1 /* $NetBSD: tcp_input.c,v 1.230 2005/06/30 02:58:28 christos 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 1.230 2005/06/30 02:58:28 christos 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 uint32_t ts_rtt; 926 927 MCLAIM(m, &tcp_rx_mowner); 928 va_start(ap, m); 929 toff = va_arg(ap, int); 930 (void)va_arg(ap, int); /* ignore value, advance ap */ 931 va_end(ap); 932 933 tcpstat.tcps_rcvtotal++; 934 935 bzero(&opti, sizeof(opti)); 936 opti.ts_present = 0; 937 opti.maxseg = 0; 938 939 /* 940 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 941 * 942 * TCP is, by definition, unicast, so we reject all 943 * multicast outright. 944 * 945 * Note, there are additional src/dst address checks in 946 * the AF-specific code below. 947 */ 948 if (m->m_flags & (M_BCAST|M_MCAST)) { 949 /* XXX stat */ 950 goto drop; 951 } 952 #ifdef INET6 953 if (m->m_flags & M_ANYCAST6) { 954 /* XXX stat */ 955 goto drop; 956 } 957 #endif 958 959 /* 960 * Get IP and TCP header. 961 * Note: IP leaves IP header in first mbuf. 962 */ 963 ip = mtod(m, struct ip *); 964 #ifdef INET6 965 ip6 = NULL; 966 #endif 967 switch (ip->ip_v) { 968 #ifdef INET 969 case 4: 970 af = AF_INET; 971 iphlen = sizeof(struct ip); 972 ip = mtod(m, struct ip *); 973 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 974 sizeof(struct tcphdr)); 975 if (th == NULL) { 976 tcpstat.tcps_rcvshort++; 977 return; 978 } 979 /* We do the checksum after PCB lookup... */ 980 len = ntohs(ip->ip_len); 981 tlen = len - toff; 982 break; 983 #endif 984 #ifdef INET6 985 case 6: 986 ip = NULL; 987 iphlen = sizeof(struct ip6_hdr); 988 af = AF_INET6; 989 ip6 = mtod(m, struct ip6_hdr *); 990 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 991 sizeof(struct tcphdr)); 992 if (th == NULL) { 993 tcpstat.tcps_rcvshort++; 994 return; 995 } 996 997 /* Be proactive about malicious use of IPv4 mapped address */ 998 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 999 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 1000 /* XXX stat */ 1001 goto drop; 1002 } 1003 1004 /* 1005 * Be proactive about unspecified IPv6 address in source. 1006 * As we use all-zero to indicate unbounded/unconnected pcb, 1007 * unspecified IPv6 address can be used to confuse us. 1008 * 1009 * Note that packets with unspecified IPv6 destination is 1010 * already dropped in ip6_input. 1011 */ 1012 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1013 /* XXX stat */ 1014 goto drop; 1015 } 1016 1017 /* 1018 * Make sure destination address is not multicast. 1019 * Source address checked in ip6_input(). 1020 */ 1021 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1022 /* XXX stat */ 1023 goto drop; 1024 } 1025 1026 /* We do the checksum after PCB lookup... */ 1027 len = m->m_pkthdr.len; 1028 tlen = len - toff; 1029 break; 1030 #endif 1031 default: 1032 m_freem(m); 1033 return; 1034 } 1035 1036 KASSERT(TCP_HDR_ALIGNED_P(th)); 1037 1038 /* 1039 * Check that TCP offset makes sense, 1040 * pull out TCP options and adjust length. XXX 1041 */ 1042 off = th->th_off << 2; 1043 if (off < sizeof (struct tcphdr) || off > tlen) { 1044 tcpstat.tcps_rcvbadoff++; 1045 goto drop; 1046 } 1047 tlen -= off; 1048 1049 /* 1050 * tcp_input() has been modified to use tlen to mean the TCP data 1051 * length throughout the function. Other functions can use 1052 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1053 * rja 1054 */ 1055 1056 if (off > sizeof (struct tcphdr)) { 1057 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1058 if (th == NULL) { 1059 tcpstat.tcps_rcvshort++; 1060 return; 1061 } 1062 /* 1063 * NOTE: ip/ip6 will not be affected by m_pulldown() 1064 * (as they're before toff) and we don't need to update those. 1065 */ 1066 KASSERT(TCP_HDR_ALIGNED_P(th)); 1067 optlen = off - sizeof (struct tcphdr); 1068 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1069 /* 1070 * Do quick retrieval of timestamp options ("options 1071 * prediction?"). If timestamp is the only option and it's 1072 * formatted as recommended in RFC 1323 appendix A, we 1073 * quickly get the values now and not bother calling 1074 * tcp_dooptions(), etc. 1075 */ 1076 if ((optlen == TCPOLEN_TSTAMP_APPA || 1077 (optlen > TCPOLEN_TSTAMP_APPA && 1078 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1079 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1080 (th->th_flags & TH_SYN) == 0) { 1081 opti.ts_present = 1; 1082 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1083 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1084 optp = NULL; /* we've parsed the options */ 1085 } 1086 } 1087 tiflags = th->th_flags; 1088 1089 /* 1090 * Locate pcb for segment. 1091 */ 1092 findpcb: 1093 inp = NULL; 1094 #ifdef INET6 1095 in6p = NULL; 1096 #endif 1097 switch (af) { 1098 #ifdef INET 1099 case AF_INET: 1100 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1101 ip->ip_dst, th->th_dport); 1102 if (inp == 0) { 1103 ++tcpstat.tcps_pcbhashmiss; 1104 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1105 } 1106 #ifdef INET6 1107 if (inp == 0) { 1108 struct in6_addr s, d; 1109 1110 /* mapped addr case */ 1111 bzero(&s, sizeof(s)); 1112 s.s6_addr16[5] = htons(0xffff); 1113 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1114 bzero(&d, sizeof(d)); 1115 d.s6_addr16[5] = htons(0xffff); 1116 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1117 in6p = in6_pcblookup_connect(&tcbtable, &s, 1118 th->th_sport, &d, th->th_dport, 0); 1119 if (in6p == 0) { 1120 ++tcpstat.tcps_pcbhashmiss; 1121 in6p = in6_pcblookup_bind(&tcbtable, &d, 1122 th->th_dport, 0); 1123 } 1124 } 1125 #endif 1126 #ifndef INET6 1127 if (inp == 0) 1128 #else 1129 if (inp == 0 && in6p == 0) 1130 #endif 1131 { 1132 ++tcpstat.tcps_noport; 1133 if (tcp_log_refused && 1134 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1135 tcp4_log_refused(ip, th); 1136 } 1137 TCP_FIELDS_TO_HOST(th); 1138 goto dropwithreset_ratelim; 1139 } 1140 #if defined(IPSEC) || defined(FAST_IPSEC) 1141 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1142 ipsec4_in_reject(m, inp)) { 1143 ipsecstat.in_polvio++; 1144 goto drop; 1145 } 1146 #ifdef INET6 1147 else if (in6p && 1148 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1149 ipsec4_in_reject_so(m, in6p->in6p_socket)) { 1150 ipsecstat.in_polvio++; 1151 goto drop; 1152 } 1153 #endif 1154 #endif /*IPSEC*/ 1155 break; 1156 #endif /*INET*/ 1157 #ifdef INET6 1158 case AF_INET6: 1159 { 1160 int faith; 1161 1162 #if defined(NFAITH) && NFAITH > 0 1163 faith = faithprefix(&ip6->ip6_dst); 1164 #else 1165 faith = 0; 1166 #endif 1167 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1168 th->th_sport, &ip6->ip6_dst, th->th_dport, faith); 1169 if (in6p == NULL) { 1170 ++tcpstat.tcps_pcbhashmiss; 1171 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1172 th->th_dport, faith); 1173 } 1174 if (in6p == NULL) { 1175 ++tcpstat.tcps_noport; 1176 if (tcp_log_refused && 1177 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1178 tcp6_log_refused(ip6, th); 1179 } 1180 TCP_FIELDS_TO_HOST(th); 1181 goto dropwithreset_ratelim; 1182 } 1183 #if defined(IPSEC) || defined(FAST_IPSEC) 1184 if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1185 ipsec6_in_reject(m, in6p)) { 1186 ipsec6stat.in_polvio++; 1187 goto drop; 1188 } 1189 #endif /*IPSEC*/ 1190 break; 1191 } 1192 #endif 1193 } 1194 1195 /* 1196 * If the state is CLOSED (i.e., TCB does not exist) then 1197 * all data in the incoming segment is discarded. 1198 * If the TCB exists but is in CLOSED state, it is embryonic, 1199 * but should either do a listen or a connect soon. 1200 */ 1201 tp = NULL; 1202 so = NULL; 1203 if (inp) { 1204 tp = intotcpcb(inp); 1205 so = inp->inp_socket; 1206 } 1207 #ifdef INET6 1208 else if (in6p) { 1209 tp = in6totcpcb(in6p); 1210 so = in6p->in6p_socket; 1211 } 1212 #endif 1213 if (tp == 0) { 1214 TCP_FIELDS_TO_HOST(th); 1215 goto dropwithreset_ratelim; 1216 } 1217 if (tp->t_state == TCPS_CLOSED) 1218 goto drop; 1219 1220 /* 1221 * Checksum extended TCP header and data. 1222 */ 1223 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1224 goto badcsum; 1225 1226 TCP_FIELDS_TO_HOST(th); 1227 1228 /* Unscale the window into a 32-bit value. */ 1229 if ((tiflags & TH_SYN) == 0) 1230 tiwin = th->th_win << tp->snd_scale; 1231 else 1232 tiwin = th->th_win; 1233 1234 #ifdef INET6 1235 /* save packet options if user wanted */ 1236 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1237 if (in6p->in6p_options) { 1238 m_freem(in6p->in6p_options); 1239 in6p->in6p_options = 0; 1240 } 1241 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1242 } 1243 #endif 1244 1245 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1246 union syn_cache_sa src; 1247 union syn_cache_sa dst; 1248 1249 bzero(&src, sizeof(src)); 1250 bzero(&dst, sizeof(dst)); 1251 switch (af) { 1252 #ifdef INET 1253 case AF_INET: 1254 src.sin.sin_len = sizeof(struct sockaddr_in); 1255 src.sin.sin_family = AF_INET; 1256 src.sin.sin_addr = ip->ip_src; 1257 src.sin.sin_port = th->th_sport; 1258 1259 dst.sin.sin_len = sizeof(struct sockaddr_in); 1260 dst.sin.sin_family = AF_INET; 1261 dst.sin.sin_addr = ip->ip_dst; 1262 dst.sin.sin_port = th->th_dport; 1263 break; 1264 #endif 1265 #ifdef INET6 1266 case AF_INET6: 1267 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1268 src.sin6.sin6_family = AF_INET6; 1269 src.sin6.sin6_addr = ip6->ip6_src; 1270 src.sin6.sin6_port = th->th_sport; 1271 1272 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1273 dst.sin6.sin6_family = AF_INET6; 1274 dst.sin6.sin6_addr = ip6->ip6_dst; 1275 dst.sin6.sin6_port = th->th_dport; 1276 break; 1277 #endif /* INET6 */ 1278 default: 1279 goto badsyn; /*sanity*/ 1280 } 1281 1282 if (so->so_options & SO_DEBUG) { 1283 #ifdef TCP_DEBUG 1284 ostate = tp->t_state; 1285 #endif 1286 1287 tcp_saveti = NULL; 1288 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1289 goto nosave; 1290 1291 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1292 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1293 if (!tcp_saveti) 1294 goto nosave; 1295 } else { 1296 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1297 if (!tcp_saveti) 1298 goto nosave; 1299 MCLAIM(m, &tcp_mowner); 1300 tcp_saveti->m_len = iphlen; 1301 m_copydata(m, 0, iphlen, 1302 mtod(tcp_saveti, caddr_t)); 1303 } 1304 1305 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1306 m_freem(tcp_saveti); 1307 tcp_saveti = NULL; 1308 } else { 1309 tcp_saveti->m_len += sizeof(struct tcphdr); 1310 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1311 sizeof(struct tcphdr)); 1312 } 1313 nosave:; 1314 } 1315 if (so->so_options & SO_ACCEPTCONN) { 1316 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1317 if (tiflags & TH_RST) { 1318 syn_cache_reset(&src.sa, &dst.sa, th); 1319 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1320 (TH_ACK|TH_SYN)) { 1321 /* 1322 * Received a SYN,ACK. This should 1323 * never happen while we are in 1324 * LISTEN. Send an RST. 1325 */ 1326 goto badsyn; 1327 } else if (tiflags & TH_ACK) { 1328 so = syn_cache_get(&src.sa, &dst.sa, 1329 th, toff, tlen, so, m); 1330 if (so == NULL) { 1331 /* 1332 * We don't have a SYN for 1333 * this ACK; send an RST. 1334 */ 1335 goto badsyn; 1336 } else if (so == 1337 (struct socket *)(-1)) { 1338 /* 1339 * We were unable to create 1340 * the connection. If the 1341 * 3-way handshake was 1342 * completed, and RST has 1343 * been sent to the peer. 1344 * Since the mbuf might be 1345 * in use for the reply, 1346 * do not free it. 1347 */ 1348 m = NULL; 1349 } else { 1350 /* 1351 * We have created a 1352 * full-blown connection. 1353 */ 1354 tp = NULL; 1355 inp = NULL; 1356 #ifdef INET6 1357 in6p = NULL; 1358 #endif 1359 switch (so->so_proto->pr_domain->dom_family) { 1360 #ifdef INET 1361 case AF_INET: 1362 inp = sotoinpcb(so); 1363 tp = intotcpcb(inp); 1364 break; 1365 #endif 1366 #ifdef INET6 1367 case AF_INET6: 1368 in6p = sotoin6pcb(so); 1369 tp = in6totcpcb(in6p); 1370 break; 1371 #endif 1372 } 1373 if (tp == NULL) 1374 goto badsyn; /*XXX*/ 1375 tiwin <<= tp->snd_scale; 1376 goto after_listen; 1377 } 1378 } else { 1379 /* 1380 * None of RST, SYN or ACK was set. 1381 * This is an invalid packet for a 1382 * TCB in LISTEN state. Send a RST. 1383 */ 1384 goto badsyn; 1385 } 1386 } else { 1387 /* 1388 * Received a SYN. 1389 */ 1390 1391 #ifdef INET6 1392 /* 1393 * If deprecated address is forbidden, we do 1394 * not accept SYN to deprecated interface 1395 * address to prevent any new inbound 1396 * connection from getting established. 1397 * When we do not accept SYN, we send a TCP 1398 * RST, with deprecated source address (instead 1399 * of dropping it). We compromise it as it is 1400 * much better for peer to send a RST, and 1401 * RST will be the final packet for the 1402 * exchange. 1403 * 1404 * If we do not forbid deprecated addresses, we 1405 * accept the SYN packet. RFC2462 does not 1406 * suggest dropping SYN in this case. 1407 * If we decipher RFC2462 5.5.4, it says like 1408 * this: 1409 * 1. use of deprecated addr with existing 1410 * communication is okay - "SHOULD continue 1411 * to be used" 1412 * 2. use of it with new communication: 1413 * (2a) "SHOULD NOT be used if alternate 1414 * address with sufficient scope is 1415 * available" 1416 * (2b) nothing mentioned otherwise. 1417 * Here we fall into (2b) case as we have no 1418 * choice in our source address selection - we 1419 * must obey the peer. 1420 * 1421 * The wording in RFC2462 is confusing, and 1422 * there are multiple description text for 1423 * deprecated address handling - worse, they 1424 * are not exactly the same. I believe 5.5.4 1425 * is the best one, so we follow 5.5.4. 1426 */ 1427 if (af == AF_INET6 && !ip6_use_deprecated) { 1428 struct in6_ifaddr *ia6; 1429 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1430 &ip6->ip6_dst)) && 1431 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1432 tp = NULL; 1433 goto dropwithreset; 1434 } 1435 } 1436 #endif 1437 1438 #ifdef IPSEC 1439 switch (af) { 1440 #ifdef INET 1441 case AF_INET: 1442 if (ipsec4_in_reject_so(m, so)) { 1443 ipsecstat.in_polvio++; 1444 tp = NULL; 1445 goto dropwithreset; 1446 } 1447 break; 1448 #endif 1449 #ifdef INET6 1450 case AF_INET6: 1451 if (ipsec6_in_reject_so(m, so)) { 1452 ipsec6stat.in_polvio++; 1453 tp = NULL; 1454 goto dropwithreset; 1455 } 1456 break; 1457 #endif 1458 } 1459 #endif 1460 1461 /* 1462 * LISTEN socket received a SYN 1463 * from itself? This can't possibly 1464 * be valid; drop the packet. 1465 */ 1466 if (th->th_sport == th->th_dport) { 1467 int i; 1468 1469 switch (af) { 1470 #ifdef INET 1471 case AF_INET: 1472 i = in_hosteq(ip->ip_src, ip->ip_dst); 1473 break; 1474 #endif 1475 #ifdef INET6 1476 case AF_INET6: 1477 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1478 break; 1479 #endif 1480 default: 1481 i = 1; 1482 } 1483 if (i) { 1484 tcpstat.tcps_badsyn++; 1485 goto drop; 1486 } 1487 } 1488 1489 /* 1490 * SYN looks ok; create compressed TCP 1491 * state for it. 1492 */ 1493 if (so->so_qlen <= so->so_qlimit && 1494 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1495 so, m, optp, optlen, &opti)) 1496 m = NULL; 1497 } 1498 goto drop; 1499 } 1500 } 1501 1502 after_listen: 1503 #ifdef DIAGNOSTIC 1504 /* 1505 * Should not happen now that all embryonic connections 1506 * are handled with compressed state. 1507 */ 1508 if (tp->t_state == TCPS_LISTEN) 1509 panic("tcp_input: TCPS_LISTEN"); 1510 #endif 1511 1512 /* 1513 * Segment received on connection. 1514 * Reset idle time and keep-alive timer. 1515 */ 1516 tp->t_rcvtime = tcp_now; 1517 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1518 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1519 1520 /* 1521 * Process options. 1522 */ 1523 #ifdef TCP_SIGNATURE 1524 if (optp || (tp->t_flags & TF_SIGNATURE)) 1525 #else 1526 if (optp) 1527 #endif 1528 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1529 goto drop; 1530 1531 if (TCP_SACK_ENABLED(tp)) { 1532 tcp_del_sackholes(tp, th); 1533 } 1534 1535 if (opti.ts_present && opti.ts_ecr) { 1536 /* 1537 * Calculate the RTT from the returned time stamp and the 1538 * connection's time base. If the time stamp is later than 1539 * the current time, or is extremely old, fall back to non-1323 1540 * RTT calculation. Since ts_ecr is unsigned, we can test both 1541 * at the same time. 1542 */ 1543 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1544 if (ts_rtt > TCP_PAWS_IDLE) 1545 ts_rtt = 0; 1546 } else { 1547 ts_rtt = 0; 1548 } 1549 1550 /* 1551 * Header prediction: check for the two common cases 1552 * of a uni-directional data xfer. If the packet has 1553 * no control flags, is in-sequence, the window didn't 1554 * change and we're not retransmitting, it's a 1555 * candidate. If the length is zero and the ack moved 1556 * forward, we're the sender side of the xfer. Just 1557 * free the data acked & wake any higher level process 1558 * that was blocked waiting for space. If the length 1559 * is non-zero and the ack didn't move, we're the 1560 * receiver side. If we're getting packets in-order 1561 * (the reassembly queue is empty), add the data to 1562 * the socket buffer and note that we need a delayed ack. 1563 */ 1564 if (tp->t_state == TCPS_ESTABLISHED && 1565 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1566 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1567 th->th_seq == tp->rcv_nxt && 1568 tiwin && tiwin == tp->snd_wnd && 1569 tp->snd_nxt == tp->snd_max) { 1570 1571 /* 1572 * If last ACK falls within this segment's sequence numbers, 1573 * record the timestamp. 1574 * NOTE: 1575 * 1) That the test incorporates suggestions from the latest 1576 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1577 * 2) That updating only on newer timestamps interferes with 1578 * our earlier PAWS tests, so this check should be solely 1579 * predicated on the sequence space of this segment. 1580 * 3) That we modify the segment boundary check to be 1581 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1582 * instead of RFC1323's 1583 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1584 * This modified check allows us to overcome RFC1323's 1585 * limitations as described in Stevens TCP/IP Illustrated 1586 * Vol. 2 p.869. In such cases, we can still calculate the 1587 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1588 */ 1589 if (opti.ts_present && 1590 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1591 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1592 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1593 tp->ts_recent_age = tcp_now; 1594 tp->ts_recent = opti.ts_val; 1595 } 1596 1597 if (tlen == 0) { 1598 /* Ack prediction. */ 1599 if (SEQ_GT(th->th_ack, tp->snd_una) && 1600 SEQ_LEQ(th->th_ack, tp->snd_max) && 1601 tp->snd_cwnd >= tp->snd_wnd && 1602 tp->t_partialacks < 0) { 1603 /* 1604 * this is a pure ack for outstanding data. 1605 */ 1606 ++tcpstat.tcps_predack; 1607 if (ts_rtt) 1608 tcp_xmit_timer(tp, ts_rtt); 1609 else if (tp->t_rtttime && 1610 SEQ_GT(th->th_ack, tp->t_rtseq)) 1611 tcp_xmit_timer(tp, 1612 tcp_now - tp->t_rtttime); 1613 acked = th->th_ack - tp->snd_una; 1614 tcpstat.tcps_rcvackpack++; 1615 tcpstat.tcps_rcvackbyte += acked; 1616 ND6_HINT(tp); 1617 1618 if (acked > (tp->t_lastoff - tp->t_inoff)) 1619 tp->t_lastm = NULL; 1620 sbdrop(&so->so_snd, acked); 1621 tp->t_lastoff -= acked; 1622 1623 tp->snd_una = th->th_ack; 1624 tp->snd_fack = tp->snd_una; 1625 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1626 tp->snd_high = tp->snd_una; 1627 m_freem(m); 1628 1629 /* 1630 * If all outstanding data are acked, stop 1631 * retransmit timer, otherwise restart timer 1632 * using current (possibly backed-off) value. 1633 * If process is waiting for space, 1634 * wakeup/selwakeup/signal. If data 1635 * are ready to send, let tcp_output 1636 * decide between more output or persist. 1637 */ 1638 if (tp->snd_una == tp->snd_max) 1639 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1640 else if (TCP_TIMER_ISARMED(tp, 1641 TCPT_PERSIST) == 0) 1642 TCP_TIMER_ARM(tp, TCPT_REXMT, 1643 tp->t_rxtcur); 1644 1645 sowwakeup(so); 1646 if (so->so_snd.sb_cc) 1647 (void) tcp_output(tp); 1648 if (tcp_saveti) 1649 m_freem(tcp_saveti); 1650 return; 1651 } 1652 } else if (th->th_ack == tp->snd_una && 1653 TAILQ_FIRST(&tp->segq) == NULL && 1654 tlen <= sbspace(&so->so_rcv)) { 1655 /* 1656 * this is a pure, in-sequence data packet 1657 * with nothing on the reassembly queue and 1658 * we have enough buffer space to take it. 1659 */ 1660 ++tcpstat.tcps_preddat; 1661 tp->rcv_nxt += tlen; 1662 tcpstat.tcps_rcvpack++; 1663 tcpstat.tcps_rcvbyte += tlen; 1664 ND6_HINT(tp); 1665 /* 1666 * Drop TCP, IP headers and TCP options then add data 1667 * to socket buffer. 1668 */ 1669 if (so->so_state & SS_CANTRCVMORE) 1670 m_freem(m); 1671 else { 1672 m_adj(m, toff + off); 1673 sbappendstream(&so->so_rcv, m); 1674 } 1675 sorwakeup(so); 1676 TCP_SETUP_ACK(tp, th); 1677 if (tp->t_flags & TF_ACKNOW) 1678 (void) tcp_output(tp); 1679 if (tcp_saveti) 1680 m_freem(tcp_saveti); 1681 return; 1682 } 1683 } 1684 1685 /* 1686 * Compute mbuf offset to TCP data segment. 1687 */ 1688 hdroptlen = toff + off; 1689 1690 /* 1691 * Calculate amount of space in receive window, 1692 * and then do TCP input processing. 1693 * Receive window is amount of space in rcv queue, 1694 * but not less than advertised window. 1695 */ 1696 { int win; 1697 1698 win = sbspace(&so->so_rcv); 1699 if (win < 0) 1700 win = 0; 1701 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1702 } 1703 1704 switch (tp->t_state) { 1705 case TCPS_LISTEN: 1706 /* 1707 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1708 */ 1709 if (m->m_flags & (M_BCAST|M_MCAST)) 1710 goto drop; 1711 switch (af) { 1712 #ifdef INET6 1713 case AF_INET6: 1714 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1715 goto drop; 1716 break; 1717 #endif /* INET6 */ 1718 case AF_INET: 1719 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1720 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1721 goto drop; 1722 break; 1723 } 1724 break; 1725 1726 /* 1727 * If the state is SYN_SENT: 1728 * if seg contains an ACK, but not for our SYN, drop the input. 1729 * if seg contains a RST, then drop the connection. 1730 * if seg does not contain SYN, then drop it. 1731 * Otherwise this is an acceptable SYN segment 1732 * initialize tp->rcv_nxt and tp->irs 1733 * if seg contains ack then advance tp->snd_una 1734 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1735 * arrange for segment to be acked (eventually) 1736 * continue processing rest of data/controls, beginning with URG 1737 */ 1738 case TCPS_SYN_SENT: 1739 if ((tiflags & TH_ACK) && 1740 (SEQ_LEQ(th->th_ack, tp->iss) || 1741 SEQ_GT(th->th_ack, tp->snd_max))) 1742 goto dropwithreset; 1743 if (tiflags & TH_RST) { 1744 if (tiflags & TH_ACK) 1745 tp = tcp_drop(tp, ECONNREFUSED); 1746 goto drop; 1747 } 1748 if ((tiflags & TH_SYN) == 0) 1749 goto drop; 1750 if (tiflags & TH_ACK) { 1751 tp->snd_una = th->th_ack; 1752 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1753 tp->snd_nxt = tp->snd_una; 1754 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1755 tp->snd_high = tp->snd_una; 1756 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1757 } 1758 tp->irs = th->th_seq; 1759 tcp_rcvseqinit(tp); 1760 tp->t_flags |= TF_ACKNOW; 1761 tcp_mss_from_peer(tp, opti.maxseg); 1762 1763 /* 1764 * Initialize the initial congestion window. If we 1765 * had to retransmit the SYN, we must initialize cwnd 1766 * to 1 segment (i.e. the Loss Window). 1767 */ 1768 if (tp->t_flags & TF_SYN_REXMT) 1769 tp->snd_cwnd = tp->t_peermss; 1770 else { 1771 int ss = tcp_init_win; 1772 #ifdef INET 1773 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1774 ss = tcp_init_win_local; 1775 #endif 1776 #ifdef INET6 1777 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1778 ss = tcp_init_win_local; 1779 #endif 1780 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1781 } 1782 1783 tcp_rmx_rtt(tp); 1784 if (tiflags & TH_ACK) { 1785 tcpstat.tcps_connects++; 1786 soisconnected(so); 1787 tcp_established(tp); 1788 /* Do window scaling on this connection? */ 1789 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1790 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1791 tp->snd_scale = tp->requested_s_scale; 1792 tp->rcv_scale = tp->request_r_scale; 1793 } 1794 TCP_REASS_LOCK(tp); 1795 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1796 TCP_REASS_UNLOCK(tp); 1797 /* 1798 * if we didn't have to retransmit the SYN, 1799 * use its rtt as our initial srtt & rtt var. 1800 */ 1801 if (tp->t_rtttime) 1802 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1803 } else 1804 tp->t_state = TCPS_SYN_RECEIVED; 1805 1806 /* 1807 * Advance th->th_seq to correspond to first data byte. 1808 * If data, trim to stay within window, 1809 * dropping FIN if necessary. 1810 */ 1811 th->th_seq++; 1812 if (tlen > tp->rcv_wnd) { 1813 todrop = tlen - tp->rcv_wnd; 1814 m_adj(m, -todrop); 1815 tlen = tp->rcv_wnd; 1816 tiflags &= ~TH_FIN; 1817 tcpstat.tcps_rcvpackafterwin++; 1818 tcpstat.tcps_rcvbyteafterwin += todrop; 1819 } 1820 tp->snd_wl1 = th->th_seq - 1; 1821 tp->rcv_up = th->th_seq; 1822 goto step6; 1823 1824 /* 1825 * If the state is SYN_RECEIVED: 1826 * If seg contains an ACK, but not for our SYN, drop the input 1827 * and generate an RST. See page 36, rfc793 1828 */ 1829 case TCPS_SYN_RECEIVED: 1830 if ((tiflags & TH_ACK) && 1831 (SEQ_LEQ(th->th_ack, tp->iss) || 1832 SEQ_GT(th->th_ack, tp->snd_max))) 1833 goto dropwithreset; 1834 break; 1835 } 1836 1837 /* 1838 * States other than LISTEN or SYN_SENT. 1839 * First check timestamp, if present. 1840 * Then check that at least some bytes of segment are within 1841 * receive window. If segment begins before rcv_nxt, 1842 * drop leading data (and SYN); if nothing left, just ack. 1843 * 1844 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1845 * and it's less than ts_recent, drop it. 1846 */ 1847 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1848 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1849 1850 /* Check to see if ts_recent is over 24 days old. */ 1851 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 1852 /* 1853 * Invalidate ts_recent. If this segment updates 1854 * ts_recent, the age will be reset later and ts_recent 1855 * will get a valid value. If it does not, setting 1856 * ts_recent to zero will at least satisfy the 1857 * requirement that zero be placed in the timestamp 1858 * echo reply when ts_recent isn't valid. The 1859 * age isn't reset until we get a valid ts_recent 1860 * because we don't want out-of-order segments to be 1861 * dropped when ts_recent is old. 1862 */ 1863 tp->ts_recent = 0; 1864 } else { 1865 tcpstat.tcps_rcvduppack++; 1866 tcpstat.tcps_rcvdupbyte += tlen; 1867 tcpstat.tcps_pawsdrop++; 1868 tcp_new_dsack(tp, th->th_seq, tlen); 1869 goto dropafterack; 1870 } 1871 } 1872 1873 todrop = tp->rcv_nxt - th->th_seq; 1874 dupseg = FALSE; 1875 if (todrop > 0) { 1876 if (tiflags & TH_SYN) { 1877 tiflags &= ~TH_SYN; 1878 th->th_seq++; 1879 if (th->th_urp > 1) 1880 th->th_urp--; 1881 else { 1882 tiflags &= ~TH_URG; 1883 th->th_urp = 0; 1884 } 1885 todrop--; 1886 } 1887 if (todrop > tlen || 1888 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1889 /* 1890 * Any valid FIN or RST must be to the left of the 1891 * window. At this point the FIN or RST must be a 1892 * duplicate or out of sequence; drop it. 1893 */ 1894 if (tiflags & TH_RST) 1895 goto drop; 1896 tiflags &= ~(TH_FIN|TH_RST); 1897 /* 1898 * Send an ACK to resynchronize and drop any data. 1899 * But keep on processing for RST or ACK. 1900 */ 1901 tp->t_flags |= TF_ACKNOW; 1902 todrop = tlen; 1903 dupseg = TRUE; 1904 tcpstat.tcps_rcvdupbyte += todrop; 1905 tcpstat.tcps_rcvduppack++; 1906 } else if ((tiflags & TH_RST) && 1907 th->th_seq != tp->last_ack_sent) { 1908 /* 1909 * Test for reset before adjusting the sequence 1910 * number for overlapping data. 1911 */ 1912 goto dropafterack_ratelim; 1913 } else { 1914 tcpstat.tcps_rcvpartduppack++; 1915 tcpstat.tcps_rcvpartdupbyte += todrop; 1916 } 1917 tcp_new_dsack(tp, th->th_seq, todrop); 1918 hdroptlen += todrop; /*drop from head afterwards*/ 1919 th->th_seq += todrop; 1920 tlen -= todrop; 1921 if (th->th_urp > todrop) 1922 th->th_urp -= todrop; 1923 else { 1924 tiflags &= ~TH_URG; 1925 th->th_urp = 0; 1926 } 1927 } 1928 1929 /* 1930 * If new data are received on a connection after the 1931 * user processes are gone, then RST the other end. 1932 */ 1933 if ((so->so_state & SS_NOFDREF) && 1934 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1935 tp = tcp_close(tp); 1936 tcpstat.tcps_rcvafterclose++; 1937 goto dropwithreset; 1938 } 1939 1940 /* 1941 * If segment ends after window, drop trailing data 1942 * (and PUSH and FIN); if nothing left, just ACK. 1943 */ 1944 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1945 if (todrop > 0) { 1946 tcpstat.tcps_rcvpackafterwin++; 1947 if (todrop >= tlen) { 1948 /* 1949 * The segment actually starts after the window. 1950 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 1951 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 1952 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 1953 */ 1954 tcpstat.tcps_rcvbyteafterwin += tlen; 1955 /* 1956 * If a new connection request is received 1957 * while in TIME_WAIT, drop the old connection 1958 * and start over if the sequence numbers 1959 * are above the previous ones. 1960 * 1961 * NOTE: We will checksum the packet again, and 1962 * so we need to put the header fields back into 1963 * network order! 1964 * XXX This kind of sucks, but we don't expect 1965 * XXX this to happen very often, so maybe it 1966 * XXX doesn't matter so much. 1967 */ 1968 if (tiflags & TH_SYN && 1969 tp->t_state == TCPS_TIME_WAIT && 1970 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1971 iss = tcp_new_iss(tp, tp->snd_nxt); 1972 tp = tcp_close(tp); 1973 TCP_FIELDS_TO_NET(th); 1974 goto findpcb; 1975 } 1976 /* 1977 * If window is closed can only take segments at 1978 * window edge, and have to drop data and PUSH from 1979 * incoming segments. Continue processing, but 1980 * remember to ack. Otherwise, drop segment 1981 * and (if not RST) ack. 1982 */ 1983 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1984 tp->t_flags |= TF_ACKNOW; 1985 tcpstat.tcps_rcvwinprobe++; 1986 } else 1987 goto dropafterack; 1988 } else 1989 tcpstat.tcps_rcvbyteafterwin += todrop; 1990 m_adj(m, -todrop); 1991 tlen -= todrop; 1992 tiflags &= ~(TH_PUSH|TH_FIN); 1993 } 1994 1995 /* 1996 * If last ACK falls within this segment's sequence numbers, 1997 * and the timestamp is newer, record it. 1998 */ 1999 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 2000 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2001 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 2002 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2003 tp->ts_recent_age = tcp_now; 2004 tp->ts_recent = opti.ts_val; 2005 } 2006 2007 /* 2008 * If the RST bit is set examine the state: 2009 * SYN_RECEIVED STATE: 2010 * If passive open, return to LISTEN state. 2011 * If active open, inform user that connection was refused. 2012 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 2013 * Inform user that connection was reset, and close tcb. 2014 * CLOSING, LAST_ACK, TIME_WAIT STATES 2015 * Close the tcb. 2016 */ 2017 if (tiflags & TH_RST) { 2018 if (th->th_seq != tp->last_ack_sent) 2019 goto dropafterack_ratelim; 2020 2021 switch (tp->t_state) { 2022 case TCPS_SYN_RECEIVED: 2023 so->so_error = ECONNREFUSED; 2024 goto close; 2025 2026 case TCPS_ESTABLISHED: 2027 case TCPS_FIN_WAIT_1: 2028 case TCPS_FIN_WAIT_2: 2029 case TCPS_CLOSE_WAIT: 2030 so->so_error = ECONNRESET; 2031 close: 2032 tp->t_state = TCPS_CLOSED; 2033 tcpstat.tcps_drops++; 2034 tp = tcp_close(tp); 2035 goto drop; 2036 2037 case TCPS_CLOSING: 2038 case TCPS_LAST_ACK: 2039 case TCPS_TIME_WAIT: 2040 tp = tcp_close(tp); 2041 goto drop; 2042 } 2043 } 2044 2045 /* 2046 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2047 * we must be in a synchronized state. RFC791 states (under RST 2048 * generation) that any unacceptable segment (an out-of-order SYN 2049 * qualifies) received in a synchronized state must elicit only an 2050 * empty acknowledgment segment ... and the connection remains in 2051 * the same state. 2052 */ 2053 if (tiflags & TH_SYN) { 2054 if (tp->rcv_nxt == th->th_seq) { 2055 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2056 TH_ACK); 2057 if (tcp_saveti) 2058 m_freem(tcp_saveti); 2059 return; 2060 } 2061 2062 goto dropafterack_ratelim; 2063 } 2064 2065 /* 2066 * If the ACK bit is off we drop the segment and return. 2067 */ 2068 if ((tiflags & TH_ACK) == 0) { 2069 if (tp->t_flags & TF_ACKNOW) 2070 goto dropafterack; 2071 else 2072 goto drop; 2073 } 2074 2075 /* 2076 * Ack processing. 2077 */ 2078 switch (tp->t_state) { 2079 2080 /* 2081 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2082 * ESTABLISHED state and continue processing, otherwise 2083 * send an RST. 2084 */ 2085 case TCPS_SYN_RECEIVED: 2086 if (SEQ_GT(tp->snd_una, th->th_ack) || 2087 SEQ_GT(th->th_ack, tp->snd_max)) 2088 goto dropwithreset; 2089 tcpstat.tcps_connects++; 2090 soisconnected(so); 2091 tcp_established(tp); 2092 /* Do window scaling? */ 2093 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2094 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2095 tp->snd_scale = tp->requested_s_scale; 2096 tp->rcv_scale = tp->request_r_scale; 2097 } 2098 TCP_REASS_LOCK(tp); 2099 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2100 TCP_REASS_UNLOCK(tp); 2101 tp->snd_wl1 = th->th_seq - 1; 2102 /* fall into ... */ 2103 2104 /* 2105 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2106 * ACKs. If the ack is in the range 2107 * tp->snd_una < th->th_ack <= tp->snd_max 2108 * then advance tp->snd_una to th->th_ack and drop 2109 * data from the retransmission queue. If this ACK reflects 2110 * more up to date window information we update our window information. 2111 */ 2112 case TCPS_ESTABLISHED: 2113 case TCPS_FIN_WAIT_1: 2114 case TCPS_FIN_WAIT_2: 2115 case TCPS_CLOSE_WAIT: 2116 case TCPS_CLOSING: 2117 case TCPS_LAST_ACK: 2118 case TCPS_TIME_WAIT: 2119 2120 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2121 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2122 tcpstat.tcps_rcvdupack++; 2123 /* 2124 * If we have outstanding data (other than 2125 * a window probe), this is a completely 2126 * duplicate ack (ie, window info didn't 2127 * change), the ack is the biggest we've 2128 * seen and we've seen exactly our rexmt 2129 * threshhold of them, assume a packet 2130 * has been dropped and retransmit it. 2131 * Kludge snd_nxt & the congestion 2132 * window so we send only this one 2133 * packet. 2134 * 2135 * We know we're losing at the current 2136 * window size so do congestion avoidance 2137 * (set ssthresh to half the current window 2138 * and pull our congestion window back to 2139 * the new ssthresh). 2140 * 2141 * Dup acks mean that packets have left the 2142 * network (they're now cached at the receiver) 2143 * so bump cwnd by the amount in the receiver 2144 * to keep a constant cwnd packets in the 2145 * network. 2146 * 2147 * If we are using TCP/SACK, then enter 2148 * Fast Recovery if the receiver SACKs 2149 * data that is tcprexmtthresh * MSS 2150 * bytes past the last ACKed segment, 2151 * irrespective of the number of DupAcks. 2152 */ 2153 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2154 th->th_ack != tp->snd_una) 2155 tp->t_dupacks = 0; 2156 else if (tp->t_partialacks < 0 && 2157 (++tp->t_dupacks == tcprexmtthresh || 2158 TCP_FACK_FASTRECOV(tp))) { 2159 tcp_seq onxt; 2160 u_int win; 2161 2162 if (tcp_do_newreno && 2163 SEQ_LT(th->th_ack, tp->snd_high)) { 2164 /* 2165 * False fast retransmit after 2166 * timeout. Do not enter fast 2167 * recovery. 2168 */ 2169 tp->t_dupacks = 0; 2170 break; 2171 } 2172 2173 onxt = tp->snd_nxt; 2174 win = min(tp->snd_wnd, tp->snd_cwnd) / 2175 2 / tp->t_segsz; 2176 if (win < 2) 2177 win = 2; 2178 tp->snd_ssthresh = win * tp->t_segsz; 2179 tp->snd_recover = tp->snd_max; 2180 tp->t_partialacks = 0; 2181 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2182 tp->t_rtttime = 0; 2183 if (TCP_SACK_ENABLED(tp)) { 2184 tp->t_dupacks = tcprexmtthresh; 2185 tp->sack_newdata = tp->snd_nxt; 2186 tp->snd_cwnd = tp->t_segsz; 2187 (void) tcp_output(tp); 2188 goto drop; 2189 } 2190 tp->snd_nxt = th->th_ack; 2191 tp->snd_cwnd = tp->t_segsz; 2192 (void) tcp_output(tp); 2193 tp->snd_cwnd = tp->snd_ssthresh + 2194 tp->t_segsz * tp->t_dupacks; 2195 if (SEQ_GT(onxt, tp->snd_nxt)) 2196 tp->snd_nxt = onxt; 2197 goto drop; 2198 } else if (tp->t_dupacks > tcprexmtthresh) { 2199 tp->snd_cwnd += tp->t_segsz; 2200 (void) tcp_output(tp); 2201 goto drop; 2202 } 2203 } else { 2204 /* 2205 * If the ack appears to be very old, only 2206 * allow data that is in-sequence. This 2207 * makes it somewhat more difficult to insert 2208 * forged data by guessing sequence numbers. 2209 * Sent an ack to try to update the send 2210 * sequence number on the other side. 2211 */ 2212 if (tlen && th->th_seq != tp->rcv_nxt && 2213 SEQ_LT(th->th_ack, 2214 tp->snd_una - tp->max_sndwnd)) 2215 goto dropafterack; 2216 } 2217 break; 2218 } 2219 /* 2220 * If the congestion window was inflated to account 2221 * for the other side's cached packets, retract it. 2222 */ 2223 if (TCP_SACK_ENABLED(tp)) 2224 tcp_sack_newack(tp, th); 2225 else if (tcp_do_newreno) 2226 tcp_newreno_newack(tp, th); 2227 else 2228 tcp_reno_newack(tp, th); 2229 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2230 tcpstat.tcps_rcvacktoomuch++; 2231 goto dropafterack; 2232 } 2233 acked = th->th_ack - tp->snd_una; 2234 tcpstat.tcps_rcvackpack++; 2235 tcpstat.tcps_rcvackbyte += acked; 2236 2237 /* 2238 * If we have a timestamp reply, update smoothed 2239 * round trip time. If no timestamp is present but 2240 * transmit timer is running and timed sequence 2241 * number was acked, update smoothed round trip time. 2242 * Since we now have an rtt measurement, cancel the 2243 * timer backoff (cf., Phil Karn's retransmit alg.). 2244 * Recompute the initial retransmit timer. 2245 */ 2246 if (ts_rtt) 2247 tcp_xmit_timer(tp, ts_rtt); 2248 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2249 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2250 2251 /* 2252 * If all outstanding data is acked, stop retransmit 2253 * timer and remember to restart (more output or persist). 2254 * If there is more data to be acked, restart retransmit 2255 * timer, using current (possibly backed-off) value. 2256 */ 2257 if (th->th_ack == tp->snd_max) { 2258 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2259 needoutput = 1; 2260 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2261 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2262 /* 2263 * When new data is acked, open the congestion window. 2264 * If the window gives us less than ssthresh packets 2265 * in flight, open exponentially (segsz per packet). 2266 * Otherwise open linearly: segsz per window 2267 * (segsz^2 / cwnd per packet). 2268 * 2269 * If we are still in fast recovery (meaning we are using 2270 * NewReno and we have only received partial acks), do not 2271 * inflate the window yet. 2272 */ 2273 if (tp->t_partialacks < 0) { 2274 u_int cw = tp->snd_cwnd; 2275 u_int incr = tp->t_segsz; 2276 2277 if (cw >= tp->snd_ssthresh) 2278 incr = incr * incr / cw; 2279 tp->snd_cwnd = min(cw + incr, 2280 TCP_MAXWIN << tp->snd_scale); 2281 } 2282 ND6_HINT(tp); 2283 if (acked > so->so_snd.sb_cc) { 2284 tp->snd_wnd -= so->so_snd.sb_cc; 2285 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2286 ourfinisacked = 1; 2287 } else { 2288 if (acked > (tp->t_lastoff - tp->t_inoff)) 2289 tp->t_lastm = NULL; 2290 sbdrop(&so->so_snd, acked); 2291 tp->t_lastoff -= acked; 2292 tp->snd_wnd -= acked; 2293 ourfinisacked = 0; 2294 } 2295 sowwakeup(so); 2296 tp->snd_una = th->th_ack; 2297 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2298 tp->snd_fack = tp->snd_una; 2299 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2300 tp->snd_nxt = tp->snd_una; 2301 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2302 tp->snd_high = tp->snd_una; 2303 2304 switch (tp->t_state) { 2305 2306 /* 2307 * In FIN_WAIT_1 STATE in addition to the processing 2308 * for the ESTABLISHED state if our FIN is now acknowledged 2309 * then enter FIN_WAIT_2. 2310 */ 2311 case TCPS_FIN_WAIT_1: 2312 if (ourfinisacked) { 2313 /* 2314 * If we can't receive any more 2315 * data, then closing user can proceed. 2316 * Starting the timer is contrary to the 2317 * specification, but if we don't get a FIN 2318 * we'll hang forever. 2319 */ 2320 if (so->so_state & SS_CANTRCVMORE) { 2321 soisdisconnected(so); 2322 if (tcp_maxidle > 0) 2323 TCP_TIMER_ARM(tp, TCPT_2MSL, 2324 tcp_maxidle); 2325 } 2326 tp->t_state = TCPS_FIN_WAIT_2; 2327 } 2328 break; 2329 2330 /* 2331 * In CLOSING STATE in addition to the processing for 2332 * the ESTABLISHED state if the ACK acknowledges our FIN 2333 * then enter the TIME-WAIT state, otherwise ignore 2334 * the segment. 2335 */ 2336 case TCPS_CLOSING: 2337 if (ourfinisacked) { 2338 tp->t_state = TCPS_TIME_WAIT; 2339 tcp_canceltimers(tp); 2340 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2341 soisdisconnected(so); 2342 } 2343 break; 2344 2345 /* 2346 * In LAST_ACK, we may still be waiting for data to drain 2347 * and/or to be acked, as well as for the ack of our FIN. 2348 * If our FIN is now acknowledged, delete the TCB, 2349 * enter the closed state and return. 2350 */ 2351 case TCPS_LAST_ACK: 2352 if (ourfinisacked) { 2353 tp = tcp_close(tp); 2354 goto drop; 2355 } 2356 break; 2357 2358 /* 2359 * In TIME_WAIT state the only thing that should arrive 2360 * is a retransmission of the remote FIN. Acknowledge 2361 * it and restart the finack timer. 2362 */ 2363 case TCPS_TIME_WAIT: 2364 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2365 goto dropafterack; 2366 } 2367 } 2368 2369 step6: 2370 /* 2371 * Update window information. 2372 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2373 */ 2374 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2375 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 2376 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 2377 /* keep track of pure window updates */ 2378 if (tlen == 0 && 2379 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2380 tcpstat.tcps_rcvwinupd++; 2381 tp->snd_wnd = tiwin; 2382 tp->snd_wl1 = th->th_seq; 2383 tp->snd_wl2 = th->th_ack; 2384 if (tp->snd_wnd > tp->max_sndwnd) 2385 tp->max_sndwnd = tp->snd_wnd; 2386 needoutput = 1; 2387 } 2388 2389 /* 2390 * Process segments with URG. 2391 */ 2392 if ((tiflags & TH_URG) && th->th_urp && 2393 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2394 /* 2395 * This is a kludge, but if we receive and accept 2396 * random urgent pointers, we'll crash in 2397 * soreceive. It's hard to imagine someone 2398 * actually wanting to send this much urgent data. 2399 */ 2400 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2401 th->th_urp = 0; /* XXX */ 2402 tiflags &= ~TH_URG; /* XXX */ 2403 goto dodata; /* XXX */ 2404 } 2405 /* 2406 * If this segment advances the known urgent pointer, 2407 * then mark the data stream. This should not happen 2408 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2409 * a FIN has been received from the remote side. 2410 * In these states we ignore the URG. 2411 * 2412 * According to RFC961 (Assigned Protocols), 2413 * the urgent pointer points to the last octet 2414 * of urgent data. We continue, however, 2415 * to consider it to indicate the first octet 2416 * of data past the urgent section as the original 2417 * spec states (in one of two places). 2418 */ 2419 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2420 tp->rcv_up = th->th_seq + th->th_urp; 2421 so->so_oobmark = so->so_rcv.sb_cc + 2422 (tp->rcv_up - tp->rcv_nxt) - 1; 2423 if (so->so_oobmark == 0) 2424 so->so_state |= SS_RCVATMARK; 2425 sohasoutofband(so); 2426 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2427 } 2428 /* 2429 * Remove out of band data so doesn't get presented to user. 2430 * This can happen independent of advancing the URG pointer, 2431 * but if two URG's are pending at once, some out-of-band 2432 * data may creep in... ick. 2433 */ 2434 if (th->th_urp <= (u_int16_t) tlen 2435 #ifdef SO_OOBINLINE 2436 && (so->so_options & SO_OOBINLINE) == 0 2437 #endif 2438 ) 2439 tcp_pulloutofband(so, th, m, hdroptlen); 2440 } else 2441 /* 2442 * If no out of band data is expected, 2443 * pull receive urgent pointer along 2444 * with the receive window. 2445 */ 2446 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2447 tp->rcv_up = tp->rcv_nxt; 2448 dodata: /* XXX */ 2449 2450 /* 2451 * Process the segment text, merging it into the TCP sequencing queue, 2452 * and arranging for acknowledgement of receipt if necessary. 2453 * This process logically involves adjusting tp->rcv_wnd as data 2454 * is presented to the user (this happens in tcp_usrreq.c, 2455 * case PRU_RCVD). If a FIN has already been received on this 2456 * connection then we just ignore the text. 2457 */ 2458 if ((tlen || (tiflags & TH_FIN)) && 2459 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2460 /* 2461 * Insert segment ti into reassembly queue of tcp with 2462 * control block tp. Return TH_FIN if reassembly now includes 2463 * a segment with FIN. The macro form does the common case 2464 * inline (segment is the next to be received on an 2465 * established connection, and the queue is empty), 2466 * avoiding linkage into and removal from the queue and 2467 * repetition of various conversions. 2468 * Set DELACK for segments received in order, but ack 2469 * immediately when segments are out of order 2470 * (so fast retransmit can work). 2471 */ 2472 /* NOTE: this was TCP_REASS() macro, but used only once */ 2473 TCP_REASS_LOCK(tp); 2474 if (th->th_seq == tp->rcv_nxt && 2475 TAILQ_FIRST(&tp->segq) == NULL && 2476 tp->t_state == TCPS_ESTABLISHED) { 2477 TCP_SETUP_ACK(tp, th); 2478 tp->rcv_nxt += tlen; 2479 tiflags = th->th_flags & TH_FIN; 2480 tcpstat.tcps_rcvpack++; 2481 tcpstat.tcps_rcvbyte += tlen; 2482 ND6_HINT(tp); 2483 if (so->so_state & SS_CANTRCVMORE) 2484 m_freem(m); 2485 else { 2486 m_adj(m, hdroptlen); 2487 sbappendstream(&(so)->so_rcv, m); 2488 } 2489 sorwakeup(so); 2490 } else { 2491 m_adj(m, hdroptlen); 2492 tiflags = tcp_reass(tp, th, m, &tlen); 2493 tp->t_flags |= TF_ACKNOW; 2494 } 2495 TCP_REASS_UNLOCK(tp); 2496 2497 /* 2498 * Note the amount of data that peer has sent into 2499 * our window, in order to estimate the sender's 2500 * buffer size. 2501 */ 2502 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2503 } else { 2504 m_freem(m); 2505 m = NULL; 2506 tiflags &= ~TH_FIN; 2507 } 2508 2509 /* 2510 * If FIN is received ACK the FIN and let the user know 2511 * that the connection is closing. Ignore a FIN received before 2512 * the connection is fully established. 2513 */ 2514 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2515 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2516 socantrcvmore(so); 2517 tp->t_flags |= TF_ACKNOW; 2518 tp->rcv_nxt++; 2519 } 2520 switch (tp->t_state) { 2521 2522 /* 2523 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2524 */ 2525 case TCPS_ESTABLISHED: 2526 tp->t_state = TCPS_CLOSE_WAIT; 2527 break; 2528 2529 /* 2530 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2531 * enter the CLOSING state. 2532 */ 2533 case TCPS_FIN_WAIT_1: 2534 tp->t_state = TCPS_CLOSING; 2535 break; 2536 2537 /* 2538 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2539 * starting the time-wait timer, turning off the other 2540 * standard timers. 2541 */ 2542 case TCPS_FIN_WAIT_2: 2543 tp->t_state = TCPS_TIME_WAIT; 2544 tcp_canceltimers(tp); 2545 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2546 soisdisconnected(so); 2547 break; 2548 2549 /* 2550 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2551 */ 2552 case TCPS_TIME_WAIT: 2553 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2554 break; 2555 } 2556 } 2557 #ifdef TCP_DEBUG 2558 if (so->so_options & SO_DEBUG) 2559 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2560 #endif 2561 2562 /* 2563 * Return any desired output. 2564 */ 2565 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2566 (void) tcp_output(tp); 2567 } 2568 if (tcp_saveti) 2569 m_freem(tcp_saveti); 2570 return; 2571 2572 badsyn: 2573 /* 2574 * Received a bad SYN. Increment counters and dropwithreset. 2575 */ 2576 tcpstat.tcps_badsyn++; 2577 tp = NULL; 2578 goto dropwithreset; 2579 2580 dropafterack: 2581 /* 2582 * Generate an ACK dropping incoming segment if it occupies 2583 * sequence space, where the ACK reflects our state. 2584 */ 2585 if (tiflags & TH_RST) 2586 goto drop; 2587 goto dropafterack2; 2588 2589 dropafterack_ratelim: 2590 /* 2591 * We may want to rate-limit ACKs against SYN/RST attack. 2592 */ 2593 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2594 tcp_ackdrop_ppslim) == 0) { 2595 /* XXX stat */ 2596 goto drop; 2597 } 2598 /* ...fall into dropafterack2... */ 2599 2600 dropafterack2: 2601 m_freem(m); 2602 tp->t_flags |= TF_ACKNOW; 2603 (void) tcp_output(tp); 2604 if (tcp_saveti) 2605 m_freem(tcp_saveti); 2606 return; 2607 2608 dropwithreset_ratelim: 2609 /* 2610 * We may want to rate-limit RSTs in certain situations, 2611 * particularly if we are sending an RST in response to 2612 * an attempt to connect to or otherwise communicate with 2613 * a port for which we have no socket. 2614 */ 2615 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2616 tcp_rst_ppslim) == 0) { 2617 /* XXX stat */ 2618 goto drop; 2619 } 2620 /* ...fall into dropwithreset... */ 2621 2622 dropwithreset: 2623 /* 2624 * Generate a RST, dropping incoming segment. 2625 * Make ACK acceptable to originator of segment. 2626 */ 2627 if (tiflags & TH_RST) 2628 goto drop; 2629 2630 switch (af) { 2631 #ifdef INET6 2632 case AF_INET6: 2633 /* For following calls to tcp_respond */ 2634 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2635 goto drop; 2636 break; 2637 #endif /* INET6 */ 2638 case AF_INET: 2639 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2640 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2641 goto drop; 2642 } 2643 2644 if (tiflags & TH_ACK) 2645 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2646 else { 2647 if (tiflags & TH_SYN) 2648 tlen++; 2649 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2650 TH_RST|TH_ACK); 2651 } 2652 if (tcp_saveti) 2653 m_freem(tcp_saveti); 2654 return; 2655 2656 badcsum: 2657 drop: 2658 /* 2659 * Drop space held by incoming segment and return. 2660 */ 2661 if (tp) { 2662 if (tp->t_inpcb) 2663 so = tp->t_inpcb->inp_socket; 2664 #ifdef INET6 2665 else if (tp->t_in6pcb) 2666 so = tp->t_in6pcb->in6p_socket; 2667 #endif 2668 else 2669 so = NULL; 2670 #ifdef TCP_DEBUG 2671 if (so && (so->so_options & SO_DEBUG) != 0) 2672 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2673 #endif 2674 } 2675 if (tcp_saveti) 2676 m_freem(tcp_saveti); 2677 m_freem(m); 2678 return; 2679 } 2680 2681 #ifdef TCP_SIGNATURE 2682 int 2683 tcp_signature_apply(void *fstate, caddr_t data, u_int len) 2684 { 2685 2686 MD5Update(fstate, (u_char *)data, len); 2687 return (0); 2688 } 2689 2690 struct secasvar * 2691 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 2692 { 2693 struct secasvar *sav; 2694 #ifdef FAST_IPSEC 2695 union sockaddr_union dst; 2696 #endif 2697 struct ip *ip; 2698 struct ip6_hdr *ip6; 2699 2700 ip = mtod(m, struct ip *); 2701 switch (ip->ip_v) { 2702 case 4: 2703 ip = mtod(m, struct ip *); 2704 ip6 = NULL; 2705 break; 2706 case 6: 2707 ip = NULL; 2708 ip6 = mtod(m, struct ip6_hdr *); 2709 break; 2710 default: 2711 return (NULL); 2712 } 2713 2714 #ifdef FAST_IPSEC 2715 /* Extract the destination from the IP header in the mbuf. */ 2716 bzero(&dst, sizeof(union sockaddr_union)); 2717 dst.sa.sa_len = sizeof(struct sockaddr_in); 2718 dst.sa.sa_family = AF_INET; 2719 dst.sin.sin_addr = ip->ip_dst; 2720 2721 /* 2722 * Look up an SADB entry which matches the address of the peer. 2723 */ 2724 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2725 #else 2726 if (ip) 2727 sav = key_allocsa(AF_INET, (caddr_t)&ip->ip_src, 2728 (caddr_t)&ip->ip_dst, IPPROTO_TCP, 2729 htonl(TCP_SIG_SPI), 0, 0); 2730 else 2731 sav = key_allocsa(AF_INET6, (caddr_t)&ip6->ip6_src, 2732 (caddr_t)&ip6->ip6_dst, IPPROTO_TCP, 2733 htonl(TCP_SIG_SPI), 0, 0); 2734 #endif 2735 2736 return (sav); /* freesav must be performed by caller */ 2737 } 2738 2739 int 2740 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 2741 struct secasvar *sav, char *sig) 2742 { 2743 MD5_CTX ctx; 2744 struct ip *ip; 2745 struct ipovly *ipovly; 2746 struct ip6_hdr *ip6; 2747 struct ippseudo ippseudo; 2748 struct ip6_hdr_pseudo ip6pseudo; 2749 struct tcphdr th0; 2750 int l, tcphdrlen; 2751 2752 if (sav == NULL) 2753 return (-1); 2754 2755 tcphdrlen = th->th_off * 4; 2756 2757 switch (mtod(m, struct ip *)->ip_v) { 2758 case 4: 2759 ip = mtod(m, struct ip *); 2760 ip6 = NULL; 2761 break; 2762 case 6: 2763 ip = NULL; 2764 ip6 = mtod(m, struct ip6_hdr *); 2765 break; 2766 default: 2767 return (-1); 2768 } 2769 2770 MD5Init(&ctx); 2771 2772 if (ip) { 2773 memset(&ippseudo, 0, sizeof(ippseudo)); 2774 ipovly = (struct ipovly *)ip; 2775 ippseudo.ippseudo_src = ipovly->ih_src; 2776 ippseudo.ippseudo_dst = ipovly->ih_dst; 2777 ippseudo.ippseudo_pad = 0; 2778 ippseudo.ippseudo_p = IPPROTO_TCP; 2779 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 2780 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 2781 } else { 2782 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 2783 ip6pseudo.ip6ph_src = ip6->ip6_src; 2784 in6_clearscope(&ip6pseudo.ip6ph_src); 2785 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 2786 in6_clearscope(&ip6pseudo.ip6ph_dst); 2787 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 2788 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 2789 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 2790 } 2791 2792 th0 = *th; 2793 th0.th_sum = 0; 2794 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 2795 2796 l = m->m_pkthdr.len - thoff - tcphdrlen; 2797 if (l > 0) 2798 m_apply(m, thoff + tcphdrlen, 2799 m->m_pkthdr.len - thoff - tcphdrlen, 2800 tcp_signature_apply, &ctx); 2801 2802 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2803 MD5Final(sig, &ctx); 2804 2805 return (0); 2806 } 2807 #endif 2808 2809 int 2810 tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcphdr *th, 2811 struct mbuf *m, int toff, struct tcp_opt_info *oi) 2812 { 2813 u_int16_t mss; 2814 int opt, optlen = 0; 2815 #ifdef TCP_SIGNATURE 2816 caddr_t sigp = NULL; 2817 char sigbuf[TCP_SIGLEN]; 2818 struct secasvar *sav = NULL; 2819 #endif 2820 2821 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 2822 opt = cp[0]; 2823 if (opt == TCPOPT_EOL) 2824 break; 2825 if (opt == TCPOPT_NOP) 2826 optlen = 1; 2827 else { 2828 if (cnt < 2) 2829 break; 2830 optlen = cp[1]; 2831 if (optlen < 2 || optlen > cnt) 2832 break; 2833 } 2834 switch (opt) { 2835 2836 default: 2837 continue; 2838 2839 case TCPOPT_MAXSEG: 2840 if (optlen != TCPOLEN_MAXSEG) 2841 continue; 2842 if (!(th->th_flags & TH_SYN)) 2843 continue; 2844 bcopy(cp + 2, &mss, sizeof(mss)); 2845 oi->maxseg = ntohs(mss); 2846 break; 2847 2848 case TCPOPT_WINDOW: 2849 if (optlen != TCPOLEN_WINDOW) 2850 continue; 2851 if (!(th->th_flags & TH_SYN)) 2852 continue; 2853 tp->t_flags |= TF_RCVD_SCALE; 2854 tp->requested_s_scale = cp[2]; 2855 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2856 #if 0 /*XXX*/ 2857 char *p; 2858 2859 if (ip) 2860 p = ntohl(ip->ip_src); 2861 #ifdef INET6 2862 else if (ip6) 2863 p = ip6_sprintf(&ip6->ip6_src); 2864 #endif 2865 else 2866 p = "(unknown)"; 2867 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2868 "assuming %d\n", 2869 tp->requested_s_scale, p, 2870 TCP_MAX_WINSHIFT); 2871 #else 2872 log(LOG_ERR, "TCP: invalid wscale %d, " 2873 "assuming %d\n", 2874 tp->requested_s_scale, 2875 TCP_MAX_WINSHIFT); 2876 #endif 2877 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2878 } 2879 break; 2880 2881 case TCPOPT_TIMESTAMP: 2882 if (optlen != TCPOLEN_TIMESTAMP) 2883 continue; 2884 oi->ts_present = 1; 2885 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2886 NTOHL(oi->ts_val); 2887 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2888 NTOHL(oi->ts_ecr); 2889 2890 /* 2891 * A timestamp received in a SYN makes 2892 * it ok to send timestamp requests and replies. 2893 */ 2894 if (th->th_flags & TH_SYN) { 2895 tp->t_flags |= TF_RCVD_TSTMP; 2896 tp->ts_recent = oi->ts_val; 2897 tp->ts_recent_age = tcp_now; 2898 } 2899 break; 2900 2901 case TCPOPT_SACK_PERMITTED: 2902 if (optlen != TCPOLEN_SACK_PERMITTED) 2903 continue; 2904 if (!(th->th_flags & TH_SYN)) 2905 continue; 2906 if (tcp_do_sack) { 2907 tp->t_flags |= TF_SACK_PERMIT; 2908 tp->t_flags |= TF_WILL_SACK; 2909 } 2910 break; 2911 2912 case TCPOPT_SACK: 2913 tcp_sack_option(tp, th, cp, optlen); 2914 break; 2915 #ifdef TCP_SIGNATURE 2916 case TCPOPT_SIGNATURE: 2917 if (optlen != TCPOLEN_SIGNATURE) 2918 continue; 2919 if (sigp && bcmp(sigp, cp + 2, TCP_SIGLEN)) 2920 return (-1); 2921 2922 sigp = sigbuf; 2923 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 2924 memset(cp + 2, 0, TCP_SIGLEN); 2925 tp->t_flags |= TF_SIGNATURE; 2926 break; 2927 #endif 2928 } 2929 } 2930 2931 #ifdef TCP_SIGNATURE 2932 if (tp->t_flags & TF_SIGNATURE) { 2933 2934 sav = tcp_signature_getsav(m, th); 2935 2936 if (sav == NULL && tp->t_state == TCPS_LISTEN) 2937 return (-1); 2938 } 2939 2940 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 2941 if (sav == NULL) 2942 return (-1); 2943 #ifdef FAST_IPSEC 2944 KEY_FREESAV(&sav); 2945 #else 2946 key_freesav(sav); 2947 #endif 2948 return (-1); 2949 } 2950 2951 if (sigp) { 2952 char sig[TCP_SIGLEN]; 2953 2954 TCP_FIELDS_TO_NET(th); 2955 if (tcp_signature(m, th, toff, sav, sig) < 0) { 2956 TCP_FIELDS_TO_HOST(th); 2957 if (sav == NULL) 2958 return (-1); 2959 #ifdef FAST_IPSEC 2960 KEY_FREESAV(&sav); 2961 #else 2962 key_freesav(sav); 2963 #endif 2964 return (-1); 2965 } 2966 TCP_FIELDS_TO_HOST(th); 2967 2968 if (bcmp(sig, sigp, TCP_SIGLEN)) { 2969 tcpstat.tcps_badsig++; 2970 if (sav == NULL) 2971 return (-1); 2972 #ifdef FAST_IPSEC 2973 KEY_FREESAV(&sav); 2974 #else 2975 key_freesav(sav); 2976 #endif 2977 return (-1); 2978 } else 2979 tcpstat.tcps_goodsig++; 2980 2981 key_sa_recordxfer(sav, m); 2982 #ifdef FAST_IPSEC 2983 KEY_FREESAV(&sav); 2984 #else 2985 key_freesav(sav); 2986 #endif 2987 } 2988 #endif 2989 2990 return (0); 2991 } 2992 2993 /* 2994 * Pull out of band byte out of a segment so 2995 * it doesn't appear in the user's data queue. 2996 * It is still reflected in the segment length for 2997 * sequencing purposes. 2998 */ 2999 void 3000 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3001 struct mbuf *m, int off) 3002 { 3003 int cnt = off + th->th_urp - 1; 3004 3005 while (cnt >= 0) { 3006 if (m->m_len > cnt) { 3007 char *cp = mtod(m, caddr_t) + cnt; 3008 struct tcpcb *tp = sototcpcb(so); 3009 3010 tp->t_iobc = *cp; 3011 tp->t_oobflags |= TCPOOB_HAVEDATA; 3012 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3013 m->m_len--; 3014 return; 3015 } 3016 cnt -= m->m_len; 3017 m = m->m_next; 3018 if (m == 0) 3019 break; 3020 } 3021 panic("tcp_pulloutofband"); 3022 } 3023 3024 /* 3025 * Collect new round-trip time estimate 3026 * and update averages and current timeout. 3027 */ 3028 void 3029 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3030 { 3031 int32_t delta; 3032 3033 tcpstat.tcps_rttupdated++; 3034 if (tp->t_srtt != 0) { 3035 /* 3036 * srtt is stored as fixed point with 3 bits after the 3037 * binary point (i.e., scaled by 8). The following magic 3038 * is equivalent to the smoothing algorithm in rfc793 with 3039 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3040 * point). Adjust rtt to origin 0. 3041 */ 3042 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3043 if ((tp->t_srtt += delta) <= 0) 3044 tp->t_srtt = 1 << 2; 3045 /* 3046 * We accumulate a smoothed rtt variance (actually, a 3047 * smoothed mean difference), then set the retransmit 3048 * timer to smoothed rtt + 4 times the smoothed variance. 3049 * rttvar is stored as fixed point with 2 bits after the 3050 * binary point (scaled by 4). The following is 3051 * equivalent to rfc793 smoothing with an alpha of .75 3052 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3053 * rfc793's wired-in beta. 3054 */ 3055 if (delta < 0) 3056 delta = -delta; 3057 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3058 if ((tp->t_rttvar += delta) <= 0) 3059 tp->t_rttvar = 1 << 2; 3060 } else { 3061 /* 3062 * No rtt measurement yet - use the unsmoothed rtt. 3063 * Set the variance to half the rtt (so our first 3064 * retransmit happens at 3*rtt). 3065 */ 3066 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3067 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3068 } 3069 tp->t_rtttime = 0; 3070 tp->t_rxtshift = 0; 3071 3072 /* 3073 * the retransmit should happen at rtt + 4 * rttvar. 3074 * Because of the way we do the smoothing, srtt and rttvar 3075 * will each average +1/2 tick of bias. When we compute 3076 * the retransmit timer, we want 1/2 tick of rounding and 3077 * 1 extra tick because of +-1/2 tick uncertainty in the 3078 * firing of the timer. The bias will give us exactly the 3079 * 1.5 tick we need. But, because the bias is 3080 * statistical, we have to test that we don't drop below 3081 * the minimum feasible timer (which is 2 ticks). 3082 */ 3083 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3084 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3085 3086 /* 3087 * We received an ack for a packet that wasn't retransmitted; 3088 * it is probably safe to discard any error indications we've 3089 * received recently. This isn't quite right, but close enough 3090 * for now (a route might have failed after we sent a segment, 3091 * and the return path might not be symmetrical). 3092 */ 3093 tp->t_softerror = 0; 3094 } 3095 3096 void 3097 tcp_reno_newack(struct tcpcb *tp, struct tcphdr *th) 3098 { 3099 if (tp->t_partialacks < 0) { 3100 /* 3101 * We were not in fast recovery. Reset the duplicate ack 3102 * counter. 3103 */ 3104 tp->t_dupacks = 0; 3105 } else { 3106 /* 3107 * Clamp the congestion window to the crossover point and 3108 * exit fast recovery. 3109 */ 3110 if (tp->snd_cwnd > tp->snd_ssthresh) 3111 tp->snd_cwnd = tp->snd_ssthresh; 3112 tp->t_partialacks = -1; 3113 tp->t_dupacks = 0; 3114 } 3115 } 3116 3117 /* 3118 * Implement the NewReno response to a new ack, checking for partial acks in 3119 * fast recovery. 3120 */ 3121 void 3122 tcp_newreno_newack(struct tcpcb *tp, struct tcphdr *th) 3123 { 3124 if (tp->t_partialacks < 0) { 3125 /* 3126 * We were not in fast recovery. Reset the duplicate ack 3127 * counter. 3128 */ 3129 tp->t_dupacks = 0; 3130 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) { 3131 /* 3132 * This is a partial ack. Retransmit the first unacknowledged 3133 * segment and deflate the congestion window by the amount of 3134 * acknowledged data. Do not exit fast recovery. 3135 */ 3136 tcp_seq onxt = tp->snd_nxt; 3137 u_long ocwnd = tp->snd_cwnd; 3138 3139 /* 3140 * snd_una has not yet been updated and the socket's send 3141 * buffer has not yet drained off the ACK'd data, so we 3142 * have to leave snd_una as it was to get the correct data 3143 * offset in tcp_output(). 3144 */ 3145 if (++tp->t_partialacks == 1) 3146 TCP_TIMER_DISARM(tp, TCPT_REXMT); 3147 tp->t_rtttime = 0; 3148 tp->snd_nxt = th->th_ack; 3149 /* 3150 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 3151 * is not yet updated when we're called. 3152 */ 3153 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 3154 (void) tcp_output(tp); 3155 tp->snd_cwnd = ocwnd; 3156 if (SEQ_GT(onxt, tp->snd_nxt)) 3157 tp->snd_nxt = onxt; 3158 /* 3159 * Partial window deflation. Relies on fact that tp->snd_una 3160 * not updated yet. 3161 */ 3162 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 3163 } else { 3164 /* 3165 * Complete ack. Inflate the congestion window to ssthresh 3166 * and exit fast recovery. 3167 * 3168 * Window inflation should have left us with approx. 3169 * snd_ssthresh outstanding data. But in case we 3170 * would be inclined to send a burst, better to do 3171 * it via the slow start mechanism. 3172 */ 3173 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 3174 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 3175 + tp->t_segsz; 3176 else 3177 tp->snd_cwnd = tp->snd_ssthresh; 3178 tp->t_partialacks = -1; 3179 tp->t_dupacks = 0; 3180 } 3181 } 3182 3183 3184 /* 3185 * TCP compressed state engine. Currently used to hold compressed 3186 * state for SYN_RECEIVED. 3187 */ 3188 3189 u_long syn_cache_count; 3190 u_int32_t syn_hash1, syn_hash2; 3191 3192 #define SYN_HASH(sa, sp, dp) \ 3193 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3194 ((u_int32_t)(sp)))^syn_hash2))) 3195 #ifndef INET6 3196 #define SYN_HASHALL(hash, src, dst) \ 3197 do { \ 3198 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3199 ((const struct sockaddr_in *)(src))->sin_port, \ 3200 ((const struct sockaddr_in *)(dst))->sin_port); \ 3201 } while (/*CONSTCOND*/ 0) 3202 #else 3203 #define SYN_HASH6(sa, sp, dp) \ 3204 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3205 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3206 & 0x7fffffff) 3207 3208 #define SYN_HASHALL(hash, src, dst) \ 3209 do { \ 3210 switch ((src)->sa_family) { \ 3211 case AF_INET: \ 3212 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3213 ((const struct sockaddr_in *)(src))->sin_port, \ 3214 ((const struct sockaddr_in *)(dst))->sin_port); \ 3215 break; \ 3216 case AF_INET6: \ 3217 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3218 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3219 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3220 break; \ 3221 default: \ 3222 hash = 0; \ 3223 } \ 3224 } while (/*CONSTCOND*/0) 3225 #endif /* INET6 */ 3226 3227 #define SYN_CACHE_RM(sc) \ 3228 do { \ 3229 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 3230 (sc), sc_bucketq); \ 3231 (sc)->sc_tp = NULL; \ 3232 LIST_REMOVE((sc), sc_tpq); \ 3233 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 3234 callout_stop(&(sc)->sc_timer); \ 3235 syn_cache_count--; \ 3236 } while (/*CONSTCOND*/0) 3237 3238 #define SYN_CACHE_PUT(sc) \ 3239 do { \ 3240 if ((sc)->sc_ipopts) \ 3241 (void) m_free((sc)->sc_ipopts); \ 3242 if ((sc)->sc_route4.ro_rt != NULL) \ 3243 RTFREE((sc)->sc_route4.ro_rt); \ 3244 if (callout_invoking(&(sc)->sc_timer)) \ 3245 (sc)->sc_flags |= SCF_DEAD; \ 3246 else \ 3247 pool_put(&syn_cache_pool, (sc)); \ 3248 } while (/*CONSTCOND*/0) 3249 3250 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL); 3251 3252 /* 3253 * We don't estimate RTT with SYNs, so each packet starts with the default 3254 * RTT and each timer step has a fixed timeout value. 3255 */ 3256 #define SYN_CACHE_TIMER_ARM(sc) \ 3257 do { \ 3258 TCPT_RANGESET((sc)->sc_rxtcur, \ 3259 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3260 TCPTV_REXMTMAX); \ 3261 callout_reset(&(sc)->sc_timer, \ 3262 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3263 } while (/*CONSTCOND*/0) 3264 3265 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3266 3267 void 3268 syn_cache_init(void) 3269 { 3270 int i; 3271 3272 /* Initialize the hash buckets. */ 3273 for (i = 0; i < tcp_syn_cache_size; i++) 3274 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3275 } 3276 3277 void 3278 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3279 { 3280 struct syn_cache_head *scp; 3281 struct syn_cache *sc2; 3282 int s; 3283 3284 /* 3285 * If there are no entries in the hash table, reinitialize 3286 * the hash secrets. 3287 */ 3288 if (syn_cache_count == 0) { 3289 syn_hash1 = arc4random(); 3290 syn_hash2 = arc4random(); 3291 } 3292 3293 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3294 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3295 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3296 3297 /* 3298 * Make sure that we don't overflow the per-bucket 3299 * limit or the total cache size limit. 3300 */ 3301 s = splsoftnet(); 3302 if (scp->sch_length >= tcp_syn_bucket_limit) { 3303 tcpstat.tcps_sc_bucketoverflow++; 3304 /* 3305 * The bucket is full. Toss the oldest element in the 3306 * bucket. This will be the first entry in the bucket. 3307 */ 3308 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3309 #ifdef DIAGNOSTIC 3310 /* 3311 * This should never happen; we should always find an 3312 * entry in our bucket. 3313 */ 3314 if (sc2 == NULL) 3315 panic("syn_cache_insert: bucketoverflow: impossible"); 3316 #endif 3317 SYN_CACHE_RM(sc2); 3318 SYN_CACHE_PUT(sc2); 3319 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3320 struct syn_cache_head *scp2, *sce; 3321 3322 tcpstat.tcps_sc_overflowed++; 3323 /* 3324 * The cache is full. Toss the oldest entry in the 3325 * first non-empty bucket we can find. 3326 * 3327 * XXX We would really like to toss the oldest 3328 * entry in the cache, but we hope that this 3329 * condition doesn't happen very often. 3330 */ 3331 scp2 = scp; 3332 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3333 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3334 for (++scp2; scp2 != scp; scp2++) { 3335 if (scp2 >= sce) 3336 scp2 = &tcp_syn_cache[0]; 3337 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3338 break; 3339 } 3340 #ifdef DIAGNOSTIC 3341 /* 3342 * This should never happen; we should always find a 3343 * non-empty bucket. 3344 */ 3345 if (scp2 == scp) 3346 panic("syn_cache_insert: cacheoverflow: " 3347 "impossible"); 3348 #endif 3349 } 3350 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3351 SYN_CACHE_RM(sc2); 3352 SYN_CACHE_PUT(sc2); 3353 } 3354 3355 /* 3356 * Initialize the entry's timer. 3357 */ 3358 sc->sc_rxttot = 0; 3359 sc->sc_rxtshift = 0; 3360 SYN_CACHE_TIMER_ARM(sc); 3361 3362 /* Link it from tcpcb entry */ 3363 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3364 3365 /* Put it into the bucket. */ 3366 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3367 scp->sch_length++; 3368 syn_cache_count++; 3369 3370 tcpstat.tcps_sc_added++; 3371 splx(s); 3372 } 3373 3374 /* 3375 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3376 * If we have retransmitted an entry the maximum number of times, expire 3377 * that entry. 3378 */ 3379 void 3380 syn_cache_timer(void *arg) 3381 { 3382 struct syn_cache *sc = arg; 3383 int s; 3384 3385 s = splsoftnet(); 3386 callout_ack(&sc->sc_timer); 3387 3388 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3389 tcpstat.tcps_sc_delayed_free++; 3390 pool_put(&syn_cache_pool, sc); 3391 splx(s); 3392 return; 3393 } 3394 3395 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3396 /* Drop it -- too many retransmissions. */ 3397 goto dropit; 3398 } 3399 3400 /* 3401 * Compute the total amount of time this entry has 3402 * been on a queue. If this entry has been on longer 3403 * than the keep alive timer would allow, expire it. 3404 */ 3405 sc->sc_rxttot += sc->sc_rxtcur; 3406 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) 3407 goto dropit; 3408 3409 tcpstat.tcps_sc_retransmitted++; 3410 (void) syn_cache_respond(sc, NULL); 3411 3412 /* Advance the timer back-off. */ 3413 sc->sc_rxtshift++; 3414 SYN_CACHE_TIMER_ARM(sc); 3415 3416 splx(s); 3417 return; 3418 3419 dropit: 3420 tcpstat.tcps_sc_timed_out++; 3421 SYN_CACHE_RM(sc); 3422 SYN_CACHE_PUT(sc); 3423 splx(s); 3424 } 3425 3426 /* 3427 * Remove syn cache created by the specified tcb entry, 3428 * because this does not make sense to keep them 3429 * (if there's no tcb entry, syn cache entry will never be used) 3430 */ 3431 void 3432 syn_cache_cleanup(struct tcpcb *tp) 3433 { 3434 struct syn_cache *sc, *nsc; 3435 int s; 3436 3437 s = splsoftnet(); 3438 3439 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3440 nsc = LIST_NEXT(sc, sc_tpq); 3441 3442 #ifdef DIAGNOSTIC 3443 if (sc->sc_tp != tp) 3444 panic("invalid sc_tp in syn_cache_cleanup"); 3445 #endif 3446 SYN_CACHE_RM(sc); 3447 SYN_CACHE_PUT(sc); 3448 } 3449 /* just for safety */ 3450 LIST_INIT(&tp->t_sc); 3451 3452 splx(s); 3453 } 3454 3455 /* 3456 * Find an entry in the syn cache. 3457 */ 3458 struct syn_cache * 3459 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3460 struct syn_cache_head **headp) 3461 { 3462 struct syn_cache *sc; 3463 struct syn_cache_head *scp; 3464 u_int32_t hash; 3465 int s; 3466 3467 SYN_HASHALL(hash, src, dst); 3468 3469 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3470 *headp = scp; 3471 s = splsoftnet(); 3472 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3473 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3474 if (sc->sc_hash != hash) 3475 continue; 3476 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3477 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3478 splx(s); 3479 return (sc); 3480 } 3481 } 3482 splx(s); 3483 return (NULL); 3484 } 3485 3486 /* 3487 * This function gets called when we receive an ACK for a 3488 * socket in the LISTEN state. We look up the connection 3489 * in the syn cache, and if its there, we pull it out of 3490 * the cache and turn it into a full-blown connection in 3491 * the SYN-RECEIVED state. 3492 * 3493 * The return values may not be immediately obvious, and their effects 3494 * can be subtle, so here they are: 3495 * 3496 * NULL SYN was not found in cache; caller should drop the 3497 * packet and send an RST. 3498 * 3499 * -1 We were unable to create the new connection, and are 3500 * aborting it. An ACK,RST is being sent to the peer 3501 * (unless we got screwey sequence numbners; see below), 3502 * because the 3-way handshake has been completed. Caller 3503 * should not free the mbuf, since we may be using it. If 3504 * we are not, we will free it. 3505 * 3506 * Otherwise, the return value is a pointer to the new socket 3507 * associated with the connection. 3508 */ 3509 struct socket * 3510 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3511 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3512 struct socket *so, struct mbuf *m) 3513 { 3514 struct syn_cache *sc; 3515 struct syn_cache_head *scp; 3516 struct inpcb *inp = NULL; 3517 #ifdef INET6 3518 struct in6pcb *in6p = NULL; 3519 #endif 3520 struct tcpcb *tp = 0; 3521 struct mbuf *am; 3522 int s; 3523 struct socket *oso; 3524 3525 s = splsoftnet(); 3526 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3527 splx(s); 3528 return (NULL); 3529 } 3530 3531 /* 3532 * Verify the sequence and ack numbers. Try getting the correct 3533 * response again. 3534 */ 3535 if ((th->th_ack != sc->sc_iss + 1) || 3536 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3537 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3538 (void) syn_cache_respond(sc, m); 3539 splx(s); 3540 return ((struct socket *)(-1)); 3541 } 3542 3543 /* Remove this cache entry */ 3544 SYN_CACHE_RM(sc); 3545 splx(s); 3546 3547 /* 3548 * Ok, create the full blown connection, and set things up 3549 * as they would have been set up if we had created the 3550 * connection when the SYN arrived. If we can't create 3551 * the connection, abort it. 3552 */ 3553 /* 3554 * inp still has the OLD in_pcb stuff, set the 3555 * v6-related flags on the new guy, too. This is 3556 * done particularly for the case where an AF_INET6 3557 * socket is bound only to a port, and a v4 connection 3558 * comes in on that port. 3559 * we also copy the flowinfo from the original pcb 3560 * to the new one. 3561 */ 3562 oso = so; 3563 so = sonewconn(so, SS_ISCONNECTED); 3564 if (so == NULL) 3565 goto resetandabort; 3566 3567 switch (so->so_proto->pr_domain->dom_family) { 3568 #ifdef INET 3569 case AF_INET: 3570 inp = sotoinpcb(so); 3571 break; 3572 #endif 3573 #ifdef INET6 3574 case AF_INET6: 3575 in6p = sotoin6pcb(so); 3576 break; 3577 #endif 3578 } 3579 switch (src->sa_family) { 3580 #ifdef INET 3581 case AF_INET: 3582 if (inp) { 3583 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3584 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3585 inp->inp_options = ip_srcroute(); 3586 in_pcbstate(inp, INP_BOUND); 3587 if (inp->inp_options == NULL) { 3588 inp->inp_options = sc->sc_ipopts; 3589 sc->sc_ipopts = NULL; 3590 } 3591 } 3592 #ifdef INET6 3593 else if (in6p) { 3594 /* IPv4 packet to AF_INET6 socket */ 3595 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3596 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3597 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3598 &in6p->in6p_laddr.s6_addr32[3], 3599 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3600 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3601 in6totcpcb(in6p)->t_family = AF_INET; 3602 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3603 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3604 else 3605 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3606 in6_pcbstate(in6p, IN6P_BOUND); 3607 } 3608 #endif 3609 break; 3610 #endif 3611 #ifdef INET6 3612 case AF_INET6: 3613 if (in6p) { 3614 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3615 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3616 in6_pcbstate(in6p, IN6P_BOUND); 3617 } 3618 break; 3619 #endif 3620 } 3621 #ifdef INET6 3622 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3623 struct in6pcb *oin6p = sotoin6pcb(oso); 3624 /* inherit socket options from the listening socket */ 3625 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3626 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3627 m_freem(in6p->in6p_options); 3628 in6p->in6p_options = 0; 3629 } 3630 ip6_savecontrol(in6p, &in6p->in6p_options, 3631 mtod(m, struct ip6_hdr *), m); 3632 } 3633 #endif 3634 3635 #if defined(IPSEC) || defined(FAST_IPSEC) 3636 /* 3637 * we make a copy of policy, instead of sharing the policy, 3638 * for better behavior in terms of SA lookup and dead SA removal. 3639 */ 3640 if (inp) { 3641 /* copy old policy into new socket's */ 3642 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3643 printf("tcp_input: could not copy policy\n"); 3644 } 3645 #ifdef INET6 3646 else if (in6p) { 3647 /* copy old policy into new socket's */ 3648 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3649 in6p->in6p_sp)) 3650 printf("tcp_input: could not copy policy\n"); 3651 } 3652 #endif 3653 #endif 3654 3655 /* 3656 * Give the new socket our cached route reference. 3657 */ 3658 if (inp) 3659 inp->inp_route = sc->sc_route4; /* struct assignment */ 3660 #ifdef INET6 3661 else 3662 in6p->in6p_route = sc->sc_route6; 3663 #endif 3664 sc->sc_route4.ro_rt = NULL; 3665 3666 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3667 if (am == NULL) 3668 goto resetandabort; 3669 MCLAIM(am, &tcp_mowner); 3670 am->m_len = src->sa_len; 3671 bcopy(src, mtod(am, caddr_t), src->sa_len); 3672 if (inp) { 3673 if (in_pcbconnect(inp, am)) { 3674 (void) m_free(am); 3675 goto resetandabort; 3676 } 3677 } 3678 #ifdef INET6 3679 else if (in6p) { 3680 if (src->sa_family == AF_INET) { 3681 /* IPv4 packet to AF_INET6 socket */ 3682 struct sockaddr_in6 *sin6; 3683 sin6 = mtod(am, struct sockaddr_in6 *); 3684 am->m_len = sizeof(*sin6); 3685 bzero(sin6, sizeof(*sin6)); 3686 sin6->sin6_family = AF_INET6; 3687 sin6->sin6_len = sizeof(*sin6); 3688 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3689 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3690 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3691 &sin6->sin6_addr.s6_addr32[3], 3692 sizeof(sin6->sin6_addr.s6_addr32[3])); 3693 } 3694 if (in6_pcbconnect(in6p, am)) { 3695 (void) m_free(am); 3696 goto resetandabort; 3697 } 3698 } 3699 #endif 3700 else { 3701 (void) m_free(am); 3702 goto resetandabort; 3703 } 3704 (void) m_free(am); 3705 3706 if (inp) 3707 tp = intotcpcb(inp); 3708 #ifdef INET6 3709 else if (in6p) 3710 tp = in6totcpcb(in6p); 3711 #endif 3712 else 3713 tp = NULL; 3714 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3715 if (sc->sc_request_r_scale != 15) { 3716 tp->requested_s_scale = sc->sc_requested_s_scale; 3717 tp->request_r_scale = sc->sc_request_r_scale; 3718 tp->snd_scale = sc->sc_requested_s_scale; 3719 tp->rcv_scale = sc->sc_request_r_scale; 3720 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3721 } 3722 if (sc->sc_flags & SCF_TIMESTAMP) 3723 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3724 tp->ts_timebase = sc->sc_timebase; 3725 3726 tp->t_template = tcp_template(tp); 3727 if (tp->t_template == 0) { 3728 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3729 so = NULL; 3730 m_freem(m); 3731 goto abort; 3732 } 3733 3734 tp->iss = sc->sc_iss; 3735 tp->irs = sc->sc_irs; 3736 tcp_sendseqinit(tp); 3737 tcp_rcvseqinit(tp); 3738 tp->t_state = TCPS_SYN_RECEIVED; 3739 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3740 tcpstat.tcps_accepts++; 3741 3742 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 3743 tp->t_flags |= TF_WILL_SACK; 3744 3745 #ifdef TCP_SIGNATURE 3746 if (sc->sc_flags & SCF_SIGNATURE) 3747 tp->t_flags |= TF_SIGNATURE; 3748 #endif 3749 3750 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3751 tp->t_ourmss = sc->sc_ourmaxseg; 3752 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3753 3754 /* 3755 * Initialize the initial congestion window. If we 3756 * had to retransmit the SYN,ACK, we must initialize cwnd 3757 * to 1 segment (i.e. the Loss Window). 3758 */ 3759 if (sc->sc_rxtshift) 3760 tp->snd_cwnd = tp->t_peermss; 3761 else { 3762 int ss = tcp_init_win; 3763 #ifdef INET 3764 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3765 ss = tcp_init_win_local; 3766 #endif 3767 #ifdef INET6 3768 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3769 ss = tcp_init_win_local; 3770 #endif 3771 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3772 } 3773 3774 tcp_rmx_rtt(tp); 3775 tp->snd_wl1 = sc->sc_irs; 3776 tp->rcv_up = sc->sc_irs + 1; 3777 3778 /* 3779 * This is what whould have happened in tcp_output() when 3780 * the SYN,ACK was sent. 3781 */ 3782 tp->snd_up = tp->snd_una; 3783 tp->snd_max = tp->snd_nxt = tp->iss+1; 3784 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3785 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3786 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3787 tp->last_ack_sent = tp->rcv_nxt; 3788 tp->t_partialacks = -1; 3789 tp->t_dupacks = 0; 3790 3791 tcpstat.tcps_sc_completed++; 3792 SYN_CACHE_PUT(sc); 3793 return (so); 3794 3795 resetandabort: 3796 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3797 abort: 3798 if (so != NULL) 3799 (void) soabort(so); 3800 SYN_CACHE_PUT(sc); 3801 tcpstat.tcps_sc_aborted++; 3802 return ((struct socket *)(-1)); 3803 } 3804 3805 /* 3806 * This function is called when we get a RST for a 3807 * non-existent connection, so that we can see if the 3808 * connection is in the syn cache. If it is, zap it. 3809 */ 3810 3811 void 3812 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 3813 { 3814 struct syn_cache *sc; 3815 struct syn_cache_head *scp; 3816 int s = splsoftnet(); 3817 3818 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3819 splx(s); 3820 return; 3821 } 3822 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3823 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3824 splx(s); 3825 return; 3826 } 3827 SYN_CACHE_RM(sc); 3828 splx(s); 3829 tcpstat.tcps_sc_reset++; 3830 SYN_CACHE_PUT(sc); 3831 } 3832 3833 void 3834 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 3835 struct tcphdr *th) 3836 { 3837 struct syn_cache *sc; 3838 struct syn_cache_head *scp; 3839 int s; 3840 3841 s = splsoftnet(); 3842 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3843 splx(s); 3844 return; 3845 } 3846 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3847 if (ntohl (th->th_seq) != sc->sc_iss) { 3848 splx(s); 3849 return; 3850 } 3851 3852 /* 3853 * If we've retransmitted 3 times and this is our second error, 3854 * we remove the entry. Otherwise, we allow it to continue on. 3855 * This prevents us from incorrectly nuking an entry during a 3856 * spurious network outage. 3857 * 3858 * See tcp_notify(). 3859 */ 3860 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3861 sc->sc_flags |= SCF_UNREACH; 3862 splx(s); 3863 return; 3864 } 3865 3866 SYN_CACHE_RM(sc); 3867 splx(s); 3868 tcpstat.tcps_sc_unreach++; 3869 SYN_CACHE_PUT(sc); 3870 } 3871 3872 /* 3873 * Given a LISTEN socket and an inbound SYN request, add 3874 * this to the syn cache, and send back a segment: 3875 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3876 * to the source. 3877 * 3878 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3879 * Doing so would require that we hold onto the data and deliver it 3880 * to the application. However, if we are the target of a SYN-flood 3881 * DoS attack, an attacker could send data which would eventually 3882 * consume all available buffer space if it were ACKed. By not ACKing 3883 * the data, we avoid this DoS scenario. 3884 */ 3885 3886 int 3887 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3888 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 3889 int optlen, struct tcp_opt_info *oi) 3890 { 3891 struct tcpcb tb, *tp; 3892 long win; 3893 struct syn_cache *sc; 3894 struct syn_cache_head *scp; 3895 struct mbuf *ipopts; 3896 struct tcp_opt_info opti; 3897 3898 tp = sototcpcb(so); 3899 3900 bzero(&opti, sizeof(opti)); 3901 3902 /* 3903 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3904 * 3905 * Note this check is performed in tcp_input() very early on. 3906 */ 3907 3908 /* 3909 * Initialize some local state. 3910 */ 3911 win = sbspace(&so->so_rcv); 3912 if (win > TCP_MAXWIN) 3913 win = TCP_MAXWIN; 3914 3915 switch (src->sa_family) { 3916 #ifdef INET 3917 case AF_INET: 3918 /* 3919 * Remember the IP options, if any. 3920 */ 3921 ipopts = ip_srcroute(); 3922 break; 3923 #endif 3924 default: 3925 ipopts = NULL; 3926 } 3927 3928 #ifdef TCP_SIGNATURE 3929 if (optp || (tp->t_flags & TF_SIGNATURE)) 3930 #else 3931 if (optp) 3932 #endif 3933 { 3934 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3935 #ifdef TCP_SIGNATURE 3936 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 3937 #endif 3938 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 3939 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 3940 return (0); 3941 } else 3942 tb.t_flags = 0; 3943 3944 /* 3945 * See if we already have an entry for this connection. 3946 * If we do, resend the SYN,ACK. We do not count this 3947 * as a retransmission (XXX though maybe we should). 3948 */ 3949 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3950 tcpstat.tcps_sc_dupesyn++; 3951 if (ipopts) { 3952 /* 3953 * If we were remembering a previous source route, 3954 * forget it and use the new one we've been given. 3955 */ 3956 if (sc->sc_ipopts) 3957 (void) m_free(sc->sc_ipopts); 3958 sc->sc_ipopts = ipopts; 3959 } 3960 sc->sc_timestamp = tb.ts_recent; 3961 if (syn_cache_respond(sc, m) == 0) { 3962 tcpstat.tcps_sndacks++; 3963 tcpstat.tcps_sndtotal++; 3964 } 3965 return (1); 3966 } 3967 3968 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3969 if (sc == NULL) { 3970 if (ipopts) 3971 (void) m_free(ipopts); 3972 return (0); 3973 } 3974 3975 /* 3976 * Fill in the cache, and put the necessary IP and TCP 3977 * options into the reply. 3978 */ 3979 bzero(sc, sizeof(struct syn_cache)); 3980 callout_init(&sc->sc_timer); 3981 bcopy(src, &sc->sc_src, src->sa_len); 3982 bcopy(dst, &sc->sc_dst, dst->sa_len); 3983 sc->sc_flags = 0; 3984 sc->sc_ipopts = ipopts; 3985 sc->sc_irs = th->th_seq; 3986 switch (src->sa_family) { 3987 #ifdef INET 3988 case AF_INET: 3989 { 3990 struct sockaddr_in *srcin = (void *) src; 3991 struct sockaddr_in *dstin = (void *) dst; 3992 3993 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 3994 &srcin->sin_addr, dstin->sin_port, 3995 srcin->sin_port, sizeof(dstin->sin_addr), 0); 3996 break; 3997 } 3998 #endif /* INET */ 3999 #ifdef INET6 4000 case AF_INET6: 4001 { 4002 struct sockaddr_in6 *srcin6 = (void *) src; 4003 struct sockaddr_in6 *dstin6 = (void *) dst; 4004 4005 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4006 &srcin6->sin6_addr, dstin6->sin6_port, 4007 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4008 break; 4009 } 4010 #endif /* INET6 */ 4011 } 4012 sc->sc_peermaxseg = oi->maxseg; 4013 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4014 m->m_pkthdr.rcvif : NULL, 4015 sc->sc_src.sa.sa_family); 4016 sc->sc_win = win; 4017 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 4018 sc->sc_timestamp = tb.ts_recent; 4019 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4020 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4021 sc->sc_flags |= SCF_TIMESTAMP; 4022 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4023 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4024 sc->sc_requested_s_scale = tb.requested_s_scale; 4025 sc->sc_request_r_scale = 0; 4026 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4027 TCP_MAXWIN << sc->sc_request_r_scale < 4028 so->so_rcv.sb_hiwat) 4029 sc->sc_request_r_scale++; 4030 } else { 4031 sc->sc_requested_s_scale = 15; 4032 sc->sc_request_r_scale = 15; 4033 } 4034 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4035 sc->sc_flags |= SCF_SACK_PERMIT; 4036 #ifdef TCP_SIGNATURE 4037 if (tb.t_flags & TF_SIGNATURE) 4038 sc->sc_flags |= SCF_SIGNATURE; 4039 #endif 4040 sc->sc_tp = tp; 4041 if (syn_cache_respond(sc, m) == 0) { 4042 syn_cache_insert(sc, tp); 4043 tcpstat.tcps_sndacks++; 4044 tcpstat.tcps_sndtotal++; 4045 } else { 4046 SYN_CACHE_PUT(sc); 4047 tcpstat.tcps_sc_dropped++; 4048 } 4049 return (1); 4050 } 4051 4052 int 4053 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4054 { 4055 struct route *ro; 4056 u_int8_t *optp; 4057 int optlen, error; 4058 u_int16_t tlen; 4059 struct ip *ip = NULL; 4060 #ifdef INET6 4061 struct ip6_hdr *ip6 = NULL; 4062 #endif 4063 struct tcpcb *tp; 4064 struct tcphdr *th; 4065 u_int hlen; 4066 struct socket *so; 4067 4068 switch (sc->sc_src.sa.sa_family) { 4069 case AF_INET: 4070 hlen = sizeof(struct ip); 4071 ro = &sc->sc_route4; 4072 break; 4073 #ifdef INET6 4074 case AF_INET6: 4075 hlen = sizeof(struct ip6_hdr); 4076 ro = (struct route *)&sc->sc_route6; 4077 break; 4078 #endif 4079 default: 4080 if (m) 4081 m_freem(m); 4082 return (EAFNOSUPPORT); 4083 } 4084 4085 /* Compute the size of the TCP options. */ 4086 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4087 ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) + 4088 #ifdef TCP_SIGNATURE 4089 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4090 #endif 4091 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4092 4093 tlen = hlen + sizeof(struct tcphdr) + optlen; 4094 4095 /* 4096 * Create the IP+TCP header from scratch. 4097 */ 4098 if (m) 4099 m_freem(m); 4100 #ifdef DIAGNOSTIC 4101 if (max_linkhdr + tlen > MCLBYTES) 4102 return (ENOBUFS); 4103 #endif 4104 MGETHDR(m, M_DONTWAIT, MT_DATA); 4105 if (m && tlen > MHLEN) { 4106 MCLGET(m, M_DONTWAIT); 4107 if ((m->m_flags & M_EXT) == 0) { 4108 m_freem(m); 4109 m = NULL; 4110 } 4111 } 4112 if (m == NULL) 4113 return (ENOBUFS); 4114 MCLAIM(m, &tcp_tx_mowner); 4115 4116 /* Fixup the mbuf. */ 4117 m->m_data += max_linkhdr; 4118 m->m_len = m->m_pkthdr.len = tlen; 4119 if (sc->sc_tp) { 4120 tp = sc->sc_tp; 4121 if (tp->t_inpcb) 4122 so = tp->t_inpcb->inp_socket; 4123 #ifdef INET6 4124 else if (tp->t_in6pcb) 4125 so = tp->t_in6pcb->in6p_socket; 4126 #endif 4127 else 4128 so = NULL; 4129 } else 4130 so = NULL; 4131 m->m_pkthdr.rcvif = NULL; 4132 memset(mtod(m, u_char *), 0, tlen); 4133 4134 switch (sc->sc_src.sa.sa_family) { 4135 case AF_INET: 4136 ip = mtod(m, struct ip *); 4137 ip->ip_v = 4; 4138 ip->ip_dst = sc->sc_src.sin.sin_addr; 4139 ip->ip_src = sc->sc_dst.sin.sin_addr; 4140 ip->ip_p = IPPROTO_TCP; 4141 th = (struct tcphdr *)(ip + 1); 4142 th->th_dport = sc->sc_src.sin.sin_port; 4143 th->th_sport = sc->sc_dst.sin.sin_port; 4144 break; 4145 #ifdef INET6 4146 case AF_INET6: 4147 ip6 = mtod(m, struct ip6_hdr *); 4148 ip6->ip6_vfc = IPV6_VERSION; 4149 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4150 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4151 ip6->ip6_nxt = IPPROTO_TCP; 4152 /* ip6_plen will be updated in ip6_output() */ 4153 th = (struct tcphdr *)(ip6 + 1); 4154 th->th_dport = sc->sc_src.sin6.sin6_port; 4155 th->th_sport = sc->sc_dst.sin6.sin6_port; 4156 break; 4157 #endif 4158 default: 4159 th = NULL; 4160 } 4161 4162 th->th_seq = htonl(sc->sc_iss); 4163 th->th_ack = htonl(sc->sc_irs + 1); 4164 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4165 th->th_flags = TH_SYN|TH_ACK; 4166 th->th_win = htons(sc->sc_win); 4167 /* th_sum already 0 */ 4168 /* th_urp already 0 */ 4169 4170 /* Tack on the TCP options. */ 4171 optp = (u_int8_t *)(th + 1); 4172 *optp++ = TCPOPT_MAXSEG; 4173 *optp++ = 4; 4174 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4175 *optp++ = sc->sc_ourmaxseg & 0xff; 4176 4177 if (sc->sc_request_r_scale != 15) { 4178 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4179 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4180 sc->sc_request_r_scale); 4181 optp += 4; 4182 } 4183 4184 if (sc->sc_flags & SCF_TIMESTAMP) { 4185 u_int32_t *lp = (u_int32_t *)(optp); 4186 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4187 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4188 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4189 *lp = htonl(sc->sc_timestamp); 4190 optp += TCPOLEN_TSTAMP_APPA; 4191 } 4192 4193 if (sc->sc_flags & SCF_SACK_PERMIT) { 4194 u_int8_t *p = optp; 4195 4196 /* Let the peer know that we will SACK. */ 4197 p[0] = TCPOPT_SACK_PERMITTED; 4198 p[1] = 2; 4199 p[2] = TCPOPT_NOP; 4200 p[3] = TCPOPT_NOP; 4201 optp += 4; 4202 } 4203 4204 #ifdef TCP_SIGNATURE 4205 if (sc->sc_flags & SCF_SIGNATURE) { 4206 struct secasvar *sav; 4207 u_int8_t *sigp; 4208 4209 sav = tcp_signature_getsav(m, th); 4210 4211 if (sav == NULL) { 4212 if (m) 4213 m_freem(m); 4214 return (EPERM); 4215 } 4216 4217 *optp++ = TCPOPT_SIGNATURE; 4218 *optp++ = TCPOLEN_SIGNATURE; 4219 sigp = optp; 4220 bzero(optp, TCP_SIGLEN); 4221 optp += TCP_SIGLEN; 4222 *optp++ = TCPOPT_NOP; 4223 *optp++ = TCPOPT_EOL; 4224 4225 (void)tcp_signature(m, th, hlen, sav, sigp); 4226 4227 key_sa_recordxfer(sav, m); 4228 #ifdef FAST_IPSEC 4229 KEY_FREESAV(&sav); 4230 #else 4231 key_freesav(sav); 4232 #endif 4233 } 4234 #endif 4235 4236 /* Compute the packet's checksum. */ 4237 switch (sc->sc_src.sa.sa_family) { 4238 case AF_INET: 4239 ip->ip_len = htons(tlen - hlen); 4240 th->th_sum = 0; 4241 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4242 break; 4243 #ifdef INET6 4244 case AF_INET6: 4245 ip6->ip6_plen = htons(tlen - hlen); 4246 th->th_sum = 0; 4247 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4248 break; 4249 #endif 4250 } 4251 4252 /* 4253 * Fill in some straggling IP bits. Note the stack expects 4254 * ip_len to be in host order, for convenience. 4255 */ 4256 switch (sc->sc_src.sa.sa_family) { 4257 #ifdef INET 4258 case AF_INET: 4259 ip->ip_len = htons(tlen); 4260 ip->ip_ttl = ip_defttl; 4261 /* XXX tos? */ 4262 break; 4263 #endif 4264 #ifdef INET6 4265 case AF_INET6: 4266 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4267 ip6->ip6_vfc |= IPV6_VERSION; 4268 ip6->ip6_plen = htons(tlen - hlen); 4269 /* ip6_hlim will be initialized afterwards */ 4270 /* XXX flowlabel? */ 4271 break; 4272 #endif 4273 } 4274 4275 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4276 tp = sc->sc_tp; 4277 4278 switch (sc->sc_src.sa.sa_family) { 4279 #ifdef INET 4280 case AF_INET: 4281 error = ip_output(m, sc->sc_ipopts, ro, 4282 (ip_mtudisc ? IP_MTUDISC : 0), 4283 (struct ip_moptions *)NULL, so); 4284 break; 4285 #endif 4286 #ifdef INET6 4287 case AF_INET6: 4288 ip6->ip6_hlim = in6_selecthlim(NULL, 4289 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 4290 4291 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0, 4292 (struct ip6_moptions *)0, so, NULL); 4293 break; 4294 #endif 4295 default: 4296 error = EAFNOSUPPORT; 4297 break; 4298 } 4299 return (error); 4300 } 4301