1 /* $NetBSD: tcp_input.c,v 1.221 2005/02/26 22:45:12 perry 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 1.221 2005/02/26 22:45:12 perry 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 (24U * 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(const struct ip *, const struct tcphdr *); 342 #endif 343 #ifdef INET6 344 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 345 #endif 346 347 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 348 349 POOL_INIT(tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", NULL); 350 351 int 352 tcp_reass(struct tcpcb *tp, struct tcphdr *th, struct mbuf *m, int *tlen) 353 { 354 struct ipqent *p, *q, *nq, *tiqe = NULL; 355 struct socket *so = NULL; 356 int pkt_flags; 357 tcp_seq pkt_seq; 358 unsigned pkt_len; 359 u_long rcvpartdupbyte = 0; 360 u_long rcvoobyte; 361 #ifdef TCP_REASS_COUNTERS 362 u_int count = 0; 363 #endif 364 365 if (tp->t_inpcb) 366 so = tp->t_inpcb->inp_socket; 367 #ifdef INET6 368 else if (tp->t_in6pcb) 369 so = tp->t_in6pcb->in6p_socket; 370 #endif 371 372 TCP_REASS_LOCK_CHECK(tp); 373 374 /* 375 * Call with th==0 after become established to 376 * force pre-ESTABLISHED data up to user socket. 377 */ 378 if (th == 0) 379 goto present; 380 381 rcvoobyte = *tlen; 382 /* 383 * Copy these to local variables because the tcpiphdr 384 * gets munged while we are collapsing mbufs. 385 */ 386 pkt_seq = th->th_seq; 387 pkt_len = *tlen; 388 pkt_flags = th->th_flags; 389 390 TCP_REASS_COUNTER_INCR(&tcp_reass_); 391 392 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 393 /* 394 * When we miss a packet, the vast majority of time we get 395 * packets that follow it in order. So optimize for that. 396 */ 397 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 398 p->ipqe_len += pkt_len; 399 p->ipqe_flags |= pkt_flags; 400 m_cat(p->ipre_mlast, m); 401 TRAVERSE(p->ipre_mlast); 402 m = NULL; 403 tiqe = p; 404 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 405 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 406 goto skip_replacement; 407 } 408 /* 409 * While we're here, if the pkt is completely beyond 410 * anything we have, just insert it at the tail. 411 */ 412 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 413 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 414 goto insert_it; 415 } 416 } 417 418 q = TAILQ_FIRST(&tp->segq); 419 420 if (q != NULL) { 421 /* 422 * If this segment immediately precedes the first out-of-order 423 * block, simply slap the segment in front of it and (mostly) 424 * skip the complicated logic. 425 */ 426 if (pkt_seq + pkt_len == q->ipqe_seq) { 427 q->ipqe_seq = pkt_seq; 428 q->ipqe_len += pkt_len; 429 q->ipqe_flags |= pkt_flags; 430 m_cat(m, q->ipqe_m); 431 q->ipqe_m = m; 432 q->ipre_mlast = m; /* last mbuf may have changed */ 433 TRAVERSE(q->ipre_mlast); 434 tiqe = q; 435 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 436 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 437 goto skip_replacement; 438 } 439 } else { 440 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 441 } 442 443 /* 444 * Find a segment which begins after this one does. 445 */ 446 for (p = NULL; q != NULL; q = nq) { 447 nq = TAILQ_NEXT(q, ipqe_q); 448 #ifdef TCP_REASS_COUNTERS 449 count++; 450 #endif 451 /* 452 * If the received segment is just right after this 453 * fragment, merge the two together and then check 454 * for further overlaps. 455 */ 456 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 457 #ifdef TCPREASS_DEBUG 458 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 459 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 460 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 461 #endif 462 pkt_len += q->ipqe_len; 463 pkt_flags |= q->ipqe_flags; 464 pkt_seq = q->ipqe_seq; 465 m_cat(q->ipre_mlast, m); 466 TRAVERSE(q->ipre_mlast); 467 m = q->ipqe_m; 468 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 469 goto free_ipqe; 470 } 471 /* 472 * If the received segment is completely past this 473 * fragment, we need to go the next fragment. 474 */ 475 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 476 p = q; 477 continue; 478 } 479 /* 480 * If the fragment is past the received segment, 481 * it (or any following) can't be concatenated. 482 */ 483 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 484 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 485 break; 486 } 487 488 /* 489 * We've received all the data in this segment before. 490 * mark it as a duplicate and return. 491 */ 492 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 493 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 494 tcpstat.tcps_rcvduppack++; 495 tcpstat.tcps_rcvdupbyte += pkt_len; 496 m_freem(m); 497 if (tiqe != NULL) 498 pool_put(&tcpipqent_pool, tiqe); 499 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 500 return (0); 501 } 502 /* 503 * Received segment completely overlaps this fragment 504 * so we drop the fragment (this keeps the temporal 505 * ordering of segments correct). 506 */ 507 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 508 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 509 rcvpartdupbyte += q->ipqe_len; 510 m_freem(q->ipqe_m); 511 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 512 goto free_ipqe; 513 } 514 /* 515 * RX'ed segment extends past the end of the 516 * fragment. Drop the overlapping bytes. Then 517 * merge the fragment and segment then treat as 518 * a longer received packet. 519 */ 520 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 521 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 522 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 523 #ifdef TCPREASS_DEBUG 524 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 525 tp, overlap, 526 pkt_seq, pkt_seq + pkt_len, pkt_len); 527 #endif 528 m_adj(m, overlap); 529 rcvpartdupbyte += overlap; 530 m_cat(q->ipre_mlast, m); 531 TRAVERSE(q->ipre_mlast); 532 m = q->ipqe_m; 533 pkt_seq = q->ipqe_seq; 534 pkt_len += q->ipqe_len - overlap; 535 rcvoobyte -= overlap; 536 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 537 goto free_ipqe; 538 } 539 /* 540 * RX'ed segment extends past the front of the 541 * fragment. Drop the overlapping bytes on the 542 * received packet. The packet will then be 543 * contatentated with this fragment a bit later. 544 */ 545 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 546 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 547 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 548 #ifdef TCPREASS_DEBUG 549 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 550 tp, overlap, 551 pkt_seq, pkt_seq + pkt_len, pkt_len); 552 #endif 553 m_adj(m, -overlap); 554 pkt_len -= overlap; 555 rcvpartdupbyte += overlap; 556 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 557 rcvoobyte -= overlap; 558 } 559 /* 560 * If the received segment immediates precedes this 561 * fragment then tack the fragment onto this segment 562 * and reinsert the data. 563 */ 564 if (q->ipqe_seq == pkt_seq + pkt_len) { 565 #ifdef TCPREASS_DEBUG 566 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 567 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 568 pkt_seq, pkt_seq + pkt_len, pkt_len); 569 #endif 570 pkt_len += q->ipqe_len; 571 pkt_flags |= q->ipqe_flags; 572 m_cat(m, q->ipqe_m); 573 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 574 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 575 if (tiqe == NULL) 576 tiqe = q; 577 else 578 pool_put(&tcpipqent_pool, q); 579 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 580 break; 581 } 582 /* 583 * If the fragment is before the segment, remember it. 584 * When this loop is terminated, p will contain the 585 * pointer to fragment that is right before the received 586 * segment. 587 */ 588 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 589 p = q; 590 591 continue; 592 593 /* 594 * This is a common operation. It also will allow 595 * to save doing a malloc/free in most instances. 596 */ 597 free_ipqe: 598 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 599 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 600 if (tiqe == NULL) 601 tiqe = q; 602 else 603 pool_put(&tcpipqent_pool, q); 604 } 605 606 #ifdef TCP_REASS_COUNTERS 607 if (count > 7) 608 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 609 else if (count > 0) 610 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 611 #endif 612 613 insert_it: 614 615 /* 616 * Allocate a new queue entry since the received segment did not 617 * collapse onto any other out-of-order block; thus we are allocating 618 * a new block. If it had collapsed, tiqe would not be NULL and 619 * we would be reusing it. 620 * XXX If we can't, just drop the packet. XXX 621 */ 622 if (tiqe == NULL) { 623 tiqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 624 if (tiqe == NULL) { 625 tcpstat.tcps_rcvmemdrop++; 626 m_freem(m); 627 return (0); 628 } 629 } 630 631 /* 632 * Update the counters. 633 */ 634 tcpstat.tcps_rcvoopack++; 635 tcpstat.tcps_rcvoobyte += rcvoobyte; 636 if (rcvpartdupbyte) { 637 tcpstat.tcps_rcvpartduppack++; 638 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 639 } 640 641 /* 642 * Insert the new fragment queue entry into both queues. 643 */ 644 tiqe->ipqe_m = m; 645 tiqe->ipre_mlast = m; 646 tiqe->ipqe_seq = pkt_seq; 647 tiqe->ipqe_len = pkt_len; 648 tiqe->ipqe_flags = pkt_flags; 649 if (p == NULL) { 650 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 651 #ifdef TCPREASS_DEBUG 652 if (tiqe->ipqe_seq != tp->rcv_nxt) 653 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 654 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 655 #endif 656 } else { 657 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 658 #ifdef TCPREASS_DEBUG 659 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 660 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 661 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 662 #endif 663 } 664 665 skip_replacement: 666 667 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 668 669 present: 670 /* 671 * Present data to user, advancing rcv_nxt through 672 * completed sequence space. 673 */ 674 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 675 return (0); 676 q = TAILQ_FIRST(&tp->segq); 677 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 678 return (0); 679 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 680 return (0); 681 682 tp->rcv_nxt += q->ipqe_len; 683 pkt_flags = q->ipqe_flags & TH_FIN; 684 ND6_HINT(tp); 685 686 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 687 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 688 if (so->so_state & SS_CANTRCVMORE) 689 m_freem(q->ipqe_m); 690 else 691 sbappendstream(&so->so_rcv, q->ipqe_m); 692 pool_put(&tcpipqent_pool, q); 693 sorwakeup(so); 694 return (pkt_flags); 695 } 696 697 #ifdef INET6 698 int 699 tcp6_input(struct mbuf **mp, int *offp, int proto) 700 { 701 struct mbuf *m = *mp; 702 703 /* 704 * draft-itojun-ipv6-tcp-to-anycast 705 * better place to put this in? 706 */ 707 if (m->m_flags & M_ANYCAST6) { 708 struct ip6_hdr *ip6; 709 if (m->m_len < sizeof(struct ip6_hdr)) { 710 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 711 tcpstat.tcps_rcvshort++; 712 return IPPROTO_DONE; 713 } 714 } 715 ip6 = mtod(m, struct ip6_hdr *); 716 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 717 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 718 return IPPROTO_DONE; 719 } 720 721 tcp_input(m, *offp, proto); 722 return IPPROTO_DONE; 723 } 724 #endif 725 726 #ifdef INET 727 static void 728 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 729 { 730 char src[4*sizeof "123"]; 731 char dst[4*sizeof "123"]; 732 733 if (ip) { 734 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 735 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 736 } 737 else { 738 strlcpy(src, "(unknown)", sizeof(src)); 739 strlcpy(dst, "(unknown)", sizeof(dst)); 740 } 741 log(LOG_INFO, 742 "Connection attempt to TCP %s:%d from %s:%d\n", 743 dst, ntohs(th->th_dport), 744 src, ntohs(th->th_sport)); 745 } 746 #endif 747 748 #ifdef INET6 749 static void 750 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 751 { 752 char src[INET6_ADDRSTRLEN]; 753 char dst[INET6_ADDRSTRLEN]; 754 755 if (ip6) { 756 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 757 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 758 } 759 else { 760 strlcpy(src, "(unknown v6)", sizeof(src)); 761 strlcpy(dst, "(unknown v6)", sizeof(dst)); 762 } 763 log(LOG_INFO, 764 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 765 dst, ntohs(th->th_dport), 766 src, ntohs(th->th_sport)); 767 } 768 #endif 769 770 /* 771 * Checksum extended TCP header and data. 772 */ 773 int 774 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, int toff, 775 int off, int tlen) 776 { 777 778 /* 779 * XXX it's better to record and check if this mbuf is 780 * already checked. 781 */ 782 783 switch (af) { 784 #ifdef INET 785 case AF_INET: 786 switch (m->m_pkthdr.csum_flags & 787 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 788 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 789 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 790 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 791 goto badcsum; 792 793 case M_CSUM_TCPv4|M_CSUM_DATA: { 794 u_int32_t hw_csum = m->m_pkthdr.csum_data; 795 796 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 797 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 798 const struct ip *ip = 799 mtod(m, const struct ip *); 800 801 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 802 ip->ip_dst.s_addr, 803 htons(hw_csum + tlen + off + IPPROTO_TCP)); 804 } 805 if ((hw_csum ^ 0xffff) != 0) 806 goto badcsum; 807 break; 808 } 809 810 case M_CSUM_TCPv4: 811 /* Checksum was okay. */ 812 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 813 break; 814 815 default: 816 /* 817 * Must compute it ourselves. Maybe skip checksum 818 * on loopback interfaces. 819 */ 820 if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & 821 IFF_LOOPBACK) || 822 tcp_do_loopback_cksum)) { 823 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 824 if (in4_cksum(m, IPPROTO_TCP, toff, 825 tlen + off) != 0) 826 goto badcsum; 827 } 828 break; 829 } 830 break; 831 #endif /* INET4 */ 832 833 #ifdef INET6 834 case AF_INET6: 835 if (__predict_true((m->m_flags & M_LOOP) == 0 || 836 tcp_do_loopback_cksum)) { 837 if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 838 goto badcsum; 839 } 840 break; 841 #endif /* INET6 */ 842 } 843 844 return 0; 845 846 badcsum: 847 tcpstat.tcps_rcvbadsum++; 848 return -1; 849 } 850 851 /* 852 * TCP input routine, follows pages 65-76 of the 853 * protocol specification dated September, 1981 very closely. 854 */ 855 void 856 tcp_input(struct mbuf *m, ...) 857 { 858 struct tcphdr *th; 859 struct ip *ip; 860 struct inpcb *inp; 861 #ifdef INET6 862 struct ip6_hdr *ip6; 863 struct in6pcb *in6p; 864 #endif 865 u_int8_t *optp = NULL; 866 int optlen = 0; 867 int len, tlen, toff, hdroptlen = 0; 868 struct tcpcb *tp = 0; 869 int tiflags; 870 struct socket *so = NULL; 871 int todrop, dupseg, acked, ourfinisacked, needoutput = 0; 872 #ifdef TCP_DEBUG 873 short ostate = 0; 874 #endif 875 int iss = 0; 876 u_long tiwin; 877 struct tcp_opt_info opti; 878 int off, iphlen; 879 va_list ap; 880 int af; /* af on the wire */ 881 struct mbuf *tcp_saveti = NULL; 882 883 MCLAIM(m, &tcp_rx_mowner); 884 va_start(ap, m); 885 toff = va_arg(ap, int); 886 (void)va_arg(ap, int); /* ignore value, advance ap */ 887 va_end(ap); 888 889 tcpstat.tcps_rcvtotal++; 890 891 bzero(&opti, sizeof(opti)); 892 opti.ts_present = 0; 893 opti.maxseg = 0; 894 895 /* 896 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 897 * 898 * TCP is, by definition, unicast, so we reject all 899 * multicast outright. 900 * 901 * Note, there are additional src/dst address checks in 902 * the AF-specific code below. 903 */ 904 if (m->m_flags & (M_BCAST|M_MCAST)) { 905 /* XXX stat */ 906 goto drop; 907 } 908 #ifdef INET6 909 if (m->m_flags & M_ANYCAST6) { 910 /* XXX stat */ 911 goto drop; 912 } 913 #endif 914 915 /* 916 * Get IP and TCP header. 917 * Note: IP leaves IP header in first mbuf. 918 */ 919 ip = mtod(m, struct ip *); 920 #ifdef INET6 921 ip6 = NULL; 922 #endif 923 switch (ip->ip_v) { 924 #ifdef INET 925 case 4: 926 af = AF_INET; 927 iphlen = sizeof(struct ip); 928 ip = mtod(m, struct ip *); 929 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 930 sizeof(struct tcphdr)); 931 if (th == NULL) { 932 tcpstat.tcps_rcvshort++; 933 return; 934 } 935 /* We do the checksum after PCB lookup... */ 936 len = ntohs(ip->ip_len); 937 tlen = len - toff; 938 break; 939 #endif 940 #ifdef INET6 941 case 6: 942 ip = NULL; 943 iphlen = sizeof(struct ip6_hdr); 944 af = AF_INET6; 945 ip6 = mtod(m, struct ip6_hdr *); 946 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 947 sizeof(struct tcphdr)); 948 if (th == NULL) { 949 tcpstat.tcps_rcvshort++; 950 return; 951 } 952 953 /* Be proactive about malicious use of IPv4 mapped address */ 954 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 955 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 956 /* XXX stat */ 957 goto drop; 958 } 959 960 /* 961 * Be proactive about unspecified IPv6 address in source. 962 * As we use all-zero to indicate unbounded/unconnected pcb, 963 * unspecified IPv6 address can be used to confuse us. 964 * 965 * Note that packets with unspecified IPv6 destination is 966 * already dropped in ip6_input. 967 */ 968 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 969 /* XXX stat */ 970 goto drop; 971 } 972 973 /* 974 * Make sure destination address is not multicast. 975 * Source address checked in ip6_input(). 976 */ 977 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 978 /* XXX stat */ 979 goto drop; 980 } 981 982 /* We do the checksum after PCB lookup... */ 983 len = m->m_pkthdr.len; 984 tlen = len - toff; 985 break; 986 #endif 987 default: 988 m_freem(m); 989 return; 990 } 991 992 KASSERT(TCP_HDR_ALIGNED_P(th)); 993 994 /* 995 * Check that TCP offset makes sense, 996 * pull out TCP options and adjust length. XXX 997 */ 998 off = th->th_off << 2; 999 if (off < sizeof (struct tcphdr) || off > tlen) { 1000 tcpstat.tcps_rcvbadoff++; 1001 goto drop; 1002 } 1003 tlen -= off; 1004 1005 /* 1006 * tcp_input() has been modified to use tlen to mean the TCP data 1007 * length throughout the function. Other functions can use 1008 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1009 * rja 1010 */ 1011 1012 if (off > sizeof (struct tcphdr)) { 1013 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1014 if (th == NULL) { 1015 tcpstat.tcps_rcvshort++; 1016 return; 1017 } 1018 /* 1019 * NOTE: ip/ip6 will not be affected by m_pulldown() 1020 * (as they're before toff) and we don't need to update those. 1021 */ 1022 KASSERT(TCP_HDR_ALIGNED_P(th)); 1023 optlen = off - sizeof (struct tcphdr); 1024 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1025 /* 1026 * Do quick retrieval of timestamp options ("options 1027 * prediction?"). If timestamp is the only option and it's 1028 * formatted as recommended in RFC 1323 appendix A, we 1029 * quickly get the values now and not bother calling 1030 * tcp_dooptions(), etc. 1031 */ 1032 if ((optlen == TCPOLEN_TSTAMP_APPA || 1033 (optlen > TCPOLEN_TSTAMP_APPA && 1034 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1035 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1036 (th->th_flags & TH_SYN) == 0) { 1037 opti.ts_present = 1; 1038 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1039 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1040 optp = NULL; /* we've parsed the options */ 1041 } 1042 } 1043 tiflags = th->th_flags; 1044 1045 /* 1046 * Locate pcb for segment. 1047 */ 1048 findpcb: 1049 inp = NULL; 1050 #ifdef INET6 1051 in6p = NULL; 1052 #endif 1053 switch (af) { 1054 #ifdef INET 1055 case AF_INET: 1056 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1057 ip->ip_dst, th->th_dport); 1058 if (inp == 0) { 1059 ++tcpstat.tcps_pcbhashmiss; 1060 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1061 } 1062 #ifdef INET6 1063 if (inp == 0) { 1064 struct in6_addr s, d; 1065 1066 /* mapped addr case */ 1067 bzero(&s, sizeof(s)); 1068 s.s6_addr16[5] = htons(0xffff); 1069 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1070 bzero(&d, sizeof(d)); 1071 d.s6_addr16[5] = htons(0xffff); 1072 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1073 in6p = in6_pcblookup_connect(&tcbtable, &s, 1074 th->th_sport, &d, th->th_dport, 0); 1075 if (in6p == 0) { 1076 ++tcpstat.tcps_pcbhashmiss; 1077 in6p = in6_pcblookup_bind(&tcbtable, &d, 1078 th->th_dport, 0); 1079 } 1080 } 1081 #endif 1082 #ifndef INET6 1083 if (inp == 0) 1084 #else 1085 if (inp == 0 && in6p == 0) 1086 #endif 1087 { 1088 ++tcpstat.tcps_noport; 1089 if (tcp_log_refused && 1090 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1091 tcp4_log_refused(ip, th); 1092 } 1093 TCP_FIELDS_TO_HOST(th); 1094 goto dropwithreset_ratelim; 1095 } 1096 #if defined(IPSEC) || defined(FAST_IPSEC) 1097 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1098 ipsec4_in_reject(m, inp)) { 1099 ipsecstat.in_polvio++; 1100 goto drop; 1101 } 1102 #ifdef INET6 1103 else if (in6p && 1104 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1105 ipsec4_in_reject_so(m, in6p->in6p_socket)) { 1106 ipsecstat.in_polvio++; 1107 goto drop; 1108 } 1109 #endif 1110 #endif /*IPSEC*/ 1111 break; 1112 #endif /*INET*/ 1113 #ifdef INET6 1114 case AF_INET6: 1115 { 1116 int faith; 1117 1118 #if defined(NFAITH) && NFAITH > 0 1119 faith = faithprefix(&ip6->ip6_dst); 1120 #else 1121 faith = 0; 1122 #endif 1123 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1124 th->th_sport, &ip6->ip6_dst, th->th_dport, faith); 1125 if (in6p == NULL) { 1126 ++tcpstat.tcps_pcbhashmiss; 1127 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1128 th->th_dport, faith); 1129 } 1130 if (in6p == NULL) { 1131 ++tcpstat.tcps_noport; 1132 if (tcp_log_refused && 1133 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1134 tcp6_log_refused(ip6, th); 1135 } 1136 TCP_FIELDS_TO_HOST(th); 1137 goto dropwithreset_ratelim; 1138 } 1139 #if defined(IPSEC) || defined(FAST_IPSEC) 1140 if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1141 ipsec6_in_reject(m, in6p)) { 1142 ipsec6stat.in_polvio++; 1143 goto drop; 1144 } 1145 #endif /*IPSEC*/ 1146 break; 1147 } 1148 #endif 1149 } 1150 1151 /* 1152 * If the state is CLOSED (i.e., TCB does not exist) then 1153 * all data in the incoming segment is discarded. 1154 * If the TCB exists but is in CLOSED state, it is embryonic, 1155 * but should either do a listen or a connect soon. 1156 */ 1157 tp = NULL; 1158 so = NULL; 1159 if (inp) { 1160 tp = intotcpcb(inp); 1161 so = inp->inp_socket; 1162 } 1163 #ifdef INET6 1164 else if (in6p) { 1165 tp = in6totcpcb(in6p); 1166 so = in6p->in6p_socket; 1167 } 1168 #endif 1169 if (tp == 0) { 1170 TCP_FIELDS_TO_HOST(th); 1171 goto dropwithreset_ratelim; 1172 } 1173 if (tp->t_state == TCPS_CLOSED) 1174 goto drop; 1175 1176 /* 1177 * Checksum extended TCP header and data. 1178 */ 1179 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1180 goto badcsum; 1181 1182 TCP_FIELDS_TO_HOST(th); 1183 1184 /* Unscale the window into a 32-bit value. */ 1185 if ((tiflags & TH_SYN) == 0) 1186 tiwin = th->th_win << tp->snd_scale; 1187 else 1188 tiwin = th->th_win; 1189 1190 #ifdef INET6 1191 /* save packet options if user wanted */ 1192 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1193 if (in6p->in6p_options) { 1194 m_freem(in6p->in6p_options); 1195 in6p->in6p_options = 0; 1196 } 1197 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1198 } 1199 #endif 1200 1201 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1202 union syn_cache_sa src; 1203 union syn_cache_sa dst; 1204 1205 bzero(&src, sizeof(src)); 1206 bzero(&dst, sizeof(dst)); 1207 switch (af) { 1208 #ifdef INET 1209 case AF_INET: 1210 src.sin.sin_len = sizeof(struct sockaddr_in); 1211 src.sin.sin_family = AF_INET; 1212 src.sin.sin_addr = ip->ip_src; 1213 src.sin.sin_port = th->th_sport; 1214 1215 dst.sin.sin_len = sizeof(struct sockaddr_in); 1216 dst.sin.sin_family = AF_INET; 1217 dst.sin.sin_addr = ip->ip_dst; 1218 dst.sin.sin_port = th->th_dport; 1219 break; 1220 #endif 1221 #ifdef INET6 1222 case AF_INET6: 1223 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1224 src.sin6.sin6_family = AF_INET6; 1225 src.sin6.sin6_addr = ip6->ip6_src; 1226 src.sin6.sin6_port = th->th_sport; 1227 1228 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1229 dst.sin6.sin6_family = AF_INET6; 1230 dst.sin6.sin6_addr = ip6->ip6_dst; 1231 dst.sin6.sin6_port = th->th_dport; 1232 break; 1233 #endif /* INET6 */ 1234 default: 1235 goto badsyn; /*sanity*/ 1236 } 1237 1238 if (so->so_options & SO_DEBUG) { 1239 #ifdef TCP_DEBUG 1240 ostate = tp->t_state; 1241 #endif 1242 1243 tcp_saveti = NULL; 1244 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1245 goto nosave; 1246 1247 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1248 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1249 if (!tcp_saveti) 1250 goto nosave; 1251 } else { 1252 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1253 if (!tcp_saveti) 1254 goto nosave; 1255 MCLAIM(m, &tcp_mowner); 1256 tcp_saveti->m_len = iphlen; 1257 m_copydata(m, 0, iphlen, 1258 mtod(tcp_saveti, caddr_t)); 1259 } 1260 1261 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1262 m_freem(tcp_saveti); 1263 tcp_saveti = NULL; 1264 } else { 1265 tcp_saveti->m_len += sizeof(struct tcphdr); 1266 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1267 sizeof(struct tcphdr)); 1268 } 1269 nosave:; 1270 } 1271 if (so->so_options & SO_ACCEPTCONN) { 1272 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1273 if (tiflags & TH_RST) { 1274 syn_cache_reset(&src.sa, &dst.sa, th); 1275 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1276 (TH_ACK|TH_SYN)) { 1277 /* 1278 * Received a SYN,ACK. This should 1279 * never happen while we are in 1280 * LISTEN. Send an RST. 1281 */ 1282 goto badsyn; 1283 } else if (tiflags & TH_ACK) { 1284 so = syn_cache_get(&src.sa, &dst.sa, 1285 th, toff, tlen, so, m); 1286 if (so == NULL) { 1287 /* 1288 * We don't have a SYN for 1289 * this ACK; send an RST. 1290 */ 1291 goto badsyn; 1292 } else if (so == 1293 (struct socket *)(-1)) { 1294 /* 1295 * We were unable to create 1296 * the connection. If the 1297 * 3-way handshake was 1298 * completed, and RST has 1299 * been sent to the peer. 1300 * Since the mbuf might be 1301 * in use for the reply, 1302 * do not free it. 1303 */ 1304 m = NULL; 1305 } else { 1306 /* 1307 * We have created a 1308 * full-blown connection. 1309 */ 1310 tp = NULL; 1311 inp = NULL; 1312 #ifdef INET6 1313 in6p = NULL; 1314 #endif 1315 switch (so->so_proto->pr_domain->dom_family) { 1316 #ifdef INET 1317 case AF_INET: 1318 inp = sotoinpcb(so); 1319 tp = intotcpcb(inp); 1320 break; 1321 #endif 1322 #ifdef INET6 1323 case AF_INET6: 1324 in6p = sotoin6pcb(so); 1325 tp = in6totcpcb(in6p); 1326 break; 1327 #endif 1328 } 1329 if (tp == NULL) 1330 goto badsyn; /*XXX*/ 1331 tiwin <<= tp->snd_scale; 1332 goto after_listen; 1333 } 1334 } else { 1335 /* 1336 * None of RST, SYN or ACK was set. 1337 * This is an invalid packet for a 1338 * TCB in LISTEN state. Send a RST. 1339 */ 1340 goto badsyn; 1341 } 1342 } else { 1343 /* 1344 * Received a SYN. 1345 */ 1346 1347 #ifdef INET6 1348 /* 1349 * If deprecated address is forbidden, we do 1350 * not accept SYN to deprecated interface 1351 * address to prevent any new inbound 1352 * connection from getting established. 1353 * When we do not accept SYN, we send a TCP 1354 * RST, with deprecated source address (instead 1355 * of dropping it). We compromise it as it is 1356 * much better for peer to send a RST, and 1357 * RST will be the final packet for the 1358 * exchange. 1359 * 1360 * If we do not forbid deprecated addresses, we 1361 * accept the SYN packet. RFC2462 does not 1362 * suggest dropping SYN in this case. 1363 * If we decipher RFC2462 5.5.4, it says like 1364 * this: 1365 * 1. use of deprecated addr with existing 1366 * communication is okay - "SHOULD continue 1367 * to be used" 1368 * 2. use of it with new communication: 1369 * (2a) "SHOULD NOT be used if alternate 1370 * address with sufficient scope is 1371 * available" 1372 * (2b) nothing mentioned otherwise. 1373 * Here we fall into (2b) case as we have no 1374 * choice in our source address selection - we 1375 * must obey the peer. 1376 * 1377 * The wording in RFC2462 is confusing, and 1378 * there are multiple description text for 1379 * deprecated address handling - worse, they 1380 * are not exactly the same. I believe 5.5.4 1381 * is the best one, so we follow 5.5.4. 1382 */ 1383 if (af == AF_INET6 && !ip6_use_deprecated) { 1384 struct in6_ifaddr *ia6; 1385 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1386 &ip6->ip6_dst)) && 1387 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1388 tp = NULL; 1389 goto dropwithreset; 1390 } 1391 } 1392 #endif 1393 1394 #ifdef IPSEC 1395 switch (af) { 1396 #ifdef INET 1397 case AF_INET: 1398 if (ipsec4_in_reject_so(m, so)) { 1399 ipsecstat.in_polvio++; 1400 tp = NULL; 1401 goto dropwithreset; 1402 } 1403 break; 1404 #endif 1405 #ifdef INET6 1406 case AF_INET6: 1407 if (ipsec6_in_reject_so(m, so)) { 1408 ipsec6stat.in_polvio++; 1409 tp = NULL; 1410 goto dropwithreset; 1411 } 1412 break; 1413 #endif 1414 } 1415 #endif 1416 1417 /* 1418 * LISTEN socket received a SYN 1419 * from itself? This can't possibly 1420 * be valid; drop the packet. 1421 */ 1422 if (th->th_sport == th->th_dport) { 1423 int i; 1424 1425 switch (af) { 1426 #ifdef INET 1427 case AF_INET: 1428 i = in_hosteq(ip->ip_src, ip->ip_dst); 1429 break; 1430 #endif 1431 #ifdef INET6 1432 case AF_INET6: 1433 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1434 break; 1435 #endif 1436 default: 1437 i = 1; 1438 } 1439 if (i) { 1440 tcpstat.tcps_badsyn++; 1441 goto drop; 1442 } 1443 } 1444 1445 /* 1446 * SYN looks ok; create compressed TCP 1447 * state for it. 1448 */ 1449 if (so->so_qlen <= so->so_qlimit && 1450 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1451 so, m, optp, optlen, &opti)) 1452 m = NULL; 1453 } 1454 goto drop; 1455 } 1456 } 1457 1458 after_listen: 1459 #ifdef DIAGNOSTIC 1460 /* 1461 * Should not happen now that all embryonic connections 1462 * are handled with compressed state. 1463 */ 1464 if (tp->t_state == TCPS_LISTEN) 1465 panic("tcp_input: TCPS_LISTEN"); 1466 #endif 1467 1468 /* 1469 * Segment received on connection. 1470 * Reset idle time and keep-alive timer. 1471 */ 1472 tp->t_rcvtime = tcp_now; 1473 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1474 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1475 1476 /* 1477 * Process options. 1478 */ 1479 #ifdef TCP_SIGNATURE 1480 if (optp || (tp->t_flags & TF_SIGNATURE)) 1481 #else 1482 if (optp) 1483 #endif 1484 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1485 goto drop; 1486 1487 if (opti.ts_present && opti.ts_ecr) { 1488 /* 1489 * Calculate the RTT from the returned time stamp and the 1490 * connection's time base. If the time stamp is later than 1491 * the current time, or is extremely old, fall back to non-1323 1492 * RTT calculation. Since ts_ecr is unsigned, we can test both 1493 * at the same time. 1494 */ 1495 opti.ts_ecr = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1496 if (opti.ts_ecr > TCP_PAWS_IDLE) 1497 opti.ts_ecr = 0; 1498 } 1499 1500 /* 1501 * Header prediction: check for the two common cases 1502 * of a uni-directional data xfer. If the packet has 1503 * no control flags, is in-sequence, the window didn't 1504 * change and we're not retransmitting, it's a 1505 * candidate. If the length is zero and the ack moved 1506 * forward, we're the sender side of the xfer. Just 1507 * free the data acked & wake any higher level process 1508 * that was blocked waiting for space. If the length 1509 * is non-zero and the ack didn't move, we're the 1510 * receiver side. If we're getting packets in-order 1511 * (the reassembly queue is empty), add the data to 1512 * the socket buffer and note that we need a delayed ack. 1513 */ 1514 if (tp->t_state == TCPS_ESTABLISHED && 1515 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1516 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1517 th->th_seq == tp->rcv_nxt && 1518 tiwin && tiwin == tp->snd_wnd && 1519 tp->snd_nxt == tp->snd_max) { 1520 1521 /* 1522 * If last ACK falls within this segment's sequence numbers, 1523 * record the timestamp. 1524 */ 1525 if (opti.ts_present && 1526 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1527 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1528 tp->ts_recent_age = tcp_now; 1529 tp->ts_recent = opti.ts_val; 1530 } 1531 1532 if (tlen == 0) { 1533 /* Ack prediction. */ 1534 if (SEQ_GT(th->th_ack, tp->snd_una) && 1535 SEQ_LEQ(th->th_ack, tp->snd_max) && 1536 tp->snd_cwnd >= tp->snd_wnd && 1537 tp->t_partialacks < 0) { 1538 /* 1539 * this is a pure ack for outstanding data. 1540 */ 1541 ++tcpstat.tcps_predack; 1542 if (opti.ts_present && opti.ts_ecr) 1543 tcp_xmit_timer(tp, opti.ts_ecr); 1544 else if (tp->t_rtttime && 1545 SEQ_GT(th->th_ack, tp->t_rtseq)) 1546 tcp_xmit_timer(tp, 1547 tcp_now - tp->t_rtttime); 1548 acked = th->th_ack - tp->snd_una; 1549 tcpstat.tcps_rcvackpack++; 1550 tcpstat.tcps_rcvackbyte += acked; 1551 ND6_HINT(tp); 1552 1553 if (acked > (tp->t_lastoff - tp->t_inoff)) 1554 tp->t_lastm = NULL; 1555 sbdrop(&so->so_snd, acked); 1556 tp->t_lastoff -= acked; 1557 1558 tp->snd_una = th->th_ack; 1559 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1560 tp->snd_high = tp->snd_una; 1561 m_freem(m); 1562 1563 /* 1564 * If all outstanding data are acked, stop 1565 * retransmit timer, otherwise restart timer 1566 * using current (possibly backed-off) value. 1567 * If process is waiting for space, 1568 * wakeup/selwakeup/signal. If data 1569 * are ready to send, let tcp_output 1570 * decide between more output or persist. 1571 */ 1572 if (tp->snd_una == tp->snd_max) 1573 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1574 else if (TCP_TIMER_ISARMED(tp, 1575 TCPT_PERSIST) == 0) 1576 TCP_TIMER_ARM(tp, TCPT_REXMT, 1577 tp->t_rxtcur); 1578 1579 sowwakeup(so); 1580 if (so->so_snd.sb_cc) 1581 (void) tcp_output(tp); 1582 if (tcp_saveti) 1583 m_freem(tcp_saveti); 1584 return; 1585 } 1586 } else if (th->th_ack == tp->snd_una && 1587 TAILQ_FIRST(&tp->segq) == NULL && 1588 tlen <= sbspace(&so->so_rcv)) { 1589 /* 1590 * this is a pure, in-sequence data packet 1591 * with nothing on the reassembly queue and 1592 * we have enough buffer space to take it. 1593 */ 1594 ++tcpstat.tcps_preddat; 1595 tp->rcv_nxt += tlen; 1596 tcpstat.tcps_rcvpack++; 1597 tcpstat.tcps_rcvbyte += tlen; 1598 ND6_HINT(tp); 1599 /* 1600 * Drop TCP, IP headers and TCP options then add data 1601 * to socket buffer. 1602 */ 1603 if (so->so_state & SS_CANTRCVMORE) 1604 m_freem(m); 1605 else { 1606 m_adj(m, toff + off); 1607 sbappendstream(&so->so_rcv, m); 1608 } 1609 sorwakeup(so); 1610 TCP_SETUP_ACK(tp, th); 1611 if (tp->t_flags & TF_ACKNOW) 1612 (void) tcp_output(tp); 1613 if (tcp_saveti) 1614 m_freem(tcp_saveti); 1615 return; 1616 } 1617 } 1618 1619 /* 1620 * Compute mbuf offset to TCP data segment. 1621 */ 1622 hdroptlen = toff + off; 1623 1624 /* 1625 * Calculate amount of space in receive window, 1626 * and then do TCP input processing. 1627 * Receive window is amount of space in rcv queue, 1628 * but not less than advertised window. 1629 */ 1630 { int win; 1631 1632 win = sbspace(&so->so_rcv); 1633 if (win < 0) 1634 win = 0; 1635 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1636 } 1637 1638 switch (tp->t_state) { 1639 case TCPS_LISTEN: 1640 /* 1641 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1642 */ 1643 if (m->m_flags & (M_BCAST|M_MCAST)) 1644 goto drop; 1645 switch (af) { 1646 #ifdef INET6 1647 case AF_INET6: 1648 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1649 goto drop; 1650 break; 1651 #endif /* INET6 */ 1652 case AF_INET: 1653 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1654 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1655 goto drop; 1656 break; 1657 } 1658 break; 1659 1660 /* 1661 * If the state is SYN_SENT: 1662 * if seg contains an ACK, but not for our SYN, drop the input. 1663 * if seg contains a RST, then drop the connection. 1664 * if seg does not contain SYN, then drop it. 1665 * Otherwise this is an acceptable SYN segment 1666 * initialize tp->rcv_nxt and tp->irs 1667 * if seg contains ack then advance tp->snd_una 1668 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1669 * arrange for segment to be acked (eventually) 1670 * continue processing rest of data/controls, beginning with URG 1671 */ 1672 case TCPS_SYN_SENT: 1673 if ((tiflags & TH_ACK) && 1674 (SEQ_LEQ(th->th_ack, tp->iss) || 1675 SEQ_GT(th->th_ack, tp->snd_max))) 1676 goto dropwithreset; 1677 if (tiflags & TH_RST) { 1678 if (tiflags & TH_ACK) 1679 tp = tcp_drop(tp, ECONNREFUSED); 1680 goto drop; 1681 } 1682 if ((tiflags & TH_SYN) == 0) 1683 goto drop; 1684 if (tiflags & TH_ACK) { 1685 tp->snd_una = th->th_ack; 1686 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1687 tp->snd_nxt = tp->snd_una; 1688 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1689 tp->snd_high = tp->snd_una; 1690 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1691 } 1692 tp->irs = th->th_seq; 1693 tcp_rcvseqinit(tp); 1694 tp->t_flags |= TF_ACKNOW; 1695 tcp_mss_from_peer(tp, opti.maxseg); 1696 1697 /* 1698 * Initialize the initial congestion window. If we 1699 * had to retransmit the SYN, we must initialize cwnd 1700 * to 1 segment (i.e. the Loss Window). 1701 */ 1702 if (tp->t_flags & TF_SYN_REXMT) 1703 tp->snd_cwnd = tp->t_peermss; 1704 else { 1705 int ss = tcp_init_win; 1706 #ifdef INET 1707 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1708 ss = tcp_init_win_local; 1709 #endif 1710 #ifdef INET6 1711 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1712 ss = tcp_init_win_local; 1713 #endif 1714 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1715 } 1716 1717 tcp_rmx_rtt(tp); 1718 if (tiflags & TH_ACK) { 1719 tcpstat.tcps_connects++; 1720 soisconnected(so); 1721 tcp_established(tp); 1722 /* Do window scaling on this connection? */ 1723 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1724 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1725 tp->snd_scale = tp->requested_s_scale; 1726 tp->rcv_scale = tp->request_r_scale; 1727 } 1728 TCP_REASS_LOCK(tp); 1729 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1730 TCP_REASS_UNLOCK(tp); 1731 /* 1732 * if we didn't have to retransmit the SYN, 1733 * use its rtt as our initial srtt & rtt var. 1734 */ 1735 if (tp->t_rtttime) 1736 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1737 } else 1738 tp->t_state = TCPS_SYN_RECEIVED; 1739 1740 /* 1741 * Advance th->th_seq to correspond to first data byte. 1742 * If data, trim to stay within window, 1743 * dropping FIN if necessary. 1744 */ 1745 th->th_seq++; 1746 if (tlen > tp->rcv_wnd) { 1747 todrop = tlen - tp->rcv_wnd; 1748 m_adj(m, -todrop); 1749 tlen = tp->rcv_wnd; 1750 tiflags &= ~TH_FIN; 1751 tcpstat.tcps_rcvpackafterwin++; 1752 tcpstat.tcps_rcvbyteafterwin += todrop; 1753 } 1754 tp->snd_wl1 = th->th_seq - 1; 1755 tp->rcv_up = th->th_seq; 1756 goto step6; 1757 1758 /* 1759 * If the state is SYN_RECEIVED: 1760 * If seg contains an ACK, but not for our SYN, drop the input 1761 * and generate an RST. See page 36, rfc793 1762 */ 1763 case TCPS_SYN_RECEIVED: 1764 if ((tiflags & TH_ACK) && 1765 (SEQ_LEQ(th->th_ack, tp->iss) || 1766 SEQ_GT(th->th_ack, tp->snd_max))) 1767 goto dropwithreset; 1768 break; 1769 } 1770 1771 /* 1772 * States other than LISTEN or SYN_SENT. 1773 * First check timestamp, if present. 1774 * Then check that at least some bytes of segment are within 1775 * receive window. If segment begins before rcv_nxt, 1776 * drop leading data (and SYN); if nothing left, just ack. 1777 * 1778 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1779 * and it's less than ts_recent, drop it. 1780 */ 1781 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1782 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1783 1784 /* Check to see if ts_recent is over 24 days old. */ 1785 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 1786 /* 1787 * Invalidate ts_recent. If this segment updates 1788 * ts_recent, the age will be reset later and ts_recent 1789 * will get a valid value. If it does not, setting 1790 * ts_recent to zero will at least satisfy the 1791 * requirement that zero be placed in the timestamp 1792 * echo reply when ts_recent isn't valid. The 1793 * age isn't reset until we get a valid ts_recent 1794 * because we don't want out-of-order segments to be 1795 * dropped when ts_recent is old. 1796 */ 1797 tp->ts_recent = 0; 1798 } else { 1799 tcpstat.tcps_rcvduppack++; 1800 tcpstat.tcps_rcvdupbyte += tlen; 1801 tcpstat.tcps_pawsdrop++; 1802 goto dropafterack; 1803 } 1804 } 1805 1806 todrop = tp->rcv_nxt - th->th_seq; 1807 dupseg = FALSE; 1808 if (todrop > 0) { 1809 if (tiflags & TH_SYN) { 1810 tiflags &= ~TH_SYN; 1811 th->th_seq++; 1812 if (th->th_urp > 1) 1813 th->th_urp--; 1814 else { 1815 tiflags &= ~TH_URG; 1816 th->th_urp = 0; 1817 } 1818 todrop--; 1819 } 1820 if (todrop > tlen || 1821 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1822 /* 1823 * Any valid FIN or RST must be to the left of the 1824 * window. At this point the FIN or RST must be a 1825 * duplicate or out of sequence; drop it. 1826 */ 1827 if (tiflags & TH_RST) 1828 goto drop; 1829 tiflags &= ~(TH_FIN|TH_RST); 1830 /* 1831 * Send an ACK to resynchronize and drop any data. 1832 * But keep on processing for RST or ACK. 1833 */ 1834 tp->t_flags |= TF_ACKNOW; 1835 todrop = tlen; 1836 dupseg = TRUE; 1837 tcpstat.tcps_rcvdupbyte += todrop; 1838 tcpstat.tcps_rcvduppack++; 1839 } else if ((tiflags & TH_RST) && 1840 th->th_seq != tp->last_ack_sent) { 1841 /* 1842 * Test for reset before adjusting the sequence 1843 * number for overlapping data. 1844 */ 1845 goto dropafterack_ratelim; 1846 } else { 1847 tcpstat.tcps_rcvpartduppack++; 1848 tcpstat.tcps_rcvpartdupbyte += todrop; 1849 } 1850 hdroptlen += todrop; /*drop from head afterwards*/ 1851 th->th_seq += todrop; 1852 tlen -= todrop; 1853 if (th->th_urp > todrop) 1854 th->th_urp -= todrop; 1855 else { 1856 tiflags &= ~TH_URG; 1857 th->th_urp = 0; 1858 } 1859 } 1860 1861 /* 1862 * If new data are received on a connection after the 1863 * user processes are gone, then RST the other end. 1864 */ 1865 if ((so->so_state & SS_NOFDREF) && 1866 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1867 tp = tcp_close(tp); 1868 tcpstat.tcps_rcvafterclose++; 1869 goto dropwithreset; 1870 } 1871 1872 /* 1873 * If segment ends after window, drop trailing data 1874 * (and PUSH and FIN); if nothing left, just ACK. 1875 */ 1876 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1877 if (todrop > 0) { 1878 tcpstat.tcps_rcvpackafterwin++; 1879 if (todrop >= tlen) { 1880 /* 1881 * The segment actually starts after the window. 1882 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 1883 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 1884 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 1885 */ 1886 tcpstat.tcps_rcvbyteafterwin += tlen; 1887 /* 1888 * If a new connection request is received 1889 * while in TIME_WAIT, drop the old connection 1890 * and start over if the sequence numbers 1891 * are above the previous ones. 1892 * 1893 * NOTE: We will checksum the packet again, and 1894 * so we need to put the header fields back into 1895 * network order! 1896 * XXX This kind of sucks, but we don't expect 1897 * XXX this to happen very often, so maybe it 1898 * XXX doesn't matter so much. 1899 */ 1900 if (tiflags & TH_SYN && 1901 tp->t_state == TCPS_TIME_WAIT && 1902 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1903 iss = tcp_new_iss(tp, tp->snd_nxt); 1904 tp = tcp_close(tp); 1905 TCP_FIELDS_TO_NET(th); 1906 goto findpcb; 1907 } 1908 /* 1909 * If window is closed can only take segments at 1910 * window edge, and have to drop data and PUSH from 1911 * incoming segments. Continue processing, but 1912 * remember to ack. Otherwise, drop segment 1913 * and (if not RST) ack. 1914 */ 1915 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1916 tp->t_flags |= TF_ACKNOW; 1917 tcpstat.tcps_rcvwinprobe++; 1918 } else 1919 goto dropafterack; 1920 } else 1921 tcpstat.tcps_rcvbyteafterwin += todrop; 1922 m_adj(m, -todrop); 1923 tlen -= todrop; 1924 tiflags &= ~(TH_PUSH|TH_FIN); 1925 } 1926 1927 /* 1928 * If last ACK falls within this segment's sequence numbers, 1929 * and the timestamp is newer, record it. 1930 */ 1931 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1932 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1933 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1934 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1935 tp->ts_recent_age = tcp_now; 1936 tp->ts_recent = opti.ts_val; 1937 } 1938 1939 /* 1940 * If the RST bit is set examine the state: 1941 * SYN_RECEIVED STATE: 1942 * If passive open, return to LISTEN state. 1943 * If active open, inform user that connection was refused. 1944 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1945 * Inform user that connection was reset, and close tcb. 1946 * CLOSING, LAST_ACK, TIME_WAIT STATES 1947 * Close the tcb. 1948 */ 1949 if (tiflags & TH_RST) { 1950 if (th->th_seq != tp->last_ack_sent) 1951 goto dropafterack_ratelim; 1952 1953 switch (tp->t_state) { 1954 case TCPS_SYN_RECEIVED: 1955 so->so_error = ECONNREFUSED; 1956 goto close; 1957 1958 case TCPS_ESTABLISHED: 1959 case TCPS_FIN_WAIT_1: 1960 case TCPS_FIN_WAIT_2: 1961 case TCPS_CLOSE_WAIT: 1962 so->so_error = ECONNRESET; 1963 close: 1964 tp->t_state = TCPS_CLOSED; 1965 tcpstat.tcps_drops++; 1966 tp = tcp_close(tp); 1967 goto drop; 1968 1969 case TCPS_CLOSING: 1970 case TCPS_LAST_ACK: 1971 case TCPS_TIME_WAIT: 1972 tp = tcp_close(tp); 1973 goto drop; 1974 } 1975 } 1976 1977 /* 1978 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 1979 * we must be in a synchronized state. RFC791 states (under RST 1980 * generation) that any unacceptable segment (an out-of-order SYN 1981 * qualifies) received in a synchronized state must elicit only an 1982 * empty acknowledgment segment ... and the connection remains in 1983 * the same state. 1984 */ 1985 if (tiflags & TH_SYN) { 1986 if (tp->rcv_nxt == th->th_seq) { 1987 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 1988 TH_ACK); 1989 if (tcp_saveti) 1990 m_freem(tcp_saveti); 1991 return; 1992 } 1993 1994 goto dropafterack_ratelim; 1995 } 1996 1997 /* 1998 * If the ACK bit is off we drop the segment and return. 1999 */ 2000 if ((tiflags & TH_ACK) == 0) { 2001 if (tp->t_flags & TF_ACKNOW) 2002 goto dropafterack; 2003 else 2004 goto drop; 2005 } 2006 2007 /* 2008 * Ack processing. 2009 */ 2010 switch (tp->t_state) { 2011 2012 /* 2013 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2014 * ESTABLISHED state and continue processing, otherwise 2015 * send an RST. 2016 */ 2017 case TCPS_SYN_RECEIVED: 2018 if (SEQ_GT(tp->snd_una, th->th_ack) || 2019 SEQ_GT(th->th_ack, tp->snd_max)) 2020 goto dropwithreset; 2021 tcpstat.tcps_connects++; 2022 soisconnected(so); 2023 tcp_established(tp); 2024 /* Do window scaling? */ 2025 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2026 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2027 tp->snd_scale = tp->requested_s_scale; 2028 tp->rcv_scale = tp->request_r_scale; 2029 } 2030 TCP_REASS_LOCK(tp); 2031 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2032 TCP_REASS_UNLOCK(tp); 2033 tp->snd_wl1 = th->th_seq - 1; 2034 /* fall into ... */ 2035 2036 /* 2037 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2038 * ACKs. If the ack is in the range 2039 * tp->snd_una < th->th_ack <= tp->snd_max 2040 * then advance tp->snd_una to th->th_ack and drop 2041 * data from the retransmission queue. If this ACK reflects 2042 * more up to date window information we update our window information. 2043 */ 2044 case TCPS_ESTABLISHED: 2045 case TCPS_FIN_WAIT_1: 2046 case TCPS_FIN_WAIT_2: 2047 case TCPS_CLOSE_WAIT: 2048 case TCPS_CLOSING: 2049 case TCPS_LAST_ACK: 2050 case TCPS_TIME_WAIT: 2051 2052 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2053 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2054 tcpstat.tcps_rcvdupack++; 2055 /* 2056 * If we have outstanding data (other than 2057 * a window probe), this is a completely 2058 * duplicate ack (ie, window info didn't 2059 * change), the ack is the biggest we've 2060 * seen and we've seen exactly our rexmt 2061 * threshhold of them, assume a packet 2062 * has been dropped and retransmit it. 2063 * Kludge snd_nxt & the congestion 2064 * window so we send only this one 2065 * packet. 2066 * 2067 * We know we're losing at the current 2068 * window size so do congestion avoidance 2069 * (set ssthresh to half the current window 2070 * and pull our congestion window back to 2071 * the new ssthresh). 2072 * 2073 * Dup acks mean that packets have left the 2074 * network (they're now cached at the receiver) 2075 * so bump cwnd by the amount in the receiver 2076 * to keep a constant cwnd packets in the 2077 * network. 2078 */ 2079 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2080 th->th_ack != tp->snd_una) 2081 tp->t_dupacks = 0; 2082 else if (++tp->t_dupacks == tcprexmtthresh && 2083 tp->t_partialacks < 0) { 2084 tcp_seq onxt; 2085 u_int win; 2086 2087 if (tcp_do_newreno && 2088 SEQ_LT(th->th_ack, tp->snd_high)) { 2089 /* 2090 * False fast retransmit after 2091 * timeout. Do not enter fast 2092 * recovery. 2093 */ 2094 tp->t_dupacks = 0; 2095 break; 2096 } 2097 2098 onxt = tp->snd_nxt; 2099 win = min(tp->snd_wnd, tp->snd_cwnd) / 2100 2 / tp->t_segsz; 2101 if (win < 2) 2102 win = 2; 2103 tp->snd_ssthresh = win * tp->t_segsz; 2104 tp->snd_recover = tp->snd_max; 2105 tp->t_partialacks = 0; 2106 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2107 tp->t_rtttime = 0; 2108 tp->snd_nxt = th->th_ack; 2109 tp->snd_cwnd = tp->t_segsz; 2110 (void) tcp_output(tp); 2111 tp->snd_cwnd = tp->snd_ssthresh + 2112 tp->t_segsz * tp->t_dupacks; 2113 if (SEQ_GT(onxt, tp->snd_nxt)) 2114 tp->snd_nxt = onxt; 2115 goto drop; 2116 } else if (tp->t_dupacks > tcprexmtthresh) { 2117 tp->snd_cwnd += tp->t_segsz; 2118 (void) tcp_output(tp); 2119 goto drop; 2120 } 2121 } else { 2122 /* 2123 * If the ack appears to be very old, only 2124 * allow data that is in-sequence. This 2125 * makes it somewhat more difficult to insert 2126 * forged data by guessing sequence numbers. 2127 * Sent an ack to try to update the send 2128 * sequence number on the other side. 2129 */ 2130 if (tlen && th->th_seq != tp->rcv_nxt && 2131 SEQ_LT(th->th_ack, 2132 tp->snd_una - tp->max_sndwnd)) 2133 goto dropafterack; 2134 } 2135 break; 2136 } 2137 /* 2138 * If the congestion window was inflated to account 2139 * for the other side's cached packets, retract it. 2140 */ 2141 if (!tcp_do_newreno) 2142 tcp_reno_newack(tp, th); 2143 else 2144 tcp_newreno_newack(tp, th); 2145 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2146 tcpstat.tcps_rcvacktoomuch++; 2147 goto dropafterack; 2148 } 2149 acked = th->th_ack - tp->snd_una; 2150 tcpstat.tcps_rcvackpack++; 2151 tcpstat.tcps_rcvackbyte += acked; 2152 2153 /* 2154 * If we have a timestamp reply, update smoothed 2155 * round trip time. If no timestamp is present but 2156 * transmit timer is running and timed sequence 2157 * number was acked, update smoothed round trip time. 2158 * Since we now have an rtt measurement, cancel the 2159 * timer backoff (cf., Phil Karn's retransmit alg.). 2160 * Recompute the initial retransmit timer. 2161 */ 2162 if (opti.ts_present && opti.ts_ecr) 2163 tcp_xmit_timer(tp, opti.ts_ecr); 2164 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2165 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2166 2167 /* 2168 * If all outstanding data is acked, stop retransmit 2169 * timer and remember to restart (more output or persist). 2170 * If there is more data to be acked, restart retransmit 2171 * timer, using current (possibly backed-off) value. 2172 */ 2173 if (th->th_ack == tp->snd_max) { 2174 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2175 needoutput = 1; 2176 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2177 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2178 /* 2179 * When new data is acked, open the congestion window. 2180 * If the window gives us less than ssthresh packets 2181 * in flight, open exponentially (segsz per packet). 2182 * Otherwise open linearly: segsz per window 2183 * (segsz^2 / cwnd per packet), plus a constant 2184 * fraction of a packet (segsz/8) to help larger windows 2185 * open quickly enough. 2186 * 2187 * If we are still in fast recovery (meaning we are using 2188 * NewReno and we have only received partial acks), do not 2189 * inflate the window yet. 2190 */ 2191 if (tp->t_partialacks < 0) { 2192 u_int cw = tp->snd_cwnd; 2193 u_int incr = tp->t_segsz; 2194 2195 if (cw >= tp->snd_ssthresh) 2196 incr = incr * incr / cw; 2197 tp->snd_cwnd = min(cw + incr, 2198 TCP_MAXWIN << tp->snd_scale); 2199 } 2200 ND6_HINT(tp); 2201 if (acked > so->so_snd.sb_cc) { 2202 tp->snd_wnd -= so->so_snd.sb_cc; 2203 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2204 ourfinisacked = 1; 2205 } else { 2206 if (acked > (tp->t_lastoff - tp->t_inoff)) 2207 tp->t_lastm = NULL; 2208 sbdrop(&so->so_snd, acked); 2209 tp->t_lastoff -= acked; 2210 tp->snd_wnd -= acked; 2211 ourfinisacked = 0; 2212 } 2213 sowwakeup(so); 2214 tp->snd_una = th->th_ack; 2215 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2216 tp->snd_nxt = tp->snd_una; 2217 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2218 tp->snd_high = tp->snd_una; 2219 2220 switch (tp->t_state) { 2221 2222 /* 2223 * In FIN_WAIT_1 STATE in addition to the processing 2224 * for the ESTABLISHED state if our FIN is now acknowledged 2225 * then enter FIN_WAIT_2. 2226 */ 2227 case TCPS_FIN_WAIT_1: 2228 if (ourfinisacked) { 2229 /* 2230 * If we can't receive any more 2231 * data, then closing user can proceed. 2232 * Starting the timer is contrary to the 2233 * specification, but if we don't get a FIN 2234 * we'll hang forever. 2235 */ 2236 if (so->so_state & SS_CANTRCVMORE) { 2237 soisdisconnected(so); 2238 if (tcp_maxidle > 0) 2239 TCP_TIMER_ARM(tp, TCPT_2MSL, 2240 tcp_maxidle); 2241 } 2242 tp->t_state = TCPS_FIN_WAIT_2; 2243 } 2244 break; 2245 2246 /* 2247 * In CLOSING STATE in addition to the processing for 2248 * the ESTABLISHED state if the ACK acknowledges our FIN 2249 * then enter the TIME-WAIT state, otherwise ignore 2250 * the segment. 2251 */ 2252 case TCPS_CLOSING: 2253 if (ourfinisacked) { 2254 tp->t_state = TCPS_TIME_WAIT; 2255 tcp_canceltimers(tp); 2256 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2257 soisdisconnected(so); 2258 } 2259 break; 2260 2261 /* 2262 * In LAST_ACK, we may still be waiting for data to drain 2263 * and/or to be acked, as well as for the ack of our FIN. 2264 * If our FIN is now acknowledged, delete the TCB, 2265 * enter the closed state and return. 2266 */ 2267 case TCPS_LAST_ACK: 2268 if (ourfinisacked) { 2269 tp = tcp_close(tp); 2270 goto drop; 2271 } 2272 break; 2273 2274 /* 2275 * In TIME_WAIT state the only thing that should arrive 2276 * is a retransmission of the remote FIN. Acknowledge 2277 * it and restart the finack timer. 2278 */ 2279 case TCPS_TIME_WAIT: 2280 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2281 goto dropafterack; 2282 } 2283 } 2284 2285 step6: 2286 /* 2287 * Update window information. 2288 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2289 */ 2290 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2291 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 2292 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 2293 /* keep track of pure window updates */ 2294 if (tlen == 0 && 2295 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2296 tcpstat.tcps_rcvwinupd++; 2297 tp->snd_wnd = tiwin; 2298 tp->snd_wl1 = th->th_seq; 2299 tp->snd_wl2 = th->th_ack; 2300 if (tp->snd_wnd > tp->max_sndwnd) 2301 tp->max_sndwnd = tp->snd_wnd; 2302 needoutput = 1; 2303 } 2304 2305 /* 2306 * Process segments with URG. 2307 */ 2308 if ((tiflags & TH_URG) && th->th_urp && 2309 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2310 /* 2311 * This is a kludge, but if we receive and accept 2312 * random urgent pointers, we'll crash in 2313 * soreceive. It's hard to imagine someone 2314 * actually wanting to send this much urgent data. 2315 */ 2316 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2317 th->th_urp = 0; /* XXX */ 2318 tiflags &= ~TH_URG; /* XXX */ 2319 goto dodata; /* XXX */ 2320 } 2321 /* 2322 * If this segment advances the known urgent pointer, 2323 * then mark the data stream. This should not happen 2324 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2325 * a FIN has been received from the remote side. 2326 * In these states we ignore the URG. 2327 * 2328 * According to RFC961 (Assigned Protocols), 2329 * the urgent pointer points to the last octet 2330 * of urgent data. We continue, however, 2331 * to consider it to indicate the first octet 2332 * of data past the urgent section as the original 2333 * spec states (in one of two places). 2334 */ 2335 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2336 tp->rcv_up = th->th_seq + th->th_urp; 2337 so->so_oobmark = so->so_rcv.sb_cc + 2338 (tp->rcv_up - tp->rcv_nxt) - 1; 2339 if (so->so_oobmark == 0) 2340 so->so_state |= SS_RCVATMARK; 2341 sohasoutofband(so); 2342 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2343 } 2344 /* 2345 * Remove out of band data so doesn't get presented to user. 2346 * This can happen independent of advancing the URG pointer, 2347 * but if two URG's are pending at once, some out-of-band 2348 * data may creep in... ick. 2349 */ 2350 if (th->th_urp <= (u_int16_t) tlen 2351 #ifdef SO_OOBINLINE 2352 && (so->so_options & SO_OOBINLINE) == 0 2353 #endif 2354 ) 2355 tcp_pulloutofband(so, th, m, hdroptlen); 2356 } else 2357 /* 2358 * If no out of band data is expected, 2359 * pull receive urgent pointer along 2360 * with the receive window. 2361 */ 2362 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2363 tp->rcv_up = tp->rcv_nxt; 2364 dodata: /* XXX */ 2365 2366 /* 2367 * Process the segment text, merging it into the TCP sequencing queue, 2368 * and arranging for acknowledgement of receipt if necessary. 2369 * This process logically involves adjusting tp->rcv_wnd as data 2370 * is presented to the user (this happens in tcp_usrreq.c, 2371 * case PRU_RCVD). If a FIN has already been received on this 2372 * connection then we just ignore the text. 2373 */ 2374 if ((tlen || (tiflags & TH_FIN)) && 2375 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2376 /* 2377 * Insert segment ti into reassembly queue of tcp with 2378 * control block tp. Return TH_FIN if reassembly now includes 2379 * a segment with FIN. The macro form does the common case 2380 * inline (segment is the next to be received on an 2381 * established connection, and the queue is empty), 2382 * avoiding linkage into and removal from the queue and 2383 * repetition of various conversions. 2384 * Set DELACK for segments received in order, but ack 2385 * immediately when segments are out of order 2386 * (so fast retransmit can work). 2387 */ 2388 /* NOTE: this was TCP_REASS() macro, but used only once */ 2389 TCP_REASS_LOCK(tp); 2390 if (th->th_seq == tp->rcv_nxt && 2391 TAILQ_FIRST(&tp->segq) == NULL && 2392 tp->t_state == TCPS_ESTABLISHED) { 2393 TCP_SETUP_ACK(tp, th); 2394 tp->rcv_nxt += tlen; 2395 tiflags = th->th_flags & TH_FIN; 2396 tcpstat.tcps_rcvpack++; 2397 tcpstat.tcps_rcvbyte += tlen; 2398 ND6_HINT(tp); 2399 if (so->so_state & SS_CANTRCVMORE) 2400 m_freem(m); 2401 else { 2402 m_adj(m, hdroptlen); 2403 sbappendstream(&(so)->so_rcv, m); 2404 } 2405 sorwakeup(so); 2406 } else { 2407 m_adj(m, hdroptlen); 2408 tiflags = tcp_reass(tp, th, m, &tlen); 2409 tp->t_flags |= TF_ACKNOW; 2410 } 2411 TCP_REASS_UNLOCK(tp); 2412 2413 /* 2414 * Note the amount of data that peer has sent into 2415 * our window, in order to estimate the sender's 2416 * buffer size. 2417 */ 2418 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2419 } else { 2420 m_freem(m); 2421 m = NULL; 2422 tiflags &= ~TH_FIN; 2423 } 2424 2425 /* 2426 * If FIN is received ACK the FIN and let the user know 2427 * that the connection is closing. Ignore a FIN received before 2428 * the connection is fully established. 2429 */ 2430 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2431 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2432 socantrcvmore(so); 2433 tp->t_flags |= TF_ACKNOW; 2434 tp->rcv_nxt++; 2435 } 2436 switch (tp->t_state) { 2437 2438 /* 2439 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2440 */ 2441 case TCPS_ESTABLISHED: 2442 tp->t_state = TCPS_CLOSE_WAIT; 2443 break; 2444 2445 /* 2446 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2447 * enter the CLOSING state. 2448 */ 2449 case TCPS_FIN_WAIT_1: 2450 tp->t_state = TCPS_CLOSING; 2451 break; 2452 2453 /* 2454 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2455 * starting the time-wait timer, turning off the other 2456 * standard timers. 2457 */ 2458 case TCPS_FIN_WAIT_2: 2459 tp->t_state = TCPS_TIME_WAIT; 2460 tcp_canceltimers(tp); 2461 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2462 soisdisconnected(so); 2463 break; 2464 2465 /* 2466 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2467 */ 2468 case TCPS_TIME_WAIT: 2469 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2470 break; 2471 } 2472 } 2473 #ifdef TCP_DEBUG 2474 if (so->so_options & SO_DEBUG) 2475 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2476 #endif 2477 2478 /* 2479 * Return any desired output. 2480 */ 2481 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2482 (void) tcp_output(tp); 2483 if (tcp_saveti) 2484 m_freem(tcp_saveti); 2485 return; 2486 2487 badsyn: 2488 /* 2489 * Received a bad SYN. Increment counters and dropwithreset. 2490 */ 2491 tcpstat.tcps_badsyn++; 2492 tp = NULL; 2493 goto dropwithreset; 2494 2495 dropafterack: 2496 /* 2497 * Generate an ACK dropping incoming segment if it occupies 2498 * sequence space, where the ACK reflects our state. 2499 */ 2500 if (tiflags & TH_RST) 2501 goto drop; 2502 goto dropafterack2; 2503 2504 dropafterack_ratelim: 2505 /* 2506 * We may want to rate-limit ACKs against SYN/RST attack. 2507 */ 2508 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2509 tcp_ackdrop_ppslim) == 0) { 2510 /* XXX stat */ 2511 goto drop; 2512 } 2513 /* ...fall into dropafterack2... */ 2514 2515 dropafterack2: 2516 m_freem(m); 2517 tp->t_flags |= TF_ACKNOW; 2518 (void) tcp_output(tp); 2519 if (tcp_saveti) 2520 m_freem(tcp_saveti); 2521 return; 2522 2523 dropwithreset_ratelim: 2524 /* 2525 * We may want to rate-limit RSTs in certain situations, 2526 * particularly if we are sending an RST in response to 2527 * an attempt to connect to or otherwise communicate with 2528 * a port for which we have no socket. 2529 */ 2530 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2531 tcp_rst_ppslim) == 0) { 2532 /* XXX stat */ 2533 goto drop; 2534 } 2535 /* ...fall into dropwithreset... */ 2536 2537 dropwithreset: 2538 /* 2539 * Generate a RST, dropping incoming segment. 2540 * Make ACK acceptable to originator of segment. 2541 */ 2542 if (tiflags & TH_RST) 2543 goto drop; 2544 2545 switch (af) { 2546 #ifdef INET6 2547 case AF_INET6: 2548 /* For following calls to tcp_respond */ 2549 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2550 goto drop; 2551 break; 2552 #endif /* INET6 */ 2553 case AF_INET: 2554 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2555 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2556 goto drop; 2557 } 2558 2559 if (tiflags & TH_ACK) 2560 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2561 else { 2562 if (tiflags & TH_SYN) 2563 tlen++; 2564 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2565 TH_RST|TH_ACK); 2566 } 2567 if (tcp_saveti) 2568 m_freem(tcp_saveti); 2569 return; 2570 2571 badcsum: 2572 drop: 2573 /* 2574 * Drop space held by incoming segment and return. 2575 */ 2576 if (tp) { 2577 if (tp->t_inpcb) 2578 so = tp->t_inpcb->inp_socket; 2579 #ifdef INET6 2580 else if (tp->t_in6pcb) 2581 so = tp->t_in6pcb->in6p_socket; 2582 #endif 2583 else 2584 so = NULL; 2585 #ifdef TCP_DEBUG 2586 if (so && (so->so_options & SO_DEBUG) != 0) 2587 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2588 #endif 2589 } 2590 if (tcp_saveti) 2591 m_freem(tcp_saveti); 2592 m_freem(m); 2593 return; 2594 } 2595 2596 #ifdef TCP_SIGNATURE 2597 int 2598 tcp_signature_apply(void *fstate, caddr_t data, u_int len) 2599 { 2600 2601 MD5Update(fstate, (u_char *)data, len); 2602 return (0); 2603 } 2604 2605 struct secasvar * 2606 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 2607 { 2608 struct secasvar *sav; 2609 #ifdef FAST_IPSEC 2610 union sockaddr_union dst; 2611 #endif 2612 struct ip *ip; 2613 struct ip6_hdr *ip6; 2614 2615 ip = mtod(m, struct ip *); 2616 switch (ip->ip_v) { 2617 case 4: 2618 ip = mtod(m, struct ip *); 2619 ip6 = NULL; 2620 break; 2621 case 6: 2622 ip = NULL; 2623 ip6 = mtod(m, struct ip6_hdr *); 2624 break; 2625 default: 2626 return (NULL); 2627 } 2628 2629 #ifdef FAST_IPSEC 2630 /* Extract the destination from the IP header in the mbuf. */ 2631 bzero(&dst, sizeof(union sockaddr_union)); 2632 dst.sa.sa_len = sizeof(struct sockaddr_in); 2633 dst.sa.sa_family = AF_INET; 2634 dst.sin.sin_addr = ip->ip_dst; 2635 2636 /* 2637 * Look up an SADB entry which matches the address of the peer. 2638 */ 2639 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2640 #else 2641 if (ip) 2642 sav = key_allocsa(AF_INET, (caddr_t)&ip->ip_src, 2643 (caddr_t)&ip->ip_dst, IPPROTO_TCP, 2644 htonl(TCP_SIG_SPI)); 2645 else 2646 sav = key_allocsa(AF_INET6, (caddr_t)&ip6->ip6_src, 2647 (caddr_t)&ip6->ip6_dst, IPPROTO_TCP, 2648 htonl(TCP_SIG_SPI)); 2649 #endif 2650 2651 return (sav); /* freesav must be performed by caller */ 2652 } 2653 2654 int 2655 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 2656 struct secasvar *sav, char *sig) 2657 { 2658 MD5_CTX ctx; 2659 struct ip *ip; 2660 struct ipovly *ipovly; 2661 struct ip6_hdr *ip6; 2662 struct ippseudo ippseudo; 2663 struct ip6_hdr_pseudo ip6pseudo; 2664 struct tcphdr th0; 2665 int l, tcphdrlen; 2666 2667 if (sav == NULL) 2668 return (-1); 2669 2670 tcphdrlen = th->th_off * 4; 2671 2672 switch (mtod(m, struct ip *)->ip_v) { 2673 case 4: 2674 ip = mtod(m, struct ip *); 2675 ip6 = NULL; 2676 break; 2677 case 6: 2678 ip = NULL; 2679 ip6 = mtod(m, struct ip6_hdr *); 2680 break; 2681 default: 2682 return (-1); 2683 } 2684 2685 MD5Init(&ctx); 2686 2687 if (ip) { 2688 memset(&ippseudo, 0, sizeof(ippseudo)); 2689 ipovly = (struct ipovly *)ip; 2690 ippseudo.ippseudo_src = ipovly->ih_src; 2691 ippseudo.ippseudo_dst = ipovly->ih_dst; 2692 ippseudo.ippseudo_pad = 0; 2693 ippseudo.ippseudo_p = IPPROTO_TCP; 2694 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 2695 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 2696 } else { 2697 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 2698 ip6pseudo.ip6ph_src = ip6->ip6_src; 2699 in6_clearscope(&ip6pseudo.ip6ph_src); 2700 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 2701 in6_clearscope(&ip6pseudo.ip6ph_dst); 2702 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 2703 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 2704 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 2705 } 2706 2707 th0 = *th; 2708 th0.th_sum = 0; 2709 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 2710 2711 l = m->m_pkthdr.len - thoff - tcphdrlen; 2712 if (l > 0) 2713 m_apply(m, thoff + tcphdrlen, 2714 m->m_pkthdr.len - thoff - tcphdrlen, 2715 tcp_signature_apply, &ctx); 2716 2717 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2718 MD5Final(sig, &ctx); 2719 2720 return (0); 2721 } 2722 #endif 2723 2724 int 2725 tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcphdr *th, 2726 struct mbuf *m, int toff, struct tcp_opt_info *oi) 2727 { 2728 u_int16_t mss; 2729 int opt, optlen = 0; 2730 #ifdef TCP_SIGNATURE 2731 caddr_t sigp = NULL; 2732 char sigbuf[TCP_SIGLEN]; 2733 struct secasvar *sav = NULL; 2734 #endif 2735 2736 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 2737 opt = cp[0]; 2738 if (opt == TCPOPT_EOL) 2739 break; 2740 if (opt == TCPOPT_NOP) 2741 optlen = 1; 2742 else { 2743 if (cnt < 2) 2744 break; 2745 optlen = cp[1]; 2746 if (optlen < 2 || optlen > cnt) 2747 break; 2748 } 2749 switch (opt) { 2750 2751 default: 2752 continue; 2753 2754 case TCPOPT_MAXSEG: 2755 if (optlen != TCPOLEN_MAXSEG) 2756 continue; 2757 if (!(th->th_flags & TH_SYN)) 2758 continue; 2759 bcopy(cp + 2, &mss, sizeof(mss)); 2760 oi->maxseg = ntohs(mss); 2761 break; 2762 2763 case TCPOPT_WINDOW: 2764 if (optlen != TCPOLEN_WINDOW) 2765 continue; 2766 if (!(th->th_flags & TH_SYN)) 2767 continue; 2768 tp->t_flags |= TF_RCVD_SCALE; 2769 tp->requested_s_scale = cp[2]; 2770 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2771 #if 0 /*XXX*/ 2772 char *p; 2773 2774 if (ip) 2775 p = ntohl(ip->ip_src); 2776 #ifdef INET6 2777 else if (ip6) 2778 p = ip6_sprintf(&ip6->ip6_src); 2779 #endif 2780 else 2781 p = "(unknown)"; 2782 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2783 "assuming %d\n", 2784 tp->requested_s_scale, p, 2785 TCP_MAX_WINSHIFT); 2786 #else 2787 log(LOG_ERR, "TCP: invalid wscale %d, " 2788 "assuming %d\n", 2789 tp->requested_s_scale, 2790 TCP_MAX_WINSHIFT); 2791 #endif 2792 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2793 } 2794 break; 2795 2796 case TCPOPT_TIMESTAMP: 2797 if (optlen != TCPOLEN_TIMESTAMP) 2798 continue; 2799 oi->ts_present = 1; 2800 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2801 NTOHL(oi->ts_val); 2802 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2803 NTOHL(oi->ts_ecr); 2804 2805 /* 2806 * A timestamp received in a SYN makes 2807 * it ok to send timestamp requests and replies. 2808 */ 2809 if (th->th_flags & TH_SYN) { 2810 tp->t_flags |= TF_RCVD_TSTMP; 2811 tp->ts_recent = oi->ts_val; 2812 tp->ts_recent_age = tcp_now; 2813 } 2814 break; 2815 case TCPOPT_SACK_PERMITTED: 2816 if (optlen != TCPOLEN_SACK_PERMITTED) 2817 continue; 2818 if (!(th->th_flags & TH_SYN)) 2819 continue; 2820 tp->t_flags &= ~TF_CANT_TXSACK; 2821 break; 2822 2823 case TCPOPT_SACK: 2824 if (tp->t_flags & TF_IGNR_RXSACK) 2825 continue; 2826 if (optlen % 8 != 2 || optlen < 10) 2827 continue; 2828 cp += 2; 2829 optlen -= 2; 2830 for (; optlen > 0; cp -= 8, optlen -= 8) { 2831 tcp_seq lwe, rwe; 2832 bcopy((char *)cp, (char *) &lwe, sizeof(lwe)); 2833 NTOHL(lwe); 2834 bcopy((char *)cp, (char *) &rwe, sizeof(rwe)); 2835 NTOHL(rwe); 2836 /* tcp_mark_sacked(tp, lwe, rwe); */ 2837 } 2838 break; 2839 #ifdef TCP_SIGNATURE 2840 case TCPOPT_SIGNATURE: 2841 if (optlen != TCPOLEN_SIGNATURE) 2842 continue; 2843 if (sigp && bcmp(sigp, cp + 2, TCP_SIGLEN)) 2844 return (-1); 2845 2846 sigp = sigbuf; 2847 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 2848 memset(cp + 2, 0, TCP_SIGLEN); 2849 tp->t_flags |= TF_SIGNATURE; 2850 break; 2851 #endif 2852 } 2853 } 2854 2855 #ifdef TCP_SIGNATURE 2856 if (tp->t_flags & TF_SIGNATURE) { 2857 2858 sav = tcp_signature_getsav(m, th); 2859 2860 if (sav == NULL && tp->t_state == TCPS_LISTEN) 2861 return (-1); 2862 } 2863 2864 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 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 2875 if (sigp) { 2876 char sig[TCP_SIGLEN]; 2877 2878 TCP_FIELDS_TO_NET(th); 2879 if (tcp_signature(m, th, toff, sav, sig) < 0) { 2880 TCP_FIELDS_TO_HOST(th); 2881 if (sav == NULL) 2882 return (-1); 2883 #ifdef FAST_IPSEC 2884 KEY_FREESAV(&sav); 2885 #else 2886 key_freesav(sav); 2887 #endif 2888 return (-1); 2889 } 2890 TCP_FIELDS_TO_HOST(th); 2891 2892 if (bcmp(sig, sigp, TCP_SIGLEN)) { 2893 tcpstat.tcps_badsig++; 2894 if (sav == NULL) 2895 return (-1); 2896 #ifdef FAST_IPSEC 2897 KEY_FREESAV(&sav); 2898 #else 2899 key_freesav(sav); 2900 #endif 2901 return (-1); 2902 } else 2903 tcpstat.tcps_goodsig++; 2904 2905 key_sa_recordxfer(sav, m); 2906 #ifdef FAST_IPSEC 2907 KEY_FREESAV(&sav); 2908 #else 2909 key_freesav(sav); 2910 #endif 2911 } 2912 #endif 2913 2914 return (0); 2915 } 2916 2917 /* 2918 * Pull out of band byte out of a segment so 2919 * it doesn't appear in the user's data queue. 2920 * It is still reflected in the segment length for 2921 * sequencing purposes. 2922 */ 2923 void 2924 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 2925 struct mbuf *m, int off) 2926 { 2927 int cnt = off + th->th_urp - 1; 2928 2929 while (cnt >= 0) { 2930 if (m->m_len > cnt) { 2931 char *cp = mtod(m, caddr_t) + cnt; 2932 struct tcpcb *tp = sototcpcb(so); 2933 2934 tp->t_iobc = *cp; 2935 tp->t_oobflags |= TCPOOB_HAVEDATA; 2936 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2937 m->m_len--; 2938 return; 2939 } 2940 cnt -= m->m_len; 2941 m = m->m_next; 2942 if (m == 0) 2943 break; 2944 } 2945 panic("tcp_pulloutofband"); 2946 } 2947 2948 /* 2949 * Collect new round-trip time estimate 2950 * and update averages and current timeout. 2951 */ 2952 void 2953 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 2954 { 2955 int32_t delta; 2956 2957 tcpstat.tcps_rttupdated++; 2958 if (tp->t_srtt != 0) { 2959 /* 2960 * srtt is stored as fixed point with 3 bits after the 2961 * binary point (i.e., scaled by 8). The following magic 2962 * is equivalent to the smoothing algorithm in rfc793 with 2963 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2964 * point). Adjust rtt to origin 0. 2965 */ 2966 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2967 if ((tp->t_srtt += delta) <= 0) 2968 tp->t_srtt = 1 << 2; 2969 /* 2970 * We accumulate a smoothed rtt variance (actually, a 2971 * smoothed mean difference), then set the retransmit 2972 * timer to smoothed rtt + 4 times the smoothed variance. 2973 * rttvar is stored as fixed point with 2 bits after the 2974 * binary point (scaled by 4). The following is 2975 * equivalent to rfc793 smoothing with an alpha of .75 2976 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2977 * rfc793's wired-in beta. 2978 */ 2979 if (delta < 0) 2980 delta = -delta; 2981 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2982 if ((tp->t_rttvar += delta) <= 0) 2983 tp->t_rttvar = 1 << 2; 2984 } else { 2985 /* 2986 * No rtt measurement yet - use the unsmoothed rtt. 2987 * Set the variance to half the rtt (so our first 2988 * retransmit happens at 3*rtt). 2989 */ 2990 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 2991 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 2992 } 2993 tp->t_rtttime = 0; 2994 tp->t_rxtshift = 0; 2995 2996 /* 2997 * the retransmit should happen at rtt + 4 * rttvar. 2998 * Because of the way we do the smoothing, srtt and rttvar 2999 * will each average +1/2 tick of bias. When we compute 3000 * the retransmit timer, we want 1/2 tick of rounding and 3001 * 1 extra tick because of +-1/2 tick uncertainty in the 3002 * firing of the timer. The bias will give us exactly the 3003 * 1.5 tick we need. But, because the bias is 3004 * statistical, we have to test that we don't drop below 3005 * the minimum feasible timer (which is 2 ticks). 3006 */ 3007 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3008 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3009 3010 /* 3011 * We received an ack for a packet that wasn't retransmitted; 3012 * it is probably safe to discard any error indications we've 3013 * received recently. This isn't quite right, but close enough 3014 * for now (a route might have failed after we sent a segment, 3015 * and the return path might not be symmetrical). 3016 */ 3017 tp->t_softerror = 0; 3018 } 3019 3020 void 3021 tcp_reno_newack(struct tcpcb *tp, struct tcphdr *th) 3022 { 3023 if (tp->t_partialacks < 0) { 3024 /* 3025 * We were not in fast recovery. Reset the duplicate ack 3026 * counter. 3027 */ 3028 tp->t_dupacks = 0; 3029 } else { 3030 /* 3031 * Clamp the congestion window to the crossover point and 3032 * exit fast recovery. 3033 */ 3034 if (tp->snd_cwnd > tp->snd_ssthresh) 3035 tp->snd_cwnd = tp->snd_ssthresh; 3036 tp->t_partialacks = -1; 3037 tp->t_dupacks = 0; 3038 } 3039 } 3040 3041 /* 3042 * Implement the NewReno response to a new ack, checking for partial acks in 3043 * fast recovery. 3044 */ 3045 void 3046 tcp_newreno_newack(struct tcpcb *tp, struct tcphdr *th) 3047 { 3048 if (tp->t_partialacks < 0) { 3049 /* 3050 * We were not in fast recovery. Reset the duplicate ack 3051 * counter. 3052 */ 3053 tp->t_dupacks = 0; 3054 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) { 3055 /* 3056 * This is a partial ack. Retransmit the first unacknowledged 3057 * segment and deflate the congestion window by the amount of 3058 * acknowledged data. Do not exit fast recovery. 3059 */ 3060 tcp_seq onxt = tp->snd_nxt; 3061 u_long ocwnd = tp->snd_cwnd; 3062 3063 /* 3064 * snd_una has not yet been updated and the socket's send 3065 * buffer has not yet drained off the ACK'd data, so we 3066 * have to leave snd_una as it was to get the correct data 3067 * offset in tcp_output(). 3068 */ 3069 if (++tp->t_partialacks == 1) 3070 TCP_TIMER_DISARM(tp, TCPT_REXMT); 3071 tp->t_rtttime = 0; 3072 tp->snd_nxt = th->th_ack; 3073 /* 3074 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 3075 * is not yet updated when we're called. 3076 */ 3077 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 3078 (void) tcp_output(tp); 3079 tp->snd_cwnd = ocwnd; 3080 if (SEQ_GT(onxt, tp->snd_nxt)) 3081 tp->snd_nxt = onxt; 3082 /* 3083 * Partial window deflation. Relies on fact that tp->snd_una 3084 * not updated yet. 3085 */ 3086 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 3087 } else { 3088 /* 3089 * Complete ack. Inflate the congestion window to ssthresh 3090 * and exit fast recovery. 3091 * 3092 * Window inflation should have left us with approx. 3093 * snd_ssthresh outstanding data. But in case we 3094 * would be inclined to send a burst, better to do 3095 * it via the slow start mechanism. 3096 */ 3097 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 3098 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 3099 + tp->t_segsz; 3100 else 3101 tp->snd_cwnd = tp->snd_ssthresh; 3102 tp->t_partialacks = -1; 3103 tp->t_dupacks = 0; 3104 } 3105 } 3106 3107 3108 /* 3109 * TCP compressed state engine. Currently used to hold compressed 3110 * state for SYN_RECEIVED. 3111 */ 3112 3113 u_long syn_cache_count; 3114 u_int32_t syn_hash1, syn_hash2; 3115 3116 #define SYN_HASH(sa, sp, dp) \ 3117 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3118 ((u_int32_t)(sp)))^syn_hash2))) 3119 #ifndef INET6 3120 #define SYN_HASHALL(hash, src, dst) \ 3121 do { \ 3122 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 3123 ((struct sockaddr_in *)(src))->sin_port, \ 3124 ((struct sockaddr_in *)(dst))->sin_port); \ 3125 } while (/*CONSTCOND*/ 0) 3126 #else 3127 #define SYN_HASH6(sa, sp, dp) \ 3128 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3129 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3130 & 0x7fffffff) 3131 3132 #define SYN_HASHALL(hash, src, dst) \ 3133 do { \ 3134 switch ((src)->sa_family) { \ 3135 case AF_INET: \ 3136 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 3137 ((struct sockaddr_in *)(src))->sin_port, \ 3138 ((struct sockaddr_in *)(dst))->sin_port); \ 3139 break; \ 3140 case AF_INET6: \ 3141 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 3142 ((struct sockaddr_in6 *)(src))->sin6_port, \ 3143 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 3144 break; \ 3145 default: \ 3146 hash = 0; \ 3147 } \ 3148 } while (/*CONSTCOND*/0) 3149 #endif /* INET6 */ 3150 3151 #define SYN_CACHE_RM(sc) \ 3152 do { \ 3153 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 3154 (sc), sc_bucketq); \ 3155 (sc)->sc_tp = NULL; \ 3156 LIST_REMOVE((sc), sc_tpq); \ 3157 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 3158 callout_stop(&(sc)->sc_timer); \ 3159 syn_cache_count--; \ 3160 } while (/*CONSTCOND*/0) 3161 3162 #define SYN_CACHE_PUT(sc) \ 3163 do { \ 3164 if ((sc)->sc_ipopts) \ 3165 (void) m_free((sc)->sc_ipopts); \ 3166 if ((sc)->sc_route4.ro_rt != NULL) \ 3167 RTFREE((sc)->sc_route4.ro_rt); \ 3168 if (callout_invoking(&(sc)->sc_timer)) \ 3169 (sc)->sc_flags |= SCF_DEAD; \ 3170 else \ 3171 pool_put(&syn_cache_pool, (sc)); \ 3172 } while (/*CONSTCOND*/0) 3173 3174 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL); 3175 3176 /* 3177 * We don't estimate RTT with SYNs, so each packet starts with the default 3178 * RTT and each timer step has a fixed timeout value. 3179 */ 3180 #define SYN_CACHE_TIMER_ARM(sc) \ 3181 do { \ 3182 TCPT_RANGESET((sc)->sc_rxtcur, \ 3183 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3184 TCPTV_REXMTMAX); \ 3185 callout_reset(&(sc)->sc_timer, \ 3186 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3187 } while (/*CONSTCOND*/0) 3188 3189 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3190 3191 void 3192 syn_cache_init(void) 3193 { 3194 int i; 3195 3196 /* Initialize the hash buckets. */ 3197 for (i = 0; i < tcp_syn_cache_size; i++) 3198 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3199 } 3200 3201 void 3202 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3203 { 3204 struct syn_cache_head *scp; 3205 struct syn_cache *sc2; 3206 int s; 3207 3208 /* 3209 * If there are no entries in the hash table, reinitialize 3210 * the hash secrets. 3211 */ 3212 if (syn_cache_count == 0) { 3213 syn_hash1 = arc4random(); 3214 syn_hash2 = arc4random(); 3215 } 3216 3217 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3218 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3219 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3220 3221 /* 3222 * Make sure that we don't overflow the per-bucket 3223 * limit or the total cache size limit. 3224 */ 3225 s = splsoftnet(); 3226 if (scp->sch_length >= tcp_syn_bucket_limit) { 3227 tcpstat.tcps_sc_bucketoverflow++; 3228 /* 3229 * The bucket is full. Toss the oldest element in the 3230 * bucket. This will be the first entry in the bucket. 3231 */ 3232 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3233 #ifdef DIAGNOSTIC 3234 /* 3235 * This should never happen; we should always find an 3236 * entry in our bucket. 3237 */ 3238 if (sc2 == NULL) 3239 panic("syn_cache_insert: bucketoverflow: impossible"); 3240 #endif 3241 SYN_CACHE_RM(sc2); 3242 SYN_CACHE_PUT(sc2); 3243 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3244 struct syn_cache_head *scp2, *sce; 3245 3246 tcpstat.tcps_sc_overflowed++; 3247 /* 3248 * The cache is full. Toss the oldest entry in the 3249 * first non-empty bucket we can find. 3250 * 3251 * XXX We would really like to toss the oldest 3252 * entry in the cache, but we hope that this 3253 * condition doesn't happen very often. 3254 */ 3255 scp2 = scp; 3256 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3257 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3258 for (++scp2; scp2 != scp; scp2++) { 3259 if (scp2 >= sce) 3260 scp2 = &tcp_syn_cache[0]; 3261 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3262 break; 3263 } 3264 #ifdef DIAGNOSTIC 3265 /* 3266 * This should never happen; we should always find a 3267 * non-empty bucket. 3268 */ 3269 if (scp2 == scp) 3270 panic("syn_cache_insert: cacheoverflow: " 3271 "impossible"); 3272 #endif 3273 } 3274 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3275 SYN_CACHE_RM(sc2); 3276 SYN_CACHE_PUT(sc2); 3277 } 3278 3279 /* 3280 * Initialize the entry's timer. 3281 */ 3282 sc->sc_rxttot = 0; 3283 sc->sc_rxtshift = 0; 3284 SYN_CACHE_TIMER_ARM(sc); 3285 3286 /* Link it from tcpcb entry */ 3287 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3288 3289 /* Put it into the bucket. */ 3290 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3291 scp->sch_length++; 3292 syn_cache_count++; 3293 3294 tcpstat.tcps_sc_added++; 3295 splx(s); 3296 } 3297 3298 /* 3299 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3300 * If we have retransmitted an entry the maximum number of times, expire 3301 * that entry. 3302 */ 3303 void 3304 syn_cache_timer(void *arg) 3305 { 3306 struct syn_cache *sc = arg; 3307 int s; 3308 3309 s = splsoftnet(); 3310 callout_ack(&sc->sc_timer); 3311 3312 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3313 tcpstat.tcps_sc_delayed_free++; 3314 pool_put(&syn_cache_pool, sc); 3315 splx(s); 3316 return; 3317 } 3318 3319 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3320 /* Drop it -- too many retransmissions. */ 3321 goto dropit; 3322 } 3323 3324 /* 3325 * Compute the total amount of time this entry has 3326 * been on a queue. If this entry has been on longer 3327 * than the keep alive timer would allow, expire it. 3328 */ 3329 sc->sc_rxttot += sc->sc_rxtcur; 3330 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) 3331 goto dropit; 3332 3333 tcpstat.tcps_sc_retransmitted++; 3334 (void) syn_cache_respond(sc, NULL); 3335 3336 /* Advance the timer back-off. */ 3337 sc->sc_rxtshift++; 3338 SYN_CACHE_TIMER_ARM(sc); 3339 3340 splx(s); 3341 return; 3342 3343 dropit: 3344 tcpstat.tcps_sc_timed_out++; 3345 SYN_CACHE_RM(sc); 3346 SYN_CACHE_PUT(sc); 3347 splx(s); 3348 } 3349 3350 /* 3351 * Remove syn cache created by the specified tcb entry, 3352 * because this does not make sense to keep them 3353 * (if there's no tcb entry, syn cache entry will never be used) 3354 */ 3355 void 3356 syn_cache_cleanup(struct tcpcb *tp) 3357 { 3358 struct syn_cache *sc, *nsc; 3359 int s; 3360 3361 s = splsoftnet(); 3362 3363 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3364 nsc = LIST_NEXT(sc, sc_tpq); 3365 3366 #ifdef DIAGNOSTIC 3367 if (sc->sc_tp != tp) 3368 panic("invalid sc_tp in syn_cache_cleanup"); 3369 #endif 3370 SYN_CACHE_RM(sc); 3371 SYN_CACHE_PUT(sc); 3372 } 3373 /* just for safety */ 3374 LIST_INIT(&tp->t_sc); 3375 3376 splx(s); 3377 } 3378 3379 /* 3380 * Find an entry in the syn cache. 3381 */ 3382 struct syn_cache * 3383 syn_cache_lookup(struct sockaddr *src, struct sockaddr *dst, 3384 struct syn_cache_head **headp) 3385 { 3386 struct syn_cache *sc; 3387 struct syn_cache_head *scp; 3388 u_int32_t hash; 3389 int s; 3390 3391 SYN_HASHALL(hash, src, dst); 3392 3393 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3394 *headp = scp; 3395 s = splsoftnet(); 3396 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3397 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3398 if (sc->sc_hash != hash) 3399 continue; 3400 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3401 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3402 splx(s); 3403 return (sc); 3404 } 3405 } 3406 splx(s); 3407 return (NULL); 3408 } 3409 3410 /* 3411 * This function gets called when we receive an ACK for a 3412 * socket in the LISTEN state. We look up the connection 3413 * in the syn cache, and if its there, we pull it out of 3414 * the cache and turn it into a full-blown connection in 3415 * the SYN-RECEIVED state. 3416 * 3417 * The return values may not be immediately obvious, and their effects 3418 * can be subtle, so here they are: 3419 * 3420 * NULL SYN was not found in cache; caller should drop the 3421 * packet and send an RST. 3422 * 3423 * -1 We were unable to create the new connection, and are 3424 * aborting it. An ACK,RST is being sent to the peer 3425 * (unless we got screwey sequence numbners; see below), 3426 * because the 3-way handshake has been completed. Caller 3427 * should not free the mbuf, since we may be using it. If 3428 * we are not, we will free it. 3429 * 3430 * Otherwise, the return value is a pointer to the new socket 3431 * associated with the connection. 3432 */ 3433 struct socket * 3434 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3435 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3436 struct socket *so, struct mbuf *m) 3437 { 3438 struct syn_cache *sc; 3439 struct syn_cache_head *scp; 3440 struct inpcb *inp = NULL; 3441 #ifdef INET6 3442 struct in6pcb *in6p = NULL; 3443 #endif 3444 struct tcpcb *tp = 0; 3445 struct mbuf *am; 3446 int s; 3447 struct socket *oso; 3448 3449 s = splsoftnet(); 3450 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3451 splx(s); 3452 return (NULL); 3453 } 3454 3455 /* 3456 * Verify the sequence and ack numbers. Try getting the correct 3457 * response again. 3458 */ 3459 if ((th->th_ack != sc->sc_iss + 1) || 3460 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3461 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3462 (void) syn_cache_respond(sc, m); 3463 splx(s); 3464 return ((struct socket *)(-1)); 3465 } 3466 3467 /* Remove this cache entry */ 3468 SYN_CACHE_RM(sc); 3469 splx(s); 3470 3471 /* 3472 * Ok, create the full blown connection, and set things up 3473 * as they would have been set up if we had created the 3474 * connection when the SYN arrived. If we can't create 3475 * the connection, abort it. 3476 */ 3477 /* 3478 * inp still has the OLD in_pcb stuff, set the 3479 * v6-related flags on the new guy, too. This is 3480 * done particularly for the case where an AF_INET6 3481 * socket is bound only to a port, and a v4 connection 3482 * comes in on that port. 3483 * we also copy the flowinfo from the original pcb 3484 * to the new one. 3485 */ 3486 oso = so; 3487 so = sonewconn(so, SS_ISCONNECTED); 3488 if (so == NULL) 3489 goto resetandabort; 3490 3491 switch (so->so_proto->pr_domain->dom_family) { 3492 #ifdef INET 3493 case AF_INET: 3494 inp = sotoinpcb(so); 3495 break; 3496 #endif 3497 #ifdef INET6 3498 case AF_INET6: 3499 in6p = sotoin6pcb(so); 3500 break; 3501 #endif 3502 } 3503 switch (src->sa_family) { 3504 #ifdef INET 3505 case AF_INET: 3506 if (inp) { 3507 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3508 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3509 inp->inp_options = ip_srcroute(); 3510 in_pcbstate(inp, INP_BOUND); 3511 if (inp->inp_options == NULL) { 3512 inp->inp_options = sc->sc_ipopts; 3513 sc->sc_ipopts = NULL; 3514 } 3515 } 3516 #ifdef INET6 3517 else if (in6p) { 3518 /* IPv4 packet to AF_INET6 socket */ 3519 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3520 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3521 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3522 &in6p->in6p_laddr.s6_addr32[3], 3523 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3524 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3525 in6totcpcb(in6p)->t_family = AF_INET; 3526 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3527 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3528 else 3529 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3530 in6_pcbstate(in6p, IN6P_BOUND); 3531 } 3532 #endif 3533 break; 3534 #endif 3535 #ifdef INET6 3536 case AF_INET6: 3537 if (in6p) { 3538 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3539 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3540 in6_pcbstate(in6p, IN6P_BOUND); 3541 } 3542 break; 3543 #endif 3544 } 3545 #ifdef INET6 3546 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3547 struct in6pcb *oin6p = sotoin6pcb(oso); 3548 /* inherit socket options from the listening socket */ 3549 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3550 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3551 m_freem(in6p->in6p_options); 3552 in6p->in6p_options = 0; 3553 } 3554 ip6_savecontrol(in6p, &in6p->in6p_options, 3555 mtod(m, struct ip6_hdr *), m); 3556 } 3557 #endif 3558 3559 #if defined(IPSEC) || defined(FAST_IPSEC) 3560 /* 3561 * we make a copy of policy, instead of sharing the policy, 3562 * for better behavior in terms of SA lookup and dead SA removal. 3563 */ 3564 if (inp) { 3565 /* copy old policy into new socket's */ 3566 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3567 printf("tcp_input: could not copy policy\n"); 3568 } 3569 #ifdef INET6 3570 else if (in6p) { 3571 /* copy old policy into new socket's */ 3572 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3573 in6p->in6p_sp)) 3574 printf("tcp_input: could not copy policy\n"); 3575 } 3576 #endif 3577 #endif 3578 3579 /* 3580 * Give the new socket our cached route reference. 3581 */ 3582 if (inp) 3583 inp->inp_route = sc->sc_route4; /* struct assignment */ 3584 #ifdef INET6 3585 else 3586 in6p->in6p_route = sc->sc_route6; 3587 #endif 3588 sc->sc_route4.ro_rt = NULL; 3589 3590 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3591 if (am == NULL) 3592 goto resetandabort; 3593 MCLAIM(am, &tcp_mowner); 3594 am->m_len = src->sa_len; 3595 bcopy(src, mtod(am, caddr_t), src->sa_len); 3596 if (inp) { 3597 if (in_pcbconnect(inp, am)) { 3598 (void) m_free(am); 3599 goto resetandabort; 3600 } 3601 } 3602 #ifdef INET6 3603 else if (in6p) { 3604 if (src->sa_family == AF_INET) { 3605 /* IPv4 packet to AF_INET6 socket */ 3606 struct sockaddr_in6 *sin6; 3607 sin6 = mtod(am, struct sockaddr_in6 *); 3608 am->m_len = sizeof(*sin6); 3609 bzero(sin6, sizeof(*sin6)); 3610 sin6->sin6_family = AF_INET6; 3611 sin6->sin6_len = sizeof(*sin6); 3612 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3613 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3614 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3615 &sin6->sin6_addr.s6_addr32[3], 3616 sizeof(sin6->sin6_addr.s6_addr32[3])); 3617 } 3618 if (in6_pcbconnect(in6p, am)) { 3619 (void) m_free(am); 3620 goto resetandabort; 3621 } 3622 } 3623 #endif 3624 else { 3625 (void) m_free(am); 3626 goto resetandabort; 3627 } 3628 (void) m_free(am); 3629 3630 if (inp) 3631 tp = intotcpcb(inp); 3632 #ifdef INET6 3633 else if (in6p) 3634 tp = in6totcpcb(in6p); 3635 #endif 3636 else 3637 tp = NULL; 3638 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3639 if (sc->sc_request_r_scale != 15) { 3640 tp->requested_s_scale = sc->sc_requested_s_scale; 3641 tp->request_r_scale = sc->sc_request_r_scale; 3642 tp->snd_scale = sc->sc_requested_s_scale; 3643 tp->rcv_scale = sc->sc_request_r_scale; 3644 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3645 } 3646 if (sc->sc_flags & SCF_TIMESTAMP) 3647 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3648 tp->ts_timebase = sc->sc_timebase; 3649 3650 tp->t_template = tcp_template(tp); 3651 if (tp->t_template == 0) { 3652 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3653 so = NULL; 3654 m_freem(m); 3655 goto abort; 3656 } 3657 3658 tp->iss = sc->sc_iss; 3659 tp->irs = sc->sc_irs; 3660 tcp_sendseqinit(tp); 3661 tcp_rcvseqinit(tp); 3662 tp->t_state = TCPS_SYN_RECEIVED; 3663 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3664 tcpstat.tcps_accepts++; 3665 3666 #ifdef TCP_SIGNATURE 3667 if (sc->sc_flags & SCF_SIGNATURE) 3668 tp->t_flags |= TF_SIGNATURE; 3669 #endif 3670 3671 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3672 tp->t_ourmss = sc->sc_ourmaxseg; 3673 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3674 3675 /* 3676 * Initialize the initial congestion window. If we 3677 * had to retransmit the SYN,ACK, we must initialize cwnd 3678 * to 1 segment (i.e. the Loss Window). 3679 */ 3680 if (sc->sc_rxtshift) 3681 tp->snd_cwnd = tp->t_peermss; 3682 else { 3683 int ss = tcp_init_win; 3684 #ifdef INET 3685 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3686 ss = tcp_init_win_local; 3687 #endif 3688 #ifdef INET6 3689 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3690 ss = tcp_init_win_local; 3691 #endif 3692 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3693 } 3694 3695 tcp_rmx_rtt(tp); 3696 tp->snd_wl1 = sc->sc_irs; 3697 tp->rcv_up = sc->sc_irs + 1; 3698 3699 /* 3700 * This is what whould have happened in tcp_output() when 3701 * the SYN,ACK was sent. 3702 */ 3703 tp->snd_up = tp->snd_una; 3704 tp->snd_max = tp->snd_nxt = tp->iss+1; 3705 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3706 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3707 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3708 tp->last_ack_sent = tp->rcv_nxt; 3709 tp->t_partialacks = -1; 3710 tp->t_dupacks = 0; 3711 3712 tcpstat.tcps_sc_completed++; 3713 SYN_CACHE_PUT(sc); 3714 return (so); 3715 3716 resetandabort: 3717 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3718 abort: 3719 if (so != NULL) 3720 (void) soabort(so); 3721 SYN_CACHE_PUT(sc); 3722 tcpstat.tcps_sc_aborted++; 3723 return ((struct socket *)(-1)); 3724 } 3725 3726 /* 3727 * This function is called when we get a RST for a 3728 * non-existent connection, so that we can see if the 3729 * connection is in the syn cache. If it is, zap it. 3730 */ 3731 3732 void 3733 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 3734 { 3735 struct syn_cache *sc; 3736 struct syn_cache_head *scp; 3737 int s = splsoftnet(); 3738 3739 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3740 splx(s); 3741 return; 3742 } 3743 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3744 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3745 splx(s); 3746 return; 3747 } 3748 SYN_CACHE_RM(sc); 3749 splx(s); 3750 tcpstat.tcps_sc_reset++; 3751 SYN_CACHE_PUT(sc); 3752 } 3753 3754 void 3755 syn_cache_unreach(struct sockaddr *src, struct sockaddr *dst, 3756 struct tcphdr *th) 3757 { 3758 struct syn_cache *sc; 3759 struct syn_cache_head *scp; 3760 int s; 3761 3762 s = splsoftnet(); 3763 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3764 splx(s); 3765 return; 3766 } 3767 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3768 if (ntohl (th->th_seq) != sc->sc_iss) { 3769 splx(s); 3770 return; 3771 } 3772 3773 /* 3774 * If we've retransmitted 3 times and this is our second error, 3775 * we remove the entry. Otherwise, we allow it to continue on. 3776 * This prevents us from incorrectly nuking an entry during a 3777 * spurious network outage. 3778 * 3779 * See tcp_notify(). 3780 */ 3781 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3782 sc->sc_flags |= SCF_UNREACH; 3783 splx(s); 3784 return; 3785 } 3786 3787 SYN_CACHE_RM(sc); 3788 splx(s); 3789 tcpstat.tcps_sc_unreach++; 3790 SYN_CACHE_PUT(sc); 3791 } 3792 3793 /* 3794 * Given a LISTEN socket and an inbound SYN request, add 3795 * this to the syn cache, and send back a segment: 3796 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3797 * to the source. 3798 * 3799 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3800 * Doing so would require that we hold onto the data and deliver it 3801 * to the application. However, if we are the target of a SYN-flood 3802 * DoS attack, an attacker could send data which would eventually 3803 * consume all available buffer space if it were ACKed. By not ACKing 3804 * the data, we avoid this DoS scenario. 3805 */ 3806 3807 int 3808 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3809 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 3810 int optlen, struct tcp_opt_info *oi) 3811 { 3812 struct tcpcb tb, *tp; 3813 long win; 3814 struct syn_cache *sc; 3815 struct syn_cache_head *scp; 3816 struct mbuf *ipopts; 3817 struct tcp_opt_info opti; 3818 3819 tp = sototcpcb(so); 3820 3821 bzero(&opti, sizeof(opti)); 3822 3823 /* 3824 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3825 * 3826 * Note this check is performed in tcp_input() very early on. 3827 */ 3828 3829 /* 3830 * Initialize some local state. 3831 */ 3832 win = sbspace(&so->so_rcv); 3833 if (win > TCP_MAXWIN) 3834 win = TCP_MAXWIN; 3835 3836 switch (src->sa_family) { 3837 #ifdef INET 3838 case AF_INET: 3839 /* 3840 * Remember the IP options, if any. 3841 */ 3842 ipopts = ip_srcroute(); 3843 break; 3844 #endif 3845 default: 3846 ipopts = NULL; 3847 } 3848 3849 #ifdef TCP_SIGNATURE 3850 if (optp || (tp->t_flags & TF_SIGNATURE)) 3851 #else 3852 if (optp) 3853 #endif 3854 { 3855 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3856 #ifdef TCP_SIGNATURE 3857 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 3858 #endif 3859 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 3860 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 3861 return (0); 3862 } else 3863 tb.t_flags = 0; 3864 3865 /* 3866 * See if we already have an entry for this connection. 3867 * If we do, resend the SYN,ACK. We do not count this 3868 * as a retransmission (XXX though maybe we should). 3869 */ 3870 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3871 tcpstat.tcps_sc_dupesyn++; 3872 if (ipopts) { 3873 /* 3874 * If we were remembering a previous source route, 3875 * forget it and use the new one we've been given. 3876 */ 3877 if (sc->sc_ipopts) 3878 (void) m_free(sc->sc_ipopts); 3879 sc->sc_ipopts = ipopts; 3880 } 3881 sc->sc_timestamp = tb.ts_recent; 3882 if (syn_cache_respond(sc, m) == 0) { 3883 tcpstat.tcps_sndacks++; 3884 tcpstat.tcps_sndtotal++; 3885 } 3886 return (1); 3887 } 3888 3889 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3890 if (sc == NULL) { 3891 if (ipopts) 3892 (void) m_free(ipopts); 3893 return (0); 3894 } 3895 3896 /* 3897 * Fill in the cache, and put the necessary IP and TCP 3898 * options into the reply. 3899 */ 3900 bzero(sc, sizeof(struct syn_cache)); 3901 callout_init(&sc->sc_timer); 3902 bcopy(src, &sc->sc_src, src->sa_len); 3903 bcopy(dst, &sc->sc_dst, dst->sa_len); 3904 sc->sc_flags = 0; 3905 sc->sc_ipopts = ipopts; 3906 sc->sc_irs = th->th_seq; 3907 switch (src->sa_family) { 3908 #ifdef INET 3909 case AF_INET: 3910 { 3911 struct sockaddr_in *srcin = (void *) src; 3912 struct sockaddr_in *dstin = (void *) dst; 3913 3914 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 3915 &srcin->sin_addr, dstin->sin_port, 3916 srcin->sin_port, sizeof(dstin->sin_addr), 0); 3917 break; 3918 } 3919 #endif /* INET */ 3920 #ifdef INET6 3921 case AF_INET6: 3922 { 3923 struct sockaddr_in6 *srcin6 = (void *) src; 3924 struct sockaddr_in6 *dstin6 = (void *) dst; 3925 3926 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 3927 &srcin6->sin6_addr, dstin6->sin6_port, 3928 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 3929 break; 3930 } 3931 #endif /* INET6 */ 3932 } 3933 sc->sc_peermaxseg = oi->maxseg; 3934 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3935 m->m_pkthdr.rcvif : NULL, 3936 sc->sc_src.sa.sa_family); 3937 sc->sc_win = win; 3938 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 3939 sc->sc_timestamp = tb.ts_recent; 3940 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 3941 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 3942 sc->sc_flags |= SCF_TIMESTAMP; 3943 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3944 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3945 sc->sc_requested_s_scale = tb.requested_s_scale; 3946 sc->sc_request_r_scale = 0; 3947 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3948 TCP_MAXWIN << sc->sc_request_r_scale < 3949 so->so_rcv.sb_hiwat) 3950 sc->sc_request_r_scale++; 3951 } else { 3952 sc->sc_requested_s_scale = 15; 3953 sc->sc_request_r_scale = 15; 3954 } 3955 #ifdef TCP_SIGNATURE 3956 if (tb.t_flags & TF_SIGNATURE) 3957 sc->sc_flags |= SCF_SIGNATURE; 3958 #endif 3959 sc->sc_tp = tp; 3960 if (syn_cache_respond(sc, m) == 0) { 3961 syn_cache_insert(sc, tp); 3962 tcpstat.tcps_sndacks++; 3963 tcpstat.tcps_sndtotal++; 3964 } else { 3965 SYN_CACHE_PUT(sc); 3966 tcpstat.tcps_sc_dropped++; 3967 } 3968 return (1); 3969 } 3970 3971 int 3972 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 3973 { 3974 struct route *ro; 3975 u_int8_t *optp; 3976 int optlen, error; 3977 u_int16_t tlen; 3978 struct ip *ip = NULL; 3979 #ifdef INET6 3980 struct ip6_hdr *ip6 = NULL; 3981 #endif 3982 struct tcpcb *tp; 3983 struct tcphdr *th; 3984 u_int hlen; 3985 struct socket *so; 3986 3987 switch (sc->sc_src.sa.sa_family) { 3988 case AF_INET: 3989 hlen = sizeof(struct ip); 3990 ro = &sc->sc_route4; 3991 break; 3992 #ifdef INET6 3993 case AF_INET6: 3994 hlen = sizeof(struct ip6_hdr); 3995 ro = (struct route *)&sc->sc_route6; 3996 break; 3997 #endif 3998 default: 3999 if (m) 4000 m_freem(m); 4001 return (EAFNOSUPPORT); 4002 } 4003 4004 /* Compute the size of the TCP options. */ 4005 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4006 #ifdef TCP_SIGNATURE 4007 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4008 #endif 4009 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4010 4011 tlen = hlen + sizeof(struct tcphdr) + optlen; 4012 4013 /* 4014 * Create the IP+TCP header from scratch. 4015 */ 4016 if (m) 4017 m_freem(m); 4018 #ifdef DIAGNOSTIC 4019 if (max_linkhdr + tlen > MCLBYTES) 4020 return (ENOBUFS); 4021 #endif 4022 MGETHDR(m, M_DONTWAIT, MT_DATA); 4023 if (m && tlen > MHLEN) { 4024 MCLGET(m, M_DONTWAIT); 4025 if ((m->m_flags & M_EXT) == 0) { 4026 m_freem(m); 4027 m = NULL; 4028 } 4029 } 4030 if (m == NULL) 4031 return (ENOBUFS); 4032 MCLAIM(m, &tcp_tx_mowner); 4033 4034 /* Fixup the mbuf. */ 4035 m->m_data += max_linkhdr; 4036 m->m_len = m->m_pkthdr.len = tlen; 4037 if (sc->sc_tp) { 4038 tp = sc->sc_tp; 4039 if (tp->t_inpcb) 4040 so = tp->t_inpcb->inp_socket; 4041 #ifdef INET6 4042 else if (tp->t_in6pcb) 4043 so = tp->t_in6pcb->in6p_socket; 4044 #endif 4045 else 4046 so = NULL; 4047 } else 4048 so = NULL; 4049 m->m_pkthdr.rcvif = NULL; 4050 memset(mtod(m, u_char *), 0, tlen); 4051 4052 switch (sc->sc_src.sa.sa_family) { 4053 case AF_INET: 4054 ip = mtod(m, struct ip *); 4055 ip->ip_v = 4; 4056 ip->ip_dst = sc->sc_src.sin.sin_addr; 4057 ip->ip_src = sc->sc_dst.sin.sin_addr; 4058 ip->ip_p = IPPROTO_TCP; 4059 th = (struct tcphdr *)(ip + 1); 4060 th->th_dport = sc->sc_src.sin.sin_port; 4061 th->th_sport = sc->sc_dst.sin.sin_port; 4062 break; 4063 #ifdef INET6 4064 case AF_INET6: 4065 ip6 = mtod(m, struct ip6_hdr *); 4066 ip6->ip6_vfc = IPV6_VERSION; 4067 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4068 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4069 ip6->ip6_nxt = IPPROTO_TCP; 4070 /* ip6_plen will be updated in ip6_output() */ 4071 th = (struct tcphdr *)(ip6 + 1); 4072 th->th_dport = sc->sc_src.sin6.sin6_port; 4073 th->th_sport = sc->sc_dst.sin6.sin6_port; 4074 break; 4075 #endif 4076 default: 4077 th = NULL; 4078 } 4079 4080 th->th_seq = htonl(sc->sc_iss); 4081 th->th_ack = htonl(sc->sc_irs + 1); 4082 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4083 th->th_flags = TH_SYN|TH_ACK; 4084 th->th_win = htons(sc->sc_win); 4085 /* th_sum already 0 */ 4086 /* th_urp already 0 */ 4087 4088 /* Tack on the TCP options. */ 4089 optp = (u_int8_t *)(th + 1); 4090 *optp++ = TCPOPT_MAXSEG; 4091 *optp++ = 4; 4092 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4093 *optp++ = sc->sc_ourmaxseg & 0xff; 4094 4095 if (sc->sc_request_r_scale != 15) { 4096 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4097 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4098 sc->sc_request_r_scale); 4099 optp += 4; 4100 } 4101 4102 if (sc->sc_flags & SCF_TIMESTAMP) { 4103 u_int32_t *lp = (u_int32_t *)(optp); 4104 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4105 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4106 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4107 *lp = htonl(sc->sc_timestamp); 4108 optp += TCPOLEN_TSTAMP_APPA; 4109 } 4110 4111 #ifdef TCP_SIGNATURE 4112 if (sc->sc_flags & SCF_SIGNATURE) { 4113 struct secasvar *sav; 4114 u_int8_t *sigp; 4115 4116 sav = tcp_signature_getsav(m, th); 4117 4118 if (sav == NULL) { 4119 if (m) 4120 m_freem(m); 4121 return (EPERM); 4122 } 4123 4124 *optp++ = TCPOPT_SIGNATURE; 4125 *optp++ = TCPOLEN_SIGNATURE; 4126 sigp = optp; 4127 bzero(optp, TCP_SIGLEN); 4128 optp += TCP_SIGLEN; 4129 *optp++ = TCPOPT_NOP; 4130 *optp++ = TCPOPT_EOL; 4131 4132 (void)tcp_signature(m, th, hlen, sav, sigp); 4133 4134 key_sa_recordxfer(sav, m); 4135 #ifdef FAST_IPSEC 4136 KEY_FREESAV(&sav); 4137 #else 4138 key_freesav(sav); 4139 #endif 4140 } 4141 #endif 4142 4143 /* Compute the packet's checksum. */ 4144 switch (sc->sc_src.sa.sa_family) { 4145 case AF_INET: 4146 ip->ip_len = htons(tlen - hlen); 4147 th->th_sum = 0; 4148 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4149 break; 4150 #ifdef INET6 4151 case AF_INET6: 4152 ip6->ip6_plen = htons(tlen - hlen); 4153 th->th_sum = 0; 4154 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4155 break; 4156 #endif 4157 } 4158 4159 /* 4160 * Fill in some straggling IP bits. Note the stack expects 4161 * ip_len to be in host order, for convenience. 4162 */ 4163 switch (sc->sc_src.sa.sa_family) { 4164 #ifdef INET 4165 case AF_INET: 4166 ip->ip_len = htons(tlen); 4167 ip->ip_ttl = ip_defttl; 4168 /* XXX tos? */ 4169 break; 4170 #endif 4171 #ifdef INET6 4172 case AF_INET6: 4173 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4174 ip6->ip6_vfc |= IPV6_VERSION; 4175 ip6->ip6_plen = htons(tlen - hlen); 4176 /* ip6_hlim will be initialized afterwards */ 4177 /* XXX flowlabel? */ 4178 break; 4179 #endif 4180 } 4181 4182 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4183 tp = sc->sc_tp; 4184 4185 switch (sc->sc_src.sa.sa_family) { 4186 #ifdef INET 4187 case AF_INET: 4188 error = ip_output(m, sc->sc_ipopts, ro, 4189 (ip_mtudisc ? IP_MTUDISC : 0), 4190 (struct ip_moptions *)NULL, so); 4191 break; 4192 #endif 4193 #ifdef INET6 4194 case AF_INET6: 4195 ip6->ip6_hlim = in6_selecthlim(NULL, 4196 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 4197 4198 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0, 4199 (struct ip6_moptions *)0, so, NULL); 4200 break; 4201 #endif 4202 default: 4203 error = EAFNOSUPPORT; 4204 break; 4205 } 4206 return (error); 4207 } 4208