1 /* $NetBSD: tcp_input.c,v 1.134 2002/01/24 02:12:29 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 1.134 2002/01/24 02:12:29 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 int 296 tcp_reass(tp, th, m, tlen) 297 struct tcpcb *tp; 298 struct tcphdr *th; 299 struct mbuf *m; 300 int *tlen; 301 { 302 struct ipqent *p, *q, *nq, *tiqe = NULL; 303 struct socket *so = NULL; 304 int pkt_flags; 305 tcp_seq pkt_seq; 306 unsigned pkt_len; 307 u_long rcvpartdupbyte = 0; 308 u_long rcvoobyte; 309 310 if (tp->t_inpcb) 311 so = tp->t_inpcb->inp_socket; 312 #ifdef INET6 313 else if (tp->t_in6pcb) 314 so = tp->t_in6pcb->in6p_socket; 315 #endif 316 317 TCP_REASS_LOCK_CHECK(tp); 318 319 /* 320 * Call with th==0 after become established to 321 * force pre-ESTABLISHED data up to user socket. 322 */ 323 if (th == 0) 324 goto present; 325 326 rcvoobyte = *tlen; 327 /* 328 * Copy these to local variables because the tcpiphdr 329 * gets munged while we are collapsing mbufs. 330 */ 331 pkt_seq = th->th_seq; 332 pkt_len = *tlen; 333 pkt_flags = th->th_flags; 334 /* 335 * Find a segment which begins after this one does. 336 */ 337 for (p = NULL, q = LIST_FIRST(&tp->segq); q != NULL; q = nq) { 338 nq = LIST_NEXT(q, ipqe_q); 339 /* 340 * If the received segment is just right after this 341 * fragment, merge the two together and then check 342 * for further overlaps. 343 */ 344 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 345 #ifdef TCPREASS_DEBUG 346 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 347 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 348 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 349 #endif 350 pkt_len += q->ipqe_len; 351 pkt_flags |= q->ipqe_flags; 352 pkt_seq = q->ipqe_seq; 353 m_cat(q->ipqe_m, m); 354 m = q->ipqe_m; 355 goto free_ipqe; 356 } 357 /* 358 * If the received segment is completely past this 359 * fragment, we need to go the next fragment. 360 */ 361 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 362 p = q; 363 continue; 364 } 365 /* 366 * If the fragment is past the received segment, 367 * it (or any following) can't be concatenated. 368 */ 369 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) 370 break; 371 /* 372 * We've received all the data in this segment before. 373 * mark it as a duplicate and return. 374 */ 375 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 376 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 377 tcpstat.tcps_rcvduppack++; 378 tcpstat.tcps_rcvdupbyte += pkt_len; 379 m_freem(m); 380 if (tiqe != NULL) 381 pool_put(&ipqent_pool, tiqe); 382 return (0); 383 } 384 /* 385 * Received segment completely overlaps this fragment 386 * so we drop the fragment (this keeps the temporal 387 * ordering of segments correct). 388 */ 389 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 390 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 391 rcvpartdupbyte += q->ipqe_len; 392 m_freem(q->ipqe_m); 393 goto free_ipqe; 394 } 395 /* 396 * RX'ed segment extends past the end of the 397 * fragment. Drop the overlapping bytes. Then 398 * merge the fragment and segment then treat as 399 * a longer received packet. 400 */ 401 if (SEQ_LT(q->ipqe_seq, pkt_seq) 402 && SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 403 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 404 #ifdef TCPREASS_DEBUG 405 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 406 tp, overlap, 407 pkt_seq, pkt_seq + pkt_len, pkt_len); 408 #endif 409 m_adj(m, overlap); 410 rcvpartdupbyte += overlap; 411 m_cat(q->ipqe_m, m); 412 m = q->ipqe_m; 413 pkt_seq = q->ipqe_seq; 414 pkt_len += q->ipqe_len - overlap; 415 rcvoobyte -= overlap; 416 goto free_ipqe; 417 } 418 /* 419 * RX'ed segment extends past the front of the 420 * fragment. Drop the overlapping bytes on the 421 * received packet. The packet will then be 422 * contatentated with this fragment a bit later. 423 */ 424 if (SEQ_GT(q->ipqe_seq, pkt_seq) 425 && SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 426 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 427 #ifdef TCPREASS_DEBUG 428 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 429 tp, overlap, 430 pkt_seq, pkt_seq + pkt_len, pkt_len); 431 #endif 432 m_adj(m, -overlap); 433 pkt_len -= overlap; 434 rcvpartdupbyte += overlap; 435 rcvoobyte -= overlap; 436 } 437 /* 438 * If the received segment immediates precedes this 439 * fragment then tack the fragment onto this segment 440 * and reinsert the data. 441 */ 442 if (q->ipqe_seq == pkt_seq + pkt_len) { 443 #ifdef TCPREASS_DEBUG 444 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 445 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 446 pkt_seq, pkt_seq + pkt_len, pkt_len); 447 #endif 448 pkt_len += q->ipqe_len; 449 pkt_flags |= q->ipqe_flags; 450 m_cat(m, q->ipqe_m); 451 LIST_REMOVE(q, ipqe_q); 452 LIST_REMOVE(q, ipqe_timeq); 453 if (tiqe == NULL) { 454 tiqe = q; 455 } else { 456 pool_put(&ipqent_pool, q); 457 } 458 break; 459 } 460 /* 461 * If the fragment is before the segment, remember it. 462 * When this loop is terminated, p will contain the 463 * pointer to fragment that is right before the received 464 * segment. 465 */ 466 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 467 p = q; 468 469 continue; 470 471 /* 472 * This is a common operation. It also will allow 473 * to save doing a malloc/free in most instances. 474 */ 475 free_ipqe: 476 LIST_REMOVE(q, ipqe_q); 477 LIST_REMOVE(q, ipqe_timeq); 478 if (tiqe == NULL) { 479 tiqe = q; 480 } else { 481 pool_put(&ipqent_pool, q); 482 } 483 } 484 485 /* 486 * Allocate a new queue entry since the received segment did not 487 * collapse onto any other out-of-order block; thus we are allocating 488 * a new block. If it had collapsed, tiqe would not be NULL and 489 * we would be reusing it. 490 * XXX If we can't, just drop the packet. XXX 491 */ 492 if (tiqe == NULL) { 493 tiqe = pool_get(&ipqent_pool, PR_NOWAIT); 494 if (tiqe == NULL) { 495 tcpstat.tcps_rcvmemdrop++; 496 m_freem(m); 497 return (0); 498 } 499 } 500 501 /* 502 * Update the counters. 503 */ 504 tcpstat.tcps_rcvoopack++; 505 tcpstat.tcps_rcvoobyte += rcvoobyte; 506 if (rcvpartdupbyte) { 507 tcpstat.tcps_rcvpartduppack++; 508 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 509 } 510 511 /* 512 * Insert the new fragment queue entry into both queues. 513 */ 514 tiqe->ipqe_m = m; 515 tiqe->ipqe_seq = pkt_seq; 516 tiqe->ipqe_len = pkt_len; 517 tiqe->ipqe_flags = pkt_flags; 518 if (p == NULL) { 519 LIST_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 520 #ifdef TCPREASS_DEBUG 521 if (tiqe->ipqe_seq != tp->rcv_nxt) 522 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 523 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 524 #endif 525 } else { 526 LIST_INSERT_AFTER(p, tiqe, ipqe_q); 527 #ifdef TCPREASS_DEBUG 528 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 529 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 530 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 531 #endif 532 } 533 534 LIST_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 535 536 present: 537 /* 538 * Present data to user, advancing rcv_nxt through 539 * completed sequence space. 540 */ 541 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 542 return (0); 543 q = LIST_FIRST(&tp->segq); 544 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 545 return (0); 546 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 547 return (0); 548 549 tp->rcv_nxt += q->ipqe_len; 550 pkt_flags = q->ipqe_flags & TH_FIN; 551 ND6_HINT(tp); 552 553 LIST_REMOVE(q, ipqe_q); 554 LIST_REMOVE(q, ipqe_timeq); 555 if (so->so_state & SS_CANTRCVMORE) 556 m_freem(q->ipqe_m); 557 else 558 sbappend(&so->so_rcv, q->ipqe_m); 559 pool_put(&ipqent_pool, q); 560 sorwakeup(so); 561 return (pkt_flags); 562 } 563 564 #ifdef INET6 565 int 566 tcp6_input(mp, offp, proto) 567 struct mbuf **mp; 568 int *offp, proto; 569 { 570 struct mbuf *m = *mp; 571 572 /* 573 * draft-itojun-ipv6-tcp-to-anycast 574 * better place to put this in? 575 */ 576 if (m->m_flags & M_ANYCAST6) { 577 struct ip6_hdr *ip6; 578 if (m->m_len < sizeof(struct ip6_hdr)) { 579 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 580 tcpstat.tcps_rcvshort++; 581 return IPPROTO_DONE; 582 } 583 } 584 ip6 = mtod(m, struct ip6_hdr *); 585 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 586 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 587 return IPPROTO_DONE; 588 } 589 590 tcp_input(m, *offp, proto); 591 return IPPROTO_DONE; 592 } 593 #endif 594 595 /* 596 * TCP input routine, follows pages 65-76 of the 597 * protocol specification dated September, 1981 very closely. 598 */ 599 void 600 #if __STDC__ 601 tcp_input(struct mbuf *m, ...) 602 #else 603 tcp_input(m, va_alist) 604 struct mbuf *m; 605 #endif 606 { 607 int proto; 608 struct tcphdr *th; 609 struct ip *ip; 610 struct inpcb *inp; 611 #ifdef INET6 612 struct ip6_hdr *ip6; 613 struct in6pcb *in6p; 614 #endif 615 caddr_t optp = NULL; 616 int optlen = 0; 617 int len, tlen, toff, hdroptlen = 0; 618 struct tcpcb *tp = 0; 619 int tiflags; 620 struct socket *so = NULL; 621 int todrop, acked, ourfinisacked, needoutput = 0; 622 short ostate = 0; 623 int iss = 0; 624 u_long tiwin; 625 struct tcp_opt_info opti; 626 int off, iphlen; 627 va_list ap; 628 int af; /* af on the wire */ 629 struct mbuf *tcp_saveti = NULL; 630 631 va_start(ap, m); 632 toff = va_arg(ap, int); 633 proto = va_arg(ap, int); 634 va_end(ap); 635 636 tcpstat.tcps_rcvtotal++; 637 638 bzero(&opti, sizeof(opti)); 639 opti.ts_present = 0; 640 opti.maxseg = 0; 641 642 /* 643 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 644 * 645 * TCP is, by definition, unicast, so we reject all 646 * multicast outright. 647 * 648 * Note, there are additional src/dst address checks in 649 * the AF-specific code below. 650 */ 651 if (m->m_flags & (M_BCAST|M_MCAST)) { 652 /* XXX stat */ 653 goto drop; 654 } 655 #ifdef INET6 656 if (m->m_flags & M_ANYCAST6) { 657 /* XXX stat */ 658 goto drop; 659 } 660 #endif 661 662 /* 663 * Get IP and TCP header together in first mbuf. 664 * Note: IP leaves IP header in first mbuf. 665 */ 666 ip = mtod(m, struct ip *); 667 #ifdef INET6 668 ip6 = NULL; 669 #endif 670 switch (ip->ip_v) { 671 #ifdef INET 672 case 4: 673 af = AF_INET; 674 iphlen = sizeof(struct ip); 675 #ifndef PULLDOWN_TEST 676 /* would like to get rid of this... */ 677 if (toff > sizeof (struct ip)) { 678 ip_stripoptions(m, (struct mbuf *)0); 679 toff = sizeof(struct ip); 680 } 681 if (m->m_len < toff + sizeof (struct tcphdr)) { 682 if ((m = m_pullup(m, toff + sizeof (struct tcphdr))) == 0) { 683 tcpstat.tcps_rcvshort++; 684 return; 685 } 686 } 687 ip = mtod(m, struct ip *); 688 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 689 #else 690 ip = mtod(m, struct ip *); 691 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 692 sizeof(struct tcphdr)); 693 if (th == NULL) { 694 tcpstat.tcps_rcvshort++; 695 return; 696 } 697 #endif 698 699 /* 700 * Make sure destination address is not multicast. 701 * Source address checked in ip_input(). 702 */ 703 if (IN_MULTICAST(ip->ip_dst.s_addr)) { 704 /* XXX stat */ 705 goto drop; 706 } 707 708 /* We do the checksum after PCB lookup... */ 709 len = ip->ip_len; 710 tlen = len - toff; 711 break; 712 #endif 713 #ifdef INET6 714 case 6: 715 ip = NULL; 716 iphlen = sizeof(struct ip6_hdr); 717 af = AF_INET6; 718 #ifndef PULLDOWN_TEST 719 if (m->m_len < toff + sizeof(struct tcphdr)) { 720 m = m_pullup(m, toff + sizeof(struct tcphdr)); /*XXX*/ 721 if (m == NULL) { 722 tcpstat.tcps_rcvshort++; 723 return; 724 } 725 } 726 ip6 = mtod(m, struct ip6_hdr *); 727 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 728 #else 729 ip6 = mtod(m, struct ip6_hdr *); 730 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 731 sizeof(struct tcphdr)); 732 if (th == NULL) { 733 tcpstat.tcps_rcvshort++; 734 return; 735 } 736 #endif 737 738 /* Be proactive about malicious use of IPv4 mapped address */ 739 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 740 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 741 /* XXX stat */ 742 goto drop; 743 } 744 745 /* 746 * Be proactive about unspecified IPv6 address in source. 747 * As we use all-zero to indicate unbounded/unconnected pcb, 748 * unspecified IPv6 address can be used to confuse us. 749 * 750 * Note that packets with unspecified IPv6 destination is 751 * already dropped in ip6_input. 752 */ 753 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 754 /* XXX stat */ 755 goto drop; 756 } 757 758 /* 759 * Make sure destination address is not multicast. 760 * Source address checked in ip6_input(). 761 */ 762 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 763 /* XXX stat */ 764 goto drop; 765 } 766 767 /* We do the checksum after PCB lookup... */ 768 len = m->m_pkthdr.len; 769 tlen = len - toff; 770 break; 771 #endif 772 default: 773 m_freem(m); 774 return; 775 } 776 777 /* 778 * Check that TCP offset makes sense, 779 * pull out TCP options and adjust length. XXX 780 */ 781 off = th->th_off << 2; 782 if (off < sizeof (struct tcphdr) || off > tlen) { 783 tcpstat.tcps_rcvbadoff++; 784 goto drop; 785 } 786 tlen -= off; 787 788 /* 789 * tcp_input() has been modified to use tlen to mean the TCP data 790 * length throughout the function. Other functions can use 791 * m->m_pkthdr.len as the basis for calculating the TCP data length. 792 * rja 793 */ 794 795 if (off > sizeof (struct tcphdr)) { 796 #ifndef PULLDOWN_TEST 797 if (m->m_len < toff + off) { 798 if ((m = m_pullup(m, toff + off)) == 0) { 799 tcpstat.tcps_rcvshort++; 800 return; 801 } 802 switch (af) { 803 #ifdef INET 804 case AF_INET: 805 ip = mtod(m, struct ip *); 806 break; 807 #endif 808 #ifdef INET6 809 case AF_INET6: 810 ip6 = mtod(m, struct ip6_hdr *); 811 break; 812 #endif 813 } 814 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 815 } 816 #else 817 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 818 if (th == NULL) { 819 tcpstat.tcps_rcvshort++; 820 return; 821 } 822 /* 823 * NOTE: ip/ip6 will not be affected by m_pulldown() 824 * (as they're before toff) and we don't need to update those. 825 */ 826 #endif 827 optlen = off - sizeof (struct tcphdr); 828 optp = ((caddr_t)th) + sizeof(struct tcphdr); 829 /* 830 * Do quick retrieval of timestamp options ("options 831 * prediction?"). If timestamp is the only option and it's 832 * formatted as recommended in RFC 1323 appendix A, we 833 * quickly get the values now and not bother calling 834 * tcp_dooptions(), etc. 835 */ 836 if ((optlen == TCPOLEN_TSTAMP_APPA || 837 (optlen > TCPOLEN_TSTAMP_APPA && 838 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 839 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 840 (th->th_flags & TH_SYN) == 0) { 841 opti.ts_present = 1; 842 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 843 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 844 optp = NULL; /* we've parsed the options */ 845 } 846 } 847 tiflags = th->th_flags; 848 849 /* 850 * Locate pcb for segment. 851 */ 852 findpcb: 853 inp = NULL; 854 #ifdef INET6 855 in6p = NULL; 856 #endif 857 switch (af) { 858 #ifdef INET 859 case AF_INET: 860 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 861 ip->ip_dst, th->th_dport); 862 if (inp == 0) { 863 ++tcpstat.tcps_pcbhashmiss; 864 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 865 } 866 #ifdef INET6 867 if (inp == 0) { 868 struct in6_addr s, d; 869 870 /* mapped addr case */ 871 bzero(&s, sizeof(s)); 872 s.s6_addr16[5] = htons(0xffff); 873 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 874 bzero(&d, sizeof(d)); 875 d.s6_addr16[5] = htons(0xffff); 876 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 877 in6p = in6_pcblookup_connect(&tcb6, &s, th->th_sport, 878 &d, th->th_dport, 0); 879 if (in6p == 0) { 880 ++tcpstat.tcps_pcbhashmiss; 881 in6p = in6_pcblookup_bind(&tcb6, &d, 882 th->th_dport, 0); 883 } 884 } 885 #endif 886 #ifndef INET6 887 if (inp == 0) 888 #else 889 if (inp == 0 && in6p == 0) 890 #endif 891 { 892 ++tcpstat.tcps_noport; 893 if (tcp_log_refused && (tiflags & TH_SYN)) { 894 #ifndef INET6 895 char src[4*sizeof "123"]; 896 char dst[4*sizeof "123"]; 897 #else 898 char src[INET6_ADDRSTRLEN]; 899 char dst[INET6_ADDRSTRLEN]; 900 #endif 901 if (ip) { 902 strcpy(src, inet_ntoa(ip->ip_src)); 903 strcpy(dst, inet_ntoa(ip->ip_dst)); 904 } 905 #ifdef INET6 906 else if (ip6) { 907 strcpy(src, ip6_sprintf(&ip6->ip6_src)); 908 strcpy(dst, ip6_sprintf(&ip6->ip6_dst)); 909 } 910 #endif 911 else { 912 strcpy(src, "(unknown)"); 913 strcpy(dst, "(unknown)"); 914 } 915 log(LOG_INFO, 916 "Connection attempt to TCP %s:%d from %s:%d\n", 917 dst, ntohs(th->th_dport), 918 src, ntohs(th->th_sport)); 919 } 920 TCP_FIELDS_TO_HOST(th); 921 goto dropwithreset_ratelim; 922 } 923 #ifdef IPSEC 924 if (inp && ipsec4_in_reject(m, inp)) { 925 ipsecstat.in_polvio++; 926 goto drop; 927 } 928 #ifdef INET6 929 else if (in6p && ipsec4_in_reject_so(m, in6p->in6p_socket)) { 930 ipsecstat.in_polvio++; 931 goto drop; 932 } 933 #endif 934 #endif /*IPSEC*/ 935 break; 936 #endif /*INET*/ 937 #ifdef INET6 938 case AF_INET6: 939 { 940 int faith; 941 942 #if defined(NFAITH) && NFAITH > 0 943 faith = faithprefix(&ip6->ip6_dst); 944 #else 945 faith = 0; 946 #endif 947 in6p = in6_pcblookup_connect(&tcb6, &ip6->ip6_src, th->th_sport, 948 &ip6->ip6_dst, th->th_dport, faith); 949 if (in6p == NULL) { 950 ++tcpstat.tcps_pcbhashmiss; 951 in6p = in6_pcblookup_bind(&tcb6, &ip6->ip6_dst, 952 th->th_dport, faith); 953 } 954 if (in6p == NULL) { 955 ++tcpstat.tcps_noport; 956 TCP_FIELDS_TO_HOST(th); 957 goto dropwithreset_ratelim; 958 } 959 #ifdef IPSEC 960 if (ipsec6_in_reject(m, in6p)) { 961 ipsec6stat.in_polvio++; 962 goto drop; 963 } 964 #endif /*IPSEC*/ 965 break; 966 } 967 #endif 968 } 969 970 /* 971 * If the state is CLOSED (i.e., TCB does not exist) then 972 * all data in the incoming segment is discarded. 973 * If the TCB exists but is in CLOSED state, it is embryonic, 974 * but should either do a listen or a connect soon. 975 */ 976 tp = NULL; 977 so = NULL; 978 if (inp) { 979 tp = intotcpcb(inp); 980 so = inp->inp_socket; 981 } 982 #ifdef INET6 983 else if (in6p) { 984 tp = in6totcpcb(in6p); 985 so = in6p->in6p_socket; 986 } 987 #endif 988 if (tp == 0) { 989 TCP_FIELDS_TO_HOST(th); 990 goto dropwithreset_ratelim; 991 } 992 if (tp->t_state == TCPS_CLOSED) 993 goto drop; 994 995 /* 996 * Checksum extended TCP header and data. 997 */ 998 switch (af) { 999 #ifdef INET 1000 case AF_INET: 1001 switch (m->m_pkthdr.csum_flags & 1002 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 1003 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 1004 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 1005 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 1006 goto badcsum; 1007 1008 case M_CSUM_TCPv4|M_CSUM_DATA: 1009 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 1010 if ((m->m_pkthdr.csum_data ^ 0xffff) != 0) 1011 goto badcsum; 1012 break; 1013 1014 case M_CSUM_TCPv4: 1015 /* Checksum was okay. */ 1016 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 1017 break; 1018 1019 default: 1020 /* Must compute it ourselves. */ 1021 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 1022 #ifndef PULLDOWN_TEST 1023 { 1024 struct ipovly *ipov; 1025 ipov = (struct ipovly *)ip; 1026 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 1027 ipov->ih_len = htons(tlen + off); 1028 1029 if (in_cksum(m, len) != 0) 1030 goto badcsum; 1031 } 1032 #else 1033 if (in4_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 1034 goto badcsum; 1035 #endif /* ! PULLDOWN_TEST */ 1036 break; 1037 } 1038 break; 1039 #endif /* INET4 */ 1040 1041 #ifdef INET6 1042 case AF_INET6: 1043 if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 1044 goto badcsum; 1045 break; 1046 #endif /* INET6 */ 1047 } 1048 1049 TCP_FIELDS_TO_HOST(th); 1050 1051 /* Unscale the window into a 32-bit value. */ 1052 if ((tiflags & TH_SYN) == 0) 1053 tiwin = th->th_win << tp->snd_scale; 1054 else 1055 tiwin = th->th_win; 1056 1057 #ifdef INET6 1058 /* save packet options if user wanted */ 1059 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1060 if (in6p->in6p_options) { 1061 m_freem(in6p->in6p_options); 1062 in6p->in6p_options = 0; 1063 } 1064 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1065 } 1066 #endif 1067 1068 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1069 union syn_cache_sa src; 1070 union syn_cache_sa dst; 1071 1072 bzero(&src, sizeof(src)); 1073 bzero(&dst, sizeof(dst)); 1074 switch (af) { 1075 #ifdef INET 1076 case AF_INET: 1077 src.sin.sin_len = sizeof(struct sockaddr_in); 1078 src.sin.sin_family = AF_INET; 1079 src.sin.sin_addr = ip->ip_src; 1080 src.sin.sin_port = th->th_sport; 1081 1082 dst.sin.sin_len = sizeof(struct sockaddr_in); 1083 dst.sin.sin_family = AF_INET; 1084 dst.sin.sin_addr = ip->ip_dst; 1085 dst.sin.sin_port = th->th_dport; 1086 break; 1087 #endif 1088 #ifdef INET6 1089 case AF_INET6: 1090 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1091 src.sin6.sin6_family = AF_INET6; 1092 src.sin6.sin6_addr = ip6->ip6_src; 1093 src.sin6.sin6_port = th->th_sport; 1094 1095 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1096 dst.sin6.sin6_family = AF_INET6; 1097 dst.sin6.sin6_addr = ip6->ip6_dst; 1098 dst.sin6.sin6_port = th->th_dport; 1099 break; 1100 #endif /* INET6 */ 1101 default: 1102 goto badsyn; /*sanity*/ 1103 } 1104 1105 if (so->so_options & SO_DEBUG) { 1106 ostate = tp->t_state; 1107 1108 tcp_saveti = NULL; 1109 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1110 goto nosave; 1111 1112 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1113 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1114 if (!tcp_saveti) 1115 goto nosave; 1116 } else { 1117 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1118 if (!tcp_saveti) 1119 goto nosave; 1120 tcp_saveti->m_len = iphlen; 1121 m_copydata(m, 0, iphlen, 1122 mtod(tcp_saveti, caddr_t)); 1123 } 1124 1125 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1126 m_freem(tcp_saveti); 1127 tcp_saveti = NULL; 1128 } else { 1129 tcp_saveti->m_len += sizeof(struct tcphdr); 1130 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1131 sizeof(struct tcphdr)); 1132 } 1133 if (tcp_saveti) { 1134 /* 1135 * need to recover version # field, which was 1136 * overwritten on ip_cksum computation. 1137 */ 1138 struct ip *sip; 1139 sip = mtod(tcp_saveti, struct ip *); 1140 switch (af) { 1141 #ifdef INET 1142 case AF_INET: 1143 sip->ip_v = 4; 1144 break; 1145 #endif 1146 #ifdef INET6 1147 case AF_INET6: 1148 sip->ip_v = 6; 1149 break; 1150 #endif 1151 } 1152 } 1153 nosave:; 1154 } 1155 if (so->so_options & SO_ACCEPTCONN) { 1156 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1157 if (tiflags & TH_RST) { 1158 syn_cache_reset(&src.sa, &dst.sa, th); 1159 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1160 (TH_ACK|TH_SYN)) { 1161 /* 1162 * Received a SYN,ACK. This should 1163 * never happen while we are in 1164 * LISTEN. Send an RST. 1165 */ 1166 goto badsyn; 1167 } else if (tiflags & TH_ACK) { 1168 so = syn_cache_get(&src.sa, &dst.sa, 1169 th, toff, tlen, so, m); 1170 if (so == NULL) { 1171 /* 1172 * We don't have a SYN for 1173 * this ACK; send an RST. 1174 */ 1175 goto badsyn; 1176 } else if (so == 1177 (struct socket *)(-1)) { 1178 /* 1179 * We were unable to create 1180 * the connection. If the 1181 * 3-way handshake was 1182 * completed, and RST has 1183 * been sent to the peer. 1184 * Since the mbuf might be 1185 * in use for the reply, 1186 * do not free it. 1187 */ 1188 m = NULL; 1189 } else { 1190 /* 1191 * We have created a 1192 * full-blown connection. 1193 */ 1194 tp = NULL; 1195 inp = NULL; 1196 #ifdef INET6 1197 in6p = NULL; 1198 #endif 1199 switch (so->so_proto->pr_domain->dom_family) { 1200 #ifdef INET 1201 case AF_INET: 1202 inp = sotoinpcb(so); 1203 tp = intotcpcb(inp); 1204 break; 1205 #endif 1206 #ifdef INET6 1207 case AF_INET6: 1208 in6p = sotoin6pcb(so); 1209 tp = in6totcpcb(in6p); 1210 break; 1211 #endif 1212 } 1213 if (tp == NULL) 1214 goto badsyn; /*XXX*/ 1215 tiwin <<= tp->snd_scale; 1216 goto after_listen; 1217 } 1218 } else { 1219 /* 1220 * None of RST, SYN or ACK was set. 1221 * This is an invalid packet for a 1222 * TCB in LISTEN state. Send a RST. 1223 */ 1224 goto badsyn; 1225 } 1226 } else { 1227 /* 1228 * Received a SYN. 1229 */ 1230 1231 /* 1232 * LISTEN socket received a SYN 1233 * from itself? This can't possibly 1234 * be valid; drop the packet. 1235 */ 1236 if (th->th_sport == th->th_dport) { 1237 int i; 1238 1239 switch (af) { 1240 #ifdef INET 1241 case AF_INET: 1242 i = in_hosteq(ip->ip_src, ip->ip_dst); 1243 break; 1244 #endif 1245 #ifdef INET6 1246 case AF_INET6: 1247 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1248 break; 1249 #endif 1250 default: 1251 i = 1; 1252 } 1253 if (i) { 1254 tcpstat.tcps_badsyn++; 1255 goto drop; 1256 } 1257 } 1258 1259 /* 1260 * SYN looks ok; create compressed TCP 1261 * state for it. 1262 */ 1263 if (so->so_qlen <= so->so_qlimit && 1264 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1265 so, m, optp, optlen, &opti)) 1266 m = NULL; 1267 } 1268 goto drop; 1269 } 1270 } 1271 1272 after_listen: 1273 #ifdef DIAGNOSTIC 1274 /* 1275 * Should not happen now that all embryonic connections 1276 * are handled with compressed state. 1277 */ 1278 if (tp->t_state == TCPS_LISTEN) 1279 panic("tcp_input: TCPS_LISTEN"); 1280 #endif 1281 1282 /* 1283 * Segment received on connection. 1284 * Reset idle time and keep-alive timer. 1285 */ 1286 tp->t_rcvtime = tcp_now; 1287 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1288 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1289 1290 /* 1291 * Process options. 1292 */ 1293 if (optp) 1294 tcp_dooptions(tp, optp, optlen, th, &opti); 1295 1296 /* 1297 * Header prediction: check for the two common cases 1298 * of a uni-directional data xfer. If the packet has 1299 * no control flags, is in-sequence, the window didn't 1300 * change and we're not retransmitting, it's a 1301 * candidate. If the length is zero and the ack moved 1302 * forward, we're the sender side of the xfer. Just 1303 * free the data acked & wake any higher level process 1304 * that was blocked waiting for space. If the length 1305 * is non-zero and the ack didn't move, we're the 1306 * receiver side. If we're getting packets in-order 1307 * (the reassembly queue is empty), add the data to 1308 * the socket buffer and note that we need a delayed ack. 1309 */ 1310 if (tp->t_state == TCPS_ESTABLISHED && 1311 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1312 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1313 th->th_seq == tp->rcv_nxt && 1314 tiwin && tiwin == tp->snd_wnd && 1315 tp->snd_nxt == tp->snd_max) { 1316 1317 /* 1318 * If last ACK falls within this segment's sequence numbers, 1319 * record the timestamp. 1320 */ 1321 if (opti.ts_present && 1322 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1323 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1324 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1325 tp->ts_recent = opti.ts_val; 1326 } 1327 1328 if (tlen == 0) { 1329 if (SEQ_GT(th->th_ack, tp->snd_una) && 1330 SEQ_LEQ(th->th_ack, tp->snd_max) && 1331 tp->snd_cwnd >= tp->snd_wnd && 1332 tp->t_dupacks < tcprexmtthresh) { 1333 /* 1334 * this is a pure ack for outstanding data. 1335 */ 1336 ++tcpstat.tcps_predack; 1337 if (opti.ts_present && opti.ts_ecr) 1338 tcp_xmit_timer(tp, 1339 TCP_TIMESTAMP(tp) - opti.ts_ecr + 1); 1340 else if (tp->t_rtttime && 1341 SEQ_GT(th->th_ack, tp->t_rtseq)) 1342 tcp_xmit_timer(tp, 1343 tcp_now - tp->t_rtttime); 1344 acked = th->th_ack - tp->snd_una; 1345 tcpstat.tcps_rcvackpack++; 1346 tcpstat.tcps_rcvackbyte += acked; 1347 ND6_HINT(tp); 1348 sbdrop(&so->so_snd, acked); 1349 /* 1350 * We want snd_recover to track snd_una to 1351 * avoid sequence wraparound problems for 1352 * very large transfers. 1353 */ 1354 tp->snd_una = tp->snd_recover = th->th_ack; 1355 m_freem(m); 1356 1357 /* 1358 * If all outstanding data are acked, stop 1359 * retransmit timer, otherwise restart timer 1360 * using current (possibly backed-off) value. 1361 * If process is waiting for space, 1362 * wakeup/selwakeup/signal. If data 1363 * are ready to send, let tcp_output 1364 * decide between more output or persist. 1365 */ 1366 if (tp->snd_una == tp->snd_max) 1367 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1368 else if (TCP_TIMER_ISARMED(tp, 1369 TCPT_PERSIST) == 0) 1370 TCP_TIMER_ARM(tp, TCPT_REXMT, 1371 tp->t_rxtcur); 1372 1373 sowwakeup(so); 1374 if (so->so_snd.sb_cc) 1375 (void) tcp_output(tp); 1376 if (tcp_saveti) 1377 m_freem(tcp_saveti); 1378 return; 1379 } 1380 } else if (th->th_ack == tp->snd_una && 1381 LIST_FIRST(&tp->segq) == NULL && 1382 tlen <= sbspace(&so->so_rcv)) { 1383 /* 1384 * this is a pure, in-sequence data packet 1385 * with nothing on the reassembly queue and 1386 * we have enough buffer space to take it. 1387 */ 1388 ++tcpstat.tcps_preddat; 1389 tp->rcv_nxt += tlen; 1390 tcpstat.tcps_rcvpack++; 1391 tcpstat.tcps_rcvbyte += tlen; 1392 ND6_HINT(tp); 1393 /* 1394 * Drop TCP, IP headers and TCP options then add data 1395 * to socket buffer. 1396 */ 1397 m_adj(m, toff + off); 1398 sbappend(&so->so_rcv, m); 1399 sorwakeup(so); 1400 TCP_SETUP_ACK(tp, th); 1401 if (tp->t_flags & TF_ACKNOW) 1402 (void) tcp_output(tp); 1403 if (tcp_saveti) 1404 m_freem(tcp_saveti); 1405 return; 1406 } 1407 } 1408 1409 /* 1410 * Compute mbuf offset to TCP data segment. 1411 */ 1412 hdroptlen = toff + off; 1413 1414 /* 1415 * Calculate amount of space in receive window, 1416 * and then do TCP input processing. 1417 * Receive window is amount of space in rcv queue, 1418 * but not less than advertised window. 1419 */ 1420 { int win; 1421 1422 win = sbspace(&so->so_rcv); 1423 if (win < 0) 1424 win = 0; 1425 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1426 } 1427 1428 switch (tp->t_state) { 1429 1430 /* 1431 * If the state is SYN_SENT: 1432 * if seg contains an ACK, but not for our SYN, drop the input. 1433 * if seg contains a RST, then drop the connection. 1434 * if seg does not contain SYN, then drop it. 1435 * Otherwise this is an acceptable SYN segment 1436 * initialize tp->rcv_nxt and tp->irs 1437 * if seg contains ack then advance tp->snd_una 1438 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1439 * arrange for segment to be acked (eventually) 1440 * continue processing rest of data/controls, beginning with URG 1441 */ 1442 case TCPS_SYN_SENT: 1443 if ((tiflags & TH_ACK) && 1444 (SEQ_LEQ(th->th_ack, tp->iss) || 1445 SEQ_GT(th->th_ack, tp->snd_max))) 1446 goto dropwithreset; 1447 if (tiflags & TH_RST) { 1448 if (tiflags & TH_ACK) 1449 tp = tcp_drop(tp, ECONNREFUSED); 1450 goto drop; 1451 } 1452 if ((tiflags & TH_SYN) == 0) 1453 goto drop; 1454 if (tiflags & TH_ACK) { 1455 tp->snd_una = tp->snd_recover = th->th_ack; 1456 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1457 tp->snd_nxt = tp->snd_una; 1458 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1459 } 1460 tp->irs = th->th_seq; 1461 tcp_rcvseqinit(tp); 1462 tp->t_flags |= TF_ACKNOW; 1463 tcp_mss_from_peer(tp, opti.maxseg); 1464 1465 /* 1466 * Initialize the initial congestion window. If we 1467 * had to retransmit the SYN, we must initialize cwnd 1468 * to 1 segment (i.e. the Loss Window). 1469 */ 1470 if (tp->t_flags & TF_SYN_REXMT) 1471 tp->snd_cwnd = tp->t_peermss; 1472 else 1473 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, 1474 tp->t_peermss); 1475 1476 tcp_rmx_rtt(tp); 1477 if (tiflags & TH_ACK) { 1478 tcpstat.tcps_connects++; 1479 soisconnected(so); 1480 tcp_established(tp); 1481 /* Do window scaling on this connection? */ 1482 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1483 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1484 tp->snd_scale = tp->requested_s_scale; 1485 tp->rcv_scale = tp->request_r_scale; 1486 } 1487 TCP_REASS_LOCK(tp); 1488 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1489 TCP_REASS_UNLOCK(tp); 1490 /* 1491 * if we didn't have to retransmit the SYN, 1492 * use its rtt as our initial srtt & rtt var. 1493 */ 1494 if (tp->t_rtttime) 1495 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1496 } else 1497 tp->t_state = TCPS_SYN_RECEIVED; 1498 1499 /* 1500 * Advance th->th_seq to correspond to first data byte. 1501 * If data, trim to stay within window, 1502 * dropping FIN if necessary. 1503 */ 1504 th->th_seq++; 1505 if (tlen > tp->rcv_wnd) { 1506 todrop = tlen - tp->rcv_wnd; 1507 m_adj(m, -todrop); 1508 tlen = tp->rcv_wnd; 1509 tiflags &= ~TH_FIN; 1510 tcpstat.tcps_rcvpackafterwin++; 1511 tcpstat.tcps_rcvbyteafterwin += todrop; 1512 } 1513 tp->snd_wl1 = th->th_seq - 1; 1514 tp->rcv_up = th->th_seq; 1515 goto step6; 1516 1517 /* 1518 * If the state is SYN_RECEIVED: 1519 * If seg contains an ACK, but not for our SYN, drop the input 1520 * and generate an RST. See page 36, rfc793 1521 */ 1522 case TCPS_SYN_RECEIVED: 1523 if ((tiflags & TH_ACK) && 1524 (SEQ_LEQ(th->th_ack, tp->iss) || 1525 SEQ_GT(th->th_ack, tp->snd_max))) 1526 goto dropwithreset; 1527 break; 1528 } 1529 1530 /* 1531 * States other than LISTEN or SYN_SENT. 1532 * First check timestamp, if present. 1533 * Then check that at least some bytes of segment are within 1534 * receive window. If segment begins before rcv_nxt, 1535 * drop leading data (and SYN); if nothing left, just ack. 1536 * 1537 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1538 * and it's less than ts_recent, drop it. 1539 */ 1540 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1541 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1542 1543 /* Check to see if ts_recent is over 24 days old. */ 1544 if ((int)(TCP_TIMESTAMP(tp) - tp->ts_recent_age) > 1545 TCP_PAWS_IDLE) { 1546 /* 1547 * Invalidate ts_recent. If this segment updates 1548 * ts_recent, the age will be reset later and ts_recent 1549 * will get a valid value. If it does not, setting 1550 * ts_recent to zero will at least satisfy the 1551 * requirement that zero be placed in the timestamp 1552 * echo reply when ts_recent isn't valid. The 1553 * age isn't reset until we get a valid ts_recent 1554 * because we don't want out-of-order segments to be 1555 * dropped when ts_recent is old. 1556 */ 1557 tp->ts_recent = 0; 1558 } else { 1559 tcpstat.tcps_rcvduppack++; 1560 tcpstat.tcps_rcvdupbyte += tlen; 1561 tcpstat.tcps_pawsdrop++; 1562 goto dropafterack; 1563 } 1564 } 1565 1566 todrop = tp->rcv_nxt - th->th_seq; 1567 if (todrop > 0) { 1568 if (tiflags & TH_SYN) { 1569 tiflags &= ~TH_SYN; 1570 th->th_seq++; 1571 if (th->th_urp > 1) 1572 th->th_urp--; 1573 else { 1574 tiflags &= ~TH_URG; 1575 th->th_urp = 0; 1576 } 1577 todrop--; 1578 } 1579 if (todrop > tlen || 1580 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1581 /* 1582 * Any valid FIN must be to the left of the window. 1583 * At this point the FIN must be a duplicate or 1584 * out of sequence; drop it. 1585 */ 1586 tiflags &= ~TH_FIN; 1587 /* 1588 * Send an ACK to resynchronize and drop any data. 1589 * But keep on processing for RST or ACK. 1590 */ 1591 tp->t_flags |= TF_ACKNOW; 1592 todrop = tlen; 1593 tcpstat.tcps_rcvdupbyte += todrop; 1594 tcpstat.tcps_rcvduppack++; 1595 } else { 1596 tcpstat.tcps_rcvpartduppack++; 1597 tcpstat.tcps_rcvpartdupbyte += todrop; 1598 } 1599 hdroptlen += todrop; /*drop from head afterwards*/ 1600 th->th_seq += todrop; 1601 tlen -= todrop; 1602 if (th->th_urp > todrop) 1603 th->th_urp -= todrop; 1604 else { 1605 tiflags &= ~TH_URG; 1606 th->th_urp = 0; 1607 } 1608 } 1609 1610 /* 1611 * If new data are received on a connection after the 1612 * user processes are gone, then RST the other end. 1613 */ 1614 if ((so->so_state & SS_NOFDREF) && 1615 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1616 tp = tcp_close(tp); 1617 tcpstat.tcps_rcvafterclose++; 1618 goto dropwithreset; 1619 } 1620 1621 /* 1622 * If segment ends after window, drop trailing data 1623 * (and PUSH and FIN); if nothing left, just ACK. 1624 */ 1625 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1626 if (todrop > 0) { 1627 tcpstat.tcps_rcvpackafterwin++; 1628 if (todrop >= tlen) { 1629 tcpstat.tcps_rcvbyteafterwin += tlen; 1630 /* 1631 * If a new connection request is received 1632 * while in TIME_WAIT, drop the old connection 1633 * and start over if the sequence numbers 1634 * are above the previous ones. 1635 */ 1636 if (tiflags & TH_SYN && 1637 tp->t_state == TCPS_TIME_WAIT && 1638 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1639 iss = tcp_new_iss(tp, tp->snd_nxt); 1640 tp = tcp_close(tp); 1641 goto findpcb; 1642 } 1643 /* 1644 * If window is closed can only take segments at 1645 * window edge, and have to drop data and PUSH from 1646 * incoming segments. Continue processing, but 1647 * remember to ack. Otherwise, drop segment 1648 * and ack. 1649 */ 1650 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1651 tp->t_flags |= TF_ACKNOW; 1652 tcpstat.tcps_rcvwinprobe++; 1653 } else 1654 goto dropafterack; 1655 } else 1656 tcpstat.tcps_rcvbyteafterwin += todrop; 1657 m_adj(m, -todrop); 1658 tlen -= todrop; 1659 tiflags &= ~(TH_PUSH|TH_FIN); 1660 } 1661 1662 /* 1663 * If last ACK falls within this segment's sequence numbers, 1664 * and the timestamp is newer, record it. 1665 */ 1666 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1667 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1668 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1669 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1670 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1671 tp->ts_recent = opti.ts_val; 1672 } 1673 1674 /* 1675 * If the RST bit is set examine the state: 1676 * SYN_RECEIVED STATE: 1677 * If passive open, return to LISTEN state. 1678 * If active open, inform user that connection was refused. 1679 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1680 * Inform user that connection was reset, and close tcb. 1681 * CLOSING, LAST_ACK, TIME_WAIT STATES 1682 * Close the tcb. 1683 */ 1684 if (tiflags&TH_RST) switch (tp->t_state) { 1685 1686 case TCPS_SYN_RECEIVED: 1687 so->so_error = ECONNREFUSED; 1688 goto close; 1689 1690 case TCPS_ESTABLISHED: 1691 case TCPS_FIN_WAIT_1: 1692 case TCPS_FIN_WAIT_2: 1693 case TCPS_CLOSE_WAIT: 1694 so->so_error = ECONNRESET; 1695 close: 1696 tp->t_state = TCPS_CLOSED; 1697 tcpstat.tcps_drops++; 1698 tp = tcp_close(tp); 1699 goto drop; 1700 1701 case TCPS_CLOSING: 1702 case TCPS_LAST_ACK: 1703 case TCPS_TIME_WAIT: 1704 tp = tcp_close(tp); 1705 goto drop; 1706 } 1707 1708 /* 1709 * If a SYN is in the window, then this is an 1710 * error and we send an RST and drop the connection. 1711 */ 1712 if (tiflags & TH_SYN) { 1713 tp = tcp_drop(tp, ECONNRESET); 1714 goto dropwithreset; 1715 } 1716 1717 /* 1718 * If the ACK bit is off we drop the segment and return. 1719 */ 1720 if ((tiflags & TH_ACK) == 0) { 1721 if (tp->t_flags & TF_ACKNOW) 1722 goto dropafterack; 1723 else 1724 goto drop; 1725 } 1726 1727 /* 1728 * Ack processing. 1729 */ 1730 switch (tp->t_state) { 1731 1732 /* 1733 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1734 * ESTABLISHED state and continue processing, otherwise 1735 * send an RST. 1736 */ 1737 case TCPS_SYN_RECEIVED: 1738 if (SEQ_GT(tp->snd_una, th->th_ack) || 1739 SEQ_GT(th->th_ack, tp->snd_max)) 1740 goto dropwithreset; 1741 tcpstat.tcps_connects++; 1742 soisconnected(so); 1743 tcp_established(tp); 1744 /* Do window scaling? */ 1745 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1746 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1747 tp->snd_scale = tp->requested_s_scale; 1748 tp->rcv_scale = tp->request_r_scale; 1749 } 1750 TCP_REASS_LOCK(tp); 1751 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1752 TCP_REASS_UNLOCK(tp); 1753 tp->snd_wl1 = th->th_seq - 1; 1754 /* fall into ... */ 1755 1756 /* 1757 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1758 * ACKs. If the ack is in the range 1759 * tp->snd_una < th->th_ack <= tp->snd_max 1760 * then advance tp->snd_una to th->th_ack and drop 1761 * data from the retransmission queue. If this ACK reflects 1762 * more up to date window information we update our window information. 1763 */ 1764 case TCPS_ESTABLISHED: 1765 case TCPS_FIN_WAIT_1: 1766 case TCPS_FIN_WAIT_2: 1767 case TCPS_CLOSE_WAIT: 1768 case TCPS_CLOSING: 1769 case TCPS_LAST_ACK: 1770 case TCPS_TIME_WAIT: 1771 1772 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1773 if (tlen == 0 && tiwin == tp->snd_wnd) { 1774 tcpstat.tcps_rcvdupack++; 1775 /* 1776 * If we have outstanding data (other than 1777 * a window probe), this is a completely 1778 * duplicate ack (ie, window info didn't 1779 * change), the ack is the biggest we've 1780 * seen and we've seen exactly our rexmt 1781 * threshhold of them, assume a packet 1782 * has been dropped and retransmit it. 1783 * Kludge snd_nxt & the congestion 1784 * window so we send only this one 1785 * packet. 1786 * 1787 * We know we're losing at the current 1788 * window size so do congestion avoidance 1789 * (set ssthresh to half the current window 1790 * and pull our congestion window back to 1791 * the new ssthresh). 1792 * 1793 * Dup acks mean that packets have left the 1794 * network (they're now cached at the receiver) 1795 * so bump cwnd by the amount in the receiver 1796 * to keep a constant cwnd packets in the 1797 * network. 1798 */ 1799 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 1800 th->th_ack != tp->snd_una) 1801 tp->t_dupacks = 0; 1802 else if (++tp->t_dupacks == tcprexmtthresh) { 1803 tcp_seq onxt = tp->snd_nxt; 1804 u_int win = 1805 min(tp->snd_wnd, tp->snd_cwnd) / 1806 2 / tp->t_segsz; 1807 if (tcp_do_newreno && SEQ_LT(th->th_ack, 1808 tp->snd_recover)) { 1809 /* 1810 * False fast retransmit after 1811 * timeout. Do not cut window. 1812 */ 1813 tp->snd_cwnd += tp->t_segsz; 1814 tp->t_dupacks = 0; 1815 (void) tcp_output(tp); 1816 goto drop; 1817 } 1818 1819 if (win < 2) 1820 win = 2; 1821 tp->snd_ssthresh = win * tp->t_segsz; 1822 tp->snd_recover = tp->snd_max; 1823 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1824 tp->t_rtttime = 0; 1825 tp->snd_nxt = th->th_ack; 1826 tp->snd_cwnd = tp->t_segsz; 1827 (void) tcp_output(tp); 1828 tp->snd_cwnd = tp->snd_ssthresh + 1829 tp->t_segsz * tp->t_dupacks; 1830 if (SEQ_GT(onxt, tp->snd_nxt)) 1831 tp->snd_nxt = onxt; 1832 goto drop; 1833 } else if (tp->t_dupacks > tcprexmtthresh) { 1834 tp->snd_cwnd += tp->t_segsz; 1835 (void) tcp_output(tp); 1836 goto drop; 1837 } 1838 } else 1839 tp->t_dupacks = 0; 1840 break; 1841 } 1842 /* 1843 * If the congestion window was inflated to account 1844 * for the other side's cached packets, retract it. 1845 */ 1846 if (tcp_do_newreno == 0) { 1847 if (tp->t_dupacks >= tcprexmtthresh && 1848 tp->snd_cwnd > tp->snd_ssthresh) 1849 tp->snd_cwnd = tp->snd_ssthresh; 1850 tp->t_dupacks = 0; 1851 } else if (tp->t_dupacks >= tcprexmtthresh && 1852 tcp_newreno(tp, th) == 0) { 1853 tp->snd_cwnd = tp->snd_ssthresh; 1854 /* 1855 * Window inflation should have left us with approx. 1856 * snd_ssthresh outstanding data. But in case we 1857 * would be inclined to send a burst, better to do 1858 * it via the slow start mechanism. 1859 */ 1860 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 1861 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 1862 + tp->t_segsz; 1863 tp->t_dupacks = 0; 1864 } 1865 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1866 tcpstat.tcps_rcvacktoomuch++; 1867 goto dropafterack; 1868 } 1869 acked = th->th_ack - tp->snd_una; 1870 tcpstat.tcps_rcvackpack++; 1871 tcpstat.tcps_rcvackbyte += acked; 1872 1873 /* 1874 * If we have a timestamp reply, update smoothed 1875 * round trip time. If no timestamp is present but 1876 * transmit timer is running and timed sequence 1877 * number was acked, update smoothed round trip time. 1878 * Since we now have an rtt measurement, cancel the 1879 * timer backoff (cf., Phil Karn's retransmit alg.). 1880 * Recompute the initial retransmit timer. 1881 */ 1882 if (opti.ts_present && opti.ts_ecr) 1883 tcp_xmit_timer(tp, TCP_TIMESTAMP(tp) - opti.ts_ecr + 1); 1884 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 1885 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1886 1887 /* 1888 * If all outstanding data is acked, stop retransmit 1889 * timer and remember to restart (more output or persist). 1890 * If there is more data to be acked, restart retransmit 1891 * timer, using current (possibly backed-off) value. 1892 */ 1893 if (th->th_ack == tp->snd_max) { 1894 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1895 needoutput = 1; 1896 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 1897 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1898 /* 1899 * When new data is acked, open the congestion window. 1900 * If the window gives us less than ssthresh packets 1901 * in flight, open exponentially (segsz per packet). 1902 * Otherwise open linearly: segsz per window 1903 * (segsz^2 / cwnd per packet), plus a constant 1904 * fraction of a packet (segsz/8) to help larger windows 1905 * open quickly enough. 1906 */ 1907 { 1908 u_int cw = tp->snd_cwnd; 1909 u_int incr = tp->t_segsz; 1910 1911 if (cw > tp->snd_ssthresh) 1912 incr = incr * incr / cw; 1913 if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover)) 1914 tp->snd_cwnd = min(cw + incr, 1915 TCP_MAXWIN << tp->snd_scale); 1916 } 1917 ND6_HINT(tp); 1918 if (acked > so->so_snd.sb_cc) { 1919 tp->snd_wnd -= so->so_snd.sb_cc; 1920 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1921 ourfinisacked = 1; 1922 } else { 1923 sbdrop(&so->so_snd, acked); 1924 tp->snd_wnd -= acked; 1925 ourfinisacked = 0; 1926 } 1927 sowwakeup(so); 1928 /* 1929 * We want snd_recover to track snd_una to 1930 * avoid sequence wraparound problems for 1931 * very large transfers. 1932 */ 1933 tp->snd_una = tp->snd_recover = th->th_ack; 1934 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1935 tp->snd_nxt = tp->snd_una; 1936 1937 switch (tp->t_state) { 1938 1939 /* 1940 * In FIN_WAIT_1 STATE in addition to the processing 1941 * for the ESTABLISHED state if our FIN is now acknowledged 1942 * then enter FIN_WAIT_2. 1943 */ 1944 case TCPS_FIN_WAIT_1: 1945 if (ourfinisacked) { 1946 /* 1947 * If we can't receive any more 1948 * data, then closing user can proceed. 1949 * Starting the timer is contrary to the 1950 * specification, but if we don't get a FIN 1951 * we'll hang forever. 1952 */ 1953 if (so->so_state & SS_CANTRCVMORE) { 1954 soisdisconnected(so); 1955 if (tcp_maxidle > 0) 1956 TCP_TIMER_ARM(tp, TCPT_2MSL, 1957 tcp_maxidle); 1958 } 1959 tp->t_state = TCPS_FIN_WAIT_2; 1960 } 1961 break; 1962 1963 /* 1964 * In CLOSING STATE in addition to the processing for 1965 * the ESTABLISHED state if the ACK acknowledges our FIN 1966 * then enter the TIME-WAIT state, otherwise ignore 1967 * the segment. 1968 */ 1969 case TCPS_CLOSING: 1970 if (ourfinisacked) { 1971 tp->t_state = TCPS_TIME_WAIT; 1972 tcp_canceltimers(tp); 1973 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1974 soisdisconnected(so); 1975 } 1976 break; 1977 1978 /* 1979 * In LAST_ACK, we may still be waiting for data to drain 1980 * and/or to be acked, as well as for the ack of our FIN. 1981 * If our FIN is now acknowledged, delete the TCB, 1982 * enter the closed state and return. 1983 */ 1984 case TCPS_LAST_ACK: 1985 if (ourfinisacked) { 1986 tp = tcp_close(tp); 1987 goto drop; 1988 } 1989 break; 1990 1991 /* 1992 * In TIME_WAIT state the only thing that should arrive 1993 * is a retransmission of the remote FIN. Acknowledge 1994 * it and restart the finack timer. 1995 */ 1996 case TCPS_TIME_WAIT: 1997 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1998 goto dropafterack; 1999 } 2000 } 2001 2002 step6: 2003 /* 2004 * Update window information. 2005 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2006 */ 2007 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2008 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 2009 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 2010 /* keep track of pure window updates */ 2011 if (tlen == 0 && 2012 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2013 tcpstat.tcps_rcvwinupd++; 2014 tp->snd_wnd = tiwin; 2015 tp->snd_wl1 = th->th_seq; 2016 tp->snd_wl2 = th->th_ack; 2017 if (tp->snd_wnd > tp->max_sndwnd) 2018 tp->max_sndwnd = tp->snd_wnd; 2019 needoutput = 1; 2020 } 2021 2022 /* 2023 * Process segments with URG. 2024 */ 2025 if ((tiflags & TH_URG) && th->th_urp && 2026 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2027 /* 2028 * This is a kludge, but if we receive and accept 2029 * random urgent pointers, we'll crash in 2030 * soreceive. It's hard to imagine someone 2031 * actually wanting to send this much urgent data. 2032 */ 2033 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2034 th->th_urp = 0; /* XXX */ 2035 tiflags &= ~TH_URG; /* XXX */ 2036 goto dodata; /* XXX */ 2037 } 2038 /* 2039 * If this segment advances the known urgent pointer, 2040 * then mark the data stream. This should not happen 2041 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2042 * a FIN has been received from the remote side. 2043 * In these states we ignore the URG. 2044 * 2045 * According to RFC961 (Assigned Protocols), 2046 * the urgent pointer points to the last octet 2047 * of urgent data. We continue, however, 2048 * to consider it to indicate the first octet 2049 * of data past the urgent section as the original 2050 * spec states (in one of two places). 2051 */ 2052 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2053 tp->rcv_up = th->th_seq + th->th_urp; 2054 so->so_oobmark = so->so_rcv.sb_cc + 2055 (tp->rcv_up - tp->rcv_nxt) - 1; 2056 if (so->so_oobmark == 0) 2057 so->so_state |= SS_RCVATMARK; 2058 sohasoutofband(so); 2059 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2060 } 2061 /* 2062 * Remove out of band data so doesn't get presented to user. 2063 * This can happen independent of advancing the URG pointer, 2064 * but if two URG's are pending at once, some out-of-band 2065 * data may creep in... ick. 2066 */ 2067 if (th->th_urp <= (u_int16_t) tlen 2068 #ifdef SO_OOBINLINE 2069 && (so->so_options & SO_OOBINLINE) == 0 2070 #endif 2071 ) 2072 tcp_pulloutofband(so, th, m, hdroptlen); 2073 } else 2074 /* 2075 * If no out of band data is expected, 2076 * pull receive urgent pointer along 2077 * with the receive window. 2078 */ 2079 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2080 tp->rcv_up = tp->rcv_nxt; 2081 dodata: /* XXX */ 2082 2083 /* 2084 * Process the segment text, merging it into the TCP sequencing queue, 2085 * and arranging for acknowledgement of receipt if necessary. 2086 * This process logically involves adjusting tp->rcv_wnd as data 2087 * is presented to the user (this happens in tcp_usrreq.c, 2088 * case PRU_RCVD). If a FIN has already been received on this 2089 * connection then we just ignore the text. 2090 */ 2091 if ((tlen || (tiflags & TH_FIN)) && 2092 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2093 /* 2094 * Insert segment ti into reassembly queue of tcp with 2095 * control block tp. Return TH_FIN if reassembly now includes 2096 * a segment with FIN. The macro form does the common case 2097 * inline (segment is the next to be received on an 2098 * established connection, and the queue is empty), 2099 * avoiding linkage into and removal from the queue and 2100 * repetition of various conversions. 2101 * Set DELACK for segments received in order, but ack 2102 * immediately when segments are out of order 2103 * (so fast retransmit can work). 2104 */ 2105 /* NOTE: this was TCP_REASS() macro, but used only once */ 2106 TCP_REASS_LOCK(tp); 2107 if (th->th_seq == tp->rcv_nxt && 2108 LIST_FIRST(&tp->segq) == NULL && 2109 tp->t_state == TCPS_ESTABLISHED) { 2110 TCP_SETUP_ACK(tp, th); 2111 tp->rcv_nxt += tlen; 2112 tiflags = th->th_flags & TH_FIN; 2113 tcpstat.tcps_rcvpack++; 2114 tcpstat.tcps_rcvbyte += tlen; 2115 ND6_HINT(tp); 2116 m_adj(m, hdroptlen); 2117 sbappend(&(so)->so_rcv, m); 2118 sorwakeup(so); 2119 } else { 2120 m_adj(m, hdroptlen); 2121 tiflags = tcp_reass(tp, th, m, &tlen); 2122 tp->t_flags |= TF_ACKNOW; 2123 } 2124 TCP_REASS_UNLOCK(tp); 2125 2126 /* 2127 * Note the amount of data that peer has sent into 2128 * our window, in order to estimate the sender's 2129 * buffer size. 2130 */ 2131 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2132 } else { 2133 m_freem(m); 2134 m = NULL; 2135 tiflags &= ~TH_FIN; 2136 } 2137 2138 /* 2139 * If FIN is received ACK the FIN and let the user know 2140 * that the connection is closing. Ignore a FIN received before 2141 * the connection is fully established. 2142 */ 2143 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2144 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2145 socantrcvmore(so); 2146 tp->t_flags |= TF_ACKNOW; 2147 tp->rcv_nxt++; 2148 } 2149 switch (tp->t_state) { 2150 2151 /* 2152 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2153 */ 2154 case TCPS_ESTABLISHED: 2155 tp->t_state = TCPS_CLOSE_WAIT; 2156 break; 2157 2158 /* 2159 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2160 * enter the CLOSING state. 2161 */ 2162 case TCPS_FIN_WAIT_1: 2163 tp->t_state = TCPS_CLOSING; 2164 break; 2165 2166 /* 2167 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2168 * starting the time-wait timer, turning off the other 2169 * standard timers. 2170 */ 2171 case TCPS_FIN_WAIT_2: 2172 tp->t_state = TCPS_TIME_WAIT; 2173 tcp_canceltimers(tp); 2174 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2175 soisdisconnected(so); 2176 break; 2177 2178 /* 2179 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2180 */ 2181 case TCPS_TIME_WAIT: 2182 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2183 break; 2184 } 2185 } 2186 #ifdef TCP_DEBUG 2187 if (so->so_options & SO_DEBUG) 2188 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2189 #endif 2190 2191 /* 2192 * Return any desired output. 2193 */ 2194 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2195 (void) tcp_output(tp); 2196 if (tcp_saveti) 2197 m_freem(tcp_saveti); 2198 return; 2199 2200 badsyn: 2201 /* 2202 * Received a bad SYN. Increment counters and dropwithreset. 2203 */ 2204 tcpstat.tcps_badsyn++; 2205 tp = NULL; 2206 goto dropwithreset; 2207 2208 dropafterack: 2209 /* 2210 * Generate an ACK dropping incoming segment if it occupies 2211 * sequence space, where the ACK reflects our state. 2212 */ 2213 if (tiflags & TH_RST) 2214 goto drop; 2215 m_freem(m); 2216 tp->t_flags |= TF_ACKNOW; 2217 (void) tcp_output(tp); 2218 if (tcp_saveti) 2219 m_freem(tcp_saveti); 2220 return; 2221 2222 dropwithreset_ratelim: 2223 /* 2224 * We may want to rate-limit RSTs in certain situations, 2225 * particularly if we are sending an RST in response to 2226 * an attempt to connect to or otherwise communicate with 2227 * a port for which we have no socket. 2228 */ 2229 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2230 tcp_rst_ppslim) == 0) { 2231 /* XXX stat */ 2232 goto drop; 2233 } 2234 /* ...fall into dropwithreset... */ 2235 2236 dropwithreset: 2237 /* 2238 * Generate a RST, dropping incoming segment. 2239 * Make ACK acceptable to originator of segment. 2240 */ 2241 if (tiflags & TH_RST) 2242 goto drop; 2243 { 2244 /* 2245 * need to recover version # field, which was overwritten on 2246 * ip_cksum computation. 2247 */ 2248 struct ip *sip; 2249 sip = mtod(m, struct ip *); 2250 switch (af) { 2251 #ifdef INET 2252 case AF_INET: 2253 sip->ip_v = 4; 2254 break; 2255 #endif 2256 #ifdef INET6 2257 case AF_INET6: 2258 sip->ip_v = 6; 2259 break; 2260 #endif 2261 } 2262 } 2263 if (tiflags & TH_ACK) 2264 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2265 else { 2266 if (tiflags & TH_SYN) 2267 tlen++; 2268 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2269 TH_RST|TH_ACK); 2270 } 2271 if (tcp_saveti) 2272 m_freem(tcp_saveti); 2273 return; 2274 2275 badcsum: 2276 tcpstat.tcps_rcvbadsum++; 2277 drop: 2278 /* 2279 * Drop space held by incoming segment and return. 2280 */ 2281 if (tp) { 2282 if (tp->t_inpcb) 2283 so = tp->t_inpcb->inp_socket; 2284 #ifdef INET6 2285 else if (tp->t_in6pcb) 2286 so = tp->t_in6pcb->in6p_socket; 2287 #endif 2288 else 2289 so = NULL; 2290 #ifdef TCP_DEBUG 2291 if (so && (so->so_options & SO_DEBUG) != 0) 2292 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2293 #endif 2294 } 2295 if (tcp_saveti) 2296 m_freem(tcp_saveti); 2297 m_freem(m); 2298 return; 2299 } 2300 2301 void 2302 tcp_dooptions(tp, cp, cnt, th, oi) 2303 struct tcpcb *tp; 2304 u_char *cp; 2305 int cnt; 2306 struct tcphdr *th; 2307 struct tcp_opt_info *oi; 2308 { 2309 u_int16_t mss; 2310 int opt, optlen; 2311 2312 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2313 opt = cp[0]; 2314 if (opt == TCPOPT_EOL) 2315 break; 2316 if (opt == TCPOPT_NOP) 2317 optlen = 1; 2318 else { 2319 if (cnt < 2) 2320 break; 2321 optlen = cp[1]; 2322 if (optlen < 2 || optlen > cnt) 2323 break; 2324 } 2325 switch (opt) { 2326 2327 default: 2328 continue; 2329 2330 case TCPOPT_MAXSEG: 2331 if (optlen != TCPOLEN_MAXSEG) 2332 continue; 2333 if (!(th->th_flags & TH_SYN)) 2334 continue; 2335 bcopy(cp + 2, &mss, sizeof(mss)); 2336 oi->maxseg = ntohs(mss); 2337 break; 2338 2339 case TCPOPT_WINDOW: 2340 if (optlen != TCPOLEN_WINDOW) 2341 continue; 2342 if (!(th->th_flags & TH_SYN)) 2343 continue; 2344 tp->t_flags |= TF_RCVD_SCALE; 2345 tp->requested_s_scale = cp[2]; 2346 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2347 #if 0 /*XXX*/ 2348 char *p; 2349 2350 if (ip) 2351 p = ntohl(ip->ip_src); 2352 #ifdef INET6 2353 else if (ip6) 2354 p = ip6_sprintf(&ip6->ip6_src); 2355 #endif 2356 else 2357 p = "(unknown)"; 2358 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2359 "assuming %d\n", 2360 tp->requested_s_scale, p, 2361 TCP_MAX_WINSHIFT); 2362 #else 2363 log(LOG_ERR, "TCP: invalid wscale %d, " 2364 "assuming %d\n", 2365 tp->requested_s_scale, 2366 TCP_MAX_WINSHIFT); 2367 #endif 2368 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2369 } 2370 break; 2371 2372 case TCPOPT_TIMESTAMP: 2373 if (optlen != TCPOLEN_TIMESTAMP) 2374 continue; 2375 oi->ts_present = 1; 2376 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2377 NTOHL(oi->ts_val); 2378 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2379 NTOHL(oi->ts_ecr); 2380 2381 /* 2382 * A timestamp received in a SYN makes 2383 * it ok to send timestamp requests and replies. 2384 */ 2385 if (th->th_flags & TH_SYN) { 2386 tp->t_flags |= TF_RCVD_TSTMP; 2387 tp->ts_recent = oi->ts_val; 2388 tp->ts_recent_age = TCP_TIMESTAMP(tp); 2389 } 2390 break; 2391 case TCPOPT_SACK_PERMITTED: 2392 if (optlen != TCPOLEN_SACK_PERMITTED) 2393 continue; 2394 if (!(th->th_flags & TH_SYN)) 2395 continue; 2396 tp->t_flags &= ~TF_CANT_TXSACK; 2397 break; 2398 2399 case TCPOPT_SACK: 2400 if (tp->t_flags & TF_IGNR_RXSACK) 2401 continue; 2402 if (optlen % 8 != 2 || optlen < 10) 2403 continue; 2404 cp += 2; 2405 optlen -= 2; 2406 for (; optlen > 0; cp -= 8, optlen -= 8) { 2407 tcp_seq lwe, rwe; 2408 bcopy((char *)cp, (char *) &lwe, sizeof(lwe)); 2409 NTOHL(lwe); 2410 bcopy((char *)cp, (char *) &rwe, sizeof(rwe)); 2411 NTOHL(rwe); 2412 /* tcp_mark_sacked(tp, lwe, rwe); */ 2413 } 2414 break; 2415 } 2416 } 2417 } 2418 2419 /* 2420 * Pull out of band byte out of a segment so 2421 * it doesn't appear in the user's data queue. 2422 * It is still reflected in the segment length for 2423 * sequencing purposes. 2424 */ 2425 void 2426 tcp_pulloutofband(so, th, m, off) 2427 struct socket *so; 2428 struct tcphdr *th; 2429 struct mbuf *m; 2430 int off; 2431 { 2432 int cnt = off + th->th_urp - 1; 2433 2434 while (cnt >= 0) { 2435 if (m->m_len > cnt) { 2436 char *cp = mtod(m, caddr_t) + cnt; 2437 struct tcpcb *tp = sototcpcb(so); 2438 2439 tp->t_iobc = *cp; 2440 tp->t_oobflags |= TCPOOB_HAVEDATA; 2441 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2442 m->m_len--; 2443 return; 2444 } 2445 cnt -= m->m_len; 2446 m = m->m_next; 2447 if (m == 0) 2448 break; 2449 } 2450 panic("tcp_pulloutofband"); 2451 } 2452 2453 /* 2454 * Collect new round-trip time estimate 2455 * and update averages and current timeout. 2456 */ 2457 void 2458 tcp_xmit_timer(tp, rtt) 2459 struct tcpcb *tp; 2460 uint32_t rtt; 2461 { 2462 int32_t delta; 2463 2464 tcpstat.tcps_rttupdated++; 2465 if (tp->t_srtt != 0) { 2466 /* 2467 * srtt is stored as fixed point with 3 bits after the 2468 * binary point (i.e., scaled by 8). The following magic 2469 * is equivalent to the smoothing algorithm in rfc793 with 2470 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2471 * point). Adjust rtt to origin 0. 2472 */ 2473 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2474 if ((tp->t_srtt += delta) <= 0) 2475 tp->t_srtt = 1 << 2; 2476 /* 2477 * We accumulate a smoothed rtt variance (actually, a 2478 * smoothed mean difference), then set the retransmit 2479 * timer to smoothed rtt + 4 times the smoothed variance. 2480 * rttvar is stored as fixed point with 2 bits after the 2481 * binary point (scaled by 4). The following is 2482 * equivalent to rfc793 smoothing with an alpha of .75 2483 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2484 * rfc793's wired-in beta. 2485 */ 2486 if (delta < 0) 2487 delta = -delta; 2488 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2489 if ((tp->t_rttvar += delta) <= 0) 2490 tp->t_rttvar = 1 << 2; 2491 } else { 2492 /* 2493 * No rtt measurement yet - use the unsmoothed rtt. 2494 * Set the variance to half the rtt (so our first 2495 * retransmit happens at 3*rtt). 2496 */ 2497 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 2498 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 2499 } 2500 tp->t_rtttime = 0; 2501 tp->t_rxtshift = 0; 2502 2503 /* 2504 * the retransmit should happen at rtt + 4 * rttvar. 2505 * Because of the way we do the smoothing, srtt and rttvar 2506 * will each average +1/2 tick of bias. When we compute 2507 * the retransmit timer, we want 1/2 tick of rounding and 2508 * 1 extra tick because of +-1/2 tick uncertainty in the 2509 * firing of the timer. The bias will give us exactly the 2510 * 1.5 tick we need. But, because the bias is 2511 * statistical, we have to test that we don't drop below 2512 * the minimum feasible timer (which is 2 ticks). 2513 */ 2514 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2515 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2516 2517 /* 2518 * We received an ack for a packet that wasn't retransmitted; 2519 * it is probably safe to discard any error indications we've 2520 * received recently. This isn't quite right, but close enough 2521 * for now (a route might have failed after we sent a segment, 2522 * and the return path might not be symmetrical). 2523 */ 2524 tp->t_softerror = 0; 2525 } 2526 2527 /* 2528 * Checks for partial ack. If partial ack arrives, force the retransmission 2529 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2530 * 1. By setting snd_nxt to th_ack, this forces retransmission timer to 2531 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2532 */ 2533 int 2534 tcp_newreno(tp, th) 2535 struct tcpcb *tp; 2536 struct tcphdr *th; 2537 { 2538 tcp_seq onxt = tp->snd_nxt; 2539 u_long ocwnd = tp->snd_cwnd; 2540 2541 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2542 /* 2543 * snd_una has not yet been updated and the socket's send 2544 * buffer has not yet drained off the ACK'd data, so we 2545 * have to leave snd_una as it was to get the correct data 2546 * offset in tcp_output(). 2547 */ 2548 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2549 tp->t_rtttime = 0; 2550 tp->snd_nxt = th->th_ack; 2551 /* 2552 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 2553 * is not yet updated when we're called. 2554 */ 2555 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 2556 (void) tcp_output(tp); 2557 tp->snd_cwnd = ocwnd; 2558 if (SEQ_GT(onxt, tp->snd_nxt)) 2559 tp->snd_nxt = onxt; 2560 /* 2561 * Partial window deflation. Relies on fact that tp->snd_una 2562 * not updated yet. 2563 */ 2564 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 2565 return 1; 2566 } 2567 return 0; 2568 } 2569 2570 2571 /* 2572 * TCP compressed state engine. Currently used to hold compressed 2573 * state for SYN_RECEIVED. 2574 */ 2575 2576 u_long syn_cache_count; 2577 u_int32_t syn_hash1, syn_hash2; 2578 2579 #define SYN_HASH(sa, sp, dp) \ 2580 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 2581 ((u_int32_t)(sp)))^syn_hash2))) 2582 #ifndef INET6 2583 #define SYN_HASHALL(hash, src, dst) \ 2584 do { \ 2585 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2586 ((struct sockaddr_in *)(src))->sin_port, \ 2587 ((struct sockaddr_in *)(dst))->sin_port); \ 2588 } while (0) 2589 #else 2590 #define SYN_HASH6(sa, sp, dp) \ 2591 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 2592 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 2593 & 0x7fffffff) 2594 2595 #define SYN_HASHALL(hash, src, dst) \ 2596 do { \ 2597 switch ((src)->sa_family) { \ 2598 case AF_INET: \ 2599 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2600 ((struct sockaddr_in *)(src))->sin_port, \ 2601 ((struct sockaddr_in *)(dst))->sin_port); \ 2602 break; \ 2603 case AF_INET6: \ 2604 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 2605 ((struct sockaddr_in6 *)(src))->sin6_port, \ 2606 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 2607 break; \ 2608 default: \ 2609 hash = 0; \ 2610 } \ 2611 } while (/*CONSTCOND*/0) 2612 #endif /* INET6 */ 2613 2614 #define SYN_CACHE_RM(sc) \ 2615 do { \ 2616 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 2617 (sc), sc_bucketq); \ 2618 (sc)->sc_tp = NULL; \ 2619 LIST_REMOVE((sc), sc_tpq); \ 2620 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 2621 callout_stop(&(sc)->sc_timer); \ 2622 syn_cache_count--; \ 2623 } while (/*CONSTCOND*/0) 2624 2625 #define SYN_CACHE_PUT(sc) \ 2626 do { \ 2627 if ((sc)->sc_ipopts) \ 2628 (void) m_free((sc)->sc_ipopts); \ 2629 if ((sc)->sc_route4.ro_rt != NULL) \ 2630 RTFREE((sc)->sc_route4.ro_rt); \ 2631 pool_put(&syn_cache_pool, (sc)); \ 2632 } while (/*CONSTCOND*/0) 2633 2634 struct pool syn_cache_pool; 2635 2636 /* 2637 * We don't estimate RTT with SYNs, so each packet starts with the default 2638 * RTT and each timer step has a fixed timeout value. 2639 */ 2640 #define SYN_CACHE_TIMER_ARM(sc) \ 2641 do { \ 2642 TCPT_RANGESET((sc)->sc_rxtcur, \ 2643 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 2644 TCPTV_REXMTMAX); \ 2645 callout_reset(&(sc)->sc_timer, \ 2646 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 2647 } while (/*CONSTCOND*/0) 2648 2649 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 2650 2651 void 2652 syn_cache_init() 2653 { 2654 int i; 2655 2656 /* Initialize the hash buckets. */ 2657 for (i = 0; i < tcp_syn_cache_size; i++) 2658 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 2659 2660 /* Initialize the syn cache pool. */ 2661 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 2662 "synpl", 0, NULL, NULL, M_PCB); 2663 } 2664 2665 void 2666 syn_cache_insert(sc, tp) 2667 struct syn_cache *sc; 2668 struct tcpcb *tp; 2669 { 2670 struct syn_cache_head *scp; 2671 struct syn_cache *sc2; 2672 int s; 2673 2674 /* 2675 * If there are no entries in the hash table, reinitialize 2676 * the hash secrets. 2677 */ 2678 if (syn_cache_count == 0) { 2679 struct timeval tv; 2680 microtime(&tv); 2681 syn_hash1 = random() ^ (u_long)≻ 2682 syn_hash2 = random() ^ tv.tv_usec; 2683 } 2684 2685 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 2686 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 2687 scp = &tcp_syn_cache[sc->sc_bucketidx]; 2688 2689 /* 2690 * Make sure that we don't overflow the per-bucket 2691 * limit or the total cache size limit. 2692 */ 2693 s = splsoftnet(); 2694 if (scp->sch_length >= tcp_syn_bucket_limit) { 2695 tcpstat.tcps_sc_bucketoverflow++; 2696 /* 2697 * The bucket is full. Toss the oldest element in the 2698 * bucket. This will be the first entry in the bucket. 2699 */ 2700 sc2 = TAILQ_FIRST(&scp->sch_bucket); 2701 #ifdef DIAGNOSTIC 2702 /* 2703 * This should never happen; we should always find an 2704 * entry in our bucket. 2705 */ 2706 if (sc2 == NULL) 2707 panic("syn_cache_insert: bucketoverflow: impossible"); 2708 #endif 2709 SYN_CACHE_RM(sc2); 2710 SYN_CACHE_PUT(sc2); 2711 } else if (syn_cache_count >= tcp_syn_cache_limit) { 2712 struct syn_cache_head *scp2, *sce; 2713 2714 tcpstat.tcps_sc_overflowed++; 2715 /* 2716 * The cache is full. Toss the oldest entry in the 2717 * first non-empty bucket we can find. 2718 * 2719 * XXX We would really like to toss the oldest 2720 * entry in the cache, but we hope that this 2721 * condition doesn't happen very often. 2722 */ 2723 scp2 = scp; 2724 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 2725 sce = &tcp_syn_cache[tcp_syn_cache_size]; 2726 for (++scp2; scp2 != scp; scp2++) { 2727 if (scp2 >= sce) 2728 scp2 = &tcp_syn_cache[0]; 2729 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 2730 break; 2731 } 2732 #ifdef DIAGNOSTIC 2733 /* 2734 * This should never happen; we should always find a 2735 * non-empty bucket. 2736 */ 2737 if (scp2 == scp) 2738 panic("syn_cache_insert: cacheoverflow: " 2739 "impossible"); 2740 #endif 2741 } 2742 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 2743 SYN_CACHE_RM(sc2); 2744 SYN_CACHE_PUT(sc2); 2745 } 2746 2747 /* 2748 * Initialize the entry's timer. 2749 */ 2750 sc->sc_rxttot = 0; 2751 sc->sc_rxtshift = 0; 2752 SYN_CACHE_TIMER_ARM(sc); 2753 2754 /* Link it from tcpcb entry */ 2755 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 2756 2757 /* Put it into the bucket. */ 2758 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 2759 scp->sch_length++; 2760 syn_cache_count++; 2761 2762 tcpstat.tcps_sc_added++; 2763 splx(s); 2764 } 2765 2766 /* 2767 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 2768 * If we have retransmitted an entry the maximum number of times, expire 2769 * that entry. 2770 */ 2771 void 2772 syn_cache_timer(void *arg) 2773 { 2774 struct syn_cache *sc = arg; 2775 int s; 2776 2777 s = splsoftnet(); 2778 2779 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 2780 /* Drop it -- too many retransmissions. */ 2781 goto dropit; 2782 } 2783 2784 /* 2785 * Compute the total amount of time this entry has 2786 * been on a queue. If this entry has been on longer 2787 * than the keep alive timer would allow, expire it. 2788 */ 2789 sc->sc_rxttot += sc->sc_rxtcur; 2790 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) 2791 goto dropit; 2792 2793 tcpstat.tcps_sc_retransmitted++; 2794 (void) syn_cache_respond(sc, NULL); 2795 2796 /* Advance the timer back-off. */ 2797 sc->sc_rxtshift++; 2798 SYN_CACHE_TIMER_ARM(sc); 2799 2800 splx(s); 2801 return; 2802 2803 dropit: 2804 tcpstat.tcps_sc_timed_out++; 2805 SYN_CACHE_RM(sc); 2806 SYN_CACHE_PUT(sc); 2807 splx(s); 2808 } 2809 2810 /* 2811 * Remove syn cache created by the specified tcb entry, 2812 * because this does not make sense to keep them 2813 * (if there's no tcb entry, syn cache entry will never be used) 2814 */ 2815 void 2816 syn_cache_cleanup(tp) 2817 struct tcpcb *tp; 2818 { 2819 struct syn_cache *sc, *nsc; 2820 int s; 2821 2822 s = splsoftnet(); 2823 2824 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 2825 nsc = LIST_NEXT(sc, sc_tpq); 2826 2827 #ifdef DIAGNOSTIC 2828 if (sc->sc_tp != tp) 2829 panic("invalid sc_tp in syn_cache_cleanup"); 2830 #endif 2831 SYN_CACHE_RM(sc); 2832 SYN_CACHE_PUT(sc); 2833 } 2834 /* just for safety */ 2835 LIST_INIT(&tp->t_sc); 2836 2837 splx(s); 2838 } 2839 2840 /* 2841 * Find an entry in the syn cache. 2842 */ 2843 struct syn_cache * 2844 syn_cache_lookup(src, dst, headp) 2845 struct sockaddr *src; 2846 struct sockaddr *dst; 2847 struct syn_cache_head **headp; 2848 { 2849 struct syn_cache *sc; 2850 struct syn_cache_head *scp; 2851 u_int32_t hash; 2852 int s; 2853 2854 SYN_HASHALL(hash, src, dst); 2855 2856 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 2857 *headp = scp; 2858 s = splsoftnet(); 2859 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 2860 sc = TAILQ_NEXT(sc, sc_bucketq)) { 2861 if (sc->sc_hash != hash) 2862 continue; 2863 if (!bcmp(&sc->sc_src, src, src->sa_len) && 2864 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 2865 splx(s); 2866 return (sc); 2867 } 2868 } 2869 splx(s); 2870 return (NULL); 2871 } 2872 2873 /* 2874 * This function gets called when we receive an ACK for a 2875 * socket in the LISTEN state. We look up the connection 2876 * in the syn cache, and if its there, we pull it out of 2877 * the cache and turn it into a full-blown connection in 2878 * the SYN-RECEIVED state. 2879 * 2880 * The return values may not be immediately obvious, and their effects 2881 * can be subtle, so here they are: 2882 * 2883 * NULL SYN was not found in cache; caller should drop the 2884 * packet and send an RST. 2885 * 2886 * -1 We were unable to create the new connection, and are 2887 * aborting it. An ACK,RST is being sent to the peer 2888 * (unless we got screwey sequence numbners; see below), 2889 * because the 3-way handshake has been completed. Caller 2890 * should not free the mbuf, since we may be using it. If 2891 * we are not, we will free it. 2892 * 2893 * Otherwise, the return value is a pointer to the new socket 2894 * associated with the connection. 2895 */ 2896 struct socket * 2897 syn_cache_get(src, dst, th, hlen, tlen, so, m) 2898 struct sockaddr *src; 2899 struct sockaddr *dst; 2900 struct tcphdr *th; 2901 unsigned int hlen, tlen; 2902 struct socket *so; 2903 struct mbuf *m; 2904 { 2905 struct syn_cache *sc; 2906 struct syn_cache_head *scp; 2907 struct inpcb *inp = NULL; 2908 #ifdef INET6 2909 struct in6pcb *in6p = NULL; 2910 #endif 2911 struct tcpcb *tp = 0; 2912 struct mbuf *am; 2913 int s; 2914 struct socket *oso; 2915 2916 s = splsoftnet(); 2917 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 2918 splx(s); 2919 return (NULL); 2920 } 2921 2922 /* 2923 * Verify the sequence and ack numbers. Try getting the correct 2924 * response again. 2925 */ 2926 if ((th->th_ack != sc->sc_iss + 1) || 2927 SEQ_LEQ(th->th_seq, sc->sc_irs) || 2928 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 2929 (void) syn_cache_respond(sc, m); 2930 splx(s); 2931 return ((struct socket *)(-1)); 2932 } 2933 2934 /* Remove this cache entry */ 2935 SYN_CACHE_RM(sc); 2936 splx(s); 2937 2938 /* 2939 * Ok, create the full blown connection, and set things up 2940 * as they would have been set up if we had created the 2941 * connection when the SYN arrived. If we can't create 2942 * the connection, abort it. 2943 */ 2944 /* 2945 * inp still has the OLD in_pcb stuff, set the 2946 * v6-related flags on the new guy, too. This is 2947 * done particularly for the case where an AF_INET6 2948 * socket is bound only to a port, and a v4 connection 2949 * comes in on that port. 2950 * we also copy the flowinfo from the original pcb 2951 * to the new one. 2952 */ 2953 { 2954 struct inpcb *parentinpcb; 2955 2956 parentinpcb = (struct inpcb *)so->so_pcb; 2957 2958 oso = so; 2959 so = sonewconn(so, SS_ISCONNECTED); 2960 if (so == NULL) 2961 goto resetandabort; 2962 2963 switch (so->so_proto->pr_domain->dom_family) { 2964 #ifdef INET 2965 case AF_INET: 2966 inp = sotoinpcb(so); 2967 break; 2968 #endif 2969 #ifdef INET6 2970 case AF_INET6: 2971 in6p = sotoin6pcb(so); 2972 break; 2973 #endif 2974 } 2975 } 2976 switch (src->sa_family) { 2977 #ifdef INET 2978 case AF_INET: 2979 if (inp) { 2980 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 2981 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 2982 inp->inp_options = ip_srcroute(); 2983 in_pcbstate(inp, INP_BOUND); 2984 if (inp->inp_options == NULL) { 2985 inp->inp_options = sc->sc_ipopts; 2986 sc->sc_ipopts = NULL; 2987 } 2988 } 2989 #ifdef INET6 2990 else if (in6p) { 2991 /* IPv4 packet to AF_INET6 socket */ 2992 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 2993 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 2994 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 2995 &in6p->in6p_laddr.s6_addr32[3], 2996 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 2997 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 2998 in6totcpcb(in6p)->t_family = AF_INET; 2999 } 3000 #endif 3001 break; 3002 #endif 3003 #ifdef INET6 3004 case AF_INET6: 3005 if (in6p) { 3006 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3007 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3008 #if 0 3009 in6p->in6p_flowinfo = ip6->ip6_flow & IPV6_FLOWINFO_MASK; 3010 /*inp->inp_options = ip6_srcroute();*/ /* soon. */ 3011 #endif 3012 } 3013 break; 3014 #endif 3015 } 3016 #ifdef INET6 3017 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3018 struct in6pcb *oin6p = sotoin6pcb(oso); 3019 /* inherit socket options from the listening socket */ 3020 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3021 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3022 m_freem(in6p->in6p_options); 3023 in6p->in6p_options = 0; 3024 } 3025 ip6_savecontrol(in6p, &in6p->in6p_options, 3026 mtod(m, struct ip6_hdr *), m); 3027 } 3028 #endif 3029 3030 #ifdef IPSEC 3031 /* 3032 * we make a copy of policy, instead of sharing the policy, 3033 * for better behavior in terms of SA lookup and dead SA removal. 3034 */ 3035 if (inp) { 3036 /* copy old policy into new socket's */ 3037 if (ipsec_copy_policy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3038 printf("tcp_input: could not copy policy\n"); 3039 } 3040 #ifdef INET6 3041 else if (in6p) { 3042 /* copy old policy into new socket's */ 3043 if (ipsec_copy_policy(sotoin6pcb(oso)->in6p_sp, in6p->in6p_sp)) 3044 printf("tcp_input: could not copy policy\n"); 3045 } 3046 #endif 3047 #endif 3048 3049 /* 3050 * Give the new socket our cached route reference. 3051 */ 3052 if (inp) 3053 inp->inp_route = sc->sc_route4; /* struct assignment */ 3054 #ifdef INET6 3055 else 3056 in6p->in6p_route = sc->sc_route6; 3057 #endif 3058 sc->sc_route4.ro_rt = NULL; 3059 3060 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3061 if (am == NULL) 3062 goto resetandabort; 3063 am->m_len = src->sa_len; 3064 bcopy(src, mtod(am, caddr_t), src->sa_len); 3065 if (inp) { 3066 if (in_pcbconnect(inp, am)) { 3067 (void) m_free(am); 3068 goto resetandabort; 3069 } 3070 } 3071 #ifdef INET6 3072 else if (in6p) { 3073 if (src->sa_family == AF_INET) { 3074 /* IPv4 packet to AF_INET6 socket */ 3075 struct sockaddr_in6 *sin6; 3076 sin6 = mtod(am, struct sockaddr_in6 *); 3077 am->m_len = sizeof(*sin6); 3078 bzero(sin6, sizeof(*sin6)); 3079 sin6->sin6_family = AF_INET6; 3080 sin6->sin6_len = sizeof(*sin6); 3081 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3082 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3083 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3084 &sin6->sin6_addr.s6_addr32[3], 3085 sizeof(sin6->sin6_addr.s6_addr32[3])); 3086 } 3087 if (in6_pcbconnect(in6p, am)) { 3088 (void) m_free(am); 3089 goto resetandabort; 3090 } 3091 } 3092 #endif 3093 else { 3094 (void) m_free(am); 3095 goto resetandabort; 3096 } 3097 (void) m_free(am); 3098 3099 if (inp) 3100 tp = intotcpcb(inp); 3101 #ifdef INET6 3102 else if (in6p) 3103 tp = in6totcpcb(in6p); 3104 #endif 3105 else 3106 tp = NULL; 3107 if (sc->sc_request_r_scale != 15) { 3108 tp->requested_s_scale = sc->sc_requested_s_scale; 3109 tp->request_r_scale = sc->sc_request_r_scale; 3110 tp->snd_scale = sc->sc_requested_s_scale; 3111 tp->rcv_scale = sc->sc_request_r_scale; 3112 tp->t_flags |= TF_RCVD_SCALE; 3113 } 3114 if (sc->sc_flags & SCF_TIMESTAMP) 3115 tp->t_flags |= TF_RCVD_TSTMP; 3116 tp->ts_timebase = sc->sc_timebase; 3117 3118 tp->t_template = tcp_template(tp); 3119 if (tp->t_template == 0) { 3120 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3121 so = NULL; 3122 m_freem(m); 3123 goto abort; 3124 } 3125 3126 tp->iss = sc->sc_iss; 3127 tp->irs = sc->sc_irs; 3128 tcp_sendseqinit(tp); 3129 tcp_rcvseqinit(tp); 3130 tp->t_state = TCPS_SYN_RECEIVED; 3131 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3132 tcpstat.tcps_accepts++; 3133 3134 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3135 tp->t_ourmss = sc->sc_ourmaxseg; 3136 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3137 3138 /* 3139 * Initialize the initial congestion window. If we 3140 * had to retransmit the SYN,ACK, we must initialize cwnd 3141 * to 1 segment (i.e. the Loss Window). 3142 */ 3143 if (sc->sc_rxtshift) 3144 tp->snd_cwnd = tp->t_peermss; 3145 else 3146 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, tp->t_peermss); 3147 3148 tcp_rmx_rtt(tp); 3149 tp->snd_wl1 = sc->sc_irs; 3150 tp->rcv_up = sc->sc_irs + 1; 3151 3152 /* 3153 * This is what whould have happened in tcp_ouput() when 3154 * the SYN,ACK was sent. 3155 */ 3156 tp->snd_up = tp->snd_una; 3157 tp->snd_max = tp->snd_nxt = tp->iss+1; 3158 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3159 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3160 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3161 tp->last_ack_sent = tp->rcv_nxt; 3162 3163 tcpstat.tcps_sc_completed++; 3164 SYN_CACHE_PUT(sc); 3165 return (so); 3166 3167 resetandabort: 3168 (void) tcp_respond(NULL, m, m, th, 3169 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 3170 abort: 3171 if (so != NULL) 3172 (void) soabort(so); 3173 SYN_CACHE_PUT(sc); 3174 tcpstat.tcps_sc_aborted++; 3175 return ((struct socket *)(-1)); 3176 } 3177 3178 /* 3179 * This function is called when we get a RST for a 3180 * non-existent connection, so that we can see if the 3181 * connection is in the syn cache. If it is, zap it. 3182 */ 3183 3184 void 3185 syn_cache_reset(src, dst, th) 3186 struct sockaddr *src; 3187 struct sockaddr *dst; 3188 struct tcphdr *th; 3189 { 3190 struct syn_cache *sc; 3191 struct syn_cache_head *scp; 3192 int s = splsoftnet(); 3193 3194 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3195 splx(s); 3196 return; 3197 } 3198 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3199 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3200 splx(s); 3201 return; 3202 } 3203 SYN_CACHE_RM(sc); 3204 splx(s); 3205 tcpstat.tcps_sc_reset++; 3206 SYN_CACHE_PUT(sc); 3207 } 3208 3209 void 3210 syn_cache_unreach(src, dst, th) 3211 struct sockaddr *src; 3212 struct sockaddr *dst; 3213 struct tcphdr *th; 3214 { 3215 struct syn_cache *sc; 3216 struct syn_cache_head *scp; 3217 int s; 3218 3219 s = splsoftnet(); 3220 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3221 splx(s); 3222 return; 3223 } 3224 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3225 if (ntohl (th->th_seq) != sc->sc_iss) { 3226 splx(s); 3227 return; 3228 } 3229 3230 /* 3231 * If we've rertransmitted 3 times and this is our second error, 3232 * we remove the entry. Otherwise, we allow it to continue on. 3233 * This prevents us from incorrectly nuking an entry during a 3234 * spurious network outage. 3235 * 3236 * See tcp_notify(). 3237 */ 3238 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3239 sc->sc_flags |= SCF_UNREACH; 3240 splx(s); 3241 return; 3242 } 3243 3244 SYN_CACHE_RM(sc); 3245 splx(s); 3246 tcpstat.tcps_sc_unreach++; 3247 SYN_CACHE_PUT(sc); 3248 } 3249 3250 /* 3251 * Given a LISTEN socket and an inbound SYN request, add 3252 * this to the syn cache, and send back a segment: 3253 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3254 * to the source. 3255 * 3256 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3257 * Doing so would require that we hold onto the data and deliver it 3258 * to the application. However, if we are the target of a SYN-flood 3259 * DoS attack, an attacker could send data which would eventually 3260 * consume all available buffer space if it were ACKed. By not ACKing 3261 * the data, we avoid this DoS scenario. 3262 */ 3263 3264 int 3265 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi) 3266 struct sockaddr *src; 3267 struct sockaddr *dst; 3268 struct tcphdr *th; 3269 unsigned int hlen; 3270 struct socket *so; 3271 struct mbuf *m; 3272 u_char *optp; 3273 int optlen; 3274 struct tcp_opt_info *oi; 3275 { 3276 struct tcpcb tb, *tp; 3277 long win; 3278 struct syn_cache *sc; 3279 struct syn_cache_head *scp; 3280 struct mbuf *ipopts; 3281 3282 tp = sototcpcb(so); 3283 3284 /* 3285 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3286 * 3287 * Note this check is performed in tcp_input() very early on. 3288 */ 3289 3290 /* 3291 * Initialize some local state. 3292 */ 3293 win = sbspace(&so->so_rcv); 3294 if (win > TCP_MAXWIN) 3295 win = TCP_MAXWIN; 3296 3297 switch (src->sa_family) { 3298 #ifdef INET 3299 case AF_INET: 3300 /* 3301 * Remember the IP options, if any. 3302 */ 3303 ipopts = ip_srcroute(); 3304 break; 3305 #endif 3306 default: 3307 ipopts = NULL; 3308 } 3309 3310 if (optp) { 3311 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3312 tcp_dooptions(&tb, optp, optlen, th, oi); 3313 } else 3314 tb.t_flags = 0; 3315 3316 /* 3317 * See if we already have an entry for this connection. 3318 * If we do, resend the SYN,ACK. We do not count this 3319 * as a retransmission (XXX though maybe we should). 3320 */ 3321 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3322 tcpstat.tcps_sc_dupesyn++; 3323 if (ipopts) { 3324 /* 3325 * If we were remembering a previous source route, 3326 * forget it and use the new one we've been given. 3327 */ 3328 if (sc->sc_ipopts) 3329 (void) m_free(sc->sc_ipopts); 3330 sc->sc_ipopts = ipopts; 3331 } 3332 sc->sc_timestamp = tb.ts_recent; 3333 if (syn_cache_respond(sc, m) == 0) { 3334 tcpstat.tcps_sndacks++; 3335 tcpstat.tcps_sndtotal++; 3336 } 3337 return (1); 3338 } 3339 3340 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3341 if (sc == NULL) { 3342 if (ipopts) 3343 (void) m_free(ipopts); 3344 return (0); 3345 } 3346 3347 /* 3348 * Fill in the cache, and put the necessary IP and TCP 3349 * options into the reply. 3350 */ 3351 callout_init(&sc->sc_timer); 3352 bzero(sc, sizeof(struct syn_cache)); 3353 bcopy(src, &sc->sc_src, src->sa_len); 3354 bcopy(dst, &sc->sc_dst, dst->sa_len); 3355 sc->sc_flags = 0; 3356 sc->sc_ipopts = ipopts; 3357 sc->sc_irs = th->th_seq; 3358 switch (src->sa_family) { 3359 #ifdef INET 3360 case AF_INET: 3361 { 3362 struct sockaddr_in *srcin = (void *) src; 3363 struct sockaddr_in *dstin = (void *) dst; 3364 3365 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 3366 &srcin->sin_addr, dstin->sin_port, 3367 srcin->sin_port, sizeof(dstin->sin_addr), 0); 3368 break; 3369 } 3370 #endif /* INET */ 3371 #ifdef INET6 3372 case AF_INET6: 3373 { 3374 struct sockaddr_in6 *srcin6 = (void *) src; 3375 struct sockaddr_in6 *dstin6 = (void *) dst; 3376 3377 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 3378 &srcin6->sin6_addr, dstin6->sin6_port, 3379 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 3380 break; 3381 } 3382 #endif /* INET6 */ 3383 } 3384 sc->sc_peermaxseg = oi->maxseg; 3385 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3386 m->m_pkthdr.rcvif : NULL, 3387 sc->sc_src.sa.sa_family); 3388 sc->sc_win = win; 3389 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 3390 sc->sc_timestamp = tb.ts_recent; 3391 if (tcp_do_rfc1323 && (tb.t_flags & TF_RCVD_TSTMP)) 3392 sc->sc_flags |= SCF_TIMESTAMP; 3393 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3394 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3395 sc->sc_requested_s_scale = tb.requested_s_scale; 3396 sc->sc_request_r_scale = 0; 3397 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3398 TCP_MAXWIN << sc->sc_request_r_scale < 3399 so->so_rcv.sb_hiwat) 3400 sc->sc_request_r_scale++; 3401 } else { 3402 sc->sc_requested_s_scale = 15; 3403 sc->sc_request_r_scale = 15; 3404 } 3405 sc->sc_tp = tp; 3406 if (syn_cache_respond(sc, m) == 0) { 3407 syn_cache_insert(sc, tp); 3408 tcpstat.tcps_sndacks++; 3409 tcpstat.tcps_sndtotal++; 3410 } else { 3411 SYN_CACHE_PUT(sc); 3412 tcpstat.tcps_sc_dropped++; 3413 } 3414 return (1); 3415 } 3416 3417 int 3418 syn_cache_respond(sc, m) 3419 struct syn_cache *sc; 3420 struct mbuf *m; 3421 { 3422 struct route *ro; 3423 u_int8_t *optp; 3424 int optlen, error; 3425 u_int16_t tlen; 3426 struct ip *ip = NULL; 3427 #ifdef INET6 3428 struct ip6_hdr *ip6 = NULL; 3429 #endif 3430 struct tcphdr *th; 3431 u_int hlen; 3432 3433 switch (sc->sc_src.sa.sa_family) { 3434 case AF_INET: 3435 hlen = sizeof(struct ip); 3436 ro = &sc->sc_route4; 3437 break; 3438 #ifdef INET6 3439 case AF_INET6: 3440 hlen = sizeof(struct ip6_hdr); 3441 ro = (struct route *)&sc->sc_route6; 3442 break; 3443 #endif 3444 default: 3445 if (m) 3446 m_freem(m); 3447 return EAFNOSUPPORT; 3448 } 3449 3450 /* Compute the size of the TCP options. */ 3451 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 3452 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 3453 3454 tlen = hlen + sizeof(struct tcphdr) + optlen; 3455 3456 /* 3457 * Create the IP+TCP header from scratch. 3458 */ 3459 if (m) 3460 m_freem(m); 3461 #ifdef DIAGNOSTIC 3462 if (max_linkhdr + tlen > MCLBYTES) 3463 return (ENOBUFS); 3464 #endif 3465 MGETHDR(m, M_DONTWAIT, MT_DATA); 3466 if (m && tlen > MHLEN) { 3467 MCLGET(m, M_DONTWAIT); 3468 if ((m->m_flags & M_EXT) == 0) { 3469 m_freem(m); 3470 m = NULL; 3471 } 3472 } 3473 if (m == NULL) 3474 return (ENOBUFS); 3475 3476 /* Fixup the mbuf. */ 3477 m->m_data += max_linkhdr; 3478 m->m_len = m->m_pkthdr.len = tlen; 3479 #ifdef IPSEC 3480 if (sc->sc_tp) { 3481 struct tcpcb *tp; 3482 struct socket *so; 3483 3484 tp = sc->sc_tp; 3485 if (tp->t_inpcb) 3486 so = tp->t_inpcb->inp_socket; 3487 #ifdef INET6 3488 else if (tp->t_in6pcb) 3489 so = tp->t_in6pcb->in6p_socket; 3490 #endif 3491 else 3492 so = NULL; 3493 /* use IPsec policy on listening socket, on SYN ACK */ 3494 if (ipsec_setsocket(m, so) != 0) { 3495 m_freem(m); 3496 return ENOBUFS; 3497 } 3498 } 3499 #endif 3500 m->m_pkthdr.rcvif = NULL; 3501 memset(mtod(m, u_char *), 0, tlen); 3502 3503 switch (sc->sc_src.sa.sa_family) { 3504 case AF_INET: 3505 ip = mtod(m, struct ip *); 3506 ip->ip_dst = sc->sc_src.sin.sin_addr; 3507 ip->ip_src = sc->sc_dst.sin.sin_addr; 3508 ip->ip_p = IPPROTO_TCP; 3509 th = (struct tcphdr *)(ip + 1); 3510 th->th_dport = sc->sc_src.sin.sin_port; 3511 th->th_sport = sc->sc_dst.sin.sin_port; 3512 break; 3513 #ifdef INET6 3514 case AF_INET6: 3515 ip6 = mtod(m, struct ip6_hdr *); 3516 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 3517 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 3518 ip6->ip6_nxt = IPPROTO_TCP; 3519 /* ip6_plen will be updated in ip6_output() */ 3520 th = (struct tcphdr *)(ip6 + 1); 3521 th->th_dport = sc->sc_src.sin6.sin6_port; 3522 th->th_sport = sc->sc_dst.sin6.sin6_port; 3523 break; 3524 #endif 3525 default: 3526 th = NULL; 3527 } 3528 3529 th->th_seq = htonl(sc->sc_iss); 3530 th->th_ack = htonl(sc->sc_irs + 1); 3531 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 3532 th->th_flags = TH_SYN|TH_ACK; 3533 th->th_win = htons(sc->sc_win); 3534 /* th_sum already 0 */ 3535 /* th_urp already 0 */ 3536 3537 /* Tack on the TCP options. */ 3538 optp = (u_int8_t *)(th + 1); 3539 *optp++ = TCPOPT_MAXSEG; 3540 *optp++ = 4; 3541 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 3542 *optp++ = sc->sc_ourmaxseg & 0xff; 3543 3544 if (sc->sc_request_r_scale != 15) { 3545 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 3546 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 3547 sc->sc_request_r_scale); 3548 optp += 4; 3549 } 3550 3551 if (sc->sc_flags & SCF_TIMESTAMP) { 3552 u_int32_t *lp = (u_int32_t *)(optp); 3553 /* Form timestamp option as shown in appendix A of RFC 1323. */ 3554 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 3555 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 3556 *lp = htonl(sc->sc_timestamp); 3557 optp += TCPOLEN_TSTAMP_APPA; 3558 } 3559 3560 /* Compute the packet's checksum. */ 3561 switch (sc->sc_src.sa.sa_family) { 3562 case AF_INET: 3563 ip->ip_len = htons(tlen - hlen); 3564 th->th_sum = 0; 3565 th->th_sum = in_cksum(m, tlen); 3566 break; 3567 #ifdef INET6 3568 case AF_INET6: 3569 ip6->ip6_plen = htons(tlen - hlen); 3570 th->th_sum = 0; 3571 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 3572 break; 3573 #endif 3574 } 3575 3576 /* 3577 * Fill in some straggling IP bits. Note the stack expects 3578 * ip_len to be in host order, for convenience. 3579 */ 3580 switch (sc->sc_src.sa.sa_family) { 3581 #ifdef INET 3582 case AF_INET: 3583 ip->ip_len = tlen; 3584 ip->ip_ttl = ip_defttl; 3585 /* XXX tos? */ 3586 break; 3587 #endif 3588 #ifdef INET6 3589 case AF_INET6: 3590 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 3591 ip6->ip6_vfc |= IPV6_VERSION; 3592 ip6->ip6_plen = htons(tlen - hlen); 3593 /* ip6_hlim will be initialized afterwards */ 3594 /* XXX flowlabel? */ 3595 break; 3596 #endif 3597 } 3598 3599 switch (sc->sc_src.sa.sa_family) { 3600 #ifdef INET 3601 case AF_INET: 3602 error = ip_output(m, sc->sc_ipopts, ro, 3603 (ip_mtudisc ? IP_MTUDISC : 0), 3604 NULL); 3605 break; 3606 #endif 3607 #ifdef INET6 3608 case AF_INET6: 3609 ip6->ip6_hlim = in6_selecthlim(NULL, 3610 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 3611 3612 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 3613 0, NULL, NULL); 3614 break; 3615 #endif 3616 default: 3617 error = EAFNOSUPPORT; 3618 break; 3619 } 3620 return (error); 3621 } 3622