tcp_input.c revision 1.256
1 /* $NetBSD: tcp_input.c,v 1.256 2006/12/06 09:08:27 yamt Exp $ */ 2 3 /* 4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the project nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 /* 33 * @(#)COPYRIGHT 1.1 (NRL) 17 January 1995 34 * 35 * NRL grants permission for redistribution and use in source and binary 36 * forms, with or without modification, of the software and documentation 37 * created at NRL provided that the following conditions are met: 38 * 39 * 1. Redistributions of source code must retain the above copyright 40 * notice, this list of conditions and the following disclaimer. 41 * 2. Redistributions in binary form must reproduce the above copyright 42 * notice, this list of conditions and the following disclaimer in the 43 * documentation and/or other materials provided with the distribution. 44 * 3. All advertising materials mentioning features or use of this software 45 * must display the following acknowledgements: 46 * This product includes software developed by the University of 47 * California, Berkeley and its contributors. 48 * This product includes software developed at the Information 49 * Technology Division, US Naval Research Laboratory. 50 * 4. Neither the name of the NRL nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS 55 * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 56 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A 57 * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NRL OR 58 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 59 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 60 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 61 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 62 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 63 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 64 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 65 * 66 * The views and conclusions contained in the software and documentation 67 * are those of the authors and should not be interpreted as representing 68 * official policies, either expressed or implied, of the US Naval 69 * Research Laboratory (NRL). 70 */ 71 72 /*- 73 * Copyright (c) 1997, 1998, 1999, 2001, 2005, 2006 The NetBSD Foundation, Inc. 74 * All rights reserved. 75 * 76 * This code is derived from software contributed to The NetBSD Foundation 77 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 78 * Facility, NASA Ames Research Center. 79 * This code is derived from software contributed to The NetBSD Foundation 80 * by Charles M. Hannum. 81 * This code is derived from software contributed to The NetBSD Foundation 82 * by Rui Paulo. 83 * 84 * Redistribution and use in source and binary forms, with or without 85 * modification, are permitted provided that the following conditions 86 * are met: 87 * 1. Redistributions of source code must retain the above copyright 88 * notice, this list of conditions and the following disclaimer. 89 * 2. Redistributions in binary form must reproduce the above copyright 90 * notice, this list of conditions and the following disclaimer in the 91 * documentation and/or other materials provided with the distribution. 92 * 3. All advertising materials mentioning features or use of this software 93 * must display the following acknowledgement: 94 * This product includes software developed by the NetBSD 95 * Foundation, Inc. and its contributors. 96 * 4. Neither the name of The NetBSD Foundation nor the names of its 97 * contributors may be used to endorse or promote products derived 98 * from this software without specific prior written permission. 99 * 100 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 101 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 102 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 103 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 104 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 105 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 106 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 107 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 108 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 109 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 110 * POSSIBILITY OF SUCH DAMAGE. 111 */ 112 113 /* 114 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 115 * The Regents of the University of California. All rights reserved. 116 * 117 * Redistribution and use in source and binary forms, with or without 118 * modification, are permitted provided that the following conditions 119 * are met: 120 * 1. Redistributions of source code must retain the above copyright 121 * notice, this list of conditions and the following disclaimer. 122 * 2. Redistributions in binary form must reproduce the above copyright 123 * notice, this list of conditions and the following disclaimer in the 124 * documentation and/or other materials provided with the distribution. 125 * 3. Neither the name of the University nor the names of its contributors 126 * may be used to endorse or promote products derived from this software 127 * without specific prior written permission. 128 * 129 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 130 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 131 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 132 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 133 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 134 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 135 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 136 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 137 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 138 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 139 * SUCH DAMAGE. 140 * 141 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 142 */ 143 144 /* 145 * TODO list for SYN cache stuff: 146 * 147 * Find room for a "state" field, which is needed to keep a 148 * compressed state for TIME_WAIT TCBs. It's been noted already 149 * that this is fairly important for very high-volume web and 150 * mail servers, which use a large number of short-lived 151 * connections. 152 */ 153 154 #include <sys/cdefs.h> 155 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.256 2006/12/06 09:08:27 yamt Exp $"); 156 157 #include "opt_inet.h" 158 #include "opt_ipsec.h" 159 #include "opt_inet_csum.h" 160 #include "opt_tcp_debug.h" 161 162 #include <sys/param.h> 163 #include <sys/systm.h> 164 #include <sys/malloc.h> 165 #include <sys/mbuf.h> 166 #include <sys/protosw.h> 167 #include <sys/socket.h> 168 #include <sys/socketvar.h> 169 #include <sys/errno.h> 170 #include <sys/syslog.h> 171 #include <sys/pool.h> 172 #include <sys/domain.h> 173 #include <sys/kernel.h> 174 #ifdef TCP_SIGNATURE 175 #include <sys/md5.h> 176 #endif 177 178 #include <net/if.h> 179 #include <net/route.h> 180 #include <net/if_types.h> 181 182 #include <netinet/in.h> 183 #include <netinet/in_systm.h> 184 #include <netinet/ip.h> 185 #include <netinet/in_pcb.h> 186 #include <netinet/in_var.h> 187 #include <netinet/ip_var.h> 188 #include <netinet/in_offload.h> 189 190 #ifdef INET6 191 #ifndef INET 192 #include <netinet/in.h> 193 #endif 194 #include <netinet/ip6.h> 195 #include <netinet6/ip6_var.h> 196 #include <netinet6/in6_pcb.h> 197 #include <netinet6/ip6_var.h> 198 #include <netinet6/in6_var.h> 199 #include <netinet/icmp6.h> 200 #include <netinet6/nd6.h> 201 #ifdef TCP_SIGNATURE 202 #include <netinet6/scope6_var.h> 203 #endif 204 #endif 205 206 #ifndef INET6 207 /* always need ip6.h for IP6_EXTHDR_GET */ 208 #include <netinet/ip6.h> 209 #endif 210 211 #include <netinet/tcp.h> 212 #include <netinet/tcp_fsm.h> 213 #include <netinet/tcp_seq.h> 214 #include <netinet/tcp_timer.h> 215 #include <netinet/tcp_var.h> 216 #include <netinet/tcpip.h> 217 #include <netinet/tcp_congctl.h> 218 #include <netinet/tcp_debug.h> 219 220 #include <machine/stdarg.h> 221 222 #ifdef IPSEC 223 #include <netinet6/ipsec.h> 224 #include <netkey/key.h> 225 #endif /*IPSEC*/ 226 #ifdef INET6 227 #include "faith.h" 228 #if defined(NFAITH) && NFAITH > 0 229 #include <net/if_faith.h> 230 #endif 231 #endif /* IPSEC */ 232 233 #ifdef FAST_IPSEC 234 #include <netipsec/ipsec.h> 235 #include <netipsec/ipsec_var.h> /* XXX ipsecstat namespace */ 236 #include <netipsec/key.h> 237 #ifdef INET6 238 #include <netipsec/ipsec6.h> 239 #endif 240 #endif /* FAST_IPSEC*/ 241 242 int tcprexmtthresh = 3; 243 int tcp_log_refused; 244 245 static int tcp_rst_ppslim_count = 0; 246 static struct timeval tcp_rst_ppslim_last; 247 static int tcp_ackdrop_ppslim_count = 0; 248 static struct timeval tcp_ackdrop_ppslim_last; 249 250 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 251 252 /* for modulo comparisons of timestamps */ 253 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 254 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 255 256 /* 257 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 258 */ 259 #ifdef INET6 260 #define ND6_HINT(tp) \ 261 do { \ 262 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 && \ 263 tp->t_in6pcb->in6p_route.ro_rt) { \ 264 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \ 265 } \ 266 } while (/*CONSTCOND*/ 0) 267 #else 268 #define ND6_HINT(tp) 269 #endif 270 271 /* 272 * Macro to compute ACK transmission behavior. Delay the ACK unless 273 * we have already delayed an ACK (must send an ACK every two segments). 274 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 275 * option is enabled. 276 */ 277 #define TCP_SETUP_ACK(tp, th) \ 278 do { \ 279 if ((tp)->t_flags & TF_DELACK || \ 280 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 281 tp->t_flags |= TF_ACKNOW; \ 282 else \ 283 TCP_SET_DELACK(tp); \ 284 } while (/*CONSTCOND*/ 0) 285 286 #define ICMP_CHECK(tp, th, acked) \ 287 do { \ 288 /* \ 289 * If we had a pending ICMP message that \ 290 * refers to data that have just been \ 291 * acknowledged, disregard the recorded ICMP \ 292 * message. \ 293 */ \ 294 if (((tp)->t_flags & TF_PMTUD_PEND) && \ 295 SEQ_GT((th)->th_ack, (tp)->t_pmtud_th_seq)) \ 296 (tp)->t_flags &= ~TF_PMTUD_PEND; \ 297 \ 298 /* \ 299 * Keep track of the largest chunk of data \ 300 * acknowledged since last PMTU update \ 301 */ \ 302 if ((tp)->t_pmtud_mss_acked < (acked)) \ 303 (tp)->t_pmtud_mss_acked = (acked); \ 304 } while (/*CONSTCOND*/ 0) 305 306 /* 307 * Convert TCP protocol fields to host order for easier processing. 308 */ 309 #define TCP_FIELDS_TO_HOST(th) \ 310 do { \ 311 NTOHL((th)->th_seq); \ 312 NTOHL((th)->th_ack); \ 313 NTOHS((th)->th_win); \ 314 NTOHS((th)->th_urp); \ 315 } while (/*CONSTCOND*/ 0) 316 317 /* 318 * ... and reverse the above. 319 */ 320 #define TCP_FIELDS_TO_NET(th) \ 321 do { \ 322 HTONL((th)->th_seq); \ 323 HTONL((th)->th_ack); \ 324 HTONS((th)->th_win); \ 325 HTONS((th)->th_urp); \ 326 } while (/*CONSTCOND*/ 0) 327 328 #ifdef TCP_CSUM_COUNTERS 329 #include <sys/device.h> 330 331 #if defined(INET) 332 extern struct evcnt tcp_hwcsum_ok; 333 extern struct evcnt tcp_hwcsum_bad; 334 extern struct evcnt tcp_hwcsum_data; 335 extern struct evcnt tcp_swcsum; 336 #endif /* defined(INET) */ 337 #if defined(INET6) 338 extern struct evcnt tcp6_hwcsum_ok; 339 extern struct evcnt tcp6_hwcsum_bad; 340 extern struct evcnt tcp6_hwcsum_data; 341 extern struct evcnt tcp6_swcsum; 342 #endif /* defined(INET6) */ 343 344 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 345 346 #else 347 348 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 349 350 #endif /* TCP_CSUM_COUNTERS */ 351 352 #ifdef TCP_REASS_COUNTERS 353 #include <sys/device.h> 354 355 extern struct evcnt tcp_reass_; 356 extern struct evcnt tcp_reass_empty; 357 extern struct evcnt tcp_reass_iteration[8]; 358 extern struct evcnt tcp_reass_prependfirst; 359 extern struct evcnt tcp_reass_prepend; 360 extern struct evcnt tcp_reass_insert; 361 extern struct evcnt tcp_reass_inserttail; 362 extern struct evcnt tcp_reass_append; 363 extern struct evcnt tcp_reass_appendtail; 364 extern struct evcnt tcp_reass_overlaptail; 365 extern struct evcnt tcp_reass_overlapfront; 366 extern struct evcnt tcp_reass_segdup; 367 extern struct evcnt tcp_reass_fragdup; 368 369 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 370 371 #else 372 373 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 374 375 #endif /* TCP_REASS_COUNTERS */ 376 377 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *, 378 int *); 379 static int tcp_dooptions(struct tcpcb *, const u_char *, int, 380 const struct tcphdr *, struct mbuf *, int, struct tcp_opt_info *); 381 382 #ifdef INET 383 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 384 #endif 385 #ifdef INET6 386 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 387 #endif 388 389 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 390 391 static POOL_INIT(tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", 392 NULL); 393 394 struct ipqent * 395 tcpipqent_alloc() 396 { 397 struct ipqent *ipqe; 398 int s; 399 400 s = splvm(); 401 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 402 splx(s); 403 404 return ipqe; 405 } 406 407 void 408 tcpipqent_free(struct ipqent *ipqe) 409 { 410 int s; 411 412 s = splvm(); 413 pool_put(&tcpipqent_pool, ipqe); 414 splx(s); 415 } 416 417 static int 418 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int *tlen) 419 { 420 struct ipqent *p, *q, *nq, *tiqe = NULL; 421 struct socket *so = NULL; 422 int pkt_flags; 423 tcp_seq pkt_seq; 424 unsigned pkt_len; 425 u_long rcvpartdupbyte = 0; 426 u_long rcvoobyte; 427 #ifdef TCP_REASS_COUNTERS 428 u_int count = 0; 429 #endif 430 431 if (tp->t_inpcb) 432 so = tp->t_inpcb->inp_socket; 433 #ifdef INET6 434 else if (tp->t_in6pcb) 435 so = tp->t_in6pcb->in6p_socket; 436 #endif 437 438 TCP_REASS_LOCK_CHECK(tp); 439 440 /* 441 * Call with th==0 after become established to 442 * force pre-ESTABLISHED data up to user socket. 443 */ 444 if (th == 0) 445 goto present; 446 447 rcvoobyte = *tlen; 448 /* 449 * Copy these to local variables because the tcpiphdr 450 * gets munged while we are collapsing mbufs. 451 */ 452 pkt_seq = th->th_seq; 453 pkt_len = *tlen; 454 pkt_flags = th->th_flags; 455 456 TCP_REASS_COUNTER_INCR(&tcp_reass_); 457 458 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 459 /* 460 * When we miss a packet, the vast majority of time we get 461 * packets that follow it in order. So optimize for that. 462 */ 463 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 464 p->ipqe_len += pkt_len; 465 p->ipqe_flags |= pkt_flags; 466 m_cat(p->ipre_mlast, m); 467 TRAVERSE(p->ipre_mlast); 468 m = NULL; 469 tiqe = p; 470 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 471 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 472 goto skip_replacement; 473 } 474 /* 475 * While we're here, if the pkt is completely beyond 476 * anything we have, just insert it at the tail. 477 */ 478 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 479 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 480 goto insert_it; 481 } 482 } 483 484 q = TAILQ_FIRST(&tp->segq); 485 486 if (q != NULL) { 487 /* 488 * If this segment immediately precedes the first out-of-order 489 * block, simply slap the segment in front of it and (mostly) 490 * skip the complicated logic. 491 */ 492 if (pkt_seq + pkt_len == q->ipqe_seq) { 493 q->ipqe_seq = pkt_seq; 494 q->ipqe_len += pkt_len; 495 q->ipqe_flags |= pkt_flags; 496 m_cat(m, q->ipqe_m); 497 q->ipqe_m = m; 498 q->ipre_mlast = m; /* last mbuf may have changed */ 499 TRAVERSE(q->ipre_mlast); 500 tiqe = q; 501 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 502 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 503 goto skip_replacement; 504 } 505 } else { 506 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 507 } 508 509 /* 510 * Find a segment which begins after this one does. 511 */ 512 for (p = NULL; q != NULL; q = nq) { 513 nq = TAILQ_NEXT(q, ipqe_q); 514 #ifdef TCP_REASS_COUNTERS 515 count++; 516 #endif 517 /* 518 * If the received segment is just right after this 519 * fragment, merge the two together and then check 520 * for further overlaps. 521 */ 522 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 523 #ifdef TCPREASS_DEBUG 524 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 525 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 526 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 527 #endif 528 pkt_len += q->ipqe_len; 529 pkt_flags |= q->ipqe_flags; 530 pkt_seq = q->ipqe_seq; 531 m_cat(q->ipre_mlast, m); 532 TRAVERSE(q->ipre_mlast); 533 m = q->ipqe_m; 534 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 535 goto free_ipqe; 536 } 537 /* 538 * If the received segment is completely past this 539 * fragment, we need to go the next fragment. 540 */ 541 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 542 p = q; 543 continue; 544 } 545 /* 546 * If the fragment is past the received segment, 547 * it (or any following) can't be concatenated. 548 */ 549 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 550 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 551 break; 552 } 553 554 /* 555 * We've received all the data in this segment before. 556 * mark it as a duplicate and return. 557 */ 558 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 559 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 560 tcpstat.tcps_rcvduppack++; 561 tcpstat.tcps_rcvdupbyte += pkt_len; 562 tcp_new_dsack(tp, pkt_seq, pkt_len); 563 m_freem(m); 564 if (tiqe != NULL) { 565 tcpipqent_free(tiqe); 566 } 567 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 568 return (0); 569 } 570 /* 571 * Received segment completely overlaps this fragment 572 * so we drop the fragment (this keeps the temporal 573 * ordering of segments correct). 574 */ 575 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 576 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 577 rcvpartdupbyte += q->ipqe_len; 578 m_freem(q->ipqe_m); 579 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 580 goto free_ipqe; 581 } 582 /* 583 * RX'ed segment extends past the end of the 584 * fragment. Drop the overlapping bytes. Then 585 * merge the fragment and segment then treat as 586 * a longer received packet. 587 */ 588 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 589 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 590 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 591 #ifdef TCPREASS_DEBUG 592 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 593 tp, overlap, 594 pkt_seq, pkt_seq + pkt_len, pkt_len); 595 #endif 596 m_adj(m, overlap); 597 rcvpartdupbyte += overlap; 598 m_cat(q->ipre_mlast, m); 599 TRAVERSE(q->ipre_mlast); 600 m = q->ipqe_m; 601 pkt_seq = q->ipqe_seq; 602 pkt_len += q->ipqe_len - overlap; 603 rcvoobyte -= overlap; 604 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 605 goto free_ipqe; 606 } 607 /* 608 * RX'ed segment extends past the front of the 609 * fragment. Drop the overlapping bytes on the 610 * received packet. The packet will then be 611 * contatentated with this fragment a bit later. 612 */ 613 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 614 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 615 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 616 #ifdef TCPREASS_DEBUG 617 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 618 tp, overlap, 619 pkt_seq, pkt_seq + pkt_len, pkt_len); 620 #endif 621 m_adj(m, -overlap); 622 pkt_len -= overlap; 623 rcvpartdupbyte += overlap; 624 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 625 rcvoobyte -= overlap; 626 } 627 /* 628 * If the received segment immediates precedes this 629 * fragment then tack the fragment onto this segment 630 * and reinsert the data. 631 */ 632 if (q->ipqe_seq == pkt_seq + pkt_len) { 633 #ifdef TCPREASS_DEBUG 634 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 635 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 636 pkt_seq, pkt_seq + pkt_len, pkt_len); 637 #endif 638 pkt_len += q->ipqe_len; 639 pkt_flags |= q->ipqe_flags; 640 m_cat(m, q->ipqe_m); 641 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 642 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 643 tp->t_segqlen--; 644 KASSERT(tp->t_segqlen >= 0); 645 KASSERT(tp->t_segqlen != 0 || 646 (TAILQ_EMPTY(&tp->segq) && 647 TAILQ_EMPTY(&tp->timeq))); 648 if (tiqe == NULL) { 649 tiqe = q; 650 } else { 651 tcpipqent_free(q); 652 } 653 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 654 break; 655 } 656 /* 657 * If the fragment is before the segment, remember it. 658 * When this loop is terminated, p will contain the 659 * pointer to fragment that is right before the received 660 * segment. 661 */ 662 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 663 p = q; 664 665 continue; 666 667 /* 668 * This is a common operation. It also will allow 669 * to save doing a malloc/free in most instances. 670 */ 671 free_ipqe: 672 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 673 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 674 tp->t_segqlen--; 675 KASSERT(tp->t_segqlen >= 0); 676 KASSERT(tp->t_segqlen != 0 || 677 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 678 if (tiqe == NULL) { 679 tiqe = q; 680 } else { 681 tcpipqent_free(q); 682 } 683 } 684 685 #ifdef TCP_REASS_COUNTERS 686 if (count > 7) 687 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 688 else if (count > 0) 689 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 690 #endif 691 692 insert_it: 693 694 /* 695 * Allocate a new queue entry since the received segment did not 696 * collapse onto any other out-of-order block; thus we are allocating 697 * a new block. If it had collapsed, tiqe would not be NULL and 698 * we would be reusing it. 699 * XXX If we can't, just drop the packet. XXX 700 */ 701 if (tiqe == NULL) { 702 tiqe = tcpipqent_alloc(); 703 if (tiqe == NULL) { 704 tcpstat.tcps_rcvmemdrop++; 705 m_freem(m); 706 return (0); 707 } 708 } 709 710 /* 711 * Update the counters. 712 */ 713 tcpstat.tcps_rcvoopack++; 714 tcpstat.tcps_rcvoobyte += rcvoobyte; 715 if (rcvpartdupbyte) { 716 tcpstat.tcps_rcvpartduppack++; 717 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 718 } 719 720 /* 721 * Insert the new fragment queue entry into both queues. 722 */ 723 tiqe->ipqe_m = m; 724 tiqe->ipre_mlast = m; 725 tiqe->ipqe_seq = pkt_seq; 726 tiqe->ipqe_len = pkt_len; 727 tiqe->ipqe_flags = pkt_flags; 728 if (p == NULL) { 729 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 730 #ifdef TCPREASS_DEBUG 731 if (tiqe->ipqe_seq != tp->rcv_nxt) 732 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 733 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 734 #endif 735 } else { 736 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 737 #ifdef TCPREASS_DEBUG 738 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 739 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 740 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 741 #endif 742 } 743 tp->t_segqlen++; 744 745 skip_replacement: 746 747 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 748 749 present: 750 /* 751 * Present data to user, advancing rcv_nxt through 752 * completed sequence space. 753 */ 754 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 755 return (0); 756 q = TAILQ_FIRST(&tp->segq); 757 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 758 return (0); 759 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 760 return (0); 761 762 tp->rcv_nxt += q->ipqe_len; 763 pkt_flags = q->ipqe_flags & TH_FIN; 764 ND6_HINT(tp); 765 766 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 767 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 768 tp->t_segqlen--; 769 KASSERT(tp->t_segqlen >= 0); 770 KASSERT(tp->t_segqlen != 0 || 771 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 772 if (so->so_state & SS_CANTRCVMORE) 773 m_freem(q->ipqe_m); 774 else 775 sbappendstream(&so->so_rcv, q->ipqe_m); 776 tcpipqent_free(q); 777 sorwakeup(so); 778 return (pkt_flags); 779 } 780 781 #ifdef INET6 782 int 783 tcp6_input(struct mbuf **mp, int *offp, int proto) 784 { 785 struct mbuf *m = *mp; 786 787 /* 788 * draft-itojun-ipv6-tcp-to-anycast 789 * better place to put this in? 790 */ 791 if (m->m_flags & M_ANYCAST6) { 792 struct ip6_hdr *ip6; 793 if (m->m_len < sizeof(struct ip6_hdr)) { 794 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 795 tcpstat.tcps_rcvshort++; 796 return IPPROTO_DONE; 797 } 798 } 799 ip6 = mtod(m, struct ip6_hdr *); 800 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 801 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 802 return IPPROTO_DONE; 803 } 804 805 tcp_input(m, *offp, proto); 806 return IPPROTO_DONE; 807 } 808 #endif 809 810 #ifdef INET 811 static void 812 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 813 { 814 char src[4*sizeof "123"]; 815 char dst[4*sizeof "123"]; 816 817 if (ip) { 818 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 819 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 820 } 821 else { 822 strlcpy(src, "(unknown)", sizeof(src)); 823 strlcpy(dst, "(unknown)", sizeof(dst)); 824 } 825 log(LOG_INFO, 826 "Connection attempt to TCP %s:%d from %s:%d\n", 827 dst, ntohs(th->th_dport), 828 src, ntohs(th->th_sport)); 829 } 830 #endif 831 832 #ifdef INET6 833 static void 834 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 835 { 836 char src[INET6_ADDRSTRLEN]; 837 char dst[INET6_ADDRSTRLEN]; 838 839 if (ip6) { 840 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 841 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 842 } 843 else { 844 strlcpy(src, "(unknown v6)", sizeof(src)); 845 strlcpy(dst, "(unknown v6)", sizeof(dst)); 846 } 847 log(LOG_INFO, 848 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 849 dst, ntohs(th->th_dport), 850 src, ntohs(th->th_sport)); 851 } 852 #endif 853 854 /* 855 * Checksum extended TCP header and data. 856 */ 857 int 858 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, 859 int toff, int off, int tlen) 860 { 861 862 /* 863 * XXX it's better to record and check if this mbuf is 864 * already checked. 865 */ 866 867 switch (af) { 868 #ifdef INET 869 case AF_INET: 870 switch (m->m_pkthdr.csum_flags & 871 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 872 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 873 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 874 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 875 goto badcsum; 876 877 case M_CSUM_TCPv4|M_CSUM_DATA: { 878 u_int32_t hw_csum = m->m_pkthdr.csum_data; 879 880 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 881 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 882 const struct ip *ip = 883 mtod(m, const struct ip *); 884 885 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 886 ip->ip_dst.s_addr, 887 htons(hw_csum + tlen + off + IPPROTO_TCP)); 888 } 889 if ((hw_csum ^ 0xffff) != 0) 890 goto badcsum; 891 break; 892 } 893 894 case M_CSUM_TCPv4: 895 /* Checksum was okay. */ 896 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 897 break; 898 899 default: 900 /* 901 * Must compute it ourselves. Maybe skip checksum 902 * on loopback interfaces. 903 */ 904 if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & 905 IFF_LOOPBACK) || 906 tcp_do_loopback_cksum)) { 907 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 908 if (in4_cksum(m, IPPROTO_TCP, toff, 909 tlen + off) != 0) 910 goto badcsum; 911 } 912 break; 913 } 914 break; 915 #endif /* INET4 */ 916 917 #ifdef INET6 918 case AF_INET6: 919 switch (m->m_pkthdr.csum_flags & 920 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 921 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 922 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 923 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 924 goto badcsum; 925 926 #if 0 /* notyet */ 927 case M_CSUM_TCPv6|M_CSUM_DATA: 928 #endif 929 930 case M_CSUM_TCPv6: 931 /* Checksum was okay. */ 932 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 933 break; 934 935 default: 936 /* 937 * Must compute it ourselves. Maybe skip checksum 938 * on loopback interfaces. 939 */ 940 if (__predict_true((m->m_flags & M_LOOP) == 0 || 941 tcp_do_loopback_cksum)) { 942 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 943 if (in6_cksum(m, IPPROTO_TCP, toff, 944 tlen + off) != 0) 945 goto badcsum; 946 } 947 } 948 break; 949 #endif /* INET6 */ 950 } 951 952 return 0; 953 954 badcsum: 955 tcpstat.tcps_rcvbadsum++; 956 return -1; 957 } 958 959 /* 960 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 961 */ 962 void 963 tcp_input(struct mbuf *m, ...) 964 { 965 struct tcphdr *th; 966 struct ip *ip; 967 struct inpcb *inp; 968 #ifdef INET6 969 struct ip6_hdr *ip6; 970 struct in6pcb *in6p; 971 #endif 972 u_int8_t *optp = NULL; 973 int optlen = 0; 974 int len, tlen, toff, hdroptlen = 0; 975 struct tcpcb *tp = 0; 976 int tiflags; 977 struct socket *so = NULL; 978 int todrop, dupseg, acked, ourfinisacked, needoutput = 0; 979 #ifdef TCP_DEBUG 980 short ostate = 0; 981 #endif 982 u_long tiwin; 983 struct tcp_opt_info opti; 984 int off, iphlen; 985 va_list ap; 986 int af; /* af on the wire */ 987 struct mbuf *tcp_saveti = NULL; 988 uint32_t ts_rtt; 989 uint8_t iptos; 990 991 MCLAIM(m, &tcp_rx_mowner); 992 va_start(ap, m); 993 toff = va_arg(ap, int); 994 (void)va_arg(ap, int); /* ignore value, advance ap */ 995 va_end(ap); 996 997 tcpstat.tcps_rcvtotal++; 998 999 bzero(&opti, sizeof(opti)); 1000 opti.ts_present = 0; 1001 opti.maxseg = 0; 1002 1003 /* 1004 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 1005 * 1006 * TCP is, by definition, unicast, so we reject all 1007 * multicast outright. 1008 * 1009 * Note, there are additional src/dst address checks in 1010 * the AF-specific code below. 1011 */ 1012 if (m->m_flags & (M_BCAST|M_MCAST)) { 1013 /* XXX stat */ 1014 goto drop; 1015 } 1016 #ifdef INET6 1017 if (m->m_flags & M_ANYCAST6) { 1018 /* XXX stat */ 1019 goto drop; 1020 } 1021 #endif 1022 1023 /* 1024 * Get IP and TCP header. 1025 * Note: IP leaves IP header in first mbuf. 1026 */ 1027 ip = mtod(m, struct ip *); 1028 #ifdef INET6 1029 ip6 = NULL; 1030 #endif 1031 switch (ip->ip_v) { 1032 #ifdef INET 1033 case 4: 1034 af = AF_INET; 1035 iphlen = sizeof(struct ip); 1036 ip = mtod(m, struct ip *); 1037 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1038 sizeof(struct tcphdr)); 1039 if (th == NULL) { 1040 tcpstat.tcps_rcvshort++; 1041 return; 1042 } 1043 /* We do the checksum after PCB lookup... */ 1044 len = ntohs(ip->ip_len); 1045 tlen = len - toff; 1046 iptos = ip->ip_tos; 1047 break; 1048 #endif 1049 #ifdef INET6 1050 case 6: 1051 ip = NULL; 1052 iphlen = sizeof(struct ip6_hdr); 1053 af = AF_INET6; 1054 ip6 = mtod(m, struct ip6_hdr *); 1055 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1056 sizeof(struct tcphdr)); 1057 if (th == NULL) { 1058 tcpstat.tcps_rcvshort++; 1059 return; 1060 } 1061 1062 /* Be proactive about malicious use of IPv4 mapped address */ 1063 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 1064 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 1065 /* XXX stat */ 1066 goto drop; 1067 } 1068 1069 /* 1070 * Be proactive about unspecified IPv6 address in source. 1071 * As we use all-zero to indicate unbounded/unconnected pcb, 1072 * unspecified IPv6 address can be used to confuse us. 1073 * 1074 * Note that packets with unspecified IPv6 destination is 1075 * already dropped in ip6_input. 1076 */ 1077 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1078 /* XXX stat */ 1079 goto drop; 1080 } 1081 1082 /* 1083 * Make sure destination address is not multicast. 1084 * Source address checked in ip6_input(). 1085 */ 1086 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1087 /* XXX stat */ 1088 goto drop; 1089 } 1090 1091 /* We do the checksum after PCB lookup... */ 1092 len = m->m_pkthdr.len; 1093 tlen = len - toff; 1094 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1095 break; 1096 #endif 1097 default: 1098 m_freem(m); 1099 return; 1100 } 1101 1102 KASSERT(TCP_HDR_ALIGNED_P(th)); 1103 1104 /* 1105 * Check that TCP offset makes sense, 1106 * pull out TCP options and adjust length. XXX 1107 */ 1108 off = th->th_off << 2; 1109 if (off < sizeof (struct tcphdr) || off > tlen) { 1110 tcpstat.tcps_rcvbadoff++; 1111 goto drop; 1112 } 1113 tlen -= off; 1114 1115 /* 1116 * tcp_input() has been modified to use tlen to mean the TCP data 1117 * length throughout the function. Other functions can use 1118 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1119 * rja 1120 */ 1121 1122 if (off > sizeof (struct tcphdr)) { 1123 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1124 if (th == NULL) { 1125 tcpstat.tcps_rcvshort++; 1126 return; 1127 } 1128 /* 1129 * NOTE: ip/ip6 will not be affected by m_pulldown() 1130 * (as they're before toff) and we don't need to update those. 1131 */ 1132 KASSERT(TCP_HDR_ALIGNED_P(th)); 1133 optlen = off - sizeof (struct tcphdr); 1134 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1135 /* 1136 * Do quick retrieval of timestamp options ("options 1137 * prediction?"). If timestamp is the only option and it's 1138 * formatted as recommended in RFC 1323 appendix A, we 1139 * quickly get the values now and not bother calling 1140 * tcp_dooptions(), etc. 1141 */ 1142 if ((optlen == TCPOLEN_TSTAMP_APPA || 1143 (optlen > TCPOLEN_TSTAMP_APPA && 1144 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1145 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1146 (th->th_flags & TH_SYN) == 0) { 1147 opti.ts_present = 1; 1148 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1149 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1150 optp = NULL; /* we've parsed the options */ 1151 } 1152 } 1153 tiflags = th->th_flags; 1154 1155 /* 1156 * Locate pcb for segment. 1157 */ 1158 findpcb: 1159 inp = NULL; 1160 #ifdef INET6 1161 in6p = NULL; 1162 #endif 1163 switch (af) { 1164 #ifdef INET 1165 case AF_INET: 1166 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1167 ip->ip_dst, th->th_dport); 1168 if (inp == 0) { 1169 ++tcpstat.tcps_pcbhashmiss; 1170 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1171 } 1172 #ifdef INET6 1173 if (inp == 0) { 1174 struct in6_addr s, d; 1175 1176 /* mapped addr case */ 1177 bzero(&s, sizeof(s)); 1178 s.s6_addr16[5] = htons(0xffff); 1179 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1180 bzero(&d, sizeof(d)); 1181 d.s6_addr16[5] = htons(0xffff); 1182 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1183 in6p = in6_pcblookup_connect(&tcbtable, &s, 1184 th->th_sport, &d, th->th_dport, 0); 1185 if (in6p == 0) { 1186 ++tcpstat.tcps_pcbhashmiss; 1187 in6p = in6_pcblookup_bind(&tcbtable, &d, 1188 th->th_dport, 0); 1189 } 1190 } 1191 #endif 1192 #ifndef INET6 1193 if (inp == 0) 1194 #else 1195 if (inp == 0 && in6p == 0) 1196 #endif 1197 { 1198 ++tcpstat.tcps_noport; 1199 if (tcp_log_refused && 1200 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1201 tcp4_log_refused(ip, th); 1202 } 1203 TCP_FIELDS_TO_HOST(th); 1204 goto dropwithreset_ratelim; 1205 } 1206 #if defined(IPSEC) || defined(FAST_IPSEC) 1207 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1208 ipsec4_in_reject(m, inp)) { 1209 ipsecstat.in_polvio++; 1210 goto drop; 1211 } 1212 #ifdef INET6 1213 else if (in6p && 1214 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1215 ipsec4_in_reject_so(m, in6p->in6p_socket)) { 1216 ipsecstat.in_polvio++; 1217 goto drop; 1218 } 1219 #endif 1220 #endif /*IPSEC*/ 1221 break; 1222 #endif /*INET*/ 1223 #ifdef INET6 1224 case AF_INET6: 1225 { 1226 int faith; 1227 1228 #if defined(NFAITH) && NFAITH > 0 1229 faith = faithprefix(&ip6->ip6_dst); 1230 #else 1231 faith = 0; 1232 #endif 1233 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1234 th->th_sport, &ip6->ip6_dst, th->th_dport, faith); 1235 if (in6p == NULL) { 1236 ++tcpstat.tcps_pcbhashmiss; 1237 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1238 th->th_dport, faith); 1239 } 1240 if (in6p == NULL) { 1241 ++tcpstat.tcps_noport; 1242 if (tcp_log_refused && 1243 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1244 tcp6_log_refused(ip6, th); 1245 } 1246 TCP_FIELDS_TO_HOST(th); 1247 goto dropwithreset_ratelim; 1248 } 1249 #if defined(IPSEC) || defined(FAST_IPSEC) 1250 if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1251 ipsec6_in_reject(m, in6p)) { 1252 ipsec6stat.in_polvio++; 1253 goto drop; 1254 } 1255 #endif /*IPSEC*/ 1256 break; 1257 } 1258 #endif 1259 } 1260 1261 /* 1262 * If the state is CLOSED (i.e., TCB does not exist) then 1263 * all data in the incoming segment is discarded. 1264 * If the TCB exists but is in CLOSED state, it is embryonic, 1265 * but should either do a listen or a connect soon. 1266 */ 1267 tp = NULL; 1268 so = NULL; 1269 if (inp) { 1270 tp = intotcpcb(inp); 1271 so = inp->inp_socket; 1272 } 1273 #ifdef INET6 1274 else if (in6p) { 1275 tp = in6totcpcb(in6p); 1276 so = in6p->in6p_socket; 1277 } 1278 #endif 1279 if (tp == 0) { 1280 TCP_FIELDS_TO_HOST(th); 1281 goto dropwithreset_ratelim; 1282 } 1283 if (tp->t_state == TCPS_CLOSED) 1284 goto drop; 1285 1286 /* 1287 * Checksum extended TCP header and data. 1288 */ 1289 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1290 goto badcsum; 1291 1292 TCP_FIELDS_TO_HOST(th); 1293 1294 /* Unscale the window into a 32-bit value. */ 1295 if ((tiflags & TH_SYN) == 0) 1296 tiwin = th->th_win << tp->snd_scale; 1297 else 1298 tiwin = th->th_win; 1299 1300 #ifdef INET6 1301 /* save packet options if user wanted */ 1302 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1303 if (in6p->in6p_options) { 1304 m_freem(in6p->in6p_options); 1305 in6p->in6p_options = 0; 1306 } 1307 KASSERT(ip6 != NULL); 1308 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1309 } 1310 #endif 1311 1312 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1313 union syn_cache_sa src; 1314 union syn_cache_sa dst; 1315 1316 bzero(&src, sizeof(src)); 1317 bzero(&dst, sizeof(dst)); 1318 switch (af) { 1319 #ifdef INET 1320 case AF_INET: 1321 src.sin.sin_len = sizeof(struct sockaddr_in); 1322 src.sin.sin_family = AF_INET; 1323 src.sin.sin_addr = ip->ip_src; 1324 src.sin.sin_port = th->th_sport; 1325 1326 dst.sin.sin_len = sizeof(struct sockaddr_in); 1327 dst.sin.sin_family = AF_INET; 1328 dst.sin.sin_addr = ip->ip_dst; 1329 dst.sin.sin_port = th->th_dport; 1330 break; 1331 #endif 1332 #ifdef INET6 1333 case AF_INET6: 1334 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1335 src.sin6.sin6_family = AF_INET6; 1336 src.sin6.sin6_addr = ip6->ip6_src; 1337 src.sin6.sin6_port = th->th_sport; 1338 1339 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1340 dst.sin6.sin6_family = AF_INET6; 1341 dst.sin6.sin6_addr = ip6->ip6_dst; 1342 dst.sin6.sin6_port = th->th_dport; 1343 break; 1344 #endif /* INET6 */ 1345 default: 1346 goto badsyn; /*sanity*/ 1347 } 1348 1349 if (so->so_options & SO_DEBUG) { 1350 #ifdef TCP_DEBUG 1351 ostate = tp->t_state; 1352 #endif 1353 1354 tcp_saveti = NULL; 1355 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1356 goto nosave; 1357 1358 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1359 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1360 if (!tcp_saveti) 1361 goto nosave; 1362 } else { 1363 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1364 if (!tcp_saveti) 1365 goto nosave; 1366 MCLAIM(m, &tcp_mowner); 1367 tcp_saveti->m_len = iphlen; 1368 m_copydata(m, 0, iphlen, 1369 mtod(tcp_saveti, caddr_t)); 1370 } 1371 1372 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1373 m_freem(tcp_saveti); 1374 tcp_saveti = NULL; 1375 } else { 1376 tcp_saveti->m_len += sizeof(struct tcphdr); 1377 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1378 sizeof(struct tcphdr)); 1379 } 1380 nosave:; 1381 } 1382 if (so->so_options & SO_ACCEPTCONN) { 1383 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1384 if (tiflags & TH_RST) { 1385 syn_cache_reset(&src.sa, &dst.sa, th); 1386 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1387 (TH_ACK|TH_SYN)) { 1388 /* 1389 * Received a SYN,ACK. This should 1390 * never happen while we are in 1391 * LISTEN. Send an RST. 1392 */ 1393 goto badsyn; 1394 } else if (tiflags & TH_ACK) { 1395 so = syn_cache_get(&src.sa, &dst.sa, 1396 th, toff, tlen, so, m); 1397 if (so == NULL) { 1398 /* 1399 * We don't have a SYN for 1400 * this ACK; send an RST. 1401 */ 1402 goto badsyn; 1403 } else if (so == 1404 (struct socket *)(-1)) { 1405 /* 1406 * We were unable to create 1407 * the connection. If the 1408 * 3-way handshake was 1409 * completed, and RST has 1410 * been sent to the peer. 1411 * Since the mbuf might be 1412 * in use for the reply, 1413 * do not free it. 1414 */ 1415 m = NULL; 1416 } else { 1417 /* 1418 * We have created a 1419 * full-blown connection. 1420 */ 1421 tp = NULL; 1422 inp = NULL; 1423 #ifdef INET6 1424 in6p = NULL; 1425 #endif 1426 switch (so->so_proto->pr_domain->dom_family) { 1427 #ifdef INET 1428 case AF_INET: 1429 inp = sotoinpcb(so); 1430 tp = intotcpcb(inp); 1431 break; 1432 #endif 1433 #ifdef INET6 1434 case AF_INET6: 1435 in6p = sotoin6pcb(so); 1436 tp = in6totcpcb(in6p); 1437 break; 1438 #endif 1439 } 1440 if (tp == NULL) 1441 goto badsyn; /*XXX*/ 1442 tiwin <<= tp->snd_scale; 1443 goto after_listen; 1444 } 1445 } else { 1446 /* 1447 * None of RST, SYN or ACK was set. 1448 * This is an invalid packet for a 1449 * TCB in LISTEN state. Send a RST. 1450 */ 1451 goto badsyn; 1452 } 1453 } else { 1454 /* 1455 * Received a SYN. 1456 * 1457 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1458 */ 1459 if (m->m_flags & (M_BCAST|M_MCAST)) 1460 goto drop; 1461 1462 switch (af) { 1463 #ifdef INET6 1464 case AF_INET6: 1465 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1466 goto drop; 1467 break; 1468 #endif /* INET6 */ 1469 case AF_INET: 1470 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1471 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1472 goto drop; 1473 break; 1474 } 1475 1476 #ifdef INET6 1477 /* 1478 * If deprecated address is forbidden, we do 1479 * not accept SYN to deprecated interface 1480 * address to prevent any new inbound 1481 * connection from getting established. 1482 * When we do not accept SYN, we send a TCP 1483 * RST, with deprecated source address (instead 1484 * of dropping it). We compromise it as it is 1485 * much better for peer to send a RST, and 1486 * RST will be the final packet for the 1487 * exchange. 1488 * 1489 * If we do not forbid deprecated addresses, we 1490 * accept the SYN packet. RFC2462 does not 1491 * suggest dropping SYN in this case. 1492 * If we decipher RFC2462 5.5.4, it says like 1493 * this: 1494 * 1. use of deprecated addr with existing 1495 * communication is okay - "SHOULD continue 1496 * to be used" 1497 * 2. use of it with new communication: 1498 * (2a) "SHOULD NOT be used if alternate 1499 * address with sufficient scope is 1500 * available" 1501 * (2b) nothing mentioned otherwise. 1502 * Here we fall into (2b) case as we have no 1503 * choice in our source address selection - we 1504 * must obey the peer. 1505 * 1506 * The wording in RFC2462 is confusing, and 1507 * there are multiple description text for 1508 * deprecated address handling - worse, they 1509 * are not exactly the same. I believe 5.5.4 1510 * is the best one, so we follow 5.5.4. 1511 */ 1512 if (af == AF_INET6 && !ip6_use_deprecated) { 1513 struct in6_ifaddr *ia6; 1514 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1515 &ip6->ip6_dst)) && 1516 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1517 tp = NULL; 1518 goto dropwithreset; 1519 } 1520 } 1521 #endif 1522 1523 #ifdef IPSEC 1524 switch (af) { 1525 #ifdef INET 1526 case AF_INET: 1527 if (ipsec4_in_reject_so(m, so)) { 1528 ipsecstat.in_polvio++; 1529 tp = NULL; 1530 goto dropwithreset; 1531 } 1532 break; 1533 #endif 1534 #ifdef INET6 1535 case AF_INET6: 1536 if (ipsec6_in_reject_so(m, so)) { 1537 ipsec6stat.in_polvio++; 1538 tp = NULL; 1539 goto dropwithreset; 1540 } 1541 break; 1542 #endif 1543 } 1544 #endif 1545 1546 /* 1547 * LISTEN socket received a SYN 1548 * from itself? This can't possibly 1549 * be valid; drop the packet. 1550 */ 1551 if (th->th_sport == th->th_dport) { 1552 int i; 1553 1554 switch (af) { 1555 #ifdef INET 1556 case AF_INET: 1557 i = in_hosteq(ip->ip_src, ip->ip_dst); 1558 break; 1559 #endif 1560 #ifdef INET6 1561 case AF_INET6: 1562 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1563 break; 1564 #endif 1565 default: 1566 i = 1; 1567 } 1568 if (i) { 1569 tcpstat.tcps_badsyn++; 1570 goto drop; 1571 } 1572 } 1573 1574 /* 1575 * SYN looks ok; create compressed TCP 1576 * state for it. 1577 */ 1578 if (so->so_qlen <= so->so_qlimit && 1579 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1580 so, m, optp, optlen, &opti)) 1581 m = NULL; 1582 } 1583 goto drop; 1584 } 1585 } 1586 1587 after_listen: 1588 #ifdef DIAGNOSTIC 1589 /* 1590 * Should not happen now that all embryonic connections 1591 * are handled with compressed state. 1592 */ 1593 if (tp->t_state == TCPS_LISTEN) 1594 panic("tcp_input: TCPS_LISTEN"); 1595 #endif 1596 1597 /* 1598 * Segment received on connection. 1599 * Reset idle time and keep-alive timer. 1600 */ 1601 tp->t_rcvtime = tcp_now; 1602 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1603 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1604 1605 /* 1606 * Process options. 1607 */ 1608 #ifdef TCP_SIGNATURE 1609 if (optp || (tp->t_flags & TF_SIGNATURE)) 1610 #else 1611 if (optp) 1612 #endif 1613 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1614 goto drop; 1615 1616 if (TCP_SACK_ENABLED(tp)) { 1617 tcp_del_sackholes(tp, th); 1618 } 1619 1620 if (TCP_ECN_ALLOWED(tp)) { 1621 switch (iptos & IPTOS_ECN_MASK) { 1622 case IPTOS_ECN_CE: 1623 tp->t_flags |= TF_ECN_SND_ECE; 1624 tcpstat.tcps_ecn_ce++; 1625 break; 1626 case IPTOS_ECN_ECT0: 1627 tcpstat.tcps_ecn_ect++; 1628 break; 1629 case IPTOS_ECN_ECT1: 1630 /* XXX: ignore for now -- rpaulo */ 1631 break; 1632 } 1633 1634 if (tiflags & TH_CWR) 1635 tp->t_flags &= ~TF_ECN_SND_ECE; 1636 1637 /* 1638 * Congestion experienced. 1639 * Ignore if we are already trying to recover. 1640 */ 1641 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1642 tp->t_congctl->cong_exp(tp); 1643 } 1644 1645 if (opti.ts_present && opti.ts_ecr) { 1646 /* 1647 * Calculate the RTT from the returned time stamp and the 1648 * connection's time base. If the time stamp is later than 1649 * the current time, or is extremely old, fall back to non-1323 1650 * RTT calculation. Since ts_ecr is unsigned, we can test both 1651 * at the same time. 1652 */ 1653 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1654 if (ts_rtt > TCP_PAWS_IDLE) 1655 ts_rtt = 0; 1656 } else { 1657 ts_rtt = 0; 1658 } 1659 1660 /* 1661 * Header prediction: check for the two common cases 1662 * of a uni-directional data xfer. If the packet has 1663 * no control flags, is in-sequence, the window didn't 1664 * change and we're not retransmitting, it's a 1665 * candidate. If the length is zero and the ack moved 1666 * forward, we're the sender side of the xfer. Just 1667 * free the data acked & wake any higher level process 1668 * that was blocked waiting for space. If the length 1669 * is non-zero and the ack didn't move, we're the 1670 * receiver side. If we're getting packets in-order 1671 * (the reassembly queue is empty), add the data to 1672 * the socket buffer and note that we need a delayed ack. 1673 */ 1674 if (tp->t_state == TCPS_ESTABLISHED && 1675 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1676 == TH_ACK && 1677 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1678 th->th_seq == tp->rcv_nxt && 1679 tiwin && tiwin == tp->snd_wnd && 1680 tp->snd_nxt == tp->snd_max) { 1681 1682 /* 1683 * If last ACK falls within this segment's sequence numbers, 1684 * record the timestamp. 1685 * NOTE: 1686 * 1) That the test incorporates suggestions from the latest 1687 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 1688 * 2) That updating only on newer timestamps interferes with 1689 * our earlier PAWS tests, so this check should be solely 1690 * predicated on the sequence space of this segment. 1691 * 3) That we modify the segment boundary check to be 1692 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1693 * instead of RFC1323's 1694 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1695 * This modified check allows us to overcome RFC1323's 1696 * limitations as described in Stevens TCP/IP Illustrated 1697 * Vol. 2 p.869. In such cases, we can still calculate the 1698 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1699 */ 1700 if (opti.ts_present && 1701 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1702 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1703 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1704 tp->ts_recent_age = tcp_now; 1705 tp->ts_recent = opti.ts_val; 1706 } 1707 1708 if (tlen == 0) { 1709 /* Ack prediction. */ 1710 if (SEQ_GT(th->th_ack, tp->snd_una) && 1711 SEQ_LEQ(th->th_ack, tp->snd_max) && 1712 tp->snd_cwnd >= tp->snd_wnd && 1713 tp->t_partialacks < 0) { 1714 /* 1715 * this is a pure ack for outstanding data. 1716 */ 1717 ++tcpstat.tcps_predack; 1718 if (ts_rtt) 1719 tcp_xmit_timer(tp, ts_rtt); 1720 else if (tp->t_rtttime && 1721 SEQ_GT(th->th_ack, tp->t_rtseq)) 1722 tcp_xmit_timer(tp, 1723 tcp_now - tp->t_rtttime); 1724 acked = th->th_ack - tp->snd_una; 1725 tcpstat.tcps_rcvackpack++; 1726 tcpstat.tcps_rcvackbyte += acked; 1727 ND6_HINT(tp); 1728 1729 if (acked > (tp->t_lastoff - tp->t_inoff)) 1730 tp->t_lastm = NULL; 1731 sbdrop(&so->so_snd, acked); 1732 tp->t_lastoff -= acked; 1733 1734 ICMP_CHECK(tp, th, acked); 1735 1736 tp->snd_una = th->th_ack; 1737 tp->snd_fack = tp->snd_una; 1738 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1739 tp->snd_high = tp->snd_una; 1740 m_freem(m); 1741 1742 /* 1743 * If all outstanding data are acked, stop 1744 * retransmit timer, otherwise restart timer 1745 * using current (possibly backed-off) value. 1746 * If process is waiting for space, 1747 * wakeup/selwakeup/signal. If data 1748 * are ready to send, let tcp_output 1749 * decide between more output or persist. 1750 */ 1751 if (tp->snd_una == tp->snd_max) 1752 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1753 else if (TCP_TIMER_ISARMED(tp, 1754 TCPT_PERSIST) == 0) 1755 TCP_TIMER_ARM(tp, TCPT_REXMT, 1756 tp->t_rxtcur); 1757 1758 sowwakeup(so); 1759 if (so->so_snd.sb_cc) 1760 (void) tcp_output(tp); 1761 if (tcp_saveti) 1762 m_freem(tcp_saveti); 1763 return; 1764 } 1765 } else if (th->th_ack == tp->snd_una && 1766 TAILQ_FIRST(&tp->segq) == NULL && 1767 tlen <= sbspace(&so->so_rcv)) { 1768 /* 1769 * this is a pure, in-sequence data packet 1770 * with nothing on the reassembly queue and 1771 * we have enough buffer space to take it. 1772 */ 1773 ++tcpstat.tcps_preddat; 1774 tp->rcv_nxt += tlen; 1775 tcpstat.tcps_rcvpack++; 1776 tcpstat.tcps_rcvbyte += tlen; 1777 ND6_HINT(tp); 1778 /* 1779 * Drop TCP, IP headers and TCP options then add data 1780 * to socket buffer. 1781 */ 1782 if (so->so_state & SS_CANTRCVMORE) 1783 m_freem(m); 1784 else { 1785 m_adj(m, toff + off); 1786 sbappendstream(&so->so_rcv, m); 1787 } 1788 sorwakeup(so); 1789 TCP_SETUP_ACK(tp, th); 1790 if (tp->t_flags & TF_ACKNOW) 1791 (void) tcp_output(tp); 1792 if (tcp_saveti) 1793 m_freem(tcp_saveti); 1794 return; 1795 } 1796 } 1797 1798 /* 1799 * Compute mbuf offset to TCP data segment. 1800 */ 1801 hdroptlen = toff + off; 1802 1803 /* 1804 * Calculate amount of space in receive window, 1805 * and then do TCP input processing. 1806 * Receive window is amount of space in rcv queue, 1807 * but not less than advertised window. 1808 */ 1809 { int win; 1810 1811 win = sbspace(&so->so_rcv); 1812 if (win < 0) 1813 win = 0; 1814 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1815 } 1816 1817 switch (tp->t_state) { 1818 /* 1819 * If the state is SYN_SENT: 1820 * if seg contains an ACK, but not for our SYN, drop the input. 1821 * if seg contains a RST, then drop the connection. 1822 * if seg does not contain SYN, then drop it. 1823 * Otherwise this is an acceptable SYN segment 1824 * initialize tp->rcv_nxt and tp->irs 1825 * if seg contains ack then advance tp->snd_una 1826 * if seg contains a ECE and ECN support is enabled, the stream 1827 * is ECN capable. 1828 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1829 * arrange for segment to be acked (eventually) 1830 * continue processing rest of data/controls, beginning with URG 1831 */ 1832 case TCPS_SYN_SENT: 1833 if ((tiflags & TH_ACK) && 1834 (SEQ_LEQ(th->th_ack, tp->iss) || 1835 SEQ_GT(th->th_ack, tp->snd_max))) 1836 goto dropwithreset; 1837 if (tiflags & TH_RST) { 1838 if (tiflags & TH_ACK) 1839 tp = tcp_drop(tp, ECONNREFUSED); 1840 goto drop; 1841 } 1842 if ((tiflags & TH_SYN) == 0) 1843 goto drop; 1844 if (tiflags & TH_ACK) { 1845 tp->snd_una = th->th_ack; 1846 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1847 tp->snd_nxt = tp->snd_una; 1848 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1849 tp->snd_high = tp->snd_una; 1850 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1851 1852 if ((tiflags & TH_ECE) && tcp_do_ecn) { 1853 tp->t_flags |= TF_ECN_PERMIT; 1854 tcpstat.tcps_ecn_shs++; 1855 } 1856 1857 } 1858 tp->irs = th->th_seq; 1859 tcp_rcvseqinit(tp); 1860 tp->t_flags |= TF_ACKNOW; 1861 tcp_mss_from_peer(tp, opti.maxseg); 1862 1863 /* 1864 * Initialize the initial congestion window. If we 1865 * had to retransmit the SYN, we must initialize cwnd 1866 * to 1 segment (i.e. the Loss Window). 1867 */ 1868 if (tp->t_flags & TF_SYN_REXMT) 1869 tp->snd_cwnd = tp->t_peermss; 1870 else { 1871 int ss = tcp_init_win; 1872 #ifdef INET 1873 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1874 ss = tcp_init_win_local; 1875 #endif 1876 #ifdef INET6 1877 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1878 ss = tcp_init_win_local; 1879 #endif 1880 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1881 } 1882 1883 tcp_rmx_rtt(tp); 1884 if (tiflags & TH_ACK) { 1885 tcpstat.tcps_connects++; 1886 soisconnected(so); 1887 tcp_established(tp); 1888 /* Do window scaling on this connection? */ 1889 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1890 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1891 tp->snd_scale = tp->requested_s_scale; 1892 tp->rcv_scale = tp->request_r_scale; 1893 } 1894 TCP_REASS_LOCK(tp); 1895 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1896 TCP_REASS_UNLOCK(tp); 1897 /* 1898 * if we didn't have to retransmit the SYN, 1899 * use its rtt as our initial srtt & rtt var. 1900 */ 1901 if (tp->t_rtttime) 1902 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1903 } else 1904 tp->t_state = TCPS_SYN_RECEIVED; 1905 1906 /* 1907 * Advance th->th_seq to correspond to first data byte. 1908 * If data, trim to stay within window, 1909 * dropping FIN if necessary. 1910 */ 1911 th->th_seq++; 1912 if (tlen > tp->rcv_wnd) { 1913 todrop = tlen - tp->rcv_wnd; 1914 m_adj(m, -todrop); 1915 tlen = tp->rcv_wnd; 1916 tiflags &= ~TH_FIN; 1917 tcpstat.tcps_rcvpackafterwin++; 1918 tcpstat.tcps_rcvbyteafterwin += todrop; 1919 } 1920 tp->snd_wl1 = th->th_seq - 1; 1921 tp->rcv_up = th->th_seq; 1922 goto step6; 1923 1924 /* 1925 * If the state is SYN_RECEIVED: 1926 * If seg contains an ACK, but not for our SYN, drop the input 1927 * and generate an RST. See page 36, rfc793 1928 */ 1929 case TCPS_SYN_RECEIVED: 1930 if ((tiflags & TH_ACK) && 1931 (SEQ_LEQ(th->th_ack, tp->iss) || 1932 SEQ_GT(th->th_ack, tp->snd_max))) 1933 goto dropwithreset; 1934 break; 1935 } 1936 1937 /* 1938 * States other than LISTEN or SYN_SENT. 1939 * First check timestamp, if present. 1940 * Then check that at least some bytes of segment are within 1941 * receive window. If segment begins before rcv_nxt, 1942 * drop leading data (and SYN); if nothing left, just ack. 1943 * 1944 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1945 * and it's less than ts_recent, drop it. 1946 */ 1947 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1948 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1949 1950 /* Check to see if ts_recent is over 24 days old. */ 1951 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 1952 /* 1953 * Invalidate ts_recent. If this segment updates 1954 * ts_recent, the age will be reset later and ts_recent 1955 * will get a valid value. If it does not, setting 1956 * ts_recent to zero will at least satisfy the 1957 * requirement that zero be placed in the timestamp 1958 * echo reply when ts_recent isn't valid. The 1959 * age isn't reset until we get a valid ts_recent 1960 * because we don't want out-of-order segments to be 1961 * dropped when ts_recent is old. 1962 */ 1963 tp->ts_recent = 0; 1964 } else { 1965 tcpstat.tcps_rcvduppack++; 1966 tcpstat.tcps_rcvdupbyte += tlen; 1967 tcpstat.tcps_pawsdrop++; 1968 tcp_new_dsack(tp, th->th_seq, tlen); 1969 goto dropafterack; 1970 } 1971 } 1972 1973 todrop = tp->rcv_nxt - th->th_seq; 1974 dupseg = FALSE; 1975 if (todrop > 0) { 1976 if (tiflags & TH_SYN) { 1977 tiflags &= ~TH_SYN; 1978 th->th_seq++; 1979 if (th->th_urp > 1) 1980 th->th_urp--; 1981 else { 1982 tiflags &= ~TH_URG; 1983 th->th_urp = 0; 1984 } 1985 todrop--; 1986 } 1987 if (todrop > tlen || 1988 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1989 /* 1990 * Any valid FIN or RST must be to the left of the 1991 * window. At this point the FIN or RST must be a 1992 * duplicate or out of sequence; drop it. 1993 */ 1994 if (tiflags & TH_RST) 1995 goto drop; 1996 tiflags &= ~(TH_FIN|TH_RST); 1997 /* 1998 * Send an ACK to resynchronize and drop any data. 1999 * But keep on processing for RST or ACK. 2000 */ 2001 tp->t_flags |= TF_ACKNOW; 2002 todrop = tlen; 2003 dupseg = TRUE; 2004 tcpstat.tcps_rcvdupbyte += todrop; 2005 tcpstat.tcps_rcvduppack++; 2006 } else if ((tiflags & TH_RST) && 2007 th->th_seq != tp->last_ack_sent) { 2008 /* 2009 * Test for reset before adjusting the sequence 2010 * number for overlapping data. 2011 */ 2012 goto dropafterack_ratelim; 2013 } else { 2014 tcpstat.tcps_rcvpartduppack++; 2015 tcpstat.tcps_rcvpartdupbyte += todrop; 2016 } 2017 tcp_new_dsack(tp, th->th_seq, todrop); 2018 hdroptlen += todrop; /*drop from head afterwards*/ 2019 th->th_seq += todrop; 2020 tlen -= todrop; 2021 if (th->th_urp > todrop) 2022 th->th_urp -= todrop; 2023 else { 2024 tiflags &= ~TH_URG; 2025 th->th_urp = 0; 2026 } 2027 } 2028 2029 /* 2030 * If new data are received on a connection after the 2031 * user processes are gone, then RST the other end. 2032 */ 2033 if ((so->so_state & SS_NOFDREF) && 2034 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2035 tp = tcp_close(tp); 2036 tcpstat.tcps_rcvafterclose++; 2037 goto dropwithreset; 2038 } 2039 2040 /* 2041 * If segment ends after window, drop trailing data 2042 * (and PUSH and FIN); if nothing left, just ACK. 2043 */ 2044 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 2045 if (todrop > 0) { 2046 tcpstat.tcps_rcvpackafterwin++; 2047 if (todrop >= tlen) { 2048 /* 2049 * The segment actually starts after the window. 2050 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2051 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2052 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2053 */ 2054 tcpstat.tcps_rcvbyteafterwin += tlen; 2055 /* 2056 * If a new connection request is received 2057 * while in TIME_WAIT, drop the old connection 2058 * and start over if the sequence numbers 2059 * are above the previous ones. 2060 * 2061 * NOTE: We will checksum the packet again, and 2062 * so we need to put the header fields back into 2063 * network order! 2064 * XXX This kind of sucks, but we don't expect 2065 * XXX this to happen very often, so maybe it 2066 * XXX doesn't matter so much. 2067 */ 2068 if (tiflags & TH_SYN && 2069 tp->t_state == TCPS_TIME_WAIT && 2070 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2071 tp = tcp_close(tp); 2072 TCP_FIELDS_TO_NET(th); 2073 goto findpcb; 2074 } 2075 /* 2076 * If window is closed can only take segments at 2077 * window edge, and have to drop data and PUSH from 2078 * incoming segments. Continue processing, but 2079 * remember to ack. Otherwise, drop segment 2080 * and (if not RST) ack. 2081 */ 2082 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2083 tp->t_flags |= TF_ACKNOW; 2084 tcpstat.tcps_rcvwinprobe++; 2085 } else 2086 goto dropafterack; 2087 } else 2088 tcpstat.tcps_rcvbyteafterwin += todrop; 2089 m_adj(m, -todrop); 2090 tlen -= todrop; 2091 tiflags &= ~(TH_PUSH|TH_FIN); 2092 } 2093 2094 /* 2095 * If last ACK falls within this segment's sequence numbers, 2096 * and the timestamp is newer, record it. 2097 */ 2098 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 2099 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2100 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 2101 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2102 tp->ts_recent_age = tcp_now; 2103 tp->ts_recent = opti.ts_val; 2104 } 2105 2106 /* 2107 * If the RST bit is set examine the state: 2108 * SYN_RECEIVED STATE: 2109 * If passive open, return to LISTEN state. 2110 * If active open, inform user that connection was refused. 2111 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 2112 * Inform user that connection was reset, and close tcb. 2113 * CLOSING, LAST_ACK, TIME_WAIT STATES 2114 * Close the tcb. 2115 */ 2116 if (tiflags & TH_RST) { 2117 if (th->th_seq != tp->last_ack_sent) 2118 goto dropafterack_ratelim; 2119 2120 switch (tp->t_state) { 2121 case TCPS_SYN_RECEIVED: 2122 so->so_error = ECONNREFUSED; 2123 goto close; 2124 2125 case TCPS_ESTABLISHED: 2126 case TCPS_FIN_WAIT_1: 2127 case TCPS_FIN_WAIT_2: 2128 case TCPS_CLOSE_WAIT: 2129 so->so_error = ECONNRESET; 2130 close: 2131 tp->t_state = TCPS_CLOSED; 2132 tcpstat.tcps_drops++; 2133 tp = tcp_close(tp); 2134 goto drop; 2135 2136 case TCPS_CLOSING: 2137 case TCPS_LAST_ACK: 2138 case TCPS_TIME_WAIT: 2139 tp = tcp_close(tp); 2140 goto drop; 2141 } 2142 } 2143 2144 /* 2145 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2146 * we must be in a synchronized state. RFC791 states (under RST 2147 * generation) that any unacceptable segment (an out-of-order SYN 2148 * qualifies) received in a synchronized state must elicit only an 2149 * empty acknowledgment segment ... and the connection remains in 2150 * the same state. 2151 */ 2152 if (tiflags & TH_SYN) { 2153 if (tp->rcv_nxt == th->th_seq) { 2154 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2155 TH_ACK); 2156 if (tcp_saveti) 2157 m_freem(tcp_saveti); 2158 return; 2159 } 2160 2161 goto dropafterack_ratelim; 2162 } 2163 2164 /* 2165 * If the ACK bit is off we drop the segment and return. 2166 */ 2167 if ((tiflags & TH_ACK) == 0) { 2168 if (tp->t_flags & TF_ACKNOW) 2169 goto dropafterack; 2170 else 2171 goto drop; 2172 } 2173 2174 /* 2175 * Ack processing. 2176 */ 2177 switch (tp->t_state) { 2178 2179 /* 2180 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2181 * ESTABLISHED state and continue processing, otherwise 2182 * send an RST. 2183 */ 2184 case TCPS_SYN_RECEIVED: 2185 if (SEQ_GT(tp->snd_una, th->th_ack) || 2186 SEQ_GT(th->th_ack, tp->snd_max)) 2187 goto dropwithreset; 2188 tcpstat.tcps_connects++; 2189 soisconnected(so); 2190 tcp_established(tp); 2191 /* Do window scaling? */ 2192 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2193 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2194 tp->snd_scale = tp->requested_s_scale; 2195 tp->rcv_scale = tp->request_r_scale; 2196 } 2197 TCP_REASS_LOCK(tp); 2198 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2199 TCP_REASS_UNLOCK(tp); 2200 tp->snd_wl1 = th->th_seq - 1; 2201 /* fall into ... */ 2202 2203 /* 2204 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2205 * ACKs. If the ack is in the range 2206 * tp->snd_una < th->th_ack <= tp->snd_max 2207 * then advance tp->snd_una to th->th_ack and drop 2208 * data from the retransmission queue. If this ACK reflects 2209 * more up to date window information we update our window information. 2210 */ 2211 case TCPS_ESTABLISHED: 2212 case TCPS_FIN_WAIT_1: 2213 case TCPS_FIN_WAIT_2: 2214 case TCPS_CLOSE_WAIT: 2215 case TCPS_CLOSING: 2216 case TCPS_LAST_ACK: 2217 case TCPS_TIME_WAIT: 2218 2219 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2220 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2221 tcpstat.tcps_rcvdupack++; 2222 /* 2223 * If we have outstanding data (other than 2224 * a window probe), this is a completely 2225 * duplicate ack (ie, window info didn't 2226 * change), the ack is the biggest we've 2227 * seen and we've seen exactly our rexmt 2228 * threshhold of them, assume a packet 2229 * has been dropped and retransmit it. 2230 * Kludge snd_nxt & the congestion 2231 * window so we send only this one 2232 * packet. 2233 */ 2234 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2235 th->th_ack != tp->snd_una) 2236 tp->t_dupacks = 0; 2237 else if (tp->t_partialacks < 0 && 2238 ((!TCP_SACK_ENABLED(tp) && 2239 ++tp->t_dupacks == tcprexmtthresh) || 2240 TCP_FACK_FASTRECOV(tp))) { 2241 /* 2242 * Do the fast retransmit, and adjust 2243 * congestion control paramenters. 2244 */ 2245 if (tp->t_congctl->fast_retransmit(tp, th)) { 2246 /* False fast retransmit */ 2247 break; 2248 } else 2249 goto drop; 2250 } else if (tp->t_dupacks > tcprexmtthresh) { 2251 tp->snd_cwnd += tp->t_segsz; 2252 (void) tcp_output(tp); 2253 goto drop; 2254 } 2255 } else { 2256 /* 2257 * If the ack appears to be very old, only 2258 * allow data that is in-sequence. This 2259 * makes it somewhat more difficult to insert 2260 * forged data by guessing sequence numbers. 2261 * Sent an ack to try to update the send 2262 * sequence number on the other side. 2263 */ 2264 if (tlen && th->th_seq != tp->rcv_nxt && 2265 SEQ_LT(th->th_ack, 2266 tp->snd_una - tp->max_sndwnd)) 2267 goto dropafterack; 2268 } 2269 break; 2270 } 2271 /* 2272 * If the congestion window was inflated to account 2273 * for the other side's cached packets, retract it. 2274 */ 2275 /* XXX: make SACK have his own congestion control 2276 * struct -- rpaulo */ 2277 if (TCP_SACK_ENABLED(tp)) 2278 tcp_sack_newack(tp, th); 2279 else 2280 tp->t_congctl->fast_retransmit_newack(tp, th); 2281 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2282 tcpstat.tcps_rcvacktoomuch++; 2283 goto dropafterack; 2284 } 2285 acked = th->th_ack - tp->snd_una; 2286 tcpstat.tcps_rcvackpack++; 2287 tcpstat.tcps_rcvackbyte += acked; 2288 2289 /* 2290 * If we have a timestamp reply, update smoothed 2291 * round trip time. If no timestamp is present but 2292 * transmit timer is running and timed sequence 2293 * number was acked, update smoothed round trip time. 2294 * Since we now have an rtt measurement, cancel the 2295 * timer backoff (cf., Phil Karn's retransmit alg.). 2296 * Recompute the initial retransmit timer. 2297 */ 2298 if (ts_rtt) 2299 tcp_xmit_timer(tp, ts_rtt); 2300 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2301 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2302 2303 /* 2304 * If all outstanding data is acked, stop retransmit 2305 * timer and remember to restart (more output or persist). 2306 * If there is more data to be acked, restart retransmit 2307 * timer, using current (possibly backed-off) value. 2308 */ 2309 if (th->th_ack == tp->snd_max) { 2310 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2311 needoutput = 1; 2312 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2313 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2314 2315 /* 2316 * New data has been acked, adjust the congestion window. 2317 */ 2318 tp->t_congctl->newack(tp, th); 2319 2320 ND6_HINT(tp); 2321 if (acked > so->so_snd.sb_cc) { 2322 tp->snd_wnd -= so->so_snd.sb_cc; 2323 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2324 ourfinisacked = 1; 2325 } else { 2326 if (acked > (tp->t_lastoff - tp->t_inoff)) 2327 tp->t_lastm = NULL; 2328 sbdrop(&so->so_snd, acked); 2329 tp->t_lastoff -= acked; 2330 tp->snd_wnd -= acked; 2331 ourfinisacked = 0; 2332 } 2333 sowwakeup(so); 2334 2335 ICMP_CHECK(tp, th, acked); 2336 2337 tp->snd_una = th->th_ack; 2338 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2339 tp->snd_fack = tp->snd_una; 2340 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2341 tp->snd_nxt = tp->snd_una; 2342 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2343 tp->snd_high = tp->snd_una; 2344 2345 switch (tp->t_state) { 2346 2347 /* 2348 * In FIN_WAIT_1 STATE in addition to the processing 2349 * for the ESTABLISHED state if our FIN is now acknowledged 2350 * then enter FIN_WAIT_2. 2351 */ 2352 case TCPS_FIN_WAIT_1: 2353 if (ourfinisacked) { 2354 /* 2355 * If we can't receive any more 2356 * data, then closing user can proceed. 2357 * Starting the timer is contrary to the 2358 * specification, but if we don't get a FIN 2359 * we'll hang forever. 2360 */ 2361 if (so->so_state & SS_CANTRCVMORE) { 2362 soisdisconnected(so); 2363 if (tcp_maxidle > 0) 2364 TCP_TIMER_ARM(tp, TCPT_2MSL, 2365 tcp_maxidle); 2366 } 2367 tp->t_state = TCPS_FIN_WAIT_2; 2368 } 2369 break; 2370 2371 /* 2372 * In CLOSING STATE in addition to the processing for 2373 * the ESTABLISHED state if the ACK acknowledges our FIN 2374 * then enter the TIME-WAIT state, otherwise ignore 2375 * the segment. 2376 */ 2377 case TCPS_CLOSING: 2378 if (ourfinisacked) { 2379 tp->t_state = TCPS_TIME_WAIT; 2380 tcp_canceltimers(tp); 2381 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2382 soisdisconnected(so); 2383 } 2384 break; 2385 2386 /* 2387 * In LAST_ACK, we may still be waiting for data to drain 2388 * and/or to be acked, as well as for the ack of our FIN. 2389 * If our FIN is now acknowledged, delete the TCB, 2390 * enter the closed state and return. 2391 */ 2392 case TCPS_LAST_ACK: 2393 if (ourfinisacked) { 2394 tp = tcp_close(tp); 2395 goto drop; 2396 } 2397 break; 2398 2399 /* 2400 * In TIME_WAIT state the only thing that should arrive 2401 * is a retransmission of the remote FIN. Acknowledge 2402 * it and restart the finack timer. 2403 */ 2404 case TCPS_TIME_WAIT: 2405 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2406 goto dropafterack; 2407 } 2408 } 2409 2410 step6: 2411 /* 2412 * Update window information. 2413 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2414 */ 2415 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2416 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2417 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2418 /* keep track of pure window updates */ 2419 if (tlen == 0 && 2420 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2421 tcpstat.tcps_rcvwinupd++; 2422 tp->snd_wnd = tiwin; 2423 tp->snd_wl1 = th->th_seq; 2424 tp->snd_wl2 = th->th_ack; 2425 if (tp->snd_wnd > tp->max_sndwnd) 2426 tp->max_sndwnd = tp->snd_wnd; 2427 needoutput = 1; 2428 } 2429 2430 /* 2431 * Process segments with URG. 2432 */ 2433 if ((tiflags & TH_URG) && th->th_urp && 2434 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2435 /* 2436 * This is a kludge, but if we receive and accept 2437 * random urgent pointers, we'll crash in 2438 * soreceive. It's hard to imagine someone 2439 * actually wanting to send this much urgent data. 2440 */ 2441 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2442 th->th_urp = 0; /* XXX */ 2443 tiflags &= ~TH_URG; /* XXX */ 2444 goto dodata; /* XXX */ 2445 } 2446 /* 2447 * If this segment advances the known urgent pointer, 2448 * then mark the data stream. This should not happen 2449 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2450 * a FIN has been received from the remote side. 2451 * In these states we ignore the URG. 2452 * 2453 * According to RFC961 (Assigned Protocols), 2454 * the urgent pointer points to the last octet 2455 * of urgent data. We continue, however, 2456 * to consider it to indicate the first octet 2457 * of data past the urgent section as the original 2458 * spec states (in one of two places). 2459 */ 2460 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2461 tp->rcv_up = th->th_seq + th->th_urp; 2462 so->so_oobmark = so->so_rcv.sb_cc + 2463 (tp->rcv_up - tp->rcv_nxt) - 1; 2464 if (so->so_oobmark == 0) 2465 so->so_state |= SS_RCVATMARK; 2466 sohasoutofband(so); 2467 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2468 } 2469 /* 2470 * Remove out of band data so doesn't get presented to user. 2471 * This can happen independent of advancing the URG pointer, 2472 * but if two URG's are pending at once, some out-of-band 2473 * data may creep in... ick. 2474 */ 2475 if (th->th_urp <= (u_int16_t) tlen 2476 #ifdef SO_OOBINLINE 2477 && (so->so_options & SO_OOBINLINE) == 0 2478 #endif 2479 ) 2480 tcp_pulloutofband(so, th, m, hdroptlen); 2481 } else 2482 /* 2483 * If no out of band data is expected, 2484 * pull receive urgent pointer along 2485 * with the receive window. 2486 */ 2487 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2488 tp->rcv_up = tp->rcv_nxt; 2489 dodata: /* XXX */ 2490 2491 /* 2492 * Process the segment text, merging it into the TCP sequencing queue, 2493 * and arranging for acknowledgement of receipt if necessary. 2494 * This process logically involves adjusting tp->rcv_wnd as data 2495 * is presented to the user (this happens in tcp_usrreq.c, 2496 * case PRU_RCVD). If a FIN has already been received on this 2497 * connection then we just ignore the text. 2498 */ 2499 if ((tlen || (tiflags & TH_FIN)) && 2500 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2501 /* 2502 * Insert segment ti into reassembly queue of tcp with 2503 * control block tp. Return TH_FIN if reassembly now includes 2504 * a segment with FIN. The macro form does the common case 2505 * inline (segment is the next to be received on an 2506 * established connection, and the queue is empty), 2507 * avoiding linkage into and removal from the queue and 2508 * repetition of various conversions. 2509 * Set DELACK for segments received in order, but ack 2510 * immediately when segments are out of order 2511 * (so fast retransmit can work). 2512 */ 2513 /* NOTE: this was TCP_REASS() macro, but used only once */ 2514 TCP_REASS_LOCK(tp); 2515 if (th->th_seq == tp->rcv_nxt && 2516 TAILQ_FIRST(&tp->segq) == NULL && 2517 tp->t_state == TCPS_ESTABLISHED) { 2518 TCP_SETUP_ACK(tp, th); 2519 tp->rcv_nxt += tlen; 2520 tiflags = th->th_flags & TH_FIN; 2521 tcpstat.tcps_rcvpack++; 2522 tcpstat.tcps_rcvbyte += tlen; 2523 ND6_HINT(tp); 2524 if (so->so_state & SS_CANTRCVMORE) 2525 m_freem(m); 2526 else { 2527 m_adj(m, hdroptlen); 2528 sbappendstream(&(so)->so_rcv, m); 2529 } 2530 sorwakeup(so); 2531 } else { 2532 m_adj(m, hdroptlen); 2533 tiflags = tcp_reass(tp, th, m, &tlen); 2534 tp->t_flags |= TF_ACKNOW; 2535 } 2536 TCP_REASS_UNLOCK(tp); 2537 2538 /* 2539 * Note the amount of data that peer has sent into 2540 * our window, in order to estimate the sender's 2541 * buffer size. 2542 */ 2543 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2544 } else { 2545 m_freem(m); 2546 m = NULL; 2547 tiflags &= ~TH_FIN; 2548 } 2549 2550 /* 2551 * If FIN is received ACK the FIN and let the user know 2552 * that the connection is closing. Ignore a FIN received before 2553 * the connection is fully established. 2554 */ 2555 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2556 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2557 socantrcvmore(so); 2558 tp->t_flags |= TF_ACKNOW; 2559 tp->rcv_nxt++; 2560 } 2561 switch (tp->t_state) { 2562 2563 /* 2564 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2565 */ 2566 case TCPS_ESTABLISHED: 2567 tp->t_state = TCPS_CLOSE_WAIT; 2568 break; 2569 2570 /* 2571 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2572 * enter the CLOSING state. 2573 */ 2574 case TCPS_FIN_WAIT_1: 2575 tp->t_state = TCPS_CLOSING; 2576 break; 2577 2578 /* 2579 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2580 * starting the time-wait timer, turning off the other 2581 * standard timers. 2582 */ 2583 case TCPS_FIN_WAIT_2: 2584 tp->t_state = TCPS_TIME_WAIT; 2585 tcp_canceltimers(tp); 2586 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2587 soisdisconnected(so); 2588 break; 2589 2590 /* 2591 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2592 */ 2593 case TCPS_TIME_WAIT: 2594 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2595 break; 2596 } 2597 } 2598 #ifdef TCP_DEBUG 2599 if (so->so_options & SO_DEBUG) 2600 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2601 #endif 2602 2603 /* 2604 * Return any desired output. 2605 */ 2606 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2607 (void) tcp_output(tp); 2608 } 2609 if (tcp_saveti) 2610 m_freem(tcp_saveti); 2611 return; 2612 2613 badsyn: 2614 /* 2615 * Received a bad SYN. Increment counters and dropwithreset. 2616 */ 2617 tcpstat.tcps_badsyn++; 2618 tp = NULL; 2619 goto dropwithreset; 2620 2621 dropafterack: 2622 /* 2623 * Generate an ACK dropping incoming segment if it occupies 2624 * sequence space, where the ACK reflects our state. 2625 */ 2626 if (tiflags & TH_RST) 2627 goto drop; 2628 goto dropafterack2; 2629 2630 dropafterack_ratelim: 2631 /* 2632 * We may want to rate-limit ACKs against SYN/RST attack. 2633 */ 2634 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2635 tcp_ackdrop_ppslim) == 0) { 2636 /* XXX stat */ 2637 goto drop; 2638 } 2639 /* ...fall into dropafterack2... */ 2640 2641 dropafterack2: 2642 m_freem(m); 2643 tp->t_flags |= TF_ACKNOW; 2644 (void) tcp_output(tp); 2645 if (tcp_saveti) 2646 m_freem(tcp_saveti); 2647 return; 2648 2649 dropwithreset_ratelim: 2650 /* 2651 * We may want to rate-limit RSTs in certain situations, 2652 * particularly if we are sending an RST in response to 2653 * an attempt to connect to or otherwise communicate with 2654 * a port for which we have no socket. 2655 */ 2656 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2657 tcp_rst_ppslim) == 0) { 2658 /* XXX stat */ 2659 goto drop; 2660 } 2661 /* ...fall into dropwithreset... */ 2662 2663 dropwithreset: 2664 /* 2665 * Generate a RST, dropping incoming segment. 2666 * Make ACK acceptable to originator of segment. 2667 */ 2668 if (tiflags & TH_RST) 2669 goto drop; 2670 2671 switch (af) { 2672 #ifdef INET6 2673 case AF_INET6: 2674 /* For following calls to tcp_respond */ 2675 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2676 goto drop; 2677 break; 2678 #endif /* INET6 */ 2679 case AF_INET: 2680 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2681 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2682 goto drop; 2683 } 2684 2685 if (tiflags & TH_ACK) 2686 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2687 else { 2688 if (tiflags & TH_SYN) 2689 tlen++; 2690 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2691 TH_RST|TH_ACK); 2692 } 2693 if (tcp_saveti) 2694 m_freem(tcp_saveti); 2695 return; 2696 2697 badcsum: 2698 drop: 2699 /* 2700 * Drop space held by incoming segment and return. 2701 */ 2702 if (tp) { 2703 if (tp->t_inpcb) 2704 so = tp->t_inpcb->inp_socket; 2705 #ifdef INET6 2706 else if (tp->t_in6pcb) 2707 so = tp->t_in6pcb->in6p_socket; 2708 #endif 2709 else 2710 so = NULL; 2711 #ifdef TCP_DEBUG 2712 if (so && (so->so_options & SO_DEBUG) != 0) 2713 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2714 #endif 2715 } 2716 if (tcp_saveti) 2717 m_freem(tcp_saveti); 2718 m_freem(m); 2719 return; 2720 } 2721 2722 #ifdef TCP_SIGNATURE 2723 int 2724 tcp_signature_apply(void *fstate, caddr_t data, u_int len) 2725 { 2726 2727 MD5Update(fstate, (u_char *)data, len); 2728 return (0); 2729 } 2730 2731 struct secasvar * 2732 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 2733 { 2734 struct secasvar *sav; 2735 #ifdef FAST_IPSEC 2736 union sockaddr_union dst; 2737 #endif 2738 struct ip *ip; 2739 struct ip6_hdr *ip6; 2740 2741 ip = mtod(m, struct ip *); 2742 switch (ip->ip_v) { 2743 case 4: 2744 ip = mtod(m, struct ip *); 2745 ip6 = NULL; 2746 break; 2747 case 6: 2748 ip = NULL; 2749 ip6 = mtod(m, struct ip6_hdr *); 2750 break; 2751 default: 2752 return (NULL); 2753 } 2754 2755 #ifdef FAST_IPSEC 2756 /* Extract the destination from the IP header in the mbuf. */ 2757 bzero(&dst, sizeof(union sockaddr_union)); 2758 dst.sa.sa_len = sizeof(struct sockaddr_in); 2759 dst.sa.sa_family = AF_INET; 2760 dst.sin.sin_addr = ip->ip_dst; 2761 2762 /* 2763 * Look up an SADB entry which matches the address of the peer. 2764 */ 2765 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2766 #else 2767 if (ip) 2768 sav = key_allocsa(AF_INET, (caddr_t)&ip->ip_src, 2769 (caddr_t)&ip->ip_dst, IPPROTO_TCP, 2770 htonl(TCP_SIG_SPI), 0, 0); 2771 else 2772 sav = key_allocsa(AF_INET6, (caddr_t)&ip6->ip6_src, 2773 (caddr_t)&ip6->ip6_dst, IPPROTO_TCP, 2774 htonl(TCP_SIG_SPI), 0, 0); 2775 #endif 2776 2777 return (sav); /* freesav must be performed by caller */ 2778 } 2779 2780 int 2781 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 2782 struct secasvar *sav, char *sig) 2783 { 2784 MD5_CTX ctx; 2785 struct ip *ip; 2786 struct ipovly *ipovly; 2787 struct ip6_hdr *ip6; 2788 struct ippseudo ippseudo; 2789 struct ip6_hdr_pseudo ip6pseudo; 2790 struct tcphdr th0; 2791 int l, tcphdrlen; 2792 2793 if (sav == NULL) 2794 return (-1); 2795 2796 tcphdrlen = th->th_off * 4; 2797 2798 switch (mtod(m, struct ip *)->ip_v) { 2799 case 4: 2800 ip = mtod(m, struct ip *); 2801 ip6 = NULL; 2802 break; 2803 case 6: 2804 ip = NULL; 2805 ip6 = mtod(m, struct ip6_hdr *); 2806 break; 2807 default: 2808 return (-1); 2809 } 2810 2811 MD5Init(&ctx); 2812 2813 if (ip) { 2814 memset(&ippseudo, 0, sizeof(ippseudo)); 2815 ipovly = (struct ipovly *)ip; 2816 ippseudo.ippseudo_src = ipovly->ih_src; 2817 ippseudo.ippseudo_dst = ipovly->ih_dst; 2818 ippseudo.ippseudo_pad = 0; 2819 ippseudo.ippseudo_p = IPPROTO_TCP; 2820 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 2821 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 2822 } else { 2823 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 2824 ip6pseudo.ip6ph_src = ip6->ip6_src; 2825 in6_clearscope(&ip6pseudo.ip6ph_src); 2826 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 2827 in6_clearscope(&ip6pseudo.ip6ph_dst); 2828 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 2829 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 2830 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 2831 } 2832 2833 th0 = *th; 2834 th0.th_sum = 0; 2835 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 2836 2837 l = m->m_pkthdr.len - thoff - tcphdrlen; 2838 if (l > 0) 2839 m_apply(m, thoff + tcphdrlen, 2840 m->m_pkthdr.len - thoff - tcphdrlen, 2841 tcp_signature_apply, &ctx); 2842 2843 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2844 MD5Final(sig, &ctx); 2845 2846 return (0); 2847 } 2848 #endif 2849 2850 static int 2851 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, 2852 const struct tcphdr *th, 2853 struct mbuf *m, int toff, struct tcp_opt_info *oi) 2854 { 2855 u_int16_t mss; 2856 int opt, optlen = 0; 2857 #ifdef TCP_SIGNATURE 2858 caddr_t sigp = NULL; 2859 char sigbuf[TCP_SIGLEN]; 2860 struct secasvar *sav = NULL; 2861 #endif 2862 2863 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 2864 opt = cp[0]; 2865 if (opt == TCPOPT_EOL) 2866 break; 2867 if (opt == TCPOPT_NOP) 2868 optlen = 1; 2869 else { 2870 if (cnt < 2) 2871 break; 2872 optlen = cp[1]; 2873 if (optlen < 2 || optlen > cnt) 2874 break; 2875 } 2876 switch (opt) { 2877 2878 default: 2879 continue; 2880 2881 case TCPOPT_MAXSEG: 2882 if (optlen != TCPOLEN_MAXSEG) 2883 continue; 2884 if (!(th->th_flags & TH_SYN)) 2885 continue; 2886 if (TCPS_HAVERCVDSYN(tp->t_state)) 2887 continue; 2888 bcopy(cp + 2, &mss, sizeof(mss)); 2889 oi->maxseg = ntohs(mss); 2890 break; 2891 2892 case TCPOPT_WINDOW: 2893 if (optlen != TCPOLEN_WINDOW) 2894 continue; 2895 if (!(th->th_flags & TH_SYN)) 2896 continue; 2897 if (TCPS_HAVERCVDSYN(tp->t_state)) 2898 continue; 2899 tp->t_flags |= TF_RCVD_SCALE; 2900 tp->requested_s_scale = cp[2]; 2901 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2902 #if 0 /*XXX*/ 2903 char *p; 2904 2905 if (ip) 2906 p = ntohl(ip->ip_src); 2907 #ifdef INET6 2908 else if (ip6) 2909 p = ip6_sprintf(&ip6->ip6_src); 2910 #endif 2911 else 2912 p = "(unknown)"; 2913 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2914 "assuming %d\n", 2915 tp->requested_s_scale, p, 2916 TCP_MAX_WINSHIFT); 2917 #else 2918 log(LOG_ERR, "TCP: invalid wscale %d, " 2919 "assuming %d\n", 2920 tp->requested_s_scale, 2921 TCP_MAX_WINSHIFT); 2922 #endif 2923 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2924 } 2925 break; 2926 2927 case TCPOPT_TIMESTAMP: 2928 if (optlen != TCPOLEN_TIMESTAMP) 2929 continue; 2930 oi->ts_present = 1; 2931 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2932 NTOHL(oi->ts_val); 2933 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2934 NTOHL(oi->ts_ecr); 2935 2936 if (!(th->th_flags & TH_SYN)) 2937 continue; 2938 if (TCPS_HAVERCVDSYN(tp->t_state)) 2939 continue; 2940 /* 2941 * A timestamp received in a SYN makes 2942 * it ok to send timestamp requests and replies. 2943 */ 2944 tp->t_flags |= TF_RCVD_TSTMP; 2945 tp->ts_recent = oi->ts_val; 2946 tp->ts_recent_age = tcp_now; 2947 break; 2948 2949 case TCPOPT_SACK_PERMITTED: 2950 if (optlen != TCPOLEN_SACK_PERMITTED) 2951 continue; 2952 if (!(th->th_flags & TH_SYN)) 2953 continue; 2954 if (TCPS_HAVERCVDSYN(tp->t_state)) 2955 continue; 2956 if (tcp_do_sack) { 2957 tp->t_flags |= TF_SACK_PERMIT; 2958 tp->t_flags |= TF_WILL_SACK; 2959 } 2960 break; 2961 2962 case TCPOPT_SACK: 2963 tcp_sack_option(tp, th, cp, optlen); 2964 break; 2965 #ifdef TCP_SIGNATURE 2966 case TCPOPT_SIGNATURE: 2967 if (optlen != TCPOLEN_SIGNATURE) 2968 continue; 2969 if (sigp && bcmp(sigp, cp + 2, TCP_SIGLEN)) 2970 return (-1); 2971 2972 sigp = sigbuf; 2973 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 2974 memset(cp + 2, 0, TCP_SIGLEN); 2975 tp->t_flags |= TF_SIGNATURE; 2976 break; 2977 #endif 2978 } 2979 } 2980 2981 #ifdef TCP_SIGNATURE 2982 if (tp->t_flags & TF_SIGNATURE) { 2983 2984 sav = tcp_signature_getsav(m, th); 2985 2986 if (sav == NULL && tp->t_state == TCPS_LISTEN) 2987 return (-1); 2988 } 2989 2990 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 2991 if (sav == NULL) 2992 return (-1); 2993 #ifdef FAST_IPSEC 2994 KEY_FREESAV(&sav); 2995 #else 2996 key_freesav(sav); 2997 #endif 2998 return (-1); 2999 } 3000 3001 if (sigp) { 3002 char sig[TCP_SIGLEN]; 3003 3004 TCP_FIELDS_TO_NET(th); 3005 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3006 TCP_FIELDS_TO_HOST(th); 3007 if (sav == NULL) 3008 return (-1); 3009 #ifdef FAST_IPSEC 3010 KEY_FREESAV(&sav); 3011 #else 3012 key_freesav(sav); 3013 #endif 3014 return (-1); 3015 } 3016 TCP_FIELDS_TO_HOST(th); 3017 3018 if (bcmp(sig, sigp, TCP_SIGLEN)) { 3019 tcpstat.tcps_badsig++; 3020 if (sav == NULL) 3021 return (-1); 3022 #ifdef FAST_IPSEC 3023 KEY_FREESAV(&sav); 3024 #else 3025 key_freesav(sav); 3026 #endif 3027 return (-1); 3028 } else 3029 tcpstat.tcps_goodsig++; 3030 3031 key_sa_recordxfer(sav, m); 3032 #ifdef FAST_IPSEC 3033 KEY_FREESAV(&sav); 3034 #else 3035 key_freesav(sav); 3036 #endif 3037 } 3038 #endif 3039 3040 return (0); 3041 } 3042 3043 /* 3044 * Pull out of band byte out of a segment so 3045 * it doesn't appear in the user's data queue. 3046 * It is still reflected in the segment length for 3047 * sequencing purposes. 3048 */ 3049 void 3050 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3051 struct mbuf *m, int off) 3052 { 3053 int cnt = off + th->th_urp - 1; 3054 3055 while (cnt >= 0) { 3056 if (m->m_len > cnt) { 3057 char *cp = mtod(m, caddr_t) + cnt; 3058 struct tcpcb *tp = sototcpcb(so); 3059 3060 tp->t_iobc = *cp; 3061 tp->t_oobflags |= TCPOOB_HAVEDATA; 3062 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3063 m->m_len--; 3064 return; 3065 } 3066 cnt -= m->m_len; 3067 m = m->m_next; 3068 if (m == 0) 3069 break; 3070 } 3071 panic("tcp_pulloutofband"); 3072 } 3073 3074 /* 3075 * Collect new round-trip time estimate 3076 * and update averages and current timeout. 3077 */ 3078 void 3079 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3080 { 3081 int32_t delta; 3082 3083 tcpstat.tcps_rttupdated++; 3084 if (tp->t_srtt != 0) { 3085 /* 3086 * srtt is stored as fixed point with 3 bits after the 3087 * binary point (i.e., scaled by 8). The following magic 3088 * is equivalent to the smoothing algorithm in rfc793 with 3089 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3090 * point). Adjust rtt to origin 0. 3091 */ 3092 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3093 if ((tp->t_srtt += delta) <= 0) 3094 tp->t_srtt = 1 << 2; 3095 /* 3096 * We accumulate a smoothed rtt variance (actually, a 3097 * smoothed mean difference), then set the retransmit 3098 * timer to smoothed rtt + 4 times the smoothed variance. 3099 * rttvar is stored as fixed point with 2 bits after the 3100 * binary point (scaled by 4). The following is 3101 * equivalent to rfc793 smoothing with an alpha of .75 3102 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3103 * rfc793's wired-in beta. 3104 */ 3105 if (delta < 0) 3106 delta = -delta; 3107 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3108 if ((tp->t_rttvar += delta) <= 0) 3109 tp->t_rttvar = 1 << 2; 3110 } else { 3111 /* 3112 * No rtt measurement yet - use the unsmoothed rtt. 3113 * Set the variance to half the rtt (so our first 3114 * retransmit happens at 3*rtt). 3115 */ 3116 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3117 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3118 } 3119 tp->t_rtttime = 0; 3120 tp->t_rxtshift = 0; 3121 3122 /* 3123 * the retransmit should happen at rtt + 4 * rttvar. 3124 * Because of the way we do the smoothing, srtt and rttvar 3125 * will each average +1/2 tick of bias. When we compute 3126 * the retransmit timer, we want 1/2 tick of rounding and 3127 * 1 extra tick because of +-1/2 tick uncertainty in the 3128 * firing of the timer. The bias will give us exactly the 3129 * 1.5 tick we need. But, because the bias is 3130 * statistical, we have to test that we don't drop below 3131 * the minimum feasible timer (which is 2 ticks). 3132 */ 3133 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3134 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3135 3136 /* 3137 * We received an ack for a packet that wasn't retransmitted; 3138 * it is probably safe to discard any error indications we've 3139 * received recently. This isn't quite right, but close enough 3140 * for now (a route might have failed after we sent a segment, 3141 * and the return path might not be symmetrical). 3142 */ 3143 tp->t_softerror = 0; 3144 } 3145 3146 3147 /* 3148 * TCP compressed state engine. Currently used to hold compressed 3149 * state for SYN_RECEIVED. 3150 */ 3151 3152 u_long syn_cache_count; 3153 u_int32_t syn_hash1, syn_hash2; 3154 3155 #define SYN_HASH(sa, sp, dp) \ 3156 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3157 ((u_int32_t)(sp)))^syn_hash2))) 3158 #ifndef INET6 3159 #define SYN_HASHALL(hash, src, dst) \ 3160 do { \ 3161 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3162 ((const struct sockaddr_in *)(src))->sin_port, \ 3163 ((const struct sockaddr_in *)(dst))->sin_port); \ 3164 } while (/*CONSTCOND*/ 0) 3165 #else 3166 #define SYN_HASH6(sa, sp, dp) \ 3167 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3168 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3169 & 0x7fffffff) 3170 3171 #define SYN_HASHALL(hash, src, dst) \ 3172 do { \ 3173 switch ((src)->sa_family) { \ 3174 case AF_INET: \ 3175 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3176 ((const struct sockaddr_in *)(src))->sin_port, \ 3177 ((const struct sockaddr_in *)(dst))->sin_port); \ 3178 break; \ 3179 case AF_INET6: \ 3180 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3181 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3182 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3183 break; \ 3184 default: \ 3185 hash = 0; \ 3186 } \ 3187 } while (/*CONSTCOND*/0) 3188 #endif /* INET6 */ 3189 3190 #define SYN_CACHE_RM(sc) \ 3191 do { \ 3192 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 3193 (sc), sc_bucketq); \ 3194 (sc)->sc_tp = NULL; \ 3195 LIST_REMOVE((sc), sc_tpq); \ 3196 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 3197 callout_stop(&(sc)->sc_timer); \ 3198 syn_cache_count--; \ 3199 } while (/*CONSTCOND*/0) 3200 3201 #define SYN_CACHE_PUT(sc) \ 3202 do { \ 3203 if ((sc)->sc_ipopts) \ 3204 (void) m_free((sc)->sc_ipopts); \ 3205 if ((sc)->sc_route4.ro_rt != NULL) \ 3206 RTFREE((sc)->sc_route4.ro_rt); \ 3207 if (callout_invoking(&(sc)->sc_timer)) \ 3208 (sc)->sc_flags |= SCF_DEAD; \ 3209 else \ 3210 pool_put(&syn_cache_pool, (sc)); \ 3211 } while (/*CONSTCOND*/0) 3212 3213 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL); 3214 3215 /* 3216 * We don't estimate RTT with SYNs, so each packet starts with the default 3217 * RTT and each timer step has a fixed timeout value. 3218 */ 3219 #define SYN_CACHE_TIMER_ARM(sc) \ 3220 do { \ 3221 TCPT_RANGESET((sc)->sc_rxtcur, \ 3222 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3223 TCPTV_REXMTMAX); \ 3224 callout_reset(&(sc)->sc_timer, \ 3225 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3226 } while (/*CONSTCOND*/0) 3227 3228 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3229 3230 void 3231 syn_cache_init(void) 3232 { 3233 int i; 3234 3235 /* Initialize the hash buckets. */ 3236 for (i = 0; i < tcp_syn_cache_size; i++) 3237 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3238 } 3239 3240 void 3241 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3242 { 3243 struct syn_cache_head *scp; 3244 struct syn_cache *sc2; 3245 int s; 3246 3247 /* 3248 * If there are no entries in the hash table, reinitialize 3249 * the hash secrets. 3250 */ 3251 if (syn_cache_count == 0) { 3252 syn_hash1 = arc4random(); 3253 syn_hash2 = arc4random(); 3254 } 3255 3256 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3257 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3258 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3259 3260 /* 3261 * Make sure that we don't overflow the per-bucket 3262 * limit or the total cache size limit. 3263 */ 3264 s = splsoftnet(); 3265 if (scp->sch_length >= tcp_syn_bucket_limit) { 3266 tcpstat.tcps_sc_bucketoverflow++; 3267 /* 3268 * The bucket is full. Toss the oldest element in the 3269 * bucket. This will be the first entry in the bucket. 3270 */ 3271 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3272 #ifdef DIAGNOSTIC 3273 /* 3274 * This should never happen; we should always find an 3275 * entry in our bucket. 3276 */ 3277 if (sc2 == NULL) 3278 panic("syn_cache_insert: bucketoverflow: impossible"); 3279 #endif 3280 SYN_CACHE_RM(sc2); 3281 SYN_CACHE_PUT(sc2); /* calls pool_put but see spl above */ 3282 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3283 struct syn_cache_head *scp2, *sce; 3284 3285 tcpstat.tcps_sc_overflowed++; 3286 /* 3287 * The cache is full. Toss the oldest entry in the 3288 * first non-empty bucket we can find. 3289 * 3290 * XXX We would really like to toss the oldest 3291 * entry in the cache, but we hope that this 3292 * condition doesn't happen very often. 3293 */ 3294 scp2 = scp; 3295 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3296 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3297 for (++scp2; scp2 != scp; scp2++) { 3298 if (scp2 >= sce) 3299 scp2 = &tcp_syn_cache[0]; 3300 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3301 break; 3302 } 3303 #ifdef DIAGNOSTIC 3304 /* 3305 * This should never happen; we should always find a 3306 * non-empty bucket. 3307 */ 3308 if (scp2 == scp) 3309 panic("syn_cache_insert: cacheoverflow: " 3310 "impossible"); 3311 #endif 3312 } 3313 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3314 SYN_CACHE_RM(sc2); 3315 SYN_CACHE_PUT(sc2); /* calls pool_put but see spl above */ 3316 } 3317 3318 /* 3319 * Initialize the entry's timer. 3320 */ 3321 sc->sc_rxttot = 0; 3322 sc->sc_rxtshift = 0; 3323 SYN_CACHE_TIMER_ARM(sc); 3324 3325 /* Link it from tcpcb entry */ 3326 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3327 3328 /* Put it into the bucket. */ 3329 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3330 scp->sch_length++; 3331 syn_cache_count++; 3332 3333 tcpstat.tcps_sc_added++; 3334 splx(s); 3335 } 3336 3337 /* 3338 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3339 * If we have retransmitted an entry the maximum number of times, expire 3340 * that entry. 3341 */ 3342 void 3343 syn_cache_timer(void *arg) 3344 { 3345 struct syn_cache *sc = arg; 3346 int s; 3347 3348 s = splsoftnet(); 3349 callout_ack(&sc->sc_timer); 3350 3351 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3352 tcpstat.tcps_sc_delayed_free++; 3353 pool_put(&syn_cache_pool, sc); 3354 splx(s); 3355 return; 3356 } 3357 3358 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3359 /* Drop it -- too many retransmissions. */ 3360 goto dropit; 3361 } 3362 3363 /* 3364 * Compute the total amount of time this entry has 3365 * been on a queue. If this entry has been on longer 3366 * than the keep alive timer would allow, expire it. 3367 */ 3368 sc->sc_rxttot += sc->sc_rxtcur; 3369 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) 3370 goto dropit; 3371 3372 tcpstat.tcps_sc_retransmitted++; 3373 (void) syn_cache_respond(sc, NULL); 3374 3375 /* Advance the timer back-off. */ 3376 sc->sc_rxtshift++; 3377 SYN_CACHE_TIMER_ARM(sc); 3378 3379 splx(s); 3380 return; 3381 3382 dropit: 3383 tcpstat.tcps_sc_timed_out++; 3384 SYN_CACHE_RM(sc); 3385 SYN_CACHE_PUT(sc); /* calls pool_put but see spl above */ 3386 splx(s); 3387 } 3388 3389 /* 3390 * Remove syn cache created by the specified tcb entry, 3391 * because this does not make sense to keep them 3392 * (if there's no tcb entry, syn cache entry will never be used) 3393 */ 3394 void 3395 syn_cache_cleanup(struct tcpcb *tp) 3396 { 3397 struct syn_cache *sc, *nsc; 3398 int s; 3399 3400 s = splsoftnet(); 3401 3402 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3403 nsc = LIST_NEXT(sc, sc_tpq); 3404 3405 #ifdef DIAGNOSTIC 3406 if (sc->sc_tp != tp) 3407 panic("invalid sc_tp in syn_cache_cleanup"); 3408 #endif 3409 SYN_CACHE_RM(sc); 3410 SYN_CACHE_PUT(sc); /* calls pool_put but see spl above */ 3411 } 3412 /* just for safety */ 3413 LIST_INIT(&tp->t_sc); 3414 3415 splx(s); 3416 } 3417 3418 /* 3419 * Find an entry in the syn cache. 3420 */ 3421 struct syn_cache * 3422 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3423 struct syn_cache_head **headp) 3424 { 3425 struct syn_cache *sc; 3426 struct syn_cache_head *scp; 3427 u_int32_t hash; 3428 int s; 3429 3430 SYN_HASHALL(hash, src, dst); 3431 3432 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3433 *headp = scp; 3434 s = splsoftnet(); 3435 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3436 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3437 if (sc->sc_hash != hash) 3438 continue; 3439 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3440 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3441 splx(s); 3442 return (sc); 3443 } 3444 } 3445 splx(s); 3446 return (NULL); 3447 } 3448 3449 /* 3450 * This function gets called when we receive an ACK for a 3451 * socket in the LISTEN state. We look up the connection 3452 * in the syn cache, and if its there, we pull it out of 3453 * the cache and turn it into a full-blown connection in 3454 * the SYN-RECEIVED state. 3455 * 3456 * The return values may not be immediately obvious, and their effects 3457 * can be subtle, so here they are: 3458 * 3459 * NULL SYN was not found in cache; caller should drop the 3460 * packet and send an RST. 3461 * 3462 * -1 We were unable to create the new connection, and are 3463 * aborting it. An ACK,RST is being sent to the peer 3464 * (unless we got screwey sequence numbners; see below), 3465 * because the 3-way handshake has been completed. Caller 3466 * should not free the mbuf, since we may be using it. If 3467 * we are not, we will free it. 3468 * 3469 * Otherwise, the return value is a pointer to the new socket 3470 * associated with the connection. 3471 */ 3472 struct socket * 3473 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3474 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3475 struct socket *so, struct mbuf *m) 3476 { 3477 struct syn_cache *sc; 3478 struct syn_cache_head *scp; 3479 struct inpcb *inp = NULL; 3480 #ifdef INET6 3481 struct in6pcb *in6p = NULL; 3482 #endif 3483 struct tcpcb *tp = 0; 3484 struct mbuf *am; 3485 int s; 3486 struct socket *oso; 3487 3488 s = splsoftnet(); 3489 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3490 splx(s); 3491 return (NULL); 3492 } 3493 3494 /* 3495 * Verify the sequence and ack numbers. Try getting the correct 3496 * response again. 3497 */ 3498 if ((th->th_ack != sc->sc_iss + 1) || 3499 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3500 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3501 (void) syn_cache_respond(sc, m); 3502 splx(s); 3503 return ((struct socket *)(-1)); 3504 } 3505 3506 /* Remove this cache entry */ 3507 SYN_CACHE_RM(sc); 3508 splx(s); 3509 3510 /* 3511 * Ok, create the full blown connection, and set things up 3512 * as they would have been set up if we had created the 3513 * connection when the SYN arrived. If we can't create 3514 * the connection, abort it. 3515 */ 3516 /* 3517 * inp still has the OLD in_pcb stuff, set the 3518 * v6-related flags on the new guy, too. This is 3519 * done particularly for the case where an AF_INET6 3520 * socket is bound only to a port, and a v4 connection 3521 * comes in on that port. 3522 * we also copy the flowinfo from the original pcb 3523 * to the new one. 3524 */ 3525 oso = so; 3526 so = sonewconn(so, SS_ISCONNECTED); 3527 if (so == NULL) 3528 goto resetandabort; 3529 3530 switch (so->so_proto->pr_domain->dom_family) { 3531 #ifdef INET 3532 case AF_INET: 3533 inp = sotoinpcb(so); 3534 break; 3535 #endif 3536 #ifdef INET6 3537 case AF_INET6: 3538 in6p = sotoin6pcb(so); 3539 break; 3540 #endif 3541 } 3542 switch (src->sa_family) { 3543 #ifdef INET 3544 case AF_INET: 3545 if (inp) { 3546 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3547 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3548 inp->inp_options = ip_srcroute(); 3549 in_pcbstate(inp, INP_BOUND); 3550 if (inp->inp_options == NULL) { 3551 inp->inp_options = sc->sc_ipopts; 3552 sc->sc_ipopts = NULL; 3553 } 3554 } 3555 #ifdef INET6 3556 else if (in6p) { 3557 /* IPv4 packet to AF_INET6 socket */ 3558 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3559 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3560 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3561 &in6p->in6p_laddr.s6_addr32[3], 3562 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3563 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3564 in6totcpcb(in6p)->t_family = AF_INET; 3565 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3566 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3567 else 3568 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3569 in6_pcbstate(in6p, IN6P_BOUND); 3570 } 3571 #endif 3572 break; 3573 #endif 3574 #ifdef INET6 3575 case AF_INET6: 3576 if (in6p) { 3577 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3578 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3579 in6_pcbstate(in6p, IN6P_BOUND); 3580 } 3581 break; 3582 #endif 3583 } 3584 #ifdef INET6 3585 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3586 struct in6pcb *oin6p = sotoin6pcb(oso); 3587 /* inherit socket options from the listening socket */ 3588 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3589 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3590 m_freem(in6p->in6p_options); 3591 in6p->in6p_options = 0; 3592 } 3593 ip6_savecontrol(in6p, &in6p->in6p_options, 3594 mtod(m, struct ip6_hdr *), m); 3595 } 3596 #endif 3597 3598 #if defined(IPSEC) || defined(FAST_IPSEC) 3599 /* 3600 * we make a copy of policy, instead of sharing the policy, 3601 * for better behavior in terms of SA lookup and dead SA removal. 3602 */ 3603 if (inp) { 3604 /* copy old policy into new socket's */ 3605 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3606 printf("tcp_input: could not copy policy\n"); 3607 } 3608 #ifdef INET6 3609 else if (in6p) { 3610 /* copy old policy into new socket's */ 3611 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3612 in6p->in6p_sp)) 3613 printf("tcp_input: could not copy policy\n"); 3614 } 3615 #endif 3616 #endif 3617 3618 /* 3619 * Give the new socket our cached route reference. 3620 */ 3621 if (inp) 3622 inp->inp_route = sc->sc_route4; /* struct assignment */ 3623 #ifdef INET6 3624 else 3625 in6p->in6p_route = sc->sc_route6; 3626 #endif 3627 sc->sc_route4.ro_rt = NULL; 3628 3629 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3630 if (am == NULL) 3631 goto resetandabort; 3632 MCLAIM(am, &tcp_mowner); 3633 am->m_len = src->sa_len; 3634 bcopy(src, mtod(am, caddr_t), src->sa_len); 3635 if (inp) { 3636 if (in_pcbconnect(inp, am, NULL)) { 3637 (void) m_free(am); 3638 goto resetandabort; 3639 } 3640 } 3641 #ifdef INET6 3642 else if (in6p) { 3643 if (src->sa_family == AF_INET) { 3644 /* IPv4 packet to AF_INET6 socket */ 3645 struct sockaddr_in6 *sin6; 3646 sin6 = mtod(am, struct sockaddr_in6 *); 3647 am->m_len = sizeof(*sin6); 3648 bzero(sin6, sizeof(*sin6)); 3649 sin6->sin6_family = AF_INET6; 3650 sin6->sin6_len = sizeof(*sin6); 3651 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3652 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3653 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3654 &sin6->sin6_addr.s6_addr32[3], 3655 sizeof(sin6->sin6_addr.s6_addr32[3])); 3656 } 3657 if (in6_pcbconnect(in6p, am, NULL)) { 3658 (void) m_free(am); 3659 goto resetandabort; 3660 } 3661 } 3662 #endif 3663 else { 3664 (void) m_free(am); 3665 goto resetandabort; 3666 } 3667 (void) m_free(am); 3668 3669 if (inp) 3670 tp = intotcpcb(inp); 3671 #ifdef INET6 3672 else if (in6p) 3673 tp = in6totcpcb(in6p); 3674 #endif 3675 else 3676 tp = NULL; 3677 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3678 if (sc->sc_request_r_scale != 15) { 3679 tp->requested_s_scale = sc->sc_requested_s_scale; 3680 tp->request_r_scale = sc->sc_request_r_scale; 3681 tp->snd_scale = sc->sc_requested_s_scale; 3682 tp->rcv_scale = sc->sc_request_r_scale; 3683 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3684 } 3685 if (sc->sc_flags & SCF_TIMESTAMP) 3686 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3687 tp->ts_timebase = sc->sc_timebase; 3688 3689 tp->t_template = tcp_template(tp); 3690 if (tp->t_template == 0) { 3691 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3692 so = NULL; 3693 m_freem(m); 3694 goto abort; 3695 } 3696 3697 tp->iss = sc->sc_iss; 3698 tp->irs = sc->sc_irs; 3699 tcp_sendseqinit(tp); 3700 tcp_rcvseqinit(tp); 3701 tp->t_state = TCPS_SYN_RECEIVED; 3702 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3703 tcpstat.tcps_accepts++; 3704 3705 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 3706 tp->t_flags |= TF_WILL_SACK; 3707 3708 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 3709 tp->t_flags |= TF_ECN_PERMIT; 3710 3711 #ifdef TCP_SIGNATURE 3712 if (sc->sc_flags & SCF_SIGNATURE) 3713 tp->t_flags |= TF_SIGNATURE; 3714 #endif 3715 3716 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3717 tp->t_ourmss = sc->sc_ourmaxseg; 3718 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3719 3720 /* 3721 * Initialize the initial congestion window. If we 3722 * had to retransmit the SYN,ACK, we must initialize cwnd 3723 * to 1 segment (i.e. the Loss Window). 3724 */ 3725 if (sc->sc_rxtshift) 3726 tp->snd_cwnd = tp->t_peermss; 3727 else { 3728 int ss = tcp_init_win; 3729 #ifdef INET 3730 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3731 ss = tcp_init_win_local; 3732 #endif 3733 #ifdef INET6 3734 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3735 ss = tcp_init_win_local; 3736 #endif 3737 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3738 } 3739 3740 tcp_rmx_rtt(tp); 3741 tp->snd_wl1 = sc->sc_irs; 3742 tp->rcv_up = sc->sc_irs + 1; 3743 3744 /* 3745 * This is what whould have happened in tcp_output() when 3746 * the SYN,ACK was sent. 3747 */ 3748 tp->snd_up = tp->snd_una; 3749 tp->snd_max = tp->snd_nxt = tp->iss+1; 3750 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3751 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3752 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3753 tp->last_ack_sent = tp->rcv_nxt; 3754 tp->t_partialacks = -1; 3755 tp->t_dupacks = 0; 3756 3757 tcpstat.tcps_sc_completed++; 3758 s = splsoftnet(); 3759 SYN_CACHE_PUT(sc); 3760 splx(s); 3761 return (so); 3762 3763 resetandabort: 3764 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3765 abort: 3766 if (so != NULL) 3767 (void) soabort(so); 3768 s = splsoftnet(); 3769 SYN_CACHE_PUT(sc); 3770 splx(s); 3771 tcpstat.tcps_sc_aborted++; 3772 return ((struct socket *)(-1)); 3773 } 3774 3775 /* 3776 * This function is called when we get a RST for a 3777 * non-existent connection, so that we can see if the 3778 * connection is in the syn cache. If it is, zap it. 3779 */ 3780 3781 void 3782 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 3783 { 3784 struct syn_cache *sc; 3785 struct syn_cache_head *scp; 3786 int s = splsoftnet(); 3787 3788 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3789 splx(s); 3790 return; 3791 } 3792 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3793 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3794 splx(s); 3795 return; 3796 } 3797 SYN_CACHE_RM(sc); 3798 tcpstat.tcps_sc_reset++; 3799 SYN_CACHE_PUT(sc); /* calls pool_put but see spl above */ 3800 splx(s); 3801 } 3802 3803 void 3804 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 3805 struct tcphdr *th) 3806 { 3807 struct syn_cache *sc; 3808 struct syn_cache_head *scp; 3809 int s; 3810 3811 s = splsoftnet(); 3812 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3813 splx(s); 3814 return; 3815 } 3816 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3817 if (ntohl (th->th_seq) != sc->sc_iss) { 3818 splx(s); 3819 return; 3820 } 3821 3822 /* 3823 * If we've retransmitted 3 times and this is our second error, 3824 * we remove the entry. Otherwise, we allow it to continue on. 3825 * This prevents us from incorrectly nuking an entry during a 3826 * spurious network outage. 3827 * 3828 * See tcp_notify(). 3829 */ 3830 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3831 sc->sc_flags |= SCF_UNREACH; 3832 splx(s); 3833 return; 3834 } 3835 3836 SYN_CACHE_RM(sc); 3837 tcpstat.tcps_sc_unreach++; 3838 SYN_CACHE_PUT(sc); /* calls pool_put but see spl above */ 3839 splx(s); 3840 } 3841 3842 /* 3843 * Given a LISTEN socket and an inbound SYN request, add 3844 * this to the syn cache, and send back a segment: 3845 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3846 * to the source. 3847 * 3848 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3849 * Doing so would require that we hold onto the data and deliver it 3850 * to the application. However, if we are the target of a SYN-flood 3851 * DoS attack, an attacker could send data which would eventually 3852 * consume all available buffer space if it were ACKed. By not ACKing 3853 * the data, we avoid this DoS scenario. 3854 */ 3855 3856 int 3857 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3858 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 3859 int optlen, struct tcp_opt_info *oi) 3860 { 3861 struct tcpcb tb, *tp; 3862 long win; 3863 struct syn_cache *sc; 3864 struct syn_cache_head *scp; 3865 struct mbuf *ipopts; 3866 struct tcp_opt_info opti; 3867 int s; 3868 3869 tp = sototcpcb(so); 3870 3871 bzero(&opti, sizeof(opti)); 3872 3873 /* 3874 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3875 * 3876 * Note this check is performed in tcp_input() very early on. 3877 */ 3878 3879 /* 3880 * Initialize some local state. 3881 */ 3882 win = sbspace(&so->so_rcv); 3883 if (win > TCP_MAXWIN) 3884 win = TCP_MAXWIN; 3885 3886 switch (src->sa_family) { 3887 #ifdef INET 3888 case AF_INET: 3889 /* 3890 * Remember the IP options, if any. 3891 */ 3892 ipopts = ip_srcroute(); 3893 break; 3894 #endif 3895 default: 3896 ipopts = NULL; 3897 } 3898 3899 #ifdef TCP_SIGNATURE 3900 if (optp || (tp->t_flags & TF_SIGNATURE)) 3901 #else 3902 if (optp) 3903 #endif 3904 { 3905 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3906 #ifdef TCP_SIGNATURE 3907 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 3908 #endif 3909 tb.t_state = TCPS_LISTEN; 3910 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 3911 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 3912 return (0); 3913 } else 3914 tb.t_flags = 0; 3915 3916 /* 3917 * See if we already have an entry for this connection. 3918 * If we do, resend the SYN,ACK. We do not count this 3919 * as a retransmission (XXX though maybe we should). 3920 */ 3921 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3922 tcpstat.tcps_sc_dupesyn++; 3923 if (ipopts) { 3924 /* 3925 * If we were remembering a previous source route, 3926 * forget it and use the new one we've been given. 3927 */ 3928 if (sc->sc_ipopts) 3929 (void) m_free(sc->sc_ipopts); 3930 sc->sc_ipopts = ipopts; 3931 } 3932 sc->sc_timestamp = tb.ts_recent; 3933 if (syn_cache_respond(sc, m) == 0) { 3934 tcpstat.tcps_sndacks++; 3935 tcpstat.tcps_sndtotal++; 3936 } 3937 return (1); 3938 } 3939 3940 s = splsoftnet(); 3941 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3942 splx(s); 3943 if (sc == NULL) { 3944 if (ipopts) 3945 (void) m_free(ipopts); 3946 return (0); 3947 } 3948 3949 /* 3950 * Fill in the cache, and put the necessary IP and TCP 3951 * options into the reply. 3952 */ 3953 bzero(sc, sizeof(struct syn_cache)); 3954 callout_init(&sc->sc_timer); 3955 bcopy(src, &sc->sc_src, src->sa_len); 3956 bcopy(dst, &sc->sc_dst, dst->sa_len); 3957 sc->sc_flags = 0; 3958 sc->sc_ipopts = ipopts; 3959 sc->sc_irs = th->th_seq; 3960 switch (src->sa_family) { 3961 #ifdef INET 3962 case AF_INET: 3963 { 3964 struct sockaddr_in *srcin = (void *) src; 3965 struct sockaddr_in *dstin = (void *) dst; 3966 3967 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 3968 &srcin->sin_addr, dstin->sin_port, 3969 srcin->sin_port, sizeof(dstin->sin_addr), 0); 3970 break; 3971 } 3972 #endif /* INET */ 3973 #ifdef INET6 3974 case AF_INET6: 3975 { 3976 struct sockaddr_in6 *srcin6 = (void *) src; 3977 struct sockaddr_in6 *dstin6 = (void *) dst; 3978 3979 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 3980 &srcin6->sin6_addr, dstin6->sin6_port, 3981 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 3982 break; 3983 } 3984 #endif /* INET6 */ 3985 } 3986 sc->sc_peermaxseg = oi->maxseg; 3987 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3988 m->m_pkthdr.rcvif : NULL, 3989 sc->sc_src.sa.sa_family); 3990 sc->sc_win = win; 3991 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 3992 sc->sc_timestamp = tb.ts_recent; 3993 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 3994 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 3995 sc->sc_flags |= SCF_TIMESTAMP; 3996 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3997 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3998 sc->sc_requested_s_scale = tb.requested_s_scale; 3999 sc->sc_request_r_scale = 0; 4000 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4001 TCP_MAXWIN << sc->sc_request_r_scale < 4002 so->so_rcv.sb_hiwat) 4003 sc->sc_request_r_scale++; 4004 } else { 4005 sc->sc_requested_s_scale = 15; 4006 sc->sc_request_r_scale = 15; 4007 } 4008 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4009 sc->sc_flags |= SCF_SACK_PERMIT; 4010 4011 /* 4012 * ECN setup packet recieved. 4013 */ 4014 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4015 sc->sc_flags |= SCF_ECN_PERMIT; 4016 4017 #ifdef TCP_SIGNATURE 4018 if (tb.t_flags & TF_SIGNATURE) 4019 sc->sc_flags |= SCF_SIGNATURE; 4020 #endif 4021 sc->sc_tp = tp; 4022 if (syn_cache_respond(sc, m) == 0) { 4023 syn_cache_insert(sc, tp); 4024 tcpstat.tcps_sndacks++; 4025 tcpstat.tcps_sndtotal++; 4026 } else { 4027 s = splsoftnet(); 4028 SYN_CACHE_PUT(sc); 4029 splx(s); 4030 tcpstat.tcps_sc_dropped++; 4031 } 4032 return (1); 4033 } 4034 4035 int 4036 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4037 { 4038 struct route *ro; 4039 u_int8_t *optp; 4040 int optlen, error; 4041 u_int16_t tlen; 4042 struct ip *ip = NULL; 4043 #ifdef INET6 4044 struct ip6_hdr *ip6 = NULL; 4045 #endif 4046 struct tcpcb *tp = NULL; 4047 struct tcphdr *th; 4048 u_int hlen; 4049 struct socket *so; 4050 4051 switch (sc->sc_src.sa.sa_family) { 4052 case AF_INET: 4053 hlen = sizeof(struct ip); 4054 ro = &sc->sc_route4; 4055 break; 4056 #ifdef INET6 4057 case AF_INET6: 4058 hlen = sizeof(struct ip6_hdr); 4059 ro = (struct route *)&sc->sc_route6; 4060 break; 4061 #endif 4062 default: 4063 if (m) 4064 m_freem(m); 4065 return (EAFNOSUPPORT); 4066 } 4067 4068 /* Compute the size of the TCP options. */ 4069 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4070 ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) + 4071 #ifdef TCP_SIGNATURE 4072 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4073 #endif 4074 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4075 4076 tlen = hlen + sizeof(struct tcphdr) + optlen; 4077 4078 /* 4079 * Create the IP+TCP header from scratch. 4080 */ 4081 if (m) 4082 m_freem(m); 4083 #ifdef DIAGNOSTIC 4084 if (max_linkhdr + tlen > MCLBYTES) 4085 return (ENOBUFS); 4086 #endif 4087 MGETHDR(m, M_DONTWAIT, MT_DATA); 4088 if (m && tlen > MHLEN) { 4089 MCLGET(m, M_DONTWAIT); 4090 if ((m->m_flags & M_EXT) == 0) { 4091 m_freem(m); 4092 m = NULL; 4093 } 4094 } 4095 if (m == NULL) 4096 return (ENOBUFS); 4097 MCLAIM(m, &tcp_tx_mowner); 4098 4099 /* Fixup the mbuf. */ 4100 m->m_data += max_linkhdr; 4101 m->m_len = m->m_pkthdr.len = tlen; 4102 if (sc->sc_tp) { 4103 tp = sc->sc_tp; 4104 if (tp->t_inpcb) 4105 so = tp->t_inpcb->inp_socket; 4106 #ifdef INET6 4107 else if (tp->t_in6pcb) 4108 so = tp->t_in6pcb->in6p_socket; 4109 #endif 4110 else 4111 so = NULL; 4112 } else 4113 so = NULL; 4114 m->m_pkthdr.rcvif = NULL; 4115 memset(mtod(m, u_char *), 0, tlen); 4116 4117 switch (sc->sc_src.sa.sa_family) { 4118 case AF_INET: 4119 ip = mtod(m, struct ip *); 4120 ip->ip_v = 4; 4121 ip->ip_dst = sc->sc_src.sin.sin_addr; 4122 ip->ip_src = sc->sc_dst.sin.sin_addr; 4123 ip->ip_p = IPPROTO_TCP; 4124 th = (struct tcphdr *)(ip + 1); 4125 th->th_dport = sc->sc_src.sin.sin_port; 4126 th->th_sport = sc->sc_dst.sin.sin_port; 4127 break; 4128 #ifdef INET6 4129 case AF_INET6: 4130 ip6 = mtod(m, struct ip6_hdr *); 4131 ip6->ip6_vfc = IPV6_VERSION; 4132 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4133 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4134 ip6->ip6_nxt = IPPROTO_TCP; 4135 /* ip6_plen will be updated in ip6_output() */ 4136 th = (struct tcphdr *)(ip6 + 1); 4137 th->th_dport = sc->sc_src.sin6.sin6_port; 4138 th->th_sport = sc->sc_dst.sin6.sin6_port; 4139 break; 4140 #endif 4141 default: 4142 th = NULL; 4143 } 4144 4145 th->th_seq = htonl(sc->sc_iss); 4146 th->th_ack = htonl(sc->sc_irs + 1); 4147 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4148 th->th_flags = TH_SYN|TH_ACK; 4149 th->th_win = htons(sc->sc_win); 4150 /* th_sum already 0 */ 4151 /* th_urp already 0 */ 4152 4153 /* Tack on the TCP options. */ 4154 optp = (u_int8_t *)(th + 1); 4155 *optp++ = TCPOPT_MAXSEG; 4156 *optp++ = 4; 4157 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4158 *optp++ = sc->sc_ourmaxseg & 0xff; 4159 4160 if (sc->sc_request_r_scale != 15) { 4161 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4162 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4163 sc->sc_request_r_scale); 4164 optp += 4; 4165 } 4166 4167 if (sc->sc_flags & SCF_TIMESTAMP) { 4168 u_int32_t *lp = (u_int32_t *)(optp); 4169 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4170 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4171 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4172 *lp = htonl(sc->sc_timestamp); 4173 optp += TCPOLEN_TSTAMP_APPA; 4174 } 4175 4176 if (sc->sc_flags & SCF_SACK_PERMIT) { 4177 u_int8_t *p = optp; 4178 4179 /* Let the peer know that we will SACK. */ 4180 p[0] = TCPOPT_SACK_PERMITTED; 4181 p[1] = 2; 4182 p[2] = TCPOPT_NOP; 4183 p[3] = TCPOPT_NOP; 4184 optp += 4; 4185 } 4186 4187 /* 4188 * Send ECN SYN-ACK setup packet. 4189 * Routes can be asymetric, so, even if we receive a packet 4190 * with ECE and CWR set, we must not assume no one will block 4191 * the ECE packet we are about to send. 4192 */ 4193 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4194 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4195 th->th_flags |= TH_ECE; 4196 tcpstat.tcps_ecn_shs++; 4197 4198 /* 4199 * draft-ietf-tcpm-ecnsyn-00.txt 4200 * 4201 * "[...] a TCP node MAY respond to an ECN-setup 4202 * SYN packet by setting ECT in the responding 4203 * ECN-setup SYN/ACK packet, indicating to routers 4204 * that the SYN/ACK packet is ECN-Capable. 4205 * This allows a congested router along the path 4206 * to mark the packet instead of dropping the 4207 * packet as an indication of congestion." 4208 * 4209 * "[...] There can be a great benefit in setting 4210 * an ECN-capable codepoint in SYN/ACK packets [...] 4211 * Congestion is most likely to occur in 4212 * the server-to-client direction. As a result, 4213 * setting an ECN-capable codepoint in SYN/ACK 4214 * packets can reduce the occurence of three-second 4215 * retransmit timeouts resulting from the drop 4216 * of SYN/ACK packets." 4217 * 4218 * Page 4 and 6, January 2006. 4219 */ 4220 4221 switch (sc->sc_src.sa.sa_family) { 4222 #ifdef INET 4223 case AF_INET: 4224 ip->ip_tos |= IPTOS_ECN_ECT0; 4225 break; 4226 #endif 4227 #ifdef INET6 4228 case AF_INET6: 4229 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4230 break; 4231 #endif 4232 } 4233 tcpstat.tcps_ecn_ect++; 4234 } 4235 4236 #ifdef TCP_SIGNATURE 4237 if (sc->sc_flags & SCF_SIGNATURE) { 4238 struct secasvar *sav; 4239 u_int8_t *sigp; 4240 4241 sav = tcp_signature_getsav(m, th); 4242 4243 if (sav == NULL) { 4244 if (m) 4245 m_freem(m); 4246 return (EPERM); 4247 } 4248 4249 *optp++ = TCPOPT_SIGNATURE; 4250 *optp++ = TCPOLEN_SIGNATURE; 4251 sigp = optp; 4252 bzero(optp, TCP_SIGLEN); 4253 optp += TCP_SIGLEN; 4254 *optp++ = TCPOPT_NOP; 4255 *optp++ = TCPOPT_EOL; 4256 4257 (void)tcp_signature(m, th, hlen, sav, sigp); 4258 4259 key_sa_recordxfer(sav, m); 4260 #ifdef FAST_IPSEC 4261 KEY_FREESAV(&sav); 4262 #else 4263 key_freesav(sav); 4264 #endif 4265 } 4266 #endif 4267 4268 /* Compute the packet's checksum. */ 4269 switch (sc->sc_src.sa.sa_family) { 4270 case AF_INET: 4271 ip->ip_len = htons(tlen - hlen); 4272 th->th_sum = 0; 4273 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4274 break; 4275 #ifdef INET6 4276 case AF_INET6: 4277 ip6->ip6_plen = htons(tlen - hlen); 4278 th->th_sum = 0; 4279 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4280 break; 4281 #endif 4282 } 4283 4284 /* 4285 * Fill in some straggling IP bits. Note the stack expects 4286 * ip_len to be in host order, for convenience. 4287 */ 4288 switch (sc->sc_src.sa.sa_family) { 4289 #ifdef INET 4290 case AF_INET: 4291 ip->ip_len = htons(tlen); 4292 ip->ip_ttl = ip_defttl; 4293 /* XXX tos? */ 4294 break; 4295 #endif 4296 #ifdef INET6 4297 case AF_INET6: 4298 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4299 ip6->ip6_vfc |= IPV6_VERSION; 4300 ip6->ip6_plen = htons(tlen - hlen); 4301 /* ip6_hlim will be initialized afterwards */ 4302 /* XXX flowlabel? */ 4303 break; 4304 #endif 4305 } 4306 4307 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4308 tp = sc->sc_tp; 4309 4310 switch (sc->sc_src.sa.sa_family) { 4311 #ifdef INET 4312 case AF_INET: 4313 error = ip_output(m, sc->sc_ipopts, ro, 4314 (ip_mtudisc ? IP_MTUDISC : 0), 4315 (struct ip_moptions *)NULL, so); 4316 break; 4317 #endif 4318 #ifdef INET6 4319 case AF_INET6: 4320 ip6->ip6_hlim = in6_selecthlim(NULL, 4321 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 4322 4323 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0, 4324 (struct ip6_moptions *)0, so, NULL); 4325 break; 4326 #endif 4327 default: 4328 error = EAFNOSUPPORT; 4329 break; 4330 } 4331 return (error); 4332 } 4333
Indexes created Mon Sep 22 13:09:51 GMT 2025