1 /* $NetBSD: tcp_input.c,v 1.142 2002/05/28 10:11:51 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. All advertising materials mentioning features or use of this software 122 * must display the following acknowledgement: 123 * This product includes software developed by the University of 124 * California, Berkeley and its contributors. 125 * 4. Neither the name of the University nor the names of its contributors 126 * may be used to endorse or promote products derived from this software 127 * without specific prior written permission. 128 * 129 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 130 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 131 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 132 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 133 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 134 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 135 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 136 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 137 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 138 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 139 * SUCH DAMAGE. 140 * 141 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 142 */ 143 144 /* 145 * TODO list for SYN cache stuff: 146 * 147 * Find room for a "state" field, which is needed to keep a 148 * compressed state for TIME_WAIT TCBs. It's been noted already 149 * that this is fairly important for very high-volume web and 150 * mail servers, which use a large number of short-lived 151 * connections. 152 */ 153 154 #include <sys/cdefs.h> 155 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.142 2002/05/28 10:11:51 itojun Exp $"); 156 157 #include "opt_inet.h" 158 #include "opt_ipsec.h" 159 #include "opt_inet_csum.h" 160 #include "opt_tcp_debug.h" 161 162 #include <sys/param.h> 163 #include <sys/systm.h> 164 #include <sys/malloc.h> 165 #include <sys/mbuf.h> 166 #include <sys/protosw.h> 167 #include <sys/socket.h> 168 #include <sys/socketvar.h> 169 #include <sys/errno.h> 170 #include <sys/syslog.h> 171 #include <sys/pool.h> 172 #include <sys/domain.h> 173 #include <sys/kernel.h> 174 175 #include <net/if.h> 176 #include <net/route.h> 177 #include <net/if_types.h> 178 179 #include <netinet/in.h> 180 #include <netinet/in_systm.h> 181 #include <netinet/ip.h> 182 #include <netinet/in_pcb.h> 183 #include <netinet/ip_var.h> 184 185 #ifdef INET6 186 #ifndef INET 187 #include <netinet/in.h> 188 #endif 189 #include <netinet/ip6.h> 190 #include <netinet6/ip6_var.h> 191 #include <netinet6/in6_pcb.h> 192 #include <netinet6/ip6_var.h> 193 #include <netinet6/in6_var.h> 194 #include <netinet/icmp6.h> 195 #include <netinet6/nd6.h> 196 #endif 197 198 #ifdef PULLDOWN_TEST 199 #ifndef INET6 200 /* always need ip6.h for IP6_EXTHDR_GET */ 201 #include <netinet/ip6.h> 202 #endif 203 #endif 204 205 #include <netinet/tcp.h> 206 #include <netinet/tcp_fsm.h> 207 #include <netinet/tcp_seq.h> 208 #include <netinet/tcp_timer.h> 209 #include <netinet/tcp_var.h> 210 #include <netinet/tcpip.h> 211 #include <netinet/tcp_debug.h> 212 213 #include <machine/stdarg.h> 214 215 #ifdef IPSEC 216 #include <netinet6/ipsec.h> 217 #include <netkey/key.h> 218 #endif /*IPSEC*/ 219 #ifdef INET6 220 #include "faith.h" 221 #if defined(NFAITH) && NFAITH > 0 222 #include <net/if_faith.h> 223 #endif 224 #endif 225 226 int tcprexmtthresh = 3; 227 int tcp_log_refused; 228 229 static int tcp_rst_ppslim_count = 0; 230 static struct timeval tcp_rst_ppslim_last; 231 232 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 233 234 /* for modulo comparisons of timestamps */ 235 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 236 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 237 238 /* 239 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 240 */ 241 #ifdef INET6 242 #define ND6_HINT(tp) \ 243 do { \ 244 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 \ 245 && tp->t_in6pcb->in6p_route.ro_rt) { \ 246 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \ 247 } \ 248 } while (0) 249 #else 250 #define ND6_HINT(tp) 251 #endif 252 253 /* 254 * Macro to compute ACK transmission behavior. Delay the ACK unless 255 * we have already delayed an ACK (must send an ACK every two segments). 256 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 257 * option is enabled. 258 */ 259 #define TCP_SETUP_ACK(tp, th) \ 260 do { \ 261 if ((tp)->t_flags & TF_DELACK || \ 262 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 263 tp->t_flags |= TF_ACKNOW; \ 264 else \ 265 TCP_SET_DELACK(tp); \ 266 } while (0) 267 268 /* 269 * Convert TCP protocol fields to host order for easier processing. 270 */ 271 #define TCP_FIELDS_TO_HOST(th) \ 272 do { \ 273 NTOHL((th)->th_seq); \ 274 NTOHL((th)->th_ack); \ 275 NTOHS((th)->th_win); \ 276 NTOHS((th)->th_urp); \ 277 } while (0) 278 279 #ifdef TCP_CSUM_COUNTERS 280 #include <sys/device.h> 281 282 extern struct evcnt tcp_hwcsum_ok; 283 extern struct evcnt tcp_hwcsum_bad; 284 extern struct evcnt tcp_hwcsum_data; 285 extern struct evcnt tcp_swcsum; 286 287 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 288 289 #else 290 291 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 292 293 #endif /* TCP_CSUM_COUNTERS */ 294 295 #ifdef TCP_REASS_COUNTERS 296 #include <sys/device.h> 297 298 extern struct evcnt tcp_reass_; 299 extern struct evcnt tcp_reass_empty; 300 extern struct evcnt tcp_reass_iteration[8]; 301 extern struct evcnt tcp_reass_prependfirst; 302 extern struct evcnt tcp_reass_prepend; 303 extern struct evcnt tcp_reass_insert; 304 extern struct evcnt tcp_reass_inserttail; 305 extern struct evcnt tcp_reass_append; 306 extern struct evcnt tcp_reass_appendtail; 307 extern struct evcnt tcp_reass_overlaptail; 308 extern struct evcnt tcp_reass_overlapfront; 309 extern struct evcnt tcp_reass_segdup; 310 extern struct evcnt tcp_reass_fragdup; 311 312 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 313 314 #else 315 316 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 317 318 #endif /* TCP_REASS_COUNTERS */ 319 320 int 321 tcp_reass(tp, th, m, tlen) 322 struct tcpcb *tp; 323 struct tcphdr *th; 324 struct mbuf *m; 325 int *tlen; 326 { 327 struct ipqent *p, *q, *nq, *tiqe = NULL; 328 struct socket *so = NULL; 329 int pkt_flags; 330 tcp_seq pkt_seq; 331 unsigned pkt_len; 332 u_long rcvpartdupbyte = 0; 333 u_long rcvoobyte; 334 #ifdef TCP_REASS_COUNTERS 335 u_int count = 0; 336 #endif 337 338 if (tp->t_inpcb) 339 so = tp->t_inpcb->inp_socket; 340 #ifdef INET6 341 else if (tp->t_in6pcb) 342 so = tp->t_in6pcb->in6p_socket; 343 #endif 344 345 TCP_REASS_LOCK_CHECK(tp); 346 347 /* 348 * Call with th==0 after become established to 349 * force pre-ESTABLISHED data up to user socket. 350 */ 351 if (th == 0) 352 goto present; 353 354 rcvoobyte = *tlen; 355 /* 356 * Copy these to local variables because the tcpiphdr 357 * gets munged while we are collapsing mbufs. 358 */ 359 pkt_seq = th->th_seq; 360 pkt_len = *tlen; 361 pkt_flags = th->th_flags; 362 363 TCP_REASS_COUNTER_INCR(&tcp_reass_); 364 365 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 366 /* 367 * When we miss a packet, the vast majority of time we get 368 * packets that follow it in order. So optimize for that. 369 */ 370 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 371 p->ipqe_len += pkt_len; 372 p->ipqe_flags |= pkt_flags; 373 m_cat(p->ipqe_m, m); 374 tiqe = p; 375 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 376 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 377 goto skip_replacement; 378 } 379 /* 380 * While we're here, if the pkt is completely beyond 381 * anything we have, just insert it at the tail. 382 */ 383 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 384 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 385 goto insert_it; 386 } 387 } 388 389 q = TAILQ_FIRST(&tp->segq); 390 391 if (q != NULL) { 392 /* 393 * If this segment immediately precedes the first out-of-order 394 * block, simply slap the segment in front of it and (mostly) 395 * skip the complicated logic. 396 */ 397 if (pkt_seq + pkt_len == q->ipqe_seq) { 398 q->ipqe_seq = pkt_seq; 399 q->ipqe_len += pkt_len; 400 q->ipqe_flags |= pkt_flags; 401 m_cat(m, q->ipqe_m); 402 q->ipqe_m = m; 403 tiqe = q; 404 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 405 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 406 goto skip_replacement; 407 } 408 } else { 409 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 410 } 411 412 /* 413 * Find a segment which begins after this one does. 414 */ 415 for (p = NULL; q != NULL; q = nq) { 416 nq = TAILQ_NEXT(q, ipqe_q); 417 #ifdef TCP_REASS_COUNTERS 418 count++; 419 #endif 420 /* 421 * If the received segment is just right after this 422 * fragment, merge the two together and then check 423 * for further overlaps. 424 */ 425 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 426 #ifdef TCPREASS_DEBUG 427 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 428 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 429 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 430 #endif 431 pkt_len += q->ipqe_len; 432 pkt_flags |= q->ipqe_flags; 433 pkt_seq = q->ipqe_seq; 434 m_cat(q->ipqe_m, m); 435 m = q->ipqe_m; 436 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 437 goto free_ipqe; 438 } 439 /* 440 * If the received segment is completely past this 441 * fragment, we need to go the next fragment. 442 */ 443 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 444 p = q; 445 continue; 446 } 447 /* 448 * If the fragment is past the received segment, 449 * it (or any following) can't be concatenated. 450 */ 451 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 452 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 453 break; 454 } 455 456 /* 457 * We've received all the data in this segment before. 458 * mark it as a duplicate and return. 459 */ 460 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 461 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 462 tcpstat.tcps_rcvduppack++; 463 tcpstat.tcps_rcvdupbyte += pkt_len; 464 m_freem(m); 465 if (tiqe != NULL) 466 pool_put(&ipqent_pool, tiqe); 467 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 468 return (0); 469 } 470 /* 471 * Received segment completely overlaps this fragment 472 * so we drop the fragment (this keeps the temporal 473 * ordering of segments correct). 474 */ 475 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 476 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 477 rcvpartdupbyte += q->ipqe_len; 478 m_freem(q->ipqe_m); 479 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 480 goto free_ipqe; 481 } 482 /* 483 * RX'ed segment extends past the end of the 484 * fragment. Drop the overlapping bytes. Then 485 * merge the fragment and segment then treat as 486 * a longer received packet. 487 */ 488 if (SEQ_LT(q->ipqe_seq, pkt_seq) 489 && SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 490 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 491 #ifdef TCPREASS_DEBUG 492 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 493 tp, overlap, 494 pkt_seq, pkt_seq + pkt_len, pkt_len); 495 #endif 496 m_adj(m, overlap); 497 rcvpartdupbyte += overlap; 498 m_cat(q->ipqe_m, m); 499 m = q->ipqe_m; 500 pkt_seq = q->ipqe_seq; 501 pkt_len += q->ipqe_len - overlap; 502 rcvoobyte -= overlap; 503 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 504 goto free_ipqe; 505 } 506 /* 507 * RX'ed segment extends past the front of the 508 * fragment. Drop the overlapping bytes on the 509 * received packet. The packet will then be 510 * contatentated with this fragment a bit later. 511 */ 512 if (SEQ_GT(q->ipqe_seq, pkt_seq) 513 && SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 514 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 515 #ifdef TCPREASS_DEBUG 516 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 517 tp, overlap, 518 pkt_seq, pkt_seq + pkt_len, pkt_len); 519 #endif 520 m_adj(m, -overlap); 521 pkt_len -= overlap; 522 rcvpartdupbyte += overlap; 523 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 524 rcvoobyte -= overlap; 525 } 526 /* 527 * If the received segment immediates precedes this 528 * fragment then tack the fragment onto this segment 529 * and reinsert the data. 530 */ 531 if (q->ipqe_seq == pkt_seq + pkt_len) { 532 #ifdef TCPREASS_DEBUG 533 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 534 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 535 pkt_seq, pkt_seq + pkt_len, pkt_len); 536 #endif 537 pkt_len += q->ipqe_len; 538 pkt_flags |= q->ipqe_flags; 539 m_cat(m, q->ipqe_m); 540 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 541 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 542 if (tiqe == NULL) { 543 tiqe = q; 544 } else { 545 pool_put(&ipqent_pool, q); 546 } 547 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 548 break; 549 } 550 /* 551 * If the fragment is before the segment, remember it. 552 * When this loop is terminated, p will contain the 553 * pointer to fragment that is right before the received 554 * segment. 555 */ 556 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 557 p = q; 558 559 continue; 560 561 /* 562 * This is a common operation. It also will allow 563 * to save doing a malloc/free in most instances. 564 */ 565 free_ipqe: 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 } 573 } 574 575 #ifdef TCP_REASS_COUNTERS 576 if (count > 7) 577 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 578 else if (count > 0) 579 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 580 #endif 581 582 insert_it: 583 584 /* 585 * Allocate a new queue entry since the received segment did not 586 * collapse onto any other out-of-order block; thus we are allocating 587 * a new block. If it had collapsed, tiqe would not be NULL and 588 * we would be reusing it. 589 * XXX If we can't, just drop the packet. XXX 590 */ 591 if (tiqe == NULL) { 592 tiqe = pool_get(&ipqent_pool, PR_NOWAIT); 593 if (tiqe == NULL) { 594 tcpstat.tcps_rcvmemdrop++; 595 m_freem(m); 596 return (0); 597 } 598 } 599 600 /* 601 * Update the counters. 602 */ 603 tcpstat.tcps_rcvoopack++; 604 tcpstat.tcps_rcvoobyte += rcvoobyte; 605 if (rcvpartdupbyte) { 606 tcpstat.tcps_rcvpartduppack++; 607 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 608 } 609 610 /* 611 * Insert the new fragment queue entry into both queues. 612 */ 613 tiqe->ipqe_m = m; 614 tiqe->ipqe_seq = pkt_seq; 615 tiqe->ipqe_len = pkt_len; 616 tiqe->ipqe_flags = pkt_flags; 617 if (p == NULL) { 618 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 619 #ifdef TCPREASS_DEBUG 620 if (tiqe->ipqe_seq != tp->rcv_nxt) 621 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 622 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 623 #endif 624 } else { 625 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 626 #ifdef TCPREASS_DEBUG 627 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 628 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 629 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 630 #endif 631 } 632 633 skip_replacement: 634 635 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 636 637 present: 638 /* 639 * Present data to user, advancing rcv_nxt through 640 * completed sequence space. 641 */ 642 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 643 return (0); 644 q = TAILQ_FIRST(&tp->segq); 645 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 646 return (0); 647 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 648 return (0); 649 650 tp->rcv_nxt += q->ipqe_len; 651 pkt_flags = q->ipqe_flags & TH_FIN; 652 ND6_HINT(tp); 653 654 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 655 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 656 if (so->so_state & SS_CANTRCVMORE) 657 m_freem(q->ipqe_m); 658 else 659 sbappend(&so->so_rcv, q->ipqe_m); 660 pool_put(&ipqent_pool, q); 661 sorwakeup(so); 662 return (pkt_flags); 663 } 664 665 #ifdef INET6 666 int 667 tcp6_input(mp, offp, proto) 668 struct mbuf **mp; 669 int *offp, proto; 670 { 671 struct mbuf *m = *mp; 672 673 /* 674 * draft-itojun-ipv6-tcp-to-anycast 675 * better place to put this in? 676 */ 677 if (m->m_flags & M_ANYCAST6) { 678 struct ip6_hdr *ip6; 679 if (m->m_len < sizeof(struct ip6_hdr)) { 680 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 681 tcpstat.tcps_rcvshort++; 682 return IPPROTO_DONE; 683 } 684 } 685 ip6 = mtod(m, struct ip6_hdr *); 686 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 687 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 688 return IPPROTO_DONE; 689 } 690 691 tcp_input(m, *offp, proto); 692 return IPPROTO_DONE; 693 } 694 #endif 695 696 /* 697 * TCP input routine, follows pages 65-76 of the 698 * protocol specification dated September, 1981 very closely. 699 */ 700 void 701 #if __STDC__ 702 tcp_input(struct mbuf *m, ...) 703 #else 704 tcp_input(m, va_alist) 705 struct mbuf *m; 706 #endif 707 { 708 int proto; 709 struct tcphdr *th; 710 struct ip *ip; 711 struct inpcb *inp; 712 #ifdef INET6 713 struct ip6_hdr *ip6; 714 struct in6pcb *in6p; 715 #endif 716 caddr_t optp = NULL; 717 int optlen = 0; 718 int len, tlen, toff, hdroptlen = 0; 719 struct tcpcb *tp = 0; 720 int tiflags; 721 struct socket *so = NULL; 722 int todrop, acked, ourfinisacked, needoutput = 0; 723 short ostate = 0; 724 int iss = 0; 725 u_long tiwin; 726 struct tcp_opt_info opti; 727 int off, iphlen; 728 va_list ap; 729 int af; /* af on the wire */ 730 struct mbuf *tcp_saveti = NULL; 731 732 va_start(ap, m); 733 toff = va_arg(ap, int); 734 proto = va_arg(ap, int); 735 va_end(ap); 736 737 tcpstat.tcps_rcvtotal++; 738 739 bzero(&opti, sizeof(opti)); 740 opti.ts_present = 0; 741 opti.maxseg = 0; 742 743 /* 744 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 745 * 746 * TCP is, by definition, unicast, so we reject all 747 * multicast outright. 748 * 749 * Note, there are additional src/dst address checks in 750 * the AF-specific code below. 751 */ 752 if (m->m_flags & (M_BCAST|M_MCAST)) { 753 /* XXX stat */ 754 goto drop; 755 } 756 #ifdef INET6 757 if (m->m_flags & M_ANYCAST6) { 758 /* XXX stat */ 759 goto drop; 760 } 761 #endif 762 763 /* 764 * Get IP and TCP header together in first mbuf. 765 * Note: IP leaves IP header in first mbuf. 766 */ 767 ip = mtod(m, struct ip *); 768 #ifdef INET6 769 ip6 = NULL; 770 #endif 771 switch (ip->ip_v) { 772 #ifdef INET 773 case 4: 774 af = AF_INET; 775 iphlen = sizeof(struct ip); 776 #ifndef PULLDOWN_TEST 777 /* would like to get rid of this... */ 778 if (toff > sizeof (struct ip)) { 779 ip_stripoptions(m, (struct mbuf *)0); 780 toff = sizeof(struct ip); 781 } 782 if (m->m_len < toff + sizeof (struct tcphdr)) { 783 if ((m = m_pullup(m, toff + sizeof (struct tcphdr))) == 0) { 784 tcpstat.tcps_rcvshort++; 785 return; 786 } 787 } 788 ip = mtod(m, struct ip *); 789 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 790 #else 791 ip = mtod(m, struct ip *); 792 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 793 sizeof(struct tcphdr)); 794 if (th == NULL) { 795 tcpstat.tcps_rcvshort++; 796 return; 797 } 798 #endif 799 /* We do the checksum after PCB lookup... */ 800 len = ip->ip_len; 801 tlen = len - toff; 802 break; 803 #endif 804 #ifdef INET6 805 case 6: 806 ip = NULL; 807 iphlen = sizeof(struct ip6_hdr); 808 af = AF_INET6; 809 #ifndef PULLDOWN_TEST 810 if (m->m_len < toff + sizeof(struct tcphdr)) { 811 m = m_pullup(m, toff + sizeof(struct tcphdr)); /*XXX*/ 812 if (m == NULL) { 813 tcpstat.tcps_rcvshort++; 814 return; 815 } 816 } 817 ip6 = mtod(m, struct ip6_hdr *); 818 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 819 #else 820 ip6 = mtod(m, struct ip6_hdr *); 821 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 822 sizeof(struct tcphdr)); 823 if (th == NULL) { 824 tcpstat.tcps_rcvshort++; 825 return; 826 } 827 #endif 828 829 /* Be proactive about malicious use of IPv4 mapped address */ 830 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 831 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 832 /* XXX stat */ 833 goto drop; 834 } 835 836 /* 837 * Be proactive about unspecified IPv6 address in source. 838 * As we use all-zero to indicate unbounded/unconnected pcb, 839 * unspecified IPv6 address can be used to confuse us. 840 * 841 * Note that packets with unspecified IPv6 destination is 842 * already dropped in ip6_input. 843 */ 844 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 845 /* XXX stat */ 846 goto drop; 847 } 848 849 /* 850 * Make sure destination address is not multicast. 851 * Source address checked in ip6_input(). 852 */ 853 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 854 /* XXX stat */ 855 goto drop; 856 } 857 858 /* We do the checksum after PCB lookup... */ 859 len = m->m_pkthdr.len; 860 tlen = len - toff; 861 break; 862 #endif 863 default: 864 m_freem(m); 865 return; 866 } 867 868 /* 869 * Check that TCP offset makes sense, 870 * pull out TCP options and adjust length. XXX 871 */ 872 off = th->th_off << 2; 873 if (off < sizeof (struct tcphdr) || off > tlen) { 874 tcpstat.tcps_rcvbadoff++; 875 goto drop; 876 } 877 tlen -= off; 878 879 /* 880 * tcp_input() has been modified to use tlen to mean the TCP data 881 * length throughout the function. Other functions can use 882 * m->m_pkthdr.len as the basis for calculating the TCP data length. 883 * rja 884 */ 885 886 if (off > sizeof (struct tcphdr)) { 887 #ifndef PULLDOWN_TEST 888 if (m->m_len < toff + off) { 889 if ((m = m_pullup(m, toff + off)) == 0) { 890 tcpstat.tcps_rcvshort++; 891 return; 892 } 893 switch (af) { 894 #ifdef INET 895 case AF_INET: 896 ip = mtod(m, struct ip *); 897 break; 898 #endif 899 #ifdef INET6 900 case AF_INET6: 901 ip6 = mtod(m, struct ip6_hdr *); 902 break; 903 #endif 904 } 905 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 906 } 907 #else 908 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 909 if (th == NULL) { 910 tcpstat.tcps_rcvshort++; 911 return; 912 } 913 /* 914 * NOTE: ip/ip6 will not be affected by m_pulldown() 915 * (as they're before toff) and we don't need to update those. 916 */ 917 #endif 918 optlen = off - sizeof (struct tcphdr); 919 optp = ((caddr_t)th) + sizeof(struct tcphdr); 920 /* 921 * Do quick retrieval of timestamp options ("options 922 * prediction?"). If timestamp is the only option and it's 923 * formatted as recommended in RFC 1323 appendix A, we 924 * quickly get the values now and not bother calling 925 * tcp_dooptions(), etc. 926 */ 927 if ((optlen == TCPOLEN_TSTAMP_APPA || 928 (optlen > TCPOLEN_TSTAMP_APPA && 929 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 930 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 931 (th->th_flags & TH_SYN) == 0) { 932 opti.ts_present = 1; 933 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 934 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 935 optp = NULL; /* we've parsed the options */ 936 } 937 } 938 tiflags = th->th_flags; 939 940 /* 941 * Locate pcb for segment. 942 */ 943 findpcb: 944 inp = NULL; 945 #ifdef INET6 946 in6p = NULL; 947 #endif 948 switch (af) { 949 #ifdef INET 950 case AF_INET: 951 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 952 ip->ip_dst, th->th_dport); 953 if (inp == 0) { 954 ++tcpstat.tcps_pcbhashmiss; 955 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 956 } 957 #ifdef INET6 958 if (inp == 0) { 959 struct in6_addr s, d; 960 961 /* mapped addr case */ 962 bzero(&s, sizeof(s)); 963 s.s6_addr16[5] = htons(0xffff); 964 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 965 bzero(&d, sizeof(d)); 966 d.s6_addr16[5] = htons(0xffff); 967 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 968 in6p = in6_pcblookup_connect(&tcb6, &s, th->th_sport, 969 &d, th->th_dport, 0); 970 if (in6p == 0) { 971 ++tcpstat.tcps_pcbhashmiss; 972 in6p = in6_pcblookup_bind(&tcb6, &d, 973 th->th_dport, 0); 974 } 975 } 976 #endif 977 #ifndef INET6 978 if (inp == 0) 979 #else 980 if (inp == 0 && in6p == 0) 981 #endif 982 { 983 ++tcpstat.tcps_noport; 984 if (tcp_log_refused && (tiflags & TH_SYN)) { 985 char src[4*sizeof "123"]; 986 char dst[4*sizeof "123"]; 987 988 if (ip) { 989 strcpy(src, inet_ntoa(ip->ip_src)); 990 strcpy(dst, inet_ntoa(ip->ip_dst)); 991 } 992 else { 993 strcpy(src, "(unknown)"); 994 strcpy(dst, "(unknown)"); 995 } 996 log(LOG_INFO, 997 "Connection attempt to TCP %s:%d from %s:%d\n", 998 dst, ntohs(th->th_dport), 999 src, ntohs(th->th_sport)); 1000 } 1001 TCP_FIELDS_TO_HOST(th); 1002 goto dropwithreset_ratelim; 1003 } 1004 #ifdef IPSEC 1005 if (inp && ipsec4_in_reject(m, inp)) { 1006 ipsecstat.in_polvio++; 1007 goto drop; 1008 } 1009 #ifdef INET6 1010 else if (in6p && ipsec4_in_reject_so(m, in6p->in6p_socket)) { 1011 ipsecstat.in_polvio++; 1012 goto drop; 1013 } 1014 #endif 1015 #endif /*IPSEC*/ 1016 break; 1017 #endif /*INET*/ 1018 #ifdef INET6 1019 case AF_INET6: 1020 { 1021 int faith; 1022 1023 #if defined(NFAITH) && NFAITH > 0 1024 faith = faithprefix(&ip6->ip6_dst); 1025 #else 1026 faith = 0; 1027 #endif 1028 in6p = in6_pcblookup_connect(&tcb6, &ip6->ip6_src, th->th_sport, 1029 &ip6->ip6_dst, th->th_dport, faith); 1030 if (in6p == NULL) { 1031 ++tcpstat.tcps_pcbhashmiss; 1032 in6p = in6_pcblookup_bind(&tcb6, &ip6->ip6_dst, 1033 th->th_dport, faith); 1034 } 1035 if (in6p == NULL) { 1036 ++tcpstat.tcps_noport; 1037 if (tcp_log_refused && (tiflags & TH_SYN)) { 1038 char src[INET6_ADDRSTRLEN]; 1039 char dst[INET6_ADDRSTRLEN]; 1040 1041 if (ip6) { 1042 strcpy(src, ip6_sprintf(&ip6->ip6_src)); 1043 strcpy(dst, ip6_sprintf(&ip6->ip6_dst)); 1044 } 1045 else { 1046 strcpy(src, "(unknown v6)"); 1047 strcpy(dst, "(unknown v6)"); 1048 } 1049 log(LOG_INFO, 1050 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 1051 dst, ntohs(th->th_dport), 1052 src, ntohs(th->th_sport)); 1053 } 1054 TCP_FIELDS_TO_HOST(th); 1055 goto dropwithreset_ratelim; 1056 } 1057 #ifdef IPSEC 1058 if (ipsec6_in_reject(m, in6p)) { 1059 ipsec6stat.in_polvio++; 1060 goto drop; 1061 } 1062 #endif /*IPSEC*/ 1063 break; 1064 } 1065 #endif 1066 } 1067 1068 /* 1069 * If the state is CLOSED (i.e., TCB does not exist) then 1070 * all data in the incoming segment is discarded. 1071 * If the TCB exists but is in CLOSED state, it is embryonic, 1072 * but should either do a listen or a connect soon. 1073 */ 1074 tp = NULL; 1075 so = NULL; 1076 if (inp) { 1077 tp = intotcpcb(inp); 1078 so = inp->inp_socket; 1079 } 1080 #ifdef INET6 1081 else if (in6p) { 1082 tp = in6totcpcb(in6p); 1083 so = in6p->in6p_socket; 1084 } 1085 #endif 1086 if (tp == 0) { 1087 TCP_FIELDS_TO_HOST(th); 1088 goto dropwithreset_ratelim; 1089 } 1090 if (tp->t_state == TCPS_CLOSED) 1091 goto drop; 1092 1093 /* 1094 * Checksum extended TCP header and data. 1095 */ 1096 switch (af) { 1097 #ifdef INET 1098 case AF_INET: 1099 switch (m->m_pkthdr.csum_flags & 1100 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 1101 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 1102 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 1103 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 1104 goto badcsum; 1105 1106 case M_CSUM_TCPv4|M_CSUM_DATA: 1107 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 1108 if ((m->m_pkthdr.csum_data ^ 0xffff) != 0) 1109 goto badcsum; 1110 break; 1111 1112 case M_CSUM_TCPv4: 1113 /* Checksum was okay. */ 1114 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 1115 break; 1116 1117 default: 1118 /* Must compute it ourselves. */ 1119 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 1120 #ifndef PULLDOWN_TEST 1121 { 1122 struct ipovly *ipov; 1123 ipov = (struct ipovly *)ip; 1124 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 1125 ipov->ih_len = htons(tlen + off); 1126 1127 if (in_cksum(m, len) != 0) 1128 goto badcsum; 1129 } 1130 #else 1131 if (in4_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 1132 goto badcsum; 1133 #endif /* ! PULLDOWN_TEST */ 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 ostate = tp->t_state; 1205 1206 tcp_saveti = NULL; 1207 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1208 goto nosave; 1209 1210 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1211 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1212 if (!tcp_saveti) 1213 goto nosave; 1214 } else { 1215 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1216 if (!tcp_saveti) 1217 goto nosave; 1218 tcp_saveti->m_len = iphlen; 1219 m_copydata(m, 0, iphlen, 1220 mtod(tcp_saveti, caddr_t)); 1221 } 1222 1223 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1224 m_freem(tcp_saveti); 1225 tcp_saveti = NULL; 1226 } else { 1227 tcp_saveti->m_len += sizeof(struct tcphdr); 1228 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1229 sizeof(struct tcphdr)); 1230 } 1231 if (tcp_saveti) { 1232 /* 1233 * need to recover version # field, which was 1234 * overwritten on ip_cksum computation. 1235 */ 1236 struct ip *sip; 1237 sip = mtod(tcp_saveti, struct ip *); 1238 switch (af) { 1239 #ifdef INET 1240 case AF_INET: 1241 sip->ip_v = 4; 1242 break; 1243 #endif 1244 #ifdef INET6 1245 case AF_INET6: 1246 sip->ip_v = 6; 1247 break; 1248 #endif 1249 } 1250 } 1251 nosave:; 1252 } 1253 if (so->so_options & SO_ACCEPTCONN) { 1254 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1255 if (tiflags & TH_RST) { 1256 syn_cache_reset(&src.sa, &dst.sa, th); 1257 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1258 (TH_ACK|TH_SYN)) { 1259 /* 1260 * Received a SYN,ACK. This should 1261 * never happen while we are in 1262 * LISTEN. Send an RST. 1263 */ 1264 goto badsyn; 1265 } else if (tiflags & TH_ACK) { 1266 so = syn_cache_get(&src.sa, &dst.sa, 1267 th, toff, tlen, so, m); 1268 if (so == NULL) { 1269 /* 1270 * We don't have a SYN for 1271 * this ACK; send an RST. 1272 */ 1273 goto badsyn; 1274 } else if (so == 1275 (struct socket *)(-1)) { 1276 /* 1277 * We were unable to create 1278 * the connection. If the 1279 * 3-way handshake was 1280 * completed, and RST has 1281 * been sent to the peer. 1282 * Since the mbuf might be 1283 * in use for the reply, 1284 * do not free it. 1285 */ 1286 m = NULL; 1287 } else { 1288 /* 1289 * We have created a 1290 * full-blown connection. 1291 */ 1292 tp = NULL; 1293 inp = NULL; 1294 #ifdef INET6 1295 in6p = NULL; 1296 #endif 1297 switch (so->so_proto->pr_domain->dom_family) { 1298 #ifdef INET 1299 case AF_INET: 1300 inp = sotoinpcb(so); 1301 tp = intotcpcb(inp); 1302 break; 1303 #endif 1304 #ifdef INET6 1305 case AF_INET6: 1306 in6p = sotoin6pcb(so); 1307 tp = in6totcpcb(in6p); 1308 break; 1309 #endif 1310 } 1311 if (tp == NULL) 1312 goto badsyn; /*XXX*/ 1313 tiwin <<= tp->snd_scale; 1314 goto after_listen; 1315 } 1316 } else { 1317 /* 1318 * None of RST, SYN or ACK was set. 1319 * This is an invalid packet for a 1320 * TCB in LISTEN state. Send a RST. 1321 */ 1322 goto badsyn; 1323 } 1324 } else { 1325 /* 1326 * Received a SYN. 1327 */ 1328 1329 /* 1330 * LISTEN socket received a SYN 1331 * from itself? This can't possibly 1332 * be valid; drop the packet. 1333 */ 1334 if (th->th_sport == th->th_dport) { 1335 int i; 1336 1337 switch (af) { 1338 #ifdef INET 1339 case AF_INET: 1340 i = in_hosteq(ip->ip_src, ip->ip_dst); 1341 break; 1342 #endif 1343 #ifdef INET6 1344 case AF_INET6: 1345 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1346 break; 1347 #endif 1348 default: 1349 i = 1; 1350 } 1351 if (i) { 1352 tcpstat.tcps_badsyn++; 1353 goto drop; 1354 } 1355 } 1356 1357 /* 1358 * SYN looks ok; create compressed TCP 1359 * state for it. 1360 */ 1361 if (so->so_qlen <= so->so_qlimit && 1362 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1363 so, m, optp, optlen, &opti)) 1364 m = NULL; 1365 } 1366 goto drop; 1367 } 1368 } 1369 1370 after_listen: 1371 #ifdef DIAGNOSTIC 1372 /* 1373 * Should not happen now that all embryonic connections 1374 * are handled with compressed state. 1375 */ 1376 if (tp->t_state == TCPS_LISTEN) 1377 panic("tcp_input: TCPS_LISTEN"); 1378 #endif 1379 1380 /* 1381 * Segment received on connection. 1382 * Reset idle time and keep-alive timer. 1383 */ 1384 tp->t_rcvtime = tcp_now; 1385 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1386 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1387 1388 /* 1389 * Process options. 1390 */ 1391 if (optp) 1392 tcp_dooptions(tp, optp, optlen, th, &opti); 1393 1394 /* 1395 * Header prediction: check for the two common cases 1396 * of a uni-directional data xfer. If the packet has 1397 * no control flags, is in-sequence, the window didn't 1398 * change and we're not retransmitting, it's a 1399 * candidate. If the length is zero and the ack moved 1400 * forward, we're the sender side of the xfer. Just 1401 * free the data acked & wake any higher level process 1402 * that was blocked waiting for space. If the length 1403 * is non-zero and the ack didn't move, we're the 1404 * receiver side. If we're getting packets in-order 1405 * (the reassembly queue is empty), add the data to 1406 * the socket buffer and note that we need a delayed ack. 1407 */ 1408 if (tp->t_state == TCPS_ESTABLISHED && 1409 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1410 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1411 th->th_seq == tp->rcv_nxt && 1412 tiwin && tiwin == tp->snd_wnd && 1413 tp->snd_nxt == tp->snd_max) { 1414 1415 /* 1416 * If last ACK falls within this segment's sequence numbers, 1417 * record the timestamp. 1418 */ 1419 if (opti.ts_present && 1420 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1421 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1422 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1423 tp->ts_recent = opti.ts_val; 1424 } 1425 1426 if (tlen == 0) { 1427 if (SEQ_GT(th->th_ack, tp->snd_una) && 1428 SEQ_LEQ(th->th_ack, tp->snd_max) && 1429 tp->snd_cwnd >= tp->snd_wnd && 1430 tp->t_dupacks < tcprexmtthresh) { 1431 /* 1432 * this is a pure ack for outstanding data. 1433 */ 1434 ++tcpstat.tcps_predack; 1435 if (opti.ts_present && opti.ts_ecr) 1436 tcp_xmit_timer(tp, 1437 TCP_TIMESTAMP(tp) - opti.ts_ecr + 1); 1438 else if (tp->t_rtttime && 1439 SEQ_GT(th->th_ack, tp->t_rtseq)) 1440 tcp_xmit_timer(tp, 1441 tcp_now - tp->t_rtttime); 1442 acked = th->th_ack - tp->snd_una; 1443 tcpstat.tcps_rcvackpack++; 1444 tcpstat.tcps_rcvackbyte += acked; 1445 ND6_HINT(tp); 1446 sbdrop(&so->so_snd, acked); 1447 /* 1448 * We want snd_recover to track snd_una to 1449 * avoid sequence wraparound problems for 1450 * very large transfers. 1451 */ 1452 tp->snd_una = tp->snd_recover = th->th_ack; 1453 m_freem(m); 1454 1455 /* 1456 * If all outstanding data are acked, stop 1457 * retransmit timer, otherwise restart timer 1458 * using current (possibly backed-off) value. 1459 * If process is waiting for space, 1460 * wakeup/selwakeup/signal. If data 1461 * are ready to send, let tcp_output 1462 * decide between more output or persist. 1463 */ 1464 if (tp->snd_una == tp->snd_max) 1465 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1466 else if (TCP_TIMER_ISARMED(tp, 1467 TCPT_PERSIST) == 0) 1468 TCP_TIMER_ARM(tp, TCPT_REXMT, 1469 tp->t_rxtcur); 1470 1471 sowwakeup(so); 1472 if (so->so_snd.sb_cc) 1473 (void) tcp_output(tp); 1474 if (tcp_saveti) 1475 m_freem(tcp_saveti); 1476 return; 1477 } 1478 } else if (th->th_ack == tp->snd_una && 1479 TAILQ_FIRST(&tp->segq) == NULL && 1480 tlen <= sbspace(&so->so_rcv)) { 1481 /* 1482 * this is a pure, in-sequence data packet 1483 * with nothing on the reassembly queue and 1484 * we have enough buffer space to take it. 1485 */ 1486 ++tcpstat.tcps_preddat; 1487 tp->rcv_nxt += tlen; 1488 tcpstat.tcps_rcvpack++; 1489 tcpstat.tcps_rcvbyte += tlen; 1490 ND6_HINT(tp); 1491 /* 1492 * Drop TCP, IP headers and TCP options then add data 1493 * to socket buffer. 1494 */ 1495 m_adj(m, toff + off); 1496 sbappend(&so->so_rcv, m); 1497 sorwakeup(so); 1498 TCP_SETUP_ACK(tp, th); 1499 if (tp->t_flags & TF_ACKNOW) 1500 (void) tcp_output(tp); 1501 if (tcp_saveti) 1502 m_freem(tcp_saveti); 1503 return; 1504 } 1505 } 1506 1507 /* 1508 * Compute mbuf offset to TCP data segment. 1509 */ 1510 hdroptlen = toff + off; 1511 1512 /* 1513 * Calculate amount of space in receive window, 1514 * and then do TCP input processing. 1515 * Receive window is amount of space in rcv queue, 1516 * but not less than advertised window. 1517 */ 1518 { int win; 1519 1520 win = sbspace(&so->so_rcv); 1521 if (win < 0) 1522 win = 0; 1523 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1524 } 1525 1526 switch (tp->t_state) { 1527 case TCPS_LISTEN: 1528 /* 1529 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1530 */ 1531 if (m->m_flags & (M_BCAST|M_MCAST)) 1532 goto drop; 1533 switch (af) { 1534 #ifdef INET6 1535 case AF_INET6: 1536 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1537 goto drop; 1538 break; 1539 #endif /* INET6 */ 1540 case AF_INET: 1541 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1542 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1543 goto drop; 1544 break; 1545 } 1546 break; 1547 1548 /* 1549 * If the state is SYN_SENT: 1550 * if seg contains an ACK, but not for our SYN, drop the input. 1551 * if seg contains a RST, then drop the connection. 1552 * if seg does not contain SYN, then drop it. 1553 * Otherwise this is an acceptable SYN segment 1554 * initialize tp->rcv_nxt and tp->irs 1555 * if seg contains ack then advance tp->snd_una 1556 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1557 * arrange for segment to be acked (eventually) 1558 * continue processing rest of data/controls, beginning with URG 1559 */ 1560 case TCPS_SYN_SENT: 1561 if ((tiflags & TH_ACK) && 1562 (SEQ_LEQ(th->th_ack, tp->iss) || 1563 SEQ_GT(th->th_ack, tp->snd_max))) 1564 goto dropwithreset; 1565 if (tiflags & TH_RST) { 1566 if (tiflags & TH_ACK) 1567 tp = tcp_drop(tp, ECONNREFUSED); 1568 goto drop; 1569 } 1570 if ((tiflags & TH_SYN) == 0) 1571 goto drop; 1572 if (tiflags & TH_ACK) { 1573 tp->snd_una = tp->snd_recover = th->th_ack; 1574 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1575 tp->snd_nxt = tp->snd_una; 1576 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1577 } 1578 tp->irs = th->th_seq; 1579 tcp_rcvseqinit(tp); 1580 tp->t_flags |= TF_ACKNOW; 1581 tcp_mss_from_peer(tp, opti.maxseg); 1582 1583 /* 1584 * Initialize the initial congestion window. If we 1585 * had to retransmit the SYN, we must initialize cwnd 1586 * to 1 segment (i.e. the Loss Window). 1587 */ 1588 if (tp->t_flags & TF_SYN_REXMT) 1589 tp->snd_cwnd = tp->t_peermss; 1590 else 1591 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, 1592 tp->t_peermss); 1593 1594 tcp_rmx_rtt(tp); 1595 if (tiflags & TH_ACK) { 1596 tcpstat.tcps_connects++; 1597 soisconnected(so); 1598 tcp_established(tp); 1599 /* Do window scaling on this connection? */ 1600 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1601 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1602 tp->snd_scale = tp->requested_s_scale; 1603 tp->rcv_scale = tp->request_r_scale; 1604 } 1605 TCP_REASS_LOCK(tp); 1606 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1607 TCP_REASS_UNLOCK(tp); 1608 /* 1609 * if we didn't have to retransmit the SYN, 1610 * use its rtt as our initial srtt & rtt var. 1611 */ 1612 if (tp->t_rtttime) 1613 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1614 } else 1615 tp->t_state = TCPS_SYN_RECEIVED; 1616 1617 /* 1618 * Advance th->th_seq to correspond to first data byte. 1619 * If data, trim to stay within window, 1620 * dropping FIN if necessary. 1621 */ 1622 th->th_seq++; 1623 if (tlen > tp->rcv_wnd) { 1624 todrop = tlen - tp->rcv_wnd; 1625 m_adj(m, -todrop); 1626 tlen = tp->rcv_wnd; 1627 tiflags &= ~TH_FIN; 1628 tcpstat.tcps_rcvpackafterwin++; 1629 tcpstat.tcps_rcvbyteafterwin += todrop; 1630 } 1631 tp->snd_wl1 = th->th_seq - 1; 1632 tp->rcv_up = th->th_seq; 1633 goto step6; 1634 1635 /* 1636 * If the state is SYN_RECEIVED: 1637 * If seg contains an ACK, but not for our SYN, drop the input 1638 * and generate an RST. See page 36, rfc793 1639 */ 1640 case TCPS_SYN_RECEIVED: 1641 if ((tiflags & TH_ACK) && 1642 (SEQ_LEQ(th->th_ack, tp->iss) || 1643 SEQ_GT(th->th_ack, tp->snd_max))) 1644 goto dropwithreset; 1645 break; 1646 } 1647 1648 /* 1649 * States other than LISTEN or SYN_SENT. 1650 * First check timestamp, if present. 1651 * Then check that at least some bytes of segment are within 1652 * receive window. If segment begins before rcv_nxt, 1653 * drop leading data (and SYN); if nothing left, just ack. 1654 * 1655 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1656 * and it's less than ts_recent, drop it. 1657 */ 1658 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1659 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1660 1661 /* Check to see if ts_recent is over 24 days old. */ 1662 if ((int)(TCP_TIMESTAMP(tp) - tp->ts_recent_age) > 1663 TCP_PAWS_IDLE) { 1664 /* 1665 * Invalidate ts_recent. If this segment updates 1666 * ts_recent, the age will be reset later and ts_recent 1667 * will get a valid value. If it does not, setting 1668 * ts_recent to zero will at least satisfy the 1669 * requirement that zero be placed in the timestamp 1670 * echo reply when ts_recent isn't valid. The 1671 * age isn't reset until we get a valid ts_recent 1672 * because we don't want out-of-order segments to be 1673 * dropped when ts_recent is old. 1674 */ 1675 tp->ts_recent = 0; 1676 } else { 1677 tcpstat.tcps_rcvduppack++; 1678 tcpstat.tcps_rcvdupbyte += tlen; 1679 tcpstat.tcps_pawsdrop++; 1680 goto dropafterack; 1681 } 1682 } 1683 1684 todrop = tp->rcv_nxt - th->th_seq; 1685 if (todrop > 0) { 1686 if (tiflags & TH_SYN) { 1687 tiflags &= ~TH_SYN; 1688 th->th_seq++; 1689 if (th->th_urp > 1) 1690 th->th_urp--; 1691 else { 1692 tiflags &= ~TH_URG; 1693 th->th_urp = 0; 1694 } 1695 todrop--; 1696 } 1697 if (todrop > tlen || 1698 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1699 /* 1700 * Any valid FIN must be to the left of the window. 1701 * At this point the FIN must be a duplicate or 1702 * out of sequence; drop it. 1703 */ 1704 tiflags &= ~TH_FIN; 1705 /* 1706 * Send an ACK to resynchronize and drop any data. 1707 * But keep on processing for RST or ACK. 1708 */ 1709 tp->t_flags |= TF_ACKNOW; 1710 todrop = tlen; 1711 tcpstat.tcps_rcvdupbyte += todrop; 1712 tcpstat.tcps_rcvduppack++; 1713 } else { 1714 tcpstat.tcps_rcvpartduppack++; 1715 tcpstat.tcps_rcvpartdupbyte += todrop; 1716 } 1717 hdroptlen += todrop; /*drop from head afterwards*/ 1718 th->th_seq += todrop; 1719 tlen -= todrop; 1720 if (th->th_urp > todrop) 1721 th->th_urp -= todrop; 1722 else { 1723 tiflags &= ~TH_URG; 1724 th->th_urp = 0; 1725 } 1726 } 1727 1728 /* 1729 * If new data are received on a connection after the 1730 * user processes are gone, then RST the other end. 1731 */ 1732 if ((so->so_state & SS_NOFDREF) && 1733 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1734 tp = tcp_close(tp); 1735 tcpstat.tcps_rcvafterclose++; 1736 goto dropwithreset; 1737 } 1738 1739 /* 1740 * If segment ends after window, drop trailing data 1741 * (and PUSH and FIN); if nothing left, just ACK. 1742 */ 1743 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1744 if (todrop > 0) { 1745 tcpstat.tcps_rcvpackafterwin++; 1746 if (todrop >= tlen) { 1747 tcpstat.tcps_rcvbyteafterwin += tlen; 1748 /* 1749 * If a new connection request is received 1750 * while in TIME_WAIT, drop the old connection 1751 * and start over if the sequence numbers 1752 * are above the previous ones. 1753 */ 1754 if (tiflags & TH_SYN && 1755 tp->t_state == TCPS_TIME_WAIT && 1756 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1757 iss = tcp_new_iss(tp, tp->snd_nxt); 1758 tp = tcp_close(tp); 1759 goto findpcb; 1760 } 1761 /* 1762 * If window is closed can only take segments at 1763 * window edge, and have to drop data and PUSH from 1764 * incoming segments. Continue processing, but 1765 * remember to ack. Otherwise, drop segment 1766 * and ack. 1767 */ 1768 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1769 tp->t_flags |= TF_ACKNOW; 1770 tcpstat.tcps_rcvwinprobe++; 1771 } else 1772 goto dropafterack; 1773 } else 1774 tcpstat.tcps_rcvbyteafterwin += todrop; 1775 m_adj(m, -todrop); 1776 tlen -= todrop; 1777 tiflags &= ~(TH_PUSH|TH_FIN); 1778 } 1779 1780 /* 1781 * If last ACK falls within this segment's sequence numbers, 1782 * and the timestamp is newer, record it. 1783 */ 1784 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1785 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1786 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1787 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1788 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1789 tp->ts_recent = opti.ts_val; 1790 } 1791 1792 /* 1793 * If the RST bit is set examine the state: 1794 * SYN_RECEIVED STATE: 1795 * If passive open, return to LISTEN state. 1796 * If active open, inform user that connection was refused. 1797 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1798 * Inform user that connection was reset, and close tcb. 1799 * CLOSING, LAST_ACK, TIME_WAIT STATES 1800 * Close the tcb. 1801 */ 1802 if (tiflags&TH_RST) switch (tp->t_state) { 1803 1804 case TCPS_SYN_RECEIVED: 1805 so->so_error = ECONNREFUSED; 1806 goto close; 1807 1808 case TCPS_ESTABLISHED: 1809 case TCPS_FIN_WAIT_1: 1810 case TCPS_FIN_WAIT_2: 1811 case TCPS_CLOSE_WAIT: 1812 so->so_error = ECONNRESET; 1813 close: 1814 tp->t_state = TCPS_CLOSED; 1815 tcpstat.tcps_drops++; 1816 tp = tcp_close(tp); 1817 goto drop; 1818 1819 case TCPS_CLOSING: 1820 case TCPS_LAST_ACK: 1821 case TCPS_TIME_WAIT: 1822 tp = tcp_close(tp); 1823 goto drop; 1824 } 1825 1826 /* 1827 * If a SYN is in the window, then this is an 1828 * error and we send an RST and drop the connection. 1829 */ 1830 if (tiflags & TH_SYN) { 1831 tp = tcp_drop(tp, ECONNRESET); 1832 goto dropwithreset; 1833 } 1834 1835 /* 1836 * If the ACK bit is off we drop the segment and return. 1837 */ 1838 if ((tiflags & TH_ACK) == 0) { 1839 if (tp->t_flags & TF_ACKNOW) 1840 goto dropafterack; 1841 else 1842 goto drop; 1843 } 1844 1845 /* 1846 * Ack processing. 1847 */ 1848 switch (tp->t_state) { 1849 1850 /* 1851 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1852 * ESTABLISHED state and continue processing, otherwise 1853 * send an RST. 1854 */ 1855 case TCPS_SYN_RECEIVED: 1856 if (SEQ_GT(tp->snd_una, th->th_ack) || 1857 SEQ_GT(th->th_ack, tp->snd_max)) 1858 goto dropwithreset; 1859 tcpstat.tcps_connects++; 1860 soisconnected(so); 1861 tcp_established(tp); 1862 /* Do window scaling? */ 1863 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1864 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1865 tp->snd_scale = tp->requested_s_scale; 1866 tp->rcv_scale = tp->request_r_scale; 1867 } 1868 TCP_REASS_LOCK(tp); 1869 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1870 TCP_REASS_UNLOCK(tp); 1871 tp->snd_wl1 = th->th_seq - 1; 1872 /* fall into ... */ 1873 1874 /* 1875 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1876 * ACKs. If the ack is in the range 1877 * tp->snd_una < th->th_ack <= tp->snd_max 1878 * then advance tp->snd_una to th->th_ack and drop 1879 * data from the retransmission queue. If this ACK reflects 1880 * more up to date window information we update our window information. 1881 */ 1882 case TCPS_ESTABLISHED: 1883 case TCPS_FIN_WAIT_1: 1884 case TCPS_FIN_WAIT_2: 1885 case TCPS_CLOSE_WAIT: 1886 case TCPS_CLOSING: 1887 case TCPS_LAST_ACK: 1888 case TCPS_TIME_WAIT: 1889 1890 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1891 if (tlen == 0 && tiwin == tp->snd_wnd) { 1892 tcpstat.tcps_rcvdupack++; 1893 /* 1894 * If we have outstanding data (other than 1895 * a window probe), this is a completely 1896 * duplicate ack (ie, window info didn't 1897 * change), the ack is the biggest we've 1898 * seen and we've seen exactly our rexmt 1899 * threshhold of them, assume a packet 1900 * has been dropped and retransmit it. 1901 * Kludge snd_nxt & the congestion 1902 * window so we send only this one 1903 * packet. 1904 * 1905 * We know we're losing at the current 1906 * window size so do congestion avoidance 1907 * (set ssthresh to half the current window 1908 * and pull our congestion window back to 1909 * the new ssthresh). 1910 * 1911 * Dup acks mean that packets have left the 1912 * network (they're now cached at the receiver) 1913 * so bump cwnd by the amount in the receiver 1914 * to keep a constant cwnd packets in the 1915 * network. 1916 */ 1917 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 1918 th->th_ack != tp->snd_una) 1919 tp->t_dupacks = 0; 1920 else if (++tp->t_dupacks == tcprexmtthresh) { 1921 tcp_seq onxt = tp->snd_nxt; 1922 u_int win = 1923 min(tp->snd_wnd, tp->snd_cwnd) / 1924 2 / tp->t_segsz; 1925 if (tcp_do_newreno && SEQ_LT(th->th_ack, 1926 tp->snd_recover)) { 1927 /* 1928 * False fast retransmit after 1929 * timeout. Do not cut window. 1930 */ 1931 tp->snd_cwnd += tp->t_segsz; 1932 tp->t_dupacks = 0; 1933 (void) tcp_output(tp); 1934 goto drop; 1935 } 1936 1937 if (win < 2) 1938 win = 2; 1939 tp->snd_ssthresh = win * tp->t_segsz; 1940 tp->snd_recover = tp->snd_max; 1941 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1942 tp->t_rtttime = 0; 1943 tp->snd_nxt = th->th_ack; 1944 tp->snd_cwnd = tp->t_segsz; 1945 (void) tcp_output(tp); 1946 tp->snd_cwnd = tp->snd_ssthresh + 1947 tp->t_segsz * tp->t_dupacks; 1948 if (SEQ_GT(onxt, tp->snd_nxt)) 1949 tp->snd_nxt = onxt; 1950 goto drop; 1951 } else if (tp->t_dupacks > tcprexmtthresh) { 1952 tp->snd_cwnd += tp->t_segsz; 1953 (void) tcp_output(tp); 1954 goto drop; 1955 } 1956 } else 1957 tp->t_dupacks = 0; 1958 break; 1959 } 1960 /* 1961 * If the congestion window was inflated to account 1962 * for the other side's cached packets, retract it. 1963 */ 1964 if (tcp_do_newreno == 0) { 1965 if (tp->t_dupacks >= tcprexmtthresh && 1966 tp->snd_cwnd > tp->snd_ssthresh) 1967 tp->snd_cwnd = tp->snd_ssthresh; 1968 tp->t_dupacks = 0; 1969 } else if (tp->t_dupacks >= tcprexmtthresh && 1970 tcp_newreno(tp, th) == 0) { 1971 tp->snd_cwnd = tp->snd_ssthresh; 1972 /* 1973 * Window inflation should have left us with approx. 1974 * snd_ssthresh outstanding data. But in case we 1975 * would be inclined to send a burst, better to do 1976 * it via the slow start mechanism. 1977 */ 1978 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 1979 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 1980 + tp->t_segsz; 1981 tp->t_dupacks = 0; 1982 } 1983 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1984 tcpstat.tcps_rcvacktoomuch++; 1985 goto dropafterack; 1986 } 1987 acked = th->th_ack - tp->snd_una; 1988 tcpstat.tcps_rcvackpack++; 1989 tcpstat.tcps_rcvackbyte += acked; 1990 1991 /* 1992 * If we have a timestamp reply, update smoothed 1993 * round trip time. If no timestamp is present but 1994 * transmit timer is running and timed sequence 1995 * number was acked, update smoothed round trip time. 1996 * Since we now have an rtt measurement, cancel the 1997 * timer backoff (cf., Phil Karn's retransmit alg.). 1998 * Recompute the initial retransmit timer. 1999 */ 2000 if (opti.ts_present && opti.ts_ecr) 2001 tcp_xmit_timer(tp, TCP_TIMESTAMP(tp) - opti.ts_ecr + 1); 2002 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2003 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2004 2005 /* 2006 * If all outstanding data is acked, stop retransmit 2007 * timer and remember to restart (more output or persist). 2008 * If there is more data to be acked, restart retransmit 2009 * timer, using current (possibly backed-off) value. 2010 */ 2011 if (th->th_ack == tp->snd_max) { 2012 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2013 needoutput = 1; 2014 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2015 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2016 /* 2017 * When new data is acked, open the congestion window. 2018 * If the window gives us less than ssthresh packets 2019 * in flight, open exponentially (segsz per packet). 2020 * Otherwise open linearly: segsz per window 2021 * (segsz^2 / cwnd per packet), plus a constant 2022 * fraction of a packet (segsz/8) to help larger windows 2023 * open quickly enough. 2024 */ 2025 { 2026 u_int cw = tp->snd_cwnd; 2027 u_int incr = tp->t_segsz; 2028 2029 if (cw > tp->snd_ssthresh) 2030 incr = incr * incr / cw; 2031 if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover)) 2032 tp->snd_cwnd = min(cw + incr, 2033 TCP_MAXWIN << tp->snd_scale); 2034 } 2035 ND6_HINT(tp); 2036 if (acked > so->so_snd.sb_cc) { 2037 tp->snd_wnd -= so->so_snd.sb_cc; 2038 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2039 ourfinisacked = 1; 2040 } else { 2041 sbdrop(&so->so_snd, acked); 2042 tp->snd_wnd -= acked; 2043 ourfinisacked = 0; 2044 } 2045 sowwakeup(so); 2046 /* 2047 * We want snd_recover to track snd_una to 2048 * avoid sequence wraparound problems for 2049 * very large transfers. 2050 */ 2051 tp->snd_una = tp->snd_recover = th->th_ack; 2052 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2053 tp->snd_nxt = tp->snd_una; 2054 2055 switch (tp->t_state) { 2056 2057 /* 2058 * In FIN_WAIT_1 STATE in addition to the processing 2059 * for the ESTABLISHED state if our FIN is now acknowledged 2060 * then enter FIN_WAIT_2. 2061 */ 2062 case TCPS_FIN_WAIT_1: 2063 if (ourfinisacked) { 2064 /* 2065 * If we can't receive any more 2066 * data, then closing user can proceed. 2067 * Starting the timer is contrary to the 2068 * specification, but if we don't get a FIN 2069 * we'll hang forever. 2070 */ 2071 if (so->so_state & SS_CANTRCVMORE) { 2072 soisdisconnected(so); 2073 if (tcp_maxidle > 0) 2074 TCP_TIMER_ARM(tp, TCPT_2MSL, 2075 tcp_maxidle); 2076 } 2077 tp->t_state = TCPS_FIN_WAIT_2; 2078 } 2079 break; 2080 2081 /* 2082 * In CLOSING STATE in addition to the processing for 2083 * the ESTABLISHED state if the ACK acknowledges our FIN 2084 * then enter the TIME-WAIT state, otherwise ignore 2085 * the segment. 2086 */ 2087 case TCPS_CLOSING: 2088 if (ourfinisacked) { 2089 tp->t_state = TCPS_TIME_WAIT; 2090 tcp_canceltimers(tp); 2091 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2092 soisdisconnected(so); 2093 } 2094 break; 2095 2096 /* 2097 * In LAST_ACK, we may still be waiting for data to drain 2098 * and/or to be acked, as well as for the ack of our FIN. 2099 * If our FIN is now acknowledged, delete the TCB, 2100 * enter the closed state and return. 2101 */ 2102 case TCPS_LAST_ACK: 2103 if (ourfinisacked) { 2104 tp = tcp_close(tp); 2105 goto drop; 2106 } 2107 break; 2108 2109 /* 2110 * In TIME_WAIT state the only thing that should arrive 2111 * is a retransmission of the remote FIN. Acknowledge 2112 * it and restart the finack timer. 2113 */ 2114 case TCPS_TIME_WAIT: 2115 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2116 goto dropafterack; 2117 } 2118 } 2119 2120 step6: 2121 /* 2122 * Update window information. 2123 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2124 */ 2125 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2126 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 2127 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 2128 /* keep track of pure window updates */ 2129 if (tlen == 0 && 2130 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2131 tcpstat.tcps_rcvwinupd++; 2132 tp->snd_wnd = tiwin; 2133 tp->snd_wl1 = th->th_seq; 2134 tp->snd_wl2 = th->th_ack; 2135 if (tp->snd_wnd > tp->max_sndwnd) 2136 tp->max_sndwnd = tp->snd_wnd; 2137 needoutput = 1; 2138 } 2139 2140 /* 2141 * Process segments with URG. 2142 */ 2143 if ((tiflags & TH_URG) && th->th_urp && 2144 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2145 /* 2146 * This is a kludge, but if we receive and accept 2147 * random urgent pointers, we'll crash in 2148 * soreceive. It's hard to imagine someone 2149 * actually wanting to send this much urgent data. 2150 */ 2151 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2152 th->th_urp = 0; /* XXX */ 2153 tiflags &= ~TH_URG; /* XXX */ 2154 goto dodata; /* XXX */ 2155 } 2156 /* 2157 * If this segment advances the known urgent pointer, 2158 * then mark the data stream. This should not happen 2159 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2160 * a FIN has been received from the remote side. 2161 * In these states we ignore the URG. 2162 * 2163 * According to RFC961 (Assigned Protocols), 2164 * the urgent pointer points to the last octet 2165 * of urgent data. We continue, however, 2166 * to consider it to indicate the first octet 2167 * of data past the urgent section as the original 2168 * spec states (in one of two places). 2169 */ 2170 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2171 tp->rcv_up = th->th_seq + th->th_urp; 2172 so->so_oobmark = so->so_rcv.sb_cc + 2173 (tp->rcv_up - tp->rcv_nxt) - 1; 2174 if (so->so_oobmark == 0) 2175 so->so_state |= SS_RCVATMARK; 2176 sohasoutofband(so); 2177 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2178 } 2179 /* 2180 * Remove out of band data so doesn't get presented to user. 2181 * This can happen independent of advancing the URG pointer, 2182 * but if two URG's are pending at once, some out-of-band 2183 * data may creep in... ick. 2184 */ 2185 if (th->th_urp <= (u_int16_t) tlen 2186 #ifdef SO_OOBINLINE 2187 && (so->so_options & SO_OOBINLINE) == 0 2188 #endif 2189 ) 2190 tcp_pulloutofband(so, th, m, hdroptlen); 2191 } else 2192 /* 2193 * If no out of band data is expected, 2194 * pull receive urgent pointer along 2195 * with the receive window. 2196 */ 2197 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2198 tp->rcv_up = tp->rcv_nxt; 2199 dodata: /* XXX */ 2200 2201 /* 2202 * Process the segment text, merging it into the TCP sequencing queue, 2203 * and arranging for acknowledgement of receipt if necessary. 2204 * This process logically involves adjusting tp->rcv_wnd as data 2205 * is presented to the user (this happens in tcp_usrreq.c, 2206 * case PRU_RCVD). If a FIN has already been received on this 2207 * connection then we just ignore the text. 2208 */ 2209 if ((tlen || (tiflags & TH_FIN)) && 2210 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2211 /* 2212 * Insert segment ti into reassembly queue of tcp with 2213 * control block tp. Return TH_FIN if reassembly now includes 2214 * a segment with FIN. The macro form does the common case 2215 * inline (segment is the next to be received on an 2216 * established connection, and the queue is empty), 2217 * avoiding linkage into and removal from the queue and 2218 * repetition of various conversions. 2219 * Set DELACK for segments received in order, but ack 2220 * immediately when segments are out of order 2221 * (so fast retransmit can work). 2222 */ 2223 /* NOTE: this was TCP_REASS() macro, but used only once */ 2224 TCP_REASS_LOCK(tp); 2225 if (th->th_seq == tp->rcv_nxt && 2226 TAILQ_FIRST(&tp->segq) == NULL && 2227 tp->t_state == TCPS_ESTABLISHED) { 2228 TCP_SETUP_ACK(tp, th); 2229 tp->rcv_nxt += tlen; 2230 tiflags = th->th_flags & TH_FIN; 2231 tcpstat.tcps_rcvpack++; 2232 tcpstat.tcps_rcvbyte += tlen; 2233 ND6_HINT(tp); 2234 m_adj(m, hdroptlen); 2235 sbappend(&(so)->so_rcv, m); 2236 sorwakeup(so); 2237 } else { 2238 m_adj(m, hdroptlen); 2239 tiflags = tcp_reass(tp, th, m, &tlen); 2240 tp->t_flags |= TF_ACKNOW; 2241 } 2242 TCP_REASS_UNLOCK(tp); 2243 2244 /* 2245 * Note the amount of data that peer has sent into 2246 * our window, in order to estimate the sender's 2247 * buffer size. 2248 */ 2249 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2250 } else { 2251 m_freem(m); 2252 m = NULL; 2253 tiflags &= ~TH_FIN; 2254 } 2255 2256 /* 2257 * If FIN is received ACK the FIN and let the user know 2258 * that the connection is closing. Ignore a FIN received before 2259 * the connection is fully established. 2260 */ 2261 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2262 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2263 socantrcvmore(so); 2264 tp->t_flags |= TF_ACKNOW; 2265 tp->rcv_nxt++; 2266 } 2267 switch (tp->t_state) { 2268 2269 /* 2270 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2271 */ 2272 case TCPS_ESTABLISHED: 2273 tp->t_state = TCPS_CLOSE_WAIT; 2274 break; 2275 2276 /* 2277 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2278 * enter the CLOSING state. 2279 */ 2280 case TCPS_FIN_WAIT_1: 2281 tp->t_state = TCPS_CLOSING; 2282 break; 2283 2284 /* 2285 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2286 * starting the time-wait timer, turning off the other 2287 * standard timers. 2288 */ 2289 case TCPS_FIN_WAIT_2: 2290 tp->t_state = TCPS_TIME_WAIT; 2291 tcp_canceltimers(tp); 2292 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2293 soisdisconnected(so); 2294 break; 2295 2296 /* 2297 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2298 */ 2299 case TCPS_TIME_WAIT: 2300 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2301 break; 2302 } 2303 } 2304 #ifdef TCP_DEBUG 2305 if (so->so_options & SO_DEBUG) 2306 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2307 #endif 2308 2309 /* 2310 * Return any desired output. 2311 */ 2312 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2313 (void) tcp_output(tp); 2314 if (tcp_saveti) 2315 m_freem(tcp_saveti); 2316 return; 2317 2318 badsyn: 2319 /* 2320 * Received a bad SYN. Increment counters and dropwithreset. 2321 */ 2322 tcpstat.tcps_badsyn++; 2323 tp = NULL; 2324 goto dropwithreset; 2325 2326 dropafterack: 2327 /* 2328 * Generate an ACK dropping incoming segment if it occupies 2329 * sequence space, where the ACK reflects our state. 2330 */ 2331 if (tiflags & TH_RST) 2332 goto drop; 2333 m_freem(m); 2334 tp->t_flags |= TF_ACKNOW; 2335 (void) tcp_output(tp); 2336 if (tcp_saveti) 2337 m_freem(tcp_saveti); 2338 return; 2339 2340 dropwithreset_ratelim: 2341 /* 2342 * We may want to rate-limit RSTs in certain situations, 2343 * particularly if we are sending an RST in response to 2344 * an attempt to connect to or otherwise communicate with 2345 * a port for which we have no socket. 2346 */ 2347 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2348 tcp_rst_ppslim) == 0) { 2349 /* XXX stat */ 2350 goto drop; 2351 } 2352 /* ...fall into dropwithreset... */ 2353 2354 dropwithreset: 2355 /* 2356 * Generate a RST, dropping incoming segment. 2357 * Make ACK acceptable to originator of segment. 2358 */ 2359 if (tiflags & TH_RST) 2360 goto drop; 2361 2362 switch (af) { 2363 #ifdef INET6 2364 case AF_INET6: 2365 /* For following calls to tcp_respond */ 2366 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2367 goto drop; 2368 break; 2369 #endif /* INET6 */ 2370 case AF_INET: 2371 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2372 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2373 goto drop; 2374 } 2375 2376 { 2377 /* 2378 * need to recover version # field, which was overwritten on 2379 * ip_cksum computation. 2380 */ 2381 struct ip *sip; 2382 sip = mtod(m, struct ip *); 2383 switch (af) { 2384 #ifdef INET 2385 case AF_INET: 2386 sip->ip_v = 4; 2387 break; 2388 #endif 2389 #ifdef INET6 2390 case AF_INET6: 2391 sip->ip_v = 6; 2392 break; 2393 #endif 2394 } 2395 } 2396 if (tiflags & TH_ACK) 2397 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2398 else { 2399 if (tiflags & TH_SYN) 2400 tlen++; 2401 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2402 TH_RST|TH_ACK); 2403 } 2404 if (tcp_saveti) 2405 m_freem(tcp_saveti); 2406 return; 2407 2408 badcsum: 2409 tcpstat.tcps_rcvbadsum++; 2410 drop: 2411 /* 2412 * Drop space held by incoming segment and return. 2413 */ 2414 if (tp) { 2415 if (tp->t_inpcb) 2416 so = tp->t_inpcb->inp_socket; 2417 #ifdef INET6 2418 else if (tp->t_in6pcb) 2419 so = tp->t_in6pcb->in6p_socket; 2420 #endif 2421 else 2422 so = NULL; 2423 #ifdef TCP_DEBUG 2424 if (so && (so->so_options & SO_DEBUG) != 0) 2425 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2426 #endif 2427 } 2428 if (tcp_saveti) 2429 m_freem(tcp_saveti); 2430 m_freem(m); 2431 return; 2432 } 2433 2434 void 2435 tcp_dooptions(tp, cp, cnt, th, oi) 2436 struct tcpcb *tp; 2437 u_char *cp; 2438 int cnt; 2439 struct tcphdr *th; 2440 struct tcp_opt_info *oi; 2441 { 2442 u_int16_t mss; 2443 int opt, optlen; 2444 2445 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2446 opt = cp[0]; 2447 if (opt == TCPOPT_EOL) 2448 break; 2449 if (opt == TCPOPT_NOP) 2450 optlen = 1; 2451 else { 2452 if (cnt < 2) 2453 break; 2454 optlen = cp[1]; 2455 if (optlen < 2 || optlen > cnt) 2456 break; 2457 } 2458 switch (opt) { 2459 2460 default: 2461 continue; 2462 2463 case TCPOPT_MAXSEG: 2464 if (optlen != TCPOLEN_MAXSEG) 2465 continue; 2466 if (!(th->th_flags & TH_SYN)) 2467 continue; 2468 bcopy(cp + 2, &mss, sizeof(mss)); 2469 oi->maxseg = ntohs(mss); 2470 break; 2471 2472 case TCPOPT_WINDOW: 2473 if (optlen != TCPOLEN_WINDOW) 2474 continue; 2475 if (!(th->th_flags & TH_SYN)) 2476 continue; 2477 tp->t_flags |= TF_RCVD_SCALE; 2478 tp->requested_s_scale = cp[2]; 2479 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2480 #if 0 /*XXX*/ 2481 char *p; 2482 2483 if (ip) 2484 p = ntohl(ip->ip_src); 2485 #ifdef INET6 2486 else if (ip6) 2487 p = ip6_sprintf(&ip6->ip6_src); 2488 #endif 2489 else 2490 p = "(unknown)"; 2491 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2492 "assuming %d\n", 2493 tp->requested_s_scale, p, 2494 TCP_MAX_WINSHIFT); 2495 #else 2496 log(LOG_ERR, "TCP: invalid wscale %d, " 2497 "assuming %d\n", 2498 tp->requested_s_scale, 2499 TCP_MAX_WINSHIFT); 2500 #endif 2501 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2502 } 2503 break; 2504 2505 case TCPOPT_TIMESTAMP: 2506 if (optlen != TCPOLEN_TIMESTAMP) 2507 continue; 2508 oi->ts_present = 1; 2509 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2510 NTOHL(oi->ts_val); 2511 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2512 NTOHL(oi->ts_ecr); 2513 2514 /* 2515 * A timestamp received in a SYN makes 2516 * it ok to send timestamp requests and replies. 2517 */ 2518 if (th->th_flags & TH_SYN) { 2519 tp->t_flags |= TF_RCVD_TSTMP; 2520 tp->ts_recent = oi->ts_val; 2521 tp->ts_recent_age = TCP_TIMESTAMP(tp); 2522 } 2523 break; 2524 case TCPOPT_SACK_PERMITTED: 2525 if (optlen != TCPOLEN_SACK_PERMITTED) 2526 continue; 2527 if (!(th->th_flags & TH_SYN)) 2528 continue; 2529 tp->t_flags &= ~TF_CANT_TXSACK; 2530 break; 2531 2532 case TCPOPT_SACK: 2533 if (tp->t_flags & TF_IGNR_RXSACK) 2534 continue; 2535 if (optlen % 8 != 2 || optlen < 10) 2536 continue; 2537 cp += 2; 2538 optlen -= 2; 2539 for (; optlen > 0; cp -= 8, optlen -= 8) { 2540 tcp_seq lwe, rwe; 2541 bcopy((char *)cp, (char *) &lwe, sizeof(lwe)); 2542 NTOHL(lwe); 2543 bcopy((char *)cp, (char *) &rwe, sizeof(rwe)); 2544 NTOHL(rwe); 2545 /* tcp_mark_sacked(tp, lwe, rwe); */ 2546 } 2547 break; 2548 } 2549 } 2550 } 2551 2552 /* 2553 * Pull out of band byte out of a segment so 2554 * it doesn't appear in the user's data queue. 2555 * It is still reflected in the segment length for 2556 * sequencing purposes. 2557 */ 2558 void 2559 tcp_pulloutofband(so, th, m, off) 2560 struct socket *so; 2561 struct tcphdr *th; 2562 struct mbuf *m; 2563 int off; 2564 { 2565 int cnt = off + th->th_urp - 1; 2566 2567 while (cnt >= 0) { 2568 if (m->m_len > cnt) { 2569 char *cp = mtod(m, caddr_t) + cnt; 2570 struct tcpcb *tp = sototcpcb(so); 2571 2572 tp->t_iobc = *cp; 2573 tp->t_oobflags |= TCPOOB_HAVEDATA; 2574 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2575 m->m_len--; 2576 return; 2577 } 2578 cnt -= m->m_len; 2579 m = m->m_next; 2580 if (m == 0) 2581 break; 2582 } 2583 panic("tcp_pulloutofband"); 2584 } 2585 2586 /* 2587 * Collect new round-trip time estimate 2588 * and update averages and current timeout. 2589 */ 2590 void 2591 tcp_xmit_timer(tp, rtt) 2592 struct tcpcb *tp; 2593 uint32_t rtt; 2594 { 2595 int32_t delta; 2596 2597 tcpstat.tcps_rttupdated++; 2598 if (tp->t_srtt != 0) { 2599 /* 2600 * srtt is stored as fixed point with 3 bits after the 2601 * binary point (i.e., scaled by 8). The following magic 2602 * is equivalent to the smoothing algorithm in rfc793 with 2603 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2604 * point). Adjust rtt to origin 0. 2605 */ 2606 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2607 if ((tp->t_srtt += delta) <= 0) 2608 tp->t_srtt = 1 << 2; 2609 /* 2610 * We accumulate a smoothed rtt variance (actually, a 2611 * smoothed mean difference), then set the retransmit 2612 * timer to smoothed rtt + 4 times the smoothed variance. 2613 * rttvar is stored as fixed point with 2 bits after the 2614 * binary point (scaled by 4). The following is 2615 * equivalent to rfc793 smoothing with an alpha of .75 2616 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2617 * rfc793's wired-in beta. 2618 */ 2619 if (delta < 0) 2620 delta = -delta; 2621 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2622 if ((tp->t_rttvar += delta) <= 0) 2623 tp->t_rttvar = 1 << 2; 2624 } else { 2625 /* 2626 * No rtt measurement yet - use the unsmoothed rtt. 2627 * Set the variance to half the rtt (so our first 2628 * retransmit happens at 3*rtt). 2629 */ 2630 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 2631 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 2632 } 2633 tp->t_rtttime = 0; 2634 tp->t_rxtshift = 0; 2635 2636 /* 2637 * the retransmit should happen at rtt + 4 * rttvar. 2638 * Because of the way we do the smoothing, srtt and rttvar 2639 * will each average +1/2 tick of bias. When we compute 2640 * the retransmit timer, we want 1/2 tick of rounding and 2641 * 1 extra tick because of +-1/2 tick uncertainty in the 2642 * firing of the timer. The bias will give us exactly the 2643 * 1.5 tick we need. But, because the bias is 2644 * statistical, we have to test that we don't drop below 2645 * the minimum feasible timer (which is 2 ticks). 2646 */ 2647 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2648 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2649 2650 /* 2651 * We received an ack for a packet that wasn't retransmitted; 2652 * it is probably safe to discard any error indications we've 2653 * received recently. This isn't quite right, but close enough 2654 * for now (a route might have failed after we sent a segment, 2655 * and the return path might not be symmetrical). 2656 */ 2657 tp->t_softerror = 0; 2658 } 2659 2660 /* 2661 * Checks for partial ack. If partial ack arrives, force the retransmission 2662 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2663 * 1. By setting snd_nxt to th_ack, this forces retransmission timer to 2664 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2665 */ 2666 int 2667 tcp_newreno(tp, th) 2668 struct tcpcb *tp; 2669 struct tcphdr *th; 2670 { 2671 tcp_seq onxt = tp->snd_nxt; 2672 u_long ocwnd = tp->snd_cwnd; 2673 2674 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2675 /* 2676 * snd_una has not yet been updated and the socket's send 2677 * buffer has not yet drained off the ACK'd data, so we 2678 * have to leave snd_una as it was to get the correct data 2679 * offset in tcp_output(). 2680 */ 2681 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2682 tp->t_rtttime = 0; 2683 tp->snd_nxt = th->th_ack; 2684 /* 2685 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 2686 * is not yet updated when we're called. 2687 */ 2688 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 2689 (void) tcp_output(tp); 2690 tp->snd_cwnd = ocwnd; 2691 if (SEQ_GT(onxt, tp->snd_nxt)) 2692 tp->snd_nxt = onxt; 2693 /* 2694 * Partial window deflation. Relies on fact that tp->snd_una 2695 * not updated yet. 2696 */ 2697 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 2698 return 1; 2699 } 2700 return 0; 2701 } 2702 2703 2704 /* 2705 * TCP compressed state engine. Currently used to hold compressed 2706 * state for SYN_RECEIVED. 2707 */ 2708 2709 u_long syn_cache_count; 2710 u_int32_t syn_hash1, syn_hash2; 2711 2712 #define SYN_HASH(sa, sp, dp) \ 2713 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 2714 ((u_int32_t)(sp)))^syn_hash2))) 2715 #ifndef INET6 2716 #define SYN_HASHALL(hash, src, dst) \ 2717 do { \ 2718 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2719 ((struct sockaddr_in *)(src))->sin_port, \ 2720 ((struct sockaddr_in *)(dst))->sin_port); \ 2721 } while (0) 2722 #else 2723 #define SYN_HASH6(sa, sp, dp) \ 2724 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 2725 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 2726 & 0x7fffffff) 2727 2728 #define SYN_HASHALL(hash, src, dst) \ 2729 do { \ 2730 switch ((src)->sa_family) { \ 2731 case AF_INET: \ 2732 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2733 ((struct sockaddr_in *)(src))->sin_port, \ 2734 ((struct sockaddr_in *)(dst))->sin_port); \ 2735 break; \ 2736 case AF_INET6: \ 2737 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 2738 ((struct sockaddr_in6 *)(src))->sin6_port, \ 2739 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 2740 break; \ 2741 default: \ 2742 hash = 0; \ 2743 } \ 2744 } while (/*CONSTCOND*/0) 2745 #endif /* INET6 */ 2746 2747 #define SYN_CACHE_RM(sc) \ 2748 do { \ 2749 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 2750 (sc), sc_bucketq); \ 2751 (sc)->sc_tp = NULL; \ 2752 LIST_REMOVE((sc), sc_tpq); \ 2753 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 2754 callout_stop(&(sc)->sc_timer); \ 2755 syn_cache_count--; \ 2756 } while (/*CONSTCOND*/0) 2757 2758 #define SYN_CACHE_PUT(sc) \ 2759 do { \ 2760 if ((sc)->sc_ipopts) \ 2761 (void) m_free((sc)->sc_ipopts); \ 2762 if ((sc)->sc_route4.ro_rt != NULL) \ 2763 RTFREE((sc)->sc_route4.ro_rt); \ 2764 pool_put(&syn_cache_pool, (sc)); \ 2765 } while (/*CONSTCOND*/0) 2766 2767 struct pool syn_cache_pool; 2768 2769 /* 2770 * We don't estimate RTT with SYNs, so each packet starts with the default 2771 * RTT and each timer step has a fixed timeout value. 2772 */ 2773 #define SYN_CACHE_TIMER_ARM(sc) \ 2774 do { \ 2775 TCPT_RANGESET((sc)->sc_rxtcur, \ 2776 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 2777 TCPTV_REXMTMAX); \ 2778 callout_reset(&(sc)->sc_timer, \ 2779 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 2780 } while (/*CONSTCOND*/0) 2781 2782 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 2783 2784 void 2785 syn_cache_init() 2786 { 2787 int i; 2788 2789 /* Initialize the hash buckets. */ 2790 for (i = 0; i < tcp_syn_cache_size; i++) 2791 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 2792 2793 /* Initialize the syn cache pool. */ 2794 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 2795 "synpl", NULL); 2796 } 2797 2798 void 2799 syn_cache_insert(sc, tp) 2800 struct syn_cache *sc; 2801 struct tcpcb *tp; 2802 { 2803 struct syn_cache_head *scp; 2804 struct syn_cache *sc2; 2805 int s; 2806 2807 /* 2808 * If there are no entries in the hash table, reinitialize 2809 * the hash secrets. 2810 */ 2811 if (syn_cache_count == 0) { 2812 struct timeval tv; 2813 microtime(&tv); 2814 syn_hash1 = arc4random() ^ (u_long)≻ 2815 syn_hash2 = arc4random() ^ tv.tv_usec; 2816 } 2817 2818 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 2819 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 2820 scp = &tcp_syn_cache[sc->sc_bucketidx]; 2821 2822 /* 2823 * Make sure that we don't overflow the per-bucket 2824 * limit or the total cache size limit. 2825 */ 2826 s = splsoftnet(); 2827 if (scp->sch_length >= tcp_syn_bucket_limit) { 2828 tcpstat.tcps_sc_bucketoverflow++; 2829 /* 2830 * The bucket is full. Toss the oldest element in the 2831 * bucket. This will be the first entry in the bucket. 2832 */ 2833 sc2 = TAILQ_FIRST(&scp->sch_bucket); 2834 #ifdef DIAGNOSTIC 2835 /* 2836 * This should never happen; we should always find an 2837 * entry in our bucket. 2838 */ 2839 if (sc2 == NULL) 2840 panic("syn_cache_insert: bucketoverflow: impossible"); 2841 #endif 2842 SYN_CACHE_RM(sc2); 2843 SYN_CACHE_PUT(sc2); 2844 } else if (syn_cache_count >= tcp_syn_cache_limit) { 2845 struct syn_cache_head *scp2, *sce; 2846 2847 tcpstat.tcps_sc_overflowed++; 2848 /* 2849 * The cache is full. Toss the oldest entry in the 2850 * first non-empty bucket we can find. 2851 * 2852 * XXX We would really like to toss the oldest 2853 * entry in the cache, but we hope that this 2854 * condition doesn't happen very often. 2855 */ 2856 scp2 = scp; 2857 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 2858 sce = &tcp_syn_cache[tcp_syn_cache_size]; 2859 for (++scp2; scp2 != scp; scp2++) { 2860 if (scp2 >= sce) 2861 scp2 = &tcp_syn_cache[0]; 2862 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 2863 break; 2864 } 2865 #ifdef DIAGNOSTIC 2866 /* 2867 * This should never happen; we should always find a 2868 * non-empty bucket. 2869 */ 2870 if (scp2 == scp) 2871 panic("syn_cache_insert: cacheoverflow: " 2872 "impossible"); 2873 #endif 2874 } 2875 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 2876 SYN_CACHE_RM(sc2); 2877 SYN_CACHE_PUT(sc2); 2878 } 2879 2880 /* 2881 * Initialize the entry's timer. 2882 */ 2883 sc->sc_rxttot = 0; 2884 sc->sc_rxtshift = 0; 2885 SYN_CACHE_TIMER_ARM(sc); 2886 2887 /* Link it from tcpcb entry */ 2888 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 2889 2890 /* Put it into the bucket. */ 2891 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 2892 scp->sch_length++; 2893 syn_cache_count++; 2894 2895 tcpstat.tcps_sc_added++; 2896 splx(s); 2897 } 2898 2899 /* 2900 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 2901 * If we have retransmitted an entry the maximum number of times, expire 2902 * that entry. 2903 */ 2904 void 2905 syn_cache_timer(void *arg) 2906 { 2907 struct syn_cache *sc = arg; 2908 int s; 2909 2910 s = splsoftnet(); 2911 2912 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 2913 /* Drop it -- too many retransmissions. */ 2914 goto dropit; 2915 } 2916 2917 /* 2918 * Compute the total amount of time this entry has 2919 * been on a queue. If this entry has been on longer 2920 * than the keep alive timer would allow, expire it. 2921 */ 2922 sc->sc_rxttot += sc->sc_rxtcur; 2923 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) 2924 goto dropit; 2925 2926 tcpstat.tcps_sc_retransmitted++; 2927 (void) syn_cache_respond(sc, NULL); 2928 2929 /* Advance the timer back-off. */ 2930 sc->sc_rxtshift++; 2931 SYN_CACHE_TIMER_ARM(sc); 2932 2933 splx(s); 2934 return; 2935 2936 dropit: 2937 tcpstat.tcps_sc_timed_out++; 2938 SYN_CACHE_RM(sc); 2939 SYN_CACHE_PUT(sc); 2940 splx(s); 2941 } 2942 2943 /* 2944 * Remove syn cache created by the specified tcb entry, 2945 * because this does not make sense to keep them 2946 * (if there's no tcb entry, syn cache entry will never be used) 2947 */ 2948 void 2949 syn_cache_cleanup(tp) 2950 struct tcpcb *tp; 2951 { 2952 struct syn_cache *sc, *nsc; 2953 int s; 2954 2955 s = splsoftnet(); 2956 2957 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 2958 nsc = LIST_NEXT(sc, sc_tpq); 2959 2960 #ifdef DIAGNOSTIC 2961 if (sc->sc_tp != tp) 2962 panic("invalid sc_tp in syn_cache_cleanup"); 2963 #endif 2964 SYN_CACHE_RM(sc); 2965 SYN_CACHE_PUT(sc); 2966 } 2967 /* just for safety */ 2968 LIST_INIT(&tp->t_sc); 2969 2970 splx(s); 2971 } 2972 2973 /* 2974 * Find an entry in the syn cache. 2975 */ 2976 struct syn_cache * 2977 syn_cache_lookup(src, dst, headp) 2978 struct sockaddr *src; 2979 struct sockaddr *dst; 2980 struct syn_cache_head **headp; 2981 { 2982 struct syn_cache *sc; 2983 struct syn_cache_head *scp; 2984 u_int32_t hash; 2985 int s; 2986 2987 SYN_HASHALL(hash, src, dst); 2988 2989 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 2990 *headp = scp; 2991 s = splsoftnet(); 2992 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 2993 sc = TAILQ_NEXT(sc, sc_bucketq)) { 2994 if (sc->sc_hash != hash) 2995 continue; 2996 if (!bcmp(&sc->sc_src, src, src->sa_len) && 2997 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 2998 splx(s); 2999 return (sc); 3000 } 3001 } 3002 splx(s); 3003 return (NULL); 3004 } 3005 3006 /* 3007 * This function gets called when we receive an ACK for a 3008 * socket in the LISTEN state. We look up the connection 3009 * in the syn cache, and if its there, we pull it out of 3010 * the cache and turn it into a full-blown connection in 3011 * the SYN-RECEIVED state. 3012 * 3013 * The return values may not be immediately obvious, and their effects 3014 * can be subtle, so here they are: 3015 * 3016 * NULL SYN was not found in cache; caller should drop the 3017 * packet and send an RST. 3018 * 3019 * -1 We were unable to create the new connection, and are 3020 * aborting it. An ACK,RST is being sent to the peer 3021 * (unless we got screwey sequence numbners; see below), 3022 * because the 3-way handshake has been completed. Caller 3023 * should not free the mbuf, since we may be using it. If 3024 * we are not, we will free it. 3025 * 3026 * Otherwise, the return value is a pointer to the new socket 3027 * associated with the connection. 3028 */ 3029 struct socket * 3030 syn_cache_get(src, dst, th, hlen, tlen, so, m) 3031 struct sockaddr *src; 3032 struct sockaddr *dst; 3033 struct tcphdr *th; 3034 unsigned int hlen, tlen; 3035 struct socket *so; 3036 struct mbuf *m; 3037 { 3038 struct syn_cache *sc; 3039 struct syn_cache_head *scp; 3040 struct inpcb *inp = NULL; 3041 #ifdef INET6 3042 struct in6pcb *in6p = NULL; 3043 #endif 3044 struct tcpcb *tp = 0; 3045 struct mbuf *am; 3046 int s; 3047 struct socket *oso; 3048 3049 s = splsoftnet(); 3050 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3051 splx(s); 3052 return (NULL); 3053 } 3054 3055 /* 3056 * Verify the sequence and ack numbers. Try getting the correct 3057 * response again. 3058 */ 3059 if ((th->th_ack != sc->sc_iss + 1) || 3060 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3061 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3062 (void) syn_cache_respond(sc, m); 3063 splx(s); 3064 return ((struct socket *)(-1)); 3065 } 3066 3067 /* Remove this cache entry */ 3068 SYN_CACHE_RM(sc); 3069 splx(s); 3070 3071 /* 3072 * Ok, create the full blown connection, and set things up 3073 * as they would have been set up if we had created the 3074 * connection when the SYN arrived. If we can't create 3075 * the connection, abort it. 3076 */ 3077 /* 3078 * inp still has the OLD in_pcb stuff, set the 3079 * v6-related flags on the new guy, too. This is 3080 * done particularly for the case where an AF_INET6 3081 * socket is bound only to a port, and a v4 connection 3082 * comes in on that port. 3083 * we also copy the flowinfo from the original pcb 3084 * to the new one. 3085 */ 3086 { 3087 struct inpcb *parentinpcb; 3088 3089 parentinpcb = (struct inpcb *)so->so_pcb; 3090 3091 oso = so; 3092 so = sonewconn(so, SS_ISCONNECTED); 3093 if (so == NULL) 3094 goto resetandabort; 3095 3096 switch (so->so_proto->pr_domain->dom_family) { 3097 #ifdef INET 3098 case AF_INET: 3099 inp = sotoinpcb(so); 3100 break; 3101 #endif 3102 #ifdef INET6 3103 case AF_INET6: 3104 in6p = sotoin6pcb(so); 3105 break; 3106 #endif 3107 } 3108 } 3109 switch (src->sa_family) { 3110 #ifdef INET 3111 case AF_INET: 3112 if (inp) { 3113 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3114 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3115 inp->inp_options = ip_srcroute(); 3116 in_pcbstate(inp, INP_BOUND); 3117 if (inp->inp_options == NULL) { 3118 inp->inp_options = sc->sc_ipopts; 3119 sc->sc_ipopts = NULL; 3120 } 3121 } 3122 #ifdef INET6 3123 else if (in6p) { 3124 /* IPv4 packet to AF_INET6 socket */ 3125 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3126 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3127 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3128 &in6p->in6p_laddr.s6_addr32[3], 3129 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3130 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3131 in6totcpcb(in6p)->t_family = AF_INET; 3132 } 3133 #endif 3134 break; 3135 #endif 3136 #ifdef INET6 3137 case AF_INET6: 3138 if (in6p) { 3139 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3140 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3141 #if 0 3142 in6p->in6p_flowinfo = ip6->ip6_flow & IPV6_FLOWINFO_MASK; 3143 /*inp->inp_options = ip6_srcroute();*/ /* soon. */ 3144 #endif 3145 } 3146 break; 3147 #endif 3148 } 3149 #ifdef INET6 3150 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3151 struct in6pcb *oin6p = sotoin6pcb(oso); 3152 /* inherit socket options from the listening socket */ 3153 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3154 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3155 m_freem(in6p->in6p_options); 3156 in6p->in6p_options = 0; 3157 } 3158 ip6_savecontrol(in6p, &in6p->in6p_options, 3159 mtod(m, struct ip6_hdr *), m); 3160 } 3161 #endif 3162 3163 #ifdef IPSEC 3164 /* 3165 * we make a copy of policy, instead of sharing the policy, 3166 * for better behavior in terms of SA lookup and dead SA removal. 3167 */ 3168 if (inp) { 3169 /* copy old policy into new socket's */ 3170 if (ipsec_copy_policy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3171 printf("tcp_input: could not copy policy\n"); 3172 } 3173 #ifdef INET6 3174 else if (in6p) { 3175 /* copy old policy into new socket's */ 3176 if (ipsec_copy_policy(sotoin6pcb(oso)->in6p_sp, in6p->in6p_sp)) 3177 printf("tcp_input: could not copy policy\n"); 3178 } 3179 #endif 3180 #endif 3181 3182 /* 3183 * Give the new socket our cached route reference. 3184 */ 3185 if (inp) 3186 inp->inp_route = sc->sc_route4; /* struct assignment */ 3187 #ifdef INET6 3188 else 3189 in6p->in6p_route = sc->sc_route6; 3190 #endif 3191 sc->sc_route4.ro_rt = NULL; 3192 3193 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3194 if (am == NULL) 3195 goto resetandabort; 3196 am->m_len = src->sa_len; 3197 bcopy(src, mtod(am, caddr_t), src->sa_len); 3198 if (inp) { 3199 if (in_pcbconnect(inp, am)) { 3200 (void) m_free(am); 3201 goto resetandabort; 3202 } 3203 } 3204 #ifdef INET6 3205 else if (in6p) { 3206 if (src->sa_family == AF_INET) { 3207 /* IPv4 packet to AF_INET6 socket */ 3208 struct sockaddr_in6 *sin6; 3209 sin6 = mtod(am, struct sockaddr_in6 *); 3210 am->m_len = sizeof(*sin6); 3211 bzero(sin6, sizeof(*sin6)); 3212 sin6->sin6_family = AF_INET6; 3213 sin6->sin6_len = sizeof(*sin6); 3214 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3215 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3216 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3217 &sin6->sin6_addr.s6_addr32[3], 3218 sizeof(sin6->sin6_addr.s6_addr32[3])); 3219 } 3220 if (in6_pcbconnect(in6p, am)) { 3221 (void) m_free(am); 3222 goto resetandabort; 3223 } 3224 } 3225 #endif 3226 else { 3227 (void) m_free(am); 3228 goto resetandabort; 3229 } 3230 (void) m_free(am); 3231 3232 if (inp) 3233 tp = intotcpcb(inp); 3234 #ifdef INET6 3235 else if (in6p) 3236 tp = in6totcpcb(in6p); 3237 #endif 3238 else 3239 tp = NULL; 3240 if (sc->sc_request_r_scale != 15) { 3241 tp->requested_s_scale = sc->sc_requested_s_scale; 3242 tp->request_r_scale = sc->sc_request_r_scale; 3243 tp->snd_scale = sc->sc_requested_s_scale; 3244 tp->rcv_scale = sc->sc_request_r_scale; 3245 tp->t_flags |= TF_RCVD_SCALE; 3246 } 3247 if (sc->sc_flags & SCF_TIMESTAMP) 3248 tp->t_flags |= TF_RCVD_TSTMP; 3249 tp->ts_timebase = sc->sc_timebase; 3250 3251 tp->t_template = tcp_template(tp); 3252 if (tp->t_template == 0) { 3253 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3254 so = NULL; 3255 m_freem(m); 3256 goto abort; 3257 } 3258 3259 tp->iss = sc->sc_iss; 3260 tp->irs = sc->sc_irs; 3261 tcp_sendseqinit(tp); 3262 tcp_rcvseqinit(tp); 3263 tp->t_state = TCPS_SYN_RECEIVED; 3264 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3265 tcpstat.tcps_accepts++; 3266 3267 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3268 tp->t_ourmss = sc->sc_ourmaxseg; 3269 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3270 3271 /* 3272 * Initialize the initial congestion window. If we 3273 * had to retransmit the SYN,ACK, we must initialize cwnd 3274 * to 1 segment (i.e. the Loss Window). 3275 */ 3276 if (sc->sc_rxtshift) 3277 tp->snd_cwnd = tp->t_peermss; 3278 else 3279 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, tp->t_peermss); 3280 3281 tcp_rmx_rtt(tp); 3282 tp->snd_wl1 = sc->sc_irs; 3283 tp->rcv_up = sc->sc_irs + 1; 3284 3285 /* 3286 * This is what whould have happened in tcp_ouput() when 3287 * the SYN,ACK was sent. 3288 */ 3289 tp->snd_up = tp->snd_una; 3290 tp->snd_max = tp->snd_nxt = tp->iss+1; 3291 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3292 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3293 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3294 tp->last_ack_sent = tp->rcv_nxt; 3295 3296 tcpstat.tcps_sc_completed++; 3297 SYN_CACHE_PUT(sc); 3298 return (so); 3299 3300 resetandabort: 3301 (void) tcp_respond(NULL, m, m, th, 3302 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 3303 abort: 3304 if (so != NULL) 3305 (void) soabort(so); 3306 SYN_CACHE_PUT(sc); 3307 tcpstat.tcps_sc_aborted++; 3308 return ((struct socket *)(-1)); 3309 } 3310 3311 /* 3312 * This function is called when we get a RST for a 3313 * non-existent connection, so that we can see if the 3314 * connection is in the syn cache. If it is, zap it. 3315 */ 3316 3317 void 3318 syn_cache_reset(src, dst, th) 3319 struct sockaddr *src; 3320 struct sockaddr *dst; 3321 struct tcphdr *th; 3322 { 3323 struct syn_cache *sc; 3324 struct syn_cache_head *scp; 3325 int s = splsoftnet(); 3326 3327 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3328 splx(s); 3329 return; 3330 } 3331 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3332 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3333 splx(s); 3334 return; 3335 } 3336 SYN_CACHE_RM(sc); 3337 splx(s); 3338 tcpstat.tcps_sc_reset++; 3339 SYN_CACHE_PUT(sc); 3340 } 3341 3342 void 3343 syn_cache_unreach(src, dst, th) 3344 struct sockaddr *src; 3345 struct sockaddr *dst; 3346 struct tcphdr *th; 3347 { 3348 struct syn_cache *sc; 3349 struct syn_cache_head *scp; 3350 int s; 3351 3352 s = splsoftnet(); 3353 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3354 splx(s); 3355 return; 3356 } 3357 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3358 if (ntohl (th->th_seq) != sc->sc_iss) { 3359 splx(s); 3360 return; 3361 } 3362 3363 /* 3364 * If we've rertransmitted 3 times and this is our second error, 3365 * we remove the entry. Otherwise, we allow it to continue on. 3366 * This prevents us from incorrectly nuking an entry during a 3367 * spurious network outage. 3368 * 3369 * See tcp_notify(). 3370 */ 3371 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3372 sc->sc_flags |= SCF_UNREACH; 3373 splx(s); 3374 return; 3375 } 3376 3377 SYN_CACHE_RM(sc); 3378 splx(s); 3379 tcpstat.tcps_sc_unreach++; 3380 SYN_CACHE_PUT(sc); 3381 } 3382 3383 /* 3384 * Given a LISTEN socket and an inbound SYN request, add 3385 * this to the syn cache, and send back a segment: 3386 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3387 * to the source. 3388 * 3389 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3390 * Doing so would require that we hold onto the data and deliver it 3391 * to the application. However, if we are the target of a SYN-flood 3392 * DoS attack, an attacker could send data which would eventually 3393 * consume all available buffer space if it were ACKed. By not ACKing 3394 * the data, we avoid this DoS scenario. 3395 */ 3396 3397 int 3398 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi) 3399 struct sockaddr *src; 3400 struct sockaddr *dst; 3401 struct tcphdr *th; 3402 unsigned int hlen; 3403 struct socket *so; 3404 struct mbuf *m; 3405 u_char *optp; 3406 int optlen; 3407 struct tcp_opt_info *oi; 3408 { 3409 struct tcpcb tb, *tp; 3410 long win; 3411 struct syn_cache *sc; 3412 struct syn_cache_head *scp; 3413 struct mbuf *ipopts; 3414 3415 tp = sototcpcb(so); 3416 3417 /* 3418 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3419 * 3420 * Note this check is performed in tcp_input() very early on. 3421 */ 3422 3423 /* 3424 * Initialize some local state. 3425 */ 3426 win = sbspace(&so->so_rcv); 3427 if (win > TCP_MAXWIN) 3428 win = TCP_MAXWIN; 3429 3430 switch (src->sa_family) { 3431 #ifdef INET 3432 case AF_INET: 3433 /* 3434 * Remember the IP options, if any. 3435 */ 3436 ipopts = ip_srcroute(); 3437 break; 3438 #endif 3439 default: 3440 ipopts = NULL; 3441 } 3442 3443 if (optp) { 3444 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3445 tcp_dooptions(&tb, optp, optlen, th, oi); 3446 } else 3447 tb.t_flags = 0; 3448 3449 /* 3450 * See if we already have an entry for this connection. 3451 * If we do, resend the SYN,ACK. We do not count this 3452 * as a retransmission (XXX though maybe we should). 3453 */ 3454 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3455 tcpstat.tcps_sc_dupesyn++; 3456 if (ipopts) { 3457 /* 3458 * If we were remembering a previous source route, 3459 * forget it and use the new one we've been given. 3460 */ 3461 if (sc->sc_ipopts) 3462 (void) m_free(sc->sc_ipopts); 3463 sc->sc_ipopts = ipopts; 3464 } 3465 sc->sc_timestamp = tb.ts_recent; 3466 if (syn_cache_respond(sc, m) == 0) { 3467 tcpstat.tcps_sndacks++; 3468 tcpstat.tcps_sndtotal++; 3469 } 3470 return (1); 3471 } 3472 3473 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3474 if (sc == NULL) { 3475 if (ipopts) 3476 (void) m_free(ipopts); 3477 return (0); 3478 } 3479 3480 /* 3481 * Fill in the cache, and put the necessary IP and TCP 3482 * options into the reply. 3483 */ 3484 callout_init(&sc->sc_timer); 3485 bzero(sc, sizeof(struct syn_cache)); 3486 bcopy(src, &sc->sc_src, src->sa_len); 3487 bcopy(dst, &sc->sc_dst, dst->sa_len); 3488 sc->sc_flags = 0; 3489 sc->sc_ipopts = ipopts; 3490 sc->sc_irs = th->th_seq; 3491 switch (src->sa_family) { 3492 #ifdef INET 3493 case AF_INET: 3494 { 3495 struct sockaddr_in *srcin = (void *) src; 3496 struct sockaddr_in *dstin = (void *) dst; 3497 3498 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 3499 &srcin->sin_addr, dstin->sin_port, 3500 srcin->sin_port, sizeof(dstin->sin_addr), 0); 3501 break; 3502 } 3503 #endif /* INET */ 3504 #ifdef INET6 3505 case AF_INET6: 3506 { 3507 struct sockaddr_in6 *srcin6 = (void *) src; 3508 struct sockaddr_in6 *dstin6 = (void *) dst; 3509 3510 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 3511 &srcin6->sin6_addr, dstin6->sin6_port, 3512 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 3513 break; 3514 } 3515 #endif /* INET6 */ 3516 } 3517 sc->sc_peermaxseg = oi->maxseg; 3518 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3519 m->m_pkthdr.rcvif : NULL, 3520 sc->sc_src.sa.sa_family); 3521 sc->sc_win = win; 3522 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 3523 sc->sc_timestamp = tb.ts_recent; 3524 if (tcp_do_rfc1323 && (tb.t_flags & TF_RCVD_TSTMP)) 3525 sc->sc_flags |= SCF_TIMESTAMP; 3526 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3527 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3528 sc->sc_requested_s_scale = tb.requested_s_scale; 3529 sc->sc_request_r_scale = 0; 3530 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3531 TCP_MAXWIN << sc->sc_request_r_scale < 3532 so->so_rcv.sb_hiwat) 3533 sc->sc_request_r_scale++; 3534 } else { 3535 sc->sc_requested_s_scale = 15; 3536 sc->sc_request_r_scale = 15; 3537 } 3538 sc->sc_tp = tp; 3539 if (syn_cache_respond(sc, m) == 0) { 3540 syn_cache_insert(sc, tp); 3541 tcpstat.tcps_sndacks++; 3542 tcpstat.tcps_sndtotal++; 3543 } else { 3544 SYN_CACHE_PUT(sc); 3545 tcpstat.tcps_sc_dropped++; 3546 } 3547 return (1); 3548 } 3549 3550 int 3551 syn_cache_respond(sc, m) 3552 struct syn_cache *sc; 3553 struct mbuf *m; 3554 { 3555 struct route *ro; 3556 u_int8_t *optp; 3557 int optlen, error; 3558 u_int16_t tlen; 3559 struct ip *ip = NULL; 3560 #ifdef INET6 3561 struct ip6_hdr *ip6 = NULL; 3562 #endif 3563 struct tcphdr *th; 3564 u_int hlen; 3565 3566 switch (sc->sc_src.sa.sa_family) { 3567 case AF_INET: 3568 hlen = sizeof(struct ip); 3569 ro = &sc->sc_route4; 3570 break; 3571 #ifdef INET6 3572 case AF_INET6: 3573 hlen = sizeof(struct ip6_hdr); 3574 ro = (struct route *)&sc->sc_route6; 3575 break; 3576 #endif 3577 default: 3578 if (m) 3579 m_freem(m); 3580 return EAFNOSUPPORT; 3581 } 3582 3583 /* Compute the size of the TCP options. */ 3584 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 3585 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 3586 3587 tlen = hlen + sizeof(struct tcphdr) + optlen; 3588 3589 /* 3590 * Create the IP+TCP header from scratch. 3591 */ 3592 if (m) 3593 m_freem(m); 3594 #ifdef DIAGNOSTIC 3595 if (max_linkhdr + tlen > MCLBYTES) 3596 return (ENOBUFS); 3597 #endif 3598 MGETHDR(m, M_DONTWAIT, MT_DATA); 3599 if (m && tlen > MHLEN) { 3600 MCLGET(m, M_DONTWAIT); 3601 if ((m->m_flags & M_EXT) == 0) { 3602 m_freem(m); 3603 m = NULL; 3604 } 3605 } 3606 if (m == NULL) 3607 return (ENOBUFS); 3608 3609 /* Fixup the mbuf. */ 3610 m->m_data += max_linkhdr; 3611 m->m_len = m->m_pkthdr.len = tlen; 3612 #ifdef IPSEC 3613 if (sc->sc_tp) { 3614 struct tcpcb *tp; 3615 struct socket *so; 3616 3617 tp = sc->sc_tp; 3618 if (tp->t_inpcb) 3619 so = tp->t_inpcb->inp_socket; 3620 #ifdef INET6 3621 else if (tp->t_in6pcb) 3622 so = tp->t_in6pcb->in6p_socket; 3623 #endif 3624 else 3625 so = NULL; 3626 /* use IPsec policy on listening socket, on SYN ACK */ 3627 if (ipsec_setsocket(m, so) != 0) { 3628 m_freem(m); 3629 return ENOBUFS; 3630 } 3631 } 3632 #endif 3633 m->m_pkthdr.rcvif = NULL; 3634 memset(mtod(m, u_char *), 0, tlen); 3635 3636 switch (sc->sc_src.sa.sa_family) { 3637 case AF_INET: 3638 ip = mtod(m, struct ip *); 3639 ip->ip_dst = sc->sc_src.sin.sin_addr; 3640 ip->ip_src = sc->sc_dst.sin.sin_addr; 3641 ip->ip_p = IPPROTO_TCP; 3642 th = (struct tcphdr *)(ip + 1); 3643 th->th_dport = sc->sc_src.sin.sin_port; 3644 th->th_sport = sc->sc_dst.sin.sin_port; 3645 break; 3646 #ifdef INET6 3647 case AF_INET6: 3648 ip6 = mtod(m, struct ip6_hdr *); 3649 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 3650 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 3651 ip6->ip6_nxt = IPPROTO_TCP; 3652 /* ip6_plen will be updated in ip6_output() */ 3653 th = (struct tcphdr *)(ip6 + 1); 3654 th->th_dport = sc->sc_src.sin6.sin6_port; 3655 th->th_sport = sc->sc_dst.sin6.sin6_port; 3656 break; 3657 #endif 3658 default: 3659 th = NULL; 3660 } 3661 3662 th->th_seq = htonl(sc->sc_iss); 3663 th->th_ack = htonl(sc->sc_irs + 1); 3664 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 3665 th->th_flags = TH_SYN|TH_ACK; 3666 th->th_win = htons(sc->sc_win); 3667 /* th_sum already 0 */ 3668 /* th_urp already 0 */ 3669 3670 /* Tack on the TCP options. */ 3671 optp = (u_int8_t *)(th + 1); 3672 *optp++ = TCPOPT_MAXSEG; 3673 *optp++ = 4; 3674 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 3675 *optp++ = sc->sc_ourmaxseg & 0xff; 3676 3677 if (sc->sc_request_r_scale != 15) { 3678 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 3679 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 3680 sc->sc_request_r_scale); 3681 optp += 4; 3682 } 3683 3684 if (sc->sc_flags & SCF_TIMESTAMP) { 3685 u_int32_t *lp = (u_int32_t *)(optp); 3686 /* Form timestamp option as shown in appendix A of RFC 1323. */ 3687 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 3688 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 3689 *lp = htonl(sc->sc_timestamp); 3690 optp += TCPOLEN_TSTAMP_APPA; 3691 } 3692 3693 /* Compute the packet's checksum. */ 3694 switch (sc->sc_src.sa.sa_family) { 3695 case AF_INET: 3696 ip->ip_len = htons(tlen - hlen); 3697 th->th_sum = 0; 3698 th->th_sum = in_cksum(m, tlen); 3699 break; 3700 #ifdef INET6 3701 case AF_INET6: 3702 ip6->ip6_plen = htons(tlen - hlen); 3703 th->th_sum = 0; 3704 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 3705 break; 3706 #endif 3707 } 3708 3709 /* 3710 * Fill in some straggling IP bits. Note the stack expects 3711 * ip_len to be in host order, for convenience. 3712 */ 3713 switch (sc->sc_src.sa.sa_family) { 3714 #ifdef INET 3715 case AF_INET: 3716 ip->ip_len = tlen; 3717 ip->ip_ttl = ip_defttl; 3718 /* XXX tos? */ 3719 break; 3720 #endif 3721 #ifdef INET6 3722 case AF_INET6: 3723 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 3724 ip6->ip6_vfc |= IPV6_VERSION; 3725 ip6->ip6_plen = htons(tlen - hlen); 3726 /* ip6_hlim will be initialized afterwards */ 3727 /* XXX flowlabel? */ 3728 break; 3729 #endif 3730 } 3731 3732 switch (sc->sc_src.sa.sa_family) { 3733 #ifdef INET 3734 case AF_INET: 3735 error = ip_output(m, sc->sc_ipopts, ro, 3736 (ip_mtudisc ? IP_MTUDISC : 0), 3737 NULL); 3738 break; 3739 #endif 3740 #ifdef INET6 3741 case AF_INET6: 3742 ip6->ip6_hlim = in6_selecthlim(NULL, 3743 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 3744 3745 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 3746 0, NULL, NULL); 3747 break; 3748 #endif 3749 default: 3750 error = EAFNOSUPPORT; 3751 break; 3752 } 3753 return (error); 3754 } 3755