1 /* $NetBSD: tcp_input.c,v 1.226 2005/04/03 05:02:46 yamt Exp $ */ 2 3 /* 4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the project nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 /* 33 * @(#)COPYRIGHT 1.1 (NRL) 17 January 1995 34 * 35 * NRL grants permission for redistribution and use in source and binary 36 * forms, with or without modification, of the software and documentation 37 * created at NRL provided that the following conditions are met: 38 * 39 * 1. Redistributions of source code must retain the above copyright 40 * notice, this list of conditions and the following disclaimer. 41 * 2. Redistributions in binary form must reproduce the above copyright 42 * notice, this list of conditions and the following disclaimer in the 43 * documentation and/or other materials provided with the distribution. 44 * 3. All advertising materials mentioning features or use of this software 45 * must display the following acknowledgements: 46 * This product includes software developed by the University of 47 * California, Berkeley and its contributors. 48 * This product includes software developed at the Information 49 * Technology Division, US Naval Research Laboratory. 50 * 4. Neither the name of the NRL nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS 55 * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 56 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A 57 * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NRL OR 58 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 59 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 60 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 61 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 62 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 63 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 64 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 65 * 66 * The views and conclusions contained in the software and documentation 67 * are those of the authors and should not be interpreted as representing 68 * official policies, either expressed or implied, of the US Naval 69 * Research Laboratory (NRL). 70 */ 71 72 /*- 73 * Copyright (c) 1997, 1998, 1999, 2001, 2005 The NetBSD Foundation, Inc. 74 * All rights reserved. 75 * 76 * This code is derived from software contributed to The NetBSD Foundation 77 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 78 * Facility, NASA Ames Research Center. 79 * This code is derived from software contributed to The NetBSD Foundation 80 * by Charles M. Hannum. 81 * 82 * Redistribution and use in source and binary forms, with or without 83 * modification, are permitted provided that the following conditions 84 * are met: 85 * 1. Redistributions of source code must retain the above copyright 86 * notice, this list of conditions and the following disclaimer. 87 * 2. Redistributions in binary form must reproduce the above copyright 88 * notice, this list of conditions and the following disclaimer in the 89 * documentation and/or other materials provided with the distribution. 90 * 3. All advertising materials mentioning features or use of this software 91 * must display the following acknowledgement: 92 * This product includes software developed by the NetBSD 93 * Foundation, Inc. and its contributors. 94 * 4. Neither the name of The NetBSD Foundation nor the names of its 95 * contributors may be used to endorse or promote products derived 96 * from this software without specific prior written permission. 97 * 98 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 99 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 100 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 101 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 102 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 103 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 104 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 105 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 106 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 107 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 108 * POSSIBILITY OF SUCH DAMAGE. 109 */ 110 111 /* 112 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 113 * The Regents of the University of California. All rights reserved. 114 * 115 * Redistribution and use in source and binary forms, with or without 116 * modification, are permitted provided that the following conditions 117 * are met: 118 * 1. Redistributions of source code must retain the above copyright 119 * notice, this list of conditions and the following disclaimer. 120 * 2. Redistributions in binary form must reproduce the above copyright 121 * notice, this list of conditions and the following disclaimer in the 122 * documentation and/or other materials provided with the distribution. 123 * 3. Neither the name of the University nor the names of its contributors 124 * may be used to endorse or promote products derived from this software 125 * without specific prior written permission. 126 * 127 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 128 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 129 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 130 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 131 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 132 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 133 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 134 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 135 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 136 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 137 * SUCH DAMAGE. 138 * 139 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 140 */ 141 142 /* 143 * TODO list for SYN cache stuff: 144 * 145 * Find room for a "state" field, which is needed to keep a 146 * compressed state for TIME_WAIT TCBs. It's been noted already 147 * that this is fairly important for very high-volume web and 148 * mail servers, which use a large number of short-lived 149 * connections. 150 */ 151 152 #include <sys/cdefs.h> 153 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.226 2005/04/03 05:02:46 yamt Exp $"); 154 155 #include "opt_inet.h" 156 #include "opt_ipsec.h" 157 #include "opt_inet_csum.h" 158 #include "opt_tcp_debug.h" 159 160 #include <sys/param.h> 161 #include <sys/systm.h> 162 #include <sys/malloc.h> 163 #include <sys/mbuf.h> 164 #include <sys/protosw.h> 165 #include <sys/socket.h> 166 #include <sys/socketvar.h> 167 #include <sys/errno.h> 168 #include <sys/syslog.h> 169 #include <sys/pool.h> 170 #include <sys/domain.h> 171 #include <sys/kernel.h> 172 #ifdef TCP_SIGNATURE 173 #include <sys/md5.h> 174 #endif 175 176 #include <net/if.h> 177 #include <net/route.h> 178 #include <net/if_types.h> 179 180 #include <netinet/in.h> 181 #include <netinet/in_systm.h> 182 #include <netinet/ip.h> 183 #include <netinet/in_pcb.h> 184 #include <netinet/in_var.h> 185 #include <netinet/ip_var.h> 186 187 #ifdef INET6 188 #ifndef INET 189 #include <netinet/in.h> 190 #endif 191 #include <netinet/ip6.h> 192 #include <netinet6/ip6_var.h> 193 #include <netinet6/in6_pcb.h> 194 #include <netinet6/ip6_var.h> 195 #include <netinet6/in6_var.h> 196 #include <netinet/icmp6.h> 197 #include <netinet6/nd6.h> 198 #endif 199 200 #ifndef INET6 201 /* always need ip6.h for IP6_EXTHDR_GET */ 202 #include <netinet/ip6.h> 203 #endif 204 205 #include <netinet/tcp.h> 206 #include <netinet/tcp_fsm.h> 207 #include <netinet/tcp_seq.h> 208 #include <netinet/tcp_timer.h> 209 #include <netinet/tcp_var.h> 210 #include <netinet/tcpip.h> 211 #include <netinet/tcp_debug.h> 212 213 #include <machine/stdarg.h> 214 215 #ifdef IPSEC 216 #include <netinet6/ipsec.h> 217 #include <netkey/key.h> 218 #endif /*IPSEC*/ 219 #ifdef INET6 220 #include "faith.h" 221 #if defined(NFAITH) && NFAITH > 0 222 #include <net/if_faith.h> 223 #endif 224 #endif /* IPSEC */ 225 226 #ifdef FAST_IPSEC 227 #include <netipsec/ipsec.h> 228 #include <netipsec/ipsec_var.h> /* XXX ipsecstat namespace */ 229 #include <netipsec/key.h> 230 #ifdef INET6 231 #include <netipsec/ipsec6.h> 232 #endif 233 #endif /* FAST_IPSEC*/ 234 235 int tcprexmtthresh = 3; 236 int tcp_log_refused; 237 238 static int tcp_rst_ppslim_count = 0; 239 static struct timeval tcp_rst_ppslim_last; 240 static int tcp_ackdrop_ppslim_count = 0; 241 static struct timeval tcp_ackdrop_ppslim_last; 242 243 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 244 245 /* for modulo comparisons of timestamps */ 246 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 247 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 248 249 /* 250 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 251 */ 252 #ifdef INET6 253 #define ND6_HINT(tp) \ 254 do { \ 255 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 && \ 256 tp->t_in6pcb->in6p_route.ro_rt) { \ 257 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \ 258 } \ 259 } while (/*CONSTCOND*/ 0) 260 #else 261 #define ND6_HINT(tp) 262 #endif 263 264 /* 265 * Macro to compute ACK transmission behavior. Delay the ACK unless 266 * we have already delayed an ACK (must send an ACK every two segments). 267 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 268 * option is enabled. 269 */ 270 #define TCP_SETUP_ACK(tp, th) \ 271 do { \ 272 if ((tp)->t_flags & TF_DELACK || \ 273 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 274 tp->t_flags |= TF_ACKNOW; \ 275 else \ 276 TCP_SET_DELACK(tp); \ 277 } while (/*CONSTCOND*/ 0) 278 279 /* 280 * Convert TCP protocol fields to host order for easier processing. 281 */ 282 #define TCP_FIELDS_TO_HOST(th) \ 283 do { \ 284 NTOHL((th)->th_seq); \ 285 NTOHL((th)->th_ack); \ 286 NTOHS((th)->th_win); \ 287 NTOHS((th)->th_urp); \ 288 } while (/*CONSTCOND*/ 0) 289 290 /* 291 * ... and reverse the above. 292 */ 293 #define TCP_FIELDS_TO_NET(th) \ 294 do { \ 295 HTONL((th)->th_seq); \ 296 HTONL((th)->th_ack); \ 297 HTONS((th)->th_win); \ 298 HTONS((th)->th_urp); \ 299 } while (/*CONSTCOND*/ 0) 300 301 #ifdef TCP_CSUM_COUNTERS 302 #include <sys/device.h> 303 304 extern struct evcnt tcp_hwcsum_ok; 305 extern struct evcnt tcp_hwcsum_bad; 306 extern struct evcnt tcp_hwcsum_data; 307 extern struct evcnt tcp_swcsum; 308 309 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 310 311 #else 312 313 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 314 315 #endif /* TCP_CSUM_COUNTERS */ 316 317 #ifdef TCP_REASS_COUNTERS 318 #include <sys/device.h> 319 320 extern struct evcnt tcp_reass_; 321 extern struct evcnt tcp_reass_empty; 322 extern struct evcnt tcp_reass_iteration[8]; 323 extern struct evcnt tcp_reass_prependfirst; 324 extern struct evcnt tcp_reass_prepend; 325 extern struct evcnt tcp_reass_insert; 326 extern struct evcnt tcp_reass_inserttail; 327 extern struct evcnt tcp_reass_append; 328 extern struct evcnt tcp_reass_appendtail; 329 extern struct evcnt tcp_reass_overlaptail; 330 extern struct evcnt tcp_reass_overlapfront; 331 extern struct evcnt tcp_reass_segdup; 332 extern struct evcnt tcp_reass_fragdup; 333 334 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 335 336 #else 337 338 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 339 340 #endif /* TCP_REASS_COUNTERS */ 341 342 #ifdef INET 343 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 344 #endif 345 #ifdef INET6 346 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 347 #endif 348 349 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 350 351 POOL_INIT(tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", NULL); 352 353 struct ipqent * 354 tcpipqent_alloc() 355 { 356 struct ipqent *ipqe; 357 int s; 358 359 s = splvm(); 360 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 361 splx(s); 362 363 return ipqe; 364 } 365 366 void 367 tcpipqent_free(struct ipqent *ipqe) 368 { 369 int s; 370 371 s = splvm(); 372 pool_put(&tcpipqent_pool, ipqe); 373 splx(s); 374 } 375 376 int 377 tcp_reass(struct tcpcb *tp, struct tcphdr *th, struct mbuf *m, int *tlen) 378 { 379 struct ipqent *p, *q, *nq, *tiqe = NULL; 380 struct socket *so = NULL; 381 int pkt_flags; 382 tcp_seq pkt_seq; 383 unsigned pkt_len; 384 u_long rcvpartdupbyte = 0; 385 u_long rcvoobyte; 386 #ifdef TCP_REASS_COUNTERS 387 u_int count = 0; 388 #endif 389 390 if (tp->t_inpcb) 391 so = tp->t_inpcb->inp_socket; 392 #ifdef INET6 393 else if (tp->t_in6pcb) 394 so = tp->t_in6pcb->in6p_socket; 395 #endif 396 397 TCP_REASS_LOCK_CHECK(tp); 398 399 /* 400 * Call with th==0 after become established to 401 * force pre-ESTABLISHED data up to user socket. 402 */ 403 if (th == 0) 404 goto present; 405 406 rcvoobyte = *tlen; 407 /* 408 * Copy these to local variables because the tcpiphdr 409 * gets munged while we are collapsing mbufs. 410 */ 411 pkt_seq = th->th_seq; 412 pkt_len = *tlen; 413 pkt_flags = th->th_flags; 414 415 TCP_REASS_COUNTER_INCR(&tcp_reass_); 416 417 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 418 /* 419 * When we miss a packet, the vast majority of time we get 420 * packets that follow it in order. So optimize for that. 421 */ 422 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 423 p->ipqe_len += pkt_len; 424 p->ipqe_flags |= pkt_flags; 425 m_cat(p->ipre_mlast, m); 426 TRAVERSE(p->ipre_mlast); 427 m = NULL; 428 tiqe = p; 429 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 430 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 431 goto skip_replacement; 432 } 433 /* 434 * While we're here, if the pkt is completely beyond 435 * anything we have, just insert it at the tail. 436 */ 437 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 438 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 439 goto insert_it; 440 } 441 } 442 443 q = TAILQ_FIRST(&tp->segq); 444 445 if (q != NULL) { 446 /* 447 * If this segment immediately precedes the first out-of-order 448 * block, simply slap the segment in front of it and (mostly) 449 * skip the complicated logic. 450 */ 451 if (pkt_seq + pkt_len == q->ipqe_seq) { 452 q->ipqe_seq = pkt_seq; 453 q->ipqe_len += pkt_len; 454 q->ipqe_flags |= pkt_flags; 455 m_cat(m, q->ipqe_m); 456 q->ipqe_m = m; 457 q->ipre_mlast = m; /* last mbuf may have changed */ 458 TRAVERSE(q->ipre_mlast); 459 tiqe = q; 460 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 461 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 462 goto skip_replacement; 463 } 464 } else { 465 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 466 } 467 468 /* 469 * Find a segment which begins after this one does. 470 */ 471 for (p = NULL; q != NULL; q = nq) { 472 nq = TAILQ_NEXT(q, ipqe_q); 473 #ifdef TCP_REASS_COUNTERS 474 count++; 475 #endif 476 /* 477 * If the received segment is just right after this 478 * fragment, merge the two together and then check 479 * for further overlaps. 480 */ 481 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 482 #ifdef TCPREASS_DEBUG 483 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 484 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 485 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 486 #endif 487 pkt_len += q->ipqe_len; 488 pkt_flags |= q->ipqe_flags; 489 pkt_seq = q->ipqe_seq; 490 m_cat(q->ipre_mlast, m); 491 TRAVERSE(q->ipre_mlast); 492 m = q->ipqe_m; 493 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 494 goto free_ipqe; 495 } 496 /* 497 * If the received segment is completely past this 498 * fragment, we need to go the next fragment. 499 */ 500 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 501 p = q; 502 continue; 503 } 504 /* 505 * If the fragment is past the received segment, 506 * it (or any following) can't be concatenated. 507 */ 508 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 509 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 510 break; 511 } 512 513 /* 514 * We've received all the data in this segment before. 515 * mark it as a duplicate and return. 516 */ 517 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 518 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 519 tcpstat.tcps_rcvduppack++; 520 tcpstat.tcps_rcvdupbyte += pkt_len; 521 tcp_new_dsack(tp, pkt_seq, pkt_len); 522 m_freem(m); 523 if (tiqe != NULL) { 524 tcpipqent_free(tiqe); 525 } 526 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 527 return (0); 528 } 529 /* 530 * Received segment completely overlaps this fragment 531 * so we drop the fragment (this keeps the temporal 532 * ordering of segments correct). 533 */ 534 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 535 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 536 rcvpartdupbyte += q->ipqe_len; 537 m_freem(q->ipqe_m); 538 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 539 goto free_ipqe; 540 } 541 /* 542 * RX'ed segment extends past the end of the 543 * fragment. Drop the overlapping bytes. Then 544 * merge the fragment and segment then treat as 545 * a longer received packet. 546 */ 547 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 548 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 549 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 550 #ifdef TCPREASS_DEBUG 551 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 552 tp, overlap, 553 pkt_seq, pkt_seq + pkt_len, pkt_len); 554 #endif 555 m_adj(m, overlap); 556 rcvpartdupbyte += overlap; 557 m_cat(q->ipre_mlast, m); 558 TRAVERSE(q->ipre_mlast); 559 m = q->ipqe_m; 560 pkt_seq = q->ipqe_seq; 561 pkt_len += q->ipqe_len - overlap; 562 rcvoobyte -= overlap; 563 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 564 goto free_ipqe; 565 } 566 /* 567 * RX'ed segment extends past the front of the 568 * fragment. Drop the overlapping bytes on the 569 * received packet. The packet will then be 570 * contatentated with this fragment a bit later. 571 */ 572 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 573 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 574 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 575 #ifdef TCPREASS_DEBUG 576 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 577 tp, overlap, 578 pkt_seq, pkt_seq + pkt_len, pkt_len); 579 #endif 580 m_adj(m, -overlap); 581 pkt_len -= overlap; 582 rcvpartdupbyte += overlap; 583 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 584 rcvoobyte -= overlap; 585 } 586 /* 587 * If the received segment immediates precedes this 588 * fragment then tack the fragment onto this segment 589 * and reinsert the data. 590 */ 591 if (q->ipqe_seq == pkt_seq + pkt_len) { 592 #ifdef TCPREASS_DEBUG 593 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 594 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 595 pkt_seq, pkt_seq + pkt_len, pkt_len); 596 #endif 597 pkt_len += q->ipqe_len; 598 pkt_flags |= q->ipqe_flags; 599 m_cat(m, q->ipqe_m); 600 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 601 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 602 tp->t_segqlen--; 603 KASSERT(tp->t_segqlen >= 0); 604 KASSERT(tp->t_segqlen != 0 || 605 (TAILQ_EMPTY(&tp->segq) && 606 TAILQ_EMPTY(&tp->timeq))); 607 if (tiqe == NULL) { 608 tiqe = q; 609 } else { 610 tcpipqent_free(q); 611 } 612 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 613 break; 614 } 615 /* 616 * If the fragment is before the segment, remember it. 617 * When this loop is terminated, p will contain the 618 * pointer to fragment that is right before the received 619 * segment. 620 */ 621 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 622 p = q; 623 624 continue; 625 626 /* 627 * This is a common operation. It also will allow 628 * to save doing a malloc/free in most instances. 629 */ 630 free_ipqe: 631 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 632 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 633 tp->t_segqlen--; 634 KASSERT(tp->t_segqlen >= 0); 635 KASSERT(tp->t_segqlen != 0 || 636 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 637 if (tiqe == NULL) { 638 tiqe = q; 639 } else { 640 tcpipqent_free(q); 641 } 642 } 643 644 #ifdef TCP_REASS_COUNTERS 645 if (count > 7) 646 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 647 else if (count > 0) 648 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 649 #endif 650 651 insert_it: 652 653 /* 654 * Allocate a new queue entry since the received segment did not 655 * collapse onto any other out-of-order block; thus we are allocating 656 * a new block. If it had collapsed, tiqe would not be NULL and 657 * we would be reusing it. 658 * XXX If we can't, just drop the packet. XXX 659 */ 660 if (tiqe == NULL) { 661 tiqe = tcpipqent_alloc(); 662 if (tiqe == NULL) { 663 tcpstat.tcps_rcvmemdrop++; 664 m_freem(m); 665 return (0); 666 } 667 } 668 669 /* 670 * Update the counters. 671 */ 672 tcpstat.tcps_rcvoopack++; 673 tcpstat.tcps_rcvoobyte += rcvoobyte; 674 if (rcvpartdupbyte) { 675 tcpstat.tcps_rcvpartduppack++; 676 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 677 } 678 679 /* 680 * Insert the new fragment queue entry into both queues. 681 */ 682 tiqe->ipqe_m = m; 683 tiqe->ipre_mlast = m; 684 tiqe->ipqe_seq = pkt_seq; 685 tiqe->ipqe_len = pkt_len; 686 tiqe->ipqe_flags = pkt_flags; 687 if (p == NULL) { 688 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 689 #ifdef TCPREASS_DEBUG 690 if (tiqe->ipqe_seq != tp->rcv_nxt) 691 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 692 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 693 #endif 694 } else { 695 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 696 #ifdef TCPREASS_DEBUG 697 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 698 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 699 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 700 #endif 701 } 702 tp->t_segqlen++; 703 704 skip_replacement: 705 706 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 707 708 present: 709 /* 710 * Present data to user, advancing rcv_nxt through 711 * completed sequence space. 712 */ 713 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 714 return (0); 715 q = TAILQ_FIRST(&tp->segq); 716 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 717 return (0); 718 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 719 return (0); 720 721 tp->rcv_nxt += q->ipqe_len; 722 pkt_flags = q->ipqe_flags & TH_FIN; 723 ND6_HINT(tp); 724 725 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 726 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 727 tp->t_segqlen--; 728 KASSERT(tp->t_segqlen >= 0); 729 KASSERT(tp->t_segqlen != 0 || 730 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 731 if (so->so_state & SS_CANTRCVMORE) 732 m_freem(q->ipqe_m); 733 else 734 sbappendstream(&so->so_rcv, q->ipqe_m); 735 tcpipqent_free(q); 736 sorwakeup(so); 737 return (pkt_flags); 738 } 739 740 #ifdef INET6 741 int 742 tcp6_input(struct mbuf **mp, int *offp, int proto) 743 { 744 struct mbuf *m = *mp; 745 746 /* 747 * draft-itojun-ipv6-tcp-to-anycast 748 * better place to put this in? 749 */ 750 if (m->m_flags & M_ANYCAST6) { 751 struct ip6_hdr *ip6; 752 if (m->m_len < sizeof(struct ip6_hdr)) { 753 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 754 tcpstat.tcps_rcvshort++; 755 return IPPROTO_DONE; 756 } 757 } 758 ip6 = mtod(m, struct ip6_hdr *); 759 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 760 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 761 return IPPROTO_DONE; 762 } 763 764 tcp_input(m, *offp, proto); 765 return IPPROTO_DONE; 766 } 767 #endif 768 769 #ifdef INET 770 static void 771 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 772 { 773 char src[4*sizeof "123"]; 774 char dst[4*sizeof "123"]; 775 776 if (ip) { 777 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 778 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 779 } 780 else { 781 strlcpy(src, "(unknown)", sizeof(src)); 782 strlcpy(dst, "(unknown)", sizeof(dst)); 783 } 784 log(LOG_INFO, 785 "Connection attempt to TCP %s:%d from %s:%d\n", 786 dst, ntohs(th->th_dport), 787 src, ntohs(th->th_sport)); 788 } 789 #endif 790 791 #ifdef INET6 792 static void 793 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 794 { 795 char src[INET6_ADDRSTRLEN]; 796 char dst[INET6_ADDRSTRLEN]; 797 798 if (ip6) { 799 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 800 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 801 } 802 else { 803 strlcpy(src, "(unknown v6)", sizeof(src)); 804 strlcpy(dst, "(unknown v6)", sizeof(dst)); 805 } 806 log(LOG_INFO, 807 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 808 dst, ntohs(th->th_dport), 809 src, ntohs(th->th_sport)); 810 } 811 #endif 812 813 /* 814 * Checksum extended TCP header and data. 815 */ 816 int 817 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, int toff, 818 int off, int tlen) 819 { 820 821 /* 822 * XXX it's better to record and check if this mbuf is 823 * already checked. 824 */ 825 826 switch (af) { 827 #ifdef INET 828 case AF_INET: 829 switch (m->m_pkthdr.csum_flags & 830 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 831 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 832 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 833 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 834 goto badcsum; 835 836 case M_CSUM_TCPv4|M_CSUM_DATA: { 837 u_int32_t hw_csum = m->m_pkthdr.csum_data; 838 839 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 840 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 841 const struct ip *ip = 842 mtod(m, const struct ip *); 843 844 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 845 ip->ip_dst.s_addr, 846 htons(hw_csum + tlen + off + IPPROTO_TCP)); 847 } 848 if ((hw_csum ^ 0xffff) != 0) 849 goto badcsum; 850 break; 851 } 852 853 case M_CSUM_TCPv4: 854 /* Checksum was okay. */ 855 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 856 break; 857 858 default: 859 /* 860 * Must compute it ourselves. Maybe skip checksum 861 * on loopback interfaces. 862 */ 863 if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & 864 IFF_LOOPBACK) || 865 tcp_do_loopback_cksum)) { 866 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 867 if (in4_cksum(m, IPPROTO_TCP, toff, 868 tlen + off) != 0) 869 goto badcsum; 870 } 871 break; 872 } 873 break; 874 #endif /* INET4 */ 875 876 #ifdef INET6 877 case AF_INET6: 878 if (__predict_true((m->m_flags & M_LOOP) == 0 || 879 tcp_do_loopback_cksum)) { 880 if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 881 goto badcsum; 882 } 883 break; 884 #endif /* INET6 */ 885 } 886 887 return 0; 888 889 badcsum: 890 tcpstat.tcps_rcvbadsum++; 891 return -1; 892 } 893 894 /* 895 * TCP input routine, follows pages 65-76 of the 896 * protocol specification dated September, 1981 very closely. 897 */ 898 void 899 tcp_input(struct mbuf *m, ...) 900 { 901 struct tcphdr *th; 902 struct ip *ip; 903 struct inpcb *inp; 904 #ifdef INET6 905 struct ip6_hdr *ip6; 906 struct in6pcb *in6p; 907 #endif 908 u_int8_t *optp = NULL; 909 int optlen = 0; 910 int len, tlen, toff, hdroptlen = 0; 911 struct tcpcb *tp = 0; 912 int tiflags; 913 struct socket *so = NULL; 914 int todrop, dupseg, acked, ourfinisacked, needoutput = 0; 915 #ifdef TCP_DEBUG 916 short ostate = 0; 917 #endif 918 int iss = 0; 919 u_long tiwin; 920 struct tcp_opt_info opti; 921 int off, iphlen; 922 va_list ap; 923 int af; /* af on the wire */ 924 struct mbuf *tcp_saveti = NULL; 925 926 MCLAIM(m, &tcp_rx_mowner); 927 va_start(ap, m); 928 toff = va_arg(ap, int); 929 (void)va_arg(ap, int); /* ignore value, advance ap */ 930 va_end(ap); 931 932 tcpstat.tcps_rcvtotal++; 933 934 bzero(&opti, sizeof(opti)); 935 opti.ts_present = 0; 936 opti.maxseg = 0; 937 938 /* 939 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 940 * 941 * TCP is, by definition, unicast, so we reject all 942 * multicast outright. 943 * 944 * Note, there are additional src/dst address checks in 945 * the AF-specific code below. 946 */ 947 if (m->m_flags & (M_BCAST|M_MCAST)) { 948 /* XXX stat */ 949 goto drop; 950 } 951 #ifdef INET6 952 if (m->m_flags & M_ANYCAST6) { 953 /* XXX stat */ 954 goto drop; 955 } 956 #endif 957 958 /* 959 * Get IP and TCP header. 960 * Note: IP leaves IP header in first mbuf. 961 */ 962 ip = mtod(m, struct ip *); 963 #ifdef INET6 964 ip6 = NULL; 965 #endif 966 switch (ip->ip_v) { 967 #ifdef INET 968 case 4: 969 af = AF_INET; 970 iphlen = sizeof(struct ip); 971 ip = mtod(m, struct ip *); 972 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 973 sizeof(struct tcphdr)); 974 if (th == NULL) { 975 tcpstat.tcps_rcvshort++; 976 return; 977 } 978 /* We do the checksum after PCB lookup... */ 979 len = ntohs(ip->ip_len); 980 tlen = len - toff; 981 break; 982 #endif 983 #ifdef INET6 984 case 6: 985 ip = NULL; 986 iphlen = sizeof(struct ip6_hdr); 987 af = AF_INET6; 988 ip6 = mtod(m, struct ip6_hdr *); 989 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 990 sizeof(struct tcphdr)); 991 if (th == NULL) { 992 tcpstat.tcps_rcvshort++; 993 return; 994 } 995 996 /* Be proactive about malicious use of IPv4 mapped address */ 997 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 998 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 999 /* XXX stat */ 1000 goto drop; 1001 } 1002 1003 /* 1004 * Be proactive about unspecified IPv6 address in source. 1005 * As we use all-zero to indicate unbounded/unconnected pcb, 1006 * unspecified IPv6 address can be used to confuse us. 1007 * 1008 * Note that packets with unspecified IPv6 destination is 1009 * already dropped in ip6_input. 1010 */ 1011 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1012 /* XXX stat */ 1013 goto drop; 1014 } 1015 1016 /* 1017 * Make sure destination address is not multicast. 1018 * Source address checked in ip6_input(). 1019 */ 1020 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1021 /* XXX stat */ 1022 goto drop; 1023 } 1024 1025 /* We do the checksum after PCB lookup... */ 1026 len = m->m_pkthdr.len; 1027 tlen = len - toff; 1028 break; 1029 #endif 1030 default: 1031 m_freem(m); 1032 return; 1033 } 1034 1035 KASSERT(TCP_HDR_ALIGNED_P(th)); 1036 1037 /* 1038 * Check that TCP offset makes sense, 1039 * pull out TCP options and adjust length. XXX 1040 */ 1041 off = th->th_off << 2; 1042 if (off < sizeof (struct tcphdr) || off > tlen) { 1043 tcpstat.tcps_rcvbadoff++; 1044 goto drop; 1045 } 1046 tlen -= off; 1047 1048 /* 1049 * tcp_input() has been modified to use tlen to mean the TCP data 1050 * length throughout the function. Other functions can use 1051 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1052 * rja 1053 */ 1054 1055 if (off > sizeof (struct tcphdr)) { 1056 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1057 if (th == NULL) { 1058 tcpstat.tcps_rcvshort++; 1059 return; 1060 } 1061 /* 1062 * NOTE: ip/ip6 will not be affected by m_pulldown() 1063 * (as they're before toff) and we don't need to update those. 1064 */ 1065 KASSERT(TCP_HDR_ALIGNED_P(th)); 1066 optlen = off - sizeof (struct tcphdr); 1067 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1068 /* 1069 * Do quick retrieval of timestamp options ("options 1070 * prediction?"). If timestamp is the only option and it's 1071 * formatted as recommended in RFC 1323 appendix A, we 1072 * quickly get the values now and not bother calling 1073 * tcp_dooptions(), etc. 1074 */ 1075 if ((optlen == TCPOLEN_TSTAMP_APPA || 1076 (optlen > TCPOLEN_TSTAMP_APPA && 1077 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1078 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1079 (th->th_flags & TH_SYN) == 0) { 1080 opti.ts_present = 1; 1081 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1082 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1083 optp = NULL; /* we've parsed the options */ 1084 } 1085 } 1086 tiflags = th->th_flags; 1087 1088 /* 1089 * Locate pcb for segment. 1090 */ 1091 findpcb: 1092 inp = NULL; 1093 #ifdef INET6 1094 in6p = NULL; 1095 #endif 1096 switch (af) { 1097 #ifdef INET 1098 case AF_INET: 1099 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1100 ip->ip_dst, th->th_dport); 1101 if (inp == 0) { 1102 ++tcpstat.tcps_pcbhashmiss; 1103 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1104 } 1105 #ifdef INET6 1106 if (inp == 0) { 1107 struct in6_addr s, d; 1108 1109 /* mapped addr case */ 1110 bzero(&s, sizeof(s)); 1111 s.s6_addr16[5] = htons(0xffff); 1112 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1113 bzero(&d, sizeof(d)); 1114 d.s6_addr16[5] = htons(0xffff); 1115 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1116 in6p = in6_pcblookup_connect(&tcbtable, &s, 1117 th->th_sport, &d, th->th_dport, 0); 1118 if (in6p == 0) { 1119 ++tcpstat.tcps_pcbhashmiss; 1120 in6p = in6_pcblookup_bind(&tcbtable, &d, 1121 th->th_dport, 0); 1122 } 1123 } 1124 #endif 1125 #ifndef INET6 1126 if (inp == 0) 1127 #else 1128 if (inp == 0 && in6p == 0) 1129 #endif 1130 { 1131 ++tcpstat.tcps_noport; 1132 if (tcp_log_refused && 1133 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1134 tcp4_log_refused(ip, th); 1135 } 1136 TCP_FIELDS_TO_HOST(th); 1137 goto dropwithreset_ratelim; 1138 } 1139 #if defined(IPSEC) || defined(FAST_IPSEC) 1140 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1141 ipsec4_in_reject(m, inp)) { 1142 ipsecstat.in_polvio++; 1143 goto drop; 1144 } 1145 #ifdef INET6 1146 else if (in6p && 1147 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1148 ipsec4_in_reject_so(m, in6p->in6p_socket)) { 1149 ipsecstat.in_polvio++; 1150 goto drop; 1151 } 1152 #endif 1153 #endif /*IPSEC*/ 1154 break; 1155 #endif /*INET*/ 1156 #ifdef INET6 1157 case AF_INET6: 1158 { 1159 int faith; 1160 1161 #if defined(NFAITH) && NFAITH > 0 1162 faith = faithprefix(&ip6->ip6_dst); 1163 #else 1164 faith = 0; 1165 #endif 1166 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1167 th->th_sport, &ip6->ip6_dst, th->th_dport, faith); 1168 if (in6p == NULL) { 1169 ++tcpstat.tcps_pcbhashmiss; 1170 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1171 th->th_dport, faith); 1172 } 1173 if (in6p == NULL) { 1174 ++tcpstat.tcps_noport; 1175 if (tcp_log_refused && 1176 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1177 tcp6_log_refused(ip6, th); 1178 } 1179 TCP_FIELDS_TO_HOST(th); 1180 goto dropwithreset_ratelim; 1181 } 1182 #if defined(IPSEC) || defined(FAST_IPSEC) 1183 if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1184 ipsec6_in_reject(m, in6p)) { 1185 ipsec6stat.in_polvio++; 1186 goto drop; 1187 } 1188 #endif /*IPSEC*/ 1189 break; 1190 } 1191 #endif 1192 } 1193 1194 /* 1195 * If the state is CLOSED (i.e., TCB does not exist) then 1196 * all data in the incoming segment is discarded. 1197 * If the TCB exists but is in CLOSED state, it is embryonic, 1198 * but should either do a listen or a connect soon. 1199 */ 1200 tp = NULL; 1201 so = NULL; 1202 if (inp) { 1203 tp = intotcpcb(inp); 1204 so = inp->inp_socket; 1205 } 1206 #ifdef INET6 1207 else if (in6p) { 1208 tp = in6totcpcb(in6p); 1209 so = in6p->in6p_socket; 1210 } 1211 #endif 1212 if (tp == 0) { 1213 TCP_FIELDS_TO_HOST(th); 1214 goto dropwithreset_ratelim; 1215 } 1216 if (tp->t_state == TCPS_CLOSED) 1217 goto drop; 1218 1219 /* 1220 * Checksum extended TCP header and data. 1221 */ 1222 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1223 goto badcsum; 1224 1225 TCP_FIELDS_TO_HOST(th); 1226 1227 /* Unscale the window into a 32-bit value. */ 1228 if ((tiflags & TH_SYN) == 0) 1229 tiwin = th->th_win << tp->snd_scale; 1230 else 1231 tiwin = th->th_win; 1232 1233 #ifdef INET6 1234 /* save packet options if user wanted */ 1235 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1236 if (in6p->in6p_options) { 1237 m_freem(in6p->in6p_options); 1238 in6p->in6p_options = 0; 1239 } 1240 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1241 } 1242 #endif 1243 1244 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1245 union syn_cache_sa src; 1246 union syn_cache_sa dst; 1247 1248 bzero(&src, sizeof(src)); 1249 bzero(&dst, sizeof(dst)); 1250 switch (af) { 1251 #ifdef INET 1252 case AF_INET: 1253 src.sin.sin_len = sizeof(struct sockaddr_in); 1254 src.sin.sin_family = AF_INET; 1255 src.sin.sin_addr = ip->ip_src; 1256 src.sin.sin_port = th->th_sport; 1257 1258 dst.sin.sin_len = sizeof(struct sockaddr_in); 1259 dst.sin.sin_family = AF_INET; 1260 dst.sin.sin_addr = ip->ip_dst; 1261 dst.sin.sin_port = th->th_dport; 1262 break; 1263 #endif 1264 #ifdef INET6 1265 case AF_INET6: 1266 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1267 src.sin6.sin6_family = AF_INET6; 1268 src.sin6.sin6_addr = ip6->ip6_src; 1269 src.sin6.sin6_port = th->th_sport; 1270 1271 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1272 dst.sin6.sin6_family = AF_INET6; 1273 dst.sin6.sin6_addr = ip6->ip6_dst; 1274 dst.sin6.sin6_port = th->th_dport; 1275 break; 1276 #endif /* INET6 */ 1277 default: 1278 goto badsyn; /*sanity*/ 1279 } 1280 1281 if (so->so_options & SO_DEBUG) { 1282 #ifdef TCP_DEBUG 1283 ostate = tp->t_state; 1284 #endif 1285 1286 tcp_saveti = NULL; 1287 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1288 goto nosave; 1289 1290 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1291 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1292 if (!tcp_saveti) 1293 goto nosave; 1294 } else { 1295 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1296 if (!tcp_saveti) 1297 goto nosave; 1298 MCLAIM(m, &tcp_mowner); 1299 tcp_saveti->m_len = iphlen; 1300 m_copydata(m, 0, iphlen, 1301 mtod(tcp_saveti, caddr_t)); 1302 } 1303 1304 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1305 m_freem(tcp_saveti); 1306 tcp_saveti = NULL; 1307 } else { 1308 tcp_saveti->m_len += sizeof(struct tcphdr); 1309 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1310 sizeof(struct tcphdr)); 1311 } 1312 nosave:; 1313 } 1314 if (so->so_options & SO_ACCEPTCONN) { 1315 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1316 if (tiflags & TH_RST) { 1317 syn_cache_reset(&src.sa, &dst.sa, th); 1318 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1319 (TH_ACK|TH_SYN)) { 1320 /* 1321 * Received a SYN,ACK. This should 1322 * never happen while we are in 1323 * LISTEN. Send an RST. 1324 */ 1325 goto badsyn; 1326 } else if (tiflags & TH_ACK) { 1327 so = syn_cache_get(&src.sa, &dst.sa, 1328 th, toff, tlen, so, m); 1329 if (so == NULL) { 1330 /* 1331 * We don't have a SYN for 1332 * this ACK; send an RST. 1333 */ 1334 goto badsyn; 1335 } else if (so == 1336 (struct socket *)(-1)) { 1337 /* 1338 * We were unable to create 1339 * the connection. If the 1340 * 3-way handshake was 1341 * completed, and RST has 1342 * been sent to the peer. 1343 * Since the mbuf might be 1344 * in use for the reply, 1345 * do not free it. 1346 */ 1347 m = NULL; 1348 } else { 1349 /* 1350 * We have created a 1351 * full-blown connection. 1352 */ 1353 tp = NULL; 1354 inp = NULL; 1355 #ifdef INET6 1356 in6p = NULL; 1357 #endif 1358 switch (so->so_proto->pr_domain->dom_family) { 1359 #ifdef INET 1360 case AF_INET: 1361 inp = sotoinpcb(so); 1362 tp = intotcpcb(inp); 1363 break; 1364 #endif 1365 #ifdef INET6 1366 case AF_INET6: 1367 in6p = sotoin6pcb(so); 1368 tp = in6totcpcb(in6p); 1369 break; 1370 #endif 1371 } 1372 if (tp == NULL) 1373 goto badsyn; /*XXX*/ 1374 tiwin <<= tp->snd_scale; 1375 goto after_listen; 1376 } 1377 } else { 1378 /* 1379 * None of RST, SYN or ACK was set. 1380 * This is an invalid packet for a 1381 * TCB in LISTEN state. Send a RST. 1382 */ 1383 goto badsyn; 1384 } 1385 } else { 1386 /* 1387 * Received a SYN. 1388 */ 1389 1390 #ifdef INET6 1391 /* 1392 * If deprecated address is forbidden, we do 1393 * not accept SYN to deprecated interface 1394 * address to prevent any new inbound 1395 * connection from getting established. 1396 * When we do not accept SYN, we send a TCP 1397 * RST, with deprecated source address (instead 1398 * of dropping it). We compromise it as it is 1399 * much better for peer to send a RST, and 1400 * RST will be the final packet for the 1401 * exchange. 1402 * 1403 * If we do not forbid deprecated addresses, we 1404 * accept the SYN packet. RFC2462 does not 1405 * suggest dropping SYN in this case. 1406 * If we decipher RFC2462 5.5.4, it says like 1407 * this: 1408 * 1. use of deprecated addr with existing 1409 * communication is okay - "SHOULD continue 1410 * to be used" 1411 * 2. use of it with new communication: 1412 * (2a) "SHOULD NOT be used if alternate 1413 * address with sufficient scope is 1414 * available" 1415 * (2b) nothing mentioned otherwise. 1416 * Here we fall into (2b) case as we have no 1417 * choice in our source address selection - we 1418 * must obey the peer. 1419 * 1420 * The wording in RFC2462 is confusing, and 1421 * there are multiple description text for 1422 * deprecated address handling - worse, they 1423 * are not exactly the same. I believe 5.5.4 1424 * is the best one, so we follow 5.5.4. 1425 */ 1426 if (af == AF_INET6 && !ip6_use_deprecated) { 1427 struct in6_ifaddr *ia6; 1428 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1429 &ip6->ip6_dst)) && 1430 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1431 tp = NULL; 1432 goto dropwithreset; 1433 } 1434 } 1435 #endif 1436 1437 #ifdef IPSEC 1438 switch (af) { 1439 #ifdef INET 1440 case AF_INET: 1441 if (ipsec4_in_reject_so(m, so)) { 1442 ipsecstat.in_polvio++; 1443 tp = NULL; 1444 goto dropwithreset; 1445 } 1446 break; 1447 #endif 1448 #ifdef INET6 1449 case AF_INET6: 1450 if (ipsec6_in_reject_so(m, so)) { 1451 ipsec6stat.in_polvio++; 1452 tp = NULL; 1453 goto dropwithreset; 1454 } 1455 break; 1456 #endif 1457 } 1458 #endif 1459 1460 /* 1461 * LISTEN socket received a SYN 1462 * from itself? This can't possibly 1463 * be valid; drop the packet. 1464 */ 1465 if (th->th_sport == th->th_dport) { 1466 int i; 1467 1468 switch (af) { 1469 #ifdef INET 1470 case AF_INET: 1471 i = in_hosteq(ip->ip_src, ip->ip_dst); 1472 break; 1473 #endif 1474 #ifdef INET6 1475 case AF_INET6: 1476 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1477 break; 1478 #endif 1479 default: 1480 i = 1; 1481 } 1482 if (i) { 1483 tcpstat.tcps_badsyn++; 1484 goto drop; 1485 } 1486 } 1487 1488 /* 1489 * SYN looks ok; create compressed TCP 1490 * state for it. 1491 */ 1492 if (so->so_qlen <= so->so_qlimit && 1493 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1494 so, m, optp, optlen, &opti)) 1495 m = NULL; 1496 } 1497 goto drop; 1498 } 1499 } 1500 1501 after_listen: 1502 #ifdef DIAGNOSTIC 1503 /* 1504 * Should not happen now that all embryonic connections 1505 * are handled with compressed state. 1506 */ 1507 if (tp->t_state == TCPS_LISTEN) 1508 panic("tcp_input: TCPS_LISTEN"); 1509 #endif 1510 1511 /* 1512 * Segment received on connection. 1513 * Reset idle time and keep-alive timer. 1514 */ 1515 tp->t_rcvtime = tcp_now; 1516 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1517 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1518 1519 /* 1520 * Process options. 1521 */ 1522 #ifdef TCP_SIGNATURE 1523 if (optp || (tp->t_flags & TF_SIGNATURE)) 1524 #else 1525 if (optp) 1526 #endif 1527 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1528 goto drop; 1529 1530 if (TCP_SACK_ENABLED(tp)) { 1531 tcp_del_sackholes(tp, th); 1532 } 1533 1534 if (opti.ts_present && opti.ts_ecr) { 1535 /* 1536 * Calculate the RTT from the returned time stamp and the 1537 * connection's time base. If the time stamp is later than 1538 * the current time, or is extremely old, fall back to non-1323 1539 * RTT calculation. Since ts_ecr is unsigned, we can test both 1540 * at the same time. 1541 */ 1542 opti.ts_ecr = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1543 if (opti.ts_ecr > TCP_PAWS_IDLE) 1544 opti.ts_ecr = 0; 1545 } 1546 1547 /* 1548 * Header prediction: check for the two common cases 1549 * of a uni-directional data xfer. If the packet has 1550 * no control flags, is in-sequence, the window didn't 1551 * change and we're not retransmitting, it's a 1552 * candidate. If the length is zero and the ack moved 1553 * forward, we're the sender side of the xfer. Just 1554 * free the data acked & wake any higher level process 1555 * that was blocked waiting for space. If the length 1556 * is non-zero and the ack didn't move, we're the 1557 * receiver side. If we're getting packets in-order 1558 * (the reassembly queue is empty), add the data to 1559 * the socket buffer and note that we need a delayed ack. 1560 */ 1561 if (tp->t_state == TCPS_ESTABLISHED && 1562 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1563 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1564 th->th_seq == tp->rcv_nxt && 1565 tiwin && tiwin == tp->snd_wnd && 1566 tp->snd_nxt == tp->snd_max) { 1567 1568 /* 1569 * If last ACK falls within this segment's sequence numbers, 1570 * record the timestamp. 1571 */ 1572 if (opti.ts_present && 1573 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1574 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1575 tp->ts_recent_age = tcp_now; 1576 tp->ts_recent = opti.ts_val; 1577 } 1578 1579 if (tlen == 0) { 1580 /* Ack prediction. */ 1581 if (SEQ_GT(th->th_ack, tp->snd_una) && 1582 SEQ_LEQ(th->th_ack, tp->snd_max) && 1583 tp->snd_cwnd >= tp->snd_wnd && 1584 tp->t_partialacks < 0) { 1585 /* 1586 * this is a pure ack for outstanding data. 1587 */ 1588 ++tcpstat.tcps_predack; 1589 if (opti.ts_present && opti.ts_ecr) 1590 tcp_xmit_timer(tp, opti.ts_ecr); 1591 else if (tp->t_rtttime && 1592 SEQ_GT(th->th_ack, tp->t_rtseq)) 1593 tcp_xmit_timer(tp, 1594 tcp_now - tp->t_rtttime); 1595 acked = th->th_ack - tp->snd_una; 1596 tcpstat.tcps_rcvackpack++; 1597 tcpstat.tcps_rcvackbyte += acked; 1598 ND6_HINT(tp); 1599 1600 if (acked > (tp->t_lastoff - tp->t_inoff)) 1601 tp->t_lastm = NULL; 1602 sbdrop(&so->so_snd, acked); 1603 tp->t_lastoff -= acked; 1604 1605 tp->snd_una = th->th_ack; 1606 tp->snd_fack = tp->snd_una; 1607 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1608 tp->snd_high = tp->snd_una; 1609 m_freem(m); 1610 1611 /* 1612 * If all outstanding data are acked, stop 1613 * retransmit timer, otherwise restart timer 1614 * using current (possibly backed-off) value. 1615 * If process is waiting for space, 1616 * wakeup/selwakeup/signal. If data 1617 * are ready to send, let tcp_output 1618 * decide between more output or persist. 1619 */ 1620 if (tp->snd_una == tp->snd_max) 1621 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1622 else if (TCP_TIMER_ISARMED(tp, 1623 TCPT_PERSIST) == 0) 1624 TCP_TIMER_ARM(tp, TCPT_REXMT, 1625 tp->t_rxtcur); 1626 1627 sowwakeup(so); 1628 if (so->so_snd.sb_cc) 1629 (void) tcp_output(tp); 1630 if (tcp_saveti) 1631 m_freem(tcp_saveti); 1632 return; 1633 } 1634 } else if (th->th_ack == tp->snd_una && 1635 TAILQ_FIRST(&tp->segq) == NULL && 1636 tlen <= sbspace(&so->so_rcv)) { 1637 /* 1638 * this is a pure, in-sequence data packet 1639 * with nothing on the reassembly queue and 1640 * we have enough buffer space to take it. 1641 */ 1642 ++tcpstat.tcps_preddat; 1643 tp->rcv_nxt += tlen; 1644 tcpstat.tcps_rcvpack++; 1645 tcpstat.tcps_rcvbyte += tlen; 1646 ND6_HINT(tp); 1647 /* 1648 * Drop TCP, IP headers and TCP options then add data 1649 * to socket buffer. 1650 */ 1651 if (so->so_state & SS_CANTRCVMORE) 1652 m_freem(m); 1653 else { 1654 m_adj(m, toff + off); 1655 sbappendstream(&so->so_rcv, m); 1656 } 1657 sorwakeup(so); 1658 TCP_SETUP_ACK(tp, th); 1659 if (tp->t_flags & TF_ACKNOW) 1660 (void) tcp_output(tp); 1661 if (tcp_saveti) 1662 m_freem(tcp_saveti); 1663 return; 1664 } 1665 } 1666 1667 /* 1668 * Compute mbuf offset to TCP data segment. 1669 */ 1670 hdroptlen = toff + off; 1671 1672 /* 1673 * Calculate amount of space in receive window, 1674 * and then do TCP input processing. 1675 * Receive window is amount of space in rcv queue, 1676 * but not less than advertised window. 1677 */ 1678 { int win; 1679 1680 win = sbspace(&so->so_rcv); 1681 if (win < 0) 1682 win = 0; 1683 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1684 } 1685 1686 switch (tp->t_state) { 1687 case TCPS_LISTEN: 1688 /* 1689 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1690 */ 1691 if (m->m_flags & (M_BCAST|M_MCAST)) 1692 goto drop; 1693 switch (af) { 1694 #ifdef INET6 1695 case AF_INET6: 1696 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1697 goto drop; 1698 break; 1699 #endif /* INET6 */ 1700 case AF_INET: 1701 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1702 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1703 goto drop; 1704 break; 1705 } 1706 break; 1707 1708 /* 1709 * If the state is SYN_SENT: 1710 * if seg contains an ACK, but not for our SYN, drop the input. 1711 * if seg contains a RST, then drop the connection. 1712 * if seg does not contain SYN, then drop it. 1713 * Otherwise this is an acceptable SYN segment 1714 * initialize tp->rcv_nxt and tp->irs 1715 * if seg contains ack then advance tp->snd_una 1716 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1717 * arrange for segment to be acked (eventually) 1718 * continue processing rest of data/controls, beginning with URG 1719 */ 1720 case TCPS_SYN_SENT: 1721 if ((tiflags & TH_ACK) && 1722 (SEQ_LEQ(th->th_ack, tp->iss) || 1723 SEQ_GT(th->th_ack, tp->snd_max))) 1724 goto dropwithreset; 1725 if (tiflags & TH_RST) { 1726 if (tiflags & TH_ACK) 1727 tp = tcp_drop(tp, ECONNREFUSED); 1728 goto drop; 1729 } 1730 if ((tiflags & TH_SYN) == 0) 1731 goto drop; 1732 if (tiflags & TH_ACK) { 1733 tp->snd_una = th->th_ack; 1734 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1735 tp->snd_nxt = tp->snd_una; 1736 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1737 tp->snd_high = tp->snd_una; 1738 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1739 } 1740 tp->irs = th->th_seq; 1741 tcp_rcvseqinit(tp); 1742 tp->t_flags |= TF_ACKNOW; 1743 tcp_mss_from_peer(tp, opti.maxseg); 1744 1745 /* 1746 * Initialize the initial congestion window. If we 1747 * had to retransmit the SYN, we must initialize cwnd 1748 * to 1 segment (i.e. the Loss Window). 1749 */ 1750 if (tp->t_flags & TF_SYN_REXMT) 1751 tp->snd_cwnd = tp->t_peermss; 1752 else { 1753 int ss = tcp_init_win; 1754 #ifdef INET 1755 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1756 ss = tcp_init_win_local; 1757 #endif 1758 #ifdef INET6 1759 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1760 ss = tcp_init_win_local; 1761 #endif 1762 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1763 } 1764 1765 tcp_rmx_rtt(tp); 1766 if (tiflags & TH_ACK) { 1767 tcpstat.tcps_connects++; 1768 soisconnected(so); 1769 tcp_established(tp); 1770 /* Do window scaling on this connection? */ 1771 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1772 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1773 tp->snd_scale = tp->requested_s_scale; 1774 tp->rcv_scale = tp->request_r_scale; 1775 } 1776 TCP_REASS_LOCK(tp); 1777 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1778 TCP_REASS_UNLOCK(tp); 1779 /* 1780 * if we didn't have to retransmit the SYN, 1781 * use its rtt as our initial srtt & rtt var. 1782 */ 1783 if (tp->t_rtttime) 1784 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1785 } else 1786 tp->t_state = TCPS_SYN_RECEIVED; 1787 1788 /* 1789 * Advance th->th_seq to correspond to first data byte. 1790 * If data, trim to stay within window, 1791 * dropping FIN if necessary. 1792 */ 1793 th->th_seq++; 1794 if (tlen > tp->rcv_wnd) { 1795 todrop = tlen - tp->rcv_wnd; 1796 m_adj(m, -todrop); 1797 tlen = tp->rcv_wnd; 1798 tiflags &= ~TH_FIN; 1799 tcpstat.tcps_rcvpackafterwin++; 1800 tcpstat.tcps_rcvbyteafterwin += todrop; 1801 } 1802 tp->snd_wl1 = th->th_seq - 1; 1803 tp->rcv_up = th->th_seq; 1804 goto step6; 1805 1806 /* 1807 * If the state is SYN_RECEIVED: 1808 * If seg contains an ACK, but not for our SYN, drop the input 1809 * and generate an RST. See page 36, rfc793 1810 */ 1811 case TCPS_SYN_RECEIVED: 1812 if ((tiflags & TH_ACK) && 1813 (SEQ_LEQ(th->th_ack, tp->iss) || 1814 SEQ_GT(th->th_ack, tp->snd_max))) 1815 goto dropwithreset; 1816 break; 1817 } 1818 1819 /* 1820 * States other than LISTEN or SYN_SENT. 1821 * First check timestamp, if present. 1822 * Then check that at least some bytes of segment are within 1823 * receive window. If segment begins before rcv_nxt, 1824 * drop leading data (and SYN); if nothing left, just ack. 1825 * 1826 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1827 * and it's less than ts_recent, drop it. 1828 */ 1829 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1830 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1831 1832 /* Check to see if ts_recent is over 24 days old. */ 1833 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 1834 /* 1835 * Invalidate ts_recent. If this segment updates 1836 * ts_recent, the age will be reset later and ts_recent 1837 * will get a valid value. If it does not, setting 1838 * ts_recent to zero will at least satisfy the 1839 * requirement that zero be placed in the timestamp 1840 * echo reply when ts_recent isn't valid. The 1841 * age isn't reset until we get a valid ts_recent 1842 * because we don't want out-of-order segments to be 1843 * dropped when ts_recent is old. 1844 */ 1845 tp->ts_recent = 0; 1846 } else { 1847 tcpstat.tcps_rcvduppack++; 1848 tcpstat.tcps_rcvdupbyte += tlen; 1849 tcpstat.tcps_pawsdrop++; 1850 tcp_new_dsack(tp, th->th_seq, tlen); 1851 goto dropafterack; 1852 } 1853 } 1854 1855 todrop = tp->rcv_nxt - th->th_seq; 1856 dupseg = FALSE; 1857 if (todrop > 0) { 1858 if (tiflags & TH_SYN) { 1859 tiflags &= ~TH_SYN; 1860 th->th_seq++; 1861 if (th->th_urp > 1) 1862 th->th_urp--; 1863 else { 1864 tiflags &= ~TH_URG; 1865 th->th_urp = 0; 1866 } 1867 todrop--; 1868 } 1869 if (todrop > tlen || 1870 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1871 /* 1872 * Any valid FIN or RST must be to the left of the 1873 * window. At this point the FIN or RST must be a 1874 * duplicate or out of sequence; drop it. 1875 */ 1876 if (tiflags & TH_RST) 1877 goto drop; 1878 tiflags &= ~(TH_FIN|TH_RST); 1879 /* 1880 * Send an ACK to resynchronize and drop any data. 1881 * But keep on processing for RST or ACK. 1882 */ 1883 tp->t_flags |= TF_ACKNOW; 1884 todrop = tlen; 1885 dupseg = TRUE; 1886 tcpstat.tcps_rcvdupbyte += todrop; 1887 tcpstat.tcps_rcvduppack++; 1888 } else if ((tiflags & TH_RST) && 1889 th->th_seq != tp->last_ack_sent) { 1890 /* 1891 * Test for reset before adjusting the sequence 1892 * number for overlapping data. 1893 */ 1894 goto dropafterack_ratelim; 1895 } else { 1896 tcpstat.tcps_rcvpartduppack++; 1897 tcpstat.tcps_rcvpartdupbyte += todrop; 1898 } 1899 tcp_new_dsack(tp, th->th_seq, todrop); 1900 hdroptlen += todrop; /*drop from head afterwards*/ 1901 th->th_seq += todrop; 1902 tlen -= todrop; 1903 if (th->th_urp > todrop) 1904 th->th_urp -= todrop; 1905 else { 1906 tiflags &= ~TH_URG; 1907 th->th_urp = 0; 1908 } 1909 } 1910 1911 /* 1912 * If new data are received on a connection after the 1913 * user processes are gone, then RST the other end. 1914 */ 1915 if ((so->so_state & SS_NOFDREF) && 1916 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1917 tp = tcp_close(tp); 1918 tcpstat.tcps_rcvafterclose++; 1919 goto dropwithreset; 1920 } 1921 1922 /* 1923 * If segment ends after window, drop trailing data 1924 * (and PUSH and FIN); if nothing left, just ACK. 1925 */ 1926 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1927 if (todrop > 0) { 1928 tcpstat.tcps_rcvpackafterwin++; 1929 if (todrop >= tlen) { 1930 /* 1931 * The segment actually starts after the window. 1932 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 1933 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 1934 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 1935 */ 1936 tcpstat.tcps_rcvbyteafterwin += tlen; 1937 /* 1938 * If a new connection request is received 1939 * while in TIME_WAIT, drop the old connection 1940 * and start over if the sequence numbers 1941 * are above the previous ones. 1942 * 1943 * NOTE: We will checksum the packet again, and 1944 * so we need to put the header fields back into 1945 * network order! 1946 * XXX This kind of sucks, but we don't expect 1947 * XXX this to happen very often, so maybe it 1948 * XXX doesn't matter so much. 1949 */ 1950 if (tiflags & TH_SYN && 1951 tp->t_state == TCPS_TIME_WAIT && 1952 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1953 iss = tcp_new_iss(tp, tp->snd_nxt); 1954 tp = tcp_close(tp); 1955 TCP_FIELDS_TO_NET(th); 1956 goto findpcb; 1957 } 1958 /* 1959 * If window is closed can only take segments at 1960 * window edge, and have to drop data and PUSH from 1961 * incoming segments. Continue processing, but 1962 * remember to ack. Otherwise, drop segment 1963 * and (if not RST) ack. 1964 */ 1965 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1966 tp->t_flags |= TF_ACKNOW; 1967 tcpstat.tcps_rcvwinprobe++; 1968 } else 1969 goto dropafterack; 1970 } else 1971 tcpstat.tcps_rcvbyteafterwin += todrop; 1972 m_adj(m, -todrop); 1973 tlen -= todrop; 1974 tiflags &= ~(TH_PUSH|TH_FIN); 1975 } 1976 1977 /* 1978 * If last ACK falls within this segment's sequence numbers, 1979 * and the timestamp is newer, record it. 1980 */ 1981 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1982 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1983 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1984 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1985 tp->ts_recent_age = tcp_now; 1986 tp->ts_recent = opti.ts_val; 1987 } 1988 1989 /* 1990 * If the RST bit is set examine the state: 1991 * SYN_RECEIVED STATE: 1992 * If passive open, return to LISTEN state. 1993 * If active open, inform user that connection was refused. 1994 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1995 * Inform user that connection was reset, and close tcb. 1996 * CLOSING, LAST_ACK, TIME_WAIT STATES 1997 * Close the tcb. 1998 */ 1999 if (tiflags & TH_RST) { 2000 if (th->th_seq != tp->last_ack_sent) 2001 goto dropafterack_ratelim; 2002 2003 switch (tp->t_state) { 2004 case TCPS_SYN_RECEIVED: 2005 so->so_error = ECONNREFUSED; 2006 goto close; 2007 2008 case TCPS_ESTABLISHED: 2009 case TCPS_FIN_WAIT_1: 2010 case TCPS_FIN_WAIT_2: 2011 case TCPS_CLOSE_WAIT: 2012 so->so_error = ECONNRESET; 2013 close: 2014 tp->t_state = TCPS_CLOSED; 2015 tcpstat.tcps_drops++; 2016 tp = tcp_close(tp); 2017 goto drop; 2018 2019 case TCPS_CLOSING: 2020 case TCPS_LAST_ACK: 2021 case TCPS_TIME_WAIT: 2022 tp = tcp_close(tp); 2023 goto drop; 2024 } 2025 } 2026 2027 /* 2028 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2029 * we must be in a synchronized state. RFC791 states (under RST 2030 * generation) that any unacceptable segment (an out-of-order SYN 2031 * qualifies) received in a synchronized state must elicit only an 2032 * empty acknowledgment segment ... and the connection remains in 2033 * the same state. 2034 */ 2035 if (tiflags & TH_SYN) { 2036 if (tp->rcv_nxt == th->th_seq) { 2037 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2038 TH_ACK); 2039 if (tcp_saveti) 2040 m_freem(tcp_saveti); 2041 return; 2042 } 2043 2044 goto dropafterack_ratelim; 2045 } 2046 2047 /* 2048 * If the ACK bit is off we drop the segment and return. 2049 */ 2050 if ((tiflags & TH_ACK) == 0) { 2051 if (tp->t_flags & TF_ACKNOW) 2052 goto dropafterack; 2053 else 2054 goto drop; 2055 } 2056 2057 /* 2058 * Ack processing. 2059 */ 2060 switch (tp->t_state) { 2061 2062 /* 2063 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2064 * ESTABLISHED state and continue processing, otherwise 2065 * send an RST. 2066 */ 2067 case TCPS_SYN_RECEIVED: 2068 if (SEQ_GT(tp->snd_una, th->th_ack) || 2069 SEQ_GT(th->th_ack, tp->snd_max)) 2070 goto dropwithreset; 2071 tcpstat.tcps_connects++; 2072 soisconnected(so); 2073 tcp_established(tp); 2074 /* Do window scaling? */ 2075 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2076 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2077 tp->snd_scale = tp->requested_s_scale; 2078 tp->rcv_scale = tp->request_r_scale; 2079 } 2080 TCP_REASS_LOCK(tp); 2081 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2082 TCP_REASS_UNLOCK(tp); 2083 tp->snd_wl1 = th->th_seq - 1; 2084 /* fall into ... */ 2085 2086 /* 2087 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2088 * ACKs. If the ack is in the range 2089 * tp->snd_una < th->th_ack <= tp->snd_max 2090 * then advance tp->snd_una to th->th_ack and drop 2091 * data from the retransmission queue. If this ACK reflects 2092 * more up to date window information we update our window information. 2093 */ 2094 case TCPS_ESTABLISHED: 2095 case TCPS_FIN_WAIT_1: 2096 case TCPS_FIN_WAIT_2: 2097 case TCPS_CLOSE_WAIT: 2098 case TCPS_CLOSING: 2099 case TCPS_LAST_ACK: 2100 case TCPS_TIME_WAIT: 2101 2102 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2103 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2104 tcpstat.tcps_rcvdupack++; 2105 /* 2106 * If we have outstanding data (other than 2107 * a window probe), this is a completely 2108 * duplicate ack (ie, window info didn't 2109 * change), the ack is the biggest we've 2110 * seen and we've seen exactly our rexmt 2111 * threshhold of them, assume a packet 2112 * has been dropped and retransmit it. 2113 * Kludge snd_nxt & the congestion 2114 * window so we send only this one 2115 * packet. 2116 * 2117 * We know we're losing at the current 2118 * window size so do congestion avoidance 2119 * (set ssthresh to half the current window 2120 * and pull our congestion window back to 2121 * the new ssthresh). 2122 * 2123 * Dup acks mean that packets have left the 2124 * network (they're now cached at the receiver) 2125 * so bump cwnd by the amount in the receiver 2126 * to keep a constant cwnd packets in the 2127 * network. 2128 * 2129 * If we are using TCP/SACK, then enter 2130 * Fast Recovery if the receiver SACKs 2131 * data that is tcprexmtthresh * MSS 2132 * bytes past the last ACKed segment, 2133 * irrespective of the number of DupAcks. 2134 */ 2135 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2136 th->th_ack != tp->snd_una) 2137 tp->t_dupacks = 0; 2138 else if (tp->t_partialacks < 0 && 2139 (++tp->t_dupacks == tcprexmtthresh || 2140 TCP_FACK_FASTRECOV(tp))) { 2141 tcp_seq onxt; 2142 u_int win; 2143 2144 if (tcp_do_newreno && 2145 SEQ_LT(th->th_ack, tp->snd_high)) { 2146 /* 2147 * False fast retransmit after 2148 * timeout. Do not enter fast 2149 * recovery. 2150 */ 2151 tp->t_dupacks = 0; 2152 break; 2153 } 2154 2155 onxt = tp->snd_nxt; 2156 win = min(tp->snd_wnd, tp->snd_cwnd) / 2157 2 / tp->t_segsz; 2158 if (win < 2) 2159 win = 2; 2160 tp->snd_ssthresh = win * tp->t_segsz; 2161 tp->snd_recover = tp->snd_max; 2162 tp->t_partialacks = 0; 2163 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2164 tp->t_rtttime = 0; 2165 if (TCP_SACK_ENABLED(tp)) { 2166 tp->t_dupacks = tcprexmtthresh; 2167 tp->sack_newdata = tp->snd_nxt; 2168 tp->snd_cwnd = tp->t_segsz; 2169 (void) tcp_output(tp); 2170 goto drop; 2171 } 2172 tp->snd_nxt = th->th_ack; 2173 tp->snd_cwnd = tp->t_segsz; 2174 (void) tcp_output(tp); 2175 tp->snd_cwnd = tp->snd_ssthresh + 2176 tp->t_segsz * tp->t_dupacks; 2177 if (SEQ_GT(onxt, tp->snd_nxt)) 2178 tp->snd_nxt = onxt; 2179 goto drop; 2180 } else if (tp->t_dupacks > tcprexmtthresh) { 2181 tp->snd_cwnd += tp->t_segsz; 2182 (void) tcp_output(tp); 2183 goto drop; 2184 } 2185 } else { 2186 /* 2187 * If the ack appears to be very old, only 2188 * allow data that is in-sequence. This 2189 * makes it somewhat more difficult to insert 2190 * forged data by guessing sequence numbers. 2191 * Sent an ack to try to update the send 2192 * sequence number on the other side. 2193 */ 2194 if (tlen && th->th_seq != tp->rcv_nxt && 2195 SEQ_LT(th->th_ack, 2196 tp->snd_una - tp->max_sndwnd)) 2197 goto dropafterack; 2198 } 2199 break; 2200 } 2201 /* 2202 * If the congestion window was inflated to account 2203 * for the other side's cached packets, retract it. 2204 */ 2205 if (TCP_SACK_ENABLED(tp)) 2206 tcp_sack_newack(tp, th); 2207 else if (tcp_do_newreno) 2208 tcp_newreno_newack(tp, th); 2209 else 2210 tcp_reno_newack(tp, th); 2211 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2212 tcpstat.tcps_rcvacktoomuch++; 2213 goto dropafterack; 2214 } 2215 acked = th->th_ack - tp->snd_una; 2216 tcpstat.tcps_rcvackpack++; 2217 tcpstat.tcps_rcvackbyte += acked; 2218 2219 /* 2220 * If we have a timestamp reply, update smoothed 2221 * round trip time. If no timestamp is present but 2222 * transmit timer is running and timed sequence 2223 * number was acked, update smoothed round trip time. 2224 * Since we now have an rtt measurement, cancel the 2225 * timer backoff (cf., Phil Karn's retransmit alg.). 2226 * Recompute the initial retransmit timer. 2227 */ 2228 if (opti.ts_present && opti.ts_ecr) 2229 tcp_xmit_timer(tp, opti.ts_ecr); 2230 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2231 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2232 2233 /* 2234 * If all outstanding data is acked, stop retransmit 2235 * timer and remember to restart (more output or persist). 2236 * If there is more data to be acked, restart retransmit 2237 * timer, using current (possibly backed-off) value. 2238 */ 2239 if (th->th_ack == tp->snd_max) { 2240 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2241 needoutput = 1; 2242 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2243 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2244 /* 2245 * When new data is acked, open the congestion window. 2246 * If the window gives us less than ssthresh packets 2247 * in flight, open exponentially (segsz per packet). 2248 * Otherwise open linearly: segsz per window 2249 * (segsz^2 / cwnd per packet). 2250 * 2251 * If we are still in fast recovery (meaning we are using 2252 * NewReno and we have only received partial acks), do not 2253 * inflate the window yet. 2254 */ 2255 if (tp->t_partialacks < 0) { 2256 u_int cw = tp->snd_cwnd; 2257 u_int incr = tp->t_segsz; 2258 2259 if (cw >= tp->snd_ssthresh) 2260 incr = incr * incr / cw; 2261 tp->snd_cwnd = min(cw + incr, 2262 TCP_MAXWIN << tp->snd_scale); 2263 } 2264 ND6_HINT(tp); 2265 if (acked > so->so_snd.sb_cc) { 2266 tp->snd_wnd -= so->so_snd.sb_cc; 2267 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2268 ourfinisacked = 1; 2269 } else { 2270 if (acked > (tp->t_lastoff - tp->t_inoff)) 2271 tp->t_lastm = NULL; 2272 sbdrop(&so->so_snd, acked); 2273 tp->t_lastoff -= acked; 2274 tp->snd_wnd -= acked; 2275 ourfinisacked = 0; 2276 } 2277 sowwakeup(so); 2278 tp->snd_una = th->th_ack; 2279 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2280 tp->snd_fack = tp->snd_una; 2281 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2282 tp->snd_nxt = tp->snd_una; 2283 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2284 tp->snd_high = tp->snd_una; 2285 2286 switch (tp->t_state) { 2287 2288 /* 2289 * In FIN_WAIT_1 STATE in addition to the processing 2290 * for the ESTABLISHED state if our FIN is now acknowledged 2291 * then enter FIN_WAIT_2. 2292 */ 2293 case TCPS_FIN_WAIT_1: 2294 if (ourfinisacked) { 2295 /* 2296 * If we can't receive any more 2297 * data, then closing user can proceed. 2298 * Starting the timer is contrary to the 2299 * specification, but if we don't get a FIN 2300 * we'll hang forever. 2301 */ 2302 if (so->so_state & SS_CANTRCVMORE) { 2303 soisdisconnected(so); 2304 if (tcp_maxidle > 0) 2305 TCP_TIMER_ARM(tp, TCPT_2MSL, 2306 tcp_maxidle); 2307 } 2308 tp->t_state = TCPS_FIN_WAIT_2; 2309 } 2310 break; 2311 2312 /* 2313 * In CLOSING STATE in addition to the processing for 2314 * the ESTABLISHED state if the ACK acknowledges our FIN 2315 * then enter the TIME-WAIT state, otherwise ignore 2316 * the segment. 2317 */ 2318 case TCPS_CLOSING: 2319 if (ourfinisacked) { 2320 tp->t_state = TCPS_TIME_WAIT; 2321 tcp_canceltimers(tp); 2322 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2323 soisdisconnected(so); 2324 } 2325 break; 2326 2327 /* 2328 * In LAST_ACK, we may still be waiting for data to drain 2329 * and/or to be acked, as well as for the ack of our FIN. 2330 * If our FIN is now acknowledged, delete the TCB, 2331 * enter the closed state and return. 2332 */ 2333 case TCPS_LAST_ACK: 2334 if (ourfinisacked) { 2335 tp = tcp_close(tp); 2336 goto drop; 2337 } 2338 break; 2339 2340 /* 2341 * In TIME_WAIT state the only thing that should arrive 2342 * is a retransmission of the remote FIN. Acknowledge 2343 * it and restart the finack timer. 2344 */ 2345 case TCPS_TIME_WAIT: 2346 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2347 goto dropafterack; 2348 } 2349 } 2350 2351 step6: 2352 /* 2353 * Update window information. 2354 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2355 */ 2356 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2357 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 2358 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 2359 /* keep track of pure window updates */ 2360 if (tlen == 0 && 2361 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2362 tcpstat.tcps_rcvwinupd++; 2363 tp->snd_wnd = tiwin; 2364 tp->snd_wl1 = th->th_seq; 2365 tp->snd_wl2 = th->th_ack; 2366 if (tp->snd_wnd > tp->max_sndwnd) 2367 tp->max_sndwnd = tp->snd_wnd; 2368 needoutput = 1; 2369 } 2370 2371 /* 2372 * Process segments with URG. 2373 */ 2374 if ((tiflags & TH_URG) && th->th_urp && 2375 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2376 /* 2377 * This is a kludge, but if we receive and accept 2378 * random urgent pointers, we'll crash in 2379 * soreceive. It's hard to imagine someone 2380 * actually wanting to send this much urgent data. 2381 */ 2382 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2383 th->th_urp = 0; /* XXX */ 2384 tiflags &= ~TH_URG; /* XXX */ 2385 goto dodata; /* XXX */ 2386 } 2387 /* 2388 * If this segment advances the known urgent pointer, 2389 * then mark the data stream. This should not happen 2390 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2391 * a FIN has been received from the remote side. 2392 * In these states we ignore the URG. 2393 * 2394 * According to RFC961 (Assigned Protocols), 2395 * the urgent pointer points to the last octet 2396 * of urgent data. We continue, however, 2397 * to consider it to indicate the first octet 2398 * of data past the urgent section as the original 2399 * spec states (in one of two places). 2400 */ 2401 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2402 tp->rcv_up = th->th_seq + th->th_urp; 2403 so->so_oobmark = so->so_rcv.sb_cc + 2404 (tp->rcv_up - tp->rcv_nxt) - 1; 2405 if (so->so_oobmark == 0) 2406 so->so_state |= SS_RCVATMARK; 2407 sohasoutofband(so); 2408 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2409 } 2410 /* 2411 * Remove out of band data so doesn't get presented to user. 2412 * This can happen independent of advancing the URG pointer, 2413 * but if two URG's are pending at once, some out-of-band 2414 * data may creep in... ick. 2415 */ 2416 if (th->th_urp <= (u_int16_t) tlen 2417 #ifdef SO_OOBINLINE 2418 && (so->so_options & SO_OOBINLINE) == 0 2419 #endif 2420 ) 2421 tcp_pulloutofband(so, th, m, hdroptlen); 2422 } else 2423 /* 2424 * If no out of band data is expected, 2425 * pull receive urgent pointer along 2426 * with the receive window. 2427 */ 2428 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2429 tp->rcv_up = tp->rcv_nxt; 2430 dodata: /* XXX */ 2431 2432 /* 2433 * Process the segment text, merging it into the TCP sequencing queue, 2434 * and arranging for acknowledgement of receipt if necessary. 2435 * This process logically involves adjusting tp->rcv_wnd as data 2436 * is presented to the user (this happens in tcp_usrreq.c, 2437 * case PRU_RCVD). If a FIN has already been received on this 2438 * connection then we just ignore the text. 2439 */ 2440 if ((tlen || (tiflags & TH_FIN)) && 2441 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2442 /* 2443 * Insert segment ti into reassembly queue of tcp with 2444 * control block tp. Return TH_FIN if reassembly now includes 2445 * a segment with FIN. The macro form does the common case 2446 * inline (segment is the next to be received on an 2447 * established connection, and the queue is empty), 2448 * avoiding linkage into and removal from the queue and 2449 * repetition of various conversions. 2450 * Set DELACK for segments received in order, but ack 2451 * immediately when segments are out of order 2452 * (so fast retransmit can work). 2453 */ 2454 /* NOTE: this was TCP_REASS() macro, but used only once */ 2455 TCP_REASS_LOCK(tp); 2456 if (th->th_seq == tp->rcv_nxt && 2457 TAILQ_FIRST(&tp->segq) == NULL && 2458 tp->t_state == TCPS_ESTABLISHED) { 2459 TCP_SETUP_ACK(tp, th); 2460 tp->rcv_nxt += tlen; 2461 tiflags = th->th_flags & TH_FIN; 2462 tcpstat.tcps_rcvpack++; 2463 tcpstat.tcps_rcvbyte += tlen; 2464 ND6_HINT(tp); 2465 if (so->so_state & SS_CANTRCVMORE) 2466 m_freem(m); 2467 else { 2468 m_adj(m, hdroptlen); 2469 sbappendstream(&(so)->so_rcv, m); 2470 } 2471 sorwakeup(so); 2472 } else { 2473 m_adj(m, hdroptlen); 2474 tiflags = tcp_reass(tp, th, m, &tlen); 2475 tp->t_flags |= TF_ACKNOW; 2476 } 2477 TCP_REASS_UNLOCK(tp); 2478 2479 /* 2480 * Note the amount of data that peer has sent into 2481 * our window, in order to estimate the sender's 2482 * buffer size. 2483 */ 2484 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2485 } else { 2486 m_freem(m); 2487 m = NULL; 2488 tiflags &= ~TH_FIN; 2489 } 2490 2491 /* 2492 * If FIN is received ACK the FIN and let the user know 2493 * that the connection is closing. Ignore a FIN received before 2494 * the connection is fully established. 2495 */ 2496 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2497 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2498 socantrcvmore(so); 2499 tp->t_flags |= TF_ACKNOW; 2500 tp->rcv_nxt++; 2501 } 2502 switch (tp->t_state) { 2503 2504 /* 2505 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2506 */ 2507 case TCPS_ESTABLISHED: 2508 tp->t_state = TCPS_CLOSE_WAIT; 2509 break; 2510 2511 /* 2512 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2513 * enter the CLOSING state. 2514 */ 2515 case TCPS_FIN_WAIT_1: 2516 tp->t_state = TCPS_CLOSING; 2517 break; 2518 2519 /* 2520 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2521 * starting the time-wait timer, turning off the other 2522 * standard timers. 2523 */ 2524 case TCPS_FIN_WAIT_2: 2525 tp->t_state = TCPS_TIME_WAIT; 2526 tcp_canceltimers(tp); 2527 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2528 soisdisconnected(so); 2529 break; 2530 2531 /* 2532 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2533 */ 2534 case TCPS_TIME_WAIT: 2535 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2536 break; 2537 } 2538 } 2539 #ifdef TCP_DEBUG 2540 if (so->so_options & SO_DEBUG) 2541 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2542 #endif 2543 2544 /* 2545 * Return any desired output. 2546 */ 2547 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2548 (void) tcp_output(tp); 2549 } 2550 if (tcp_saveti) 2551 m_freem(tcp_saveti); 2552 return; 2553 2554 badsyn: 2555 /* 2556 * Received a bad SYN. Increment counters and dropwithreset. 2557 */ 2558 tcpstat.tcps_badsyn++; 2559 tp = NULL; 2560 goto dropwithreset; 2561 2562 dropafterack: 2563 /* 2564 * Generate an ACK dropping incoming segment if it occupies 2565 * sequence space, where the ACK reflects our state. 2566 */ 2567 if (tiflags & TH_RST) 2568 goto drop; 2569 goto dropafterack2; 2570 2571 dropafterack_ratelim: 2572 /* 2573 * We may want to rate-limit ACKs against SYN/RST attack. 2574 */ 2575 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2576 tcp_ackdrop_ppslim) == 0) { 2577 /* XXX stat */ 2578 goto drop; 2579 } 2580 /* ...fall into dropafterack2... */ 2581 2582 dropafterack2: 2583 m_freem(m); 2584 tp->t_flags |= TF_ACKNOW; 2585 (void) tcp_output(tp); 2586 if (tcp_saveti) 2587 m_freem(tcp_saveti); 2588 return; 2589 2590 dropwithreset_ratelim: 2591 /* 2592 * We may want to rate-limit RSTs in certain situations, 2593 * particularly if we are sending an RST in response to 2594 * an attempt to connect to or otherwise communicate with 2595 * a port for which we have no socket. 2596 */ 2597 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2598 tcp_rst_ppslim) == 0) { 2599 /* XXX stat */ 2600 goto drop; 2601 } 2602 /* ...fall into dropwithreset... */ 2603 2604 dropwithreset: 2605 /* 2606 * Generate a RST, dropping incoming segment. 2607 * Make ACK acceptable to originator of segment. 2608 */ 2609 if (tiflags & TH_RST) 2610 goto drop; 2611 2612 switch (af) { 2613 #ifdef INET6 2614 case AF_INET6: 2615 /* For following calls to tcp_respond */ 2616 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2617 goto drop; 2618 break; 2619 #endif /* INET6 */ 2620 case AF_INET: 2621 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2622 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2623 goto drop; 2624 } 2625 2626 if (tiflags & TH_ACK) 2627 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2628 else { 2629 if (tiflags & TH_SYN) 2630 tlen++; 2631 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2632 TH_RST|TH_ACK); 2633 } 2634 if (tcp_saveti) 2635 m_freem(tcp_saveti); 2636 return; 2637 2638 badcsum: 2639 drop: 2640 /* 2641 * Drop space held by incoming segment and return. 2642 */ 2643 if (tp) { 2644 if (tp->t_inpcb) 2645 so = tp->t_inpcb->inp_socket; 2646 #ifdef INET6 2647 else if (tp->t_in6pcb) 2648 so = tp->t_in6pcb->in6p_socket; 2649 #endif 2650 else 2651 so = NULL; 2652 #ifdef TCP_DEBUG 2653 if (so && (so->so_options & SO_DEBUG) != 0) 2654 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2655 #endif 2656 } 2657 if (tcp_saveti) 2658 m_freem(tcp_saveti); 2659 m_freem(m); 2660 return; 2661 } 2662 2663 #ifdef TCP_SIGNATURE 2664 int 2665 tcp_signature_apply(void *fstate, caddr_t data, u_int len) 2666 { 2667 2668 MD5Update(fstate, (u_char *)data, len); 2669 return (0); 2670 } 2671 2672 struct secasvar * 2673 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 2674 { 2675 struct secasvar *sav; 2676 #ifdef FAST_IPSEC 2677 union sockaddr_union dst; 2678 #endif 2679 struct ip *ip; 2680 struct ip6_hdr *ip6; 2681 2682 ip = mtod(m, struct ip *); 2683 switch (ip->ip_v) { 2684 case 4: 2685 ip = mtod(m, struct ip *); 2686 ip6 = NULL; 2687 break; 2688 case 6: 2689 ip = NULL; 2690 ip6 = mtod(m, struct ip6_hdr *); 2691 break; 2692 default: 2693 return (NULL); 2694 } 2695 2696 #ifdef FAST_IPSEC 2697 /* Extract the destination from the IP header in the mbuf. */ 2698 bzero(&dst, sizeof(union sockaddr_union)); 2699 dst.sa.sa_len = sizeof(struct sockaddr_in); 2700 dst.sa.sa_family = AF_INET; 2701 dst.sin.sin_addr = ip->ip_dst; 2702 2703 /* 2704 * Look up an SADB entry which matches the address of the peer. 2705 */ 2706 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2707 #else 2708 if (ip) 2709 sav = key_allocsa(AF_INET, (caddr_t)&ip->ip_src, 2710 (caddr_t)&ip->ip_dst, IPPROTO_TCP, 2711 htonl(TCP_SIG_SPI)); 2712 else 2713 sav = key_allocsa(AF_INET6, (caddr_t)&ip6->ip6_src, 2714 (caddr_t)&ip6->ip6_dst, IPPROTO_TCP, 2715 htonl(TCP_SIG_SPI)); 2716 #endif 2717 2718 return (sav); /* freesav must be performed by caller */ 2719 } 2720 2721 int 2722 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 2723 struct secasvar *sav, char *sig) 2724 { 2725 MD5_CTX ctx; 2726 struct ip *ip; 2727 struct ipovly *ipovly; 2728 struct ip6_hdr *ip6; 2729 struct ippseudo ippseudo; 2730 struct ip6_hdr_pseudo ip6pseudo; 2731 struct tcphdr th0; 2732 int l, tcphdrlen; 2733 2734 if (sav == NULL) 2735 return (-1); 2736 2737 tcphdrlen = th->th_off * 4; 2738 2739 switch (mtod(m, struct ip *)->ip_v) { 2740 case 4: 2741 ip = mtod(m, struct ip *); 2742 ip6 = NULL; 2743 break; 2744 case 6: 2745 ip = NULL; 2746 ip6 = mtod(m, struct ip6_hdr *); 2747 break; 2748 default: 2749 return (-1); 2750 } 2751 2752 MD5Init(&ctx); 2753 2754 if (ip) { 2755 memset(&ippseudo, 0, sizeof(ippseudo)); 2756 ipovly = (struct ipovly *)ip; 2757 ippseudo.ippseudo_src = ipovly->ih_src; 2758 ippseudo.ippseudo_dst = ipovly->ih_dst; 2759 ippseudo.ippseudo_pad = 0; 2760 ippseudo.ippseudo_p = IPPROTO_TCP; 2761 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 2762 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 2763 } else { 2764 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 2765 ip6pseudo.ip6ph_src = ip6->ip6_src; 2766 in6_clearscope(&ip6pseudo.ip6ph_src); 2767 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 2768 in6_clearscope(&ip6pseudo.ip6ph_dst); 2769 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 2770 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 2771 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 2772 } 2773 2774 th0 = *th; 2775 th0.th_sum = 0; 2776 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 2777 2778 l = m->m_pkthdr.len - thoff - tcphdrlen; 2779 if (l > 0) 2780 m_apply(m, thoff + tcphdrlen, 2781 m->m_pkthdr.len - thoff - tcphdrlen, 2782 tcp_signature_apply, &ctx); 2783 2784 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2785 MD5Final(sig, &ctx); 2786 2787 return (0); 2788 } 2789 #endif 2790 2791 int 2792 tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcphdr *th, 2793 struct mbuf *m, int toff, struct tcp_opt_info *oi) 2794 { 2795 u_int16_t mss; 2796 int opt, optlen = 0; 2797 #ifdef TCP_SIGNATURE 2798 caddr_t sigp = NULL; 2799 char sigbuf[TCP_SIGLEN]; 2800 struct secasvar *sav = NULL; 2801 #endif 2802 2803 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 2804 opt = cp[0]; 2805 if (opt == TCPOPT_EOL) 2806 break; 2807 if (opt == TCPOPT_NOP) 2808 optlen = 1; 2809 else { 2810 if (cnt < 2) 2811 break; 2812 optlen = cp[1]; 2813 if (optlen < 2 || optlen > cnt) 2814 break; 2815 } 2816 switch (opt) { 2817 2818 default: 2819 continue; 2820 2821 case TCPOPT_MAXSEG: 2822 if (optlen != TCPOLEN_MAXSEG) 2823 continue; 2824 if (!(th->th_flags & TH_SYN)) 2825 continue; 2826 bcopy(cp + 2, &mss, sizeof(mss)); 2827 oi->maxseg = ntohs(mss); 2828 break; 2829 2830 case TCPOPT_WINDOW: 2831 if (optlen != TCPOLEN_WINDOW) 2832 continue; 2833 if (!(th->th_flags & TH_SYN)) 2834 continue; 2835 tp->t_flags |= TF_RCVD_SCALE; 2836 tp->requested_s_scale = cp[2]; 2837 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2838 #if 0 /*XXX*/ 2839 char *p; 2840 2841 if (ip) 2842 p = ntohl(ip->ip_src); 2843 #ifdef INET6 2844 else if (ip6) 2845 p = ip6_sprintf(&ip6->ip6_src); 2846 #endif 2847 else 2848 p = "(unknown)"; 2849 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2850 "assuming %d\n", 2851 tp->requested_s_scale, p, 2852 TCP_MAX_WINSHIFT); 2853 #else 2854 log(LOG_ERR, "TCP: invalid wscale %d, " 2855 "assuming %d\n", 2856 tp->requested_s_scale, 2857 TCP_MAX_WINSHIFT); 2858 #endif 2859 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2860 } 2861 break; 2862 2863 case TCPOPT_TIMESTAMP: 2864 if (optlen != TCPOLEN_TIMESTAMP) 2865 continue; 2866 oi->ts_present = 1; 2867 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2868 NTOHL(oi->ts_val); 2869 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2870 NTOHL(oi->ts_ecr); 2871 2872 /* 2873 * A timestamp received in a SYN makes 2874 * it ok to send timestamp requests and replies. 2875 */ 2876 if (th->th_flags & TH_SYN) { 2877 tp->t_flags |= TF_RCVD_TSTMP; 2878 tp->ts_recent = oi->ts_val; 2879 tp->ts_recent_age = tcp_now; 2880 } 2881 break; 2882 case TCPOPT_SACK_PERMITTED: 2883 if (optlen != TCPOLEN_SACK_PERMITTED) 2884 continue; 2885 if (!(th->th_flags & TH_SYN)) 2886 continue; 2887 if (tcp_do_sack) { 2888 tp->t_flags |= TF_SACK_PERMIT; 2889 tp->t_flags |= TF_WILL_SACK; 2890 } 2891 break; 2892 2893 case TCPOPT_SACK: 2894 tcp_sack_option(tp, th, cp, optlen); 2895 break; 2896 #ifdef TCP_SIGNATURE 2897 case TCPOPT_SIGNATURE: 2898 if (optlen != TCPOLEN_SIGNATURE) 2899 continue; 2900 if (sigp && bcmp(sigp, cp + 2, TCP_SIGLEN)) 2901 return (-1); 2902 2903 sigp = sigbuf; 2904 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 2905 memset(cp + 2, 0, TCP_SIGLEN); 2906 tp->t_flags |= TF_SIGNATURE; 2907 break; 2908 #endif 2909 } 2910 } 2911 2912 #ifdef TCP_SIGNATURE 2913 if (tp->t_flags & TF_SIGNATURE) { 2914 2915 sav = tcp_signature_getsav(m, th); 2916 2917 if (sav == NULL && tp->t_state == TCPS_LISTEN) 2918 return (-1); 2919 } 2920 2921 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 2922 if (sav == NULL) 2923 return (-1); 2924 #ifdef FAST_IPSEC 2925 KEY_FREESAV(&sav); 2926 #else 2927 key_freesav(sav); 2928 #endif 2929 return (-1); 2930 } 2931 2932 if (sigp) { 2933 char sig[TCP_SIGLEN]; 2934 2935 TCP_FIELDS_TO_NET(th); 2936 if (tcp_signature(m, th, toff, sav, sig) < 0) { 2937 TCP_FIELDS_TO_HOST(th); 2938 if (sav == NULL) 2939 return (-1); 2940 #ifdef FAST_IPSEC 2941 KEY_FREESAV(&sav); 2942 #else 2943 key_freesav(sav); 2944 #endif 2945 return (-1); 2946 } 2947 TCP_FIELDS_TO_HOST(th); 2948 2949 if (bcmp(sig, sigp, TCP_SIGLEN)) { 2950 tcpstat.tcps_badsig++; 2951 if (sav == NULL) 2952 return (-1); 2953 #ifdef FAST_IPSEC 2954 KEY_FREESAV(&sav); 2955 #else 2956 key_freesav(sav); 2957 #endif 2958 return (-1); 2959 } else 2960 tcpstat.tcps_goodsig++; 2961 2962 key_sa_recordxfer(sav, m); 2963 #ifdef FAST_IPSEC 2964 KEY_FREESAV(&sav); 2965 #else 2966 key_freesav(sav); 2967 #endif 2968 } 2969 #endif 2970 2971 return (0); 2972 } 2973 2974 /* 2975 * Pull out of band byte out of a segment so 2976 * it doesn't appear in the user's data queue. 2977 * It is still reflected in the segment length for 2978 * sequencing purposes. 2979 */ 2980 void 2981 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 2982 struct mbuf *m, int off) 2983 { 2984 int cnt = off + th->th_urp - 1; 2985 2986 while (cnt >= 0) { 2987 if (m->m_len > cnt) { 2988 char *cp = mtod(m, caddr_t) + cnt; 2989 struct tcpcb *tp = sototcpcb(so); 2990 2991 tp->t_iobc = *cp; 2992 tp->t_oobflags |= TCPOOB_HAVEDATA; 2993 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2994 m->m_len--; 2995 return; 2996 } 2997 cnt -= m->m_len; 2998 m = m->m_next; 2999 if (m == 0) 3000 break; 3001 } 3002 panic("tcp_pulloutofband"); 3003 } 3004 3005 /* 3006 * Collect new round-trip time estimate 3007 * and update averages and current timeout. 3008 */ 3009 void 3010 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3011 { 3012 int32_t delta; 3013 3014 tcpstat.tcps_rttupdated++; 3015 if (tp->t_srtt != 0) { 3016 /* 3017 * srtt is stored as fixed point with 3 bits after the 3018 * binary point (i.e., scaled by 8). The following magic 3019 * is equivalent to the smoothing algorithm in rfc793 with 3020 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3021 * point). Adjust rtt to origin 0. 3022 */ 3023 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3024 if ((tp->t_srtt += delta) <= 0) 3025 tp->t_srtt = 1 << 2; 3026 /* 3027 * We accumulate a smoothed rtt variance (actually, a 3028 * smoothed mean difference), then set the retransmit 3029 * timer to smoothed rtt + 4 times the smoothed variance. 3030 * rttvar is stored as fixed point with 2 bits after the 3031 * binary point (scaled by 4). The following is 3032 * equivalent to rfc793 smoothing with an alpha of .75 3033 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3034 * rfc793's wired-in beta. 3035 */ 3036 if (delta < 0) 3037 delta = -delta; 3038 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3039 if ((tp->t_rttvar += delta) <= 0) 3040 tp->t_rttvar = 1 << 2; 3041 } else { 3042 /* 3043 * No rtt measurement yet - use the unsmoothed rtt. 3044 * Set the variance to half the rtt (so our first 3045 * retransmit happens at 3*rtt). 3046 */ 3047 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3048 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3049 } 3050 tp->t_rtttime = 0; 3051 tp->t_rxtshift = 0; 3052 3053 /* 3054 * the retransmit should happen at rtt + 4 * rttvar. 3055 * Because of the way we do the smoothing, srtt and rttvar 3056 * will each average +1/2 tick of bias. When we compute 3057 * the retransmit timer, we want 1/2 tick of rounding and 3058 * 1 extra tick because of +-1/2 tick uncertainty in the 3059 * firing of the timer. The bias will give us exactly the 3060 * 1.5 tick we need. But, because the bias is 3061 * statistical, we have to test that we don't drop below 3062 * the minimum feasible timer (which is 2 ticks). 3063 */ 3064 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3065 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3066 3067 /* 3068 * We received an ack for a packet that wasn't retransmitted; 3069 * it is probably safe to discard any error indications we've 3070 * received recently. This isn't quite right, but close enough 3071 * for now (a route might have failed after we sent a segment, 3072 * and the return path might not be symmetrical). 3073 */ 3074 tp->t_softerror = 0; 3075 } 3076 3077 void 3078 tcp_reno_newack(struct tcpcb *tp, struct tcphdr *th) 3079 { 3080 if (tp->t_partialacks < 0) { 3081 /* 3082 * We were not in fast recovery. Reset the duplicate ack 3083 * counter. 3084 */ 3085 tp->t_dupacks = 0; 3086 } else { 3087 /* 3088 * Clamp the congestion window to the crossover point and 3089 * exit fast recovery. 3090 */ 3091 if (tp->snd_cwnd > tp->snd_ssthresh) 3092 tp->snd_cwnd = tp->snd_ssthresh; 3093 tp->t_partialacks = -1; 3094 tp->t_dupacks = 0; 3095 } 3096 } 3097 3098 /* 3099 * Implement the NewReno response to a new ack, checking for partial acks in 3100 * fast recovery. 3101 */ 3102 void 3103 tcp_newreno_newack(struct tcpcb *tp, struct tcphdr *th) 3104 { 3105 if (tp->t_partialacks < 0) { 3106 /* 3107 * We were not in fast recovery. Reset the duplicate ack 3108 * counter. 3109 */ 3110 tp->t_dupacks = 0; 3111 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) { 3112 /* 3113 * This is a partial ack. Retransmit the first unacknowledged 3114 * segment and deflate the congestion window by the amount of 3115 * acknowledged data. Do not exit fast recovery. 3116 */ 3117 tcp_seq onxt = tp->snd_nxt; 3118 u_long ocwnd = tp->snd_cwnd; 3119 3120 /* 3121 * snd_una has not yet been updated and the socket's send 3122 * buffer has not yet drained off the ACK'd data, so we 3123 * have to leave snd_una as it was to get the correct data 3124 * offset in tcp_output(). 3125 */ 3126 if (++tp->t_partialacks == 1) 3127 TCP_TIMER_DISARM(tp, TCPT_REXMT); 3128 tp->t_rtttime = 0; 3129 tp->snd_nxt = th->th_ack; 3130 /* 3131 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 3132 * is not yet updated when we're called. 3133 */ 3134 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 3135 (void) tcp_output(tp); 3136 tp->snd_cwnd = ocwnd; 3137 if (SEQ_GT(onxt, tp->snd_nxt)) 3138 tp->snd_nxt = onxt; 3139 /* 3140 * Partial window deflation. Relies on fact that tp->snd_una 3141 * not updated yet. 3142 */ 3143 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 3144 } else { 3145 /* 3146 * Complete ack. Inflate the congestion window to ssthresh 3147 * and exit fast recovery. 3148 * 3149 * Window inflation should have left us with approx. 3150 * snd_ssthresh outstanding data. But in case we 3151 * would be inclined to send a burst, better to do 3152 * it via the slow start mechanism. 3153 */ 3154 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 3155 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 3156 + tp->t_segsz; 3157 else 3158 tp->snd_cwnd = tp->snd_ssthresh; 3159 tp->t_partialacks = -1; 3160 tp->t_dupacks = 0; 3161 } 3162 } 3163 3164 3165 /* 3166 * TCP compressed state engine. Currently used to hold compressed 3167 * state for SYN_RECEIVED. 3168 */ 3169 3170 u_long syn_cache_count; 3171 u_int32_t syn_hash1, syn_hash2; 3172 3173 #define SYN_HASH(sa, sp, dp) \ 3174 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3175 ((u_int32_t)(sp)))^syn_hash2))) 3176 #ifndef INET6 3177 #define SYN_HASHALL(hash, src, dst) \ 3178 do { \ 3179 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 3180 ((struct sockaddr_in *)(src))->sin_port, \ 3181 ((struct sockaddr_in *)(dst))->sin_port); \ 3182 } while (/*CONSTCOND*/ 0) 3183 #else 3184 #define SYN_HASH6(sa, sp, dp) \ 3185 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3186 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3187 & 0x7fffffff) 3188 3189 #define SYN_HASHALL(hash, src, dst) \ 3190 do { \ 3191 switch ((src)->sa_family) { \ 3192 case AF_INET: \ 3193 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 3194 ((struct sockaddr_in *)(src))->sin_port, \ 3195 ((struct sockaddr_in *)(dst))->sin_port); \ 3196 break; \ 3197 case AF_INET6: \ 3198 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 3199 ((struct sockaddr_in6 *)(src))->sin6_port, \ 3200 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 3201 break; \ 3202 default: \ 3203 hash = 0; \ 3204 } \ 3205 } while (/*CONSTCOND*/0) 3206 #endif /* INET6 */ 3207 3208 #define SYN_CACHE_RM(sc) \ 3209 do { \ 3210 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 3211 (sc), sc_bucketq); \ 3212 (sc)->sc_tp = NULL; \ 3213 LIST_REMOVE((sc), sc_tpq); \ 3214 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 3215 callout_stop(&(sc)->sc_timer); \ 3216 syn_cache_count--; \ 3217 } while (/*CONSTCOND*/0) 3218 3219 #define SYN_CACHE_PUT(sc) \ 3220 do { \ 3221 if ((sc)->sc_ipopts) \ 3222 (void) m_free((sc)->sc_ipopts); \ 3223 if ((sc)->sc_route4.ro_rt != NULL) \ 3224 RTFREE((sc)->sc_route4.ro_rt); \ 3225 if (callout_invoking(&(sc)->sc_timer)) \ 3226 (sc)->sc_flags |= SCF_DEAD; \ 3227 else \ 3228 pool_put(&syn_cache_pool, (sc)); \ 3229 } while (/*CONSTCOND*/0) 3230 3231 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL); 3232 3233 /* 3234 * We don't estimate RTT with SYNs, so each packet starts with the default 3235 * RTT and each timer step has a fixed timeout value. 3236 */ 3237 #define SYN_CACHE_TIMER_ARM(sc) \ 3238 do { \ 3239 TCPT_RANGESET((sc)->sc_rxtcur, \ 3240 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3241 TCPTV_REXMTMAX); \ 3242 callout_reset(&(sc)->sc_timer, \ 3243 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3244 } while (/*CONSTCOND*/0) 3245 3246 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3247 3248 void 3249 syn_cache_init(void) 3250 { 3251 int i; 3252 3253 /* Initialize the hash buckets. */ 3254 for (i = 0; i < tcp_syn_cache_size; i++) 3255 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3256 } 3257 3258 void 3259 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3260 { 3261 struct syn_cache_head *scp; 3262 struct syn_cache *sc2; 3263 int s; 3264 3265 /* 3266 * If there are no entries in the hash table, reinitialize 3267 * the hash secrets. 3268 */ 3269 if (syn_cache_count == 0) { 3270 syn_hash1 = arc4random(); 3271 syn_hash2 = arc4random(); 3272 } 3273 3274 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3275 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3276 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3277 3278 /* 3279 * Make sure that we don't overflow the per-bucket 3280 * limit or the total cache size limit. 3281 */ 3282 s = splsoftnet(); 3283 if (scp->sch_length >= tcp_syn_bucket_limit) { 3284 tcpstat.tcps_sc_bucketoverflow++; 3285 /* 3286 * The bucket is full. Toss the oldest element in the 3287 * bucket. This will be the first entry in the bucket. 3288 */ 3289 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3290 #ifdef DIAGNOSTIC 3291 /* 3292 * This should never happen; we should always find an 3293 * entry in our bucket. 3294 */ 3295 if (sc2 == NULL) 3296 panic("syn_cache_insert: bucketoverflow: impossible"); 3297 #endif 3298 SYN_CACHE_RM(sc2); 3299 SYN_CACHE_PUT(sc2); 3300 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3301 struct syn_cache_head *scp2, *sce; 3302 3303 tcpstat.tcps_sc_overflowed++; 3304 /* 3305 * The cache is full. Toss the oldest entry in the 3306 * first non-empty bucket we can find. 3307 * 3308 * XXX We would really like to toss the oldest 3309 * entry in the cache, but we hope that this 3310 * condition doesn't happen very often. 3311 */ 3312 scp2 = scp; 3313 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3314 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3315 for (++scp2; scp2 != scp; scp2++) { 3316 if (scp2 >= sce) 3317 scp2 = &tcp_syn_cache[0]; 3318 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3319 break; 3320 } 3321 #ifdef DIAGNOSTIC 3322 /* 3323 * This should never happen; we should always find a 3324 * non-empty bucket. 3325 */ 3326 if (scp2 == scp) 3327 panic("syn_cache_insert: cacheoverflow: " 3328 "impossible"); 3329 #endif 3330 } 3331 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3332 SYN_CACHE_RM(sc2); 3333 SYN_CACHE_PUT(sc2); 3334 } 3335 3336 /* 3337 * Initialize the entry's timer. 3338 */ 3339 sc->sc_rxttot = 0; 3340 sc->sc_rxtshift = 0; 3341 SYN_CACHE_TIMER_ARM(sc); 3342 3343 /* Link it from tcpcb entry */ 3344 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3345 3346 /* Put it into the bucket. */ 3347 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3348 scp->sch_length++; 3349 syn_cache_count++; 3350 3351 tcpstat.tcps_sc_added++; 3352 splx(s); 3353 } 3354 3355 /* 3356 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3357 * If we have retransmitted an entry the maximum number of times, expire 3358 * that entry. 3359 */ 3360 void 3361 syn_cache_timer(void *arg) 3362 { 3363 struct syn_cache *sc = arg; 3364 int s; 3365 3366 s = splsoftnet(); 3367 callout_ack(&sc->sc_timer); 3368 3369 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3370 tcpstat.tcps_sc_delayed_free++; 3371 pool_put(&syn_cache_pool, sc); 3372 splx(s); 3373 return; 3374 } 3375 3376 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3377 /* Drop it -- too many retransmissions. */ 3378 goto dropit; 3379 } 3380 3381 /* 3382 * Compute the total amount of time this entry has 3383 * been on a queue. If this entry has been on longer 3384 * than the keep alive timer would allow, expire it. 3385 */ 3386 sc->sc_rxttot += sc->sc_rxtcur; 3387 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) 3388 goto dropit; 3389 3390 tcpstat.tcps_sc_retransmitted++; 3391 (void) syn_cache_respond(sc, NULL); 3392 3393 /* Advance the timer back-off. */ 3394 sc->sc_rxtshift++; 3395 SYN_CACHE_TIMER_ARM(sc); 3396 3397 splx(s); 3398 return; 3399 3400 dropit: 3401 tcpstat.tcps_sc_timed_out++; 3402 SYN_CACHE_RM(sc); 3403 SYN_CACHE_PUT(sc); 3404 splx(s); 3405 } 3406 3407 /* 3408 * Remove syn cache created by the specified tcb entry, 3409 * because this does not make sense to keep them 3410 * (if there's no tcb entry, syn cache entry will never be used) 3411 */ 3412 void 3413 syn_cache_cleanup(struct tcpcb *tp) 3414 { 3415 struct syn_cache *sc, *nsc; 3416 int s; 3417 3418 s = splsoftnet(); 3419 3420 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3421 nsc = LIST_NEXT(sc, sc_tpq); 3422 3423 #ifdef DIAGNOSTIC 3424 if (sc->sc_tp != tp) 3425 panic("invalid sc_tp in syn_cache_cleanup"); 3426 #endif 3427 SYN_CACHE_RM(sc); 3428 SYN_CACHE_PUT(sc); 3429 } 3430 /* just for safety */ 3431 LIST_INIT(&tp->t_sc); 3432 3433 splx(s); 3434 } 3435 3436 /* 3437 * Find an entry in the syn cache. 3438 */ 3439 struct syn_cache * 3440 syn_cache_lookup(struct sockaddr *src, struct sockaddr *dst, 3441 struct syn_cache_head **headp) 3442 { 3443 struct syn_cache *sc; 3444 struct syn_cache_head *scp; 3445 u_int32_t hash; 3446 int s; 3447 3448 SYN_HASHALL(hash, src, dst); 3449 3450 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3451 *headp = scp; 3452 s = splsoftnet(); 3453 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3454 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3455 if (sc->sc_hash != hash) 3456 continue; 3457 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3458 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3459 splx(s); 3460 return (sc); 3461 } 3462 } 3463 splx(s); 3464 return (NULL); 3465 } 3466 3467 /* 3468 * This function gets called when we receive an ACK for a 3469 * socket in the LISTEN state. We look up the connection 3470 * in the syn cache, and if its there, we pull it out of 3471 * the cache and turn it into a full-blown connection in 3472 * the SYN-RECEIVED state. 3473 * 3474 * The return values may not be immediately obvious, and their effects 3475 * can be subtle, so here they are: 3476 * 3477 * NULL SYN was not found in cache; caller should drop the 3478 * packet and send an RST. 3479 * 3480 * -1 We were unable to create the new connection, and are 3481 * aborting it. An ACK,RST is being sent to the peer 3482 * (unless we got screwey sequence numbners; see below), 3483 * because the 3-way handshake has been completed. Caller 3484 * should not free the mbuf, since we may be using it. If 3485 * we are not, we will free it. 3486 * 3487 * Otherwise, the return value is a pointer to the new socket 3488 * associated with the connection. 3489 */ 3490 struct socket * 3491 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3492 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3493 struct socket *so, struct mbuf *m) 3494 { 3495 struct syn_cache *sc; 3496 struct syn_cache_head *scp; 3497 struct inpcb *inp = NULL; 3498 #ifdef INET6 3499 struct in6pcb *in6p = NULL; 3500 #endif 3501 struct tcpcb *tp = 0; 3502 struct mbuf *am; 3503 int s; 3504 struct socket *oso; 3505 3506 s = splsoftnet(); 3507 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3508 splx(s); 3509 return (NULL); 3510 } 3511 3512 /* 3513 * Verify the sequence and ack numbers. Try getting the correct 3514 * response again. 3515 */ 3516 if ((th->th_ack != sc->sc_iss + 1) || 3517 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3518 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3519 (void) syn_cache_respond(sc, m); 3520 splx(s); 3521 return ((struct socket *)(-1)); 3522 } 3523 3524 /* Remove this cache entry */ 3525 SYN_CACHE_RM(sc); 3526 splx(s); 3527 3528 /* 3529 * Ok, create the full blown connection, and set things up 3530 * as they would have been set up if we had created the 3531 * connection when the SYN arrived. If we can't create 3532 * the connection, abort it. 3533 */ 3534 /* 3535 * inp still has the OLD in_pcb stuff, set the 3536 * v6-related flags on the new guy, too. This is 3537 * done particularly for the case where an AF_INET6 3538 * socket is bound only to a port, and a v4 connection 3539 * comes in on that port. 3540 * we also copy the flowinfo from the original pcb 3541 * to the new one. 3542 */ 3543 oso = so; 3544 so = sonewconn(so, SS_ISCONNECTED); 3545 if (so == NULL) 3546 goto resetandabort; 3547 3548 switch (so->so_proto->pr_domain->dom_family) { 3549 #ifdef INET 3550 case AF_INET: 3551 inp = sotoinpcb(so); 3552 break; 3553 #endif 3554 #ifdef INET6 3555 case AF_INET6: 3556 in6p = sotoin6pcb(so); 3557 break; 3558 #endif 3559 } 3560 switch (src->sa_family) { 3561 #ifdef INET 3562 case AF_INET: 3563 if (inp) { 3564 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3565 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3566 inp->inp_options = ip_srcroute(); 3567 in_pcbstate(inp, INP_BOUND); 3568 if (inp->inp_options == NULL) { 3569 inp->inp_options = sc->sc_ipopts; 3570 sc->sc_ipopts = NULL; 3571 } 3572 } 3573 #ifdef INET6 3574 else if (in6p) { 3575 /* IPv4 packet to AF_INET6 socket */ 3576 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3577 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3578 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3579 &in6p->in6p_laddr.s6_addr32[3], 3580 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3581 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3582 in6totcpcb(in6p)->t_family = AF_INET; 3583 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3584 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3585 else 3586 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3587 in6_pcbstate(in6p, IN6P_BOUND); 3588 } 3589 #endif 3590 break; 3591 #endif 3592 #ifdef INET6 3593 case AF_INET6: 3594 if (in6p) { 3595 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3596 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3597 in6_pcbstate(in6p, IN6P_BOUND); 3598 } 3599 break; 3600 #endif 3601 } 3602 #ifdef INET6 3603 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3604 struct in6pcb *oin6p = sotoin6pcb(oso); 3605 /* inherit socket options from the listening socket */ 3606 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3607 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3608 m_freem(in6p->in6p_options); 3609 in6p->in6p_options = 0; 3610 } 3611 ip6_savecontrol(in6p, &in6p->in6p_options, 3612 mtod(m, struct ip6_hdr *), m); 3613 } 3614 #endif 3615 3616 #if defined(IPSEC) || defined(FAST_IPSEC) 3617 /* 3618 * we make a copy of policy, instead of sharing the policy, 3619 * for better behavior in terms of SA lookup and dead SA removal. 3620 */ 3621 if (inp) { 3622 /* copy old policy into new socket's */ 3623 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3624 printf("tcp_input: could not copy policy\n"); 3625 } 3626 #ifdef INET6 3627 else if (in6p) { 3628 /* copy old policy into new socket's */ 3629 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3630 in6p->in6p_sp)) 3631 printf("tcp_input: could not copy policy\n"); 3632 } 3633 #endif 3634 #endif 3635 3636 /* 3637 * Give the new socket our cached route reference. 3638 */ 3639 if (inp) 3640 inp->inp_route = sc->sc_route4; /* struct assignment */ 3641 #ifdef INET6 3642 else 3643 in6p->in6p_route = sc->sc_route6; 3644 #endif 3645 sc->sc_route4.ro_rt = NULL; 3646 3647 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3648 if (am == NULL) 3649 goto resetandabort; 3650 MCLAIM(am, &tcp_mowner); 3651 am->m_len = src->sa_len; 3652 bcopy(src, mtod(am, caddr_t), src->sa_len); 3653 if (inp) { 3654 if (in_pcbconnect(inp, am)) { 3655 (void) m_free(am); 3656 goto resetandabort; 3657 } 3658 } 3659 #ifdef INET6 3660 else if (in6p) { 3661 if (src->sa_family == AF_INET) { 3662 /* IPv4 packet to AF_INET6 socket */ 3663 struct sockaddr_in6 *sin6; 3664 sin6 = mtod(am, struct sockaddr_in6 *); 3665 am->m_len = sizeof(*sin6); 3666 bzero(sin6, sizeof(*sin6)); 3667 sin6->sin6_family = AF_INET6; 3668 sin6->sin6_len = sizeof(*sin6); 3669 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3670 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3671 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3672 &sin6->sin6_addr.s6_addr32[3], 3673 sizeof(sin6->sin6_addr.s6_addr32[3])); 3674 } 3675 if (in6_pcbconnect(in6p, am)) { 3676 (void) m_free(am); 3677 goto resetandabort; 3678 } 3679 } 3680 #endif 3681 else { 3682 (void) m_free(am); 3683 goto resetandabort; 3684 } 3685 (void) m_free(am); 3686 3687 if (inp) 3688 tp = intotcpcb(inp); 3689 #ifdef INET6 3690 else if (in6p) 3691 tp = in6totcpcb(in6p); 3692 #endif 3693 else 3694 tp = NULL; 3695 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3696 if (sc->sc_request_r_scale != 15) { 3697 tp->requested_s_scale = sc->sc_requested_s_scale; 3698 tp->request_r_scale = sc->sc_request_r_scale; 3699 tp->snd_scale = sc->sc_requested_s_scale; 3700 tp->rcv_scale = sc->sc_request_r_scale; 3701 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3702 } 3703 if (sc->sc_flags & SCF_TIMESTAMP) 3704 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3705 tp->ts_timebase = sc->sc_timebase; 3706 3707 tp->t_template = tcp_template(tp); 3708 if (tp->t_template == 0) { 3709 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3710 so = NULL; 3711 m_freem(m); 3712 goto abort; 3713 } 3714 3715 tp->iss = sc->sc_iss; 3716 tp->irs = sc->sc_irs; 3717 tcp_sendseqinit(tp); 3718 tcp_rcvseqinit(tp); 3719 tp->t_state = TCPS_SYN_RECEIVED; 3720 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3721 tcpstat.tcps_accepts++; 3722 3723 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 3724 tp->t_flags |= TF_WILL_SACK; 3725 3726 #ifdef TCP_SIGNATURE 3727 if (sc->sc_flags & SCF_SIGNATURE) 3728 tp->t_flags |= TF_SIGNATURE; 3729 #endif 3730 3731 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3732 tp->t_ourmss = sc->sc_ourmaxseg; 3733 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3734 3735 /* 3736 * Initialize the initial congestion window. If we 3737 * had to retransmit the SYN,ACK, we must initialize cwnd 3738 * to 1 segment (i.e. the Loss Window). 3739 */ 3740 if (sc->sc_rxtshift) 3741 tp->snd_cwnd = tp->t_peermss; 3742 else { 3743 int ss = tcp_init_win; 3744 #ifdef INET 3745 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3746 ss = tcp_init_win_local; 3747 #endif 3748 #ifdef INET6 3749 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3750 ss = tcp_init_win_local; 3751 #endif 3752 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3753 } 3754 3755 tcp_rmx_rtt(tp); 3756 tp->snd_wl1 = sc->sc_irs; 3757 tp->rcv_up = sc->sc_irs + 1; 3758 3759 /* 3760 * This is what whould have happened in tcp_output() when 3761 * the SYN,ACK was sent. 3762 */ 3763 tp->snd_up = tp->snd_una; 3764 tp->snd_max = tp->snd_nxt = tp->iss+1; 3765 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3766 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3767 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3768 tp->last_ack_sent = tp->rcv_nxt; 3769 tp->t_partialacks = -1; 3770 tp->t_dupacks = 0; 3771 3772 tcpstat.tcps_sc_completed++; 3773 SYN_CACHE_PUT(sc); 3774 return (so); 3775 3776 resetandabort: 3777 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3778 abort: 3779 if (so != NULL) 3780 (void) soabort(so); 3781 SYN_CACHE_PUT(sc); 3782 tcpstat.tcps_sc_aborted++; 3783 return ((struct socket *)(-1)); 3784 } 3785 3786 /* 3787 * This function is called when we get a RST for a 3788 * non-existent connection, so that we can see if the 3789 * connection is in the syn cache. If it is, zap it. 3790 */ 3791 3792 void 3793 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 3794 { 3795 struct syn_cache *sc; 3796 struct syn_cache_head *scp; 3797 int s = splsoftnet(); 3798 3799 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3800 splx(s); 3801 return; 3802 } 3803 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3804 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3805 splx(s); 3806 return; 3807 } 3808 SYN_CACHE_RM(sc); 3809 splx(s); 3810 tcpstat.tcps_sc_reset++; 3811 SYN_CACHE_PUT(sc); 3812 } 3813 3814 void 3815 syn_cache_unreach(struct sockaddr *src, struct sockaddr *dst, 3816 struct tcphdr *th) 3817 { 3818 struct syn_cache *sc; 3819 struct syn_cache_head *scp; 3820 int s; 3821 3822 s = splsoftnet(); 3823 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3824 splx(s); 3825 return; 3826 } 3827 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3828 if (ntohl (th->th_seq) != sc->sc_iss) { 3829 splx(s); 3830 return; 3831 } 3832 3833 /* 3834 * If we've retransmitted 3 times and this is our second error, 3835 * we remove the entry. Otherwise, we allow it to continue on. 3836 * This prevents us from incorrectly nuking an entry during a 3837 * spurious network outage. 3838 * 3839 * See tcp_notify(). 3840 */ 3841 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3842 sc->sc_flags |= SCF_UNREACH; 3843 splx(s); 3844 return; 3845 } 3846 3847 SYN_CACHE_RM(sc); 3848 splx(s); 3849 tcpstat.tcps_sc_unreach++; 3850 SYN_CACHE_PUT(sc); 3851 } 3852 3853 /* 3854 * Given a LISTEN socket and an inbound SYN request, add 3855 * this to the syn cache, and send back a segment: 3856 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3857 * to the source. 3858 * 3859 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3860 * Doing so would require that we hold onto the data and deliver it 3861 * to the application. However, if we are the target of a SYN-flood 3862 * DoS attack, an attacker could send data which would eventually 3863 * consume all available buffer space if it were ACKed. By not ACKing 3864 * the data, we avoid this DoS scenario. 3865 */ 3866 3867 int 3868 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3869 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 3870 int optlen, struct tcp_opt_info *oi) 3871 { 3872 struct tcpcb tb, *tp; 3873 long win; 3874 struct syn_cache *sc; 3875 struct syn_cache_head *scp; 3876 struct mbuf *ipopts; 3877 struct tcp_opt_info opti; 3878 3879 tp = sototcpcb(so); 3880 3881 bzero(&opti, sizeof(opti)); 3882 3883 /* 3884 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3885 * 3886 * Note this check is performed in tcp_input() very early on. 3887 */ 3888 3889 /* 3890 * Initialize some local state. 3891 */ 3892 win = sbspace(&so->so_rcv); 3893 if (win > TCP_MAXWIN) 3894 win = TCP_MAXWIN; 3895 3896 switch (src->sa_family) { 3897 #ifdef INET 3898 case AF_INET: 3899 /* 3900 * Remember the IP options, if any. 3901 */ 3902 ipopts = ip_srcroute(); 3903 break; 3904 #endif 3905 default: 3906 ipopts = NULL; 3907 } 3908 3909 #ifdef TCP_SIGNATURE 3910 if (optp || (tp->t_flags & TF_SIGNATURE)) 3911 #else 3912 if (optp) 3913 #endif 3914 { 3915 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3916 #ifdef TCP_SIGNATURE 3917 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 3918 #endif 3919 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 3920 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 3921 return (0); 3922 } else 3923 tb.t_flags = 0; 3924 3925 /* 3926 * See if we already have an entry for this connection. 3927 * If we do, resend the SYN,ACK. We do not count this 3928 * as a retransmission (XXX though maybe we should). 3929 */ 3930 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3931 tcpstat.tcps_sc_dupesyn++; 3932 if (ipopts) { 3933 /* 3934 * If we were remembering a previous source route, 3935 * forget it and use the new one we've been given. 3936 */ 3937 if (sc->sc_ipopts) 3938 (void) m_free(sc->sc_ipopts); 3939 sc->sc_ipopts = ipopts; 3940 } 3941 sc->sc_timestamp = tb.ts_recent; 3942 if (syn_cache_respond(sc, m) == 0) { 3943 tcpstat.tcps_sndacks++; 3944 tcpstat.tcps_sndtotal++; 3945 } 3946 return (1); 3947 } 3948 3949 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3950 if (sc == NULL) { 3951 if (ipopts) 3952 (void) m_free(ipopts); 3953 return (0); 3954 } 3955 3956 /* 3957 * Fill in the cache, and put the necessary IP and TCP 3958 * options into the reply. 3959 */ 3960 bzero(sc, sizeof(struct syn_cache)); 3961 callout_init(&sc->sc_timer); 3962 bcopy(src, &sc->sc_src, src->sa_len); 3963 bcopy(dst, &sc->sc_dst, dst->sa_len); 3964 sc->sc_flags = 0; 3965 sc->sc_ipopts = ipopts; 3966 sc->sc_irs = th->th_seq; 3967 switch (src->sa_family) { 3968 #ifdef INET 3969 case AF_INET: 3970 { 3971 struct sockaddr_in *srcin = (void *) src; 3972 struct sockaddr_in *dstin = (void *) dst; 3973 3974 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 3975 &srcin->sin_addr, dstin->sin_port, 3976 srcin->sin_port, sizeof(dstin->sin_addr), 0); 3977 break; 3978 } 3979 #endif /* INET */ 3980 #ifdef INET6 3981 case AF_INET6: 3982 { 3983 struct sockaddr_in6 *srcin6 = (void *) src; 3984 struct sockaddr_in6 *dstin6 = (void *) dst; 3985 3986 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 3987 &srcin6->sin6_addr, dstin6->sin6_port, 3988 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 3989 break; 3990 } 3991 #endif /* INET6 */ 3992 } 3993 sc->sc_peermaxseg = oi->maxseg; 3994 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3995 m->m_pkthdr.rcvif : NULL, 3996 sc->sc_src.sa.sa_family); 3997 sc->sc_win = win; 3998 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 3999 sc->sc_timestamp = tb.ts_recent; 4000 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4001 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4002 sc->sc_flags |= SCF_TIMESTAMP; 4003 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4004 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4005 sc->sc_requested_s_scale = tb.requested_s_scale; 4006 sc->sc_request_r_scale = 0; 4007 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4008 TCP_MAXWIN << sc->sc_request_r_scale < 4009 so->so_rcv.sb_hiwat) 4010 sc->sc_request_r_scale++; 4011 } else { 4012 sc->sc_requested_s_scale = 15; 4013 sc->sc_request_r_scale = 15; 4014 } 4015 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4016 sc->sc_flags |= SCF_SACK_PERMIT; 4017 #ifdef TCP_SIGNATURE 4018 if (tb.t_flags & TF_SIGNATURE) 4019 sc->sc_flags |= SCF_SIGNATURE; 4020 #endif 4021 sc->sc_tp = tp; 4022 if (syn_cache_respond(sc, m) == 0) { 4023 syn_cache_insert(sc, tp); 4024 tcpstat.tcps_sndacks++; 4025 tcpstat.tcps_sndtotal++; 4026 } else { 4027 SYN_CACHE_PUT(sc); 4028 tcpstat.tcps_sc_dropped++; 4029 } 4030 return (1); 4031 } 4032 4033 int 4034 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4035 { 4036 struct route *ro; 4037 u_int8_t *optp; 4038 int optlen, error; 4039 u_int16_t tlen; 4040 struct ip *ip = NULL; 4041 #ifdef INET6 4042 struct ip6_hdr *ip6 = NULL; 4043 #endif 4044 struct tcpcb *tp; 4045 struct tcphdr *th; 4046 u_int hlen; 4047 struct socket *so; 4048 4049 switch (sc->sc_src.sa.sa_family) { 4050 case AF_INET: 4051 hlen = sizeof(struct ip); 4052 ro = &sc->sc_route4; 4053 break; 4054 #ifdef INET6 4055 case AF_INET6: 4056 hlen = sizeof(struct ip6_hdr); 4057 ro = (struct route *)&sc->sc_route6; 4058 break; 4059 #endif 4060 default: 4061 if (m) 4062 m_freem(m); 4063 return (EAFNOSUPPORT); 4064 } 4065 4066 /* Compute the size of the TCP options. */ 4067 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4068 ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) + 4069 #ifdef TCP_SIGNATURE 4070 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4071 #endif 4072 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4073 4074 tlen = hlen + sizeof(struct tcphdr) + optlen; 4075 4076 /* 4077 * Create the IP+TCP header from scratch. 4078 */ 4079 if (m) 4080 m_freem(m); 4081 #ifdef DIAGNOSTIC 4082 if (max_linkhdr + tlen > MCLBYTES) 4083 return (ENOBUFS); 4084 #endif 4085 MGETHDR(m, M_DONTWAIT, MT_DATA); 4086 if (m && tlen > MHLEN) { 4087 MCLGET(m, M_DONTWAIT); 4088 if ((m->m_flags & M_EXT) == 0) { 4089 m_freem(m); 4090 m = NULL; 4091 } 4092 } 4093 if (m == NULL) 4094 return (ENOBUFS); 4095 MCLAIM(m, &tcp_tx_mowner); 4096 4097 /* Fixup the mbuf. */ 4098 m->m_data += max_linkhdr; 4099 m->m_len = m->m_pkthdr.len = tlen; 4100 if (sc->sc_tp) { 4101 tp = sc->sc_tp; 4102 if (tp->t_inpcb) 4103 so = tp->t_inpcb->inp_socket; 4104 #ifdef INET6 4105 else if (tp->t_in6pcb) 4106 so = tp->t_in6pcb->in6p_socket; 4107 #endif 4108 else 4109 so = NULL; 4110 } else 4111 so = NULL; 4112 m->m_pkthdr.rcvif = NULL; 4113 memset(mtod(m, u_char *), 0, tlen); 4114 4115 switch (sc->sc_src.sa.sa_family) { 4116 case AF_INET: 4117 ip = mtod(m, struct ip *); 4118 ip->ip_v = 4; 4119 ip->ip_dst = sc->sc_src.sin.sin_addr; 4120 ip->ip_src = sc->sc_dst.sin.sin_addr; 4121 ip->ip_p = IPPROTO_TCP; 4122 th = (struct tcphdr *)(ip + 1); 4123 th->th_dport = sc->sc_src.sin.sin_port; 4124 th->th_sport = sc->sc_dst.sin.sin_port; 4125 break; 4126 #ifdef INET6 4127 case AF_INET6: 4128 ip6 = mtod(m, struct ip6_hdr *); 4129 ip6->ip6_vfc = IPV6_VERSION; 4130 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4131 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4132 ip6->ip6_nxt = IPPROTO_TCP; 4133 /* ip6_plen will be updated in ip6_output() */ 4134 th = (struct tcphdr *)(ip6 + 1); 4135 th->th_dport = sc->sc_src.sin6.sin6_port; 4136 th->th_sport = sc->sc_dst.sin6.sin6_port; 4137 break; 4138 #endif 4139 default: 4140 th = NULL; 4141 } 4142 4143 th->th_seq = htonl(sc->sc_iss); 4144 th->th_ack = htonl(sc->sc_irs + 1); 4145 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4146 th->th_flags = TH_SYN|TH_ACK; 4147 th->th_win = htons(sc->sc_win); 4148 /* th_sum already 0 */ 4149 /* th_urp already 0 */ 4150 4151 /* Tack on the TCP options. */ 4152 optp = (u_int8_t *)(th + 1); 4153 *optp++ = TCPOPT_MAXSEG; 4154 *optp++ = 4; 4155 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4156 *optp++ = sc->sc_ourmaxseg & 0xff; 4157 4158 if (sc->sc_request_r_scale != 15) { 4159 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4160 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4161 sc->sc_request_r_scale); 4162 optp += 4; 4163 } 4164 4165 if (sc->sc_flags & SCF_TIMESTAMP) { 4166 u_int32_t *lp = (u_int32_t *)(optp); 4167 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4168 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4169 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4170 *lp = htonl(sc->sc_timestamp); 4171 optp += TCPOLEN_TSTAMP_APPA; 4172 } 4173 4174 if (sc->sc_flags & SCF_SACK_PERMIT) { 4175 u_int8_t *p = optp; 4176 4177 /* Let the peer know that we will SACK. */ 4178 p[0] = TCPOPT_SACK_PERMITTED; 4179 p[1] = 2; 4180 p[2] = TCPOPT_NOP; 4181 p[3] = TCPOPT_NOP; 4182 optp += 4; 4183 } 4184 4185 #ifdef TCP_SIGNATURE 4186 if (sc->sc_flags & SCF_SIGNATURE) { 4187 struct secasvar *sav; 4188 u_int8_t *sigp; 4189 4190 sav = tcp_signature_getsav(m, th); 4191 4192 if (sav == NULL) { 4193 if (m) 4194 m_freem(m); 4195 return (EPERM); 4196 } 4197 4198 *optp++ = TCPOPT_SIGNATURE; 4199 *optp++ = TCPOLEN_SIGNATURE; 4200 sigp = optp; 4201 bzero(optp, TCP_SIGLEN); 4202 optp += TCP_SIGLEN; 4203 *optp++ = TCPOPT_NOP; 4204 *optp++ = TCPOPT_EOL; 4205 4206 (void)tcp_signature(m, th, hlen, sav, sigp); 4207 4208 key_sa_recordxfer(sav, m); 4209 #ifdef FAST_IPSEC 4210 KEY_FREESAV(&sav); 4211 #else 4212 key_freesav(sav); 4213 #endif 4214 } 4215 #endif 4216 4217 /* Compute the packet's checksum. */ 4218 switch (sc->sc_src.sa.sa_family) { 4219 case AF_INET: 4220 ip->ip_len = htons(tlen - hlen); 4221 th->th_sum = 0; 4222 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4223 break; 4224 #ifdef INET6 4225 case AF_INET6: 4226 ip6->ip6_plen = htons(tlen - hlen); 4227 th->th_sum = 0; 4228 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4229 break; 4230 #endif 4231 } 4232 4233 /* 4234 * Fill in some straggling IP bits. Note the stack expects 4235 * ip_len to be in host order, for convenience. 4236 */ 4237 switch (sc->sc_src.sa.sa_family) { 4238 #ifdef INET 4239 case AF_INET: 4240 ip->ip_len = htons(tlen); 4241 ip->ip_ttl = ip_defttl; 4242 /* XXX tos? */ 4243 break; 4244 #endif 4245 #ifdef INET6 4246 case AF_INET6: 4247 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4248 ip6->ip6_vfc |= IPV6_VERSION; 4249 ip6->ip6_plen = htons(tlen - hlen); 4250 /* ip6_hlim will be initialized afterwards */ 4251 /* XXX flowlabel? */ 4252 break; 4253 #endif 4254 } 4255 4256 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4257 tp = sc->sc_tp; 4258 4259 switch (sc->sc_src.sa.sa_family) { 4260 #ifdef INET 4261 case AF_INET: 4262 error = ip_output(m, sc->sc_ipopts, ro, 4263 (ip_mtudisc ? IP_MTUDISC : 0), 4264 (struct ip_moptions *)NULL, so); 4265 break; 4266 #endif 4267 #ifdef INET6 4268 case AF_INET6: 4269 ip6->ip6_hlim = in6_selecthlim(NULL, 4270 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 4271 4272 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0, 4273 (struct ip6_moptions *)0, so, NULL); 4274 break; 4275 #endif 4276 default: 4277 error = EAFNOSUPPORT; 4278 break; 4279 } 4280 return (error); 4281 } 4282