tcp_input.c revision 1.273
1 /* $NetBSD: tcp_input.c,v 1.273 2007/12/16 14:12:34 elad 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 1.273 2007/12/16 14:12:34 elad 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 #include <sys/lwp.h> /* for lwp0 */ 178 179 #include <net/if.h> 180 #include <net/route.h> 181 #include <net/if_types.h> 182 183 #include <netinet/in.h> 184 #include <netinet/in_systm.h> 185 #include <netinet/ip.h> 186 #include <netinet/in_pcb.h> 187 #include <netinet/in_var.h> 188 #include <netinet/ip_var.h> 189 #include <netinet/in_offload.h> 190 191 #ifdef INET6 192 #ifndef INET 193 #include <netinet/in.h> 194 #endif 195 #include <netinet/ip6.h> 196 #include <netinet6/ip6_var.h> 197 #include <netinet6/in6_pcb.h> 198 #include <netinet6/ip6_var.h> 199 #include <netinet6/in6_var.h> 200 #include <netinet/icmp6.h> 201 #include <netinet6/nd6.h> 202 #ifdef TCP_SIGNATURE 203 #include <netinet6/scope6_var.h> 204 #endif 205 #endif 206 207 #ifndef INET6 208 /* always need ip6.h for IP6_EXTHDR_GET */ 209 #include <netinet/ip6.h> 210 #endif 211 212 #include <netinet/tcp.h> 213 #include <netinet/tcp_fsm.h> 214 #include <netinet/tcp_seq.h> 215 #include <netinet/tcp_timer.h> 216 #include <netinet/tcp_var.h> 217 #include <netinet/tcpip.h> 218 #include <netinet/tcp_congctl.h> 219 #include <netinet/tcp_debug.h> 220 221 #include <machine/stdarg.h> 222 223 #ifdef IPSEC 224 #include <netinet6/ipsec.h> 225 #include <netkey/key.h> 226 #endif /*IPSEC*/ 227 #ifdef INET6 228 #include "faith.h" 229 #if defined(NFAITH) && NFAITH > 0 230 #include <net/if_faith.h> 231 #endif 232 #endif /* IPSEC */ 233 234 #ifdef FAST_IPSEC 235 #include <netipsec/ipsec.h> 236 #include <netipsec/ipsec_var.h> /* XXX ipsecstat namespace */ 237 #include <netipsec/key.h> 238 #ifdef INET6 239 #include <netipsec/ipsec6.h> 240 #endif 241 #endif /* FAST_IPSEC*/ 242 243 int tcprexmtthresh = 3; 244 int tcp_log_refused; 245 246 int tcp_do_autorcvbuf = 0; 247 int tcp_autorcvbuf_inc = 16 * 1024; 248 int tcp_autorcvbuf_max = 256 * 1024; 249 250 static int tcp_rst_ppslim_count = 0; 251 static struct timeval tcp_rst_ppslim_last; 252 static int tcp_ackdrop_ppslim_count = 0; 253 static struct timeval tcp_ackdrop_ppslim_last; 254 255 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 256 257 /* for modulo comparisons of timestamps */ 258 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 259 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 260 261 /* 262 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 263 */ 264 #ifdef INET6 265 #define ND6_HINT(tp) \ 266 do { \ 267 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 && \ 268 tp->t_in6pcb->in6p_route.ro_rt) { \ 269 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \ 270 } \ 271 } while (/*CONSTCOND*/ 0) 272 #else 273 #define ND6_HINT(tp) 274 #endif 275 276 /* 277 * Macro to compute ACK transmission behavior. Delay the ACK unless 278 * we have already delayed an ACK (must send an ACK every two segments). 279 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 280 * option is enabled. 281 */ 282 #define TCP_SETUP_ACK(tp, th) \ 283 do { \ 284 if ((tp)->t_flags & TF_DELACK || \ 285 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 286 tp->t_flags |= TF_ACKNOW; \ 287 else \ 288 TCP_SET_DELACK(tp); \ 289 } while (/*CONSTCOND*/ 0) 290 291 #define ICMP_CHECK(tp, th, acked) \ 292 do { \ 293 /* \ 294 * If we had a pending ICMP message that \ 295 * refers to data that have just been \ 296 * acknowledged, disregard the recorded ICMP \ 297 * message. \ 298 */ \ 299 if (((tp)->t_flags & TF_PMTUD_PEND) && \ 300 SEQ_GT((th)->th_ack, (tp)->t_pmtud_th_seq)) \ 301 (tp)->t_flags &= ~TF_PMTUD_PEND; \ 302 \ 303 /* \ 304 * Keep track of the largest chunk of data \ 305 * acknowledged since last PMTU update \ 306 */ \ 307 if ((tp)->t_pmtud_mss_acked < (acked)) \ 308 (tp)->t_pmtud_mss_acked = (acked); \ 309 } while (/*CONSTCOND*/ 0) 310 311 /* 312 * Convert TCP protocol fields to host order for easier processing. 313 */ 314 #define TCP_FIELDS_TO_HOST(th) \ 315 do { \ 316 NTOHL((th)->th_seq); \ 317 NTOHL((th)->th_ack); \ 318 NTOHS((th)->th_win); \ 319 NTOHS((th)->th_urp); \ 320 } while (/*CONSTCOND*/ 0) 321 322 /* 323 * ... and reverse the above. 324 */ 325 #define TCP_FIELDS_TO_NET(th) \ 326 do { \ 327 HTONL((th)->th_seq); \ 328 HTONL((th)->th_ack); \ 329 HTONS((th)->th_win); \ 330 HTONS((th)->th_urp); \ 331 } while (/*CONSTCOND*/ 0) 332 333 #ifdef TCP_CSUM_COUNTERS 334 #include <sys/device.h> 335 336 #if defined(INET) 337 extern struct evcnt tcp_hwcsum_ok; 338 extern struct evcnt tcp_hwcsum_bad; 339 extern struct evcnt tcp_hwcsum_data; 340 extern struct evcnt tcp_swcsum; 341 #endif /* defined(INET) */ 342 #if defined(INET6) 343 extern struct evcnt tcp6_hwcsum_ok; 344 extern struct evcnt tcp6_hwcsum_bad; 345 extern struct evcnt tcp6_hwcsum_data; 346 extern struct evcnt tcp6_swcsum; 347 #endif /* defined(INET6) */ 348 349 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 350 351 #else 352 353 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 354 355 #endif /* TCP_CSUM_COUNTERS */ 356 357 #ifdef TCP_REASS_COUNTERS 358 #include <sys/device.h> 359 360 extern struct evcnt tcp_reass_; 361 extern struct evcnt tcp_reass_empty; 362 extern struct evcnt tcp_reass_iteration[8]; 363 extern struct evcnt tcp_reass_prependfirst; 364 extern struct evcnt tcp_reass_prepend; 365 extern struct evcnt tcp_reass_insert; 366 extern struct evcnt tcp_reass_inserttail; 367 extern struct evcnt tcp_reass_append; 368 extern struct evcnt tcp_reass_appendtail; 369 extern struct evcnt tcp_reass_overlaptail; 370 extern struct evcnt tcp_reass_overlapfront; 371 extern struct evcnt tcp_reass_segdup; 372 extern struct evcnt tcp_reass_fragdup; 373 374 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 375 376 #else 377 378 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 379 380 #endif /* TCP_REASS_COUNTERS */ 381 382 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *, 383 int *); 384 static int tcp_dooptions(struct tcpcb *, const u_char *, int, 385 struct tcphdr *, struct mbuf *, int, struct tcp_opt_info *); 386 387 #ifdef INET 388 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 389 #endif 390 #ifdef INET6 391 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 392 #endif 393 394 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 395 396 #if defined(MBUFTRACE) 397 struct mowner tcp_reass_mowner = MOWNER_INIT("tcp", "reass"); 398 #endif /* defined(MBUFTRACE) */ 399 400 static POOL_INIT(tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", 401 NULL, IPL_VM); 402 403 struct ipqent * 404 tcpipqent_alloc() 405 { 406 struct ipqent *ipqe; 407 int s; 408 409 s = splvm(); 410 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 411 splx(s); 412 413 return ipqe; 414 } 415 416 void 417 tcpipqent_free(struct ipqent *ipqe) 418 { 419 int s; 420 421 s = splvm(); 422 pool_put(&tcpipqent_pool, ipqe); 423 splx(s); 424 } 425 426 static int 427 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int *tlen) 428 { 429 struct ipqent *p, *q, *nq, *tiqe = NULL; 430 struct socket *so = NULL; 431 int pkt_flags; 432 tcp_seq pkt_seq; 433 unsigned pkt_len; 434 u_long rcvpartdupbyte = 0; 435 u_long rcvoobyte; 436 #ifdef TCP_REASS_COUNTERS 437 u_int count = 0; 438 #endif 439 440 if (tp->t_inpcb) 441 so = tp->t_inpcb->inp_socket; 442 #ifdef INET6 443 else if (tp->t_in6pcb) 444 so = tp->t_in6pcb->in6p_socket; 445 #endif 446 447 TCP_REASS_LOCK_CHECK(tp); 448 449 /* 450 * Call with th==0 after become established to 451 * force pre-ESTABLISHED data up to user socket. 452 */ 453 if (th == 0) 454 goto present; 455 456 m_claimm(m, &tcp_reass_mowner); 457 458 rcvoobyte = *tlen; 459 /* 460 * Copy these to local variables because the tcpiphdr 461 * gets munged while we are collapsing mbufs. 462 */ 463 pkt_seq = th->th_seq; 464 pkt_len = *tlen; 465 pkt_flags = th->th_flags; 466 467 TCP_REASS_COUNTER_INCR(&tcp_reass_); 468 469 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 470 /* 471 * When we miss a packet, the vast majority of time we get 472 * packets that follow it in order. So optimize for that. 473 */ 474 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 475 p->ipqe_len += pkt_len; 476 p->ipqe_flags |= pkt_flags; 477 m_cat(p->ipre_mlast, m); 478 TRAVERSE(p->ipre_mlast); 479 m = NULL; 480 tiqe = p; 481 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 482 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 483 goto skip_replacement; 484 } 485 /* 486 * While we're here, if the pkt is completely beyond 487 * anything we have, just insert it at the tail. 488 */ 489 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 490 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 491 goto insert_it; 492 } 493 } 494 495 q = TAILQ_FIRST(&tp->segq); 496 497 if (q != NULL) { 498 /* 499 * If this segment immediately precedes the first out-of-order 500 * block, simply slap the segment in front of it and (mostly) 501 * skip the complicated logic. 502 */ 503 if (pkt_seq + pkt_len == q->ipqe_seq) { 504 q->ipqe_seq = pkt_seq; 505 q->ipqe_len += pkt_len; 506 q->ipqe_flags |= pkt_flags; 507 m_cat(m, q->ipqe_m); 508 q->ipqe_m = m; 509 q->ipre_mlast = m; /* last mbuf may have changed */ 510 TRAVERSE(q->ipre_mlast); 511 tiqe = q; 512 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 513 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 514 goto skip_replacement; 515 } 516 } else { 517 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 518 } 519 520 /* 521 * Find a segment which begins after this one does. 522 */ 523 for (p = NULL; q != NULL; q = nq) { 524 nq = TAILQ_NEXT(q, ipqe_q); 525 #ifdef TCP_REASS_COUNTERS 526 count++; 527 #endif 528 /* 529 * If the received segment is just right after this 530 * fragment, merge the two together and then check 531 * for further overlaps. 532 */ 533 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 534 #ifdef TCPREASS_DEBUG 535 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 536 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 537 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 538 #endif 539 pkt_len += q->ipqe_len; 540 pkt_flags |= q->ipqe_flags; 541 pkt_seq = q->ipqe_seq; 542 m_cat(q->ipre_mlast, m); 543 TRAVERSE(q->ipre_mlast); 544 m = q->ipqe_m; 545 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 546 goto free_ipqe; 547 } 548 /* 549 * If the received segment is completely past this 550 * fragment, we need to go the next fragment. 551 */ 552 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 553 p = q; 554 continue; 555 } 556 /* 557 * If the fragment is past the received segment, 558 * it (or any following) can't be concatenated. 559 */ 560 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 561 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 562 break; 563 } 564 565 /* 566 * We've received all the data in this segment before. 567 * mark it as a duplicate and return. 568 */ 569 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 570 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 571 tcpstat.tcps_rcvduppack++; 572 tcpstat.tcps_rcvdupbyte += pkt_len; 573 tcp_new_dsack(tp, pkt_seq, pkt_len); 574 m_freem(m); 575 if (tiqe != NULL) { 576 tcpipqent_free(tiqe); 577 } 578 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 579 return (0); 580 } 581 /* 582 * Received segment completely overlaps this fragment 583 * so we drop the fragment (this keeps the temporal 584 * ordering of segments correct). 585 */ 586 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 587 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 588 rcvpartdupbyte += q->ipqe_len; 589 m_freem(q->ipqe_m); 590 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 591 goto free_ipqe; 592 } 593 /* 594 * RX'ed segment extends past the end of the 595 * fragment. Drop the overlapping bytes. Then 596 * merge the fragment and segment then treat as 597 * a longer received packet. 598 */ 599 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 600 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 601 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 602 #ifdef TCPREASS_DEBUG 603 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 604 tp, overlap, 605 pkt_seq, pkt_seq + pkt_len, pkt_len); 606 #endif 607 m_adj(m, overlap); 608 rcvpartdupbyte += overlap; 609 m_cat(q->ipre_mlast, m); 610 TRAVERSE(q->ipre_mlast); 611 m = q->ipqe_m; 612 pkt_seq = q->ipqe_seq; 613 pkt_len += q->ipqe_len - overlap; 614 rcvoobyte -= overlap; 615 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 616 goto free_ipqe; 617 } 618 /* 619 * RX'ed segment extends past the front of the 620 * fragment. Drop the overlapping bytes on the 621 * received packet. The packet will then be 622 * contatentated with this fragment a bit later. 623 */ 624 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 625 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 626 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 627 #ifdef TCPREASS_DEBUG 628 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 629 tp, overlap, 630 pkt_seq, pkt_seq + pkt_len, pkt_len); 631 #endif 632 m_adj(m, -overlap); 633 pkt_len -= overlap; 634 rcvpartdupbyte += overlap; 635 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 636 rcvoobyte -= overlap; 637 } 638 /* 639 * If the received segment immediates precedes this 640 * fragment then tack the fragment onto this segment 641 * and reinsert the data. 642 */ 643 if (q->ipqe_seq == pkt_seq + pkt_len) { 644 #ifdef TCPREASS_DEBUG 645 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 646 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 647 pkt_seq, pkt_seq + pkt_len, pkt_len); 648 #endif 649 pkt_len += q->ipqe_len; 650 pkt_flags |= q->ipqe_flags; 651 m_cat(m, q->ipqe_m); 652 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 653 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 654 tp->t_segqlen--; 655 KASSERT(tp->t_segqlen >= 0); 656 KASSERT(tp->t_segqlen != 0 || 657 (TAILQ_EMPTY(&tp->segq) && 658 TAILQ_EMPTY(&tp->timeq))); 659 if (tiqe == NULL) { 660 tiqe = q; 661 } else { 662 tcpipqent_free(q); 663 } 664 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 665 break; 666 } 667 /* 668 * If the fragment is before the segment, remember it. 669 * When this loop is terminated, p will contain the 670 * pointer to fragment that is right before the received 671 * segment. 672 */ 673 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 674 p = q; 675 676 continue; 677 678 /* 679 * This is a common operation. It also will allow 680 * to save doing a malloc/free in most instances. 681 */ 682 free_ipqe: 683 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 684 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 685 tp->t_segqlen--; 686 KASSERT(tp->t_segqlen >= 0); 687 KASSERT(tp->t_segqlen != 0 || 688 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 689 if (tiqe == NULL) { 690 tiqe = q; 691 } else { 692 tcpipqent_free(q); 693 } 694 } 695 696 #ifdef TCP_REASS_COUNTERS 697 if (count > 7) 698 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 699 else if (count > 0) 700 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 701 #endif 702 703 insert_it: 704 705 /* 706 * Allocate a new queue entry since the received segment did not 707 * collapse onto any other out-of-order block; thus we are allocating 708 * a new block. If it had collapsed, tiqe would not be NULL and 709 * we would be reusing it. 710 * XXX If we can't, just drop the packet. XXX 711 */ 712 if (tiqe == NULL) { 713 tiqe = tcpipqent_alloc(); 714 if (tiqe == NULL) { 715 tcpstat.tcps_rcvmemdrop++; 716 m_freem(m); 717 return (0); 718 } 719 } 720 721 /* 722 * Update the counters. 723 */ 724 tcpstat.tcps_rcvoopack++; 725 tcpstat.tcps_rcvoobyte += rcvoobyte; 726 if (rcvpartdupbyte) { 727 tcpstat.tcps_rcvpartduppack++; 728 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 729 } 730 731 /* 732 * Insert the new fragment queue entry into both queues. 733 */ 734 tiqe->ipqe_m = m; 735 tiqe->ipre_mlast = m; 736 tiqe->ipqe_seq = pkt_seq; 737 tiqe->ipqe_len = pkt_len; 738 tiqe->ipqe_flags = pkt_flags; 739 if (p == NULL) { 740 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 741 #ifdef TCPREASS_DEBUG 742 if (tiqe->ipqe_seq != tp->rcv_nxt) 743 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 744 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 745 #endif 746 } else { 747 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 748 #ifdef TCPREASS_DEBUG 749 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 750 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 751 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 752 #endif 753 } 754 tp->t_segqlen++; 755 756 skip_replacement: 757 758 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 759 760 present: 761 /* 762 * Present data to user, advancing rcv_nxt through 763 * completed sequence space. 764 */ 765 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 766 return (0); 767 q = TAILQ_FIRST(&tp->segq); 768 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 769 return (0); 770 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 771 return (0); 772 773 tp->rcv_nxt += q->ipqe_len; 774 pkt_flags = q->ipqe_flags & TH_FIN; 775 ND6_HINT(tp); 776 777 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 778 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 779 tp->t_segqlen--; 780 KASSERT(tp->t_segqlen >= 0); 781 KASSERT(tp->t_segqlen != 0 || 782 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 783 if (so->so_state & SS_CANTRCVMORE) 784 m_freem(q->ipqe_m); 785 else 786 sbappendstream(&so->so_rcv, q->ipqe_m); 787 tcpipqent_free(q); 788 sorwakeup(so); 789 return (pkt_flags); 790 } 791 792 #ifdef INET6 793 int 794 tcp6_input(struct mbuf **mp, int *offp, int proto) 795 { 796 struct mbuf *m = *mp; 797 798 /* 799 * draft-itojun-ipv6-tcp-to-anycast 800 * better place to put this in? 801 */ 802 if (m->m_flags & M_ANYCAST6) { 803 struct ip6_hdr *ip6; 804 if (m->m_len < sizeof(struct ip6_hdr)) { 805 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 806 tcpstat.tcps_rcvshort++; 807 return IPPROTO_DONE; 808 } 809 } 810 ip6 = mtod(m, struct ip6_hdr *); 811 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 812 (char *)&ip6->ip6_dst - (char *)ip6); 813 return IPPROTO_DONE; 814 } 815 816 tcp_input(m, *offp, proto); 817 return IPPROTO_DONE; 818 } 819 #endif 820 821 #ifdef INET 822 static void 823 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 824 { 825 char src[4*sizeof "123"]; 826 char dst[4*sizeof "123"]; 827 828 if (ip) { 829 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 830 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 831 } 832 else { 833 strlcpy(src, "(unknown)", sizeof(src)); 834 strlcpy(dst, "(unknown)", sizeof(dst)); 835 } 836 log(LOG_INFO, 837 "Connection attempt to TCP %s:%d from %s:%d\n", 838 dst, ntohs(th->th_dport), 839 src, ntohs(th->th_sport)); 840 } 841 #endif 842 843 #ifdef INET6 844 static void 845 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 846 { 847 char src[INET6_ADDRSTRLEN]; 848 char dst[INET6_ADDRSTRLEN]; 849 850 if (ip6) { 851 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 852 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 853 } 854 else { 855 strlcpy(src, "(unknown v6)", sizeof(src)); 856 strlcpy(dst, "(unknown v6)", sizeof(dst)); 857 } 858 log(LOG_INFO, 859 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 860 dst, ntohs(th->th_dport), 861 src, ntohs(th->th_sport)); 862 } 863 #endif 864 865 /* 866 * Checksum extended TCP header and data. 867 */ 868 int 869 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, 870 int toff, int off, int tlen) 871 { 872 873 /* 874 * XXX it's better to record and check if this mbuf is 875 * already checked. 876 */ 877 878 switch (af) { 879 #ifdef INET 880 case AF_INET: 881 switch (m->m_pkthdr.csum_flags & 882 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 883 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 884 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 885 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 886 goto badcsum; 887 888 case M_CSUM_TCPv4|M_CSUM_DATA: { 889 u_int32_t hw_csum = m->m_pkthdr.csum_data; 890 891 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 892 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 893 const struct ip *ip = 894 mtod(m, const struct ip *); 895 896 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 897 ip->ip_dst.s_addr, 898 htons(hw_csum + tlen + off + IPPROTO_TCP)); 899 } 900 if ((hw_csum ^ 0xffff) != 0) 901 goto badcsum; 902 break; 903 } 904 905 case M_CSUM_TCPv4: 906 /* Checksum was okay. */ 907 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 908 break; 909 910 default: 911 /* 912 * Must compute it ourselves. Maybe skip checksum 913 * on loopback interfaces. 914 */ 915 if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & 916 IFF_LOOPBACK) || 917 tcp_do_loopback_cksum)) { 918 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 919 if (in4_cksum(m, IPPROTO_TCP, toff, 920 tlen + off) != 0) 921 goto badcsum; 922 } 923 break; 924 } 925 break; 926 #endif /* INET4 */ 927 928 #ifdef INET6 929 case AF_INET6: 930 switch (m->m_pkthdr.csum_flags & 931 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 932 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 933 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 934 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 935 goto badcsum; 936 937 #if 0 /* notyet */ 938 case M_CSUM_TCPv6|M_CSUM_DATA: 939 #endif 940 941 case M_CSUM_TCPv6: 942 /* Checksum was okay. */ 943 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 944 break; 945 946 default: 947 /* 948 * Must compute it ourselves. Maybe skip checksum 949 * on loopback interfaces. 950 */ 951 if (__predict_true((m->m_flags & M_LOOP) == 0 || 952 tcp_do_loopback_cksum)) { 953 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 954 if (in6_cksum(m, IPPROTO_TCP, toff, 955 tlen + off) != 0) 956 goto badcsum; 957 } 958 } 959 break; 960 #endif /* INET6 */ 961 } 962 963 return 0; 964 965 badcsum: 966 tcpstat.tcps_rcvbadsum++; 967 return -1; 968 } 969 970 /* 971 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 972 */ 973 void 974 tcp_input(struct mbuf *m, ...) 975 { 976 struct tcphdr *th; 977 struct ip *ip; 978 struct inpcb *inp; 979 #ifdef INET6 980 struct ip6_hdr *ip6; 981 struct in6pcb *in6p; 982 #endif 983 u_int8_t *optp = NULL; 984 int optlen = 0; 985 int len, tlen, toff, hdroptlen = 0; 986 struct tcpcb *tp = 0; 987 int tiflags; 988 struct socket *so = NULL; 989 int todrop, dupseg, acked, ourfinisacked, needoutput = 0; 990 #ifdef TCP_DEBUG 991 short ostate = 0; 992 #endif 993 u_long tiwin; 994 struct tcp_opt_info opti; 995 int off, iphlen; 996 va_list ap; 997 int af; /* af on the wire */ 998 struct mbuf *tcp_saveti = NULL; 999 uint32_t ts_rtt; 1000 uint8_t iptos; 1001 1002 MCLAIM(m, &tcp_rx_mowner); 1003 va_start(ap, m); 1004 toff = va_arg(ap, int); 1005 (void)va_arg(ap, int); /* ignore value, advance ap */ 1006 va_end(ap); 1007 1008 tcpstat.tcps_rcvtotal++; 1009 1010 bzero(&opti, sizeof(opti)); 1011 opti.ts_present = 0; 1012 opti.maxseg = 0; 1013 1014 /* 1015 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 1016 * 1017 * TCP is, by definition, unicast, so we reject all 1018 * multicast outright. 1019 * 1020 * Note, there are additional src/dst address checks in 1021 * the AF-specific code below. 1022 */ 1023 if (m->m_flags & (M_BCAST|M_MCAST)) { 1024 /* XXX stat */ 1025 goto drop; 1026 } 1027 #ifdef INET6 1028 if (m->m_flags & M_ANYCAST6) { 1029 /* XXX stat */ 1030 goto drop; 1031 } 1032 #endif 1033 1034 /* 1035 * Get IP and TCP header. 1036 * Note: IP leaves IP header in first mbuf. 1037 */ 1038 ip = mtod(m, struct ip *); 1039 #ifdef INET6 1040 ip6 = NULL; 1041 #endif 1042 switch (ip->ip_v) { 1043 #ifdef INET 1044 case 4: 1045 af = AF_INET; 1046 iphlen = sizeof(struct ip); 1047 ip = mtod(m, struct ip *); 1048 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1049 sizeof(struct tcphdr)); 1050 if (th == NULL) { 1051 tcpstat.tcps_rcvshort++; 1052 return; 1053 } 1054 /* We do the checksum after PCB lookup... */ 1055 len = ntohs(ip->ip_len); 1056 tlen = len - toff; 1057 iptos = ip->ip_tos; 1058 break; 1059 #endif 1060 #ifdef INET6 1061 case 6: 1062 ip = NULL; 1063 iphlen = sizeof(struct ip6_hdr); 1064 af = AF_INET6; 1065 ip6 = mtod(m, struct ip6_hdr *); 1066 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1067 sizeof(struct tcphdr)); 1068 if (th == NULL) { 1069 tcpstat.tcps_rcvshort++; 1070 return; 1071 } 1072 1073 /* Be proactive about malicious use of IPv4 mapped address */ 1074 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 1075 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 1076 /* XXX stat */ 1077 goto drop; 1078 } 1079 1080 /* 1081 * Be proactive about unspecified IPv6 address in source. 1082 * As we use all-zero to indicate unbounded/unconnected pcb, 1083 * unspecified IPv6 address can be used to confuse us. 1084 * 1085 * Note that packets with unspecified IPv6 destination is 1086 * already dropped in ip6_input. 1087 */ 1088 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1089 /* XXX stat */ 1090 goto drop; 1091 } 1092 1093 /* 1094 * Make sure destination address is not multicast. 1095 * Source address checked in ip6_input(). 1096 */ 1097 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1098 /* XXX stat */ 1099 goto drop; 1100 } 1101 1102 /* We do the checksum after PCB lookup... */ 1103 len = m->m_pkthdr.len; 1104 tlen = len - toff; 1105 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1106 break; 1107 #endif 1108 default: 1109 m_freem(m); 1110 return; 1111 } 1112 1113 KASSERT(TCP_HDR_ALIGNED_P(th)); 1114 1115 /* 1116 * Check that TCP offset makes sense, 1117 * pull out TCP options and adjust length. XXX 1118 */ 1119 off = th->th_off << 2; 1120 if (off < sizeof (struct tcphdr) || off > tlen) { 1121 tcpstat.tcps_rcvbadoff++; 1122 goto drop; 1123 } 1124 tlen -= off; 1125 1126 /* 1127 * tcp_input() has been modified to use tlen to mean the TCP data 1128 * length throughout the function. Other functions can use 1129 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1130 * rja 1131 */ 1132 1133 if (off > sizeof (struct tcphdr)) { 1134 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1135 if (th == NULL) { 1136 tcpstat.tcps_rcvshort++; 1137 return; 1138 } 1139 /* 1140 * NOTE: ip/ip6 will not be affected by m_pulldown() 1141 * (as they're before toff) and we don't need to update those. 1142 */ 1143 KASSERT(TCP_HDR_ALIGNED_P(th)); 1144 optlen = off - sizeof (struct tcphdr); 1145 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1146 /* 1147 * Do quick retrieval of timestamp options ("options 1148 * prediction?"). If timestamp is the only option and it's 1149 * formatted as recommended in RFC 1323 appendix A, we 1150 * quickly get the values now and not bother calling 1151 * tcp_dooptions(), etc. 1152 */ 1153 if ((optlen == TCPOLEN_TSTAMP_APPA || 1154 (optlen > TCPOLEN_TSTAMP_APPA && 1155 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1156 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1157 (th->th_flags & TH_SYN) == 0) { 1158 opti.ts_present = 1; 1159 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1160 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1161 optp = NULL; /* we've parsed the options */ 1162 } 1163 } 1164 tiflags = th->th_flags; 1165 1166 /* 1167 * Locate pcb for segment. 1168 */ 1169 findpcb: 1170 inp = NULL; 1171 #ifdef INET6 1172 in6p = NULL; 1173 #endif 1174 switch (af) { 1175 #ifdef INET 1176 case AF_INET: 1177 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1178 ip->ip_dst, th->th_dport); 1179 if (inp == 0) { 1180 ++tcpstat.tcps_pcbhashmiss; 1181 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1182 } 1183 #ifdef INET6 1184 if (inp == 0) { 1185 struct in6_addr s, d; 1186 1187 /* mapped addr case */ 1188 bzero(&s, sizeof(s)); 1189 s.s6_addr16[5] = htons(0xffff); 1190 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1191 bzero(&d, sizeof(d)); 1192 d.s6_addr16[5] = htons(0xffff); 1193 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1194 in6p = in6_pcblookup_connect(&tcbtable, &s, 1195 th->th_sport, &d, th->th_dport, 0); 1196 if (in6p == 0) { 1197 ++tcpstat.tcps_pcbhashmiss; 1198 in6p = in6_pcblookup_bind(&tcbtable, &d, 1199 th->th_dport, 0); 1200 } 1201 } 1202 #endif 1203 #ifndef INET6 1204 if (inp == 0) 1205 #else 1206 if (inp == 0 && in6p == 0) 1207 #endif 1208 { 1209 ++tcpstat.tcps_noport; 1210 if (tcp_log_refused && 1211 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1212 tcp4_log_refused(ip, th); 1213 } 1214 TCP_FIELDS_TO_HOST(th); 1215 goto dropwithreset_ratelim; 1216 } 1217 #if defined(IPSEC) || defined(FAST_IPSEC) 1218 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1219 ipsec4_in_reject(m, inp)) { 1220 ipsecstat.in_polvio++; 1221 goto drop; 1222 } 1223 #ifdef INET6 1224 else if (in6p && 1225 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1226 ipsec6_in_reject_so(m, in6p->in6p_socket)) { 1227 ipsecstat.in_polvio++; 1228 goto drop; 1229 } 1230 #endif 1231 #endif /*IPSEC*/ 1232 break; 1233 #endif /*INET*/ 1234 #ifdef INET6 1235 case AF_INET6: 1236 { 1237 int faith; 1238 1239 #if defined(NFAITH) && NFAITH > 0 1240 faith = faithprefix(&ip6->ip6_dst); 1241 #else 1242 faith = 0; 1243 #endif 1244 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1245 th->th_sport, &ip6->ip6_dst, th->th_dport, faith); 1246 if (in6p == NULL) { 1247 ++tcpstat.tcps_pcbhashmiss; 1248 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1249 th->th_dport, faith); 1250 } 1251 if (in6p == NULL) { 1252 ++tcpstat.tcps_noport; 1253 if (tcp_log_refused && 1254 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1255 tcp6_log_refused(ip6, th); 1256 } 1257 TCP_FIELDS_TO_HOST(th); 1258 goto dropwithreset_ratelim; 1259 } 1260 #if defined(IPSEC) || defined(FAST_IPSEC) 1261 if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1262 ipsec6_in_reject(m, in6p)) { 1263 ipsec6stat.in_polvio++; 1264 goto drop; 1265 } 1266 #endif /*IPSEC*/ 1267 break; 1268 } 1269 #endif 1270 } 1271 1272 /* 1273 * If the state is CLOSED (i.e., TCB does not exist) then 1274 * all data in the incoming segment is discarded. 1275 * If the TCB exists but is in CLOSED state, it is embryonic, 1276 * but should either do a listen or a connect soon. 1277 */ 1278 tp = NULL; 1279 so = NULL; 1280 if (inp) { 1281 tp = intotcpcb(inp); 1282 so = inp->inp_socket; 1283 } 1284 #ifdef INET6 1285 else if (in6p) { 1286 tp = in6totcpcb(in6p); 1287 so = in6p->in6p_socket; 1288 } 1289 #endif 1290 if (tp == 0) { 1291 TCP_FIELDS_TO_HOST(th); 1292 goto dropwithreset_ratelim; 1293 } 1294 if (tp->t_state == TCPS_CLOSED) 1295 goto drop; 1296 1297 /* 1298 * Checksum extended TCP header and data. 1299 */ 1300 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1301 goto badcsum; 1302 1303 TCP_FIELDS_TO_HOST(th); 1304 1305 /* Unscale the window into a 32-bit value. */ 1306 if ((tiflags & TH_SYN) == 0) 1307 tiwin = th->th_win << tp->snd_scale; 1308 else 1309 tiwin = th->th_win; 1310 1311 #ifdef INET6 1312 /* save packet options if user wanted */ 1313 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1314 if (in6p->in6p_options) { 1315 m_freem(in6p->in6p_options); 1316 in6p->in6p_options = 0; 1317 } 1318 KASSERT(ip6 != NULL); 1319 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1320 } 1321 #endif 1322 1323 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1324 union syn_cache_sa src; 1325 union syn_cache_sa dst; 1326 1327 bzero(&src, sizeof(src)); 1328 bzero(&dst, sizeof(dst)); 1329 switch (af) { 1330 #ifdef INET 1331 case AF_INET: 1332 src.sin.sin_len = sizeof(struct sockaddr_in); 1333 src.sin.sin_family = AF_INET; 1334 src.sin.sin_addr = ip->ip_src; 1335 src.sin.sin_port = th->th_sport; 1336 1337 dst.sin.sin_len = sizeof(struct sockaddr_in); 1338 dst.sin.sin_family = AF_INET; 1339 dst.sin.sin_addr = ip->ip_dst; 1340 dst.sin.sin_port = th->th_dport; 1341 break; 1342 #endif 1343 #ifdef INET6 1344 case AF_INET6: 1345 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1346 src.sin6.sin6_family = AF_INET6; 1347 src.sin6.sin6_addr = ip6->ip6_src; 1348 src.sin6.sin6_port = th->th_sport; 1349 1350 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1351 dst.sin6.sin6_family = AF_INET6; 1352 dst.sin6.sin6_addr = ip6->ip6_dst; 1353 dst.sin6.sin6_port = th->th_dport; 1354 break; 1355 #endif /* INET6 */ 1356 default: 1357 goto badsyn; /*sanity*/ 1358 } 1359 1360 if (so->so_options & SO_DEBUG) { 1361 #ifdef TCP_DEBUG 1362 ostate = tp->t_state; 1363 #endif 1364 1365 tcp_saveti = NULL; 1366 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1367 goto nosave; 1368 1369 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1370 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1371 if (!tcp_saveti) 1372 goto nosave; 1373 } else { 1374 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1375 if (!tcp_saveti) 1376 goto nosave; 1377 MCLAIM(m, &tcp_mowner); 1378 tcp_saveti->m_len = iphlen; 1379 m_copydata(m, 0, iphlen, 1380 mtod(tcp_saveti, void *)); 1381 } 1382 1383 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1384 m_freem(tcp_saveti); 1385 tcp_saveti = NULL; 1386 } else { 1387 tcp_saveti->m_len += sizeof(struct tcphdr); 1388 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1389 sizeof(struct tcphdr)); 1390 } 1391 nosave:; 1392 } 1393 if (so->so_options & SO_ACCEPTCONN) { 1394 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1395 if (tiflags & TH_RST) { 1396 syn_cache_reset(&src.sa, &dst.sa, th); 1397 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1398 (TH_ACK|TH_SYN)) { 1399 /* 1400 * Received a SYN,ACK. This should 1401 * never happen while we are in 1402 * LISTEN. Send an RST. 1403 */ 1404 goto badsyn; 1405 } else if (tiflags & TH_ACK) { 1406 so = syn_cache_get(&src.sa, &dst.sa, 1407 th, toff, tlen, so, m); 1408 if (so == NULL) { 1409 /* 1410 * We don't have a SYN for 1411 * this ACK; send an RST. 1412 */ 1413 goto badsyn; 1414 } else if (so == 1415 (struct socket *)(-1)) { 1416 /* 1417 * We were unable to create 1418 * the connection. If the 1419 * 3-way handshake was 1420 * completed, and RST has 1421 * been sent to the peer. 1422 * Since the mbuf might be 1423 * in use for the reply, 1424 * do not free it. 1425 */ 1426 m = NULL; 1427 } else { 1428 /* 1429 * We have created a 1430 * full-blown connection. 1431 */ 1432 tp = NULL; 1433 inp = NULL; 1434 #ifdef INET6 1435 in6p = NULL; 1436 #endif 1437 switch (so->so_proto->pr_domain->dom_family) { 1438 #ifdef INET 1439 case AF_INET: 1440 inp = sotoinpcb(so); 1441 tp = intotcpcb(inp); 1442 break; 1443 #endif 1444 #ifdef INET6 1445 case AF_INET6: 1446 in6p = sotoin6pcb(so); 1447 tp = in6totcpcb(in6p); 1448 break; 1449 #endif 1450 } 1451 if (tp == NULL) 1452 goto badsyn; /*XXX*/ 1453 tiwin <<= tp->snd_scale; 1454 goto after_listen; 1455 } 1456 } else { 1457 /* 1458 * None of RST, SYN or ACK was set. 1459 * This is an invalid packet for a 1460 * TCB in LISTEN state. Send a RST. 1461 */ 1462 goto badsyn; 1463 } 1464 } else { 1465 /* 1466 * Received a SYN. 1467 * 1468 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1469 */ 1470 if (m->m_flags & (M_BCAST|M_MCAST)) 1471 goto drop; 1472 1473 switch (af) { 1474 #ifdef INET6 1475 case AF_INET6: 1476 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1477 goto drop; 1478 break; 1479 #endif /* INET6 */ 1480 case AF_INET: 1481 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1482 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1483 goto drop; 1484 break; 1485 } 1486 1487 #ifdef INET6 1488 /* 1489 * If deprecated address is forbidden, we do 1490 * not accept SYN to deprecated interface 1491 * address to prevent any new inbound 1492 * connection from getting established. 1493 * When we do not accept SYN, we send a TCP 1494 * RST, with deprecated source address (instead 1495 * of dropping it). We compromise it as it is 1496 * much better for peer to send a RST, and 1497 * RST will be the final packet for the 1498 * exchange. 1499 * 1500 * If we do not forbid deprecated addresses, we 1501 * accept the SYN packet. RFC2462 does not 1502 * suggest dropping SYN in this case. 1503 * If we decipher RFC2462 5.5.4, it says like 1504 * this: 1505 * 1. use of deprecated addr with existing 1506 * communication is okay - "SHOULD continue 1507 * to be used" 1508 * 2. use of it with new communication: 1509 * (2a) "SHOULD NOT be used if alternate 1510 * address with sufficient scope is 1511 * available" 1512 * (2b) nothing mentioned otherwise. 1513 * Here we fall into (2b) case as we have no 1514 * choice in our source address selection - we 1515 * must obey the peer. 1516 * 1517 * The wording in RFC2462 is confusing, and 1518 * there are multiple description text for 1519 * deprecated address handling - worse, they 1520 * are not exactly the same. I believe 5.5.4 1521 * is the best one, so we follow 5.5.4. 1522 */ 1523 if (af == AF_INET6 && !ip6_use_deprecated) { 1524 struct in6_ifaddr *ia6; 1525 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1526 &ip6->ip6_dst)) && 1527 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1528 tp = NULL; 1529 goto dropwithreset; 1530 } 1531 } 1532 #endif 1533 1534 #if defined(IPSEC) || defined(FAST_IPSEC) 1535 switch (af) { 1536 #ifdef INET 1537 case AF_INET: 1538 if (ipsec4_in_reject_so(m, so)) { 1539 ipsecstat.in_polvio++; 1540 tp = NULL; 1541 goto dropwithreset; 1542 } 1543 break; 1544 #endif 1545 #ifdef INET6 1546 case AF_INET6: 1547 if (ipsec6_in_reject_so(m, so)) { 1548 ipsec6stat.in_polvio++; 1549 tp = NULL; 1550 goto dropwithreset; 1551 } 1552 break; 1553 #endif /*INET6*/ 1554 } 1555 #endif /*IPSEC*/ 1556 1557 /* 1558 * LISTEN socket received a SYN 1559 * from itself? This can't possibly 1560 * be valid; drop the packet. 1561 */ 1562 if (th->th_sport == th->th_dport) { 1563 int i; 1564 1565 switch (af) { 1566 #ifdef INET 1567 case AF_INET: 1568 i = in_hosteq(ip->ip_src, ip->ip_dst); 1569 break; 1570 #endif 1571 #ifdef INET6 1572 case AF_INET6: 1573 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1574 break; 1575 #endif 1576 default: 1577 i = 1; 1578 } 1579 if (i) { 1580 tcpstat.tcps_badsyn++; 1581 goto drop; 1582 } 1583 } 1584 1585 /* 1586 * SYN looks ok; create compressed TCP 1587 * state for it. 1588 */ 1589 if (so->so_qlen <= so->so_qlimit && 1590 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1591 so, m, optp, optlen, &opti)) 1592 m = NULL; 1593 } 1594 goto drop; 1595 } 1596 } 1597 1598 after_listen: 1599 #ifdef DIAGNOSTIC 1600 /* 1601 * Should not happen now that all embryonic connections 1602 * are handled with compressed state. 1603 */ 1604 if (tp->t_state == TCPS_LISTEN) 1605 panic("tcp_input: TCPS_LISTEN"); 1606 #endif 1607 1608 /* 1609 * Segment received on connection. 1610 * Reset idle time and keep-alive timer. 1611 */ 1612 tp->t_rcvtime = tcp_now; 1613 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1614 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1615 1616 /* 1617 * Process options. 1618 */ 1619 #ifdef TCP_SIGNATURE 1620 if (optp || (tp->t_flags & TF_SIGNATURE)) 1621 #else 1622 if (optp) 1623 #endif 1624 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1625 goto drop; 1626 1627 if (TCP_SACK_ENABLED(tp)) { 1628 tcp_del_sackholes(tp, th); 1629 } 1630 1631 if (TCP_ECN_ALLOWED(tp)) { 1632 switch (iptos & IPTOS_ECN_MASK) { 1633 case IPTOS_ECN_CE: 1634 tp->t_flags |= TF_ECN_SND_ECE; 1635 tcpstat.tcps_ecn_ce++; 1636 break; 1637 case IPTOS_ECN_ECT0: 1638 tcpstat.tcps_ecn_ect++; 1639 break; 1640 case IPTOS_ECN_ECT1: 1641 /* XXX: ignore for now -- rpaulo */ 1642 break; 1643 } 1644 1645 if (tiflags & TH_CWR) 1646 tp->t_flags &= ~TF_ECN_SND_ECE; 1647 1648 /* 1649 * Congestion experienced. 1650 * Ignore if we are already trying to recover. 1651 */ 1652 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1653 tp->t_congctl->cong_exp(tp); 1654 } 1655 1656 if (opti.ts_present && opti.ts_ecr) { 1657 /* 1658 * Calculate the RTT from the returned time stamp and the 1659 * connection's time base. If the time stamp is later than 1660 * the current time, or is extremely old, fall back to non-1323 1661 * RTT calculation. Since ts_ecr is unsigned, we can test both 1662 * at the same time. 1663 */ 1664 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1665 if (ts_rtt > TCP_PAWS_IDLE) 1666 ts_rtt = 0; 1667 } else { 1668 ts_rtt = 0; 1669 } 1670 1671 /* 1672 * Header prediction: check for the two common cases 1673 * of a uni-directional data xfer. If the packet has 1674 * no control flags, is in-sequence, the window didn't 1675 * change and we're not retransmitting, it's a 1676 * candidate. If the length is zero and the ack moved 1677 * forward, we're the sender side of the xfer. Just 1678 * free the data acked & wake any higher level process 1679 * that was blocked waiting for space. If the length 1680 * is non-zero and the ack didn't move, we're the 1681 * receiver side. If we're getting packets in-order 1682 * (the reassembly queue is empty), add the data to 1683 * the socket buffer and note that we need a delayed ack. 1684 */ 1685 if (tp->t_state == TCPS_ESTABLISHED && 1686 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1687 == TH_ACK && 1688 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1689 th->th_seq == tp->rcv_nxt && 1690 tiwin && tiwin == tp->snd_wnd && 1691 tp->snd_nxt == tp->snd_max) { 1692 1693 /* 1694 * If last ACK falls within this segment's sequence numbers, 1695 * record the timestamp. 1696 * NOTE: 1697 * 1) That the test incorporates suggestions from the latest 1698 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 1699 * 2) That updating only on newer timestamps interferes with 1700 * our earlier PAWS tests, so this check should be solely 1701 * predicated on the sequence space of this segment. 1702 * 3) That we modify the segment boundary check to be 1703 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1704 * instead of RFC1323's 1705 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1706 * This modified check allows us to overcome RFC1323's 1707 * limitations as described in Stevens TCP/IP Illustrated 1708 * Vol. 2 p.869. In such cases, we can still calculate the 1709 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1710 */ 1711 if (opti.ts_present && 1712 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1713 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1714 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1715 tp->ts_recent_age = tcp_now; 1716 tp->ts_recent = opti.ts_val; 1717 } 1718 1719 if (tlen == 0) { 1720 /* Ack prediction. */ 1721 if (SEQ_GT(th->th_ack, tp->snd_una) && 1722 SEQ_LEQ(th->th_ack, tp->snd_max) && 1723 tp->snd_cwnd >= tp->snd_wnd && 1724 tp->t_partialacks < 0) { 1725 /* 1726 * this is a pure ack for outstanding data. 1727 */ 1728 ++tcpstat.tcps_predack; 1729 if (ts_rtt) 1730 tcp_xmit_timer(tp, ts_rtt); 1731 else if (tp->t_rtttime && 1732 SEQ_GT(th->th_ack, tp->t_rtseq)) 1733 tcp_xmit_timer(tp, 1734 tcp_now - tp->t_rtttime); 1735 acked = th->th_ack - tp->snd_una; 1736 tcpstat.tcps_rcvackpack++; 1737 tcpstat.tcps_rcvackbyte += acked; 1738 ND6_HINT(tp); 1739 1740 if (acked > (tp->t_lastoff - tp->t_inoff)) 1741 tp->t_lastm = NULL; 1742 sbdrop(&so->so_snd, acked); 1743 tp->t_lastoff -= acked; 1744 1745 ICMP_CHECK(tp, th, acked); 1746 1747 tp->snd_una = th->th_ack; 1748 tp->snd_fack = tp->snd_una; 1749 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1750 tp->snd_high = tp->snd_una; 1751 m_freem(m); 1752 1753 /* 1754 * If all outstanding data are acked, stop 1755 * retransmit timer, otherwise restart timer 1756 * using current (possibly backed-off) value. 1757 * If process is waiting for space, 1758 * wakeup/selwakeup/signal. If data 1759 * are ready to send, let tcp_output 1760 * decide between more output or persist. 1761 */ 1762 if (tp->snd_una == tp->snd_max) 1763 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1764 else if (TCP_TIMER_ISARMED(tp, 1765 TCPT_PERSIST) == 0) 1766 TCP_TIMER_ARM(tp, TCPT_REXMT, 1767 tp->t_rxtcur); 1768 1769 sowwakeup(so); 1770 if (so->so_snd.sb_cc) 1771 (void) tcp_output(tp); 1772 if (tcp_saveti) 1773 m_freem(tcp_saveti); 1774 return; 1775 } 1776 } else if (th->th_ack == tp->snd_una && 1777 TAILQ_FIRST(&tp->segq) == NULL && 1778 tlen <= sbspace(&so->so_rcv)) { 1779 int newsize = 0; /* automatic sockbuf scaling */ 1780 1781 /* 1782 * this is a pure, in-sequence data packet 1783 * with nothing on the reassembly queue and 1784 * we have enough buffer space to take it. 1785 */ 1786 ++tcpstat.tcps_preddat; 1787 tp->rcv_nxt += tlen; 1788 tcpstat.tcps_rcvpack++; 1789 tcpstat.tcps_rcvbyte += tlen; 1790 ND6_HINT(tp); 1791 1792 /* 1793 * Automatic sizing enables the performance of large buffers 1794 * and most of the efficiency of small ones by only allocating 1795 * space when it is needed. 1796 * 1797 * On the receive side the socket buffer memory is only rarely 1798 * used to any significant extent. This allows us to be much 1799 * more aggressive in scaling the receive socket buffer. For 1800 * the case that the buffer space is actually used to a large 1801 * extent and we run out of kernel memory we can simply drop 1802 * the new segments; TCP on the sender will just retransmit it 1803 * later. Setting the buffer size too big may only consume too 1804 * much kernel memory if the application doesn't read() from 1805 * the socket or packet loss or reordering makes use of the 1806 * reassembly queue. 1807 * 1808 * The criteria to step up the receive buffer one notch are: 1809 * 1. the number of bytes received during the time it takes 1810 * one timestamp to be reflected back to us (the RTT); 1811 * 2. received bytes per RTT is within seven eighth of the 1812 * current socket buffer size; 1813 * 3. receive buffer size has not hit maximal automatic size; 1814 * 1815 * This algorithm does one step per RTT at most and only if 1816 * we receive a bulk stream w/o packet losses or reorderings. 1817 * Shrinking the buffer during idle times is not necessary as 1818 * it doesn't consume any memory when idle. 1819 * 1820 * TODO: Only step up if the application is actually serving 1821 * the buffer to better manage the socket buffer resources. 1822 */ 1823 if (tcp_do_autorcvbuf && 1824 opti.ts_ecr && 1825 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1826 if (opti.ts_ecr > tp->rfbuf_ts && 1827 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 1828 if (tp->rfbuf_cnt > 1829 (so->so_rcv.sb_hiwat / 8 * 7) && 1830 so->so_rcv.sb_hiwat < 1831 tcp_autorcvbuf_max) { 1832 newsize = 1833 min(so->so_rcv.sb_hiwat + 1834 tcp_autorcvbuf_inc, 1835 tcp_autorcvbuf_max); 1836 } 1837 /* Start over with next RTT. */ 1838 tp->rfbuf_ts = 0; 1839 tp->rfbuf_cnt = 0; 1840 } else 1841 tp->rfbuf_cnt += tlen; /* add up */ 1842 } 1843 1844 /* 1845 * Drop TCP, IP headers and TCP options then add data 1846 * to socket buffer. 1847 */ 1848 if (so->so_state & SS_CANTRCVMORE) 1849 m_freem(m); 1850 else { 1851 /* 1852 * Set new socket buffer size. 1853 * Give up when limit is reached. 1854 */ 1855 if (newsize) 1856 if (!sbreserve(&so->so_rcv, 1857 newsize, so)) 1858 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1859 m_adj(m, toff + off); 1860 sbappendstream(&so->so_rcv, m); 1861 } 1862 sorwakeup(so); 1863 TCP_SETUP_ACK(tp, th); 1864 if (tp->t_flags & TF_ACKNOW) 1865 (void) tcp_output(tp); 1866 if (tcp_saveti) 1867 m_freem(tcp_saveti); 1868 return; 1869 } 1870 } 1871 1872 /* 1873 * Compute mbuf offset to TCP data segment. 1874 */ 1875 hdroptlen = toff + off; 1876 1877 /* 1878 * Calculate amount of space in receive window, 1879 * and then do TCP input processing. 1880 * Receive window is amount of space in rcv queue, 1881 * but not less than advertised window. 1882 */ 1883 { int win; 1884 1885 win = sbspace(&so->so_rcv); 1886 if (win < 0) 1887 win = 0; 1888 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1889 } 1890 1891 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1892 tp->rfbuf_ts = 0; 1893 tp->rfbuf_cnt = 0; 1894 1895 switch (tp->t_state) { 1896 /* 1897 * If the state is SYN_SENT: 1898 * if seg contains an ACK, but not for our SYN, drop the input. 1899 * if seg contains a RST, then drop the connection. 1900 * if seg does not contain SYN, then drop it. 1901 * Otherwise this is an acceptable SYN segment 1902 * initialize tp->rcv_nxt and tp->irs 1903 * if seg contains ack then advance tp->snd_una 1904 * if seg contains a ECE and ECN support is enabled, the stream 1905 * is ECN capable. 1906 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1907 * arrange for segment to be acked (eventually) 1908 * continue processing rest of data/controls, beginning with URG 1909 */ 1910 case TCPS_SYN_SENT: 1911 if ((tiflags & TH_ACK) && 1912 (SEQ_LEQ(th->th_ack, tp->iss) || 1913 SEQ_GT(th->th_ack, tp->snd_max))) 1914 goto dropwithreset; 1915 if (tiflags & TH_RST) { 1916 if (tiflags & TH_ACK) 1917 tp = tcp_drop(tp, ECONNREFUSED); 1918 goto drop; 1919 } 1920 if ((tiflags & TH_SYN) == 0) 1921 goto drop; 1922 if (tiflags & TH_ACK) { 1923 tp->snd_una = th->th_ack; 1924 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1925 tp->snd_nxt = tp->snd_una; 1926 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1927 tp->snd_high = tp->snd_una; 1928 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1929 1930 if ((tiflags & TH_ECE) && tcp_do_ecn) { 1931 tp->t_flags |= TF_ECN_PERMIT; 1932 tcpstat.tcps_ecn_shs++; 1933 } 1934 1935 } 1936 tp->irs = th->th_seq; 1937 tcp_rcvseqinit(tp); 1938 tp->t_flags |= TF_ACKNOW; 1939 tcp_mss_from_peer(tp, opti.maxseg); 1940 1941 /* 1942 * Initialize the initial congestion window. If we 1943 * had to retransmit the SYN, we must initialize cwnd 1944 * to 1 segment (i.e. the Loss Window). 1945 */ 1946 if (tp->t_flags & TF_SYN_REXMT) 1947 tp->snd_cwnd = tp->t_peermss; 1948 else { 1949 int ss = tcp_init_win; 1950 #ifdef INET 1951 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1952 ss = tcp_init_win_local; 1953 #endif 1954 #ifdef INET6 1955 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1956 ss = tcp_init_win_local; 1957 #endif 1958 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1959 } 1960 1961 tcp_rmx_rtt(tp); 1962 if (tiflags & TH_ACK) { 1963 tcpstat.tcps_connects++; 1964 soisconnected(so); 1965 tcp_established(tp); 1966 /* Do window scaling on this connection? */ 1967 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1968 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1969 tp->snd_scale = tp->requested_s_scale; 1970 tp->rcv_scale = tp->request_r_scale; 1971 } 1972 TCP_REASS_LOCK(tp); 1973 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1974 TCP_REASS_UNLOCK(tp); 1975 /* 1976 * if we didn't have to retransmit the SYN, 1977 * use its rtt as our initial srtt & rtt var. 1978 */ 1979 if (tp->t_rtttime) 1980 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1981 } else 1982 tp->t_state = TCPS_SYN_RECEIVED; 1983 1984 /* 1985 * Advance th->th_seq to correspond to first data byte. 1986 * If data, trim to stay within window, 1987 * dropping FIN if necessary. 1988 */ 1989 th->th_seq++; 1990 if (tlen > tp->rcv_wnd) { 1991 todrop = tlen - tp->rcv_wnd; 1992 m_adj(m, -todrop); 1993 tlen = tp->rcv_wnd; 1994 tiflags &= ~TH_FIN; 1995 tcpstat.tcps_rcvpackafterwin++; 1996 tcpstat.tcps_rcvbyteafterwin += todrop; 1997 } 1998 tp->snd_wl1 = th->th_seq - 1; 1999 tp->rcv_up = th->th_seq; 2000 goto step6; 2001 2002 /* 2003 * If the state is SYN_RECEIVED: 2004 * If seg contains an ACK, but not for our SYN, drop the input 2005 * and generate an RST. See page 36, rfc793 2006 */ 2007 case TCPS_SYN_RECEIVED: 2008 if ((tiflags & TH_ACK) && 2009 (SEQ_LEQ(th->th_ack, tp->iss) || 2010 SEQ_GT(th->th_ack, tp->snd_max))) 2011 goto dropwithreset; 2012 break; 2013 } 2014 2015 /* 2016 * States other than LISTEN or SYN_SENT. 2017 * First check timestamp, if present. 2018 * Then check that at least some bytes of segment are within 2019 * receive window. If segment begins before rcv_nxt, 2020 * drop leading data (and SYN); if nothing left, just ack. 2021 * 2022 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2023 * and it's less than ts_recent, drop it. 2024 */ 2025 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2026 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2027 2028 /* Check to see if ts_recent is over 24 days old. */ 2029 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2030 /* 2031 * Invalidate ts_recent. If this segment updates 2032 * ts_recent, the age will be reset later and ts_recent 2033 * will get a valid value. If it does not, setting 2034 * ts_recent to zero will at least satisfy the 2035 * requirement that zero be placed in the timestamp 2036 * echo reply when ts_recent isn't valid. The 2037 * age isn't reset until we get a valid ts_recent 2038 * because we don't want out-of-order segments to be 2039 * dropped when ts_recent is old. 2040 */ 2041 tp->ts_recent = 0; 2042 } else { 2043 tcpstat.tcps_rcvduppack++; 2044 tcpstat.tcps_rcvdupbyte += tlen; 2045 tcpstat.tcps_pawsdrop++; 2046 tcp_new_dsack(tp, th->th_seq, tlen); 2047 goto dropafterack; 2048 } 2049 } 2050 2051 todrop = tp->rcv_nxt - th->th_seq; 2052 dupseg = false; 2053 if (todrop > 0) { 2054 if (tiflags & TH_SYN) { 2055 tiflags &= ~TH_SYN; 2056 th->th_seq++; 2057 if (th->th_urp > 1) 2058 th->th_urp--; 2059 else { 2060 tiflags &= ~TH_URG; 2061 th->th_urp = 0; 2062 } 2063 todrop--; 2064 } 2065 if (todrop > tlen || 2066 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2067 /* 2068 * Any valid FIN or RST must be to the left of the 2069 * window. At this point the FIN or RST must be a 2070 * duplicate or out of sequence; drop it. 2071 */ 2072 if (tiflags & TH_RST) 2073 goto drop; 2074 tiflags &= ~(TH_FIN|TH_RST); 2075 /* 2076 * Send an ACK to resynchronize and drop any data. 2077 * But keep on processing for RST or ACK. 2078 */ 2079 tp->t_flags |= TF_ACKNOW; 2080 todrop = tlen; 2081 dupseg = true; 2082 tcpstat.tcps_rcvdupbyte += todrop; 2083 tcpstat.tcps_rcvduppack++; 2084 } else if ((tiflags & TH_RST) && 2085 th->th_seq != tp->last_ack_sent) { 2086 /* 2087 * Test for reset before adjusting the sequence 2088 * number for overlapping data. 2089 */ 2090 goto dropafterack_ratelim; 2091 } else { 2092 tcpstat.tcps_rcvpartduppack++; 2093 tcpstat.tcps_rcvpartdupbyte += todrop; 2094 } 2095 tcp_new_dsack(tp, th->th_seq, todrop); 2096 hdroptlen += todrop; /*drop from head afterwards*/ 2097 th->th_seq += todrop; 2098 tlen -= todrop; 2099 if (th->th_urp > todrop) 2100 th->th_urp -= todrop; 2101 else { 2102 tiflags &= ~TH_URG; 2103 th->th_urp = 0; 2104 } 2105 } 2106 2107 /* 2108 * If new data are received on a connection after the 2109 * user processes are gone, then RST the other end. 2110 */ 2111 if ((so->so_state & SS_NOFDREF) && 2112 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2113 tp = tcp_close(tp); 2114 tcpstat.tcps_rcvafterclose++; 2115 goto dropwithreset; 2116 } 2117 2118 /* 2119 * If segment ends after window, drop trailing data 2120 * (and PUSH and FIN); if nothing left, just ACK. 2121 */ 2122 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 2123 if (todrop > 0) { 2124 tcpstat.tcps_rcvpackafterwin++; 2125 if (todrop >= tlen) { 2126 /* 2127 * The segment actually starts after the window. 2128 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2129 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2130 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2131 */ 2132 tcpstat.tcps_rcvbyteafterwin += tlen; 2133 /* 2134 * If a new connection request is received 2135 * while in TIME_WAIT, drop the old connection 2136 * and start over if the sequence numbers 2137 * are above the previous ones. 2138 * 2139 * NOTE: We will checksum the packet again, and 2140 * so we need to put the header fields back into 2141 * network order! 2142 * XXX This kind of sucks, but we don't expect 2143 * XXX this to happen very often, so maybe it 2144 * XXX doesn't matter so much. 2145 */ 2146 if (tiflags & TH_SYN && 2147 tp->t_state == TCPS_TIME_WAIT && 2148 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2149 tp = tcp_close(tp); 2150 TCP_FIELDS_TO_NET(th); 2151 goto findpcb; 2152 } 2153 /* 2154 * If window is closed can only take segments at 2155 * window edge, and have to drop data and PUSH from 2156 * incoming segments. Continue processing, but 2157 * remember to ack. Otherwise, drop segment 2158 * and (if not RST) ack. 2159 */ 2160 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2161 tp->t_flags |= TF_ACKNOW; 2162 tcpstat.tcps_rcvwinprobe++; 2163 } else 2164 goto dropafterack; 2165 } else 2166 tcpstat.tcps_rcvbyteafterwin += todrop; 2167 m_adj(m, -todrop); 2168 tlen -= todrop; 2169 tiflags &= ~(TH_PUSH|TH_FIN); 2170 } 2171 2172 /* 2173 * If last ACK falls within this segment's sequence numbers, 2174 * and the timestamp is newer, record it. 2175 */ 2176 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 2177 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2178 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 2179 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2180 tp->ts_recent_age = tcp_now; 2181 tp->ts_recent = opti.ts_val; 2182 } 2183 2184 /* 2185 * If the RST bit is set examine the state: 2186 * SYN_RECEIVED STATE: 2187 * If passive open, return to LISTEN state. 2188 * If active open, inform user that connection was refused. 2189 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 2190 * Inform user that connection was reset, and close tcb. 2191 * CLOSING, LAST_ACK, TIME_WAIT STATES 2192 * Close the tcb. 2193 */ 2194 if (tiflags & TH_RST) { 2195 if (th->th_seq != tp->last_ack_sent) 2196 goto dropafterack_ratelim; 2197 2198 switch (tp->t_state) { 2199 case TCPS_SYN_RECEIVED: 2200 so->so_error = ECONNREFUSED; 2201 goto close; 2202 2203 case TCPS_ESTABLISHED: 2204 case TCPS_FIN_WAIT_1: 2205 case TCPS_FIN_WAIT_2: 2206 case TCPS_CLOSE_WAIT: 2207 so->so_error = ECONNRESET; 2208 close: 2209 tp->t_state = TCPS_CLOSED; 2210 tcpstat.tcps_drops++; 2211 tp = tcp_close(tp); 2212 goto drop; 2213 2214 case TCPS_CLOSING: 2215 case TCPS_LAST_ACK: 2216 case TCPS_TIME_WAIT: 2217 tp = tcp_close(tp); 2218 goto drop; 2219 } 2220 } 2221 2222 /* 2223 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2224 * we must be in a synchronized state. RFC791 states (under RST 2225 * generation) that any unacceptable segment (an out-of-order SYN 2226 * qualifies) received in a synchronized state must elicit only an 2227 * empty acknowledgment segment ... and the connection remains in 2228 * the same state. 2229 */ 2230 if (tiflags & TH_SYN) { 2231 if (tp->rcv_nxt == th->th_seq) { 2232 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2233 TH_ACK); 2234 if (tcp_saveti) 2235 m_freem(tcp_saveti); 2236 return; 2237 } 2238 2239 goto dropafterack_ratelim; 2240 } 2241 2242 /* 2243 * If the ACK bit is off we drop the segment and return. 2244 */ 2245 if ((tiflags & TH_ACK) == 0) { 2246 if (tp->t_flags & TF_ACKNOW) 2247 goto dropafterack; 2248 else 2249 goto drop; 2250 } 2251 2252 /* 2253 * Ack processing. 2254 */ 2255 switch (tp->t_state) { 2256 2257 /* 2258 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2259 * ESTABLISHED state and continue processing, otherwise 2260 * send an RST. 2261 */ 2262 case TCPS_SYN_RECEIVED: 2263 if (SEQ_GT(tp->snd_una, th->th_ack) || 2264 SEQ_GT(th->th_ack, tp->snd_max)) 2265 goto dropwithreset; 2266 tcpstat.tcps_connects++; 2267 soisconnected(so); 2268 tcp_established(tp); 2269 /* Do window scaling? */ 2270 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2271 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2272 tp->snd_scale = tp->requested_s_scale; 2273 tp->rcv_scale = tp->request_r_scale; 2274 } 2275 TCP_REASS_LOCK(tp); 2276 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2277 TCP_REASS_UNLOCK(tp); 2278 tp->snd_wl1 = th->th_seq - 1; 2279 /* fall into ... */ 2280 2281 /* 2282 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2283 * ACKs. If the ack is in the range 2284 * tp->snd_una < th->th_ack <= tp->snd_max 2285 * then advance tp->snd_una to th->th_ack and drop 2286 * data from the retransmission queue. If this ACK reflects 2287 * more up to date window information we update our window information. 2288 */ 2289 case TCPS_ESTABLISHED: 2290 case TCPS_FIN_WAIT_1: 2291 case TCPS_FIN_WAIT_2: 2292 case TCPS_CLOSE_WAIT: 2293 case TCPS_CLOSING: 2294 case TCPS_LAST_ACK: 2295 case TCPS_TIME_WAIT: 2296 2297 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2298 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2299 tcpstat.tcps_rcvdupack++; 2300 /* 2301 * If we have outstanding data (other than 2302 * a window probe), this is a completely 2303 * duplicate ack (ie, window info didn't 2304 * change), the ack is the biggest we've 2305 * seen and we've seen exactly our rexmt 2306 * threshhold of them, assume a packet 2307 * has been dropped and retransmit it. 2308 * Kludge snd_nxt & the congestion 2309 * window so we send only this one 2310 * packet. 2311 */ 2312 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2313 th->th_ack != tp->snd_una) 2314 tp->t_dupacks = 0; 2315 else if (tp->t_partialacks < 0 && 2316 ((!TCP_SACK_ENABLED(tp) && 2317 ++tp->t_dupacks == tcprexmtthresh) || 2318 TCP_FACK_FASTRECOV(tp))) { 2319 /* 2320 * Do the fast retransmit, and adjust 2321 * congestion control paramenters. 2322 */ 2323 if (tp->t_congctl->fast_retransmit(tp, th)) { 2324 /* False fast retransmit */ 2325 break; 2326 } else 2327 goto drop; 2328 } else if (tp->t_dupacks > tcprexmtthresh) { 2329 tp->snd_cwnd += tp->t_segsz; 2330 (void) tcp_output(tp); 2331 goto drop; 2332 } 2333 } else { 2334 /* 2335 * If the ack appears to be very old, only 2336 * allow data that is in-sequence. This 2337 * makes it somewhat more difficult to insert 2338 * forged data by guessing sequence numbers. 2339 * Sent an ack to try to update the send 2340 * sequence number on the other side. 2341 */ 2342 if (tlen && th->th_seq != tp->rcv_nxt && 2343 SEQ_LT(th->th_ack, 2344 tp->snd_una - tp->max_sndwnd)) 2345 goto dropafterack; 2346 } 2347 break; 2348 } 2349 /* 2350 * If the congestion window was inflated to account 2351 * for the other side's cached packets, retract it. 2352 */ 2353 /* XXX: make SACK have his own congestion control 2354 * struct -- rpaulo */ 2355 if (TCP_SACK_ENABLED(tp)) 2356 tcp_sack_newack(tp, th); 2357 else 2358 tp->t_congctl->fast_retransmit_newack(tp, th); 2359 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2360 tcpstat.tcps_rcvacktoomuch++; 2361 goto dropafterack; 2362 } 2363 acked = th->th_ack - tp->snd_una; 2364 tcpstat.tcps_rcvackpack++; 2365 tcpstat.tcps_rcvackbyte += acked; 2366 2367 /* 2368 * If we have a timestamp reply, update smoothed 2369 * round trip time. If no timestamp is present but 2370 * transmit timer is running and timed sequence 2371 * number was acked, update smoothed round trip time. 2372 * Since we now have an rtt measurement, cancel the 2373 * timer backoff (cf., Phil Karn's retransmit alg.). 2374 * Recompute the initial retransmit timer. 2375 */ 2376 if (ts_rtt) 2377 tcp_xmit_timer(tp, ts_rtt); 2378 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2379 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2380 2381 /* 2382 * If all outstanding data is acked, stop retransmit 2383 * timer and remember to restart (more output or persist). 2384 * If there is more data to be acked, restart retransmit 2385 * timer, using current (possibly backed-off) value. 2386 */ 2387 if (th->th_ack == tp->snd_max) { 2388 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2389 needoutput = 1; 2390 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2391 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2392 2393 /* 2394 * New data has been acked, adjust the congestion window. 2395 */ 2396 tp->t_congctl->newack(tp, th); 2397 2398 ND6_HINT(tp); 2399 if (acked > so->so_snd.sb_cc) { 2400 tp->snd_wnd -= so->so_snd.sb_cc; 2401 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2402 ourfinisacked = 1; 2403 } else { 2404 if (acked > (tp->t_lastoff - tp->t_inoff)) 2405 tp->t_lastm = NULL; 2406 sbdrop(&so->so_snd, acked); 2407 tp->t_lastoff -= acked; 2408 tp->snd_wnd -= acked; 2409 ourfinisacked = 0; 2410 } 2411 sowwakeup(so); 2412 2413 ICMP_CHECK(tp, th, acked); 2414 2415 tp->snd_una = th->th_ack; 2416 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2417 tp->snd_fack = tp->snd_una; 2418 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2419 tp->snd_nxt = tp->snd_una; 2420 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2421 tp->snd_high = tp->snd_una; 2422 2423 switch (tp->t_state) { 2424 2425 /* 2426 * In FIN_WAIT_1 STATE in addition to the processing 2427 * for the ESTABLISHED state if our FIN is now acknowledged 2428 * then enter FIN_WAIT_2. 2429 */ 2430 case TCPS_FIN_WAIT_1: 2431 if (ourfinisacked) { 2432 /* 2433 * If we can't receive any more 2434 * data, then closing user can proceed. 2435 * Starting the timer is contrary to the 2436 * specification, but if we don't get a FIN 2437 * we'll hang forever. 2438 */ 2439 if (so->so_state & SS_CANTRCVMORE) { 2440 soisdisconnected(so); 2441 if (tp->t_maxidle > 0) 2442 TCP_TIMER_ARM(tp, TCPT_2MSL, 2443 tp->t_maxidle); 2444 } 2445 tp->t_state = TCPS_FIN_WAIT_2; 2446 } 2447 break; 2448 2449 /* 2450 * In CLOSING STATE in addition to the processing for 2451 * the ESTABLISHED state if the ACK acknowledges our FIN 2452 * then enter the TIME-WAIT state, otherwise ignore 2453 * the segment. 2454 */ 2455 case TCPS_CLOSING: 2456 if (ourfinisacked) { 2457 tp->t_state = TCPS_TIME_WAIT; 2458 tcp_canceltimers(tp); 2459 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2460 soisdisconnected(so); 2461 } 2462 break; 2463 2464 /* 2465 * In LAST_ACK, we may still be waiting for data to drain 2466 * and/or to be acked, as well as for the ack of our FIN. 2467 * If our FIN is now acknowledged, delete the TCB, 2468 * enter the closed state and return. 2469 */ 2470 case TCPS_LAST_ACK: 2471 if (ourfinisacked) { 2472 tp = tcp_close(tp); 2473 goto drop; 2474 } 2475 break; 2476 2477 /* 2478 * In TIME_WAIT state the only thing that should arrive 2479 * is a retransmission of the remote FIN. Acknowledge 2480 * it and restart the finack timer. 2481 */ 2482 case TCPS_TIME_WAIT: 2483 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2484 goto dropafterack; 2485 } 2486 } 2487 2488 step6: 2489 /* 2490 * Update window information. 2491 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2492 */ 2493 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2494 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2495 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2496 /* keep track of pure window updates */ 2497 if (tlen == 0 && 2498 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2499 tcpstat.tcps_rcvwinupd++; 2500 tp->snd_wnd = tiwin; 2501 tp->snd_wl1 = th->th_seq; 2502 tp->snd_wl2 = th->th_ack; 2503 if (tp->snd_wnd > tp->max_sndwnd) 2504 tp->max_sndwnd = tp->snd_wnd; 2505 needoutput = 1; 2506 } 2507 2508 /* 2509 * Process segments with URG. 2510 */ 2511 if ((tiflags & TH_URG) && th->th_urp && 2512 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2513 /* 2514 * This is a kludge, but if we receive and accept 2515 * random urgent pointers, we'll crash in 2516 * soreceive. It's hard to imagine someone 2517 * actually wanting to send this much urgent data. 2518 */ 2519 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2520 th->th_urp = 0; /* XXX */ 2521 tiflags &= ~TH_URG; /* XXX */ 2522 goto dodata; /* XXX */ 2523 } 2524 /* 2525 * If this segment advances the known urgent pointer, 2526 * then mark the data stream. This should not happen 2527 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2528 * a FIN has been received from the remote side. 2529 * In these states we ignore the URG. 2530 * 2531 * According to RFC961 (Assigned Protocols), 2532 * the urgent pointer points to the last octet 2533 * of urgent data. We continue, however, 2534 * to consider it to indicate the first octet 2535 * of data past the urgent section as the original 2536 * spec states (in one of two places). 2537 */ 2538 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2539 tp->rcv_up = th->th_seq + th->th_urp; 2540 so->so_oobmark = so->so_rcv.sb_cc + 2541 (tp->rcv_up - tp->rcv_nxt) - 1; 2542 if (so->so_oobmark == 0) 2543 so->so_state |= SS_RCVATMARK; 2544 sohasoutofband(so); 2545 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2546 } 2547 /* 2548 * Remove out of band data so doesn't get presented to user. 2549 * This can happen independent of advancing the URG pointer, 2550 * but if two URG's are pending at once, some out-of-band 2551 * data may creep in... ick. 2552 */ 2553 if (th->th_urp <= (u_int16_t) tlen 2554 #ifdef SO_OOBINLINE 2555 && (so->so_options & SO_OOBINLINE) == 0 2556 #endif 2557 ) 2558 tcp_pulloutofband(so, th, m, hdroptlen); 2559 } else 2560 /* 2561 * If no out of band data is expected, 2562 * pull receive urgent pointer along 2563 * with the receive window. 2564 */ 2565 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2566 tp->rcv_up = tp->rcv_nxt; 2567 dodata: /* XXX */ 2568 2569 /* 2570 * Process the segment text, merging it into the TCP sequencing queue, 2571 * and arranging for acknowledgement of receipt if necessary. 2572 * This process logically involves adjusting tp->rcv_wnd as data 2573 * is presented to the user (this happens in tcp_usrreq.c, 2574 * case PRU_RCVD). If a FIN has already been received on this 2575 * connection then we just ignore the text. 2576 */ 2577 if ((tlen || (tiflags & TH_FIN)) && 2578 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2579 /* 2580 * Insert segment ti into reassembly queue of tcp with 2581 * control block tp. Return TH_FIN if reassembly now includes 2582 * a segment with FIN. The macro form does the common case 2583 * inline (segment is the next to be received on an 2584 * established connection, and the queue is empty), 2585 * avoiding linkage into and removal from the queue and 2586 * repetition of various conversions. 2587 * Set DELACK for segments received in order, but ack 2588 * immediately when segments are out of order 2589 * (so fast retransmit can work). 2590 */ 2591 /* NOTE: this was TCP_REASS() macro, but used only once */ 2592 TCP_REASS_LOCK(tp); 2593 if (th->th_seq == tp->rcv_nxt && 2594 TAILQ_FIRST(&tp->segq) == NULL && 2595 tp->t_state == TCPS_ESTABLISHED) { 2596 TCP_SETUP_ACK(tp, th); 2597 tp->rcv_nxt += tlen; 2598 tiflags = th->th_flags & TH_FIN; 2599 tcpstat.tcps_rcvpack++; 2600 tcpstat.tcps_rcvbyte += tlen; 2601 ND6_HINT(tp); 2602 if (so->so_state & SS_CANTRCVMORE) 2603 m_freem(m); 2604 else { 2605 m_adj(m, hdroptlen); 2606 sbappendstream(&(so)->so_rcv, m); 2607 } 2608 sorwakeup(so); 2609 } else { 2610 m_adj(m, hdroptlen); 2611 tiflags = tcp_reass(tp, th, m, &tlen); 2612 tp->t_flags |= TF_ACKNOW; 2613 } 2614 TCP_REASS_UNLOCK(tp); 2615 2616 /* 2617 * Note the amount of data that peer has sent into 2618 * our window, in order to estimate the sender's 2619 * buffer size. 2620 */ 2621 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2622 } else { 2623 m_freem(m); 2624 m = NULL; 2625 tiflags &= ~TH_FIN; 2626 } 2627 2628 /* 2629 * If FIN is received ACK the FIN and let the user know 2630 * that the connection is closing. Ignore a FIN received before 2631 * the connection is fully established. 2632 */ 2633 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2634 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2635 socantrcvmore(so); 2636 tp->t_flags |= TF_ACKNOW; 2637 tp->rcv_nxt++; 2638 } 2639 switch (tp->t_state) { 2640 2641 /* 2642 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2643 */ 2644 case TCPS_ESTABLISHED: 2645 tp->t_state = TCPS_CLOSE_WAIT; 2646 break; 2647 2648 /* 2649 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2650 * enter the CLOSING state. 2651 */ 2652 case TCPS_FIN_WAIT_1: 2653 tp->t_state = TCPS_CLOSING; 2654 break; 2655 2656 /* 2657 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2658 * starting the time-wait timer, turning off the other 2659 * standard timers. 2660 */ 2661 case TCPS_FIN_WAIT_2: 2662 tp->t_state = TCPS_TIME_WAIT; 2663 tcp_canceltimers(tp); 2664 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2665 soisdisconnected(so); 2666 break; 2667 2668 /* 2669 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2670 */ 2671 case TCPS_TIME_WAIT: 2672 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2673 break; 2674 } 2675 } 2676 #ifdef TCP_DEBUG 2677 if (so->so_options & SO_DEBUG) 2678 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2679 #endif 2680 2681 /* 2682 * Return any desired output. 2683 */ 2684 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2685 (void) tcp_output(tp); 2686 } 2687 if (tcp_saveti) 2688 m_freem(tcp_saveti); 2689 return; 2690 2691 badsyn: 2692 /* 2693 * Received a bad SYN. Increment counters and dropwithreset. 2694 */ 2695 tcpstat.tcps_badsyn++; 2696 tp = NULL; 2697 goto dropwithreset; 2698 2699 dropafterack: 2700 /* 2701 * Generate an ACK dropping incoming segment if it occupies 2702 * sequence space, where the ACK reflects our state. 2703 */ 2704 if (tiflags & TH_RST) 2705 goto drop; 2706 goto dropafterack2; 2707 2708 dropafterack_ratelim: 2709 /* 2710 * We may want to rate-limit ACKs against SYN/RST attack. 2711 */ 2712 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2713 tcp_ackdrop_ppslim) == 0) { 2714 /* XXX stat */ 2715 goto drop; 2716 } 2717 /* ...fall into dropafterack2... */ 2718 2719 dropafterack2: 2720 m_freem(m); 2721 tp->t_flags |= TF_ACKNOW; 2722 (void) tcp_output(tp); 2723 if (tcp_saveti) 2724 m_freem(tcp_saveti); 2725 return; 2726 2727 dropwithreset_ratelim: 2728 /* 2729 * We may want to rate-limit RSTs in certain situations, 2730 * particularly if we are sending an RST in response to 2731 * an attempt to connect to or otherwise communicate with 2732 * a port for which we have no socket. 2733 */ 2734 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2735 tcp_rst_ppslim) == 0) { 2736 /* XXX stat */ 2737 goto drop; 2738 } 2739 /* ...fall into dropwithreset... */ 2740 2741 dropwithreset: 2742 /* 2743 * Generate a RST, dropping incoming segment. 2744 * Make ACK acceptable to originator of segment. 2745 */ 2746 if (tiflags & TH_RST) 2747 goto drop; 2748 2749 switch (af) { 2750 #ifdef INET6 2751 case AF_INET6: 2752 /* For following calls to tcp_respond */ 2753 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2754 goto drop; 2755 break; 2756 #endif /* INET6 */ 2757 case AF_INET: 2758 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2759 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2760 goto drop; 2761 } 2762 2763 if (tiflags & TH_ACK) 2764 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2765 else { 2766 if (tiflags & TH_SYN) 2767 tlen++; 2768 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2769 TH_RST|TH_ACK); 2770 } 2771 if (tcp_saveti) 2772 m_freem(tcp_saveti); 2773 return; 2774 2775 badcsum: 2776 drop: 2777 /* 2778 * Drop space held by incoming segment and return. 2779 */ 2780 if (tp) { 2781 if (tp->t_inpcb) 2782 so = tp->t_inpcb->inp_socket; 2783 #ifdef INET6 2784 else if (tp->t_in6pcb) 2785 so = tp->t_in6pcb->in6p_socket; 2786 #endif 2787 else 2788 so = NULL; 2789 #ifdef TCP_DEBUG 2790 if (so && (so->so_options & SO_DEBUG) != 0) 2791 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2792 #endif 2793 } 2794 if (tcp_saveti) 2795 m_freem(tcp_saveti); 2796 m_freem(m); 2797 return; 2798 } 2799 2800 #ifdef TCP_SIGNATURE 2801 int 2802 tcp_signature_apply(void *fstate, void *data, u_int len) 2803 { 2804 2805 MD5Update(fstate, (u_char *)data, len); 2806 return (0); 2807 } 2808 2809 struct secasvar * 2810 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 2811 { 2812 struct secasvar *sav; 2813 #ifdef FAST_IPSEC 2814 union sockaddr_union dst; 2815 #endif 2816 struct ip *ip; 2817 struct ip6_hdr *ip6; 2818 2819 ip = mtod(m, struct ip *); 2820 switch (ip->ip_v) { 2821 case 4: 2822 ip = mtod(m, struct ip *); 2823 ip6 = NULL; 2824 break; 2825 case 6: 2826 ip = NULL; 2827 ip6 = mtod(m, struct ip6_hdr *); 2828 break; 2829 default: 2830 return (NULL); 2831 } 2832 2833 #ifdef FAST_IPSEC 2834 /* Extract the destination from the IP header in the mbuf. */ 2835 bzero(&dst, sizeof(union sockaddr_union)); 2836 if (ip !=NULL) { 2837 dst.sa.sa_len = sizeof(struct sockaddr_in); 2838 dst.sa.sa_family = AF_INET; 2839 dst.sin.sin_addr = ip->ip_dst; 2840 } else { 2841 dst.sa.sa_len = sizeof(struct sockaddr_in6); 2842 dst.sa.sa_family = AF_INET6; 2843 dst.sin6.sin6_addr = ip6->ip6_dst; 2844 } 2845 2846 /* 2847 * Look up an SADB entry which matches the address of the peer. 2848 */ 2849 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2850 #else 2851 if (ip) 2852 sav = key_allocsa(AF_INET, (void *)&ip->ip_src, 2853 (void *)&ip->ip_dst, IPPROTO_TCP, 2854 htonl(TCP_SIG_SPI), 0, 0); 2855 else 2856 sav = key_allocsa(AF_INET6, (void *)&ip6->ip6_src, 2857 (void *)&ip6->ip6_dst, IPPROTO_TCP, 2858 htonl(TCP_SIG_SPI), 0, 0); 2859 #endif 2860 2861 return (sav); /* freesav must be performed by caller */ 2862 } 2863 2864 int 2865 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 2866 struct secasvar *sav, char *sig) 2867 { 2868 MD5_CTX ctx; 2869 struct ip *ip; 2870 struct ipovly *ipovly; 2871 struct ip6_hdr *ip6; 2872 struct ippseudo ippseudo; 2873 struct ip6_hdr_pseudo ip6pseudo; 2874 struct tcphdr th0; 2875 int l, tcphdrlen; 2876 2877 if (sav == NULL) 2878 return (-1); 2879 2880 tcphdrlen = th->th_off * 4; 2881 2882 switch (mtod(m, struct ip *)->ip_v) { 2883 case 4: 2884 ip = mtod(m, struct ip *); 2885 ip6 = NULL; 2886 break; 2887 case 6: 2888 ip = NULL; 2889 ip6 = mtod(m, struct ip6_hdr *); 2890 break; 2891 default: 2892 return (-1); 2893 } 2894 2895 MD5Init(&ctx); 2896 2897 if (ip) { 2898 memset(&ippseudo, 0, sizeof(ippseudo)); 2899 ipovly = (struct ipovly *)ip; 2900 ippseudo.ippseudo_src = ipovly->ih_src; 2901 ippseudo.ippseudo_dst = ipovly->ih_dst; 2902 ippseudo.ippseudo_pad = 0; 2903 ippseudo.ippseudo_p = IPPROTO_TCP; 2904 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 2905 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 2906 } else { 2907 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 2908 ip6pseudo.ip6ph_src = ip6->ip6_src; 2909 in6_clearscope(&ip6pseudo.ip6ph_src); 2910 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 2911 in6_clearscope(&ip6pseudo.ip6ph_dst); 2912 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 2913 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 2914 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 2915 } 2916 2917 th0 = *th; 2918 th0.th_sum = 0; 2919 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 2920 2921 l = m->m_pkthdr.len - thoff - tcphdrlen; 2922 if (l > 0) 2923 m_apply(m, thoff + tcphdrlen, 2924 m->m_pkthdr.len - thoff - tcphdrlen, 2925 tcp_signature_apply, &ctx); 2926 2927 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2928 MD5Final(sig, &ctx); 2929 2930 return (0); 2931 } 2932 #endif 2933 2934 static int 2935 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, 2936 struct tcphdr *th, 2937 struct mbuf *m, int toff, struct tcp_opt_info *oi) 2938 { 2939 u_int16_t mss; 2940 int opt, optlen = 0; 2941 #ifdef TCP_SIGNATURE 2942 void *sigp = NULL; 2943 char sigbuf[TCP_SIGLEN]; 2944 struct secasvar *sav = NULL; 2945 #endif 2946 2947 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 2948 opt = cp[0]; 2949 if (opt == TCPOPT_EOL) 2950 break; 2951 if (opt == TCPOPT_NOP) 2952 optlen = 1; 2953 else { 2954 if (cnt < 2) 2955 break; 2956 optlen = cp[1]; 2957 if (optlen < 2 || optlen > cnt) 2958 break; 2959 } 2960 switch (opt) { 2961 2962 default: 2963 continue; 2964 2965 case TCPOPT_MAXSEG: 2966 if (optlen != TCPOLEN_MAXSEG) 2967 continue; 2968 if (!(th->th_flags & TH_SYN)) 2969 continue; 2970 if (TCPS_HAVERCVDSYN(tp->t_state)) 2971 continue; 2972 bcopy(cp + 2, &mss, sizeof(mss)); 2973 oi->maxseg = ntohs(mss); 2974 break; 2975 2976 case TCPOPT_WINDOW: 2977 if (optlen != TCPOLEN_WINDOW) 2978 continue; 2979 if (!(th->th_flags & TH_SYN)) 2980 continue; 2981 if (TCPS_HAVERCVDSYN(tp->t_state)) 2982 continue; 2983 tp->t_flags |= TF_RCVD_SCALE; 2984 tp->requested_s_scale = cp[2]; 2985 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2986 #if 0 /*XXX*/ 2987 char *p; 2988 2989 if (ip) 2990 p = ntohl(ip->ip_src); 2991 #ifdef INET6 2992 else if (ip6) 2993 p = ip6_sprintf(&ip6->ip6_src); 2994 #endif 2995 else 2996 p = "(unknown)"; 2997 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2998 "assuming %d\n", 2999 tp->requested_s_scale, p, 3000 TCP_MAX_WINSHIFT); 3001 #else 3002 log(LOG_ERR, "TCP: invalid wscale %d, " 3003 "assuming %d\n", 3004 tp->requested_s_scale, 3005 TCP_MAX_WINSHIFT); 3006 #endif 3007 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3008 } 3009 break; 3010 3011 case TCPOPT_TIMESTAMP: 3012 if (optlen != TCPOLEN_TIMESTAMP) 3013 continue; 3014 oi->ts_present = 1; 3015 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 3016 NTOHL(oi->ts_val); 3017 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 3018 NTOHL(oi->ts_ecr); 3019 3020 if (!(th->th_flags & TH_SYN)) 3021 continue; 3022 if (TCPS_HAVERCVDSYN(tp->t_state)) 3023 continue; 3024 /* 3025 * A timestamp received in a SYN makes 3026 * it ok to send timestamp requests and replies. 3027 */ 3028 tp->t_flags |= TF_RCVD_TSTMP; 3029 tp->ts_recent = oi->ts_val; 3030 tp->ts_recent_age = tcp_now; 3031 break; 3032 3033 case TCPOPT_SACK_PERMITTED: 3034 if (optlen != TCPOLEN_SACK_PERMITTED) 3035 continue; 3036 if (!(th->th_flags & TH_SYN)) 3037 continue; 3038 if (TCPS_HAVERCVDSYN(tp->t_state)) 3039 continue; 3040 if (tcp_do_sack) { 3041 tp->t_flags |= TF_SACK_PERMIT; 3042 tp->t_flags |= TF_WILL_SACK; 3043 } 3044 break; 3045 3046 case TCPOPT_SACK: 3047 tcp_sack_option(tp, th, cp, optlen); 3048 break; 3049 #ifdef TCP_SIGNATURE 3050 case TCPOPT_SIGNATURE: 3051 if (optlen != TCPOLEN_SIGNATURE) 3052 continue; 3053 if (sigp && bcmp(sigp, cp + 2, TCP_SIGLEN)) 3054 return (-1); 3055 3056 sigp = sigbuf; 3057 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3058 tp->t_flags |= TF_SIGNATURE; 3059 break; 3060 #endif 3061 } 3062 } 3063 3064 #ifdef TCP_SIGNATURE 3065 if (tp->t_flags & TF_SIGNATURE) { 3066 3067 sav = tcp_signature_getsav(m, th); 3068 3069 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3070 return (-1); 3071 } 3072 3073 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 3074 if (sav == NULL) 3075 return (-1); 3076 #ifdef FAST_IPSEC 3077 KEY_FREESAV(&sav); 3078 #else 3079 key_freesav(sav); 3080 #endif 3081 return (-1); 3082 } 3083 3084 if (sigp) { 3085 char sig[TCP_SIGLEN]; 3086 3087 TCP_FIELDS_TO_NET(th); 3088 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3089 TCP_FIELDS_TO_HOST(th); 3090 if (sav == NULL) 3091 return (-1); 3092 #ifdef FAST_IPSEC 3093 KEY_FREESAV(&sav); 3094 #else 3095 key_freesav(sav); 3096 #endif 3097 return (-1); 3098 } 3099 TCP_FIELDS_TO_HOST(th); 3100 3101 if (bcmp(sig, sigp, TCP_SIGLEN)) { 3102 tcpstat.tcps_badsig++; 3103 if (sav == NULL) 3104 return (-1); 3105 #ifdef FAST_IPSEC 3106 KEY_FREESAV(&sav); 3107 #else 3108 key_freesav(sav); 3109 #endif 3110 return (-1); 3111 } else 3112 tcpstat.tcps_goodsig++; 3113 3114 key_sa_recordxfer(sav, m); 3115 #ifdef FAST_IPSEC 3116 KEY_FREESAV(&sav); 3117 #else 3118 key_freesav(sav); 3119 #endif 3120 } 3121 #endif 3122 3123 return (0); 3124 } 3125 3126 /* 3127 * Pull out of band byte out of a segment so 3128 * it doesn't appear in the user's data queue. 3129 * It is still reflected in the segment length for 3130 * sequencing purposes. 3131 */ 3132 void 3133 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3134 struct mbuf *m, int off) 3135 { 3136 int cnt = off + th->th_urp - 1; 3137 3138 while (cnt >= 0) { 3139 if (m->m_len > cnt) { 3140 char *cp = mtod(m, char *) + cnt; 3141 struct tcpcb *tp = sototcpcb(so); 3142 3143 tp->t_iobc = *cp; 3144 tp->t_oobflags |= TCPOOB_HAVEDATA; 3145 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3146 m->m_len--; 3147 return; 3148 } 3149 cnt -= m->m_len; 3150 m = m->m_next; 3151 if (m == 0) 3152 break; 3153 } 3154 panic("tcp_pulloutofband"); 3155 } 3156 3157 /* 3158 * Collect new round-trip time estimate 3159 * and update averages and current timeout. 3160 */ 3161 void 3162 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3163 { 3164 int32_t delta; 3165 3166 tcpstat.tcps_rttupdated++; 3167 if (tp->t_srtt != 0) { 3168 /* 3169 * srtt is stored as fixed point with 3 bits after the 3170 * binary point (i.e., scaled by 8). The following magic 3171 * is equivalent to the smoothing algorithm in rfc793 with 3172 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3173 * point). Adjust rtt to origin 0. 3174 */ 3175 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3176 if ((tp->t_srtt += delta) <= 0) 3177 tp->t_srtt = 1 << 2; 3178 /* 3179 * We accumulate a smoothed rtt variance (actually, a 3180 * smoothed mean difference), then set the retransmit 3181 * timer to smoothed rtt + 4 times the smoothed variance. 3182 * rttvar is stored as fixed point with 2 bits after the 3183 * binary point (scaled by 4). The following is 3184 * equivalent to rfc793 smoothing with an alpha of .75 3185 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3186 * rfc793's wired-in beta. 3187 */ 3188 if (delta < 0) 3189 delta = -delta; 3190 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3191 if ((tp->t_rttvar += delta) <= 0) 3192 tp->t_rttvar = 1 << 2; 3193 } else { 3194 /* 3195 * No rtt measurement yet - use the unsmoothed rtt. 3196 * Set the variance to half the rtt (so our first 3197 * retransmit happens at 3*rtt). 3198 */ 3199 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3200 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3201 } 3202 tp->t_rtttime = 0; 3203 tp->t_rxtshift = 0; 3204 3205 /* 3206 * the retransmit should happen at rtt + 4 * rttvar. 3207 * Because of the way we do the smoothing, srtt and rttvar 3208 * will each average +1/2 tick of bias. When we compute 3209 * the retransmit timer, we want 1/2 tick of rounding and 3210 * 1 extra tick because of +-1/2 tick uncertainty in the 3211 * firing of the timer. The bias will give us exactly the 3212 * 1.5 tick we need. But, because the bias is 3213 * statistical, we have to test that we don't drop below 3214 * the minimum feasible timer (which is 2 ticks). 3215 */ 3216 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3217 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3218 3219 /* 3220 * We received an ack for a packet that wasn't retransmitted; 3221 * it is probably safe to discard any error indications we've 3222 * received recently. This isn't quite right, but close enough 3223 * for now (a route might have failed after we sent a segment, 3224 * and the return path might not be symmetrical). 3225 */ 3226 tp->t_softerror = 0; 3227 } 3228 3229 3230 /* 3231 * TCP compressed state engine. Currently used to hold compressed 3232 * state for SYN_RECEIVED. 3233 */ 3234 3235 u_long syn_cache_count; 3236 u_int32_t syn_hash1, syn_hash2; 3237 3238 #define SYN_HASH(sa, sp, dp) \ 3239 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3240 ((u_int32_t)(sp)))^syn_hash2))) 3241 #ifndef INET6 3242 #define SYN_HASHALL(hash, src, dst) \ 3243 do { \ 3244 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3245 ((const struct sockaddr_in *)(src))->sin_port, \ 3246 ((const struct sockaddr_in *)(dst))->sin_port); \ 3247 } while (/*CONSTCOND*/ 0) 3248 #else 3249 #define SYN_HASH6(sa, sp, dp) \ 3250 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3251 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3252 & 0x7fffffff) 3253 3254 #define SYN_HASHALL(hash, src, dst) \ 3255 do { \ 3256 switch ((src)->sa_family) { \ 3257 case AF_INET: \ 3258 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3259 ((const struct sockaddr_in *)(src))->sin_port, \ 3260 ((const struct sockaddr_in *)(dst))->sin_port); \ 3261 break; \ 3262 case AF_INET6: \ 3263 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3264 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3265 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3266 break; \ 3267 default: \ 3268 hash = 0; \ 3269 } \ 3270 } while (/*CONSTCOND*/0) 3271 #endif /* INET6 */ 3272 3273 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL, 3274 IPL_SOFTNET); 3275 3276 /* 3277 * We don't estimate RTT with SYNs, so each packet starts with the default 3278 * RTT and each timer step has a fixed timeout value. 3279 */ 3280 #define SYN_CACHE_TIMER_ARM(sc) \ 3281 do { \ 3282 TCPT_RANGESET((sc)->sc_rxtcur, \ 3283 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3284 TCPTV_REXMTMAX); \ 3285 callout_reset(&(sc)->sc_timer, \ 3286 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3287 } while (/*CONSTCOND*/0) 3288 3289 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3290 3291 static inline void 3292 syn_cache_rm(struct syn_cache *sc) 3293 { 3294 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3295 sc, sc_bucketq); 3296 sc->sc_tp = NULL; 3297 LIST_REMOVE(sc, sc_tpq); 3298 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3299 callout_stop(&sc->sc_timer); 3300 syn_cache_count--; 3301 } 3302 3303 static inline void 3304 syn_cache_put(struct syn_cache *sc) 3305 { 3306 if (sc->sc_ipopts) 3307 (void) m_free(sc->sc_ipopts); 3308 rtcache_free(&sc->sc_route); 3309 if (callout_invoking(&sc->sc_timer)) 3310 sc->sc_flags |= SCF_DEAD; 3311 else { 3312 callout_destroy(&sc->sc_timer); 3313 pool_put(&syn_cache_pool, sc); 3314 } 3315 } 3316 3317 void 3318 syn_cache_init(void) 3319 { 3320 int i; 3321 3322 /* Initialize the hash buckets. */ 3323 for (i = 0; i < tcp_syn_cache_size; i++) 3324 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3325 } 3326 3327 void 3328 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3329 { 3330 struct syn_cache_head *scp; 3331 struct syn_cache *sc2; 3332 int s; 3333 3334 /* 3335 * If there are no entries in the hash table, reinitialize 3336 * the hash secrets. 3337 */ 3338 if (syn_cache_count == 0) { 3339 syn_hash1 = arc4random(); 3340 syn_hash2 = arc4random(); 3341 } 3342 3343 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3344 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3345 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3346 3347 /* 3348 * Make sure that we don't overflow the per-bucket 3349 * limit or the total cache size limit. 3350 */ 3351 s = splsoftnet(); 3352 if (scp->sch_length >= tcp_syn_bucket_limit) { 3353 tcpstat.tcps_sc_bucketoverflow++; 3354 /* 3355 * The bucket is full. Toss the oldest element in the 3356 * bucket. This will be the first entry in the bucket. 3357 */ 3358 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3359 #ifdef DIAGNOSTIC 3360 /* 3361 * This should never happen; we should always find an 3362 * entry in our bucket. 3363 */ 3364 if (sc2 == NULL) 3365 panic("syn_cache_insert: bucketoverflow: impossible"); 3366 #endif 3367 syn_cache_rm(sc2); 3368 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3369 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3370 struct syn_cache_head *scp2, *sce; 3371 3372 tcpstat.tcps_sc_overflowed++; 3373 /* 3374 * The cache is full. Toss the oldest entry in the 3375 * first non-empty bucket we can find. 3376 * 3377 * XXX We would really like to toss the oldest 3378 * entry in the cache, but we hope that this 3379 * condition doesn't happen very often. 3380 */ 3381 scp2 = scp; 3382 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3383 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3384 for (++scp2; scp2 != scp; scp2++) { 3385 if (scp2 >= sce) 3386 scp2 = &tcp_syn_cache[0]; 3387 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3388 break; 3389 } 3390 #ifdef DIAGNOSTIC 3391 /* 3392 * This should never happen; we should always find a 3393 * non-empty bucket. 3394 */ 3395 if (scp2 == scp) 3396 panic("syn_cache_insert: cacheoverflow: " 3397 "impossible"); 3398 #endif 3399 } 3400 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3401 syn_cache_rm(sc2); 3402 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3403 } 3404 3405 /* 3406 * Initialize the entry's timer. 3407 */ 3408 sc->sc_rxttot = 0; 3409 sc->sc_rxtshift = 0; 3410 SYN_CACHE_TIMER_ARM(sc); 3411 3412 /* Link it from tcpcb entry */ 3413 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3414 3415 /* Put it into the bucket. */ 3416 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3417 scp->sch_length++; 3418 syn_cache_count++; 3419 3420 tcpstat.tcps_sc_added++; 3421 splx(s); 3422 } 3423 3424 /* 3425 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3426 * If we have retransmitted an entry the maximum number of times, expire 3427 * that entry. 3428 */ 3429 void 3430 syn_cache_timer(void *arg) 3431 { 3432 struct syn_cache *sc = arg; 3433 int s; 3434 3435 s = splsoftnet(); 3436 callout_ack(&sc->sc_timer); 3437 3438 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3439 tcpstat.tcps_sc_delayed_free++; 3440 callout_destroy(&sc->sc_timer); 3441 pool_put(&syn_cache_pool, sc); 3442 splx(s); 3443 return; 3444 } 3445 3446 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3447 /* Drop it -- too many retransmissions. */ 3448 goto dropit; 3449 } 3450 3451 /* 3452 * Compute the total amount of time this entry has 3453 * been on a queue. If this entry has been on longer 3454 * than the keep alive timer would allow, expire it. 3455 */ 3456 sc->sc_rxttot += sc->sc_rxtcur; 3457 if (sc->sc_rxttot >= tcp_keepinit) 3458 goto dropit; 3459 3460 tcpstat.tcps_sc_retransmitted++; 3461 (void) syn_cache_respond(sc, NULL); 3462 3463 /* Advance the timer back-off. */ 3464 sc->sc_rxtshift++; 3465 SYN_CACHE_TIMER_ARM(sc); 3466 3467 splx(s); 3468 return; 3469 3470 dropit: 3471 tcpstat.tcps_sc_timed_out++; 3472 syn_cache_rm(sc); 3473 syn_cache_put(sc); /* calls pool_put but see spl above */ 3474 splx(s); 3475 } 3476 3477 /* 3478 * Remove syn cache created by the specified tcb entry, 3479 * because this does not make sense to keep them 3480 * (if there's no tcb entry, syn cache entry will never be used) 3481 */ 3482 void 3483 syn_cache_cleanup(struct tcpcb *tp) 3484 { 3485 struct syn_cache *sc, *nsc; 3486 int s; 3487 3488 s = splsoftnet(); 3489 3490 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3491 nsc = LIST_NEXT(sc, sc_tpq); 3492 3493 #ifdef DIAGNOSTIC 3494 if (sc->sc_tp != tp) 3495 panic("invalid sc_tp in syn_cache_cleanup"); 3496 #endif 3497 syn_cache_rm(sc); 3498 syn_cache_put(sc); /* calls pool_put but see spl above */ 3499 } 3500 /* just for safety */ 3501 LIST_INIT(&tp->t_sc); 3502 3503 splx(s); 3504 } 3505 3506 /* 3507 * Find an entry in the syn cache. 3508 */ 3509 struct syn_cache * 3510 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3511 struct syn_cache_head **headp) 3512 { 3513 struct syn_cache *sc; 3514 struct syn_cache_head *scp; 3515 u_int32_t hash; 3516 int s; 3517 3518 SYN_HASHALL(hash, src, dst); 3519 3520 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3521 *headp = scp; 3522 s = splsoftnet(); 3523 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3524 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3525 if (sc->sc_hash != hash) 3526 continue; 3527 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3528 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3529 splx(s); 3530 return (sc); 3531 } 3532 } 3533 splx(s); 3534 return (NULL); 3535 } 3536 3537 /* 3538 * This function gets called when we receive an ACK for a 3539 * socket in the LISTEN state. We look up the connection 3540 * in the syn cache, and if its there, we pull it out of 3541 * the cache and turn it into a full-blown connection in 3542 * the SYN-RECEIVED state. 3543 * 3544 * The return values may not be immediately obvious, and their effects 3545 * can be subtle, so here they are: 3546 * 3547 * NULL SYN was not found in cache; caller should drop the 3548 * packet and send an RST. 3549 * 3550 * -1 We were unable to create the new connection, and are 3551 * aborting it. An ACK,RST is being sent to the peer 3552 * (unless we got screwey sequence numbners; see below), 3553 * because the 3-way handshake has been completed. Caller 3554 * should not free the mbuf, since we may be using it. If 3555 * we are not, we will free it. 3556 * 3557 * Otherwise, the return value is a pointer to the new socket 3558 * associated with the connection. 3559 */ 3560 struct socket * 3561 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3562 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3563 struct socket *so, struct mbuf *m) 3564 { 3565 struct syn_cache *sc; 3566 struct syn_cache_head *scp; 3567 struct inpcb *inp = NULL; 3568 #ifdef INET6 3569 struct in6pcb *in6p = NULL; 3570 #endif 3571 struct tcpcb *tp = 0; 3572 struct mbuf *am; 3573 int s; 3574 struct socket *oso; 3575 3576 s = splsoftnet(); 3577 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3578 splx(s); 3579 return (NULL); 3580 } 3581 3582 /* 3583 * Verify the sequence and ack numbers. Try getting the correct 3584 * response again. 3585 */ 3586 if ((th->th_ack != sc->sc_iss + 1) || 3587 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3588 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3589 (void) syn_cache_respond(sc, m); 3590 splx(s); 3591 return ((struct socket *)(-1)); 3592 } 3593 3594 /* Remove this cache entry */ 3595 syn_cache_rm(sc); 3596 splx(s); 3597 3598 /* 3599 * Ok, create the full blown connection, and set things up 3600 * as they would have been set up if we had created the 3601 * connection when the SYN arrived. If we can't create 3602 * the connection, abort it. 3603 */ 3604 /* 3605 * inp still has the OLD in_pcb stuff, set the 3606 * v6-related flags on the new guy, too. This is 3607 * done particularly for the case where an AF_INET6 3608 * socket is bound only to a port, and a v4 connection 3609 * comes in on that port. 3610 * we also copy the flowinfo from the original pcb 3611 * to the new one. 3612 */ 3613 oso = so; 3614 so = sonewconn(so, SS_ISCONNECTED); 3615 if (so == NULL) 3616 goto resetandabort; 3617 3618 switch (so->so_proto->pr_domain->dom_family) { 3619 #ifdef INET 3620 case AF_INET: 3621 inp = sotoinpcb(so); 3622 break; 3623 #endif 3624 #ifdef INET6 3625 case AF_INET6: 3626 in6p = sotoin6pcb(so); 3627 break; 3628 #endif 3629 } 3630 switch (src->sa_family) { 3631 #ifdef INET 3632 case AF_INET: 3633 if (inp) { 3634 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3635 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3636 inp->inp_options = ip_srcroute(); 3637 in_pcbstate(inp, INP_BOUND); 3638 if (inp->inp_options == NULL) { 3639 inp->inp_options = sc->sc_ipopts; 3640 sc->sc_ipopts = NULL; 3641 } 3642 } 3643 #ifdef INET6 3644 else if (in6p) { 3645 /* IPv4 packet to AF_INET6 socket */ 3646 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3647 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3648 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3649 &in6p->in6p_laddr.s6_addr32[3], 3650 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3651 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3652 in6totcpcb(in6p)->t_family = AF_INET; 3653 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3654 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3655 else 3656 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3657 in6_pcbstate(in6p, IN6P_BOUND); 3658 } 3659 #endif 3660 break; 3661 #endif 3662 #ifdef INET6 3663 case AF_INET6: 3664 if (in6p) { 3665 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3666 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3667 in6_pcbstate(in6p, IN6P_BOUND); 3668 } 3669 break; 3670 #endif 3671 } 3672 #ifdef INET6 3673 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3674 struct in6pcb *oin6p = sotoin6pcb(oso); 3675 /* inherit socket options from the listening socket */ 3676 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3677 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3678 m_freem(in6p->in6p_options); 3679 in6p->in6p_options = 0; 3680 } 3681 ip6_savecontrol(in6p, &in6p->in6p_options, 3682 mtod(m, struct ip6_hdr *), m); 3683 } 3684 #endif 3685 3686 #if defined(IPSEC) || defined(FAST_IPSEC) 3687 /* 3688 * we make a copy of policy, instead of sharing the policy, 3689 * for better behavior in terms of SA lookup and dead SA removal. 3690 */ 3691 if (inp) { 3692 /* copy old policy into new socket's */ 3693 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3694 printf("tcp_input: could not copy policy\n"); 3695 } 3696 #ifdef INET6 3697 else if (in6p) { 3698 /* copy old policy into new socket's */ 3699 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3700 in6p->in6p_sp)) 3701 printf("tcp_input: could not copy policy\n"); 3702 } 3703 #endif 3704 #endif 3705 3706 /* 3707 * Give the new socket our cached route reference. 3708 */ 3709 if (inp) { 3710 rtcache_copy(&inp->inp_route, &sc->sc_route); 3711 rtcache_free(&sc->sc_route); 3712 } 3713 #ifdef INET6 3714 else { 3715 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 3716 rtcache_free(&sc->sc_route); 3717 } 3718 #endif 3719 3720 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3721 if (am == NULL) 3722 goto resetandabort; 3723 MCLAIM(am, &tcp_mowner); 3724 am->m_len = src->sa_len; 3725 bcopy(src, mtod(am, void *), src->sa_len); 3726 if (inp) { 3727 if (in_pcbconnect(inp, am, &lwp0)) { 3728 (void) m_free(am); 3729 goto resetandabort; 3730 } 3731 } 3732 #ifdef INET6 3733 else if (in6p) { 3734 if (src->sa_family == AF_INET) { 3735 /* IPv4 packet to AF_INET6 socket */ 3736 struct sockaddr_in6 *sin6; 3737 sin6 = mtod(am, struct sockaddr_in6 *); 3738 am->m_len = sizeof(*sin6); 3739 bzero(sin6, sizeof(*sin6)); 3740 sin6->sin6_family = AF_INET6; 3741 sin6->sin6_len = sizeof(*sin6); 3742 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3743 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3744 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3745 &sin6->sin6_addr.s6_addr32[3], 3746 sizeof(sin6->sin6_addr.s6_addr32[3])); 3747 } 3748 if (in6_pcbconnect(in6p, am, NULL)) { 3749 (void) m_free(am); 3750 goto resetandabort; 3751 } 3752 } 3753 #endif 3754 else { 3755 (void) m_free(am); 3756 goto resetandabort; 3757 } 3758 (void) m_free(am); 3759 3760 if (inp) 3761 tp = intotcpcb(inp); 3762 #ifdef INET6 3763 else if (in6p) 3764 tp = in6totcpcb(in6p); 3765 #endif 3766 else 3767 tp = NULL; 3768 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3769 if (sc->sc_request_r_scale != 15) { 3770 tp->requested_s_scale = sc->sc_requested_s_scale; 3771 tp->request_r_scale = sc->sc_request_r_scale; 3772 tp->snd_scale = sc->sc_requested_s_scale; 3773 tp->rcv_scale = sc->sc_request_r_scale; 3774 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3775 } 3776 if (sc->sc_flags & SCF_TIMESTAMP) 3777 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3778 tp->ts_timebase = sc->sc_timebase; 3779 3780 tp->t_template = tcp_template(tp); 3781 if (tp->t_template == 0) { 3782 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3783 so = NULL; 3784 m_freem(m); 3785 goto abort; 3786 } 3787 3788 tp->iss = sc->sc_iss; 3789 tp->irs = sc->sc_irs; 3790 tcp_sendseqinit(tp); 3791 tcp_rcvseqinit(tp); 3792 tp->t_state = TCPS_SYN_RECEIVED; 3793 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 3794 tcpstat.tcps_accepts++; 3795 3796 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 3797 tp->t_flags |= TF_WILL_SACK; 3798 3799 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 3800 tp->t_flags |= TF_ECN_PERMIT; 3801 3802 #ifdef TCP_SIGNATURE 3803 if (sc->sc_flags & SCF_SIGNATURE) 3804 tp->t_flags |= TF_SIGNATURE; 3805 #endif 3806 3807 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3808 tp->t_ourmss = sc->sc_ourmaxseg; 3809 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3810 3811 /* 3812 * Initialize the initial congestion window. If we 3813 * had to retransmit the SYN,ACK, we must initialize cwnd 3814 * to 1 segment (i.e. the Loss Window). 3815 */ 3816 if (sc->sc_rxtshift) 3817 tp->snd_cwnd = tp->t_peermss; 3818 else { 3819 int ss = tcp_init_win; 3820 #ifdef INET 3821 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3822 ss = tcp_init_win_local; 3823 #endif 3824 #ifdef INET6 3825 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3826 ss = tcp_init_win_local; 3827 #endif 3828 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3829 } 3830 3831 tcp_rmx_rtt(tp); 3832 tp->snd_wl1 = sc->sc_irs; 3833 tp->rcv_up = sc->sc_irs + 1; 3834 3835 /* 3836 * This is what whould have happened in tcp_output() when 3837 * the SYN,ACK was sent. 3838 */ 3839 tp->snd_up = tp->snd_una; 3840 tp->snd_max = tp->snd_nxt = tp->iss+1; 3841 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3842 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3843 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3844 tp->last_ack_sent = tp->rcv_nxt; 3845 tp->t_partialacks = -1; 3846 tp->t_dupacks = 0; 3847 3848 tcpstat.tcps_sc_completed++; 3849 s = splsoftnet(); 3850 syn_cache_put(sc); 3851 splx(s); 3852 return (so); 3853 3854 resetandabort: 3855 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3856 abort: 3857 if (so != NULL) 3858 (void) soabort(so); 3859 s = splsoftnet(); 3860 syn_cache_put(sc); 3861 splx(s); 3862 tcpstat.tcps_sc_aborted++; 3863 return ((struct socket *)(-1)); 3864 } 3865 3866 /* 3867 * This function is called when we get a RST for a 3868 * non-existent connection, so that we can see if the 3869 * connection is in the syn cache. If it is, zap it. 3870 */ 3871 3872 void 3873 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 3874 { 3875 struct syn_cache *sc; 3876 struct syn_cache_head *scp; 3877 int s = splsoftnet(); 3878 3879 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3880 splx(s); 3881 return; 3882 } 3883 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3884 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3885 splx(s); 3886 return; 3887 } 3888 syn_cache_rm(sc); 3889 tcpstat.tcps_sc_reset++; 3890 syn_cache_put(sc); /* calls pool_put but see spl above */ 3891 splx(s); 3892 } 3893 3894 void 3895 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 3896 struct tcphdr *th) 3897 { 3898 struct syn_cache *sc; 3899 struct syn_cache_head *scp; 3900 int s; 3901 3902 s = splsoftnet(); 3903 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3904 splx(s); 3905 return; 3906 } 3907 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3908 if (ntohl (th->th_seq) != sc->sc_iss) { 3909 splx(s); 3910 return; 3911 } 3912 3913 /* 3914 * If we've retransmitted 3 times and this is our second error, 3915 * we remove the entry. Otherwise, we allow it to continue on. 3916 * This prevents us from incorrectly nuking an entry during a 3917 * spurious network outage. 3918 * 3919 * See tcp_notify(). 3920 */ 3921 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3922 sc->sc_flags |= SCF_UNREACH; 3923 splx(s); 3924 return; 3925 } 3926 3927 syn_cache_rm(sc); 3928 tcpstat.tcps_sc_unreach++; 3929 syn_cache_put(sc); /* calls pool_put but see spl above */ 3930 splx(s); 3931 } 3932 3933 /* 3934 * Given a LISTEN socket and an inbound SYN request, add 3935 * this to the syn cache, and send back a segment: 3936 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3937 * to the source. 3938 * 3939 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3940 * Doing so would require that we hold onto the data and deliver it 3941 * to the application. However, if we are the target of a SYN-flood 3942 * DoS attack, an attacker could send data which would eventually 3943 * consume all available buffer space if it were ACKed. By not ACKing 3944 * the data, we avoid this DoS scenario. 3945 */ 3946 3947 int 3948 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3949 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 3950 int optlen, struct tcp_opt_info *oi) 3951 { 3952 struct tcpcb tb, *tp; 3953 long win; 3954 struct syn_cache *sc; 3955 struct syn_cache_head *scp; 3956 struct mbuf *ipopts; 3957 struct tcp_opt_info opti; 3958 int s; 3959 3960 tp = sototcpcb(so); 3961 3962 bzero(&opti, sizeof(opti)); 3963 3964 /* 3965 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3966 * 3967 * Note this check is performed in tcp_input() very early on. 3968 */ 3969 3970 /* 3971 * Initialize some local state. 3972 */ 3973 win = sbspace(&so->so_rcv); 3974 if (win > TCP_MAXWIN) 3975 win = TCP_MAXWIN; 3976 3977 switch (src->sa_family) { 3978 #ifdef INET 3979 case AF_INET: 3980 /* 3981 * Remember the IP options, if any. 3982 */ 3983 ipopts = ip_srcroute(); 3984 break; 3985 #endif 3986 default: 3987 ipopts = NULL; 3988 } 3989 3990 #ifdef TCP_SIGNATURE 3991 if (optp || (tp->t_flags & TF_SIGNATURE)) 3992 #else 3993 if (optp) 3994 #endif 3995 { 3996 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3997 #ifdef TCP_SIGNATURE 3998 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 3999 #endif 4000 tb.t_state = TCPS_LISTEN; 4001 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 4002 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 4003 return (0); 4004 } else 4005 tb.t_flags = 0; 4006 4007 /* 4008 * See if we already have an entry for this connection. 4009 * If we do, resend the SYN,ACK. We do not count this 4010 * as a retransmission (XXX though maybe we should). 4011 */ 4012 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4013 tcpstat.tcps_sc_dupesyn++; 4014 if (ipopts) { 4015 /* 4016 * If we were remembering a previous source route, 4017 * forget it and use the new one we've been given. 4018 */ 4019 if (sc->sc_ipopts) 4020 (void) m_free(sc->sc_ipopts); 4021 sc->sc_ipopts = ipopts; 4022 } 4023 sc->sc_timestamp = tb.ts_recent; 4024 if (syn_cache_respond(sc, m) == 0) { 4025 tcpstat.tcps_sndacks++; 4026 tcpstat.tcps_sndtotal++; 4027 } 4028 return (1); 4029 } 4030 4031 s = splsoftnet(); 4032 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4033 splx(s); 4034 if (sc == NULL) { 4035 if (ipopts) 4036 (void) m_free(ipopts); 4037 return (0); 4038 } 4039 4040 /* 4041 * Fill in the cache, and put the necessary IP and TCP 4042 * options into the reply. 4043 */ 4044 bzero(sc, sizeof(struct syn_cache)); 4045 callout_init(&sc->sc_timer, 0); 4046 bcopy(src, &sc->sc_src, src->sa_len); 4047 bcopy(dst, &sc->sc_dst, dst->sa_len); 4048 sc->sc_flags = 0; 4049 sc->sc_ipopts = ipopts; 4050 sc->sc_irs = th->th_seq; 4051 switch (src->sa_family) { 4052 #ifdef INET 4053 case AF_INET: 4054 { 4055 struct sockaddr_in *srcin = (void *) src; 4056 struct sockaddr_in *dstin = (void *) dst; 4057 4058 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4059 &srcin->sin_addr, dstin->sin_port, 4060 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4061 break; 4062 } 4063 #endif /* INET */ 4064 #ifdef INET6 4065 case AF_INET6: 4066 { 4067 struct sockaddr_in6 *srcin6 = (void *) src; 4068 struct sockaddr_in6 *dstin6 = (void *) dst; 4069 4070 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4071 &srcin6->sin6_addr, dstin6->sin6_port, 4072 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4073 break; 4074 } 4075 #endif /* INET6 */ 4076 } 4077 sc->sc_peermaxseg = oi->maxseg; 4078 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4079 m->m_pkthdr.rcvif : NULL, 4080 sc->sc_src.sa.sa_family); 4081 sc->sc_win = win; 4082 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 4083 sc->sc_timestamp = tb.ts_recent; 4084 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4085 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4086 sc->sc_flags |= SCF_TIMESTAMP; 4087 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4088 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4089 sc->sc_requested_s_scale = tb.requested_s_scale; 4090 sc->sc_request_r_scale = 0; 4091 /* 4092 * Pick the smallest possible scaling factor that 4093 * will still allow us to scale up to sb_max. 4094 * 4095 * We do this because there are broken firewalls that 4096 * will corrupt the window scale option, leading to 4097 * the other endpoint believing that our advertised 4098 * window is unscaled. At scale factors larger than 4099 * 5 the unscaled window will drop below 1500 bytes, 4100 * leading to serious problems when traversing these 4101 * broken firewalls. 4102 * 4103 * With the default sbmax of 256K, a scale factor 4104 * of 3 will be chosen by this algorithm. Those who 4105 * choose a larger sbmax should watch out 4106 * for the compatiblity problems mentioned above. 4107 * 4108 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4109 * or <SYN,ACK>) segment itself is never scaled. 4110 */ 4111 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4112 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4113 sc->sc_request_r_scale++; 4114 } else { 4115 sc->sc_requested_s_scale = 15; 4116 sc->sc_request_r_scale = 15; 4117 } 4118 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4119 sc->sc_flags |= SCF_SACK_PERMIT; 4120 4121 /* 4122 * ECN setup packet recieved. 4123 */ 4124 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4125 sc->sc_flags |= SCF_ECN_PERMIT; 4126 4127 #ifdef TCP_SIGNATURE 4128 if (tb.t_flags & TF_SIGNATURE) 4129 sc->sc_flags |= SCF_SIGNATURE; 4130 #endif 4131 sc->sc_tp = tp; 4132 if (syn_cache_respond(sc, m) == 0) { 4133 syn_cache_insert(sc, tp); 4134 tcpstat.tcps_sndacks++; 4135 tcpstat.tcps_sndtotal++; 4136 } else { 4137 s = splsoftnet(); 4138 syn_cache_put(sc); 4139 splx(s); 4140 tcpstat.tcps_sc_dropped++; 4141 } 4142 return (1); 4143 } 4144 4145 int 4146 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4147 { 4148 struct route *ro; 4149 u_int8_t *optp; 4150 int optlen, error; 4151 u_int16_t tlen; 4152 struct ip *ip = NULL; 4153 #ifdef INET6 4154 struct ip6_hdr *ip6 = NULL; 4155 #endif 4156 struct tcpcb *tp = NULL; 4157 struct tcphdr *th; 4158 u_int hlen; 4159 struct socket *so; 4160 4161 ro = &sc->sc_route; 4162 switch (sc->sc_src.sa.sa_family) { 4163 case AF_INET: 4164 hlen = sizeof(struct ip); 4165 break; 4166 #ifdef INET6 4167 case AF_INET6: 4168 hlen = sizeof(struct ip6_hdr); 4169 break; 4170 #endif 4171 default: 4172 if (m) 4173 m_freem(m); 4174 return (EAFNOSUPPORT); 4175 } 4176 4177 /* Compute the size of the TCP options. */ 4178 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4179 ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) + 4180 #ifdef TCP_SIGNATURE 4181 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4182 #endif 4183 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4184 4185 tlen = hlen + sizeof(struct tcphdr) + optlen; 4186 4187 /* 4188 * Create the IP+TCP header from scratch. 4189 */ 4190 if (m) 4191 m_freem(m); 4192 #ifdef DIAGNOSTIC 4193 if (max_linkhdr + tlen > MCLBYTES) 4194 return (ENOBUFS); 4195 #endif 4196 MGETHDR(m, M_DONTWAIT, MT_DATA); 4197 if (m && tlen > MHLEN) { 4198 MCLGET(m, M_DONTWAIT); 4199 if ((m->m_flags & M_EXT) == 0) { 4200 m_freem(m); 4201 m = NULL; 4202 } 4203 } 4204 if (m == NULL) 4205 return (ENOBUFS); 4206 MCLAIM(m, &tcp_tx_mowner); 4207 4208 /* Fixup the mbuf. */ 4209 m->m_data += max_linkhdr; 4210 m->m_len = m->m_pkthdr.len = tlen; 4211 if (sc->sc_tp) { 4212 tp = sc->sc_tp; 4213 if (tp->t_inpcb) 4214 so = tp->t_inpcb->inp_socket; 4215 #ifdef INET6 4216 else if (tp->t_in6pcb) 4217 so = tp->t_in6pcb->in6p_socket; 4218 #endif 4219 else 4220 so = NULL; 4221 } else 4222 so = NULL; 4223 m->m_pkthdr.rcvif = NULL; 4224 memset(mtod(m, u_char *), 0, tlen); 4225 4226 switch (sc->sc_src.sa.sa_family) { 4227 case AF_INET: 4228 ip = mtod(m, struct ip *); 4229 ip->ip_v = 4; 4230 ip->ip_dst = sc->sc_src.sin.sin_addr; 4231 ip->ip_src = sc->sc_dst.sin.sin_addr; 4232 ip->ip_p = IPPROTO_TCP; 4233 th = (struct tcphdr *)(ip + 1); 4234 th->th_dport = sc->sc_src.sin.sin_port; 4235 th->th_sport = sc->sc_dst.sin.sin_port; 4236 break; 4237 #ifdef INET6 4238 case AF_INET6: 4239 ip6 = mtod(m, struct ip6_hdr *); 4240 ip6->ip6_vfc = IPV6_VERSION; 4241 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4242 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4243 ip6->ip6_nxt = IPPROTO_TCP; 4244 /* ip6_plen will be updated in ip6_output() */ 4245 th = (struct tcphdr *)(ip6 + 1); 4246 th->th_dport = sc->sc_src.sin6.sin6_port; 4247 th->th_sport = sc->sc_dst.sin6.sin6_port; 4248 break; 4249 #endif 4250 default: 4251 th = NULL; 4252 } 4253 4254 th->th_seq = htonl(sc->sc_iss); 4255 th->th_ack = htonl(sc->sc_irs + 1); 4256 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4257 th->th_flags = TH_SYN|TH_ACK; 4258 th->th_win = htons(sc->sc_win); 4259 /* th_sum already 0 */ 4260 /* th_urp already 0 */ 4261 4262 /* Tack on the TCP options. */ 4263 optp = (u_int8_t *)(th + 1); 4264 *optp++ = TCPOPT_MAXSEG; 4265 *optp++ = 4; 4266 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4267 *optp++ = sc->sc_ourmaxseg & 0xff; 4268 4269 if (sc->sc_request_r_scale != 15) { 4270 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4271 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4272 sc->sc_request_r_scale); 4273 optp += 4; 4274 } 4275 4276 if (sc->sc_flags & SCF_TIMESTAMP) { 4277 u_int32_t *lp = (u_int32_t *)(optp); 4278 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4279 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4280 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4281 *lp = htonl(sc->sc_timestamp); 4282 optp += TCPOLEN_TSTAMP_APPA; 4283 } 4284 4285 if (sc->sc_flags & SCF_SACK_PERMIT) { 4286 u_int8_t *p = optp; 4287 4288 /* Let the peer know that we will SACK. */ 4289 p[0] = TCPOPT_SACK_PERMITTED; 4290 p[1] = 2; 4291 p[2] = TCPOPT_NOP; 4292 p[3] = TCPOPT_NOP; 4293 optp += 4; 4294 } 4295 4296 /* 4297 * Send ECN SYN-ACK setup packet. 4298 * Routes can be asymetric, so, even if we receive a packet 4299 * with ECE and CWR set, we must not assume no one will block 4300 * the ECE packet we are about to send. 4301 */ 4302 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4303 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4304 th->th_flags |= TH_ECE; 4305 tcpstat.tcps_ecn_shs++; 4306 4307 /* 4308 * draft-ietf-tcpm-ecnsyn-00.txt 4309 * 4310 * "[...] a TCP node MAY respond to an ECN-setup 4311 * SYN packet by setting ECT in the responding 4312 * ECN-setup SYN/ACK packet, indicating to routers 4313 * that the SYN/ACK packet is ECN-Capable. 4314 * This allows a congested router along the path 4315 * to mark the packet instead of dropping the 4316 * packet as an indication of congestion." 4317 * 4318 * "[...] There can be a great benefit in setting 4319 * an ECN-capable codepoint in SYN/ACK packets [...] 4320 * Congestion is most likely to occur in 4321 * the server-to-client direction. As a result, 4322 * setting an ECN-capable codepoint in SYN/ACK 4323 * packets can reduce the occurence of three-second 4324 * retransmit timeouts resulting from the drop 4325 * of SYN/ACK packets." 4326 * 4327 * Page 4 and 6, January 2006. 4328 */ 4329 4330 switch (sc->sc_src.sa.sa_family) { 4331 #ifdef INET 4332 case AF_INET: 4333 ip->ip_tos |= IPTOS_ECN_ECT0; 4334 break; 4335 #endif 4336 #ifdef INET6 4337 case AF_INET6: 4338 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4339 break; 4340 #endif 4341 } 4342 tcpstat.tcps_ecn_ect++; 4343 } 4344 4345 #ifdef TCP_SIGNATURE 4346 if (sc->sc_flags & SCF_SIGNATURE) { 4347 struct secasvar *sav; 4348 u_int8_t *sigp; 4349 4350 sav = tcp_signature_getsav(m, th); 4351 4352 if (sav == NULL) { 4353 if (m) 4354 m_freem(m); 4355 return (EPERM); 4356 } 4357 4358 *optp++ = TCPOPT_SIGNATURE; 4359 *optp++ = TCPOLEN_SIGNATURE; 4360 sigp = optp; 4361 bzero(optp, TCP_SIGLEN); 4362 optp += TCP_SIGLEN; 4363 *optp++ = TCPOPT_NOP; 4364 *optp++ = TCPOPT_EOL; 4365 4366 (void)tcp_signature(m, th, hlen, sav, sigp); 4367 4368 key_sa_recordxfer(sav, m); 4369 #ifdef FAST_IPSEC 4370 KEY_FREESAV(&sav); 4371 #else 4372 key_freesav(sav); 4373 #endif 4374 } 4375 #endif 4376 4377 /* Compute the packet's checksum. */ 4378 switch (sc->sc_src.sa.sa_family) { 4379 case AF_INET: 4380 ip->ip_len = htons(tlen - hlen); 4381 th->th_sum = 0; 4382 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4383 break; 4384 #ifdef INET6 4385 case AF_INET6: 4386 ip6->ip6_plen = htons(tlen - hlen); 4387 th->th_sum = 0; 4388 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4389 break; 4390 #endif 4391 } 4392 4393 /* 4394 * Fill in some straggling IP bits. Note the stack expects 4395 * ip_len to be in host order, for convenience. 4396 */ 4397 switch (sc->sc_src.sa.sa_family) { 4398 #ifdef INET 4399 case AF_INET: 4400 ip->ip_len = htons(tlen); 4401 ip->ip_ttl = ip_defttl; 4402 /* XXX tos? */ 4403 break; 4404 #endif 4405 #ifdef INET6 4406 case AF_INET6: 4407 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4408 ip6->ip6_vfc |= IPV6_VERSION; 4409 ip6->ip6_plen = htons(tlen - hlen); 4410 /* ip6_hlim will be initialized afterwards */ 4411 /* XXX flowlabel? */ 4412 break; 4413 #endif 4414 } 4415 4416 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4417 tp = sc->sc_tp; 4418 4419 switch (sc->sc_src.sa.sa_family) { 4420 #ifdef INET 4421 case AF_INET: 4422 error = ip_output(m, sc->sc_ipopts, ro, 4423 (ip_mtudisc ? IP_MTUDISC : 0), 4424 (struct ip_moptions *)NULL, so); 4425 break; 4426 #endif 4427 #ifdef INET6 4428 case AF_INET6: 4429 ip6->ip6_hlim = in6_selecthlim(NULL, 4430 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 4431 4432 error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, so, NULL); 4433 break; 4434 #endif 4435 default: 4436 error = EAFNOSUPPORT; 4437 break; 4438 } 4439 return (error); 4440 } 4441
Indexes created Tue Sep 23 13:09:54 GMT 2025