1 /* $NetBSD: tcp_input.c,v 1.210 2004/12/15 04:25:19 thorpej 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 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 * 80 * Redistribution and use in source and binary forms, with or without 81 * modification, are permitted provided that the following conditions 82 * are met: 83 * 1. Redistributions of source code must retain the above copyright 84 * notice, this list of conditions and the following disclaimer. 85 * 2. Redistributions in binary form must reproduce the above copyright 86 * notice, this list of conditions and the following disclaimer in the 87 * documentation and/or other materials provided with the distribution. 88 * 3. All advertising materials mentioning features or use of this software 89 * must display the following acknowledgement: 90 * This product includes software developed by the NetBSD 91 * Foundation, Inc. and its contributors. 92 * 4. Neither the name of The NetBSD Foundation nor the names of its 93 * contributors may be used to endorse or promote products derived 94 * from this software without specific prior written permission. 95 * 96 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 97 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 98 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 99 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 100 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 101 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 102 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 103 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 104 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 105 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 106 * POSSIBILITY OF SUCH DAMAGE. 107 */ 108 109 /* 110 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 111 * The Regents of the University of California. All rights reserved. 112 * 113 * Redistribution and use in source and binary forms, with or without 114 * modification, are permitted provided that the following conditions 115 * are met: 116 * 1. Redistributions of source code must retain the above copyright 117 * notice, this list of conditions and the following disclaimer. 118 * 2. Redistributions in binary form must reproduce the above copyright 119 * notice, this list of conditions and the following disclaimer in the 120 * documentation and/or other materials provided with the distribution. 121 * 3. Neither the name of the University nor the names of its contributors 122 * may be used to endorse or promote products derived from this software 123 * without specific prior written permission. 124 * 125 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 126 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 127 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 128 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 129 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 130 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 131 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 132 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 133 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 134 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 135 * SUCH DAMAGE. 136 * 137 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 138 */ 139 140 /* 141 * TODO list for SYN cache stuff: 142 * 143 * Find room for a "state" field, which is needed to keep a 144 * compressed state for TIME_WAIT TCBs. It's been noted already 145 * that this is fairly important for very high-volume web and 146 * mail servers, which use a large number of short-lived 147 * connections. 148 */ 149 150 #include <sys/cdefs.h> 151 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.210 2004/12/15 04:25:19 thorpej Exp $"); 152 153 #include "opt_inet.h" 154 #include "opt_ipsec.h" 155 #include "opt_inet_csum.h" 156 #include "opt_tcp_debug.h" 157 158 #include <sys/param.h> 159 #include <sys/systm.h> 160 #include <sys/malloc.h> 161 #include <sys/mbuf.h> 162 #include <sys/protosw.h> 163 #include <sys/socket.h> 164 #include <sys/socketvar.h> 165 #include <sys/errno.h> 166 #include <sys/syslog.h> 167 #include <sys/pool.h> 168 #include <sys/domain.h> 169 #include <sys/kernel.h> 170 #ifdef TCP_SIGNATURE 171 #include <sys/md5.h> 172 #endif 173 174 #include <net/if.h> 175 #include <net/route.h> 176 #include <net/if_types.h> 177 178 #include <netinet/in.h> 179 #include <netinet/in_systm.h> 180 #include <netinet/ip.h> 181 #include <netinet/in_pcb.h> 182 #include <netinet/in_var.h> 183 #include <netinet/ip_var.h> 184 185 #ifdef INET6 186 #ifndef INET 187 #include <netinet/in.h> 188 #endif 189 #include <netinet/ip6.h> 190 #include <netinet6/ip6_var.h> 191 #include <netinet6/in6_pcb.h> 192 #include <netinet6/ip6_var.h> 193 #include <netinet6/in6_var.h> 194 #include <netinet/icmp6.h> 195 #include <netinet6/nd6.h> 196 #endif 197 198 #ifndef INET6 199 /* always need ip6.h for IP6_EXTHDR_GET */ 200 #include <netinet/ip6.h> 201 #endif 202 203 #include <netinet/tcp.h> 204 #include <netinet/tcp_fsm.h> 205 #include <netinet/tcp_seq.h> 206 #include <netinet/tcp_timer.h> 207 #include <netinet/tcp_var.h> 208 #include <netinet/tcpip.h> 209 #include <netinet/tcp_debug.h> 210 211 #include <machine/stdarg.h> 212 213 #ifdef IPSEC 214 #include <netinet6/ipsec.h> 215 #include <netkey/key.h> 216 #endif /*IPSEC*/ 217 #ifdef INET6 218 #include "faith.h" 219 #if defined(NFAITH) && NFAITH > 0 220 #include <net/if_faith.h> 221 #endif 222 #endif /* IPSEC */ 223 224 #ifdef FAST_IPSEC 225 #include <netipsec/ipsec.h> 226 #include <netipsec/ipsec_var.h> /* XXX ipsecstat namespace */ 227 #include <netipsec/key.h> 228 #ifdef INET6 229 #include <netipsec/ipsec6.h> 230 #endif 231 #endif /* FAST_IPSEC*/ 232 233 int tcprexmtthresh = 3; 234 int tcp_log_refused; 235 236 static int tcp_rst_ppslim_count = 0; 237 static struct timeval tcp_rst_ppslim_last; 238 static int tcp_ackdrop_ppslim_count = 0; 239 static struct timeval tcp_ackdrop_ppslim_last; 240 241 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 242 243 /* for modulo comparisons of timestamps */ 244 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 245 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 246 247 /* 248 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 249 */ 250 #ifdef INET6 251 #define ND6_HINT(tp) \ 252 do { \ 253 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 && \ 254 tp->t_in6pcb->in6p_route.ro_rt) { \ 255 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \ 256 } \ 257 } while (/*CONSTCOND*/ 0) 258 #else 259 #define ND6_HINT(tp) 260 #endif 261 262 /* 263 * Macro to compute ACK transmission behavior. Delay the ACK unless 264 * we have already delayed an ACK (must send an ACK every two segments). 265 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 266 * option is enabled. 267 */ 268 #define TCP_SETUP_ACK(tp, th) \ 269 do { \ 270 if ((tp)->t_flags & TF_DELACK || \ 271 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 272 tp->t_flags |= TF_ACKNOW; \ 273 else \ 274 TCP_SET_DELACK(tp); \ 275 } while (/*CONSTCOND*/ 0) 276 277 /* 278 * Convert TCP protocol fields to host order for easier processing. 279 */ 280 #define TCP_FIELDS_TO_HOST(th) \ 281 do { \ 282 NTOHL((th)->th_seq); \ 283 NTOHL((th)->th_ack); \ 284 NTOHS((th)->th_win); \ 285 NTOHS((th)->th_urp); \ 286 } while (/*CONSTCOND*/ 0) 287 288 /* 289 * ... and reverse the above. 290 */ 291 #define TCP_FIELDS_TO_NET(th) \ 292 do { \ 293 HTONL((th)->th_seq); \ 294 HTONL((th)->th_ack); \ 295 HTONS((th)->th_win); \ 296 HTONS((th)->th_urp); \ 297 } while (/*CONSTCOND*/ 0) 298 299 #ifdef TCP_CSUM_COUNTERS 300 #include <sys/device.h> 301 302 extern struct evcnt tcp_hwcsum_ok; 303 extern struct evcnt tcp_hwcsum_bad; 304 extern struct evcnt tcp_hwcsum_data; 305 extern struct evcnt tcp_swcsum; 306 307 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 308 309 #else 310 311 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 312 313 #endif /* TCP_CSUM_COUNTERS */ 314 315 #ifdef TCP_REASS_COUNTERS 316 #include <sys/device.h> 317 318 extern struct evcnt tcp_reass_; 319 extern struct evcnt tcp_reass_empty; 320 extern struct evcnt tcp_reass_iteration[8]; 321 extern struct evcnt tcp_reass_prependfirst; 322 extern struct evcnt tcp_reass_prepend; 323 extern struct evcnt tcp_reass_insert; 324 extern struct evcnt tcp_reass_inserttail; 325 extern struct evcnt tcp_reass_append; 326 extern struct evcnt tcp_reass_appendtail; 327 extern struct evcnt tcp_reass_overlaptail; 328 extern struct evcnt tcp_reass_overlapfront; 329 extern struct evcnt tcp_reass_segdup; 330 extern struct evcnt tcp_reass_fragdup; 331 332 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 333 334 #else 335 336 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 337 338 #endif /* TCP_REASS_COUNTERS */ 339 340 #ifdef INET 341 static void tcp4_log_refused __P((const struct ip *, const struct tcphdr *)); 342 #endif 343 #ifdef INET6 344 static void tcp6_log_refused 345 __P((const struct ip6_hdr *, const struct tcphdr *)); 346 #endif 347 348 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 349 350 POOL_INIT(tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", NULL); 351 352 int 353 tcp_reass(tp, th, m, tlen) 354 struct tcpcb *tp; 355 struct tcphdr *th; 356 struct mbuf *m; 357 int *tlen; 358 { 359 struct ipqent *p, *q, *nq, *tiqe = NULL; 360 struct socket *so = NULL; 361 int pkt_flags; 362 tcp_seq pkt_seq; 363 unsigned pkt_len; 364 u_long rcvpartdupbyte = 0; 365 u_long rcvoobyte; 366 #ifdef TCP_REASS_COUNTERS 367 u_int count = 0; 368 #endif 369 370 if (tp->t_inpcb) 371 so = tp->t_inpcb->inp_socket; 372 #ifdef INET6 373 else if (tp->t_in6pcb) 374 so = tp->t_in6pcb->in6p_socket; 375 #endif 376 377 TCP_REASS_LOCK_CHECK(tp); 378 379 /* 380 * Call with th==0 after become established to 381 * force pre-ESTABLISHED data up to user socket. 382 */ 383 if (th == 0) 384 goto present; 385 386 rcvoobyte = *tlen; 387 /* 388 * Copy these to local variables because the tcpiphdr 389 * gets munged while we are collapsing mbufs. 390 */ 391 pkt_seq = th->th_seq; 392 pkt_len = *tlen; 393 pkt_flags = th->th_flags; 394 395 TCP_REASS_COUNTER_INCR(&tcp_reass_); 396 397 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 398 /* 399 * When we miss a packet, the vast majority of time we get 400 * packets that follow it in order. So optimize for that. 401 */ 402 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 403 p->ipqe_len += pkt_len; 404 p->ipqe_flags |= pkt_flags; 405 m_cat(p->ipre_mlast, m); 406 TRAVERSE(p->ipre_mlast); 407 m = NULL; 408 tiqe = p; 409 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 410 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 411 goto skip_replacement; 412 } 413 /* 414 * While we're here, if the pkt is completely beyond 415 * anything we have, just insert it at the tail. 416 */ 417 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 418 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 419 goto insert_it; 420 } 421 } 422 423 q = TAILQ_FIRST(&tp->segq); 424 425 if (q != NULL) { 426 /* 427 * If this segment immediately precedes the first out-of-order 428 * block, simply slap the segment in front of it and (mostly) 429 * skip the complicated logic. 430 */ 431 if (pkt_seq + pkt_len == q->ipqe_seq) { 432 q->ipqe_seq = pkt_seq; 433 q->ipqe_len += pkt_len; 434 q->ipqe_flags |= pkt_flags; 435 m_cat(m, q->ipqe_m); 436 q->ipqe_m = m; 437 q->ipre_mlast = m; /* last mbuf may have changed */ 438 TRAVERSE(q->ipre_mlast); 439 tiqe = q; 440 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 441 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 442 goto skip_replacement; 443 } 444 } else { 445 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 446 } 447 448 /* 449 * Find a segment which begins after this one does. 450 */ 451 for (p = NULL; q != NULL; q = nq) { 452 nq = TAILQ_NEXT(q, ipqe_q); 453 #ifdef TCP_REASS_COUNTERS 454 count++; 455 #endif 456 /* 457 * If the received segment is just right after this 458 * fragment, merge the two together and then check 459 * for further overlaps. 460 */ 461 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 462 #ifdef TCPREASS_DEBUG 463 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 464 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 465 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 466 #endif 467 pkt_len += q->ipqe_len; 468 pkt_flags |= q->ipqe_flags; 469 pkt_seq = q->ipqe_seq; 470 m_cat(q->ipre_mlast, m); 471 TRAVERSE(q->ipre_mlast); 472 m = q->ipqe_m; 473 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 474 goto free_ipqe; 475 } 476 /* 477 * If the received segment is completely past this 478 * fragment, we need to go the next fragment. 479 */ 480 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 481 p = q; 482 continue; 483 } 484 /* 485 * If the fragment is past the received segment, 486 * it (or any following) can't be concatenated. 487 */ 488 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 489 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 490 break; 491 } 492 493 /* 494 * We've received all the data in this segment before. 495 * mark it as a duplicate and return. 496 */ 497 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 498 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 499 tcpstat.tcps_rcvduppack++; 500 tcpstat.tcps_rcvdupbyte += pkt_len; 501 m_freem(m); 502 if (tiqe != NULL) 503 pool_put(&tcpipqent_pool, tiqe); 504 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 505 return (0); 506 } 507 /* 508 * Received segment completely overlaps this fragment 509 * so we drop the fragment (this keeps the temporal 510 * ordering of segments correct). 511 */ 512 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 513 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 514 rcvpartdupbyte += q->ipqe_len; 515 m_freem(q->ipqe_m); 516 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 517 goto free_ipqe; 518 } 519 /* 520 * RX'ed segment extends past the end of the 521 * fragment. Drop the overlapping bytes. Then 522 * merge the fragment and segment then treat as 523 * a longer received packet. 524 */ 525 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 526 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 527 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 528 #ifdef TCPREASS_DEBUG 529 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 530 tp, overlap, 531 pkt_seq, pkt_seq + pkt_len, pkt_len); 532 #endif 533 m_adj(m, overlap); 534 rcvpartdupbyte += overlap; 535 m_cat(q->ipre_mlast, m); 536 TRAVERSE(q->ipre_mlast); 537 m = q->ipqe_m; 538 pkt_seq = q->ipqe_seq; 539 pkt_len += q->ipqe_len - overlap; 540 rcvoobyte -= overlap; 541 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 542 goto free_ipqe; 543 } 544 /* 545 * RX'ed segment extends past the front of the 546 * fragment. Drop the overlapping bytes on the 547 * received packet. The packet will then be 548 * contatentated with this fragment a bit later. 549 */ 550 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 551 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 552 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 553 #ifdef TCPREASS_DEBUG 554 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 555 tp, overlap, 556 pkt_seq, pkt_seq + pkt_len, pkt_len); 557 #endif 558 m_adj(m, -overlap); 559 pkt_len -= overlap; 560 rcvpartdupbyte += overlap; 561 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 562 rcvoobyte -= overlap; 563 } 564 /* 565 * If the received segment immediates precedes this 566 * fragment then tack the fragment onto this segment 567 * and reinsert the data. 568 */ 569 if (q->ipqe_seq == pkt_seq + pkt_len) { 570 #ifdef TCPREASS_DEBUG 571 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 572 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 573 pkt_seq, pkt_seq + pkt_len, pkt_len); 574 #endif 575 pkt_len += q->ipqe_len; 576 pkt_flags |= q->ipqe_flags; 577 m_cat(m, q->ipqe_m); 578 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 579 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 580 if (tiqe == NULL) 581 tiqe = q; 582 else 583 pool_put(&tcpipqent_pool, q); 584 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 585 break; 586 } 587 /* 588 * If the fragment is before the segment, remember it. 589 * When this loop is terminated, p will contain the 590 * pointer to fragment that is right before the received 591 * segment. 592 */ 593 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 594 p = q; 595 596 continue; 597 598 /* 599 * This is a common operation. It also will allow 600 * to save doing a malloc/free in most instances. 601 */ 602 free_ipqe: 603 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 604 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 605 if (tiqe == NULL) 606 tiqe = q; 607 else 608 pool_put(&tcpipqent_pool, q); 609 } 610 611 #ifdef TCP_REASS_COUNTERS 612 if (count > 7) 613 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 614 else if (count > 0) 615 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 616 #endif 617 618 insert_it: 619 620 /* 621 * Allocate a new queue entry since the received segment did not 622 * collapse onto any other out-of-order block; thus we are allocating 623 * a new block. If it had collapsed, tiqe would not be NULL and 624 * we would be reusing it. 625 * XXX If we can't, just drop the packet. XXX 626 */ 627 if (tiqe == NULL) { 628 tiqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 629 if (tiqe == NULL) { 630 tcpstat.tcps_rcvmemdrop++; 631 m_freem(m); 632 return (0); 633 } 634 } 635 636 /* 637 * Update the counters. 638 */ 639 tcpstat.tcps_rcvoopack++; 640 tcpstat.tcps_rcvoobyte += rcvoobyte; 641 if (rcvpartdupbyte) { 642 tcpstat.tcps_rcvpartduppack++; 643 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 644 } 645 646 /* 647 * Insert the new fragment queue entry into both queues. 648 */ 649 tiqe->ipqe_m = m; 650 tiqe->ipre_mlast = m; 651 tiqe->ipqe_seq = pkt_seq; 652 tiqe->ipqe_len = pkt_len; 653 tiqe->ipqe_flags = pkt_flags; 654 if (p == NULL) { 655 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 656 #ifdef TCPREASS_DEBUG 657 if (tiqe->ipqe_seq != tp->rcv_nxt) 658 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 659 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 660 #endif 661 } else { 662 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 663 #ifdef TCPREASS_DEBUG 664 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 665 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 666 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 667 #endif 668 } 669 670 skip_replacement: 671 672 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 673 674 present: 675 /* 676 * Present data to user, advancing rcv_nxt through 677 * completed sequence space. 678 */ 679 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 680 return (0); 681 q = TAILQ_FIRST(&tp->segq); 682 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 683 return (0); 684 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 685 return (0); 686 687 tp->rcv_nxt += q->ipqe_len; 688 pkt_flags = q->ipqe_flags & TH_FIN; 689 ND6_HINT(tp); 690 691 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 692 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 693 if (so->so_state & SS_CANTRCVMORE) 694 m_freem(q->ipqe_m); 695 else 696 sbappendstream(&so->so_rcv, q->ipqe_m); 697 pool_put(&tcpipqent_pool, q); 698 sorwakeup(so); 699 return (pkt_flags); 700 } 701 702 #ifdef INET6 703 int 704 tcp6_input(mp, offp, proto) 705 struct mbuf **mp; 706 int *offp, proto; 707 { 708 struct mbuf *m = *mp; 709 710 /* 711 * draft-itojun-ipv6-tcp-to-anycast 712 * better place to put this in? 713 */ 714 if (m->m_flags & M_ANYCAST6) { 715 struct ip6_hdr *ip6; 716 if (m->m_len < sizeof(struct ip6_hdr)) { 717 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 718 tcpstat.tcps_rcvshort++; 719 return IPPROTO_DONE; 720 } 721 } 722 ip6 = mtod(m, struct ip6_hdr *); 723 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 724 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 725 return IPPROTO_DONE; 726 } 727 728 tcp_input(m, *offp, proto); 729 return IPPROTO_DONE; 730 } 731 #endif 732 733 #ifdef INET 734 static void 735 tcp4_log_refused(ip, th) 736 const struct ip *ip; 737 const struct tcphdr *th; 738 { 739 char src[4*sizeof "123"]; 740 char dst[4*sizeof "123"]; 741 742 if (ip) { 743 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 744 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 745 } 746 else { 747 strlcpy(src, "(unknown)", sizeof(src)); 748 strlcpy(dst, "(unknown)", sizeof(dst)); 749 } 750 log(LOG_INFO, 751 "Connection attempt to TCP %s:%d from %s:%d\n", 752 dst, ntohs(th->th_dport), 753 src, ntohs(th->th_sport)); 754 } 755 #endif 756 757 #ifdef INET6 758 static void 759 tcp6_log_refused(ip6, th) 760 const struct ip6_hdr *ip6; 761 const struct tcphdr *th; 762 { 763 char src[INET6_ADDRSTRLEN]; 764 char dst[INET6_ADDRSTRLEN]; 765 766 if (ip6) { 767 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 768 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 769 } 770 else { 771 strlcpy(src, "(unknown v6)", sizeof(src)); 772 strlcpy(dst, "(unknown v6)", sizeof(dst)); 773 } 774 log(LOG_INFO, 775 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 776 dst, ntohs(th->th_dport), 777 src, ntohs(th->th_sport)); 778 } 779 #endif 780 781 /* 782 * TCP input routine, follows pages 65-76 of the 783 * protocol specification dated September, 1981 very closely. 784 */ 785 void 786 tcp_input(struct mbuf *m, ...) 787 { 788 struct tcphdr *th; 789 struct ip *ip; 790 struct inpcb *inp; 791 #ifdef INET6 792 struct ip6_hdr *ip6; 793 struct in6pcb *in6p; 794 #endif 795 u_int8_t *optp = NULL; 796 int optlen = 0; 797 int len, tlen, toff, hdroptlen = 0; 798 struct tcpcb *tp = 0; 799 int tiflags; 800 struct socket *so = NULL; 801 int todrop, acked, ourfinisacked, needoutput = 0; 802 #ifdef TCP_DEBUG 803 short ostate = 0; 804 #endif 805 int iss = 0; 806 u_long tiwin; 807 struct tcp_opt_info opti; 808 int off, iphlen; 809 va_list ap; 810 int af; /* af on the wire */ 811 struct mbuf *tcp_saveti = NULL; 812 813 MCLAIM(m, &tcp_rx_mowner); 814 va_start(ap, m); 815 toff = va_arg(ap, int); 816 (void)va_arg(ap, int); /* ignore value, advance ap */ 817 va_end(ap); 818 819 tcpstat.tcps_rcvtotal++; 820 821 bzero(&opti, sizeof(opti)); 822 opti.ts_present = 0; 823 opti.maxseg = 0; 824 825 /* 826 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 827 * 828 * TCP is, by definition, unicast, so we reject all 829 * multicast outright. 830 * 831 * Note, there are additional src/dst address checks in 832 * the AF-specific code below. 833 */ 834 if (m->m_flags & (M_BCAST|M_MCAST)) { 835 /* XXX stat */ 836 goto drop; 837 } 838 #ifdef INET6 839 if (m->m_flags & M_ANYCAST6) { 840 /* XXX stat */ 841 goto drop; 842 } 843 #endif 844 845 /* 846 * Get IP and TCP header. 847 * Note: IP leaves IP header in first mbuf. 848 */ 849 ip = mtod(m, struct ip *); 850 #ifdef INET6 851 ip6 = NULL; 852 #endif 853 switch (ip->ip_v) { 854 #ifdef INET 855 case 4: 856 af = AF_INET; 857 iphlen = sizeof(struct ip); 858 ip = mtod(m, struct ip *); 859 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 860 sizeof(struct tcphdr)); 861 if (th == NULL) { 862 tcpstat.tcps_rcvshort++; 863 return; 864 } 865 /* We do the checksum after PCB lookup... */ 866 len = ntohs(ip->ip_len); 867 tlen = len - toff; 868 break; 869 #endif 870 #ifdef INET6 871 case 6: 872 ip = NULL; 873 iphlen = sizeof(struct ip6_hdr); 874 af = AF_INET6; 875 ip6 = mtod(m, struct ip6_hdr *); 876 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 877 sizeof(struct tcphdr)); 878 if (th == NULL) { 879 tcpstat.tcps_rcvshort++; 880 return; 881 } 882 883 /* Be proactive about malicious use of IPv4 mapped address */ 884 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 885 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 886 /* XXX stat */ 887 goto drop; 888 } 889 890 /* 891 * Be proactive about unspecified IPv6 address in source. 892 * As we use all-zero to indicate unbounded/unconnected pcb, 893 * unspecified IPv6 address can be used to confuse us. 894 * 895 * Note that packets with unspecified IPv6 destination is 896 * already dropped in ip6_input. 897 */ 898 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 899 /* XXX stat */ 900 goto drop; 901 } 902 903 /* 904 * Make sure destination address is not multicast. 905 * Source address checked in ip6_input(). 906 */ 907 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 908 /* XXX stat */ 909 goto drop; 910 } 911 912 /* We do the checksum after PCB lookup... */ 913 len = m->m_pkthdr.len; 914 tlen = len - toff; 915 break; 916 #endif 917 default: 918 m_freem(m); 919 return; 920 } 921 922 KASSERT(TCP_HDR_ALIGNED_P(th)); 923 924 /* 925 * Check that TCP offset makes sense, 926 * pull out TCP options and adjust length. XXX 927 */ 928 off = th->th_off << 2; 929 if (off < sizeof (struct tcphdr) || off > tlen) { 930 tcpstat.tcps_rcvbadoff++; 931 goto drop; 932 } 933 tlen -= off; 934 935 /* 936 * tcp_input() has been modified to use tlen to mean the TCP data 937 * length throughout the function. Other functions can use 938 * m->m_pkthdr.len as the basis for calculating the TCP data length. 939 * rja 940 */ 941 942 if (off > sizeof (struct tcphdr)) { 943 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 944 if (th == NULL) { 945 tcpstat.tcps_rcvshort++; 946 return; 947 } 948 /* 949 * NOTE: ip/ip6 will not be affected by m_pulldown() 950 * (as they're before toff) and we don't need to update those. 951 */ 952 KASSERT(TCP_HDR_ALIGNED_P(th)); 953 optlen = off - sizeof (struct tcphdr); 954 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 955 /* 956 * Do quick retrieval of timestamp options ("options 957 * prediction?"). If timestamp is the only option and it's 958 * formatted as recommended in RFC 1323 appendix A, we 959 * quickly get the values now and not bother calling 960 * tcp_dooptions(), etc. 961 */ 962 if ((optlen == TCPOLEN_TSTAMP_APPA || 963 (optlen > TCPOLEN_TSTAMP_APPA && 964 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 965 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 966 (th->th_flags & TH_SYN) == 0) { 967 opti.ts_present = 1; 968 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 969 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 970 optp = NULL; /* we've parsed the options */ 971 } 972 } 973 tiflags = th->th_flags; 974 975 /* 976 * Locate pcb for segment. 977 */ 978 findpcb: 979 inp = NULL; 980 #ifdef INET6 981 in6p = NULL; 982 #endif 983 switch (af) { 984 #ifdef INET 985 case AF_INET: 986 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 987 ip->ip_dst, th->th_dport); 988 if (inp == 0) { 989 ++tcpstat.tcps_pcbhashmiss; 990 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 991 } 992 #ifdef INET6 993 if (inp == 0) { 994 struct in6_addr s, d; 995 996 /* mapped addr case */ 997 bzero(&s, sizeof(s)); 998 s.s6_addr16[5] = htons(0xffff); 999 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1000 bzero(&d, sizeof(d)); 1001 d.s6_addr16[5] = htons(0xffff); 1002 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1003 in6p = in6_pcblookup_connect(&tcbtable, &s, 1004 th->th_sport, &d, th->th_dport, 0); 1005 if (in6p == 0) { 1006 ++tcpstat.tcps_pcbhashmiss; 1007 in6p = in6_pcblookup_bind(&tcbtable, &d, 1008 th->th_dport, 0); 1009 } 1010 } 1011 #endif 1012 #ifndef INET6 1013 if (inp == 0) 1014 #else 1015 if (inp == 0 && in6p == 0) 1016 #endif 1017 { 1018 ++tcpstat.tcps_noport; 1019 if (tcp_log_refused && 1020 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1021 tcp4_log_refused(ip, th); 1022 } 1023 TCP_FIELDS_TO_HOST(th); 1024 goto dropwithreset_ratelim; 1025 } 1026 #if defined(IPSEC) || defined(FAST_IPSEC) 1027 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1028 ipsec4_in_reject(m, inp)) { 1029 ipsecstat.in_polvio++; 1030 goto drop; 1031 } 1032 #ifdef INET6 1033 else if (in6p && 1034 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1035 ipsec4_in_reject_so(m, in6p->in6p_socket)) { 1036 ipsecstat.in_polvio++; 1037 goto drop; 1038 } 1039 #endif 1040 #endif /*IPSEC*/ 1041 break; 1042 #endif /*INET*/ 1043 #ifdef INET6 1044 case AF_INET6: 1045 { 1046 int faith; 1047 1048 #if defined(NFAITH) && NFAITH > 0 1049 faith = faithprefix(&ip6->ip6_dst); 1050 #else 1051 faith = 0; 1052 #endif 1053 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1054 th->th_sport, &ip6->ip6_dst, th->th_dport, faith); 1055 if (in6p == NULL) { 1056 ++tcpstat.tcps_pcbhashmiss; 1057 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1058 th->th_dport, faith); 1059 } 1060 if (in6p == NULL) { 1061 ++tcpstat.tcps_noport; 1062 if (tcp_log_refused && 1063 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1064 tcp6_log_refused(ip6, th); 1065 } 1066 TCP_FIELDS_TO_HOST(th); 1067 goto dropwithreset_ratelim; 1068 } 1069 #if defined(IPSEC) || defined(FAST_IPSEC) 1070 if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1071 ipsec6_in_reject(m, in6p)) { 1072 ipsec6stat.in_polvio++; 1073 goto drop; 1074 } 1075 #endif /*IPSEC*/ 1076 break; 1077 } 1078 #endif 1079 } 1080 1081 /* 1082 * If the state is CLOSED (i.e., TCB does not exist) then 1083 * all data in the incoming segment is discarded. 1084 * If the TCB exists but is in CLOSED state, it is embryonic, 1085 * but should either do a listen or a connect soon. 1086 */ 1087 tp = NULL; 1088 so = NULL; 1089 if (inp) { 1090 tp = intotcpcb(inp); 1091 so = inp->inp_socket; 1092 } 1093 #ifdef INET6 1094 else if (in6p) { 1095 tp = in6totcpcb(in6p); 1096 so = in6p->in6p_socket; 1097 } 1098 #endif 1099 if (tp == 0) { 1100 TCP_FIELDS_TO_HOST(th); 1101 goto dropwithreset_ratelim; 1102 } 1103 if (tp->t_state == TCPS_CLOSED) 1104 goto drop; 1105 1106 /* 1107 * Checksum extended TCP header and data. 1108 */ 1109 switch (af) { 1110 #ifdef INET 1111 case AF_INET: 1112 switch (m->m_pkthdr.csum_flags & 1113 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 1114 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 1115 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 1116 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 1117 goto badcsum; 1118 1119 case M_CSUM_TCPv4|M_CSUM_DATA: { 1120 u_int32_t hw_csum = m->m_pkthdr.csum_data; 1121 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 1122 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 1123 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 1124 ip->ip_dst.s_addr, 1125 htons(hw_csum + tlen + off + IPPROTO_TCP)); 1126 } 1127 if ((hw_csum ^ 0xffff) != 0) 1128 goto badcsum; 1129 break; 1130 } 1131 1132 case M_CSUM_TCPv4: 1133 /* Checksum was okay. */ 1134 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 1135 break; 1136 1137 default: 1138 /* 1139 * Must compute it ourselves. Maybe skip checksum 1140 * on loopback interfaces. 1141 */ 1142 if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & 1143 IFF_LOOPBACK) || 1144 tcp_do_loopback_cksum)) { 1145 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 1146 if (in4_cksum(m, IPPROTO_TCP, toff, 1147 tlen + off) != 0) 1148 goto badcsum; 1149 } 1150 break; 1151 } 1152 break; 1153 #endif /* INET4 */ 1154 1155 #ifdef INET6 1156 case AF_INET6: 1157 if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 1158 goto badcsum; 1159 break; 1160 #endif /* INET6 */ 1161 } 1162 1163 TCP_FIELDS_TO_HOST(th); 1164 1165 /* Unscale the window into a 32-bit value. */ 1166 if ((tiflags & TH_SYN) == 0) 1167 tiwin = th->th_win << tp->snd_scale; 1168 else 1169 tiwin = th->th_win; 1170 1171 #ifdef INET6 1172 /* save packet options if user wanted */ 1173 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1174 if (in6p->in6p_options) { 1175 m_freem(in6p->in6p_options); 1176 in6p->in6p_options = 0; 1177 } 1178 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1179 } 1180 #endif 1181 1182 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1183 union syn_cache_sa src; 1184 union syn_cache_sa dst; 1185 1186 bzero(&src, sizeof(src)); 1187 bzero(&dst, sizeof(dst)); 1188 switch (af) { 1189 #ifdef INET 1190 case AF_INET: 1191 src.sin.sin_len = sizeof(struct sockaddr_in); 1192 src.sin.sin_family = AF_INET; 1193 src.sin.sin_addr = ip->ip_src; 1194 src.sin.sin_port = th->th_sport; 1195 1196 dst.sin.sin_len = sizeof(struct sockaddr_in); 1197 dst.sin.sin_family = AF_INET; 1198 dst.sin.sin_addr = ip->ip_dst; 1199 dst.sin.sin_port = th->th_dport; 1200 break; 1201 #endif 1202 #ifdef INET6 1203 case AF_INET6: 1204 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1205 src.sin6.sin6_family = AF_INET6; 1206 src.sin6.sin6_addr = ip6->ip6_src; 1207 src.sin6.sin6_port = th->th_sport; 1208 1209 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1210 dst.sin6.sin6_family = AF_INET6; 1211 dst.sin6.sin6_addr = ip6->ip6_dst; 1212 dst.sin6.sin6_port = th->th_dport; 1213 break; 1214 #endif /* INET6 */ 1215 default: 1216 goto badsyn; /*sanity*/ 1217 } 1218 1219 if (so->so_options & SO_DEBUG) { 1220 #ifdef TCP_DEBUG 1221 ostate = tp->t_state; 1222 #endif 1223 1224 tcp_saveti = NULL; 1225 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1226 goto nosave; 1227 1228 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1229 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1230 if (!tcp_saveti) 1231 goto nosave; 1232 } else { 1233 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1234 if (!tcp_saveti) 1235 goto nosave; 1236 MCLAIM(m, &tcp_mowner); 1237 tcp_saveti->m_len = iphlen; 1238 m_copydata(m, 0, iphlen, 1239 mtod(tcp_saveti, caddr_t)); 1240 } 1241 1242 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1243 m_freem(tcp_saveti); 1244 tcp_saveti = NULL; 1245 } else { 1246 tcp_saveti->m_len += sizeof(struct tcphdr); 1247 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1248 sizeof(struct tcphdr)); 1249 } 1250 nosave:; 1251 } 1252 if (so->so_options & SO_ACCEPTCONN) { 1253 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1254 if (tiflags & TH_RST) { 1255 syn_cache_reset(&src.sa, &dst.sa, th); 1256 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1257 (TH_ACK|TH_SYN)) { 1258 /* 1259 * Received a SYN,ACK. This should 1260 * never happen while we are in 1261 * LISTEN. Send an RST. 1262 */ 1263 goto badsyn; 1264 } else if (tiflags & TH_ACK) { 1265 so = syn_cache_get(&src.sa, &dst.sa, 1266 th, toff, tlen, so, m); 1267 if (so == NULL) { 1268 /* 1269 * We don't have a SYN for 1270 * this ACK; send an RST. 1271 */ 1272 goto badsyn; 1273 } else if (so == 1274 (struct socket *)(-1)) { 1275 /* 1276 * We were unable to create 1277 * the connection. If the 1278 * 3-way handshake was 1279 * completed, and RST has 1280 * been sent to the peer. 1281 * Since the mbuf might be 1282 * in use for the reply, 1283 * do not free it. 1284 */ 1285 m = NULL; 1286 } else { 1287 /* 1288 * We have created a 1289 * full-blown connection. 1290 */ 1291 tp = NULL; 1292 inp = NULL; 1293 #ifdef INET6 1294 in6p = NULL; 1295 #endif 1296 switch (so->so_proto->pr_domain->dom_family) { 1297 #ifdef INET 1298 case AF_INET: 1299 inp = sotoinpcb(so); 1300 tp = intotcpcb(inp); 1301 break; 1302 #endif 1303 #ifdef INET6 1304 case AF_INET6: 1305 in6p = sotoin6pcb(so); 1306 tp = in6totcpcb(in6p); 1307 break; 1308 #endif 1309 } 1310 if (tp == NULL) 1311 goto badsyn; /*XXX*/ 1312 tiwin <<= tp->snd_scale; 1313 goto after_listen; 1314 } 1315 } else { 1316 /* 1317 * None of RST, SYN or ACK was set. 1318 * This is an invalid packet for a 1319 * TCB in LISTEN state. Send a RST. 1320 */ 1321 goto badsyn; 1322 } 1323 } else { 1324 /* 1325 * Received a SYN. 1326 */ 1327 1328 #ifdef INET6 1329 /* 1330 * If deprecated address is forbidden, we do 1331 * not accept SYN to deprecated interface 1332 * address to prevent any new inbound 1333 * connection from getting established. 1334 * When we do not accept SYN, we send a TCP 1335 * RST, with deprecated source address (instead 1336 * of dropping it). We compromise it as it is 1337 * much better for peer to send a RST, and 1338 * RST will be the final packet for the 1339 * exchange. 1340 * 1341 * If we do not forbid deprecated addresses, we 1342 * accept the SYN packet. RFC2462 does not 1343 * suggest dropping SYN in this case. 1344 * If we decipher RFC2462 5.5.4, it says like 1345 * this: 1346 * 1. use of deprecated addr with existing 1347 * communication is okay - "SHOULD continue 1348 * to be used" 1349 * 2. use of it with new communication: 1350 * (2a) "SHOULD NOT be used if alternate 1351 * address with sufficient scope is 1352 * available" 1353 * (2b) nothing mentioned otherwise. 1354 * Here we fall into (2b) case as we have no 1355 * choice in our source address selection - we 1356 * must obey the peer. 1357 * 1358 * The wording in RFC2462 is confusing, and 1359 * there are multiple description text for 1360 * deprecated address handling - worse, they 1361 * are not exactly the same. I believe 5.5.4 1362 * is the best one, so we follow 5.5.4. 1363 */ 1364 if (af == AF_INET6 && !ip6_use_deprecated) { 1365 struct in6_ifaddr *ia6; 1366 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1367 &ip6->ip6_dst)) && 1368 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1369 tp = NULL; 1370 goto dropwithreset; 1371 } 1372 } 1373 #endif 1374 1375 #ifdef IPSEC 1376 switch (af) { 1377 #ifdef INET 1378 case AF_INET: 1379 if (ipsec4_in_reject_so(m, so)) { 1380 ipsecstat.in_polvio++; 1381 tp = NULL; 1382 goto dropwithreset; 1383 } 1384 break; 1385 #endif 1386 #ifdef INET6 1387 case AF_INET6: 1388 if (ipsec6_in_reject_so(m, so)) { 1389 ipsec6stat.in_polvio++; 1390 tp = NULL; 1391 goto dropwithreset; 1392 } 1393 break; 1394 #endif 1395 } 1396 #endif 1397 1398 /* 1399 * LISTEN socket received a SYN 1400 * from itself? This can't possibly 1401 * be valid; drop the packet. 1402 */ 1403 if (th->th_sport == th->th_dport) { 1404 int i; 1405 1406 switch (af) { 1407 #ifdef INET 1408 case AF_INET: 1409 i = in_hosteq(ip->ip_src, ip->ip_dst); 1410 break; 1411 #endif 1412 #ifdef INET6 1413 case AF_INET6: 1414 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1415 break; 1416 #endif 1417 default: 1418 i = 1; 1419 } 1420 if (i) { 1421 tcpstat.tcps_badsyn++; 1422 goto drop; 1423 } 1424 } 1425 1426 /* 1427 * SYN looks ok; create compressed TCP 1428 * state for it. 1429 */ 1430 if (so->so_qlen <= so->so_qlimit && 1431 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1432 so, m, optp, optlen, &opti)) 1433 m = NULL; 1434 } 1435 goto drop; 1436 } 1437 } 1438 1439 after_listen: 1440 #ifdef DIAGNOSTIC 1441 /* 1442 * Should not happen now that all embryonic connections 1443 * are handled with compressed state. 1444 */ 1445 if (tp->t_state == TCPS_LISTEN) 1446 panic("tcp_input: TCPS_LISTEN"); 1447 #endif 1448 1449 /* 1450 * Segment received on connection. 1451 * Reset idle time and keep-alive timer. 1452 */ 1453 tp->t_rcvtime = tcp_now; 1454 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1455 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1456 1457 /* 1458 * Process options. 1459 */ 1460 #ifdef TCP_SIGNATURE 1461 if (optp || (tp->t_flags & TF_SIGNATURE)) 1462 #else 1463 if (optp) 1464 #endif 1465 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1466 goto drop; 1467 1468 /* 1469 * Header prediction: check for the two common cases 1470 * of a uni-directional data xfer. If the packet has 1471 * no control flags, is in-sequence, the window didn't 1472 * change and we're not retransmitting, it's a 1473 * candidate. If the length is zero and the ack moved 1474 * forward, we're the sender side of the xfer. Just 1475 * free the data acked & wake any higher level process 1476 * that was blocked waiting for space. If the length 1477 * is non-zero and the ack didn't move, we're the 1478 * receiver side. If we're getting packets in-order 1479 * (the reassembly queue is empty), add the data to 1480 * the socket buffer and note that we need a delayed ack. 1481 */ 1482 if (tp->t_state == TCPS_ESTABLISHED && 1483 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1484 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1485 th->th_seq == tp->rcv_nxt && 1486 tiwin && tiwin == tp->snd_wnd && 1487 tp->snd_nxt == tp->snd_max) { 1488 1489 /* 1490 * If last ACK falls within this segment's sequence numbers, 1491 * record the timestamp. 1492 */ 1493 if (opti.ts_present && 1494 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1495 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1496 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1497 tp->ts_recent = opti.ts_val; 1498 } 1499 1500 if (tlen == 0) { 1501 if (SEQ_GT(th->th_ack, tp->snd_una) && 1502 SEQ_LEQ(th->th_ack, tp->snd_max) && 1503 tp->snd_cwnd >= tp->snd_wnd && 1504 tp->t_dupacks < tcprexmtthresh) { 1505 /* 1506 * this is a pure ack for outstanding data. 1507 */ 1508 ++tcpstat.tcps_predack; 1509 if (opti.ts_present && opti.ts_ecr) 1510 tcp_xmit_timer(tp, 1511 TCP_TIMESTAMP(tp) - opti.ts_ecr + 1); 1512 else if (tp->t_rtttime && 1513 SEQ_GT(th->th_ack, tp->t_rtseq)) 1514 tcp_xmit_timer(tp, 1515 tcp_now - tp->t_rtttime); 1516 acked = th->th_ack - tp->snd_una; 1517 tcpstat.tcps_rcvackpack++; 1518 tcpstat.tcps_rcvackbyte += acked; 1519 ND6_HINT(tp); 1520 1521 if (acked > (tp->t_lastoff - tp->t_inoff)) 1522 tp->t_lastm = NULL; 1523 sbdrop(&so->so_snd, acked); 1524 tp->t_lastoff -= acked; 1525 1526 /* 1527 * We want snd_recover to track snd_una to 1528 * avoid sequence wraparound problems for 1529 * very large transfers. 1530 */ 1531 tp->snd_una = tp->snd_recover = th->th_ack; 1532 m_freem(m); 1533 1534 /* 1535 * If all outstanding data are acked, stop 1536 * retransmit timer, otherwise restart timer 1537 * using current (possibly backed-off) value. 1538 * If process is waiting for space, 1539 * wakeup/selwakeup/signal. If data 1540 * are ready to send, let tcp_output 1541 * decide between more output or persist. 1542 */ 1543 if (tp->snd_una == tp->snd_max) 1544 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1545 else if (TCP_TIMER_ISARMED(tp, 1546 TCPT_PERSIST) == 0) 1547 TCP_TIMER_ARM(tp, TCPT_REXMT, 1548 tp->t_rxtcur); 1549 1550 sowwakeup(so); 1551 if (so->so_snd.sb_cc) 1552 (void) tcp_output(tp); 1553 if (tcp_saveti) 1554 m_freem(tcp_saveti); 1555 return; 1556 } 1557 } else if (th->th_ack == tp->snd_una && 1558 TAILQ_FIRST(&tp->segq) == NULL && 1559 tlen <= sbspace(&so->so_rcv)) { 1560 /* 1561 * this is a pure, in-sequence data packet 1562 * with nothing on the reassembly queue and 1563 * we have enough buffer space to take it. 1564 */ 1565 ++tcpstat.tcps_preddat; 1566 tp->rcv_nxt += tlen; 1567 tcpstat.tcps_rcvpack++; 1568 tcpstat.tcps_rcvbyte += tlen; 1569 ND6_HINT(tp); 1570 /* 1571 * Drop TCP, IP headers and TCP options then add data 1572 * to socket buffer. 1573 */ 1574 if (so->so_state & SS_CANTRCVMORE) 1575 m_freem(m); 1576 else { 1577 m_adj(m, toff + off); 1578 sbappendstream(&so->so_rcv, m); 1579 } 1580 sorwakeup(so); 1581 TCP_SETUP_ACK(tp, th); 1582 if (tp->t_flags & TF_ACKNOW) 1583 (void) tcp_output(tp); 1584 if (tcp_saveti) 1585 m_freem(tcp_saveti); 1586 return; 1587 } 1588 } 1589 1590 /* 1591 * Compute mbuf offset to TCP data segment. 1592 */ 1593 hdroptlen = toff + off; 1594 1595 /* 1596 * Calculate amount of space in receive window, 1597 * and then do TCP input processing. 1598 * Receive window is amount of space in rcv queue, 1599 * but not less than advertised window. 1600 */ 1601 { int win; 1602 1603 win = sbspace(&so->so_rcv); 1604 if (win < 0) 1605 win = 0; 1606 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1607 } 1608 1609 switch (tp->t_state) { 1610 case TCPS_LISTEN: 1611 /* 1612 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1613 */ 1614 if (m->m_flags & (M_BCAST|M_MCAST)) 1615 goto drop; 1616 switch (af) { 1617 #ifdef INET6 1618 case AF_INET6: 1619 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1620 goto drop; 1621 break; 1622 #endif /* INET6 */ 1623 case AF_INET: 1624 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1625 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1626 goto drop; 1627 break; 1628 } 1629 break; 1630 1631 /* 1632 * If the state is SYN_SENT: 1633 * if seg contains an ACK, but not for our SYN, drop the input. 1634 * if seg contains a RST, then drop the connection. 1635 * if seg does not contain SYN, then drop it. 1636 * Otherwise this is an acceptable SYN segment 1637 * initialize tp->rcv_nxt and tp->irs 1638 * if seg contains ack then advance tp->snd_una 1639 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1640 * arrange for segment to be acked (eventually) 1641 * continue processing rest of data/controls, beginning with URG 1642 */ 1643 case TCPS_SYN_SENT: 1644 if ((tiflags & TH_ACK) && 1645 (SEQ_LEQ(th->th_ack, tp->iss) || 1646 SEQ_GT(th->th_ack, tp->snd_max))) 1647 goto dropwithreset; 1648 if (tiflags & TH_RST) { 1649 if (tiflags & TH_ACK) 1650 tp = tcp_drop(tp, ECONNREFUSED); 1651 goto drop; 1652 } 1653 if ((tiflags & TH_SYN) == 0) 1654 goto drop; 1655 if (tiflags & TH_ACK) { 1656 tp->snd_una = tp->snd_recover = th->th_ack; 1657 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1658 tp->snd_nxt = tp->snd_una; 1659 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1660 } 1661 tp->irs = th->th_seq; 1662 tcp_rcvseqinit(tp); 1663 tp->t_flags |= TF_ACKNOW; 1664 tcp_mss_from_peer(tp, opti.maxseg); 1665 1666 /* 1667 * Initialize the initial congestion window. If we 1668 * had to retransmit the SYN, we must initialize cwnd 1669 * to 1 segment (i.e. the Loss Window). 1670 */ 1671 if (tp->t_flags & TF_SYN_REXMT) 1672 tp->snd_cwnd = tp->t_peermss; 1673 else { 1674 int ss = tcp_init_win; 1675 #ifdef INET 1676 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1677 ss = tcp_init_win_local; 1678 #endif 1679 #ifdef INET6 1680 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1681 ss = tcp_init_win_local; 1682 #endif 1683 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1684 } 1685 1686 tcp_rmx_rtt(tp); 1687 if (tiflags & TH_ACK) { 1688 tcpstat.tcps_connects++; 1689 soisconnected(so); 1690 tcp_established(tp); 1691 /* Do window scaling on this connection? */ 1692 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1693 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1694 tp->snd_scale = tp->requested_s_scale; 1695 tp->rcv_scale = tp->request_r_scale; 1696 } 1697 TCP_REASS_LOCK(tp); 1698 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1699 TCP_REASS_UNLOCK(tp); 1700 /* 1701 * if we didn't have to retransmit the SYN, 1702 * use its rtt as our initial srtt & rtt var. 1703 */ 1704 if (tp->t_rtttime) 1705 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1706 } else 1707 tp->t_state = TCPS_SYN_RECEIVED; 1708 1709 /* 1710 * Advance th->th_seq to correspond to first data byte. 1711 * If data, trim to stay within window, 1712 * dropping FIN if necessary. 1713 */ 1714 th->th_seq++; 1715 if (tlen > tp->rcv_wnd) { 1716 todrop = tlen - tp->rcv_wnd; 1717 m_adj(m, -todrop); 1718 tlen = tp->rcv_wnd; 1719 tiflags &= ~TH_FIN; 1720 tcpstat.tcps_rcvpackafterwin++; 1721 tcpstat.tcps_rcvbyteafterwin += todrop; 1722 } 1723 tp->snd_wl1 = th->th_seq - 1; 1724 tp->rcv_up = th->th_seq; 1725 goto step6; 1726 1727 /* 1728 * If the state is SYN_RECEIVED: 1729 * If seg contains an ACK, but not for our SYN, drop the input 1730 * and generate an RST. See page 36, rfc793 1731 */ 1732 case TCPS_SYN_RECEIVED: 1733 if ((tiflags & TH_ACK) && 1734 (SEQ_LEQ(th->th_ack, tp->iss) || 1735 SEQ_GT(th->th_ack, tp->snd_max))) 1736 goto dropwithreset; 1737 break; 1738 } 1739 1740 /* 1741 * States other than LISTEN or SYN_SENT. 1742 * First check timestamp, if present. 1743 * Then check that at least some bytes of segment are within 1744 * receive window. If segment begins before rcv_nxt, 1745 * drop leading data (and SYN); if nothing left, just ack. 1746 * 1747 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1748 * and it's less than ts_recent, drop it. 1749 */ 1750 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1751 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1752 1753 /* Check to see if ts_recent is over 24 days old. */ 1754 if ((int)(TCP_TIMESTAMP(tp) - tp->ts_recent_age) > 1755 TCP_PAWS_IDLE) { 1756 /* 1757 * Invalidate ts_recent. If this segment updates 1758 * ts_recent, the age will be reset later and ts_recent 1759 * will get a valid value. If it does not, setting 1760 * ts_recent to zero will at least satisfy the 1761 * requirement that zero be placed in the timestamp 1762 * echo reply when ts_recent isn't valid. The 1763 * age isn't reset until we get a valid ts_recent 1764 * because we don't want out-of-order segments to be 1765 * dropped when ts_recent is old. 1766 */ 1767 tp->ts_recent = 0; 1768 } else { 1769 tcpstat.tcps_rcvduppack++; 1770 tcpstat.tcps_rcvdupbyte += tlen; 1771 tcpstat.tcps_pawsdrop++; 1772 goto dropafterack; 1773 } 1774 } 1775 1776 todrop = tp->rcv_nxt - th->th_seq; 1777 if (todrop > 0) { 1778 if (tiflags & TH_SYN) { 1779 tiflags &= ~TH_SYN; 1780 th->th_seq++; 1781 if (th->th_urp > 1) 1782 th->th_urp--; 1783 else { 1784 tiflags &= ~TH_URG; 1785 th->th_urp = 0; 1786 } 1787 todrop--; 1788 } 1789 if (todrop > tlen || 1790 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1791 /* 1792 * Any valid FIN or RST must be to the left of the 1793 * window. At this point the FIN or RST must be a 1794 * duplicate or out of sequence; drop it. 1795 */ 1796 if (tiflags & TH_RST) 1797 goto drop; 1798 tiflags &= ~(TH_FIN|TH_RST); 1799 /* 1800 * Send an ACK to resynchronize and drop any data. 1801 * But keep on processing for RST or ACK. 1802 */ 1803 tp->t_flags |= TF_ACKNOW; 1804 todrop = tlen; 1805 tcpstat.tcps_rcvdupbyte += todrop; 1806 tcpstat.tcps_rcvduppack++; 1807 } else if ((tiflags & TH_RST) && 1808 th->th_seq != tp->last_ack_sent) { 1809 /* 1810 * Test for reset before adjusting the sequence 1811 * number for overlapping data. 1812 */ 1813 goto dropafterack_ratelim; 1814 } else { 1815 tcpstat.tcps_rcvpartduppack++; 1816 tcpstat.tcps_rcvpartdupbyte += todrop; 1817 } 1818 hdroptlen += todrop; /*drop from head afterwards*/ 1819 th->th_seq += todrop; 1820 tlen -= todrop; 1821 if (th->th_urp > todrop) 1822 th->th_urp -= todrop; 1823 else { 1824 tiflags &= ~TH_URG; 1825 th->th_urp = 0; 1826 } 1827 } 1828 1829 /* 1830 * If new data are received on a connection after the 1831 * user processes are gone, then RST the other end. 1832 */ 1833 if ((so->so_state & SS_NOFDREF) && 1834 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1835 tp = tcp_close(tp); 1836 tcpstat.tcps_rcvafterclose++; 1837 goto dropwithreset; 1838 } 1839 1840 /* 1841 * If segment ends after window, drop trailing data 1842 * (and PUSH and FIN); if nothing left, just ACK. 1843 */ 1844 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1845 if (todrop > 0) { 1846 tcpstat.tcps_rcvpackafterwin++; 1847 if (todrop >= tlen) { 1848 /* 1849 * The segment actually starts after the window. 1850 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 1851 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 1852 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 1853 */ 1854 tcpstat.tcps_rcvbyteafterwin += tlen; 1855 /* 1856 * If a new connection request is received 1857 * while in TIME_WAIT, drop the old connection 1858 * and start over if the sequence numbers 1859 * are above the previous ones. 1860 * 1861 * NOTE: We will checksum the packet again, and 1862 * so we need to put the header fields back into 1863 * network order! 1864 * XXX This kind of sucks, but we don't expect 1865 * XXX this to happen very often, so maybe it 1866 * XXX doesn't matter so much. 1867 */ 1868 if (tiflags & TH_SYN && 1869 tp->t_state == TCPS_TIME_WAIT && 1870 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1871 iss = tcp_new_iss(tp, tp->snd_nxt); 1872 tp = tcp_close(tp); 1873 TCP_FIELDS_TO_NET(th); 1874 goto findpcb; 1875 } 1876 /* 1877 * If window is closed can only take segments at 1878 * window edge, and have to drop data and PUSH from 1879 * incoming segments. Continue processing, but 1880 * remember to ack. Otherwise, drop segment 1881 * and (if not RST) ack. 1882 */ 1883 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1884 tp->t_flags |= TF_ACKNOW; 1885 tcpstat.tcps_rcvwinprobe++; 1886 } else 1887 goto dropafterack; 1888 } else 1889 tcpstat.tcps_rcvbyteafterwin += todrop; 1890 m_adj(m, -todrop); 1891 tlen -= todrop; 1892 tiflags &= ~(TH_PUSH|TH_FIN); 1893 } 1894 1895 /* 1896 * If last ACK falls within this segment's sequence numbers, 1897 * and the timestamp is newer, record it. 1898 */ 1899 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1900 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1901 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1902 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1903 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1904 tp->ts_recent = opti.ts_val; 1905 } 1906 1907 /* 1908 * If the RST bit is set examine the state: 1909 * SYN_RECEIVED STATE: 1910 * If passive open, return to LISTEN state. 1911 * If active open, inform user that connection was refused. 1912 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1913 * Inform user that connection was reset, and close tcb. 1914 * CLOSING, LAST_ACK, TIME_WAIT STATES 1915 * Close the tcb. 1916 */ 1917 if (tiflags & TH_RST) { 1918 if (th->th_seq != tp->last_ack_sent) 1919 goto dropafterack_ratelim; 1920 1921 switch (tp->t_state) { 1922 case TCPS_SYN_RECEIVED: 1923 so->so_error = ECONNREFUSED; 1924 goto close; 1925 1926 case TCPS_ESTABLISHED: 1927 case TCPS_FIN_WAIT_1: 1928 case TCPS_FIN_WAIT_2: 1929 case TCPS_CLOSE_WAIT: 1930 so->so_error = ECONNRESET; 1931 close: 1932 tp->t_state = TCPS_CLOSED; 1933 tcpstat.tcps_drops++; 1934 tp = tcp_close(tp); 1935 goto drop; 1936 1937 case TCPS_CLOSING: 1938 case TCPS_LAST_ACK: 1939 case TCPS_TIME_WAIT: 1940 tp = tcp_close(tp); 1941 goto drop; 1942 } 1943 } 1944 1945 /* 1946 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 1947 * we must be in a synchronized state. RFC791 states (under RST 1948 * generation) that any unacceptable segment (an out-of-order SYN 1949 * qualifies) received in a synchronized state must elicit only an 1950 * empty acknowledgment segment ... and the connection remains in 1951 * the same state. 1952 */ 1953 if (tiflags & TH_SYN) { 1954 if (tp->rcv_nxt == th->th_seq) { 1955 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 1956 TH_ACK); 1957 if (tcp_saveti) 1958 m_freem(tcp_saveti); 1959 return; 1960 } 1961 1962 goto dropafterack_ratelim; 1963 } 1964 1965 /* 1966 * If the ACK bit is off we drop the segment and return. 1967 */ 1968 if ((tiflags & TH_ACK) == 0) { 1969 if (tp->t_flags & TF_ACKNOW) 1970 goto dropafterack; 1971 else 1972 goto drop; 1973 } 1974 1975 /* 1976 * Ack processing. 1977 */ 1978 switch (tp->t_state) { 1979 1980 /* 1981 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1982 * ESTABLISHED state and continue processing, otherwise 1983 * send an RST. 1984 */ 1985 case TCPS_SYN_RECEIVED: 1986 if (SEQ_GT(tp->snd_una, th->th_ack) || 1987 SEQ_GT(th->th_ack, tp->snd_max)) 1988 goto dropwithreset; 1989 tcpstat.tcps_connects++; 1990 soisconnected(so); 1991 tcp_established(tp); 1992 /* Do window scaling? */ 1993 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1994 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1995 tp->snd_scale = tp->requested_s_scale; 1996 tp->rcv_scale = tp->request_r_scale; 1997 } 1998 TCP_REASS_LOCK(tp); 1999 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2000 TCP_REASS_UNLOCK(tp); 2001 tp->snd_wl1 = th->th_seq - 1; 2002 /* fall into ... */ 2003 2004 /* 2005 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2006 * ACKs. If the ack is in the range 2007 * tp->snd_una < th->th_ack <= tp->snd_max 2008 * then advance tp->snd_una to th->th_ack and drop 2009 * data from the retransmission queue. If this ACK reflects 2010 * more up to date window information we update our window information. 2011 */ 2012 case TCPS_ESTABLISHED: 2013 case TCPS_FIN_WAIT_1: 2014 case TCPS_FIN_WAIT_2: 2015 case TCPS_CLOSE_WAIT: 2016 case TCPS_CLOSING: 2017 case TCPS_LAST_ACK: 2018 case TCPS_TIME_WAIT: 2019 2020 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2021 if (tlen == 0 && tiwin == tp->snd_wnd) { 2022 tcpstat.tcps_rcvdupack++; 2023 /* 2024 * If we have outstanding data (other than 2025 * a window probe), this is a completely 2026 * duplicate ack (ie, window info didn't 2027 * change), the ack is the biggest we've 2028 * seen and we've seen exactly our rexmt 2029 * threshhold of them, assume a packet 2030 * has been dropped and retransmit it. 2031 * Kludge snd_nxt & the congestion 2032 * window so we send only this one 2033 * packet. 2034 * 2035 * We know we're losing at the current 2036 * window size so do congestion avoidance 2037 * (set ssthresh to half the current window 2038 * and pull our congestion window back to 2039 * the new ssthresh). 2040 * 2041 * Dup acks mean that packets have left the 2042 * network (they're now cached at the receiver) 2043 * so bump cwnd by the amount in the receiver 2044 * to keep a constant cwnd packets in the 2045 * network. 2046 */ 2047 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2048 th->th_ack != tp->snd_una) 2049 tp->t_dupacks = 0; 2050 else if (++tp->t_dupacks == tcprexmtthresh) { 2051 tcp_seq onxt = tp->snd_nxt; 2052 u_int win = 2053 min(tp->snd_wnd, tp->snd_cwnd) / 2054 2 / tp->t_segsz; 2055 if (tcp_do_newreno && SEQ_LT(th->th_ack, 2056 tp->snd_recover)) { 2057 /* 2058 * False fast retransmit after 2059 * timeout. Do not cut window. 2060 */ 2061 tp->snd_cwnd += tp->t_segsz; 2062 tp->t_dupacks = 0; 2063 (void) tcp_output(tp); 2064 goto drop; 2065 } 2066 2067 if (win < 2) 2068 win = 2; 2069 tp->snd_ssthresh = win * tp->t_segsz; 2070 tp->snd_recover = tp->snd_max; 2071 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2072 tp->t_rtttime = 0; 2073 tp->snd_nxt = th->th_ack; 2074 tp->snd_cwnd = tp->t_segsz; 2075 (void) tcp_output(tp); 2076 tp->snd_cwnd = tp->snd_ssthresh + 2077 tp->t_segsz * tp->t_dupacks; 2078 if (SEQ_GT(onxt, tp->snd_nxt)) 2079 tp->snd_nxt = onxt; 2080 goto drop; 2081 } else if (tp->t_dupacks > tcprexmtthresh) { 2082 tp->snd_cwnd += tp->t_segsz; 2083 (void) tcp_output(tp); 2084 goto drop; 2085 } 2086 } else if (tlen) { 2087 tp->t_dupacks = 0; /*XXX*/ 2088 /* drop very old ACKs unless th_seq matches */ 2089 if (th->th_seq != tp->rcv_nxt && 2090 SEQ_LT(th->th_ack, 2091 tp->snd_una - tp->max_sndwnd)) { 2092 goto drop; 2093 } 2094 break; 2095 } else 2096 tp->t_dupacks = 0; 2097 break; 2098 } 2099 /* 2100 * If the congestion window was inflated to account 2101 * for the other side's cached packets, retract it. 2102 */ 2103 if (tcp_do_newreno == 0) { 2104 if (tp->t_dupacks >= tcprexmtthresh && 2105 tp->snd_cwnd > tp->snd_ssthresh) 2106 tp->snd_cwnd = tp->snd_ssthresh; 2107 tp->t_dupacks = 0; 2108 } else if (tp->t_dupacks >= tcprexmtthresh && 2109 tcp_newreno(tp, th) == 0) { 2110 tp->snd_cwnd = tp->snd_ssthresh; 2111 /* 2112 * Window inflation should have left us with approx. 2113 * snd_ssthresh outstanding data. But in case we 2114 * would be inclined to send a burst, better to do 2115 * it via the slow start mechanism. 2116 */ 2117 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 2118 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 2119 + tp->t_segsz; 2120 tp->t_dupacks = 0; 2121 } 2122 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2123 tcpstat.tcps_rcvacktoomuch++; 2124 goto dropafterack; 2125 } 2126 acked = th->th_ack - tp->snd_una; 2127 tcpstat.tcps_rcvackpack++; 2128 tcpstat.tcps_rcvackbyte += acked; 2129 2130 /* 2131 * If we have a timestamp reply, update smoothed 2132 * round trip time. If no timestamp is present but 2133 * transmit timer is running and timed sequence 2134 * number was acked, update smoothed round trip time. 2135 * Since we now have an rtt measurement, cancel the 2136 * timer backoff (cf., Phil Karn's retransmit alg.). 2137 * Recompute the initial retransmit timer. 2138 */ 2139 if (opti.ts_present && opti.ts_ecr) 2140 tcp_xmit_timer(tp, TCP_TIMESTAMP(tp) - opti.ts_ecr + 1); 2141 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2142 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2143 2144 /* 2145 * If all outstanding data is acked, stop retransmit 2146 * timer and remember to restart (more output or persist). 2147 * If there is more data to be acked, restart retransmit 2148 * timer, using current (possibly backed-off) value. 2149 */ 2150 if (th->th_ack == tp->snd_max) { 2151 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2152 needoutput = 1; 2153 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2154 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2155 /* 2156 * When new data is acked, open the congestion window. 2157 * If the window gives us less than ssthresh packets 2158 * in flight, open exponentially (segsz per packet). 2159 * Otherwise open linearly: segsz per window 2160 * (segsz^2 / cwnd per packet), plus a constant 2161 * fraction of a packet (segsz/8) to help larger windows 2162 * open quickly enough. 2163 */ 2164 { 2165 u_int cw = tp->snd_cwnd; 2166 u_int incr = tp->t_segsz; 2167 2168 if (cw > tp->snd_ssthresh) 2169 incr = incr * incr / cw; 2170 if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover)) 2171 tp->snd_cwnd = min(cw + incr, 2172 TCP_MAXWIN << tp->snd_scale); 2173 } 2174 ND6_HINT(tp); 2175 if (acked > so->so_snd.sb_cc) { 2176 tp->snd_wnd -= so->so_snd.sb_cc; 2177 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2178 ourfinisacked = 1; 2179 } else { 2180 if (acked > (tp->t_lastoff - tp->t_inoff)) 2181 tp->t_lastm = NULL; 2182 sbdrop(&so->so_snd, acked); 2183 tp->t_lastoff -= acked; 2184 tp->snd_wnd -= acked; 2185 ourfinisacked = 0; 2186 } 2187 sowwakeup(so); 2188 /* 2189 * We want snd_recover to track snd_una to 2190 * avoid sequence wraparound problems for 2191 * very large transfers. 2192 */ 2193 tp->snd_una = tp->snd_recover = th->th_ack; 2194 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2195 tp->snd_nxt = tp->snd_una; 2196 2197 switch (tp->t_state) { 2198 2199 /* 2200 * In FIN_WAIT_1 STATE in addition to the processing 2201 * for the ESTABLISHED state if our FIN is now acknowledged 2202 * then enter FIN_WAIT_2. 2203 */ 2204 case TCPS_FIN_WAIT_1: 2205 if (ourfinisacked) { 2206 /* 2207 * If we can't receive any more 2208 * data, then closing user can proceed. 2209 * Starting the timer is contrary to the 2210 * specification, but if we don't get a FIN 2211 * we'll hang forever. 2212 */ 2213 if (so->so_state & SS_CANTRCVMORE) { 2214 soisdisconnected(so); 2215 if (tcp_maxidle > 0) 2216 TCP_TIMER_ARM(tp, TCPT_2MSL, 2217 tcp_maxidle); 2218 } 2219 tp->t_state = TCPS_FIN_WAIT_2; 2220 } 2221 break; 2222 2223 /* 2224 * In CLOSING STATE in addition to the processing for 2225 * the ESTABLISHED state if the ACK acknowledges our FIN 2226 * then enter the TIME-WAIT state, otherwise ignore 2227 * the segment. 2228 */ 2229 case TCPS_CLOSING: 2230 if (ourfinisacked) { 2231 tp->t_state = TCPS_TIME_WAIT; 2232 tcp_canceltimers(tp); 2233 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2234 soisdisconnected(so); 2235 } 2236 break; 2237 2238 /* 2239 * In LAST_ACK, we may still be waiting for data to drain 2240 * and/or to be acked, as well as for the ack of our FIN. 2241 * If our FIN is now acknowledged, delete the TCB, 2242 * enter the closed state and return. 2243 */ 2244 case TCPS_LAST_ACK: 2245 if (ourfinisacked) { 2246 tp = tcp_close(tp); 2247 goto drop; 2248 } 2249 break; 2250 2251 /* 2252 * In TIME_WAIT state the only thing that should arrive 2253 * is a retransmission of the remote FIN. Acknowledge 2254 * it and restart the finack timer. 2255 */ 2256 case TCPS_TIME_WAIT: 2257 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2258 goto dropafterack; 2259 } 2260 } 2261 2262 step6: 2263 /* 2264 * Update window information. 2265 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2266 */ 2267 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2268 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 2269 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 2270 /* keep track of pure window updates */ 2271 if (tlen == 0 && 2272 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2273 tcpstat.tcps_rcvwinupd++; 2274 tp->snd_wnd = tiwin; 2275 tp->snd_wl1 = th->th_seq; 2276 tp->snd_wl2 = th->th_ack; 2277 if (tp->snd_wnd > tp->max_sndwnd) 2278 tp->max_sndwnd = tp->snd_wnd; 2279 needoutput = 1; 2280 } 2281 2282 /* 2283 * Process segments with URG. 2284 */ 2285 if ((tiflags & TH_URG) && th->th_urp && 2286 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2287 /* 2288 * This is a kludge, but if we receive and accept 2289 * random urgent pointers, we'll crash in 2290 * soreceive. It's hard to imagine someone 2291 * actually wanting to send this much urgent data. 2292 */ 2293 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2294 th->th_urp = 0; /* XXX */ 2295 tiflags &= ~TH_URG; /* XXX */ 2296 goto dodata; /* XXX */ 2297 } 2298 /* 2299 * If this segment advances the known urgent pointer, 2300 * then mark the data stream. This should not happen 2301 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2302 * a FIN has been received from the remote side. 2303 * In these states we ignore the URG. 2304 * 2305 * According to RFC961 (Assigned Protocols), 2306 * the urgent pointer points to the last octet 2307 * of urgent data. We continue, however, 2308 * to consider it to indicate the first octet 2309 * of data past the urgent section as the original 2310 * spec states (in one of two places). 2311 */ 2312 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2313 tp->rcv_up = th->th_seq + th->th_urp; 2314 so->so_oobmark = so->so_rcv.sb_cc + 2315 (tp->rcv_up - tp->rcv_nxt) - 1; 2316 if (so->so_oobmark == 0) 2317 so->so_state |= SS_RCVATMARK; 2318 sohasoutofband(so); 2319 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2320 } 2321 /* 2322 * Remove out of band data so doesn't get presented to user. 2323 * This can happen independent of advancing the URG pointer, 2324 * but if two URG's are pending at once, some out-of-band 2325 * data may creep in... ick. 2326 */ 2327 if (th->th_urp <= (u_int16_t) tlen 2328 #ifdef SO_OOBINLINE 2329 && (so->so_options & SO_OOBINLINE) == 0 2330 #endif 2331 ) 2332 tcp_pulloutofband(so, th, m, hdroptlen); 2333 } else 2334 /* 2335 * If no out of band data is expected, 2336 * pull receive urgent pointer along 2337 * with the receive window. 2338 */ 2339 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2340 tp->rcv_up = tp->rcv_nxt; 2341 dodata: /* XXX */ 2342 2343 /* 2344 * Process the segment text, merging it into the TCP sequencing queue, 2345 * and arranging for acknowledgement of receipt if necessary. 2346 * This process logically involves adjusting tp->rcv_wnd as data 2347 * is presented to the user (this happens in tcp_usrreq.c, 2348 * case PRU_RCVD). If a FIN has already been received on this 2349 * connection then we just ignore the text. 2350 */ 2351 if ((tlen || (tiflags & TH_FIN)) && 2352 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2353 /* 2354 * Insert segment ti into reassembly queue of tcp with 2355 * control block tp. Return TH_FIN if reassembly now includes 2356 * a segment with FIN. The macro form does the common case 2357 * inline (segment is the next to be received on an 2358 * established connection, and the queue is empty), 2359 * avoiding linkage into and removal from the queue and 2360 * repetition of various conversions. 2361 * Set DELACK for segments received in order, but ack 2362 * immediately when segments are out of order 2363 * (so fast retransmit can work). 2364 */ 2365 /* NOTE: this was TCP_REASS() macro, but used only once */ 2366 TCP_REASS_LOCK(tp); 2367 if (th->th_seq == tp->rcv_nxt && 2368 TAILQ_FIRST(&tp->segq) == NULL && 2369 tp->t_state == TCPS_ESTABLISHED) { 2370 TCP_SETUP_ACK(tp, th); 2371 tp->rcv_nxt += tlen; 2372 tiflags = th->th_flags & TH_FIN; 2373 tcpstat.tcps_rcvpack++; 2374 tcpstat.tcps_rcvbyte += tlen; 2375 ND6_HINT(tp); 2376 if (so->so_state & SS_CANTRCVMORE) 2377 m_freem(m); 2378 else { 2379 m_adj(m, hdroptlen); 2380 sbappendstream(&(so)->so_rcv, m); 2381 } 2382 sorwakeup(so); 2383 } else { 2384 m_adj(m, hdroptlen); 2385 tiflags = tcp_reass(tp, th, m, &tlen); 2386 tp->t_flags |= TF_ACKNOW; 2387 } 2388 TCP_REASS_UNLOCK(tp); 2389 2390 /* 2391 * Note the amount of data that peer has sent into 2392 * our window, in order to estimate the sender's 2393 * buffer size. 2394 */ 2395 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2396 } else { 2397 m_freem(m); 2398 m = NULL; 2399 tiflags &= ~TH_FIN; 2400 } 2401 2402 /* 2403 * If FIN is received ACK the FIN and let the user know 2404 * that the connection is closing. Ignore a FIN received before 2405 * the connection is fully established. 2406 */ 2407 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2408 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2409 socantrcvmore(so); 2410 tp->t_flags |= TF_ACKNOW; 2411 tp->rcv_nxt++; 2412 } 2413 switch (tp->t_state) { 2414 2415 /* 2416 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2417 */ 2418 case TCPS_ESTABLISHED: 2419 tp->t_state = TCPS_CLOSE_WAIT; 2420 break; 2421 2422 /* 2423 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2424 * enter the CLOSING state. 2425 */ 2426 case TCPS_FIN_WAIT_1: 2427 tp->t_state = TCPS_CLOSING; 2428 break; 2429 2430 /* 2431 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2432 * starting the time-wait timer, turning off the other 2433 * standard timers. 2434 */ 2435 case TCPS_FIN_WAIT_2: 2436 tp->t_state = TCPS_TIME_WAIT; 2437 tcp_canceltimers(tp); 2438 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2439 soisdisconnected(so); 2440 break; 2441 2442 /* 2443 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2444 */ 2445 case TCPS_TIME_WAIT: 2446 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2447 break; 2448 } 2449 } 2450 #ifdef TCP_DEBUG 2451 if (so->so_options & SO_DEBUG) 2452 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2453 #endif 2454 2455 /* 2456 * Return any desired output. 2457 */ 2458 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2459 (void) tcp_output(tp); 2460 if (tcp_saveti) 2461 m_freem(tcp_saveti); 2462 return; 2463 2464 badsyn: 2465 /* 2466 * Received a bad SYN. Increment counters and dropwithreset. 2467 */ 2468 tcpstat.tcps_badsyn++; 2469 tp = NULL; 2470 goto dropwithreset; 2471 2472 dropafterack: 2473 /* 2474 * Generate an ACK dropping incoming segment if it occupies 2475 * sequence space, where the ACK reflects our state. 2476 */ 2477 if (tiflags & TH_RST) 2478 goto drop; 2479 goto dropafterack2; 2480 2481 dropafterack_ratelim: 2482 /* 2483 * We may want to rate-limit ACKs against SYN/RST attack. 2484 */ 2485 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2486 tcp_ackdrop_ppslim) == 0) { 2487 /* XXX stat */ 2488 goto drop; 2489 } 2490 /* ...fall into dropafterack2... */ 2491 2492 dropafterack2: 2493 m_freem(m); 2494 tp->t_flags |= TF_ACKNOW; 2495 (void) tcp_output(tp); 2496 if (tcp_saveti) 2497 m_freem(tcp_saveti); 2498 return; 2499 2500 dropwithreset_ratelim: 2501 /* 2502 * We may want to rate-limit RSTs in certain situations, 2503 * particularly if we are sending an RST in response to 2504 * an attempt to connect to or otherwise communicate with 2505 * a port for which we have no socket. 2506 */ 2507 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2508 tcp_rst_ppslim) == 0) { 2509 /* XXX stat */ 2510 goto drop; 2511 } 2512 /* ...fall into dropwithreset... */ 2513 2514 dropwithreset: 2515 /* 2516 * Generate a RST, dropping incoming segment. 2517 * Make ACK acceptable to originator of segment. 2518 */ 2519 if (tiflags & TH_RST) 2520 goto drop; 2521 2522 switch (af) { 2523 #ifdef INET6 2524 case AF_INET6: 2525 /* For following calls to tcp_respond */ 2526 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2527 goto drop; 2528 break; 2529 #endif /* INET6 */ 2530 case AF_INET: 2531 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2532 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2533 goto drop; 2534 } 2535 2536 if (tiflags & TH_ACK) 2537 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2538 else { 2539 if (tiflags & TH_SYN) 2540 tlen++; 2541 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2542 TH_RST|TH_ACK); 2543 } 2544 if (tcp_saveti) 2545 m_freem(tcp_saveti); 2546 return; 2547 2548 badcsum: 2549 tcpstat.tcps_rcvbadsum++; 2550 drop: 2551 /* 2552 * Drop space held by incoming segment and return. 2553 */ 2554 if (tp) { 2555 if (tp->t_inpcb) 2556 so = tp->t_inpcb->inp_socket; 2557 #ifdef INET6 2558 else if (tp->t_in6pcb) 2559 so = tp->t_in6pcb->in6p_socket; 2560 #endif 2561 else 2562 so = NULL; 2563 #ifdef TCP_DEBUG 2564 if (so && (so->so_options & SO_DEBUG) != 0) 2565 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2566 #endif 2567 } 2568 if (tcp_saveti) 2569 m_freem(tcp_saveti); 2570 m_freem(m); 2571 return; 2572 } 2573 2574 #ifdef TCP_SIGNATURE 2575 int 2576 tcp_signature_apply(void *fstate, caddr_t data, u_int len) 2577 { 2578 2579 MD5Update(fstate, (u_char *)data, len); 2580 return (0); 2581 } 2582 2583 struct secasvar * 2584 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 2585 { 2586 struct secasvar *sav; 2587 #ifdef FAST_IPSEC 2588 union sockaddr_union dst; 2589 #endif 2590 struct ip *ip; 2591 struct ip6_hdr *ip6; 2592 2593 ip = mtod(m, struct ip *); 2594 switch (ip->ip_v) { 2595 case 4: 2596 ip = mtod(m, struct ip *); 2597 ip6 = NULL; 2598 break; 2599 case 6: 2600 ip = NULL; 2601 ip6 = mtod(m, struct ip6_hdr *); 2602 break; 2603 default: 2604 return (NULL); 2605 } 2606 2607 #ifdef FAST_IPSEC 2608 /* Extract the destination from the IP header in the mbuf. */ 2609 bzero(&dst, sizeof(union sockaddr_union)); 2610 dst.sa.sa_len = sizeof(struct sockaddr_in); 2611 dst.sa.sa_family = AF_INET; 2612 dst.sin.sin_addr = ip->ip_dst; 2613 2614 /* 2615 * Look up an SADB entry which matches the address of the peer. 2616 */ 2617 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2618 #else 2619 if (ip) 2620 sav = key_allocsa(AF_INET, (caddr_t)&ip->ip_src, 2621 (caddr_t)&ip->ip_dst, IPPROTO_TCP, 2622 htonl(TCP_SIG_SPI)); 2623 else 2624 sav = key_allocsa(AF_INET6, (caddr_t)&ip6->ip6_src, 2625 (caddr_t)&ip6->ip6_dst, IPPROTO_TCP, 2626 htonl(TCP_SIG_SPI)); 2627 #endif 2628 2629 return (sav); /* freesav must be performed by caller */ 2630 } 2631 2632 int 2633 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 2634 struct secasvar *sav, char *sig) 2635 { 2636 MD5_CTX ctx; 2637 struct ip *ip; 2638 struct ipovly *ipovly; 2639 struct ip6_hdr *ip6; 2640 struct ippseudo ippseudo; 2641 struct ip6_hdr_pseudo ip6pseudo; 2642 struct tcphdr th0; 2643 int l, tcphdrlen; 2644 2645 if (sav == NULL) 2646 return (-1); 2647 2648 tcphdrlen = th->th_off * 4; 2649 2650 switch (mtod(m, struct ip *)->ip_v) { 2651 case 4: 2652 ip = mtod(m, struct ip *); 2653 ip6 = NULL; 2654 break; 2655 case 6: 2656 ip = NULL; 2657 ip6 = mtod(m, struct ip6_hdr *); 2658 break; 2659 default: 2660 return (-1); 2661 } 2662 2663 MD5Init(&ctx); 2664 2665 if (ip) { 2666 memset(&ippseudo, 0, sizeof(ippseudo)); 2667 ipovly = (struct ipovly *)ip; 2668 ippseudo.ippseudo_src = ipovly->ih_src; 2669 ippseudo.ippseudo_dst = ipovly->ih_dst; 2670 ippseudo.ippseudo_pad = 0; 2671 ippseudo.ippseudo_p = IPPROTO_TCP; 2672 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 2673 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 2674 } else { 2675 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 2676 ip6pseudo.ip6ph_src = ip6->ip6_src; 2677 in6_clearscope(&ip6pseudo.ip6ph_src); 2678 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 2679 in6_clearscope(&ip6pseudo.ip6ph_dst); 2680 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 2681 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 2682 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 2683 } 2684 2685 th0 = *th; 2686 th0.th_sum = 0; 2687 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 2688 2689 l = m->m_pkthdr.len - thoff - tcphdrlen; 2690 if (l > 0) 2691 m_apply(m, thoff + tcphdrlen, 2692 m->m_pkthdr.len - thoff - tcphdrlen, 2693 tcp_signature_apply, &ctx); 2694 2695 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2696 MD5Final(sig, &ctx); 2697 2698 return (0); 2699 } 2700 #endif 2701 2702 int 2703 tcp_dooptions(tp, cp, cnt, th, m, toff, oi) 2704 struct tcpcb *tp; 2705 u_char *cp; 2706 int cnt; 2707 struct tcphdr *th; 2708 struct mbuf *m; 2709 int toff; 2710 struct tcp_opt_info *oi; 2711 { 2712 u_int16_t mss; 2713 int opt, optlen = 0; 2714 #ifdef TCP_SIGNATURE 2715 caddr_t sigp = NULL; 2716 char sigbuf[TCP_SIGLEN]; 2717 struct secasvar *sav = NULL; 2718 #endif 2719 2720 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 2721 opt = cp[0]; 2722 if (opt == TCPOPT_EOL) 2723 break; 2724 if (opt == TCPOPT_NOP) 2725 optlen = 1; 2726 else { 2727 if (cnt < 2) 2728 break; 2729 optlen = cp[1]; 2730 if (optlen < 2 || optlen > cnt) 2731 break; 2732 } 2733 switch (opt) { 2734 2735 default: 2736 continue; 2737 2738 case TCPOPT_MAXSEG: 2739 if (optlen != TCPOLEN_MAXSEG) 2740 continue; 2741 if (!(th->th_flags & TH_SYN)) 2742 continue; 2743 bcopy(cp + 2, &mss, sizeof(mss)); 2744 oi->maxseg = ntohs(mss); 2745 break; 2746 2747 case TCPOPT_WINDOW: 2748 if (optlen != TCPOLEN_WINDOW) 2749 continue; 2750 if (!(th->th_flags & TH_SYN)) 2751 continue; 2752 tp->t_flags |= TF_RCVD_SCALE; 2753 tp->requested_s_scale = cp[2]; 2754 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2755 #if 0 /*XXX*/ 2756 char *p; 2757 2758 if (ip) 2759 p = ntohl(ip->ip_src); 2760 #ifdef INET6 2761 else if (ip6) 2762 p = ip6_sprintf(&ip6->ip6_src); 2763 #endif 2764 else 2765 p = "(unknown)"; 2766 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2767 "assuming %d\n", 2768 tp->requested_s_scale, p, 2769 TCP_MAX_WINSHIFT); 2770 #else 2771 log(LOG_ERR, "TCP: invalid wscale %d, " 2772 "assuming %d\n", 2773 tp->requested_s_scale, 2774 TCP_MAX_WINSHIFT); 2775 #endif 2776 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2777 } 2778 break; 2779 2780 case TCPOPT_TIMESTAMP: 2781 if (optlen != TCPOLEN_TIMESTAMP) 2782 continue; 2783 oi->ts_present = 1; 2784 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2785 NTOHL(oi->ts_val); 2786 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2787 NTOHL(oi->ts_ecr); 2788 2789 /* 2790 * A timestamp received in a SYN makes 2791 * it ok to send timestamp requests and replies. 2792 */ 2793 if (th->th_flags & TH_SYN) { 2794 tp->t_flags |= TF_RCVD_TSTMP; 2795 tp->ts_recent = oi->ts_val; 2796 tp->ts_recent_age = TCP_TIMESTAMP(tp); 2797 } 2798 break; 2799 case TCPOPT_SACK_PERMITTED: 2800 if (optlen != TCPOLEN_SACK_PERMITTED) 2801 continue; 2802 if (!(th->th_flags & TH_SYN)) 2803 continue; 2804 tp->t_flags &= ~TF_CANT_TXSACK; 2805 break; 2806 2807 case TCPOPT_SACK: 2808 if (tp->t_flags & TF_IGNR_RXSACK) 2809 continue; 2810 if (optlen % 8 != 2 || optlen < 10) 2811 continue; 2812 cp += 2; 2813 optlen -= 2; 2814 for (; optlen > 0; cp -= 8, optlen -= 8) { 2815 tcp_seq lwe, rwe; 2816 bcopy((char *)cp, (char *) &lwe, sizeof(lwe)); 2817 NTOHL(lwe); 2818 bcopy((char *)cp, (char *) &rwe, sizeof(rwe)); 2819 NTOHL(rwe); 2820 /* tcp_mark_sacked(tp, lwe, rwe); */ 2821 } 2822 break; 2823 #ifdef TCP_SIGNATURE 2824 case TCPOPT_SIGNATURE: 2825 if (optlen != TCPOLEN_SIGNATURE) 2826 continue; 2827 if (sigp && bcmp(sigp, cp + 2, TCP_SIGLEN)) 2828 return (-1); 2829 2830 sigp = sigbuf; 2831 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 2832 memset(cp + 2, 0, TCP_SIGLEN); 2833 tp->t_flags |= TF_SIGNATURE; 2834 break; 2835 #endif 2836 } 2837 } 2838 2839 #ifdef TCP_SIGNATURE 2840 if (tp->t_flags & TF_SIGNATURE) { 2841 2842 sav = tcp_signature_getsav(m, th); 2843 2844 if (sav == NULL && tp->t_state == TCPS_LISTEN) 2845 return (-1); 2846 } 2847 2848 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 2849 if (sav == NULL) 2850 return (-1); 2851 #ifdef FAST_IPSEC 2852 KEY_FREESAV(&sav); 2853 #else 2854 key_freesav(sav); 2855 #endif 2856 return (-1); 2857 } 2858 2859 if (sigp) { 2860 char sig[TCP_SIGLEN]; 2861 2862 TCP_FIELDS_TO_NET(th); 2863 if (tcp_signature(m, th, toff, sav, sig) < 0) { 2864 TCP_FIELDS_TO_HOST(th); 2865 if (sav == NULL) 2866 return (-1); 2867 #ifdef FAST_IPSEC 2868 KEY_FREESAV(&sav); 2869 #else 2870 key_freesav(sav); 2871 #endif 2872 return (-1); 2873 } 2874 TCP_FIELDS_TO_HOST(th); 2875 2876 if (bcmp(sig, sigp, TCP_SIGLEN)) { 2877 tcpstat.tcps_badsig++; 2878 if (sav == NULL) 2879 return (-1); 2880 #ifdef FAST_IPSEC 2881 KEY_FREESAV(&sav); 2882 #else 2883 key_freesav(sav); 2884 #endif 2885 return (-1); 2886 } else 2887 tcpstat.tcps_goodsig++; 2888 2889 key_sa_recordxfer(sav, m); 2890 #ifdef FAST_IPSEC 2891 KEY_FREESAV(&sav); 2892 #else 2893 key_freesav(sav); 2894 #endif 2895 } 2896 #endif 2897 2898 return (0); 2899 } 2900 2901 /* 2902 * Pull out of band byte out of a segment so 2903 * it doesn't appear in the user's data queue. 2904 * It is still reflected in the segment length for 2905 * sequencing purposes. 2906 */ 2907 void 2908 tcp_pulloutofband(so, th, m, off) 2909 struct socket *so; 2910 struct tcphdr *th; 2911 struct mbuf *m; 2912 int off; 2913 { 2914 int cnt = off + th->th_urp - 1; 2915 2916 while (cnt >= 0) { 2917 if (m->m_len > cnt) { 2918 char *cp = mtod(m, caddr_t) + cnt; 2919 struct tcpcb *tp = sototcpcb(so); 2920 2921 tp->t_iobc = *cp; 2922 tp->t_oobflags |= TCPOOB_HAVEDATA; 2923 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2924 m->m_len--; 2925 return; 2926 } 2927 cnt -= m->m_len; 2928 m = m->m_next; 2929 if (m == 0) 2930 break; 2931 } 2932 panic("tcp_pulloutofband"); 2933 } 2934 2935 /* 2936 * Collect new round-trip time estimate 2937 * and update averages and current timeout. 2938 */ 2939 void 2940 tcp_xmit_timer(tp, rtt) 2941 struct tcpcb *tp; 2942 uint32_t rtt; 2943 { 2944 int32_t delta; 2945 2946 tcpstat.tcps_rttupdated++; 2947 if (tp->t_srtt != 0) { 2948 /* 2949 * srtt is stored as fixed point with 3 bits after the 2950 * binary point (i.e., scaled by 8). The following magic 2951 * is equivalent to the smoothing algorithm in rfc793 with 2952 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2953 * point). Adjust rtt to origin 0. 2954 */ 2955 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2956 if ((tp->t_srtt += delta) <= 0) 2957 tp->t_srtt = 1 << 2; 2958 /* 2959 * We accumulate a smoothed rtt variance (actually, a 2960 * smoothed mean difference), then set the retransmit 2961 * timer to smoothed rtt + 4 times the smoothed variance. 2962 * rttvar is stored as fixed point with 2 bits after the 2963 * binary point (scaled by 4). The following is 2964 * equivalent to rfc793 smoothing with an alpha of .75 2965 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2966 * rfc793's wired-in beta. 2967 */ 2968 if (delta < 0) 2969 delta = -delta; 2970 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2971 if ((tp->t_rttvar += delta) <= 0) 2972 tp->t_rttvar = 1 << 2; 2973 } else { 2974 /* 2975 * No rtt measurement yet - use the unsmoothed rtt. 2976 * Set the variance to half the rtt (so our first 2977 * retransmit happens at 3*rtt). 2978 */ 2979 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 2980 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 2981 } 2982 tp->t_rtttime = 0; 2983 tp->t_rxtshift = 0; 2984 2985 /* 2986 * the retransmit should happen at rtt + 4 * rttvar. 2987 * Because of the way we do the smoothing, srtt and rttvar 2988 * will each average +1/2 tick of bias. When we compute 2989 * the retransmit timer, we want 1/2 tick of rounding and 2990 * 1 extra tick because of +-1/2 tick uncertainty in the 2991 * firing of the timer. The bias will give us exactly the 2992 * 1.5 tick we need. But, because the bias is 2993 * statistical, we have to test that we don't drop below 2994 * the minimum feasible timer (which is 2 ticks). 2995 */ 2996 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2997 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2998 2999 /* 3000 * We received an ack for a packet that wasn't retransmitted; 3001 * it is probably safe to discard any error indications we've 3002 * received recently. This isn't quite right, but close enough 3003 * for now (a route might have failed after we sent a segment, 3004 * and the return path might not be symmetrical). 3005 */ 3006 tp->t_softerror = 0; 3007 } 3008 3009 /* 3010 * Checks for partial ack. If partial ack arrives, force the retransmission 3011 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 3012 * 1. By setting snd_nxt to th_ack, this forces retransmission timer to 3013 * be started again. If the ack advances at least to tp->snd_recover, return 0. 3014 */ 3015 int 3016 tcp_newreno(tp, th) 3017 struct tcpcb *tp; 3018 struct tcphdr *th; 3019 { 3020 tcp_seq onxt = tp->snd_nxt; 3021 u_long ocwnd = tp->snd_cwnd; 3022 3023 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 3024 /* 3025 * snd_una has not yet been updated and the socket's send 3026 * buffer has not yet drained off the ACK'd data, so we 3027 * have to leave snd_una as it was to get the correct data 3028 * offset in tcp_output(). 3029 */ 3030 TCP_TIMER_DISARM(tp, TCPT_REXMT); 3031 tp->t_rtttime = 0; 3032 tp->snd_nxt = th->th_ack; 3033 /* 3034 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 3035 * is not yet updated when we're called. 3036 */ 3037 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 3038 (void) tcp_output(tp); 3039 tp->snd_cwnd = ocwnd; 3040 if (SEQ_GT(onxt, tp->snd_nxt)) 3041 tp->snd_nxt = onxt; 3042 /* 3043 * Partial window deflation. Relies on fact that tp->snd_una 3044 * not updated yet. 3045 */ 3046 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 3047 return 1; 3048 } 3049 return 0; 3050 } 3051 3052 3053 /* 3054 * TCP compressed state engine. Currently used to hold compressed 3055 * state for SYN_RECEIVED. 3056 */ 3057 3058 u_long syn_cache_count; 3059 u_int32_t syn_hash1, syn_hash2; 3060 3061 #define SYN_HASH(sa, sp, dp) \ 3062 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3063 ((u_int32_t)(sp)))^syn_hash2))) 3064 #ifndef INET6 3065 #define SYN_HASHALL(hash, src, dst) \ 3066 do { \ 3067 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 3068 ((struct sockaddr_in *)(src))->sin_port, \ 3069 ((struct sockaddr_in *)(dst))->sin_port); \ 3070 } while (/*CONSTCOND*/ 0) 3071 #else 3072 #define SYN_HASH6(sa, sp, dp) \ 3073 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3074 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3075 & 0x7fffffff) 3076 3077 #define SYN_HASHALL(hash, src, dst) \ 3078 do { \ 3079 switch ((src)->sa_family) { \ 3080 case AF_INET: \ 3081 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 3082 ((struct sockaddr_in *)(src))->sin_port, \ 3083 ((struct sockaddr_in *)(dst))->sin_port); \ 3084 break; \ 3085 case AF_INET6: \ 3086 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 3087 ((struct sockaddr_in6 *)(src))->sin6_port, \ 3088 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 3089 break; \ 3090 default: \ 3091 hash = 0; \ 3092 } \ 3093 } while (/*CONSTCOND*/0) 3094 #endif /* INET6 */ 3095 3096 #define SYN_CACHE_RM(sc) \ 3097 do { \ 3098 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 3099 (sc), sc_bucketq); \ 3100 (sc)->sc_tp = NULL; \ 3101 LIST_REMOVE((sc), sc_tpq); \ 3102 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 3103 callout_stop(&(sc)->sc_timer); \ 3104 syn_cache_count--; \ 3105 } while (/*CONSTCOND*/0) 3106 3107 #define SYN_CACHE_PUT(sc) \ 3108 do { \ 3109 if ((sc)->sc_ipopts) \ 3110 (void) m_free((sc)->sc_ipopts); \ 3111 if ((sc)->sc_route4.ro_rt != NULL) \ 3112 RTFREE((sc)->sc_route4.ro_rt); \ 3113 if (callout_invoking(&(sc)->sc_timer)) \ 3114 (sc)->sc_flags |= SCF_DEAD; \ 3115 else \ 3116 pool_put(&syn_cache_pool, (sc)); \ 3117 } while (/*CONSTCOND*/0) 3118 3119 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL); 3120 3121 /* 3122 * We don't estimate RTT with SYNs, so each packet starts with the default 3123 * RTT and each timer step has a fixed timeout value. 3124 */ 3125 #define SYN_CACHE_TIMER_ARM(sc) \ 3126 do { \ 3127 TCPT_RANGESET((sc)->sc_rxtcur, \ 3128 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3129 TCPTV_REXMTMAX); \ 3130 callout_reset(&(sc)->sc_timer, \ 3131 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3132 } while (/*CONSTCOND*/0) 3133 3134 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3135 3136 void 3137 syn_cache_init() 3138 { 3139 int i; 3140 3141 /* Initialize the hash buckets. */ 3142 for (i = 0; i < tcp_syn_cache_size; i++) 3143 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3144 } 3145 3146 void 3147 syn_cache_insert(sc, tp) 3148 struct syn_cache *sc; 3149 struct tcpcb *tp; 3150 { 3151 struct syn_cache_head *scp; 3152 struct syn_cache *sc2; 3153 int s; 3154 3155 /* 3156 * If there are no entries in the hash table, reinitialize 3157 * the hash secrets. 3158 */ 3159 if (syn_cache_count == 0) { 3160 syn_hash1 = arc4random(); 3161 syn_hash2 = arc4random(); 3162 } 3163 3164 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3165 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3166 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3167 3168 /* 3169 * Make sure that we don't overflow the per-bucket 3170 * limit or the total cache size limit. 3171 */ 3172 s = splsoftnet(); 3173 if (scp->sch_length >= tcp_syn_bucket_limit) { 3174 tcpstat.tcps_sc_bucketoverflow++; 3175 /* 3176 * The bucket is full. Toss the oldest element in the 3177 * bucket. This will be the first entry in the bucket. 3178 */ 3179 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3180 #ifdef DIAGNOSTIC 3181 /* 3182 * This should never happen; we should always find an 3183 * entry in our bucket. 3184 */ 3185 if (sc2 == NULL) 3186 panic("syn_cache_insert: bucketoverflow: impossible"); 3187 #endif 3188 SYN_CACHE_RM(sc2); 3189 SYN_CACHE_PUT(sc2); 3190 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3191 struct syn_cache_head *scp2, *sce; 3192 3193 tcpstat.tcps_sc_overflowed++; 3194 /* 3195 * The cache is full. Toss the oldest entry in the 3196 * first non-empty bucket we can find. 3197 * 3198 * XXX We would really like to toss the oldest 3199 * entry in the cache, but we hope that this 3200 * condition doesn't happen very often. 3201 */ 3202 scp2 = scp; 3203 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3204 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3205 for (++scp2; scp2 != scp; scp2++) { 3206 if (scp2 >= sce) 3207 scp2 = &tcp_syn_cache[0]; 3208 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3209 break; 3210 } 3211 #ifdef DIAGNOSTIC 3212 /* 3213 * This should never happen; we should always find a 3214 * non-empty bucket. 3215 */ 3216 if (scp2 == scp) 3217 panic("syn_cache_insert: cacheoverflow: " 3218 "impossible"); 3219 #endif 3220 } 3221 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3222 SYN_CACHE_RM(sc2); 3223 SYN_CACHE_PUT(sc2); 3224 } 3225 3226 /* 3227 * Initialize the entry's timer. 3228 */ 3229 sc->sc_rxttot = 0; 3230 sc->sc_rxtshift = 0; 3231 SYN_CACHE_TIMER_ARM(sc); 3232 3233 /* Link it from tcpcb entry */ 3234 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3235 3236 /* Put it into the bucket. */ 3237 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3238 scp->sch_length++; 3239 syn_cache_count++; 3240 3241 tcpstat.tcps_sc_added++; 3242 splx(s); 3243 } 3244 3245 /* 3246 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3247 * If we have retransmitted an entry the maximum number of times, expire 3248 * that entry. 3249 */ 3250 void 3251 syn_cache_timer(void *arg) 3252 { 3253 struct syn_cache *sc = arg; 3254 int s; 3255 3256 s = splsoftnet(); 3257 callout_ack(&sc->sc_timer); 3258 3259 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3260 tcpstat.tcps_sc_delayed_free++; 3261 pool_put(&syn_cache_pool, sc); 3262 splx(s); 3263 return; 3264 } 3265 3266 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3267 /* Drop it -- too many retransmissions. */ 3268 goto dropit; 3269 } 3270 3271 /* 3272 * Compute the total amount of time this entry has 3273 * been on a queue. If this entry has been on longer 3274 * than the keep alive timer would allow, expire it. 3275 */ 3276 sc->sc_rxttot += sc->sc_rxtcur; 3277 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) 3278 goto dropit; 3279 3280 tcpstat.tcps_sc_retransmitted++; 3281 (void) syn_cache_respond(sc, NULL); 3282 3283 /* Advance the timer back-off. */ 3284 sc->sc_rxtshift++; 3285 SYN_CACHE_TIMER_ARM(sc); 3286 3287 splx(s); 3288 return; 3289 3290 dropit: 3291 tcpstat.tcps_sc_timed_out++; 3292 SYN_CACHE_RM(sc); 3293 SYN_CACHE_PUT(sc); 3294 splx(s); 3295 } 3296 3297 /* 3298 * Remove syn cache created by the specified tcb entry, 3299 * because this does not make sense to keep them 3300 * (if there's no tcb entry, syn cache entry will never be used) 3301 */ 3302 void 3303 syn_cache_cleanup(tp) 3304 struct tcpcb *tp; 3305 { 3306 struct syn_cache *sc, *nsc; 3307 int s; 3308 3309 s = splsoftnet(); 3310 3311 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3312 nsc = LIST_NEXT(sc, sc_tpq); 3313 3314 #ifdef DIAGNOSTIC 3315 if (sc->sc_tp != tp) 3316 panic("invalid sc_tp in syn_cache_cleanup"); 3317 #endif 3318 SYN_CACHE_RM(sc); 3319 SYN_CACHE_PUT(sc); 3320 } 3321 /* just for safety */ 3322 LIST_INIT(&tp->t_sc); 3323 3324 splx(s); 3325 } 3326 3327 /* 3328 * Find an entry in the syn cache. 3329 */ 3330 struct syn_cache * 3331 syn_cache_lookup(src, dst, headp) 3332 struct sockaddr *src; 3333 struct sockaddr *dst; 3334 struct syn_cache_head **headp; 3335 { 3336 struct syn_cache *sc; 3337 struct syn_cache_head *scp; 3338 u_int32_t hash; 3339 int s; 3340 3341 SYN_HASHALL(hash, src, dst); 3342 3343 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3344 *headp = scp; 3345 s = splsoftnet(); 3346 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3347 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3348 if (sc->sc_hash != hash) 3349 continue; 3350 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3351 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3352 splx(s); 3353 return (sc); 3354 } 3355 } 3356 splx(s); 3357 return (NULL); 3358 } 3359 3360 /* 3361 * This function gets called when we receive an ACK for a 3362 * socket in the LISTEN state. We look up the connection 3363 * in the syn cache, and if its there, we pull it out of 3364 * the cache and turn it into a full-blown connection in 3365 * the SYN-RECEIVED state. 3366 * 3367 * The return values may not be immediately obvious, and their effects 3368 * can be subtle, so here they are: 3369 * 3370 * NULL SYN was not found in cache; caller should drop the 3371 * packet and send an RST. 3372 * 3373 * -1 We were unable to create the new connection, and are 3374 * aborting it. An ACK,RST is being sent to the peer 3375 * (unless we got screwey sequence numbners; see below), 3376 * because the 3-way handshake has been completed. Caller 3377 * should not free the mbuf, since we may be using it. If 3378 * we are not, we will free it. 3379 * 3380 * Otherwise, the return value is a pointer to the new socket 3381 * associated with the connection. 3382 */ 3383 struct socket * 3384 syn_cache_get(src, dst, th, hlen, tlen, so, m) 3385 struct sockaddr *src; 3386 struct sockaddr *dst; 3387 struct tcphdr *th; 3388 unsigned int hlen, tlen; 3389 struct socket *so; 3390 struct mbuf *m; 3391 { 3392 struct syn_cache *sc; 3393 struct syn_cache_head *scp; 3394 struct inpcb *inp = NULL; 3395 #ifdef INET6 3396 struct in6pcb *in6p = NULL; 3397 #endif 3398 struct tcpcb *tp = 0; 3399 struct mbuf *am; 3400 int s; 3401 struct socket *oso; 3402 3403 s = splsoftnet(); 3404 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3405 splx(s); 3406 return (NULL); 3407 } 3408 3409 /* 3410 * Verify the sequence and ack numbers. Try getting the correct 3411 * response again. 3412 */ 3413 if ((th->th_ack != sc->sc_iss + 1) || 3414 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3415 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3416 (void) syn_cache_respond(sc, m); 3417 splx(s); 3418 return ((struct socket *)(-1)); 3419 } 3420 3421 /* Remove this cache entry */ 3422 SYN_CACHE_RM(sc); 3423 splx(s); 3424 3425 /* 3426 * Ok, create the full blown connection, and set things up 3427 * as they would have been set up if we had created the 3428 * connection when the SYN arrived. If we can't create 3429 * the connection, abort it. 3430 */ 3431 /* 3432 * inp still has the OLD in_pcb stuff, set the 3433 * v6-related flags on the new guy, too. This is 3434 * done particularly for the case where an AF_INET6 3435 * socket is bound only to a port, and a v4 connection 3436 * comes in on that port. 3437 * we also copy the flowinfo from the original pcb 3438 * to the new one. 3439 */ 3440 oso = so; 3441 so = sonewconn(so, SS_ISCONNECTED); 3442 if (so == NULL) 3443 goto resetandabort; 3444 3445 switch (so->so_proto->pr_domain->dom_family) { 3446 #ifdef INET 3447 case AF_INET: 3448 inp = sotoinpcb(so); 3449 break; 3450 #endif 3451 #ifdef INET6 3452 case AF_INET6: 3453 in6p = sotoin6pcb(so); 3454 break; 3455 #endif 3456 } 3457 switch (src->sa_family) { 3458 #ifdef INET 3459 case AF_INET: 3460 if (inp) { 3461 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3462 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3463 inp->inp_options = ip_srcroute(); 3464 in_pcbstate(inp, INP_BOUND); 3465 if (inp->inp_options == NULL) { 3466 inp->inp_options = sc->sc_ipopts; 3467 sc->sc_ipopts = NULL; 3468 } 3469 } 3470 #ifdef INET6 3471 else if (in6p) { 3472 /* IPv4 packet to AF_INET6 socket */ 3473 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3474 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3475 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3476 &in6p->in6p_laddr.s6_addr32[3], 3477 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3478 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3479 in6totcpcb(in6p)->t_family = AF_INET; 3480 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3481 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3482 else 3483 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3484 in6_pcbstate(in6p, IN6P_BOUND); 3485 } 3486 #endif 3487 break; 3488 #endif 3489 #ifdef INET6 3490 case AF_INET6: 3491 if (in6p) { 3492 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3493 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3494 in6_pcbstate(in6p, IN6P_BOUND); 3495 } 3496 break; 3497 #endif 3498 } 3499 #ifdef INET6 3500 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3501 struct in6pcb *oin6p = sotoin6pcb(oso); 3502 /* inherit socket options from the listening socket */ 3503 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3504 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3505 m_freem(in6p->in6p_options); 3506 in6p->in6p_options = 0; 3507 } 3508 ip6_savecontrol(in6p, &in6p->in6p_options, 3509 mtod(m, struct ip6_hdr *), m); 3510 } 3511 #endif 3512 3513 #if defined(IPSEC) || defined(FAST_IPSEC) 3514 /* 3515 * we make a copy of policy, instead of sharing the policy, 3516 * for better behavior in terms of SA lookup and dead SA removal. 3517 */ 3518 if (inp) { 3519 /* copy old policy into new socket's */ 3520 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3521 printf("tcp_input: could not copy policy\n"); 3522 } 3523 #ifdef INET6 3524 else if (in6p) { 3525 /* copy old policy into new socket's */ 3526 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3527 in6p->in6p_sp)) 3528 printf("tcp_input: could not copy policy\n"); 3529 } 3530 #endif 3531 #endif 3532 3533 /* 3534 * Give the new socket our cached route reference. 3535 */ 3536 if (inp) 3537 inp->inp_route = sc->sc_route4; /* struct assignment */ 3538 #ifdef INET6 3539 else 3540 in6p->in6p_route = sc->sc_route6; 3541 #endif 3542 sc->sc_route4.ro_rt = NULL; 3543 3544 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3545 if (am == NULL) 3546 goto resetandabort; 3547 MCLAIM(am, &tcp_mowner); 3548 am->m_len = src->sa_len; 3549 bcopy(src, mtod(am, caddr_t), src->sa_len); 3550 if (inp) { 3551 if (in_pcbconnect(inp, am)) { 3552 (void) m_free(am); 3553 goto resetandabort; 3554 } 3555 } 3556 #ifdef INET6 3557 else if (in6p) { 3558 if (src->sa_family == AF_INET) { 3559 /* IPv4 packet to AF_INET6 socket */ 3560 struct sockaddr_in6 *sin6; 3561 sin6 = mtod(am, struct sockaddr_in6 *); 3562 am->m_len = sizeof(*sin6); 3563 bzero(sin6, sizeof(*sin6)); 3564 sin6->sin6_family = AF_INET6; 3565 sin6->sin6_len = sizeof(*sin6); 3566 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3567 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3568 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3569 &sin6->sin6_addr.s6_addr32[3], 3570 sizeof(sin6->sin6_addr.s6_addr32[3])); 3571 } 3572 if (in6_pcbconnect(in6p, am)) { 3573 (void) m_free(am); 3574 goto resetandabort; 3575 } 3576 } 3577 #endif 3578 else { 3579 (void) m_free(am); 3580 goto resetandabort; 3581 } 3582 (void) m_free(am); 3583 3584 if (inp) 3585 tp = intotcpcb(inp); 3586 #ifdef INET6 3587 else if (in6p) 3588 tp = in6totcpcb(in6p); 3589 #endif 3590 else 3591 tp = NULL; 3592 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3593 if (sc->sc_request_r_scale != 15) { 3594 tp->requested_s_scale = sc->sc_requested_s_scale; 3595 tp->request_r_scale = sc->sc_request_r_scale; 3596 tp->snd_scale = sc->sc_requested_s_scale; 3597 tp->rcv_scale = sc->sc_request_r_scale; 3598 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3599 } 3600 if (sc->sc_flags & SCF_TIMESTAMP) 3601 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3602 tp->ts_timebase = sc->sc_timebase; 3603 3604 tp->t_template = tcp_template(tp); 3605 if (tp->t_template == 0) { 3606 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3607 so = NULL; 3608 m_freem(m); 3609 goto abort; 3610 } 3611 3612 tp->iss = sc->sc_iss; 3613 tp->irs = sc->sc_irs; 3614 tcp_sendseqinit(tp); 3615 tcp_rcvseqinit(tp); 3616 tp->t_state = TCPS_SYN_RECEIVED; 3617 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3618 tcpstat.tcps_accepts++; 3619 3620 #ifdef TCP_SIGNATURE 3621 if (sc->sc_flags & SCF_SIGNATURE) 3622 tp->t_flags |= TF_SIGNATURE; 3623 #endif 3624 3625 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3626 tp->t_ourmss = sc->sc_ourmaxseg; 3627 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3628 3629 /* 3630 * Initialize the initial congestion window. If we 3631 * had to retransmit the SYN,ACK, we must initialize cwnd 3632 * to 1 segment (i.e. the Loss Window). 3633 */ 3634 if (sc->sc_rxtshift) 3635 tp->snd_cwnd = tp->t_peermss; 3636 else { 3637 int ss = tcp_init_win; 3638 #ifdef INET 3639 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3640 ss = tcp_init_win_local; 3641 #endif 3642 #ifdef INET6 3643 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3644 ss = tcp_init_win_local; 3645 #endif 3646 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3647 } 3648 3649 tcp_rmx_rtt(tp); 3650 tp->snd_wl1 = sc->sc_irs; 3651 tp->rcv_up = sc->sc_irs + 1; 3652 3653 /* 3654 * This is what whould have happened in tcp_output() when 3655 * the SYN,ACK was sent. 3656 */ 3657 tp->snd_up = tp->snd_una; 3658 tp->snd_max = tp->snd_nxt = tp->iss+1; 3659 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3660 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3661 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3662 tp->last_ack_sent = tp->rcv_nxt; 3663 3664 tcpstat.tcps_sc_completed++; 3665 SYN_CACHE_PUT(sc); 3666 return (so); 3667 3668 resetandabort: 3669 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3670 abort: 3671 if (so != NULL) 3672 (void) soabort(so); 3673 SYN_CACHE_PUT(sc); 3674 tcpstat.tcps_sc_aborted++; 3675 return ((struct socket *)(-1)); 3676 } 3677 3678 /* 3679 * This function is called when we get a RST for a 3680 * non-existent connection, so that we can see if the 3681 * connection is in the syn cache. If it is, zap it. 3682 */ 3683 3684 void 3685 syn_cache_reset(src, dst, th) 3686 struct sockaddr *src; 3687 struct sockaddr *dst; 3688 struct tcphdr *th; 3689 { 3690 struct syn_cache *sc; 3691 struct syn_cache_head *scp; 3692 int s = splsoftnet(); 3693 3694 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3695 splx(s); 3696 return; 3697 } 3698 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3699 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3700 splx(s); 3701 return; 3702 } 3703 SYN_CACHE_RM(sc); 3704 splx(s); 3705 tcpstat.tcps_sc_reset++; 3706 SYN_CACHE_PUT(sc); 3707 } 3708 3709 void 3710 syn_cache_unreach(src, dst, th) 3711 struct sockaddr *src; 3712 struct sockaddr *dst; 3713 struct tcphdr *th; 3714 { 3715 struct syn_cache *sc; 3716 struct syn_cache_head *scp; 3717 int s; 3718 3719 s = splsoftnet(); 3720 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3721 splx(s); 3722 return; 3723 } 3724 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3725 if (ntohl (th->th_seq) != sc->sc_iss) { 3726 splx(s); 3727 return; 3728 } 3729 3730 /* 3731 * If we've retransmitted 3 times and this is our second error, 3732 * we remove the entry. Otherwise, we allow it to continue on. 3733 * This prevents us from incorrectly nuking an entry during a 3734 * spurious network outage. 3735 * 3736 * See tcp_notify(). 3737 */ 3738 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3739 sc->sc_flags |= SCF_UNREACH; 3740 splx(s); 3741 return; 3742 } 3743 3744 SYN_CACHE_RM(sc); 3745 splx(s); 3746 tcpstat.tcps_sc_unreach++; 3747 SYN_CACHE_PUT(sc); 3748 } 3749 3750 /* 3751 * Given a LISTEN socket and an inbound SYN request, add 3752 * this to the syn cache, and send back a segment: 3753 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3754 * to the source. 3755 * 3756 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3757 * Doing so would require that we hold onto the data and deliver it 3758 * to the application. However, if we are the target of a SYN-flood 3759 * DoS attack, an attacker could send data which would eventually 3760 * consume all available buffer space if it were ACKed. By not ACKing 3761 * the data, we avoid this DoS scenario. 3762 */ 3763 3764 int 3765 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi) 3766 struct sockaddr *src; 3767 struct sockaddr *dst; 3768 struct tcphdr *th; 3769 unsigned int hlen; 3770 struct socket *so; 3771 struct mbuf *m; 3772 u_char *optp; 3773 int optlen; 3774 struct tcp_opt_info *oi; 3775 { 3776 struct tcpcb tb, *tp; 3777 long win; 3778 struct syn_cache *sc; 3779 struct syn_cache_head *scp; 3780 struct mbuf *ipopts; 3781 struct tcp_opt_info opti; 3782 3783 tp = sototcpcb(so); 3784 3785 bzero(&opti, sizeof(opti)); 3786 3787 /* 3788 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3789 * 3790 * Note this check is performed in tcp_input() very early on. 3791 */ 3792 3793 /* 3794 * Initialize some local state. 3795 */ 3796 win = sbspace(&so->so_rcv); 3797 if (win > TCP_MAXWIN) 3798 win = TCP_MAXWIN; 3799 3800 switch (src->sa_family) { 3801 #ifdef INET 3802 case AF_INET: 3803 /* 3804 * Remember the IP options, if any. 3805 */ 3806 ipopts = ip_srcroute(); 3807 break; 3808 #endif 3809 default: 3810 ipopts = NULL; 3811 } 3812 3813 #ifdef TCP_SIGNATURE 3814 if (optp || (tp->t_flags & TF_SIGNATURE)) 3815 #else 3816 if (optp) 3817 #endif 3818 { 3819 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3820 #ifdef TCP_SIGNATURE 3821 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 3822 #endif 3823 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 3824 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 3825 return (0); 3826 } else 3827 tb.t_flags = 0; 3828 3829 /* 3830 * See if we already have an entry for this connection. 3831 * If we do, resend the SYN,ACK. We do not count this 3832 * as a retransmission (XXX though maybe we should). 3833 */ 3834 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3835 tcpstat.tcps_sc_dupesyn++; 3836 if (ipopts) { 3837 /* 3838 * If we were remembering a previous source route, 3839 * forget it and use the new one we've been given. 3840 */ 3841 if (sc->sc_ipopts) 3842 (void) m_free(sc->sc_ipopts); 3843 sc->sc_ipopts = ipopts; 3844 } 3845 sc->sc_timestamp = tb.ts_recent; 3846 if (syn_cache_respond(sc, m) == 0) { 3847 tcpstat.tcps_sndacks++; 3848 tcpstat.tcps_sndtotal++; 3849 } 3850 return (1); 3851 } 3852 3853 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3854 if (sc == NULL) { 3855 if (ipopts) 3856 (void) m_free(ipopts); 3857 return (0); 3858 } 3859 3860 /* 3861 * Fill in the cache, and put the necessary IP and TCP 3862 * options into the reply. 3863 */ 3864 bzero(sc, sizeof(struct syn_cache)); 3865 callout_init(&sc->sc_timer); 3866 bcopy(src, &sc->sc_src, src->sa_len); 3867 bcopy(dst, &sc->sc_dst, dst->sa_len); 3868 sc->sc_flags = 0; 3869 sc->sc_ipopts = ipopts; 3870 sc->sc_irs = th->th_seq; 3871 switch (src->sa_family) { 3872 #ifdef INET 3873 case AF_INET: 3874 { 3875 struct sockaddr_in *srcin = (void *) src; 3876 struct sockaddr_in *dstin = (void *) dst; 3877 3878 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 3879 &srcin->sin_addr, dstin->sin_port, 3880 srcin->sin_port, sizeof(dstin->sin_addr), 0); 3881 break; 3882 } 3883 #endif /* INET */ 3884 #ifdef INET6 3885 case AF_INET6: 3886 { 3887 struct sockaddr_in6 *srcin6 = (void *) src; 3888 struct sockaddr_in6 *dstin6 = (void *) dst; 3889 3890 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 3891 &srcin6->sin6_addr, dstin6->sin6_port, 3892 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 3893 break; 3894 } 3895 #endif /* INET6 */ 3896 } 3897 sc->sc_peermaxseg = oi->maxseg; 3898 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3899 m->m_pkthdr.rcvif : NULL, 3900 sc->sc_src.sa.sa_family); 3901 sc->sc_win = win; 3902 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 3903 sc->sc_timestamp = tb.ts_recent; 3904 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 3905 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 3906 sc->sc_flags |= SCF_TIMESTAMP; 3907 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3908 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3909 sc->sc_requested_s_scale = tb.requested_s_scale; 3910 sc->sc_request_r_scale = 0; 3911 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3912 TCP_MAXWIN << sc->sc_request_r_scale < 3913 so->so_rcv.sb_hiwat) 3914 sc->sc_request_r_scale++; 3915 } else { 3916 sc->sc_requested_s_scale = 15; 3917 sc->sc_request_r_scale = 15; 3918 } 3919 #ifdef TCP_SIGNATURE 3920 if (tb.t_flags & TF_SIGNATURE) 3921 sc->sc_flags |= SCF_SIGNATURE; 3922 #endif 3923 sc->sc_tp = tp; 3924 if (syn_cache_respond(sc, m) == 0) { 3925 syn_cache_insert(sc, tp); 3926 tcpstat.tcps_sndacks++; 3927 tcpstat.tcps_sndtotal++; 3928 } else { 3929 SYN_CACHE_PUT(sc); 3930 tcpstat.tcps_sc_dropped++; 3931 } 3932 return (1); 3933 } 3934 3935 int 3936 syn_cache_respond(sc, m) 3937 struct syn_cache *sc; 3938 struct mbuf *m; 3939 { 3940 struct route *ro; 3941 u_int8_t *optp; 3942 int optlen, error; 3943 u_int16_t tlen; 3944 struct ip *ip = NULL; 3945 #ifdef INET6 3946 struct ip6_hdr *ip6 = NULL; 3947 #endif 3948 struct tcpcb *tp; 3949 struct tcphdr *th; 3950 u_int hlen; 3951 struct socket *so; 3952 3953 switch (sc->sc_src.sa.sa_family) { 3954 case AF_INET: 3955 hlen = sizeof(struct ip); 3956 ro = &sc->sc_route4; 3957 break; 3958 #ifdef INET6 3959 case AF_INET6: 3960 hlen = sizeof(struct ip6_hdr); 3961 ro = (struct route *)&sc->sc_route6; 3962 break; 3963 #endif 3964 default: 3965 if (m) 3966 m_freem(m); 3967 return (EAFNOSUPPORT); 3968 } 3969 3970 /* Compute the size of the TCP options. */ 3971 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 3972 #ifdef TCP_SIGNATURE 3973 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 3974 #endif 3975 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 3976 3977 tlen = hlen + sizeof(struct tcphdr) + optlen; 3978 3979 /* 3980 * Create the IP+TCP header from scratch. 3981 */ 3982 if (m) 3983 m_freem(m); 3984 #ifdef DIAGNOSTIC 3985 if (max_linkhdr + tlen > MCLBYTES) 3986 return (ENOBUFS); 3987 #endif 3988 MGETHDR(m, M_DONTWAIT, MT_DATA); 3989 if (m && tlen > MHLEN) { 3990 MCLGET(m, M_DONTWAIT); 3991 if ((m->m_flags & M_EXT) == 0) { 3992 m_freem(m); 3993 m = NULL; 3994 } 3995 } 3996 if (m == NULL) 3997 return (ENOBUFS); 3998 MCLAIM(m, &tcp_tx_mowner); 3999 4000 /* Fixup the mbuf. */ 4001 m->m_data += max_linkhdr; 4002 m->m_len = m->m_pkthdr.len = tlen; 4003 if (sc->sc_tp) { 4004 tp = sc->sc_tp; 4005 if (tp->t_inpcb) 4006 so = tp->t_inpcb->inp_socket; 4007 #ifdef INET6 4008 else if (tp->t_in6pcb) 4009 so = tp->t_in6pcb->in6p_socket; 4010 #endif 4011 else 4012 so = NULL; 4013 } else 4014 so = NULL; 4015 m->m_pkthdr.rcvif = NULL; 4016 memset(mtod(m, u_char *), 0, tlen); 4017 4018 switch (sc->sc_src.sa.sa_family) { 4019 case AF_INET: 4020 ip = mtod(m, struct ip *); 4021 ip->ip_v = 4; 4022 ip->ip_dst = sc->sc_src.sin.sin_addr; 4023 ip->ip_src = sc->sc_dst.sin.sin_addr; 4024 ip->ip_p = IPPROTO_TCP; 4025 th = (struct tcphdr *)(ip + 1); 4026 th->th_dport = sc->sc_src.sin.sin_port; 4027 th->th_sport = sc->sc_dst.sin.sin_port; 4028 break; 4029 #ifdef INET6 4030 case AF_INET6: 4031 ip6 = mtod(m, struct ip6_hdr *); 4032 ip6->ip6_vfc = IPV6_VERSION; 4033 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4034 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4035 ip6->ip6_nxt = IPPROTO_TCP; 4036 /* ip6_plen will be updated in ip6_output() */ 4037 th = (struct tcphdr *)(ip6 + 1); 4038 th->th_dport = sc->sc_src.sin6.sin6_port; 4039 th->th_sport = sc->sc_dst.sin6.sin6_port; 4040 break; 4041 #endif 4042 default: 4043 th = NULL; 4044 } 4045 4046 th->th_seq = htonl(sc->sc_iss); 4047 th->th_ack = htonl(sc->sc_irs + 1); 4048 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4049 th->th_flags = TH_SYN|TH_ACK; 4050 th->th_win = htons(sc->sc_win); 4051 /* th_sum already 0 */ 4052 /* th_urp already 0 */ 4053 4054 /* Tack on the TCP options. */ 4055 optp = (u_int8_t *)(th + 1); 4056 *optp++ = TCPOPT_MAXSEG; 4057 *optp++ = 4; 4058 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4059 *optp++ = sc->sc_ourmaxseg & 0xff; 4060 4061 if (sc->sc_request_r_scale != 15) { 4062 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4063 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4064 sc->sc_request_r_scale); 4065 optp += 4; 4066 } 4067 4068 if (sc->sc_flags & SCF_TIMESTAMP) { 4069 u_int32_t *lp = (u_int32_t *)(optp); 4070 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4071 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4072 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4073 *lp = htonl(sc->sc_timestamp); 4074 optp += TCPOLEN_TSTAMP_APPA; 4075 } 4076 4077 #ifdef TCP_SIGNATURE 4078 if (sc->sc_flags & SCF_SIGNATURE) { 4079 struct secasvar *sav; 4080 u_int8_t *sigp; 4081 4082 sav = tcp_signature_getsav(m, th); 4083 4084 if (sav == NULL) { 4085 if (m) 4086 m_freem(m); 4087 return (EPERM); 4088 } 4089 4090 *optp++ = TCPOPT_SIGNATURE; 4091 *optp++ = TCPOLEN_SIGNATURE; 4092 sigp = optp; 4093 bzero(optp, TCP_SIGLEN); 4094 optp += TCP_SIGLEN; 4095 *optp++ = TCPOPT_NOP; 4096 *optp++ = TCPOPT_EOL; 4097 4098 (void)tcp_signature(m, th, hlen, sav, sigp); 4099 4100 key_sa_recordxfer(sav, m); 4101 #ifdef FAST_IPSEC 4102 KEY_FREESAV(&sav); 4103 #else 4104 key_freesav(sav); 4105 #endif 4106 } 4107 #endif 4108 4109 /* Compute the packet's checksum. */ 4110 switch (sc->sc_src.sa.sa_family) { 4111 case AF_INET: 4112 ip->ip_len = htons(tlen - hlen); 4113 th->th_sum = 0; 4114 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4115 break; 4116 #ifdef INET6 4117 case AF_INET6: 4118 ip6->ip6_plen = htons(tlen - hlen); 4119 th->th_sum = 0; 4120 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4121 break; 4122 #endif 4123 } 4124 4125 /* 4126 * Fill in some straggling IP bits. Note the stack expects 4127 * ip_len to be in host order, for convenience. 4128 */ 4129 switch (sc->sc_src.sa.sa_family) { 4130 #ifdef INET 4131 case AF_INET: 4132 ip->ip_len = htons(tlen); 4133 ip->ip_ttl = ip_defttl; 4134 /* XXX tos? */ 4135 break; 4136 #endif 4137 #ifdef INET6 4138 case AF_INET6: 4139 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4140 ip6->ip6_vfc |= IPV6_VERSION; 4141 ip6->ip6_plen = htons(tlen - hlen); 4142 /* ip6_hlim will be initialized afterwards */ 4143 /* XXX flowlabel? */ 4144 break; 4145 #endif 4146 } 4147 4148 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4149 tp = sc->sc_tp; 4150 4151 switch (sc->sc_src.sa.sa_family) { 4152 #ifdef INET 4153 case AF_INET: 4154 error = ip_output(m, sc->sc_ipopts, ro, 4155 (ip_mtudisc ? IP_MTUDISC : 0), 4156 (struct ip_moptions *)NULL, so); 4157 break; 4158 #endif 4159 #ifdef INET6 4160 case AF_INET6: 4161 ip6->ip6_hlim = in6_selecthlim(NULL, 4162 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 4163 4164 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0, 4165 (struct ip6_moptions *)0, so, NULL); 4166 break; 4167 #endif 4168 default: 4169 error = EAFNOSUPPORT; 4170 break; 4171 } 4172 return (error); 4173 } 4174