1 /* $NetBSD: tcp_input.c,v 1.207 2004/05/23 00:37:27 jonathan 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 1.207 2004/05/23 00:37:27 jonathan Exp $"); 152 153 #include "opt_inet.h" 154 #include "opt_ipsec.h" 155 #include "opt_inet_csum.h" 156 #include "opt_tcp_debug.h" 157 158 #include <sys/param.h> 159 #include <sys/systm.h> 160 #include <sys/malloc.h> 161 #include <sys/mbuf.h> 162 #include <sys/protosw.h> 163 #include <sys/socket.h> 164 #include <sys/socketvar.h> 165 #include <sys/errno.h> 166 #include <sys/syslog.h> 167 #include <sys/pool.h> 168 #include <sys/domain.h> 169 #include <sys/kernel.h> 170 #ifdef TCP_SIGNATURE 171 #include <sys/md5.h> 172 #endif 173 174 #include <net/if.h> 175 #include <net/route.h> 176 #include <net/if_types.h> 177 178 #include <netinet/in.h> 179 #include <netinet/in_systm.h> 180 #include <netinet/ip.h> 181 #include <netinet/in_pcb.h> 182 #include <netinet/in_var.h> 183 #include <netinet/ip_var.h> 184 185 #ifdef INET6 186 #ifndef INET 187 #include <netinet/in.h> 188 #endif 189 #include <netinet/ip6.h> 190 #include <netinet6/ip6_var.h> 191 #include <netinet6/in6_pcb.h> 192 #include <netinet6/ip6_var.h> 193 #include <netinet6/in6_var.h> 194 #include <netinet/icmp6.h> 195 #include <netinet6/nd6.h> 196 #endif 197 198 #ifndef INET6 199 /* always need ip6.h for IP6_EXTHDR_GET */ 200 #include <netinet/ip6.h> 201 #endif 202 203 #include <netinet/tcp.h> 204 #include <netinet/tcp_fsm.h> 205 #include <netinet/tcp_seq.h> 206 #include <netinet/tcp_timer.h> 207 #include <netinet/tcp_var.h> 208 #include <netinet/tcpip.h> 209 #include <netinet/tcp_debug.h> 210 211 #include <machine/stdarg.h> 212 213 #ifdef IPSEC 214 #include <netinet6/ipsec.h> 215 #include <netkey/key.h> 216 #endif /*IPSEC*/ 217 #ifdef INET6 218 #include "faith.h" 219 #if defined(NFAITH) && NFAITH > 0 220 #include <net/if_faith.h> 221 #endif 222 #endif /* IPSEC */ 223 224 #ifdef FAST_IPSEC 225 #include <netipsec/ipsec.h> 226 #include <netipsec/ipsec_var.h> /* XXX ipsecstat namespace */ 227 #include <netipsec/key.h> 228 #ifdef INET6 229 #include <netipsec/ipsec6.h> 230 #endif 231 #endif /* FAST_IPSEC*/ 232 233 int tcprexmtthresh = 3; 234 int tcp_log_refused; 235 236 static int tcp_rst_ppslim_count = 0; 237 static struct timeval tcp_rst_ppslim_last; 238 static int tcp_ackdrop_ppslim_count = 0; 239 static struct timeval tcp_ackdrop_ppslim_last; 240 241 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 242 243 /* for modulo comparisons of timestamps */ 244 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 245 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 246 247 /* 248 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 249 */ 250 #ifdef INET6 251 #define ND6_HINT(tp) \ 252 do { \ 253 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 && \ 254 tp->t_in6pcb->in6p_route.ro_rt) { \ 255 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \ 256 } \ 257 } while (/*CONSTCOND*/ 0) 258 #else 259 #define ND6_HINT(tp) 260 #endif 261 262 /* 263 * Macro to compute ACK transmission behavior. Delay the ACK unless 264 * we have already delayed an ACK (must send an ACK every two segments). 265 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 266 * option is enabled. 267 */ 268 #define TCP_SETUP_ACK(tp, th) \ 269 do { \ 270 if ((tp)->t_flags & TF_DELACK || \ 271 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 272 tp->t_flags |= TF_ACKNOW; \ 273 else \ 274 TCP_SET_DELACK(tp); \ 275 } while (/*CONSTCOND*/ 0) 276 277 /* 278 * Convert TCP protocol fields to host order for easier processing. 279 */ 280 #define TCP_FIELDS_TO_HOST(th) \ 281 do { \ 282 NTOHL((th)->th_seq); \ 283 NTOHL((th)->th_ack); \ 284 NTOHS((th)->th_win); \ 285 NTOHS((th)->th_urp); \ 286 } while (/*CONSTCOND*/ 0) 287 288 /* 289 * ... and reverse the above. 290 */ 291 #define TCP_FIELDS_TO_NET(th) \ 292 do { \ 293 HTONL((th)->th_seq); \ 294 HTONL((th)->th_ack); \ 295 HTONS((th)->th_win); \ 296 HTONS((th)->th_urp); \ 297 } while (/*CONSTCOND*/ 0) 298 299 #ifdef TCP_CSUM_COUNTERS 300 #include <sys/device.h> 301 302 extern struct evcnt tcp_hwcsum_ok; 303 extern struct evcnt tcp_hwcsum_bad; 304 extern struct evcnt tcp_hwcsum_data; 305 extern struct evcnt tcp_swcsum; 306 307 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 308 309 #else 310 311 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 312 313 #endif /* TCP_CSUM_COUNTERS */ 314 315 #ifdef TCP_REASS_COUNTERS 316 #include <sys/device.h> 317 318 extern struct evcnt tcp_reass_; 319 extern struct evcnt tcp_reass_empty; 320 extern struct evcnt tcp_reass_iteration[8]; 321 extern struct evcnt tcp_reass_prependfirst; 322 extern struct evcnt tcp_reass_prepend; 323 extern struct evcnt tcp_reass_insert; 324 extern struct evcnt tcp_reass_inserttail; 325 extern struct evcnt tcp_reass_append; 326 extern struct evcnt tcp_reass_appendtail; 327 extern struct evcnt tcp_reass_overlaptail; 328 extern struct evcnt tcp_reass_overlapfront; 329 extern struct evcnt tcp_reass_segdup; 330 extern struct evcnt tcp_reass_fragdup; 331 332 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 333 334 #else 335 336 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 337 338 #endif /* TCP_REASS_COUNTERS */ 339 340 #ifdef INET 341 static void tcp4_log_refused __P((const struct ip *, const struct tcphdr *)); 342 #endif 343 #ifdef INET6 344 static void tcp6_log_refused 345 __P((const struct ip6_hdr *, const struct tcphdr *)); 346 #endif 347 348 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 349 350 int 351 tcp_reass(tp, th, m, tlen) 352 struct tcpcb *tp; 353 struct tcphdr *th; 354 struct mbuf *m; 355 int *tlen; 356 { 357 struct ipqent *p, *q, *nq, *tiqe = NULL; 358 struct socket *so = NULL; 359 int pkt_flags; 360 tcp_seq pkt_seq; 361 unsigned pkt_len; 362 u_long rcvpartdupbyte = 0; 363 u_long rcvoobyte; 364 #ifdef TCP_REASS_COUNTERS 365 u_int count = 0; 366 #endif 367 368 if (tp->t_inpcb) 369 so = tp->t_inpcb->inp_socket; 370 #ifdef INET6 371 else if (tp->t_in6pcb) 372 so = tp->t_in6pcb->in6p_socket; 373 #endif 374 375 TCP_REASS_LOCK_CHECK(tp); 376 377 /* 378 * Call with th==0 after become established to 379 * force pre-ESTABLISHED data up to user socket. 380 */ 381 if (th == 0) 382 goto present; 383 384 rcvoobyte = *tlen; 385 /* 386 * Copy these to local variables because the tcpiphdr 387 * gets munged while we are collapsing mbufs. 388 */ 389 pkt_seq = th->th_seq; 390 pkt_len = *tlen; 391 pkt_flags = th->th_flags; 392 393 TCP_REASS_COUNTER_INCR(&tcp_reass_); 394 395 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 396 /* 397 * When we miss a packet, the vast majority of time we get 398 * packets that follow it in order. So optimize for that. 399 */ 400 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 401 p->ipqe_len += pkt_len; 402 p->ipqe_flags |= pkt_flags; 403 m_cat(p->ipre_mlast, m); 404 TRAVERSE(p->ipre_mlast); 405 m = NULL; 406 tiqe = p; 407 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 408 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 409 goto skip_replacement; 410 } 411 /* 412 * While we're here, if the pkt is completely beyond 413 * anything we have, just insert it at the tail. 414 */ 415 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 416 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 417 goto insert_it; 418 } 419 } 420 421 q = TAILQ_FIRST(&tp->segq); 422 423 if (q != NULL) { 424 /* 425 * If this segment immediately precedes the first out-of-order 426 * block, simply slap the segment in front of it and (mostly) 427 * skip the complicated logic. 428 */ 429 if (pkt_seq + pkt_len == q->ipqe_seq) { 430 q->ipqe_seq = pkt_seq; 431 q->ipqe_len += pkt_len; 432 q->ipqe_flags |= pkt_flags; 433 m_cat(m, q->ipqe_m); 434 q->ipqe_m = m; 435 q->ipre_mlast = m; /* last mbuf may have changed */ 436 TRAVERSE(q->ipre_mlast); 437 tiqe = q; 438 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 439 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 440 goto skip_replacement; 441 } 442 } else { 443 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 444 } 445 446 /* 447 * Find a segment which begins after this one does. 448 */ 449 for (p = NULL; q != NULL; q = nq) { 450 nq = TAILQ_NEXT(q, ipqe_q); 451 #ifdef TCP_REASS_COUNTERS 452 count++; 453 #endif 454 /* 455 * If the received segment is just right after this 456 * fragment, merge the two together and then check 457 * for further overlaps. 458 */ 459 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 460 #ifdef TCPREASS_DEBUG 461 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 462 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 463 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 464 #endif 465 pkt_len += q->ipqe_len; 466 pkt_flags |= q->ipqe_flags; 467 pkt_seq = q->ipqe_seq; 468 m_cat(q->ipre_mlast, m); 469 TRAVERSE(q->ipre_mlast); 470 m = q->ipqe_m; 471 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 472 goto free_ipqe; 473 } 474 /* 475 * If the received segment is completely past this 476 * fragment, we need to go the next fragment. 477 */ 478 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 479 p = q; 480 continue; 481 } 482 /* 483 * If the fragment is past the received segment, 484 * it (or any following) can't be concatenated. 485 */ 486 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 487 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 488 break; 489 } 490 491 /* 492 * We've received all the data in this segment before. 493 * mark it as a duplicate and return. 494 */ 495 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 496 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 497 tcpstat.tcps_rcvduppack++; 498 tcpstat.tcps_rcvdupbyte += pkt_len; 499 m_freem(m); 500 if (tiqe != NULL) 501 pool_put(&ipqent_pool, tiqe); 502 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 503 return (0); 504 } 505 /* 506 * Received segment completely overlaps this fragment 507 * so we drop the fragment (this keeps the temporal 508 * ordering of segments correct). 509 */ 510 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 511 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 512 rcvpartdupbyte += q->ipqe_len; 513 m_freem(q->ipqe_m); 514 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 515 goto free_ipqe; 516 } 517 /* 518 * RX'ed segment extends past the end of the 519 * fragment. Drop the overlapping bytes. Then 520 * merge the fragment and segment then treat as 521 * a longer received packet. 522 */ 523 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 524 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 525 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 526 #ifdef TCPREASS_DEBUG 527 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 528 tp, overlap, 529 pkt_seq, pkt_seq + pkt_len, pkt_len); 530 #endif 531 m_adj(m, overlap); 532 rcvpartdupbyte += overlap; 533 m_cat(q->ipre_mlast, m); 534 TRAVERSE(q->ipre_mlast); 535 m = q->ipqe_m; 536 pkt_seq = q->ipqe_seq; 537 pkt_len += q->ipqe_len - overlap; 538 rcvoobyte -= overlap; 539 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 540 goto free_ipqe; 541 } 542 /* 543 * RX'ed segment extends past the front of the 544 * fragment. Drop the overlapping bytes on the 545 * received packet. The packet will then be 546 * contatentated with this fragment a bit later. 547 */ 548 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 549 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 550 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 551 #ifdef TCPREASS_DEBUG 552 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 553 tp, overlap, 554 pkt_seq, pkt_seq + pkt_len, pkt_len); 555 #endif 556 m_adj(m, -overlap); 557 pkt_len -= overlap; 558 rcvpartdupbyte += overlap; 559 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 560 rcvoobyte -= overlap; 561 } 562 /* 563 * If the received segment immediates precedes this 564 * fragment then tack the fragment onto this segment 565 * and reinsert the data. 566 */ 567 if (q->ipqe_seq == pkt_seq + pkt_len) { 568 #ifdef TCPREASS_DEBUG 569 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 570 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 571 pkt_seq, pkt_seq + pkt_len, pkt_len); 572 #endif 573 pkt_len += q->ipqe_len; 574 pkt_flags |= q->ipqe_flags; 575 m_cat(m, q->ipqe_m); 576 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 577 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 578 if (tiqe == NULL) 579 tiqe = q; 580 else 581 pool_put(&ipqent_pool, q); 582 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 583 break; 584 } 585 /* 586 * If the fragment is before the segment, remember it. 587 * When this loop is terminated, p will contain the 588 * pointer to fragment that is right before the received 589 * segment. 590 */ 591 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 592 p = q; 593 594 continue; 595 596 /* 597 * This is a common operation. It also will allow 598 * to save doing a malloc/free in most instances. 599 */ 600 free_ipqe: 601 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 602 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 603 if (tiqe == NULL) 604 tiqe = q; 605 else 606 pool_put(&ipqent_pool, q); 607 } 608 609 #ifdef TCP_REASS_COUNTERS 610 if (count > 7) 611 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 612 else if (count > 0) 613 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 614 #endif 615 616 insert_it: 617 618 /* 619 * Allocate a new queue entry since the received segment did not 620 * collapse onto any other out-of-order block; thus we are allocating 621 * a new block. If it had collapsed, tiqe would not be NULL and 622 * we would be reusing it. 623 * XXX If we can't, just drop the packet. XXX 624 */ 625 if (tiqe == NULL) { 626 tiqe = pool_get(&ipqent_pool, PR_NOWAIT); 627 if (tiqe == NULL) { 628 tcpstat.tcps_rcvmemdrop++; 629 m_freem(m); 630 return (0); 631 } 632 } 633 634 /* 635 * Update the counters. 636 */ 637 tcpstat.tcps_rcvoopack++; 638 tcpstat.tcps_rcvoobyte += rcvoobyte; 639 if (rcvpartdupbyte) { 640 tcpstat.tcps_rcvpartduppack++; 641 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 642 } 643 644 /* 645 * Insert the new fragment queue entry into both queues. 646 */ 647 tiqe->ipqe_m = m; 648 tiqe->ipre_mlast = m; 649 tiqe->ipqe_seq = pkt_seq; 650 tiqe->ipqe_len = pkt_len; 651 tiqe->ipqe_flags = pkt_flags; 652 if (p == NULL) { 653 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 654 #ifdef TCPREASS_DEBUG 655 if (tiqe->ipqe_seq != tp->rcv_nxt) 656 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 657 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 658 #endif 659 } else { 660 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 661 #ifdef TCPREASS_DEBUG 662 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 663 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 664 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 665 #endif 666 } 667 668 skip_replacement: 669 670 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 671 672 present: 673 /* 674 * Present data to user, advancing rcv_nxt through 675 * completed sequence space. 676 */ 677 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 678 return (0); 679 q = TAILQ_FIRST(&tp->segq); 680 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 681 return (0); 682 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 683 return (0); 684 685 tp->rcv_nxt += q->ipqe_len; 686 pkt_flags = q->ipqe_flags & TH_FIN; 687 ND6_HINT(tp); 688 689 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 690 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 691 if (so->so_state & SS_CANTRCVMORE) 692 m_freem(q->ipqe_m); 693 else 694 sbappendstream(&so->so_rcv, q->ipqe_m); 695 pool_put(&ipqent_pool, q); 696 sorwakeup(so); 697 return (pkt_flags); 698 } 699 700 #ifdef INET6 701 int 702 tcp6_input(mp, offp, proto) 703 struct mbuf **mp; 704 int *offp, proto; 705 { 706 struct mbuf *m = *mp; 707 708 /* 709 * draft-itojun-ipv6-tcp-to-anycast 710 * better place to put this in? 711 */ 712 if (m->m_flags & M_ANYCAST6) { 713 struct ip6_hdr *ip6; 714 if (m->m_len < sizeof(struct ip6_hdr)) { 715 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 716 tcpstat.tcps_rcvshort++; 717 return IPPROTO_DONE; 718 } 719 } 720 ip6 = mtod(m, struct ip6_hdr *); 721 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 722 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 723 return IPPROTO_DONE; 724 } 725 726 tcp_input(m, *offp, proto); 727 return IPPROTO_DONE; 728 } 729 #endif 730 731 #ifdef INET 732 static void 733 tcp4_log_refused(ip, th) 734 const struct ip *ip; 735 const struct tcphdr *th; 736 { 737 char src[4*sizeof "123"]; 738 char dst[4*sizeof "123"]; 739 740 if (ip) { 741 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 742 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 743 } 744 else { 745 strlcpy(src, "(unknown)", sizeof(src)); 746 strlcpy(dst, "(unknown)", sizeof(dst)); 747 } 748 log(LOG_INFO, 749 "Connection attempt to TCP %s:%d from %s:%d\n", 750 dst, ntohs(th->th_dport), 751 src, ntohs(th->th_sport)); 752 } 753 #endif 754 755 #ifdef INET6 756 static void 757 tcp6_log_refused(ip6, th) 758 const struct ip6_hdr *ip6; 759 const struct tcphdr *th; 760 { 761 char src[INET6_ADDRSTRLEN]; 762 char dst[INET6_ADDRSTRLEN]; 763 764 if (ip6) { 765 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 766 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 767 } 768 else { 769 strlcpy(src, "(unknown v6)", sizeof(src)); 770 strlcpy(dst, "(unknown v6)", sizeof(dst)); 771 } 772 log(LOG_INFO, 773 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 774 dst, ntohs(th->th_dport), 775 src, ntohs(th->th_sport)); 776 } 777 #endif 778 779 /* 780 * TCP input routine, follows pages 65-76 of the 781 * protocol specification dated September, 1981 very closely. 782 */ 783 void 784 tcp_input(struct mbuf *m, ...) 785 { 786 struct tcphdr *th; 787 struct ip *ip; 788 struct inpcb *inp; 789 #ifdef INET6 790 struct ip6_hdr *ip6; 791 struct in6pcb *in6p; 792 #endif 793 u_int8_t *optp = NULL; 794 int optlen = 0; 795 int len, tlen, toff, hdroptlen = 0; 796 struct tcpcb *tp = 0; 797 int tiflags; 798 struct socket *so = NULL; 799 int todrop, acked, ourfinisacked, needoutput = 0; 800 #ifdef TCP_DEBUG 801 short ostate = 0; 802 #endif 803 int iss = 0; 804 u_long tiwin; 805 struct tcp_opt_info opti; 806 int off, iphlen; 807 va_list ap; 808 int af; /* af on the wire */ 809 struct mbuf *tcp_saveti = NULL; 810 811 MCLAIM(m, &tcp_rx_mowner); 812 va_start(ap, m); 813 toff = va_arg(ap, int); 814 (void)va_arg(ap, int); /* ignore value, advance ap */ 815 va_end(ap); 816 817 tcpstat.tcps_rcvtotal++; 818 819 bzero(&opti, sizeof(opti)); 820 opti.ts_present = 0; 821 opti.maxseg = 0; 822 823 /* 824 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 825 * 826 * TCP is, by definition, unicast, so we reject all 827 * multicast outright. 828 * 829 * Note, there are additional src/dst address checks in 830 * the AF-specific code below. 831 */ 832 if (m->m_flags & (M_BCAST|M_MCAST)) { 833 /* XXX stat */ 834 goto drop; 835 } 836 #ifdef INET6 837 if (m->m_flags & M_ANYCAST6) { 838 /* XXX stat */ 839 goto drop; 840 } 841 #endif 842 843 /* 844 * Get IP and TCP header. 845 * Note: IP leaves IP header in first mbuf. 846 */ 847 ip = mtod(m, struct ip *); 848 #ifdef INET6 849 ip6 = NULL; 850 #endif 851 switch (ip->ip_v) { 852 #ifdef INET 853 case 4: 854 af = AF_INET; 855 iphlen = sizeof(struct ip); 856 ip = mtod(m, struct ip *); 857 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 858 sizeof(struct tcphdr)); 859 if (th == NULL) { 860 tcpstat.tcps_rcvshort++; 861 return; 862 } 863 /* We do the checksum after PCB lookup... */ 864 len = ntohs(ip->ip_len); 865 tlen = len - toff; 866 break; 867 #endif 868 #ifdef INET6 869 case 6: 870 ip = NULL; 871 iphlen = sizeof(struct ip6_hdr); 872 af = AF_INET6; 873 ip6 = mtod(m, struct ip6_hdr *); 874 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 875 sizeof(struct tcphdr)); 876 if (th == NULL) { 877 tcpstat.tcps_rcvshort++; 878 return; 879 } 880 881 /* Be proactive about malicious use of IPv4 mapped address */ 882 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 883 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 884 /* XXX stat */ 885 goto drop; 886 } 887 888 /* 889 * Be proactive about unspecified IPv6 address in source. 890 * As we use all-zero to indicate unbounded/unconnected pcb, 891 * unspecified IPv6 address can be used to confuse us. 892 * 893 * Note that packets with unspecified IPv6 destination is 894 * already dropped in ip6_input. 895 */ 896 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 897 /* XXX stat */ 898 goto drop; 899 } 900 901 /* 902 * Make sure destination address is not multicast. 903 * Source address checked in ip6_input(). 904 */ 905 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 906 /* XXX stat */ 907 goto drop; 908 } 909 910 /* We do the checksum after PCB lookup... */ 911 len = m->m_pkthdr.len; 912 tlen = len - toff; 913 break; 914 #endif 915 default: 916 m_freem(m); 917 return; 918 } 919 920 KASSERT(TCP_HDR_ALIGNED_P(th)); 921 922 /* 923 * Check that TCP offset makes sense, 924 * pull out TCP options and adjust length. XXX 925 */ 926 off = th->th_off << 2; 927 if (off < sizeof (struct tcphdr) || off > tlen) { 928 tcpstat.tcps_rcvbadoff++; 929 goto drop; 930 } 931 tlen -= off; 932 933 /* 934 * tcp_input() has been modified to use tlen to mean the TCP data 935 * length throughout the function. Other functions can use 936 * m->m_pkthdr.len as the basis for calculating the TCP data length. 937 * rja 938 */ 939 940 if (off > sizeof (struct tcphdr)) { 941 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 942 if (th == NULL) { 943 tcpstat.tcps_rcvshort++; 944 return; 945 } 946 /* 947 * NOTE: ip/ip6 will not be affected by m_pulldown() 948 * (as they're before toff) and we don't need to update those. 949 */ 950 KASSERT(TCP_HDR_ALIGNED_P(th)); 951 optlen = off - sizeof (struct tcphdr); 952 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 953 /* 954 * Do quick retrieval of timestamp options ("options 955 * prediction?"). If timestamp is the only option and it's 956 * formatted as recommended in RFC 1323 appendix A, we 957 * quickly get the values now and not bother calling 958 * tcp_dooptions(), etc. 959 */ 960 if ((optlen == TCPOLEN_TSTAMP_APPA || 961 (optlen > TCPOLEN_TSTAMP_APPA && 962 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 963 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 964 (th->th_flags & TH_SYN) == 0) { 965 opti.ts_present = 1; 966 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 967 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 968 optp = NULL; /* we've parsed the options */ 969 } 970 } 971 tiflags = th->th_flags; 972 973 /* 974 * Locate pcb for segment. 975 */ 976 findpcb: 977 inp = NULL; 978 #ifdef INET6 979 in6p = NULL; 980 #endif 981 switch (af) { 982 #ifdef INET 983 case AF_INET: 984 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 985 ip->ip_dst, th->th_dport); 986 if (inp == 0) { 987 ++tcpstat.tcps_pcbhashmiss; 988 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 989 } 990 #ifdef INET6 991 if (inp == 0) { 992 struct in6_addr s, d; 993 994 /* mapped addr case */ 995 bzero(&s, sizeof(s)); 996 s.s6_addr16[5] = htons(0xffff); 997 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 998 bzero(&d, sizeof(d)); 999 d.s6_addr16[5] = htons(0xffff); 1000 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1001 in6p = in6_pcblookup_connect(&tcbtable, &s, 1002 th->th_sport, &d, th->th_dport, 0); 1003 if (in6p == 0) { 1004 ++tcpstat.tcps_pcbhashmiss; 1005 in6p = in6_pcblookup_bind(&tcbtable, &d, 1006 th->th_dport, 0); 1007 } 1008 } 1009 #endif 1010 #ifndef INET6 1011 if (inp == 0) 1012 #else 1013 if (inp == 0 && in6p == 0) 1014 #endif 1015 { 1016 ++tcpstat.tcps_noport; 1017 if (tcp_log_refused && 1018 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1019 tcp4_log_refused(ip, th); 1020 } 1021 TCP_FIELDS_TO_HOST(th); 1022 goto dropwithreset_ratelim; 1023 } 1024 #if defined(IPSEC) || defined(FAST_IPSEC) 1025 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1026 ipsec4_in_reject(m, inp)) { 1027 ipsecstat.in_polvio++; 1028 goto drop; 1029 } 1030 #ifdef INET6 1031 else if (in6p && 1032 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1033 ipsec4_in_reject_so(m, in6p->in6p_socket)) { 1034 ipsecstat.in_polvio++; 1035 goto drop; 1036 } 1037 #endif 1038 #endif /*IPSEC*/ 1039 break; 1040 #endif /*INET*/ 1041 #ifdef INET6 1042 case AF_INET6: 1043 { 1044 int faith; 1045 1046 #if defined(NFAITH) && NFAITH > 0 1047 faith = faithprefix(&ip6->ip6_dst); 1048 #else 1049 faith = 0; 1050 #endif 1051 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1052 th->th_sport, &ip6->ip6_dst, th->th_dport, faith); 1053 if (in6p == NULL) { 1054 ++tcpstat.tcps_pcbhashmiss; 1055 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1056 th->th_dport, faith); 1057 } 1058 if (in6p == NULL) { 1059 ++tcpstat.tcps_noport; 1060 if (tcp_log_refused && 1061 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1062 tcp6_log_refused(ip6, th); 1063 } 1064 TCP_FIELDS_TO_HOST(th); 1065 goto dropwithreset_ratelim; 1066 } 1067 #if defined(IPSEC) || defined(FAST_IPSEC) 1068 if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1069 ipsec6_in_reject(m, in6p)) { 1070 ipsec6stat.in_polvio++; 1071 goto drop; 1072 } 1073 #endif /*IPSEC*/ 1074 break; 1075 } 1076 #endif 1077 } 1078 1079 /* 1080 * If the state is CLOSED (i.e., TCB does not exist) then 1081 * all data in the incoming segment is discarded. 1082 * If the TCB exists but is in CLOSED state, it is embryonic, 1083 * but should either do a listen or a connect soon. 1084 */ 1085 tp = NULL; 1086 so = NULL; 1087 if (inp) { 1088 tp = intotcpcb(inp); 1089 so = inp->inp_socket; 1090 } 1091 #ifdef INET6 1092 else if (in6p) { 1093 tp = in6totcpcb(in6p); 1094 so = in6p->in6p_socket; 1095 } 1096 #endif 1097 if (tp == 0) { 1098 TCP_FIELDS_TO_HOST(th); 1099 goto dropwithreset_ratelim; 1100 } 1101 if (tp->t_state == TCPS_CLOSED) 1102 goto drop; 1103 1104 /* 1105 * Checksum extended TCP header and data. 1106 */ 1107 switch (af) { 1108 #ifdef INET 1109 case AF_INET: 1110 switch (m->m_pkthdr.csum_flags & 1111 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 1112 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 1113 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 1114 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 1115 goto badcsum; 1116 1117 case M_CSUM_TCPv4|M_CSUM_DATA: { 1118 u_int32_t hw_csum = m->m_pkthdr.csum_data; 1119 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 1120 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 1121 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 1122 ip->ip_dst.s_addr, 1123 htons(hw_csum + tlen + off + IPPROTO_TCP)); 1124 } 1125 if ((hw_csum ^ 0xffff) != 0) 1126 goto badcsum; 1127 break; 1128 } 1129 1130 case M_CSUM_TCPv4: 1131 /* Checksum was okay. */ 1132 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 1133 break; 1134 1135 default: 1136 /* Must compute it ourselves. */ 1137 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 1138 if (in4_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 1139 goto badcsum; 1140 break; 1141 } 1142 break; 1143 #endif /* INET4 */ 1144 1145 #ifdef INET6 1146 case AF_INET6: 1147 if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 1148 goto badcsum; 1149 break; 1150 #endif /* INET6 */ 1151 } 1152 1153 TCP_FIELDS_TO_HOST(th); 1154 1155 /* Unscale the window into a 32-bit value. */ 1156 if ((tiflags & TH_SYN) == 0) 1157 tiwin = th->th_win << tp->snd_scale; 1158 else 1159 tiwin = th->th_win; 1160 1161 #ifdef INET6 1162 /* save packet options if user wanted */ 1163 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1164 if (in6p->in6p_options) { 1165 m_freem(in6p->in6p_options); 1166 in6p->in6p_options = 0; 1167 } 1168 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1169 } 1170 #endif 1171 1172 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1173 union syn_cache_sa src; 1174 union syn_cache_sa dst; 1175 1176 bzero(&src, sizeof(src)); 1177 bzero(&dst, sizeof(dst)); 1178 switch (af) { 1179 #ifdef INET 1180 case AF_INET: 1181 src.sin.sin_len = sizeof(struct sockaddr_in); 1182 src.sin.sin_family = AF_INET; 1183 src.sin.sin_addr = ip->ip_src; 1184 src.sin.sin_port = th->th_sport; 1185 1186 dst.sin.sin_len = sizeof(struct sockaddr_in); 1187 dst.sin.sin_family = AF_INET; 1188 dst.sin.sin_addr = ip->ip_dst; 1189 dst.sin.sin_port = th->th_dport; 1190 break; 1191 #endif 1192 #ifdef INET6 1193 case AF_INET6: 1194 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1195 src.sin6.sin6_family = AF_INET6; 1196 src.sin6.sin6_addr = ip6->ip6_src; 1197 src.sin6.sin6_port = th->th_sport; 1198 1199 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1200 dst.sin6.sin6_family = AF_INET6; 1201 dst.sin6.sin6_addr = ip6->ip6_dst; 1202 dst.sin6.sin6_port = th->th_dport; 1203 break; 1204 #endif /* INET6 */ 1205 default: 1206 goto badsyn; /*sanity*/ 1207 } 1208 1209 if (so->so_options & SO_DEBUG) { 1210 #ifdef TCP_DEBUG 1211 ostate = tp->t_state; 1212 #endif 1213 1214 tcp_saveti = NULL; 1215 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1216 goto nosave; 1217 1218 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1219 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1220 if (!tcp_saveti) 1221 goto nosave; 1222 } else { 1223 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1224 if (!tcp_saveti) 1225 goto nosave; 1226 MCLAIM(m, &tcp_mowner); 1227 tcp_saveti->m_len = iphlen; 1228 m_copydata(m, 0, iphlen, 1229 mtod(tcp_saveti, caddr_t)); 1230 } 1231 1232 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1233 m_freem(tcp_saveti); 1234 tcp_saveti = NULL; 1235 } else { 1236 tcp_saveti->m_len += sizeof(struct tcphdr); 1237 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1238 sizeof(struct tcphdr)); 1239 } 1240 nosave:; 1241 } 1242 if (so->so_options & SO_ACCEPTCONN) { 1243 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1244 if (tiflags & TH_RST) { 1245 syn_cache_reset(&src.sa, &dst.sa, th); 1246 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1247 (TH_ACK|TH_SYN)) { 1248 /* 1249 * Received a SYN,ACK. This should 1250 * never happen while we are in 1251 * LISTEN. Send an RST. 1252 */ 1253 goto badsyn; 1254 } else if (tiflags & TH_ACK) { 1255 so = syn_cache_get(&src.sa, &dst.sa, 1256 th, toff, tlen, so, m); 1257 if (so == NULL) { 1258 /* 1259 * We don't have a SYN for 1260 * this ACK; send an RST. 1261 */ 1262 goto badsyn; 1263 } else if (so == 1264 (struct socket *)(-1)) { 1265 /* 1266 * We were unable to create 1267 * the connection. If the 1268 * 3-way handshake was 1269 * completed, and RST has 1270 * been sent to the peer. 1271 * Since the mbuf might be 1272 * in use for the reply, 1273 * do not free it. 1274 */ 1275 m = NULL; 1276 } else { 1277 /* 1278 * We have created a 1279 * full-blown connection. 1280 */ 1281 tp = NULL; 1282 inp = NULL; 1283 #ifdef INET6 1284 in6p = NULL; 1285 #endif 1286 switch (so->so_proto->pr_domain->dom_family) { 1287 #ifdef INET 1288 case AF_INET: 1289 inp = sotoinpcb(so); 1290 tp = intotcpcb(inp); 1291 break; 1292 #endif 1293 #ifdef INET6 1294 case AF_INET6: 1295 in6p = sotoin6pcb(so); 1296 tp = in6totcpcb(in6p); 1297 break; 1298 #endif 1299 } 1300 if (tp == NULL) 1301 goto badsyn; /*XXX*/ 1302 tiwin <<= tp->snd_scale; 1303 goto after_listen; 1304 } 1305 } else { 1306 /* 1307 * None of RST, SYN or ACK was set. 1308 * This is an invalid packet for a 1309 * TCB in LISTEN state. Send a RST. 1310 */ 1311 goto badsyn; 1312 } 1313 } else { 1314 /* 1315 * Received a SYN. 1316 */ 1317 1318 #ifdef INET6 1319 /* 1320 * If deprecated address is forbidden, we do 1321 * not accept SYN to deprecated interface 1322 * address to prevent any new inbound 1323 * connection from getting established. 1324 * When we do not accept SYN, we send a TCP 1325 * RST, with deprecated source address (instead 1326 * of dropping it). We compromise it as it is 1327 * much better for peer to send a RST, and 1328 * RST will be the final packet for the 1329 * exchange. 1330 * 1331 * If we do not forbid deprecated addresses, we 1332 * accept the SYN packet. RFC2462 does not 1333 * suggest dropping SYN in this case. 1334 * If we decipher RFC2462 5.5.4, it says like 1335 * this: 1336 * 1. use of deprecated addr with existing 1337 * communication is okay - "SHOULD continue 1338 * to be used" 1339 * 2. use of it with new communication: 1340 * (2a) "SHOULD NOT be used if alternate 1341 * address with sufficient scope is 1342 * available" 1343 * (2b) nothing mentioned otherwise. 1344 * Here we fall into (2b) case as we have no 1345 * choice in our source address selection - we 1346 * must obey the peer. 1347 * 1348 * The wording in RFC2462 is confusing, and 1349 * there are multiple description text for 1350 * deprecated address handling - worse, they 1351 * are not exactly the same. I believe 5.5.4 1352 * is the best one, so we follow 5.5.4. 1353 */ 1354 if (af == AF_INET6 && !ip6_use_deprecated) { 1355 struct in6_ifaddr *ia6; 1356 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1357 &ip6->ip6_dst)) && 1358 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1359 tp = NULL; 1360 goto dropwithreset; 1361 } 1362 } 1363 #endif 1364 1365 #ifdef IPSEC 1366 switch (af) { 1367 #ifdef INET 1368 case AF_INET: 1369 if (ipsec4_in_reject_so(m, so)) { 1370 ipsecstat.in_polvio++; 1371 tp = NULL; 1372 goto dropwithreset; 1373 } 1374 break; 1375 #endif 1376 #ifdef INET6 1377 case AF_INET6: 1378 if (ipsec6_in_reject_so(m, so)) { 1379 ipsec6stat.in_polvio++; 1380 tp = NULL; 1381 goto dropwithreset; 1382 } 1383 break; 1384 #endif 1385 } 1386 #endif 1387 1388 /* 1389 * LISTEN socket received a SYN 1390 * from itself? This can't possibly 1391 * be valid; drop the packet. 1392 */ 1393 if (th->th_sport == th->th_dport) { 1394 int i; 1395 1396 switch (af) { 1397 #ifdef INET 1398 case AF_INET: 1399 i = in_hosteq(ip->ip_src, ip->ip_dst); 1400 break; 1401 #endif 1402 #ifdef INET6 1403 case AF_INET6: 1404 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1405 break; 1406 #endif 1407 default: 1408 i = 1; 1409 } 1410 if (i) { 1411 tcpstat.tcps_badsyn++; 1412 goto drop; 1413 } 1414 } 1415 1416 /* 1417 * SYN looks ok; create compressed TCP 1418 * state for it. 1419 */ 1420 if (so->so_qlen <= so->so_qlimit && 1421 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1422 so, m, optp, optlen, &opti)) 1423 m = NULL; 1424 } 1425 goto drop; 1426 } 1427 } 1428 1429 after_listen: 1430 #ifdef DIAGNOSTIC 1431 /* 1432 * Should not happen now that all embryonic connections 1433 * are handled with compressed state. 1434 */ 1435 if (tp->t_state == TCPS_LISTEN) 1436 panic("tcp_input: TCPS_LISTEN"); 1437 #endif 1438 1439 /* 1440 * Segment received on connection. 1441 * Reset idle time and keep-alive timer. 1442 */ 1443 tp->t_rcvtime = tcp_now; 1444 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1445 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1446 1447 /* 1448 * Process options. 1449 */ 1450 #ifdef TCP_SIGNATURE 1451 if (optp || (tp->t_flags & TF_SIGNATURE)) 1452 #else 1453 if (optp) 1454 #endif 1455 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1456 goto drop; 1457 1458 /* 1459 * Header prediction: check for the two common cases 1460 * of a uni-directional data xfer. If the packet has 1461 * no control flags, is in-sequence, the window didn't 1462 * change and we're not retransmitting, it's a 1463 * candidate. If the length is zero and the ack moved 1464 * forward, we're the sender side of the xfer. Just 1465 * free the data acked & wake any higher level process 1466 * that was blocked waiting for space. If the length 1467 * is non-zero and the ack didn't move, we're the 1468 * receiver side. If we're getting packets in-order 1469 * (the reassembly queue is empty), add the data to 1470 * the socket buffer and note that we need a delayed ack. 1471 */ 1472 if (tp->t_state == TCPS_ESTABLISHED && 1473 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1474 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1475 th->th_seq == tp->rcv_nxt && 1476 tiwin && tiwin == tp->snd_wnd && 1477 tp->snd_nxt == tp->snd_max) { 1478 1479 /* 1480 * If last ACK falls within this segment's sequence numbers, 1481 * record the timestamp. 1482 */ 1483 if (opti.ts_present && 1484 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1485 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1486 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1487 tp->ts_recent = opti.ts_val; 1488 } 1489 1490 if (tlen == 0) { 1491 if (SEQ_GT(th->th_ack, tp->snd_una) && 1492 SEQ_LEQ(th->th_ack, tp->snd_max) && 1493 tp->snd_cwnd >= tp->snd_wnd && 1494 tp->t_dupacks < tcprexmtthresh) { 1495 /* 1496 * this is a pure ack for outstanding data. 1497 */ 1498 ++tcpstat.tcps_predack; 1499 if (opti.ts_present && opti.ts_ecr) 1500 tcp_xmit_timer(tp, 1501 TCP_TIMESTAMP(tp) - opti.ts_ecr + 1); 1502 else if (tp->t_rtttime && 1503 SEQ_GT(th->th_ack, tp->t_rtseq)) 1504 tcp_xmit_timer(tp, 1505 tcp_now - tp->t_rtttime); 1506 acked = th->th_ack - tp->snd_una; 1507 tcpstat.tcps_rcvackpack++; 1508 tcpstat.tcps_rcvackbyte += acked; 1509 ND6_HINT(tp); 1510 1511 if (acked > (tp->t_lastoff - tp->t_inoff)) 1512 tp->t_lastm = NULL; 1513 sbdrop(&so->so_snd, acked); 1514 tp->t_lastoff -= acked; 1515 1516 /* 1517 * We want snd_recover to track snd_una to 1518 * avoid sequence wraparound problems for 1519 * very large transfers. 1520 */ 1521 tp->snd_una = tp->snd_recover = th->th_ack; 1522 m_freem(m); 1523 1524 /* 1525 * If all outstanding data are acked, stop 1526 * retransmit timer, otherwise restart timer 1527 * using current (possibly backed-off) value. 1528 * If process is waiting for space, 1529 * wakeup/selwakeup/signal. If data 1530 * are ready to send, let tcp_output 1531 * decide between more output or persist. 1532 */ 1533 if (tp->snd_una == tp->snd_max) 1534 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1535 else if (TCP_TIMER_ISARMED(tp, 1536 TCPT_PERSIST) == 0) 1537 TCP_TIMER_ARM(tp, TCPT_REXMT, 1538 tp->t_rxtcur); 1539 1540 sowwakeup(so); 1541 if (so->so_snd.sb_cc) 1542 (void) tcp_output(tp); 1543 if (tcp_saveti) 1544 m_freem(tcp_saveti); 1545 return; 1546 } 1547 } else if (th->th_ack == tp->snd_una && 1548 TAILQ_FIRST(&tp->segq) == NULL && 1549 tlen <= sbspace(&so->so_rcv)) { 1550 /* 1551 * this is a pure, in-sequence data packet 1552 * with nothing on the reassembly queue and 1553 * we have enough buffer space to take it. 1554 */ 1555 ++tcpstat.tcps_preddat; 1556 tp->rcv_nxt += tlen; 1557 tcpstat.tcps_rcvpack++; 1558 tcpstat.tcps_rcvbyte += tlen; 1559 ND6_HINT(tp); 1560 /* 1561 * Drop TCP, IP headers and TCP options then add data 1562 * to socket buffer. 1563 */ 1564 if (so->so_state & SS_CANTRCVMORE) 1565 m_freem(m); 1566 else { 1567 m_adj(m, toff + off); 1568 sbappendstream(&so->so_rcv, m); 1569 } 1570 sorwakeup(so); 1571 TCP_SETUP_ACK(tp, th); 1572 if (tp->t_flags & TF_ACKNOW) 1573 (void) tcp_output(tp); 1574 if (tcp_saveti) 1575 m_freem(tcp_saveti); 1576 return; 1577 } 1578 } 1579 1580 /* 1581 * Compute mbuf offset to TCP data segment. 1582 */ 1583 hdroptlen = toff + off; 1584 1585 /* 1586 * Calculate amount of space in receive window, 1587 * and then do TCP input processing. 1588 * Receive window is amount of space in rcv queue, 1589 * but not less than advertised window. 1590 */ 1591 { int win; 1592 1593 win = sbspace(&so->so_rcv); 1594 if (win < 0) 1595 win = 0; 1596 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1597 } 1598 1599 switch (tp->t_state) { 1600 case TCPS_LISTEN: 1601 /* 1602 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1603 */ 1604 if (m->m_flags & (M_BCAST|M_MCAST)) 1605 goto drop; 1606 switch (af) { 1607 #ifdef INET6 1608 case AF_INET6: 1609 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1610 goto drop; 1611 break; 1612 #endif /* INET6 */ 1613 case AF_INET: 1614 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1615 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1616 goto drop; 1617 break; 1618 } 1619 break; 1620 1621 /* 1622 * If the state is SYN_SENT: 1623 * if seg contains an ACK, but not for our SYN, drop the input. 1624 * if seg contains a RST, then drop the connection. 1625 * if seg does not contain SYN, then drop it. 1626 * Otherwise this is an acceptable SYN segment 1627 * initialize tp->rcv_nxt and tp->irs 1628 * if seg contains ack then advance tp->snd_una 1629 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1630 * arrange for segment to be acked (eventually) 1631 * continue processing rest of data/controls, beginning with URG 1632 */ 1633 case TCPS_SYN_SENT: 1634 if ((tiflags & TH_ACK) && 1635 (SEQ_LEQ(th->th_ack, tp->iss) || 1636 SEQ_GT(th->th_ack, tp->snd_max))) 1637 goto dropwithreset; 1638 if (tiflags & TH_RST) { 1639 if (tiflags & TH_ACK) 1640 tp = tcp_drop(tp, ECONNREFUSED); 1641 goto drop; 1642 } 1643 if ((tiflags & TH_SYN) == 0) 1644 goto drop; 1645 if (tiflags & TH_ACK) { 1646 tp->snd_una = tp->snd_recover = th->th_ack; 1647 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1648 tp->snd_nxt = tp->snd_una; 1649 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1650 } 1651 tp->irs = th->th_seq; 1652 tcp_rcvseqinit(tp); 1653 tp->t_flags |= TF_ACKNOW; 1654 tcp_mss_from_peer(tp, opti.maxseg); 1655 1656 /* 1657 * Initialize the initial congestion window. If we 1658 * had to retransmit the SYN, we must initialize cwnd 1659 * to 1 segment (i.e. the Loss Window). 1660 */ 1661 if (tp->t_flags & TF_SYN_REXMT) 1662 tp->snd_cwnd = tp->t_peermss; 1663 else { 1664 int ss = tcp_init_win; 1665 #ifdef INET 1666 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1667 ss = tcp_init_win_local; 1668 #endif 1669 #ifdef INET6 1670 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1671 ss = tcp_init_win_local; 1672 #endif 1673 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1674 } 1675 1676 tcp_rmx_rtt(tp); 1677 if (tiflags & TH_ACK) { 1678 tcpstat.tcps_connects++; 1679 soisconnected(so); 1680 tcp_established(tp); 1681 /* Do window scaling on this connection? */ 1682 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1683 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1684 tp->snd_scale = tp->requested_s_scale; 1685 tp->rcv_scale = tp->request_r_scale; 1686 } 1687 TCP_REASS_LOCK(tp); 1688 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1689 TCP_REASS_UNLOCK(tp); 1690 /* 1691 * if we didn't have to retransmit the SYN, 1692 * use its rtt as our initial srtt & rtt var. 1693 */ 1694 if (tp->t_rtttime) 1695 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1696 } else 1697 tp->t_state = TCPS_SYN_RECEIVED; 1698 1699 /* 1700 * Advance th->th_seq to correspond to first data byte. 1701 * If data, trim to stay within window, 1702 * dropping FIN if necessary. 1703 */ 1704 th->th_seq++; 1705 if (tlen > tp->rcv_wnd) { 1706 todrop = tlen - tp->rcv_wnd; 1707 m_adj(m, -todrop); 1708 tlen = tp->rcv_wnd; 1709 tiflags &= ~TH_FIN; 1710 tcpstat.tcps_rcvpackafterwin++; 1711 tcpstat.tcps_rcvbyteafterwin += todrop; 1712 } 1713 tp->snd_wl1 = th->th_seq - 1; 1714 tp->rcv_up = th->th_seq; 1715 goto step6; 1716 1717 /* 1718 * If the state is SYN_RECEIVED: 1719 * If seg contains an ACK, but not for our SYN, drop the input 1720 * and generate an RST. See page 36, rfc793 1721 */ 1722 case TCPS_SYN_RECEIVED: 1723 if ((tiflags & TH_ACK) && 1724 (SEQ_LEQ(th->th_ack, tp->iss) || 1725 SEQ_GT(th->th_ack, tp->snd_max))) 1726 goto dropwithreset; 1727 break; 1728 } 1729 1730 /* 1731 * States other than LISTEN or SYN_SENT. 1732 * First check timestamp, if present. 1733 * Then check that at least some bytes of segment are within 1734 * receive window. If segment begins before rcv_nxt, 1735 * drop leading data (and SYN); if nothing left, just ack. 1736 * 1737 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1738 * and it's less than ts_recent, drop it. 1739 */ 1740 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1741 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1742 1743 /* Check to see if ts_recent is over 24 days old. */ 1744 if ((int)(TCP_TIMESTAMP(tp) - tp->ts_recent_age) > 1745 TCP_PAWS_IDLE) { 1746 /* 1747 * Invalidate ts_recent. If this segment updates 1748 * ts_recent, the age will be reset later and ts_recent 1749 * will get a valid value. If it does not, setting 1750 * ts_recent to zero will at least satisfy the 1751 * requirement that zero be placed in the timestamp 1752 * echo reply when ts_recent isn't valid. The 1753 * age isn't reset until we get a valid ts_recent 1754 * because we don't want out-of-order segments to be 1755 * dropped when ts_recent is old. 1756 */ 1757 tp->ts_recent = 0; 1758 } else { 1759 tcpstat.tcps_rcvduppack++; 1760 tcpstat.tcps_rcvdupbyte += tlen; 1761 tcpstat.tcps_pawsdrop++; 1762 goto dropafterack; 1763 } 1764 } 1765 1766 todrop = tp->rcv_nxt - th->th_seq; 1767 if (todrop > 0) { 1768 if (tiflags & TH_SYN) { 1769 tiflags &= ~TH_SYN; 1770 th->th_seq++; 1771 if (th->th_urp > 1) 1772 th->th_urp--; 1773 else { 1774 tiflags &= ~TH_URG; 1775 th->th_urp = 0; 1776 } 1777 todrop--; 1778 } 1779 if (todrop > tlen || 1780 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1781 /* 1782 * Any valid FIN or RST must be to the left of the 1783 * window. At this point the FIN or RST must be a 1784 * duplicate or out of sequence; drop it. 1785 */ 1786 if (tiflags & TH_RST) 1787 goto drop; 1788 tiflags &= ~(TH_FIN|TH_RST); 1789 /* 1790 * Send an ACK to resynchronize and drop any data. 1791 * But keep on processing for RST or ACK. 1792 */ 1793 tp->t_flags |= TF_ACKNOW; 1794 todrop = tlen; 1795 tcpstat.tcps_rcvdupbyte += todrop; 1796 tcpstat.tcps_rcvduppack++; 1797 } else if ((tiflags & TH_RST) && 1798 th->th_seq != tp->last_ack_sent) { 1799 /* 1800 * Test for reset before adjusting the sequence 1801 * number for overlapping data. 1802 */ 1803 goto dropafterack_ratelim; 1804 } else { 1805 tcpstat.tcps_rcvpartduppack++; 1806 tcpstat.tcps_rcvpartdupbyte += todrop; 1807 } 1808 hdroptlen += todrop; /*drop from head afterwards*/ 1809 th->th_seq += todrop; 1810 tlen -= todrop; 1811 if (th->th_urp > todrop) 1812 th->th_urp -= todrop; 1813 else { 1814 tiflags &= ~TH_URG; 1815 th->th_urp = 0; 1816 } 1817 } 1818 1819 /* 1820 * If new data are received on a connection after the 1821 * user processes are gone, then RST the other end. 1822 */ 1823 if ((so->so_state & SS_NOFDREF) && 1824 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1825 tp = tcp_close(tp); 1826 tcpstat.tcps_rcvafterclose++; 1827 goto dropwithreset; 1828 } 1829 1830 /* 1831 * If segment ends after window, drop trailing data 1832 * (and PUSH and FIN); if nothing left, just ACK. 1833 */ 1834 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1835 if (todrop > 0) { 1836 tcpstat.tcps_rcvpackafterwin++; 1837 if (todrop >= tlen) { 1838 /* 1839 * The segment actually starts after the window. 1840 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 1841 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 1842 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 1843 */ 1844 tcpstat.tcps_rcvbyteafterwin += tlen; 1845 /* 1846 * If a new connection request is received 1847 * while in TIME_WAIT, drop the old connection 1848 * and start over if the sequence numbers 1849 * are above the previous ones. 1850 * 1851 * NOTE: We will checksum the packet again, and 1852 * so we need to put the header fields back into 1853 * network order! 1854 * XXX This kind of sucks, but we don't expect 1855 * XXX this to happen very often, so maybe it 1856 * XXX doesn't matter so much. 1857 */ 1858 if (tiflags & TH_SYN && 1859 tp->t_state == TCPS_TIME_WAIT && 1860 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1861 iss = tcp_new_iss(tp, tp->snd_nxt); 1862 tp = tcp_close(tp); 1863 TCP_FIELDS_TO_NET(th); 1864 goto findpcb; 1865 } 1866 /* 1867 * If window is closed can only take segments at 1868 * window edge, and have to drop data and PUSH from 1869 * incoming segments. Continue processing, but 1870 * remember to ack. Otherwise, drop segment 1871 * and (if not RST) ack. 1872 */ 1873 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1874 tp->t_flags |= TF_ACKNOW; 1875 tcpstat.tcps_rcvwinprobe++; 1876 } else 1877 goto dropafterack; 1878 } else 1879 tcpstat.tcps_rcvbyteafterwin += todrop; 1880 m_adj(m, -todrop); 1881 tlen -= todrop; 1882 tiflags &= ~(TH_PUSH|TH_FIN); 1883 } 1884 1885 /* 1886 * If last ACK falls within this segment's sequence numbers, 1887 * and the timestamp is newer, record it. 1888 */ 1889 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1890 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1891 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1892 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1893 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1894 tp->ts_recent = opti.ts_val; 1895 } 1896 1897 /* 1898 * If the RST bit is set examine the state: 1899 * SYN_RECEIVED STATE: 1900 * If passive open, return to LISTEN state. 1901 * If active open, inform user that connection was refused. 1902 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1903 * Inform user that connection was reset, and close tcb. 1904 * CLOSING, LAST_ACK, TIME_WAIT STATES 1905 * Close the tcb. 1906 */ 1907 if (tiflags & TH_RST) { 1908 if (th->th_seq != tp->last_ack_sent) 1909 goto dropafterack_ratelim; 1910 1911 switch (tp->t_state) { 1912 case TCPS_SYN_RECEIVED: 1913 so->so_error = ECONNREFUSED; 1914 goto close; 1915 1916 case TCPS_ESTABLISHED: 1917 case TCPS_FIN_WAIT_1: 1918 case TCPS_FIN_WAIT_2: 1919 case TCPS_CLOSE_WAIT: 1920 so->so_error = ECONNRESET; 1921 close: 1922 tp->t_state = TCPS_CLOSED; 1923 tcpstat.tcps_drops++; 1924 tp = tcp_close(tp); 1925 goto drop; 1926 1927 case TCPS_CLOSING: 1928 case TCPS_LAST_ACK: 1929 case TCPS_TIME_WAIT: 1930 tp = tcp_close(tp); 1931 goto drop; 1932 } 1933 } 1934 1935 /* 1936 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 1937 * we must be in a synchronized state. RFC791 states (under RST 1938 * generation) that any unacceptable segment (an out-of-order SYN 1939 * qualifies) received in a synchronized state must elicit only an 1940 * empty acknowledgment segment ... and the connection remains in 1941 * the same state. 1942 */ 1943 if (tiflags & TH_SYN) { 1944 if (tp->rcv_nxt == th->th_seq) { 1945 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 1946 TH_ACK); 1947 if (tcp_saveti) 1948 m_freem(tcp_saveti); 1949 return; 1950 } 1951 1952 goto dropafterack_ratelim; 1953 } 1954 1955 /* 1956 * If the ACK bit is off we drop the segment and return. 1957 */ 1958 if ((tiflags & TH_ACK) == 0) { 1959 if (tp->t_flags & TF_ACKNOW) 1960 goto dropafterack; 1961 else 1962 goto drop; 1963 } 1964 1965 /* 1966 * Ack processing. 1967 */ 1968 switch (tp->t_state) { 1969 1970 /* 1971 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1972 * ESTABLISHED state and continue processing, otherwise 1973 * send an RST. 1974 */ 1975 case TCPS_SYN_RECEIVED: 1976 if (SEQ_GT(tp->snd_una, th->th_ack) || 1977 SEQ_GT(th->th_ack, tp->snd_max)) 1978 goto dropwithreset; 1979 tcpstat.tcps_connects++; 1980 soisconnected(so); 1981 tcp_established(tp); 1982 /* Do window scaling? */ 1983 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1984 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1985 tp->snd_scale = tp->requested_s_scale; 1986 tp->rcv_scale = tp->request_r_scale; 1987 } 1988 TCP_REASS_LOCK(tp); 1989 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1990 TCP_REASS_UNLOCK(tp); 1991 tp->snd_wl1 = th->th_seq - 1; 1992 /* fall into ... */ 1993 1994 /* 1995 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1996 * ACKs. If the ack is in the range 1997 * tp->snd_una < th->th_ack <= tp->snd_max 1998 * then advance tp->snd_una to th->th_ack and drop 1999 * data from the retransmission queue. If this ACK reflects 2000 * more up to date window information we update our window information. 2001 */ 2002 case TCPS_ESTABLISHED: 2003 case TCPS_FIN_WAIT_1: 2004 case TCPS_FIN_WAIT_2: 2005 case TCPS_CLOSE_WAIT: 2006 case TCPS_CLOSING: 2007 case TCPS_LAST_ACK: 2008 case TCPS_TIME_WAIT: 2009 2010 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2011 if (tlen == 0 && tiwin == tp->snd_wnd) { 2012 tcpstat.tcps_rcvdupack++; 2013 /* 2014 * If we have outstanding data (other than 2015 * a window probe), this is a completely 2016 * duplicate ack (ie, window info didn't 2017 * change), the ack is the biggest we've 2018 * seen and we've seen exactly our rexmt 2019 * threshhold of them, assume a packet 2020 * has been dropped and retransmit it. 2021 * Kludge snd_nxt & the congestion 2022 * window so we send only this one 2023 * packet. 2024 * 2025 * We know we're losing at the current 2026 * window size so do congestion avoidance 2027 * (set ssthresh to half the current window 2028 * and pull our congestion window back to 2029 * the new ssthresh). 2030 * 2031 * Dup acks mean that packets have left the 2032 * network (they're now cached at the receiver) 2033 * so bump cwnd by the amount in the receiver 2034 * to keep a constant cwnd packets in the 2035 * network. 2036 */ 2037 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2038 th->th_ack != tp->snd_una) 2039 tp->t_dupacks = 0; 2040 else if (++tp->t_dupacks == tcprexmtthresh) { 2041 tcp_seq onxt = tp->snd_nxt; 2042 u_int win = 2043 min(tp->snd_wnd, tp->snd_cwnd) / 2044 2 / tp->t_segsz; 2045 if (tcp_do_newreno && SEQ_LT(th->th_ack, 2046 tp->snd_recover)) { 2047 /* 2048 * False fast retransmit after 2049 * timeout. Do not cut window. 2050 */ 2051 tp->snd_cwnd += tp->t_segsz; 2052 tp->t_dupacks = 0; 2053 (void) tcp_output(tp); 2054 goto drop; 2055 } 2056 2057 if (win < 2) 2058 win = 2; 2059 tp->snd_ssthresh = win * tp->t_segsz; 2060 tp->snd_recover = tp->snd_max; 2061 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2062 tp->t_rtttime = 0; 2063 tp->snd_nxt = th->th_ack; 2064 tp->snd_cwnd = tp->t_segsz; 2065 (void) tcp_output(tp); 2066 tp->snd_cwnd = tp->snd_ssthresh + 2067 tp->t_segsz * tp->t_dupacks; 2068 if (SEQ_GT(onxt, tp->snd_nxt)) 2069 tp->snd_nxt = onxt; 2070 goto drop; 2071 } else if (tp->t_dupacks > tcprexmtthresh) { 2072 tp->snd_cwnd += tp->t_segsz; 2073 (void) tcp_output(tp); 2074 goto drop; 2075 } 2076 } else if (tlen) { 2077 tp->t_dupacks = 0; /*XXX*/ 2078 /* drop very old ACKs unless th_seq matches */ 2079 if (th->th_seq != tp->rcv_nxt && 2080 SEQ_LT(th->th_ack, 2081 tp->snd_una - tp->max_sndwnd)) { 2082 goto drop; 2083 } 2084 break; 2085 } else 2086 tp->t_dupacks = 0; 2087 break; 2088 } 2089 /* 2090 * If the congestion window was inflated to account 2091 * for the other side's cached packets, retract it. 2092 */ 2093 if (tcp_do_newreno == 0) { 2094 if (tp->t_dupacks >= tcprexmtthresh && 2095 tp->snd_cwnd > tp->snd_ssthresh) 2096 tp->snd_cwnd = tp->snd_ssthresh; 2097 tp->t_dupacks = 0; 2098 } else if (tp->t_dupacks >= tcprexmtthresh && 2099 tcp_newreno(tp, th) == 0) { 2100 tp->snd_cwnd = tp->snd_ssthresh; 2101 /* 2102 * Window inflation should have left us with approx. 2103 * snd_ssthresh outstanding data. But in case we 2104 * would be inclined to send a burst, better to do 2105 * it via the slow start mechanism. 2106 */ 2107 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 2108 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 2109 + tp->t_segsz; 2110 tp->t_dupacks = 0; 2111 } 2112 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2113 tcpstat.tcps_rcvacktoomuch++; 2114 goto dropafterack; 2115 } 2116 acked = th->th_ack - tp->snd_una; 2117 tcpstat.tcps_rcvackpack++; 2118 tcpstat.tcps_rcvackbyte += acked; 2119 2120 /* 2121 * If we have a timestamp reply, update smoothed 2122 * round trip time. If no timestamp is present but 2123 * transmit timer is running and timed sequence 2124 * number was acked, update smoothed round trip time. 2125 * Since we now have an rtt measurement, cancel the 2126 * timer backoff (cf., Phil Karn's retransmit alg.). 2127 * Recompute the initial retransmit timer. 2128 */ 2129 if (opti.ts_present && opti.ts_ecr) 2130 tcp_xmit_timer(tp, TCP_TIMESTAMP(tp) - opti.ts_ecr + 1); 2131 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2132 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2133 2134 /* 2135 * If all outstanding data is acked, stop retransmit 2136 * timer and remember to restart (more output or persist). 2137 * If there is more data to be acked, restart retransmit 2138 * timer, using current (possibly backed-off) value. 2139 */ 2140 if (th->th_ack == tp->snd_max) { 2141 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2142 needoutput = 1; 2143 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2144 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2145 /* 2146 * When new data is acked, open the congestion window. 2147 * If the window gives us less than ssthresh packets 2148 * in flight, open exponentially (segsz per packet). 2149 * Otherwise open linearly: segsz per window 2150 * (segsz^2 / cwnd per packet), plus a constant 2151 * fraction of a packet (segsz/8) to help larger windows 2152 * open quickly enough. 2153 */ 2154 { 2155 u_int cw = tp->snd_cwnd; 2156 u_int incr = tp->t_segsz; 2157 2158 if (cw > tp->snd_ssthresh) 2159 incr = incr * incr / cw; 2160 if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover)) 2161 tp->snd_cwnd = min(cw + incr, 2162 TCP_MAXWIN << tp->snd_scale); 2163 } 2164 ND6_HINT(tp); 2165 if (acked > so->so_snd.sb_cc) { 2166 tp->snd_wnd -= so->so_snd.sb_cc; 2167 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2168 ourfinisacked = 1; 2169 } else { 2170 if (acked > (tp->t_lastoff - tp->t_inoff)) 2171 tp->t_lastm = NULL; 2172 sbdrop(&so->so_snd, acked); 2173 tp->t_lastoff -= acked; 2174 tp->snd_wnd -= acked; 2175 ourfinisacked = 0; 2176 } 2177 sowwakeup(so); 2178 /* 2179 * We want snd_recover to track snd_una to 2180 * avoid sequence wraparound problems for 2181 * very large transfers. 2182 */ 2183 tp->snd_una = tp->snd_recover = th->th_ack; 2184 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2185 tp->snd_nxt = tp->snd_una; 2186 2187 switch (tp->t_state) { 2188 2189 /* 2190 * In FIN_WAIT_1 STATE in addition to the processing 2191 * for the ESTABLISHED state if our FIN is now acknowledged 2192 * then enter FIN_WAIT_2. 2193 */ 2194 case TCPS_FIN_WAIT_1: 2195 if (ourfinisacked) { 2196 /* 2197 * If we can't receive any more 2198 * data, then closing user can proceed. 2199 * Starting the timer is contrary to the 2200 * specification, but if we don't get a FIN 2201 * we'll hang forever. 2202 */ 2203 if (so->so_state & SS_CANTRCVMORE) { 2204 soisdisconnected(so); 2205 if (tcp_maxidle > 0) 2206 TCP_TIMER_ARM(tp, TCPT_2MSL, 2207 tcp_maxidle); 2208 } 2209 tp->t_state = TCPS_FIN_WAIT_2; 2210 } 2211 break; 2212 2213 /* 2214 * In CLOSING STATE in addition to the processing for 2215 * the ESTABLISHED state if the ACK acknowledges our FIN 2216 * then enter the TIME-WAIT state, otherwise ignore 2217 * the segment. 2218 */ 2219 case TCPS_CLOSING: 2220 if (ourfinisacked) { 2221 tp->t_state = TCPS_TIME_WAIT; 2222 tcp_canceltimers(tp); 2223 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2224 soisdisconnected(so); 2225 } 2226 break; 2227 2228 /* 2229 * In LAST_ACK, we may still be waiting for data to drain 2230 * and/or to be acked, as well as for the ack of our FIN. 2231 * If our FIN is now acknowledged, delete the TCB, 2232 * enter the closed state and return. 2233 */ 2234 case TCPS_LAST_ACK: 2235 if (ourfinisacked) { 2236 tp = tcp_close(tp); 2237 goto drop; 2238 } 2239 break; 2240 2241 /* 2242 * In TIME_WAIT state the only thing that should arrive 2243 * is a retransmission of the remote FIN. Acknowledge 2244 * it and restart the finack timer. 2245 */ 2246 case TCPS_TIME_WAIT: 2247 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2248 goto dropafterack; 2249 } 2250 } 2251 2252 step6: 2253 /* 2254 * Update window information. 2255 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2256 */ 2257 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2258 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 2259 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 2260 /* keep track of pure window updates */ 2261 if (tlen == 0 && 2262 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2263 tcpstat.tcps_rcvwinupd++; 2264 tp->snd_wnd = tiwin; 2265 tp->snd_wl1 = th->th_seq; 2266 tp->snd_wl2 = th->th_ack; 2267 if (tp->snd_wnd > tp->max_sndwnd) 2268 tp->max_sndwnd = tp->snd_wnd; 2269 needoutput = 1; 2270 } 2271 2272 /* 2273 * Process segments with URG. 2274 */ 2275 if ((tiflags & TH_URG) && th->th_urp && 2276 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2277 /* 2278 * This is a kludge, but if we receive and accept 2279 * random urgent pointers, we'll crash in 2280 * soreceive. It's hard to imagine someone 2281 * actually wanting to send this much urgent data. 2282 */ 2283 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2284 th->th_urp = 0; /* XXX */ 2285 tiflags &= ~TH_URG; /* XXX */ 2286 goto dodata; /* XXX */ 2287 } 2288 /* 2289 * If this segment advances the known urgent pointer, 2290 * then mark the data stream. This should not happen 2291 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2292 * a FIN has been received from the remote side. 2293 * In these states we ignore the URG. 2294 * 2295 * According to RFC961 (Assigned Protocols), 2296 * the urgent pointer points to the last octet 2297 * of urgent data. We continue, however, 2298 * to consider it to indicate the first octet 2299 * of data past the urgent section as the original 2300 * spec states (in one of two places). 2301 */ 2302 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2303 tp->rcv_up = th->th_seq + th->th_urp; 2304 so->so_oobmark = so->so_rcv.sb_cc + 2305 (tp->rcv_up - tp->rcv_nxt) - 1; 2306 if (so->so_oobmark == 0) 2307 so->so_state |= SS_RCVATMARK; 2308 sohasoutofband(so); 2309 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2310 } 2311 /* 2312 * Remove out of band data so doesn't get presented to user. 2313 * This can happen independent of advancing the URG pointer, 2314 * but if two URG's are pending at once, some out-of-band 2315 * data may creep in... ick. 2316 */ 2317 if (th->th_urp <= (u_int16_t) tlen 2318 #ifdef SO_OOBINLINE 2319 && (so->so_options & SO_OOBINLINE) == 0 2320 #endif 2321 ) 2322 tcp_pulloutofband(so, th, m, hdroptlen); 2323 } else 2324 /* 2325 * If no out of band data is expected, 2326 * pull receive urgent pointer along 2327 * with the receive window. 2328 */ 2329 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2330 tp->rcv_up = tp->rcv_nxt; 2331 dodata: /* XXX */ 2332 2333 /* 2334 * Process the segment text, merging it into the TCP sequencing queue, 2335 * and arranging for acknowledgement of receipt if necessary. 2336 * This process logically involves adjusting tp->rcv_wnd as data 2337 * is presented to the user (this happens in tcp_usrreq.c, 2338 * case PRU_RCVD). If a FIN has already been received on this 2339 * connection then we just ignore the text. 2340 */ 2341 if ((tlen || (tiflags & TH_FIN)) && 2342 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2343 /* 2344 * Insert segment ti into reassembly queue of tcp with 2345 * control block tp. Return TH_FIN if reassembly now includes 2346 * a segment with FIN. The macro form does the common case 2347 * inline (segment is the next to be received on an 2348 * established connection, and the queue is empty), 2349 * avoiding linkage into and removal from the queue and 2350 * repetition of various conversions. 2351 * Set DELACK for segments received in order, but ack 2352 * immediately when segments are out of order 2353 * (so fast retransmit can work). 2354 */ 2355 /* NOTE: this was TCP_REASS() macro, but used only once */ 2356 TCP_REASS_LOCK(tp); 2357 if (th->th_seq == tp->rcv_nxt && 2358 TAILQ_FIRST(&tp->segq) == NULL && 2359 tp->t_state == TCPS_ESTABLISHED) { 2360 TCP_SETUP_ACK(tp, th); 2361 tp->rcv_nxt += tlen; 2362 tiflags = th->th_flags & TH_FIN; 2363 tcpstat.tcps_rcvpack++; 2364 tcpstat.tcps_rcvbyte += tlen; 2365 ND6_HINT(tp); 2366 if (so->so_state & SS_CANTRCVMORE) 2367 m_freem(m); 2368 else { 2369 m_adj(m, hdroptlen); 2370 sbappendstream(&(so)->so_rcv, m); 2371 } 2372 sorwakeup(so); 2373 } else { 2374 m_adj(m, hdroptlen); 2375 tiflags = tcp_reass(tp, th, m, &tlen); 2376 tp->t_flags |= TF_ACKNOW; 2377 } 2378 TCP_REASS_UNLOCK(tp); 2379 2380 /* 2381 * Note the amount of data that peer has sent into 2382 * our window, in order to estimate the sender's 2383 * buffer size. 2384 */ 2385 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2386 } else { 2387 m_freem(m); 2388 m = NULL; 2389 tiflags &= ~TH_FIN; 2390 } 2391 2392 /* 2393 * If FIN is received ACK the FIN and let the user know 2394 * that the connection is closing. Ignore a FIN received before 2395 * the connection is fully established. 2396 */ 2397 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2398 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2399 socantrcvmore(so); 2400 tp->t_flags |= TF_ACKNOW; 2401 tp->rcv_nxt++; 2402 } 2403 switch (tp->t_state) { 2404 2405 /* 2406 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2407 */ 2408 case TCPS_ESTABLISHED: 2409 tp->t_state = TCPS_CLOSE_WAIT; 2410 break; 2411 2412 /* 2413 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2414 * enter the CLOSING state. 2415 */ 2416 case TCPS_FIN_WAIT_1: 2417 tp->t_state = TCPS_CLOSING; 2418 break; 2419 2420 /* 2421 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2422 * starting the time-wait timer, turning off the other 2423 * standard timers. 2424 */ 2425 case TCPS_FIN_WAIT_2: 2426 tp->t_state = TCPS_TIME_WAIT; 2427 tcp_canceltimers(tp); 2428 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2429 soisdisconnected(so); 2430 break; 2431 2432 /* 2433 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2434 */ 2435 case TCPS_TIME_WAIT: 2436 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2437 break; 2438 } 2439 } 2440 #ifdef TCP_DEBUG 2441 if (so->so_options & SO_DEBUG) 2442 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2443 #endif 2444 2445 /* 2446 * Return any desired output. 2447 */ 2448 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2449 (void) tcp_output(tp); 2450 if (tcp_saveti) 2451 m_freem(tcp_saveti); 2452 return; 2453 2454 badsyn: 2455 /* 2456 * Received a bad SYN. Increment counters and dropwithreset. 2457 */ 2458 tcpstat.tcps_badsyn++; 2459 tp = NULL; 2460 goto dropwithreset; 2461 2462 dropafterack: 2463 /* 2464 * Generate an ACK dropping incoming segment if it occupies 2465 * sequence space, where the ACK reflects our state. 2466 */ 2467 if (tiflags & TH_RST) 2468 goto drop; 2469 goto dropafterack2; 2470 2471 dropafterack_ratelim: 2472 /* 2473 * We may want to rate-limit ACKs against SYN/RST attack. 2474 */ 2475 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2476 tcp_ackdrop_ppslim) == 0) { 2477 /* XXX stat */ 2478 goto drop; 2479 } 2480 /* ...fall into dropafterack2... */ 2481 2482 dropafterack2: 2483 m_freem(m); 2484 tp->t_flags |= TF_ACKNOW; 2485 (void) tcp_output(tp); 2486 if (tcp_saveti) 2487 m_freem(tcp_saveti); 2488 return; 2489 2490 dropwithreset_ratelim: 2491 /* 2492 * We may want to rate-limit RSTs in certain situations, 2493 * particularly if we are sending an RST in response to 2494 * an attempt to connect to or otherwise communicate with 2495 * a port for which we have no socket. 2496 */ 2497 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2498 tcp_rst_ppslim) == 0) { 2499 /* XXX stat */ 2500 goto drop; 2501 } 2502 /* ...fall into dropwithreset... */ 2503 2504 dropwithreset: 2505 /* 2506 * Generate a RST, dropping incoming segment. 2507 * Make ACK acceptable to originator of segment. 2508 */ 2509 if (tiflags & TH_RST) 2510 goto drop; 2511 2512 switch (af) { 2513 #ifdef INET6 2514 case AF_INET6: 2515 /* For following calls to tcp_respond */ 2516 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2517 goto drop; 2518 break; 2519 #endif /* INET6 */ 2520 case AF_INET: 2521 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2522 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2523 goto drop; 2524 } 2525 2526 if (tiflags & TH_ACK) 2527 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2528 else { 2529 if (tiflags & TH_SYN) 2530 tlen++; 2531 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2532 TH_RST|TH_ACK); 2533 } 2534 if (tcp_saveti) 2535 m_freem(tcp_saveti); 2536 return; 2537 2538 badcsum: 2539 tcpstat.tcps_rcvbadsum++; 2540 drop: 2541 /* 2542 * Drop space held by incoming segment and return. 2543 */ 2544 if (tp) { 2545 if (tp->t_inpcb) 2546 so = tp->t_inpcb->inp_socket; 2547 #ifdef INET6 2548 else if (tp->t_in6pcb) 2549 so = tp->t_in6pcb->in6p_socket; 2550 #endif 2551 else 2552 so = NULL; 2553 #ifdef TCP_DEBUG 2554 if (so && (so->so_options & SO_DEBUG) != 0) 2555 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2556 #endif 2557 } 2558 if (tcp_saveti) 2559 m_freem(tcp_saveti); 2560 m_freem(m); 2561 return; 2562 } 2563 2564 #ifdef TCP_SIGNATURE 2565 int 2566 tcp_signature_apply(void *fstate, caddr_t data, u_int len) 2567 { 2568 2569 MD5Update(fstate, (u_char *)data, len); 2570 return (0); 2571 } 2572 2573 struct secasvar * 2574 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 2575 { 2576 struct secasvar *sav; 2577 #ifdef FAST_IPSEC 2578 union sockaddr_union dst; 2579 #endif 2580 struct ip *ip; 2581 struct ip6_hdr *ip6; 2582 2583 ip = mtod(m, struct ip *); 2584 switch (ip->ip_v) { 2585 case 4: 2586 ip = mtod(m, struct ip *); 2587 ip6 = NULL; 2588 break; 2589 case 6: 2590 ip = NULL; 2591 ip6 = mtod(m, struct ip6_hdr *); 2592 break; 2593 default: 2594 return (NULL); 2595 } 2596 2597 #ifdef FAST_IPSEC 2598 /* Extract the destination from the IP header in the mbuf. */ 2599 bzero(&dst, sizeof(union sockaddr_union)); 2600 dst.sa.sa_len = sizeof(struct sockaddr_in); 2601 dst.sa.sa_family = AF_INET; 2602 dst.sin.sin_addr = ip->ip_dst; 2603 2604 /* 2605 * Look up an SADB entry which matches the address of the peer. 2606 */ 2607 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2608 #else 2609 if (ip) 2610 sav = key_allocsa(AF_INET, (caddr_t)&ip->ip_src, 2611 (caddr_t)&ip->ip_dst, IPPROTO_TCP, 2612 htonl(TCP_SIG_SPI)); 2613 else 2614 sav = key_allocsa(AF_INET6, (caddr_t)&ip6->ip6_src, 2615 (caddr_t)&ip6->ip6_dst, IPPROTO_TCP, 2616 htonl(TCP_SIG_SPI)); 2617 #endif 2618 2619 return (sav); /* freesav must be performed by caller */ 2620 } 2621 2622 int 2623 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 2624 struct secasvar *sav, char *sig) 2625 { 2626 MD5_CTX ctx; 2627 struct ip *ip; 2628 struct ipovly *ipovly; 2629 struct ip6_hdr *ip6; 2630 struct ippseudo ippseudo; 2631 struct ip6_hdr_pseudo ip6pseudo; 2632 struct tcphdr th0; 2633 int l; 2634 2635 if (sav == NULL) 2636 return (-1); 2637 2638 switch (mtod(m, struct ip *)->ip_v) { 2639 case 4: 2640 ip = mtod(m, struct ip *); 2641 ip6 = NULL; 2642 break; 2643 case 6: 2644 ip = NULL; 2645 ip6 = mtod(m, struct ip6_hdr *); 2646 break; 2647 default: 2648 return (-1); 2649 } 2650 2651 MD5Init(&ctx); 2652 2653 if (ip) { 2654 memset(&ippseudo, 0, sizeof(ippseudo)); 2655 ipovly = (struct ipovly *)ip; 2656 ippseudo.ippseudo_src = ipovly->ih_src; 2657 ippseudo.ippseudo_dst = ipovly->ih_dst; 2658 ippseudo.ippseudo_pad = 0; 2659 ippseudo.ippseudo_p = IPPROTO_TCP; 2660 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 2661 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 2662 } else { 2663 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 2664 ip6pseudo.ip6ph_src = ip6->ip6_src; 2665 in6_clearscope(&ip6pseudo.ip6ph_src); 2666 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 2667 in6_clearscope(&ip6pseudo.ip6ph_dst); 2668 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 2669 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 2670 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 2671 } 2672 2673 th0 = *th; 2674 th0.th_sum = 0; 2675 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 2676 2677 l = m->m_pkthdr.len - thoff - sizeof(struct tcphdr); 2678 if (l > 0) 2679 m_apply(m, thoff + sizeof(struct tcphdr), 2680 m->m_pkthdr.len - thoff - sizeof(struct tcphdr), 2681 tcp_signature_apply, &ctx); 2682 2683 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2684 MD5Final(sig, &ctx); 2685 2686 return (0); 2687 } 2688 #endif 2689 2690 int 2691 tcp_dooptions(tp, cp, cnt, th, m, toff, oi) 2692 struct tcpcb *tp; 2693 u_char *cp; 2694 int cnt; 2695 struct tcphdr *th; 2696 struct mbuf *m; 2697 int toff; 2698 struct tcp_opt_info *oi; 2699 { 2700 u_int16_t mss; 2701 int opt, optlen = 0; 2702 #ifdef TCP_SIGNATURE 2703 caddr_t sigp = NULL; 2704 char sigbuf[TCP_SIGLEN]; 2705 struct secasvar *sav = NULL; 2706 #endif 2707 2708 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 2709 opt = cp[0]; 2710 if (opt == TCPOPT_EOL) 2711 break; 2712 if (opt == TCPOPT_NOP) 2713 optlen = 1; 2714 else { 2715 if (cnt < 2) 2716 break; 2717 optlen = cp[1]; 2718 if (optlen < 2 || optlen > cnt) 2719 break; 2720 } 2721 switch (opt) { 2722 2723 default: 2724 continue; 2725 2726 case TCPOPT_MAXSEG: 2727 if (optlen != TCPOLEN_MAXSEG) 2728 continue; 2729 if (!(th->th_flags & TH_SYN)) 2730 continue; 2731 bcopy(cp + 2, &mss, sizeof(mss)); 2732 oi->maxseg = ntohs(mss); 2733 break; 2734 2735 case TCPOPT_WINDOW: 2736 if (optlen != TCPOLEN_WINDOW) 2737 continue; 2738 if (!(th->th_flags & TH_SYN)) 2739 continue; 2740 tp->t_flags |= TF_RCVD_SCALE; 2741 tp->requested_s_scale = cp[2]; 2742 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2743 #if 0 /*XXX*/ 2744 char *p; 2745 2746 if (ip) 2747 p = ntohl(ip->ip_src); 2748 #ifdef INET6 2749 else if (ip6) 2750 p = ip6_sprintf(&ip6->ip6_src); 2751 #endif 2752 else 2753 p = "(unknown)"; 2754 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2755 "assuming %d\n", 2756 tp->requested_s_scale, p, 2757 TCP_MAX_WINSHIFT); 2758 #else 2759 log(LOG_ERR, "TCP: invalid wscale %d, " 2760 "assuming %d\n", 2761 tp->requested_s_scale, 2762 TCP_MAX_WINSHIFT); 2763 #endif 2764 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2765 } 2766 break; 2767 2768 case TCPOPT_TIMESTAMP: 2769 if (optlen != TCPOLEN_TIMESTAMP) 2770 continue; 2771 oi->ts_present = 1; 2772 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2773 NTOHL(oi->ts_val); 2774 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2775 NTOHL(oi->ts_ecr); 2776 2777 /* 2778 * A timestamp received in a SYN makes 2779 * it ok to send timestamp requests and replies. 2780 */ 2781 if (th->th_flags & TH_SYN) { 2782 tp->t_flags |= TF_RCVD_TSTMP; 2783 tp->ts_recent = oi->ts_val; 2784 tp->ts_recent_age = TCP_TIMESTAMP(tp); 2785 } 2786 break; 2787 case TCPOPT_SACK_PERMITTED: 2788 if (optlen != TCPOLEN_SACK_PERMITTED) 2789 continue; 2790 if (!(th->th_flags & TH_SYN)) 2791 continue; 2792 tp->t_flags &= ~TF_CANT_TXSACK; 2793 break; 2794 2795 case TCPOPT_SACK: 2796 if (tp->t_flags & TF_IGNR_RXSACK) 2797 continue; 2798 if (optlen % 8 != 2 || optlen < 10) 2799 continue; 2800 cp += 2; 2801 optlen -= 2; 2802 for (; optlen > 0; cp -= 8, optlen -= 8) { 2803 tcp_seq lwe, rwe; 2804 bcopy((char *)cp, (char *) &lwe, sizeof(lwe)); 2805 NTOHL(lwe); 2806 bcopy((char *)cp, (char *) &rwe, sizeof(rwe)); 2807 NTOHL(rwe); 2808 /* tcp_mark_sacked(tp, lwe, rwe); */ 2809 } 2810 break; 2811 #ifdef TCP_SIGNATURE 2812 case TCPOPT_SIGNATURE: 2813 if (optlen != TCPOLEN_SIGNATURE) 2814 continue; 2815 if (sigp && bcmp(sigp, cp + 2, TCP_SIGLEN)) 2816 return (-1); 2817 2818 sigp = sigbuf; 2819 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 2820 memset(cp + 2, 0, TCP_SIGLEN); 2821 tp->t_flags |= TF_SIGNATURE; 2822 break; 2823 #endif 2824 } 2825 } 2826 2827 #ifdef TCP_SIGNATURE 2828 if (tp->t_flags & TF_SIGNATURE) { 2829 2830 sav = tcp_signature_getsav(m, th); 2831 2832 if (sav == NULL && tp->t_state == TCPS_LISTEN) 2833 return (-1); 2834 } 2835 2836 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 2837 #ifdef FAST_IPSEC 2838 KEY_FREESAV(&sav); 2839 #else 2840 key_freesav(sav); 2841 #endif 2842 return (-1); 2843 } 2844 2845 if (sigp) { 2846 char sig[TCP_SIGLEN]; 2847 2848 TCP_FIELDS_TO_NET(th); 2849 if (tcp_signature(m, th, toff, sav, sig) < 0) { 2850 TCP_FIELDS_TO_HOST(th); 2851 #ifdef FAST_IPSEC 2852 KEY_FREESAV(&sav); 2853 #else 2854 key_freesav(sav); 2855 #endif 2856 return (-1); 2857 } 2858 TCP_FIELDS_TO_HOST(th); 2859 2860 if (bcmp(sig, sigp, TCP_SIGLEN)) { 2861 tcpstat.tcps_badsig++; 2862 #ifdef FAST_IPSEC 2863 KEY_FREESAV(&sav); 2864 #else 2865 key_freesav(sav); 2866 #endif 2867 return (-1); 2868 } else 2869 tcpstat.tcps_goodsig++; 2870 2871 key_sa_recordxfer(sav, m); 2872 #ifdef FAST_IPSEC 2873 KEY_FREESAV(&sav); 2874 #else 2875 key_freesav(sav); 2876 #endif 2877 } 2878 #endif 2879 2880 return (0); 2881 } 2882 2883 /* 2884 * Pull out of band byte out of a segment so 2885 * it doesn't appear in the user's data queue. 2886 * It is still reflected in the segment length for 2887 * sequencing purposes. 2888 */ 2889 void 2890 tcp_pulloutofband(so, th, m, off) 2891 struct socket *so; 2892 struct tcphdr *th; 2893 struct mbuf *m; 2894 int off; 2895 { 2896 int cnt = off + th->th_urp - 1; 2897 2898 while (cnt >= 0) { 2899 if (m->m_len > cnt) { 2900 char *cp = mtod(m, caddr_t) + cnt; 2901 struct tcpcb *tp = sototcpcb(so); 2902 2903 tp->t_iobc = *cp; 2904 tp->t_oobflags |= TCPOOB_HAVEDATA; 2905 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2906 m->m_len--; 2907 return; 2908 } 2909 cnt -= m->m_len; 2910 m = m->m_next; 2911 if (m == 0) 2912 break; 2913 } 2914 panic("tcp_pulloutofband"); 2915 } 2916 2917 /* 2918 * Collect new round-trip time estimate 2919 * and update averages and current timeout. 2920 */ 2921 void 2922 tcp_xmit_timer(tp, rtt) 2923 struct tcpcb *tp; 2924 uint32_t rtt; 2925 { 2926 int32_t delta; 2927 2928 tcpstat.tcps_rttupdated++; 2929 if (tp->t_srtt != 0) { 2930 /* 2931 * srtt is stored as fixed point with 3 bits after the 2932 * binary point (i.e., scaled by 8). The following magic 2933 * is equivalent to the smoothing algorithm in rfc793 with 2934 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2935 * point). Adjust rtt to origin 0. 2936 */ 2937 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2938 if ((tp->t_srtt += delta) <= 0) 2939 tp->t_srtt = 1 << 2; 2940 /* 2941 * We accumulate a smoothed rtt variance (actually, a 2942 * smoothed mean difference), then set the retransmit 2943 * timer to smoothed rtt + 4 times the smoothed variance. 2944 * rttvar is stored as fixed point with 2 bits after the 2945 * binary point (scaled by 4). The following is 2946 * equivalent to rfc793 smoothing with an alpha of .75 2947 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2948 * rfc793's wired-in beta. 2949 */ 2950 if (delta < 0) 2951 delta = -delta; 2952 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2953 if ((tp->t_rttvar += delta) <= 0) 2954 tp->t_rttvar = 1 << 2; 2955 } else { 2956 /* 2957 * No rtt measurement yet - use the unsmoothed rtt. 2958 * Set the variance to half the rtt (so our first 2959 * retransmit happens at 3*rtt). 2960 */ 2961 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 2962 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 2963 } 2964 tp->t_rtttime = 0; 2965 tp->t_rxtshift = 0; 2966 2967 /* 2968 * the retransmit should happen at rtt + 4 * rttvar. 2969 * Because of the way we do the smoothing, srtt and rttvar 2970 * will each average +1/2 tick of bias. When we compute 2971 * the retransmit timer, we want 1/2 tick of rounding and 2972 * 1 extra tick because of +-1/2 tick uncertainty in the 2973 * firing of the timer. The bias will give us exactly the 2974 * 1.5 tick we need. But, because the bias is 2975 * statistical, we have to test that we don't drop below 2976 * the minimum feasible timer (which is 2 ticks). 2977 */ 2978 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2979 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2980 2981 /* 2982 * We received an ack for a packet that wasn't retransmitted; 2983 * it is probably safe to discard any error indications we've 2984 * received recently. This isn't quite right, but close enough 2985 * for now (a route might have failed after we sent a segment, 2986 * and the return path might not be symmetrical). 2987 */ 2988 tp->t_softerror = 0; 2989 } 2990 2991 /* 2992 * Checks for partial ack. If partial ack arrives, force the retransmission 2993 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2994 * 1. By setting snd_nxt to th_ack, this forces retransmission timer to 2995 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2996 */ 2997 int 2998 tcp_newreno(tp, th) 2999 struct tcpcb *tp; 3000 struct tcphdr *th; 3001 { 3002 tcp_seq onxt = tp->snd_nxt; 3003 u_long ocwnd = tp->snd_cwnd; 3004 3005 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 3006 /* 3007 * snd_una has not yet been updated and the socket's send 3008 * buffer has not yet drained off the ACK'd data, so we 3009 * have to leave snd_una as it was to get the correct data 3010 * offset in tcp_output(). 3011 */ 3012 TCP_TIMER_DISARM(tp, TCPT_REXMT); 3013 tp->t_rtttime = 0; 3014 tp->snd_nxt = th->th_ack; 3015 /* 3016 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 3017 * is not yet updated when we're called. 3018 */ 3019 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 3020 (void) tcp_output(tp); 3021 tp->snd_cwnd = ocwnd; 3022 if (SEQ_GT(onxt, tp->snd_nxt)) 3023 tp->snd_nxt = onxt; 3024 /* 3025 * Partial window deflation. Relies on fact that tp->snd_una 3026 * not updated yet. 3027 */ 3028 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 3029 return 1; 3030 } 3031 return 0; 3032 } 3033 3034 3035 /* 3036 * TCP compressed state engine. Currently used to hold compressed 3037 * state for SYN_RECEIVED. 3038 */ 3039 3040 u_long syn_cache_count; 3041 u_int32_t syn_hash1, syn_hash2; 3042 3043 #define SYN_HASH(sa, sp, dp) \ 3044 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3045 ((u_int32_t)(sp)))^syn_hash2))) 3046 #ifndef INET6 3047 #define SYN_HASHALL(hash, src, dst) \ 3048 do { \ 3049 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 3050 ((struct sockaddr_in *)(src))->sin_port, \ 3051 ((struct sockaddr_in *)(dst))->sin_port); \ 3052 } while (/*CONSTCOND*/ 0) 3053 #else 3054 #define SYN_HASH6(sa, sp, dp) \ 3055 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3056 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3057 & 0x7fffffff) 3058 3059 #define SYN_HASHALL(hash, src, dst) \ 3060 do { \ 3061 switch ((src)->sa_family) { \ 3062 case AF_INET: \ 3063 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 3064 ((struct sockaddr_in *)(src))->sin_port, \ 3065 ((struct sockaddr_in *)(dst))->sin_port); \ 3066 break; \ 3067 case AF_INET6: \ 3068 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 3069 ((struct sockaddr_in6 *)(src))->sin6_port, \ 3070 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 3071 break; \ 3072 default: \ 3073 hash = 0; \ 3074 } \ 3075 } while (/*CONSTCOND*/0) 3076 #endif /* INET6 */ 3077 3078 #define SYN_CACHE_RM(sc) \ 3079 do { \ 3080 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 3081 (sc), sc_bucketq); \ 3082 (sc)->sc_tp = NULL; \ 3083 LIST_REMOVE((sc), sc_tpq); \ 3084 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 3085 callout_stop(&(sc)->sc_timer); \ 3086 syn_cache_count--; \ 3087 } while (/*CONSTCOND*/0) 3088 3089 #define SYN_CACHE_PUT(sc) \ 3090 do { \ 3091 if ((sc)->sc_ipopts) \ 3092 (void) m_free((sc)->sc_ipopts); \ 3093 if ((sc)->sc_route4.ro_rt != NULL) \ 3094 RTFREE((sc)->sc_route4.ro_rt); \ 3095 if (callout_invoking(&(sc)->sc_timer)) \ 3096 (sc)->sc_flags |= SCF_DEAD; \ 3097 else \ 3098 pool_put(&syn_cache_pool, (sc)); \ 3099 } while (/*CONSTCOND*/0) 3100 3101 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL); 3102 3103 /* 3104 * We don't estimate RTT with SYNs, so each packet starts with the default 3105 * RTT and each timer step has a fixed timeout value. 3106 */ 3107 #define SYN_CACHE_TIMER_ARM(sc) \ 3108 do { \ 3109 TCPT_RANGESET((sc)->sc_rxtcur, \ 3110 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3111 TCPTV_REXMTMAX); \ 3112 callout_reset(&(sc)->sc_timer, \ 3113 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3114 } while (/*CONSTCOND*/0) 3115 3116 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3117 3118 void 3119 syn_cache_init() 3120 { 3121 int i; 3122 3123 /* Initialize the hash buckets. */ 3124 for (i = 0; i < tcp_syn_cache_size; i++) 3125 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3126 } 3127 3128 void 3129 syn_cache_insert(sc, tp) 3130 struct syn_cache *sc; 3131 struct tcpcb *tp; 3132 { 3133 struct syn_cache_head *scp; 3134 struct syn_cache *sc2; 3135 int s; 3136 3137 /* 3138 * If there are no entries in the hash table, reinitialize 3139 * the hash secrets. 3140 */ 3141 if (syn_cache_count == 0) { 3142 syn_hash1 = arc4random(); 3143 syn_hash2 = arc4random(); 3144 } 3145 3146 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3147 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3148 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3149 3150 /* 3151 * Make sure that we don't overflow the per-bucket 3152 * limit or the total cache size limit. 3153 */ 3154 s = splsoftnet(); 3155 if (scp->sch_length >= tcp_syn_bucket_limit) { 3156 tcpstat.tcps_sc_bucketoverflow++; 3157 /* 3158 * The bucket is full. Toss the oldest element in the 3159 * bucket. This will be the first entry in the bucket. 3160 */ 3161 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3162 #ifdef DIAGNOSTIC 3163 /* 3164 * This should never happen; we should always find an 3165 * entry in our bucket. 3166 */ 3167 if (sc2 == NULL) 3168 panic("syn_cache_insert: bucketoverflow: impossible"); 3169 #endif 3170 SYN_CACHE_RM(sc2); 3171 SYN_CACHE_PUT(sc2); 3172 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3173 struct syn_cache_head *scp2, *sce; 3174 3175 tcpstat.tcps_sc_overflowed++; 3176 /* 3177 * The cache is full. Toss the oldest entry in the 3178 * first non-empty bucket we can find. 3179 * 3180 * XXX We would really like to toss the oldest 3181 * entry in the cache, but we hope that this 3182 * condition doesn't happen very often. 3183 */ 3184 scp2 = scp; 3185 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3186 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3187 for (++scp2; scp2 != scp; scp2++) { 3188 if (scp2 >= sce) 3189 scp2 = &tcp_syn_cache[0]; 3190 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3191 break; 3192 } 3193 #ifdef DIAGNOSTIC 3194 /* 3195 * This should never happen; we should always find a 3196 * non-empty bucket. 3197 */ 3198 if (scp2 == scp) 3199 panic("syn_cache_insert: cacheoverflow: " 3200 "impossible"); 3201 #endif 3202 } 3203 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3204 SYN_CACHE_RM(sc2); 3205 SYN_CACHE_PUT(sc2); 3206 } 3207 3208 /* 3209 * Initialize the entry's timer. 3210 */ 3211 sc->sc_rxttot = 0; 3212 sc->sc_rxtshift = 0; 3213 SYN_CACHE_TIMER_ARM(sc); 3214 3215 /* Link it from tcpcb entry */ 3216 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3217 3218 /* Put it into the bucket. */ 3219 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3220 scp->sch_length++; 3221 syn_cache_count++; 3222 3223 tcpstat.tcps_sc_added++; 3224 splx(s); 3225 } 3226 3227 /* 3228 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3229 * If we have retransmitted an entry the maximum number of times, expire 3230 * that entry. 3231 */ 3232 void 3233 syn_cache_timer(void *arg) 3234 { 3235 struct syn_cache *sc = arg; 3236 int s; 3237 3238 s = splsoftnet(); 3239 callout_ack(&sc->sc_timer); 3240 3241 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3242 tcpstat.tcps_sc_delayed_free++; 3243 pool_put(&syn_cache_pool, sc); 3244 splx(s); 3245 return; 3246 } 3247 3248 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3249 /* Drop it -- too many retransmissions. */ 3250 goto dropit; 3251 } 3252 3253 /* 3254 * Compute the total amount of time this entry has 3255 * been on a queue. If this entry has been on longer 3256 * than the keep alive timer would allow, expire it. 3257 */ 3258 sc->sc_rxttot += sc->sc_rxtcur; 3259 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) 3260 goto dropit; 3261 3262 tcpstat.tcps_sc_retransmitted++; 3263 (void) syn_cache_respond(sc, NULL); 3264 3265 /* Advance the timer back-off. */ 3266 sc->sc_rxtshift++; 3267 SYN_CACHE_TIMER_ARM(sc); 3268 3269 splx(s); 3270 return; 3271 3272 dropit: 3273 tcpstat.tcps_sc_timed_out++; 3274 SYN_CACHE_RM(sc); 3275 SYN_CACHE_PUT(sc); 3276 splx(s); 3277 } 3278 3279 /* 3280 * Remove syn cache created by the specified tcb entry, 3281 * because this does not make sense to keep them 3282 * (if there's no tcb entry, syn cache entry will never be used) 3283 */ 3284 void 3285 syn_cache_cleanup(tp) 3286 struct tcpcb *tp; 3287 { 3288 struct syn_cache *sc, *nsc; 3289 int s; 3290 3291 s = splsoftnet(); 3292 3293 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3294 nsc = LIST_NEXT(sc, sc_tpq); 3295 3296 #ifdef DIAGNOSTIC 3297 if (sc->sc_tp != tp) 3298 panic("invalid sc_tp in syn_cache_cleanup"); 3299 #endif 3300 SYN_CACHE_RM(sc); 3301 SYN_CACHE_PUT(sc); 3302 } 3303 /* just for safety */ 3304 LIST_INIT(&tp->t_sc); 3305 3306 splx(s); 3307 } 3308 3309 /* 3310 * Find an entry in the syn cache. 3311 */ 3312 struct syn_cache * 3313 syn_cache_lookup(src, dst, headp) 3314 struct sockaddr *src; 3315 struct sockaddr *dst; 3316 struct syn_cache_head **headp; 3317 { 3318 struct syn_cache *sc; 3319 struct syn_cache_head *scp; 3320 u_int32_t hash; 3321 int s; 3322 3323 SYN_HASHALL(hash, src, dst); 3324 3325 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3326 *headp = scp; 3327 s = splsoftnet(); 3328 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3329 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3330 if (sc->sc_hash != hash) 3331 continue; 3332 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3333 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3334 splx(s); 3335 return (sc); 3336 } 3337 } 3338 splx(s); 3339 return (NULL); 3340 } 3341 3342 /* 3343 * This function gets called when we receive an ACK for a 3344 * socket in the LISTEN state. We look up the connection 3345 * in the syn cache, and if its there, we pull it out of 3346 * the cache and turn it into a full-blown connection in 3347 * the SYN-RECEIVED state. 3348 * 3349 * The return values may not be immediately obvious, and their effects 3350 * can be subtle, so here they are: 3351 * 3352 * NULL SYN was not found in cache; caller should drop the 3353 * packet and send an RST. 3354 * 3355 * -1 We were unable to create the new connection, and are 3356 * aborting it. An ACK,RST is being sent to the peer 3357 * (unless we got screwey sequence numbners; see below), 3358 * because the 3-way handshake has been completed. Caller 3359 * should not free the mbuf, since we may be using it. If 3360 * we are not, we will free it. 3361 * 3362 * Otherwise, the return value is a pointer to the new socket 3363 * associated with the connection. 3364 */ 3365 struct socket * 3366 syn_cache_get(src, dst, th, hlen, tlen, so, m) 3367 struct sockaddr *src; 3368 struct sockaddr *dst; 3369 struct tcphdr *th; 3370 unsigned int hlen, tlen; 3371 struct socket *so; 3372 struct mbuf *m; 3373 { 3374 struct syn_cache *sc; 3375 struct syn_cache_head *scp; 3376 struct inpcb *inp = NULL; 3377 #ifdef INET6 3378 struct in6pcb *in6p = NULL; 3379 #endif 3380 struct tcpcb *tp = 0; 3381 struct mbuf *am; 3382 int s; 3383 struct socket *oso; 3384 3385 s = splsoftnet(); 3386 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3387 splx(s); 3388 return (NULL); 3389 } 3390 3391 /* 3392 * Verify the sequence and ack numbers. Try getting the correct 3393 * response again. 3394 */ 3395 if ((th->th_ack != sc->sc_iss + 1) || 3396 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3397 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3398 (void) syn_cache_respond(sc, m); 3399 splx(s); 3400 return ((struct socket *)(-1)); 3401 } 3402 3403 /* Remove this cache entry */ 3404 SYN_CACHE_RM(sc); 3405 splx(s); 3406 3407 /* 3408 * Ok, create the full blown connection, and set things up 3409 * as they would have been set up if we had created the 3410 * connection when the SYN arrived. If we can't create 3411 * the connection, abort it. 3412 */ 3413 /* 3414 * inp still has the OLD in_pcb stuff, set the 3415 * v6-related flags on the new guy, too. This is 3416 * done particularly for the case where an AF_INET6 3417 * socket is bound only to a port, and a v4 connection 3418 * comes in on that port. 3419 * we also copy the flowinfo from the original pcb 3420 * to the new one. 3421 */ 3422 oso = so; 3423 so = sonewconn(so, SS_ISCONNECTED); 3424 if (so == NULL) 3425 goto resetandabort; 3426 3427 switch (so->so_proto->pr_domain->dom_family) { 3428 #ifdef INET 3429 case AF_INET: 3430 inp = sotoinpcb(so); 3431 break; 3432 #endif 3433 #ifdef INET6 3434 case AF_INET6: 3435 in6p = sotoin6pcb(so); 3436 break; 3437 #endif 3438 } 3439 switch (src->sa_family) { 3440 #ifdef INET 3441 case AF_INET: 3442 if (inp) { 3443 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3444 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3445 inp->inp_options = ip_srcroute(); 3446 in_pcbstate(inp, INP_BOUND); 3447 if (inp->inp_options == NULL) { 3448 inp->inp_options = sc->sc_ipopts; 3449 sc->sc_ipopts = NULL; 3450 } 3451 } 3452 #ifdef INET6 3453 else if (in6p) { 3454 /* IPv4 packet to AF_INET6 socket */ 3455 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3456 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3457 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3458 &in6p->in6p_laddr.s6_addr32[3], 3459 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3460 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3461 in6totcpcb(in6p)->t_family = AF_INET; 3462 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3463 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3464 else 3465 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3466 in6_pcbstate(in6p, IN6P_BOUND); 3467 } 3468 #endif 3469 break; 3470 #endif 3471 #ifdef INET6 3472 case AF_INET6: 3473 if (in6p) { 3474 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3475 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3476 in6_pcbstate(in6p, IN6P_BOUND); 3477 } 3478 break; 3479 #endif 3480 } 3481 #ifdef INET6 3482 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3483 struct in6pcb *oin6p = sotoin6pcb(oso); 3484 /* inherit socket options from the listening socket */ 3485 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3486 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3487 m_freem(in6p->in6p_options); 3488 in6p->in6p_options = 0; 3489 } 3490 ip6_savecontrol(in6p, &in6p->in6p_options, 3491 mtod(m, struct ip6_hdr *), m); 3492 } 3493 #endif 3494 3495 #if defined(IPSEC) || defined(FAST_IPSEC) 3496 /* 3497 * we make a copy of policy, instead of sharing the policy, 3498 * for better behavior in terms of SA lookup and dead SA removal. 3499 */ 3500 if (inp) { 3501 /* copy old policy into new socket's */ 3502 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3503 printf("tcp_input: could not copy policy\n"); 3504 } 3505 #ifdef INET6 3506 else if (in6p) { 3507 /* copy old policy into new socket's */ 3508 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3509 in6p->in6p_sp)) 3510 printf("tcp_input: could not copy policy\n"); 3511 } 3512 #endif 3513 #endif 3514 3515 /* 3516 * Give the new socket our cached route reference. 3517 */ 3518 if (inp) 3519 inp->inp_route = sc->sc_route4; /* struct assignment */ 3520 #ifdef INET6 3521 else 3522 in6p->in6p_route = sc->sc_route6; 3523 #endif 3524 sc->sc_route4.ro_rt = NULL; 3525 3526 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3527 if (am == NULL) 3528 goto resetandabort; 3529 MCLAIM(am, &tcp_mowner); 3530 am->m_len = src->sa_len; 3531 bcopy(src, mtod(am, caddr_t), src->sa_len); 3532 if (inp) { 3533 if (in_pcbconnect(inp, am)) { 3534 (void) m_free(am); 3535 goto resetandabort; 3536 } 3537 } 3538 #ifdef INET6 3539 else if (in6p) { 3540 if (src->sa_family == AF_INET) { 3541 /* IPv4 packet to AF_INET6 socket */ 3542 struct sockaddr_in6 *sin6; 3543 sin6 = mtod(am, struct sockaddr_in6 *); 3544 am->m_len = sizeof(*sin6); 3545 bzero(sin6, sizeof(*sin6)); 3546 sin6->sin6_family = AF_INET6; 3547 sin6->sin6_len = sizeof(*sin6); 3548 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3549 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3550 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3551 &sin6->sin6_addr.s6_addr32[3], 3552 sizeof(sin6->sin6_addr.s6_addr32[3])); 3553 } 3554 if (in6_pcbconnect(in6p, am)) { 3555 (void) m_free(am); 3556 goto resetandabort; 3557 } 3558 } 3559 #endif 3560 else { 3561 (void) m_free(am); 3562 goto resetandabort; 3563 } 3564 (void) m_free(am); 3565 3566 if (inp) 3567 tp = intotcpcb(inp); 3568 #ifdef INET6 3569 else if (in6p) 3570 tp = in6totcpcb(in6p); 3571 #endif 3572 else 3573 tp = NULL; 3574 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3575 if (sc->sc_request_r_scale != 15) { 3576 tp->requested_s_scale = sc->sc_requested_s_scale; 3577 tp->request_r_scale = sc->sc_request_r_scale; 3578 tp->snd_scale = sc->sc_requested_s_scale; 3579 tp->rcv_scale = sc->sc_request_r_scale; 3580 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3581 } 3582 if (sc->sc_flags & SCF_TIMESTAMP) 3583 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3584 tp->ts_timebase = sc->sc_timebase; 3585 3586 tp->t_template = tcp_template(tp); 3587 if (tp->t_template == 0) { 3588 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3589 so = NULL; 3590 m_freem(m); 3591 goto abort; 3592 } 3593 3594 tp->iss = sc->sc_iss; 3595 tp->irs = sc->sc_irs; 3596 tcp_sendseqinit(tp); 3597 tcp_rcvseqinit(tp); 3598 tp->t_state = TCPS_SYN_RECEIVED; 3599 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3600 tcpstat.tcps_accepts++; 3601 3602 #ifdef TCP_SIGNATURE 3603 if (sc->sc_flags & SCF_SIGNATURE) 3604 tp->t_flags |= TF_SIGNATURE; 3605 #endif 3606 3607 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3608 tp->t_ourmss = sc->sc_ourmaxseg; 3609 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3610 3611 /* 3612 * Initialize the initial congestion window. If we 3613 * had to retransmit the SYN,ACK, we must initialize cwnd 3614 * to 1 segment (i.e. the Loss Window). 3615 */ 3616 if (sc->sc_rxtshift) 3617 tp->snd_cwnd = tp->t_peermss; 3618 else { 3619 int ss = tcp_init_win; 3620 #ifdef INET 3621 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3622 ss = tcp_init_win_local; 3623 #endif 3624 #ifdef INET6 3625 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3626 ss = tcp_init_win_local; 3627 #endif 3628 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3629 } 3630 3631 tcp_rmx_rtt(tp); 3632 tp->snd_wl1 = sc->sc_irs; 3633 tp->rcv_up = sc->sc_irs + 1; 3634 3635 /* 3636 * This is what whould have happened in tcp_output() when 3637 * the SYN,ACK was sent. 3638 */ 3639 tp->snd_up = tp->snd_una; 3640 tp->snd_max = tp->snd_nxt = tp->iss+1; 3641 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3642 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3643 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3644 tp->last_ack_sent = tp->rcv_nxt; 3645 3646 tcpstat.tcps_sc_completed++; 3647 SYN_CACHE_PUT(sc); 3648 return (so); 3649 3650 resetandabort: 3651 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3652 abort: 3653 if (so != NULL) 3654 (void) soabort(so); 3655 SYN_CACHE_PUT(sc); 3656 tcpstat.tcps_sc_aborted++; 3657 return ((struct socket *)(-1)); 3658 } 3659 3660 /* 3661 * This function is called when we get a RST for a 3662 * non-existent connection, so that we can see if the 3663 * connection is in the syn cache. If it is, zap it. 3664 */ 3665 3666 void 3667 syn_cache_reset(src, dst, th) 3668 struct sockaddr *src; 3669 struct sockaddr *dst; 3670 struct tcphdr *th; 3671 { 3672 struct syn_cache *sc; 3673 struct syn_cache_head *scp; 3674 int s = splsoftnet(); 3675 3676 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3677 splx(s); 3678 return; 3679 } 3680 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3681 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3682 splx(s); 3683 return; 3684 } 3685 SYN_CACHE_RM(sc); 3686 splx(s); 3687 tcpstat.tcps_sc_reset++; 3688 SYN_CACHE_PUT(sc); 3689 } 3690 3691 void 3692 syn_cache_unreach(src, dst, th) 3693 struct sockaddr *src; 3694 struct sockaddr *dst; 3695 struct tcphdr *th; 3696 { 3697 struct syn_cache *sc; 3698 struct syn_cache_head *scp; 3699 int s; 3700 3701 s = splsoftnet(); 3702 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3703 splx(s); 3704 return; 3705 } 3706 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3707 if (ntohl (th->th_seq) != sc->sc_iss) { 3708 splx(s); 3709 return; 3710 } 3711 3712 /* 3713 * If we've retransmitted 3 times and this is our second error, 3714 * we remove the entry. Otherwise, we allow it to continue on. 3715 * This prevents us from incorrectly nuking an entry during a 3716 * spurious network outage. 3717 * 3718 * See tcp_notify(). 3719 */ 3720 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3721 sc->sc_flags |= SCF_UNREACH; 3722 splx(s); 3723 return; 3724 } 3725 3726 SYN_CACHE_RM(sc); 3727 splx(s); 3728 tcpstat.tcps_sc_unreach++; 3729 SYN_CACHE_PUT(sc); 3730 } 3731 3732 /* 3733 * Given a LISTEN socket and an inbound SYN request, add 3734 * this to the syn cache, and send back a segment: 3735 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3736 * to the source. 3737 * 3738 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3739 * Doing so would require that we hold onto the data and deliver it 3740 * to the application. However, if we are the target of a SYN-flood 3741 * DoS attack, an attacker could send data which would eventually 3742 * consume all available buffer space if it were ACKed. By not ACKing 3743 * the data, we avoid this DoS scenario. 3744 */ 3745 3746 int 3747 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi) 3748 struct sockaddr *src; 3749 struct sockaddr *dst; 3750 struct tcphdr *th; 3751 unsigned int hlen; 3752 struct socket *so; 3753 struct mbuf *m; 3754 u_char *optp; 3755 int optlen; 3756 struct tcp_opt_info *oi; 3757 { 3758 struct tcpcb tb, *tp; 3759 long win; 3760 struct syn_cache *sc; 3761 struct syn_cache_head *scp; 3762 struct mbuf *ipopts; 3763 struct tcp_opt_info opti; 3764 3765 tp = sototcpcb(so); 3766 3767 bzero(&opti, sizeof(opti)); 3768 3769 /* 3770 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3771 * 3772 * Note this check is performed in tcp_input() very early on. 3773 */ 3774 3775 /* 3776 * Initialize some local state. 3777 */ 3778 win = sbspace(&so->so_rcv); 3779 if (win > TCP_MAXWIN) 3780 win = TCP_MAXWIN; 3781 3782 switch (src->sa_family) { 3783 #ifdef INET 3784 case AF_INET: 3785 /* 3786 * Remember the IP options, if any. 3787 */ 3788 ipopts = ip_srcroute(); 3789 break; 3790 #endif 3791 default: 3792 ipopts = NULL; 3793 } 3794 3795 #ifdef TCP_SIGNATURE 3796 if (optp || (tp->t_flags & TF_SIGNATURE)) 3797 #else 3798 if (optp) 3799 #endif 3800 { 3801 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 3802 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 3803 return (0); 3804 } else 3805 tb.t_flags = 0; 3806 3807 /* 3808 * See if we already have an entry for this connection. 3809 * If we do, resend the SYN,ACK. We do not count this 3810 * as a retransmission (XXX though maybe we should). 3811 */ 3812 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3813 tcpstat.tcps_sc_dupesyn++; 3814 if (ipopts) { 3815 /* 3816 * If we were remembering a previous source route, 3817 * forget it and use the new one we've been given. 3818 */ 3819 if (sc->sc_ipopts) 3820 (void) m_free(sc->sc_ipopts); 3821 sc->sc_ipopts = ipopts; 3822 } 3823 sc->sc_timestamp = tb.ts_recent; 3824 if (syn_cache_respond(sc, m) == 0) { 3825 tcpstat.tcps_sndacks++; 3826 tcpstat.tcps_sndtotal++; 3827 } 3828 return (1); 3829 } 3830 3831 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3832 if (sc == NULL) { 3833 if (ipopts) 3834 (void) m_free(ipopts); 3835 return (0); 3836 } 3837 3838 /* 3839 * Fill in the cache, and put the necessary IP and TCP 3840 * options into the reply. 3841 */ 3842 bzero(sc, sizeof(struct syn_cache)); 3843 callout_init(&sc->sc_timer); 3844 bcopy(src, &sc->sc_src, src->sa_len); 3845 bcopy(dst, &sc->sc_dst, dst->sa_len); 3846 sc->sc_flags = 0; 3847 sc->sc_ipopts = ipopts; 3848 sc->sc_irs = th->th_seq; 3849 switch (src->sa_family) { 3850 #ifdef INET 3851 case AF_INET: 3852 { 3853 struct sockaddr_in *srcin = (void *) src; 3854 struct sockaddr_in *dstin = (void *) dst; 3855 3856 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 3857 &srcin->sin_addr, dstin->sin_port, 3858 srcin->sin_port, sizeof(dstin->sin_addr), 0); 3859 break; 3860 } 3861 #endif /* INET */ 3862 #ifdef INET6 3863 case AF_INET6: 3864 { 3865 struct sockaddr_in6 *srcin6 = (void *) src; 3866 struct sockaddr_in6 *dstin6 = (void *) dst; 3867 3868 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 3869 &srcin6->sin6_addr, dstin6->sin6_port, 3870 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 3871 break; 3872 } 3873 #endif /* INET6 */ 3874 } 3875 sc->sc_peermaxseg = oi->maxseg; 3876 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3877 m->m_pkthdr.rcvif : NULL, 3878 sc->sc_src.sa.sa_family); 3879 sc->sc_win = win; 3880 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 3881 sc->sc_timestamp = tb.ts_recent; 3882 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 3883 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 3884 sc->sc_flags |= SCF_TIMESTAMP; 3885 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3886 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3887 sc->sc_requested_s_scale = tb.requested_s_scale; 3888 sc->sc_request_r_scale = 0; 3889 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3890 TCP_MAXWIN << sc->sc_request_r_scale < 3891 so->so_rcv.sb_hiwat) 3892 sc->sc_request_r_scale++; 3893 } else { 3894 sc->sc_requested_s_scale = 15; 3895 sc->sc_request_r_scale = 15; 3896 } 3897 #ifdef TCP_SIGNATURE 3898 if (tb.t_flags & TF_SIGNATURE) 3899 sc->sc_flags |= SCF_SIGNATURE; 3900 #endif 3901 sc->sc_tp = tp; 3902 if (syn_cache_respond(sc, m) == 0) { 3903 syn_cache_insert(sc, tp); 3904 tcpstat.tcps_sndacks++; 3905 tcpstat.tcps_sndtotal++; 3906 } else { 3907 SYN_CACHE_PUT(sc); 3908 tcpstat.tcps_sc_dropped++; 3909 } 3910 return (1); 3911 } 3912 3913 int 3914 syn_cache_respond(sc, m) 3915 struct syn_cache *sc; 3916 struct mbuf *m; 3917 { 3918 struct route *ro; 3919 u_int8_t *optp; 3920 int optlen, error; 3921 u_int16_t tlen; 3922 struct ip *ip = NULL; 3923 #ifdef INET6 3924 struct ip6_hdr *ip6 = NULL; 3925 #endif 3926 struct tcpcb *tp; 3927 struct tcphdr *th; 3928 u_int hlen; 3929 struct socket *so; 3930 3931 switch (sc->sc_src.sa.sa_family) { 3932 case AF_INET: 3933 hlen = sizeof(struct ip); 3934 ro = &sc->sc_route4; 3935 break; 3936 #ifdef INET6 3937 case AF_INET6: 3938 hlen = sizeof(struct ip6_hdr); 3939 ro = (struct route *)&sc->sc_route6; 3940 break; 3941 #endif 3942 default: 3943 if (m) 3944 m_freem(m); 3945 return (EAFNOSUPPORT); 3946 } 3947 3948 /* Compute the size of the TCP options. */ 3949 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 3950 #ifdef TCP_SIGNATURE 3951 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 3952 #endif 3953 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 3954 3955 tlen = hlen + sizeof(struct tcphdr) + optlen; 3956 3957 /* 3958 * Create the IP+TCP header from scratch. 3959 */ 3960 if (m) 3961 m_freem(m); 3962 #ifdef DIAGNOSTIC 3963 if (max_linkhdr + tlen > MCLBYTES) 3964 return (ENOBUFS); 3965 #endif 3966 MGETHDR(m, M_DONTWAIT, MT_DATA); 3967 if (m && tlen > MHLEN) { 3968 MCLGET(m, M_DONTWAIT); 3969 if ((m->m_flags & M_EXT) == 0) { 3970 m_freem(m); 3971 m = NULL; 3972 } 3973 } 3974 if (m == NULL) 3975 return (ENOBUFS); 3976 MCLAIM(m, &tcp_tx_mowner); 3977 3978 /* Fixup the mbuf. */ 3979 m->m_data += max_linkhdr; 3980 m->m_len = m->m_pkthdr.len = tlen; 3981 if (sc->sc_tp) { 3982 tp = sc->sc_tp; 3983 if (tp->t_inpcb) 3984 so = tp->t_inpcb->inp_socket; 3985 #ifdef INET6 3986 else if (tp->t_in6pcb) 3987 so = tp->t_in6pcb->in6p_socket; 3988 #endif 3989 else 3990 so = NULL; 3991 } else 3992 so = NULL; 3993 m->m_pkthdr.rcvif = NULL; 3994 memset(mtod(m, u_char *), 0, tlen); 3995 3996 switch (sc->sc_src.sa.sa_family) { 3997 case AF_INET: 3998 ip = mtod(m, struct ip *); 3999 ip->ip_v = 4; 4000 ip->ip_dst = sc->sc_src.sin.sin_addr; 4001 ip->ip_src = sc->sc_dst.sin.sin_addr; 4002 ip->ip_p = IPPROTO_TCP; 4003 th = (struct tcphdr *)(ip + 1); 4004 th->th_dport = sc->sc_src.sin.sin_port; 4005 th->th_sport = sc->sc_dst.sin.sin_port; 4006 break; 4007 #ifdef INET6 4008 case AF_INET6: 4009 ip6 = mtod(m, struct ip6_hdr *); 4010 ip6->ip6_vfc = IPV6_VERSION; 4011 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4012 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4013 ip6->ip6_nxt = IPPROTO_TCP; 4014 /* ip6_plen will be updated in ip6_output() */ 4015 th = (struct tcphdr *)(ip6 + 1); 4016 th->th_dport = sc->sc_src.sin6.sin6_port; 4017 th->th_sport = sc->sc_dst.sin6.sin6_port; 4018 break; 4019 #endif 4020 default: 4021 th = NULL; 4022 } 4023 4024 th->th_seq = htonl(sc->sc_iss); 4025 th->th_ack = htonl(sc->sc_irs + 1); 4026 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4027 th->th_flags = TH_SYN|TH_ACK; 4028 th->th_win = htons(sc->sc_win); 4029 /* th_sum already 0 */ 4030 /* th_urp already 0 */ 4031 4032 /* Tack on the TCP options. */ 4033 optp = (u_int8_t *)(th + 1); 4034 *optp++ = TCPOPT_MAXSEG; 4035 *optp++ = 4; 4036 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4037 *optp++ = sc->sc_ourmaxseg & 0xff; 4038 4039 if (sc->sc_request_r_scale != 15) { 4040 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4041 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4042 sc->sc_request_r_scale); 4043 optp += 4; 4044 } 4045 4046 if (sc->sc_flags & SCF_TIMESTAMP) { 4047 u_int32_t *lp = (u_int32_t *)(optp); 4048 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4049 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4050 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4051 *lp = htonl(sc->sc_timestamp); 4052 optp += TCPOLEN_TSTAMP_APPA; 4053 } 4054 4055 #ifdef TCP_SIGNATURE 4056 if (sc->sc_flags & SCF_SIGNATURE) { 4057 struct secasvar *sav; 4058 u_int8_t *sigp; 4059 4060 sav = tcp_signature_getsav(m, th); 4061 4062 if (sav == NULL) { 4063 if (m) 4064 m_freem(m); 4065 return (EPERM); 4066 } 4067 4068 *optp++ = TCPOPT_SIGNATURE; 4069 *optp++ = TCPOLEN_SIGNATURE; 4070 sigp = optp; 4071 bzero(optp, TCP_SIGLEN); 4072 optp += TCP_SIGLEN; 4073 *optp++ = TCPOPT_NOP; 4074 *optp++ = TCPOPT_EOL; 4075 4076 (void)tcp_signature(m, th, hlen, sav, sigp); 4077 4078 key_sa_recordxfer(sav, m); 4079 #ifdef FAST_IPSEC 4080 KEY_FREESAV(&sav); 4081 #else 4082 key_freesav(sav); 4083 #endif 4084 } 4085 #endif 4086 4087 /* Compute the packet's checksum. */ 4088 switch (sc->sc_src.sa.sa_family) { 4089 case AF_INET: 4090 ip->ip_len = htons(tlen - hlen); 4091 th->th_sum = 0; 4092 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4093 break; 4094 #ifdef INET6 4095 case AF_INET6: 4096 ip6->ip6_plen = htons(tlen - hlen); 4097 th->th_sum = 0; 4098 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4099 break; 4100 #endif 4101 } 4102 4103 /* 4104 * Fill in some straggling IP bits. Note the stack expects 4105 * ip_len to be in host order, for convenience. 4106 */ 4107 switch (sc->sc_src.sa.sa_family) { 4108 #ifdef INET 4109 case AF_INET: 4110 ip->ip_len = htons(tlen); 4111 ip->ip_ttl = ip_defttl; 4112 /* XXX tos? */ 4113 break; 4114 #endif 4115 #ifdef INET6 4116 case AF_INET6: 4117 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4118 ip6->ip6_vfc |= IPV6_VERSION; 4119 ip6->ip6_plen = htons(tlen - hlen); 4120 /* ip6_hlim will be initialized afterwards */ 4121 /* XXX flowlabel? */ 4122 break; 4123 #endif 4124 } 4125 4126 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4127 tp = sc->sc_tp; 4128 4129 switch (sc->sc_src.sa.sa_family) { 4130 #ifdef INET 4131 case AF_INET: 4132 error = ip_output(m, sc->sc_ipopts, ro, 4133 (ip_mtudisc ? IP_MTUDISC : 0), 4134 (struct ip_moptions *)NULL, so); 4135 break; 4136 #endif 4137 #ifdef INET6 4138 case AF_INET6: 4139 ip6->ip6_hlim = in6_selecthlim(NULL, 4140 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 4141 4142 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0, 4143 (struct ip6_moptions *)0, so, NULL); 4144 break; 4145 #endif 4146 default: 4147 error = EAFNOSUPPORT; 4148 break; 4149 } 4150 return (error); 4151 } 4152