tcp_input.c revision 1.216
1 /* $NetBSD: tcp_input.c,v 1.216 2005/01/27 17:10:07 mycroft Exp $ */ 2 3 /* 4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the project nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 /* 33 * @(#)COPYRIGHT 1.1 (NRL) 17 January 1995 34 * 35 * NRL grants permission for redistribution and use in source and binary 36 * forms, with or without modification, of the software and documentation 37 * created at NRL provided that the following conditions are met: 38 * 39 * 1. Redistributions of source code must retain the above copyright 40 * notice, this list of conditions and the following disclaimer. 41 * 2. Redistributions in binary form must reproduce the above copyright 42 * notice, this list of conditions and the following disclaimer in the 43 * documentation and/or other materials provided with the distribution. 44 * 3. All advertising materials mentioning features or use of this software 45 * must display the following acknowledgements: 46 * This product includes software developed by the University of 47 * California, Berkeley and its contributors. 48 * This product includes software developed at the Information 49 * Technology Division, US Naval Research Laboratory. 50 * 4. Neither the name of the NRL nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS 55 * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 56 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A 57 * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NRL OR 58 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 59 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 60 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 61 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 62 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 63 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 64 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 65 * 66 * The views and conclusions contained in the software and documentation 67 * are those of the authors and should not be interpreted as representing 68 * official policies, either expressed or implied, of the US Naval 69 * Research Laboratory (NRL). 70 */ 71 72 /*- 73 * Copyright (c) 1997, 1998, 1999, 2001 The NetBSD Foundation, Inc. 74 * All rights reserved. 75 * 76 * This code is derived from software contributed to The NetBSD Foundation 77 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 78 * Facility, NASA Ames Research Center. 79 * 80 * Redistribution and use in source and binary forms, with or without 81 * modification, are permitted provided that the following conditions 82 * are met: 83 * 1. Redistributions of source code must retain the above copyright 84 * notice, this list of conditions and the following disclaimer. 85 * 2. Redistributions in binary form must reproduce the above copyright 86 * notice, this list of conditions and the following disclaimer in the 87 * documentation and/or other materials provided with the distribution. 88 * 3. All advertising materials mentioning features or use of this software 89 * must display the following acknowledgement: 90 * This product includes software developed by the NetBSD 91 * Foundation, Inc. and its contributors. 92 * 4. Neither the name of The NetBSD Foundation nor the names of its 93 * contributors may be used to endorse or promote products derived 94 * from this software without specific prior written permission. 95 * 96 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 97 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 98 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 99 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 100 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 101 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 102 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 103 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 104 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 105 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 106 * POSSIBILITY OF SUCH DAMAGE. 107 */ 108 109 /* 110 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 111 * The Regents of the University of California. All rights reserved. 112 * 113 * Redistribution and use in source and binary forms, with or without 114 * modification, are permitted provided that the following conditions 115 * are met: 116 * 1. Redistributions of source code must retain the above copyright 117 * notice, this list of conditions and the following disclaimer. 118 * 2. Redistributions in binary form must reproduce the above copyright 119 * notice, this list of conditions and the following disclaimer in the 120 * documentation and/or other materials provided with the distribution. 121 * 3. Neither the name of the University nor the names of its contributors 122 * may be used to endorse or promote products derived from this software 123 * without specific prior written permission. 124 * 125 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 126 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 127 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 128 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 129 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 130 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 131 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 132 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 133 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 134 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 135 * SUCH DAMAGE. 136 * 137 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 138 */ 139 140 /* 141 * TODO list for SYN cache stuff: 142 * 143 * Find room for a "state" field, which is needed to keep a 144 * compressed state for TIME_WAIT TCBs. It's been noted already 145 * that this is fairly important for very high-volume web and 146 * mail servers, which use a large number of short-lived 147 * connections. 148 */ 149 150 #include <sys/cdefs.h> 151 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.216 2005/01/27 17:10:07 mycroft Exp $"); 152 153 #include "opt_inet.h" 154 #include "opt_ipsec.h" 155 #include "opt_inet_csum.h" 156 #include "opt_tcp_debug.h" 157 158 #include <sys/param.h> 159 #include <sys/systm.h> 160 #include <sys/malloc.h> 161 #include <sys/mbuf.h> 162 #include <sys/protosw.h> 163 #include <sys/socket.h> 164 #include <sys/socketvar.h> 165 #include <sys/errno.h> 166 #include <sys/syslog.h> 167 #include <sys/pool.h> 168 #include <sys/domain.h> 169 #include <sys/kernel.h> 170 #ifdef TCP_SIGNATURE 171 #include <sys/md5.h> 172 #endif 173 174 #include <net/if.h> 175 #include <net/route.h> 176 #include <net/if_types.h> 177 178 #include <netinet/in.h> 179 #include <netinet/in_systm.h> 180 #include <netinet/ip.h> 181 #include <netinet/in_pcb.h> 182 #include <netinet/in_var.h> 183 #include <netinet/ip_var.h> 184 185 #ifdef INET6 186 #ifndef INET 187 #include <netinet/in.h> 188 #endif 189 #include <netinet/ip6.h> 190 #include <netinet6/ip6_var.h> 191 #include <netinet6/in6_pcb.h> 192 #include <netinet6/ip6_var.h> 193 #include <netinet6/in6_var.h> 194 #include <netinet/icmp6.h> 195 #include <netinet6/nd6.h> 196 #endif 197 198 #ifndef INET6 199 /* always need ip6.h for IP6_EXTHDR_GET */ 200 #include <netinet/ip6.h> 201 #endif 202 203 #include <netinet/tcp.h> 204 #include <netinet/tcp_fsm.h> 205 #include <netinet/tcp_seq.h> 206 #include <netinet/tcp_timer.h> 207 #include <netinet/tcp_var.h> 208 #include <netinet/tcpip.h> 209 #include <netinet/tcp_debug.h> 210 211 #include <machine/stdarg.h> 212 213 #ifdef IPSEC 214 #include <netinet6/ipsec.h> 215 #include <netkey/key.h> 216 #endif /*IPSEC*/ 217 #ifdef INET6 218 #include "faith.h" 219 #if defined(NFAITH) && NFAITH > 0 220 #include <net/if_faith.h> 221 #endif 222 #endif /* IPSEC */ 223 224 #ifdef FAST_IPSEC 225 #include <netipsec/ipsec.h> 226 #include <netipsec/ipsec_var.h> /* XXX ipsecstat namespace */ 227 #include <netipsec/key.h> 228 #ifdef INET6 229 #include <netipsec/ipsec6.h> 230 #endif 231 #endif /* FAST_IPSEC*/ 232 233 int tcprexmtthresh = 3; 234 int tcp_log_refused; 235 236 static int tcp_rst_ppslim_count = 0; 237 static struct timeval tcp_rst_ppslim_last; 238 static int tcp_ackdrop_ppslim_count = 0; 239 static struct timeval tcp_ackdrop_ppslim_last; 240 241 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 242 243 /* for modulo comparisons of timestamps */ 244 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 245 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 246 247 /* 248 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 249 */ 250 #ifdef INET6 251 #define ND6_HINT(tp) \ 252 do { \ 253 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 && \ 254 tp->t_in6pcb->in6p_route.ro_rt) { \ 255 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \ 256 } \ 257 } while (/*CONSTCOND*/ 0) 258 #else 259 #define ND6_HINT(tp) 260 #endif 261 262 /* 263 * Macro to compute ACK transmission behavior. Delay the ACK unless 264 * we have already delayed an ACK (must send an ACK every two segments). 265 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 266 * option is enabled. 267 */ 268 #define TCP_SETUP_ACK(tp, th) \ 269 do { \ 270 if ((tp)->t_flags & TF_DELACK || \ 271 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 272 tp->t_flags |= TF_ACKNOW; \ 273 else \ 274 TCP_SET_DELACK(tp); \ 275 } while (/*CONSTCOND*/ 0) 276 277 /* 278 * Convert TCP protocol fields to host order for easier processing. 279 */ 280 #define TCP_FIELDS_TO_HOST(th) \ 281 do { \ 282 NTOHL((th)->th_seq); \ 283 NTOHL((th)->th_ack); \ 284 NTOHS((th)->th_win); \ 285 NTOHS((th)->th_urp); \ 286 } while (/*CONSTCOND*/ 0) 287 288 /* 289 * ... and reverse the above. 290 */ 291 #define TCP_FIELDS_TO_NET(th) \ 292 do { \ 293 HTONL((th)->th_seq); \ 294 HTONL((th)->th_ack); \ 295 HTONS((th)->th_win); \ 296 HTONS((th)->th_urp); \ 297 } while (/*CONSTCOND*/ 0) 298 299 #ifdef TCP_CSUM_COUNTERS 300 #include <sys/device.h> 301 302 extern struct evcnt tcp_hwcsum_ok; 303 extern struct evcnt tcp_hwcsum_bad; 304 extern struct evcnt tcp_hwcsum_data; 305 extern struct evcnt tcp_swcsum; 306 307 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 308 309 #else 310 311 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 312 313 #endif /* TCP_CSUM_COUNTERS */ 314 315 #ifdef TCP_REASS_COUNTERS 316 #include <sys/device.h> 317 318 extern struct evcnt tcp_reass_; 319 extern struct evcnt tcp_reass_empty; 320 extern struct evcnt tcp_reass_iteration[8]; 321 extern struct evcnt tcp_reass_prependfirst; 322 extern struct evcnt tcp_reass_prepend; 323 extern struct evcnt tcp_reass_insert; 324 extern struct evcnt tcp_reass_inserttail; 325 extern struct evcnt tcp_reass_append; 326 extern struct evcnt tcp_reass_appendtail; 327 extern struct evcnt tcp_reass_overlaptail; 328 extern struct evcnt tcp_reass_overlapfront; 329 extern struct evcnt tcp_reass_segdup; 330 extern struct evcnt tcp_reass_fragdup; 331 332 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 333 334 #else 335 336 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 337 338 #endif /* TCP_REASS_COUNTERS */ 339 340 #ifdef INET 341 static void tcp4_log_refused __P((const struct ip *, const struct tcphdr *)); 342 #endif 343 #ifdef INET6 344 static void tcp6_log_refused 345 __P((const struct ip6_hdr *, const struct tcphdr *)); 346 #endif 347 348 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 349 350 POOL_INIT(tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", NULL); 351 352 int 353 tcp_reass(tp, th, m, tlen) 354 struct tcpcb *tp; 355 struct tcphdr *th; 356 struct mbuf *m; 357 int *tlen; 358 { 359 struct ipqent *p, *q, *nq, *tiqe = NULL; 360 struct socket *so = NULL; 361 int pkt_flags; 362 tcp_seq pkt_seq; 363 unsigned pkt_len; 364 u_long rcvpartdupbyte = 0; 365 u_long rcvoobyte; 366 #ifdef TCP_REASS_COUNTERS 367 u_int count = 0; 368 #endif 369 370 if (tp->t_inpcb) 371 so = tp->t_inpcb->inp_socket; 372 #ifdef INET6 373 else if (tp->t_in6pcb) 374 so = tp->t_in6pcb->in6p_socket; 375 #endif 376 377 TCP_REASS_LOCK_CHECK(tp); 378 379 /* 380 * Call with th==0 after become established to 381 * force pre-ESTABLISHED data up to user socket. 382 */ 383 if (th == 0) 384 goto present; 385 386 rcvoobyte = *tlen; 387 /* 388 * Copy these to local variables because the tcpiphdr 389 * gets munged while we are collapsing mbufs. 390 */ 391 pkt_seq = th->th_seq; 392 pkt_len = *tlen; 393 pkt_flags = th->th_flags; 394 395 TCP_REASS_COUNTER_INCR(&tcp_reass_); 396 397 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 398 /* 399 * When we miss a packet, the vast majority of time we get 400 * packets that follow it in order. So optimize for that. 401 */ 402 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 403 p->ipqe_len += pkt_len; 404 p->ipqe_flags |= pkt_flags; 405 m_cat(p->ipre_mlast, m); 406 TRAVERSE(p->ipre_mlast); 407 m = NULL; 408 tiqe = p; 409 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 410 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 411 goto skip_replacement; 412 } 413 /* 414 * While we're here, if the pkt is completely beyond 415 * anything we have, just insert it at the tail. 416 */ 417 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 418 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 419 goto insert_it; 420 } 421 } 422 423 q = TAILQ_FIRST(&tp->segq); 424 425 if (q != NULL) { 426 /* 427 * If this segment immediately precedes the first out-of-order 428 * block, simply slap the segment in front of it and (mostly) 429 * skip the complicated logic. 430 */ 431 if (pkt_seq + pkt_len == q->ipqe_seq) { 432 q->ipqe_seq = pkt_seq; 433 q->ipqe_len += pkt_len; 434 q->ipqe_flags |= pkt_flags; 435 m_cat(m, q->ipqe_m); 436 q->ipqe_m = m; 437 q->ipre_mlast = m; /* last mbuf may have changed */ 438 TRAVERSE(q->ipre_mlast); 439 tiqe = q; 440 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 441 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 442 goto skip_replacement; 443 } 444 } else { 445 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 446 } 447 448 /* 449 * Find a segment which begins after this one does. 450 */ 451 for (p = NULL; q != NULL; q = nq) { 452 nq = TAILQ_NEXT(q, ipqe_q); 453 #ifdef TCP_REASS_COUNTERS 454 count++; 455 #endif 456 /* 457 * If the received segment is just right after this 458 * fragment, merge the two together and then check 459 * for further overlaps. 460 */ 461 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 462 #ifdef TCPREASS_DEBUG 463 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 464 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 465 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 466 #endif 467 pkt_len += q->ipqe_len; 468 pkt_flags |= q->ipqe_flags; 469 pkt_seq = q->ipqe_seq; 470 m_cat(q->ipre_mlast, m); 471 TRAVERSE(q->ipre_mlast); 472 m = q->ipqe_m; 473 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 474 goto free_ipqe; 475 } 476 /* 477 * If the received segment is completely past this 478 * fragment, we need to go the next fragment. 479 */ 480 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 481 p = q; 482 continue; 483 } 484 /* 485 * If the fragment is past the received segment, 486 * it (or any following) can't be concatenated. 487 */ 488 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 489 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 490 break; 491 } 492 493 /* 494 * We've received all the data in this segment before. 495 * mark it as a duplicate and return. 496 */ 497 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 498 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 499 tcpstat.tcps_rcvduppack++; 500 tcpstat.tcps_rcvdupbyte += pkt_len; 501 m_freem(m); 502 if (tiqe != NULL) 503 pool_put(&tcpipqent_pool, tiqe); 504 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 505 return (0); 506 } 507 /* 508 * Received segment completely overlaps this fragment 509 * so we drop the fragment (this keeps the temporal 510 * ordering of segments correct). 511 */ 512 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 513 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 514 rcvpartdupbyte += q->ipqe_len; 515 m_freem(q->ipqe_m); 516 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 517 goto free_ipqe; 518 } 519 /* 520 * RX'ed segment extends past the end of the 521 * fragment. Drop the overlapping bytes. Then 522 * merge the fragment and segment then treat as 523 * a longer received packet. 524 */ 525 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 526 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 527 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 528 #ifdef TCPREASS_DEBUG 529 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 530 tp, overlap, 531 pkt_seq, pkt_seq + pkt_len, pkt_len); 532 #endif 533 m_adj(m, overlap); 534 rcvpartdupbyte += overlap; 535 m_cat(q->ipre_mlast, m); 536 TRAVERSE(q->ipre_mlast); 537 m = q->ipqe_m; 538 pkt_seq = q->ipqe_seq; 539 pkt_len += q->ipqe_len - overlap; 540 rcvoobyte -= overlap; 541 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 542 goto free_ipqe; 543 } 544 /* 545 * RX'ed segment extends past the front of the 546 * fragment. Drop the overlapping bytes on the 547 * received packet. The packet will then be 548 * contatentated with this fragment a bit later. 549 */ 550 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 551 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 552 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 553 #ifdef TCPREASS_DEBUG 554 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 555 tp, overlap, 556 pkt_seq, pkt_seq + pkt_len, pkt_len); 557 #endif 558 m_adj(m, -overlap); 559 pkt_len -= overlap; 560 rcvpartdupbyte += overlap; 561 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 562 rcvoobyte -= overlap; 563 } 564 /* 565 * If the received segment immediates precedes this 566 * fragment then tack the fragment onto this segment 567 * and reinsert the data. 568 */ 569 if (q->ipqe_seq == pkt_seq + pkt_len) { 570 #ifdef TCPREASS_DEBUG 571 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 572 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 573 pkt_seq, pkt_seq + pkt_len, pkt_len); 574 #endif 575 pkt_len += q->ipqe_len; 576 pkt_flags |= q->ipqe_flags; 577 m_cat(m, q->ipqe_m); 578 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 579 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 580 if (tiqe == NULL) 581 tiqe = q; 582 else 583 pool_put(&tcpipqent_pool, q); 584 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 585 break; 586 } 587 /* 588 * If the fragment is before the segment, remember it. 589 * When this loop is terminated, p will contain the 590 * pointer to fragment that is right before the received 591 * segment. 592 */ 593 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 594 p = q; 595 596 continue; 597 598 /* 599 * This is a common operation. It also will allow 600 * to save doing a malloc/free in most instances. 601 */ 602 free_ipqe: 603 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 604 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 605 if (tiqe == NULL) 606 tiqe = q; 607 else 608 pool_put(&tcpipqent_pool, q); 609 } 610 611 #ifdef TCP_REASS_COUNTERS 612 if (count > 7) 613 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 614 else if (count > 0) 615 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 616 #endif 617 618 insert_it: 619 620 /* 621 * Allocate a new queue entry since the received segment did not 622 * collapse onto any other out-of-order block; thus we are allocating 623 * a new block. If it had collapsed, tiqe would not be NULL and 624 * we would be reusing it. 625 * XXX If we can't, just drop the packet. XXX 626 */ 627 if (tiqe == NULL) { 628 tiqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 629 if (tiqe == NULL) { 630 tcpstat.tcps_rcvmemdrop++; 631 m_freem(m); 632 return (0); 633 } 634 } 635 636 /* 637 * Update the counters. 638 */ 639 tcpstat.tcps_rcvoopack++; 640 tcpstat.tcps_rcvoobyte += rcvoobyte; 641 if (rcvpartdupbyte) { 642 tcpstat.tcps_rcvpartduppack++; 643 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 644 } 645 646 /* 647 * Insert the new fragment queue entry into both queues. 648 */ 649 tiqe->ipqe_m = m; 650 tiqe->ipre_mlast = m; 651 tiqe->ipqe_seq = pkt_seq; 652 tiqe->ipqe_len = pkt_len; 653 tiqe->ipqe_flags = pkt_flags; 654 if (p == NULL) { 655 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 656 #ifdef TCPREASS_DEBUG 657 if (tiqe->ipqe_seq != tp->rcv_nxt) 658 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 659 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 660 #endif 661 } else { 662 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 663 #ifdef TCPREASS_DEBUG 664 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 665 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 666 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 667 #endif 668 } 669 670 skip_replacement: 671 672 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 673 674 present: 675 /* 676 * Present data to user, advancing rcv_nxt through 677 * completed sequence space. 678 */ 679 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 680 return (0); 681 q = TAILQ_FIRST(&tp->segq); 682 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 683 return (0); 684 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 685 return (0); 686 687 tp->rcv_nxt += q->ipqe_len; 688 pkt_flags = q->ipqe_flags & TH_FIN; 689 ND6_HINT(tp); 690 691 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 692 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 693 if (so->so_state & SS_CANTRCVMORE) 694 m_freem(q->ipqe_m); 695 else 696 sbappendstream(&so->so_rcv, q->ipqe_m); 697 pool_put(&tcpipqent_pool, q); 698 sorwakeup(so); 699 return (pkt_flags); 700 } 701 702 #ifdef INET6 703 int 704 tcp6_input(mp, offp, proto) 705 struct mbuf **mp; 706 int *offp, proto; 707 { 708 struct mbuf *m = *mp; 709 710 /* 711 * draft-itojun-ipv6-tcp-to-anycast 712 * better place to put this in? 713 */ 714 if (m->m_flags & M_ANYCAST6) { 715 struct ip6_hdr *ip6; 716 if (m->m_len < sizeof(struct ip6_hdr)) { 717 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 718 tcpstat.tcps_rcvshort++; 719 return IPPROTO_DONE; 720 } 721 } 722 ip6 = mtod(m, struct ip6_hdr *); 723 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 724 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 725 return IPPROTO_DONE; 726 } 727 728 tcp_input(m, *offp, proto); 729 return IPPROTO_DONE; 730 } 731 #endif 732 733 #ifdef INET 734 static void 735 tcp4_log_refused(ip, th) 736 const struct ip *ip; 737 const struct tcphdr *th; 738 { 739 char src[4*sizeof "123"]; 740 char dst[4*sizeof "123"]; 741 742 if (ip) { 743 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 744 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 745 } 746 else { 747 strlcpy(src, "(unknown)", sizeof(src)); 748 strlcpy(dst, "(unknown)", sizeof(dst)); 749 } 750 log(LOG_INFO, 751 "Connection attempt to TCP %s:%d from %s:%d\n", 752 dst, ntohs(th->th_dport), 753 src, ntohs(th->th_sport)); 754 } 755 #endif 756 757 #ifdef INET6 758 static void 759 tcp6_log_refused(ip6, th) 760 const struct ip6_hdr *ip6; 761 const struct tcphdr *th; 762 { 763 char src[INET6_ADDRSTRLEN]; 764 char dst[INET6_ADDRSTRLEN]; 765 766 if (ip6) { 767 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 768 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 769 } 770 else { 771 strlcpy(src, "(unknown v6)", sizeof(src)); 772 strlcpy(dst, "(unknown v6)", sizeof(dst)); 773 } 774 log(LOG_INFO, 775 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 776 dst, ntohs(th->th_dport), 777 src, ntohs(th->th_sport)); 778 } 779 #endif 780 781 /* 782 * Checksum extended TCP header and data. 783 */ 784 int 785 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, int toff, 786 int off, int tlen) 787 { 788 789 /* 790 * XXX it's better to record and check if this mbuf is 791 * already checked. 792 */ 793 794 switch (af) { 795 #ifdef INET 796 case AF_INET: 797 switch (m->m_pkthdr.csum_flags & 798 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 799 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 800 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 801 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 802 goto badcsum; 803 804 case M_CSUM_TCPv4|M_CSUM_DATA: { 805 u_int32_t hw_csum = m->m_pkthdr.csum_data; 806 807 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 808 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 809 const struct ip *ip = 810 mtod(m, const struct ip *); 811 812 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 813 ip->ip_dst.s_addr, 814 htons(hw_csum + tlen + off + IPPROTO_TCP)); 815 } 816 if ((hw_csum ^ 0xffff) != 0) 817 goto badcsum; 818 break; 819 } 820 821 case M_CSUM_TCPv4: 822 /* Checksum was okay. */ 823 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 824 break; 825 826 default: 827 /* 828 * Must compute it ourselves. Maybe skip checksum 829 * on loopback interfaces. 830 */ 831 if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & 832 IFF_LOOPBACK) || 833 tcp_do_loopback_cksum)) { 834 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 835 if (in4_cksum(m, IPPROTO_TCP, toff, 836 tlen + off) != 0) 837 goto badcsum; 838 } 839 break; 840 } 841 break; 842 #endif /* INET4 */ 843 844 #ifdef INET6 845 case AF_INET6: 846 if (__predict_true((m->m_flags & M_LOOP) == 0 || 847 tcp_do_loopback_cksum)) { 848 if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 849 goto badcsum; 850 } 851 break; 852 #endif /* INET6 */ 853 } 854 855 return 0; 856 857 badcsum: 858 tcpstat.tcps_rcvbadsum++; 859 return -1; 860 } 861 862 /* 863 * TCP input routine, follows pages 65-76 of the 864 * protocol specification dated September, 1981 very closely. 865 */ 866 void 867 tcp_input(struct mbuf *m, ...) 868 { 869 struct tcphdr *th; 870 struct ip *ip; 871 struct inpcb *inp; 872 #ifdef INET6 873 struct ip6_hdr *ip6; 874 struct in6pcb *in6p; 875 #endif 876 u_int8_t *optp = NULL; 877 int optlen = 0; 878 int len, tlen, toff, hdroptlen = 0; 879 struct tcpcb *tp = 0; 880 int tiflags; 881 struct socket *so = NULL; 882 int todrop, acked, ourfinisacked, needoutput = 0; 883 #ifdef TCP_DEBUG 884 short ostate = 0; 885 #endif 886 int iss = 0; 887 u_long tiwin; 888 struct tcp_opt_info opti; 889 int off, iphlen; 890 va_list ap; 891 int af; /* af on the wire */ 892 struct mbuf *tcp_saveti = NULL; 893 894 MCLAIM(m, &tcp_rx_mowner); 895 va_start(ap, m); 896 toff = va_arg(ap, int); 897 (void)va_arg(ap, int); /* ignore value, advance ap */ 898 va_end(ap); 899 900 tcpstat.tcps_rcvtotal++; 901 902 bzero(&opti, sizeof(opti)); 903 opti.ts_present = 0; 904 opti.maxseg = 0; 905 906 /* 907 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 908 * 909 * TCP is, by definition, unicast, so we reject all 910 * multicast outright. 911 * 912 * Note, there are additional src/dst address checks in 913 * the AF-specific code below. 914 */ 915 if (m->m_flags & (M_BCAST|M_MCAST)) { 916 /* XXX stat */ 917 goto drop; 918 } 919 #ifdef INET6 920 if (m->m_flags & M_ANYCAST6) { 921 /* XXX stat */ 922 goto drop; 923 } 924 #endif 925 926 /* 927 * Get IP and TCP header. 928 * Note: IP leaves IP header in first mbuf. 929 */ 930 ip = mtod(m, struct ip *); 931 #ifdef INET6 932 ip6 = NULL; 933 #endif 934 switch (ip->ip_v) { 935 #ifdef INET 936 case 4: 937 af = AF_INET; 938 iphlen = sizeof(struct ip); 939 ip = mtod(m, struct ip *); 940 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 941 sizeof(struct tcphdr)); 942 if (th == NULL) { 943 tcpstat.tcps_rcvshort++; 944 return; 945 } 946 /* We do the checksum after PCB lookup... */ 947 len = ntohs(ip->ip_len); 948 tlen = len - toff; 949 break; 950 #endif 951 #ifdef INET6 952 case 6: 953 ip = NULL; 954 iphlen = sizeof(struct ip6_hdr); 955 af = AF_INET6; 956 ip6 = mtod(m, struct ip6_hdr *); 957 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 958 sizeof(struct tcphdr)); 959 if (th == NULL) { 960 tcpstat.tcps_rcvshort++; 961 return; 962 } 963 964 /* Be proactive about malicious use of IPv4 mapped address */ 965 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 966 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 967 /* XXX stat */ 968 goto drop; 969 } 970 971 /* 972 * Be proactive about unspecified IPv6 address in source. 973 * As we use all-zero to indicate unbounded/unconnected pcb, 974 * unspecified IPv6 address can be used to confuse us. 975 * 976 * Note that packets with unspecified IPv6 destination is 977 * already dropped in ip6_input. 978 */ 979 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 980 /* XXX stat */ 981 goto drop; 982 } 983 984 /* 985 * Make sure destination address is not multicast. 986 * Source address checked in ip6_input(). 987 */ 988 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 989 /* XXX stat */ 990 goto drop; 991 } 992 993 /* We do the checksum after PCB lookup... */ 994 len = m->m_pkthdr.len; 995 tlen = len - toff; 996 break; 997 #endif 998 default: 999 m_freem(m); 1000 return; 1001 } 1002 1003 KASSERT(TCP_HDR_ALIGNED_P(th)); 1004 1005 /* 1006 * Check that TCP offset makes sense, 1007 * pull out TCP options and adjust length. XXX 1008 */ 1009 off = th->th_off << 2; 1010 if (off < sizeof (struct tcphdr) || off > tlen) { 1011 tcpstat.tcps_rcvbadoff++; 1012 goto drop; 1013 } 1014 tlen -= off; 1015 1016 /* 1017 * tcp_input() has been modified to use tlen to mean the TCP data 1018 * length throughout the function. Other functions can use 1019 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1020 * rja 1021 */ 1022 1023 if (off > sizeof (struct tcphdr)) { 1024 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1025 if (th == NULL) { 1026 tcpstat.tcps_rcvshort++; 1027 return; 1028 } 1029 /* 1030 * NOTE: ip/ip6 will not be affected by m_pulldown() 1031 * (as they're before toff) and we don't need to update those. 1032 */ 1033 KASSERT(TCP_HDR_ALIGNED_P(th)); 1034 optlen = off - sizeof (struct tcphdr); 1035 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1036 /* 1037 * Do quick retrieval of timestamp options ("options 1038 * prediction?"). If timestamp is the only option and it's 1039 * formatted as recommended in RFC 1323 appendix A, we 1040 * quickly get the values now and not bother calling 1041 * tcp_dooptions(), etc. 1042 */ 1043 if ((optlen == TCPOLEN_TSTAMP_APPA || 1044 (optlen > TCPOLEN_TSTAMP_APPA && 1045 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1046 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1047 (th->th_flags & TH_SYN) == 0) { 1048 opti.ts_present = 1; 1049 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1050 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1051 optp = NULL; /* we've parsed the options */ 1052 } 1053 } 1054 tiflags = th->th_flags; 1055 1056 /* 1057 * Locate pcb for segment. 1058 */ 1059 findpcb: 1060 inp = NULL; 1061 #ifdef INET6 1062 in6p = NULL; 1063 #endif 1064 switch (af) { 1065 #ifdef INET 1066 case AF_INET: 1067 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1068 ip->ip_dst, th->th_dport); 1069 if (inp == 0) { 1070 ++tcpstat.tcps_pcbhashmiss; 1071 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1072 } 1073 #ifdef INET6 1074 if (inp == 0) { 1075 struct in6_addr s, d; 1076 1077 /* mapped addr case */ 1078 bzero(&s, sizeof(s)); 1079 s.s6_addr16[5] = htons(0xffff); 1080 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1081 bzero(&d, sizeof(d)); 1082 d.s6_addr16[5] = htons(0xffff); 1083 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1084 in6p = in6_pcblookup_connect(&tcbtable, &s, 1085 th->th_sport, &d, th->th_dport, 0); 1086 if (in6p == 0) { 1087 ++tcpstat.tcps_pcbhashmiss; 1088 in6p = in6_pcblookup_bind(&tcbtable, &d, 1089 th->th_dport, 0); 1090 } 1091 } 1092 #endif 1093 #ifndef INET6 1094 if (inp == 0) 1095 #else 1096 if (inp == 0 && in6p == 0) 1097 #endif 1098 { 1099 ++tcpstat.tcps_noport; 1100 if (tcp_log_refused && 1101 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1102 tcp4_log_refused(ip, th); 1103 } 1104 TCP_FIELDS_TO_HOST(th); 1105 goto dropwithreset_ratelim; 1106 } 1107 #if defined(IPSEC) || defined(FAST_IPSEC) 1108 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1109 ipsec4_in_reject(m, inp)) { 1110 ipsecstat.in_polvio++; 1111 goto drop; 1112 } 1113 #ifdef INET6 1114 else if (in6p && 1115 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1116 ipsec4_in_reject_so(m, in6p->in6p_socket)) { 1117 ipsecstat.in_polvio++; 1118 goto drop; 1119 } 1120 #endif 1121 #endif /*IPSEC*/ 1122 break; 1123 #endif /*INET*/ 1124 #ifdef INET6 1125 case AF_INET6: 1126 { 1127 int faith; 1128 1129 #if defined(NFAITH) && NFAITH > 0 1130 faith = faithprefix(&ip6->ip6_dst); 1131 #else 1132 faith = 0; 1133 #endif 1134 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1135 th->th_sport, &ip6->ip6_dst, th->th_dport, faith); 1136 if (in6p == NULL) { 1137 ++tcpstat.tcps_pcbhashmiss; 1138 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1139 th->th_dport, faith); 1140 } 1141 if (in6p == NULL) { 1142 ++tcpstat.tcps_noport; 1143 if (tcp_log_refused && 1144 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1145 tcp6_log_refused(ip6, th); 1146 } 1147 TCP_FIELDS_TO_HOST(th); 1148 goto dropwithreset_ratelim; 1149 } 1150 #if defined(IPSEC) || defined(FAST_IPSEC) 1151 if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1152 ipsec6_in_reject(m, in6p)) { 1153 ipsec6stat.in_polvio++; 1154 goto drop; 1155 } 1156 #endif /*IPSEC*/ 1157 break; 1158 } 1159 #endif 1160 } 1161 1162 /* 1163 * If the state is CLOSED (i.e., TCB does not exist) then 1164 * all data in the incoming segment is discarded. 1165 * If the TCB exists but is in CLOSED state, it is embryonic, 1166 * but should either do a listen or a connect soon. 1167 */ 1168 tp = NULL; 1169 so = NULL; 1170 if (inp) { 1171 tp = intotcpcb(inp); 1172 so = inp->inp_socket; 1173 } 1174 #ifdef INET6 1175 else if (in6p) { 1176 tp = in6totcpcb(in6p); 1177 so = in6p->in6p_socket; 1178 } 1179 #endif 1180 if (tp == 0) { 1181 TCP_FIELDS_TO_HOST(th); 1182 goto dropwithreset_ratelim; 1183 } 1184 if (tp->t_state == TCPS_CLOSED) 1185 goto drop; 1186 1187 /* 1188 * Checksum extended TCP header and data. 1189 */ 1190 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1191 goto badcsum; 1192 1193 TCP_FIELDS_TO_HOST(th); 1194 1195 /* Unscale the window into a 32-bit value. */ 1196 if ((tiflags & TH_SYN) == 0) 1197 tiwin = th->th_win << tp->snd_scale; 1198 else 1199 tiwin = th->th_win; 1200 1201 #ifdef INET6 1202 /* save packet options if user wanted */ 1203 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1204 if (in6p->in6p_options) { 1205 m_freem(in6p->in6p_options); 1206 in6p->in6p_options = 0; 1207 } 1208 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1209 } 1210 #endif 1211 1212 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1213 union syn_cache_sa src; 1214 union syn_cache_sa dst; 1215 1216 bzero(&src, sizeof(src)); 1217 bzero(&dst, sizeof(dst)); 1218 switch (af) { 1219 #ifdef INET 1220 case AF_INET: 1221 src.sin.sin_len = sizeof(struct sockaddr_in); 1222 src.sin.sin_family = AF_INET; 1223 src.sin.sin_addr = ip->ip_src; 1224 src.sin.sin_port = th->th_sport; 1225 1226 dst.sin.sin_len = sizeof(struct sockaddr_in); 1227 dst.sin.sin_family = AF_INET; 1228 dst.sin.sin_addr = ip->ip_dst; 1229 dst.sin.sin_port = th->th_dport; 1230 break; 1231 #endif 1232 #ifdef INET6 1233 case AF_INET6: 1234 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1235 src.sin6.sin6_family = AF_INET6; 1236 src.sin6.sin6_addr = ip6->ip6_src; 1237 src.sin6.sin6_port = th->th_sport; 1238 1239 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1240 dst.sin6.sin6_family = AF_INET6; 1241 dst.sin6.sin6_addr = ip6->ip6_dst; 1242 dst.sin6.sin6_port = th->th_dport; 1243 break; 1244 #endif /* INET6 */ 1245 default: 1246 goto badsyn; /*sanity*/ 1247 } 1248 1249 if (so->so_options & SO_DEBUG) { 1250 #ifdef TCP_DEBUG 1251 ostate = tp->t_state; 1252 #endif 1253 1254 tcp_saveti = NULL; 1255 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1256 goto nosave; 1257 1258 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1259 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1260 if (!tcp_saveti) 1261 goto nosave; 1262 } else { 1263 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1264 if (!tcp_saveti) 1265 goto nosave; 1266 MCLAIM(m, &tcp_mowner); 1267 tcp_saveti->m_len = iphlen; 1268 m_copydata(m, 0, iphlen, 1269 mtod(tcp_saveti, caddr_t)); 1270 } 1271 1272 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1273 m_freem(tcp_saveti); 1274 tcp_saveti = NULL; 1275 } else { 1276 tcp_saveti->m_len += sizeof(struct tcphdr); 1277 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1278 sizeof(struct tcphdr)); 1279 } 1280 nosave:; 1281 } 1282 if (so->so_options & SO_ACCEPTCONN) { 1283 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1284 if (tiflags & TH_RST) { 1285 syn_cache_reset(&src.sa, &dst.sa, th); 1286 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1287 (TH_ACK|TH_SYN)) { 1288 /* 1289 * Received a SYN,ACK. This should 1290 * never happen while we are in 1291 * LISTEN. Send an RST. 1292 */ 1293 goto badsyn; 1294 } else if (tiflags & TH_ACK) { 1295 so = syn_cache_get(&src.sa, &dst.sa, 1296 th, toff, tlen, so, m); 1297 if (so == NULL) { 1298 /* 1299 * We don't have a SYN for 1300 * this ACK; send an RST. 1301 */ 1302 goto badsyn; 1303 } else if (so == 1304 (struct socket *)(-1)) { 1305 /* 1306 * We were unable to create 1307 * the connection. If the 1308 * 3-way handshake was 1309 * completed, and RST has 1310 * been sent to the peer. 1311 * Since the mbuf might be 1312 * in use for the reply, 1313 * do not free it. 1314 */ 1315 m = NULL; 1316 } else { 1317 /* 1318 * We have created a 1319 * full-blown connection. 1320 */ 1321 tp = NULL; 1322 inp = NULL; 1323 #ifdef INET6 1324 in6p = NULL; 1325 #endif 1326 switch (so->so_proto->pr_domain->dom_family) { 1327 #ifdef INET 1328 case AF_INET: 1329 inp = sotoinpcb(so); 1330 tp = intotcpcb(inp); 1331 break; 1332 #endif 1333 #ifdef INET6 1334 case AF_INET6: 1335 in6p = sotoin6pcb(so); 1336 tp = in6totcpcb(in6p); 1337 break; 1338 #endif 1339 } 1340 if (tp == NULL) 1341 goto badsyn; /*XXX*/ 1342 tiwin <<= tp->snd_scale; 1343 goto after_listen; 1344 } 1345 } else { 1346 /* 1347 * None of RST, SYN or ACK was set. 1348 * This is an invalid packet for a 1349 * TCB in LISTEN state. Send a RST. 1350 */ 1351 goto badsyn; 1352 } 1353 } else { 1354 /* 1355 * Received a SYN. 1356 */ 1357 1358 #ifdef INET6 1359 /* 1360 * If deprecated address is forbidden, we do 1361 * not accept SYN to deprecated interface 1362 * address to prevent any new inbound 1363 * connection from getting established. 1364 * When we do not accept SYN, we send a TCP 1365 * RST, with deprecated source address (instead 1366 * of dropping it). We compromise it as it is 1367 * much better for peer to send a RST, and 1368 * RST will be the final packet for the 1369 * exchange. 1370 * 1371 * If we do not forbid deprecated addresses, we 1372 * accept the SYN packet. RFC2462 does not 1373 * suggest dropping SYN in this case. 1374 * If we decipher RFC2462 5.5.4, it says like 1375 * this: 1376 * 1. use of deprecated addr with existing 1377 * communication is okay - "SHOULD continue 1378 * to be used" 1379 * 2. use of it with new communication: 1380 * (2a) "SHOULD NOT be used if alternate 1381 * address with sufficient scope is 1382 * available" 1383 * (2b) nothing mentioned otherwise. 1384 * Here we fall into (2b) case as we have no 1385 * choice in our source address selection - we 1386 * must obey the peer. 1387 * 1388 * The wording in RFC2462 is confusing, and 1389 * there are multiple description text for 1390 * deprecated address handling - worse, they 1391 * are not exactly the same. I believe 5.5.4 1392 * is the best one, so we follow 5.5.4. 1393 */ 1394 if (af == AF_INET6 && !ip6_use_deprecated) { 1395 struct in6_ifaddr *ia6; 1396 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1397 &ip6->ip6_dst)) && 1398 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1399 tp = NULL; 1400 goto dropwithreset; 1401 } 1402 } 1403 #endif 1404 1405 #ifdef IPSEC 1406 switch (af) { 1407 #ifdef INET 1408 case AF_INET: 1409 if (ipsec4_in_reject_so(m, so)) { 1410 ipsecstat.in_polvio++; 1411 tp = NULL; 1412 goto dropwithreset; 1413 } 1414 break; 1415 #endif 1416 #ifdef INET6 1417 case AF_INET6: 1418 if (ipsec6_in_reject_so(m, so)) { 1419 ipsec6stat.in_polvio++; 1420 tp = NULL; 1421 goto dropwithreset; 1422 } 1423 break; 1424 #endif 1425 } 1426 #endif 1427 1428 /* 1429 * LISTEN socket received a SYN 1430 * from itself? This can't possibly 1431 * be valid; drop the packet. 1432 */ 1433 if (th->th_sport == th->th_dport) { 1434 int i; 1435 1436 switch (af) { 1437 #ifdef INET 1438 case AF_INET: 1439 i = in_hosteq(ip->ip_src, ip->ip_dst); 1440 break; 1441 #endif 1442 #ifdef INET6 1443 case AF_INET6: 1444 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1445 break; 1446 #endif 1447 default: 1448 i = 1; 1449 } 1450 if (i) { 1451 tcpstat.tcps_badsyn++; 1452 goto drop; 1453 } 1454 } 1455 1456 /* 1457 * SYN looks ok; create compressed TCP 1458 * state for it. 1459 */ 1460 if (so->so_qlen <= so->so_qlimit && 1461 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1462 so, m, optp, optlen, &opti)) 1463 m = NULL; 1464 } 1465 goto drop; 1466 } 1467 } 1468 1469 after_listen: 1470 #ifdef DIAGNOSTIC 1471 /* 1472 * Should not happen now that all embryonic connections 1473 * are handled with compressed state. 1474 */ 1475 if (tp->t_state == TCPS_LISTEN) 1476 panic("tcp_input: TCPS_LISTEN"); 1477 #endif 1478 1479 /* 1480 * Segment received on connection. 1481 * Reset idle time and keep-alive timer. 1482 */ 1483 tp->t_rcvtime = tcp_now; 1484 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1485 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1486 1487 /* 1488 * Process options. 1489 */ 1490 #ifdef TCP_SIGNATURE 1491 if (optp || (tp->t_flags & TF_SIGNATURE)) 1492 #else 1493 if (optp) 1494 #endif 1495 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1496 goto drop; 1497 1498 if (opti.ts_present && opti.ts_ecr) { 1499 /* 1500 * Calculate the RTT from the returned time stamp and the 1501 * connection's time base. If the time stamp is later than 1502 * the current time, or is extremely old, fall back to non-1323 1503 * RTT calculation. Since ts_ecr is unsigned, we can test both 1504 * at the same time. 1505 */ 1506 opti.ts_ecr = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1507 if (opti.ts_ecr > TCP_PAWS_IDLE) 1508 opti.ts_ecr = 0; 1509 } 1510 1511 /* 1512 * Header prediction: check for the two common cases 1513 * of a uni-directional data xfer. If the packet has 1514 * no control flags, is in-sequence, the window didn't 1515 * change and we're not retransmitting, it's a 1516 * candidate. If the length is zero and the ack moved 1517 * forward, we're the sender side of the xfer. Just 1518 * free the data acked & wake any higher level process 1519 * that was blocked waiting for space. If the length 1520 * is non-zero and the ack didn't move, we're the 1521 * receiver side. If we're getting packets in-order 1522 * (the reassembly queue is empty), add the data to 1523 * the socket buffer and note that we need a delayed ack. 1524 */ 1525 if (tp->t_state == TCPS_ESTABLISHED && 1526 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1527 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1528 th->th_seq == tp->rcv_nxt && 1529 tiwin && tiwin == tp->snd_wnd && 1530 tp->snd_nxt == tp->snd_max) { 1531 1532 /* 1533 * If last ACK falls within this segment's sequence numbers, 1534 * record the timestamp. 1535 */ 1536 if (opti.ts_present && 1537 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1538 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1539 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1540 tp->ts_recent = opti.ts_val; 1541 } 1542 1543 if (tlen == 0) { 1544 /* Ack prediction. */ 1545 if (SEQ_GT(th->th_ack, tp->snd_una) && 1546 SEQ_LEQ(th->th_ack, tp->snd_max) && 1547 tp->snd_cwnd >= tp->snd_wnd && 1548 tp->t_partialacks < 0) { 1549 /* 1550 * this is a pure ack for outstanding data. 1551 */ 1552 ++tcpstat.tcps_predack; 1553 if (opti.ts_present && opti.ts_ecr) 1554 tcp_xmit_timer(tp, opti.ts_ecr); 1555 else if (tp->t_rtttime && 1556 SEQ_GT(th->th_ack, tp->t_rtseq)) 1557 tcp_xmit_timer(tp, 1558 tcp_now - tp->t_rtttime); 1559 acked = th->th_ack - tp->snd_una; 1560 tcpstat.tcps_rcvackpack++; 1561 tcpstat.tcps_rcvackbyte += acked; 1562 ND6_HINT(tp); 1563 1564 if (acked > (tp->t_lastoff - tp->t_inoff)) 1565 tp->t_lastm = NULL; 1566 sbdrop(&so->so_snd, acked); 1567 tp->t_lastoff -= acked; 1568 1569 tp->snd_una = th->th_ack; 1570 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1571 tp->snd_high = tp->snd_una; 1572 m_freem(m); 1573 1574 /* 1575 * If all outstanding data are acked, stop 1576 * retransmit timer, otherwise restart timer 1577 * using current (possibly backed-off) value. 1578 * If process is waiting for space, 1579 * wakeup/selwakeup/signal. If data 1580 * are ready to send, let tcp_output 1581 * decide between more output or persist. 1582 */ 1583 if (tp->snd_una == tp->snd_max) 1584 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1585 else if (TCP_TIMER_ISARMED(tp, 1586 TCPT_PERSIST) == 0) 1587 TCP_TIMER_ARM(tp, TCPT_REXMT, 1588 tp->t_rxtcur); 1589 1590 sowwakeup(so); 1591 if (so->so_snd.sb_cc) 1592 (void) tcp_output(tp); 1593 if (tcp_saveti) 1594 m_freem(tcp_saveti); 1595 return; 1596 } 1597 } else if (th->th_ack == tp->snd_una && 1598 TAILQ_FIRST(&tp->segq) == NULL && 1599 tlen <= sbspace(&so->so_rcv)) { 1600 /* 1601 * this is a pure, in-sequence data packet 1602 * with nothing on the reassembly queue and 1603 * we have enough buffer space to take it. 1604 */ 1605 ++tcpstat.tcps_preddat; 1606 tp->rcv_nxt += tlen; 1607 tcpstat.tcps_rcvpack++; 1608 tcpstat.tcps_rcvbyte += tlen; 1609 ND6_HINT(tp); 1610 /* 1611 * Drop TCP, IP headers and TCP options then add data 1612 * to socket buffer. 1613 */ 1614 if (so->so_state & SS_CANTRCVMORE) 1615 m_freem(m); 1616 else { 1617 m_adj(m, toff + off); 1618 sbappendstream(&so->so_rcv, m); 1619 } 1620 sorwakeup(so); 1621 TCP_SETUP_ACK(tp, th); 1622 if (tp->t_flags & TF_ACKNOW) 1623 (void) tcp_output(tp); 1624 if (tcp_saveti) 1625 m_freem(tcp_saveti); 1626 return; 1627 } 1628 } 1629 1630 /* 1631 * Compute mbuf offset to TCP data segment. 1632 */ 1633 hdroptlen = toff + off; 1634 1635 /* 1636 * Calculate amount of space in receive window, 1637 * and then do TCP input processing. 1638 * Receive window is amount of space in rcv queue, 1639 * but not less than advertised window. 1640 */ 1641 { int win; 1642 1643 win = sbspace(&so->so_rcv); 1644 if (win < 0) 1645 win = 0; 1646 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1647 } 1648 1649 switch (tp->t_state) { 1650 case TCPS_LISTEN: 1651 /* 1652 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1653 */ 1654 if (m->m_flags & (M_BCAST|M_MCAST)) 1655 goto drop; 1656 switch (af) { 1657 #ifdef INET6 1658 case AF_INET6: 1659 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1660 goto drop; 1661 break; 1662 #endif /* INET6 */ 1663 case AF_INET: 1664 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1665 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1666 goto drop; 1667 break; 1668 } 1669 break; 1670 1671 /* 1672 * If the state is SYN_SENT: 1673 * if seg contains an ACK, but not for our SYN, drop the input. 1674 * if seg contains a RST, then drop the connection. 1675 * if seg does not contain SYN, then drop it. 1676 * Otherwise this is an acceptable SYN segment 1677 * initialize tp->rcv_nxt and tp->irs 1678 * if seg contains ack then advance tp->snd_una 1679 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1680 * arrange for segment to be acked (eventually) 1681 * continue processing rest of data/controls, beginning with URG 1682 */ 1683 case TCPS_SYN_SENT: 1684 if ((tiflags & TH_ACK) && 1685 (SEQ_LEQ(th->th_ack, tp->iss) || 1686 SEQ_GT(th->th_ack, tp->snd_max))) 1687 goto dropwithreset; 1688 if (tiflags & TH_RST) { 1689 if (tiflags & TH_ACK) 1690 tp = tcp_drop(tp, ECONNREFUSED); 1691 goto drop; 1692 } 1693 if ((tiflags & TH_SYN) == 0) 1694 goto drop; 1695 if (tiflags & TH_ACK) { 1696 tp->snd_una = th->th_ack; 1697 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1698 tp->snd_nxt = tp->snd_una; 1699 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1700 tp->snd_high = tp->snd_una; 1701 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1702 } 1703 tp->irs = th->th_seq; 1704 tcp_rcvseqinit(tp); 1705 tp->t_flags |= TF_ACKNOW; 1706 tcp_mss_from_peer(tp, opti.maxseg); 1707 1708 /* 1709 * Initialize the initial congestion window. If we 1710 * had to retransmit the SYN, we must initialize cwnd 1711 * to 1 segment (i.e. the Loss Window). 1712 */ 1713 if (tp->t_flags & TF_SYN_REXMT) 1714 tp->snd_cwnd = tp->t_peermss; 1715 else { 1716 int ss = tcp_init_win; 1717 #ifdef INET 1718 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1719 ss = tcp_init_win_local; 1720 #endif 1721 #ifdef INET6 1722 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1723 ss = tcp_init_win_local; 1724 #endif 1725 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1726 } 1727 1728 tcp_rmx_rtt(tp); 1729 if (tiflags & TH_ACK) { 1730 tcpstat.tcps_connects++; 1731 soisconnected(so); 1732 tcp_established(tp); 1733 /* Do window scaling on this connection? */ 1734 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1735 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1736 tp->snd_scale = tp->requested_s_scale; 1737 tp->rcv_scale = tp->request_r_scale; 1738 } 1739 TCP_REASS_LOCK(tp); 1740 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1741 TCP_REASS_UNLOCK(tp); 1742 /* 1743 * if we didn't have to retransmit the SYN, 1744 * use its rtt as our initial srtt & rtt var. 1745 */ 1746 if (tp->t_rtttime) 1747 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1748 } else 1749 tp->t_state = TCPS_SYN_RECEIVED; 1750 1751 /* 1752 * Advance th->th_seq to correspond to first data byte. 1753 * If data, trim to stay within window, 1754 * dropping FIN if necessary. 1755 */ 1756 th->th_seq++; 1757 if (tlen > tp->rcv_wnd) { 1758 todrop = tlen - tp->rcv_wnd; 1759 m_adj(m, -todrop); 1760 tlen = tp->rcv_wnd; 1761 tiflags &= ~TH_FIN; 1762 tcpstat.tcps_rcvpackafterwin++; 1763 tcpstat.tcps_rcvbyteafterwin += todrop; 1764 } 1765 tp->snd_wl1 = th->th_seq - 1; 1766 tp->rcv_up = th->th_seq; 1767 goto step6; 1768 1769 /* 1770 * If the state is SYN_RECEIVED: 1771 * If seg contains an ACK, but not for our SYN, drop the input 1772 * and generate an RST. See page 36, rfc793 1773 */ 1774 case TCPS_SYN_RECEIVED: 1775 if ((tiflags & TH_ACK) && 1776 (SEQ_LEQ(th->th_ack, tp->iss) || 1777 SEQ_GT(th->th_ack, tp->snd_max))) 1778 goto dropwithreset; 1779 break; 1780 } 1781 1782 /* 1783 * States other than LISTEN or SYN_SENT. 1784 * First check timestamp, if present. 1785 * Then check that at least some bytes of segment are within 1786 * receive window. If segment begins before rcv_nxt, 1787 * drop leading data (and SYN); if nothing left, just ack. 1788 * 1789 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1790 * and it's less than ts_recent, drop it. 1791 */ 1792 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1793 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1794 1795 /* Check to see if ts_recent is over 24 days old. */ 1796 if (TCP_TIMESTAMP(tp) - tp->ts_recent_age > TCP_PAWS_IDLE) { 1797 /* 1798 * Invalidate ts_recent. If this segment updates 1799 * ts_recent, the age will be reset later and ts_recent 1800 * will get a valid value. If it does not, setting 1801 * ts_recent to zero will at least satisfy the 1802 * requirement that zero be placed in the timestamp 1803 * echo reply when ts_recent isn't valid. The 1804 * age isn't reset until we get a valid ts_recent 1805 * because we don't want out-of-order segments to be 1806 * dropped when ts_recent is old. 1807 */ 1808 tp->ts_recent = 0; 1809 } else { 1810 tcpstat.tcps_rcvduppack++; 1811 tcpstat.tcps_rcvdupbyte += tlen; 1812 tcpstat.tcps_pawsdrop++; 1813 goto dropafterack; 1814 } 1815 } 1816 1817 todrop = tp->rcv_nxt - th->th_seq; 1818 if (todrop > 0) { 1819 if (tiflags & TH_SYN) { 1820 tiflags &= ~TH_SYN; 1821 th->th_seq++; 1822 if (th->th_urp > 1) 1823 th->th_urp--; 1824 else { 1825 tiflags &= ~TH_URG; 1826 th->th_urp = 0; 1827 } 1828 todrop--; 1829 } 1830 if (todrop > tlen || 1831 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1832 /* 1833 * Any valid FIN or RST must be to the left of the 1834 * window. At this point the FIN or RST must be a 1835 * duplicate or out of sequence; drop it. 1836 */ 1837 if (tiflags & TH_RST) 1838 goto drop; 1839 tiflags &= ~(TH_FIN|TH_RST); 1840 /* 1841 * Send an ACK to resynchronize and drop any data. 1842 * But keep on processing for RST or ACK. 1843 */ 1844 tp->t_flags |= TF_ACKNOW; 1845 todrop = tlen; 1846 tcpstat.tcps_rcvdupbyte += todrop; 1847 tcpstat.tcps_rcvduppack++; 1848 } else if ((tiflags & TH_RST) && 1849 th->th_seq != tp->last_ack_sent) { 1850 /* 1851 * Test for reset before adjusting the sequence 1852 * number for overlapping data. 1853 */ 1854 goto dropafterack_ratelim; 1855 } else { 1856 tcpstat.tcps_rcvpartduppack++; 1857 tcpstat.tcps_rcvpartdupbyte += todrop; 1858 } 1859 hdroptlen += todrop; /*drop from head afterwards*/ 1860 th->th_seq += todrop; 1861 tlen -= todrop; 1862 if (th->th_urp > todrop) 1863 th->th_urp -= todrop; 1864 else { 1865 tiflags &= ~TH_URG; 1866 th->th_urp = 0; 1867 } 1868 } 1869 1870 /* 1871 * If new data are received on a connection after the 1872 * user processes are gone, then RST the other end. 1873 */ 1874 if ((so->so_state & SS_NOFDREF) && 1875 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1876 tp = tcp_close(tp); 1877 tcpstat.tcps_rcvafterclose++; 1878 goto dropwithreset; 1879 } 1880 1881 /* 1882 * If segment ends after window, drop trailing data 1883 * (and PUSH and FIN); if nothing left, just ACK. 1884 */ 1885 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1886 if (todrop > 0) { 1887 tcpstat.tcps_rcvpackafterwin++; 1888 if (todrop >= tlen) { 1889 /* 1890 * The segment actually starts after the window. 1891 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 1892 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 1893 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 1894 */ 1895 tcpstat.tcps_rcvbyteafterwin += tlen; 1896 /* 1897 * If a new connection request is received 1898 * while in TIME_WAIT, drop the old connection 1899 * and start over if the sequence numbers 1900 * are above the previous ones. 1901 * 1902 * NOTE: We will checksum the packet again, and 1903 * so we need to put the header fields back into 1904 * network order! 1905 * XXX This kind of sucks, but we don't expect 1906 * XXX this to happen very often, so maybe it 1907 * XXX doesn't matter so much. 1908 */ 1909 if (tiflags & TH_SYN && 1910 tp->t_state == TCPS_TIME_WAIT && 1911 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1912 iss = tcp_new_iss(tp, tp->snd_nxt); 1913 tp = tcp_close(tp); 1914 TCP_FIELDS_TO_NET(th); 1915 goto findpcb; 1916 } 1917 /* 1918 * If window is closed can only take segments at 1919 * window edge, and have to drop data and PUSH from 1920 * incoming segments. Continue processing, but 1921 * remember to ack. Otherwise, drop segment 1922 * and (if not RST) ack. 1923 */ 1924 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1925 tp->t_flags |= TF_ACKNOW; 1926 tcpstat.tcps_rcvwinprobe++; 1927 } else 1928 goto dropafterack; 1929 } else 1930 tcpstat.tcps_rcvbyteafterwin += todrop; 1931 m_adj(m, -todrop); 1932 tlen -= todrop; 1933 tiflags &= ~(TH_PUSH|TH_FIN); 1934 } 1935 1936 /* 1937 * If last ACK falls within this segment's sequence numbers, 1938 * and the timestamp is newer, record it. 1939 */ 1940 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1941 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1942 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1943 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1944 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1945 tp->ts_recent = opti.ts_val; 1946 } 1947 1948 /* 1949 * If the RST bit is set examine the state: 1950 * SYN_RECEIVED STATE: 1951 * If passive open, return to LISTEN state. 1952 * If active open, inform user that connection was refused. 1953 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1954 * Inform user that connection was reset, and close tcb. 1955 * CLOSING, LAST_ACK, TIME_WAIT STATES 1956 * Close the tcb. 1957 */ 1958 if (tiflags & TH_RST) { 1959 if (th->th_seq != tp->last_ack_sent) 1960 goto dropafterack_ratelim; 1961 1962 switch (tp->t_state) { 1963 case TCPS_SYN_RECEIVED: 1964 so->so_error = ECONNREFUSED; 1965 goto close; 1966 1967 case TCPS_ESTABLISHED: 1968 case TCPS_FIN_WAIT_1: 1969 case TCPS_FIN_WAIT_2: 1970 case TCPS_CLOSE_WAIT: 1971 so->so_error = ECONNRESET; 1972 close: 1973 tp->t_state = TCPS_CLOSED; 1974 tcpstat.tcps_drops++; 1975 tp = tcp_close(tp); 1976 goto drop; 1977 1978 case TCPS_CLOSING: 1979 case TCPS_LAST_ACK: 1980 case TCPS_TIME_WAIT: 1981 tp = tcp_close(tp); 1982 goto drop; 1983 } 1984 } 1985 1986 /* 1987 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 1988 * we must be in a synchronized state. RFC791 states (under RST 1989 * generation) that any unacceptable segment (an out-of-order SYN 1990 * qualifies) received in a synchronized state must elicit only an 1991 * empty acknowledgment segment ... and the connection remains in 1992 * the same state. 1993 */ 1994 if (tiflags & TH_SYN) { 1995 if (tp->rcv_nxt == th->th_seq) { 1996 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 1997 TH_ACK); 1998 if (tcp_saveti) 1999 m_freem(tcp_saveti); 2000 return; 2001 } 2002 2003 goto dropafterack_ratelim; 2004 } 2005 2006 /* 2007 * If the ACK bit is off we drop the segment and return. 2008 */ 2009 if ((tiflags & TH_ACK) == 0) { 2010 if (tp->t_flags & TF_ACKNOW) 2011 goto dropafterack; 2012 else 2013 goto drop; 2014 } 2015 2016 /* 2017 * Ack processing. 2018 */ 2019 switch (tp->t_state) { 2020 2021 /* 2022 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2023 * ESTABLISHED state and continue processing, otherwise 2024 * send an RST. 2025 */ 2026 case TCPS_SYN_RECEIVED: 2027 if (SEQ_GT(tp->snd_una, th->th_ack) || 2028 SEQ_GT(th->th_ack, tp->snd_max)) 2029 goto dropwithreset; 2030 tcpstat.tcps_connects++; 2031 soisconnected(so); 2032 tcp_established(tp); 2033 /* Do window scaling? */ 2034 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2035 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2036 tp->snd_scale = tp->requested_s_scale; 2037 tp->rcv_scale = tp->request_r_scale; 2038 } 2039 TCP_REASS_LOCK(tp); 2040 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2041 TCP_REASS_UNLOCK(tp); 2042 tp->snd_wl1 = th->th_seq - 1; 2043 /* fall into ... */ 2044 2045 /* 2046 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2047 * ACKs. If the ack is in the range 2048 * tp->snd_una < th->th_ack <= tp->snd_max 2049 * then advance tp->snd_una to th->th_ack and drop 2050 * data from the retransmission queue. If this ACK reflects 2051 * more up to date window information we update our window information. 2052 */ 2053 case TCPS_ESTABLISHED: 2054 case TCPS_FIN_WAIT_1: 2055 case TCPS_FIN_WAIT_2: 2056 case TCPS_CLOSE_WAIT: 2057 case TCPS_CLOSING: 2058 case TCPS_LAST_ACK: 2059 case TCPS_TIME_WAIT: 2060 2061 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2062 if (tlen == 0 && tiwin == tp->snd_wnd) { 2063 tcpstat.tcps_rcvdupack++; 2064 /* 2065 * If we have outstanding data (other than 2066 * a window probe), this is a completely 2067 * duplicate ack (ie, window info didn't 2068 * change), the ack is the biggest we've 2069 * seen and we've seen exactly our rexmt 2070 * threshhold of them, assume a packet 2071 * has been dropped and retransmit it. 2072 * Kludge snd_nxt & the congestion 2073 * window so we send only this one 2074 * packet. 2075 * 2076 * We know we're losing at the current 2077 * window size so do congestion avoidance 2078 * (set ssthresh to half the current window 2079 * and pull our congestion window back to 2080 * the new ssthresh). 2081 * 2082 * Dup acks mean that packets have left the 2083 * network (they're now cached at the receiver) 2084 * so bump cwnd by the amount in the receiver 2085 * to keep a constant cwnd packets in the 2086 * network. 2087 */ 2088 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2089 th->th_ack != tp->snd_una) 2090 tp->t_dupacks = 0; 2091 else if (++tp->t_dupacks == tcprexmtthresh && 2092 tp->t_partialacks < 0) { 2093 tcp_seq onxt; 2094 u_int win; 2095 2096 if (tcp_do_newreno && 2097 SEQ_LT(th->th_ack, tp->snd_high)) { 2098 /* 2099 * False fast retransmit after 2100 * timeout. Do not enter fast 2101 * recovery. 2102 */ 2103 tp->t_dupacks = 0; 2104 break; 2105 } 2106 2107 onxt = tp->snd_nxt; 2108 win = min(tp->snd_wnd, tp->snd_cwnd) / 2109 2 / tp->t_segsz; 2110 if (win < 2) 2111 win = 2; 2112 tp->snd_ssthresh = win * tp->t_segsz; 2113 tp->snd_recover = tp->snd_max; 2114 tp->t_partialacks = 0; 2115 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2116 tp->t_rtttime = 0; 2117 tp->snd_nxt = th->th_ack; 2118 tp->snd_cwnd = tp->t_segsz; 2119 (void) tcp_output(tp); 2120 tp->snd_cwnd = tp->snd_ssthresh + 2121 tp->t_segsz * tp->t_dupacks; 2122 if (SEQ_GT(onxt, tp->snd_nxt)) 2123 tp->snd_nxt = onxt; 2124 goto drop; 2125 } else if (tp->t_dupacks > tcprexmtthresh) { 2126 tp->snd_cwnd += tp->t_segsz; 2127 (void) tcp_output(tp); 2128 goto drop; 2129 } 2130 } else if (tlen) { 2131 tp->t_dupacks = 0; /*XXX*/ 2132 /* drop very old ACKs unless th_seq matches */ 2133 if (th->th_seq != tp->rcv_nxt && 2134 SEQ_LT(th->th_ack, 2135 tp->snd_una - tp->max_sndwnd)) { 2136 goto drop; 2137 } 2138 break; 2139 } else 2140 tp->t_dupacks = 0; 2141 break; 2142 } 2143 /* 2144 * If the congestion window was inflated to account 2145 * for the other side's cached packets, retract it. 2146 */ 2147 if (!tcp_do_newreno) 2148 tcp_reno_newack(tp, th); 2149 else 2150 tcp_newreno_newack(tp, th); 2151 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2152 tcpstat.tcps_rcvacktoomuch++; 2153 goto dropafterack; 2154 } 2155 acked = th->th_ack - tp->snd_una; 2156 tcpstat.tcps_rcvackpack++; 2157 tcpstat.tcps_rcvackbyte += acked; 2158 2159 /* 2160 * If we have a timestamp reply, update smoothed 2161 * round trip time. If no timestamp is present but 2162 * transmit timer is running and timed sequence 2163 * number was acked, update smoothed round trip time. 2164 * Since we now have an rtt measurement, cancel the 2165 * timer backoff (cf., Phil Karn's retransmit alg.). 2166 * Recompute the initial retransmit timer. 2167 */ 2168 if (opti.ts_present && opti.ts_ecr) 2169 tcp_xmit_timer(tp, opti.ts_ecr); 2170 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2171 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2172 2173 /* 2174 * If all outstanding data is acked, stop retransmit 2175 * timer and remember to restart (more output or persist). 2176 * If there is more data to be acked, restart retransmit 2177 * timer, using current (possibly backed-off) value. 2178 */ 2179 if (th->th_ack == tp->snd_max) { 2180 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2181 needoutput = 1; 2182 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2183 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2184 /* 2185 * When new data is acked, open the congestion window. 2186 * If the window gives us less than ssthresh packets 2187 * in flight, open exponentially (segsz per packet). 2188 * Otherwise open linearly: segsz per window 2189 * (segsz^2 / cwnd per packet), plus a constant 2190 * fraction of a packet (segsz/8) to help larger windows 2191 * open quickly enough. 2192 * 2193 * If we are still in fast recovery (meaning we are using 2194 * NewReno and we have only received partial acks), do not 2195 * inflate the window yet. 2196 */ 2197 if (tp->t_partialacks < 0) { 2198 u_int cw = tp->snd_cwnd; 2199 u_int incr = tp->t_segsz; 2200 2201 if (cw > tp->snd_ssthresh) 2202 incr = incr * incr / cw; 2203 tp->snd_cwnd = min(cw + incr, 2204 TCP_MAXWIN << tp->snd_scale); 2205 } 2206 ND6_HINT(tp); 2207 if (acked > so->so_snd.sb_cc) { 2208 tp->snd_wnd -= so->so_snd.sb_cc; 2209 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2210 ourfinisacked = 1; 2211 } else { 2212 if (acked > (tp->t_lastoff - tp->t_inoff)) 2213 tp->t_lastm = NULL; 2214 sbdrop(&so->so_snd, acked); 2215 tp->t_lastoff -= acked; 2216 tp->snd_wnd -= acked; 2217 ourfinisacked = 0; 2218 } 2219 sowwakeup(so); 2220 tp->snd_una = th->th_ack; 2221 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2222 tp->snd_nxt = tp->snd_una; 2223 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2224 tp->snd_high = tp->snd_una; 2225 2226 switch (tp->t_state) { 2227 2228 /* 2229 * In FIN_WAIT_1 STATE in addition to the processing 2230 * for the ESTABLISHED state if our FIN is now acknowledged 2231 * then enter FIN_WAIT_2. 2232 */ 2233 case TCPS_FIN_WAIT_1: 2234 if (ourfinisacked) { 2235 /* 2236 * If we can't receive any more 2237 * data, then closing user can proceed. 2238 * Starting the timer is contrary to the 2239 * specification, but if we don't get a FIN 2240 * we'll hang forever. 2241 */ 2242 if (so->so_state & SS_CANTRCVMORE) { 2243 soisdisconnected(so); 2244 if (tcp_maxidle > 0) 2245 TCP_TIMER_ARM(tp, TCPT_2MSL, 2246 tcp_maxidle); 2247 } 2248 tp->t_state = TCPS_FIN_WAIT_2; 2249 } 2250 break; 2251 2252 /* 2253 * In CLOSING STATE in addition to the processing for 2254 * the ESTABLISHED state if the ACK acknowledges our FIN 2255 * then enter the TIME-WAIT state, otherwise ignore 2256 * the segment. 2257 */ 2258 case TCPS_CLOSING: 2259 if (ourfinisacked) { 2260 tp->t_state = TCPS_TIME_WAIT; 2261 tcp_canceltimers(tp); 2262 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2263 soisdisconnected(so); 2264 } 2265 break; 2266 2267 /* 2268 * In LAST_ACK, we may still be waiting for data to drain 2269 * and/or to be acked, as well as for the ack of our FIN. 2270 * If our FIN is now acknowledged, delete the TCB, 2271 * enter the closed state and return. 2272 */ 2273 case TCPS_LAST_ACK: 2274 if (ourfinisacked) { 2275 tp = tcp_close(tp); 2276 goto drop; 2277 } 2278 break; 2279 2280 /* 2281 * In TIME_WAIT state the only thing that should arrive 2282 * is a retransmission of the remote FIN. Acknowledge 2283 * it and restart the finack timer. 2284 */ 2285 case TCPS_TIME_WAIT: 2286 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2287 goto dropafterack; 2288 } 2289 } 2290 2291 step6: 2292 /* 2293 * Update window information. 2294 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2295 */ 2296 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2297 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 2298 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 2299 /* keep track of pure window updates */ 2300 if (tlen == 0 && 2301 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2302 tcpstat.tcps_rcvwinupd++; 2303 tp->snd_wnd = tiwin; 2304 tp->snd_wl1 = th->th_seq; 2305 tp->snd_wl2 = th->th_ack; 2306 if (tp->snd_wnd > tp->max_sndwnd) 2307 tp->max_sndwnd = tp->snd_wnd; 2308 needoutput = 1; 2309 } 2310 2311 /* 2312 * Process segments with URG. 2313 */ 2314 if ((tiflags & TH_URG) && th->th_urp && 2315 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2316 /* 2317 * This is a kludge, but if we receive and accept 2318 * random urgent pointers, we'll crash in 2319 * soreceive. It's hard to imagine someone 2320 * actually wanting to send this much urgent data. 2321 */ 2322 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2323 th->th_urp = 0; /* XXX */ 2324 tiflags &= ~TH_URG; /* XXX */ 2325 goto dodata; /* XXX */ 2326 } 2327 /* 2328 * If this segment advances the known urgent pointer, 2329 * then mark the data stream. This should not happen 2330 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2331 * a FIN has been received from the remote side. 2332 * In these states we ignore the URG. 2333 * 2334 * According to RFC961 (Assigned Protocols), 2335 * the urgent pointer points to the last octet 2336 * of urgent data. We continue, however, 2337 * to consider it to indicate the first octet 2338 * of data past the urgent section as the original 2339 * spec states (in one of two places). 2340 */ 2341 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2342 tp->rcv_up = th->th_seq + th->th_urp; 2343 so->so_oobmark = so->so_rcv.sb_cc + 2344 (tp->rcv_up - tp->rcv_nxt) - 1; 2345 if (so->so_oobmark == 0) 2346 so->so_state |= SS_RCVATMARK; 2347 sohasoutofband(so); 2348 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2349 } 2350 /* 2351 * Remove out of band data so doesn't get presented to user. 2352 * This can happen independent of advancing the URG pointer, 2353 * but if two URG's are pending at once, some out-of-band 2354 * data may creep in... ick. 2355 */ 2356 if (th->th_urp <= (u_int16_t) tlen 2357 #ifdef SO_OOBINLINE 2358 && (so->so_options & SO_OOBINLINE) == 0 2359 #endif 2360 ) 2361 tcp_pulloutofband(so, th, m, hdroptlen); 2362 } else 2363 /* 2364 * If no out of band data is expected, 2365 * pull receive urgent pointer along 2366 * with the receive window. 2367 */ 2368 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2369 tp->rcv_up = tp->rcv_nxt; 2370 dodata: /* XXX */ 2371 2372 /* 2373 * Process the segment text, merging it into the TCP sequencing queue, 2374 * and arranging for acknowledgement of receipt if necessary. 2375 * This process logically involves adjusting tp->rcv_wnd as data 2376 * is presented to the user (this happens in tcp_usrreq.c, 2377 * case PRU_RCVD). If a FIN has already been received on this 2378 * connection then we just ignore the text. 2379 */ 2380 if ((tlen || (tiflags & TH_FIN)) && 2381 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2382 /* 2383 * Insert segment ti into reassembly queue of tcp with 2384 * control block tp. Return TH_FIN if reassembly now includes 2385 * a segment with FIN. The macro form does the common case 2386 * inline (segment is the next to be received on an 2387 * established connection, and the queue is empty), 2388 * avoiding linkage into and removal from the queue and 2389 * repetition of various conversions. 2390 * Set DELACK for segments received in order, but ack 2391 * immediately when segments are out of order 2392 * (so fast retransmit can work). 2393 */ 2394 /* NOTE: this was TCP_REASS() macro, but used only once */ 2395 TCP_REASS_LOCK(tp); 2396 if (th->th_seq == tp->rcv_nxt && 2397 TAILQ_FIRST(&tp->segq) == NULL && 2398 tp->t_state == TCPS_ESTABLISHED) { 2399 TCP_SETUP_ACK(tp, th); 2400 tp->rcv_nxt += tlen; 2401 tiflags = th->th_flags & TH_FIN; 2402 tcpstat.tcps_rcvpack++; 2403 tcpstat.tcps_rcvbyte += tlen; 2404 ND6_HINT(tp); 2405 if (so->so_state & SS_CANTRCVMORE) 2406 m_freem(m); 2407 else { 2408 m_adj(m, hdroptlen); 2409 sbappendstream(&(so)->so_rcv, m); 2410 } 2411 sorwakeup(so); 2412 } else { 2413 m_adj(m, hdroptlen); 2414 tiflags = tcp_reass(tp, th, m, &tlen); 2415 tp->t_flags |= TF_ACKNOW; 2416 } 2417 TCP_REASS_UNLOCK(tp); 2418 2419 /* 2420 * Note the amount of data that peer has sent into 2421 * our window, in order to estimate the sender's 2422 * buffer size. 2423 */ 2424 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2425 } else { 2426 m_freem(m); 2427 m = NULL; 2428 tiflags &= ~TH_FIN; 2429 } 2430 2431 /* 2432 * If FIN is received ACK the FIN and let the user know 2433 * that the connection is closing. Ignore a FIN received before 2434 * the connection is fully established. 2435 */ 2436 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2437 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2438 socantrcvmore(so); 2439 tp->t_flags |= TF_ACKNOW; 2440 tp->rcv_nxt++; 2441 } 2442 switch (tp->t_state) { 2443 2444 /* 2445 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2446 */ 2447 case TCPS_ESTABLISHED: 2448 tp->t_state = TCPS_CLOSE_WAIT; 2449 break; 2450 2451 /* 2452 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2453 * enter the CLOSING state. 2454 */ 2455 case TCPS_FIN_WAIT_1: 2456 tp->t_state = TCPS_CLOSING; 2457 break; 2458 2459 /* 2460 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2461 * starting the time-wait timer, turning off the other 2462 * standard timers. 2463 */ 2464 case TCPS_FIN_WAIT_2: 2465 tp->t_state = TCPS_TIME_WAIT; 2466 tcp_canceltimers(tp); 2467 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2468 soisdisconnected(so); 2469 break; 2470 2471 /* 2472 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2473 */ 2474 case TCPS_TIME_WAIT: 2475 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2476 break; 2477 } 2478 } 2479 #ifdef TCP_DEBUG 2480 if (so->so_options & SO_DEBUG) 2481 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2482 #endif 2483 2484 /* 2485 * Return any desired output. 2486 */ 2487 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2488 (void) tcp_output(tp); 2489 if (tcp_saveti) 2490 m_freem(tcp_saveti); 2491 return; 2492 2493 badsyn: 2494 /* 2495 * Received a bad SYN. Increment counters and dropwithreset. 2496 */ 2497 tcpstat.tcps_badsyn++; 2498 tp = NULL; 2499 goto dropwithreset; 2500 2501 dropafterack: 2502 /* 2503 * Generate an ACK dropping incoming segment if it occupies 2504 * sequence space, where the ACK reflects our state. 2505 */ 2506 if (tiflags & TH_RST) 2507 goto drop; 2508 goto dropafterack2; 2509 2510 dropafterack_ratelim: 2511 /* 2512 * We may want to rate-limit ACKs against SYN/RST attack. 2513 */ 2514 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2515 tcp_ackdrop_ppslim) == 0) { 2516 /* XXX stat */ 2517 goto drop; 2518 } 2519 /* ...fall into dropafterack2... */ 2520 2521 dropafterack2: 2522 m_freem(m); 2523 tp->t_flags |= TF_ACKNOW; 2524 (void) tcp_output(tp); 2525 if (tcp_saveti) 2526 m_freem(tcp_saveti); 2527 return; 2528 2529 dropwithreset_ratelim: 2530 /* 2531 * We may want to rate-limit RSTs in certain situations, 2532 * particularly if we are sending an RST in response to 2533 * an attempt to connect to or otherwise communicate with 2534 * a port for which we have no socket. 2535 */ 2536 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2537 tcp_rst_ppslim) == 0) { 2538 /* XXX stat */ 2539 goto drop; 2540 } 2541 /* ...fall into dropwithreset... */ 2542 2543 dropwithreset: 2544 /* 2545 * Generate a RST, dropping incoming segment. 2546 * Make ACK acceptable to originator of segment. 2547 */ 2548 if (tiflags & TH_RST) 2549 goto drop; 2550 2551 switch (af) { 2552 #ifdef INET6 2553 case AF_INET6: 2554 /* For following calls to tcp_respond */ 2555 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2556 goto drop; 2557 break; 2558 #endif /* INET6 */ 2559 case AF_INET: 2560 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2561 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2562 goto drop; 2563 } 2564 2565 if (tiflags & TH_ACK) 2566 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2567 else { 2568 if (tiflags & TH_SYN) 2569 tlen++; 2570 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2571 TH_RST|TH_ACK); 2572 } 2573 if (tcp_saveti) 2574 m_freem(tcp_saveti); 2575 return; 2576 2577 badcsum: 2578 drop: 2579 /* 2580 * Drop space held by incoming segment and return. 2581 */ 2582 if (tp) { 2583 if (tp->t_inpcb) 2584 so = tp->t_inpcb->inp_socket; 2585 #ifdef INET6 2586 else if (tp->t_in6pcb) 2587 so = tp->t_in6pcb->in6p_socket; 2588 #endif 2589 else 2590 so = NULL; 2591 #ifdef TCP_DEBUG 2592 if (so && (so->so_options & SO_DEBUG) != 0) 2593 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2594 #endif 2595 } 2596 if (tcp_saveti) 2597 m_freem(tcp_saveti); 2598 m_freem(m); 2599 return; 2600 } 2601 2602 #ifdef TCP_SIGNATURE 2603 int 2604 tcp_signature_apply(void *fstate, caddr_t data, u_int len) 2605 { 2606 2607 MD5Update(fstate, (u_char *)data, len); 2608 return (0); 2609 } 2610 2611 struct secasvar * 2612 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 2613 { 2614 struct secasvar *sav; 2615 #ifdef FAST_IPSEC 2616 union sockaddr_union dst; 2617 #endif 2618 struct ip *ip; 2619 struct ip6_hdr *ip6; 2620 2621 ip = mtod(m, struct ip *); 2622 switch (ip->ip_v) { 2623 case 4: 2624 ip = mtod(m, struct ip *); 2625 ip6 = NULL; 2626 break; 2627 case 6: 2628 ip = NULL; 2629 ip6 = mtod(m, struct ip6_hdr *); 2630 break; 2631 default: 2632 return (NULL); 2633 } 2634 2635 #ifdef FAST_IPSEC 2636 /* Extract the destination from the IP header in the mbuf. */ 2637 bzero(&dst, sizeof(union sockaddr_union)); 2638 dst.sa.sa_len = sizeof(struct sockaddr_in); 2639 dst.sa.sa_family = AF_INET; 2640 dst.sin.sin_addr = ip->ip_dst; 2641 2642 /* 2643 * Look up an SADB entry which matches the address of the peer. 2644 */ 2645 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2646 #else 2647 if (ip) 2648 sav = key_allocsa(AF_INET, (caddr_t)&ip->ip_src, 2649 (caddr_t)&ip->ip_dst, IPPROTO_TCP, 2650 htonl(TCP_SIG_SPI)); 2651 else 2652 sav = key_allocsa(AF_INET6, (caddr_t)&ip6->ip6_src, 2653 (caddr_t)&ip6->ip6_dst, IPPROTO_TCP, 2654 htonl(TCP_SIG_SPI)); 2655 #endif 2656 2657 return (sav); /* freesav must be performed by caller */ 2658 } 2659 2660 int 2661 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 2662 struct secasvar *sav, char *sig) 2663 { 2664 MD5_CTX ctx; 2665 struct ip *ip; 2666 struct ipovly *ipovly; 2667 struct ip6_hdr *ip6; 2668 struct ippseudo ippseudo; 2669 struct ip6_hdr_pseudo ip6pseudo; 2670 struct tcphdr th0; 2671 int l, tcphdrlen; 2672 2673 if (sav == NULL) 2674 return (-1); 2675 2676 tcphdrlen = th->th_off * 4; 2677 2678 switch (mtod(m, struct ip *)->ip_v) { 2679 case 4: 2680 ip = mtod(m, struct ip *); 2681 ip6 = NULL; 2682 break; 2683 case 6: 2684 ip = NULL; 2685 ip6 = mtod(m, struct ip6_hdr *); 2686 break; 2687 default: 2688 return (-1); 2689 } 2690 2691 MD5Init(&ctx); 2692 2693 if (ip) { 2694 memset(&ippseudo, 0, sizeof(ippseudo)); 2695 ipovly = (struct ipovly *)ip; 2696 ippseudo.ippseudo_src = ipovly->ih_src; 2697 ippseudo.ippseudo_dst = ipovly->ih_dst; 2698 ippseudo.ippseudo_pad = 0; 2699 ippseudo.ippseudo_p = IPPROTO_TCP; 2700 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 2701 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 2702 } else { 2703 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 2704 ip6pseudo.ip6ph_src = ip6->ip6_src; 2705 in6_clearscope(&ip6pseudo.ip6ph_src); 2706 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 2707 in6_clearscope(&ip6pseudo.ip6ph_dst); 2708 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 2709 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 2710 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 2711 } 2712 2713 th0 = *th; 2714 th0.th_sum = 0; 2715 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 2716 2717 l = m->m_pkthdr.len - thoff - tcphdrlen; 2718 if (l > 0) 2719 m_apply(m, thoff + tcphdrlen, 2720 m->m_pkthdr.len - thoff - tcphdrlen, 2721 tcp_signature_apply, &ctx); 2722 2723 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2724 MD5Final(sig, &ctx); 2725 2726 return (0); 2727 } 2728 #endif 2729 2730 int 2731 tcp_dooptions(tp, cp, cnt, th, m, toff, oi) 2732 struct tcpcb *tp; 2733 u_char *cp; 2734 int cnt; 2735 struct tcphdr *th; 2736 struct mbuf *m; 2737 int toff; 2738 struct tcp_opt_info *oi; 2739 { 2740 u_int16_t mss; 2741 int opt, optlen = 0; 2742 #ifdef TCP_SIGNATURE 2743 caddr_t sigp = NULL; 2744 char sigbuf[TCP_SIGLEN]; 2745 struct secasvar *sav = NULL; 2746 #endif 2747 2748 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 2749 opt = cp[0]; 2750 if (opt == TCPOPT_EOL) 2751 break; 2752 if (opt == TCPOPT_NOP) 2753 optlen = 1; 2754 else { 2755 if (cnt < 2) 2756 break; 2757 optlen = cp[1]; 2758 if (optlen < 2 || optlen > cnt) 2759 break; 2760 } 2761 switch (opt) { 2762 2763 default: 2764 continue; 2765 2766 case TCPOPT_MAXSEG: 2767 if (optlen != TCPOLEN_MAXSEG) 2768 continue; 2769 if (!(th->th_flags & TH_SYN)) 2770 continue; 2771 bcopy(cp + 2, &mss, sizeof(mss)); 2772 oi->maxseg = ntohs(mss); 2773 break; 2774 2775 case TCPOPT_WINDOW: 2776 if (optlen != TCPOLEN_WINDOW) 2777 continue; 2778 if (!(th->th_flags & TH_SYN)) 2779 continue; 2780 tp->t_flags |= TF_RCVD_SCALE; 2781 tp->requested_s_scale = cp[2]; 2782 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2783 #if 0 /*XXX*/ 2784 char *p; 2785 2786 if (ip) 2787 p = ntohl(ip->ip_src); 2788 #ifdef INET6 2789 else if (ip6) 2790 p = ip6_sprintf(&ip6->ip6_src); 2791 #endif 2792 else 2793 p = "(unknown)"; 2794 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2795 "assuming %d\n", 2796 tp->requested_s_scale, p, 2797 TCP_MAX_WINSHIFT); 2798 #else 2799 log(LOG_ERR, "TCP: invalid wscale %d, " 2800 "assuming %d\n", 2801 tp->requested_s_scale, 2802 TCP_MAX_WINSHIFT); 2803 #endif 2804 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2805 } 2806 break; 2807 2808 case TCPOPT_TIMESTAMP: 2809 if (optlen != TCPOLEN_TIMESTAMP) 2810 continue; 2811 oi->ts_present = 1; 2812 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2813 NTOHL(oi->ts_val); 2814 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2815 NTOHL(oi->ts_ecr); 2816 2817 /* 2818 * A timestamp received in a SYN makes 2819 * it ok to send timestamp requests and replies. 2820 */ 2821 if (th->th_flags & TH_SYN) { 2822 tp->t_flags |= TF_RCVD_TSTMP; 2823 tp->ts_recent = oi->ts_val; 2824 tp->ts_recent_age = TCP_TIMESTAMP(tp); 2825 } 2826 break; 2827 case TCPOPT_SACK_PERMITTED: 2828 if (optlen != TCPOLEN_SACK_PERMITTED) 2829 continue; 2830 if (!(th->th_flags & TH_SYN)) 2831 continue; 2832 tp->t_flags &= ~TF_CANT_TXSACK; 2833 break; 2834 2835 case TCPOPT_SACK: 2836 if (tp->t_flags & TF_IGNR_RXSACK) 2837 continue; 2838 if (optlen % 8 != 2 || optlen < 10) 2839 continue; 2840 cp += 2; 2841 optlen -= 2; 2842 for (; optlen > 0; cp -= 8, optlen -= 8) { 2843 tcp_seq lwe, rwe; 2844 bcopy((char *)cp, (char *) &lwe, sizeof(lwe)); 2845 NTOHL(lwe); 2846 bcopy((char *)cp, (char *) &rwe, sizeof(rwe)); 2847 NTOHL(rwe); 2848 /* tcp_mark_sacked(tp, lwe, rwe); */ 2849 } 2850 break; 2851 #ifdef TCP_SIGNATURE 2852 case TCPOPT_SIGNATURE: 2853 if (optlen != TCPOLEN_SIGNATURE) 2854 continue; 2855 if (sigp && bcmp(sigp, cp + 2, TCP_SIGLEN)) 2856 return (-1); 2857 2858 sigp = sigbuf; 2859 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 2860 memset(cp + 2, 0, TCP_SIGLEN); 2861 tp->t_flags |= TF_SIGNATURE; 2862 break; 2863 #endif 2864 } 2865 } 2866 2867 #ifdef TCP_SIGNATURE 2868 if (tp->t_flags & TF_SIGNATURE) { 2869 2870 sav = tcp_signature_getsav(m, th); 2871 2872 if (sav == NULL && tp->t_state == TCPS_LISTEN) 2873 return (-1); 2874 } 2875 2876 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 2877 if (sav == NULL) 2878 return (-1); 2879 #ifdef FAST_IPSEC 2880 KEY_FREESAV(&sav); 2881 #else 2882 key_freesav(sav); 2883 #endif 2884 return (-1); 2885 } 2886 2887 if (sigp) { 2888 char sig[TCP_SIGLEN]; 2889 2890 TCP_FIELDS_TO_NET(th); 2891 if (tcp_signature(m, th, toff, sav, sig) < 0) { 2892 TCP_FIELDS_TO_HOST(th); 2893 if (sav == NULL) 2894 return (-1); 2895 #ifdef FAST_IPSEC 2896 KEY_FREESAV(&sav); 2897 #else 2898 key_freesav(sav); 2899 #endif 2900 return (-1); 2901 } 2902 TCP_FIELDS_TO_HOST(th); 2903 2904 if (bcmp(sig, sigp, TCP_SIGLEN)) { 2905 tcpstat.tcps_badsig++; 2906 if (sav == NULL) 2907 return (-1); 2908 #ifdef FAST_IPSEC 2909 KEY_FREESAV(&sav); 2910 #else 2911 key_freesav(sav); 2912 #endif 2913 return (-1); 2914 } else 2915 tcpstat.tcps_goodsig++; 2916 2917 key_sa_recordxfer(sav, m); 2918 #ifdef FAST_IPSEC 2919 KEY_FREESAV(&sav); 2920 #else 2921 key_freesav(sav); 2922 #endif 2923 } 2924 #endif 2925 2926 return (0); 2927 } 2928 2929 /* 2930 * Pull out of band byte out of a segment so 2931 * it doesn't appear in the user's data queue. 2932 * It is still reflected in the segment length for 2933 * sequencing purposes. 2934 */ 2935 void 2936 tcp_pulloutofband(so, th, m, off) 2937 struct socket *so; 2938 struct tcphdr *th; 2939 struct mbuf *m; 2940 int off; 2941 { 2942 int cnt = off + th->th_urp - 1; 2943 2944 while (cnt >= 0) { 2945 if (m->m_len > cnt) { 2946 char *cp = mtod(m, caddr_t) + cnt; 2947 struct tcpcb *tp = sototcpcb(so); 2948 2949 tp->t_iobc = *cp; 2950 tp->t_oobflags |= TCPOOB_HAVEDATA; 2951 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2952 m->m_len--; 2953 return; 2954 } 2955 cnt -= m->m_len; 2956 m = m->m_next; 2957 if (m == 0) 2958 break; 2959 } 2960 panic("tcp_pulloutofband"); 2961 } 2962 2963 /* 2964 * Collect new round-trip time estimate 2965 * and update averages and current timeout. 2966 */ 2967 void 2968 tcp_xmit_timer(tp, rtt) 2969 struct tcpcb *tp; 2970 uint32_t rtt; 2971 { 2972 int32_t delta; 2973 2974 tcpstat.tcps_rttupdated++; 2975 if (tp->t_srtt != 0) { 2976 /* 2977 * srtt is stored as fixed point with 3 bits after the 2978 * binary point (i.e., scaled by 8). The following magic 2979 * is equivalent to the smoothing algorithm in rfc793 with 2980 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2981 * point). Adjust rtt to origin 0. 2982 */ 2983 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2984 if ((tp->t_srtt += delta) <= 0) 2985 tp->t_srtt = 1 << 2; 2986 /* 2987 * We accumulate a smoothed rtt variance (actually, a 2988 * smoothed mean difference), then set the retransmit 2989 * timer to smoothed rtt + 4 times the smoothed variance. 2990 * rttvar is stored as fixed point with 2 bits after the 2991 * binary point (scaled by 4). The following is 2992 * equivalent to rfc793 smoothing with an alpha of .75 2993 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2994 * rfc793's wired-in beta. 2995 */ 2996 if (delta < 0) 2997 delta = -delta; 2998 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2999 if ((tp->t_rttvar += delta) <= 0) 3000 tp->t_rttvar = 1 << 2; 3001 } else { 3002 /* 3003 * No rtt measurement yet - use the unsmoothed rtt. 3004 * Set the variance to half the rtt (so our first 3005 * retransmit happens at 3*rtt). 3006 */ 3007 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3008 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3009 } 3010 tp->t_rtttime = 0; 3011 tp->t_rxtshift = 0; 3012 3013 /* 3014 * the retransmit should happen at rtt + 4 * rttvar. 3015 * Because of the way we do the smoothing, srtt and rttvar 3016 * will each average +1/2 tick of bias. When we compute 3017 * the retransmit timer, we want 1/2 tick of rounding and 3018 * 1 extra tick because of +-1/2 tick uncertainty in the 3019 * firing of the timer. The bias will give us exactly the 3020 * 1.5 tick we need. But, because the bias is 3021 * statistical, we have to test that we don't drop below 3022 * the minimum feasible timer (which is 2 ticks). 3023 */ 3024 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3025 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3026 3027 /* 3028 * We received an ack for a packet that wasn't retransmitted; 3029 * it is probably safe to discard any error indications we've 3030 * received recently. This isn't quite right, but close enough 3031 * for now (a route might have failed after we sent a segment, 3032 * and the return path might not be symmetrical). 3033 */ 3034 tp->t_softerror = 0; 3035 } 3036 3037 void 3038 tcp_reno_newack(tp, th) 3039 struct tcpcb *tp; 3040 struct tcphdr *th; 3041 { 3042 if (tp->t_partialacks < 0) { 3043 /* 3044 * We were not in fast recovery. Reset the duplicate ack 3045 * counter. 3046 */ 3047 tp->t_dupacks = 0; 3048 } else { 3049 /* 3050 * Clamp the congestion window to the crossover point and 3051 * exit fast recovery. 3052 */ 3053 if (tp->snd_cwnd > tp->snd_ssthresh) 3054 tp->snd_cwnd = tp->snd_ssthresh; 3055 tp->t_partialacks = -1; 3056 tp->t_dupacks = 0; 3057 } 3058 } 3059 3060 /* 3061 * Implement the NewReno response to a new ack, checking for partial acks in 3062 * fast recovery. 3063 */ 3064 void 3065 tcp_newreno_newack(tp, th) 3066 struct tcpcb *tp; 3067 struct tcphdr *th; 3068 { 3069 if (tp->t_partialacks < 0) { 3070 /* 3071 * We were not in fast recovery. Reset the duplicate ack 3072 * counter. 3073 */ 3074 tp->t_dupacks = 0; 3075 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) { 3076 /* 3077 * This is a partial ack. Retransmit the first unacknowledged 3078 * segment and deflate the congestion window by the amount of 3079 * acknowledged data. Do not exit fast recovery. 3080 */ 3081 tcp_seq onxt = tp->snd_nxt; 3082 u_long ocwnd = tp->snd_cwnd; 3083 3084 /* 3085 * snd_una has not yet been updated and the socket's send 3086 * buffer has not yet drained off the ACK'd data, so we 3087 * have to leave snd_una as it was to get the correct data 3088 * offset in tcp_output(). 3089 */ 3090 if (++tp->t_partialacks == 1) 3091 TCP_TIMER_DISARM(tp, TCPT_REXMT); 3092 tp->t_rtttime = 0; 3093 tp->snd_nxt = th->th_ack; 3094 /* 3095 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 3096 * is not yet updated when we're called. 3097 */ 3098 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 3099 (void) tcp_output(tp); 3100 tp->snd_cwnd = ocwnd; 3101 if (SEQ_GT(onxt, tp->snd_nxt)) 3102 tp->snd_nxt = onxt; 3103 /* 3104 * Partial window deflation. Relies on fact that tp->snd_una 3105 * not updated yet. 3106 */ 3107 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 3108 } else { 3109 /* 3110 * Complete ack. Inflate the congestion window to ssthresh 3111 * and exit fast recovery. 3112 * 3113 * Window inflation should have left us with approx. 3114 * snd_ssthresh outstanding data. But in case we 3115 * would be inclined to send a burst, better to do 3116 * it via the slow start mechanism. 3117 */ 3118 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 3119 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 3120 + tp->t_segsz; 3121 else 3122 tp->snd_cwnd = tp->snd_ssthresh; 3123 tp->t_partialacks = -1; 3124 tp->t_dupacks = 0; 3125 } 3126 } 3127 3128 3129 /* 3130 * TCP compressed state engine. Currently used to hold compressed 3131 * state for SYN_RECEIVED. 3132 */ 3133 3134 u_long syn_cache_count; 3135 u_int32_t syn_hash1, syn_hash2; 3136 3137 #define SYN_HASH(sa, sp, dp) \ 3138 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3139 ((u_int32_t)(sp)))^syn_hash2))) 3140 #ifndef INET6 3141 #define SYN_HASHALL(hash, src, dst) \ 3142 do { \ 3143 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 3144 ((struct sockaddr_in *)(src))->sin_port, \ 3145 ((struct sockaddr_in *)(dst))->sin_port); \ 3146 } while (/*CONSTCOND*/ 0) 3147 #else 3148 #define SYN_HASH6(sa, sp, dp) \ 3149 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3150 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3151 & 0x7fffffff) 3152 3153 #define SYN_HASHALL(hash, src, dst) \ 3154 do { \ 3155 switch ((src)->sa_family) { \ 3156 case AF_INET: \ 3157 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 3158 ((struct sockaddr_in *)(src))->sin_port, \ 3159 ((struct sockaddr_in *)(dst))->sin_port); \ 3160 break; \ 3161 case AF_INET6: \ 3162 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 3163 ((struct sockaddr_in6 *)(src))->sin6_port, \ 3164 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 3165 break; \ 3166 default: \ 3167 hash = 0; \ 3168 } \ 3169 } while (/*CONSTCOND*/0) 3170 #endif /* INET6 */ 3171 3172 #define SYN_CACHE_RM(sc) \ 3173 do { \ 3174 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 3175 (sc), sc_bucketq); \ 3176 (sc)->sc_tp = NULL; \ 3177 LIST_REMOVE((sc), sc_tpq); \ 3178 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 3179 callout_stop(&(sc)->sc_timer); \ 3180 syn_cache_count--; \ 3181 } while (/*CONSTCOND*/0) 3182 3183 #define SYN_CACHE_PUT(sc) \ 3184 do { \ 3185 if ((sc)->sc_ipopts) \ 3186 (void) m_free((sc)->sc_ipopts); \ 3187 if ((sc)->sc_route4.ro_rt != NULL) \ 3188 RTFREE((sc)->sc_route4.ro_rt); \ 3189 if (callout_invoking(&(sc)->sc_timer)) \ 3190 (sc)->sc_flags |= SCF_DEAD; \ 3191 else \ 3192 pool_put(&syn_cache_pool, (sc)); \ 3193 } while (/*CONSTCOND*/0) 3194 3195 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL); 3196 3197 /* 3198 * We don't estimate RTT with SYNs, so each packet starts with the default 3199 * RTT and each timer step has a fixed timeout value. 3200 */ 3201 #define SYN_CACHE_TIMER_ARM(sc) \ 3202 do { \ 3203 TCPT_RANGESET((sc)->sc_rxtcur, \ 3204 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3205 TCPTV_REXMTMAX); \ 3206 callout_reset(&(sc)->sc_timer, \ 3207 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3208 } while (/*CONSTCOND*/0) 3209 3210 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3211 3212 void 3213 syn_cache_init() 3214 { 3215 int i; 3216 3217 /* Initialize the hash buckets. */ 3218 for (i = 0; i < tcp_syn_cache_size; i++) 3219 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3220 } 3221 3222 void 3223 syn_cache_insert(sc, tp) 3224 struct syn_cache *sc; 3225 struct tcpcb *tp; 3226 { 3227 struct syn_cache_head *scp; 3228 struct syn_cache *sc2; 3229 int s; 3230 3231 /* 3232 * If there are no entries in the hash table, reinitialize 3233 * the hash secrets. 3234 */ 3235 if (syn_cache_count == 0) { 3236 syn_hash1 = arc4random(); 3237 syn_hash2 = arc4random(); 3238 } 3239 3240 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3241 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3242 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3243 3244 /* 3245 * Make sure that we don't overflow the per-bucket 3246 * limit or the total cache size limit. 3247 */ 3248 s = splsoftnet(); 3249 if (scp->sch_length >= tcp_syn_bucket_limit) { 3250 tcpstat.tcps_sc_bucketoverflow++; 3251 /* 3252 * The bucket is full. Toss the oldest element in the 3253 * bucket. This will be the first entry in the bucket. 3254 */ 3255 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3256 #ifdef DIAGNOSTIC 3257 /* 3258 * This should never happen; we should always find an 3259 * entry in our bucket. 3260 */ 3261 if (sc2 == NULL) 3262 panic("syn_cache_insert: bucketoverflow: impossible"); 3263 #endif 3264 SYN_CACHE_RM(sc2); 3265 SYN_CACHE_PUT(sc2); 3266 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3267 struct syn_cache_head *scp2, *sce; 3268 3269 tcpstat.tcps_sc_overflowed++; 3270 /* 3271 * The cache is full. Toss the oldest entry in the 3272 * first non-empty bucket we can find. 3273 * 3274 * XXX We would really like to toss the oldest 3275 * entry in the cache, but we hope that this 3276 * condition doesn't happen very often. 3277 */ 3278 scp2 = scp; 3279 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3280 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3281 for (++scp2; scp2 != scp; scp2++) { 3282 if (scp2 >= sce) 3283 scp2 = &tcp_syn_cache[0]; 3284 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3285 break; 3286 } 3287 #ifdef DIAGNOSTIC 3288 /* 3289 * This should never happen; we should always find a 3290 * non-empty bucket. 3291 */ 3292 if (scp2 == scp) 3293 panic("syn_cache_insert: cacheoverflow: " 3294 "impossible"); 3295 #endif 3296 } 3297 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3298 SYN_CACHE_RM(sc2); 3299 SYN_CACHE_PUT(sc2); 3300 } 3301 3302 /* 3303 * Initialize the entry's timer. 3304 */ 3305 sc->sc_rxttot = 0; 3306 sc->sc_rxtshift = 0; 3307 SYN_CACHE_TIMER_ARM(sc); 3308 3309 /* Link it from tcpcb entry */ 3310 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3311 3312 /* Put it into the bucket. */ 3313 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3314 scp->sch_length++; 3315 syn_cache_count++; 3316 3317 tcpstat.tcps_sc_added++; 3318 splx(s); 3319 } 3320 3321 /* 3322 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3323 * If we have retransmitted an entry the maximum number of times, expire 3324 * that entry. 3325 */ 3326 void 3327 syn_cache_timer(void *arg) 3328 { 3329 struct syn_cache *sc = arg; 3330 int s; 3331 3332 s = splsoftnet(); 3333 callout_ack(&sc->sc_timer); 3334 3335 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3336 tcpstat.tcps_sc_delayed_free++; 3337 pool_put(&syn_cache_pool, sc); 3338 splx(s); 3339 return; 3340 } 3341 3342 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3343 /* Drop it -- too many retransmissions. */ 3344 goto dropit; 3345 } 3346 3347 /* 3348 * Compute the total amount of time this entry has 3349 * been on a queue. If this entry has been on longer 3350 * than the keep alive timer would allow, expire it. 3351 */ 3352 sc->sc_rxttot += sc->sc_rxtcur; 3353 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) 3354 goto dropit; 3355 3356 tcpstat.tcps_sc_retransmitted++; 3357 (void) syn_cache_respond(sc, NULL); 3358 3359 /* Advance the timer back-off. */ 3360 sc->sc_rxtshift++; 3361 SYN_CACHE_TIMER_ARM(sc); 3362 3363 splx(s); 3364 return; 3365 3366 dropit: 3367 tcpstat.tcps_sc_timed_out++; 3368 SYN_CACHE_RM(sc); 3369 SYN_CACHE_PUT(sc); 3370 splx(s); 3371 } 3372 3373 /* 3374 * Remove syn cache created by the specified tcb entry, 3375 * because this does not make sense to keep them 3376 * (if there's no tcb entry, syn cache entry will never be used) 3377 */ 3378 void 3379 syn_cache_cleanup(tp) 3380 struct tcpcb *tp; 3381 { 3382 struct syn_cache *sc, *nsc; 3383 int s; 3384 3385 s = splsoftnet(); 3386 3387 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3388 nsc = LIST_NEXT(sc, sc_tpq); 3389 3390 #ifdef DIAGNOSTIC 3391 if (sc->sc_tp != tp) 3392 panic("invalid sc_tp in syn_cache_cleanup"); 3393 #endif 3394 SYN_CACHE_RM(sc); 3395 SYN_CACHE_PUT(sc); 3396 } 3397 /* just for safety */ 3398 LIST_INIT(&tp->t_sc); 3399 3400 splx(s); 3401 } 3402 3403 /* 3404 * Find an entry in the syn cache. 3405 */ 3406 struct syn_cache * 3407 syn_cache_lookup(src, dst, headp) 3408 struct sockaddr *src; 3409 struct sockaddr *dst; 3410 struct syn_cache_head **headp; 3411 { 3412 struct syn_cache *sc; 3413 struct syn_cache_head *scp; 3414 u_int32_t hash; 3415 int s; 3416 3417 SYN_HASHALL(hash, src, dst); 3418 3419 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3420 *headp = scp; 3421 s = splsoftnet(); 3422 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3423 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3424 if (sc->sc_hash != hash) 3425 continue; 3426 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3427 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3428 splx(s); 3429 return (sc); 3430 } 3431 } 3432 splx(s); 3433 return (NULL); 3434 } 3435 3436 /* 3437 * This function gets called when we receive an ACK for a 3438 * socket in the LISTEN state. We look up the connection 3439 * in the syn cache, and if its there, we pull it out of 3440 * the cache and turn it into a full-blown connection in 3441 * the SYN-RECEIVED state. 3442 * 3443 * The return values may not be immediately obvious, and their effects 3444 * can be subtle, so here they are: 3445 * 3446 * NULL SYN was not found in cache; caller should drop the 3447 * packet and send an RST. 3448 * 3449 * -1 We were unable to create the new connection, and are 3450 * aborting it. An ACK,RST is being sent to the peer 3451 * (unless we got screwey sequence numbners; see below), 3452 * because the 3-way handshake has been completed. Caller 3453 * should not free the mbuf, since we may be using it. If 3454 * we are not, we will free it. 3455 * 3456 * Otherwise, the return value is a pointer to the new socket 3457 * associated with the connection. 3458 */ 3459 struct socket * 3460 syn_cache_get(src, dst, th, hlen, tlen, so, m) 3461 struct sockaddr *src; 3462 struct sockaddr *dst; 3463 struct tcphdr *th; 3464 unsigned int hlen, tlen; 3465 struct socket *so; 3466 struct mbuf *m; 3467 { 3468 struct syn_cache *sc; 3469 struct syn_cache_head *scp; 3470 struct inpcb *inp = NULL; 3471 #ifdef INET6 3472 struct in6pcb *in6p = NULL; 3473 #endif 3474 struct tcpcb *tp = 0; 3475 struct mbuf *am; 3476 int s; 3477 struct socket *oso; 3478 3479 s = splsoftnet(); 3480 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3481 splx(s); 3482 return (NULL); 3483 } 3484 3485 /* 3486 * Verify the sequence and ack numbers. Try getting the correct 3487 * response again. 3488 */ 3489 if ((th->th_ack != sc->sc_iss + 1) || 3490 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3491 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3492 (void) syn_cache_respond(sc, m); 3493 splx(s); 3494 return ((struct socket *)(-1)); 3495 } 3496 3497 /* Remove this cache entry */ 3498 SYN_CACHE_RM(sc); 3499 splx(s); 3500 3501 /* 3502 * Ok, create the full blown connection, and set things up 3503 * as they would have been set up if we had created the 3504 * connection when the SYN arrived. If we can't create 3505 * the connection, abort it. 3506 */ 3507 /* 3508 * inp still has the OLD in_pcb stuff, set the 3509 * v6-related flags on the new guy, too. This is 3510 * done particularly for the case where an AF_INET6 3511 * socket is bound only to a port, and a v4 connection 3512 * comes in on that port. 3513 * we also copy the flowinfo from the original pcb 3514 * to the new one. 3515 */ 3516 oso = so; 3517 so = sonewconn(so, SS_ISCONNECTED); 3518 if (so == NULL) 3519 goto resetandabort; 3520 3521 switch (so->so_proto->pr_domain->dom_family) { 3522 #ifdef INET 3523 case AF_INET: 3524 inp = sotoinpcb(so); 3525 break; 3526 #endif 3527 #ifdef INET6 3528 case AF_INET6: 3529 in6p = sotoin6pcb(so); 3530 break; 3531 #endif 3532 } 3533 switch (src->sa_family) { 3534 #ifdef INET 3535 case AF_INET: 3536 if (inp) { 3537 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3538 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3539 inp->inp_options = ip_srcroute(); 3540 in_pcbstate(inp, INP_BOUND); 3541 if (inp->inp_options == NULL) { 3542 inp->inp_options = sc->sc_ipopts; 3543 sc->sc_ipopts = NULL; 3544 } 3545 } 3546 #ifdef INET6 3547 else if (in6p) { 3548 /* IPv4 packet to AF_INET6 socket */ 3549 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3550 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3551 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3552 &in6p->in6p_laddr.s6_addr32[3], 3553 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3554 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3555 in6totcpcb(in6p)->t_family = AF_INET; 3556 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3557 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3558 else 3559 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3560 in6_pcbstate(in6p, IN6P_BOUND); 3561 } 3562 #endif 3563 break; 3564 #endif 3565 #ifdef INET6 3566 case AF_INET6: 3567 if (in6p) { 3568 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3569 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3570 in6_pcbstate(in6p, IN6P_BOUND); 3571 } 3572 break; 3573 #endif 3574 } 3575 #ifdef INET6 3576 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3577 struct in6pcb *oin6p = sotoin6pcb(oso); 3578 /* inherit socket options from the listening socket */ 3579 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3580 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3581 m_freem(in6p->in6p_options); 3582 in6p->in6p_options = 0; 3583 } 3584 ip6_savecontrol(in6p, &in6p->in6p_options, 3585 mtod(m, struct ip6_hdr *), m); 3586 } 3587 #endif 3588 3589 #if defined(IPSEC) || defined(FAST_IPSEC) 3590 /* 3591 * we make a copy of policy, instead of sharing the policy, 3592 * for better behavior in terms of SA lookup and dead SA removal. 3593 */ 3594 if (inp) { 3595 /* copy old policy into new socket's */ 3596 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3597 printf("tcp_input: could not copy policy\n"); 3598 } 3599 #ifdef INET6 3600 else if (in6p) { 3601 /* copy old policy into new socket's */ 3602 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3603 in6p->in6p_sp)) 3604 printf("tcp_input: could not copy policy\n"); 3605 } 3606 #endif 3607 #endif 3608 3609 /* 3610 * Give the new socket our cached route reference. 3611 */ 3612 if (inp) 3613 inp->inp_route = sc->sc_route4; /* struct assignment */ 3614 #ifdef INET6 3615 else 3616 in6p->in6p_route = sc->sc_route6; 3617 #endif 3618 sc->sc_route4.ro_rt = NULL; 3619 3620 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3621 if (am == NULL) 3622 goto resetandabort; 3623 MCLAIM(am, &tcp_mowner); 3624 am->m_len = src->sa_len; 3625 bcopy(src, mtod(am, caddr_t), src->sa_len); 3626 if (inp) { 3627 if (in_pcbconnect(inp, am)) { 3628 (void) m_free(am); 3629 goto resetandabort; 3630 } 3631 } 3632 #ifdef INET6 3633 else if (in6p) { 3634 if (src->sa_family == AF_INET) { 3635 /* IPv4 packet to AF_INET6 socket */ 3636 struct sockaddr_in6 *sin6; 3637 sin6 = mtod(am, struct sockaddr_in6 *); 3638 am->m_len = sizeof(*sin6); 3639 bzero(sin6, sizeof(*sin6)); 3640 sin6->sin6_family = AF_INET6; 3641 sin6->sin6_len = sizeof(*sin6); 3642 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3643 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3644 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3645 &sin6->sin6_addr.s6_addr32[3], 3646 sizeof(sin6->sin6_addr.s6_addr32[3])); 3647 } 3648 if (in6_pcbconnect(in6p, am)) { 3649 (void) m_free(am); 3650 goto resetandabort; 3651 } 3652 } 3653 #endif 3654 else { 3655 (void) m_free(am); 3656 goto resetandabort; 3657 } 3658 (void) m_free(am); 3659 3660 if (inp) 3661 tp = intotcpcb(inp); 3662 #ifdef INET6 3663 else if (in6p) 3664 tp = in6totcpcb(in6p); 3665 #endif 3666 else 3667 tp = NULL; 3668 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3669 if (sc->sc_request_r_scale != 15) { 3670 tp->requested_s_scale = sc->sc_requested_s_scale; 3671 tp->request_r_scale = sc->sc_request_r_scale; 3672 tp->snd_scale = sc->sc_requested_s_scale; 3673 tp->rcv_scale = sc->sc_request_r_scale; 3674 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3675 } 3676 if (sc->sc_flags & SCF_TIMESTAMP) 3677 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3678 tp->ts_timebase = sc->sc_timebase; 3679 3680 tp->t_template = tcp_template(tp); 3681 if (tp->t_template == 0) { 3682 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3683 so = NULL; 3684 m_freem(m); 3685 goto abort; 3686 } 3687 3688 tp->iss = sc->sc_iss; 3689 tp->irs = sc->sc_irs; 3690 tcp_sendseqinit(tp); 3691 tcp_rcvseqinit(tp); 3692 tp->t_state = TCPS_SYN_RECEIVED; 3693 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3694 tcpstat.tcps_accepts++; 3695 3696 #ifdef TCP_SIGNATURE 3697 if (sc->sc_flags & SCF_SIGNATURE) 3698 tp->t_flags |= TF_SIGNATURE; 3699 #endif 3700 3701 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3702 tp->t_ourmss = sc->sc_ourmaxseg; 3703 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3704 3705 /* 3706 * Initialize the initial congestion window. If we 3707 * had to retransmit the SYN,ACK, we must initialize cwnd 3708 * to 1 segment (i.e. the Loss Window). 3709 */ 3710 if (sc->sc_rxtshift) 3711 tp->snd_cwnd = tp->t_peermss; 3712 else { 3713 int ss = tcp_init_win; 3714 #ifdef INET 3715 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3716 ss = tcp_init_win_local; 3717 #endif 3718 #ifdef INET6 3719 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3720 ss = tcp_init_win_local; 3721 #endif 3722 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3723 } 3724 3725 tcp_rmx_rtt(tp); 3726 tp->snd_wl1 = sc->sc_irs; 3727 tp->rcv_up = sc->sc_irs + 1; 3728 3729 /* 3730 * This is what whould have happened in tcp_output() when 3731 * the SYN,ACK was sent. 3732 */ 3733 tp->snd_up = tp->snd_una; 3734 tp->snd_max = tp->snd_nxt = tp->iss+1; 3735 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3736 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3737 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3738 tp->last_ack_sent = tp->rcv_nxt; 3739 tp->t_partialacks = -1; 3740 tp->t_dupacks = 0; 3741 3742 tcpstat.tcps_sc_completed++; 3743 SYN_CACHE_PUT(sc); 3744 return (so); 3745 3746 resetandabort: 3747 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3748 abort: 3749 if (so != NULL) 3750 (void) soabort(so); 3751 SYN_CACHE_PUT(sc); 3752 tcpstat.tcps_sc_aborted++; 3753 return ((struct socket *)(-1)); 3754 } 3755 3756 /* 3757 * This function is called when we get a RST for a 3758 * non-existent connection, so that we can see if the 3759 * connection is in the syn cache. If it is, zap it. 3760 */ 3761 3762 void 3763 syn_cache_reset(src, dst, th) 3764 struct sockaddr *src; 3765 struct sockaddr *dst; 3766 struct tcphdr *th; 3767 { 3768 struct syn_cache *sc; 3769 struct syn_cache_head *scp; 3770 int s = splsoftnet(); 3771 3772 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3773 splx(s); 3774 return; 3775 } 3776 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3777 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3778 splx(s); 3779 return; 3780 } 3781 SYN_CACHE_RM(sc); 3782 splx(s); 3783 tcpstat.tcps_sc_reset++; 3784 SYN_CACHE_PUT(sc); 3785 } 3786 3787 void 3788 syn_cache_unreach(src, dst, th) 3789 struct sockaddr *src; 3790 struct sockaddr *dst; 3791 struct tcphdr *th; 3792 { 3793 struct syn_cache *sc; 3794 struct syn_cache_head *scp; 3795 int s; 3796 3797 s = splsoftnet(); 3798 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3799 splx(s); 3800 return; 3801 } 3802 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3803 if (ntohl (th->th_seq) != sc->sc_iss) { 3804 splx(s); 3805 return; 3806 } 3807 3808 /* 3809 * If we've retransmitted 3 times and this is our second error, 3810 * we remove the entry. Otherwise, we allow it to continue on. 3811 * This prevents us from incorrectly nuking an entry during a 3812 * spurious network outage. 3813 * 3814 * See tcp_notify(). 3815 */ 3816 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3817 sc->sc_flags |= SCF_UNREACH; 3818 splx(s); 3819 return; 3820 } 3821 3822 SYN_CACHE_RM(sc); 3823 splx(s); 3824 tcpstat.tcps_sc_unreach++; 3825 SYN_CACHE_PUT(sc); 3826 } 3827 3828 /* 3829 * Given a LISTEN socket and an inbound SYN request, add 3830 * this to the syn cache, and send back a segment: 3831 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3832 * to the source. 3833 * 3834 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3835 * Doing so would require that we hold onto the data and deliver it 3836 * to the application. However, if we are the target of a SYN-flood 3837 * DoS attack, an attacker could send data which would eventually 3838 * consume all available buffer space if it were ACKed. By not ACKing 3839 * the data, we avoid this DoS scenario. 3840 */ 3841 3842 int 3843 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi) 3844 struct sockaddr *src; 3845 struct sockaddr *dst; 3846 struct tcphdr *th; 3847 unsigned int hlen; 3848 struct socket *so; 3849 struct mbuf *m; 3850 u_char *optp; 3851 int optlen; 3852 struct tcp_opt_info *oi; 3853 { 3854 struct tcpcb tb, *tp; 3855 long win; 3856 struct syn_cache *sc; 3857 struct syn_cache_head *scp; 3858 struct mbuf *ipopts; 3859 struct tcp_opt_info opti; 3860 3861 tp = sototcpcb(so); 3862 3863 bzero(&opti, sizeof(opti)); 3864 3865 /* 3866 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3867 * 3868 * Note this check is performed in tcp_input() very early on. 3869 */ 3870 3871 /* 3872 * Initialize some local state. 3873 */ 3874 win = sbspace(&so->so_rcv); 3875 if (win > TCP_MAXWIN) 3876 win = TCP_MAXWIN; 3877 3878 switch (src->sa_family) { 3879 #ifdef INET 3880 case AF_INET: 3881 /* 3882 * Remember the IP options, if any. 3883 */ 3884 ipopts = ip_srcroute(); 3885 break; 3886 #endif 3887 default: 3888 ipopts = NULL; 3889 } 3890 3891 #ifdef TCP_SIGNATURE 3892 if (optp || (tp->t_flags & TF_SIGNATURE)) 3893 #else 3894 if (optp) 3895 #endif 3896 { 3897 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3898 #ifdef TCP_SIGNATURE 3899 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 3900 #endif 3901 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 3902 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 3903 return (0); 3904 } else 3905 tb.t_flags = 0; 3906 3907 /* 3908 * See if we already have an entry for this connection. 3909 * If we do, resend the SYN,ACK. We do not count this 3910 * as a retransmission (XXX though maybe we should). 3911 */ 3912 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3913 tcpstat.tcps_sc_dupesyn++; 3914 if (ipopts) { 3915 /* 3916 * If we were remembering a previous source route, 3917 * forget it and use the new one we've been given. 3918 */ 3919 if (sc->sc_ipopts) 3920 (void) m_free(sc->sc_ipopts); 3921 sc->sc_ipopts = ipopts; 3922 } 3923 sc->sc_timestamp = tb.ts_recent; 3924 if (syn_cache_respond(sc, m) == 0) { 3925 tcpstat.tcps_sndacks++; 3926 tcpstat.tcps_sndtotal++; 3927 } 3928 return (1); 3929 } 3930 3931 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3932 if (sc == NULL) { 3933 if (ipopts) 3934 (void) m_free(ipopts); 3935 return (0); 3936 } 3937 3938 /* 3939 * Fill in the cache, and put the necessary IP and TCP 3940 * options into the reply. 3941 */ 3942 bzero(sc, sizeof(struct syn_cache)); 3943 callout_init(&sc->sc_timer); 3944 bcopy(src, &sc->sc_src, src->sa_len); 3945 bcopy(dst, &sc->sc_dst, dst->sa_len); 3946 sc->sc_flags = 0; 3947 sc->sc_ipopts = ipopts; 3948 sc->sc_irs = th->th_seq; 3949 switch (src->sa_family) { 3950 #ifdef INET 3951 case AF_INET: 3952 { 3953 struct sockaddr_in *srcin = (void *) src; 3954 struct sockaddr_in *dstin = (void *) dst; 3955 3956 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 3957 &srcin->sin_addr, dstin->sin_port, 3958 srcin->sin_port, sizeof(dstin->sin_addr), 0); 3959 break; 3960 } 3961 #endif /* INET */ 3962 #ifdef INET6 3963 case AF_INET6: 3964 { 3965 struct sockaddr_in6 *srcin6 = (void *) src; 3966 struct sockaddr_in6 *dstin6 = (void *) dst; 3967 3968 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 3969 &srcin6->sin6_addr, dstin6->sin6_port, 3970 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 3971 break; 3972 } 3973 #endif /* INET6 */ 3974 } 3975 sc->sc_peermaxseg = oi->maxseg; 3976 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3977 m->m_pkthdr.rcvif : NULL, 3978 sc->sc_src.sa.sa_family); 3979 sc->sc_win = win; 3980 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 3981 sc->sc_timestamp = tb.ts_recent; 3982 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 3983 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 3984 sc->sc_flags |= SCF_TIMESTAMP; 3985 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3986 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3987 sc->sc_requested_s_scale = tb.requested_s_scale; 3988 sc->sc_request_r_scale = 0; 3989 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3990 TCP_MAXWIN << sc->sc_request_r_scale < 3991 so->so_rcv.sb_hiwat) 3992 sc->sc_request_r_scale++; 3993 } else { 3994 sc->sc_requested_s_scale = 15; 3995 sc->sc_request_r_scale = 15; 3996 } 3997 #ifdef TCP_SIGNATURE 3998 if (tb.t_flags & TF_SIGNATURE) 3999 sc->sc_flags |= SCF_SIGNATURE; 4000 #endif 4001 sc->sc_tp = tp; 4002 if (syn_cache_respond(sc, m) == 0) { 4003 syn_cache_insert(sc, tp); 4004 tcpstat.tcps_sndacks++; 4005 tcpstat.tcps_sndtotal++; 4006 } else { 4007 SYN_CACHE_PUT(sc); 4008 tcpstat.tcps_sc_dropped++; 4009 } 4010 return (1); 4011 } 4012 4013 int 4014 syn_cache_respond(sc, m) 4015 struct syn_cache *sc; 4016 struct mbuf *m; 4017 { 4018 struct route *ro; 4019 u_int8_t *optp; 4020 int optlen, error; 4021 u_int16_t tlen; 4022 struct ip *ip = NULL; 4023 #ifdef INET6 4024 struct ip6_hdr *ip6 = NULL; 4025 #endif 4026 struct tcpcb *tp; 4027 struct tcphdr *th; 4028 u_int hlen; 4029 struct socket *so; 4030 4031 switch (sc->sc_src.sa.sa_family) { 4032 case AF_INET: 4033 hlen = sizeof(struct ip); 4034 ro = &sc->sc_route4; 4035 break; 4036 #ifdef INET6 4037 case AF_INET6: 4038 hlen = sizeof(struct ip6_hdr); 4039 ro = (struct route *)&sc->sc_route6; 4040 break; 4041 #endif 4042 default: 4043 if (m) 4044 m_freem(m); 4045 return (EAFNOSUPPORT); 4046 } 4047 4048 /* Compute the size of the TCP options. */ 4049 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4050 #ifdef TCP_SIGNATURE 4051 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4052 #endif 4053 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4054 4055 tlen = hlen + sizeof(struct tcphdr) + optlen; 4056 4057 /* 4058 * Create the IP+TCP header from scratch. 4059 */ 4060 if (m) 4061 m_freem(m); 4062 #ifdef DIAGNOSTIC 4063 if (max_linkhdr + tlen > MCLBYTES) 4064 return (ENOBUFS); 4065 #endif 4066 MGETHDR(m, M_DONTWAIT, MT_DATA); 4067 if (m && tlen > MHLEN) { 4068 MCLGET(m, M_DONTWAIT); 4069 if ((m->m_flags & M_EXT) == 0) { 4070 m_freem(m); 4071 m = NULL; 4072 } 4073 } 4074 if (m == NULL) 4075 return (ENOBUFS); 4076 MCLAIM(m, &tcp_tx_mowner); 4077 4078 /* Fixup the mbuf. */ 4079 m->m_data += max_linkhdr; 4080 m->m_len = m->m_pkthdr.len = tlen; 4081 if (sc->sc_tp) { 4082 tp = sc->sc_tp; 4083 if (tp->t_inpcb) 4084 so = tp->t_inpcb->inp_socket; 4085 #ifdef INET6 4086 else if (tp->t_in6pcb) 4087 so = tp->t_in6pcb->in6p_socket; 4088 #endif 4089 else 4090 so = NULL; 4091 } else 4092 so = NULL; 4093 m->m_pkthdr.rcvif = NULL; 4094 memset(mtod(m, u_char *), 0, tlen); 4095 4096 switch (sc->sc_src.sa.sa_family) { 4097 case AF_INET: 4098 ip = mtod(m, struct ip *); 4099 ip->ip_v = 4; 4100 ip->ip_dst = sc->sc_src.sin.sin_addr; 4101 ip->ip_src = sc->sc_dst.sin.sin_addr; 4102 ip->ip_p = IPPROTO_TCP; 4103 th = (struct tcphdr *)(ip + 1); 4104 th->th_dport = sc->sc_src.sin.sin_port; 4105 th->th_sport = sc->sc_dst.sin.sin_port; 4106 break; 4107 #ifdef INET6 4108 case AF_INET6: 4109 ip6 = mtod(m, struct ip6_hdr *); 4110 ip6->ip6_vfc = IPV6_VERSION; 4111 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4112 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4113 ip6->ip6_nxt = IPPROTO_TCP; 4114 /* ip6_plen will be updated in ip6_output() */ 4115 th = (struct tcphdr *)(ip6 + 1); 4116 th->th_dport = sc->sc_src.sin6.sin6_port; 4117 th->th_sport = sc->sc_dst.sin6.sin6_port; 4118 break; 4119 #endif 4120 default: 4121 th = NULL; 4122 } 4123 4124 th->th_seq = htonl(sc->sc_iss); 4125 th->th_ack = htonl(sc->sc_irs + 1); 4126 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4127 th->th_flags = TH_SYN|TH_ACK; 4128 th->th_win = htons(sc->sc_win); 4129 /* th_sum already 0 */ 4130 /* th_urp already 0 */ 4131 4132 /* Tack on the TCP options. */ 4133 optp = (u_int8_t *)(th + 1); 4134 *optp++ = TCPOPT_MAXSEG; 4135 *optp++ = 4; 4136 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4137 *optp++ = sc->sc_ourmaxseg & 0xff; 4138 4139 if (sc->sc_request_r_scale != 15) { 4140 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4141 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4142 sc->sc_request_r_scale); 4143 optp += 4; 4144 } 4145 4146 if (sc->sc_flags & SCF_TIMESTAMP) { 4147 u_int32_t *lp = (u_int32_t *)(optp); 4148 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4149 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4150 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4151 *lp = htonl(sc->sc_timestamp); 4152 optp += TCPOLEN_TSTAMP_APPA; 4153 } 4154 4155 #ifdef TCP_SIGNATURE 4156 if (sc->sc_flags & SCF_SIGNATURE) { 4157 struct secasvar *sav; 4158 u_int8_t *sigp; 4159 4160 sav = tcp_signature_getsav(m, th); 4161 4162 if (sav == NULL) { 4163 if (m) 4164 m_freem(m); 4165 return (EPERM); 4166 } 4167 4168 *optp++ = TCPOPT_SIGNATURE; 4169 *optp++ = TCPOLEN_SIGNATURE; 4170 sigp = optp; 4171 bzero(optp, TCP_SIGLEN); 4172 optp += TCP_SIGLEN; 4173 *optp++ = TCPOPT_NOP; 4174 *optp++ = TCPOPT_EOL; 4175 4176 (void)tcp_signature(m, th, hlen, sav, sigp); 4177 4178 key_sa_recordxfer(sav, m); 4179 #ifdef FAST_IPSEC 4180 KEY_FREESAV(&sav); 4181 #else 4182 key_freesav(sav); 4183 #endif 4184 } 4185 #endif 4186 4187 /* Compute the packet's checksum. */ 4188 switch (sc->sc_src.sa.sa_family) { 4189 case AF_INET: 4190 ip->ip_len = htons(tlen - hlen); 4191 th->th_sum = 0; 4192 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4193 break; 4194 #ifdef INET6 4195 case AF_INET6: 4196 ip6->ip6_plen = htons(tlen - hlen); 4197 th->th_sum = 0; 4198 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4199 break; 4200 #endif 4201 } 4202 4203 /* 4204 * Fill in some straggling IP bits. Note the stack expects 4205 * ip_len to be in host order, for convenience. 4206 */ 4207 switch (sc->sc_src.sa.sa_family) { 4208 #ifdef INET 4209 case AF_INET: 4210 ip->ip_len = htons(tlen); 4211 ip->ip_ttl = ip_defttl; 4212 /* XXX tos? */ 4213 break; 4214 #endif 4215 #ifdef INET6 4216 case AF_INET6: 4217 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4218 ip6->ip6_vfc |= IPV6_VERSION; 4219 ip6->ip6_plen = htons(tlen - hlen); 4220 /* ip6_hlim will be initialized afterwards */ 4221 /* XXX flowlabel? */ 4222 break; 4223 #endif 4224 } 4225 4226 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4227 tp = sc->sc_tp; 4228 4229 switch (sc->sc_src.sa.sa_family) { 4230 #ifdef INET 4231 case AF_INET: 4232 error = ip_output(m, sc->sc_ipopts, ro, 4233 (ip_mtudisc ? IP_MTUDISC : 0), 4234 (struct ip_moptions *)NULL, so); 4235 break; 4236 #endif 4237 #ifdef INET6 4238 case AF_INET6: 4239 ip6->ip6_hlim = in6_selecthlim(NULL, 4240 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 4241 4242 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0, 4243 (struct ip6_moptions *)0, so, NULL); 4244 break; 4245 #endif 4246 default: 4247 error = EAFNOSUPPORT; 4248 break; 4249 } 4250 return (error); 4251 } 4252
Indexes created Mon Sep 22 10:09:38 GMT 2025