 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2401 */ 2402 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 2403 /* 2404 * If a new connection request is received 2405 * while in TIME_WAIT, drop the old connection 2406 * and start over if the sequence numbers 2407 * are above the previous ones. 2408 * 2409 * NOTE: We will checksum the packet again, and 2410 * so we need to put the header fields back into 2411 * network order! 2412 * XXX This kind of sucks, but we don't expect 2413 * XXX this to happen very often, so maybe it 2414 * XXX doesn't matter so much. 2415 */ 2416 if (tiflags & TH_SYN && 2417 tp->t_state == TCPS_TIME_WAIT && 2418 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2419 tp = tcp_close(tp); 2420 tcp_fields_to_net(th); 2421 goto findpcb; 2422 } 2423 /* 2424 * If window is closed can only take segments at 2425 * window edge, and have to drop data and PUSH from 2426 * incoming segments. Continue processing, but 2427 * remember to ack. Otherwise, drop segment 2428 * and (if not RST) ack. 2429 */ 2430 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2431 tp->t_flags |= TF_ACKNOW; 2432 TCP_STATINC(TCP_STAT_RCVWINPROBE); 2433 } else 2434 goto dropafterack; 2435 } else 2436 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 2437 m_adj(m, -todrop); 2438 tlen -= todrop; 2439 tiflags &= ~(TH_PUSH|TH_FIN); 2440 } 2441 2442 /* 2443 * If last ACK falls within this segment's sequence numbers, 2444 * record the timestamp. 2445 * NOTE: 2446 * 1) That the test incorporates suggestions from the latest 2447 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 2448 * 2) That updating only on newer timestamps interferes with 2449 * our earlier PAWS tests, so this check should be solely 2450 * predicated on the sequence space of this segment. 2451 * 3) That we modify the segment boundary check to be 2452 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2453 * instead of RFC1323's 2454 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2455 * This modified check allows us to overcome RFC1323's 2456 * limitations as described in Stevens TCP/IP Illustrated 2457 * Vol. 2 p.869. In such cases, we can still calculate the 2458 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2459 */ 2460 if (opti.ts_present && 2461 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2462 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2463 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2464 tp->ts_recent_age = tcp_now; 2465 tp->ts_recent = opti.ts_val; 2466 } 2467 2468 /* 2469 * If the RST bit is set examine the state: 2470 * SYN_RECEIVED STATE: 2471 * If passive open, return to LISTEN state. 2472 * If active open, inform user that connection was refused. 2473 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 2474 * Inform user that connection was reset, and close tcb. 2475 * CLOSING, LAST_ACK, TIME_WAIT STATES 2476 * Close the tcb. 2477 */ 2478 if (tiflags & TH_RST) { 2479 if (th->th_seq != tp->rcv_nxt) 2480 goto dropafterack_ratelim; 2481 2482 switch (tp->t_state) { 2483 case TCPS_SYN_RECEIVED: 2484 so->so_error = ECONNREFUSED; 2485 goto close; 2486 2487 case TCPS_ESTABLISHED: 2488 case TCPS_FIN_WAIT_1: 2489 case TCPS_FIN_WAIT_2: 2490 case TCPS_CLOSE_WAIT: 2491 so->so_error = ECONNRESET; 2492 close: 2493 tp->t_state = TCPS_CLOSED; 2494 TCP_STATINC(TCP_STAT_DROPS); 2495 tp = tcp_close(tp); 2496 goto drop; 2497 2498 case TCPS_CLOSING: 2499 case TCPS_LAST_ACK: 2500 case TCPS_TIME_WAIT: 2501 tp = tcp_close(tp); 2502 goto drop; 2503 } 2504 } 2505 2506 /* 2507 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2508 * we must be in a synchronized state. RFC791 states (under RST 2509 * generation) that any unacceptable segment (an out-of-order SYN 2510 * qualifies) received in a synchronized state must elicit only an 2511 * empty acknowledgment segment ... and the connection remains in 2512 * the same state. 2513 */ 2514 if (tiflags & TH_SYN) { 2515 if (tp->rcv_nxt == th->th_seq) { 2516 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2517 TH_ACK); 2518 if (tcp_saveti) 2519 m_freem(tcp_saveti); 2520 return; 2521 } 2522 2523 goto dropafterack_ratelim; 2524 } 2525 2526 /* 2527 * If the ACK bit is off we drop the segment and return. 2528 */ 2529 if ((tiflags & TH_ACK) == 0) { 2530 if (tp->t_flags & TF_ACKNOW) 2531 goto dropafterack; 2532 else 2533 goto drop; 2534 } 2535 2536 /* 2537 * Ack processing. 2538 */ 2539 switch (tp->t_state) { 2540 2541 /* 2542 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2543 * ESTABLISHED state and continue processing, otherwise 2544 * send an RST. 2545 */ 2546 case TCPS_SYN_RECEIVED: 2547 if (SEQ_GT(tp->snd_una, th->th_ack) || 2548 SEQ_GT(th->th_ack, tp->snd_max)) 2549 goto dropwithreset; 2550 TCP_STATINC(TCP_STAT_CONNECTS); 2551 soisconnected(so); 2552 tcp_established(tp); 2553 /* Do window scaling? */ 2554 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2555 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2556 tp->snd_scale = tp->requested_s_scale; 2557 tp->rcv_scale = tp->request_r_scale; 2558 } 2559 TCP_REASS_LOCK(tp); 2560 (void) tcp_reass(tp, NULL, NULL, &tlen); 2561 tp->snd_wl1 = th->th_seq - 1; 2562 /* fall into ... */ 2563 2564 /* 2565 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2566 * ACKs. If the ack is in the range 2567 * tp->snd_una < th->th_ack <= tp->snd_max 2568 * then advance tp->snd_una to th->th_ack and drop 2569 * data from the retransmission queue. If this ACK reflects 2570 * more up to date window information we update our window information. 2571 */ 2572 case TCPS_ESTABLISHED: 2573 case TCPS_FIN_WAIT_1: 2574 case TCPS_FIN_WAIT_2: 2575 case TCPS_CLOSE_WAIT: 2576 case TCPS_CLOSING: 2577 case TCPS_LAST_ACK: 2578 case TCPS_TIME_WAIT: 2579 2580 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2581 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2582 TCP_STATINC(TCP_STAT_RCVDUPACK); 2583 /* 2584 * If we have outstanding data (other than 2585 * a window probe), this is a completely 2586 * duplicate ack (ie, window info didn't 2587 * change), the ack is the biggest we've 2588 * seen and we've seen exactly our rexmt 2589 * threshhold of them, assume a packet 2590 * has been dropped and retransmit it. 2591 * Kludge snd_nxt & the congestion 2592 * window so we send only this one 2593 * packet. 2594 */ 2595 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2596 th->th_ack != tp->snd_una) 2597 tp->t_dupacks = 0; 2598 else if (tp->t_partialacks < 0 && 2599 (++tp->t_dupacks == tcprexmtthresh || 2600 TCP_FACK_FASTRECOV(tp))) { 2601 /* 2602 * Do the fast retransmit, and adjust 2603 * congestion control paramenters. 2604 */ 2605 if (tp->t_congctl->fast_retransmit(tp, th)) { 2606 /* False fast retransmit */ 2607 break; 2608 } else 2609 goto drop; 2610 } else if (tp->t_dupacks > tcprexmtthresh) { 2611 tp->snd_cwnd += tp->t_segsz; 2612 KERNEL_LOCK(1, NULL); 2613 (void) tcp_output(tp); 2614 KERNEL_UNLOCK_ONE(NULL); 2615 goto drop; 2616 } 2617 } else { 2618 /* 2619 * If the ack appears to be very old, only 2620 * allow data that is in-sequence. This 2621 * makes it somewhat more difficult to insert 2622 * forged data by guessing sequence numbers. 2623 * Sent an ack to try to update the send 2624 * sequence number on the other side. 2625 */ 2626 if (tlen && th->th_seq != tp->rcv_nxt && 2627 SEQ_LT(th->th_ack, 2628 tp->snd_una - tp->max_sndwnd)) 2629 goto dropafterack; 2630 } 2631 break; 2632 } 2633 /* 2634 * If the congestion window was inflated to account 2635 * for the other side's cached packets, retract it. 2636 */ 2637 tp->t_congctl->fast_retransmit_newack(tp, th); 2638 2639 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2640 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2641 goto dropafterack; 2642 } 2643 acked = th->th_ack - tp->snd_una; 2644 tcps = TCP_STAT_GETREF(); 2645 tcps[TCP_STAT_RCVACKPACK]++; 2646 tcps[TCP_STAT_RCVACKBYTE] += acked; 2647 TCP_STAT_PUTREF(); 2648 2649 /* 2650 * If we have a timestamp reply, update smoothed 2651 * round trip time. If no timestamp is present but 2652 * transmit timer is running and timed sequence 2653 * number was acked, update smoothed round trip time. 2654 * Since we now have an rtt measurement, cancel the 2655 * timer backoff (cf., Phil Karn's retransmit alg.). 2656 * Recompute the initial retransmit timer. 2657 */ 2658 if (ts_rtt) 2659 tcp_xmit_timer(tp, ts_rtt - 1); 2660 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2661 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2662 2663 /* 2664 * If all outstanding data is acked, stop retransmit 2665 * timer and remember to restart (more output or persist). 2666 * If there is more data to be acked, restart retransmit 2667 * timer, using current (possibly backed-off) value. 2668 */ 2669 if (th->th_ack == tp->snd_max) { 2670 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2671 needoutput = 1; 2672 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2673 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2674 2675 /* 2676 * New data has been acked, adjust the congestion window. 2677 */ 2678 tp->t_congctl->newack(tp, th); 2679 2680 nd6_hint(tp); 2681 if (acked > so->so_snd.sb_cc) { 2682 tp->snd_wnd -= so->so_snd.sb_cc; 2683 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2684 ourfinisacked = 1; 2685 } else { 2686 if (acked > (tp->t_lastoff - tp->t_inoff)) 2687 tp->t_lastm = NULL; 2688 sbdrop(&so->so_snd, acked); 2689 tp->t_lastoff -= acked; 2690 if (tp->snd_wnd > acked) 2691 tp->snd_wnd -= acked; 2692 else 2693 tp->snd_wnd = 0; 2694 ourfinisacked = 0; 2695 } 2696 sowwakeup(so); 2697 2698 icmp_check(tp, th, acked); 2699 2700 tp->snd_una = th->th_ack; 2701 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2702 tp->snd_fack = tp->snd_una; 2703 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2704 tp->snd_nxt = tp->snd_una; 2705 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2706 tp->snd_high = tp->snd_una; 2707 2708 switch (tp->t_state) { 2709 2710 /* 2711 * In FIN_WAIT_1 STATE in addition to the processing 2712 * for the ESTABLISHED state if our FIN is now acknowledged 2713 * then enter FIN_WAIT_2. 2714 */ 2715 case TCPS_FIN_WAIT_1: 2716 if (ourfinisacked) { 2717 /* 2718 * If we can't receive any more 2719 * data, then closing user can proceed. 2720 * Starting the timer is contrary to the 2721 * specification, but if we don't get a FIN 2722 * we'll hang forever. 2723 */ 2724 if (so->so_state & SS_CANTRCVMORE) { 2725 soisdisconnected(so); 2726 if (tp->t_maxidle > 0) 2727 TCP_TIMER_ARM(tp, TCPT_2MSL, 2728 tp->t_maxidle); 2729 } 2730 tp->t_state = TCPS_FIN_WAIT_2; 2731 } 2732 break; 2733 2734 /* 2735 * In CLOSING STATE in addition to the processing for 2736 * the ESTABLISHED state if the ACK acknowledges our FIN 2737 * then enter the TIME-WAIT state, otherwise ignore 2738 * the segment. 2739 */ 2740 case TCPS_CLOSING: 2741 if (ourfinisacked) { 2742 tp->t_state = TCPS_TIME_WAIT; 2743 tcp_canceltimers(tp); 2744 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2745 soisdisconnected(so); 2746 } 2747 break; 2748 2749 /* 2750 * In LAST_ACK, we may still be waiting for data to drain 2751 * and/or to be acked, as well as for the ack of our FIN. 2752 * If our FIN is now acknowledged, delete the TCB, 2753 * enter the closed state and return. 2754 */ 2755 case TCPS_LAST_ACK: 2756 if (ourfinisacked) { 2757 tp = tcp_close(tp); 2758 goto drop; 2759 } 2760 break; 2761 2762 /* 2763 * In TIME_WAIT state the only thing that should arrive 2764 * is a retransmission of the remote FIN. Acknowledge 2765 * it and restart the finack timer. 2766 */ 2767 case TCPS_TIME_WAIT: 2768 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2769 goto dropafterack; 2770 } 2771 } 2772 2773 step6: 2774 /* 2775 * Update window information. 2776 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2777 */ 2778 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2779 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2780 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2781 /* keep track of pure window updates */ 2782 if (tlen == 0 && 2783 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2784 TCP_STATINC(TCP_STAT_RCVWINUPD); 2785 tp->snd_wnd = tiwin; 2786 tp->snd_wl1 = th->th_seq; 2787 tp->snd_wl2 = th->th_ack; 2788 if (tp->snd_wnd > tp->max_sndwnd) 2789 tp->max_sndwnd = tp->snd_wnd; 2790 needoutput = 1; 2791 } 2792 2793 /* 2794 * Process segments with URG. 2795 */ 2796 if ((tiflags & TH_URG) && th->th_urp && 2797 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2798 /* 2799 * This is a kludge, but if we receive and accept 2800 * random urgent pointers, we'll crash in 2801 * soreceive. It's hard to imagine someone 2802 * actually wanting to send this much urgent data. 2803 */ 2804 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2805 th->th_urp = 0; /* XXX */ 2806 tiflags &= ~TH_URG; /* XXX */ 2807 goto dodata; /* XXX */ 2808 } 2809 /* 2810 * If this segment advances the known urgent pointer, 2811 * then mark the data stream. This should not happen 2812 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2813 * a FIN has been received from the remote side. 2814 * In these states we ignore the URG. 2815 * 2816 * According to RFC961 (Assigned Protocols), 2817 * the urgent pointer points to the last octet 2818 * of urgent data. We continue, however, 2819 * to consider it to indicate the first octet 2820 * of data past the urgent section as the original 2821 * spec states (in one of two places). 2822 */ 2823 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2824 tp->rcv_up = th->th_seq + th->th_urp; 2825 so->so_oobmark = so->so_rcv.sb_cc + 2826 (tp->rcv_up - tp->rcv_nxt) - 1; 2827 if (so->so_oobmark == 0) 2828 so->so_state |= SS_RCVATMARK; 2829 sohasoutofband(so); 2830 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2831 } 2832 /* 2833 * Remove out of band data so doesn't get presented to user. 2834 * This can happen independent of advancing the URG pointer, 2835 * but if two URG's are pending at once, some out-of-band 2836 * data may creep in... ick. 2837 */ 2838 if (th->th_urp <= (u_int16_t) tlen 2839 #ifdef SO_OOBINLINE 2840 && (so->so_options & SO_OOBINLINE) == 0 2841 #endif 2842 ) 2843 tcp_pulloutofband(so, th, m, hdroptlen); 2844 } else 2845 /* 2846 * If no out of band data is expected, 2847 * pull receive urgent pointer along 2848 * with the receive window. 2849 */ 2850 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2851 tp->rcv_up = tp->rcv_nxt; 2852 dodata: /* XXX */ 2853 2854 /* 2855 * Process the segment text, merging it into the TCP sequencing queue, 2856 * and arranging for acknowledgement of receipt if necessary. 2857 * This process logically involves adjusting tp->rcv_wnd as data 2858 * is presented to the user (this happens in tcp_usrreq.c, 2859 * tcp_rcvd()). If a FIN has already been received on this 2860 * connection then we just ignore the text. 2861 */ 2862 if ((tlen || (tiflags & TH_FIN)) && 2863 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2864 /* 2865 * Insert segment ti into reassembly queue of tcp with 2866 * control block tp. Return TH_FIN if reassembly now includes 2867 * a segment with FIN. The macro form does the common case 2868 * inline (segment is the next to be received on an 2869 * established connection, and the queue is empty), 2870 * avoiding linkage into and removal from the queue and 2871 * repetition of various conversions. 2872 * Set DELACK for segments received in order, but ack 2873 * immediately when segments are out of order 2874 * (so fast retransmit can work). 2875 */ 2876 /* NOTE: this was TCP_REASS() macro, but used only once */ 2877 TCP_REASS_LOCK(tp); 2878 if (th->th_seq == tp->rcv_nxt && 2879 TAILQ_FIRST(&tp->segq) == NULL && 2880 tp->t_state == TCPS_ESTABLISHED) { 2881 tcp_setup_ack(tp, th); 2882 tp->rcv_nxt += tlen; 2883 tiflags = th->th_flags & TH_FIN; 2884 tcps = TCP_STAT_GETREF(); 2885 tcps[TCP_STAT_RCVPACK]++; 2886 tcps[TCP_STAT_RCVBYTE] += tlen; 2887 TCP_STAT_PUTREF(); 2888 nd6_hint(tp); 2889 if (so->so_state & SS_CANTRCVMORE) 2890 m_freem(m); 2891 else { 2892 m_adj(m, hdroptlen); 2893 sbappendstream(&(so)->so_rcv, m); 2894 } 2895 TCP_REASS_UNLOCK(tp); 2896 sorwakeup(so); 2897 } else { 2898 m_adj(m, hdroptlen); 2899 tiflags = tcp_reass(tp, th, m, &tlen); 2900 tp->t_flags |= TF_ACKNOW; 2901 } 2902 2903 /* 2904 * Note the amount of data that peer has sent into 2905 * our window, in order to estimate the sender's 2906 * buffer size. 2907 */ 2908 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2909 } else { 2910 m_freem(m); 2911 m = NULL; 2912 tiflags &= ~TH_FIN; 2913 } 2914 2915 /* 2916 * If FIN is received ACK the FIN and let the user know 2917 * that the connection is closing. Ignore a FIN received before 2918 * the connection is fully established. 2919 */ 2920 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2921 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2922 socantrcvmore(so); 2923 tp->t_flags |= TF_ACKNOW; 2924 tp->rcv_nxt++; 2925 } 2926 switch (tp->t_state) { 2927 2928 /* 2929 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2930 */ 2931 case TCPS_ESTABLISHED: 2932 tp->t_state = TCPS_CLOSE_WAIT; 2933 break; 2934 2935 /* 2936 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2937 * enter the CLOSING state. 2938 */ 2939 case TCPS_FIN_WAIT_1: 2940 tp->t_state = TCPS_CLOSING; 2941 break; 2942 2943 /* 2944 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2945 * starting the time-wait timer, turning off the other 2946 * standard timers. 2947 */ 2948 case TCPS_FIN_WAIT_2: 2949 tp->t_state = TCPS_TIME_WAIT; 2950 tcp_canceltimers(tp); 2951 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2952 soisdisconnected(so); 2953 break; 2954 2955 /* 2956 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2957 */ 2958 case TCPS_TIME_WAIT: 2959 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2960 break; 2961 } 2962 } 2963 #ifdef TCP_DEBUG 2964 if (so->so_options & SO_DEBUG) 2965 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2966 #endif 2967 2968 /* 2969 * Return any desired output. 2970 */ 2971 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2972 KERNEL_LOCK(1, NULL); 2973 (void) tcp_output(tp); 2974 KERNEL_UNLOCK_ONE(NULL); 2975 } 2976 if (tcp_saveti) 2977 m_freem(tcp_saveti); 2978 2979 if (tp->t_state == TCPS_TIME_WAIT 2980 && (so->so_state & SS_NOFDREF) 2981 && (tp->t_inpcb || af != AF_INET) 2982 && (tp->t_in6pcb || af != AF_INET6) 2983 && ((af == AF_INET ? tcp4_vtw_enable : tcp6_vtw_enable) & 1) != 0 2984 && TAILQ_EMPTY(&tp->segq) 2985 && vtw_add(af, tp)) { 2986 ; 2987 } 2988 return; 2989 2990 badsyn: 2991 /* 2992 * Received a bad SYN. Increment counters and dropwithreset. 2993 */ 2994 TCP_STATINC(TCP_STAT_BADSYN); 2995 tp = NULL; 2996 goto dropwithreset; 2997 2998 dropafterack: 2999 /* 3000 * Generate an ACK dropping incoming segment if it occupies 3001 * sequence space, where the ACK reflects our state. 3002 */ 3003 if (tiflags & TH_RST) 3004 goto drop; 3005 goto dropafterack2; 3006 3007 dropafterack_ratelim: 3008 /* 3009 * We may want to rate-limit ACKs against SYN/RST attack. 3010 */ 3011 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 3012 tcp_ackdrop_ppslim) == 0) { 3013 /* XXX stat */ 3014 goto drop; 3015 } 3016 /* ...fall into dropafterack2... */ 3017 3018 dropafterack2: 3019 m_freem(m); 3020 tp->t_flags |= TF_ACKNOW; 3021 KERNEL_LOCK(1, NULL); 3022 (void) tcp_output(tp); 3023 KERNEL_UNLOCK_ONE(NULL); 3024 if (tcp_saveti) 3025 m_freem(tcp_saveti); 3026 return; 3027 3028 dropwithreset_ratelim: 3029 /* 3030 * We may want to rate-limit RSTs in certain situations, 3031 * particularly if we are sending an RST in response to 3032 * an attempt to connect to or otherwise communicate with 3033 * a port for which we have no socket. 3034 */ 3035 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 3036 tcp_rst_ppslim) == 0) { 3037 /* XXX stat */ 3038 goto drop; 3039 } 3040 /* ...fall into dropwithreset... */ 3041 3042 dropwithreset: 3043 /* 3044 * Generate a RST, dropping incoming segment. 3045 * Make ACK acceptable to originator of segment. 3046 */ 3047 if (tiflags & TH_RST) 3048 goto drop; 3049 3050 switch (af) { 3051 #ifdef INET6 3052 case AF_INET6: 3053 /* For following calls to tcp_respond */ 3054 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 3055 goto drop; 3056 break; 3057 #endif /* INET6 */ 3058 case AF_INET: 3059 if (IN_MULTICAST(ip->ip_dst.s_addr) || 3060 in_broadcast(ip->ip_dst, m_get_rcvif_NOMPSAFE(m))) 3061 goto drop; 3062 } 3063 3064 if (tiflags & TH_ACK) 3065 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3066 else { 3067 if (tiflags & TH_SYN) 3068 tlen++; 3069 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 3070 TH_RST|TH_ACK); 3071 } 3072 if (tcp_saveti) 3073 m_freem(tcp_saveti); 3074 return; 3075 3076 badcsum: 3077 drop: 3078 /* 3079 * Drop space held by incoming segment and return. 3080 */ 3081 if (tp) { 3082 if (tp->t_inpcb) 3083 so = tp->t_inpcb->inp_socket; 3084 #ifdef INET6 3085 else if (tp->t_in6pcb) 3086 so = tp->t_in6pcb->in6p_socket; 3087 #endif 3088 else 3089 so = NULL; 3090 #ifdef TCP_DEBUG 3091 if (so && (so->so_options & SO_DEBUG) != 0) 3092 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 3093 #endif 3094 } 3095 if (tcp_saveti) 3096 m_freem(tcp_saveti); 3097 m_freem(m); 3098 return; 3099 } 3100 3101 #ifdef TCP_SIGNATURE 3102 int 3103 tcp_signature_apply(void *fstate, void *data, u_int len) 3104 { 3105 3106 MD5Update(fstate, (u_char *)data, len); 3107 return (0); 3108 } 3109 3110 struct secasvar * 3111 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 3112 { 3113 struct ip *ip; 3114 struct ip6_hdr *ip6; 3115 3116 ip = mtod(m, struct ip *); 3117 switch (ip->ip_v) { 3118 case 4: 3119 ip = mtod(m, struct ip *); 3120 ip6 = NULL; 3121 break; 3122 case 6: 3123 ip = NULL; 3124 ip6 = mtod(m, struct ip6_hdr *); 3125 break; 3126 default: 3127 return (NULL); 3128 } 3129 3130 #ifdef IPSEC 3131 union sockaddr_union dst; 3132 3133 /* Extract the destination from the IP header in the mbuf. */ 3134 memset(&dst, 0, sizeof(union sockaddr_union)); 3135 if (ip != NULL) { 3136 dst.sa.sa_len = sizeof(struct sockaddr_in); 3137 dst.sa.sa_family = AF_INET; 3138 dst.sin.sin_addr = ip->ip_dst; 3139 } else { 3140 dst.sa.sa_len = sizeof(struct sockaddr_in6); 3141 dst.sa.sa_family = AF_INET6; 3142 dst.sin6.sin6_addr = ip6->ip6_dst; 3143 } 3144 3145 /* 3146 * Look up an SADB entry which matches the address of the peer. 3147 */ 3148 return KEY_LOOKUP_SA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI), 0, 0); 3149 #else 3150 return NULL; 3151 #endif 3152 } 3153 3154 int 3155 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 3156 struct secasvar *sav, char *sig) 3157 { 3158 MD5_CTX ctx; 3159 struct ip *ip; 3160 struct ipovly *ipovly; 3161 #ifdef INET6 3162 struct ip6_hdr *ip6; 3163 struct ip6_hdr_pseudo ip6pseudo; 3164 #endif /* INET6 */ 3165 struct ippseudo ippseudo; 3166 struct tcphdr th0; 3167 int l, tcphdrlen; 3168 3169 if (sav == NULL) 3170 return (-1); 3171 3172 tcphdrlen = th->th_off * 4; 3173 3174 switch (mtod(m, struct ip *)->ip_v) { 3175 case 4: 3176 MD5Init(&ctx); 3177 ip = mtod(m, struct ip *); 3178 memset(&ippseudo, 0, sizeof(ippseudo)); 3179 ipovly = (struct ipovly *)ip; 3180 ippseudo.ippseudo_src = ipovly->ih_src; 3181 ippseudo.ippseudo_dst = ipovly->ih_dst; 3182 ippseudo.ippseudo_pad = 0; 3183 ippseudo.ippseudo_p = IPPROTO_TCP; 3184 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 3185 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 3186 break; 3187 #if INET6 3188 case 6: 3189 MD5Init(&ctx); 3190 ip6 = mtod(m, struct ip6_hdr *); 3191 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 3192 ip6pseudo.ip6ph_src = ip6->ip6_src; 3193 in6_clearscope(&ip6pseudo.ip6ph_src); 3194 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 3195 in6_clearscope(&ip6pseudo.ip6ph_dst); 3196 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 3197 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 3198 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 3199 break; 3200 #endif /* INET6 */ 3201 default: 3202 return (-1); 3203 } 3204 3205 th0 = *th; 3206 th0.th_sum = 0; 3207 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 3208 3209 l = m->m_pkthdr.len - thoff - tcphdrlen; 3210 if (l > 0) 3211 m_apply(m, thoff + tcphdrlen, 3212 m->m_pkthdr.len - thoff - tcphdrlen, 3213 tcp_signature_apply, &ctx); 3214 3215 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 3216 MD5Final(sig, &ctx); 3217 3218 return (0); 3219 } 3220 #endif 3221 3222 /* 3223 * tcp_dooptions: parse and process tcp options. 3224 * 3225 * returns -1 if this segment should be dropped. (eg. wrong signature) 3226 * otherwise returns 0. 3227 */ 3228 3229 static int 3230 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, 3231 struct tcphdr *th, 3232 struct mbuf *m, int toff, struct tcp_opt_info *oi) 3233 { 3234 u_int16_t mss; 3235 int opt, optlen = 0; 3236 #ifdef TCP_SIGNATURE 3237 void *sigp = NULL; 3238 char sigbuf[TCP_SIGLEN]; 3239 struct secasvar *sav = NULL; 3240 #endif 3241 3242 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 3243 opt = cp[0]; 3244 if (opt == TCPOPT_EOL) 3245 break; 3246 if (opt == TCPOPT_NOP) 3247 optlen = 1; 3248 else { 3249 if (cnt < 2) 3250 break; 3251 optlen = cp[1]; 3252 if (optlen < 2 || optlen > cnt) 3253 break; 3254 } 3255 switch (opt) { 3256 3257 default: 3258 continue; 3259 3260 case TCPOPT_MAXSEG: 3261 if (optlen != TCPOLEN_MAXSEG) 3262 continue; 3263 if (!(th->th_flags & TH_SYN)) 3264 continue; 3265 if (TCPS_HAVERCVDSYN(tp->t_state)) 3266 continue; 3267 bcopy(cp + 2, &mss, sizeof(mss)); 3268 oi->maxseg = ntohs(mss); 3269 break; 3270 3271 case TCPOPT_WINDOW: 3272 if (optlen != TCPOLEN_WINDOW) 3273 continue; 3274 if (!(th->th_flags & TH_SYN)) 3275 continue; 3276 if (TCPS_HAVERCVDSYN(tp->t_state)) 3277 continue; 3278 tp->t_flags |= TF_RCVD_SCALE; 3279 tp->requested_s_scale = cp[2]; 3280 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3281 char buf[INET6_ADDRSTRLEN]; 3282 struct ip *ip = mtod(m, struct ip *); 3283 #ifdef INET6 3284 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 3285 #endif 3286 if (ip) 3287 in_print(buf, sizeof(buf), 3288 &ip->ip_src); 3289 #ifdef INET6 3290 else if (ip6) 3291 in6_print(buf, sizeof(buf), 3292 &ip6->ip6_src); 3293 #endif 3294 else 3295 strlcpy(buf, "(unknown)", sizeof(buf)); 3296 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3297 "assuming %d\n", 3298 tp->requested_s_scale, buf, 3299 TCP_MAX_WINSHIFT); 3300 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3301 } 3302 break; 3303 3304 case TCPOPT_TIMESTAMP: 3305 if (optlen != TCPOLEN_TIMESTAMP) 3306 continue; 3307 oi->ts_present = 1; 3308 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 3309 NTOHL(oi->ts_val); 3310 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 3311 NTOHL(oi->ts_ecr); 3312 3313 if (!(th->th_flags & TH_SYN)) 3314 continue; 3315 if (TCPS_HAVERCVDSYN(tp->t_state)) 3316 continue; 3317 /* 3318 * A timestamp received in a SYN makes 3319 * it ok to send timestamp requests and replies. 3320 */ 3321 tp->t_flags |= TF_RCVD_TSTMP; 3322 tp->ts_recent = oi->ts_val; 3323 tp->ts_recent_age = tcp_now; 3324 break; 3325 3326 case TCPOPT_SACK_PERMITTED: 3327 if (optlen != TCPOLEN_SACK_PERMITTED) 3328 continue; 3329 if (!(th->th_flags & TH_SYN)) 3330 continue; 3331 if (TCPS_HAVERCVDSYN(tp->t_state)) 3332 continue; 3333 if (tcp_do_sack) { 3334 tp->t_flags |= TF_SACK_PERMIT; 3335 tp->t_flags |= TF_WILL_SACK; 3336 } 3337 break; 3338 3339 case TCPOPT_SACK: 3340 tcp_sack_option(tp, th, cp, optlen); 3341 break; 3342 #ifdef TCP_SIGNATURE 3343 case TCPOPT_SIGNATURE: 3344 if (optlen != TCPOLEN_SIGNATURE) 3345 continue; 3346 if (sigp && memcmp(sigp, cp + 2, TCP_SIGLEN)) 3347 return (-1); 3348 3349 sigp = sigbuf; 3350 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3351 tp->t_flags |= TF_SIGNATURE; 3352 break; 3353 #endif 3354 } 3355 } 3356 3357 #ifndef TCP_SIGNATURE 3358 return 0; 3359 #else 3360 if (tp->t_flags & TF_SIGNATURE) { 3361 3362 sav = tcp_signature_getsav(m, th); 3363 3364 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3365 return (-1); 3366 } 3367 3368 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) 3369 goto out; 3370 3371 if (sigp) { 3372 char sig[TCP_SIGLEN]; 3373 3374 tcp_fields_to_net(th); 3375 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3376 tcp_fields_to_host(th); 3377 goto out; 3378 } 3379 tcp_fields_to_host(th); 3380 3381 if (memcmp(sig, sigp, TCP_SIGLEN)) { 3382 TCP_STATINC(TCP_STAT_BADSIG); 3383 goto out; 3384 } else 3385 TCP_STATINC(TCP_STAT_GOODSIG); 3386 3387 key_sa_recordxfer(sav, m); 3388 KEY_SA_UNREF(&sav); 3389 } 3390 return 0; 3391 out: 3392 if (sav != NULL) 3393 KEY_SA_UNREF(&sav); 3394 return -1; 3395 #endif 3396 } 3397 3398 /* 3399 * Pull out of band byte out of a segment so 3400 * it doesn't appear in the user's data queue. 3401 * It is still reflected in the segment length for 3402 * sequencing purposes. 3403 */ 3404 void 3405 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3406 struct mbuf *m, int off) 3407 { 3408 int cnt = off + th->th_urp - 1; 3409 3410 while (cnt >= 0) { 3411 if (m->m_len > cnt) { 3412 char *cp = mtod(m, char *) + cnt; 3413 struct tcpcb *tp = sototcpcb(so); 3414 3415 tp->t_iobc = *cp; 3416 tp->t_oobflags |= TCPOOB_HAVEDATA; 3417 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3418 m->m_len--; 3419 return; 3420 } 3421 cnt -= m->m_len; 3422 m = m->m_next; 3423 if (m == 0) 3424 break; 3425 } 3426 panic("tcp_pulloutofband"); 3427 } 3428 3429 /* 3430 * Collect new round-trip time estimate 3431 * and update averages and current timeout. 3432 * 3433 * rtt is in units of slow ticks (typically 500 ms) -- essentially the 3434 * difference of two timestamps. 3435 */ 3436 void 3437 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3438 { 3439 int32_t delta; 3440 3441 TCP_STATINC(TCP_STAT_RTTUPDATED); 3442 if (tp->t_srtt != 0) { 3443 /* 3444 * Compute the amount to add to srtt for smoothing, 3445 * *alpha, or 2^(-TCP_RTT_SHIFT). Because 3446 * srtt is stored in 1/32 slow ticks, we conceptually 3447 * shift left 5 bits, subtract srtt to get the 3448 * diference, and then shift right by TCP_RTT_SHIFT 3449 * (3) to obtain 1/8 of the difference. 3450 */ 3451 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3452 /* 3453 * This can never happen, because delta's lowest 3454 * possible value is 1/8 of t_srtt. But if it does, 3455 * set srtt to some reasonable value, here chosen 3456 * as 1/8 tick. 3457 */ 3458 if ((tp->t_srtt += delta) <= 0) 3459 tp->t_srtt = 1 << 2; 3460 /* 3461 * RFC2988 requires that rttvar be updated first. 3462 * This code is compliant because "delta" is the old 3463 * srtt minus the new observation (scaled). 3464 * 3465 * RFC2988 says: 3466 * rttvar = (1-beta) * rttvar + beta * |srtt-observed| 3467 * 3468 * delta is in units of 1/32 ticks, and has then been 3469 * divided by 8. This is equivalent to being in 1/16s 3470 * units and divided by 4. Subtract from it 1/4 of 3471 * the existing rttvar to form the (signed) amount to 3472 * adjust. 3473 */ 3474 if (delta < 0) 3475 delta = -delta; 3476 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3477 /* 3478 * As with srtt, this should never happen. There is 3479 * no support in RFC2988 for this operation. But 1/4s 3480 * as rttvar when faced with something arguably wrong 3481 * is ok. 3482 */ 3483 if ((tp->t_rttvar += delta) <= 0) 3484 tp->t_rttvar = 1 << 2; 3485 3486 /* 3487 * If srtt exceeds .01 second, ensure we use the 'remote' MSL 3488 * Problem is: it doesn't work. Disabled by defaulting 3489 * tcp_rttlocal to 0; see corresponding code in 3490 * tcp_subr that selects local vs remote in a different way. 3491 * 3492 * The static branch prediction hint here should be removed 3493 * when the rtt estimator is fixed and the rtt_enable code 3494 * is turned back on. 3495 */ 3496 if (__predict_false(tcp_rttlocal) && tcp_msl_enable 3497 && tp->t_srtt > tcp_msl_remote_threshold 3498 && tp->t_msl < tcp_msl_remote) { 3499 tp->t_msl = tcp_msl_remote; 3500 } 3501 } else { 3502 /* 3503 * This is the first measurement. Per RFC2988, 2.2, 3504 * set rtt=R and srtt=R/2. 3505 * For srtt, storage representation is 1/32 ticks, 3506 * so shift left by 5. 3507 * For rttvar, storage representation is 1/16 ticks, 3508 * So shift left by 4, but then right by 1 to halve. 3509 */ 3510 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3511 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3512 } 3513 tp->t_rtttime = 0; 3514 tp->t_rxtshift = 0; 3515 3516 /* 3517 * the retransmit should happen at rtt + 4 * rttvar. 3518 * Because of the way we do the smoothing, srtt and rttvar 3519 * will each average +1/2 tick of bias. When we compute 3520 * the retransmit timer, we want 1/2 tick of rounding and 3521 * 1 extra tick because of +-1/2 tick uncertainty in the 3522 * firing of the timer. The bias will give us exactly the 3523 * 1.5 tick we need. But, because the bias is 3524 * statistical, we have to test that we don't drop below 3525 * the minimum feasible timer (which is 2 ticks). 3526 */ 3527 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3528 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3529 3530 /* 3531 * We received an ack for a packet that wasn't retransmitted; 3532 * it is probably safe to discard any error indications we've 3533 * received recently. This isn't quite right, but close enough 3534 * for now (a route might have failed after we sent a segment, 3535 * and the return path might not be symmetrical). 3536 */ 3537 tp->t_softerror = 0; 3538 } 3539 3540 3541 /* 3542 * TCP compressed state engine. Currently used to hold compressed 3543 * state for SYN_RECEIVED. 3544 */ 3545 3546 u_long syn_cache_count; 3547 u_int32_t syn_hash1, syn_hash2; 3548 3549 #define SYN_HASH(sa, sp, dp) \ 3550 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3551 ((u_int32_t)(sp)))^syn_hash2))) 3552 #ifndef INET6 3553 #define SYN_HASHALL(hash, src, dst) \ 3554 do { \ 3555 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3556 ((const struct sockaddr_in *)(src))->sin_port, \ 3557 ((const struct sockaddr_in *)(dst))->sin_port); \ 3558 } while (/*CONSTCOND*/ 0) 3559 #else 3560 #define SYN_HASH6(sa, sp, dp) \ 3561 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3562 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3563 & 0x7fffffff) 3564 3565 #define SYN_HASHALL(hash, src, dst) \ 3566 do { \ 3567 switch ((src)->sa_family) { \ 3568 case AF_INET: \ 3569 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3570 ((const struct sockaddr_in *)(src))->sin_port, \ 3571 ((const struct sockaddr_in *)(dst))->sin_port); \ 3572 break; \ 3573 case AF_INET6: \ 3574 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3575 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3576 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3577 break; \ 3578 default: \ 3579 hash = 0; \ 3580 } \ 3581 } while (/*CONSTCOND*/0) 3582 #endif /* INET6 */ 3583 3584 static struct pool syn_cache_pool; 3585 3586 /* 3587 * We don't estimate RTT with SYNs, so each packet starts with the default 3588 * RTT and each timer step has a fixed timeout value. 3589 */ 3590 static inline void 3591 syn_cache_timer_arm(struct syn_cache *sc) 3592 { 3593 3594 TCPT_RANGESET(sc->sc_rxtcur, 3595 TCPTV_SRTTDFLT * tcp_backoff[sc->sc_rxtshift], TCPTV_MIN, 3596 TCPTV_REXMTMAX); 3597 callout_reset(&sc->sc_timer, 3598 sc->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, sc); 3599 } 3600 3601 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3602 3603 static inline void 3604 syn_cache_rm(struct syn_cache *sc) 3605 { 3606 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3607 sc, sc_bucketq); 3608 sc->sc_tp = NULL; 3609 LIST_REMOVE(sc, sc_tpq); 3610 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3611 callout_stop(&sc->sc_timer); 3612 syn_cache_count--; 3613 } 3614 3615 static inline void 3616 syn_cache_put(struct syn_cache *sc) 3617 { 3618 if (sc->sc_ipopts) 3619 (void) m_free(sc->sc_ipopts); 3620 rtcache_free(&sc->sc_route); 3621 sc->sc_flags |= SCF_DEAD; 3622 if (!callout_invoking(&sc->sc_timer)) 3623 callout_schedule(&(sc)->sc_timer, 1); 3624 } 3625 3626 void 3627 syn_cache_init(void) 3628 { 3629 int i; 3630 3631 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 3632 "synpl", NULL, IPL_SOFTNET); 3633 3634 /* Initialize the hash buckets. */ 3635 for (i = 0; i < tcp_syn_cache_size; i++) 3636 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3637 } 3638 3639 void 3640 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3641 { 3642 struct syn_cache_head *scp; 3643 struct syn_cache *sc2; 3644 int s; 3645 3646 /* 3647 * If there are no entries in the hash table, reinitialize 3648 * the hash secrets. 3649 */ 3650 if (syn_cache_count == 0) { 3651 syn_hash1 = cprng_fast32(); 3652 syn_hash2 = cprng_fast32(); 3653 } 3654 3655 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3656 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3657 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3658 3659 /* 3660 * Make sure that we don't overflow the per-bucket 3661 * limit or the total cache size limit. 3662 */ 3663 s = splsoftnet(); 3664 if (scp->sch_length >= tcp_syn_bucket_limit) { 3665 TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW); 3666 /* 3667 * The bucket is full. Toss the oldest element in the 3668 * bucket. This will be the first entry in the bucket. 3669 */ 3670 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3671 #ifdef DIAGNOSTIC 3672 /* 3673 * This should never happen; we should always find an 3674 * entry in our bucket. 3675 */ 3676 if (sc2 == NULL) 3677 panic("syn_cache_insert: bucketoverflow: impossible"); 3678 #endif 3679 syn_cache_rm(sc2); 3680 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3681 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3682 struct syn_cache_head *scp2, *sce; 3683 3684 TCP_STATINC(TCP_STAT_SC_OVERFLOWED); 3685 /* 3686 * The cache is full. Toss the oldest entry in the 3687 * first non-empty bucket we can find. 3688 * 3689 * XXX We would really like to toss the oldest 3690 * entry in the cache, but we hope that this 3691 * condition doesn't happen very often. 3692 */ 3693 scp2 = scp; 3694 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3695 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3696 for (++scp2; scp2 != scp; scp2++) { 3697 if (scp2 >= sce) 3698 scp2 = &tcp_syn_cache[0]; 3699 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3700 break; 3701 } 3702 #ifdef DIAGNOSTIC 3703 /* 3704 * This should never happen; we should always find a 3705 * non-empty bucket. 3706 */ 3707 if (scp2 == scp) 3708 panic("syn_cache_insert: cacheoverflow: " 3709 "impossible"); 3710 #endif 3711 } 3712 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3713 syn_cache_rm(sc2); 3714 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3715 } 3716 3717 /* 3718 * Initialize the entry's timer. 3719 */ 3720 sc->sc_rxttot = 0; 3721 sc->sc_rxtshift = 0; 3722 syn_cache_timer_arm(sc); 3723 3724 /* Link it from tcpcb entry */ 3725 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3726 3727 /* Put it into the bucket. */ 3728 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3729 scp->sch_length++; 3730 syn_cache_count++; 3731 3732 TCP_STATINC(TCP_STAT_SC_ADDED); 3733 splx(s); 3734 } 3735 3736 /* 3737 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3738 * If we have retransmitted an entry the maximum number of times, expire 3739 * that entry. 3740 */ 3741 static void 3742 syn_cache_timer(void *arg) 3743 { 3744 struct syn_cache *sc = arg; 3745 3746 mutex_enter(softnet_lock); 3747 KERNEL_LOCK(1, NULL); 3748 3749 callout_ack(&sc->sc_timer); 3750 3751 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3752 TCP_STATINC(TCP_STAT_SC_DELAYED_FREE); 3753 goto free; 3754 } 3755 3756 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3757 /* Drop it -- too many retransmissions. */ 3758 goto dropit; 3759 } 3760 3761 /* 3762 * Compute the total amount of time this entry has 3763 * been on a queue. If this entry has been on longer 3764 * than the keep alive timer would allow, expire it. 3765 */ 3766 sc->sc_rxttot += sc->sc_rxtcur; 3767 if (sc->sc_rxttot >= tcp_keepinit) 3768 goto dropit; 3769 3770 TCP_STATINC(TCP_STAT_SC_RETRANSMITTED); 3771 (void) syn_cache_respond(sc, NULL); 3772 3773 /* Advance the timer back-off. */ 3774 sc->sc_rxtshift++; 3775 syn_cache_timer_arm(sc); 3776 3777 goto out; 3778 3779 dropit: 3780 TCP_STATINC(TCP_STAT_SC_TIMED_OUT); 3781 syn_cache_rm(sc); 3782 if (sc->sc_ipopts) 3783 (void) m_free(sc->sc_ipopts); 3784 rtcache_free(&sc->sc_route); 3785 3786 free: 3787 callout_destroy(&sc->sc_timer); 3788 pool_put(&syn_cache_pool, sc); 3789 3790 out: 3791 KERNEL_UNLOCK_ONE(NULL); 3792 mutex_exit(softnet_lock); 3793 } 3794 3795 /* 3796 * Remove syn cache created by the specified tcb entry, 3797 * because this does not make sense to keep them 3798 * (if there's no tcb entry, syn cache entry will never be used) 3799 */ 3800 void 3801 syn_cache_cleanup(struct tcpcb *tp) 3802 { 3803 struct syn_cache *sc, *nsc; 3804 int s; 3805 3806 s = splsoftnet(); 3807 3808 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3809 nsc = LIST_NEXT(sc, sc_tpq); 3810 3811 #ifdef DIAGNOSTIC 3812 if (sc->sc_tp != tp) 3813 panic("invalid sc_tp in syn_cache_cleanup"); 3814 #endif 3815 syn_cache_rm(sc); 3816 syn_cache_put(sc); /* calls pool_put but see spl above */ 3817 } 3818 /* just for safety */ 3819 LIST_INIT(&tp->t_sc); 3820 3821 splx(s); 3822 } 3823 3824 /* 3825 * Find an entry in the syn cache. 3826 */ 3827 struct syn_cache * 3828 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3829 struct syn_cache_head **headp) 3830 { 3831 struct syn_cache *sc; 3832 struct syn_cache_head *scp; 3833 u_int32_t hash; 3834 int s; 3835 3836 SYN_HASHALL(hash, src, dst); 3837 3838 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3839 *headp = scp; 3840 s = splsoftnet(); 3841 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3842 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3843 if (sc->sc_hash != hash) 3844 continue; 3845 if (!memcmp(&sc->sc_src, src, src->sa_len) && 3846 !memcmp(&sc->sc_dst, dst, dst->sa_len)) { 3847 splx(s); 3848 return (sc); 3849 } 3850 } 3851 splx(s); 3852 return (NULL); 3853 } 3854 3855 /* 3856 * This function gets called when we receive an ACK for a 3857 * socket in the LISTEN state. We look up the connection 3858 * in the syn cache, and if its there, we pull it out of 3859 * the cache and turn it into a full-blown connection in 3860 * the SYN-RECEIVED state. 3861 * 3862 * The return values may not be immediately obvious, and their effects 3863 * can be subtle, so here they are: 3864 * 3865 * NULL SYN was not found in cache; caller should drop the 3866 * packet and send an RST. 3867 * 3868 * -1 We were unable to create the new connection, and are 3869 * aborting it. An ACK,RST is being sent to the peer 3870 * (unless we got screwey sequence numbners; see below), 3871 * because the 3-way handshake has been completed. Caller 3872 * should not free the mbuf, since we may be using it. If 3873 * we are not, we will free it. 3874 * 3875 * Otherwise, the return value is a pointer to the new socket 3876 * associated with the connection. 3877 */ 3878 struct socket * 3879 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3880 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3881 struct socket *so, struct mbuf *m) 3882 { 3883 struct syn_cache *sc; 3884 struct syn_cache_head *scp; 3885 struct inpcb *inp = NULL; 3886 #ifdef INET6 3887 struct in6pcb *in6p = NULL; 3888 #endif 3889 struct tcpcb *tp = 0; 3890 int s; 3891 struct socket *oso; 3892 3893 s = splsoftnet(); 3894 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3895 splx(s); 3896 return (NULL); 3897 } 3898 3899 /* 3900 * Verify the sequence and ack numbers. Try getting the correct 3901 * response again. 3902 */ 3903 if ((th->th_ack != sc->sc_iss + 1) || 3904 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3905 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3906 (void) syn_cache_respond(sc, m); 3907 splx(s); 3908 return ((struct socket *)(-1)); 3909 } 3910 3911 /* Remove this cache entry */ 3912 syn_cache_rm(sc); 3913 splx(s); 3914 3915 /* 3916 * Ok, create the full blown connection, and set things up 3917 * as they would have been set up if we had created the 3918 * connection when the SYN arrived. If we can't create 3919 * the connection, abort it. 3920 */ 3921 /* 3922 * inp still has the OLD in_pcb stuff, set the 3923 * v6-related flags on the new guy, too. This is 3924 * done particularly for the case where an AF_INET6 3925 * socket is bound only to a port, and a v4 connection 3926 * comes in on that port. 3927 * we also copy the flowinfo from the original pcb 3928 * to the new one. 3929 */ 3930 oso = so; 3931 so = sonewconn(so, true); 3932 if (so == NULL) 3933 goto resetandabort; 3934 3935 switch (so->so_proto->pr_domain->dom_family) { 3936 #ifdef INET 3937 case AF_INET: 3938 inp = sotoinpcb(so); 3939 break; 3940 #endif 3941 #ifdef INET6 3942 case AF_INET6: 3943 in6p = sotoin6pcb(so); 3944 break; 3945 #endif 3946 } 3947 switch (src->sa_family) { 3948 #ifdef INET 3949 case AF_INET: 3950 if (inp) { 3951 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3952 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3953 inp->inp_options = ip_srcroute(m); 3954 in_pcbstate(inp, INP_BOUND); 3955 if (inp->inp_options == NULL) { 3956 inp->inp_options = sc->sc_ipopts; 3957 sc->sc_ipopts = NULL; 3958 } 3959 } 3960 #ifdef INET6 3961 else if (in6p) { 3962 /* IPv4 packet to AF_INET6 socket */ 3963 memset(&in6p->in6p_laddr, 0, sizeof(in6p->in6p_laddr)); 3964 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3965 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3966 &in6p->in6p_laddr.s6_addr32[3], 3967 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3968 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3969 in6totcpcb(in6p)->t_family = AF_INET; 3970 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3971 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3972 else 3973 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3974 in6_pcbstate(in6p, IN6P_BOUND); 3975 } 3976 #endif 3977 break; 3978 #endif 3979 #ifdef INET6 3980 case AF_INET6: 3981 if (in6p) { 3982 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3983 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3984 in6_pcbstate(in6p, IN6P_BOUND); 3985 } 3986 break; 3987 #endif 3988 } 3989 #ifdef INET6 3990 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3991 struct in6pcb *oin6p = sotoin6pcb(oso); 3992 /* inherit socket options from the listening socket */ 3993 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3994 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3995 m_freem(in6p->in6p_options); 3996 in6p->in6p_options = 0; 3997 } 3998 ip6_savecontrol(in6p, &in6p->in6p_options, 3999 mtod(m, struct ip6_hdr *), m); 4000 } 4001 #endif 4002 4003 #if defined(IPSEC) 4004 if (ipsec_used) { 4005 /* 4006 * we make a copy of policy, instead of sharing the policy, for 4007 * better behavior in terms of SA lookup and dead SA removal. 4008 */ 4009 if (inp) { 4010 /* copy old policy into new socket's */ 4011 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, 4012 inp->inp_sp)) 4013 printf("tcp_input: could not copy policy\n"); 4014 } 4015 #ifdef INET6 4016 else if (in6p) { 4017 /* copy old policy into new socket's */ 4018 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 4019 in6p->in6p_sp)) 4020 printf("tcp_input: could not copy policy\n"); 4021 } 4022 #endif 4023 } 4024 #endif 4025 4026 /* 4027 * Give the new socket our cached route reference. 4028 */ 4029 if (inp) { 4030 rtcache_copy(&inp->inp_route, &sc->sc_route); 4031 rtcache_free(&sc->sc_route); 4032 } 4033 #ifdef INET6 4034 else { 4035 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 4036 rtcache_free(&sc->sc_route); 4037 } 4038 #endif 4039 4040 if (inp) { 4041 struct sockaddr_in sin; 4042 memcpy(&sin, src, src->sa_len); 4043 if (in_pcbconnect(inp, &sin, &lwp0)) { 4044 goto resetandabort; 4045 } 4046 } 4047 #ifdef INET6 4048 else if (in6p) { 4049 struct sockaddr_in6 sin6; 4050 memcpy(&sin6, src, src->sa_len); 4051 if (src->sa_family == AF_INET) { 4052 /* IPv4 packet to AF_INET6 socket */ 4053 in6_sin_2_v4mapsin6((struct sockaddr_in *)src, &sin6); 4054 } 4055 if (in6_pcbconnect(in6p, &sin6, NULL)) { 4056 goto resetandabort; 4057 } 4058 } 4059 #endif 4060 else { 4061 goto resetandabort; 4062 } 4063 4064 if (inp) 4065 tp = intotcpcb(inp); 4066 #ifdef INET6 4067 else if (in6p) 4068 tp = in6totcpcb(in6p); 4069 #endif 4070 else 4071 tp = NULL; 4072 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 4073 if (sc->sc_request_r_scale != 15) { 4074 tp->requested_s_scale = sc->sc_requested_s_scale; 4075 tp->request_r_scale = sc->sc_request_r_scale; 4076 tp->snd_scale = sc->sc_requested_s_scale; 4077 tp->rcv_scale = sc->sc_request_r_scale; 4078 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 4079 } 4080 if (sc->sc_flags & SCF_TIMESTAMP) 4081 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 4082 tp->ts_timebase = sc->sc_timebase; 4083 4084 tp->t_template = tcp_template(tp); 4085 if (tp->t_template == 0) { 4086 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 4087 so = NULL; 4088 m_freem(m); 4089 goto abort; 4090 } 4091 4092 tp->iss = sc->sc_iss; 4093 tp->irs = sc->sc_irs; 4094 tcp_sendseqinit(tp); 4095 tcp_rcvseqinit(tp); 4096 tp->t_state = TCPS_SYN_RECEIVED; 4097 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 4098 TCP_STATINC(TCP_STAT_ACCEPTS); 4099 4100 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 4101 tp->t_flags |= TF_WILL_SACK; 4102 4103 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 4104 tp->t_flags |= TF_ECN_PERMIT; 4105 4106 #ifdef TCP_SIGNATURE 4107 if (sc->sc_flags & SCF_SIGNATURE) 4108 tp->t_flags |= TF_SIGNATURE; 4109 #endif 4110 4111 /* Initialize tp->t_ourmss before we deal with the peer's! */ 4112 tp->t_ourmss = sc->sc_ourmaxseg; 4113 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 4114 4115 /* 4116 * Initialize the initial congestion window. If we 4117 * had to retransmit the SYN,ACK, we must initialize cwnd 4118 * to 1 segment (i.e. the Loss Window). 4119 */ 4120 if (sc->sc_rxtshift) 4121 tp->snd_cwnd = tp->t_peermss; 4122 else { 4123 int ss = tcp_init_win; 4124 #ifdef INET 4125 if (inp != NULL && in_localaddr(inp->inp_faddr)) 4126 ss = tcp_init_win_local; 4127 #endif 4128 #ifdef INET6 4129 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 4130 ss = tcp_init_win_local; 4131 #endif 4132 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 4133 } 4134 4135 tcp_rmx_rtt(tp); 4136 tp->snd_wl1 = sc->sc_irs; 4137 tp->rcv_up = sc->sc_irs + 1; 4138 4139 /* 4140 * This is what whould have happened in tcp_output() when 4141 * the SYN,ACK was sent. 4142 */ 4143 tp->snd_up = tp->snd_una; 4144 tp->snd_max = tp->snd_nxt = tp->iss+1; 4145 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 4146 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 4147 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 4148 tp->last_ack_sent = tp->rcv_nxt; 4149 tp->t_partialacks = -1; 4150 tp->t_dupacks = 0; 4151 4152 TCP_STATINC(TCP_STAT_SC_COMPLETED); 4153 s = splsoftnet(); 4154 syn_cache_put(sc); 4155 splx(s); 4156 return (so); 4157 4158 resetandabort: 4159 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 4160 abort: 4161 if (so != NULL) { 4162 (void) soqremque(so, 1); 4163 (void) soabort(so); 4164 mutex_enter(softnet_lock); 4165 } 4166 s = splsoftnet(); 4167 syn_cache_put(sc); 4168 splx(s); 4169 TCP_STATINC(TCP_STAT_SC_ABORTED); 4170 return ((struct socket *)(-1)); 4171 } 4172 4173 /* 4174 * This function is called when we get a RST for a 4175 * non-existent connection, so that we can see if the 4176 * connection is in the syn cache. If it is, zap it. 4177 */ 4178 4179 void 4180 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 4181 { 4182 struct syn_cache *sc; 4183 struct syn_cache_head *scp; 4184 int s = splsoftnet(); 4185 4186 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4187 splx(s); 4188 return; 4189 } 4190 if (SEQ_LT(th->th_seq, sc->sc_irs) || 4191 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 4192 splx(s); 4193 return; 4194 } 4195 syn_cache_rm(sc); 4196 TCP_STATINC(TCP_STAT_SC_RESET); 4197 syn_cache_put(sc); /* calls pool_put but see spl above */ 4198 splx(s); 4199 } 4200 4201 void 4202 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 4203 struct tcphdr *th) 4204 { 4205 struct syn_cache *sc; 4206 struct syn_cache_head *scp; 4207 int s; 4208 4209 s = splsoftnet(); 4210 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4211 splx(s); 4212 return; 4213 } 4214 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 4215 if (ntohl (th->th_seq) != sc->sc_iss) { 4216 splx(s); 4217 return; 4218 } 4219 4220 /* 4221 * If we've retransmitted 3 times and this is our second error, 4222 * we remove the entry. Otherwise, we allow it to continue on. 4223 * This prevents us from incorrectly nuking an entry during a 4224 * spurious network outage. 4225 * 4226 * See tcp_notify(). 4227 */ 4228 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 4229 sc->sc_flags |= SCF_UNREACH; 4230 splx(s); 4231 return; 4232 } 4233 4234 syn_cache_rm(sc); 4235 TCP_STATINC(TCP_STAT_SC_UNREACH); 4236 syn_cache_put(sc); /* calls pool_put but see spl above */ 4237 splx(s); 4238 } 4239 4240 /* 4241 * Given a LISTEN socket and an inbound SYN request, add 4242 * this to the syn cache, and send back a segment: 4243 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 4244 * to the source. 4245 * 4246 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 4247 * Doing so would require that we hold onto the data and deliver it 4248 * to the application. However, if we are the target of a SYN-flood 4249 * DoS attack, an attacker could send data which would eventually 4250 * consume all available buffer space if it were ACKed. By not ACKing 4251 * the data, we avoid this DoS scenario. 4252 */ 4253 4254 int 4255 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 4256 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 4257 int optlen, struct tcp_opt_info *oi) 4258 { 4259 struct tcpcb tb, *tp; 4260 long win; 4261 struct syn_cache *sc; 4262 struct syn_cache_head *scp; 4263 struct mbuf *ipopts; 4264 struct tcp_opt_info opti; 4265 int s; 4266 4267 tp = sototcpcb(so); 4268 4269 memset(&opti, 0, sizeof(opti)); 4270 4271 /* 4272 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 4273 * 4274 * Note this check is performed in tcp_input() very early on. 4275 */ 4276 4277 /* 4278 * Initialize some local state. 4279 */ 4280 win = sbspace(&so->so_rcv); 4281 if (win > TCP_MAXWIN) 4282 win = TCP_MAXWIN; 4283 4284 switch (src->sa_family) { 4285 #ifdef INET 4286 case AF_INET: 4287 /* 4288 * Remember the IP options, if any. 4289 */ 4290 ipopts = ip_srcroute(m); 4291 break; 4292 #endif 4293 default: 4294 ipopts = NULL; 4295 } 4296 4297 #ifdef TCP_SIGNATURE 4298 if (optp || (tp->t_flags & TF_SIGNATURE)) 4299 #else 4300 if (optp) 4301 #endif 4302 { 4303 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4304 #ifdef TCP_SIGNATURE 4305 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4306 #endif 4307 tb.t_state = TCPS_LISTEN; 4308 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 4309 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 4310 return (0); 4311 } else 4312 tb.t_flags = 0; 4313 4314 /* 4315 * See if we already have an entry for this connection. 4316 * If we do, resend the SYN,ACK. We do not count this 4317 * as a retransmission (XXX though maybe we should). 4318 */ 4319 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4320 TCP_STATINC(TCP_STAT_SC_DUPESYN); 4321 if (ipopts) { 4322 /* 4323 * If we were remembering a previous source route, 4324 * forget it and use the new one we've been given. 4325 */ 4326 if (sc->sc_ipopts) 4327 (void) m_free(sc->sc_ipopts); 4328 sc->sc_ipopts = ipopts; 4329 } 4330 sc->sc_timestamp = tb.ts_recent; 4331 if (syn_cache_respond(sc, m) == 0) { 4332 uint64_t *tcps = TCP_STAT_GETREF(); 4333 tcps[TCP_STAT_SNDACKS]++; 4334 tcps[TCP_STAT_SNDTOTAL]++; 4335 TCP_STAT_PUTREF(); 4336 } 4337 return (1); 4338 } 4339 4340 s = splsoftnet(); 4341 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4342 splx(s); 4343 if (sc == NULL) { 4344 if (ipopts) 4345 (void) m_free(ipopts); 4346 return (0); 4347 } 4348 4349 /* 4350 * Fill in the cache, and put the necessary IP and TCP 4351 * options into the reply. 4352 */ 4353 memset(sc, 0, sizeof(struct syn_cache)); 4354 callout_init(&sc->sc_timer, CALLOUT_MPSAFE); 4355 bcopy(src, &sc->sc_src, src->sa_len); 4356 bcopy(dst, &sc->sc_dst, dst->sa_len); 4357 sc->sc_flags = 0; 4358 sc->sc_ipopts = ipopts; 4359 sc->sc_irs = th->th_seq; 4360 switch (src->sa_family) { 4361 #ifdef INET 4362 case AF_INET: 4363 { 4364 struct sockaddr_in *srcin = (void *) src; 4365 struct sockaddr_in *dstin = (void *) dst; 4366 4367 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4368 &srcin->sin_addr, dstin->sin_port, 4369 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4370 break; 4371 } 4372 #endif /* INET */ 4373 #ifdef INET6 4374 case AF_INET6: 4375 { 4376 struct sockaddr_in6 *srcin6 = (void *) src; 4377 struct sockaddr_in6 *dstin6 = (void *) dst; 4378 4379 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4380 &srcin6->sin6_addr, dstin6->sin6_port, 4381 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4382 break; 4383 } 4384 #endif /* INET6 */ 4385 } 4386 sc->sc_peermaxseg = oi->maxseg; 4387 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4388 m_get_rcvif_NOMPSAFE(m) : NULL, 4389 sc->sc_src.sa.sa_family); 4390 sc->sc_win = win; 4391 sc->sc_timebase = tcp_now - 1; /* see tcp_newtcpcb() */ 4392 sc->sc_timestamp = tb.ts_recent; 4393 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4394 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4395 sc->sc_flags |= SCF_TIMESTAMP; 4396 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4397 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4398 sc->sc_requested_s_scale = tb.requested_s_scale; 4399 sc->sc_request_r_scale = 0; 4400 /* 4401 * Pick the smallest possible scaling factor that 4402 * will still allow us to scale up to sb_max. 4403 * 4404 * We do this because there are broken firewalls that 4405 * will corrupt the window scale option, leading to 4406 * the other endpoint believing that our advertised 4407 * window is unscaled. At scale factors larger than 4408 * 5 the unscaled window will drop below 1500 bytes, 4409 * leading to serious problems when traversing these 4410 * broken firewalls. 4411 * 4412 * With the default sbmax of 256K, a scale factor 4413 * of 3 will be chosen by this algorithm. Those who 4414 * choose a larger sbmax should watch out 4415 * for the compatiblity problems mentioned above. 4416 * 4417 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4418 * or <SYN,ACK>) segment itself is never scaled. 4419 */ 4420 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4421 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4422 sc->sc_request_r_scale++; 4423 } else { 4424 sc->sc_requested_s_scale = 15; 4425 sc->sc_request_r_scale = 15; 4426 } 4427 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4428 sc->sc_flags |= SCF_SACK_PERMIT; 4429 4430 /* 4431 * ECN setup packet received. 4432 */ 4433 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4434 sc->sc_flags |= SCF_ECN_PERMIT; 4435 4436 #ifdef TCP_SIGNATURE 4437 if (tb.t_flags & TF_SIGNATURE) 4438 sc->sc_flags |= SCF_SIGNATURE; 4439 #endif 4440 sc->sc_tp = tp; 4441 if (syn_cache_respond(sc, m) == 0) { 4442 uint64_t *tcps = TCP_STAT_GETREF(); 4443 tcps[TCP_STAT_SNDACKS]++; 4444 tcps[TCP_STAT_SNDTOTAL]++; 4445 TCP_STAT_PUTREF(); 4446 syn_cache_insert(sc, tp); 4447 } else { 4448 s = splsoftnet(); 4449 /* 4450 * syn_cache_put() will try to schedule the timer, so 4451 * we need to initialize it 4452 */ 4453 syn_cache_timer_arm(sc); 4454 syn_cache_put(sc); 4455 splx(s); 4456 TCP_STATINC(TCP_STAT_SC_DROPPED); 4457 } 4458 return (1); 4459 } 4460 4461 /* 4462 * syn_cache_respond: (re)send SYN+ACK. 4463 * 4464 * returns 0 on success. otherwise returns an errno, typically ENOBUFS. 4465 */ 4466 4467 int 4468 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4469 { 4470 #ifdef INET6 4471 struct rtentry *rt = NULL; 4472 #endif 4473 struct route *ro; 4474 u_int8_t *optp; 4475 int optlen, error; 4476 u_int16_t tlen; 4477 struct ip *ip = NULL; 4478 #ifdef INET6 4479 struct ip6_hdr *ip6 = NULL; 4480 #endif 4481 struct tcpcb *tp = NULL; 4482 struct tcphdr *th; 4483 u_int hlen; 4484 #ifdef TCP_SIGNATURE 4485 struct secasvar *sav = NULL; 4486 u_int8_t *sigp = NULL; 4487 #endif 4488 4489 ro = &sc->sc_route; 4490 switch (sc->sc_src.sa.sa_family) { 4491 case AF_INET: 4492 hlen = sizeof(struct ip); 4493 break; 4494 #ifdef INET6 4495 case AF_INET6: 4496 hlen = sizeof(struct ip6_hdr); 4497 break; 4498 #endif 4499 default: 4500 if (m) 4501 m_freem(m); 4502 return (EAFNOSUPPORT); 4503 } 4504 4505 /* worst case scanario, since we don't know the option size yet */ 4506 tlen = hlen + sizeof(struct tcphdr) + MAX_TCPOPTLEN; 4507 4508 /* 4509 * Create the IP+TCP header from scratch. 4510 */ 4511 if (m) 4512 m_freem(m); 4513 #ifdef DIAGNOSTIC 4514 if (max_linkhdr + tlen > MCLBYTES) 4515 return ENOBUFS; 4516 #endif 4517 4518 MGETHDR(m, M_DONTWAIT, MT_DATA); 4519 if (m && (max_linkhdr + tlen) > MHLEN) { 4520 MCLGET(m, M_DONTWAIT); 4521 if ((m->m_flags & M_EXT) == 0) { 4522 m_freem(m); 4523 m = NULL; 4524 } 4525 } 4526 if (m == NULL) 4527 return ENOBUFS; 4528 MCLAIM(m, &tcp_tx_mowner); 4529 4530 /* Fixup the mbuf. */ 4531 m->m_data += max_linkhdr; 4532 if (sc->sc_tp) 4533 tp = sc->sc_tp; 4534 m_reset_rcvif(m); 4535 memset(mtod(m, u_char *), 0, tlen); 4536 4537 switch (sc->sc_src.sa.sa_family) { 4538 case AF_INET: 4539 ip = mtod(m, struct ip *); 4540 ip->ip_v = 4; 4541 ip->ip_dst = sc->sc_src.sin.sin_addr; 4542 ip->ip_src = sc->sc_dst.sin.sin_addr; 4543 ip->ip_p = IPPROTO_TCP; 4544 th = (struct tcphdr *)(ip + 1); 4545 th->th_dport = sc->sc_src.sin.sin_port; 4546 th->th_sport = sc->sc_dst.sin.sin_port; 4547 break; 4548 #ifdef INET6 4549 case AF_INET6: 4550 ip6 = mtod(m, struct ip6_hdr *); 4551 ip6->ip6_vfc = IPV6_VERSION; 4552 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4553 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4554 ip6->ip6_nxt = IPPROTO_TCP; 4555 /* ip6_plen will be updated in ip6_output() */ 4556 th = (struct tcphdr *)(ip6 + 1); 4557 th->th_dport = sc->sc_src.sin6.sin6_port; 4558 th->th_sport = sc->sc_dst.sin6.sin6_port; 4559 break; 4560 #endif 4561 default: 4562 return ENOBUFS; 4563 } 4564 4565 th->th_seq = htonl(sc->sc_iss); 4566 th->th_ack = htonl(sc->sc_irs + 1); 4567 th->th_flags = TH_SYN|TH_ACK; 4568 th->th_win = htons(sc->sc_win); 4569 /* th_x2, th_sum, th_urp already 0 from memset */ 4570 4571 /* Tack on the TCP options. */ 4572 optp = (u_int8_t *)(th + 1); 4573 optlen = 0; 4574 *optp++ = TCPOPT_MAXSEG; 4575 *optp++ = TCPOLEN_MAXSEG; 4576 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4577 *optp++ = sc->sc_ourmaxseg & 0xff; 4578 optlen += TCPOLEN_MAXSEG; 4579 4580 if (sc->sc_request_r_scale != 15) { 4581 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4582 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4583 sc->sc_request_r_scale); 4584 optp += TCPOLEN_WINDOW + TCPOLEN_NOP; 4585 optlen += TCPOLEN_WINDOW + TCPOLEN_NOP; 4586 } 4587 4588 if (sc->sc_flags & SCF_SACK_PERMIT) { 4589 /* Let the peer know that we will SACK. */ 4590 *optp++ = TCPOPT_SACK_PERMITTED; 4591 *optp++ = TCPOLEN_SACK_PERMITTED; 4592 optlen += TCPOLEN_SACK_PERMITTED; 4593 } 4594 4595 if (sc->sc_flags & SCF_TIMESTAMP) { 4596 while (optlen % 4 != 2) { 4597 optlen += TCPOLEN_NOP; 4598 *optp++ = TCPOPT_NOP; 4599 } 4600 *optp++ = TCPOPT_TIMESTAMP; 4601 *optp++ = TCPOLEN_TIMESTAMP; 4602 u_int32_t *lp = (u_int32_t *)(optp); 4603 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4604 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4605 *lp = htonl(sc->sc_timestamp); 4606 optp += TCPOLEN_TIMESTAMP - 2; 4607 optlen += TCPOLEN_TIMESTAMP; 4608 } 4609 4610 #ifdef TCP_SIGNATURE 4611 if (sc->sc_flags & SCF_SIGNATURE) { 4612 4613 sav = tcp_signature_getsav(m, th); 4614 4615 if (sav == NULL) { 4616 if (m) 4617 m_freem(m); 4618 return (EPERM); 4619 } 4620 4621 *optp++ = TCPOPT_SIGNATURE; 4622 *optp++ = TCPOLEN_SIGNATURE; 4623 sigp = optp; 4624 memset(optp, 0, TCP_SIGLEN); 4625 optp += TCP_SIGLEN; 4626 optlen += TCPOLEN_SIGNATURE; 4627 4628 } 4629 #endif 4630 /* Terminate and pad TCP options to a 4 byte boundary. */ 4631 if (optlen % 4) { 4632 optlen += TCPOLEN_EOL; 4633 *optp++ = TCPOPT_EOL; 4634 } 4635 /* 4636 * According to RFC 793 (STD0007): 4637 * "The content of the header beyond the End-of-Option option 4638 * must be header padding (i.e., zero)." 4639 * and later: "The padding is composed of zeros." 4640 */ 4641 while (optlen % 4) { 4642 optlen += TCPOLEN_PAD; 4643 *optp++ = TCPOPT_PAD; 4644 } 4645 4646 /* compute the actual values now that we've added the options */ 4647 tlen = hlen + sizeof(struct tcphdr) + optlen; 4648 m->m_len = m->m_pkthdr.len = tlen; 4649 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4650 4651 #ifdef TCP_SIGNATURE 4652 if (sav) { 4653 (void)tcp_signature(m, th, hlen, sav, sigp); 4654 key_sa_recordxfer(sav, m); 4655 KEY_SA_UNREF(&sav); 4656 } 4657 #endif 4658 4659 /* 4660 * Send ECN SYN-ACK setup packet. 4661 * Routes can be asymetric, so, even if we receive a packet 4662 * with ECE and CWR set, we must not assume no one will block 4663 * the ECE packet we are about to send. 4664 */ 4665 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4666 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4667 th->th_flags |= TH_ECE; 4668 TCP_STATINC(TCP_STAT_ECN_SHS); 4669 4670 /* 4671 * draft-ietf-tcpm-ecnsyn-00.txt 4672 * 4673 * "[...] a TCP node MAY respond to an ECN-setup 4674 * SYN packet by setting ECT in the responding 4675 * ECN-setup SYN/ACK packet, indicating to routers 4676 * that the SYN/ACK packet is ECN-Capable. 4677 * This allows a congested router along the path 4678 * to mark the packet instead of dropping the 4679 * packet as an indication of congestion." 4680 * 4681 * "[...] There can be a great benefit in setting 4682 * an ECN-capable codepoint in SYN/ACK packets [...] 4683 * Congestion is most likely to occur in 4684 * the server-to-client direction. As a result, 4685 * setting an ECN-capable codepoint in SYN/ACK 4686 * packets can reduce the occurence of three-second 4687 * retransmit timeouts resulting from the drop 4688 * of SYN/ACK packets." 4689 * 4690 * Page 4 and 6, January 2006. 4691 */ 4692 4693 switch (sc->sc_src.sa.sa_family) { 4694 #ifdef INET 4695 case AF_INET: 4696 ip->ip_tos |= IPTOS_ECN_ECT0; 4697 break; 4698 #endif 4699 #ifdef INET6 4700 case AF_INET6: 4701 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4702 break; 4703 #endif 4704 } 4705 TCP_STATINC(TCP_STAT_ECN_ECT); 4706 } 4707 4708 4709 /* Compute the packet's checksum. */ 4710 switch (sc->sc_src.sa.sa_family) { 4711 case AF_INET: 4712 ip->ip_len = htons(tlen - hlen); 4713 th->th_sum = 0; 4714 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4715 break; 4716 #ifdef INET6 4717 case AF_INET6: 4718 ip6->ip6_plen = htons(tlen - hlen); 4719 th->th_sum = 0; 4720 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4721 break; 4722 #endif 4723 } 4724 4725 /* 4726 * Fill in some straggling IP bits. Note the stack expects 4727 * ip_len to be in host order, for convenience. 4728 */ 4729 switch (sc->sc_src.sa.sa_family) { 4730 #ifdef INET 4731 case AF_INET: 4732 ip->ip_len = htons(tlen); 4733 ip->ip_ttl = ip_defttl; 4734 /* XXX tos? */ 4735 break; 4736 #endif 4737 #ifdef INET6 4738 case AF_INET6: 4739 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4740 ip6->ip6_vfc |= IPV6_VERSION; 4741 ip6->ip6_plen = htons(tlen - hlen); 4742 /* ip6_hlim will be initialized afterwards */ 4743 /* XXX flowlabel? */ 4744 break; 4745 #endif 4746 } 4747 4748 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4749 tp = sc->sc_tp; 4750 4751 switch (sc->sc_src.sa.sa_family) { 4752 #ifdef INET 4753 case AF_INET: 4754 error = ip_output(m, sc->sc_ipopts, ro, 4755 (ip_mtudisc ? IP_MTUDISC : 0), 4756 NULL, tp ? tp->t_inpcb : NULL); 4757 break; 4758 #endif 4759 #ifdef INET6 4760 case AF_INET6: 4761 ip6->ip6_hlim = in6_selecthlim(NULL, 4762 (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp : NULL); 4763 rtcache_unref(rt, ro); 4764 4765 error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, 4766 tp ? tp->t_in6pcb : NULL, NULL); 4767 break; 4768 #endif 4769 default: 4770 error = EAFNOSUPPORT; 4771 break; 4772 } 4773 return (error); 4774 } 4775
Indexes created Mon Sep 22 10:09:38 GMT 2025