tcp_input.c revision 1.194
1 /* $NetBSD: tcp_input.c,v 1.194 2004/04/20 16:52:12 itojun 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 1.194 2004/04/20 16:52:12 itojun 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> 170 171 #include <net/if.h> 172 #include <net/route.h> 173 #include <net/if_types.h> 174 175 #include <netinet/in.h> 176 #include <netinet/in_systm.h> 177 #include <netinet/ip.h> 178 #include <netinet/in_pcb.h> 179 #include <netinet/in_var.h> 180 #include <netinet/ip_var.h> 181 182 #ifdef INET6 183 #ifndef INET 184 #include <netinet/in.h> 185 #endif 186 #include <netinet/ip6.h> 187 #include <netinet6/ip6_var.h> 188 #include <netinet6/in6_pcb.h> 189 #include <netinet6/ip6_var.h> 190 #include <netinet6/in6_var.h> 191 #include <netinet/icmp6.h> 192 #include <netinet6/nd6.h> 193 #endif 194 195 #ifndef INET6 196 /* always need ip6.h for IP6_EXTHDR_GET */ 197 #include <netinet/ip6.h> 198 #endif 199 200 #include <netinet/tcp.h> 201 #include <netinet/tcp_fsm.h> 202 #include <netinet/tcp_seq.h> 203 #include <netinet/tcp_timer.h> 204 #include <netinet/tcp_var.h> 205 #include <netinet/tcpip.h> 206 #include <netinet/tcp_debug.h> 207 208 #include <machine/stdarg.h> 209 210 #ifdef IPSEC 211 #include <netinet6/ipsec.h> 212 #include <netkey/key.h> 213 #endif /*IPSEC*/ 214 #ifdef INET6 215 #include "faith.h" 216 #if defined(NFAITH) && NFAITH > 0 217 #include <net/if_faith.h> 218 #endif 219 #endif /* IPSEC */ 220 221 #ifdef FAST_IPSEC 222 #include <netipsec/ipsec.h> 223 #include <netipsec/key.h> 224 #ifdef INET6 225 #include <netipsec/ipsec6.h> 226 #endif 227 #endif /* FAST_IPSEC*/ 228 229 230 int tcprexmtthresh = 3; 231 int tcp_log_refused; 232 233 static int tcp_rst_ppslim_count = 0; 234 static struct timeval tcp_rst_ppslim_last; 235 static int tcp_ackdrop_ppslim_count = 0; 236 static struct timeval tcp_ackdrop_ppslim_last; 237 238 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 239 240 /* for modulo comparisons of timestamps */ 241 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 242 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 243 244 /* 245 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 246 */ 247 #ifdef INET6 248 #define ND6_HINT(tp) \ 249 do { \ 250 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 && \ 251 tp->t_in6pcb->in6p_route.ro_rt) { \ 252 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \ 253 } \ 254 } while (/*CONSTCOND*/ 0) 255 #else 256 #define ND6_HINT(tp) 257 #endif 258 259 /* 260 * Macro to compute ACK transmission behavior. Delay the ACK unless 261 * we have already delayed an ACK (must send an ACK every two segments). 262 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 263 * option is enabled. 264 */ 265 #define TCP_SETUP_ACK(tp, th) \ 266 do { \ 267 if ((tp)->t_flags & TF_DELACK || \ 268 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 269 tp->t_flags |= TF_ACKNOW; \ 270 else \ 271 TCP_SET_DELACK(tp); \ 272 } while (/*CONSTCOND*/ 0) 273 274 /* 275 * Convert TCP protocol fields to host order for easier processing. 276 */ 277 #define TCP_FIELDS_TO_HOST(th) \ 278 do { \ 279 NTOHL((th)->th_seq); \ 280 NTOHL((th)->th_ack); \ 281 NTOHS((th)->th_win); \ 282 NTOHS((th)->th_urp); \ 283 } while (/*CONSTCOND*/ 0) 284 285 /* 286 * ... and reverse the above. 287 */ 288 #define TCP_FIELDS_TO_NET(th) \ 289 do { \ 290 HTONL((th)->th_seq); \ 291 HTONL((th)->th_ack); \ 292 HTONS((th)->th_win); \ 293 HTONS((th)->th_urp); \ 294 } while (/*CONSTCOND*/ 0) 295 296 #ifdef TCP_CSUM_COUNTERS 297 #include <sys/device.h> 298 299 extern struct evcnt tcp_hwcsum_ok; 300 extern struct evcnt tcp_hwcsum_bad; 301 extern struct evcnt tcp_hwcsum_data; 302 extern struct evcnt tcp_swcsum; 303 304 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 305 306 #else 307 308 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 309 310 #endif /* TCP_CSUM_COUNTERS */ 311 312 #ifdef TCP_REASS_COUNTERS 313 #include <sys/device.h> 314 315 extern struct evcnt tcp_reass_; 316 extern struct evcnt tcp_reass_empty; 317 extern struct evcnt tcp_reass_iteration[8]; 318 extern struct evcnt tcp_reass_prependfirst; 319 extern struct evcnt tcp_reass_prepend; 320 extern struct evcnt tcp_reass_insert; 321 extern struct evcnt tcp_reass_inserttail; 322 extern struct evcnt tcp_reass_append; 323 extern struct evcnt tcp_reass_appendtail; 324 extern struct evcnt tcp_reass_overlaptail; 325 extern struct evcnt tcp_reass_overlapfront; 326 extern struct evcnt tcp_reass_segdup; 327 extern struct evcnt tcp_reass_fragdup; 328 329 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 330 331 #else 332 333 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 334 335 #endif /* TCP_REASS_COUNTERS */ 336 337 #ifdef INET 338 static void tcp4_log_refused __P((const struct ip *, const struct tcphdr *)); 339 #endif 340 #ifdef INET6 341 static void tcp6_log_refused 342 __P((const struct ip6_hdr *, const struct tcphdr *)); 343 #endif 344 345 int 346 tcp_reass(tp, th, m, tlen) 347 struct tcpcb *tp; 348 struct tcphdr *th; 349 struct mbuf *m; 350 int *tlen; 351 { 352 struct ipqent *p, *q, *nq, *tiqe = NULL; 353 struct socket *so = NULL; 354 int pkt_flags; 355 tcp_seq pkt_seq; 356 unsigned pkt_len; 357 u_long rcvpartdupbyte = 0; 358 u_long rcvoobyte; 359 #ifdef TCP_REASS_COUNTERS 360 u_int count = 0; 361 #endif 362 363 if (tp->t_inpcb) 364 so = tp->t_inpcb->inp_socket; 365 #ifdef INET6 366 else if (tp->t_in6pcb) 367 so = tp->t_in6pcb->in6p_socket; 368 #endif 369 370 TCP_REASS_LOCK_CHECK(tp); 371 372 /* 373 * Call with th==0 after become established to 374 * force pre-ESTABLISHED data up to user socket. 375 */ 376 if (th == 0) 377 goto present; 378 379 rcvoobyte = *tlen; 380 /* 381 * Copy these to local variables because the tcpiphdr 382 * gets munged while we are collapsing mbufs. 383 */ 384 pkt_seq = th->th_seq; 385 pkt_len = *tlen; 386 pkt_flags = th->th_flags; 387 388 TCP_REASS_COUNTER_INCR(&tcp_reass_); 389 390 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 391 /* 392 * When we miss a packet, the vast majority of time we get 393 * packets that follow it in order. So optimize for that. 394 */ 395 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 396 p->ipqe_len += pkt_len; 397 p->ipqe_flags |= pkt_flags; 398 m_cat(p->ipqe_m, m); 399 m = NULL; 400 tiqe = p; 401 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 402 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 403 goto skip_replacement; 404 } 405 /* 406 * While we're here, if the pkt is completely beyond 407 * anything we have, just insert it at the tail. 408 */ 409 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 410 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 411 goto insert_it; 412 } 413 } 414 415 q = TAILQ_FIRST(&tp->segq); 416 417 if (q != NULL) { 418 /* 419 * If this segment immediately precedes the first out-of-order 420 * block, simply slap the segment in front of it and (mostly) 421 * skip the complicated logic. 422 */ 423 if (pkt_seq + pkt_len == q->ipqe_seq) { 424 q->ipqe_seq = pkt_seq; 425 q->ipqe_len += pkt_len; 426 q->ipqe_flags |= pkt_flags; 427 m_cat(m, q->ipqe_m); 428 q->ipqe_m = m; 429 tiqe = q; 430 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 431 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 432 goto skip_replacement; 433 } 434 } else { 435 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 436 } 437 438 /* 439 * Find a segment which begins after this one does. 440 */ 441 for (p = NULL; q != NULL; q = nq) { 442 nq = TAILQ_NEXT(q, ipqe_q); 443 #ifdef TCP_REASS_COUNTERS 444 count++; 445 #endif 446 /* 447 * If the received segment is just right after this 448 * fragment, merge the two together and then check 449 * for further overlaps. 450 */ 451 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 452 #ifdef TCPREASS_DEBUG 453 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 454 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 455 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 456 #endif 457 pkt_len += q->ipqe_len; 458 pkt_flags |= q->ipqe_flags; 459 pkt_seq = q->ipqe_seq; 460 m_cat(q->ipqe_m, m); 461 m = q->ipqe_m; 462 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 463 goto free_ipqe; 464 } 465 /* 466 * If the received segment is completely past this 467 * fragment, we need to go the next fragment. 468 */ 469 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 470 p = q; 471 continue; 472 } 473 /* 474 * If the fragment is past the received segment, 475 * it (or any following) can't be concatenated. 476 */ 477 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 478 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 479 break; 480 } 481 482 /* 483 * We've received all the data in this segment before. 484 * mark it as a duplicate and return. 485 */ 486 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 487 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 488 tcpstat.tcps_rcvduppack++; 489 tcpstat.tcps_rcvdupbyte += pkt_len; 490 m_freem(m); 491 if (tiqe != NULL) 492 pool_put(&ipqent_pool, tiqe); 493 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 494 return (0); 495 } 496 /* 497 * Received segment completely overlaps this fragment 498 * so we drop the fragment (this keeps the temporal 499 * ordering of segments correct). 500 */ 501 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 502 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 503 rcvpartdupbyte += q->ipqe_len; 504 m_freem(q->ipqe_m); 505 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 506 goto free_ipqe; 507 } 508 /* 509 * RX'ed segment extends past the end of the 510 * fragment. Drop the overlapping bytes. Then 511 * merge the fragment and segment then treat as 512 * a longer received packet. 513 */ 514 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 515 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 516 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 517 #ifdef TCPREASS_DEBUG 518 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 519 tp, overlap, 520 pkt_seq, pkt_seq + pkt_len, pkt_len); 521 #endif 522 m_adj(m, overlap); 523 rcvpartdupbyte += overlap; 524 m_cat(q->ipqe_m, m); 525 m = q->ipqe_m; 526 pkt_seq = q->ipqe_seq; 527 pkt_len += q->ipqe_len - overlap; 528 rcvoobyte -= overlap; 529 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 530 goto free_ipqe; 531 } 532 /* 533 * RX'ed segment extends past the front of the 534 * fragment. Drop the overlapping bytes on the 535 * received packet. The packet will then be 536 * contatentated with this fragment a bit later. 537 */ 538 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 539 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 540 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 541 #ifdef TCPREASS_DEBUG 542 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 543 tp, overlap, 544 pkt_seq, pkt_seq + pkt_len, pkt_len); 545 #endif 546 m_adj(m, -overlap); 547 pkt_len -= overlap; 548 rcvpartdupbyte += overlap; 549 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 550 rcvoobyte -= overlap; 551 } 552 /* 553 * If the received segment immediates precedes this 554 * fragment then tack the fragment onto this segment 555 * and reinsert the data. 556 */ 557 if (q->ipqe_seq == pkt_seq + pkt_len) { 558 #ifdef TCPREASS_DEBUG 559 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 560 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 561 pkt_seq, pkt_seq + pkt_len, pkt_len); 562 #endif 563 pkt_len += q->ipqe_len; 564 pkt_flags |= q->ipqe_flags; 565 m_cat(m, q->ipqe_m); 566 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 567 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 568 if (tiqe == NULL) 569 tiqe = q; 570 else 571 pool_put(&ipqent_pool, q); 572 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 573 break; 574 } 575 /* 576 * If the fragment is before the segment, remember it. 577 * When this loop is terminated, p will contain the 578 * pointer to fragment that is right before the received 579 * segment. 580 */ 581 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 582 p = q; 583 584 continue; 585 586 /* 587 * This is a common operation. It also will allow 588 * to save doing a malloc/free in most instances. 589 */ 590 free_ipqe: 591 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 592 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 593 if (tiqe == NULL) 594 tiqe = q; 595 else 596 pool_put(&ipqent_pool, q); 597 } 598 599 #ifdef TCP_REASS_COUNTERS 600 if (count > 7) 601 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 602 else if (count > 0) 603 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 604 #endif 605 606 insert_it: 607 608 /* 609 * Allocate a new queue entry since the received segment did not 610 * collapse onto any other out-of-order block; thus we are allocating 611 * a new block. If it had collapsed, tiqe would not be NULL and 612 * we would be reusing it. 613 * XXX If we can't, just drop the packet. XXX 614 */ 615 if (tiqe == NULL) { 616 tiqe = pool_get(&ipqent_pool, PR_NOWAIT); 617 if (tiqe == NULL) { 618 tcpstat.tcps_rcvmemdrop++; 619 m_freem(m); 620 return (0); 621 } 622 } 623 624 /* 625 * Update the counters. 626 */ 627 tcpstat.tcps_rcvoopack++; 628 tcpstat.tcps_rcvoobyte += rcvoobyte; 629 if (rcvpartdupbyte) { 630 tcpstat.tcps_rcvpartduppack++; 631 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 632 } 633 634 /* 635 * Insert the new fragment queue entry into both queues. 636 */ 637 tiqe->ipqe_m = m; 638 tiqe->ipqe_seq = pkt_seq; 639 tiqe->ipqe_len = pkt_len; 640 tiqe->ipqe_flags = pkt_flags; 641 if (p == NULL) { 642 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 643 #ifdef TCPREASS_DEBUG 644 if (tiqe->ipqe_seq != tp->rcv_nxt) 645 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 646 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 647 #endif 648 } else { 649 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 650 #ifdef TCPREASS_DEBUG 651 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 652 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 653 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 654 #endif 655 } 656 657 skip_replacement: 658 659 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 660 661 present: 662 /* 663 * Present data to user, advancing rcv_nxt through 664 * completed sequence space. 665 */ 666 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 667 return (0); 668 q = TAILQ_FIRST(&tp->segq); 669 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 670 return (0); 671 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 672 return (0); 673 674 tp->rcv_nxt += q->ipqe_len; 675 pkt_flags = q->ipqe_flags & TH_FIN; 676 ND6_HINT(tp); 677 678 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 679 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 680 if (so->so_state & SS_CANTRCVMORE) 681 m_freem(q->ipqe_m); 682 else 683 sbappendstream(&so->so_rcv, q->ipqe_m); 684 pool_put(&ipqent_pool, q); 685 sorwakeup(so); 686 return (pkt_flags); 687 } 688 689 #ifdef INET6 690 int 691 tcp6_input(mp, offp, proto) 692 struct mbuf **mp; 693 int *offp, proto; 694 { 695 struct mbuf *m = *mp; 696 697 /* 698 * draft-itojun-ipv6-tcp-to-anycast 699 * better place to put this in? 700 */ 701 if (m->m_flags & M_ANYCAST6) { 702 struct ip6_hdr *ip6; 703 if (m->m_len < sizeof(struct ip6_hdr)) { 704 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 705 tcpstat.tcps_rcvshort++; 706 return IPPROTO_DONE; 707 } 708 } 709 ip6 = mtod(m, struct ip6_hdr *); 710 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 711 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 712 return IPPROTO_DONE; 713 } 714 715 tcp_input(m, *offp, proto); 716 return IPPROTO_DONE; 717 } 718 #endif 719 720 #ifdef INET 721 static void 722 tcp4_log_refused(ip, th) 723 const struct ip *ip; 724 const struct tcphdr *th; 725 { 726 char src[4*sizeof "123"]; 727 char dst[4*sizeof "123"]; 728 729 if (ip) { 730 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 731 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 732 } 733 else { 734 strlcpy(src, "(unknown)", sizeof(src)); 735 strlcpy(dst, "(unknown)", sizeof(dst)); 736 } 737 log(LOG_INFO, 738 "Connection attempt to TCP %s:%d from %s:%d\n", 739 dst, ntohs(th->th_dport), 740 src, ntohs(th->th_sport)); 741 } 742 #endif 743 744 #ifdef INET6 745 static void 746 tcp6_log_refused(ip6, th) 747 const struct ip6_hdr *ip6; 748 const struct tcphdr *th; 749 { 750 char src[INET6_ADDRSTRLEN]; 751 char dst[INET6_ADDRSTRLEN]; 752 753 if (ip6) { 754 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 755 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 756 } 757 else { 758 strlcpy(src, "(unknown v6)", sizeof(src)); 759 strlcpy(dst, "(unknown v6)", sizeof(dst)); 760 } 761 log(LOG_INFO, 762 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 763 dst, ntohs(th->th_dport), 764 src, ntohs(th->th_sport)); 765 } 766 #endif 767 768 /* 769 * TCP input routine, follows pages 65-76 of the 770 * protocol specification dated September, 1981 very closely. 771 */ 772 void 773 #if __STDC__ 774 tcp_input(struct mbuf *m, ...) 775 #else 776 tcp_input(m, va_alist) 777 struct mbuf *m; 778 #endif 779 { 780 struct tcphdr *th; 781 struct ip *ip; 782 struct inpcb *inp; 783 #ifdef INET6 784 struct ip6_hdr *ip6; 785 struct in6pcb *in6p; 786 #endif 787 u_int8_t *optp = NULL; 788 int optlen = 0; 789 int len, tlen, toff, hdroptlen = 0; 790 struct tcpcb *tp = 0; 791 int tiflags; 792 struct socket *so = NULL; 793 int todrop, acked, ourfinisacked, needoutput = 0; 794 #ifdef TCP_DEBUG 795 short ostate = 0; 796 #endif 797 int iss = 0; 798 u_long tiwin; 799 struct tcp_opt_info opti; 800 int off, iphlen; 801 va_list ap; 802 int af; /* af on the wire */ 803 struct mbuf *tcp_saveti = NULL; 804 805 MCLAIM(m, &tcp_rx_mowner); 806 va_start(ap, m); 807 toff = va_arg(ap, int); 808 (void)va_arg(ap, int); /* ignore value, advance ap */ 809 va_end(ap); 810 811 tcpstat.tcps_rcvtotal++; 812 813 bzero(&opti, sizeof(opti)); 814 opti.ts_present = 0; 815 opti.maxseg = 0; 816 817 /* 818 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 819 * 820 * TCP is, by definition, unicast, so we reject all 821 * multicast outright. 822 * 823 * Note, there are additional src/dst address checks in 824 * the AF-specific code below. 825 */ 826 if (m->m_flags & (M_BCAST|M_MCAST)) { 827 /* XXX stat */ 828 goto drop; 829 } 830 #ifdef INET6 831 if (m->m_flags & M_ANYCAST6) { 832 /* XXX stat */ 833 goto drop; 834 } 835 #endif 836 837 /* 838 * Get IP and TCP header together in first mbuf. 839 * Note: IP leaves IP header in first mbuf. 840 */ 841 ip = mtod(m, struct ip *); 842 #ifdef INET6 843 ip6 = NULL; 844 #endif 845 switch (ip->ip_v) { 846 #ifdef INET 847 case 4: 848 af = AF_INET; 849 iphlen = sizeof(struct ip); 850 ip = mtod(m, struct ip *); 851 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 852 sizeof(struct tcphdr)); 853 if (th == NULL) { 854 tcpstat.tcps_rcvshort++; 855 return; 856 } 857 /* We do the checksum after PCB lookup... */ 858 len = ntohs(ip->ip_len); 859 tlen = len - toff; 860 break; 861 #endif 862 #ifdef INET6 863 case 6: 864 ip = NULL; 865 iphlen = sizeof(struct ip6_hdr); 866 af = AF_INET6; 867 ip6 = mtod(m, struct ip6_hdr *); 868 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 869 sizeof(struct tcphdr)); 870 if (th == NULL) { 871 tcpstat.tcps_rcvshort++; 872 return; 873 } 874 875 /* Be proactive about malicious use of IPv4 mapped address */ 876 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 877 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 878 /* XXX stat */ 879 goto drop; 880 } 881 882 /* 883 * Be proactive about unspecified IPv6 address in source. 884 * As we use all-zero to indicate unbounded/unconnected pcb, 885 * unspecified IPv6 address can be used to confuse us. 886 * 887 * Note that packets with unspecified IPv6 destination is 888 * already dropped in ip6_input. 889 */ 890 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 891 /* XXX stat */ 892 goto drop; 893 } 894 895 /* 896 * Make sure destination address is not multicast. 897 * Source address checked in ip6_input(). 898 */ 899 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 900 /* XXX stat */ 901 goto drop; 902 } 903 904 /* We do the checksum after PCB lookup... */ 905 len = m->m_pkthdr.len; 906 tlen = len - toff; 907 break; 908 #endif 909 default: 910 m_freem(m); 911 return; 912 } 913 914 KASSERT(TCP_HDR_ALIGNED_P(th)); 915 916 /* 917 * Check that TCP offset makes sense, 918 * pull out TCP options and adjust length. XXX 919 */ 920 off = th->th_off << 2; 921 if (off < sizeof (struct tcphdr) || off > tlen) { 922 tcpstat.tcps_rcvbadoff++; 923 goto drop; 924 } 925 tlen -= off; 926 927 /* 928 * tcp_input() has been modified to use tlen to mean the TCP data 929 * length throughout the function. Other functions can use 930 * m->m_pkthdr.len as the basis for calculating the TCP data length. 931 * rja 932 */ 933 934 if (off > sizeof (struct tcphdr)) { 935 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 936 if (th == NULL) { 937 tcpstat.tcps_rcvshort++; 938 return; 939 } 940 /* 941 * NOTE: ip/ip6 will not be affected by m_pulldown() 942 * (as they're before toff) and we don't need to update those. 943 */ 944 KASSERT(TCP_HDR_ALIGNED_P(th)); 945 optlen = off - sizeof (struct tcphdr); 946 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 947 /* 948 * Do quick retrieval of timestamp options ("options 949 * prediction?"). If timestamp is the only option and it's 950 * formatted as recommended in RFC 1323 appendix A, we 951 * quickly get the values now and not bother calling 952 * tcp_dooptions(), etc. 953 */ 954 if ((optlen == TCPOLEN_TSTAMP_APPA || 955 (optlen > TCPOLEN_TSTAMP_APPA && 956 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 957 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 958 (th->th_flags & TH_SYN) == 0) { 959 opti.ts_present = 1; 960 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 961 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 962 optp = NULL; /* we've parsed the options */ 963 } 964 } 965 tiflags = th->th_flags; 966 967 /* 968 * Locate pcb for segment. 969 */ 970 findpcb: 971 inp = NULL; 972 #ifdef INET6 973 in6p = NULL; 974 #endif 975 switch (af) { 976 #ifdef INET 977 case AF_INET: 978 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 979 ip->ip_dst, th->th_dport); 980 if (inp == 0) { 981 ++tcpstat.tcps_pcbhashmiss; 982 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 983 } 984 #ifdef INET6 985 if (inp == 0) { 986 struct in6_addr s, d; 987 988 /* mapped addr case */ 989 bzero(&s, sizeof(s)); 990 s.s6_addr16[5] = htons(0xffff); 991 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 992 bzero(&d, sizeof(d)); 993 d.s6_addr16[5] = htons(0xffff); 994 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 995 in6p = in6_pcblookup_connect(&tcbtable, &s, 996 th->th_sport, &d, th->th_dport, 0); 997 if (in6p == 0) { 998 ++tcpstat.tcps_pcbhashmiss; 999 in6p = in6_pcblookup_bind(&tcbtable, &d, 1000 th->th_dport, 0); 1001 } 1002 } 1003 #endif 1004 #ifndef INET6 1005 if (inp == 0) 1006 #else 1007 if (inp == 0 && in6p == 0) 1008 #endif 1009 { 1010 ++tcpstat.tcps_noport; 1011 if (tcp_log_refused && 1012 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1013 tcp4_log_refused(ip, th); 1014 } 1015 TCP_FIELDS_TO_HOST(th); 1016 goto dropwithreset_ratelim; 1017 } 1018 #if defined(IPSEC) || defined(FAST_IPSEC) 1019 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1020 ipsec4_in_reject(m, inp)) { 1021 ipsecstat.in_polvio++; 1022 goto drop; 1023 } 1024 #ifdef INET6 1025 else if (in6p && 1026 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1027 ipsec4_in_reject_so(m, in6p->in6p_socket)) { 1028 ipsecstat.in_polvio++; 1029 goto drop; 1030 } 1031 #endif 1032 #endif /*IPSEC*/ 1033 break; 1034 #endif /*INET*/ 1035 #ifdef INET6 1036 case AF_INET6: 1037 { 1038 int faith; 1039 1040 #if defined(NFAITH) && NFAITH > 0 1041 faith = faithprefix(&ip6->ip6_dst); 1042 #else 1043 faith = 0; 1044 #endif 1045 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1046 th->th_sport, &ip6->ip6_dst, th->th_dport, faith); 1047 if (in6p == NULL) { 1048 ++tcpstat.tcps_pcbhashmiss; 1049 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1050 th->th_dport, faith); 1051 } 1052 if (in6p == NULL) { 1053 ++tcpstat.tcps_noport; 1054 if (tcp_log_refused && 1055 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1056 tcp6_log_refused(ip6, th); 1057 } 1058 TCP_FIELDS_TO_HOST(th); 1059 goto dropwithreset_ratelim; 1060 } 1061 #if defined(IPSEC) || defined(FAST_IPSEC) 1062 if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1063 ipsec6_in_reject(m, in6p)) { 1064 ipsec6stat.in_polvio++; 1065 goto drop; 1066 } 1067 #endif /*IPSEC*/ 1068 break; 1069 } 1070 #endif 1071 } 1072 1073 /* 1074 * If the state is CLOSED (i.e., TCB does not exist) then 1075 * all data in the incoming segment is discarded. 1076 * If the TCB exists but is in CLOSED state, it is embryonic, 1077 * but should either do a listen or a connect soon. 1078 */ 1079 tp = NULL; 1080 so = NULL; 1081 if (inp) { 1082 tp = intotcpcb(inp); 1083 so = inp->inp_socket; 1084 } 1085 #ifdef INET6 1086 else if (in6p) { 1087 tp = in6totcpcb(in6p); 1088 so = in6p->in6p_socket; 1089 } 1090 #endif 1091 if (tp == 0) { 1092 TCP_FIELDS_TO_HOST(th); 1093 goto dropwithreset_ratelim; 1094 } 1095 if (tp->t_state == TCPS_CLOSED) 1096 goto drop; 1097 1098 /* 1099 * Checksum extended TCP header and data. 1100 */ 1101 switch (af) { 1102 #ifdef INET 1103 case AF_INET: 1104 switch (m->m_pkthdr.csum_flags & 1105 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 1106 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 1107 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 1108 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 1109 goto badcsum; 1110 1111 case M_CSUM_TCPv4|M_CSUM_DATA: { 1112 u_int32_t hw_csum = m->m_pkthdr.csum_data; 1113 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 1114 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 1115 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 1116 ip->ip_dst.s_addr, 1117 htons(hw_csum + tlen + off + IPPROTO_TCP)); 1118 } 1119 if ((hw_csum ^ 0xffff) != 0) 1120 goto badcsum; 1121 break; 1122 } 1123 1124 case M_CSUM_TCPv4: 1125 /* Checksum was okay. */ 1126 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 1127 break; 1128 1129 default: 1130 /* Must compute it ourselves. */ 1131 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 1132 if (in4_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 1133 goto badcsum; 1134 break; 1135 } 1136 break; 1137 #endif /* INET4 */ 1138 1139 #ifdef INET6 1140 case AF_INET6: 1141 if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 1142 goto badcsum; 1143 break; 1144 #endif /* INET6 */ 1145 } 1146 1147 TCP_FIELDS_TO_HOST(th); 1148 1149 /* Unscale the window into a 32-bit value. */ 1150 if ((tiflags & TH_SYN) == 0) 1151 tiwin = th->th_win << tp->snd_scale; 1152 else 1153 tiwin = th->th_win; 1154 1155 #ifdef INET6 1156 /* save packet options if user wanted */ 1157 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1158 if (in6p->in6p_options) { 1159 m_freem(in6p->in6p_options); 1160 in6p->in6p_options = 0; 1161 } 1162 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1163 } 1164 #endif 1165 1166 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1167 union syn_cache_sa src; 1168 union syn_cache_sa dst; 1169 1170 bzero(&src, sizeof(src)); 1171 bzero(&dst, sizeof(dst)); 1172 switch (af) { 1173 #ifdef INET 1174 case AF_INET: 1175 src.sin.sin_len = sizeof(struct sockaddr_in); 1176 src.sin.sin_family = AF_INET; 1177 src.sin.sin_addr = ip->ip_src; 1178 src.sin.sin_port = th->th_sport; 1179 1180 dst.sin.sin_len = sizeof(struct sockaddr_in); 1181 dst.sin.sin_family = AF_INET; 1182 dst.sin.sin_addr = ip->ip_dst; 1183 dst.sin.sin_port = th->th_dport; 1184 break; 1185 #endif 1186 #ifdef INET6 1187 case AF_INET6: 1188 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1189 src.sin6.sin6_family = AF_INET6; 1190 src.sin6.sin6_addr = ip6->ip6_src; 1191 src.sin6.sin6_port = th->th_sport; 1192 1193 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1194 dst.sin6.sin6_family = AF_INET6; 1195 dst.sin6.sin6_addr = ip6->ip6_dst; 1196 dst.sin6.sin6_port = th->th_dport; 1197 break; 1198 #endif /* INET6 */ 1199 default: 1200 goto badsyn; /*sanity*/ 1201 } 1202 1203 if (so->so_options & SO_DEBUG) { 1204 #ifdef TCP_DEBUG 1205 ostate = tp->t_state; 1206 #endif 1207 1208 tcp_saveti = NULL; 1209 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1210 goto nosave; 1211 1212 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1213 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1214 if (!tcp_saveti) 1215 goto nosave; 1216 } else { 1217 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1218 if (!tcp_saveti) 1219 goto nosave; 1220 MCLAIM(m, &tcp_mowner); 1221 tcp_saveti->m_len = iphlen; 1222 m_copydata(m, 0, iphlen, 1223 mtod(tcp_saveti, caddr_t)); 1224 } 1225 1226 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1227 m_freem(tcp_saveti); 1228 tcp_saveti = NULL; 1229 } else { 1230 tcp_saveti->m_len += sizeof(struct tcphdr); 1231 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1232 sizeof(struct tcphdr)); 1233 } 1234 nosave:; 1235 } 1236 if (so->so_options & SO_ACCEPTCONN) { 1237 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1238 if (tiflags & TH_RST) { 1239 syn_cache_reset(&src.sa, &dst.sa, th); 1240 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1241 (TH_ACK|TH_SYN)) { 1242 /* 1243 * Received a SYN,ACK. This should 1244 * never happen while we are in 1245 * LISTEN. Send an RST. 1246 */ 1247 goto badsyn; 1248 } else if (tiflags & TH_ACK) { 1249 so = syn_cache_get(&src.sa, &dst.sa, 1250 th, toff, tlen, so, m); 1251 if (so == NULL) { 1252 /* 1253 * We don't have a SYN for 1254 * this ACK; send an RST. 1255 */ 1256 goto badsyn; 1257 } else if (so == 1258 (struct socket *)(-1)) { 1259 /* 1260 * We were unable to create 1261 * the connection. If the 1262 * 3-way handshake was 1263 * completed, and RST has 1264 * been sent to the peer. 1265 * Since the mbuf might be 1266 * in use for the reply, 1267 * do not free it. 1268 */ 1269 m = NULL; 1270 } else { 1271 /* 1272 * We have created a 1273 * full-blown connection. 1274 */ 1275 tp = NULL; 1276 inp = NULL; 1277 #ifdef INET6 1278 in6p = NULL; 1279 #endif 1280 switch (so->so_proto->pr_domain->dom_family) { 1281 #ifdef INET 1282 case AF_INET: 1283 inp = sotoinpcb(so); 1284 tp = intotcpcb(inp); 1285 break; 1286 #endif 1287 #ifdef INET6 1288 case AF_INET6: 1289 in6p = sotoin6pcb(so); 1290 tp = in6totcpcb(in6p); 1291 break; 1292 #endif 1293 } 1294 if (tp == NULL) 1295 goto badsyn; /*XXX*/ 1296 tiwin <<= tp->snd_scale; 1297 goto after_listen; 1298 } 1299 } else { 1300 /* 1301 * None of RST, SYN or ACK was set. 1302 * This is an invalid packet for a 1303 * TCB in LISTEN state. Send a RST. 1304 */ 1305 goto badsyn; 1306 } 1307 } else { 1308 /* 1309 * Received a SYN. 1310 */ 1311 1312 #ifdef INET6 1313 /* 1314 * If deprecated address is forbidden, we do 1315 * not accept SYN to deprecated interface 1316 * address to prevent any new inbound 1317 * connection from getting established. 1318 * When we do not accept SYN, we send a TCP 1319 * RST, with deprecated source address (instead 1320 * of dropping it). We compromise it as it is 1321 * much better for peer to send a RST, and 1322 * RST will be the final packet for the 1323 * exchange. 1324 * 1325 * If we do not forbid deprecated addresses, we 1326 * accept the SYN packet. RFC2462 does not 1327 * suggest dropping SYN in this case. 1328 * If we decipher RFC2462 5.5.4, it says like 1329 * this: 1330 * 1. use of deprecated addr with existing 1331 * communication is okay - "SHOULD continue 1332 * to be used" 1333 * 2. use of it with new communication: 1334 * (2a) "SHOULD NOT be used if alternate 1335 * address with sufficient scope is 1336 * available" 1337 * (2b) nothing mentioned otherwise. 1338 * Here we fall into (2b) case as we have no 1339 * choice in our source address selection - we 1340 * must obey the peer. 1341 * 1342 * The wording in RFC2462 is confusing, and 1343 * there are multiple description text for 1344 * deprecated address handling - worse, they 1345 * are not exactly the same. I believe 5.5.4 1346 * is the best one, so we follow 5.5.4. 1347 */ 1348 if (af == AF_INET6 && !ip6_use_deprecated) { 1349 struct in6_ifaddr *ia6; 1350 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1351 &ip6->ip6_dst)) && 1352 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1353 tp = NULL; 1354 goto dropwithreset; 1355 } 1356 } 1357 #endif 1358 1359 #ifdef IPSEC 1360 switch (af) { 1361 #ifdef INET 1362 case AF_INET: 1363 if (ipsec4_in_reject_so(m, so)) { 1364 ipsecstat.in_polvio++; 1365 tp = NULL; 1366 goto dropwithreset; 1367 } 1368 break; 1369 #endif 1370 #ifdef INET6 1371 case AF_INET6: 1372 if (ipsec6_in_reject_so(m, so)) { 1373 ipsec6stat.in_polvio++; 1374 tp = NULL; 1375 goto dropwithreset; 1376 } 1377 break; 1378 #endif 1379 } 1380 #endif 1381 1382 /* 1383 * LISTEN socket received a SYN 1384 * from itself? This can't possibly 1385 * be valid; drop the packet. 1386 */ 1387 if (th->th_sport == th->th_dport) { 1388 int i; 1389 1390 switch (af) { 1391 #ifdef INET 1392 case AF_INET: 1393 i = in_hosteq(ip->ip_src, ip->ip_dst); 1394 break; 1395 #endif 1396 #ifdef INET6 1397 case AF_INET6: 1398 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1399 break; 1400 #endif 1401 default: 1402 i = 1; 1403 } 1404 if (i) { 1405 tcpstat.tcps_badsyn++; 1406 goto drop; 1407 } 1408 } 1409 1410 /* 1411 * SYN looks ok; create compressed TCP 1412 * state for it. 1413 */ 1414 if (so->so_qlen <= so->so_qlimit && 1415 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1416 so, m, optp, optlen, &opti)) 1417 m = NULL; 1418 } 1419 goto drop; 1420 } 1421 } 1422 1423 after_listen: 1424 #ifdef DIAGNOSTIC 1425 /* 1426 * Should not happen now that all embryonic connections 1427 * are handled with compressed state. 1428 */ 1429 if (tp->t_state == TCPS_LISTEN) 1430 panic("tcp_input: TCPS_LISTEN"); 1431 #endif 1432 1433 /* 1434 * Segment received on connection. 1435 * Reset idle time and keep-alive timer. 1436 */ 1437 tp->t_rcvtime = tcp_now; 1438 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1439 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1440 1441 /* 1442 * Process options. 1443 */ 1444 if (optp) 1445 tcp_dooptions(tp, optp, optlen, th, &opti); 1446 1447 /* 1448 * Header prediction: check for the two common cases 1449 * of a uni-directional data xfer. If the packet has 1450 * no control flags, is in-sequence, the window didn't 1451 * change and we're not retransmitting, it's a 1452 * candidate. If the length is zero and the ack moved 1453 * forward, we're the sender side of the xfer. Just 1454 * free the data acked & wake any higher level process 1455 * that was blocked waiting for space. If the length 1456 * is non-zero and the ack didn't move, we're the 1457 * receiver side. If we're getting packets in-order 1458 * (the reassembly queue is empty), add the data to 1459 * the socket buffer and note that we need a delayed ack. 1460 */ 1461 if (tp->t_state == TCPS_ESTABLISHED && 1462 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1463 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1464 th->th_seq == tp->rcv_nxt && 1465 tiwin && tiwin == tp->snd_wnd && 1466 tp->snd_nxt == tp->snd_max) { 1467 1468 /* 1469 * If last ACK falls within this segment's sequence numbers, 1470 * record the timestamp. 1471 */ 1472 if (opti.ts_present && 1473 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1474 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1475 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1476 tp->ts_recent = opti.ts_val; 1477 } 1478 1479 if (tlen == 0) { 1480 if (SEQ_GT(th->th_ack, tp->snd_una) && 1481 SEQ_LEQ(th->th_ack, tp->snd_max) && 1482 tp->snd_cwnd >= tp->snd_wnd && 1483 tp->t_dupacks < tcprexmtthresh) { 1484 /* 1485 * this is a pure ack for outstanding data. 1486 */ 1487 ++tcpstat.tcps_predack; 1488 if (opti.ts_present && opti.ts_ecr) 1489 tcp_xmit_timer(tp, 1490 TCP_TIMESTAMP(tp) - opti.ts_ecr + 1); 1491 else if (tp->t_rtttime && 1492 SEQ_GT(th->th_ack, tp->t_rtseq)) 1493 tcp_xmit_timer(tp, 1494 tcp_now - tp->t_rtttime); 1495 acked = th->th_ack - tp->snd_una; 1496 tcpstat.tcps_rcvackpack++; 1497 tcpstat.tcps_rcvackbyte += acked; 1498 ND6_HINT(tp); 1499 1500 if (acked > (tp->t_lastoff - tp->t_inoff)) 1501 tp->t_lastm = NULL; 1502 sbdrop(&so->so_snd, acked); 1503 tp->t_lastoff -= acked; 1504 1505 /* 1506 * We want snd_recover to track snd_una to 1507 * avoid sequence wraparound problems for 1508 * very large transfers. 1509 */ 1510 tp->snd_una = tp->snd_recover = th->th_ack; 1511 m_freem(m); 1512 1513 /* 1514 * If all outstanding data are acked, stop 1515 * retransmit timer, otherwise restart timer 1516 * using current (possibly backed-off) value. 1517 * If process is waiting for space, 1518 * wakeup/selwakeup/signal. If data 1519 * are ready to send, let tcp_output 1520 * decide between more output or persist. 1521 */ 1522 if (tp->snd_una == tp->snd_max) 1523 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1524 else if (TCP_TIMER_ISARMED(tp, 1525 TCPT_PERSIST) == 0) 1526 TCP_TIMER_ARM(tp, TCPT_REXMT, 1527 tp->t_rxtcur); 1528 1529 sowwakeup(so); 1530 if (so->so_snd.sb_cc) 1531 (void) tcp_output(tp); 1532 if (tcp_saveti) 1533 m_freem(tcp_saveti); 1534 return; 1535 } 1536 } else if (th->th_ack == tp->snd_una && 1537 TAILQ_FIRST(&tp->segq) == NULL && 1538 tlen <= sbspace(&so->so_rcv)) { 1539 /* 1540 * this is a pure, in-sequence data packet 1541 * with nothing on the reassembly queue and 1542 * we have enough buffer space to take it. 1543 */ 1544 ++tcpstat.tcps_preddat; 1545 tp->rcv_nxt += tlen; 1546 tcpstat.tcps_rcvpack++; 1547 tcpstat.tcps_rcvbyte += tlen; 1548 ND6_HINT(tp); 1549 /* 1550 * Drop TCP, IP headers and TCP options then add data 1551 * to socket buffer. 1552 */ 1553 if (so->so_state & SS_CANTRCVMORE) 1554 m_freem(m); 1555 else { 1556 m_adj(m, toff + off); 1557 sbappendstream(&so->so_rcv, m); 1558 } 1559 sorwakeup(so); 1560 TCP_SETUP_ACK(tp, th); 1561 if (tp->t_flags & TF_ACKNOW) 1562 (void) tcp_output(tp); 1563 if (tcp_saveti) 1564 m_freem(tcp_saveti); 1565 return; 1566 } 1567 } 1568 1569 /* 1570 * Compute mbuf offset to TCP data segment. 1571 */ 1572 hdroptlen = toff + off; 1573 1574 /* 1575 * Calculate amount of space in receive window, 1576 * and then do TCP input processing. 1577 * Receive window is amount of space in rcv queue, 1578 * but not less than advertised window. 1579 */ 1580 { int win; 1581 1582 win = sbspace(&so->so_rcv); 1583 if (win < 0) 1584 win = 0; 1585 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1586 } 1587 1588 switch (tp->t_state) { 1589 case TCPS_LISTEN: 1590 /* 1591 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1592 */ 1593 if (m->m_flags & (M_BCAST|M_MCAST)) 1594 goto drop; 1595 switch (af) { 1596 #ifdef INET6 1597 case AF_INET6: 1598 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1599 goto drop; 1600 break; 1601 #endif /* INET6 */ 1602 case AF_INET: 1603 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1604 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1605 goto drop; 1606 break; 1607 } 1608 break; 1609 1610 /* 1611 * If the state is SYN_SENT: 1612 * if seg contains an ACK, but not for our SYN, drop the input. 1613 * if seg contains a RST, then drop the connection. 1614 * if seg does not contain SYN, then drop it. 1615 * Otherwise this is an acceptable SYN segment 1616 * initialize tp->rcv_nxt and tp->irs 1617 * if seg contains ack then advance tp->snd_una 1618 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1619 * arrange for segment to be acked (eventually) 1620 * continue processing rest of data/controls, beginning with URG 1621 */ 1622 case TCPS_SYN_SENT: 1623 if ((tiflags & TH_ACK) && 1624 (SEQ_LEQ(th->th_ack, tp->iss) || 1625 SEQ_GT(th->th_ack, tp->snd_max))) 1626 goto dropwithreset; 1627 if (tiflags & TH_RST) { 1628 if (tiflags & TH_ACK) 1629 tp = tcp_drop(tp, ECONNREFUSED); 1630 goto drop; 1631 } 1632 if ((tiflags & TH_SYN) == 0) 1633 goto drop; 1634 if (tiflags & TH_ACK) { 1635 tp->snd_una = tp->snd_recover = th->th_ack; 1636 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1637 tp->snd_nxt = tp->snd_una; 1638 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1639 } 1640 tp->irs = th->th_seq; 1641 tcp_rcvseqinit(tp); 1642 tp->t_flags |= TF_ACKNOW; 1643 tcp_mss_from_peer(tp, opti.maxseg); 1644 1645 /* 1646 * Initialize the initial congestion window. If we 1647 * had to retransmit the SYN, we must initialize cwnd 1648 * to 1 segment (i.e. the Loss Window). 1649 */ 1650 if (tp->t_flags & TF_SYN_REXMT) 1651 tp->snd_cwnd = tp->t_peermss; 1652 else { 1653 int ss = tcp_init_win; 1654 #ifdef INET 1655 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1656 ss = tcp_init_win_local; 1657 #endif 1658 #ifdef INET6 1659 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1660 ss = tcp_init_win_local; 1661 #endif 1662 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1663 } 1664 1665 tcp_rmx_rtt(tp); 1666 if (tiflags & TH_ACK) { 1667 tcpstat.tcps_connects++; 1668 soisconnected(so); 1669 tcp_established(tp); 1670 /* Do window scaling on this connection? */ 1671 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1672 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1673 tp->snd_scale = tp->requested_s_scale; 1674 tp->rcv_scale = tp->request_r_scale; 1675 } 1676 TCP_REASS_LOCK(tp); 1677 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1678 TCP_REASS_UNLOCK(tp); 1679 /* 1680 * if we didn't have to retransmit the SYN, 1681 * use its rtt as our initial srtt & rtt var. 1682 */ 1683 if (tp->t_rtttime) 1684 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1685 } else 1686 tp->t_state = TCPS_SYN_RECEIVED; 1687 1688 /* 1689 * Advance th->th_seq to correspond to first data byte. 1690 * If data, trim to stay within window, 1691 * dropping FIN if necessary. 1692 */ 1693 th->th_seq++; 1694 if (tlen > tp->rcv_wnd) { 1695 todrop = tlen - tp->rcv_wnd; 1696 m_adj(m, -todrop); 1697 tlen = tp->rcv_wnd; 1698 tiflags &= ~TH_FIN; 1699 tcpstat.tcps_rcvpackafterwin++; 1700 tcpstat.tcps_rcvbyteafterwin += todrop; 1701 } 1702 tp->snd_wl1 = th->th_seq - 1; 1703 tp->rcv_up = th->th_seq; 1704 goto step6; 1705 1706 /* 1707 * If the state is SYN_RECEIVED: 1708 * If seg contains an ACK, but not for our SYN, drop the input 1709 * and generate an RST. See page 36, rfc793 1710 */ 1711 case TCPS_SYN_RECEIVED: 1712 if ((tiflags & TH_ACK) && 1713 (SEQ_LEQ(th->th_ack, tp->iss) || 1714 SEQ_GT(th->th_ack, tp->snd_max))) 1715 goto dropwithreset; 1716 break; 1717 } 1718 1719 /* 1720 * States other than LISTEN or SYN_SENT. 1721 * First check timestamp, if present. 1722 * Then check that at least some bytes of segment are within 1723 * receive window. If segment begins before rcv_nxt, 1724 * drop leading data (and SYN); if nothing left, just ack. 1725 * 1726 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1727 * and it's less than ts_recent, drop it. 1728 */ 1729 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1730 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1731 1732 /* Check to see if ts_recent is over 24 days old. */ 1733 if ((int)(TCP_TIMESTAMP(tp) - tp->ts_recent_age) > 1734 TCP_PAWS_IDLE) { 1735 /* 1736 * Invalidate ts_recent. If this segment updates 1737 * ts_recent, the age will be reset later and ts_recent 1738 * will get a valid value. If it does not, setting 1739 * ts_recent to zero will at least satisfy the 1740 * requirement that zero be placed in the timestamp 1741 * echo reply when ts_recent isn't valid. The 1742 * age isn't reset until we get a valid ts_recent 1743 * because we don't want out-of-order segments to be 1744 * dropped when ts_recent is old. 1745 */ 1746 tp->ts_recent = 0; 1747 } else { 1748 tcpstat.tcps_rcvduppack++; 1749 tcpstat.tcps_rcvdupbyte += tlen; 1750 tcpstat.tcps_pawsdrop++; 1751 goto dropafterack; 1752 } 1753 } 1754 1755 todrop = tp->rcv_nxt - th->th_seq; 1756 if (todrop > 0) { 1757 if (tiflags & TH_SYN) { 1758 tiflags &= ~TH_SYN; 1759 th->th_seq++; 1760 if (th->th_urp > 1) 1761 th->th_urp--; 1762 else { 1763 tiflags &= ~TH_URG; 1764 th->th_urp = 0; 1765 } 1766 todrop--; 1767 } 1768 if (todrop > tlen || 1769 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1770 /* 1771 * Any valid FIN or RST must be to the left of the 1772 * window. At this point the FIN or RST must be a 1773 * duplicate or out of sequence; drop it. 1774 */ 1775 if (tiflags & TH_RST) 1776 goto drop; 1777 tiflags &= ~(TH_FIN|TH_RST); 1778 /* 1779 * Send an ACK to resynchronize and drop any data. 1780 * But keep on processing for RST or ACK. 1781 */ 1782 tp->t_flags |= TF_ACKNOW; 1783 todrop = tlen; 1784 tcpstat.tcps_rcvdupbyte += todrop; 1785 tcpstat.tcps_rcvduppack++; 1786 } else { 1787 tcpstat.tcps_rcvpartduppack++; 1788 tcpstat.tcps_rcvpartdupbyte += todrop; 1789 } 1790 hdroptlen += todrop; /*drop from head afterwards*/ 1791 th->th_seq += todrop; 1792 tlen -= todrop; 1793 if (th->th_urp > todrop) 1794 th->th_urp -= todrop; 1795 else { 1796 tiflags &= ~TH_URG; 1797 th->th_urp = 0; 1798 } 1799 } 1800 1801 /* 1802 * If new data are received on a connection after the 1803 * user processes are gone, then RST the other end. 1804 */ 1805 if ((so->so_state & SS_NOFDREF) && 1806 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1807 tp = tcp_close(tp); 1808 tcpstat.tcps_rcvafterclose++; 1809 goto dropwithreset; 1810 } 1811 1812 /* 1813 * If segment ends after window, drop trailing data 1814 * (and PUSH and FIN); if nothing left, just ACK. 1815 */ 1816 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1817 if (todrop > 0) { 1818 tcpstat.tcps_rcvpackafterwin++; 1819 if (todrop >= tlen) { 1820 /* 1821 * The segment actually starts after the window. 1822 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 1823 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 1824 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 1825 */ 1826 tcpstat.tcps_rcvbyteafterwin += tlen; 1827 /* 1828 * If a new connection request is received 1829 * while in TIME_WAIT, drop the old connection 1830 * and start over if the sequence numbers 1831 * are above the previous ones. 1832 * 1833 * NOTE: We will checksum the packet again, and 1834 * so we need to put the header fields back into 1835 * network order! 1836 * XXX This kind of sucks, but we don't expect 1837 * XXX this to happen very often, so maybe it 1838 * XXX doesn't matter so much. 1839 */ 1840 if (tiflags & TH_SYN && 1841 tp->t_state == TCPS_TIME_WAIT && 1842 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1843 iss = tcp_new_iss(tp, tp->snd_nxt); 1844 tp = tcp_close(tp); 1845 TCP_FIELDS_TO_NET(th); 1846 goto findpcb; 1847 } 1848 /* 1849 * If window is closed can only take segments at 1850 * window edge, and have to drop data and PUSH from 1851 * incoming segments. Continue processing, but 1852 * remember to ack. Otherwise, drop segment 1853 * and (if not RST) ack. 1854 */ 1855 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1856 tp->t_flags |= TF_ACKNOW; 1857 tcpstat.tcps_rcvwinprobe++; 1858 } else 1859 goto dropafterack; 1860 } else 1861 tcpstat.tcps_rcvbyteafterwin += todrop; 1862 m_adj(m, -todrop); 1863 tlen -= todrop; 1864 tiflags &= ~(TH_PUSH|TH_FIN); 1865 } 1866 1867 /* 1868 * If last ACK falls within this segment's sequence numbers, 1869 * and the timestamp is newer, record it. 1870 */ 1871 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1872 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1873 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1874 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1875 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1876 tp->ts_recent = opti.ts_val; 1877 } 1878 1879 /* 1880 * If the RST bit is set examine the state: 1881 * SYN_RECEIVED STATE: 1882 * If passive open, return to LISTEN state. 1883 * If active open, inform user that connection was refused. 1884 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1885 * Inform user that connection was reset, and close tcb. 1886 * CLOSING, LAST_ACK, TIME_WAIT STATES 1887 * Close the tcb. 1888 */ 1889 if (tiflags & TH_RST) { 1890 if (th->th_seq != tp->last_ack_sent) 1891 goto dropafterack_ratelim; 1892 1893 switch (tp->t_state) { 1894 case TCPS_SYN_RECEIVED: 1895 so->so_error = ECONNREFUSED; 1896 goto close; 1897 1898 case TCPS_ESTABLISHED: 1899 case TCPS_FIN_WAIT_1: 1900 case TCPS_FIN_WAIT_2: 1901 case TCPS_CLOSE_WAIT: 1902 so->so_error = ECONNRESET; 1903 close: 1904 tp->t_state = TCPS_CLOSED; 1905 tcpstat.tcps_drops++; 1906 tp = tcp_close(tp); 1907 goto drop; 1908 1909 case TCPS_CLOSING: 1910 case TCPS_LAST_ACK: 1911 case TCPS_TIME_WAIT: 1912 tp = tcp_close(tp); 1913 goto drop; 1914 } 1915 } 1916 1917 /* 1918 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 1919 * we must be in a synchronized state. RFC791 states (under RST 1920 * generation) that any unacceptable segment (an out-of-order SYN 1921 * qualifies) received in a synchronized state must elicit only an 1922 * empty acknowledgment segment ... and the connection remains in 1923 * the same state. 1924 */ 1925 if (tiflags & TH_SYN) 1926 goto dropafterack_ratelim; 1927 1928 /* 1929 * If the ACK bit is off we drop the segment and return. 1930 */ 1931 if ((tiflags & TH_ACK) == 0) { 1932 if (tp->t_flags & TF_ACKNOW) 1933 goto dropafterack; 1934 else 1935 goto drop; 1936 } 1937 1938 /* 1939 * Ack processing. 1940 */ 1941 switch (tp->t_state) { 1942 1943 /* 1944 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1945 * ESTABLISHED state and continue processing, otherwise 1946 * send an RST. 1947 */ 1948 case TCPS_SYN_RECEIVED: 1949 if (SEQ_GT(tp->snd_una, th->th_ack) || 1950 SEQ_GT(th->th_ack, tp->snd_max)) 1951 goto dropwithreset; 1952 tcpstat.tcps_connects++; 1953 soisconnected(so); 1954 tcp_established(tp); 1955 /* Do window scaling? */ 1956 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1957 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1958 tp->snd_scale = tp->requested_s_scale; 1959 tp->rcv_scale = tp->request_r_scale; 1960 } 1961 TCP_REASS_LOCK(tp); 1962 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1963 TCP_REASS_UNLOCK(tp); 1964 tp->snd_wl1 = th->th_seq - 1; 1965 /* fall into ... */ 1966 1967 /* 1968 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1969 * ACKs. If the ack is in the range 1970 * tp->snd_una < th->th_ack <= tp->snd_max 1971 * then advance tp->snd_una to th->th_ack and drop 1972 * data from the retransmission queue. If this ACK reflects 1973 * more up to date window information we update our window information. 1974 */ 1975 case TCPS_ESTABLISHED: 1976 case TCPS_FIN_WAIT_1: 1977 case TCPS_FIN_WAIT_2: 1978 case TCPS_CLOSE_WAIT: 1979 case TCPS_CLOSING: 1980 case TCPS_LAST_ACK: 1981 case TCPS_TIME_WAIT: 1982 1983 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1984 if (tlen == 0 && tiwin == tp->snd_wnd) { 1985 tcpstat.tcps_rcvdupack++; 1986 /* 1987 * If we have outstanding data (other than 1988 * a window probe), this is a completely 1989 * duplicate ack (ie, window info didn't 1990 * change), the ack is the biggest we've 1991 * seen and we've seen exactly our rexmt 1992 * threshhold of them, assume a packet 1993 * has been dropped and retransmit it. 1994 * Kludge snd_nxt & the congestion 1995 * window so we send only this one 1996 * packet. 1997 * 1998 * We know we're losing at the current 1999 * window size so do congestion avoidance 2000 * (set ssthresh to half the current window 2001 * and pull our congestion window back to 2002 * the new ssthresh). 2003 * 2004 * Dup acks mean that packets have left the 2005 * network (they're now cached at the receiver) 2006 * so bump cwnd by the amount in the receiver 2007 * to keep a constant cwnd packets in the 2008 * network. 2009 */ 2010 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2011 th->th_ack != tp->snd_una) 2012 tp->t_dupacks = 0; 2013 else if (++tp->t_dupacks == tcprexmtthresh) { 2014 tcp_seq onxt = tp->snd_nxt; 2015 u_int win = 2016 min(tp->snd_wnd, tp->snd_cwnd) / 2017 2 / tp->t_segsz; 2018 if (tcp_do_newreno && SEQ_LT(th->th_ack, 2019 tp->snd_recover)) { 2020 /* 2021 * False fast retransmit after 2022 * timeout. Do not cut window. 2023 */ 2024 tp->snd_cwnd += tp->t_segsz; 2025 tp->t_dupacks = 0; 2026 (void) tcp_output(tp); 2027 goto drop; 2028 } 2029 2030 if (win < 2) 2031 win = 2; 2032 tp->snd_ssthresh = win * tp->t_segsz; 2033 tp->snd_recover = tp->snd_max; 2034 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2035 tp->t_rtttime = 0; 2036 tp->snd_nxt = th->th_ack; 2037 tp->snd_cwnd = tp->t_segsz; 2038 (void) tcp_output(tp); 2039 tp->snd_cwnd = tp->snd_ssthresh + 2040 tp->t_segsz * tp->t_dupacks; 2041 if (SEQ_GT(onxt, tp->snd_nxt)) 2042 tp->snd_nxt = onxt; 2043 goto drop; 2044 } else if (tp->t_dupacks > tcprexmtthresh) { 2045 tp->snd_cwnd += tp->t_segsz; 2046 (void) tcp_output(tp); 2047 goto drop; 2048 } 2049 } else if (tlen) { 2050 tp->t_dupacks = 0; /*XXX*/ 2051 /* drop very old ACKs unless th_seq matches */ 2052 if (th->th_seq != tp->rcv_nxt && 2053 SEQ_LT(th->th_ack, 2054 tp->snd_una - tp->max_sndwnd)) { 2055 goto drop; 2056 } 2057 break; 2058 } else 2059 tp->t_dupacks = 0; 2060 break; 2061 } 2062 /* 2063 * If the congestion window was inflated to account 2064 * for the other side's cached packets, retract it. 2065 */ 2066 if (tcp_do_newreno == 0) { 2067 if (tp->t_dupacks >= tcprexmtthresh && 2068 tp->snd_cwnd > tp->snd_ssthresh) 2069 tp->snd_cwnd = tp->snd_ssthresh; 2070 tp->t_dupacks = 0; 2071 } else if (tp->t_dupacks >= tcprexmtthresh && 2072 tcp_newreno(tp, th) == 0) { 2073 tp->snd_cwnd = tp->snd_ssthresh; 2074 /* 2075 * Window inflation should have left us with approx. 2076 * snd_ssthresh outstanding data. But in case we 2077 * would be inclined to send a burst, better to do 2078 * it via the slow start mechanism. 2079 */ 2080 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 2081 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 2082 + tp->t_segsz; 2083 tp->t_dupacks = 0; 2084 } 2085 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2086 tcpstat.tcps_rcvacktoomuch++; 2087 goto dropafterack; 2088 } 2089 acked = th->th_ack - tp->snd_una; 2090 tcpstat.tcps_rcvackpack++; 2091 tcpstat.tcps_rcvackbyte += acked; 2092 2093 /* 2094 * If we have a timestamp reply, update smoothed 2095 * round trip time. If no timestamp is present but 2096 * transmit timer is running and timed sequence 2097 * number was acked, update smoothed round trip time. 2098 * Since we now have an rtt measurement, cancel the 2099 * timer backoff (cf., Phil Karn's retransmit alg.). 2100 * Recompute the initial retransmit timer. 2101 */ 2102 if (opti.ts_present && opti.ts_ecr) 2103 tcp_xmit_timer(tp, TCP_TIMESTAMP(tp) - opti.ts_ecr + 1); 2104 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2105 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2106 2107 /* 2108 * If all outstanding data is acked, stop retransmit 2109 * timer and remember to restart (more output or persist). 2110 * If there is more data to be acked, restart retransmit 2111 * timer, using current (possibly backed-off) value. 2112 */ 2113 if (th->th_ack == tp->snd_max) { 2114 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2115 needoutput = 1; 2116 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2117 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2118 /* 2119 * When new data is acked, open the congestion window. 2120 * If the window gives us less than ssthresh packets 2121 * in flight, open exponentially (segsz per packet). 2122 * Otherwise open linearly: segsz per window 2123 * (segsz^2 / cwnd per packet), plus a constant 2124 * fraction of a packet (segsz/8) to help larger windows 2125 * open quickly enough. 2126 */ 2127 { 2128 u_int cw = tp->snd_cwnd; 2129 u_int incr = tp->t_segsz; 2130 2131 if (cw > tp->snd_ssthresh) 2132 incr = incr * incr / cw; 2133 if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover)) 2134 tp->snd_cwnd = min(cw + incr, 2135 TCP_MAXWIN << tp->snd_scale); 2136 } 2137 ND6_HINT(tp); 2138 if (acked > so->so_snd.sb_cc) { 2139 tp->snd_wnd -= so->so_snd.sb_cc; 2140 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2141 ourfinisacked = 1; 2142 } else { 2143 if (acked > (tp->t_lastoff - tp->t_inoff)) 2144 tp->t_lastm = NULL; 2145 sbdrop(&so->so_snd, acked); 2146 tp->t_lastoff -= acked; 2147 tp->snd_wnd -= acked; 2148 ourfinisacked = 0; 2149 } 2150 sowwakeup(so); 2151 /* 2152 * We want snd_recover to track snd_una to 2153 * avoid sequence wraparound problems for 2154 * very large transfers. 2155 */ 2156 tp->snd_una = tp->snd_recover = th->th_ack; 2157 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2158 tp->snd_nxt = tp->snd_una; 2159 2160 switch (tp->t_state) { 2161 2162 /* 2163 * In FIN_WAIT_1 STATE in addition to the processing 2164 * for the ESTABLISHED state if our FIN is now acknowledged 2165 * then enter FIN_WAIT_2. 2166 */ 2167 case TCPS_FIN_WAIT_1: 2168 if (ourfinisacked) { 2169 /* 2170 * If we can't receive any more 2171 * data, then closing user can proceed. 2172 * Starting the timer is contrary to the 2173 * specification, but if we don't get a FIN 2174 * we'll hang forever. 2175 */ 2176 if (so->so_state & SS_CANTRCVMORE) { 2177 soisdisconnected(so); 2178 if (tcp_maxidle > 0) 2179 TCP_TIMER_ARM(tp, TCPT_2MSL, 2180 tcp_maxidle); 2181 } 2182 tp->t_state = TCPS_FIN_WAIT_2; 2183 } 2184 break; 2185 2186 /* 2187 * In CLOSING STATE in addition to the processing for 2188 * the ESTABLISHED state if the ACK acknowledges our FIN 2189 * then enter the TIME-WAIT state, otherwise ignore 2190 * the segment. 2191 */ 2192 case TCPS_CLOSING: 2193 if (ourfinisacked) { 2194 tp->t_state = TCPS_TIME_WAIT; 2195 tcp_canceltimers(tp); 2196 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2197 soisdisconnected(so); 2198 } 2199 break; 2200 2201 /* 2202 * In LAST_ACK, we may still be waiting for data to drain 2203 * and/or to be acked, as well as for the ack of our FIN. 2204 * If our FIN is now acknowledged, delete the TCB, 2205 * enter the closed state and return. 2206 */ 2207 case TCPS_LAST_ACK: 2208 if (ourfinisacked) { 2209 tp = tcp_close(tp); 2210 goto drop; 2211 } 2212 break; 2213 2214 /* 2215 * In TIME_WAIT state the only thing that should arrive 2216 * is a retransmission of the remote FIN. Acknowledge 2217 * it and restart the finack timer. 2218 */ 2219 case TCPS_TIME_WAIT: 2220 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2221 goto dropafterack; 2222 } 2223 } 2224 2225 step6: 2226 /* 2227 * Update window information. 2228 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2229 */ 2230 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2231 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 2232 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 2233 /* keep track of pure window updates */ 2234 if (tlen == 0 && 2235 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2236 tcpstat.tcps_rcvwinupd++; 2237 tp->snd_wnd = tiwin; 2238 tp->snd_wl1 = th->th_seq; 2239 tp->snd_wl2 = th->th_ack; 2240 if (tp->snd_wnd > tp->max_sndwnd) 2241 tp->max_sndwnd = tp->snd_wnd; 2242 needoutput = 1; 2243 } 2244 2245 /* 2246 * Process segments with URG. 2247 */ 2248 if ((tiflags & TH_URG) && th->th_urp && 2249 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2250 /* 2251 * This is a kludge, but if we receive and accept 2252 * random urgent pointers, we'll crash in 2253 * soreceive. It's hard to imagine someone 2254 * actually wanting to send this much urgent data. 2255 */ 2256 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2257 th->th_urp = 0; /* XXX */ 2258 tiflags &= ~TH_URG; /* XXX */ 2259 goto dodata; /* XXX */ 2260 } 2261 /* 2262 * If this segment advances the known urgent pointer, 2263 * then mark the data stream. This should not happen 2264 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2265 * a FIN has been received from the remote side. 2266 * In these states we ignore the URG. 2267 * 2268 * According to RFC961 (Assigned Protocols), 2269 * the urgent pointer points to the last octet 2270 * of urgent data. We continue, however, 2271 * to consider it to indicate the first octet 2272 * of data past the urgent section as the original 2273 * spec states (in one of two places). 2274 */ 2275 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2276 tp->rcv_up = th->th_seq + th->th_urp; 2277 so->so_oobmark = so->so_rcv.sb_cc + 2278 (tp->rcv_up - tp->rcv_nxt) - 1; 2279 if (so->so_oobmark == 0) 2280 so->so_state |= SS_RCVATMARK; 2281 sohasoutofband(so); 2282 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2283 } 2284 /* 2285 * Remove out of band data so doesn't get presented to user. 2286 * This can happen independent of advancing the URG pointer, 2287 * but if two URG's are pending at once, some out-of-band 2288 * data may creep in... ick. 2289 */ 2290 if (th->th_urp <= (u_int16_t) tlen 2291 #ifdef SO_OOBINLINE 2292 && (so->so_options & SO_OOBINLINE) == 0 2293 #endif 2294 ) 2295 tcp_pulloutofband(so, th, m, hdroptlen); 2296 } else 2297 /* 2298 * If no out of band data is expected, 2299 * pull receive urgent pointer along 2300 * with the receive window. 2301 */ 2302 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2303 tp->rcv_up = tp->rcv_nxt; 2304 dodata: /* XXX */ 2305 2306 /* 2307 * Process the segment text, merging it into the TCP sequencing queue, 2308 * and arranging for acknowledgement of receipt if necessary. 2309 * This process logically involves adjusting tp->rcv_wnd as data 2310 * is presented to the user (this happens in tcp_usrreq.c, 2311 * case PRU_RCVD). If a FIN has already been received on this 2312 * connection then we just ignore the text. 2313 */ 2314 if ((tlen || (tiflags & TH_FIN)) && 2315 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2316 /* 2317 * Insert segment ti into reassembly queue of tcp with 2318 * control block tp. Return TH_FIN if reassembly now includes 2319 * a segment with FIN. The macro form does the common case 2320 * inline (segment is the next to be received on an 2321 * established connection, and the queue is empty), 2322 * avoiding linkage into and removal from the queue and 2323 * repetition of various conversions. 2324 * Set DELACK for segments received in order, but ack 2325 * immediately when segments are out of order 2326 * (so fast retransmit can work). 2327 */ 2328 /* NOTE: this was TCP_REASS() macro, but used only once */ 2329 TCP_REASS_LOCK(tp); 2330 if (th->th_seq == tp->rcv_nxt && 2331 TAILQ_FIRST(&tp->segq) == NULL && 2332 tp->t_state == TCPS_ESTABLISHED) { 2333 TCP_SETUP_ACK(tp, th); 2334 tp->rcv_nxt += tlen; 2335 tiflags = th->th_flags & TH_FIN; 2336 tcpstat.tcps_rcvpack++; 2337 tcpstat.tcps_rcvbyte += tlen; 2338 ND6_HINT(tp); 2339 if (so->so_state & SS_CANTRCVMORE) 2340 m_freem(m); 2341 else { 2342 m_adj(m, hdroptlen); 2343 sbappendstream(&(so)->so_rcv, m); 2344 } 2345 sorwakeup(so); 2346 } else { 2347 m_adj(m, hdroptlen); 2348 tiflags = tcp_reass(tp, th, m, &tlen); 2349 tp->t_flags |= TF_ACKNOW; 2350 } 2351 TCP_REASS_UNLOCK(tp); 2352 2353 /* 2354 * Note the amount of data that peer has sent into 2355 * our window, in order to estimate the sender's 2356 * buffer size. 2357 */ 2358 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2359 } else { 2360 m_freem(m); 2361 m = NULL; 2362 tiflags &= ~TH_FIN; 2363 } 2364 2365 /* 2366 * If FIN is received ACK the FIN and let the user know 2367 * that the connection is closing. Ignore a FIN received before 2368 * the connection is fully established. 2369 */ 2370 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2371 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2372 socantrcvmore(so); 2373 tp->t_flags |= TF_ACKNOW; 2374 tp->rcv_nxt++; 2375 } 2376 switch (tp->t_state) { 2377 2378 /* 2379 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2380 */ 2381 case TCPS_ESTABLISHED: 2382 tp->t_state = TCPS_CLOSE_WAIT; 2383 break; 2384 2385 /* 2386 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2387 * enter the CLOSING state. 2388 */ 2389 case TCPS_FIN_WAIT_1: 2390 tp->t_state = TCPS_CLOSING; 2391 break; 2392 2393 /* 2394 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2395 * starting the time-wait timer, turning off the other 2396 * standard timers. 2397 */ 2398 case TCPS_FIN_WAIT_2: 2399 tp->t_state = TCPS_TIME_WAIT; 2400 tcp_canceltimers(tp); 2401 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2402 soisdisconnected(so); 2403 break; 2404 2405 /* 2406 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2407 */ 2408 case TCPS_TIME_WAIT: 2409 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2410 break; 2411 } 2412 } 2413 #ifdef TCP_DEBUG 2414 if (so->so_options & SO_DEBUG) 2415 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2416 #endif 2417 2418 /* 2419 * Return any desired output. 2420 */ 2421 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2422 (void) tcp_output(tp); 2423 if (tcp_saveti) 2424 m_freem(tcp_saveti); 2425 return; 2426 2427 badsyn: 2428 /* 2429 * Received a bad SYN. Increment counters and dropwithreset. 2430 */ 2431 tcpstat.tcps_badsyn++; 2432 tp = NULL; 2433 goto dropwithreset; 2434 2435 dropafterack: 2436 /* 2437 * Generate an ACK dropping incoming segment if it occupies 2438 * sequence space, where the ACK reflects our state. 2439 */ 2440 if (tiflags & TH_RST) 2441 goto drop; 2442 goto dropafterack2; 2443 2444 dropafterack_ratelim: 2445 /* 2446 * We may want to rate-limit ACKs against SYN/RST attack. 2447 */ 2448 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2449 tcp_ackdrop_ppslim) == 0) { 2450 /* XXX stat */ 2451 goto drop; 2452 } 2453 /* ...fall into dropafterack2... */ 2454 2455 dropafterack2: 2456 m_freem(m); 2457 tp->t_flags |= TF_ACKNOW; 2458 (void) tcp_output(tp); 2459 if (tcp_saveti) 2460 m_freem(tcp_saveti); 2461 return; 2462 2463 dropwithreset_ratelim: 2464 /* 2465 * We may want to rate-limit RSTs in certain situations, 2466 * particularly if we are sending an RST in response to 2467 * an attempt to connect to or otherwise communicate with 2468 * a port for which we have no socket. 2469 */ 2470 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2471 tcp_rst_ppslim) == 0) { 2472 /* XXX stat */ 2473 goto drop; 2474 } 2475 /* ...fall into dropwithreset... */ 2476 2477 dropwithreset: 2478 /* 2479 * Generate a RST, dropping incoming segment. 2480 * Make ACK acceptable to originator of segment. 2481 */ 2482 if (tiflags & TH_RST) 2483 goto drop; 2484 2485 switch (af) { 2486 #ifdef INET6 2487 case AF_INET6: 2488 /* For following calls to tcp_respond */ 2489 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2490 goto drop; 2491 break; 2492 #endif /* INET6 */ 2493 case AF_INET: 2494 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2495 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2496 goto drop; 2497 } 2498 2499 if (tiflags & TH_ACK) 2500 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2501 else { 2502 if (tiflags & TH_SYN) 2503 tlen++; 2504 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2505 TH_RST|TH_ACK); 2506 } 2507 if (tcp_saveti) 2508 m_freem(tcp_saveti); 2509 return; 2510 2511 badcsum: 2512 tcpstat.tcps_rcvbadsum++; 2513 drop: 2514 /* 2515 * Drop space held by incoming segment and return. 2516 */ 2517 if (tp) { 2518 if (tp->t_inpcb) 2519 so = tp->t_inpcb->inp_socket; 2520 #ifdef INET6 2521 else if (tp->t_in6pcb) 2522 so = tp->t_in6pcb->in6p_socket; 2523 #endif 2524 else 2525 so = NULL; 2526 #ifdef TCP_DEBUG 2527 if (so && (so->so_options & SO_DEBUG) != 0) 2528 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2529 #endif 2530 } 2531 if (tcp_saveti) 2532 m_freem(tcp_saveti); 2533 m_freem(m); 2534 return; 2535 } 2536 2537 void 2538 tcp_dooptions(tp, cp, cnt, th, oi) 2539 struct tcpcb *tp; 2540 u_char *cp; 2541 int cnt; 2542 struct tcphdr *th; 2543 struct tcp_opt_info *oi; 2544 { 2545 u_int16_t mss; 2546 int opt, optlen; 2547 2548 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2549 opt = cp[0]; 2550 if (opt == TCPOPT_EOL) 2551 break; 2552 if (opt == TCPOPT_NOP) 2553 optlen = 1; 2554 else { 2555 if (cnt < 2) 2556 break; 2557 optlen = cp[1]; 2558 if (optlen < 2 || optlen > cnt) 2559 break; 2560 } 2561 switch (opt) { 2562 2563 default: 2564 continue; 2565 2566 case TCPOPT_MAXSEG: 2567 if (optlen != TCPOLEN_MAXSEG) 2568 continue; 2569 if (!(th->th_flags & TH_SYN)) 2570 continue; 2571 bcopy(cp + 2, &mss, sizeof(mss)); 2572 oi->maxseg = ntohs(mss); 2573 break; 2574 2575 case TCPOPT_WINDOW: 2576 if (optlen != TCPOLEN_WINDOW) 2577 continue; 2578 if (!(th->th_flags & TH_SYN)) 2579 continue; 2580 tp->t_flags |= TF_RCVD_SCALE; 2581 tp->requested_s_scale = cp[2]; 2582 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2583 #if 0 /*XXX*/ 2584 char *p; 2585 2586 if (ip) 2587 p = ntohl(ip->ip_src); 2588 #ifdef INET6 2589 else if (ip6) 2590 p = ip6_sprintf(&ip6->ip6_src); 2591 #endif 2592 else 2593 p = "(unknown)"; 2594 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2595 "assuming %d\n", 2596 tp->requested_s_scale, p, 2597 TCP_MAX_WINSHIFT); 2598 #else 2599 log(LOG_ERR, "TCP: invalid wscale %d, " 2600 "assuming %d\n", 2601 tp->requested_s_scale, 2602 TCP_MAX_WINSHIFT); 2603 #endif 2604 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2605 } 2606 break; 2607 2608 case TCPOPT_TIMESTAMP: 2609 if (optlen != TCPOLEN_TIMESTAMP) 2610 continue; 2611 oi->ts_present = 1; 2612 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2613 NTOHL(oi->ts_val); 2614 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2615 NTOHL(oi->ts_ecr); 2616 2617 /* 2618 * A timestamp received in a SYN makes 2619 * it ok to send timestamp requests and replies. 2620 */ 2621 if (th->th_flags & TH_SYN) { 2622 tp->t_flags |= TF_RCVD_TSTMP; 2623 tp->ts_recent = oi->ts_val; 2624 tp->ts_recent_age = TCP_TIMESTAMP(tp); 2625 } 2626 break; 2627 case TCPOPT_SACK_PERMITTED: 2628 if (optlen != TCPOLEN_SACK_PERMITTED) 2629 continue; 2630 if (!(th->th_flags & TH_SYN)) 2631 continue; 2632 tp->t_flags &= ~TF_CANT_TXSACK; 2633 break; 2634 2635 case TCPOPT_SACK: 2636 if (tp->t_flags & TF_IGNR_RXSACK) 2637 continue; 2638 if (optlen % 8 != 2 || optlen < 10) 2639 continue; 2640 cp += 2; 2641 optlen -= 2; 2642 for (; optlen > 0; cp -= 8, optlen -= 8) { 2643 tcp_seq lwe, rwe; 2644 bcopy((char *)cp, (char *) &lwe, sizeof(lwe)); 2645 NTOHL(lwe); 2646 bcopy((char *)cp, (char *) &rwe, sizeof(rwe)); 2647 NTOHL(rwe); 2648 /* tcp_mark_sacked(tp, lwe, rwe); */ 2649 } 2650 break; 2651 } 2652 } 2653 } 2654 2655 /* 2656 * Pull out of band byte out of a segment so 2657 * it doesn't appear in the user's data queue. 2658 * It is still reflected in the segment length for 2659 * sequencing purposes. 2660 */ 2661 void 2662 tcp_pulloutofband(so, th, m, off) 2663 struct socket *so; 2664 struct tcphdr *th; 2665 struct mbuf *m; 2666 int off; 2667 { 2668 int cnt = off + th->th_urp - 1; 2669 2670 while (cnt >= 0) { 2671 if (m->m_len > cnt) { 2672 char *cp = mtod(m, caddr_t) + cnt; 2673 struct tcpcb *tp = sototcpcb(so); 2674 2675 tp->t_iobc = *cp; 2676 tp->t_oobflags |= TCPOOB_HAVEDATA; 2677 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2678 m->m_len--; 2679 return; 2680 } 2681 cnt -= m->m_len; 2682 m = m->m_next; 2683 if (m == 0) 2684 break; 2685 } 2686 panic("tcp_pulloutofband"); 2687 } 2688 2689 /* 2690 * Collect new round-trip time estimate 2691 * and update averages and current timeout. 2692 */ 2693 void 2694 tcp_xmit_timer(tp, rtt) 2695 struct tcpcb *tp; 2696 uint32_t rtt; 2697 { 2698 int32_t delta; 2699 2700 tcpstat.tcps_rttupdated++; 2701 if (tp->t_srtt != 0) { 2702 /* 2703 * srtt is stored as fixed point with 3 bits after the 2704 * binary point (i.e., scaled by 8). The following magic 2705 * is equivalent to the smoothing algorithm in rfc793 with 2706 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2707 * point). Adjust rtt to origin 0. 2708 */ 2709 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2710 if ((tp->t_srtt += delta) <= 0) 2711 tp->t_srtt = 1 << 2; 2712 /* 2713 * We accumulate a smoothed rtt variance (actually, a 2714 * smoothed mean difference), then set the retransmit 2715 * timer to smoothed rtt + 4 times the smoothed variance. 2716 * rttvar is stored as fixed point with 2 bits after the 2717 * binary point (scaled by 4). The following is 2718 * equivalent to rfc793 smoothing with an alpha of .75 2719 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2720 * rfc793's wired-in beta. 2721 */ 2722 if (delta < 0) 2723 delta = -delta; 2724 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2725 if ((tp->t_rttvar += delta) <= 0) 2726 tp->t_rttvar = 1 << 2; 2727 } else { 2728 /* 2729 * No rtt measurement yet - use the unsmoothed rtt. 2730 * Set the variance to half the rtt (so our first 2731 * retransmit happens at 3*rtt). 2732 */ 2733 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 2734 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 2735 } 2736 tp->t_rtttime = 0; 2737 tp->t_rxtshift = 0; 2738 2739 /* 2740 * the retransmit should happen at rtt + 4 * rttvar. 2741 * Because of the way we do the smoothing, srtt and rttvar 2742 * will each average +1/2 tick of bias. When we compute 2743 * the retransmit timer, we want 1/2 tick of rounding and 2744 * 1 extra tick because of +-1/2 tick uncertainty in the 2745 * firing of the timer. The bias will give us exactly the 2746 * 1.5 tick we need. But, because the bias is 2747 * statistical, we have to test that we don't drop below 2748 * the minimum feasible timer (which is 2 ticks). 2749 */ 2750 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2751 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2752 2753 /* 2754 * We received an ack for a packet that wasn't retransmitted; 2755 * it is probably safe to discard any error indications we've 2756 * received recently. This isn't quite right, but close enough 2757 * for now (a route might have failed after we sent a segment, 2758 * and the return path might not be symmetrical). 2759 */ 2760 tp->t_softerror = 0; 2761 } 2762 2763 /* 2764 * Checks for partial ack. If partial ack arrives, force the retransmission 2765 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2766 * 1. By setting snd_nxt to th_ack, this forces retransmission timer to 2767 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2768 */ 2769 int 2770 tcp_newreno(tp, th) 2771 struct tcpcb *tp; 2772 struct tcphdr *th; 2773 { 2774 tcp_seq onxt = tp->snd_nxt; 2775 u_long ocwnd = tp->snd_cwnd; 2776 2777 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2778 /* 2779 * snd_una has not yet been updated and the socket's send 2780 * buffer has not yet drained off the ACK'd data, so we 2781 * have to leave snd_una as it was to get the correct data 2782 * offset in tcp_output(). 2783 */ 2784 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2785 tp->t_rtttime = 0; 2786 tp->snd_nxt = th->th_ack; 2787 /* 2788 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 2789 * is not yet updated when we're called. 2790 */ 2791 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 2792 (void) tcp_output(tp); 2793 tp->snd_cwnd = ocwnd; 2794 if (SEQ_GT(onxt, tp->snd_nxt)) 2795 tp->snd_nxt = onxt; 2796 /* 2797 * Partial window deflation. Relies on fact that tp->snd_una 2798 * not updated yet. 2799 */ 2800 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 2801 return 1; 2802 } 2803 return 0; 2804 } 2805 2806 2807 /* 2808 * TCP compressed state engine. Currently used to hold compressed 2809 * state for SYN_RECEIVED. 2810 */ 2811 2812 u_long syn_cache_count; 2813 u_int32_t syn_hash1, syn_hash2; 2814 2815 #define SYN_HASH(sa, sp, dp) \ 2816 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 2817 ((u_int32_t)(sp)))^syn_hash2))) 2818 #ifndef INET6 2819 #define SYN_HASHALL(hash, src, dst) \ 2820 do { \ 2821 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2822 ((struct sockaddr_in *)(src))->sin_port, \ 2823 ((struct sockaddr_in *)(dst))->sin_port); \ 2824 } while (/*CONSTCOND*/ 0) 2825 #else 2826 #define SYN_HASH6(sa, sp, dp) \ 2827 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 2828 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 2829 & 0x7fffffff) 2830 2831 #define SYN_HASHALL(hash, src, dst) \ 2832 do { \ 2833 switch ((src)->sa_family) { \ 2834 case AF_INET: \ 2835 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2836 ((struct sockaddr_in *)(src))->sin_port, \ 2837 ((struct sockaddr_in *)(dst))->sin_port); \ 2838 break; \ 2839 case AF_INET6: \ 2840 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 2841 ((struct sockaddr_in6 *)(src))->sin6_port, \ 2842 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 2843 break; \ 2844 default: \ 2845 hash = 0; \ 2846 } \ 2847 } while (/*CONSTCOND*/0) 2848 #endif /* INET6 */ 2849 2850 #define SYN_CACHE_RM(sc) \ 2851 do { \ 2852 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 2853 (sc), sc_bucketq); \ 2854 (sc)->sc_tp = NULL; \ 2855 LIST_REMOVE((sc), sc_tpq); \ 2856 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 2857 callout_stop(&(sc)->sc_timer); \ 2858 syn_cache_count--; \ 2859 } while (/*CONSTCOND*/0) 2860 2861 #define SYN_CACHE_PUT(sc) \ 2862 do { \ 2863 if ((sc)->sc_ipopts) \ 2864 (void) m_free((sc)->sc_ipopts); \ 2865 if ((sc)->sc_route4.ro_rt != NULL) \ 2866 RTFREE((sc)->sc_route4.ro_rt); \ 2867 if (callout_invoking(&(sc)->sc_timer)) \ 2868 (sc)->sc_flags |= SCF_DEAD; \ 2869 else \ 2870 pool_put(&syn_cache_pool, (sc)); \ 2871 } while (/*CONSTCOND*/0) 2872 2873 struct pool syn_cache_pool; 2874 2875 /* 2876 * We don't estimate RTT with SYNs, so each packet starts with the default 2877 * RTT and each timer step has a fixed timeout value. 2878 */ 2879 #define SYN_CACHE_TIMER_ARM(sc) \ 2880 do { \ 2881 TCPT_RANGESET((sc)->sc_rxtcur, \ 2882 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 2883 TCPTV_REXMTMAX); \ 2884 callout_reset(&(sc)->sc_timer, \ 2885 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 2886 } while (/*CONSTCOND*/0) 2887 2888 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 2889 2890 void 2891 syn_cache_init() 2892 { 2893 int i; 2894 2895 /* Initialize the hash buckets. */ 2896 for (i = 0; i < tcp_syn_cache_size; i++) 2897 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 2898 2899 /* Initialize the syn cache pool. */ 2900 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 2901 "synpl", NULL); 2902 } 2903 2904 void 2905 syn_cache_insert(sc, tp) 2906 struct syn_cache *sc; 2907 struct tcpcb *tp; 2908 { 2909 struct syn_cache_head *scp; 2910 struct syn_cache *sc2; 2911 int s; 2912 2913 /* 2914 * If there are no entries in the hash table, reinitialize 2915 * the hash secrets. 2916 */ 2917 if (syn_cache_count == 0) { 2918 syn_hash1 = arc4random(); 2919 syn_hash2 = arc4random(); 2920 } 2921 2922 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 2923 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 2924 scp = &tcp_syn_cache[sc->sc_bucketidx]; 2925 2926 /* 2927 * Make sure that we don't overflow the per-bucket 2928 * limit or the total cache size limit. 2929 */ 2930 s = splsoftnet(); 2931 if (scp->sch_length >= tcp_syn_bucket_limit) { 2932 tcpstat.tcps_sc_bucketoverflow++; 2933 /* 2934 * The bucket is full. Toss the oldest element in the 2935 * bucket. This will be the first entry in the bucket. 2936 */ 2937 sc2 = TAILQ_FIRST(&scp->sch_bucket); 2938 #ifdef DIAGNOSTIC 2939 /* 2940 * This should never happen; we should always find an 2941 * entry in our bucket. 2942 */ 2943 if (sc2 == NULL) 2944 panic("syn_cache_insert: bucketoverflow: impossible"); 2945 #endif 2946 SYN_CACHE_RM(sc2); 2947 SYN_CACHE_PUT(sc2); 2948 } else if (syn_cache_count >= tcp_syn_cache_limit) { 2949 struct syn_cache_head *scp2, *sce; 2950 2951 tcpstat.tcps_sc_overflowed++; 2952 /* 2953 * The cache is full. Toss the oldest entry in the 2954 * first non-empty bucket we can find. 2955 * 2956 * XXX We would really like to toss the oldest 2957 * entry in the cache, but we hope that this 2958 * condition doesn't happen very often. 2959 */ 2960 scp2 = scp; 2961 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 2962 sce = &tcp_syn_cache[tcp_syn_cache_size]; 2963 for (++scp2; scp2 != scp; scp2++) { 2964 if (scp2 >= sce) 2965 scp2 = &tcp_syn_cache[0]; 2966 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 2967 break; 2968 } 2969 #ifdef DIAGNOSTIC 2970 /* 2971 * This should never happen; we should always find a 2972 * non-empty bucket. 2973 */ 2974 if (scp2 == scp) 2975 panic("syn_cache_insert: cacheoverflow: " 2976 "impossible"); 2977 #endif 2978 } 2979 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 2980 SYN_CACHE_RM(sc2); 2981 SYN_CACHE_PUT(sc2); 2982 } 2983 2984 /* 2985 * Initialize the entry's timer. 2986 */ 2987 sc->sc_rxttot = 0; 2988 sc->sc_rxtshift = 0; 2989 SYN_CACHE_TIMER_ARM(sc); 2990 2991 /* Link it from tcpcb entry */ 2992 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 2993 2994 /* Put it into the bucket. */ 2995 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 2996 scp->sch_length++; 2997 syn_cache_count++; 2998 2999 tcpstat.tcps_sc_added++; 3000 splx(s); 3001 } 3002 3003 /* 3004 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3005 * If we have retransmitted an entry the maximum number of times, expire 3006 * that entry. 3007 */ 3008 void 3009 syn_cache_timer(void *arg) 3010 { 3011 struct syn_cache *sc = arg; 3012 int s; 3013 3014 s = splsoftnet(); 3015 callout_ack(&sc->sc_timer); 3016 3017 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3018 tcpstat.tcps_sc_delayed_free++; 3019 pool_put(&syn_cache_pool, sc); 3020 splx(s); 3021 return; 3022 } 3023 3024 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3025 /* Drop it -- too many retransmissions. */ 3026 goto dropit; 3027 } 3028 3029 /* 3030 * Compute the total amount of time this entry has 3031 * been on a queue. If this entry has been on longer 3032 * than the keep alive timer would allow, expire it. 3033 */ 3034 sc->sc_rxttot += sc->sc_rxtcur; 3035 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) 3036 goto dropit; 3037 3038 tcpstat.tcps_sc_retransmitted++; 3039 (void) syn_cache_respond(sc, NULL); 3040 3041 /* Advance the timer back-off. */ 3042 sc->sc_rxtshift++; 3043 SYN_CACHE_TIMER_ARM(sc); 3044 3045 splx(s); 3046 return; 3047 3048 dropit: 3049 tcpstat.tcps_sc_timed_out++; 3050 SYN_CACHE_RM(sc); 3051 SYN_CACHE_PUT(sc); 3052 splx(s); 3053 } 3054 3055 /* 3056 * Remove syn cache created by the specified tcb entry, 3057 * because this does not make sense to keep them 3058 * (if there's no tcb entry, syn cache entry will never be used) 3059 */ 3060 void 3061 syn_cache_cleanup(tp) 3062 struct tcpcb *tp; 3063 { 3064 struct syn_cache *sc, *nsc; 3065 int s; 3066 3067 s = splsoftnet(); 3068 3069 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3070 nsc = LIST_NEXT(sc, sc_tpq); 3071 3072 #ifdef DIAGNOSTIC 3073 if (sc->sc_tp != tp) 3074 panic("invalid sc_tp in syn_cache_cleanup"); 3075 #endif 3076 SYN_CACHE_RM(sc); 3077 SYN_CACHE_PUT(sc); 3078 } 3079 /* just for safety */ 3080 LIST_INIT(&tp->t_sc); 3081 3082 splx(s); 3083 } 3084 3085 /* 3086 * Find an entry in the syn cache. 3087 */ 3088 struct syn_cache * 3089 syn_cache_lookup(src, dst, headp) 3090 struct sockaddr *src; 3091 struct sockaddr *dst; 3092 struct syn_cache_head **headp; 3093 { 3094 struct syn_cache *sc; 3095 struct syn_cache_head *scp; 3096 u_int32_t hash; 3097 int s; 3098 3099 SYN_HASHALL(hash, src, dst); 3100 3101 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3102 *headp = scp; 3103 s = splsoftnet(); 3104 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3105 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3106 if (sc->sc_hash != hash) 3107 continue; 3108 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3109 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3110 splx(s); 3111 return (sc); 3112 } 3113 } 3114 splx(s); 3115 return (NULL); 3116 } 3117 3118 /* 3119 * This function gets called when we receive an ACK for a 3120 * socket in the LISTEN state. We look up the connection 3121 * in the syn cache, and if its there, we pull it out of 3122 * the cache and turn it into a full-blown connection in 3123 * the SYN-RECEIVED state. 3124 * 3125 * The return values may not be immediately obvious, and their effects 3126 * can be subtle, so here they are: 3127 * 3128 * NULL SYN was not found in cache; caller should drop the 3129 * packet and send an RST. 3130 * 3131 * -1 We were unable to create the new connection, and are 3132 * aborting it. An ACK,RST is being sent to the peer 3133 * (unless we got screwey sequence numbners; see below), 3134 * because the 3-way handshake has been completed. Caller 3135 * should not free the mbuf, since we may be using it. If 3136 * we are not, we will free it. 3137 * 3138 * Otherwise, the return value is a pointer to the new socket 3139 * associated with the connection. 3140 */ 3141 struct socket * 3142 syn_cache_get(src, dst, th, hlen, tlen, so, m) 3143 struct sockaddr *src; 3144 struct sockaddr *dst; 3145 struct tcphdr *th; 3146 unsigned int hlen, tlen; 3147 struct socket *so; 3148 struct mbuf *m; 3149 { 3150 struct syn_cache *sc; 3151 struct syn_cache_head *scp; 3152 struct inpcb *inp = NULL; 3153 #ifdef INET6 3154 struct in6pcb *in6p = NULL; 3155 #endif 3156 struct tcpcb *tp = 0; 3157 struct mbuf *am; 3158 int s; 3159 struct socket *oso; 3160 3161 s = splsoftnet(); 3162 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3163 splx(s); 3164 return (NULL); 3165 } 3166 3167 /* 3168 * Verify the sequence and ack numbers. Try getting the correct 3169 * response again. 3170 */ 3171 if ((th->th_ack != sc->sc_iss + 1) || 3172 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3173 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3174 (void) syn_cache_respond(sc, m); 3175 splx(s); 3176 return ((struct socket *)(-1)); 3177 } 3178 3179 /* Remove this cache entry */ 3180 SYN_CACHE_RM(sc); 3181 splx(s); 3182 3183 /* 3184 * Ok, create the full blown connection, and set things up 3185 * as they would have been set up if we had created the 3186 * connection when the SYN arrived. If we can't create 3187 * the connection, abort it. 3188 */ 3189 /* 3190 * inp still has the OLD in_pcb stuff, set the 3191 * v6-related flags on the new guy, too. This is 3192 * done particularly for the case where an AF_INET6 3193 * socket is bound only to a port, and a v4 connection 3194 * comes in on that port. 3195 * we also copy the flowinfo from the original pcb 3196 * to the new one. 3197 */ 3198 oso = so; 3199 so = sonewconn(so, SS_ISCONNECTED); 3200 if (so == NULL) 3201 goto resetandabort; 3202 3203 switch (so->so_proto->pr_domain->dom_family) { 3204 #ifdef INET 3205 case AF_INET: 3206 inp = sotoinpcb(so); 3207 break; 3208 #endif 3209 #ifdef INET6 3210 case AF_INET6: 3211 in6p = sotoin6pcb(so); 3212 break; 3213 #endif 3214 } 3215 switch (src->sa_family) { 3216 #ifdef INET 3217 case AF_INET: 3218 if (inp) { 3219 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3220 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3221 inp->inp_options = ip_srcroute(); 3222 in_pcbstate(inp, INP_BOUND); 3223 if (inp->inp_options == NULL) { 3224 inp->inp_options = sc->sc_ipopts; 3225 sc->sc_ipopts = NULL; 3226 } 3227 } 3228 #ifdef INET6 3229 else if (in6p) { 3230 /* IPv4 packet to AF_INET6 socket */ 3231 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3232 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3233 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3234 &in6p->in6p_laddr.s6_addr32[3], 3235 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3236 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3237 in6totcpcb(in6p)->t_family = AF_INET; 3238 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3239 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3240 else 3241 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3242 in6_pcbstate(in6p, IN6P_BOUND); 3243 } 3244 #endif 3245 break; 3246 #endif 3247 #ifdef INET6 3248 case AF_INET6: 3249 if (in6p) { 3250 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3251 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3252 in6_pcbstate(in6p, IN6P_BOUND); 3253 } 3254 break; 3255 #endif 3256 } 3257 #ifdef INET6 3258 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3259 struct in6pcb *oin6p = sotoin6pcb(oso); 3260 /* inherit socket options from the listening socket */ 3261 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3262 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3263 m_freem(in6p->in6p_options); 3264 in6p->in6p_options = 0; 3265 } 3266 ip6_savecontrol(in6p, &in6p->in6p_options, 3267 mtod(m, struct ip6_hdr *), m); 3268 } 3269 #endif 3270 3271 #if defined(IPSEC) || defined(FAST_IPSEC) 3272 /* 3273 * we make a copy of policy, instead of sharing the policy, 3274 * for better behavior in terms of SA lookup and dead SA removal. 3275 */ 3276 if (inp) { 3277 /* copy old policy into new socket's */ 3278 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3279 printf("tcp_input: could not copy policy\n"); 3280 } 3281 #ifdef INET6 3282 else if (in6p) { 3283 /* copy old policy into new socket's */ 3284 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3285 in6p->in6p_sp)) 3286 printf("tcp_input: could not copy policy\n"); 3287 } 3288 #endif 3289 #endif 3290 3291 /* 3292 * Give the new socket our cached route reference. 3293 */ 3294 if (inp) 3295 inp->inp_route = sc->sc_route4; /* struct assignment */ 3296 #ifdef INET6 3297 else 3298 in6p->in6p_route = sc->sc_route6; 3299 #endif 3300 sc->sc_route4.ro_rt = NULL; 3301 3302 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3303 if (am == NULL) 3304 goto resetandabort; 3305 MCLAIM(am, &tcp_mowner); 3306 am->m_len = src->sa_len; 3307 bcopy(src, mtod(am, caddr_t), src->sa_len); 3308 if (inp) { 3309 if (in_pcbconnect(inp, am)) { 3310 (void) m_free(am); 3311 goto resetandabort; 3312 } 3313 } 3314 #ifdef INET6 3315 else if (in6p) { 3316 if (src->sa_family == AF_INET) { 3317 /* IPv4 packet to AF_INET6 socket */ 3318 struct sockaddr_in6 *sin6; 3319 sin6 = mtod(am, struct sockaddr_in6 *); 3320 am->m_len = sizeof(*sin6); 3321 bzero(sin6, sizeof(*sin6)); 3322 sin6->sin6_family = AF_INET6; 3323 sin6->sin6_len = sizeof(*sin6); 3324 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3325 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3326 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3327 &sin6->sin6_addr.s6_addr32[3], 3328 sizeof(sin6->sin6_addr.s6_addr32[3])); 3329 } 3330 if (in6_pcbconnect(in6p, am)) { 3331 (void) m_free(am); 3332 goto resetandabort; 3333 } 3334 } 3335 #endif 3336 else { 3337 (void) m_free(am); 3338 goto resetandabort; 3339 } 3340 (void) m_free(am); 3341 3342 if (inp) 3343 tp = intotcpcb(inp); 3344 #ifdef INET6 3345 else if (in6p) 3346 tp = in6totcpcb(in6p); 3347 #endif 3348 else 3349 tp = NULL; 3350 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3351 if (sc->sc_request_r_scale != 15) { 3352 tp->requested_s_scale = sc->sc_requested_s_scale; 3353 tp->request_r_scale = sc->sc_request_r_scale; 3354 tp->snd_scale = sc->sc_requested_s_scale; 3355 tp->rcv_scale = sc->sc_request_r_scale; 3356 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3357 } 3358 if (sc->sc_flags & SCF_TIMESTAMP) 3359 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3360 tp->ts_timebase = sc->sc_timebase; 3361 3362 tp->t_template = tcp_template(tp); 3363 if (tp->t_template == 0) { 3364 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3365 so = NULL; 3366 m_freem(m); 3367 goto abort; 3368 } 3369 3370 tp->iss = sc->sc_iss; 3371 tp->irs = sc->sc_irs; 3372 tcp_sendseqinit(tp); 3373 tcp_rcvseqinit(tp); 3374 tp->t_state = TCPS_SYN_RECEIVED; 3375 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3376 tcpstat.tcps_accepts++; 3377 3378 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3379 tp->t_ourmss = sc->sc_ourmaxseg; 3380 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3381 3382 /* 3383 * Initialize the initial congestion window. If we 3384 * had to retransmit the SYN,ACK, we must initialize cwnd 3385 * to 1 segment (i.e. the Loss Window). 3386 */ 3387 if (sc->sc_rxtshift) 3388 tp->snd_cwnd = tp->t_peermss; 3389 else { 3390 int ss = tcp_init_win; 3391 #ifdef INET 3392 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3393 ss = tcp_init_win_local; 3394 #endif 3395 #ifdef INET6 3396 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3397 ss = tcp_init_win_local; 3398 #endif 3399 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3400 } 3401 3402 tcp_rmx_rtt(tp); 3403 tp->snd_wl1 = sc->sc_irs; 3404 tp->rcv_up = sc->sc_irs + 1; 3405 3406 /* 3407 * This is what whould have happened in tcp_output() when 3408 * the SYN,ACK was sent. 3409 */ 3410 tp->snd_up = tp->snd_una; 3411 tp->snd_max = tp->snd_nxt = tp->iss+1; 3412 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3413 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3414 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3415 tp->last_ack_sent = tp->rcv_nxt; 3416 3417 tcpstat.tcps_sc_completed++; 3418 SYN_CACHE_PUT(sc); 3419 return (so); 3420 3421 resetandabort: 3422 (void) tcp_respond(NULL, m, m, th, 3423 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 3424 abort: 3425 if (so != NULL) 3426 (void) soabort(so); 3427 SYN_CACHE_PUT(sc); 3428 tcpstat.tcps_sc_aborted++; 3429 return ((struct socket *)(-1)); 3430 } 3431 3432 /* 3433 * This function is called when we get a RST for a 3434 * non-existent connection, so that we can see if the 3435 * connection is in the syn cache. If it is, zap it. 3436 */ 3437 3438 void 3439 syn_cache_reset(src, dst, th) 3440 struct sockaddr *src; 3441 struct sockaddr *dst; 3442 struct tcphdr *th; 3443 { 3444 struct syn_cache *sc; 3445 struct syn_cache_head *scp; 3446 int s = splsoftnet(); 3447 3448 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3449 splx(s); 3450 return; 3451 } 3452 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3453 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3454 splx(s); 3455 return; 3456 } 3457 SYN_CACHE_RM(sc); 3458 splx(s); 3459 tcpstat.tcps_sc_reset++; 3460 SYN_CACHE_PUT(sc); 3461 } 3462 3463 void 3464 syn_cache_unreach(src, dst, th) 3465 struct sockaddr *src; 3466 struct sockaddr *dst; 3467 struct tcphdr *th; 3468 { 3469 struct syn_cache *sc; 3470 struct syn_cache_head *scp; 3471 int s; 3472 3473 s = splsoftnet(); 3474 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3475 splx(s); 3476 return; 3477 } 3478 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3479 if (ntohl (th->th_seq) != sc->sc_iss) { 3480 splx(s); 3481 return; 3482 } 3483 3484 /* 3485 * If we've retransmitted 3 times and this is our second error, 3486 * we remove the entry. Otherwise, we allow it to continue on. 3487 * This prevents us from incorrectly nuking an entry during a 3488 * spurious network outage. 3489 * 3490 * See tcp_notify(). 3491 */ 3492 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3493 sc->sc_flags |= SCF_UNREACH; 3494 splx(s); 3495 return; 3496 } 3497 3498 SYN_CACHE_RM(sc); 3499 splx(s); 3500 tcpstat.tcps_sc_unreach++; 3501 SYN_CACHE_PUT(sc); 3502 } 3503 3504 /* 3505 * Given a LISTEN socket and an inbound SYN request, add 3506 * this to the syn cache, and send back a segment: 3507 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3508 * to the source. 3509 * 3510 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3511 * Doing so would require that we hold onto the data and deliver it 3512 * to the application. However, if we are the target of a SYN-flood 3513 * DoS attack, an attacker could send data which would eventually 3514 * consume all available buffer space if it were ACKed. By not ACKing 3515 * the data, we avoid this DoS scenario. 3516 */ 3517 3518 int 3519 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi) 3520 struct sockaddr *src; 3521 struct sockaddr *dst; 3522 struct tcphdr *th; 3523 unsigned int hlen; 3524 struct socket *so; 3525 struct mbuf *m; 3526 u_char *optp; 3527 int optlen; 3528 struct tcp_opt_info *oi; 3529 { 3530 struct tcpcb tb, *tp; 3531 long win; 3532 struct syn_cache *sc; 3533 struct syn_cache_head *scp; 3534 struct mbuf *ipopts; 3535 3536 tp = sototcpcb(so); 3537 3538 /* 3539 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3540 * 3541 * Note this check is performed in tcp_input() very early on. 3542 */ 3543 3544 /* 3545 * Initialize some local state. 3546 */ 3547 win = sbspace(&so->so_rcv); 3548 if (win > TCP_MAXWIN) 3549 win = TCP_MAXWIN; 3550 3551 switch (src->sa_family) { 3552 #ifdef INET 3553 case AF_INET: 3554 /* 3555 * Remember the IP options, if any. 3556 */ 3557 ipopts = ip_srcroute(); 3558 break; 3559 #endif 3560 default: 3561 ipopts = NULL; 3562 } 3563 3564 if (optp) { 3565 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3566 tcp_dooptions(&tb, optp, optlen, th, oi); 3567 } else 3568 tb.t_flags = 0; 3569 3570 /* 3571 * See if we already have an entry for this connection. 3572 * If we do, resend the SYN,ACK. We do not count this 3573 * as a retransmission (XXX though maybe we should). 3574 */ 3575 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3576 tcpstat.tcps_sc_dupesyn++; 3577 if (ipopts) { 3578 /* 3579 * If we were remembering a previous source route, 3580 * forget it and use the new one we've been given. 3581 */ 3582 if (sc->sc_ipopts) 3583 (void) m_free(sc->sc_ipopts); 3584 sc->sc_ipopts = ipopts; 3585 } 3586 sc->sc_timestamp = tb.ts_recent; 3587 if (syn_cache_respond(sc, m) == 0) { 3588 tcpstat.tcps_sndacks++; 3589 tcpstat.tcps_sndtotal++; 3590 } 3591 return (1); 3592 } 3593 3594 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3595 if (sc == NULL) { 3596 if (ipopts) 3597 (void) m_free(ipopts); 3598 return (0); 3599 } 3600 3601 /* 3602 * Fill in the cache, and put the necessary IP and TCP 3603 * options into the reply. 3604 */ 3605 bzero(sc, sizeof(struct syn_cache)); 3606 callout_init(&sc->sc_timer); 3607 bcopy(src, &sc->sc_src, src->sa_len); 3608 bcopy(dst, &sc->sc_dst, dst->sa_len); 3609 sc->sc_flags = 0; 3610 sc->sc_ipopts = ipopts; 3611 sc->sc_irs = th->th_seq; 3612 switch (src->sa_family) { 3613 #ifdef INET 3614 case AF_INET: 3615 { 3616 struct sockaddr_in *srcin = (void *) src; 3617 struct sockaddr_in *dstin = (void *) dst; 3618 3619 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 3620 &srcin->sin_addr, dstin->sin_port, 3621 srcin->sin_port, sizeof(dstin->sin_addr), 0); 3622 break; 3623 } 3624 #endif /* INET */ 3625 #ifdef INET6 3626 case AF_INET6: 3627 { 3628 struct sockaddr_in6 *srcin6 = (void *) src; 3629 struct sockaddr_in6 *dstin6 = (void *) dst; 3630 3631 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 3632 &srcin6->sin6_addr, dstin6->sin6_port, 3633 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 3634 break; 3635 } 3636 #endif /* INET6 */ 3637 } 3638 sc->sc_peermaxseg = oi->maxseg; 3639 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3640 m->m_pkthdr.rcvif : NULL, 3641 sc->sc_src.sa.sa_family); 3642 sc->sc_win = win; 3643 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 3644 sc->sc_timestamp = tb.ts_recent; 3645 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 3646 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 3647 sc->sc_flags |= SCF_TIMESTAMP; 3648 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3649 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3650 sc->sc_requested_s_scale = tb.requested_s_scale; 3651 sc->sc_request_r_scale = 0; 3652 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3653 TCP_MAXWIN << sc->sc_request_r_scale < 3654 so->so_rcv.sb_hiwat) 3655 sc->sc_request_r_scale++; 3656 } else { 3657 sc->sc_requested_s_scale = 15; 3658 sc->sc_request_r_scale = 15; 3659 } 3660 sc->sc_tp = tp; 3661 if (syn_cache_respond(sc, m) == 0) { 3662 syn_cache_insert(sc, tp); 3663 tcpstat.tcps_sndacks++; 3664 tcpstat.tcps_sndtotal++; 3665 } else { 3666 SYN_CACHE_PUT(sc); 3667 tcpstat.tcps_sc_dropped++; 3668 } 3669 return (1); 3670 } 3671 3672 int 3673 syn_cache_respond(sc, m) 3674 struct syn_cache *sc; 3675 struct mbuf *m; 3676 { 3677 struct route *ro; 3678 u_int8_t *optp; 3679 int optlen, error; 3680 u_int16_t tlen; 3681 struct ip *ip = NULL; 3682 #ifdef INET6 3683 struct ip6_hdr *ip6 = NULL; 3684 #endif 3685 struct tcpcb *tp; 3686 struct tcphdr *th; 3687 u_int hlen; 3688 struct socket *so; 3689 3690 switch (sc->sc_src.sa.sa_family) { 3691 case AF_INET: 3692 hlen = sizeof(struct ip); 3693 ro = &sc->sc_route4; 3694 break; 3695 #ifdef INET6 3696 case AF_INET6: 3697 hlen = sizeof(struct ip6_hdr); 3698 ro = (struct route *)&sc->sc_route6; 3699 break; 3700 #endif 3701 default: 3702 if (m) 3703 m_freem(m); 3704 return (EAFNOSUPPORT); 3705 } 3706 3707 /* Compute the size of the TCP options. */ 3708 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 3709 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 3710 3711 tlen = hlen + sizeof(struct tcphdr) + optlen; 3712 3713 /* 3714 * Create the IP+TCP header from scratch. 3715 */ 3716 if (m) 3717 m_freem(m); 3718 #ifdef DIAGNOSTIC 3719 if (max_linkhdr + tlen > MCLBYTES) 3720 return (ENOBUFS); 3721 #endif 3722 MGETHDR(m, M_DONTWAIT, MT_DATA); 3723 if (m && tlen > MHLEN) { 3724 MCLGET(m, M_DONTWAIT); 3725 if ((m->m_flags & M_EXT) == 0) { 3726 m_freem(m); 3727 m = NULL; 3728 } 3729 } 3730 if (m == NULL) 3731 return (ENOBUFS); 3732 MCLAIM(m, &tcp_tx_mowner); 3733 3734 /* Fixup the mbuf. */ 3735 m->m_data += max_linkhdr; 3736 m->m_len = m->m_pkthdr.len = tlen; 3737 if (sc->sc_tp) { 3738 tp = sc->sc_tp; 3739 if (tp->t_inpcb) 3740 so = tp->t_inpcb->inp_socket; 3741 #ifdef INET6 3742 else if (tp->t_in6pcb) 3743 so = tp->t_in6pcb->in6p_socket; 3744 #endif 3745 else 3746 so = NULL; 3747 } else 3748 so = NULL; 3749 m->m_pkthdr.rcvif = NULL; 3750 memset(mtod(m, u_char *), 0, tlen); 3751 3752 switch (sc->sc_src.sa.sa_family) { 3753 case AF_INET: 3754 ip = mtod(m, struct ip *); 3755 ip->ip_dst = sc->sc_src.sin.sin_addr; 3756 ip->ip_src = sc->sc_dst.sin.sin_addr; 3757 ip->ip_p = IPPROTO_TCP; 3758 th = (struct tcphdr *)(ip + 1); 3759 th->th_dport = sc->sc_src.sin.sin_port; 3760 th->th_sport = sc->sc_dst.sin.sin_port; 3761 break; 3762 #ifdef INET6 3763 case AF_INET6: 3764 ip6 = mtod(m, struct ip6_hdr *); 3765 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 3766 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 3767 ip6->ip6_nxt = IPPROTO_TCP; 3768 /* ip6_plen will be updated in ip6_output() */ 3769 th = (struct tcphdr *)(ip6 + 1); 3770 th->th_dport = sc->sc_src.sin6.sin6_port; 3771 th->th_sport = sc->sc_dst.sin6.sin6_port; 3772 break; 3773 #endif 3774 default: 3775 th = NULL; 3776 } 3777 3778 th->th_seq = htonl(sc->sc_iss); 3779 th->th_ack = htonl(sc->sc_irs + 1); 3780 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 3781 th->th_flags = TH_SYN|TH_ACK; 3782 th->th_win = htons(sc->sc_win); 3783 /* th_sum already 0 */ 3784 /* th_urp already 0 */ 3785 3786 /* Tack on the TCP options. */ 3787 optp = (u_int8_t *)(th + 1); 3788 *optp++ = TCPOPT_MAXSEG; 3789 *optp++ = 4; 3790 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 3791 *optp++ = sc->sc_ourmaxseg & 0xff; 3792 3793 if (sc->sc_request_r_scale != 15) { 3794 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 3795 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 3796 sc->sc_request_r_scale); 3797 optp += 4; 3798 } 3799 3800 if (sc->sc_flags & SCF_TIMESTAMP) { 3801 u_int32_t *lp = (u_int32_t *)(optp); 3802 /* Form timestamp option as shown in appendix A of RFC 1323. */ 3803 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 3804 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 3805 *lp = htonl(sc->sc_timestamp); 3806 optp += TCPOLEN_TSTAMP_APPA; 3807 } 3808 3809 /* Compute the packet's checksum. */ 3810 switch (sc->sc_src.sa.sa_family) { 3811 case AF_INET: 3812 ip->ip_len = htons(tlen - hlen); 3813 th->th_sum = 0; 3814 th->th_sum = in_cksum(m, tlen); 3815 break; 3816 #ifdef INET6 3817 case AF_INET6: 3818 ip6->ip6_plen = htons(tlen - hlen); 3819 th->th_sum = 0; 3820 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 3821 break; 3822 #endif 3823 } 3824 3825 /* 3826 * Fill in some straggling IP bits. Note the stack expects 3827 * ip_len to be in host order, for convenience. 3828 */ 3829 switch (sc->sc_src.sa.sa_family) { 3830 #ifdef INET 3831 case AF_INET: 3832 ip->ip_len = htons(tlen); 3833 ip->ip_ttl = ip_defttl; 3834 /* XXX tos? */ 3835 break; 3836 #endif 3837 #ifdef INET6 3838 case AF_INET6: 3839 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 3840 ip6->ip6_vfc |= IPV6_VERSION; 3841 ip6->ip6_plen = htons(tlen - hlen); 3842 /* ip6_hlim will be initialized afterwards */ 3843 /* XXX flowlabel? */ 3844 break; 3845 #endif 3846 } 3847 3848 /* XXX use IPsec policy on listening socket, on SYN ACK */ 3849 tp = sc->sc_tp; 3850 3851 switch (sc->sc_src.sa.sa_family) { 3852 #ifdef INET 3853 case AF_INET: 3854 error = ip_output(m, sc->sc_ipopts, ro, 3855 (ip_mtudisc ? IP_MTUDISC : 0), 3856 (struct ip_moptions *)NULL, so); 3857 break; 3858 #endif 3859 #ifdef INET6 3860 case AF_INET6: 3861 ip6->ip6_hlim = in6_selecthlim(NULL, 3862 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 3863 3864 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0, 3865 (struct ip6_moptions *)0, so, NULL); 3866 break; 3867 #endif 3868 default: 3869 error = EAFNOSUPPORT; 3870 break; 3871 } 3872 return (error); 3873 } 3874
Indexes created Wed Sep 24 05:09:52 GMT 2025