tcp_input.c revision 1.253
1 /* $NetBSD: tcp_input.c,v 1.253 2006/10/17 09:31:17 yamt 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 The NetBSD Foundation, Inc. 74 * All rights reserved. 75 * 76 * This code is derived from software contributed to The NetBSD Foundation 77 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 78 * Facility, NASA Ames Research Center. 79 * This code is derived from software contributed to The NetBSD Foundation 80 * by Charles M. Hannum. 81 * This code is derived from software contributed to The NetBSD Foundation 82 * by Rui Paulo. 83 * 84 * Redistribution and use in source and binary forms, with or without 85 * modification, are permitted provided that the following conditions 86 * are met: 87 * 1. Redistributions of source code must retain the above copyright 88 * notice, this list of conditions and the following disclaimer. 89 * 2. Redistributions in binary form must reproduce the above copyright 90 * notice, this list of conditions and the following disclaimer in the 91 * documentation and/or other materials provided with the distribution. 92 * 3. All advertising materials mentioning features or use of this software 93 * must display the following acknowledgement: 94 * This product includes software developed by the NetBSD 95 * Foundation, Inc. and its contributors. 96 * 4. Neither the name of The NetBSD Foundation nor the names of its 97 * contributors may be used to endorse or promote products derived 98 * from this software without specific prior written permission. 99 * 100 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 101 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 102 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 103 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 104 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 105 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 106 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 107 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 108 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 109 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 110 * POSSIBILITY OF SUCH DAMAGE. 111 */ 112 113 /* 114 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 115 * The Regents of the University of California. All rights reserved. 116 * 117 * Redistribution and use in source and binary forms, with or without 118 * modification, are permitted provided that the following conditions 119 * are met: 120 * 1. Redistributions of source code must retain the above copyright 121 * notice, this list of conditions and the following disclaimer. 122 * 2. Redistributions in binary form must reproduce the above copyright 123 * notice, this list of conditions and the following disclaimer in the 124 * documentation and/or other materials provided with the distribution. 125 * 3. Neither the name of the University nor the names of its contributors 126 * may be used to endorse or promote products derived from this software 127 * without specific prior written permission. 128 * 129 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 130 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 131 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 132 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 133 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 134 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 135 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 136 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 137 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 138 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 139 * SUCH DAMAGE. 140 * 141 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 142 */ 143 144 /* 145 * TODO list for SYN cache stuff: 146 * 147 * Find room for a "state" field, which is needed to keep a 148 * compressed state for TIME_WAIT TCBs. It's been noted already 149 * that this is fairly important for very high-volume web and 150 * mail servers, which use a large number of short-lived 151 * connections. 152 */ 153 154 #include <sys/cdefs.h> 155 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.253 2006/10/17 09:31:17 yamt Exp $"); 156 157 #include "opt_inet.h" 158 #include "opt_ipsec.h" 159 #include "opt_inet_csum.h" 160 #include "opt_tcp_debug.h" 161 162 #include <sys/param.h> 163 #include <sys/systm.h> 164 #include <sys/malloc.h> 165 #include <sys/mbuf.h> 166 #include <sys/protosw.h> 167 #include <sys/socket.h> 168 #include <sys/socketvar.h> 169 #include <sys/errno.h> 170 #include <sys/syslog.h> 171 #include <sys/pool.h> 172 #include <sys/domain.h> 173 #include <sys/kernel.h> 174 #ifdef TCP_SIGNATURE 175 #include <sys/md5.h> 176 #endif 177 178 #include <net/if.h> 179 #include <net/route.h> 180 #include <net/if_types.h> 181 182 #include <netinet/in.h> 183 #include <netinet/in_systm.h> 184 #include <netinet/ip.h> 185 #include <netinet/in_pcb.h> 186 #include <netinet/in_var.h> 187 #include <netinet/ip_var.h> 188 #include <netinet/in_offload.h> 189 190 #ifdef INET6 191 #ifndef INET 192 #include <netinet/in.h> 193 #endif 194 #include <netinet/ip6.h> 195 #include <netinet6/ip6_var.h> 196 #include <netinet6/in6_pcb.h> 197 #include <netinet6/ip6_var.h> 198 #include <netinet6/in6_var.h> 199 #include <netinet/icmp6.h> 200 #include <netinet6/nd6.h> 201 #ifdef TCP_SIGNATURE 202 #include <netinet6/scope6_var.h> 203 #endif 204 #endif 205 206 #ifndef INET6 207 /* always need ip6.h for IP6_EXTHDR_GET */ 208 #include <netinet/ip6.h> 209 #endif 210 211 #include <netinet/tcp.h> 212 #include <netinet/tcp_fsm.h> 213 #include <netinet/tcp_seq.h> 214 #include <netinet/tcp_timer.h> 215 #include <netinet/tcp_var.h> 216 #include <netinet/tcpip.h> 217 #include <netinet/tcp_congctl.h> 218 #include <netinet/tcp_debug.h> 219 220 #include <machine/stdarg.h> 221 222 #ifdef IPSEC 223 #include <netinet6/ipsec.h> 224 #include <netkey/key.h> 225 #endif /*IPSEC*/ 226 #ifdef INET6 227 #include "faith.h" 228 #if defined(NFAITH) && NFAITH > 0 229 #include <net/if_faith.h> 230 #endif 231 #endif /* IPSEC */ 232 233 #ifdef FAST_IPSEC 234 #include <netipsec/ipsec.h> 235 #include <netipsec/ipsec_var.h> /* XXX ipsecstat namespace */ 236 #include <netipsec/key.h> 237 #ifdef INET6 238 #include <netipsec/ipsec6.h> 239 #endif 240 #endif /* FAST_IPSEC*/ 241 242 int tcprexmtthresh = 3; 243 int tcp_log_refused; 244 245 static int tcp_rst_ppslim_count = 0; 246 static struct timeval tcp_rst_ppslim_last; 247 static int tcp_ackdrop_ppslim_count = 0; 248 static struct timeval tcp_ackdrop_ppslim_last; 249 250 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 251 252 /* for modulo comparisons of timestamps */ 253 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 254 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 255 256 /* 257 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 258 */ 259 #ifdef INET6 260 #define ND6_HINT(tp) \ 261 do { \ 262 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 && \ 263 tp->t_in6pcb->in6p_route.ro_rt) { \ 264 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \ 265 } \ 266 } while (/*CONSTCOND*/ 0) 267 #else 268 #define ND6_HINT(tp) 269 #endif 270 271 /* 272 * Macro to compute ACK transmission behavior. Delay the ACK unless 273 * we have already delayed an ACK (must send an ACK every two segments). 274 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 275 * option is enabled. 276 */ 277 #define TCP_SETUP_ACK(tp, th) \ 278 do { \ 279 if ((tp)->t_flags & TF_DELACK || \ 280 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 281 tp->t_flags |= TF_ACKNOW; \ 282 else \ 283 TCP_SET_DELACK(tp); \ 284 } while (/*CONSTCOND*/ 0) 285 286 #define ICMP_CHECK(tp, th, acked) \ 287 do { \ 288 /* \ 289 * If we had a pending ICMP message that \ 290 * refers to data that have just been \ 291 * acknowledged, disregard the recorded ICMP \ 292 * message. \ 293 */ \ 294 if (((tp)->t_flags & TF_PMTUD_PEND) && \ 295 SEQ_GT((th)->th_ack, (tp)->t_pmtud_th_seq)) \ 296 (tp)->t_flags &= ~TF_PMTUD_PEND; \ 297 \ 298 /* \ 299 * Keep track of the largest chunk of data \ 300 * acknowledged since last PMTU update \ 301 */ \ 302 if ((tp)->t_pmtud_mss_acked < (acked)) \ 303 (tp)->t_pmtud_mss_acked = (acked); \ 304 } while (/*CONSTCOND*/ 0) 305 306 /* 307 * Convert TCP protocol fields to host order for easier processing. 308 */ 309 #define TCP_FIELDS_TO_HOST(th) \ 310 do { \ 311 NTOHL((th)->th_seq); \ 312 NTOHL((th)->th_ack); \ 313 NTOHS((th)->th_win); \ 314 NTOHS((th)->th_urp); \ 315 } while (/*CONSTCOND*/ 0) 316 317 /* 318 * ... and reverse the above. 319 */ 320 #define TCP_FIELDS_TO_NET(th) \ 321 do { \ 322 HTONL((th)->th_seq); \ 323 HTONL((th)->th_ack); \ 324 HTONS((th)->th_win); \ 325 HTONS((th)->th_urp); \ 326 } while (/*CONSTCOND*/ 0) 327 328 #ifdef TCP_CSUM_COUNTERS 329 #include <sys/device.h> 330 331 #if defined(INET) 332 extern struct evcnt tcp_hwcsum_ok; 333 extern struct evcnt tcp_hwcsum_bad; 334 extern struct evcnt tcp_hwcsum_data; 335 extern struct evcnt tcp_swcsum; 336 #endif /* defined(INET) */ 337 #if defined(INET6) 338 extern struct evcnt tcp6_hwcsum_ok; 339 extern struct evcnt tcp6_hwcsum_bad; 340 extern struct evcnt tcp6_hwcsum_data; 341 extern struct evcnt tcp6_swcsum; 342 #endif /* defined(INET6) */ 343 344 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 345 346 #else 347 348 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 349 350 #endif /* TCP_CSUM_COUNTERS */ 351 352 #ifdef TCP_REASS_COUNTERS 353 #include <sys/device.h> 354 355 extern struct evcnt tcp_reass_; 356 extern struct evcnt tcp_reass_empty; 357 extern struct evcnt tcp_reass_iteration[8]; 358 extern struct evcnt tcp_reass_prependfirst; 359 extern struct evcnt tcp_reass_prepend; 360 extern struct evcnt tcp_reass_insert; 361 extern struct evcnt tcp_reass_inserttail; 362 extern struct evcnt tcp_reass_append; 363 extern struct evcnt tcp_reass_appendtail; 364 extern struct evcnt tcp_reass_overlaptail; 365 extern struct evcnt tcp_reass_overlapfront; 366 extern struct evcnt tcp_reass_segdup; 367 extern struct evcnt tcp_reass_fragdup; 368 369 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 370 371 #else 372 373 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 374 375 #endif /* TCP_REASS_COUNTERS */ 376 377 #ifdef INET 378 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 379 #endif 380 #ifdef INET6 381 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 382 #endif 383 384 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 385 386 POOL_INIT(tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", NULL); 387 388 struct ipqent * 389 tcpipqent_alloc() 390 { 391 struct ipqent *ipqe; 392 int s; 393 394 s = splvm(); 395 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 396 splx(s); 397 398 return ipqe; 399 } 400 401 void 402 tcpipqent_free(struct ipqent *ipqe) 403 { 404 int s; 405 406 s = splvm(); 407 pool_put(&tcpipqent_pool, ipqe); 408 splx(s); 409 } 410 411 int 412 tcp_reass(struct tcpcb *tp, struct tcphdr *th, struct mbuf *m, int *tlen) 413 { 414 struct ipqent *p, *q, *nq, *tiqe = NULL; 415 struct socket *so = NULL; 416 int pkt_flags; 417 tcp_seq pkt_seq; 418 unsigned pkt_len; 419 u_long rcvpartdupbyte = 0; 420 u_long rcvoobyte; 421 #ifdef TCP_REASS_COUNTERS 422 u_int count = 0; 423 #endif 424 425 if (tp->t_inpcb) 426 so = tp->t_inpcb->inp_socket; 427 #ifdef INET6 428 else if (tp->t_in6pcb) 429 so = tp->t_in6pcb->in6p_socket; 430 #endif 431 432 TCP_REASS_LOCK_CHECK(tp); 433 434 /* 435 * Call with th==0 after become established to 436 * force pre-ESTABLISHED data up to user socket. 437 */ 438 if (th == 0) 439 goto present; 440 441 rcvoobyte = *tlen; 442 /* 443 * Copy these to local variables because the tcpiphdr 444 * gets munged while we are collapsing mbufs. 445 */ 446 pkt_seq = th->th_seq; 447 pkt_len = *tlen; 448 pkt_flags = th->th_flags; 449 450 TCP_REASS_COUNTER_INCR(&tcp_reass_); 451 452 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 453 /* 454 * When we miss a packet, the vast majority of time we get 455 * packets that follow it in order. So optimize for that. 456 */ 457 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 458 p->ipqe_len += pkt_len; 459 p->ipqe_flags |= pkt_flags; 460 m_cat(p->ipre_mlast, m); 461 TRAVERSE(p->ipre_mlast); 462 m = NULL; 463 tiqe = p; 464 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 465 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 466 goto skip_replacement; 467 } 468 /* 469 * While we're here, if the pkt is completely beyond 470 * anything we have, just insert it at the tail. 471 */ 472 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 473 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 474 goto insert_it; 475 } 476 } 477 478 q = TAILQ_FIRST(&tp->segq); 479 480 if (q != NULL) { 481 /* 482 * If this segment immediately precedes the first out-of-order 483 * block, simply slap the segment in front of it and (mostly) 484 * skip the complicated logic. 485 */ 486 if (pkt_seq + pkt_len == q->ipqe_seq) { 487 q->ipqe_seq = pkt_seq; 488 q->ipqe_len += pkt_len; 489 q->ipqe_flags |= pkt_flags; 490 m_cat(m, q->ipqe_m); 491 q->ipqe_m = m; 492 q->ipre_mlast = m; /* last mbuf may have changed */ 493 TRAVERSE(q->ipre_mlast); 494 tiqe = q; 495 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 496 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 497 goto skip_replacement; 498 } 499 } else { 500 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 501 } 502 503 /* 504 * Find a segment which begins after this one does. 505 */ 506 for (p = NULL; q != NULL; q = nq) { 507 nq = TAILQ_NEXT(q, ipqe_q); 508 #ifdef TCP_REASS_COUNTERS 509 count++; 510 #endif 511 /* 512 * If the received segment is just right after this 513 * fragment, merge the two together and then check 514 * for further overlaps. 515 */ 516 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 517 #ifdef TCPREASS_DEBUG 518 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 519 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 520 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 521 #endif 522 pkt_len += q->ipqe_len; 523 pkt_flags |= q->ipqe_flags; 524 pkt_seq = q->ipqe_seq; 525 m_cat(q->ipre_mlast, m); 526 TRAVERSE(q->ipre_mlast); 527 m = q->ipqe_m; 528 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 529 goto free_ipqe; 530 } 531 /* 532 * If the received segment is completely past this 533 * fragment, we need to go the next fragment. 534 */ 535 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 536 p = q; 537 continue; 538 } 539 /* 540 * If the fragment is past the received segment, 541 * it (or any following) can't be concatenated. 542 */ 543 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 544 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 545 break; 546 } 547 548 /* 549 * We've received all the data in this segment before. 550 * mark it as a duplicate and return. 551 */ 552 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 553 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 554 tcpstat.tcps_rcvduppack++; 555 tcpstat.tcps_rcvdupbyte += pkt_len; 556 tcp_new_dsack(tp, pkt_seq, pkt_len); 557 m_freem(m); 558 if (tiqe != NULL) { 559 tcpipqent_free(tiqe); 560 } 561 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 562 return (0); 563 } 564 /* 565 * Received segment completely overlaps this fragment 566 * so we drop the fragment (this keeps the temporal 567 * ordering of segments correct). 568 */ 569 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 570 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 571 rcvpartdupbyte += q->ipqe_len; 572 m_freem(q->ipqe_m); 573 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 574 goto free_ipqe; 575 } 576 /* 577 * RX'ed segment extends past the end of the 578 * fragment. Drop the overlapping bytes. Then 579 * merge the fragment and segment then treat as 580 * a longer received packet. 581 */ 582 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 583 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 584 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 585 #ifdef TCPREASS_DEBUG 586 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 587 tp, overlap, 588 pkt_seq, pkt_seq + pkt_len, pkt_len); 589 #endif 590 m_adj(m, overlap); 591 rcvpartdupbyte += overlap; 592 m_cat(q->ipre_mlast, m); 593 TRAVERSE(q->ipre_mlast); 594 m = q->ipqe_m; 595 pkt_seq = q->ipqe_seq; 596 pkt_len += q->ipqe_len - overlap; 597 rcvoobyte -= overlap; 598 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 599 goto free_ipqe; 600 } 601 /* 602 * RX'ed segment extends past the front of the 603 * fragment. Drop the overlapping bytes on the 604 * received packet. The packet will then be 605 * contatentated with this fragment a bit later. 606 */ 607 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 608 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 609 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 610 #ifdef TCPREASS_DEBUG 611 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 612 tp, overlap, 613 pkt_seq, pkt_seq + pkt_len, pkt_len); 614 #endif 615 m_adj(m, -overlap); 616 pkt_len -= overlap; 617 rcvpartdupbyte += overlap; 618 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 619 rcvoobyte -= overlap; 620 } 621 /* 622 * If the received segment immediates precedes this 623 * fragment then tack the fragment onto this segment 624 * and reinsert the data. 625 */ 626 if (q->ipqe_seq == pkt_seq + pkt_len) { 627 #ifdef TCPREASS_DEBUG 628 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 629 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 630 pkt_seq, pkt_seq + pkt_len, pkt_len); 631 #endif 632 pkt_len += q->ipqe_len; 633 pkt_flags |= q->ipqe_flags; 634 m_cat(m, q->ipqe_m); 635 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 636 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 637 tp->t_segqlen--; 638 KASSERT(tp->t_segqlen >= 0); 639 KASSERT(tp->t_segqlen != 0 || 640 (TAILQ_EMPTY(&tp->segq) && 641 TAILQ_EMPTY(&tp->timeq))); 642 if (tiqe == NULL) { 643 tiqe = q; 644 } else { 645 tcpipqent_free(q); 646 } 647 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 648 break; 649 } 650 /* 651 * If the fragment is before the segment, remember it. 652 * When this loop is terminated, p will contain the 653 * pointer to fragment that is right before the received 654 * segment. 655 */ 656 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 657 p = q; 658 659 continue; 660 661 /* 662 * This is a common operation. It also will allow 663 * to save doing a malloc/free in most instances. 664 */ 665 free_ipqe: 666 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 667 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 668 tp->t_segqlen--; 669 KASSERT(tp->t_segqlen >= 0); 670 KASSERT(tp->t_segqlen != 0 || 671 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 672 if (tiqe == NULL) { 673 tiqe = q; 674 } else { 675 tcpipqent_free(q); 676 } 677 } 678 679 #ifdef TCP_REASS_COUNTERS 680 if (count > 7) 681 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 682 else if (count > 0) 683 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 684 #endif 685 686 insert_it: 687 688 /* 689 * Allocate a new queue entry since the received segment did not 690 * collapse onto any other out-of-order block; thus we are allocating 691 * a new block. If it had collapsed, tiqe would not be NULL and 692 * we would be reusing it. 693 * XXX If we can't, just drop the packet. XXX 694 */ 695 if (tiqe == NULL) { 696 tiqe = tcpipqent_alloc(); 697 if (tiqe == NULL) { 698 tcpstat.tcps_rcvmemdrop++; 699 m_freem(m); 700 return (0); 701 } 702 } 703 704 /* 705 * Update the counters. 706 */ 707 tcpstat.tcps_rcvoopack++; 708 tcpstat.tcps_rcvoobyte += rcvoobyte; 709 if (rcvpartdupbyte) { 710 tcpstat.tcps_rcvpartduppack++; 711 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 712 } 713 714 /* 715 * Insert the new fragment queue entry into both queues. 716 */ 717 tiqe->ipqe_m = m; 718 tiqe->ipre_mlast = m; 719 tiqe->ipqe_seq = pkt_seq; 720 tiqe->ipqe_len = pkt_len; 721 tiqe->ipqe_flags = pkt_flags; 722 if (p == NULL) { 723 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 724 #ifdef TCPREASS_DEBUG 725 if (tiqe->ipqe_seq != tp->rcv_nxt) 726 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 727 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 728 #endif 729 } else { 730 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 731 #ifdef TCPREASS_DEBUG 732 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 733 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 734 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 735 #endif 736 } 737 tp->t_segqlen++; 738 739 skip_replacement: 740 741 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 742 743 present: 744 /* 745 * Present data to user, advancing rcv_nxt through 746 * completed sequence space. 747 */ 748 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 749 return (0); 750 q = TAILQ_FIRST(&tp->segq); 751 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 752 return (0); 753 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 754 return (0); 755 756 tp->rcv_nxt += q->ipqe_len; 757 pkt_flags = q->ipqe_flags & TH_FIN; 758 ND6_HINT(tp); 759 760 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 761 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 762 tp->t_segqlen--; 763 KASSERT(tp->t_segqlen >= 0); 764 KASSERT(tp->t_segqlen != 0 || 765 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 766 if (so->so_state & SS_CANTRCVMORE) 767 m_freem(q->ipqe_m); 768 else 769 sbappendstream(&so->so_rcv, q->ipqe_m); 770 tcpipqent_free(q); 771 sorwakeup(so); 772 return (pkt_flags); 773 } 774 775 #ifdef INET6 776 int 777 tcp6_input(struct mbuf **mp, int *offp, int proto) 778 { 779 struct mbuf *m = *mp; 780 781 /* 782 * draft-itojun-ipv6-tcp-to-anycast 783 * better place to put this in? 784 */ 785 if (m->m_flags & M_ANYCAST6) { 786 struct ip6_hdr *ip6; 787 if (m->m_len < sizeof(struct ip6_hdr)) { 788 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 789 tcpstat.tcps_rcvshort++; 790 return IPPROTO_DONE; 791 } 792 } 793 ip6 = mtod(m, struct ip6_hdr *); 794 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 795 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 796 return IPPROTO_DONE; 797 } 798 799 tcp_input(m, *offp, proto); 800 return IPPROTO_DONE; 801 } 802 #endif 803 804 #ifdef INET 805 static void 806 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 807 { 808 char src[4*sizeof "123"]; 809 char dst[4*sizeof "123"]; 810 811 if (ip) { 812 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 813 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 814 } 815 else { 816 strlcpy(src, "(unknown)", sizeof(src)); 817 strlcpy(dst, "(unknown)", sizeof(dst)); 818 } 819 log(LOG_INFO, 820 "Connection attempt to TCP %s:%d from %s:%d\n", 821 dst, ntohs(th->th_dport), 822 src, ntohs(th->th_sport)); 823 } 824 #endif 825 826 #ifdef INET6 827 static void 828 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 829 { 830 char src[INET6_ADDRSTRLEN]; 831 char dst[INET6_ADDRSTRLEN]; 832 833 if (ip6) { 834 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 835 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 836 } 837 else { 838 strlcpy(src, "(unknown v6)", sizeof(src)); 839 strlcpy(dst, "(unknown v6)", sizeof(dst)); 840 } 841 log(LOG_INFO, 842 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 843 dst, ntohs(th->th_dport), 844 src, ntohs(th->th_sport)); 845 } 846 #endif 847 848 /* 849 * Checksum extended TCP header and data. 850 */ 851 int 852 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th __unused, 853 int toff, int off, int tlen) 854 { 855 856 /* 857 * XXX it's better to record and check if this mbuf is 858 * already checked. 859 */ 860 861 switch (af) { 862 #ifdef INET 863 case AF_INET: 864 switch (m->m_pkthdr.csum_flags & 865 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 866 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 867 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 868 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 869 goto badcsum; 870 871 case M_CSUM_TCPv4|M_CSUM_DATA: { 872 u_int32_t hw_csum = m->m_pkthdr.csum_data; 873 874 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 875 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 876 const struct ip *ip = 877 mtod(m, const struct ip *); 878 879 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 880 ip->ip_dst.s_addr, 881 htons(hw_csum + tlen + off + IPPROTO_TCP)); 882 } 883 if ((hw_csum ^ 0xffff) != 0) 884 goto badcsum; 885 break; 886 } 887 888 case M_CSUM_TCPv4: 889 /* Checksum was okay. */ 890 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 891 break; 892 893 default: 894 /* 895 * Must compute it ourselves. Maybe skip checksum 896 * on loopback interfaces. 897 */ 898 if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & 899 IFF_LOOPBACK) || 900 tcp_do_loopback_cksum)) { 901 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 902 if (in4_cksum(m, IPPROTO_TCP, toff, 903 tlen + off) != 0) 904 goto badcsum; 905 } 906 break; 907 } 908 break; 909 #endif /* INET4 */ 910 911 #ifdef INET6 912 case AF_INET6: 913 switch (m->m_pkthdr.csum_flags & 914 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 915 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 916 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 917 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 918 goto badcsum; 919 920 #if 0 /* notyet */ 921 case M_CSUM_TCPv6|M_CSUM_DATA: 922 #endif 923 924 case M_CSUM_TCPv6: 925 /* Checksum was okay. */ 926 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 927 break; 928 929 default: 930 /* 931 * Must compute it ourselves. Maybe skip checksum 932 * on loopback interfaces. 933 */ 934 if (__predict_true((m->m_flags & M_LOOP) == 0 || 935 tcp_do_loopback_cksum)) { 936 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 937 if (in6_cksum(m, IPPROTO_TCP, toff, 938 tlen + off) != 0) 939 goto badcsum; 940 } 941 } 942 break; 943 #endif /* INET6 */ 944 } 945 946 return 0; 947 948 badcsum: 949 tcpstat.tcps_rcvbadsum++; 950 return -1; 951 } 952 953 /* 954 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 955 */ 956 void 957 tcp_input(struct mbuf *m, ...) 958 { 959 struct tcphdr *th; 960 struct ip *ip; 961 struct inpcb *inp; 962 #ifdef INET6 963 struct ip6_hdr *ip6; 964 struct in6pcb *in6p; 965 #endif 966 u_int8_t *optp = NULL; 967 int optlen = 0; 968 int len, tlen, toff, hdroptlen = 0; 969 struct tcpcb *tp = 0; 970 int tiflags; 971 struct socket *so = NULL; 972 int todrop, dupseg, acked, ourfinisacked, needoutput = 0; 973 #ifdef TCP_DEBUG 974 short ostate = 0; 975 #endif 976 u_long tiwin; 977 struct tcp_opt_info opti; 978 int off, iphlen; 979 va_list ap; 980 int af; /* af on the wire */ 981 struct mbuf *tcp_saveti = NULL; 982 uint32_t ts_rtt; 983 uint8_t iptos; 984 985 MCLAIM(m, &tcp_rx_mowner); 986 va_start(ap, m); 987 toff = va_arg(ap, int); 988 (void)va_arg(ap, int); /* ignore value, advance ap */ 989 va_end(ap); 990 991 tcpstat.tcps_rcvtotal++; 992 993 bzero(&opti, sizeof(opti)); 994 opti.ts_present = 0; 995 opti.maxseg = 0; 996 997 /* 998 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 999 * 1000 * TCP is, by definition, unicast, so we reject all 1001 * multicast outright. 1002 * 1003 * Note, there are additional src/dst address checks in 1004 * the AF-specific code below. 1005 */ 1006 if (m->m_flags & (M_BCAST|M_MCAST)) { 1007 /* XXX stat */ 1008 goto drop; 1009 } 1010 #ifdef INET6 1011 if (m->m_flags & M_ANYCAST6) { 1012 /* XXX stat */ 1013 goto drop; 1014 } 1015 #endif 1016 1017 /* 1018 * Get IP and TCP header. 1019 * Note: IP leaves IP header in first mbuf. 1020 */ 1021 ip = mtod(m, struct ip *); 1022 #ifdef INET6 1023 ip6 = NULL; 1024 #endif 1025 switch (ip->ip_v) { 1026 #ifdef INET 1027 case 4: 1028 af = AF_INET; 1029 iphlen = sizeof(struct ip); 1030 ip = mtod(m, struct ip *); 1031 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1032 sizeof(struct tcphdr)); 1033 if (th == NULL) { 1034 tcpstat.tcps_rcvshort++; 1035 return; 1036 } 1037 /* We do the checksum after PCB lookup... */ 1038 len = ntohs(ip->ip_len); 1039 tlen = len - toff; 1040 iptos = ip->ip_tos; 1041 break; 1042 #endif 1043 #ifdef INET6 1044 case 6: 1045 ip = NULL; 1046 iphlen = sizeof(struct ip6_hdr); 1047 af = AF_INET6; 1048 ip6 = mtod(m, struct ip6_hdr *); 1049 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1050 sizeof(struct tcphdr)); 1051 if (th == NULL) { 1052 tcpstat.tcps_rcvshort++; 1053 return; 1054 } 1055 1056 /* Be proactive about malicious use of IPv4 mapped address */ 1057 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 1058 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 1059 /* XXX stat */ 1060 goto drop; 1061 } 1062 1063 /* 1064 * Be proactive about unspecified IPv6 address in source. 1065 * As we use all-zero to indicate unbounded/unconnected pcb, 1066 * unspecified IPv6 address can be used to confuse us. 1067 * 1068 * Note that packets with unspecified IPv6 destination is 1069 * already dropped in ip6_input. 1070 */ 1071 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1072 /* XXX stat */ 1073 goto drop; 1074 } 1075 1076 /* 1077 * Make sure destination address is not multicast. 1078 * Source address checked in ip6_input(). 1079 */ 1080 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1081 /* XXX stat */ 1082 goto drop; 1083 } 1084 1085 /* We do the checksum after PCB lookup... */ 1086 len = m->m_pkthdr.len; 1087 tlen = len - toff; 1088 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1089 break; 1090 #endif 1091 default: 1092 m_freem(m); 1093 return; 1094 } 1095 1096 KASSERT(TCP_HDR_ALIGNED_P(th)); 1097 1098 /* 1099 * Check that TCP offset makes sense, 1100 * pull out TCP options and adjust length. XXX 1101 */ 1102 off = th->th_off << 2; 1103 if (off < sizeof (struct tcphdr) || off > tlen) { 1104 tcpstat.tcps_rcvbadoff++; 1105 goto drop; 1106 } 1107 tlen -= off; 1108 1109 /* 1110 * tcp_input() has been modified to use tlen to mean the TCP data 1111 * length throughout the function. Other functions can use 1112 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1113 * rja 1114 */ 1115 1116 if (off > sizeof (struct tcphdr)) { 1117 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1118 if (th == NULL) { 1119 tcpstat.tcps_rcvshort++; 1120 return; 1121 } 1122 /* 1123 * NOTE: ip/ip6 will not be affected by m_pulldown() 1124 * (as they're before toff) and we don't need to update those. 1125 */ 1126 KASSERT(TCP_HDR_ALIGNED_P(th)); 1127 optlen = off - sizeof (struct tcphdr); 1128 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1129 /* 1130 * Do quick retrieval of timestamp options ("options 1131 * prediction?"). If timestamp is the only option and it's 1132 * formatted as recommended in RFC 1323 appendix A, we 1133 * quickly get the values now and not bother calling 1134 * tcp_dooptions(), etc. 1135 */ 1136 if ((optlen == TCPOLEN_TSTAMP_APPA || 1137 (optlen > TCPOLEN_TSTAMP_APPA && 1138 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1139 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1140 (th->th_flags & TH_SYN) == 0) { 1141 opti.ts_present = 1; 1142 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1143 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1144 optp = NULL; /* we've parsed the options */ 1145 } 1146 } 1147 tiflags = th->th_flags; 1148 1149 /* 1150 * Locate pcb for segment. 1151 */ 1152 findpcb: 1153 inp = NULL; 1154 #ifdef INET6 1155 in6p = NULL; 1156 #endif 1157 switch (af) { 1158 #ifdef INET 1159 case AF_INET: 1160 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1161 ip->ip_dst, th->th_dport); 1162 if (inp == 0) { 1163 ++tcpstat.tcps_pcbhashmiss; 1164 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1165 } 1166 #ifdef INET6 1167 if (inp == 0) { 1168 struct in6_addr s, d; 1169 1170 /* mapped addr case */ 1171 bzero(&s, sizeof(s)); 1172 s.s6_addr16[5] = htons(0xffff); 1173 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1174 bzero(&d, sizeof(d)); 1175 d.s6_addr16[5] = htons(0xffff); 1176 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1177 in6p = in6_pcblookup_connect(&tcbtable, &s, 1178 th->th_sport, &d, th->th_dport, 0); 1179 if (in6p == 0) { 1180 ++tcpstat.tcps_pcbhashmiss; 1181 in6p = in6_pcblookup_bind(&tcbtable, &d, 1182 th->th_dport, 0); 1183 } 1184 } 1185 #endif 1186 #ifndef INET6 1187 if (inp == 0) 1188 #else 1189 if (inp == 0 && in6p == 0) 1190 #endif 1191 { 1192 ++tcpstat.tcps_noport; 1193 if (tcp_log_refused && 1194 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1195 tcp4_log_refused(ip, th); 1196 } 1197 TCP_FIELDS_TO_HOST(th); 1198 goto dropwithreset_ratelim; 1199 } 1200 #if defined(IPSEC) || defined(FAST_IPSEC) 1201 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1202 ipsec4_in_reject(m, inp)) { 1203 ipsecstat.in_polvio++; 1204 goto drop; 1205 } 1206 #ifdef INET6 1207 else if (in6p && 1208 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1209 ipsec4_in_reject_so(m, in6p->in6p_socket)) { 1210 ipsecstat.in_polvio++; 1211 goto drop; 1212 } 1213 #endif 1214 #endif /*IPSEC*/ 1215 break; 1216 #endif /*INET*/ 1217 #ifdef INET6 1218 case AF_INET6: 1219 { 1220 int faith; 1221 1222 #if defined(NFAITH) && NFAITH > 0 1223 faith = faithprefix(&ip6->ip6_dst); 1224 #else 1225 faith = 0; 1226 #endif 1227 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1228 th->th_sport, &ip6->ip6_dst, th->th_dport, faith); 1229 if (in6p == NULL) { 1230 ++tcpstat.tcps_pcbhashmiss; 1231 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1232 th->th_dport, faith); 1233 } 1234 if (in6p == NULL) { 1235 ++tcpstat.tcps_noport; 1236 if (tcp_log_refused && 1237 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1238 tcp6_log_refused(ip6, th); 1239 } 1240 TCP_FIELDS_TO_HOST(th); 1241 goto dropwithreset_ratelim; 1242 } 1243 #if defined(IPSEC) || defined(FAST_IPSEC) 1244 if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1245 ipsec6_in_reject(m, in6p)) { 1246 ipsec6stat.in_polvio++; 1247 goto drop; 1248 } 1249 #endif /*IPSEC*/ 1250 break; 1251 } 1252 #endif 1253 } 1254 1255 /* 1256 * If the state is CLOSED (i.e., TCB does not exist) then 1257 * all data in the incoming segment is discarded. 1258 * If the TCB exists but is in CLOSED state, it is embryonic, 1259 * but should either do a listen or a connect soon. 1260 */ 1261 tp = NULL; 1262 so = NULL; 1263 if (inp) { 1264 tp = intotcpcb(inp); 1265 so = inp->inp_socket; 1266 } 1267 #ifdef INET6 1268 else if (in6p) { 1269 tp = in6totcpcb(in6p); 1270 so = in6p->in6p_socket; 1271 } 1272 #endif 1273 if (tp == 0) { 1274 TCP_FIELDS_TO_HOST(th); 1275 goto dropwithreset_ratelim; 1276 } 1277 if (tp->t_state == TCPS_CLOSED) 1278 goto drop; 1279 1280 /* 1281 * Checksum extended TCP header and data. 1282 */ 1283 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1284 goto badcsum; 1285 1286 TCP_FIELDS_TO_HOST(th); 1287 1288 /* Unscale the window into a 32-bit value. */ 1289 if ((tiflags & TH_SYN) == 0) 1290 tiwin = th->th_win << tp->snd_scale; 1291 else 1292 tiwin = th->th_win; 1293 1294 #ifdef INET6 1295 /* save packet options if user wanted */ 1296 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1297 if (in6p->in6p_options) { 1298 m_freem(in6p->in6p_options); 1299 in6p->in6p_options = 0; 1300 } 1301 KASSERT(ip6 != NULL); 1302 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1303 } 1304 #endif 1305 1306 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1307 union syn_cache_sa src; 1308 union syn_cache_sa dst; 1309 1310 bzero(&src, sizeof(src)); 1311 bzero(&dst, sizeof(dst)); 1312 switch (af) { 1313 #ifdef INET 1314 case AF_INET: 1315 src.sin.sin_len = sizeof(struct sockaddr_in); 1316 src.sin.sin_family = AF_INET; 1317 src.sin.sin_addr = ip->ip_src; 1318 src.sin.sin_port = th->th_sport; 1319 1320 dst.sin.sin_len = sizeof(struct sockaddr_in); 1321 dst.sin.sin_family = AF_INET; 1322 dst.sin.sin_addr = ip->ip_dst; 1323 dst.sin.sin_port = th->th_dport; 1324 break; 1325 #endif 1326 #ifdef INET6 1327 case AF_INET6: 1328 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1329 src.sin6.sin6_family = AF_INET6; 1330 src.sin6.sin6_addr = ip6->ip6_src; 1331 src.sin6.sin6_port = th->th_sport; 1332 1333 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1334 dst.sin6.sin6_family = AF_INET6; 1335 dst.sin6.sin6_addr = ip6->ip6_dst; 1336 dst.sin6.sin6_port = th->th_dport; 1337 break; 1338 #endif /* INET6 */ 1339 default: 1340 goto badsyn; /*sanity*/ 1341 } 1342 1343 if (so->so_options & SO_DEBUG) { 1344 #ifdef TCP_DEBUG 1345 ostate = tp->t_state; 1346 #endif 1347 1348 tcp_saveti = NULL; 1349 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1350 goto nosave; 1351 1352 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1353 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1354 if (!tcp_saveti) 1355 goto nosave; 1356 } else { 1357 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1358 if (!tcp_saveti) 1359 goto nosave; 1360 MCLAIM(m, &tcp_mowner); 1361 tcp_saveti->m_len = iphlen; 1362 m_copydata(m, 0, iphlen, 1363 mtod(tcp_saveti, caddr_t)); 1364 } 1365 1366 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1367 m_freem(tcp_saveti); 1368 tcp_saveti = NULL; 1369 } else { 1370 tcp_saveti->m_len += sizeof(struct tcphdr); 1371 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1372 sizeof(struct tcphdr)); 1373 } 1374 nosave:; 1375 } 1376 if (so->so_options & SO_ACCEPTCONN) { 1377 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1378 if (tiflags & TH_RST) { 1379 syn_cache_reset(&src.sa, &dst.sa, th); 1380 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1381 (TH_ACK|TH_SYN)) { 1382 /* 1383 * Received a SYN,ACK. This should 1384 * never happen while we are in 1385 * LISTEN. Send an RST. 1386 */ 1387 goto badsyn; 1388 } else if (tiflags & TH_ACK) { 1389 so = syn_cache_get(&src.sa, &dst.sa, 1390 th, toff, tlen, so, m); 1391 if (so == NULL) { 1392 /* 1393 * We don't have a SYN for 1394 * this ACK; send an RST. 1395 */ 1396 goto badsyn; 1397 } else if (so == 1398 (struct socket *)(-1)) { 1399 /* 1400 * We were unable to create 1401 * the connection. If the 1402 * 3-way handshake was 1403 * completed, and RST has 1404 * been sent to the peer. 1405 * Since the mbuf might be 1406 * in use for the reply, 1407 * do not free it. 1408 */ 1409 m = NULL; 1410 } else { 1411 /* 1412 * We have created a 1413 * full-blown connection. 1414 */ 1415 tp = NULL; 1416 inp = NULL; 1417 #ifdef INET6 1418 in6p = NULL; 1419 #endif 1420 switch (so->so_proto->pr_domain->dom_family) { 1421 #ifdef INET 1422 case AF_INET: 1423 inp = sotoinpcb(so); 1424 tp = intotcpcb(inp); 1425 break; 1426 #endif 1427 #ifdef INET6 1428 case AF_INET6: 1429 in6p = sotoin6pcb(so); 1430 tp = in6totcpcb(in6p); 1431 break; 1432 #endif 1433 } 1434 if (tp == NULL) 1435 goto badsyn; /*XXX*/ 1436 tiwin <<= tp->snd_scale; 1437 goto after_listen; 1438 } 1439 } else { 1440 /* 1441 * None of RST, SYN or ACK was set. 1442 * This is an invalid packet for a 1443 * TCB in LISTEN state. Send a RST. 1444 */ 1445 goto badsyn; 1446 } 1447 } else { 1448 /* 1449 * Received a SYN. 1450 * 1451 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1452 */ 1453 if (m->m_flags & (M_BCAST|M_MCAST)) 1454 goto drop; 1455 1456 switch (af) { 1457 #ifdef INET6 1458 case AF_INET6: 1459 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1460 goto drop; 1461 break; 1462 #endif /* INET6 */ 1463 case AF_INET: 1464 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1465 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1466 goto drop; 1467 break; 1468 } 1469 1470 #ifdef INET6 1471 /* 1472 * If deprecated address is forbidden, we do 1473 * not accept SYN to deprecated interface 1474 * address to prevent any new inbound 1475 * connection from getting established. 1476 * When we do not accept SYN, we send a TCP 1477 * RST, with deprecated source address (instead 1478 * of dropping it). We compromise it as it is 1479 * much better for peer to send a RST, and 1480 * RST will be the final packet for the 1481 * exchange. 1482 * 1483 * If we do not forbid deprecated addresses, we 1484 * accept the SYN packet. RFC2462 does not 1485 * suggest dropping SYN in this case. 1486 * If we decipher RFC2462 5.5.4, it says like 1487 * this: 1488 * 1. use of deprecated addr with existing 1489 * communication is okay - "SHOULD continue 1490 * to be used" 1491 * 2. use of it with new communication: 1492 * (2a) "SHOULD NOT be used if alternate 1493 * address with sufficient scope is 1494 * available" 1495 * (2b) nothing mentioned otherwise. 1496 * Here we fall into (2b) case as we have no 1497 * choice in our source address selection - we 1498 * must obey the peer. 1499 * 1500 * The wording in RFC2462 is confusing, and 1501 * there are multiple description text for 1502 * deprecated address handling - worse, they 1503 * are not exactly the same. I believe 5.5.4 1504 * is the best one, so we follow 5.5.4. 1505 */ 1506 if (af == AF_INET6 && !ip6_use_deprecated) { 1507 struct in6_ifaddr *ia6; 1508 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1509 &ip6->ip6_dst)) && 1510 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1511 tp = NULL; 1512 goto dropwithreset; 1513 } 1514 } 1515 #endif 1516 1517 #ifdef IPSEC 1518 switch (af) { 1519 #ifdef INET 1520 case AF_INET: 1521 if (ipsec4_in_reject_so(m, so)) { 1522 ipsecstat.in_polvio++; 1523 tp = NULL; 1524 goto dropwithreset; 1525 } 1526 break; 1527 #endif 1528 #ifdef INET6 1529 case AF_INET6: 1530 if (ipsec6_in_reject_so(m, so)) { 1531 ipsec6stat.in_polvio++; 1532 tp = NULL; 1533 goto dropwithreset; 1534 } 1535 break; 1536 #endif 1537 } 1538 #endif 1539 1540 /* 1541 * LISTEN socket received a SYN 1542 * from itself? This can't possibly 1543 * be valid; drop the packet. 1544 */ 1545 if (th->th_sport == th->th_dport) { 1546 int i; 1547 1548 switch (af) { 1549 #ifdef INET 1550 case AF_INET: 1551 i = in_hosteq(ip->ip_src, ip->ip_dst); 1552 break; 1553 #endif 1554 #ifdef INET6 1555 case AF_INET6: 1556 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1557 break; 1558 #endif 1559 default: 1560 i = 1; 1561 } 1562 if (i) { 1563 tcpstat.tcps_badsyn++; 1564 goto drop; 1565 } 1566 } 1567 1568 /* 1569 * SYN looks ok; create compressed TCP 1570 * state for it. 1571 */ 1572 if (so->so_qlen <= so->so_qlimit && 1573 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1574 so, m, optp, optlen, &opti)) 1575 m = NULL; 1576 } 1577 goto drop; 1578 } 1579 } 1580 1581 after_listen: 1582 #ifdef DIAGNOSTIC 1583 /* 1584 * Should not happen now that all embryonic connections 1585 * are handled with compressed state. 1586 */ 1587 if (tp->t_state == TCPS_LISTEN) 1588 panic("tcp_input: TCPS_LISTEN"); 1589 #endif 1590 1591 /* 1592 * Segment received on connection. 1593 * Reset idle time and keep-alive timer. 1594 */ 1595 tp->t_rcvtime = tcp_now; 1596 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1597 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1598 1599 /* 1600 * Process options. 1601 */ 1602 #ifdef TCP_SIGNATURE 1603 if (optp || (tp->t_flags & TF_SIGNATURE)) 1604 #else 1605 if (optp) 1606 #endif 1607 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1608 goto drop; 1609 1610 if (TCP_SACK_ENABLED(tp)) { 1611 tcp_del_sackholes(tp, th); 1612 } 1613 1614 if (TCP_ECN_ALLOWED(tp)) { 1615 switch (iptos & IPTOS_ECN_MASK) { 1616 case IPTOS_ECN_CE: 1617 tp->t_flags |= TF_ECN_SND_ECE; 1618 tcpstat.tcps_ecn_ce++; 1619 break; 1620 case IPTOS_ECN_ECT0: 1621 tcpstat.tcps_ecn_ect++; 1622 break; 1623 case IPTOS_ECN_ECT1: 1624 /* XXX: ignore for now -- rpaulo */ 1625 break; 1626 } 1627 1628 if (tiflags & TH_CWR) 1629 tp->t_flags &= ~TF_ECN_SND_ECE; 1630 1631 /* 1632 * Congestion experienced. 1633 * Ignore if we are already trying to recover. 1634 */ 1635 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1636 tp->t_congctl->cong_exp(tp); 1637 } 1638 1639 if (opti.ts_present && opti.ts_ecr) { 1640 /* 1641 * Calculate the RTT from the returned time stamp and the 1642 * connection's time base. If the time stamp is later than 1643 * the current time, or is extremely old, fall back to non-1323 1644 * RTT calculation. Since ts_ecr is unsigned, we can test both 1645 * at the same time. 1646 */ 1647 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1648 if (ts_rtt > TCP_PAWS_IDLE) 1649 ts_rtt = 0; 1650 } else { 1651 ts_rtt = 0; 1652 } 1653 1654 /* 1655 * Header prediction: check for the two common cases 1656 * of a uni-directional data xfer. If the packet has 1657 * no control flags, is in-sequence, the window didn't 1658 * change and we're not retransmitting, it's a 1659 * candidate. If the length is zero and the ack moved 1660 * forward, we're the sender side of the xfer. Just 1661 * free the data acked & wake any higher level process 1662 * that was blocked waiting for space. If the length 1663 * is non-zero and the ack didn't move, we're the 1664 * receiver side. If we're getting packets in-order 1665 * (the reassembly queue is empty), add the data to 1666 * the socket buffer and note that we need a delayed ack. 1667 */ 1668 if (tp->t_state == TCPS_ESTABLISHED && 1669 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1670 == TH_ACK && 1671 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1672 th->th_seq == tp->rcv_nxt && 1673 tiwin && tiwin == tp->snd_wnd && 1674 tp->snd_nxt == tp->snd_max) { 1675 1676 /* 1677 * If last ACK falls within this segment's sequence numbers, 1678 * record the timestamp. 1679 * NOTE: 1680 * 1) That the test incorporates suggestions from the latest 1681 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 1682 * 2) That updating only on newer timestamps interferes with 1683 * our earlier PAWS tests, so this check should be solely 1684 * predicated on the sequence space of this segment. 1685 * 3) That we modify the segment boundary check to be 1686 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1687 * instead of RFC1323's 1688 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1689 * This modified check allows us to overcome RFC1323's 1690 * limitations as described in Stevens TCP/IP Illustrated 1691 * Vol. 2 p.869. In such cases, we can still calculate the 1692 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1693 */ 1694 if (opti.ts_present && 1695 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1696 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1697 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1698 tp->ts_recent_age = tcp_now; 1699 tp->ts_recent = opti.ts_val; 1700 } 1701 1702 if (tlen == 0) { 1703 /* Ack prediction. */ 1704 if (SEQ_GT(th->th_ack, tp->snd_una) && 1705 SEQ_LEQ(th->th_ack, tp->snd_max) && 1706 tp->snd_cwnd >= tp->snd_wnd && 1707 tp->t_partialacks < 0) { 1708 /* 1709 * this is a pure ack for outstanding data. 1710 */ 1711 ++tcpstat.tcps_predack; 1712 if (ts_rtt) 1713 tcp_xmit_timer(tp, ts_rtt); 1714 else if (tp->t_rtttime && 1715 SEQ_GT(th->th_ack, tp->t_rtseq)) 1716 tcp_xmit_timer(tp, 1717 tcp_now - tp->t_rtttime); 1718 acked = th->th_ack - tp->snd_una; 1719 tcpstat.tcps_rcvackpack++; 1720 tcpstat.tcps_rcvackbyte += acked; 1721 ND6_HINT(tp); 1722 1723 if (acked > (tp->t_lastoff - tp->t_inoff)) 1724 tp->t_lastm = NULL; 1725 sbdrop(&so->so_snd, acked); 1726 tp->t_lastoff -= acked; 1727 1728 ICMP_CHECK(tp, th, acked); 1729 1730 tp->snd_una = th->th_ack; 1731 tp->snd_fack = tp->snd_una; 1732 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1733 tp->snd_high = tp->snd_una; 1734 m_freem(m); 1735 1736 /* 1737 * If all outstanding data are acked, stop 1738 * retransmit timer, otherwise restart timer 1739 * using current (possibly backed-off) value. 1740 * If process is waiting for space, 1741 * wakeup/selwakeup/signal. If data 1742 * are ready to send, let tcp_output 1743 * decide between more output or persist. 1744 */ 1745 if (tp->snd_una == tp->snd_max) 1746 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1747 else if (TCP_TIMER_ISARMED(tp, 1748 TCPT_PERSIST) == 0) 1749 TCP_TIMER_ARM(tp, TCPT_REXMT, 1750 tp->t_rxtcur); 1751 1752 sowwakeup(so); 1753 if (so->so_snd.sb_cc) 1754 (void) tcp_output(tp); 1755 if (tcp_saveti) 1756 m_freem(tcp_saveti); 1757 return; 1758 } 1759 } else if (th->th_ack == tp->snd_una && 1760 TAILQ_FIRST(&tp->segq) == NULL && 1761 tlen <= sbspace(&so->so_rcv)) { 1762 /* 1763 * this is a pure, in-sequence data packet 1764 * with nothing on the reassembly queue and 1765 * we have enough buffer space to take it. 1766 */ 1767 ++tcpstat.tcps_preddat; 1768 tp->rcv_nxt += tlen; 1769 tcpstat.tcps_rcvpack++; 1770 tcpstat.tcps_rcvbyte += tlen; 1771 ND6_HINT(tp); 1772 /* 1773 * Drop TCP, IP headers and TCP options then add data 1774 * to socket buffer. 1775 */ 1776 if (so->so_state & SS_CANTRCVMORE) 1777 m_freem(m); 1778 else { 1779 m_adj(m, toff + off); 1780 sbappendstream(&so->so_rcv, m); 1781 } 1782 sorwakeup(so); 1783 TCP_SETUP_ACK(tp, th); 1784 if (tp->t_flags & TF_ACKNOW) 1785 (void) tcp_output(tp); 1786 if (tcp_saveti) 1787 m_freem(tcp_saveti); 1788 return; 1789 } 1790 } 1791 1792 /* 1793 * Compute mbuf offset to TCP data segment. 1794 */ 1795 hdroptlen = toff + off; 1796 1797 /* 1798 * Calculate amount of space in receive window, 1799 * and then do TCP input processing. 1800 * Receive window is amount of space in rcv queue, 1801 * but not less than advertised window. 1802 */ 1803 { int win; 1804 1805 win = sbspace(&so->so_rcv); 1806 if (win < 0) 1807 win = 0; 1808 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1809 } 1810 1811 switch (tp->t_state) { 1812 /* 1813 * If the state is SYN_SENT: 1814 * if seg contains an ACK, but not for our SYN, drop the input. 1815 * if seg contains a RST, then drop the connection. 1816 * if seg does not contain SYN, then drop it. 1817 * Otherwise this is an acceptable SYN segment 1818 * initialize tp->rcv_nxt and tp->irs 1819 * if seg contains ack then advance tp->snd_una 1820 * if seg contains a ECE and ECN support is enabled, the stream 1821 * is ECN capable. 1822 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1823 * arrange for segment to be acked (eventually) 1824 * continue processing rest of data/controls, beginning with URG 1825 */ 1826 case TCPS_SYN_SENT: 1827 if ((tiflags & TH_ACK) && 1828 (SEQ_LEQ(th->th_ack, tp->iss) || 1829 SEQ_GT(th->th_ack, tp->snd_max))) 1830 goto dropwithreset; 1831 if (tiflags & TH_RST) { 1832 if (tiflags & TH_ACK) 1833 tp = tcp_drop(tp, ECONNREFUSED); 1834 goto drop; 1835 } 1836 if ((tiflags & TH_SYN) == 0) 1837 goto drop; 1838 if (tiflags & TH_ACK) { 1839 tp->snd_una = th->th_ack; 1840 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1841 tp->snd_nxt = tp->snd_una; 1842 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1843 tp->snd_high = tp->snd_una; 1844 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1845 1846 if ((tiflags & TH_ECE) && tcp_do_ecn) { 1847 tp->t_flags |= TF_ECN_PERMIT; 1848 tcpstat.tcps_ecn_shs++; 1849 } 1850 1851 } 1852 tp->irs = th->th_seq; 1853 tcp_rcvseqinit(tp); 1854 tp->t_flags |= TF_ACKNOW; 1855 tcp_mss_from_peer(tp, opti.maxseg); 1856 1857 /* 1858 * Initialize the initial congestion window. If we 1859 * had to retransmit the SYN, we must initialize cwnd 1860 * to 1 segment (i.e. the Loss Window). 1861 */ 1862 if (tp->t_flags & TF_SYN_REXMT) 1863 tp->snd_cwnd = tp->t_peermss; 1864 else { 1865 int ss = tcp_init_win; 1866 #ifdef INET 1867 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1868 ss = tcp_init_win_local; 1869 #endif 1870 #ifdef INET6 1871 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1872 ss = tcp_init_win_local; 1873 #endif 1874 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1875 } 1876 1877 tcp_rmx_rtt(tp); 1878 if (tiflags & TH_ACK) { 1879 tcpstat.tcps_connects++; 1880 soisconnected(so); 1881 tcp_established(tp); 1882 /* Do window scaling on this connection? */ 1883 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1884 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1885 tp->snd_scale = tp->requested_s_scale; 1886 tp->rcv_scale = tp->request_r_scale; 1887 } 1888 TCP_REASS_LOCK(tp); 1889 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1890 TCP_REASS_UNLOCK(tp); 1891 /* 1892 * if we didn't have to retransmit the SYN, 1893 * use its rtt as our initial srtt & rtt var. 1894 */ 1895 if (tp->t_rtttime) 1896 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1897 } else 1898 tp->t_state = TCPS_SYN_RECEIVED; 1899 1900 /* 1901 * Advance th->th_seq to correspond to first data byte. 1902 * If data, trim to stay within window, 1903 * dropping FIN if necessary. 1904 */ 1905 th->th_seq++; 1906 if (tlen > tp->rcv_wnd) { 1907 todrop = tlen - tp->rcv_wnd; 1908 m_adj(m, -todrop); 1909 tlen = tp->rcv_wnd; 1910 tiflags &= ~TH_FIN; 1911 tcpstat.tcps_rcvpackafterwin++; 1912 tcpstat.tcps_rcvbyteafterwin += todrop; 1913 } 1914 tp->snd_wl1 = th->th_seq - 1; 1915 tp->rcv_up = th->th_seq; 1916 goto step6; 1917 1918 /* 1919 * If the state is SYN_RECEIVED: 1920 * If seg contains an ACK, but not for our SYN, drop the input 1921 * and generate an RST. See page 36, rfc793 1922 */ 1923 case TCPS_SYN_RECEIVED: 1924 if ((tiflags & TH_ACK) && 1925 (SEQ_LEQ(th->th_ack, tp->iss) || 1926 SEQ_GT(th->th_ack, tp->snd_max))) 1927 goto dropwithreset; 1928 break; 1929 } 1930 1931 /* 1932 * States other than LISTEN or SYN_SENT. 1933 * First check timestamp, if present. 1934 * Then check that at least some bytes of segment are within 1935 * receive window. If segment begins before rcv_nxt, 1936 * drop leading data (and SYN); if nothing left, just ack. 1937 * 1938 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1939 * and it's less than ts_recent, drop it. 1940 */ 1941 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1942 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1943 1944 /* Check to see if ts_recent is over 24 days old. */ 1945 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 1946 /* 1947 * Invalidate ts_recent. If this segment updates 1948 * ts_recent, the age will be reset later and ts_recent 1949 * will get a valid value. If it does not, setting 1950 * ts_recent to zero will at least satisfy the 1951 * requirement that zero be placed in the timestamp 1952 * echo reply when ts_recent isn't valid. The 1953 * age isn't reset until we get a valid ts_recent 1954 * because we don't want out-of-order segments to be 1955 * dropped when ts_recent is old. 1956 */ 1957 tp->ts_recent = 0; 1958 } else { 1959 tcpstat.tcps_rcvduppack++; 1960 tcpstat.tcps_rcvdupbyte += tlen; 1961 tcpstat.tcps_pawsdrop++; 1962 tcp_new_dsack(tp, th->th_seq, tlen); 1963 goto dropafterack; 1964 } 1965 } 1966 1967 todrop = tp->rcv_nxt - th->th_seq; 1968 dupseg = FALSE; 1969 if (todrop > 0) { 1970 if (tiflags & TH_SYN) { 1971 tiflags &= ~TH_SYN; 1972 th->th_seq++; 1973 if (th->th_urp > 1) 1974 th->th_urp--; 1975 else { 1976 tiflags &= ~TH_URG; 1977 th->th_urp = 0; 1978 } 1979 todrop--; 1980 } 1981 if (todrop > tlen || 1982 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1983 /* 1984 * Any valid FIN or RST must be to the left of the 1985 * window. At this point the FIN or RST must be a 1986 * duplicate or out of sequence; drop it. 1987 */ 1988 if (tiflags & TH_RST) 1989 goto drop; 1990 tiflags &= ~(TH_FIN|TH_RST); 1991 /* 1992 * Send an ACK to resynchronize and drop any data. 1993 * But keep on processing for RST or ACK. 1994 */ 1995 tp->t_flags |= TF_ACKNOW; 1996 todrop = tlen; 1997 dupseg = TRUE; 1998 tcpstat.tcps_rcvdupbyte += todrop; 1999 tcpstat.tcps_rcvduppack++; 2000 } else if ((tiflags & TH_RST) && 2001 th->th_seq != tp->last_ack_sent) { 2002 /* 2003 * Test for reset before adjusting the sequence 2004 * number for overlapping data. 2005 */ 2006 goto dropafterack_ratelim; 2007 } else { 2008 tcpstat.tcps_rcvpartduppack++; 2009 tcpstat.tcps_rcvpartdupbyte += todrop; 2010 } 2011 tcp_new_dsack(tp, th->th_seq, todrop); 2012 hdroptlen += todrop; /*drop from head afterwards*/ 2013 th->th_seq += todrop; 2014 tlen -= todrop; 2015 if (th->th_urp > todrop) 2016 th->th_urp -= todrop; 2017 else { 2018 tiflags &= ~TH_URG; 2019 th->th_urp = 0; 2020 } 2021 } 2022 2023 /* 2024 * If new data are received on a connection after the 2025 * user processes are gone, then RST the other end. 2026 */ 2027 if ((so->so_state & SS_NOFDREF) && 2028 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2029 tp = tcp_close(tp); 2030 tcpstat.tcps_rcvafterclose++; 2031 goto dropwithreset; 2032 } 2033 2034 /* 2035 * If segment ends after window, drop trailing data 2036 * (and PUSH and FIN); if nothing left, just ACK. 2037 */ 2038 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 2039 if (todrop > 0) { 2040 tcpstat.tcps_rcvpackafterwin++; 2041 if (todrop >= tlen) { 2042 /* 2043 * The segment actually starts after the window. 2044 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2045 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2046 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2047 */ 2048 tcpstat.tcps_rcvbyteafterwin += tlen; 2049 /* 2050 * If a new connection request is received 2051 * while in TIME_WAIT, drop the old connection 2052 * and start over if the sequence numbers 2053 * are above the previous ones. 2054 * 2055 * NOTE: We will checksum the packet again, and 2056 * so we need to put the header fields back into 2057 * network order! 2058 * XXX This kind of sucks, but we don't expect 2059 * XXX this to happen very often, so maybe it 2060 * XXX doesn't matter so much. 2061 */ 2062 if (tiflags & TH_SYN && 2063 tp->t_state == TCPS_TIME_WAIT && 2064 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2065 tp = tcp_close(tp); 2066 TCP_FIELDS_TO_NET(th); 2067 goto findpcb; 2068 } 2069 /* 2070 * If window is closed can only take segments at 2071 * window edge, and have to drop data and PUSH from 2072 * incoming segments. Continue processing, but 2073 * remember to ack. Otherwise, drop segment 2074 * and (if not RST) ack. 2075 */ 2076 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2077 tp->t_flags |= TF_ACKNOW; 2078 tcpstat.tcps_rcvwinprobe++; 2079 } else 2080 goto dropafterack; 2081 } else 2082 tcpstat.tcps_rcvbyteafterwin += todrop; 2083 m_adj(m, -todrop); 2084 tlen -= todrop; 2085 tiflags &= ~(TH_PUSH|TH_FIN); 2086 } 2087 2088 /* 2089 * If last ACK falls within this segment's sequence numbers, 2090 * and the timestamp is newer, record it. 2091 */ 2092 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 2093 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2094 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 2095 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2096 tp->ts_recent_age = tcp_now; 2097 tp->ts_recent = opti.ts_val; 2098 } 2099 2100 /* 2101 * If the RST bit is set examine the state: 2102 * SYN_RECEIVED STATE: 2103 * If passive open, return to LISTEN state. 2104 * If active open, inform user that connection was refused. 2105 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 2106 * Inform user that connection was reset, and close tcb. 2107 * CLOSING, LAST_ACK, TIME_WAIT STATES 2108 * Close the tcb. 2109 */ 2110 if (tiflags & TH_RST) { 2111 if (th->th_seq != tp->last_ack_sent) 2112 goto dropafterack_ratelim; 2113 2114 switch (tp->t_state) { 2115 case TCPS_SYN_RECEIVED: 2116 so->so_error = ECONNREFUSED; 2117 goto close; 2118 2119 case TCPS_ESTABLISHED: 2120 case TCPS_FIN_WAIT_1: 2121 case TCPS_FIN_WAIT_2: 2122 case TCPS_CLOSE_WAIT: 2123 so->so_error = ECONNRESET; 2124 close: 2125 tp->t_state = TCPS_CLOSED; 2126 tcpstat.tcps_drops++; 2127 tp = tcp_close(tp); 2128 goto drop; 2129 2130 case TCPS_CLOSING: 2131 case TCPS_LAST_ACK: 2132 case TCPS_TIME_WAIT: 2133 tp = tcp_close(tp); 2134 goto drop; 2135 } 2136 } 2137 2138 /* 2139 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2140 * we must be in a synchronized state. RFC791 states (under RST 2141 * generation) that any unacceptable segment (an out-of-order SYN 2142 * qualifies) received in a synchronized state must elicit only an 2143 * empty acknowledgment segment ... and the connection remains in 2144 * the same state. 2145 */ 2146 if (tiflags & TH_SYN) { 2147 if (tp->rcv_nxt == th->th_seq) { 2148 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2149 TH_ACK); 2150 if (tcp_saveti) 2151 m_freem(tcp_saveti); 2152 return; 2153 } 2154 2155 goto dropafterack_ratelim; 2156 } 2157 2158 /* 2159 * If the ACK bit is off we drop the segment and return. 2160 */ 2161 if ((tiflags & TH_ACK) == 0) { 2162 if (tp->t_flags & TF_ACKNOW) 2163 goto dropafterack; 2164 else 2165 goto drop; 2166 } 2167 2168 /* 2169 * Ack processing. 2170 */ 2171 switch (tp->t_state) { 2172 2173 /* 2174 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2175 * ESTABLISHED state and continue processing, otherwise 2176 * send an RST. 2177 */ 2178 case TCPS_SYN_RECEIVED: 2179 if (SEQ_GT(tp->snd_una, th->th_ack) || 2180 SEQ_GT(th->th_ack, tp->snd_max)) 2181 goto dropwithreset; 2182 tcpstat.tcps_connects++; 2183 soisconnected(so); 2184 tcp_established(tp); 2185 /* Do window scaling? */ 2186 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2187 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2188 tp->snd_scale = tp->requested_s_scale; 2189 tp->rcv_scale = tp->request_r_scale; 2190 } 2191 TCP_REASS_LOCK(tp); 2192 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2193 TCP_REASS_UNLOCK(tp); 2194 tp->snd_wl1 = th->th_seq - 1; 2195 /* fall into ... */ 2196 2197 /* 2198 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2199 * ACKs. If the ack is in the range 2200 * tp->snd_una < th->th_ack <= tp->snd_max 2201 * then advance tp->snd_una to th->th_ack and drop 2202 * data from the retransmission queue. If this ACK reflects 2203 * more up to date window information we update our window information. 2204 */ 2205 case TCPS_ESTABLISHED: 2206 case TCPS_FIN_WAIT_1: 2207 case TCPS_FIN_WAIT_2: 2208 case TCPS_CLOSE_WAIT: 2209 case TCPS_CLOSING: 2210 case TCPS_LAST_ACK: 2211 case TCPS_TIME_WAIT: 2212 2213 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2214 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2215 tcpstat.tcps_rcvdupack++; 2216 /* 2217 * If we have outstanding data (other than 2218 * a window probe), this is a completely 2219 * duplicate ack (ie, window info didn't 2220 * change), the ack is the biggest we've 2221 * seen and we've seen exactly our rexmt 2222 * threshhold of them, assume a packet 2223 * has been dropped and retransmit it. 2224 * Kludge snd_nxt & the congestion 2225 * window so we send only this one 2226 * packet. 2227 */ 2228 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2229 th->th_ack != tp->snd_una) 2230 tp->t_dupacks = 0; 2231 else if (tp->t_partialacks < 0 && 2232 ((!TCP_SACK_ENABLED(tp) && 2233 ++tp->t_dupacks == tcprexmtthresh) || 2234 TCP_FACK_FASTRECOV(tp))) { 2235 /* 2236 * Do the fast retransmit, and adjust 2237 * congestion control paramenters. 2238 */ 2239 if (tp->t_congctl->fast_retransmit(tp, th)) { 2240 /* False fast retransmit */ 2241 break; 2242 } else 2243 goto drop; 2244 } else if (tp->t_dupacks > tcprexmtthresh) { 2245 tp->snd_cwnd += tp->t_segsz; 2246 (void) tcp_output(tp); 2247 goto drop; 2248 } 2249 } else { 2250 /* 2251 * If the ack appears to be very old, only 2252 * allow data that is in-sequence. This 2253 * makes it somewhat more difficult to insert 2254 * forged data by guessing sequence numbers. 2255 * Sent an ack to try to update the send 2256 * sequence number on the other side. 2257 */ 2258 if (tlen && th->th_seq != tp->rcv_nxt && 2259 SEQ_LT(th->th_ack, 2260 tp->snd_una - tp->max_sndwnd)) 2261 goto dropafterack; 2262 } 2263 break; 2264 } 2265 /* 2266 * If the congestion window was inflated to account 2267 * for the other side's cached packets, retract it. 2268 */ 2269 /* XXX: make SACK have his own congestion control 2270 * struct -- rpaulo */ 2271 if (TCP_SACK_ENABLED(tp)) 2272 tcp_sack_newack(tp, th); 2273 else 2274 tp->t_congctl->fast_retransmit_newack(tp, th); 2275 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2276 tcpstat.tcps_rcvacktoomuch++; 2277 goto dropafterack; 2278 } 2279 acked = th->th_ack - tp->snd_una; 2280 tcpstat.tcps_rcvackpack++; 2281 tcpstat.tcps_rcvackbyte += acked; 2282 2283 /* 2284 * If we have a timestamp reply, update smoothed 2285 * round trip time. If no timestamp is present but 2286 * transmit timer is running and timed sequence 2287 * number was acked, update smoothed round trip time. 2288 * Since we now have an rtt measurement, cancel the 2289 * timer backoff (cf., Phil Karn's retransmit alg.). 2290 * Recompute the initial retransmit timer. 2291 */ 2292 if (ts_rtt) 2293 tcp_xmit_timer(tp, ts_rtt); 2294 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2295 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2296 2297 /* 2298 * If all outstanding data is acked, stop retransmit 2299 * timer and remember to restart (more output or persist). 2300 * If there is more data to be acked, restart retransmit 2301 * timer, using current (possibly backed-off) value. 2302 */ 2303 if (th->th_ack == tp->snd_max) { 2304 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2305 needoutput = 1; 2306 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2307 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2308 2309 /* 2310 * New data has been acked, adjust the congestion window. 2311 */ 2312 tp->t_congctl->newack(tp, th); 2313 2314 ND6_HINT(tp); 2315 if (acked > so->so_snd.sb_cc) { 2316 tp->snd_wnd -= so->so_snd.sb_cc; 2317 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2318 ourfinisacked = 1; 2319 } else { 2320 if (acked > (tp->t_lastoff - tp->t_inoff)) 2321 tp->t_lastm = NULL; 2322 sbdrop(&so->so_snd, acked); 2323 tp->t_lastoff -= acked; 2324 tp->snd_wnd -= acked; 2325 ourfinisacked = 0; 2326 } 2327 sowwakeup(so); 2328 2329 ICMP_CHECK(tp, th, acked); 2330 2331 tp->snd_una = th->th_ack; 2332 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2333 tp->snd_fack = tp->snd_una; 2334 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2335 tp->snd_nxt = tp->snd_una; 2336 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2337 tp->snd_high = tp->snd_una; 2338 2339 switch (tp->t_state) { 2340 2341 /* 2342 * In FIN_WAIT_1 STATE in addition to the processing 2343 * for the ESTABLISHED state if our FIN is now acknowledged 2344 * then enter FIN_WAIT_2. 2345 */ 2346 case TCPS_FIN_WAIT_1: 2347 if (ourfinisacked) { 2348 /* 2349 * If we can't receive any more 2350 * data, then closing user can proceed. 2351 * Starting the timer is contrary to the 2352 * specification, but if we don't get a FIN 2353 * we'll hang forever. 2354 */ 2355 if (so->so_state & SS_CANTRCVMORE) { 2356 soisdisconnected(so); 2357 if (tcp_maxidle > 0) 2358 TCP_TIMER_ARM(tp, TCPT_2MSL, 2359 tcp_maxidle); 2360 } 2361 tp->t_state = TCPS_FIN_WAIT_2; 2362 } 2363 break; 2364 2365 /* 2366 * In CLOSING STATE in addition to the processing for 2367 * the ESTABLISHED state if the ACK acknowledges our FIN 2368 * then enter the TIME-WAIT state, otherwise ignore 2369 * the segment. 2370 */ 2371 case TCPS_CLOSING: 2372 if (ourfinisacked) { 2373 tp->t_state = TCPS_TIME_WAIT; 2374 tcp_canceltimers(tp); 2375 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2376 soisdisconnected(so); 2377 } 2378 break; 2379 2380 /* 2381 * In LAST_ACK, we may still be waiting for data to drain 2382 * and/or to be acked, as well as for the ack of our FIN. 2383 * If our FIN is now acknowledged, delete the TCB, 2384 * enter the closed state and return. 2385 */ 2386 case TCPS_LAST_ACK: 2387 if (ourfinisacked) { 2388 tp = tcp_close(tp); 2389 goto drop; 2390 } 2391 break; 2392 2393 /* 2394 * In TIME_WAIT state the only thing that should arrive 2395 * is a retransmission of the remote FIN. Acknowledge 2396 * it and restart the finack timer. 2397 */ 2398 case TCPS_TIME_WAIT: 2399 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2400 goto dropafterack; 2401 } 2402 } 2403 2404 step6: 2405 /* 2406 * Update window information. 2407 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2408 */ 2409 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2410 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2411 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2412 /* keep track of pure window updates */ 2413 if (tlen == 0 && 2414 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2415 tcpstat.tcps_rcvwinupd++; 2416 tp->snd_wnd = tiwin; 2417 tp->snd_wl1 = th->th_seq; 2418 tp->snd_wl2 = th->th_ack; 2419 if (tp->snd_wnd > tp->max_sndwnd) 2420 tp->max_sndwnd = tp->snd_wnd; 2421 needoutput = 1; 2422 } 2423 2424 /* 2425 * Process segments with URG. 2426 */ 2427 if ((tiflags & TH_URG) && th->th_urp && 2428 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2429 /* 2430 * This is a kludge, but if we receive and accept 2431 * random urgent pointers, we'll crash in 2432 * soreceive. It's hard to imagine someone 2433 * actually wanting to send this much urgent data. 2434 */ 2435 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2436 th->th_urp = 0; /* XXX */ 2437 tiflags &= ~TH_URG; /* XXX */ 2438 goto dodata; /* XXX */ 2439 } 2440 /* 2441 * If this segment advances the known urgent pointer, 2442 * then mark the data stream. This should not happen 2443 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2444 * a FIN has been received from the remote side. 2445 * In these states we ignore the URG. 2446 * 2447 * According to RFC961 (Assigned Protocols), 2448 * the urgent pointer points to the last octet 2449 * of urgent data. We continue, however, 2450 * to consider it to indicate the first octet 2451 * of data past the urgent section as the original 2452 * spec states (in one of two places). 2453 */ 2454 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2455 tp->rcv_up = th->th_seq + th->th_urp; 2456 so->so_oobmark = so->so_rcv.sb_cc + 2457 (tp->rcv_up - tp->rcv_nxt) - 1; 2458 if (so->so_oobmark == 0) 2459 so->so_state |= SS_RCVATMARK; 2460 sohasoutofband(so); 2461 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2462 } 2463 /* 2464 * Remove out of band data so doesn't get presented to user. 2465 * This can happen independent of advancing the URG pointer, 2466 * but if two URG's are pending at once, some out-of-band 2467 * data may creep in... ick. 2468 */ 2469 if (th->th_urp <= (u_int16_t) tlen 2470 #ifdef SO_OOBINLINE 2471 && (so->so_options & SO_OOBINLINE) == 0 2472 #endif 2473 ) 2474 tcp_pulloutofband(so, th, m, hdroptlen); 2475 } else 2476 /* 2477 * If no out of band data is expected, 2478 * pull receive urgent pointer along 2479 * with the receive window. 2480 */ 2481 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2482 tp->rcv_up = tp->rcv_nxt; 2483 dodata: /* XXX */ 2484 2485 /* 2486 * Process the segment text, merging it into the TCP sequencing queue, 2487 * and arranging for acknowledgement of receipt if necessary. 2488 * This process logically involves adjusting tp->rcv_wnd as data 2489 * is presented to the user (this happens in tcp_usrreq.c, 2490 * case PRU_RCVD). If a FIN has already been received on this 2491 * connection then we just ignore the text. 2492 */ 2493 if ((tlen || (tiflags & TH_FIN)) && 2494 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2495 /* 2496 * Insert segment ti into reassembly queue of tcp with 2497 * control block tp. Return TH_FIN if reassembly now includes 2498 * a segment with FIN. The macro form does the common case 2499 * inline (segment is the next to be received on an 2500 * established connection, and the queue is empty), 2501 * avoiding linkage into and removal from the queue and 2502 * repetition of various conversions. 2503 * Set DELACK for segments received in order, but ack 2504 * immediately when segments are out of order 2505 * (so fast retransmit can work). 2506 */ 2507 /* NOTE: this was TCP_REASS() macro, but used only once */ 2508 TCP_REASS_LOCK(tp); 2509 if (th->th_seq == tp->rcv_nxt && 2510 TAILQ_FIRST(&tp->segq) == NULL && 2511 tp->t_state == TCPS_ESTABLISHED) { 2512 TCP_SETUP_ACK(tp, th); 2513 tp->rcv_nxt += tlen; 2514 tiflags = th->th_flags & TH_FIN; 2515 tcpstat.tcps_rcvpack++; 2516 tcpstat.tcps_rcvbyte += tlen; 2517 ND6_HINT(tp); 2518 if (so->so_state & SS_CANTRCVMORE) 2519 m_freem(m); 2520 else { 2521 m_adj(m, hdroptlen); 2522 sbappendstream(&(so)->so_rcv, m); 2523 } 2524 sorwakeup(so); 2525 } else { 2526 m_adj(m, hdroptlen); 2527 tiflags = tcp_reass(tp, th, m, &tlen); 2528 tp->t_flags |= TF_ACKNOW; 2529 } 2530 TCP_REASS_UNLOCK(tp); 2531 2532 /* 2533 * Note the amount of data that peer has sent into 2534 * our window, in order to estimate the sender's 2535 * buffer size. 2536 */ 2537 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2538 } else { 2539 m_freem(m); 2540 m = NULL; 2541 tiflags &= ~TH_FIN; 2542 } 2543 2544 /* 2545 * If FIN is received ACK the FIN and let the user know 2546 * that the connection is closing. Ignore a FIN received before 2547 * the connection is fully established. 2548 */ 2549 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2550 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2551 socantrcvmore(so); 2552 tp->t_flags |= TF_ACKNOW; 2553 tp->rcv_nxt++; 2554 } 2555 switch (tp->t_state) { 2556 2557 /* 2558 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2559 */ 2560 case TCPS_ESTABLISHED: 2561 tp->t_state = TCPS_CLOSE_WAIT; 2562 break; 2563 2564 /* 2565 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2566 * enter the CLOSING state. 2567 */ 2568 case TCPS_FIN_WAIT_1: 2569 tp->t_state = TCPS_CLOSING; 2570 break; 2571 2572 /* 2573 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2574 * starting the time-wait timer, turning off the other 2575 * standard timers. 2576 */ 2577 case TCPS_FIN_WAIT_2: 2578 tp->t_state = TCPS_TIME_WAIT; 2579 tcp_canceltimers(tp); 2580 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2581 soisdisconnected(so); 2582 break; 2583 2584 /* 2585 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2586 */ 2587 case TCPS_TIME_WAIT: 2588 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2589 break; 2590 } 2591 } 2592 #ifdef TCP_DEBUG 2593 if (so->so_options & SO_DEBUG) 2594 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2595 #endif 2596 2597 /* 2598 * Return any desired output. 2599 */ 2600 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2601 (void) tcp_output(tp); 2602 } 2603 if (tcp_saveti) 2604 m_freem(tcp_saveti); 2605 return; 2606 2607 badsyn: 2608 /* 2609 * Received a bad SYN. Increment counters and dropwithreset. 2610 */ 2611 tcpstat.tcps_badsyn++; 2612 tp = NULL; 2613 goto dropwithreset; 2614 2615 dropafterack: 2616 /* 2617 * Generate an ACK dropping incoming segment if it occupies 2618 * sequence space, where the ACK reflects our state. 2619 */ 2620 if (tiflags & TH_RST) 2621 goto drop; 2622 goto dropafterack2; 2623 2624 dropafterack_ratelim: 2625 /* 2626 * We may want to rate-limit ACKs against SYN/RST attack. 2627 */ 2628 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2629 tcp_ackdrop_ppslim) == 0) { 2630 /* XXX stat */ 2631 goto drop; 2632 } 2633 /* ...fall into dropafterack2... */ 2634 2635 dropafterack2: 2636 m_freem(m); 2637 tp->t_flags |= TF_ACKNOW; 2638 (void) tcp_output(tp); 2639 if (tcp_saveti) 2640 m_freem(tcp_saveti); 2641 return; 2642 2643 dropwithreset_ratelim: 2644 /* 2645 * We may want to rate-limit RSTs in certain situations, 2646 * particularly if we are sending an RST in response to 2647 * an attempt to connect to or otherwise communicate with 2648 * a port for which we have no socket. 2649 */ 2650 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2651 tcp_rst_ppslim) == 0) { 2652 /* XXX stat */ 2653 goto drop; 2654 } 2655 /* ...fall into dropwithreset... */ 2656 2657 dropwithreset: 2658 /* 2659 * Generate a RST, dropping incoming segment. 2660 * Make ACK acceptable to originator of segment. 2661 */ 2662 if (tiflags & TH_RST) 2663 goto drop; 2664 2665 switch (af) { 2666 #ifdef INET6 2667 case AF_INET6: 2668 /* For following calls to tcp_respond */ 2669 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2670 goto drop; 2671 break; 2672 #endif /* INET6 */ 2673 case AF_INET: 2674 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2675 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2676 goto drop; 2677 } 2678 2679 if (tiflags & TH_ACK) 2680 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2681 else { 2682 if (tiflags & TH_SYN) 2683 tlen++; 2684 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2685 TH_RST|TH_ACK); 2686 } 2687 if (tcp_saveti) 2688 m_freem(tcp_saveti); 2689 return; 2690 2691 badcsum: 2692 drop: 2693 /* 2694 * Drop space held by incoming segment and return. 2695 */ 2696 if (tp) { 2697 if (tp->t_inpcb) 2698 so = tp->t_inpcb->inp_socket; 2699 #ifdef INET6 2700 else if (tp->t_in6pcb) 2701 so = tp->t_in6pcb->in6p_socket; 2702 #endif 2703 else 2704 so = NULL; 2705 #ifdef TCP_DEBUG 2706 if (so && (so->so_options & SO_DEBUG) != 0) 2707 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2708 #endif 2709 } 2710 if (tcp_saveti) 2711 m_freem(tcp_saveti); 2712 m_freem(m); 2713 return; 2714 } 2715 2716 #ifdef TCP_SIGNATURE 2717 int 2718 tcp_signature_apply(void *fstate, caddr_t data, u_int len) 2719 { 2720 2721 MD5Update(fstate, (u_char *)data, len); 2722 return (0); 2723 } 2724 2725 struct secasvar * 2726 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 2727 { 2728 struct secasvar *sav; 2729 #ifdef FAST_IPSEC 2730 union sockaddr_union dst; 2731 #endif 2732 struct ip *ip; 2733 struct ip6_hdr *ip6; 2734 2735 ip = mtod(m, struct ip *); 2736 switch (ip->ip_v) { 2737 case 4: 2738 ip = mtod(m, struct ip *); 2739 ip6 = NULL; 2740 break; 2741 case 6: 2742 ip = NULL; 2743 ip6 = mtod(m, struct ip6_hdr *); 2744 break; 2745 default: 2746 return (NULL); 2747 } 2748 2749 #ifdef FAST_IPSEC 2750 /* Extract the destination from the IP header in the mbuf. */ 2751 bzero(&dst, sizeof(union sockaddr_union)); 2752 dst.sa.sa_len = sizeof(struct sockaddr_in); 2753 dst.sa.sa_family = AF_INET; 2754 dst.sin.sin_addr = ip->ip_dst; 2755 2756 /* 2757 * Look up an SADB entry which matches the address of the peer. 2758 */ 2759 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2760 #else 2761 if (ip) 2762 sav = key_allocsa(AF_INET, (caddr_t)&ip->ip_src, 2763 (caddr_t)&ip->ip_dst, IPPROTO_TCP, 2764 htonl(TCP_SIG_SPI), 0, 0); 2765 else 2766 sav = key_allocsa(AF_INET6, (caddr_t)&ip6->ip6_src, 2767 (caddr_t)&ip6->ip6_dst, IPPROTO_TCP, 2768 htonl(TCP_SIG_SPI), 0, 0); 2769 #endif 2770 2771 return (sav); /* freesav must be performed by caller */ 2772 } 2773 2774 int 2775 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 2776 struct secasvar *sav, char *sig) 2777 { 2778 MD5_CTX ctx; 2779 struct ip *ip; 2780 struct ipovly *ipovly; 2781 struct ip6_hdr *ip6; 2782 struct ippseudo ippseudo; 2783 struct ip6_hdr_pseudo ip6pseudo; 2784 struct tcphdr th0; 2785 int l, tcphdrlen; 2786 2787 if (sav == NULL) 2788 return (-1); 2789 2790 tcphdrlen = th->th_off * 4; 2791 2792 switch (mtod(m, struct ip *)->ip_v) { 2793 case 4: 2794 ip = mtod(m, struct ip *); 2795 ip6 = NULL; 2796 break; 2797 case 6: 2798 ip = NULL; 2799 ip6 = mtod(m, struct ip6_hdr *); 2800 break; 2801 default: 2802 return (-1); 2803 } 2804 2805 MD5Init(&ctx); 2806 2807 if (ip) { 2808 memset(&ippseudo, 0, sizeof(ippseudo)); 2809 ipovly = (struct ipovly *)ip; 2810 ippseudo.ippseudo_src = ipovly->ih_src; 2811 ippseudo.ippseudo_dst = ipovly->ih_dst; 2812 ippseudo.ippseudo_pad = 0; 2813 ippseudo.ippseudo_p = IPPROTO_TCP; 2814 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 2815 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 2816 } else { 2817 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 2818 ip6pseudo.ip6ph_src = ip6->ip6_src; 2819 in6_clearscope(&ip6pseudo.ip6ph_src); 2820 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 2821 in6_clearscope(&ip6pseudo.ip6ph_dst); 2822 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 2823 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 2824 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 2825 } 2826 2827 th0 = *th; 2828 th0.th_sum = 0; 2829 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 2830 2831 l = m->m_pkthdr.len - thoff - tcphdrlen; 2832 if (l > 0) 2833 m_apply(m, thoff + tcphdrlen, 2834 m->m_pkthdr.len - thoff - tcphdrlen, 2835 tcp_signature_apply, &ctx); 2836 2837 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2838 MD5Final(sig, &ctx); 2839 2840 return (0); 2841 } 2842 #endif 2843 2844 int 2845 tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcphdr *th, 2846 struct mbuf *m __unused, int toff __unused, struct tcp_opt_info *oi) 2847 { 2848 u_int16_t mss; 2849 int opt, optlen = 0; 2850 #ifdef TCP_SIGNATURE 2851 caddr_t sigp = NULL; 2852 char sigbuf[TCP_SIGLEN]; 2853 struct secasvar *sav = NULL; 2854 #endif 2855 2856 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 2857 opt = cp[0]; 2858 if (opt == TCPOPT_EOL) 2859 break; 2860 if (opt == TCPOPT_NOP) 2861 optlen = 1; 2862 else { 2863 if (cnt < 2) 2864 break; 2865 optlen = cp[1]; 2866 if (optlen < 2 || optlen > cnt) 2867 break; 2868 } 2869 switch (opt) { 2870 2871 default: 2872 continue; 2873 2874 case TCPOPT_MAXSEG: 2875 if (optlen != TCPOLEN_MAXSEG) 2876 continue; 2877 if (!(th->th_flags & TH_SYN)) 2878 continue; 2879 if (TCPS_HAVERCVDSYN(tp->t_state)) 2880 continue; 2881 bcopy(cp + 2, &mss, sizeof(mss)); 2882 oi->maxseg = ntohs(mss); 2883 break; 2884 2885 case TCPOPT_WINDOW: 2886 if (optlen != TCPOLEN_WINDOW) 2887 continue; 2888 if (!(th->th_flags & TH_SYN)) 2889 continue; 2890 if (TCPS_HAVERCVDSYN(tp->t_state)) 2891 continue; 2892 tp->t_flags |= TF_RCVD_SCALE; 2893 tp->requested_s_scale = cp[2]; 2894 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2895 #if 0 /*XXX*/ 2896 char *p; 2897 2898 if (ip) 2899 p = ntohl(ip->ip_src); 2900 #ifdef INET6 2901 else if (ip6) 2902 p = ip6_sprintf(&ip6->ip6_src); 2903 #endif 2904 else 2905 p = "(unknown)"; 2906 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2907 "assuming %d\n", 2908 tp->requested_s_scale, p, 2909 TCP_MAX_WINSHIFT); 2910 #else 2911 log(LOG_ERR, "TCP: invalid wscale %d, " 2912 "assuming %d\n", 2913 tp->requested_s_scale, 2914 TCP_MAX_WINSHIFT); 2915 #endif 2916 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2917 } 2918 break; 2919 2920 case TCPOPT_TIMESTAMP: 2921 if (optlen != TCPOLEN_TIMESTAMP) 2922 continue; 2923 oi->ts_present = 1; 2924 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2925 NTOHL(oi->ts_val); 2926 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2927 NTOHL(oi->ts_ecr); 2928 2929 if (!(th->th_flags & TH_SYN)) 2930 continue; 2931 if (TCPS_HAVERCVDSYN(tp->t_state)) 2932 continue; 2933 /* 2934 * A timestamp received in a SYN makes 2935 * it ok to send timestamp requests and replies. 2936 */ 2937 tp->t_flags |= TF_RCVD_TSTMP; 2938 tp->ts_recent = oi->ts_val; 2939 tp->ts_recent_age = tcp_now; 2940 break; 2941 2942 case TCPOPT_SACK_PERMITTED: 2943 if (optlen != TCPOLEN_SACK_PERMITTED) 2944 continue; 2945 if (!(th->th_flags & TH_SYN)) 2946 continue; 2947 if (TCPS_HAVERCVDSYN(tp->t_state)) 2948 continue; 2949 if (tcp_do_sack) { 2950 tp->t_flags |= TF_SACK_PERMIT; 2951 tp->t_flags |= TF_WILL_SACK; 2952 } 2953 break; 2954 2955 case TCPOPT_SACK: 2956 tcp_sack_option(tp, th, cp, optlen); 2957 break; 2958 #ifdef TCP_SIGNATURE 2959 case TCPOPT_SIGNATURE: 2960 if (optlen != TCPOLEN_SIGNATURE) 2961 continue; 2962 if (sigp && bcmp(sigp, cp + 2, TCP_SIGLEN)) 2963 return (-1); 2964 2965 sigp = sigbuf; 2966 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 2967 memset(cp + 2, 0, TCP_SIGLEN); 2968 tp->t_flags |= TF_SIGNATURE; 2969 break; 2970 #endif 2971 } 2972 } 2973 2974 #ifdef TCP_SIGNATURE 2975 if (tp->t_flags & TF_SIGNATURE) { 2976 2977 sav = tcp_signature_getsav(m, th); 2978 2979 if (sav == NULL && tp->t_state == TCPS_LISTEN) 2980 return (-1); 2981 } 2982 2983 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 2984 if (sav == NULL) 2985 return (-1); 2986 #ifdef FAST_IPSEC 2987 KEY_FREESAV(&sav); 2988 #else 2989 key_freesav(sav); 2990 #endif 2991 return (-1); 2992 } 2993 2994 if (sigp) { 2995 char sig[TCP_SIGLEN]; 2996 2997 TCP_FIELDS_TO_NET(th); 2998 if (tcp_signature(m, th, toff, sav, sig) < 0) { 2999 TCP_FIELDS_TO_HOST(th); 3000 if (sav == NULL) 3001 return (-1); 3002 #ifdef FAST_IPSEC 3003 KEY_FREESAV(&sav); 3004 #else 3005 key_freesav(sav); 3006 #endif 3007 return (-1); 3008 } 3009 TCP_FIELDS_TO_HOST(th); 3010 3011 if (bcmp(sig, sigp, TCP_SIGLEN)) { 3012 tcpstat.tcps_badsig++; 3013 if (sav == NULL) 3014 return (-1); 3015 #ifdef FAST_IPSEC 3016 KEY_FREESAV(&sav); 3017 #else 3018 key_freesav(sav); 3019 #endif 3020 return (-1); 3021 } else 3022 tcpstat.tcps_goodsig++; 3023 3024 key_sa_recordxfer(sav, m); 3025 #ifdef FAST_IPSEC 3026 KEY_FREESAV(&sav); 3027 #else 3028 key_freesav(sav); 3029 #endif 3030 } 3031 #endif 3032 3033 return (0); 3034 } 3035 3036 /* 3037 * Pull out of band byte out of a segment so 3038 * it doesn't appear in the user's data queue. 3039 * It is still reflected in the segment length for 3040 * sequencing purposes. 3041 */ 3042 void 3043 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3044 struct mbuf *m, int off) 3045 { 3046 int cnt = off + th->th_urp - 1; 3047 3048 while (cnt >= 0) { 3049 if (m->m_len > cnt) { 3050 char *cp = mtod(m, caddr_t) + cnt; 3051 struct tcpcb *tp = sototcpcb(so); 3052 3053 tp->t_iobc = *cp; 3054 tp->t_oobflags |= TCPOOB_HAVEDATA; 3055 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3056 m->m_len--; 3057 return; 3058 } 3059 cnt -= m->m_len; 3060 m = m->m_next; 3061 if (m == 0) 3062 break; 3063 } 3064 panic("tcp_pulloutofband"); 3065 } 3066 3067 /* 3068 * Collect new round-trip time estimate 3069 * and update averages and current timeout. 3070 */ 3071 void 3072 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3073 { 3074 int32_t delta; 3075 3076 tcpstat.tcps_rttupdated++; 3077 if (tp->t_srtt != 0) { 3078 /* 3079 * srtt is stored as fixed point with 3 bits after the 3080 * binary point (i.e., scaled by 8). The following magic 3081 * is equivalent to the smoothing algorithm in rfc793 with 3082 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3083 * point). Adjust rtt to origin 0. 3084 */ 3085 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3086 if ((tp->t_srtt += delta) <= 0) 3087 tp->t_srtt = 1 << 2; 3088 /* 3089 * We accumulate a smoothed rtt variance (actually, a 3090 * smoothed mean difference), then set the retransmit 3091 * timer to smoothed rtt + 4 times the smoothed variance. 3092 * rttvar is stored as fixed point with 2 bits after the 3093 * binary point (scaled by 4). The following is 3094 * equivalent to rfc793 smoothing with an alpha of .75 3095 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3096 * rfc793's wired-in beta. 3097 */ 3098 if (delta < 0) 3099 delta = -delta; 3100 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3101 if ((tp->t_rttvar += delta) <= 0) 3102 tp->t_rttvar = 1 << 2; 3103 } else { 3104 /* 3105 * No rtt measurement yet - use the unsmoothed rtt. 3106 * Set the variance to half the rtt (so our first 3107 * retransmit happens at 3*rtt). 3108 */ 3109 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3110 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3111 } 3112 tp->t_rtttime = 0; 3113 tp->t_rxtshift = 0; 3114 3115 /* 3116 * the retransmit should happen at rtt + 4 * rttvar. 3117 * Because of the way we do the smoothing, srtt and rttvar 3118 * will each average +1/2 tick of bias. When we compute 3119 * the retransmit timer, we want 1/2 tick of rounding and 3120 * 1 extra tick because of +-1/2 tick uncertainty in the 3121 * firing of the timer. The bias will give us exactly the 3122 * 1.5 tick we need. But, because the bias is 3123 * statistical, we have to test that we don't drop below 3124 * the minimum feasible timer (which is 2 ticks). 3125 */ 3126 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3127 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3128 3129 /* 3130 * We received an ack for a packet that wasn't retransmitted; 3131 * it is probably safe to discard any error indications we've 3132 * received recently. This isn't quite right, but close enough 3133 * for now (a route might have failed after we sent a segment, 3134 * and the return path might not be symmetrical). 3135 */ 3136 tp->t_softerror = 0; 3137 } 3138 3139 3140 /* 3141 * TCP compressed state engine. Currently used to hold compressed 3142 * state for SYN_RECEIVED. 3143 */ 3144 3145 u_long syn_cache_count; 3146 u_int32_t syn_hash1, syn_hash2; 3147 3148 #define SYN_HASH(sa, sp, dp) \ 3149 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3150 ((u_int32_t)(sp)))^syn_hash2))) 3151 #ifndef INET6 3152 #define SYN_HASHALL(hash, src, dst) \ 3153 do { \ 3154 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3155 ((const struct sockaddr_in *)(src))->sin_port, \ 3156 ((const struct sockaddr_in *)(dst))->sin_port); \ 3157 } while (/*CONSTCOND*/ 0) 3158 #else 3159 #define SYN_HASH6(sa, sp, dp) \ 3160 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3161 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3162 & 0x7fffffff) 3163 3164 #define SYN_HASHALL(hash, src, dst) \ 3165 do { \ 3166 switch ((src)->sa_family) { \ 3167 case AF_INET: \ 3168 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3169 ((const struct sockaddr_in *)(src))->sin_port, \ 3170 ((const struct sockaddr_in *)(dst))->sin_port); \ 3171 break; \ 3172 case AF_INET6: \ 3173 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3174 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3175 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3176 break; \ 3177 default: \ 3178 hash = 0; \ 3179 } \ 3180 } while (/*CONSTCOND*/0) 3181 #endif /* INET6 */ 3182 3183 #define SYN_CACHE_RM(sc) \ 3184 do { \ 3185 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 3186 (sc), sc_bucketq); \ 3187 (sc)->sc_tp = NULL; \ 3188 LIST_REMOVE((sc), sc_tpq); \ 3189 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 3190 callout_stop(&(sc)->sc_timer); \ 3191 syn_cache_count--; \ 3192 } while (/*CONSTCOND*/0) 3193 3194 #define SYN_CACHE_PUT(sc) \ 3195 do { \ 3196 if ((sc)->sc_ipopts) \ 3197 (void) m_free((sc)->sc_ipopts); \ 3198 if ((sc)->sc_route4.ro_rt != NULL) \ 3199 RTFREE((sc)->sc_route4.ro_rt); \ 3200 if (callout_invoking(&(sc)->sc_timer)) \ 3201 (sc)->sc_flags |= SCF_DEAD; \ 3202 else \ 3203 pool_put(&syn_cache_pool, (sc)); \ 3204 } while (/*CONSTCOND*/0) 3205 3206 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL); 3207 3208 /* 3209 * We don't estimate RTT with SYNs, so each packet starts with the default 3210 * RTT and each timer step has a fixed timeout value. 3211 */ 3212 #define SYN_CACHE_TIMER_ARM(sc) \ 3213 do { \ 3214 TCPT_RANGESET((sc)->sc_rxtcur, \ 3215 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3216 TCPTV_REXMTMAX); \ 3217 callout_reset(&(sc)->sc_timer, \ 3218 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3219 } while (/*CONSTCOND*/0) 3220 3221 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3222 3223 void 3224 syn_cache_init(void) 3225 { 3226 int i; 3227 3228 /* Initialize the hash buckets. */ 3229 for (i = 0; i < tcp_syn_cache_size; i++) 3230 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3231 } 3232 3233 void 3234 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3235 { 3236 struct syn_cache_head *scp; 3237 struct syn_cache *sc2; 3238 int s; 3239 3240 /* 3241 * If there are no entries in the hash table, reinitialize 3242 * the hash secrets. 3243 */ 3244 if (syn_cache_count == 0) { 3245 syn_hash1 = arc4random(); 3246 syn_hash2 = arc4random(); 3247 } 3248 3249 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3250 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3251 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3252 3253 /* 3254 * Make sure that we don't overflow the per-bucket 3255 * limit or the total cache size limit. 3256 */ 3257 s = splsoftnet(); 3258 if (scp->sch_length >= tcp_syn_bucket_limit) { 3259 tcpstat.tcps_sc_bucketoverflow++; 3260 /* 3261 * The bucket is full. Toss the oldest element in the 3262 * bucket. This will be the first entry in the bucket. 3263 */ 3264 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3265 #ifdef DIAGNOSTIC 3266 /* 3267 * This should never happen; we should always find an 3268 * entry in our bucket. 3269 */ 3270 if (sc2 == NULL) 3271 panic("syn_cache_insert: bucketoverflow: impossible"); 3272 #endif 3273 SYN_CACHE_RM(sc2); 3274 SYN_CACHE_PUT(sc2); /* calls pool_put but see spl above */ 3275 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3276 struct syn_cache_head *scp2, *sce; 3277 3278 tcpstat.tcps_sc_overflowed++; 3279 /* 3280 * The cache is full. Toss the oldest entry in the 3281 * first non-empty bucket we can find. 3282 * 3283 * XXX We would really like to toss the oldest 3284 * entry in the cache, but we hope that this 3285 * condition doesn't happen very often. 3286 */ 3287 scp2 = scp; 3288 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3289 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3290 for (++scp2; scp2 != scp; scp2++) { 3291 if (scp2 >= sce) 3292 scp2 = &tcp_syn_cache[0]; 3293 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3294 break; 3295 } 3296 #ifdef DIAGNOSTIC 3297 /* 3298 * This should never happen; we should always find a 3299 * non-empty bucket. 3300 */ 3301 if (scp2 == scp) 3302 panic("syn_cache_insert: cacheoverflow: " 3303 "impossible"); 3304 #endif 3305 } 3306 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3307 SYN_CACHE_RM(sc2); 3308 SYN_CACHE_PUT(sc2); /* calls pool_put but see spl above */ 3309 } 3310 3311 /* 3312 * Initialize the entry's timer. 3313 */ 3314 sc->sc_rxttot = 0; 3315 sc->sc_rxtshift = 0; 3316 SYN_CACHE_TIMER_ARM(sc); 3317 3318 /* Link it from tcpcb entry */ 3319 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3320 3321 /* Put it into the bucket. */ 3322 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3323 scp->sch_length++; 3324 syn_cache_count++; 3325 3326 tcpstat.tcps_sc_added++; 3327 splx(s); 3328 } 3329 3330 /* 3331 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3332 * If we have retransmitted an entry the maximum number of times, expire 3333 * that entry. 3334 */ 3335 void 3336 syn_cache_timer(void *arg) 3337 { 3338 struct syn_cache *sc = arg; 3339 int s; 3340 3341 s = splsoftnet(); 3342 callout_ack(&sc->sc_timer); 3343 3344 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3345 tcpstat.tcps_sc_delayed_free++; 3346 pool_put(&syn_cache_pool, sc); 3347 splx(s); 3348 return; 3349 } 3350 3351 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3352 /* Drop it -- too many retransmissions. */ 3353 goto dropit; 3354 } 3355 3356 /* 3357 * Compute the total amount of time this entry has 3358 * been on a queue. If this entry has been on longer 3359 * than the keep alive timer would allow, expire it. 3360 */ 3361 sc->sc_rxttot += sc->sc_rxtcur; 3362 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) 3363 goto dropit; 3364 3365 tcpstat.tcps_sc_retransmitted++; 3366 (void) syn_cache_respond(sc, NULL); 3367 3368 /* Advance the timer back-off. */ 3369 sc->sc_rxtshift++; 3370 SYN_CACHE_TIMER_ARM(sc); 3371 3372 splx(s); 3373 return; 3374 3375 dropit: 3376 tcpstat.tcps_sc_timed_out++; 3377 SYN_CACHE_RM(sc); 3378 SYN_CACHE_PUT(sc); /* calls pool_put but see spl above */ 3379 splx(s); 3380 } 3381 3382 /* 3383 * Remove syn cache created by the specified tcb entry, 3384 * because this does not make sense to keep them 3385 * (if there's no tcb entry, syn cache entry will never be used) 3386 */ 3387 void 3388 syn_cache_cleanup(struct tcpcb *tp) 3389 { 3390 struct syn_cache *sc, *nsc; 3391 int s; 3392 3393 s = splsoftnet(); 3394 3395 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3396 nsc = LIST_NEXT(sc, sc_tpq); 3397 3398 #ifdef DIAGNOSTIC 3399 if (sc->sc_tp != tp) 3400 panic("invalid sc_tp in syn_cache_cleanup"); 3401 #endif 3402 SYN_CACHE_RM(sc); 3403 SYN_CACHE_PUT(sc); /* calls pool_put but see spl above */ 3404 } 3405 /* just for safety */ 3406 LIST_INIT(&tp->t_sc); 3407 3408 splx(s); 3409 } 3410 3411 /* 3412 * Find an entry in the syn cache. 3413 */ 3414 struct syn_cache * 3415 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3416 struct syn_cache_head **headp) 3417 { 3418 struct syn_cache *sc; 3419 struct syn_cache_head *scp; 3420 u_int32_t hash; 3421 int s; 3422 3423 SYN_HASHALL(hash, src, dst); 3424 3425 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3426 *headp = scp; 3427 s = splsoftnet(); 3428 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3429 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3430 if (sc->sc_hash != hash) 3431 continue; 3432 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3433 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3434 splx(s); 3435 return (sc); 3436 } 3437 } 3438 splx(s); 3439 return (NULL); 3440 } 3441 3442 /* 3443 * This function gets called when we receive an ACK for a 3444 * socket in the LISTEN state. We look up the connection 3445 * in the syn cache, and if its there, we pull it out of 3446 * the cache and turn it into a full-blown connection in 3447 * the SYN-RECEIVED state. 3448 * 3449 * The return values may not be immediately obvious, and their effects 3450 * can be subtle, so here they are: 3451 * 3452 * NULL SYN was not found in cache; caller should drop the 3453 * packet and send an RST. 3454 * 3455 * -1 We were unable to create the new connection, and are 3456 * aborting it. An ACK,RST is being sent to the peer 3457 * (unless we got screwey sequence numbners; see below), 3458 * because the 3-way handshake has been completed. Caller 3459 * should not free the mbuf, since we may be using it. If 3460 * we are not, we will free it. 3461 * 3462 * Otherwise, the return value is a pointer to the new socket 3463 * associated with the connection. 3464 */ 3465 struct socket * 3466 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3467 struct tcphdr *th, unsigned int hlen __unused, unsigned int tlen __unused, 3468 struct socket *so, struct mbuf *m) 3469 { 3470 struct syn_cache *sc; 3471 struct syn_cache_head *scp; 3472 struct inpcb *inp = NULL; 3473 #ifdef INET6 3474 struct in6pcb *in6p = NULL; 3475 #endif 3476 struct tcpcb *tp = 0; 3477 struct mbuf *am; 3478 int s; 3479 struct socket *oso; 3480 3481 s = splsoftnet(); 3482 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3483 splx(s); 3484 return (NULL); 3485 } 3486 3487 /* 3488 * Verify the sequence and ack numbers. Try getting the correct 3489 * response again. 3490 */ 3491 if ((th->th_ack != sc->sc_iss + 1) || 3492 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3493 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3494 (void) syn_cache_respond(sc, m); 3495 splx(s); 3496 return ((struct socket *)(-1)); 3497 } 3498 3499 /* Remove this cache entry */ 3500 SYN_CACHE_RM(sc); 3501 splx(s); 3502 3503 /* 3504 * Ok, create the full blown connection, and set things up 3505 * as they would have been set up if we had created the 3506 * connection when the SYN arrived. If we can't create 3507 * the connection, abort it. 3508 */ 3509 /* 3510 * inp still has the OLD in_pcb stuff, set the 3511 * v6-related flags on the new guy, too. This is 3512 * done particularly for the case where an AF_INET6 3513 * socket is bound only to a port, and a v4 connection 3514 * comes in on that port. 3515 * we also copy the flowinfo from the original pcb 3516 * to the new one. 3517 */ 3518 oso = so; 3519 so = sonewconn(so, SS_ISCONNECTED); 3520 if (so == NULL) 3521 goto resetandabort; 3522 3523 switch (so->so_proto->pr_domain->dom_family) { 3524 #ifdef INET 3525 case AF_INET: 3526 inp = sotoinpcb(so); 3527 break; 3528 #endif 3529 #ifdef INET6 3530 case AF_INET6: 3531 in6p = sotoin6pcb(so); 3532 break; 3533 #endif 3534 } 3535 switch (src->sa_family) { 3536 #ifdef INET 3537 case AF_INET: 3538 if (inp) { 3539 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3540 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3541 inp->inp_options = ip_srcroute(); 3542 in_pcbstate(inp, INP_BOUND); 3543 if (inp->inp_options == NULL) { 3544 inp->inp_options = sc->sc_ipopts; 3545 sc->sc_ipopts = NULL; 3546 } 3547 } 3548 #ifdef INET6 3549 else if (in6p) { 3550 /* IPv4 packet to AF_INET6 socket */ 3551 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3552 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3553 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3554 &in6p->in6p_laddr.s6_addr32[3], 3555 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3556 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3557 in6totcpcb(in6p)->t_family = AF_INET; 3558 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3559 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3560 else 3561 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3562 in6_pcbstate(in6p, IN6P_BOUND); 3563 } 3564 #endif 3565 break; 3566 #endif 3567 #ifdef INET6 3568 case AF_INET6: 3569 if (in6p) { 3570 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3571 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3572 in6_pcbstate(in6p, IN6P_BOUND); 3573 } 3574 break; 3575 #endif 3576 } 3577 #ifdef INET6 3578 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3579 struct in6pcb *oin6p = sotoin6pcb(oso); 3580 /* inherit socket options from the listening socket */ 3581 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3582 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3583 m_freem(in6p->in6p_options); 3584 in6p->in6p_options = 0; 3585 } 3586 ip6_savecontrol(in6p, &in6p->in6p_options, 3587 mtod(m, struct ip6_hdr *), m); 3588 } 3589 #endif 3590 3591 #if defined(IPSEC) || defined(FAST_IPSEC) 3592 /* 3593 * we make a copy of policy, instead of sharing the policy, 3594 * for better behavior in terms of SA lookup and dead SA removal. 3595 */ 3596 if (inp) { 3597 /* copy old policy into new socket's */ 3598 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3599 printf("tcp_input: could not copy policy\n"); 3600 } 3601 #ifdef INET6 3602 else if (in6p) { 3603 /* copy old policy into new socket's */ 3604 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3605 in6p->in6p_sp)) 3606 printf("tcp_input: could not copy policy\n"); 3607 } 3608 #endif 3609 #endif 3610 3611 /* 3612 * Give the new socket our cached route reference. 3613 */ 3614 if (inp) 3615 inp->inp_route = sc->sc_route4; /* struct assignment */ 3616 #ifdef INET6 3617 else 3618 in6p->in6p_route = sc->sc_route6; 3619 #endif 3620 sc->sc_route4.ro_rt = NULL; 3621 3622 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3623 if (am == NULL) 3624 goto resetandabort; 3625 MCLAIM(am, &tcp_mowner); 3626 am->m_len = src->sa_len; 3627 bcopy(src, mtod(am, caddr_t), src->sa_len); 3628 if (inp) { 3629 if (in_pcbconnect(inp, am, NULL)) { 3630 (void) m_free(am); 3631 goto resetandabort; 3632 } 3633 } 3634 #ifdef INET6 3635 else if (in6p) { 3636 if (src->sa_family == AF_INET) { 3637 /* IPv4 packet to AF_INET6 socket */ 3638 struct sockaddr_in6 *sin6; 3639 sin6 = mtod(am, struct sockaddr_in6 *); 3640 am->m_len = sizeof(*sin6); 3641 bzero(sin6, sizeof(*sin6)); 3642 sin6->sin6_family = AF_INET6; 3643 sin6->sin6_len = sizeof(*sin6); 3644 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3645 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3646 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3647 &sin6->sin6_addr.s6_addr32[3], 3648 sizeof(sin6->sin6_addr.s6_addr32[3])); 3649 } 3650 if (in6_pcbconnect(in6p, am, NULL)) { 3651 (void) m_free(am); 3652 goto resetandabort; 3653 } 3654 } 3655 #endif 3656 else { 3657 (void) m_free(am); 3658 goto resetandabort; 3659 } 3660 (void) m_free(am); 3661 3662 if (inp) 3663 tp = intotcpcb(inp); 3664 #ifdef INET6 3665 else if (in6p) 3666 tp = in6totcpcb(in6p); 3667 #endif 3668 else 3669 tp = NULL; 3670 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3671 if (sc->sc_request_r_scale != 15) { 3672 tp->requested_s_scale = sc->sc_requested_s_scale; 3673 tp->request_r_scale = sc->sc_request_r_scale; 3674 tp->snd_scale = sc->sc_requested_s_scale; 3675 tp->rcv_scale = sc->sc_request_r_scale; 3676 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3677 } 3678 if (sc->sc_flags & SCF_TIMESTAMP) 3679 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3680 tp->ts_timebase = sc->sc_timebase; 3681 3682 tp->t_template = tcp_template(tp); 3683 if (tp->t_template == 0) { 3684 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3685 so = NULL; 3686 m_freem(m); 3687 goto abort; 3688 } 3689 3690 tp->iss = sc->sc_iss; 3691 tp->irs = sc->sc_irs; 3692 tcp_sendseqinit(tp); 3693 tcp_rcvseqinit(tp); 3694 tp->t_state = TCPS_SYN_RECEIVED; 3695 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3696 tcpstat.tcps_accepts++; 3697 3698 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 3699 tp->t_flags |= TF_WILL_SACK; 3700 3701 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 3702 tp->t_flags |= TF_ECN_PERMIT; 3703 3704 #ifdef TCP_SIGNATURE 3705 if (sc->sc_flags & SCF_SIGNATURE) 3706 tp->t_flags |= TF_SIGNATURE; 3707 #endif 3708 3709 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3710 tp->t_ourmss = sc->sc_ourmaxseg; 3711 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3712 3713 /* 3714 * Initialize the initial congestion window. If we 3715 * had to retransmit the SYN,ACK, we must initialize cwnd 3716 * to 1 segment (i.e. the Loss Window). 3717 */ 3718 if (sc->sc_rxtshift) 3719 tp->snd_cwnd = tp->t_peermss; 3720 else { 3721 int ss = tcp_init_win; 3722 #ifdef INET 3723 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3724 ss = tcp_init_win_local; 3725 #endif 3726 #ifdef INET6 3727 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3728 ss = tcp_init_win_local; 3729 #endif 3730 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3731 } 3732 3733 tcp_rmx_rtt(tp); 3734 tp->snd_wl1 = sc->sc_irs; 3735 tp->rcv_up = sc->sc_irs + 1; 3736 3737 /* 3738 * This is what whould have happened in tcp_output() when 3739 * the SYN,ACK was sent. 3740 */ 3741 tp->snd_up = tp->snd_una; 3742 tp->snd_max = tp->snd_nxt = tp->iss+1; 3743 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3744 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3745 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3746 tp->last_ack_sent = tp->rcv_nxt; 3747 tp->t_partialacks = -1; 3748 tp->t_dupacks = 0; 3749 3750 tcpstat.tcps_sc_completed++; 3751 s = splsoftnet(); 3752 SYN_CACHE_PUT(sc); 3753 splx(s); 3754 return (so); 3755 3756 resetandabort: 3757 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3758 abort: 3759 if (so != NULL) 3760 (void) soabort(so); 3761 s = splsoftnet(); 3762 SYN_CACHE_PUT(sc); 3763 splx(s); 3764 tcpstat.tcps_sc_aborted++; 3765 return ((struct socket *)(-1)); 3766 } 3767 3768 /* 3769 * This function is called when we get a RST for a 3770 * non-existent connection, so that we can see if the 3771 * connection is in the syn cache. If it is, zap it. 3772 */ 3773 3774 void 3775 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 3776 { 3777 struct syn_cache *sc; 3778 struct syn_cache_head *scp; 3779 int s = splsoftnet(); 3780 3781 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3782 splx(s); 3783 return; 3784 } 3785 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3786 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3787 splx(s); 3788 return; 3789 } 3790 SYN_CACHE_RM(sc); 3791 tcpstat.tcps_sc_reset++; 3792 SYN_CACHE_PUT(sc); /* calls pool_put but see spl above */ 3793 splx(s); 3794 } 3795 3796 void 3797 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 3798 struct tcphdr *th) 3799 { 3800 struct syn_cache *sc; 3801 struct syn_cache_head *scp; 3802 int s; 3803 3804 s = splsoftnet(); 3805 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3806 splx(s); 3807 return; 3808 } 3809 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3810 if (ntohl (th->th_seq) != sc->sc_iss) { 3811 splx(s); 3812 return; 3813 } 3814 3815 /* 3816 * If we've retransmitted 3 times and this is our second error, 3817 * we remove the entry. Otherwise, we allow it to continue on. 3818 * This prevents us from incorrectly nuking an entry during a 3819 * spurious network outage. 3820 * 3821 * See tcp_notify(). 3822 */ 3823 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3824 sc->sc_flags |= SCF_UNREACH; 3825 splx(s); 3826 return; 3827 } 3828 3829 SYN_CACHE_RM(sc); 3830 tcpstat.tcps_sc_unreach++; 3831 SYN_CACHE_PUT(sc); /* calls pool_put but see spl above */ 3832 splx(s); 3833 } 3834 3835 /* 3836 * Given a LISTEN socket and an inbound SYN request, add 3837 * this to the syn cache, and send back a segment: 3838 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3839 * to the source. 3840 * 3841 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3842 * Doing so would require that we hold onto the data and deliver it 3843 * to the application. However, if we are the target of a SYN-flood 3844 * DoS attack, an attacker could send data which would eventually 3845 * consume all available buffer space if it were ACKed. By not ACKing 3846 * the data, we avoid this DoS scenario. 3847 */ 3848 3849 int 3850 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3851 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 3852 int optlen, struct tcp_opt_info *oi) 3853 { 3854 struct tcpcb tb, *tp; 3855 long win; 3856 struct syn_cache *sc; 3857 struct syn_cache_head *scp; 3858 struct mbuf *ipopts; 3859 struct tcp_opt_info opti; 3860 int s; 3861 3862 tp = sototcpcb(so); 3863 3864 bzero(&opti, sizeof(opti)); 3865 3866 /* 3867 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3868 * 3869 * Note this check is performed in tcp_input() very early on. 3870 */ 3871 3872 /* 3873 * Initialize some local state. 3874 */ 3875 win = sbspace(&so->so_rcv); 3876 if (win > TCP_MAXWIN) 3877 win = TCP_MAXWIN; 3878 3879 switch (src->sa_family) { 3880 #ifdef INET 3881 case AF_INET: 3882 /* 3883 * Remember the IP options, if any. 3884 */ 3885 ipopts = ip_srcroute(); 3886 break; 3887 #endif 3888 default: 3889 ipopts = NULL; 3890 } 3891 3892 #ifdef TCP_SIGNATURE 3893 if (optp || (tp->t_flags & TF_SIGNATURE)) 3894 #else 3895 if (optp) 3896 #endif 3897 { 3898 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3899 #ifdef TCP_SIGNATURE 3900 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 3901 #endif 3902 tb.t_state = TCPS_LISTEN; 3903 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 3904 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 3905 return (0); 3906 } else 3907 tb.t_flags = 0; 3908 3909 /* 3910 * See if we already have an entry for this connection. 3911 * If we do, resend the SYN,ACK. We do not count this 3912 * as a retransmission (XXX though maybe we should). 3913 */ 3914 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3915 tcpstat.tcps_sc_dupesyn++; 3916 if (ipopts) { 3917 /* 3918 * If we were remembering a previous source route, 3919 * forget it and use the new one we've been given. 3920 */ 3921 if (sc->sc_ipopts) 3922 (void) m_free(sc->sc_ipopts); 3923 sc->sc_ipopts = ipopts; 3924 } 3925 sc->sc_timestamp = tb.ts_recent; 3926 if (syn_cache_respond(sc, m) == 0) { 3927 tcpstat.tcps_sndacks++; 3928 tcpstat.tcps_sndtotal++; 3929 } 3930 return (1); 3931 } 3932 3933 s = splsoftnet(); 3934 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3935 splx(s); 3936 if (sc == NULL) { 3937 if (ipopts) 3938 (void) m_free(ipopts); 3939 return (0); 3940 } 3941 3942 /* 3943 * Fill in the cache, and put the necessary IP and TCP 3944 * options into the reply. 3945 */ 3946 bzero(sc, sizeof(struct syn_cache)); 3947 callout_init(&sc->sc_timer); 3948 bcopy(src, &sc->sc_src, src->sa_len); 3949 bcopy(dst, &sc->sc_dst, dst->sa_len); 3950 sc->sc_flags = 0; 3951 sc->sc_ipopts = ipopts; 3952 sc->sc_irs = th->th_seq; 3953 switch (src->sa_family) { 3954 #ifdef INET 3955 case AF_INET: 3956 { 3957 struct sockaddr_in *srcin = (void *) src; 3958 struct sockaddr_in *dstin = (void *) dst; 3959 3960 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 3961 &srcin->sin_addr, dstin->sin_port, 3962 srcin->sin_port, sizeof(dstin->sin_addr), 0); 3963 break; 3964 } 3965 #endif /* INET */ 3966 #ifdef INET6 3967 case AF_INET6: 3968 { 3969 struct sockaddr_in6 *srcin6 = (void *) src; 3970 struct sockaddr_in6 *dstin6 = (void *) dst; 3971 3972 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 3973 &srcin6->sin6_addr, dstin6->sin6_port, 3974 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 3975 break; 3976 } 3977 #endif /* INET6 */ 3978 } 3979 sc->sc_peermaxseg = oi->maxseg; 3980 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3981 m->m_pkthdr.rcvif : NULL, 3982 sc->sc_src.sa.sa_family); 3983 sc->sc_win = win; 3984 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 3985 sc->sc_timestamp = tb.ts_recent; 3986 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 3987 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 3988 sc->sc_flags |= SCF_TIMESTAMP; 3989 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3990 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3991 sc->sc_requested_s_scale = tb.requested_s_scale; 3992 sc->sc_request_r_scale = 0; 3993 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3994 TCP_MAXWIN << sc->sc_request_r_scale < 3995 so->so_rcv.sb_hiwat) 3996 sc->sc_request_r_scale++; 3997 } else { 3998 sc->sc_requested_s_scale = 15; 3999 sc->sc_request_r_scale = 15; 4000 } 4001 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4002 sc->sc_flags |= SCF_SACK_PERMIT; 4003 4004 /* 4005 * ECN setup packet recieved. 4006 */ 4007 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4008 sc->sc_flags |= SCF_ECN_PERMIT; 4009 4010 #ifdef TCP_SIGNATURE 4011 if (tb.t_flags & TF_SIGNATURE) 4012 sc->sc_flags |= SCF_SIGNATURE; 4013 #endif 4014 sc->sc_tp = tp; 4015 if (syn_cache_respond(sc, m) == 0) { 4016 syn_cache_insert(sc, tp); 4017 tcpstat.tcps_sndacks++; 4018 tcpstat.tcps_sndtotal++; 4019 } else { 4020 s = splsoftnet(); 4021 SYN_CACHE_PUT(sc); 4022 splx(s); 4023 tcpstat.tcps_sc_dropped++; 4024 } 4025 return (1); 4026 } 4027 4028 int 4029 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4030 { 4031 struct route *ro; 4032 u_int8_t *optp; 4033 int optlen, error; 4034 u_int16_t tlen; 4035 struct ip *ip = NULL; 4036 #ifdef INET6 4037 struct ip6_hdr *ip6 = NULL; 4038 #endif 4039 struct tcpcb *tp = NULL; 4040 struct tcphdr *th; 4041 u_int hlen; 4042 struct socket *so; 4043 4044 switch (sc->sc_src.sa.sa_family) { 4045 case AF_INET: 4046 hlen = sizeof(struct ip); 4047 ro = &sc->sc_route4; 4048 break; 4049 #ifdef INET6 4050 case AF_INET6: 4051 hlen = sizeof(struct ip6_hdr); 4052 ro = (struct route *)&sc->sc_route6; 4053 break; 4054 #endif 4055 default: 4056 if (m) 4057 m_freem(m); 4058 return (EAFNOSUPPORT); 4059 } 4060 4061 /* Compute the size of the TCP options. */ 4062 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4063 ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) + 4064 #ifdef TCP_SIGNATURE 4065 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4066 #endif 4067 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4068 4069 tlen = hlen + sizeof(struct tcphdr) + optlen; 4070 4071 /* 4072 * Create the IP+TCP header from scratch. 4073 */ 4074 if (m) 4075 m_freem(m); 4076 #ifdef DIAGNOSTIC 4077 if (max_linkhdr + tlen > MCLBYTES) 4078 return (ENOBUFS); 4079 #endif 4080 MGETHDR(m, M_DONTWAIT, MT_DATA); 4081 if (m && tlen > MHLEN) { 4082 MCLGET(m, M_DONTWAIT); 4083 if ((m->m_flags & M_EXT) == 0) { 4084 m_freem(m); 4085 m = NULL; 4086 } 4087 } 4088 if (m == NULL) 4089 return (ENOBUFS); 4090 MCLAIM(m, &tcp_tx_mowner); 4091 4092 /* Fixup the mbuf. */ 4093 m->m_data += max_linkhdr; 4094 m->m_len = m->m_pkthdr.len = tlen; 4095 if (sc->sc_tp) { 4096 tp = sc->sc_tp; 4097 if (tp->t_inpcb) 4098 so = tp->t_inpcb->inp_socket; 4099 #ifdef INET6 4100 else if (tp->t_in6pcb) 4101 so = tp->t_in6pcb->in6p_socket; 4102 #endif 4103 else 4104 so = NULL; 4105 } else 4106 so = NULL; 4107 m->m_pkthdr.rcvif = NULL; 4108 memset(mtod(m, u_char *), 0, tlen); 4109 4110 switch (sc->sc_src.sa.sa_family) { 4111 case AF_INET: 4112 ip = mtod(m, struct ip *); 4113 ip->ip_v = 4; 4114 ip->ip_dst = sc->sc_src.sin.sin_addr; 4115 ip->ip_src = sc->sc_dst.sin.sin_addr; 4116 ip->ip_p = IPPROTO_TCP; 4117 th = (struct tcphdr *)(ip + 1); 4118 th->th_dport = sc->sc_src.sin.sin_port; 4119 th->th_sport = sc->sc_dst.sin.sin_port; 4120 break; 4121 #ifdef INET6 4122 case AF_INET6: 4123 ip6 = mtod(m, struct ip6_hdr *); 4124 ip6->ip6_vfc = IPV6_VERSION; 4125 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4126 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4127 ip6->ip6_nxt = IPPROTO_TCP; 4128 /* ip6_plen will be updated in ip6_output() */ 4129 th = (struct tcphdr *)(ip6 + 1); 4130 th->th_dport = sc->sc_src.sin6.sin6_port; 4131 th->th_sport = sc->sc_dst.sin6.sin6_port; 4132 break; 4133 #endif 4134 default: 4135 th = NULL; 4136 } 4137 4138 th->th_seq = htonl(sc->sc_iss); 4139 th->th_ack = htonl(sc->sc_irs + 1); 4140 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4141 th->th_flags = TH_SYN|TH_ACK; 4142 th->th_win = htons(sc->sc_win); 4143 /* th_sum already 0 */ 4144 /* th_urp already 0 */ 4145 4146 /* Tack on the TCP options. */ 4147 optp = (u_int8_t *)(th + 1); 4148 *optp++ = TCPOPT_MAXSEG; 4149 *optp++ = 4; 4150 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4151 *optp++ = sc->sc_ourmaxseg & 0xff; 4152 4153 if (sc->sc_request_r_scale != 15) { 4154 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4155 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4156 sc->sc_request_r_scale); 4157 optp += 4; 4158 } 4159 4160 if (sc->sc_flags & SCF_TIMESTAMP) { 4161 u_int32_t *lp = (u_int32_t *)(optp); 4162 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4163 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4164 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4165 *lp = htonl(sc->sc_timestamp); 4166 optp += TCPOLEN_TSTAMP_APPA; 4167 } 4168 4169 if (sc->sc_flags & SCF_SACK_PERMIT) { 4170 u_int8_t *p = optp; 4171 4172 /* Let the peer know that we will SACK. */ 4173 p[0] = TCPOPT_SACK_PERMITTED; 4174 p[1] = 2; 4175 p[2] = TCPOPT_NOP; 4176 p[3] = TCPOPT_NOP; 4177 optp += 4; 4178 } 4179 4180 /* 4181 * Send ECN SYN-ACK setup packet. 4182 * Routes can be asymetric, so, even if we receive a packet 4183 * with ECE and CWR set, we must not assume no one will block 4184 * the ECE packet we are about to send. 4185 */ 4186 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4187 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4188 th->th_flags |= TH_ECE; 4189 tcpstat.tcps_ecn_shs++; 4190 4191 /* 4192 * draft-ietf-tcpm-ecnsyn-00.txt 4193 * 4194 * "[...] a TCP node MAY respond to an ECN-setup 4195 * SYN packet by setting ECT in the responding 4196 * ECN-setup SYN/ACK packet, indicating to routers 4197 * that the SYN/ACK packet is ECN-Capable. 4198 * This allows a congested router along the path 4199 * to mark the packet instead of dropping the 4200 * packet as an indication of congestion." 4201 * 4202 * "[...] There can be a great benefit in setting 4203 * an ECN-capable codepoint in SYN/ACK packets [...] 4204 * Congestion is most likely to occur in 4205 * the server-to-client direction. As a result, 4206 * setting an ECN-capable codepoint in SYN/ACK 4207 * packets can reduce the occurence of three-second 4208 * retransmit timeouts resulting from the drop 4209 * of SYN/ACK packets." 4210 * 4211 * Page 4 and 6, January 2006. 4212 */ 4213 4214 switch (sc->sc_src.sa.sa_family) { 4215 #ifdef INET 4216 case AF_INET: 4217 ip->ip_tos |= IPTOS_ECN_ECT0; 4218 break; 4219 #endif 4220 #ifdef INET6 4221 case AF_INET6: 4222 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4223 break; 4224 #endif 4225 } 4226 tcpstat.tcps_ecn_ect++; 4227 } 4228 4229 #ifdef TCP_SIGNATURE 4230 if (sc->sc_flags & SCF_SIGNATURE) { 4231 struct secasvar *sav; 4232 u_int8_t *sigp; 4233 4234 sav = tcp_signature_getsav(m, th); 4235 4236 if (sav == NULL) { 4237 if (m) 4238 m_freem(m); 4239 return (EPERM); 4240 } 4241 4242 *optp++ = TCPOPT_SIGNATURE; 4243 *optp++ = TCPOLEN_SIGNATURE; 4244 sigp = optp; 4245 bzero(optp, TCP_SIGLEN); 4246 optp += TCP_SIGLEN; 4247 *optp++ = TCPOPT_NOP; 4248 *optp++ = TCPOPT_EOL; 4249 4250 (void)tcp_signature(m, th, hlen, sav, sigp); 4251 4252 key_sa_recordxfer(sav, m); 4253 #ifdef FAST_IPSEC 4254 KEY_FREESAV(&sav); 4255 #else 4256 key_freesav(sav); 4257 #endif 4258 } 4259 #endif 4260 4261 /* Compute the packet's checksum. */ 4262 switch (sc->sc_src.sa.sa_family) { 4263 case AF_INET: 4264 ip->ip_len = htons(tlen - hlen); 4265 th->th_sum = 0; 4266 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4267 break; 4268 #ifdef INET6 4269 case AF_INET6: 4270 ip6->ip6_plen = htons(tlen - hlen); 4271 th->th_sum = 0; 4272 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4273 break; 4274 #endif 4275 } 4276 4277 /* 4278 * Fill in some straggling IP bits. Note the stack expects 4279 * ip_len to be in host order, for convenience. 4280 */ 4281 switch (sc->sc_src.sa.sa_family) { 4282 #ifdef INET 4283 case AF_INET: 4284 ip->ip_len = htons(tlen); 4285 ip->ip_ttl = ip_defttl; 4286 /* XXX tos? */ 4287 break; 4288 #endif 4289 #ifdef INET6 4290 case AF_INET6: 4291 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4292 ip6->ip6_vfc |= IPV6_VERSION; 4293 ip6->ip6_plen = htons(tlen - hlen); 4294 /* ip6_hlim will be initialized afterwards */ 4295 /* XXX flowlabel? */ 4296 break; 4297 #endif 4298 } 4299 4300 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4301 tp = sc->sc_tp; 4302 4303 switch (sc->sc_src.sa.sa_family) { 4304 #ifdef INET 4305 case AF_INET: 4306 error = ip_output(m, sc->sc_ipopts, ro, 4307 (ip_mtudisc ? IP_MTUDISC : 0), 4308 (struct ip_moptions *)NULL, so); 4309 break; 4310 #endif 4311 #ifdef INET6 4312 case AF_INET6: 4313 ip6->ip6_hlim = in6_selecthlim(NULL, 4314 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 4315 4316 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0, 4317 (struct ip6_moptions *)0, so, NULL); 4318 break; 4319 #endif 4320 default: 4321 error = EAFNOSUPPORT; 4322 break; 4323 } 4324 return (error); 4325 } 4326
Indexes created Wed Sep 24 05:09:52 GMT 2025