1 /* $NetBSD: tcp_input.c,v 1.311 2011/04/25 22:12:43 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 * 93 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 94 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 95 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 96 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 97 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 98 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 99 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 100 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 101 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 102 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 103 * POSSIBILITY OF SUCH DAMAGE. 104 */ 105 106 /* 107 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 108 * The Regents of the University of California. All rights reserved. 109 * 110 * Redistribution and use in source and binary forms, with or without 111 * modification, are permitted provided that the following conditions 112 * are met: 113 * 1. Redistributions of source code must retain the above copyright 114 * notice, this list of conditions and the following disclaimer. 115 * 2. Redistributions in binary form must reproduce the above copyright 116 * notice, this list of conditions and the following disclaimer in the 117 * documentation and/or other materials provided with the distribution. 118 * 3. Neither the name of the University nor the names of its contributors 119 * may be used to endorse or promote products derived from this software 120 * without specific prior written permission. 121 * 122 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 123 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 124 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 125 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 126 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 127 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 128 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 129 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 130 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 131 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 132 * SUCH DAMAGE. 133 * 134 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 135 */ 136 137 /* 138 * TODO list for SYN cache stuff: 139 * 140 * Find room for a "state" field, which is needed to keep a 141 * compressed state for TIME_WAIT TCBs. It's been noted already 142 * that this is fairly important for very high-volume web and 143 * mail servers, which use a large number of short-lived 144 * connections. 145 */ 146 147 #include <sys/cdefs.h> 148 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.311 2011/04/25 22:12:43 yamt Exp $"); 149 150 #include "opt_inet.h" 151 #include "opt_ipsec.h" 152 #include "opt_inet_csum.h" 153 #include "opt_tcp_debug.h" 154 155 #include <sys/param.h> 156 #include <sys/systm.h> 157 #include <sys/malloc.h> 158 #include <sys/mbuf.h> 159 #include <sys/protosw.h> 160 #include <sys/socket.h> 161 #include <sys/socketvar.h> 162 #include <sys/errno.h> 163 #include <sys/syslog.h> 164 #include <sys/pool.h> 165 #include <sys/domain.h> 166 #include <sys/kernel.h> 167 #ifdef TCP_SIGNATURE 168 #include <sys/md5.h> 169 #endif 170 #include <sys/lwp.h> /* for lwp0 */ 171 172 #include <net/if.h> 173 #include <net/route.h> 174 #include <net/if_types.h> 175 176 #include <netinet/in.h> 177 #include <netinet/in_systm.h> 178 #include <netinet/ip.h> 179 #include <netinet/in_pcb.h> 180 #include <netinet/in_var.h> 181 #include <netinet/ip_var.h> 182 #include <netinet/in_offload.h> 183 184 #ifdef INET6 185 #ifndef INET 186 #include <netinet/in.h> 187 #endif 188 #include <netinet/ip6.h> 189 #include <netinet6/ip6_var.h> 190 #include <netinet6/in6_pcb.h> 191 #include <netinet6/ip6_var.h> 192 #include <netinet6/in6_var.h> 193 #include <netinet/icmp6.h> 194 #include <netinet6/nd6.h> 195 #ifdef TCP_SIGNATURE 196 #include <netinet6/scope6_var.h> 197 #endif 198 #endif 199 200 #ifndef INET6 201 /* always need ip6.h for IP6_EXTHDR_GET */ 202 #include <netinet/ip6.h> 203 #endif 204 205 #include <netinet/tcp.h> 206 #include <netinet/tcp_fsm.h> 207 #include <netinet/tcp_seq.h> 208 #include <netinet/tcp_timer.h> 209 #include <netinet/tcp_var.h> 210 #include <netinet/tcp_private.h> 211 #include <netinet/tcpip.h> 212 #include <netinet/tcp_congctl.h> 213 #include <netinet/tcp_debug.h> 214 215 #include <machine/stdarg.h> 216 217 #ifdef IPSEC 218 #include <netinet6/ipsec.h> 219 #include <netinet6/ipsec_private.h> 220 #include <netkey/key.h> 221 #endif /*IPSEC*/ 222 #ifdef INET6 223 #include "faith.h" 224 #if defined(NFAITH) && NFAITH > 0 225 #include <net/if_faith.h> 226 #endif 227 #endif /* IPSEC */ 228 229 #ifdef FAST_IPSEC 230 #include <netipsec/ipsec.h> 231 #include <netipsec/ipsec_var.h> 232 #include <netipsec/ipsec_private.h> 233 #include <netipsec/key.h> 234 #ifdef INET6 235 #include <netipsec/ipsec6.h> 236 #endif 237 #endif /* FAST_IPSEC*/ 238 239 int tcprexmtthresh = 3; 240 int tcp_log_refused; 241 242 int tcp_do_autorcvbuf = 1; 243 int tcp_autorcvbuf_inc = 16 * 1024; 244 int tcp_autorcvbuf_max = 256 * 1024; 245 int tcp_msl = (TCPTV_MSL / PR_SLOWHZ); 246 247 static int tcp_rst_ppslim_count = 0; 248 static struct timeval tcp_rst_ppslim_last; 249 static int tcp_ackdrop_ppslim_count = 0; 250 static struct timeval tcp_ackdrop_ppslim_last; 251 252 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 253 254 /* for modulo comparisons of timestamps */ 255 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 256 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 257 258 /* 259 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 260 */ 261 #ifdef INET6 262 static inline void 263 nd6_hint(struct tcpcb *tp) 264 { 265 struct rtentry *rt; 266 267 if (tp != NULL && tp->t_in6pcb != NULL && tp->t_family == AF_INET6 && 268 (rt = rtcache_validate(&tp->t_in6pcb->in6p_route)) != NULL) 269 nd6_nud_hint(rt, NULL, 0); 270 } 271 #else 272 static inline void 273 nd6_hint(struct tcpcb *tp) 274 { 275 } 276 #endif 277 278 /* 279 * Compute ACK transmission behavior. Delay the ACK unless 280 * we have already delayed an ACK (must send an ACK every two segments). 281 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 282 * option is enabled. 283 */ 284 static void 285 tcp_setup_ack(struct tcpcb *tp, const struct tcphdr *th) 286 { 287 288 if (tp->t_flags & TF_DELACK || 289 (tcp_ack_on_push && th->th_flags & TH_PUSH)) 290 tp->t_flags |= TF_ACKNOW; 291 else 292 TCP_SET_DELACK(tp); 293 } 294 295 static void 296 icmp_check(struct tcpcb *tp, const struct tcphdr *th, int acked) 297 { 298 299 /* 300 * If we had a pending ICMP message that refers to data that have 301 * just been acknowledged, disregard the recorded ICMP message. 302 */ 303 if ((tp->t_flags & TF_PMTUD_PEND) && 304 SEQ_GT(th->th_ack, tp->t_pmtud_th_seq)) 305 tp->t_flags &= ~TF_PMTUD_PEND; 306 307 /* 308 * Keep track of the largest chunk of data 309 * acknowledged since last PMTU update 310 */ 311 if (tp->t_pmtud_mss_acked < acked) 312 tp->t_pmtud_mss_acked = acked; 313 } 314 315 /* 316 * Convert TCP protocol fields to host order for easier processing. 317 */ 318 static void 319 tcp_fields_to_host(struct tcphdr *th) 320 { 321 322 NTOHL(th->th_seq); 323 NTOHL(th->th_ack); 324 NTOHS(th->th_win); 325 NTOHS(th->th_urp); 326 } 327 328 /* 329 * ... and reverse the above. 330 */ 331 static void 332 tcp_fields_to_net(struct tcphdr *th) 333 { 334 335 HTONL(th->th_seq); 336 HTONL(th->th_ack); 337 HTONS(th->th_win); 338 HTONS(th->th_urp); 339 } 340 341 #ifdef TCP_CSUM_COUNTERS 342 #include <sys/device.h> 343 344 #if defined(INET) 345 extern struct evcnt tcp_hwcsum_ok; 346 extern struct evcnt tcp_hwcsum_bad; 347 extern struct evcnt tcp_hwcsum_data; 348 extern struct evcnt tcp_swcsum; 349 #endif /* defined(INET) */ 350 #if defined(INET6) 351 extern struct evcnt tcp6_hwcsum_ok; 352 extern struct evcnt tcp6_hwcsum_bad; 353 extern struct evcnt tcp6_hwcsum_data; 354 extern struct evcnt tcp6_swcsum; 355 #endif /* defined(INET6) */ 356 357 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 358 359 #else 360 361 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 362 363 #endif /* TCP_CSUM_COUNTERS */ 364 365 #ifdef TCP_REASS_COUNTERS 366 #include <sys/device.h> 367 368 extern struct evcnt tcp_reass_; 369 extern struct evcnt tcp_reass_empty; 370 extern struct evcnt tcp_reass_iteration[8]; 371 extern struct evcnt tcp_reass_prependfirst; 372 extern struct evcnt tcp_reass_prepend; 373 extern struct evcnt tcp_reass_insert; 374 extern struct evcnt tcp_reass_inserttail; 375 extern struct evcnt tcp_reass_append; 376 extern struct evcnt tcp_reass_appendtail; 377 extern struct evcnt tcp_reass_overlaptail; 378 extern struct evcnt tcp_reass_overlapfront; 379 extern struct evcnt tcp_reass_segdup; 380 extern struct evcnt tcp_reass_fragdup; 381 382 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 383 384 #else 385 386 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 387 388 #endif /* TCP_REASS_COUNTERS */ 389 390 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *, 391 int *); 392 static int tcp_dooptions(struct tcpcb *, const u_char *, int, 393 struct tcphdr *, struct mbuf *, int, struct tcp_opt_info *); 394 395 #ifdef INET 396 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 397 #endif 398 #ifdef INET6 399 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 400 #endif 401 402 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 403 404 #if defined(MBUFTRACE) 405 struct mowner tcp_reass_mowner = MOWNER_INIT("tcp", "reass"); 406 #endif /* defined(MBUFTRACE) */ 407 408 static struct pool tcpipqent_pool; 409 410 void 411 tcpipqent_init(void) 412 { 413 414 pool_init(&tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", 415 NULL, IPL_VM); 416 } 417 418 struct ipqent * 419 tcpipqent_alloc(void) 420 { 421 struct ipqent *ipqe; 422 int s; 423 424 s = splvm(); 425 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 426 splx(s); 427 428 return ipqe; 429 } 430 431 void 432 tcpipqent_free(struct ipqent *ipqe) 433 { 434 int s; 435 436 s = splvm(); 437 pool_put(&tcpipqent_pool, ipqe); 438 splx(s); 439 } 440 441 static int 442 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int *tlen) 443 { 444 struct ipqent *p, *q, *nq, *tiqe = NULL; 445 struct socket *so = NULL; 446 int pkt_flags; 447 tcp_seq pkt_seq; 448 unsigned pkt_len; 449 u_long rcvpartdupbyte = 0; 450 u_long rcvoobyte; 451 #ifdef TCP_REASS_COUNTERS 452 u_int count = 0; 453 #endif 454 uint64_t *tcps; 455 456 if (tp->t_inpcb) 457 so = tp->t_inpcb->inp_socket; 458 #ifdef INET6 459 else if (tp->t_in6pcb) 460 so = tp->t_in6pcb->in6p_socket; 461 #endif 462 463 TCP_REASS_LOCK_CHECK(tp); 464 465 /* 466 * Call with th==0 after become established to 467 * force pre-ESTABLISHED data up to user socket. 468 */ 469 if (th == 0) 470 goto present; 471 472 m_claimm(m, &tcp_reass_mowner); 473 474 rcvoobyte = *tlen; 475 /* 476 * Copy these to local variables because the tcpiphdr 477 * gets munged while we are collapsing mbufs. 478 */ 479 pkt_seq = th->th_seq; 480 pkt_len = *tlen; 481 pkt_flags = th->th_flags; 482 483 TCP_REASS_COUNTER_INCR(&tcp_reass_); 484 485 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 486 /* 487 * When we miss a packet, the vast majority of time we get 488 * packets that follow it in order. So optimize for that. 489 */ 490 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 491 p->ipqe_len += pkt_len; 492 p->ipqe_flags |= pkt_flags; 493 m_cat(p->ipre_mlast, m); 494 TRAVERSE(p->ipre_mlast); 495 m = NULL; 496 tiqe = p; 497 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 498 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 499 goto skip_replacement; 500 } 501 /* 502 * While we're here, if the pkt is completely beyond 503 * anything we have, just insert it at the tail. 504 */ 505 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 506 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 507 goto insert_it; 508 } 509 } 510 511 q = TAILQ_FIRST(&tp->segq); 512 513 if (q != NULL) { 514 /* 515 * If this segment immediately precedes the first out-of-order 516 * block, simply slap the segment in front of it and (mostly) 517 * skip the complicated logic. 518 */ 519 if (pkt_seq + pkt_len == q->ipqe_seq) { 520 q->ipqe_seq = pkt_seq; 521 q->ipqe_len += pkt_len; 522 q->ipqe_flags |= pkt_flags; 523 m_cat(m, q->ipqe_m); 524 q->ipqe_m = m; 525 q->ipre_mlast = m; /* last mbuf may have changed */ 526 TRAVERSE(q->ipre_mlast); 527 tiqe = q; 528 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 529 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 530 goto skip_replacement; 531 } 532 } else { 533 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 534 } 535 536 /* 537 * Find a segment which begins after this one does. 538 */ 539 for (p = NULL; q != NULL; q = nq) { 540 nq = TAILQ_NEXT(q, ipqe_q); 541 #ifdef TCP_REASS_COUNTERS 542 count++; 543 #endif 544 /* 545 * If the received segment is just right after this 546 * fragment, merge the two together and then check 547 * for further overlaps. 548 */ 549 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 550 #ifdef TCPREASS_DEBUG 551 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 552 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 553 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 554 #endif 555 pkt_len += q->ipqe_len; 556 pkt_flags |= q->ipqe_flags; 557 pkt_seq = q->ipqe_seq; 558 m_cat(q->ipre_mlast, m); 559 TRAVERSE(q->ipre_mlast); 560 m = q->ipqe_m; 561 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 562 goto free_ipqe; 563 } 564 /* 565 * If the received segment is completely past this 566 * fragment, we need to go the next fragment. 567 */ 568 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 569 p = q; 570 continue; 571 } 572 /* 573 * If the fragment is past the received segment, 574 * it (or any following) can't be concatenated. 575 */ 576 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 577 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 578 break; 579 } 580 581 /* 582 * We've received all the data in this segment before. 583 * mark it as a duplicate and return. 584 */ 585 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 586 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 587 tcps = TCP_STAT_GETREF(); 588 tcps[TCP_STAT_RCVDUPPACK]++; 589 tcps[TCP_STAT_RCVDUPBYTE] += pkt_len; 590 TCP_STAT_PUTREF(); 591 tcp_new_dsack(tp, pkt_seq, pkt_len); 592 m_freem(m); 593 if (tiqe != NULL) { 594 tcpipqent_free(tiqe); 595 } 596 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 597 return (0); 598 } 599 /* 600 * Received segment completely overlaps this fragment 601 * so we drop the fragment (this keeps the temporal 602 * ordering of segments correct). 603 */ 604 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 605 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 606 rcvpartdupbyte += q->ipqe_len; 607 m_freem(q->ipqe_m); 608 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 609 goto free_ipqe; 610 } 611 /* 612 * RX'ed segment extends past the end of the 613 * fragment. Drop the overlapping bytes. Then 614 * merge the fragment and segment then treat as 615 * a longer received packet. 616 */ 617 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 618 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 619 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 620 #ifdef TCPREASS_DEBUG 621 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 622 tp, overlap, 623 pkt_seq, pkt_seq + pkt_len, pkt_len); 624 #endif 625 m_adj(m, overlap); 626 rcvpartdupbyte += overlap; 627 m_cat(q->ipre_mlast, m); 628 TRAVERSE(q->ipre_mlast); 629 m = q->ipqe_m; 630 pkt_seq = q->ipqe_seq; 631 pkt_len += q->ipqe_len - overlap; 632 rcvoobyte -= overlap; 633 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 634 goto free_ipqe; 635 } 636 /* 637 * RX'ed segment extends past the front of the 638 * fragment. Drop the overlapping bytes on the 639 * received packet. The packet will then be 640 * contatentated with this fragment a bit later. 641 */ 642 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 643 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 644 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 645 #ifdef TCPREASS_DEBUG 646 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 647 tp, overlap, 648 pkt_seq, pkt_seq + pkt_len, pkt_len); 649 #endif 650 m_adj(m, -overlap); 651 pkt_len -= overlap; 652 rcvpartdupbyte += overlap; 653 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 654 rcvoobyte -= overlap; 655 } 656 /* 657 * If the received segment immediates precedes this 658 * fragment then tack the fragment onto this segment 659 * and reinsert the data. 660 */ 661 if (q->ipqe_seq == pkt_seq + pkt_len) { 662 #ifdef TCPREASS_DEBUG 663 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 664 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 665 pkt_seq, pkt_seq + pkt_len, pkt_len); 666 #endif 667 pkt_len += q->ipqe_len; 668 pkt_flags |= q->ipqe_flags; 669 m_cat(m, q->ipqe_m); 670 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 671 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 672 tp->t_segqlen--; 673 KASSERT(tp->t_segqlen >= 0); 674 KASSERT(tp->t_segqlen != 0 || 675 (TAILQ_EMPTY(&tp->segq) && 676 TAILQ_EMPTY(&tp->timeq))); 677 if (tiqe == NULL) { 678 tiqe = q; 679 } else { 680 tcpipqent_free(q); 681 } 682 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 683 break; 684 } 685 /* 686 * If the fragment is before the segment, remember it. 687 * When this loop is terminated, p will contain the 688 * pointer to fragment that is right before the received 689 * segment. 690 */ 691 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 692 p = q; 693 694 continue; 695 696 /* 697 * This is a common operation. It also will allow 698 * to save doing a malloc/free in most instances. 699 */ 700 free_ipqe: 701 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 702 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 703 tp->t_segqlen--; 704 KASSERT(tp->t_segqlen >= 0); 705 KASSERT(tp->t_segqlen != 0 || 706 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 707 if (tiqe == NULL) { 708 tiqe = q; 709 } else { 710 tcpipqent_free(q); 711 } 712 } 713 714 #ifdef TCP_REASS_COUNTERS 715 if (count > 7) 716 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 717 else if (count > 0) 718 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 719 #endif 720 721 insert_it: 722 723 /* 724 * Allocate a new queue entry since the received segment did not 725 * collapse onto any other out-of-order block; thus we are allocating 726 * a new block. If it had collapsed, tiqe would not be NULL and 727 * we would be reusing it. 728 * XXX If we can't, just drop the packet. XXX 729 */ 730 if (tiqe == NULL) { 731 tiqe = tcpipqent_alloc(); 732 if (tiqe == NULL) { 733 TCP_STATINC(TCP_STAT_RCVMEMDROP); 734 m_freem(m); 735 return (0); 736 } 737 } 738 739 /* 740 * Update the counters. 741 */ 742 tcps = TCP_STAT_GETREF(); 743 tcps[TCP_STAT_RCVOOPACK]++; 744 tcps[TCP_STAT_RCVOOBYTE] += rcvoobyte; 745 if (rcvpartdupbyte) { 746 tcps[TCP_STAT_RCVPARTDUPPACK]++; 747 tcps[TCP_STAT_RCVPARTDUPBYTE] += rcvpartdupbyte; 748 } 749 TCP_STAT_PUTREF(); 750 751 /* 752 * Insert the new fragment queue entry into both queues. 753 */ 754 tiqe->ipqe_m = m; 755 tiqe->ipre_mlast = m; 756 tiqe->ipqe_seq = pkt_seq; 757 tiqe->ipqe_len = pkt_len; 758 tiqe->ipqe_flags = pkt_flags; 759 if (p == NULL) { 760 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 761 #ifdef TCPREASS_DEBUG 762 if (tiqe->ipqe_seq != tp->rcv_nxt) 763 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 764 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 765 #endif 766 } else { 767 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 768 #ifdef TCPREASS_DEBUG 769 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 770 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 771 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 772 #endif 773 } 774 tp->t_segqlen++; 775 776 skip_replacement: 777 778 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 779 780 present: 781 /* 782 * Present data to user, advancing rcv_nxt through 783 * completed sequence space. 784 */ 785 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 786 return (0); 787 q = TAILQ_FIRST(&tp->segq); 788 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 789 return (0); 790 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 791 return (0); 792 793 tp->rcv_nxt += q->ipqe_len; 794 pkt_flags = q->ipqe_flags & TH_FIN; 795 nd6_hint(tp); 796 797 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 798 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 799 tp->t_segqlen--; 800 KASSERT(tp->t_segqlen >= 0); 801 KASSERT(tp->t_segqlen != 0 || 802 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 803 if (so->so_state & SS_CANTRCVMORE) 804 m_freem(q->ipqe_m); 805 else 806 sbappendstream(&so->so_rcv, q->ipqe_m); 807 tcpipqent_free(q); 808 sorwakeup(so); 809 return (pkt_flags); 810 } 811 812 #ifdef INET6 813 int 814 tcp6_input(struct mbuf **mp, int *offp, int proto) 815 { 816 struct mbuf *m = *mp; 817 818 /* 819 * draft-itojun-ipv6-tcp-to-anycast 820 * better place to put this in? 821 */ 822 if (m->m_flags & M_ANYCAST6) { 823 struct ip6_hdr *ip6; 824 if (m->m_len < sizeof(struct ip6_hdr)) { 825 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 826 TCP_STATINC(TCP_STAT_RCVSHORT); 827 return IPPROTO_DONE; 828 } 829 } 830 ip6 = mtod(m, struct ip6_hdr *); 831 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 832 (char *)&ip6->ip6_dst - (char *)ip6); 833 return IPPROTO_DONE; 834 } 835 836 tcp_input(m, *offp, proto); 837 return IPPROTO_DONE; 838 } 839 #endif 840 841 #ifdef INET 842 static void 843 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 844 { 845 char src[4*sizeof "123"]; 846 char dst[4*sizeof "123"]; 847 848 if (ip) { 849 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 850 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 851 } 852 else { 853 strlcpy(src, "(unknown)", sizeof(src)); 854 strlcpy(dst, "(unknown)", sizeof(dst)); 855 } 856 log(LOG_INFO, 857 "Connection attempt to TCP %s:%d from %s:%d\n", 858 dst, ntohs(th->th_dport), 859 src, ntohs(th->th_sport)); 860 } 861 #endif 862 863 #ifdef INET6 864 static void 865 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 866 { 867 char src[INET6_ADDRSTRLEN]; 868 char dst[INET6_ADDRSTRLEN]; 869 870 if (ip6) { 871 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 872 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 873 } 874 else { 875 strlcpy(src, "(unknown v6)", sizeof(src)); 876 strlcpy(dst, "(unknown v6)", sizeof(dst)); 877 } 878 log(LOG_INFO, 879 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 880 dst, ntohs(th->th_dport), 881 src, ntohs(th->th_sport)); 882 } 883 #endif 884 885 /* 886 * Checksum extended TCP header and data. 887 */ 888 int 889 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, 890 int toff, int off, int tlen) 891 { 892 893 /* 894 * XXX it's better to record and check if this mbuf is 895 * already checked. 896 */ 897 898 switch (af) { 899 #ifdef INET 900 case AF_INET: 901 switch (m->m_pkthdr.csum_flags & 902 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 903 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 904 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 905 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 906 goto badcsum; 907 908 case M_CSUM_TCPv4|M_CSUM_DATA: { 909 u_int32_t hw_csum = m->m_pkthdr.csum_data; 910 911 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 912 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 913 const struct ip *ip = 914 mtod(m, const struct ip *); 915 916 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 917 ip->ip_dst.s_addr, 918 htons(hw_csum + tlen + off + IPPROTO_TCP)); 919 } 920 if ((hw_csum ^ 0xffff) != 0) 921 goto badcsum; 922 break; 923 } 924 925 case M_CSUM_TCPv4: 926 /* Checksum was okay. */ 927 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 928 break; 929 930 default: 931 /* 932 * Must compute it ourselves. Maybe skip checksum 933 * on loopback interfaces. 934 */ 935 if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & 936 IFF_LOOPBACK) || 937 tcp_do_loopback_cksum)) { 938 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 939 if (in4_cksum(m, IPPROTO_TCP, toff, 940 tlen + off) != 0) 941 goto badcsum; 942 } 943 break; 944 } 945 break; 946 #endif /* INET4 */ 947 948 #ifdef INET6 949 case AF_INET6: 950 switch (m->m_pkthdr.csum_flags & 951 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 952 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 953 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 954 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 955 goto badcsum; 956 957 #if 0 /* notyet */ 958 case M_CSUM_TCPv6|M_CSUM_DATA: 959 #endif 960 961 case M_CSUM_TCPv6: 962 /* Checksum was okay. */ 963 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 964 break; 965 966 default: 967 /* 968 * Must compute it ourselves. Maybe skip checksum 969 * on loopback interfaces. 970 */ 971 if (__predict_true((m->m_flags & M_LOOP) == 0 || 972 tcp_do_loopback_cksum)) { 973 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 974 if (in6_cksum(m, IPPROTO_TCP, toff, 975 tlen + off) != 0) 976 goto badcsum; 977 } 978 } 979 break; 980 #endif /* INET6 */ 981 } 982 983 return 0; 984 985 badcsum: 986 TCP_STATINC(TCP_STAT_RCVBADSUM); 987 return -1; 988 } 989 990 /* 991 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 992 */ 993 void 994 tcp_input(struct mbuf *m, ...) 995 { 996 struct tcphdr *th; 997 struct ip *ip; 998 struct inpcb *inp; 999 #ifdef INET6 1000 struct ip6_hdr *ip6; 1001 struct in6pcb *in6p; 1002 #endif 1003 u_int8_t *optp = NULL; 1004 int optlen = 0; 1005 int len, tlen, toff, hdroptlen = 0; 1006 struct tcpcb *tp = 0; 1007 int tiflags; 1008 struct socket *so = NULL; 1009 int todrop, acked, ourfinisacked, needoutput = 0; 1010 bool dupseg; 1011 #ifdef TCP_DEBUG 1012 short ostate = 0; 1013 #endif 1014 u_long tiwin; 1015 struct tcp_opt_info opti; 1016 int off, iphlen; 1017 va_list ap; 1018 int af; /* af on the wire */ 1019 struct mbuf *tcp_saveti = NULL; 1020 uint32_t ts_rtt; 1021 uint8_t iptos; 1022 uint64_t *tcps; 1023 1024 MCLAIM(m, &tcp_rx_mowner); 1025 va_start(ap, m); 1026 toff = va_arg(ap, int); 1027 (void)va_arg(ap, int); /* ignore value, advance ap */ 1028 va_end(ap); 1029 1030 TCP_STATINC(TCP_STAT_RCVTOTAL); 1031 1032 memset(&opti, 0, sizeof(opti)); 1033 opti.ts_present = 0; 1034 opti.maxseg = 0; 1035 1036 /* 1037 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 1038 * 1039 * TCP is, by definition, unicast, so we reject all 1040 * multicast outright. 1041 * 1042 * Note, there are additional src/dst address checks in 1043 * the AF-specific code below. 1044 */ 1045 if (m->m_flags & (M_BCAST|M_MCAST)) { 1046 /* XXX stat */ 1047 goto drop; 1048 } 1049 #ifdef INET6 1050 if (m->m_flags & M_ANYCAST6) { 1051 /* XXX stat */ 1052 goto drop; 1053 } 1054 #endif 1055 1056 /* 1057 * Get IP and TCP header. 1058 * Note: IP leaves IP header in first mbuf. 1059 */ 1060 ip = mtod(m, struct ip *); 1061 switch (ip->ip_v) { 1062 #ifdef INET 1063 case 4: 1064 #ifdef INET6 1065 ip6 = NULL; 1066 #endif 1067 af = AF_INET; 1068 iphlen = sizeof(struct ip); 1069 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1070 sizeof(struct tcphdr)); 1071 if (th == NULL) { 1072 TCP_STATINC(TCP_STAT_RCVSHORT); 1073 return; 1074 } 1075 /* We do the checksum after PCB lookup... */ 1076 len = ntohs(ip->ip_len); 1077 tlen = len - toff; 1078 iptos = ip->ip_tos; 1079 break; 1080 #endif 1081 #ifdef INET6 1082 case 6: 1083 ip = NULL; 1084 iphlen = sizeof(struct ip6_hdr); 1085 af = AF_INET6; 1086 ip6 = mtod(m, struct ip6_hdr *); 1087 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1088 sizeof(struct tcphdr)); 1089 if (th == NULL) { 1090 TCP_STATINC(TCP_STAT_RCVSHORT); 1091 return; 1092 } 1093 1094 /* Be proactive about malicious use of IPv4 mapped address */ 1095 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 1096 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 1097 /* XXX stat */ 1098 goto drop; 1099 } 1100 1101 /* 1102 * Be proactive about unspecified IPv6 address in source. 1103 * As we use all-zero to indicate unbounded/unconnected pcb, 1104 * unspecified IPv6 address can be used to confuse us. 1105 * 1106 * Note that packets with unspecified IPv6 destination is 1107 * already dropped in ip6_input. 1108 */ 1109 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1110 /* XXX stat */ 1111 goto drop; 1112 } 1113 1114 /* 1115 * Make sure destination address is not multicast. 1116 * Source address checked in ip6_input(). 1117 */ 1118 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1119 /* XXX stat */ 1120 goto drop; 1121 } 1122 1123 /* We do the checksum after PCB lookup... */ 1124 len = m->m_pkthdr.len; 1125 tlen = len - toff; 1126 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1127 break; 1128 #endif 1129 default: 1130 m_freem(m); 1131 return; 1132 } 1133 1134 KASSERT(TCP_HDR_ALIGNED_P(th)); 1135 1136 /* 1137 * Check that TCP offset makes sense, 1138 * pull out TCP options and adjust length. XXX 1139 */ 1140 off = th->th_off << 2; 1141 if (off < sizeof (struct tcphdr) || off > tlen) { 1142 TCP_STATINC(TCP_STAT_RCVBADOFF); 1143 goto drop; 1144 } 1145 tlen -= off; 1146 1147 /* 1148 * tcp_input() has been modified to use tlen to mean the TCP data 1149 * length throughout the function. Other functions can use 1150 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1151 * rja 1152 */ 1153 1154 if (off > sizeof (struct tcphdr)) { 1155 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1156 if (th == NULL) { 1157 TCP_STATINC(TCP_STAT_RCVSHORT); 1158 return; 1159 } 1160 /* 1161 * NOTE: ip/ip6 will not be affected by m_pulldown() 1162 * (as they're before toff) and we don't need to update those. 1163 */ 1164 KASSERT(TCP_HDR_ALIGNED_P(th)); 1165 optlen = off - sizeof (struct tcphdr); 1166 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1167 /* 1168 * Do quick retrieval of timestamp options ("options 1169 * prediction?"). If timestamp is the only option and it's 1170 * formatted as recommended in RFC 1323 appendix A, we 1171 * quickly get the values now and not bother calling 1172 * tcp_dooptions(), etc. 1173 */ 1174 if ((optlen == TCPOLEN_TSTAMP_APPA || 1175 (optlen > TCPOLEN_TSTAMP_APPA && 1176 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1177 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1178 (th->th_flags & TH_SYN) == 0) { 1179 opti.ts_present = 1; 1180 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1181 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1182 optp = NULL; /* we've parsed the options */ 1183 } 1184 } 1185 tiflags = th->th_flags; 1186 1187 /* 1188 * Locate pcb for segment. 1189 */ 1190 findpcb: 1191 inp = NULL; 1192 #ifdef INET6 1193 in6p = NULL; 1194 #endif 1195 switch (af) { 1196 #ifdef INET 1197 case AF_INET: 1198 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1199 ip->ip_dst, th->th_dport); 1200 if (inp == 0) { 1201 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1202 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1203 } 1204 #ifdef INET6 1205 if (inp == 0) { 1206 struct in6_addr s, d; 1207 1208 /* mapped addr case */ 1209 memset(&s, 0, sizeof(s)); 1210 s.s6_addr16[5] = htons(0xffff); 1211 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1212 memset(&d, 0, sizeof(d)); 1213 d.s6_addr16[5] = htons(0xffff); 1214 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1215 in6p = in6_pcblookup_connect(&tcbtable, &s, 1216 th->th_sport, &d, th->th_dport, 0); 1217 if (in6p == 0) { 1218 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1219 in6p = in6_pcblookup_bind(&tcbtable, &d, 1220 th->th_dport, 0); 1221 } 1222 } 1223 #endif 1224 #ifndef INET6 1225 if (inp == 0) 1226 #else 1227 if (inp == 0 && in6p == 0) 1228 #endif 1229 { 1230 TCP_STATINC(TCP_STAT_NOPORT); 1231 if (tcp_log_refused && 1232 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1233 tcp4_log_refused(ip, th); 1234 } 1235 tcp_fields_to_host(th); 1236 goto dropwithreset_ratelim; 1237 } 1238 #if defined(IPSEC) || defined(FAST_IPSEC) 1239 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1240 ipsec4_in_reject(m, inp)) { 1241 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1242 goto drop; 1243 } 1244 #ifdef INET6 1245 else if (in6p && 1246 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1247 ipsec6_in_reject_so(m, in6p->in6p_socket)) { 1248 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1249 goto drop; 1250 } 1251 #endif 1252 #endif /*IPSEC*/ 1253 break; 1254 #endif /*INET*/ 1255 #ifdef INET6 1256 case AF_INET6: 1257 { 1258 int faith; 1259 1260 #if defined(NFAITH) && NFAITH > 0 1261 faith = faithprefix(&ip6->ip6_dst); 1262 #else 1263 faith = 0; 1264 #endif 1265 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1266 th->th_sport, &ip6->ip6_dst, th->th_dport, faith); 1267 if (in6p == NULL) { 1268 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1269 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1270 th->th_dport, faith); 1271 } 1272 if (in6p == NULL) { 1273 TCP_STATINC(TCP_STAT_NOPORT); 1274 if (tcp_log_refused && 1275 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1276 tcp6_log_refused(ip6, th); 1277 } 1278 tcp_fields_to_host(th); 1279 goto dropwithreset_ratelim; 1280 } 1281 #if defined(IPSEC) || defined(FAST_IPSEC) 1282 if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1283 ipsec6_in_reject(m, in6p)) { 1284 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1285 goto drop; 1286 } 1287 #endif /*IPSEC*/ 1288 break; 1289 } 1290 #endif 1291 } 1292 1293 /* 1294 * If the state is CLOSED (i.e., TCB does not exist) then 1295 * all data in the incoming segment is discarded. 1296 * If the TCB exists but is in CLOSED state, it is embryonic, 1297 * but should either do a listen or a connect soon. 1298 */ 1299 tp = NULL; 1300 so = NULL; 1301 if (inp) { 1302 /* Check the minimum TTL for socket. */ 1303 if (ip->ip_ttl < inp->inp_ip_minttl) 1304 goto drop; 1305 1306 tp = intotcpcb(inp); 1307 so = inp->inp_socket; 1308 } 1309 #ifdef INET6 1310 else if (in6p) { 1311 tp = in6totcpcb(in6p); 1312 so = in6p->in6p_socket; 1313 } 1314 #endif 1315 if (tp == 0) { 1316 tcp_fields_to_host(th); 1317 goto dropwithreset_ratelim; 1318 } 1319 if (tp->t_state == TCPS_CLOSED) 1320 goto drop; 1321 1322 KASSERT(so->so_lock == softnet_lock); 1323 KASSERT(solocked(so)); 1324 1325 /* 1326 * Checksum extended TCP header and data. 1327 */ 1328 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1329 goto badcsum; 1330 1331 tcp_fields_to_host(th); 1332 1333 /* Unscale the window into a 32-bit value. */ 1334 if ((tiflags & TH_SYN) == 0) 1335 tiwin = th->th_win << tp->snd_scale; 1336 else 1337 tiwin = th->th_win; 1338 1339 #ifdef INET6 1340 /* save packet options if user wanted */ 1341 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1342 if (in6p->in6p_options) { 1343 m_freem(in6p->in6p_options); 1344 in6p->in6p_options = 0; 1345 } 1346 KASSERT(ip6 != NULL); 1347 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1348 } 1349 #endif 1350 1351 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1352 union syn_cache_sa src; 1353 union syn_cache_sa dst; 1354 1355 memset(&src, 0, sizeof(src)); 1356 memset(&dst, 0, sizeof(dst)); 1357 switch (af) { 1358 #ifdef INET 1359 case AF_INET: 1360 src.sin.sin_len = sizeof(struct sockaddr_in); 1361 src.sin.sin_family = AF_INET; 1362 src.sin.sin_addr = ip->ip_src; 1363 src.sin.sin_port = th->th_sport; 1364 1365 dst.sin.sin_len = sizeof(struct sockaddr_in); 1366 dst.sin.sin_family = AF_INET; 1367 dst.sin.sin_addr = ip->ip_dst; 1368 dst.sin.sin_port = th->th_dport; 1369 break; 1370 #endif 1371 #ifdef INET6 1372 case AF_INET6: 1373 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1374 src.sin6.sin6_family = AF_INET6; 1375 src.sin6.sin6_addr = ip6->ip6_src; 1376 src.sin6.sin6_port = th->th_sport; 1377 1378 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1379 dst.sin6.sin6_family = AF_INET6; 1380 dst.sin6.sin6_addr = ip6->ip6_dst; 1381 dst.sin6.sin6_port = th->th_dport; 1382 break; 1383 #endif /* INET6 */ 1384 default: 1385 goto badsyn; /*sanity*/ 1386 } 1387 1388 if (so->so_options & SO_DEBUG) { 1389 #ifdef TCP_DEBUG 1390 ostate = tp->t_state; 1391 #endif 1392 1393 tcp_saveti = NULL; 1394 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1395 goto nosave; 1396 1397 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1398 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1399 if (!tcp_saveti) 1400 goto nosave; 1401 } else { 1402 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1403 if (!tcp_saveti) 1404 goto nosave; 1405 MCLAIM(m, &tcp_mowner); 1406 tcp_saveti->m_len = iphlen; 1407 m_copydata(m, 0, iphlen, 1408 mtod(tcp_saveti, void *)); 1409 } 1410 1411 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1412 m_freem(tcp_saveti); 1413 tcp_saveti = NULL; 1414 } else { 1415 tcp_saveti->m_len += sizeof(struct tcphdr); 1416 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1417 sizeof(struct tcphdr)); 1418 } 1419 nosave:; 1420 } 1421 if (so->so_options & SO_ACCEPTCONN) { 1422 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1423 if (tiflags & TH_RST) { 1424 syn_cache_reset(&src.sa, &dst.sa, th); 1425 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1426 (TH_ACK|TH_SYN)) { 1427 /* 1428 * Received a SYN,ACK. This should 1429 * never happen while we are in 1430 * LISTEN. Send an RST. 1431 */ 1432 goto badsyn; 1433 } else if (tiflags & TH_ACK) { 1434 so = syn_cache_get(&src.sa, &dst.sa, 1435 th, toff, tlen, so, m); 1436 if (so == NULL) { 1437 /* 1438 * We don't have a SYN for 1439 * this ACK; send an RST. 1440 */ 1441 goto badsyn; 1442 } else if (so == 1443 (struct socket *)(-1)) { 1444 /* 1445 * We were unable to create 1446 * the connection. If the 1447 * 3-way handshake was 1448 * completed, and RST has 1449 * been sent to the peer. 1450 * Since the mbuf might be 1451 * in use for the reply, 1452 * do not free it. 1453 */ 1454 m = NULL; 1455 } else { 1456 /* 1457 * We have created a 1458 * full-blown connection. 1459 */ 1460 tp = NULL; 1461 inp = NULL; 1462 #ifdef INET6 1463 in6p = NULL; 1464 #endif 1465 switch (so->so_proto->pr_domain->dom_family) { 1466 #ifdef INET 1467 case AF_INET: 1468 inp = sotoinpcb(so); 1469 tp = intotcpcb(inp); 1470 break; 1471 #endif 1472 #ifdef INET6 1473 case AF_INET6: 1474 in6p = sotoin6pcb(so); 1475 tp = in6totcpcb(in6p); 1476 break; 1477 #endif 1478 } 1479 if (tp == NULL) 1480 goto badsyn; /*XXX*/ 1481 tiwin <<= tp->snd_scale; 1482 goto after_listen; 1483 } 1484 } else { 1485 /* 1486 * None of RST, SYN or ACK was set. 1487 * This is an invalid packet for a 1488 * TCB in LISTEN state. Send a RST. 1489 */ 1490 goto badsyn; 1491 } 1492 } else { 1493 /* 1494 * Received a SYN. 1495 * 1496 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1497 */ 1498 if (m->m_flags & (M_BCAST|M_MCAST)) 1499 goto drop; 1500 1501 switch (af) { 1502 #ifdef INET6 1503 case AF_INET6: 1504 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1505 goto drop; 1506 break; 1507 #endif /* INET6 */ 1508 case AF_INET: 1509 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1510 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1511 goto drop; 1512 break; 1513 } 1514 1515 #ifdef INET6 1516 /* 1517 * If deprecated address is forbidden, we do 1518 * not accept SYN to deprecated interface 1519 * address to prevent any new inbound 1520 * connection from getting established. 1521 * When we do not accept SYN, we send a TCP 1522 * RST, with deprecated source address (instead 1523 * of dropping it). We compromise it as it is 1524 * much better for peer to send a RST, and 1525 * RST will be the final packet for the 1526 * exchange. 1527 * 1528 * If we do not forbid deprecated addresses, we 1529 * accept the SYN packet. RFC2462 does not 1530 * suggest dropping SYN in this case. 1531 * If we decipher RFC2462 5.5.4, it says like 1532 * this: 1533 * 1. use of deprecated addr with existing 1534 * communication is okay - "SHOULD continue 1535 * to be used" 1536 * 2. use of it with new communication: 1537 * (2a) "SHOULD NOT be used if alternate 1538 * address with sufficient scope is 1539 * available" 1540 * (2b) nothing mentioned otherwise. 1541 * Here we fall into (2b) case as we have no 1542 * choice in our source address selection - we 1543 * must obey the peer. 1544 * 1545 * The wording in RFC2462 is confusing, and 1546 * there are multiple description text for 1547 * deprecated address handling - worse, they 1548 * are not exactly the same. I believe 5.5.4 1549 * is the best one, so we follow 5.5.4. 1550 */ 1551 if (af == AF_INET6 && !ip6_use_deprecated) { 1552 struct in6_ifaddr *ia6; 1553 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1554 &ip6->ip6_dst)) && 1555 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1556 tp = NULL; 1557 goto dropwithreset; 1558 } 1559 } 1560 #endif 1561 1562 #if defined(IPSEC) || defined(FAST_IPSEC) 1563 switch (af) { 1564 #ifdef INET 1565 case AF_INET: 1566 if (ipsec4_in_reject_so(m, so)) { 1567 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1568 tp = NULL; 1569 goto dropwithreset; 1570 } 1571 break; 1572 #endif 1573 #ifdef INET6 1574 case AF_INET6: 1575 if (ipsec6_in_reject_so(m, so)) { 1576 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1577 tp = NULL; 1578 goto dropwithreset; 1579 } 1580 break; 1581 #endif /*INET6*/ 1582 } 1583 #endif /*IPSEC*/ 1584 1585 /* 1586 * LISTEN socket received a SYN 1587 * from itself? This can't possibly 1588 * be valid; drop the packet. 1589 */ 1590 if (th->th_sport == th->th_dport) { 1591 int i; 1592 1593 switch (af) { 1594 #ifdef INET 1595 case AF_INET: 1596 i = in_hosteq(ip->ip_src, ip->ip_dst); 1597 break; 1598 #endif 1599 #ifdef INET6 1600 case AF_INET6: 1601 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1602 break; 1603 #endif 1604 default: 1605 i = 1; 1606 } 1607 if (i) { 1608 TCP_STATINC(TCP_STAT_BADSYN); 1609 goto drop; 1610 } 1611 } 1612 1613 /* 1614 * SYN looks ok; create compressed TCP 1615 * state for it. 1616 */ 1617 if (so->so_qlen <= so->so_qlimit && 1618 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1619 so, m, optp, optlen, &opti)) 1620 m = NULL; 1621 } 1622 goto drop; 1623 } 1624 } 1625 1626 after_listen: 1627 #ifdef DIAGNOSTIC 1628 /* 1629 * Should not happen now that all embryonic connections 1630 * are handled with compressed state. 1631 */ 1632 if (tp->t_state == TCPS_LISTEN) 1633 panic("tcp_input: TCPS_LISTEN"); 1634 #endif 1635 1636 /* 1637 * Segment received on connection. 1638 * Reset idle time and keep-alive timer. 1639 */ 1640 tp->t_rcvtime = tcp_now; 1641 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1642 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1643 1644 /* 1645 * Process options. 1646 */ 1647 #ifdef TCP_SIGNATURE 1648 if (optp || (tp->t_flags & TF_SIGNATURE)) 1649 #else 1650 if (optp) 1651 #endif 1652 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1653 goto drop; 1654 1655 if (TCP_SACK_ENABLED(tp)) { 1656 tcp_del_sackholes(tp, th); 1657 } 1658 1659 if (TCP_ECN_ALLOWED(tp)) { 1660 if (tiflags & TH_CWR) { 1661 tp->t_flags &= ~TF_ECN_SND_ECE; 1662 } 1663 switch (iptos & IPTOS_ECN_MASK) { 1664 case IPTOS_ECN_CE: 1665 tp->t_flags |= TF_ECN_SND_ECE; 1666 TCP_STATINC(TCP_STAT_ECN_CE); 1667 break; 1668 case IPTOS_ECN_ECT0: 1669 TCP_STATINC(TCP_STAT_ECN_ECT); 1670 break; 1671 case IPTOS_ECN_ECT1: 1672 /* XXX: ignore for now -- rpaulo */ 1673 break; 1674 } 1675 /* 1676 * Congestion experienced. 1677 * Ignore if we are already trying to recover. 1678 */ 1679 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1680 tp->t_congctl->cong_exp(tp); 1681 } 1682 1683 if (opti.ts_present && opti.ts_ecr) { 1684 /* 1685 * Calculate the RTT from the returned time stamp and the 1686 * connection's time base. If the time stamp is later than 1687 * the current time, or is extremely old, fall back to non-1323 1688 * RTT calculation. Since ts_rtt is unsigned, we can test both 1689 * at the same time. 1690 * 1691 * Note that ts_rtt is in units of slow ticks (500 1692 * ms). Since most earthbound RTTs are < 500 ms, 1693 * observed values will have large quantization noise. 1694 * Our smoothed RTT is then the fraction of observed 1695 * samples that are 1 tick instead of 0 (times 500 1696 * ms). 1697 * 1698 * ts_rtt is increased by 1 to denote a valid sample, 1699 * with 0 indicating an invalid measurement. This 1700 * extra 1 must be removed when ts_rtt is used, or 1701 * else an an erroneous extra 500 ms will result. 1702 */ 1703 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1704 if (ts_rtt > TCP_PAWS_IDLE) 1705 ts_rtt = 0; 1706 } else { 1707 ts_rtt = 0; 1708 } 1709 1710 /* 1711 * Header prediction: check for the two common cases 1712 * of a uni-directional data xfer. If the packet has 1713 * no control flags, is in-sequence, the window didn't 1714 * change and we're not retransmitting, it's a 1715 * candidate. If the length is zero and the ack moved 1716 * forward, we're the sender side of the xfer. Just 1717 * free the data acked & wake any higher level process 1718 * that was blocked waiting for space. If the length 1719 * is non-zero and the ack didn't move, we're the 1720 * receiver side. If we're getting packets in-order 1721 * (the reassembly queue is empty), add the data to 1722 * the socket buffer and note that we need a delayed ack. 1723 */ 1724 if (tp->t_state == TCPS_ESTABLISHED && 1725 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1726 == TH_ACK && 1727 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1728 th->th_seq == tp->rcv_nxt && 1729 tiwin && tiwin == tp->snd_wnd && 1730 tp->snd_nxt == tp->snd_max) { 1731 1732 /* 1733 * If last ACK falls within this segment's sequence numbers, 1734 * record the timestamp. 1735 * NOTE that the test is modified according to the latest 1736 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1737 * 1738 * note that we already know 1739 * TSTMP_GEQ(opti.ts_val, tp->ts_recent) 1740 */ 1741 if (opti.ts_present && 1742 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1743 tp->ts_recent_age = tcp_now; 1744 tp->ts_recent = opti.ts_val; 1745 } 1746 1747 if (tlen == 0) { 1748 /* Ack prediction. */ 1749 if (SEQ_GT(th->th_ack, tp->snd_una) && 1750 SEQ_LEQ(th->th_ack, tp->snd_max) && 1751 tp->snd_cwnd >= tp->snd_wnd && 1752 tp->t_partialacks < 0) { 1753 /* 1754 * this is a pure ack for outstanding data. 1755 */ 1756 if (ts_rtt) 1757 tcp_xmit_timer(tp, ts_rtt); 1758 else if (tp->t_rtttime && 1759 SEQ_GT(th->th_ack, tp->t_rtseq)) 1760 tcp_xmit_timer(tp, 1761 tcp_now - tp->t_rtttime); 1762 acked = th->th_ack - tp->snd_una; 1763 tcps = TCP_STAT_GETREF(); 1764 tcps[TCP_STAT_PREDACK]++; 1765 tcps[TCP_STAT_RCVACKPACK]++; 1766 tcps[TCP_STAT_RCVACKBYTE] += acked; 1767 TCP_STAT_PUTREF(); 1768 nd6_hint(tp); 1769 1770 if (acked > (tp->t_lastoff - tp->t_inoff)) 1771 tp->t_lastm = NULL; 1772 sbdrop(&so->so_snd, acked); 1773 tp->t_lastoff -= acked; 1774 1775 icmp_check(tp, th, acked); 1776 1777 tp->snd_una = th->th_ack; 1778 tp->snd_fack = tp->snd_una; 1779 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1780 tp->snd_high = tp->snd_una; 1781 m_freem(m); 1782 1783 /* 1784 * If all outstanding data are acked, stop 1785 * retransmit timer, otherwise restart timer 1786 * using current (possibly backed-off) value. 1787 * If process is waiting for space, 1788 * wakeup/selnotify/signal. If data 1789 * are ready to send, let tcp_output 1790 * decide between more output or persist. 1791 */ 1792 if (tp->snd_una == tp->snd_max) 1793 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1794 else if (TCP_TIMER_ISARMED(tp, 1795 TCPT_PERSIST) == 0) 1796 TCP_TIMER_ARM(tp, TCPT_REXMT, 1797 tp->t_rxtcur); 1798 1799 sowwakeup(so); 1800 if (so->so_snd.sb_cc) { 1801 KERNEL_LOCK(1, NULL); 1802 (void) tcp_output(tp); 1803 KERNEL_UNLOCK_ONE(NULL); 1804 } 1805 if (tcp_saveti) 1806 m_freem(tcp_saveti); 1807 return; 1808 } 1809 } else if (th->th_ack == tp->snd_una && 1810 TAILQ_FIRST(&tp->segq) == NULL && 1811 tlen <= sbspace(&so->so_rcv)) { 1812 int newsize = 0; /* automatic sockbuf scaling */ 1813 1814 /* 1815 * this is a pure, in-sequence data packet 1816 * with nothing on the reassembly queue and 1817 * we have enough buffer space to take it. 1818 */ 1819 tp->rcv_nxt += tlen; 1820 tcps = TCP_STAT_GETREF(); 1821 tcps[TCP_STAT_PREDDAT]++; 1822 tcps[TCP_STAT_RCVPACK]++; 1823 tcps[TCP_STAT_RCVBYTE] += tlen; 1824 TCP_STAT_PUTREF(); 1825 nd6_hint(tp); 1826 1827 /* 1828 * Automatic sizing enables the performance of large buffers 1829 * and most of the efficiency of small ones by only allocating 1830 * space when it is needed. 1831 * 1832 * On the receive side the socket buffer memory is only rarely 1833 * used to any significant extent. This allows us to be much 1834 * more aggressive in scaling the receive socket buffer. For 1835 * the case that the buffer space is actually used to a large 1836 * extent and we run out of kernel memory we can simply drop 1837 * the new segments; TCP on the sender will just retransmit it 1838 * later. Setting the buffer size too big may only consume too 1839 * much kernel memory if the application doesn't read() from 1840 * the socket or packet loss or reordering makes use of the 1841 * reassembly queue. 1842 * 1843 * The criteria to step up the receive buffer one notch are: 1844 * 1. the number of bytes received during the time it takes 1845 * one timestamp to be reflected back to us (the RTT); 1846 * 2. received bytes per RTT is within seven eighth of the 1847 * current socket buffer size; 1848 * 3. receive buffer size has not hit maximal automatic size; 1849 * 1850 * This algorithm does one step per RTT at most and only if 1851 * we receive a bulk stream w/o packet losses or reorderings. 1852 * Shrinking the buffer during idle times is not necessary as 1853 * it doesn't consume any memory when idle. 1854 * 1855 * TODO: Only step up if the application is actually serving 1856 * the buffer to better manage the socket buffer resources. 1857 */ 1858 if (tcp_do_autorcvbuf && 1859 opti.ts_ecr && 1860 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1861 if (opti.ts_ecr > tp->rfbuf_ts && 1862 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 1863 if (tp->rfbuf_cnt > 1864 (so->so_rcv.sb_hiwat / 8 * 7) && 1865 so->so_rcv.sb_hiwat < 1866 tcp_autorcvbuf_max) { 1867 newsize = 1868 min(so->so_rcv.sb_hiwat + 1869 tcp_autorcvbuf_inc, 1870 tcp_autorcvbuf_max); 1871 } 1872 /* Start over with next RTT. */ 1873 tp->rfbuf_ts = 0; 1874 tp->rfbuf_cnt = 0; 1875 } else 1876 tp->rfbuf_cnt += tlen; /* add up */ 1877 } 1878 1879 /* 1880 * Drop TCP, IP headers and TCP options then add data 1881 * to socket buffer. 1882 */ 1883 if (so->so_state & SS_CANTRCVMORE) 1884 m_freem(m); 1885 else { 1886 /* 1887 * Set new socket buffer size. 1888 * Give up when limit is reached. 1889 */ 1890 if (newsize) 1891 if (!sbreserve(&so->so_rcv, 1892 newsize, so)) 1893 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1894 m_adj(m, toff + off); 1895 sbappendstream(&so->so_rcv, m); 1896 } 1897 sorwakeup(so); 1898 tcp_setup_ack(tp, th); 1899 if (tp->t_flags & TF_ACKNOW) { 1900 KERNEL_LOCK(1, NULL); 1901 (void) tcp_output(tp); 1902 KERNEL_UNLOCK_ONE(NULL); 1903 } 1904 if (tcp_saveti) 1905 m_freem(tcp_saveti); 1906 return; 1907 } 1908 } 1909 1910 /* 1911 * Compute mbuf offset to TCP data segment. 1912 */ 1913 hdroptlen = toff + off; 1914 1915 /* 1916 * Calculate amount of space in receive window, 1917 * and then do TCP input processing. 1918 * Receive window is amount of space in rcv queue, 1919 * but not less than advertised window. 1920 */ 1921 { int win; 1922 1923 win = sbspace(&so->so_rcv); 1924 if (win < 0) 1925 win = 0; 1926 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1927 } 1928 1929 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1930 tp->rfbuf_ts = 0; 1931 tp->rfbuf_cnt = 0; 1932 1933 switch (tp->t_state) { 1934 /* 1935 * If the state is SYN_SENT: 1936 * if seg contains an ACK, but not for our SYN, drop the input. 1937 * if seg contains a RST, then drop the connection. 1938 * if seg does not contain SYN, then drop it. 1939 * Otherwise this is an acceptable SYN segment 1940 * initialize tp->rcv_nxt and tp->irs 1941 * if seg contains ack then advance tp->snd_una 1942 * if seg contains a ECE and ECN support is enabled, the stream 1943 * is ECN capable. 1944 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1945 * arrange for segment to be acked (eventually) 1946 * continue processing rest of data/controls, beginning with URG 1947 */ 1948 case TCPS_SYN_SENT: 1949 if ((tiflags & TH_ACK) && 1950 (SEQ_LEQ(th->th_ack, tp->iss) || 1951 SEQ_GT(th->th_ack, tp->snd_max))) 1952 goto dropwithreset; 1953 if (tiflags & TH_RST) { 1954 if (tiflags & TH_ACK) 1955 tp = tcp_drop(tp, ECONNREFUSED); 1956 goto drop; 1957 } 1958 if ((tiflags & TH_SYN) == 0) 1959 goto drop; 1960 if (tiflags & TH_ACK) { 1961 tp->snd_una = th->th_ack; 1962 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1963 tp->snd_nxt = tp->snd_una; 1964 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1965 tp->snd_high = tp->snd_una; 1966 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1967 1968 if ((tiflags & TH_ECE) && tcp_do_ecn) { 1969 tp->t_flags |= TF_ECN_PERMIT; 1970 TCP_STATINC(TCP_STAT_ECN_SHS); 1971 } 1972 1973 } 1974 tp->irs = th->th_seq; 1975 tcp_rcvseqinit(tp); 1976 tp->t_flags |= TF_ACKNOW; 1977 tcp_mss_from_peer(tp, opti.maxseg); 1978 1979 /* 1980 * Initialize the initial congestion window. If we 1981 * had to retransmit the SYN, we must initialize cwnd 1982 * to 1 segment (i.e. the Loss Window). 1983 */ 1984 if (tp->t_flags & TF_SYN_REXMT) 1985 tp->snd_cwnd = tp->t_peermss; 1986 else { 1987 int ss = tcp_init_win; 1988 #ifdef INET 1989 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1990 ss = tcp_init_win_local; 1991 #endif 1992 #ifdef INET6 1993 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1994 ss = tcp_init_win_local; 1995 #endif 1996 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1997 } 1998 1999 tcp_rmx_rtt(tp); 2000 if (tiflags & TH_ACK) { 2001 TCP_STATINC(TCP_STAT_CONNECTS); 2002 soisconnected(so); 2003 tcp_established(tp); 2004 /* Do window scaling on this connection? */ 2005 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2006 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2007 tp->snd_scale = tp->requested_s_scale; 2008 tp->rcv_scale = tp->request_r_scale; 2009 } 2010 TCP_REASS_LOCK(tp); 2011 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2012 TCP_REASS_UNLOCK(tp); 2013 /* 2014 * if we didn't have to retransmit the SYN, 2015 * use its rtt as our initial srtt & rtt var. 2016 */ 2017 if (tp->t_rtttime) 2018 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2019 } else 2020 tp->t_state = TCPS_SYN_RECEIVED; 2021 2022 /* 2023 * Advance th->th_seq to correspond to first data byte. 2024 * If data, trim to stay within window, 2025 * dropping FIN if necessary. 2026 */ 2027 th->th_seq++; 2028 if (tlen > tp->rcv_wnd) { 2029 todrop = tlen - tp->rcv_wnd; 2030 m_adj(m, -todrop); 2031 tlen = tp->rcv_wnd; 2032 tiflags &= ~TH_FIN; 2033 tcps = TCP_STAT_GETREF(); 2034 tcps[TCP_STAT_RCVPACKAFTERWIN]++; 2035 tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop; 2036 TCP_STAT_PUTREF(); 2037 } 2038 tp->snd_wl1 = th->th_seq - 1; 2039 tp->rcv_up = th->th_seq; 2040 goto step6; 2041 2042 /* 2043 * If the state is SYN_RECEIVED: 2044 * If seg contains an ACK, but not for our SYN, drop the input 2045 * and generate an RST. See page 36, rfc793 2046 */ 2047 case TCPS_SYN_RECEIVED: 2048 if ((tiflags & TH_ACK) && 2049 (SEQ_LEQ(th->th_ack, tp->iss) || 2050 SEQ_GT(th->th_ack, tp->snd_max))) 2051 goto dropwithreset; 2052 break; 2053 } 2054 2055 /* 2056 * States other than LISTEN or SYN_SENT. 2057 * First check timestamp, if present. 2058 * Then check that at least some bytes of segment are within 2059 * receive window. If segment begins before rcv_nxt, 2060 * drop leading data (and SYN); if nothing left, just ack. 2061 * 2062 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2063 * and it's less than ts_recent, drop it. 2064 */ 2065 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2066 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2067 2068 /* Check to see if ts_recent is over 24 days old. */ 2069 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2070 /* 2071 * Invalidate ts_recent. If this segment updates 2072 * ts_recent, the age will be reset later and ts_recent 2073 * will get a valid value. If it does not, setting 2074 * ts_recent to zero will at least satisfy the 2075 * requirement that zero be placed in the timestamp 2076 * echo reply when ts_recent isn't valid. The 2077 * age isn't reset until we get a valid ts_recent 2078 * because we don't want out-of-order segments to be 2079 * dropped when ts_recent is old. 2080 */ 2081 tp->ts_recent = 0; 2082 } else { 2083 tcps = TCP_STAT_GETREF(); 2084 tcps[TCP_STAT_RCVDUPPACK]++; 2085 tcps[TCP_STAT_RCVDUPBYTE] += tlen; 2086 tcps[TCP_STAT_PAWSDROP]++; 2087 TCP_STAT_PUTREF(); 2088 tcp_new_dsack(tp, th->th_seq, tlen); 2089 goto dropafterack; 2090 } 2091 } 2092 2093 todrop = tp->rcv_nxt - th->th_seq; 2094 dupseg = false; 2095 if (todrop > 0) { 2096 if (tiflags & TH_SYN) { 2097 tiflags &= ~TH_SYN; 2098 th->th_seq++; 2099 if (th->th_urp > 1) 2100 th->th_urp--; 2101 else { 2102 tiflags &= ~TH_URG; 2103 th->th_urp = 0; 2104 } 2105 todrop--; 2106 } 2107 if (todrop > tlen || 2108 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2109 /* 2110 * Any valid FIN or RST must be to the left of the 2111 * window. At this point the FIN or RST must be a 2112 * duplicate or out of sequence; drop it. 2113 */ 2114 if (tiflags & TH_RST) 2115 goto drop; 2116 tiflags &= ~(TH_FIN|TH_RST); 2117 /* 2118 * Send an ACK to resynchronize and drop any data. 2119 * But keep on processing for RST or ACK. 2120 */ 2121 tp->t_flags |= TF_ACKNOW; 2122 todrop = tlen; 2123 dupseg = true; 2124 tcps = TCP_STAT_GETREF(); 2125 tcps[TCP_STAT_RCVDUPPACK]++; 2126 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 2127 TCP_STAT_PUTREF(); 2128 } else if ((tiflags & TH_RST) && 2129 th->th_seq != tp->rcv_nxt) { 2130 /* 2131 * Test for reset before adjusting the sequence 2132 * number for overlapping data. 2133 */ 2134 goto dropafterack_ratelim; 2135 } else { 2136 tcps = TCP_STAT_GETREF(); 2137 tcps[TCP_STAT_RCVPARTDUPPACK]++; 2138 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 2139 TCP_STAT_PUTREF(); 2140 } 2141 tcp_new_dsack(tp, th->th_seq, todrop); 2142 hdroptlen += todrop; /*drop from head afterwards*/ 2143 th->th_seq += todrop; 2144 tlen -= todrop; 2145 if (th->th_urp > todrop) 2146 th->th_urp -= todrop; 2147 else { 2148 tiflags &= ~TH_URG; 2149 th->th_urp = 0; 2150 } 2151 } 2152 2153 /* 2154 * If new data are received on a connection after the 2155 * user processes are gone, then RST the other end. 2156 */ 2157 if ((so->so_state & SS_NOFDREF) && 2158 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2159 tp = tcp_close(tp); 2160 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 2161 goto dropwithreset; 2162 } 2163 2164 /* 2165 * If segment ends after window, drop trailing data 2166 * (and PUSH and FIN); if nothing left, just ACK. 2167 */ 2168 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 2169 if (todrop > 0) { 2170 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 2171 if (todrop >= tlen) { 2172 /* 2173 * The segment actually starts after the window. 2174 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2175 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2176 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2177 */ 2178 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 2179 /* 2180 * If a new connection request is received 2181 * while in TIME_WAIT, drop the old connection 2182 * and start over if the sequence numbers 2183 * are above the previous ones. 2184 * 2185 * NOTE: We will checksum the packet again, and 2186 * so we need to put the header fields back into 2187 * network order! 2188 * XXX This kind of sucks, but we don't expect 2189 * XXX this to happen very often, so maybe it 2190 * XXX doesn't matter so much. 2191 */ 2192 if (tiflags & TH_SYN && 2193 tp->t_state == TCPS_TIME_WAIT && 2194 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2195 tp = tcp_close(tp); 2196 tcp_fields_to_net(th); 2197 goto findpcb; 2198 } 2199 /* 2200 * If window is closed can only take segments at 2201 * window edge, and have to drop data and PUSH from 2202 * incoming segments. Continue processing, but 2203 * remember to ack. Otherwise, drop segment 2204 * and (if not RST) ack. 2205 */ 2206 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2207 tp->t_flags |= TF_ACKNOW; 2208 TCP_STATINC(TCP_STAT_RCVWINPROBE); 2209 } else 2210 goto dropafterack; 2211 } else 2212 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 2213 m_adj(m, -todrop); 2214 tlen -= todrop; 2215 tiflags &= ~(TH_PUSH|TH_FIN); 2216 } 2217 2218 /* 2219 * If last ACK falls within this segment's sequence numbers, 2220 * record the timestamp. 2221 * NOTE: 2222 * 1) That the test incorporates suggestions from the latest 2223 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2224 * 2) That updating only on newer timestamps interferes with 2225 * our earlier PAWS tests, so this check should be solely 2226 * predicated on the sequence space of this segment. 2227 * 3) That we modify the segment boundary check to be 2228 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2229 * instead of RFC1323's 2230 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2231 * This modified check allows us to overcome RFC1323's 2232 * limitations as described in Stevens TCP/IP Illustrated 2233 * Vol. 2 p.869. In such cases, we can still calculate the 2234 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2235 */ 2236 if (opti.ts_present && 2237 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2238 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2239 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2240 tp->ts_recent_age = tcp_now; 2241 tp->ts_recent = opti.ts_val; 2242 } 2243 2244 /* 2245 * If the RST bit is set examine the state: 2246 * SYN_RECEIVED STATE: 2247 * If passive open, return to LISTEN state. 2248 * If active open, inform user that connection was refused. 2249 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 2250 * Inform user that connection was reset, and close tcb. 2251 * CLOSING, LAST_ACK, TIME_WAIT STATES 2252 * Close the tcb. 2253 */ 2254 if (tiflags & TH_RST) { 2255 if (th->th_seq != tp->rcv_nxt) 2256 goto dropafterack_ratelim; 2257 2258 switch (tp->t_state) { 2259 case TCPS_SYN_RECEIVED: 2260 so->so_error = ECONNREFUSED; 2261 goto close; 2262 2263 case TCPS_ESTABLISHED: 2264 case TCPS_FIN_WAIT_1: 2265 case TCPS_FIN_WAIT_2: 2266 case TCPS_CLOSE_WAIT: 2267 so->so_error = ECONNRESET; 2268 close: 2269 tp->t_state = TCPS_CLOSED; 2270 TCP_STATINC(TCP_STAT_DROPS); 2271 tp = tcp_close(tp); 2272 goto drop; 2273 2274 case TCPS_CLOSING: 2275 case TCPS_LAST_ACK: 2276 case TCPS_TIME_WAIT: 2277 tp = tcp_close(tp); 2278 goto drop; 2279 } 2280 } 2281 2282 /* 2283 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2284 * we must be in a synchronized state. RFC791 states (under RST 2285 * generation) that any unacceptable segment (an out-of-order SYN 2286 * qualifies) received in a synchronized state must elicit only an 2287 * empty acknowledgment segment ... and the connection remains in 2288 * the same state. 2289 */ 2290 if (tiflags & TH_SYN) { 2291 if (tp->rcv_nxt == th->th_seq) { 2292 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2293 TH_ACK); 2294 if (tcp_saveti) 2295 m_freem(tcp_saveti); 2296 return; 2297 } 2298 2299 goto dropafterack_ratelim; 2300 } 2301 2302 /* 2303 * If the ACK bit is off we drop the segment and return. 2304 */ 2305 if ((tiflags & TH_ACK) == 0) { 2306 if (tp->t_flags & TF_ACKNOW) 2307 goto dropafterack; 2308 else 2309 goto drop; 2310 } 2311 2312 /* 2313 * Ack processing. 2314 */ 2315 switch (tp->t_state) { 2316 2317 /* 2318 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2319 * ESTABLISHED state and continue processing, otherwise 2320 * send an RST. 2321 */ 2322 case TCPS_SYN_RECEIVED: 2323 if (SEQ_GT(tp->snd_una, th->th_ack) || 2324 SEQ_GT(th->th_ack, tp->snd_max)) 2325 goto dropwithreset; 2326 TCP_STATINC(TCP_STAT_CONNECTS); 2327 soisconnected(so); 2328 tcp_established(tp); 2329 /* Do window scaling? */ 2330 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2331 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2332 tp->snd_scale = tp->requested_s_scale; 2333 tp->rcv_scale = tp->request_r_scale; 2334 } 2335 TCP_REASS_LOCK(tp); 2336 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2337 TCP_REASS_UNLOCK(tp); 2338 tp->snd_wl1 = th->th_seq - 1; 2339 /* fall into ... */ 2340 2341 /* 2342 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2343 * ACKs. If the ack is in the range 2344 * tp->snd_una < th->th_ack <= tp->snd_max 2345 * then advance tp->snd_una to th->th_ack and drop 2346 * data from the retransmission queue. If this ACK reflects 2347 * more up to date window information we update our window information. 2348 */ 2349 case TCPS_ESTABLISHED: 2350 case TCPS_FIN_WAIT_1: 2351 case TCPS_FIN_WAIT_2: 2352 case TCPS_CLOSE_WAIT: 2353 case TCPS_CLOSING: 2354 case TCPS_LAST_ACK: 2355 case TCPS_TIME_WAIT: 2356 2357 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2358 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2359 TCP_STATINC(TCP_STAT_RCVDUPACK); 2360 /* 2361 * If we have outstanding data (other than 2362 * a window probe), this is a completely 2363 * duplicate ack (ie, window info didn't 2364 * change), the ack is the biggest we've 2365 * seen and we've seen exactly our rexmt 2366 * threshhold of them, assume a packet 2367 * has been dropped and retransmit it. 2368 * Kludge snd_nxt & the congestion 2369 * window so we send only this one 2370 * packet. 2371 */ 2372 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2373 th->th_ack != tp->snd_una) 2374 tp->t_dupacks = 0; 2375 else if (tp->t_partialacks < 0 && 2376 (++tp->t_dupacks == tcprexmtthresh || 2377 TCP_FACK_FASTRECOV(tp))) { 2378 /* 2379 * Do the fast retransmit, and adjust 2380 * congestion control paramenters. 2381 */ 2382 if (tp->t_congctl->fast_retransmit(tp, th)) { 2383 /* False fast retransmit */ 2384 break; 2385 } else 2386 goto drop; 2387 } else if (tp->t_dupacks > tcprexmtthresh) { 2388 tp->snd_cwnd += tp->t_segsz; 2389 KERNEL_LOCK(1, NULL); 2390 (void) tcp_output(tp); 2391 KERNEL_UNLOCK_ONE(NULL); 2392 goto drop; 2393 } 2394 } else { 2395 /* 2396 * If the ack appears to be very old, only 2397 * allow data that is in-sequence. This 2398 * makes it somewhat more difficult to insert 2399 * forged data by guessing sequence numbers. 2400 * Sent an ack to try to update the send 2401 * sequence number on the other side. 2402 */ 2403 if (tlen && th->th_seq != tp->rcv_nxt && 2404 SEQ_LT(th->th_ack, 2405 tp->snd_una - tp->max_sndwnd)) 2406 goto dropafterack; 2407 } 2408 break; 2409 } 2410 /* 2411 * If the congestion window was inflated to account 2412 * for the other side's cached packets, retract it. 2413 */ 2414 /* XXX: make SACK have his own congestion control 2415 * struct -- rpaulo */ 2416 if (TCP_SACK_ENABLED(tp)) 2417 tcp_sack_newack(tp, th); 2418 else 2419 tp->t_congctl->fast_retransmit_newack(tp, th); 2420 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2421 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2422 goto dropafterack; 2423 } 2424 acked = th->th_ack - tp->snd_una; 2425 tcps = TCP_STAT_GETREF(); 2426 tcps[TCP_STAT_RCVACKPACK]++; 2427 tcps[TCP_STAT_RCVACKBYTE] += acked; 2428 TCP_STAT_PUTREF(); 2429 2430 /* 2431 * If we have a timestamp reply, update smoothed 2432 * round trip time. If no timestamp is present but 2433 * transmit timer is running and timed sequence 2434 * number was acked, update smoothed round trip time. 2435 * Since we now have an rtt measurement, cancel the 2436 * timer backoff (cf., Phil Karn's retransmit alg.). 2437 * Recompute the initial retransmit timer. 2438 */ 2439 if (ts_rtt) 2440 tcp_xmit_timer(tp, ts_rtt); 2441 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2442 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2443 2444 /* 2445 * If all outstanding data is acked, stop retransmit 2446 * timer and remember to restart (more output or persist). 2447 * If there is more data to be acked, restart retransmit 2448 * timer, using current (possibly backed-off) value. 2449 */ 2450 if (th->th_ack == tp->snd_max) { 2451 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2452 needoutput = 1; 2453 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2454 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2455 2456 /* 2457 * New data has been acked, adjust the congestion window. 2458 */ 2459 tp->t_congctl->newack(tp, th); 2460 2461 nd6_hint(tp); 2462 if (acked > so->so_snd.sb_cc) { 2463 tp->snd_wnd -= so->so_snd.sb_cc; 2464 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2465 ourfinisacked = 1; 2466 } else { 2467 if (acked > (tp->t_lastoff - tp->t_inoff)) 2468 tp->t_lastm = NULL; 2469 sbdrop(&so->so_snd, acked); 2470 tp->t_lastoff -= acked; 2471 tp->snd_wnd -= acked; 2472 ourfinisacked = 0; 2473 } 2474 sowwakeup(so); 2475 2476 icmp_check(tp, th, acked); 2477 2478 tp->snd_una = th->th_ack; 2479 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2480 tp->snd_fack = tp->snd_una; 2481 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2482 tp->snd_nxt = tp->snd_una; 2483 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2484 tp->snd_high = tp->snd_una; 2485 2486 switch (tp->t_state) { 2487 2488 /* 2489 * In FIN_WAIT_1 STATE in addition to the processing 2490 * for the ESTABLISHED state if our FIN is now acknowledged 2491 * then enter FIN_WAIT_2. 2492 */ 2493 case TCPS_FIN_WAIT_1: 2494 if (ourfinisacked) { 2495 /* 2496 * If we can't receive any more 2497 * data, then closing user can proceed. 2498 * Starting the timer is contrary to the 2499 * specification, but if we don't get a FIN 2500 * we'll hang forever. 2501 */ 2502 if (so->so_state & SS_CANTRCVMORE) { 2503 soisdisconnected(so); 2504 if (tp->t_maxidle > 0) 2505 TCP_TIMER_ARM(tp, TCPT_2MSL, 2506 tp->t_maxidle); 2507 } 2508 tp->t_state = TCPS_FIN_WAIT_2; 2509 } 2510 break; 2511 2512 /* 2513 * In CLOSING STATE in addition to the processing for 2514 * the ESTABLISHED state if the ACK acknowledges our FIN 2515 * then enter the TIME-WAIT state, otherwise ignore 2516 * the segment. 2517 */ 2518 case TCPS_CLOSING: 2519 if (ourfinisacked) { 2520 tp->t_state = TCPS_TIME_WAIT; 2521 tcp_canceltimers(tp); 2522 TCP_TIMER_ARM(tp, TCPT_2MSL, 2523 2 * PR_SLOWHZ * tcp_msl); 2524 soisdisconnected(so); 2525 } 2526 break; 2527 2528 /* 2529 * In LAST_ACK, we may still be waiting for data to drain 2530 * and/or to be acked, as well as for the ack of our FIN. 2531 * If our FIN is now acknowledged, delete the TCB, 2532 * enter the closed state and return. 2533 */ 2534 case TCPS_LAST_ACK: 2535 if (ourfinisacked) { 2536 tp = tcp_close(tp); 2537 goto drop; 2538 } 2539 break; 2540 2541 /* 2542 * In TIME_WAIT state the only thing that should arrive 2543 * is a retransmission of the remote FIN. Acknowledge 2544 * it and restart the finack timer. 2545 */ 2546 case TCPS_TIME_WAIT: 2547 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * PR_SLOWHZ * tcp_msl); 2548 goto dropafterack; 2549 } 2550 } 2551 2552 step6: 2553 /* 2554 * Update window information. 2555 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2556 */ 2557 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2558 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2559 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2560 /* keep track of pure window updates */ 2561 if (tlen == 0 && 2562 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2563 TCP_STATINC(TCP_STAT_RCVWINUPD); 2564 tp->snd_wnd = tiwin; 2565 tp->snd_wl1 = th->th_seq; 2566 tp->snd_wl2 = th->th_ack; 2567 if (tp->snd_wnd > tp->max_sndwnd) 2568 tp->max_sndwnd = tp->snd_wnd; 2569 needoutput = 1; 2570 } 2571 2572 /* 2573 * Process segments with URG. 2574 */ 2575 if ((tiflags & TH_URG) && th->th_urp && 2576 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2577 /* 2578 * This is a kludge, but if we receive and accept 2579 * random urgent pointers, we'll crash in 2580 * soreceive. It's hard to imagine someone 2581 * actually wanting to send this much urgent data. 2582 */ 2583 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2584 th->th_urp = 0; /* XXX */ 2585 tiflags &= ~TH_URG; /* XXX */ 2586 goto dodata; /* XXX */ 2587 } 2588 /* 2589 * If this segment advances the known urgent pointer, 2590 * then mark the data stream. This should not happen 2591 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2592 * a FIN has been received from the remote side. 2593 * In these states we ignore the URG. 2594 * 2595 * According to RFC961 (Assigned Protocols), 2596 * the urgent pointer points to the last octet 2597 * of urgent data. We continue, however, 2598 * to consider it to indicate the first octet 2599 * of data past the urgent section as the original 2600 * spec states (in one of two places). 2601 */ 2602 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2603 tp->rcv_up = th->th_seq + th->th_urp; 2604 so->so_oobmark = so->so_rcv.sb_cc + 2605 (tp->rcv_up - tp->rcv_nxt) - 1; 2606 if (so->so_oobmark == 0) 2607 so->so_state |= SS_RCVATMARK; 2608 sohasoutofband(so); 2609 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2610 } 2611 /* 2612 * Remove out of band data so doesn't get presented to user. 2613 * This can happen independent of advancing the URG pointer, 2614 * but if two URG's are pending at once, some out-of-band 2615 * data may creep in... ick. 2616 */ 2617 if (th->th_urp <= (u_int16_t) tlen 2618 #ifdef SO_OOBINLINE 2619 && (so->so_options & SO_OOBINLINE) == 0 2620 #endif 2621 ) 2622 tcp_pulloutofband(so, th, m, hdroptlen); 2623 } else 2624 /* 2625 * If no out of band data is expected, 2626 * pull receive urgent pointer along 2627 * with the receive window. 2628 */ 2629 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2630 tp->rcv_up = tp->rcv_nxt; 2631 dodata: /* XXX */ 2632 2633 /* 2634 * Process the segment text, merging it into the TCP sequencing queue, 2635 * and arranging for acknowledgement of receipt if necessary. 2636 * This process logically involves adjusting tp->rcv_wnd as data 2637 * is presented to the user (this happens in tcp_usrreq.c, 2638 * case PRU_RCVD). If a FIN has already been received on this 2639 * connection then we just ignore the text. 2640 */ 2641 if ((tlen || (tiflags & TH_FIN)) && 2642 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2643 /* 2644 * Insert segment ti into reassembly queue of tcp with 2645 * control block tp. Return TH_FIN if reassembly now includes 2646 * a segment with FIN. The macro form does the common case 2647 * inline (segment is the next to be received on an 2648 * established connection, and the queue is empty), 2649 * avoiding linkage into and removal from the queue and 2650 * repetition of various conversions. 2651 * Set DELACK for segments received in order, but ack 2652 * immediately when segments are out of order 2653 * (so fast retransmit can work). 2654 */ 2655 /* NOTE: this was TCP_REASS() macro, but used only once */ 2656 TCP_REASS_LOCK(tp); 2657 if (th->th_seq == tp->rcv_nxt && 2658 TAILQ_FIRST(&tp->segq) == NULL && 2659 tp->t_state == TCPS_ESTABLISHED) { 2660 tcp_setup_ack(tp, th); 2661 tp->rcv_nxt += tlen; 2662 tiflags = th->th_flags & TH_FIN; 2663 tcps = TCP_STAT_GETREF(); 2664 tcps[TCP_STAT_RCVPACK]++; 2665 tcps[TCP_STAT_RCVBYTE] += tlen; 2666 TCP_STAT_PUTREF(); 2667 nd6_hint(tp); 2668 if (so->so_state & SS_CANTRCVMORE) 2669 m_freem(m); 2670 else { 2671 m_adj(m, hdroptlen); 2672 sbappendstream(&(so)->so_rcv, m); 2673 } 2674 TCP_REASS_UNLOCK(tp); 2675 sorwakeup(so); 2676 } else { 2677 m_adj(m, hdroptlen); 2678 tiflags = tcp_reass(tp, th, m, &tlen); 2679 tp->t_flags |= TF_ACKNOW; 2680 TCP_REASS_UNLOCK(tp); 2681 } 2682 2683 /* 2684 * Note the amount of data that peer has sent into 2685 * our window, in order to estimate the sender's 2686 * buffer size. 2687 */ 2688 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2689 } else { 2690 m_freem(m); 2691 m = NULL; 2692 tiflags &= ~TH_FIN; 2693 } 2694 2695 /* 2696 * If FIN is received ACK the FIN and let the user know 2697 * that the connection is closing. Ignore a FIN received before 2698 * the connection is fully established. 2699 */ 2700 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2701 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2702 socantrcvmore(so); 2703 tp->t_flags |= TF_ACKNOW; 2704 tp->rcv_nxt++; 2705 } 2706 switch (tp->t_state) { 2707 2708 /* 2709 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2710 */ 2711 case TCPS_ESTABLISHED: 2712 tp->t_state = TCPS_CLOSE_WAIT; 2713 break; 2714 2715 /* 2716 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2717 * enter the CLOSING state. 2718 */ 2719 case TCPS_FIN_WAIT_1: 2720 tp->t_state = TCPS_CLOSING; 2721 break; 2722 2723 /* 2724 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2725 * starting the time-wait timer, turning off the other 2726 * standard timers. 2727 */ 2728 case TCPS_FIN_WAIT_2: 2729 tp->t_state = TCPS_TIME_WAIT; 2730 tcp_canceltimers(tp); 2731 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * PR_SLOWHZ * tcp_msl); 2732 soisdisconnected(so); 2733 break; 2734 2735 /* 2736 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2737 */ 2738 case TCPS_TIME_WAIT: 2739 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * PR_SLOWHZ * tcp_msl); 2740 break; 2741 } 2742 } 2743 #ifdef TCP_DEBUG 2744 if (so->so_options & SO_DEBUG) 2745 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2746 #endif 2747 2748 /* 2749 * Return any desired output. 2750 */ 2751 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2752 KERNEL_LOCK(1, NULL); 2753 (void) tcp_output(tp); 2754 KERNEL_UNLOCK_ONE(NULL); 2755 } 2756 if (tcp_saveti) 2757 m_freem(tcp_saveti); 2758 return; 2759 2760 badsyn: 2761 /* 2762 * Received a bad SYN. Increment counters and dropwithreset. 2763 */ 2764 TCP_STATINC(TCP_STAT_BADSYN); 2765 tp = NULL; 2766 goto dropwithreset; 2767 2768 dropafterack: 2769 /* 2770 * Generate an ACK dropping incoming segment if it occupies 2771 * sequence space, where the ACK reflects our state. 2772 */ 2773 if (tiflags & TH_RST) 2774 goto drop; 2775 goto dropafterack2; 2776 2777 dropafterack_ratelim: 2778 /* 2779 * We may want to rate-limit ACKs against SYN/RST attack. 2780 */ 2781 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2782 tcp_ackdrop_ppslim) == 0) { 2783 /* XXX stat */ 2784 goto drop; 2785 } 2786 /* ...fall into dropafterack2... */ 2787 2788 dropafterack2: 2789 m_freem(m); 2790 tp->t_flags |= TF_ACKNOW; 2791 KERNEL_LOCK(1, NULL); 2792 (void) tcp_output(tp); 2793 KERNEL_UNLOCK_ONE(NULL); 2794 if (tcp_saveti) 2795 m_freem(tcp_saveti); 2796 return; 2797 2798 dropwithreset_ratelim: 2799 /* 2800 * We may want to rate-limit RSTs in certain situations, 2801 * particularly if we are sending an RST in response to 2802 * an attempt to connect to or otherwise communicate with 2803 * a port for which we have no socket. 2804 */ 2805 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2806 tcp_rst_ppslim) == 0) { 2807 /* XXX stat */ 2808 goto drop; 2809 } 2810 /* ...fall into dropwithreset... */ 2811 2812 dropwithreset: 2813 /* 2814 * Generate a RST, dropping incoming segment. 2815 * Make ACK acceptable to originator of segment. 2816 */ 2817 if (tiflags & TH_RST) 2818 goto drop; 2819 2820 switch (af) { 2821 #ifdef INET6 2822 case AF_INET6: 2823 /* For following calls to tcp_respond */ 2824 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2825 goto drop; 2826 break; 2827 #endif /* INET6 */ 2828 case AF_INET: 2829 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2830 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2831 goto drop; 2832 } 2833 2834 if (tiflags & TH_ACK) 2835 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2836 else { 2837 if (tiflags & TH_SYN) 2838 tlen++; 2839 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2840 TH_RST|TH_ACK); 2841 } 2842 if (tcp_saveti) 2843 m_freem(tcp_saveti); 2844 return; 2845 2846 badcsum: 2847 drop: 2848 /* 2849 * Drop space held by incoming segment and return. 2850 */ 2851 if (tp) { 2852 if (tp->t_inpcb) 2853 so = tp->t_inpcb->inp_socket; 2854 #ifdef INET6 2855 else if (tp->t_in6pcb) 2856 so = tp->t_in6pcb->in6p_socket; 2857 #endif 2858 else 2859 so = NULL; 2860 #ifdef TCP_DEBUG 2861 if (so && (so->so_options & SO_DEBUG) != 0) 2862 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2863 #endif 2864 } 2865 if (tcp_saveti) 2866 m_freem(tcp_saveti); 2867 m_freem(m); 2868 return; 2869 } 2870 2871 #ifdef TCP_SIGNATURE 2872 int 2873 tcp_signature_apply(void *fstate, void *data, u_int len) 2874 { 2875 2876 MD5Update(fstate, (u_char *)data, len); 2877 return (0); 2878 } 2879 2880 struct secasvar * 2881 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 2882 { 2883 struct secasvar *sav; 2884 #ifdef FAST_IPSEC 2885 union sockaddr_union dst; 2886 #endif 2887 struct ip *ip; 2888 struct ip6_hdr *ip6; 2889 2890 ip = mtod(m, struct ip *); 2891 switch (ip->ip_v) { 2892 case 4: 2893 ip = mtod(m, struct ip *); 2894 ip6 = NULL; 2895 break; 2896 case 6: 2897 ip = NULL; 2898 ip6 = mtod(m, struct ip6_hdr *); 2899 break; 2900 default: 2901 return (NULL); 2902 } 2903 2904 #ifdef FAST_IPSEC 2905 /* Extract the destination from the IP header in the mbuf. */ 2906 memset(&dst, 0, sizeof(union sockaddr_union)); 2907 if (ip !=NULL) { 2908 dst.sa.sa_len = sizeof(struct sockaddr_in); 2909 dst.sa.sa_family = AF_INET; 2910 dst.sin.sin_addr = ip->ip_dst; 2911 } else { 2912 dst.sa.sa_len = sizeof(struct sockaddr_in6); 2913 dst.sa.sa_family = AF_INET6; 2914 dst.sin6.sin6_addr = ip6->ip6_dst; 2915 } 2916 2917 /* 2918 * Look up an SADB entry which matches the address of the peer. 2919 */ 2920 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2921 #else 2922 if (ip) 2923 sav = key_allocsa(AF_INET, (void *)&ip->ip_src, 2924 (void *)&ip->ip_dst, IPPROTO_TCP, 2925 htonl(TCP_SIG_SPI), 0, 0); 2926 else 2927 sav = key_allocsa(AF_INET6, (void *)&ip6->ip6_src, 2928 (void *)&ip6->ip6_dst, IPPROTO_TCP, 2929 htonl(TCP_SIG_SPI), 0, 0); 2930 #endif 2931 2932 return (sav); /* freesav must be performed by caller */ 2933 } 2934 2935 int 2936 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 2937 struct secasvar *sav, char *sig) 2938 { 2939 MD5_CTX ctx; 2940 struct ip *ip; 2941 struct ipovly *ipovly; 2942 struct ip6_hdr *ip6; 2943 struct ippseudo ippseudo; 2944 struct ip6_hdr_pseudo ip6pseudo; 2945 struct tcphdr th0; 2946 int l, tcphdrlen; 2947 2948 if (sav == NULL) 2949 return (-1); 2950 2951 tcphdrlen = th->th_off * 4; 2952 2953 switch (mtod(m, struct ip *)->ip_v) { 2954 case 4: 2955 ip = mtod(m, struct ip *); 2956 ip6 = NULL; 2957 break; 2958 case 6: 2959 ip = NULL; 2960 ip6 = mtod(m, struct ip6_hdr *); 2961 break; 2962 default: 2963 return (-1); 2964 } 2965 2966 MD5Init(&ctx); 2967 2968 if (ip) { 2969 memset(&ippseudo, 0, sizeof(ippseudo)); 2970 ipovly = (struct ipovly *)ip; 2971 ippseudo.ippseudo_src = ipovly->ih_src; 2972 ippseudo.ippseudo_dst = ipovly->ih_dst; 2973 ippseudo.ippseudo_pad = 0; 2974 ippseudo.ippseudo_p = IPPROTO_TCP; 2975 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 2976 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 2977 } else { 2978 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 2979 ip6pseudo.ip6ph_src = ip6->ip6_src; 2980 in6_clearscope(&ip6pseudo.ip6ph_src); 2981 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 2982 in6_clearscope(&ip6pseudo.ip6ph_dst); 2983 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 2984 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 2985 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 2986 } 2987 2988 th0 = *th; 2989 th0.th_sum = 0; 2990 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 2991 2992 l = m->m_pkthdr.len - thoff - tcphdrlen; 2993 if (l > 0) 2994 m_apply(m, thoff + tcphdrlen, 2995 m->m_pkthdr.len - thoff - tcphdrlen, 2996 tcp_signature_apply, &ctx); 2997 2998 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2999 MD5Final(sig, &ctx); 3000 3001 return (0); 3002 } 3003 #endif 3004 3005 /* 3006 * tcp_dooptions: parse and process tcp options. 3007 * 3008 * returns -1 if this segment should be dropped. (eg. wrong signature) 3009 * otherwise returns 0. 3010 */ 3011 3012 static int 3013 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, 3014 struct tcphdr *th, 3015 struct mbuf *m, int toff, struct tcp_opt_info *oi) 3016 { 3017 u_int16_t mss; 3018 int opt, optlen = 0; 3019 #ifdef TCP_SIGNATURE 3020 void *sigp = NULL; 3021 char sigbuf[TCP_SIGLEN]; 3022 struct secasvar *sav = NULL; 3023 #endif 3024 3025 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 3026 opt = cp[0]; 3027 if (opt == TCPOPT_EOL) 3028 break; 3029 if (opt == TCPOPT_NOP) 3030 optlen = 1; 3031 else { 3032 if (cnt < 2) 3033 break; 3034 optlen = cp[1]; 3035 if (optlen < 2 || optlen > cnt) 3036 break; 3037 } 3038 switch (opt) { 3039 3040 default: 3041 continue; 3042 3043 case TCPOPT_MAXSEG: 3044 if (optlen != TCPOLEN_MAXSEG) 3045 continue; 3046 if (!(th->th_flags & TH_SYN)) 3047 continue; 3048 if (TCPS_HAVERCVDSYN(tp->t_state)) 3049 continue; 3050 bcopy(cp + 2, &mss, sizeof(mss)); 3051 oi->maxseg = ntohs(mss); 3052 break; 3053 3054 case TCPOPT_WINDOW: 3055 if (optlen != TCPOLEN_WINDOW) 3056 continue; 3057 if (!(th->th_flags & TH_SYN)) 3058 continue; 3059 if (TCPS_HAVERCVDSYN(tp->t_state)) 3060 continue; 3061 tp->t_flags |= TF_RCVD_SCALE; 3062 tp->requested_s_scale = cp[2]; 3063 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3064 #if 0 /*XXX*/ 3065 char *p; 3066 3067 if (ip) 3068 p = ntohl(ip->ip_src); 3069 #ifdef INET6 3070 else if (ip6) 3071 p = ip6_sprintf(&ip6->ip6_src); 3072 #endif 3073 else 3074 p = "(unknown)"; 3075 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3076 "assuming %d\n", 3077 tp->requested_s_scale, p, 3078 TCP_MAX_WINSHIFT); 3079 #else 3080 log(LOG_ERR, "TCP: invalid wscale %d, " 3081 "assuming %d\n", 3082 tp->requested_s_scale, 3083 TCP_MAX_WINSHIFT); 3084 #endif 3085 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3086 } 3087 break; 3088 3089 case TCPOPT_TIMESTAMP: 3090 if (optlen != TCPOLEN_TIMESTAMP) 3091 continue; 3092 oi->ts_present = 1; 3093 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 3094 NTOHL(oi->ts_val); 3095 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 3096 NTOHL(oi->ts_ecr); 3097 3098 if (!(th->th_flags & TH_SYN)) 3099 continue; 3100 if (TCPS_HAVERCVDSYN(tp->t_state)) 3101 continue; 3102 /* 3103 * A timestamp received in a SYN makes 3104 * it ok to send timestamp requests and replies. 3105 */ 3106 tp->t_flags |= TF_RCVD_TSTMP; 3107 tp->ts_recent = oi->ts_val; 3108 tp->ts_recent_age = tcp_now; 3109 break; 3110 3111 case TCPOPT_SACK_PERMITTED: 3112 if (optlen != TCPOLEN_SACK_PERMITTED) 3113 continue; 3114 if (!(th->th_flags & TH_SYN)) 3115 continue; 3116 if (TCPS_HAVERCVDSYN(tp->t_state)) 3117 continue; 3118 if (tcp_do_sack) { 3119 tp->t_flags |= TF_SACK_PERMIT; 3120 tp->t_flags |= TF_WILL_SACK; 3121 } 3122 break; 3123 3124 case TCPOPT_SACK: 3125 tcp_sack_option(tp, th, cp, optlen); 3126 break; 3127 #ifdef TCP_SIGNATURE 3128 case TCPOPT_SIGNATURE: 3129 if (optlen != TCPOLEN_SIGNATURE) 3130 continue; 3131 if (sigp && memcmp(sigp, cp + 2, TCP_SIGLEN)) 3132 return (-1); 3133 3134 sigp = sigbuf; 3135 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3136 tp->t_flags |= TF_SIGNATURE; 3137 break; 3138 #endif 3139 } 3140 } 3141 3142 #ifdef TCP_SIGNATURE 3143 if (tp->t_flags & TF_SIGNATURE) { 3144 3145 sav = tcp_signature_getsav(m, th); 3146 3147 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3148 return (-1); 3149 } 3150 3151 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 3152 if (sav == NULL) 3153 return (-1); 3154 #ifdef FAST_IPSEC 3155 KEY_FREESAV(&sav); 3156 #else 3157 key_freesav(sav); 3158 #endif 3159 return (-1); 3160 } 3161 3162 if (sigp) { 3163 char sig[TCP_SIGLEN]; 3164 3165 tcp_fields_to_net(th); 3166 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3167 tcp_fields_to_host(th); 3168 if (sav == NULL) 3169 return (-1); 3170 #ifdef FAST_IPSEC 3171 KEY_FREESAV(&sav); 3172 #else 3173 key_freesav(sav); 3174 #endif 3175 return (-1); 3176 } 3177 tcp_fields_to_host(th); 3178 3179 if (memcmp(sig, sigp, TCP_SIGLEN)) { 3180 TCP_STATINC(TCP_STAT_BADSIG); 3181 if (sav == NULL) 3182 return (-1); 3183 #ifdef FAST_IPSEC 3184 KEY_FREESAV(&sav); 3185 #else 3186 key_freesav(sav); 3187 #endif 3188 return (-1); 3189 } else 3190 TCP_STATINC(TCP_STAT_GOODSIG); 3191 3192 key_sa_recordxfer(sav, m); 3193 #ifdef FAST_IPSEC 3194 KEY_FREESAV(&sav); 3195 #else 3196 key_freesav(sav); 3197 #endif 3198 } 3199 #endif 3200 3201 return (0); 3202 } 3203 3204 /* 3205 * Pull out of band byte out of a segment so 3206 * it doesn't appear in the user's data queue. 3207 * It is still reflected in the segment length for 3208 * sequencing purposes. 3209 */ 3210 void 3211 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3212 struct mbuf *m, int off) 3213 { 3214 int cnt = off + th->th_urp - 1; 3215 3216 while (cnt >= 0) { 3217 if (m->m_len > cnt) { 3218 char *cp = mtod(m, char *) + cnt; 3219 struct tcpcb *tp = sototcpcb(so); 3220 3221 tp->t_iobc = *cp; 3222 tp->t_oobflags |= TCPOOB_HAVEDATA; 3223 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3224 m->m_len--; 3225 return; 3226 } 3227 cnt -= m->m_len; 3228 m = m->m_next; 3229 if (m == 0) 3230 break; 3231 } 3232 panic("tcp_pulloutofband"); 3233 } 3234 3235 /* 3236 * Collect new round-trip time estimate 3237 * and update averages and current timeout. 3238 * 3239 * rtt is in units of slow ticks (typically 500 ms) -- essentially the 3240 * difference of two timestamps. 3241 */ 3242 void 3243 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3244 { 3245 int32_t delta; 3246 3247 TCP_STATINC(TCP_STAT_RTTUPDATED); 3248 if (tp->t_srtt != 0) { 3249 /* 3250 * Compute the amount to add to srtt for smoothing, 3251 * *alpha, or 2^(-TCP_RTT_SHIFT). Because 3252 * srtt is stored in 1/32 slow ticks, we conceptually 3253 * shift left 5 bits, subtract srtt to get the 3254 * diference, and then shift right by TCP_RTT_SHIFT 3255 * (3) to obtain 1/8 of the difference. 3256 */ 3257 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3258 /* 3259 * This can never happen, because delta's lowest 3260 * possible value is 1/8 of t_srtt. But if it does, 3261 * set srtt to some reasonable value, here chosen 3262 * as 1/8 tick. 3263 */ 3264 if ((tp->t_srtt += delta) <= 0) 3265 tp->t_srtt = 1 << 2; 3266 /* 3267 * RFC2988 requires that rttvar be updated first. 3268 * This code is compliant because "delta" is the old 3269 * srtt minus the new observation (scaled). 3270 * 3271 * RFC2988 says: 3272 * rttvar = (1-beta) * rttvar + beta * |srtt-observed| 3273 * 3274 * delta is in units of 1/32 ticks, and has then been 3275 * divided by 8. This is equivalent to being in 1/16s 3276 * units and divided by 4. Subtract from it 1/4 of 3277 * the existing rttvar to form the (signed) amount to 3278 * adjust. 3279 */ 3280 if (delta < 0) 3281 delta = -delta; 3282 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3283 /* 3284 * As with srtt, this should never happen. There is 3285 * no support in RFC2988 for this operation. But 1/4s 3286 * as rttvar when faced with something arguably wrong 3287 * is ok. 3288 */ 3289 if ((tp->t_rttvar += delta) <= 0) 3290 tp->t_rttvar = 1 << 2; 3291 } else { 3292 /* 3293 * This is the first measurement. Per RFC2988, 2.2, 3294 * set rtt=R and srtt=R/2. 3295 * For srtt, storage representation is 1/32 ticks, 3296 * so shift left by 5. 3297 * For rttvar, storage representation is 1/16 ticks, 3298 * So shift left by 4, but then right by 1 to halve. 3299 */ 3300 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3301 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3302 } 3303 tp->t_rtttime = 0; 3304 tp->t_rxtshift = 0; 3305 3306 /* 3307 * the retransmit should happen at rtt + 4 * rttvar. 3308 * Because of the way we do the smoothing, srtt and rttvar 3309 * will each average +1/2 tick of bias. When we compute 3310 * the retransmit timer, we want 1/2 tick of rounding and 3311 * 1 extra tick because of +-1/2 tick uncertainty in the 3312 * firing of the timer. The bias will give us exactly the 3313 * 1.5 tick we need. But, because the bias is 3314 * statistical, we have to test that we don't drop below 3315 * the minimum feasible timer (which is 2 ticks). 3316 */ 3317 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3318 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3319 3320 /* 3321 * We received an ack for a packet that wasn't retransmitted; 3322 * it is probably safe to discard any error indications we've 3323 * received recently. This isn't quite right, but close enough 3324 * for now (a route might have failed after we sent a segment, 3325 * and the return path might not be symmetrical). 3326 */ 3327 tp->t_softerror = 0; 3328 } 3329 3330 3331 /* 3332 * TCP compressed state engine. Currently used to hold compressed 3333 * state for SYN_RECEIVED. 3334 */ 3335 3336 u_long syn_cache_count; 3337 u_int32_t syn_hash1, syn_hash2; 3338 3339 #define SYN_HASH(sa, sp, dp) \ 3340 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3341 ((u_int32_t)(sp)))^syn_hash2))) 3342 #ifndef INET6 3343 #define SYN_HASHALL(hash, src, dst) \ 3344 do { \ 3345 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3346 ((const struct sockaddr_in *)(src))->sin_port, \ 3347 ((const struct sockaddr_in *)(dst))->sin_port); \ 3348 } while (/*CONSTCOND*/ 0) 3349 #else 3350 #define SYN_HASH6(sa, sp, dp) \ 3351 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3352 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3353 & 0x7fffffff) 3354 3355 #define SYN_HASHALL(hash, src, dst) \ 3356 do { \ 3357 switch ((src)->sa_family) { \ 3358 case AF_INET: \ 3359 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3360 ((const struct sockaddr_in *)(src))->sin_port, \ 3361 ((const struct sockaddr_in *)(dst))->sin_port); \ 3362 break; \ 3363 case AF_INET6: \ 3364 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3365 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3366 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3367 break; \ 3368 default: \ 3369 hash = 0; \ 3370 } \ 3371 } while (/*CONSTCOND*/0) 3372 #endif /* INET6 */ 3373 3374 static struct pool syn_cache_pool; 3375 3376 /* 3377 * We don't estimate RTT with SYNs, so each packet starts with the default 3378 * RTT and each timer step has a fixed timeout value. 3379 */ 3380 #define SYN_CACHE_TIMER_ARM(sc) \ 3381 do { \ 3382 TCPT_RANGESET((sc)->sc_rxtcur, \ 3383 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3384 TCPTV_REXMTMAX); \ 3385 callout_reset(&(sc)->sc_timer, \ 3386 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3387 } while (/*CONSTCOND*/0) 3388 3389 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3390 3391 static inline void 3392 syn_cache_rm(struct syn_cache *sc) 3393 { 3394 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3395 sc, sc_bucketq); 3396 sc->sc_tp = NULL; 3397 LIST_REMOVE(sc, sc_tpq); 3398 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3399 callout_stop(&sc->sc_timer); 3400 syn_cache_count--; 3401 } 3402 3403 static inline void 3404 syn_cache_put(struct syn_cache *sc) 3405 { 3406 if (sc->sc_ipopts) 3407 (void) m_free(sc->sc_ipopts); 3408 rtcache_free(&sc->sc_route); 3409 sc->sc_flags |= SCF_DEAD; 3410 if (!callout_invoking(&sc->sc_timer)) 3411 callout_schedule(&(sc)->sc_timer, 1); 3412 } 3413 3414 void 3415 syn_cache_init(void) 3416 { 3417 int i; 3418 3419 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 3420 "synpl", NULL, IPL_SOFTNET); 3421 3422 /* Initialize the hash buckets. */ 3423 for (i = 0; i < tcp_syn_cache_size; i++) 3424 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3425 } 3426 3427 void 3428 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3429 { 3430 struct syn_cache_head *scp; 3431 struct syn_cache *sc2; 3432 int s; 3433 3434 /* 3435 * If there are no entries in the hash table, reinitialize 3436 * the hash secrets. 3437 */ 3438 if (syn_cache_count == 0) { 3439 syn_hash1 = arc4random(); 3440 syn_hash2 = arc4random(); 3441 } 3442 3443 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3444 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3445 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3446 3447 /* 3448 * Make sure that we don't overflow the per-bucket 3449 * limit or the total cache size limit. 3450 */ 3451 s = splsoftnet(); 3452 if (scp->sch_length >= tcp_syn_bucket_limit) { 3453 TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW); 3454 /* 3455 * The bucket is full. Toss the oldest element in the 3456 * bucket. This will be the first entry in the bucket. 3457 */ 3458 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3459 #ifdef DIAGNOSTIC 3460 /* 3461 * This should never happen; we should always find an 3462 * entry in our bucket. 3463 */ 3464 if (sc2 == NULL) 3465 panic("syn_cache_insert: bucketoverflow: impossible"); 3466 #endif 3467 syn_cache_rm(sc2); 3468 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3469 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3470 struct syn_cache_head *scp2, *sce; 3471 3472 TCP_STATINC(TCP_STAT_SC_OVERFLOWED); 3473 /* 3474 * The cache is full. Toss the oldest entry in the 3475 * first non-empty bucket we can find. 3476 * 3477 * XXX We would really like to toss the oldest 3478 * entry in the cache, but we hope that this 3479 * condition doesn't happen very often. 3480 */ 3481 scp2 = scp; 3482 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3483 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3484 for (++scp2; scp2 != scp; scp2++) { 3485 if (scp2 >= sce) 3486 scp2 = &tcp_syn_cache[0]; 3487 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3488 break; 3489 } 3490 #ifdef DIAGNOSTIC 3491 /* 3492 * This should never happen; we should always find a 3493 * non-empty bucket. 3494 */ 3495 if (scp2 == scp) 3496 panic("syn_cache_insert: cacheoverflow: " 3497 "impossible"); 3498 #endif 3499 } 3500 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3501 syn_cache_rm(sc2); 3502 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3503 } 3504 3505 /* 3506 * Initialize the entry's timer. 3507 */ 3508 sc->sc_rxttot = 0; 3509 sc->sc_rxtshift = 0; 3510 SYN_CACHE_TIMER_ARM(sc); 3511 3512 /* Link it from tcpcb entry */ 3513 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3514 3515 /* Put it into the bucket. */ 3516 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3517 scp->sch_length++; 3518 syn_cache_count++; 3519 3520 TCP_STATINC(TCP_STAT_SC_ADDED); 3521 splx(s); 3522 } 3523 3524 /* 3525 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3526 * If we have retransmitted an entry the maximum number of times, expire 3527 * that entry. 3528 */ 3529 void 3530 syn_cache_timer(void *arg) 3531 { 3532 struct syn_cache *sc = arg; 3533 3534 mutex_enter(softnet_lock); 3535 KERNEL_LOCK(1, NULL); 3536 callout_ack(&sc->sc_timer); 3537 3538 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3539 TCP_STATINC(TCP_STAT_SC_DELAYED_FREE); 3540 callout_destroy(&sc->sc_timer); 3541 pool_put(&syn_cache_pool, sc); 3542 KERNEL_UNLOCK_ONE(NULL); 3543 mutex_exit(softnet_lock); 3544 return; 3545 } 3546 3547 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3548 /* Drop it -- too many retransmissions. */ 3549 goto dropit; 3550 } 3551 3552 /* 3553 * Compute the total amount of time this entry has 3554 * been on a queue. If this entry has been on longer 3555 * than the keep alive timer would allow, expire it. 3556 */ 3557 sc->sc_rxttot += sc->sc_rxtcur; 3558 if (sc->sc_rxttot >= tcp_keepinit) 3559 goto dropit; 3560 3561 TCP_STATINC(TCP_STAT_SC_RETRANSMITTED); 3562 (void) syn_cache_respond(sc, NULL); 3563 3564 /* Advance the timer back-off. */ 3565 sc->sc_rxtshift++; 3566 SYN_CACHE_TIMER_ARM(sc); 3567 3568 KERNEL_UNLOCK_ONE(NULL); 3569 mutex_exit(softnet_lock); 3570 return; 3571 3572 dropit: 3573 TCP_STATINC(TCP_STAT_SC_TIMED_OUT); 3574 syn_cache_rm(sc); 3575 if (sc->sc_ipopts) 3576 (void) m_free(sc->sc_ipopts); 3577 rtcache_free(&sc->sc_route); 3578 callout_destroy(&sc->sc_timer); 3579 pool_put(&syn_cache_pool, sc); 3580 KERNEL_UNLOCK_ONE(NULL); 3581 mutex_exit(softnet_lock); 3582 } 3583 3584 /* 3585 * Remove syn cache created by the specified tcb entry, 3586 * because this does not make sense to keep them 3587 * (if there's no tcb entry, syn cache entry will never be used) 3588 */ 3589 void 3590 syn_cache_cleanup(struct tcpcb *tp) 3591 { 3592 struct syn_cache *sc, *nsc; 3593 int s; 3594 3595 s = splsoftnet(); 3596 3597 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3598 nsc = LIST_NEXT(sc, sc_tpq); 3599 3600 #ifdef DIAGNOSTIC 3601 if (sc->sc_tp != tp) 3602 panic("invalid sc_tp in syn_cache_cleanup"); 3603 #endif 3604 syn_cache_rm(sc); 3605 syn_cache_put(sc); /* calls pool_put but see spl above */ 3606 } 3607 /* just for safety */ 3608 LIST_INIT(&tp->t_sc); 3609 3610 splx(s); 3611 } 3612 3613 /* 3614 * Find an entry in the syn cache. 3615 */ 3616 struct syn_cache * 3617 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3618 struct syn_cache_head **headp) 3619 { 3620 struct syn_cache *sc; 3621 struct syn_cache_head *scp; 3622 u_int32_t hash; 3623 int s; 3624 3625 SYN_HASHALL(hash, src, dst); 3626 3627 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3628 *headp = scp; 3629 s = splsoftnet(); 3630 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3631 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3632 if (sc->sc_hash != hash) 3633 continue; 3634 if (!memcmp(&sc->sc_src, src, src->sa_len) && 3635 !memcmp(&sc->sc_dst, dst, dst->sa_len)) { 3636 splx(s); 3637 return (sc); 3638 } 3639 } 3640 splx(s); 3641 return (NULL); 3642 } 3643 3644 /* 3645 * This function gets called when we receive an ACK for a 3646 * socket in the LISTEN state. We look up the connection 3647 * in the syn cache, and if its there, we pull it out of 3648 * the cache and turn it into a full-blown connection in 3649 * the SYN-RECEIVED state. 3650 * 3651 * The return values may not be immediately obvious, and their effects 3652 * can be subtle, so here they are: 3653 * 3654 * NULL SYN was not found in cache; caller should drop the 3655 * packet and send an RST. 3656 * 3657 * -1 We were unable to create the new connection, and are 3658 * aborting it. An ACK,RST is being sent to the peer 3659 * (unless we got screwey sequence numbners; see below), 3660 * because the 3-way handshake has been completed. Caller 3661 * should not free the mbuf, since we may be using it. If 3662 * we are not, we will free it. 3663 * 3664 * Otherwise, the return value is a pointer to the new socket 3665 * associated with the connection. 3666 */ 3667 struct socket * 3668 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3669 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3670 struct socket *so, struct mbuf *m) 3671 { 3672 struct syn_cache *sc; 3673 struct syn_cache_head *scp; 3674 struct inpcb *inp = NULL; 3675 #ifdef INET6 3676 struct in6pcb *in6p = NULL; 3677 #endif 3678 struct tcpcb *tp = 0; 3679 struct mbuf *am; 3680 int s; 3681 struct socket *oso; 3682 3683 s = splsoftnet(); 3684 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3685 splx(s); 3686 return (NULL); 3687 } 3688 3689 /* 3690 * Verify the sequence and ack numbers. Try getting the correct 3691 * response again. 3692 */ 3693 if ((th->th_ack != sc->sc_iss + 1) || 3694 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3695 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3696 (void) syn_cache_respond(sc, m); 3697 splx(s); 3698 return ((struct socket *)(-1)); 3699 } 3700 3701 /* Remove this cache entry */ 3702 syn_cache_rm(sc); 3703 splx(s); 3704 3705 /* 3706 * Ok, create the full blown connection, and set things up 3707 * as they would have been set up if we had created the 3708 * connection when the SYN arrived. If we can't create 3709 * the connection, abort it. 3710 */ 3711 /* 3712 * inp still has the OLD in_pcb stuff, set the 3713 * v6-related flags on the new guy, too. This is 3714 * done particularly for the case where an AF_INET6 3715 * socket is bound only to a port, and a v4 connection 3716 * comes in on that port. 3717 * we also copy the flowinfo from the original pcb 3718 * to the new one. 3719 */ 3720 oso = so; 3721 so = sonewconn(so, SS_ISCONNECTED); 3722 if (so == NULL) 3723 goto resetandabort; 3724 3725 switch (so->so_proto->pr_domain->dom_family) { 3726 #ifdef INET 3727 case AF_INET: 3728 inp = sotoinpcb(so); 3729 break; 3730 #endif 3731 #ifdef INET6 3732 case AF_INET6: 3733 in6p = sotoin6pcb(so); 3734 break; 3735 #endif 3736 } 3737 switch (src->sa_family) { 3738 #ifdef INET 3739 case AF_INET: 3740 if (inp) { 3741 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3742 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3743 inp->inp_options = ip_srcroute(); 3744 in_pcbstate(inp, INP_BOUND); 3745 if (inp->inp_options == NULL) { 3746 inp->inp_options = sc->sc_ipopts; 3747 sc->sc_ipopts = NULL; 3748 } 3749 } 3750 #ifdef INET6 3751 else if (in6p) { 3752 /* IPv4 packet to AF_INET6 socket */ 3753 memset(&in6p->in6p_laddr, 0, sizeof(in6p->in6p_laddr)); 3754 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3755 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3756 &in6p->in6p_laddr.s6_addr32[3], 3757 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3758 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3759 in6totcpcb(in6p)->t_family = AF_INET; 3760 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3761 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3762 else 3763 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3764 in6_pcbstate(in6p, IN6P_BOUND); 3765 } 3766 #endif 3767 break; 3768 #endif 3769 #ifdef INET6 3770 case AF_INET6: 3771 if (in6p) { 3772 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3773 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3774 in6_pcbstate(in6p, IN6P_BOUND); 3775 } 3776 break; 3777 #endif 3778 } 3779 #ifdef INET6 3780 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3781 struct in6pcb *oin6p = sotoin6pcb(oso); 3782 /* inherit socket options from the listening socket */ 3783 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3784 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3785 m_freem(in6p->in6p_options); 3786 in6p->in6p_options = 0; 3787 } 3788 ip6_savecontrol(in6p, &in6p->in6p_options, 3789 mtod(m, struct ip6_hdr *), m); 3790 } 3791 #endif 3792 3793 #if defined(IPSEC) || defined(FAST_IPSEC) 3794 /* 3795 * we make a copy of policy, instead of sharing the policy, 3796 * for better behavior in terms of SA lookup and dead SA removal. 3797 */ 3798 if (inp) { 3799 /* copy old policy into new socket's */ 3800 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3801 printf("tcp_input: could not copy policy\n"); 3802 } 3803 #ifdef INET6 3804 else if (in6p) { 3805 /* copy old policy into new socket's */ 3806 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3807 in6p->in6p_sp)) 3808 printf("tcp_input: could not copy policy\n"); 3809 } 3810 #endif 3811 #endif 3812 3813 /* 3814 * Give the new socket our cached route reference. 3815 */ 3816 if (inp) { 3817 rtcache_copy(&inp->inp_route, &sc->sc_route); 3818 rtcache_free(&sc->sc_route); 3819 } 3820 #ifdef INET6 3821 else { 3822 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 3823 rtcache_free(&sc->sc_route); 3824 } 3825 #endif 3826 3827 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3828 if (am == NULL) 3829 goto resetandabort; 3830 MCLAIM(am, &tcp_mowner); 3831 am->m_len = src->sa_len; 3832 bcopy(src, mtod(am, void *), src->sa_len); 3833 if (inp) { 3834 if (in_pcbconnect(inp, am, &lwp0)) { 3835 (void) m_free(am); 3836 goto resetandabort; 3837 } 3838 } 3839 #ifdef INET6 3840 else if (in6p) { 3841 if (src->sa_family == AF_INET) { 3842 /* IPv4 packet to AF_INET6 socket */ 3843 struct sockaddr_in6 *sin6; 3844 sin6 = mtod(am, struct sockaddr_in6 *); 3845 am->m_len = sizeof(*sin6); 3846 memset(sin6, 0, sizeof(*sin6)); 3847 sin6->sin6_family = AF_INET6; 3848 sin6->sin6_len = sizeof(*sin6); 3849 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3850 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3851 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3852 &sin6->sin6_addr.s6_addr32[3], 3853 sizeof(sin6->sin6_addr.s6_addr32[3])); 3854 } 3855 if (in6_pcbconnect(in6p, am, NULL)) { 3856 (void) m_free(am); 3857 goto resetandabort; 3858 } 3859 } 3860 #endif 3861 else { 3862 (void) m_free(am); 3863 goto resetandabort; 3864 } 3865 (void) m_free(am); 3866 3867 if (inp) 3868 tp = intotcpcb(inp); 3869 #ifdef INET6 3870 else if (in6p) 3871 tp = in6totcpcb(in6p); 3872 #endif 3873 else 3874 tp = NULL; 3875 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3876 if (sc->sc_request_r_scale != 15) { 3877 tp->requested_s_scale = sc->sc_requested_s_scale; 3878 tp->request_r_scale = sc->sc_request_r_scale; 3879 tp->snd_scale = sc->sc_requested_s_scale; 3880 tp->rcv_scale = sc->sc_request_r_scale; 3881 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3882 } 3883 if (sc->sc_flags & SCF_TIMESTAMP) 3884 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3885 tp->ts_timebase = sc->sc_timebase; 3886 3887 tp->t_template = tcp_template(tp); 3888 if (tp->t_template == 0) { 3889 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3890 so = NULL; 3891 m_freem(m); 3892 goto abort; 3893 } 3894 3895 tp->iss = sc->sc_iss; 3896 tp->irs = sc->sc_irs; 3897 tcp_sendseqinit(tp); 3898 tcp_rcvseqinit(tp); 3899 tp->t_state = TCPS_SYN_RECEIVED; 3900 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 3901 TCP_STATINC(TCP_STAT_ACCEPTS); 3902 3903 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 3904 tp->t_flags |= TF_WILL_SACK; 3905 3906 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 3907 tp->t_flags |= TF_ECN_PERMIT; 3908 3909 #ifdef TCP_SIGNATURE 3910 if (sc->sc_flags & SCF_SIGNATURE) 3911 tp->t_flags |= TF_SIGNATURE; 3912 #endif 3913 3914 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3915 tp->t_ourmss = sc->sc_ourmaxseg; 3916 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3917 3918 /* 3919 * Initialize the initial congestion window. If we 3920 * had to retransmit the SYN,ACK, we must initialize cwnd 3921 * to 1 segment (i.e. the Loss Window). 3922 */ 3923 if (sc->sc_rxtshift) 3924 tp->snd_cwnd = tp->t_peermss; 3925 else { 3926 int ss = tcp_init_win; 3927 #ifdef INET 3928 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3929 ss = tcp_init_win_local; 3930 #endif 3931 #ifdef INET6 3932 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3933 ss = tcp_init_win_local; 3934 #endif 3935 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3936 } 3937 3938 tcp_rmx_rtt(tp); 3939 tp->snd_wl1 = sc->sc_irs; 3940 tp->rcv_up = sc->sc_irs + 1; 3941 3942 /* 3943 * This is what whould have happened in tcp_output() when 3944 * the SYN,ACK was sent. 3945 */ 3946 tp->snd_up = tp->snd_una; 3947 tp->snd_max = tp->snd_nxt = tp->iss+1; 3948 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3949 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3950 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3951 tp->last_ack_sent = tp->rcv_nxt; 3952 tp->t_partialacks = -1; 3953 tp->t_dupacks = 0; 3954 3955 TCP_STATINC(TCP_STAT_SC_COMPLETED); 3956 s = splsoftnet(); 3957 syn_cache_put(sc); 3958 splx(s); 3959 return (so); 3960 3961 resetandabort: 3962 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3963 abort: 3964 if (so != NULL) { 3965 (void) soqremque(so, 1); 3966 (void) soabort(so); 3967 mutex_enter(softnet_lock); 3968 } 3969 s = splsoftnet(); 3970 syn_cache_put(sc); 3971 splx(s); 3972 TCP_STATINC(TCP_STAT_SC_ABORTED); 3973 return ((struct socket *)(-1)); 3974 } 3975 3976 /* 3977 * This function is called when we get a RST for a 3978 * non-existent connection, so that we can see if the 3979 * connection is in the syn cache. If it is, zap it. 3980 */ 3981 3982 void 3983 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 3984 { 3985 struct syn_cache *sc; 3986 struct syn_cache_head *scp; 3987 int s = splsoftnet(); 3988 3989 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3990 splx(s); 3991 return; 3992 } 3993 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3994 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3995 splx(s); 3996 return; 3997 } 3998 syn_cache_rm(sc); 3999 TCP_STATINC(TCP_STAT_SC_RESET); 4000 syn_cache_put(sc); /* calls pool_put but see spl above */ 4001 splx(s); 4002 } 4003 4004 void 4005 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 4006 struct tcphdr *th) 4007 { 4008 struct syn_cache *sc; 4009 struct syn_cache_head *scp; 4010 int s; 4011 4012 s = splsoftnet(); 4013 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4014 splx(s); 4015 return; 4016 } 4017 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 4018 if (ntohl (th->th_seq) != sc->sc_iss) { 4019 splx(s); 4020 return; 4021 } 4022 4023 /* 4024 * If we've retransmitted 3 times and this is our second error, 4025 * we remove the entry. Otherwise, we allow it to continue on. 4026 * This prevents us from incorrectly nuking an entry during a 4027 * spurious network outage. 4028 * 4029 * See tcp_notify(). 4030 */ 4031 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 4032 sc->sc_flags |= SCF_UNREACH; 4033 splx(s); 4034 return; 4035 } 4036 4037 syn_cache_rm(sc); 4038 TCP_STATINC(TCP_STAT_SC_UNREACH); 4039 syn_cache_put(sc); /* calls pool_put but see spl above */ 4040 splx(s); 4041 } 4042 4043 /* 4044 * Given a LISTEN socket and an inbound SYN request, add 4045 * this to the syn cache, and send back a segment: 4046 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 4047 * to the source. 4048 * 4049 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 4050 * Doing so would require that we hold onto the data and deliver it 4051 * to the application. However, if we are the target of a SYN-flood 4052 * DoS attack, an attacker could send data which would eventually 4053 * consume all available buffer space if it were ACKed. By not ACKing 4054 * the data, we avoid this DoS scenario. 4055 */ 4056 4057 int 4058 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 4059 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 4060 int optlen, struct tcp_opt_info *oi) 4061 { 4062 struct tcpcb tb, *tp; 4063 long win; 4064 struct syn_cache *sc; 4065 struct syn_cache_head *scp; 4066 struct mbuf *ipopts; 4067 struct tcp_opt_info opti; 4068 int s; 4069 4070 tp = sototcpcb(so); 4071 4072 memset(&opti, 0, sizeof(opti)); 4073 4074 /* 4075 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 4076 * 4077 * Note this check is performed in tcp_input() very early on. 4078 */ 4079 4080 /* 4081 * Initialize some local state. 4082 */ 4083 win = sbspace(&so->so_rcv); 4084 if (win > TCP_MAXWIN) 4085 win = TCP_MAXWIN; 4086 4087 switch (src->sa_family) { 4088 #ifdef INET 4089 case AF_INET: 4090 /* 4091 * Remember the IP options, if any. 4092 */ 4093 ipopts = ip_srcroute(); 4094 break; 4095 #endif 4096 default: 4097 ipopts = NULL; 4098 } 4099 4100 #ifdef TCP_SIGNATURE 4101 if (optp || (tp->t_flags & TF_SIGNATURE)) 4102 #else 4103 if (optp) 4104 #endif 4105 { 4106 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4107 #ifdef TCP_SIGNATURE 4108 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4109 #endif 4110 tb.t_state = TCPS_LISTEN; 4111 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 4112 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 4113 return (0); 4114 } else 4115 tb.t_flags = 0; 4116 4117 /* 4118 * See if we already have an entry for this connection. 4119 * If we do, resend the SYN,ACK. We do not count this 4120 * as a retransmission (XXX though maybe we should). 4121 */ 4122 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4123 TCP_STATINC(TCP_STAT_SC_DUPESYN); 4124 if (ipopts) { 4125 /* 4126 * If we were remembering a previous source route, 4127 * forget it and use the new one we've been given. 4128 */ 4129 if (sc->sc_ipopts) 4130 (void) m_free(sc->sc_ipopts); 4131 sc->sc_ipopts = ipopts; 4132 } 4133 sc->sc_timestamp = tb.ts_recent; 4134 if (syn_cache_respond(sc, m) == 0) { 4135 uint64_t *tcps = TCP_STAT_GETREF(); 4136 tcps[TCP_STAT_SNDACKS]++; 4137 tcps[TCP_STAT_SNDTOTAL]++; 4138 TCP_STAT_PUTREF(); 4139 } 4140 return (1); 4141 } 4142 4143 s = splsoftnet(); 4144 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4145 splx(s); 4146 if (sc == NULL) { 4147 if (ipopts) 4148 (void) m_free(ipopts); 4149 return (0); 4150 } 4151 4152 /* 4153 * Fill in the cache, and put the necessary IP and TCP 4154 * options into the reply. 4155 */ 4156 memset(sc, 0, sizeof(struct syn_cache)); 4157 callout_init(&sc->sc_timer, CALLOUT_MPSAFE); 4158 bcopy(src, &sc->sc_src, src->sa_len); 4159 bcopy(dst, &sc->sc_dst, dst->sa_len); 4160 sc->sc_flags = 0; 4161 sc->sc_ipopts = ipopts; 4162 sc->sc_irs = th->th_seq; 4163 switch (src->sa_family) { 4164 #ifdef INET 4165 case AF_INET: 4166 { 4167 struct sockaddr_in *srcin = (void *) src; 4168 struct sockaddr_in *dstin = (void *) dst; 4169 4170 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4171 &srcin->sin_addr, dstin->sin_port, 4172 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4173 break; 4174 } 4175 #endif /* INET */ 4176 #ifdef INET6 4177 case AF_INET6: 4178 { 4179 struct sockaddr_in6 *srcin6 = (void *) src; 4180 struct sockaddr_in6 *dstin6 = (void *) dst; 4181 4182 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4183 &srcin6->sin6_addr, dstin6->sin6_port, 4184 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4185 break; 4186 } 4187 #endif /* INET6 */ 4188 } 4189 sc->sc_peermaxseg = oi->maxseg; 4190 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4191 m->m_pkthdr.rcvif : NULL, 4192 sc->sc_src.sa.sa_family); 4193 sc->sc_win = win; 4194 sc->sc_timebase = tcp_now - 1; /* see tcp_newtcpcb() */ 4195 sc->sc_timestamp = tb.ts_recent; 4196 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4197 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4198 sc->sc_flags |= SCF_TIMESTAMP; 4199 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4200 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4201 sc->sc_requested_s_scale = tb.requested_s_scale; 4202 sc->sc_request_r_scale = 0; 4203 /* 4204 * Pick the smallest possible scaling factor that 4205 * will still allow us to scale up to sb_max. 4206 * 4207 * We do this because there are broken firewalls that 4208 * will corrupt the window scale option, leading to 4209 * the other endpoint believing that our advertised 4210 * window is unscaled. At scale factors larger than 4211 * 5 the unscaled window will drop below 1500 bytes, 4212 * leading to serious problems when traversing these 4213 * broken firewalls. 4214 * 4215 * With the default sbmax of 256K, a scale factor 4216 * of 3 will be chosen by this algorithm. Those who 4217 * choose a larger sbmax should watch out 4218 * for the compatiblity problems mentioned above. 4219 * 4220 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4221 * or <SYN,ACK>) segment itself is never scaled. 4222 */ 4223 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4224 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4225 sc->sc_request_r_scale++; 4226 } else { 4227 sc->sc_requested_s_scale = 15; 4228 sc->sc_request_r_scale = 15; 4229 } 4230 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4231 sc->sc_flags |= SCF_SACK_PERMIT; 4232 4233 /* 4234 * ECN setup packet recieved. 4235 */ 4236 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4237 sc->sc_flags |= SCF_ECN_PERMIT; 4238 4239 #ifdef TCP_SIGNATURE 4240 if (tb.t_flags & TF_SIGNATURE) 4241 sc->sc_flags |= SCF_SIGNATURE; 4242 #endif 4243 sc->sc_tp = tp; 4244 if (syn_cache_respond(sc, m) == 0) { 4245 uint64_t *tcps = TCP_STAT_GETREF(); 4246 tcps[TCP_STAT_SNDACKS]++; 4247 tcps[TCP_STAT_SNDTOTAL]++; 4248 TCP_STAT_PUTREF(); 4249 syn_cache_insert(sc, tp); 4250 } else { 4251 s = splsoftnet(); 4252 /* 4253 * syn_cache_put() will try to schedule the timer, so 4254 * we need to initialize it 4255 */ 4256 SYN_CACHE_TIMER_ARM(sc); 4257 syn_cache_put(sc); 4258 splx(s); 4259 TCP_STATINC(TCP_STAT_SC_DROPPED); 4260 } 4261 return (1); 4262 } 4263 4264 int 4265 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4266 { 4267 #ifdef INET6 4268 struct rtentry *rt; 4269 #endif 4270 struct route *ro; 4271 u_int8_t *optp; 4272 int optlen, error; 4273 u_int16_t tlen; 4274 struct ip *ip = NULL; 4275 #ifdef INET6 4276 struct ip6_hdr *ip6 = NULL; 4277 #endif 4278 struct tcpcb *tp = NULL; 4279 struct tcphdr *th; 4280 u_int hlen; 4281 struct socket *so; 4282 4283 ro = &sc->sc_route; 4284 switch (sc->sc_src.sa.sa_family) { 4285 case AF_INET: 4286 hlen = sizeof(struct ip); 4287 break; 4288 #ifdef INET6 4289 case AF_INET6: 4290 hlen = sizeof(struct ip6_hdr); 4291 break; 4292 #endif 4293 default: 4294 if (m) 4295 m_freem(m); 4296 return (EAFNOSUPPORT); 4297 } 4298 4299 /* Compute the size of the TCP options. */ 4300 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4301 ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) + 4302 #ifdef TCP_SIGNATURE 4303 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4304 #endif 4305 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4306 4307 tlen = hlen + sizeof(struct tcphdr) + optlen; 4308 4309 /* 4310 * Create the IP+TCP header from scratch. 4311 */ 4312 if (m) 4313 m_freem(m); 4314 #ifdef DIAGNOSTIC 4315 if (max_linkhdr + tlen > MCLBYTES) 4316 return (ENOBUFS); 4317 #endif 4318 MGETHDR(m, M_DONTWAIT, MT_DATA); 4319 if (m && (max_linkhdr + tlen) > MHLEN) { 4320 MCLGET(m, M_DONTWAIT); 4321 if ((m->m_flags & M_EXT) == 0) { 4322 m_freem(m); 4323 m = NULL; 4324 } 4325 } 4326 if (m == NULL) 4327 return (ENOBUFS); 4328 MCLAIM(m, &tcp_tx_mowner); 4329 4330 /* Fixup the mbuf. */ 4331 m->m_data += max_linkhdr; 4332 m->m_len = m->m_pkthdr.len = tlen; 4333 if (sc->sc_tp) { 4334 tp = sc->sc_tp; 4335 if (tp->t_inpcb) 4336 so = tp->t_inpcb->inp_socket; 4337 #ifdef INET6 4338 else if (tp->t_in6pcb) 4339 so = tp->t_in6pcb->in6p_socket; 4340 #endif 4341 else 4342 so = NULL; 4343 } else 4344 so = NULL; 4345 m->m_pkthdr.rcvif = NULL; 4346 memset(mtod(m, u_char *), 0, tlen); 4347 4348 switch (sc->sc_src.sa.sa_family) { 4349 case AF_INET: 4350 ip = mtod(m, struct ip *); 4351 ip->ip_v = 4; 4352 ip->ip_dst = sc->sc_src.sin.sin_addr; 4353 ip->ip_src = sc->sc_dst.sin.sin_addr; 4354 ip->ip_p = IPPROTO_TCP; 4355 th = (struct tcphdr *)(ip + 1); 4356 th->th_dport = sc->sc_src.sin.sin_port; 4357 th->th_sport = sc->sc_dst.sin.sin_port; 4358 break; 4359 #ifdef INET6 4360 case AF_INET6: 4361 ip6 = mtod(m, struct ip6_hdr *); 4362 ip6->ip6_vfc = IPV6_VERSION; 4363 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4364 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4365 ip6->ip6_nxt = IPPROTO_TCP; 4366 /* ip6_plen will be updated in ip6_output() */ 4367 th = (struct tcphdr *)(ip6 + 1); 4368 th->th_dport = sc->sc_src.sin6.sin6_port; 4369 th->th_sport = sc->sc_dst.sin6.sin6_port; 4370 break; 4371 #endif 4372 default: 4373 th = NULL; 4374 } 4375 4376 th->th_seq = htonl(sc->sc_iss); 4377 th->th_ack = htonl(sc->sc_irs + 1); 4378 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4379 th->th_flags = TH_SYN|TH_ACK; 4380 th->th_win = htons(sc->sc_win); 4381 /* th_sum already 0 */ 4382 /* th_urp already 0 */ 4383 4384 /* Tack on the TCP options. */ 4385 optp = (u_int8_t *)(th + 1); 4386 *optp++ = TCPOPT_MAXSEG; 4387 *optp++ = 4; 4388 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4389 *optp++ = sc->sc_ourmaxseg & 0xff; 4390 4391 if (sc->sc_request_r_scale != 15) { 4392 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4393 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4394 sc->sc_request_r_scale); 4395 optp += 4; 4396 } 4397 4398 if (sc->sc_flags & SCF_TIMESTAMP) { 4399 u_int32_t *lp = (u_int32_t *)(optp); 4400 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4401 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4402 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4403 *lp = htonl(sc->sc_timestamp); 4404 optp += TCPOLEN_TSTAMP_APPA; 4405 } 4406 4407 if (sc->sc_flags & SCF_SACK_PERMIT) { 4408 u_int8_t *p = optp; 4409 4410 /* Let the peer know that we will SACK. */ 4411 p[0] = TCPOPT_SACK_PERMITTED; 4412 p[1] = 2; 4413 p[2] = TCPOPT_NOP; 4414 p[3] = TCPOPT_NOP; 4415 optp += 4; 4416 } 4417 4418 /* 4419 * Send ECN SYN-ACK setup packet. 4420 * Routes can be asymetric, so, even if we receive a packet 4421 * with ECE and CWR set, we must not assume no one will block 4422 * the ECE packet we are about to send. 4423 */ 4424 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4425 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4426 th->th_flags |= TH_ECE; 4427 TCP_STATINC(TCP_STAT_ECN_SHS); 4428 4429 /* 4430 * draft-ietf-tcpm-ecnsyn-00.txt 4431 * 4432 * "[...] a TCP node MAY respond to an ECN-setup 4433 * SYN packet by setting ECT in the responding 4434 * ECN-setup SYN/ACK packet, indicating to routers 4435 * that the SYN/ACK packet is ECN-Capable. 4436 * This allows a congested router along the path 4437 * to mark the packet instead of dropping the 4438 * packet as an indication of congestion." 4439 * 4440 * "[...] There can be a great benefit in setting 4441 * an ECN-capable codepoint in SYN/ACK packets [...] 4442 * Congestion is most likely to occur in 4443 * the server-to-client direction. As a result, 4444 * setting an ECN-capable codepoint in SYN/ACK 4445 * packets can reduce the occurence of three-second 4446 * retransmit timeouts resulting from the drop 4447 * of SYN/ACK packets." 4448 * 4449 * Page 4 and 6, January 2006. 4450 */ 4451 4452 switch (sc->sc_src.sa.sa_family) { 4453 #ifdef INET 4454 case AF_INET: 4455 ip->ip_tos |= IPTOS_ECN_ECT0; 4456 break; 4457 #endif 4458 #ifdef INET6 4459 case AF_INET6: 4460 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4461 break; 4462 #endif 4463 } 4464 TCP_STATINC(TCP_STAT_ECN_ECT); 4465 } 4466 4467 #ifdef TCP_SIGNATURE 4468 if (sc->sc_flags & SCF_SIGNATURE) { 4469 struct secasvar *sav; 4470 u_int8_t *sigp; 4471 4472 sav = tcp_signature_getsav(m, th); 4473 4474 if (sav == NULL) { 4475 if (m) 4476 m_freem(m); 4477 return (EPERM); 4478 } 4479 4480 *optp++ = TCPOPT_SIGNATURE; 4481 *optp++ = TCPOLEN_SIGNATURE; 4482 sigp = optp; 4483 memset(optp, 0, TCP_SIGLEN); 4484 optp += TCP_SIGLEN; 4485 *optp++ = TCPOPT_NOP; 4486 *optp++ = TCPOPT_EOL; 4487 4488 (void)tcp_signature(m, th, hlen, sav, sigp); 4489 4490 key_sa_recordxfer(sav, m); 4491 #ifdef FAST_IPSEC 4492 KEY_FREESAV(&sav); 4493 #else 4494 key_freesav(sav); 4495 #endif 4496 } 4497 #endif 4498 4499 /* Compute the packet's checksum. */ 4500 switch (sc->sc_src.sa.sa_family) { 4501 case AF_INET: 4502 ip->ip_len = htons(tlen - hlen); 4503 th->th_sum = 0; 4504 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4505 break; 4506 #ifdef INET6 4507 case AF_INET6: 4508 ip6->ip6_plen = htons(tlen - hlen); 4509 th->th_sum = 0; 4510 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4511 break; 4512 #endif 4513 } 4514 4515 /* 4516 * Fill in some straggling IP bits. Note the stack expects 4517 * ip_len to be in host order, for convenience. 4518 */ 4519 switch (sc->sc_src.sa.sa_family) { 4520 #ifdef INET 4521 case AF_INET: 4522 ip->ip_len = htons(tlen); 4523 ip->ip_ttl = ip_defttl; 4524 /* XXX tos? */ 4525 break; 4526 #endif 4527 #ifdef INET6 4528 case AF_INET6: 4529 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4530 ip6->ip6_vfc |= IPV6_VERSION; 4531 ip6->ip6_plen = htons(tlen - hlen); 4532 /* ip6_hlim will be initialized afterwards */ 4533 /* XXX flowlabel? */ 4534 break; 4535 #endif 4536 } 4537 4538 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4539 tp = sc->sc_tp; 4540 4541 switch (sc->sc_src.sa.sa_family) { 4542 #ifdef INET 4543 case AF_INET: 4544 error = ip_output(m, sc->sc_ipopts, ro, 4545 (ip_mtudisc ? IP_MTUDISC : 0), 4546 (struct ip_moptions *)NULL, so); 4547 break; 4548 #endif 4549 #ifdef INET6 4550 case AF_INET6: 4551 ip6->ip6_hlim = in6_selecthlim(NULL, 4552 (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp 4553 : NULL); 4554 4555 error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, so, NULL); 4556 break; 4557 #endif 4558 default: 4559 error = EAFNOSUPPORT; 4560 break; 4561 } 4562 return (error); 4563 } 4564