1 /* $NetBSD: tcp_input.c,v 1.434 2022/09/20 07:19:14 ozaki-r Exp $ */ 2 3 /* 4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the project nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 /* 33 * @(#)COPYRIGHT 1.1 (NRL) 17 January 1995 34 * 35 * NRL grants permission for redistribution and use in source and binary 36 * forms, with or without modification, of the software and documentation 37 * created at NRL provided that the following conditions are met: 38 * 39 * 1. Redistributions of source code must retain the above copyright 40 * notice, this list of conditions and the following disclaimer. 41 * 2. Redistributions in binary form must reproduce the above copyright 42 * notice, this list of conditions and the following disclaimer in the 43 * documentation and/or other materials provided with the distribution. 44 * 3. All advertising materials mentioning features or use of this software 45 * must display the following acknowledgements: 46 * This product includes software developed by the University of 47 * California, Berkeley and its contributors. 48 * This product includes software developed at the Information 49 * Technology Division, US Naval Research Laboratory. 50 * 4. Neither the name of the NRL nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS 55 * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 56 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A 57 * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NRL OR 58 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 59 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 60 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 61 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 62 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 63 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 64 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 65 * 66 * The views and conclusions contained in the software and documentation 67 * are those of the authors and should not be interpreted as representing 68 * official policies, either expressed or implied, of the US Naval 69 * Research Laboratory (NRL). 70 */ 71 72 /*- 73 * Copyright (c) 1997, 1998, 1999, 2001, 2005, 2006, 74 * 2011 The NetBSD Foundation, Inc. 75 * All rights reserved. 76 * 77 * This code is derived from software contributed to The NetBSD Foundation 78 * by Coyote Point Systems, Inc. 79 * This code is derived from software contributed to The NetBSD Foundation 80 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 81 * Facility, NASA Ames Research Center. 82 * This code is derived from software contributed to The NetBSD Foundation 83 * by Charles M. Hannum. 84 * This code is derived from software contributed to The NetBSD Foundation 85 * by Rui Paulo. 86 * 87 * Redistribution and use in source and binary forms, with or without 88 * modification, are permitted provided that the following conditions 89 * are met: 90 * 1. Redistributions of source code must retain the above copyright 91 * notice, this list of conditions and the following disclaimer. 92 * 2. Redistributions in binary form must reproduce the above copyright 93 * notice, this list of conditions and the following disclaimer in the 94 * documentation and/or other materials provided with the distribution. 95 * 96 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 97 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 98 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 99 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 100 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 101 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 102 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 103 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 104 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 105 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 106 * POSSIBILITY OF SUCH DAMAGE. 107 */ 108 109 /* 110 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 111 * The Regents of the University of California. All rights reserved. 112 * 113 * Redistribution and use in source and binary forms, with or without 114 * modification, are permitted provided that the following conditions 115 * are met: 116 * 1. Redistributions of source code must retain the above copyright 117 * notice, this list of conditions and the following disclaimer. 118 * 2. Redistributions in binary form must reproduce the above copyright 119 * notice, this list of conditions and the following disclaimer in the 120 * documentation and/or other materials provided with the distribution. 121 * 3. Neither the name of the University nor the names of its contributors 122 * may be used to endorse or promote products derived from this software 123 * without specific prior written permission. 124 * 125 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 126 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 127 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 128 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 129 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 130 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 131 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 132 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 133 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 134 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 135 * SUCH DAMAGE. 136 * 137 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 138 */ 139 140 #include <sys/cdefs.h> 141 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.434 2022/09/20 07:19:14 ozaki-r Exp $"); 142 143 #ifdef _KERNEL_OPT 144 #include "opt_inet.h" 145 #include "opt_ipsec.h" 146 #include "opt_inet_csum.h" 147 #include "opt_tcp_debug.h" 148 #endif 149 150 #include <sys/param.h> 151 #include <sys/systm.h> 152 #include <sys/malloc.h> 153 #include <sys/mbuf.h> 154 #include <sys/protosw.h> 155 #include <sys/socket.h> 156 #include <sys/socketvar.h> 157 #include <sys/errno.h> 158 #include <sys/syslog.h> 159 #include <sys/pool.h> 160 #include <sys/domain.h> 161 #include <sys/kernel.h> 162 #ifdef TCP_SIGNATURE 163 #include <sys/md5.h> 164 #endif 165 #include <sys/lwp.h> /* for lwp0 */ 166 #include <sys/cprng.h> 167 168 #include <net/if.h> 169 #include <net/if_types.h> 170 171 #include <netinet/in.h> 172 #include <netinet/in_systm.h> 173 #include <netinet/ip.h> 174 #include <netinet/in_pcb.h> 175 #include <netinet/in_var.h> 176 #include <netinet/ip_var.h> 177 #include <netinet/in_offload.h> 178 179 #if NARP > 0 180 #include <netinet/if_inarp.h> 181 #endif 182 #ifdef INET6 183 #include <netinet/ip6.h> 184 #include <netinet6/ip6_var.h> 185 #include <netinet6/in6_pcb.h> 186 #include <netinet6/ip6_var.h> 187 #include <netinet6/in6_var.h> 188 #include <netinet/icmp6.h> 189 #include <netinet6/nd6.h> 190 #ifdef TCP_SIGNATURE 191 #include <netinet6/scope6_var.h> 192 #endif 193 #endif 194 195 #ifndef INET6 196 #include <netinet/ip6.h> 197 #endif 198 199 #include <netinet/tcp.h> 200 #include <netinet/tcp_fsm.h> 201 #include <netinet/tcp_seq.h> 202 #include <netinet/tcp_timer.h> 203 #include <netinet/tcp_var.h> 204 #include <netinet/tcp_private.h> 205 #include <netinet/tcp_congctl.h> 206 #include <netinet/tcp_debug.h> 207 #include <netinet/tcp_syncache.h> 208 209 #ifdef INET6 210 #include "faith.h" 211 #if defined(NFAITH) && NFAITH > 0 212 #include <net/if_faith.h> 213 #endif 214 #endif 215 216 #ifdef IPSEC 217 #include <netipsec/ipsec.h> 218 #include <netipsec/key.h> 219 #ifdef INET6 220 #include <netipsec/ipsec6.h> 221 #endif 222 #endif /* IPSEC*/ 223 224 #include <netinet/tcp_vtw.h> 225 226 int tcprexmtthresh = 3; 227 int tcp_log_refused; 228 229 int tcp_do_autorcvbuf = 1; 230 int tcp_autorcvbuf_inc = 16 * 1024; 231 int tcp_autorcvbuf_max = 256 * 1024; 232 int tcp_msl = (TCPTV_MSL / PR_SLOWHZ); 233 234 static int tcp_rst_ppslim_count = 0; 235 static struct timeval tcp_rst_ppslim_last; 236 static int tcp_ackdrop_ppslim_count = 0; 237 static struct timeval tcp_ackdrop_ppslim_last; 238 239 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 240 241 /* for modulo comparisons of timestamps */ 242 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 243 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 244 245 /* 246 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 247 */ 248 static void 249 nd_hint(struct tcpcb *tp) 250 { 251 struct route *ro = NULL; 252 struct rtentry *rt; 253 254 if (tp == NULL) 255 return; 256 257 switch (tp->t_family) { 258 #if NARP > 0 259 case AF_INET: 260 if (tp->t_inpcb != NULL) 261 ro = &tp->t_inpcb->inp_route; 262 break; 263 #endif 264 #ifdef INET6 265 case AF_INET6: 266 if (tp->t_in6pcb != NULL) 267 ro = &tp->t_in6pcb->in6p_route; 268 break; 269 #endif 270 } 271 272 if (ro == NULL) 273 return; 274 275 rt = rtcache_validate(ro); 276 if (rt == NULL) 277 return; 278 279 switch (tp->t_family) { 280 #if NARP > 0 281 case AF_INET: 282 arp_nud_hint(rt); 283 break; 284 #endif 285 #ifdef INET6 286 case AF_INET6: 287 nd6_nud_hint(rt); 288 break; 289 #endif 290 } 291 292 rtcache_unref(rt, ro); 293 } 294 295 /* 296 * Compute ACK transmission behavior. Delay the ACK unless 297 * we have already delayed an ACK (must send an ACK every two segments). 298 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 299 * option is enabled. 300 */ 301 static void 302 tcp_setup_ack(struct tcpcb *tp, const struct tcphdr *th) 303 { 304 305 if (tp->t_flags & TF_DELACK || 306 (tcp_ack_on_push && th->th_flags & TH_PUSH)) 307 tp->t_flags |= TF_ACKNOW; 308 else 309 TCP_SET_DELACK(tp); 310 } 311 312 static void 313 icmp_check(struct tcpcb *tp, const struct tcphdr *th, int acked) 314 { 315 316 /* 317 * If we had a pending ICMP message that refers to data that have 318 * just been acknowledged, disregard the recorded ICMP message. 319 */ 320 if ((tp->t_flags & TF_PMTUD_PEND) && 321 SEQ_GT(th->th_ack, tp->t_pmtud_th_seq)) 322 tp->t_flags &= ~TF_PMTUD_PEND; 323 324 /* 325 * Keep track of the largest chunk of data 326 * acknowledged since last PMTU update 327 */ 328 if (tp->t_pmtud_mss_acked < acked) 329 tp->t_pmtud_mss_acked = acked; 330 } 331 332 /* 333 * Convert TCP protocol fields to host order for easier processing. 334 */ 335 static void 336 tcp_fields_to_host(struct tcphdr *th) 337 { 338 339 NTOHL(th->th_seq); 340 NTOHL(th->th_ack); 341 NTOHS(th->th_win); 342 NTOHS(th->th_urp); 343 } 344 345 /* 346 * ... and reverse the above. 347 */ 348 static void 349 tcp_fields_to_net(struct tcphdr *th) 350 { 351 352 HTONL(th->th_seq); 353 HTONL(th->th_ack); 354 HTONS(th->th_win); 355 HTONS(th->th_urp); 356 } 357 358 static void 359 tcp_urp_drop(struct tcphdr *th, int todrop, int *tiflags) 360 { 361 if (th->th_urp > todrop) { 362 th->th_urp -= todrop; 363 } else { 364 *tiflags &= ~TH_URG; 365 th->th_urp = 0; 366 } 367 } 368 369 #ifdef TCP_CSUM_COUNTERS 370 #include <sys/device.h> 371 372 extern struct evcnt tcp_hwcsum_ok; 373 extern struct evcnt tcp_hwcsum_bad; 374 extern struct evcnt tcp_hwcsum_data; 375 extern struct evcnt tcp_swcsum; 376 #if defined(INET6) 377 extern struct evcnt tcp6_hwcsum_ok; 378 extern struct evcnt tcp6_hwcsum_bad; 379 extern struct evcnt tcp6_hwcsum_data; 380 extern struct evcnt tcp6_swcsum; 381 #endif /* defined(INET6) */ 382 383 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 384 385 #else 386 387 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 388 389 #endif /* TCP_CSUM_COUNTERS */ 390 391 #ifdef TCP_REASS_COUNTERS 392 #include <sys/device.h> 393 394 extern struct evcnt tcp_reass_; 395 extern struct evcnt tcp_reass_empty; 396 extern struct evcnt tcp_reass_iteration[8]; 397 extern struct evcnt tcp_reass_prependfirst; 398 extern struct evcnt tcp_reass_prepend; 399 extern struct evcnt tcp_reass_insert; 400 extern struct evcnt tcp_reass_inserttail; 401 extern struct evcnt tcp_reass_append; 402 extern struct evcnt tcp_reass_appendtail; 403 extern struct evcnt tcp_reass_overlaptail; 404 extern struct evcnt tcp_reass_overlapfront; 405 extern struct evcnt tcp_reass_segdup; 406 extern struct evcnt tcp_reass_fragdup; 407 408 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 409 410 #else 411 412 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 413 414 #endif /* TCP_REASS_COUNTERS */ 415 416 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *, 417 int); 418 419 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 420 #ifdef INET6 421 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 422 #endif 423 424 #if defined(MBUFTRACE) 425 struct mowner tcp_reass_mowner = MOWNER_INIT("tcp", "reass"); 426 #endif /* defined(MBUFTRACE) */ 427 428 static struct pool tcpipqent_pool; 429 430 void 431 tcpipqent_init(void) 432 { 433 434 pool_init(&tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", 435 NULL, IPL_VM); 436 } 437 438 struct ipqent * 439 tcpipqent_alloc(void) 440 { 441 struct ipqent *ipqe; 442 int s; 443 444 s = splvm(); 445 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 446 splx(s); 447 448 return ipqe; 449 } 450 451 void 452 tcpipqent_free(struct ipqent *ipqe) 453 { 454 int s; 455 456 s = splvm(); 457 pool_put(&tcpipqent_pool, ipqe); 458 splx(s); 459 } 460 461 /* 462 * Insert segment ti into reassembly queue of tcp with 463 * control block tp. Return TH_FIN if reassembly now includes 464 * a segment with FIN. 465 */ 466 static int 467 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int tlen) 468 { 469 struct ipqent *p, *q, *nq, *tiqe = NULL; 470 struct socket *so = NULL; 471 int pkt_flags; 472 tcp_seq pkt_seq; 473 unsigned pkt_len; 474 u_long rcvpartdupbyte = 0; 475 u_long rcvoobyte; 476 #ifdef TCP_REASS_COUNTERS 477 u_int count = 0; 478 #endif 479 uint64_t *tcps; 480 481 if (tp->t_inpcb) 482 so = tp->t_inpcb->inp_socket; 483 #ifdef INET6 484 else if (tp->t_in6pcb) 485 so = tp->t_in6pcb->in6p_socket; 486 #endif 487 488 TCP_REASS_LOCK_CHECK(tp); 489 490 /* 491 * Call with th==NULL after become established to 492 * force pre-ESTABLISHED data up to user socket. 493 */ 494 if (th == NULL) 495 goto present; 496 497 m_claimm(m, &tcp_reass_mowner); 498 499 rcvoobyte = tlen; 500 /* 501 * Copy these to local variables because the TCP header gets munged 502 * while we are collapsing mbufs. 503 */ 504 pkt_seq = th->th_seq; 505 pkt_len = tlen; 506 pkt_flags = th->th_flags; 507 508 TCP_REASS_COUNTER_INCR(&tcp_reass_); 509 510 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 511 /* 512 * When we miss a packet, the vast majority of time we get 513 * packets that follow it in order. So optimize for that. 514 */ 515 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 516 p->ipqe_len += pkt_len; 517 p->ipqe_flags |= pkt_flags; 518 m_cat(p->ipqe_m, m); 519 m = NULL; 520 tiqe = p; 521 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 522 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 523 goto skip_replacement; 524 } 525 /* 526 * While we're here, if the pkt is completely beyond 527 * anything we have, just insert it at the tail. 528 */ 529 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 530 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 531 goto insert_it; 532 } 533 } 534 535 q = TAILQ_FIRST(&tp->segq); 536 537 if (q != NULL) { 538 /* 539 * If this segment immediately precedes the first out-of-order 540 * block, simply slap the segment in front of it and (mostly) 541 * skip the complicated logic. 542 */ 543 if (pkt_seq + pkt_len == q->ipqe_seq) { 544 q->ipqe_seq = pkt_seq; 545 q->ipqe_len += pkt_len; 546 q->ipqe_flags |= pkt_flags; 547 m_cat(m, q->ipqe_m); 548 q->ipqe_m = m; 549 tiqe = q; 550 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 551 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 552 goto skip_replacement; 553 } 554 } else { 555 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 556 } 557 558 /* 559 * Find a segment which begins after this one does. 560 */ 561 for (p = NULL; q != NULL; q = nq) { 562 nq = TAILQ_NEXT(q, ipqe_q); 563 #ifdef TCP_REASS_COUNTERS 564 count++; 565 #endif 566 567 /* 568 * If the received segment is just right after this 569 * fragment, merge the two together and then check 570 * for further overlaps. 571 */ 572 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 573 pkt_len += q->ipqe_len; 574 pkt_flags |= q->ipqe_flags; 575 pkt_seq = q->ipqe_seq; 576 m_cat(q->ipqe_m, m); 577 m = q->ipqe_m; 578 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 579 goto free_ipqe; 580 } 581 582 /* 583 * If the received segment is completely past this 584 * fragment, we need to go to the next fragment. 585 */ 586 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 587 p = q; 588 continue; 589 } 590 591 /* 592 * If the fragment is past the received segment, 593 * it (or any following) can't be concatenated. 594 */ 595 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 596 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 597 break; 598 } 599 600 /* 601 * We've received all the data in this segment before. 602 * Mark it as a duplicate and return. 603 */ 604 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 605 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 606 tcps = TCP_STAT_GETREF(); 607 tcps[TCP_STAT_RCVDUPPACK]++; 608 tcps[TCP_STAT_RCVDUPBYTE] += pkt_len; 609 TCP_STAT_PUTREF(); 610 tcp_new_dsack(tp, pkt_seq, pkt_len); 611 m_freem(m); 612 if (tiqe != NULL) { 613 tcpipqent_free(tiqe); 614 } 615 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 616 goto out; 617 } 618 619 /* 620 * Received segment completely overlaps this fragment 621 * so we drop the fragment (this keeps the temporal 622 * ordering of segments correct). 623 */ 624 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 625 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 626 rcvpartdupbyte += q->ipqe_len; 627 m_freem(q->ipqe_m); 628 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 629 goto free_ipqe; 630 } 631 632 /* 633 * Received segment extends past the end of the fragment. 634 * Drop the overlapping bytes, merge the fragment and 635 * segment, and treat as a longer received packet. 636 */ 637 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 638 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 639 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 640 m_adj(m, overlap); 641 rcvpartdupbyte += overlap; 642 m_cat(q->ipqe_m, m); 643 m = q->ipqe_m; 644 pkt_seq = q->ipqe_seq; 645 pkt_len += q->ipqe_len - overlap; 646 rcvoobyte -= overlap; 647 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 648 goto free_ipqe; 649 } 650 651 /* 652 * Received segment extends past the front of the fragment. 653 * Drop the overlapping bytes on the received packet. The 654 * packet will then be concatenated with this fragment a 655 * bit later. 656 */ 657 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 658 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 659 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 660 m_adj(m, -overlap); 661 pkt_len -= overlap; 662 rcvpartdupbyte += overlap; 663 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 664 rcvoobyte -= overlap; 665 } 666 667 /* 668 * If the received segment immediately precedes this 669 * fragment then tack the fragment onto this segment 670 * and reinsert the data. 671 */ 672 if (q->ipqe_seq == pkt_seq + pkt_len) { 673 pkt_len += q->ipqe_len; 674 pkt_flags |= q->ipqe_flags; 675 m_cat(m, q->ipqe_m); 676 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 677 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 678 tp->t_segqlen--; 679 KASSERT(tp->t_segqlen >= 0); 680 KASSERT(tp->t_segqlen != 0 || 681 (TAILQ_EMPTY(&tp->segq) && 682 TAILQ_EMPTY(&tp->timeq))); 683 if (tiqe == NULL) { 684 tiqe = q; 685 } else { 686 tcpipqent_free(q); 687 } 688 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 689 break; 690 } 691 692 /* 693 * If the fragment is before the segment, remember it. 694 * When this loop is terminated, p will contain the 695 * pointer to the fragment that is right before the 696 * received segment. 697 */ 698 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 699 p = q; 700 701 continue; 702 703 /* 704 * This is a common operation. It also will allow 705 * to save doing a malloc/free in most instances. 706 */ 707 free_ipqe: 708 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 709 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 710 tp->t_segqlen--; 711 KASSERT(tp->t_segqlen >= 0); 712 KASSERT(tp->t_segqlen != 0 || 713 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 714 if (tiqe == NULL) { 715 tiqe = q; 716 } else { 717 tcpipqent_free(q); 718 } 719 } 720 721 #ifdef TCP_REASS_COUNTERS 722 if (count > 7) 723 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 724 else if (count > 0) 725 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 726 #endif 727 728 insert_it: 729 /* 730 * Allocate a new queue entry (block) since the received segment 731 * did not collapse onto any other out-of-order block. If it had 732 * collapsed, tiqe would not be NULL and we would be reusing it. 733 * 734 * If the allocation fails, drop the packet. 735 */ 736 if (tiqe == NULL) { 737 tiqe = tcpipqent_alloc(); 738 if (tiqe == NULL) { 739 TCP_STATINC(TCP_STAT_RCVMEMDROP); 740 m_freem(m); 741 goto out; 742 } 743 } 744 745 /* 746 * Update the counters. 747 */ 748 tp->t_rcvoopack++; 749 tcps = TCP_STAT_GETREF(); 750 tcps[TCP_STAT_RCVOOPACK]++; 751 tcps[TCP_STAT_RCVOOBYTE] += rcvoobyte; 752 if (rcvpartdupbyte) { 753 tcps[TCP_STAT_RCVPARTDUPPACK]++; 754 tcps[TCP_STAT_RCVPARTDUPBYTE] += rcvpartdupbyte; 755 } 756 TCP_STAT_PUTREF(); 757 758 /* 759 * Insert the new fragment queue entry into both queues. 760 */ 761 tiqe->ipqe_m = m; 762 tiqe->ipqe_seq = pkt_seq; 763 tiqe->ipqe_len = pkt_len; 764 tiqe->ipqe_flags = pkt_flags; 765 if (p == NULL) { 766 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 767 } else { 768 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 769 } 770 tp->t_segqlen++; 771 772 skip_replacement: 773 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 774 775 present: 776 /* 777 * Present data to user, advancing rcv_nxt through 778 * completed sequence space. 779 */ 780 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 781 goto out; 782 q = TAILQ_FIRST(&tp->segq); 783 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 784 goto out; 785 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 786 goto out; 787 788 tp->rcv_nxt += q->ipqe_len; 789 pkt_flags = q->ipqe_flags & TH_FIN; 790 nd_hint(tp); 791 792 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 793 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 794 tp->t_segqlen--; 795 KASSERT(tp->t_segqlen >= 0); 796 KASSERT(tp->t_segqlen != 0 || 797 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 798 if (so->so_state & SS_CANTRCVMORE) 799 m_freem(q->ipqe_m); 800 else 801 sbappendstream(&so->so_rcv, q->ipqe_m); 802 tcpipqent_free(q); 803 TCP_REASS_UNLOCK(tp); 804 sorwakeup(so); 805 return pkt_flags; 806 807 out: 808 TCP_REASS_UNLOCK(tp); 809 return 0; 810 } 811 812 #ifdef INET6 813 int 814 tcp6_input(struct mbuf **mp, int *offp, int proto) 815 { 816 struct mbuf *m = *mp; 817 818 /* 819 * draft-itojun-ipv6-tcp-to-anycast 820 * better place to put this in? 821 */ 822 if (m->m_flags & M_ANYCAST6) { 823 struct ip6_hdr *ip6; 824 if (m->m_len < sizeof(struct ip6_hdr)) { 825 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 826 TCP_STATINC(TCP_STAT_RCVSHORT); 827 return IPPROTO_DONE; 828 } 829 } 830 ip6 = mtod(m, struct ip6_hdr *); 831 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 832 (char *)&ip6->ip6_dst - (char *)ip6); 833 return IPPROTO_DONE; 834 } 835 836 tcp_input(m, *offp, proto); 837 return IPPROTO_DONE; 838 } 839 #endif 840 841 static void 842 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 843 { 844 char src[INET_ADDRSTRLEN]; 845 char dst[INET_ADDRSTRLEN]; 846 847 if (ip) { 848 in_print(src, sizeof(src), &ip->ip_src); 849 in_print(dst, sizeof(dst), &ip->ip_dst); 850 } else { 851 strlcpy(src, "(unknown)", sizeof(src)); 852 strlcpy(dst, "(unknown)", sizeof(dst)); 853 } 854 log(LOG_INFO, 855 "Connection attempt to TCP %s:%d from %s:%d\n", 856 dst, ntohs(th->th_dport), 857 src, ntohs(th->th_sport)); 858 } 859 860 #ifdef INET6 861 static void 862 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 863 { 864 char src[INET6_ADDRSTRLEN]; 865 char dst[INET6_ADDRSTRLEN]; 866 867 if (ip6) { 868 in6_print(src, sizeof(src), &ip6->ip6_src); 869 in6_print(dst, sizeof(dst), &ip6->ip6_dst); 870 } else { 871 strlcpy(src, "(unknown v6)", sizeof(src)); 872 strlcpy(dst, "(unknown v6)", sizeof(dst)); 873 } 874 log(LOG_INFO, 875 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 876 dst, ntohs(th->th_dport), 877 src, ntohs(th->th_sport)); 878 } 879 #endif 880 881 /* 882 * Checksum extended TCP header and data. 883 */ 884 int 885 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, 886 int toff, int off, int tlen) 887 { 888 struct ifnet *rcvif; 889 int s; 890 891 /* 892 * XXX it's better to record and check if this mbuf is 893 * already checked. 894 */ 895 896 rcvif = m_get_rcvif(m, &s); 897 if (__predict_false(rcvif == NULL)) 898 goto badcsum; /* XXX */ 899 900 switch (af) { 901 case AF_INET: 902 switch (m->m_pkthdr.csum_flags & 903 ((rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 904 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 905 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 906 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 907 goto badcsum; 908 909 case M_CSUM_TCPv4|M_CSUM_DATA: { 910 u_int32_t hw_csum = m->m_pkthdr.csum_data; 911 912 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 913 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 914 const struct ip *ip = 915 mtod(m, const struct ip *); 916 917 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 918 ip->ip_dst.s_addr, 919 htons(hw_csum + tlen + off + IPPROTO_TCP)); 920 } 921 if ((hw_csum ^ 0xffff) != 0) 922 goto badcsum; 923 break; 924 } 925 926 case M_CSUM_TCPv4: 927 /* Checksum was okay. */ 928 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 929 break; 930 931 default: 932 /* 933 * Must compute it ourselves. Maybe skip checksum 934 * on loopback interfaces. 935 */ 936 if (__predict_true(!(rcvif->if_flags & IFF_LOOPBACK) || 937 tcp_do_loopback_cksum)) { 938 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 939 if (in4_cksum(m, IPPROTO_TCP, toff, 940 tlen + off) != 0) 941 goto badcsum; 942 } 943 break; 944 } 945 break; 946 947 #ifdef INET6 948 case AF_INET6: 949 switch (m->m_pkthdr.csum_flags & 950 ((rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 951 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 952 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 953 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 954 goto badcsum; 955 956 #if 0 /* notyet */ 957 case M_CSUM_TCPv6|M_CSUM_DATA: 958 #endif 959 960 case M_CSUM_TCPv6: 961 /* Checksum was okay. */ 962 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 963 break; 964 965 default: 966 /* 967 * Must compute it ourselves. Maybe skip checksum 968 * on loopback interfaces. 969 */ 970 if (__predict_true((m->m_flags & M_LOOP) == 0 || 971 tcp_do_loopback_cksum)) { 972 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 973 if (in6_cksum(m, IPPROTO_TCP, toff, 974 tlen + off) != 0) 975 goto badcsum; 976 } 977 } 978 break; 979 #endif /* INET6 */ 980 } 981 m_put_rcvif(rcvif, &s); 982 983 return 0; 984 985 badcsum: 986 m_put_rcvif(rcvif, &s); 987 TCP_STATINC(TCP_STAT_RCVBADSUM); 988 return -1; 989 } 990 991 /* 992 * When a packet arrives addressed to a vestigial tcpbp, we 993 * nevertheless have to respond to it per the spec. 994 * 995 * This code is duplicated from the one in tcp_input(). 996 */ 997 static void tcp_vtw_input(struct tcphdr *th, vestigial_inpcb_t *vp, 998 struct mbuf *m, int tlen) 999 { 1000 int tiflags; 1001 int todrop; 1002 uint32_t t_flags = 0; 1003 uint64_t *tcps; 1004 1005 tiflags = th->th_flags; 1006 todrop = vp->rcv_nxt - th->th_seq; 1007 1008 if (todrop > 0) { 1009 if (tiflags & TH_SYN) { 1010 tiflags &= ~TH_SYN; 1011 th->th_seq++; 1012 tcp_urp_drop(th, 1, &tiflags); 1013 todrop--; 1014 } 1015 if (todrop > tlen || 1016 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1017 /* 1018 * Any valid FIN or RST must be to the left of the 1019 * window. At this point the FIN or RST must be a 1020 * duplicate or out of sequence; drop it. 1021 */ 1022 if (tiflags & TH_RST) 1023 goto drop; 1024 tiflags &= ~(TH_FIN|TH_RST); 1025 1026 /* 1027 * Send an ACK to resynchronize and drop any data. 1028 * But keep on processing for RST or ACK. 1029 */ 1030 t_flags |= TF_ACKNOW; 1031 todrop = tlen; 1032 tcps = TCP_STAT_GETREF(); 1033 tcps[TCP_STAT_RCVDUPPACK] += 1; 1034 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 1035 TCP_STAT_PUTREF(); 1036 } else if ((tiflags & TH_RST) && 1037 th->th_seq != vp->rcv_nxt) { 1038 /* 1039 * Test for reset before adjusting the sequence 1040 * number for overlapping data. 1041 */ 1042 goto dropafterack_ratelim; 1043 } else { 1044 tcps = TCP_STAT_GETREF(); 1045 tcps[TCP_STAT_RCVPARTDUPPACK] += 1; 1046 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 1047 TCP_STAT_PUTREF(); 1048 } 1049 1050 // tcp_new_dsack(tp, th->th_seq, todrop); 1051 // hdroptlen += todrop; /*drop from head afterwards*/ 1052 1053 th->th_seq += todrop; 1054 tlen -= todrop; 1055 tcp_urp_drop(th, todrop, &tiflags); 1056 } 1057 1058 /* 1059 * If new data are received on a connection after the 1060 * user processes are gone, then RST the other end. 1061 */ 1062 if (tlen) { 1063 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 1064 goto dropwithreset; 1065 } 1066 1067 /* 1068 * If segment ends after window, drop trailing data 1069 * (and PUSH and FIN); if nothing left, just ACK. 1070 */ 1071 todrop = (th->th_seq + tlen) - (vp->rcv_nxt + vp->rcv_wnd); 1072 1073 if (todrop > 0) { 1074 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 1075 if (todrop >= tlen) { 1076 /* 1077 * The segment actually starts after the window. 1078 * th->th_seq + tlen - vp->rcv_nxt - vp->rcv_wnd >= tlen 1079 * th->th_seq - vp->rcv_nxt - vp->rcv_wnd >= 0 1080 * th->th_seq >= vp->rcv_nxt + vp->rcv_wnd 1081 */ 1082 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 1083 1084 /* 1085 * If a new connection request is received 1086 * while in TIME_WAIT, drop the old connection 1087 * and start over if the sequence numbers 1088 * are above the previous ones. 1089 */ 1090 if ((tiflags & TH_SYN) && 1091 SEQ_GT(th->th_seq, vp->rcv_nxt)) { 1092 /* 1093 * We only support this in the !NOFDREF case, which 1094 * is to say: not here. 1095 */ 1096 goto dropwithreset; 1097 } 1098 1099 /* 1100 * If window is closed can only take segments at 1101 * window edge, and have to drop data and PUSH from 1102 * incoming segments. Continue processing, but 1103 * remember to ack. Otherwise, drop segment 1104 * and (if not RST) ack. 1105 */ 1106 if (vp->rcv_wnd == 0 && th->th_seq == vp->rcv_nxt) { 1107 t_flags |= TF_ACKNOW; 1108 TCP_STATINC(TCP_STAT_RCVWINPROBE); 1109 } else { 1110 goto dropafterack; 1111 } 1112 } else { 1113 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 1114 } 1115 m_adj(m, -todrop); 1116 tlen -= todrop; 1117 tiflags &= ~(TH_PUSH|TH_FIN); 1118 } 1119 1120 if (tiflags & TH_RST) { 1121 if (th->th_seq != vp->rcv_nxt) 1122 goto dropafterack_ratelim; 1123 1124 vtw_del(vp->ctl, vp->vtw); 1125 goto drop; 1126 } 1127 1128 /* 1129 * If the ACK bit is off we drop the segment and return. 1130 */ 1131 if ((tiflags & TH_ACK) == 0) { 1132 if (t_flags & TF_ACKNOW) 1133 goto dropafterack; 1134 goto drop; 1135 } 1136 1137 /* 1138 * In TIME_WAIT state the only thing that should arrive 1139 * is a retransmission of the remote FIN. Acknowledge 1140 * it and restart the finack timer. 1141 */ 1142 vtw_restart(vp); 1143 goto dropafterack; 1144 1145 dropafterack: 1146 /* 1147 * Generate an ACK dropping incoming segment if it occupies 1148 * sequence space, where the ACK reflects our state. 1149 */ 1150 if (tiflags & TH_RST) 1151 goto drop; 1152 goto dropafterack2; 1153 1154 dropafterack_ratelim: 1155 /* 1156 * We may want to rate-limit ACKs against SYN/RST attack. 1157 */ 1158 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 1159 tcp_ackdrop_ppslim) == 0) { 1160 /* XXX stat */ 1161 goto drop; 1162 } 1163 /* ...fall into dropafterack2... */ 1164 1165 dropafterack2: 1166 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, th->th_ack, TH_ACK); 1167 return; 1168 1169 dropwithreset: 1170 /* 1171 * Generate a RST, dropping incoming segment. 1172 * Make ACK acceptable to originator of segment. 1173 */ 1174 if (tiflags & TH_RST) 1175 goto drop; 1176 1177 if (tiflags & TH_ACK) { 1178 tcp_respond(0, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 1179 } else { 1180 if (tiflags & TH_SYN) 1181 ++tlen; 1182 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, (tcp_seq)0, 1183 TH_RST|TH_ACK); 1184 } 1185 return; 1186 drop: 1187 m_freem(m); 1188 } 1189 1190 /* 1191 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 1192 */ 1193 void 1194 tcp_input(struct mbuf *m, int off, int proto) 1195 { 1196 struct tcphdr *th; 1197 struct ip *ip; 1198 struct inpcb *inp; 1199 #ifdef INET6 1200 struct ip6_hdr *ip6; 1201 struct in6pcb *in6p; 1202 #endif 1203 u_int8_t *optp = NULL; 1204 int optlen = 0; 1205 int len, tlen, hdroptlen = 0; 1206 struct tcpcb *tp = NULL; 1207 int tiflags; 1208 struct socket *so = NULL; 1209 int todrop, acked, ourfinisacked, needoutput = 0; 1210 bool dupseg; 1211 #ifdef TCP_DEBUG 1212 short ostate = 0; 1213 #endif 1214 u_long tiwin; 1215 struct tcp_opt_info opti; 1216 int thlen, iphlen; 1217 int af; /* af on the wire */ 1218 struct mbuf *tcp_saveti = NULL; 1219 uint32_t ts_rtt; 1220 uint8_t iptos; 1221 uint64_t *tcps; 1222 vestigial_inpcb_t vestige; 1223 1224 vestige.valid = 0; 1225 1226 MCLAIM(m, &tcp_rx_mowner); 1227 1228 TCP_STATINC(TCP_STAT_RCVTOTAL); 1229 1230 memset(&opti, 0, sizeof(opti)); 1231 opti.ts_present = 0; 1232 opti.maxseg = 0; 1233 1234 /* 1235 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 1236 * 1237 * TCP is, by definition, unicast, so we reject all 1238 * multicast outright. 1239 * 1240 * Note, there are additional src/dst address checks in 1241 * the AF-specific code below. 1242 */ 1243 if (m->m_flags & (M_BCAST|M_MCAST)) { 1244 /* XXX stat */ 1245 goto drop; 1246 } 1247 #ifdef INET6 1248 if (m->m_flags & M_ANYCAST6) { 1249 /* XXX stat */ 1250 goto drop; 1251 } 1252 #endif 1253 1254 M_REGION_GET(th, struct tcphdr *, m, off, sizeof(struct tcphdr)); 1255 if (th == NULL) { 1256 TCP_STATINC(TCP_STAT_RCVSHORT); 1257 return; 1258 } 1259 1260 /* 1261 * Enforce alignment requirements that are violated in 1262 * some cases, see kern/50766 for details. 1263 */ 1264 if (ACCESSIBLE_POINTER(th, struct tcphdr) == 0) { 1265 m = m_copyup(m, off + sizeof(struct tcphdr), 0); 1266 if (m == NULL) { 1267 TCP_STATINC(TCP_STAT_RCVSHORT); 1268 return; 1269 } 1270 th = (struct tcphdr *)(mtod(m, char *) + off); 1271 } 1272 KASSERT(ACCESSIBLE_POINTER(th, struct tcphdr)); 1273 1274 /* 1275 * Get IP and TCP header. 1276 * Note: IP leaves IP header in first mbuf. 1277 */ 1278 ip = mtod(m, struct ip *); 1279 #ifdef INET6 1280 ip6 = mtod(m, struct ip6_hdr *); 1281 #endif 1282 switch (ip->ip_v) { 1283 case 4: 1284 af = AF_INET; 1285 iphlen = sizeof(struct ip); 1286 1287 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1288 in_broadcast(ip->ip_dst, m_get_rcvif_NOMPSAFE(m))) 1289 goto drop; 1290 1291 /* We do the checksum after PCB lookup... */ 1292 len = ntohs(ip->ip_len); 1293 tlen = len - off; 1294 iptos = ip->ip_tos; 1295 break; 1296 #ifdef INET6 1297 case 6: 1298 iphlen = sizeof(struct ip6_hdr); 1299 af = AF_INET6; 1300 1301 /* 1302 * Be proactive about unspecified IPv6 address in source. 1303 * As we use all-zero to indicate unbounded/unconnected pcb, 1304 * unspecified IPv6 address can be used to confuse us. 1305 * 1306 * Note that packets with unspecified IPv6 destination is 1307 * already dropped in ip6_input. 1308 */ 1309 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1310 /* XXX stat */ 1311 goto drop; 1312 } 1313 1314 /* 1315 * Make sure destination address is not multicast. 1316 * Source address checked in ip6_input(). 1317 */ 1318 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1319 /* XXX stat */ 1320 goto drop; 1321 } 1322 1323 /* We do the checksum after PCB lookup... */ 1324 len = m->m_pkthdr.len; 1325 tlen = len - off; 1326 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1327 break; 1328 #endif 1329 default: 1330 m_freem(m); 1331 return; 1332 } 1333 1334 1335 /* 1336 * Check that TCP offset makes sense, pull out TCP options and 1337 * adjust length. 1338 */ 1339 thlen = th->th_off << 2; 1340 if (thlen < sizeof(struct tcphdr) || thlen > tlen) { 1341 TCP_STATINC(TCP_STAT_RCVBADOFF); 1342 goto drop; 1343 } 1344 tlen -= thlen; 1345 1346 if (thlen > sizeof(struct tcphdr)) { 1347 M_REGION_GET(th, struct tcphdr *, m, off, thlen); 1348 if (th == NULL) { 1349 TCP_STATINC(TCP_STAT_RCVSHORT); 1350 return; 1351 } 1352 KASSERT(ACCESSIBLE_POINTER(th, struct tcphdr)); 1353 optlen = thlen - sizeof(struct tcphdr); 1354 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1355 1356 /* 1357 * Do quick retrieval of timestamp options. 1358 * 1359 * If timestamp is the only option and it's formatted as 1360 * recommended in RFC 1323 appendix A, we quickly get the 1361 * values now and don't bother calling tcp_dooptions(), 1362 * etc. 1363 */ 1364 if ((optlen == TCPOLEN_TSTAMP_APPA || 1365 (optlen > TCPOLEN_TSTAMP_APPA && 1366 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1367 be32dec(optp) == TCPOPT_TSTAMP_HDR && 1368 (th->th_flags & TH_SYN) == 0) { 1369 opti.ts_present = 1; 1370 opti.ts_val = be32dec(optp + 4); 1371 opti.ts_ecr = be32dec(optp + 8); 1372 optp = NULL; /* we've parsed the options */ 1373 } 1374 } 1375 tiflags = th->th_flags; 1376 1377 /* 1378 * Checksum extended TCP header and data 1379 */ 1380 if (tcp_input_checksum(af, m, th, off, thlen, tlen)) 1381 goto badcsum; 1382 1383 /* 1384 * Locate pcb for segment. 1385 */ 1386 findpcb: 1387 inp = NULL; 1388 #ifdef INET6 1389 in6p = NULL; 1390 #endif 1391 switch (af) { 1392 case AF_INET: 1393 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1394 ip->ip_dst, th->th_dport, &vestige); 1395 if (inp == NULL && !vestige.valid) { 1396 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1397 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, 1398 th->th_dport); 1399 } 1400 #ifdef INET6 1401 if (inp == NULL && !vestige.valid) { 1402 struct in6_addr s, d; 1403 1404 /* mapped addr case */ 1405 in6_in_2_v4mapin6(&ip->ip_src, &s); 1406 in6_in_2_v4mapin6(&ip->ip_dst, &d); 1407 in6p = in6_pcblookup_connect(&tcbtable, &s, 1408 th->th_sport, &d, th->th_dport, 0, &vestige); 1409 if (in6p == 0 && !vestige.valid) { 1410 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1411 in6p = in6_pcblookup_bind(&tcbtable, &d, 1412 th->th_dport, 0); 1413 } 1414 } 1415 #endif 1416 #ifndef INET6 1417 if (inp == NULL && !vestige.valid) 1418 #else 1419 if (inp == NULL && in6p == NULL && !vestige.valid) 1420 #endif 1421 { 1422 TCP_STATINC(TCP_STAT_NOPORT); 1423 if (tcp_log_refused && 1424 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1425 tcp4_log_refused(ip, th); 1426 } 1427 tcp_fields_to_host(th); 1428 goto dropwithreset_ratelim; 1429 } 1430 #if defined(IPSEC) 1431 if (ipsec_used) { 1432 if (inp && ipsec_in_reject(m, inp)) { 1433 goto drop; 1434 } 1435 #ifdef INET6 1436 else if (in6p && ipsec_in_reject(m, in6p)) { 1437 goto drop; 1438 } 1439 #endif 1440 } 1441 #endif /*IPSEC*/ 1442 break; 1443 #ifdef INET6 1444 case AF_INET6: 1445 { 1446 int faith; 1447 1448 #if defined(NFAITH) && NFAITH > 0 1449 faith = faithprefix(&ip6->ip6_dst); 1450 #else 1451 faith = 0; 1452 #endif 1453 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1454 th->th_sport, &ip6->ip6_dst, th->th_dport, faith, &vestige); 1455 if (!in6p && !vestige.valid) { 1456 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1457 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1458 th->th_dport, faith); 1459 } 1460 if (!in6p && !vestige.valid) { 1461 TCP_STATINC(TCP_STAT_NOPORT); 1462 if (tcp_log_refused && 1463 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1464 tcp6_log_refused(ip6, th); 1465 } 1466 tcp_fields_to_host(th); 1467 goto dropwithreset_ratelim; 1468 } 1469 #if defined(IPSEC) 1470 if (ipsec_used && in6p && ipsec_in_reject(m, in6p)) { 1471 goto drop; 1472 } 1473 #endif 1474 break; 1475 } 1476 #endif 1477 } 1478 1479 tcp_fields_to_host(th); 1480 1481 /* 1482 * If the state is CLOSED (i.e., TCB does not exist) then 1483 * all data in the incoming segment is discarded. 1484 * If the TCB exists but is in CLOSED state, it is embryonic, 1485 * but should either do a listen or a connect soon. 1486 */ 1487 tp = NULL; 1488 so = NULL; 1489 if (inp) { 1490 /* Check the minimum TTL for socket. */ 1491 if (ip->ip_ttl < inp->inp_ip_minttl) 1492 goto drop; 1493 1494 tp = intotcpcb(inp); 1495 so = inp->inp_socket; 1496 } 1497 #ifdef INET6 1498 else if (in6p) { 1499 tp = in6totcpcb(in6p); 1500 so = in6p->in6p_socket; 1501 } 1502 #endif 1503 else if (vestige.valid) { 1504 /* We do not support the resurrection of vtw tcpcps. */ 1505 tcp_vtw_input(th, &vestige, m, tlen); 1506 m = NULL; 1507 goto drop; 1508 } 1509 1510 if (tp == NULL) 1511 goto dropwithreset_ratelim; 1512 if (tp->t_state == TCPS_CLOSED) 1513 goto drop; 1514 1515 KASSERT(so->so_lock == softnet_lock); 1516 KASSERT(solocked(so)); 1517 1518 /* Unscale the window into a 32-bit value. */ 1519 if ((tiflags & TH_SYN) == 0) 1520 tiwin = th->th_win << tp->snd_scale; 1521 else 1522 tiwin = th->th_win; 1523 1524 #ifdef INET6 1525 /* save packet options if user wanted */ 1526 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1527 if (in6p->in6p_options) { 1528 m_freem(in6p->in6p_options); 1529 in6p->in6p_options = NULL; 1530 } 1531 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1532 } 1533 #endif 1534 1535 if (so->so_options & SO_DEBUG) { 1536 #ifdef TCP_DEBUG 1537 ostate = tp->t_state; 1538 #endif 1539 1540 tcp_saveti = NULL; 1541 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1542 goto nosave; 1543 1544 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1545 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1546 if (tcp_saveti == NULL) 1547 goto nosave; 1548 } else { 1549 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1550 if (tcp_saveti == NULL) 1551 goto nosave; 1552 MCLAIM(m, &tcp_mowner); 1553 tcp_saveti->m_len = iphlen; 1554 m_copydata(m, 0, iphlen, 1555 mtod(tcp_saveti, void *)); 1556 } 1557 1558 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1559 m_freem(tcp_saveti); 1560 tcp_saveti = NULL; 1561 } else { 1562 tcp_saveti->m_len += sizeof(struct tcphdr); 1563 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1564 sizeof(struct tcphdr)); 1565 } 1566 nosave:; 1567 } 1568 1569 if (so->so_options & SO_ACCEPTCONN) { 1570 union syn_cache_sa src; 1571 union syn_cache_sa dst; 1572 1573 KASSERT(tp->t_state == TCPS_LISTEN); 1574 1575 memset(&src, 0, sizeof(src)); 1576 memset(&dst, 0, sizeof(dst)); 1577 switch (af) { 1578 case AF_INET: 1579 src.sin.sin_len = sizeof(struct sockaddr_in); 1580 src.sin.sin_family = AF_INET; 1581 src.sin.sin_addr = ip->ip_src; 1582 src.sin.sin_port = th->th_sport; 1583 1584 dst.sin.sin_len = sizeof(struct sockaddr_in); 1585 dst.sin.sin_family = AF_INET; 1586 dst.sin.sin_addr = ip->ip_dst; 1587 dst.sin.sin_port = th->th_dport; 1588 break; 1589 #ifdef INET6 1590 case AF_INET6: 1591 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1592 src.sin6.sin6_family = AF_INET6; 1593 src.sin6.sin6_addr = ip6->ip6_src; 1594 src.sin6.sin6_port = th->th_sport; 1595 1596 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1597 dst.sin6.sin6_family = AF_INET6; 1598 dst.sin6.sin6_addr = ip6->ip6_dst; 1599 dst.sin6.sin6_port = th->th_dport; 1600 break; 1601 #endif 1602 } 1603 1604 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1605 if (tiflags & TH_RST) { 1606 syn_cache_reset(&src.sa, &dst.sa, th); 1607 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1608 (TH_ACK|TH_SYN)) { 1609 /* 1610 * Received a SYN,ACK. This should never 1611 * happen while we are in LISTEN. Send an RST. 1612 */ 1613 goto badsyn; 1614 } else if (tiflags & TH_ACK) { 1615 so = syn_cache_get(&src.sa, &dst.sa, th, so, m); 1616 if (so == NULL) { 1617 /* 1618 * We don't have a SYN for this ACK; 1619 * send an RST. 1620 */ 1621 goto badsyn; 1622 } else if (so == (struct socket *)(-1)) { 1623 /* 1624 * We were unable to create the 1625 * connection. If the 3-way handshake 1626 * was completed, and RST has been 1627 * sent to the peer. Since the mbuf 1628 * might be in use for the reply, do 1629 * not free it. 1630 */ 1631 m = NULL; 1632 } else { 1633 /* 1634 * We have created a full-blown 1635 * connection. 1636 */ 1637 tp = NULL; 1638 inp = NULL; 1639 #ifdef INET6 1640 in6p = NULL; 1641 #endif 1642 switch (so->so_proto->pr_domain->dom_family) { 1643 case AF_INET: 1644 inp = sotoinpcb(so); 1645 tp = intotcpcb(inp); 1646 break; 1647 #ifdef INET6 1648 case AF_INET6: 1649 in6p = sotoin6pcb(so); 1650 tp = in6totcpcb(in6p); 1651 break; 1652 #endif 1653 } 1654 if (tp == NULL) 1655 goto badsyn; /*XXX*/ 1656 tiwin <<= tp->snd_scale; 1657 goto after_listen; 1658 } 1659 } else { 1660 /* 1661 * None of RST, SYN or ACK was set. 1662 * This is an invalid packet for a 1663 * TCB in LISTEN state. Send a RST. 1664 */ 1665 goto badsyn; 1666 } 1667 } else { 1668 /* 1669 * Received a SYN. 1670 */ 1671 1672 #ifdef INET6 1673 /* 1674 * If deprecated address is forbidden, we do 1675 * not accept SYN to deprecated interface 1676 * address to prevent any new inbound 1677 * connection from getting established. 1678 * When we do not accept SYN, we send a TCP 1679 * RST, with deprecated source address (instead 1680 * of dropping it). We compromise it as it is 1681 * much better for peer to send a RST, and 1682 * RST will be the final packet for the 1683 * exchange. 1684 * 1685 * If we do not forbid deprecated addresses, we 1686 * accept the SYN packet. RFC2462 does not 1687 * suggest dropping SYN in this case. 1688 * If we decipher RFC2462 5.5.4, it says like 1689 * this: 1690 * 1. use of deprecated addr with existing 1691 * communication is okay - "SHOULD continue 1692 * to be used" 1693 * 2. use of it with new communication: 1694 * (2a) "SHOULD NOT be used if alternate 1695 * address with sufficient scope is 1696 * available" 1697 * (2b) nothing mentioned otherwise. 1698 * Here we fall into (2b) case as we have no 1699 * choice in our source address selection - we 1700 * must obey the peer. 1701 * 1702 * The wording in RFC2462 is confusing, and 1703 * there are multiple description text for 1704 * deprecated address handling - worse, they 1705 * are not exactly the same. I believe 5.5.4 1706 * is the best one, so we follow 5.5.4. 1707 */ 1708 if (af == AF_INET6 && !ip6_use_deprecated) { 1709 struct in6_ifaddr *ia6; 1710 int s; 1711 struct ifnet *rcvif = m_get_rcvif(m, &s); 1712 if (rcvif == NULL) 1713 goto dropwithreset; /* XXX */ 1714 if ((ia6 = in6ifa_ifpwithaddr(rcvif, 1715 &ip6->ip6_dst)) && 1716 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1717 tp = NULL; 1718 m_put_rcvif(rcvif, &s); 1719 goto dropwithreset; 1720 } 1721 m_put_rcvif(rcvif, &s); 1722 } 1723 #endif 1724 1725 /* 1726 * LISTEN socket received a SYN from itself? This 1727 * can't possibly be valid; drop the packet. 1728 */ 1729 if (th->th_sport == th->th_dport) { 1730 int eq = 0; 1731 1732 switch (af) { 1733 case AF_INET: 1734 eq = in_hosteq(ip->ip_src, ip->ip_dst); 1735 break; 1736 #ifdef INET6 1737 case AF_INET6: 1738 eq = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, 1739 &ip6->ip6_dst); 1740 break; 1741 #endif 1742 } 1743 if (eq) { 1744 TCP_STATINC(TCP_STAT_BADSYN); 1745 goto drop; 1746 } 1747 } 1748 1749 /* 1750 * SYN looks ok; create compressed TCP 1751 * state for it. 1752 */ 1753 if (so->so_qlen <= so->so_qlimit && 1754 syn_cache_add(&src.sa, &dst.sa, th, off, 1755 so, m, optp, optlen, &opti)) 1756 m = NULL; 1757 } 1758 1759 goto drop; 1760 } 1761 1762 after_listen: 1763 /* 1764 * From here on, we're dealing with !LISTEN. 1765 */ 1766 KASSERT(tp->t_state != TCPS_LISTEN); 1767 1768 /* 1769 * Segment received on connection. 1770 * Reset idle time and keep-alive timer. 1771 */ 1772 tp->t_rcvtime = tcp_now; 1773 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1774 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1775 1776 /* 1777 * Process options. 1778 */ 1779 #ifdef TCP_SIGNATURE 1780 if (optp || (tp->t_flags & TF_SIGNATURE)) 1781 #else 1782 if (optp) 1783 #endif 1784 if (tcp_dooptions(tp, optp, optlen, th, m, off, &opti) < 0) 1785 goto drop; 1786 1787 if (TCP_SACK_ENABLED(tp)) { 1788 tcp_del_sackholes(tp, th); 1789 } 1790 1791 if (TCP_ECN_ALLOWED(tp)) { 1792 if (tiflags & TH_CWR) { 1793 tp->t_flags &= ~TF_ECN_SND_ECE; 1794 } 1795 switch (iptos & IPTOS_ECN_MASK) { 1796 case IPTOS_ECN_CE: 1797 tp->t_flags |= TF_ECN_SND_ECE; 1798 TCP_STATINC(TCP_STAT_ECN_CE); 1799 break; 1800 case IPTOS_ECN_ECT0: 1801 TCP_STATINC(TCP_STAT_ECN_ECT); 1802 break; 1803 case IPTOS_ECN_ECT1: 1804 /* XXX: ignore for now -- rpaulo */ 1805 break; 1806 } 1807 /* 1808 * Congestion experienced. 1809 * Ignore if we are already trying to recover. 1810 */ 1811 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1812 tp->t_congctl->cong_exp(tp); 1813 } 1814 1815 if (opti.ts_present && opti.ts_ecr) { 1816 /* 1817 * Calculate the RTT from the returned time stamp and the 1818 * connection's time base. If the time stamp is later than 1819 * the current time, or is extremely old, fall back to non-1323 1820 * RTT calculation. Since ts_rtt is unsigned, we can test both 1821 * at the same time. 1822 * 1823 * Note that ts_rtt is in units of slow ticks (500 1824 * ms). Since most earthbound RTTs are < 500 ms, 1825 * observed values will have large quantization noise. 1826 * Our smoothed RTT is then the fraction of observed 1827 * samples that are 1 tick instead of 0 (times 500 1828 * ms). 1829 * 1830 * ts_rtt is increased by 1 to denote a valid sample, 1831 * with 0 indicating an invalid measurement. This 1832 * extra 1 must be removed when ts_rtt is used, or 1833 * else an erroneous extra 500 ms will result. 1834 */ 1835 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1836 if (ts_rtt > TCP_PAWS_IDLE) 1837 ts_rtt = 0; 1838 } else { 1839 ts_rtt = 0; 1840 } 1841 1842 /* 1843 * Fast path: check for the two common cases of a uni-directional 1844 * data transfer. If: 1845 * o We are in the ESTABLISHED state, and 1846 * o The packet has no control flags, and 1847 * o The packet is in-sequence, and 1848 * o The window didn't change, and 1849 * o We are not retransmitting 1850 * It's a candidate. 1851 * 1852 * If the length (tlen) is zero and the ack moved forward, we're 1853 * the sender side of the transfer. Just free the data acked and 1854 * wake any higher level process that was blocked waiting for 1855 * space. 1856 * 1857 * If the length is non-zero and the ack didn't move, we're the 1858 * receiver side. If we're getting packets in-order (the reassembly 1859 * queue is empty), add the data to the socket buffer and note 1860 * that we need a delayed ack. 1861 */ 1862 if (tp->t_state == TCPS_ESTABLISHED && 1863 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1864 == TH_ACK && 1865 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1866 th->th_seq == tp->rcv_nxt && 1867 tiwin && tiwin == tp->snd_wnd && 1868 tp->snd_nxt == tp->snd_max) { 1869 1870 /* 1871 * If last ACK falls within this segment's sequence numbers, 1872 * record the timestamp. 1873 * NOTE that the test is modified according to the latest 1874 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1875 * 1876 * note that we already know 1877 * TSTMP_GEQ(opti.ts_val, tp->ts_recent) 1878 */ 1879 if (opti.ts_present && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1880 tp->ts_recent_age = tcp_now; 1881 tp->ts_recent = opti.ts_val; 1882 } 1883 1884 if (tlen == 0) { 1885 /* Ack prediction. */ 1886 if (SEQ_GT(th->th_ack, tp->snd_una) && 1887 SEQ_LEQ(th->th_ack, tp->snd_max) && 1888 tp->snd_cwnd >= tp->snd_wnd && 1889 tp->t_partialacks < 0) { 1890 /* 1891 * this is a pure ack for outstanding data. 1892 */ 1893 if (ts_rtt) 1894 tcp_xmit_timer(tp, ts_rtt - 1); 1895 else if (tp->t_rtttime && 1896 SEQ_GT(th->th_ack, tp->t_rtseq)) 1897 tcp_xmit_timer(tp, 1898 tcp_now - tp->t_rtttime); 1899 acked = th->th_ack - tp->snd_una; 1900 tcps = TCP_STAT_GETREF(); 1901 tcps[TCP_STAT_PREDACK]++; 1902 tcps[TCP_STAT_RCVACKPACK]++; 1903 tcps[TCP_STAT_RCVACKBYTE] += acked; 1904 TCP_STAT_PUTREF(); 1905 nd_hint(tp); 1906 1907 if (acked > (tp->t_lastoff - tp->t_inoff)) 1908 tp->t_lastm = NULL; 1909 sbdrop(&so->so_snd, acked); 1910 tp->t_lastoff -= acked; 1911 1912 icmp_check(tp, th, acked); 1913 1914 tp->snd_una = th->th_ack; 1915 tp->snd_fack = tp->snd_una; 1916 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1917 tp->snd_high = tp->snd_una; 1918 /* 1919 * drag snd_wl2 along so only newer 1920 * ACKs can update the window size. 1921 * also avoids the state where snd_wl2 1922 * is eventually larger than th_ack and thus 1923 * blocking the window update mechanism and 1924 * the connection gets stuck for a loooong 1925 * time in the zero sized send window state. 1926 * 1927 * see PR/kern 55567 1928 */ 1929 tp->snd_wl2 = tp->snd_una; 1930 1931 m_freem(m); 1932 1933 /* 1934 * If all outstanding data are acked, stop 1935 * retransmit timer, otherwise restart timer 1936 * using current (possibly backed-off) value. 1937 * If process is waiting for space, 1938 * wakeup/selnotify/signal. If data 1939 * are ready to send, let tcp_output 1940 * decide between more output or persist. 1941 */ 1942 if (tp->snd_una == tp->snd_max) 1943 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1944 else if (TCP_TIMER_ISARMED(tp, 1945 TCPT_PERSIST) == 0) 1946 TCP_TIMER_ARM(tp, TCPT_REXMT, 1947 tp->t_rxtcur); 1948 1949 sowwakeup(so); 1950 if (so->so_snd.sb_cc) { 1951 KERNEL_LOCK(1, NULL); 1952 (void)tcp_output(tp); 1953 KERNEL_UNLOCK_ONE(NULL); 1954 } 1955 if (tcp_saveti) 1956 m_freem(tcp_saveti); 1957 return; 1958 } 1959 } else if (th->th_ack == tp->snd_una && 1960 TAILQ_FIRST(&tp->segq) == NULL && 1961 tlen <= sbspace(&so->so_rcv)) { 1962 int newsize = 0; 1963 1964 /* 1965 * this is a pure, in-sequence data packet 1966 * with nothing on the reassembly queue and 1967 * we have enough buffer space to take it. 1968 */ 1969 tp->rcv_nxt += tlen; 1970 1971 /* 1972 * Pull rcv_up up to prevent seq wrap relative to 1973 * rcv_nxt. 1974 */ 1975 tp->rcv_up = tp->rcv_nxt; 1976 1977 /* 1978 * Pull snd_wl1 up to prevent seq wrap relative to 1979 * th_seq. 1980 */ 1981 tp->snd_wl1 = th->th_seq; 1982 1983 tcps = TCP_STAT_GETREF(); 1984 tcps[TCP_STAT_PREDDAT]++; 1985 tcps[TCP_STAT_RCVPACK]++; 1986 tcps[TCP_STAT_RCVBYTE] += tlen; 1987 TCP_STAT_PUTREF(); 1988 nd_hint(tp); 1989 /* 1990 * Automatic sizing enables the performance of large buffers 1991 * and most of the efficiency of small ones by only allocating 1992 * space when it is needed. 1993 * 1994 * On the receive side the socket buffer memory is only rarely 1995 * used to any significant extent. This allows us to be much 1996 * more aggressive in scaling the receive socket buffer. For 1997 * the case that the buffer space is actually used to a large 1998 * extent and we run out of kernel memory we can simply drop 1999 * the new segments; TCP on the sender will just retransmit it 2000 * later. Setting the buffer size too big may only consume too 2001 * much kernel memory if the application doesn't read() from 2002 * the socket or packet loss or reordering makes use of the 2003 * reassembly queue. 2004 * 2005 * The criteria to step up the receive buffer one notch are: 2006 * 1. the number of bytes received during the time it takes 2007 * one timestamp to be reflected back to us (the RTT); 2008 * 2. received bytes per RTT is within seven eighth of the 2009 * current socket buffer size; 2010 * 3. receive buffer size has not hit maximal automatic size; 2011 * 2012 * This algorithm does one step per RTT at most and only if 2013 * we receive a bulk stream w/o packet losses or reorderings. 2014 * Shrinking the buffer during idle times is not necessary as 2015 * it doesn't consume any memory when idle. 2016 * 2017 * TODO: Only step up if the application is actually serving 2018 * the buffer to better manage the socket buffer resources. 2019 */ 2020 if (tcp_do_autorcvbuf && 2021 opti.ts_ecr && 2022 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 2023 if (opti.ts_ecr > tp->rfbuf_ts && 2024 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 2025 if (tp->rfbuf_cnt > 2026 (so->so_rcv.sb_hiwat / 8 * 7) && 2027 so->so_rcv.sb_hiwat < 2028 tcp_autorcvbuf_max) { 2029 newsize = 2030 uimin(so->so_rcv.sb_hiwat + 2031 tcp_autorcvbuf_inc, 2032 tcp_autorcvbuf_max); 2033 } 2034 /* Start over with next RTT. */ 2035 tp->rfbuf_ts = 0; 2036 tp->rfbuf_cnt = 0; 2037 } else 2038 tp->rfbuf_cnt += tlen; /* add up */ 2039 } 2040 2041 /* 2042 * Drop TCP, IP headers and TCP options then add data 2043 * to socket buffer. 2044 */ 2045 if (so->so_state & SS_CANTRCVMORE) { 2046 m_freem(m); 2047 } else { 2048 /* 2049 * Set new socket buffer size. 2050 * Give up when limit is reached. 2051 */ 2052 if (newsize) 2053 if (!sbreserve(&so->so_rcv, 2054 newsize, so)) 2055 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 2056 m_adj(m, off + thlen); 2057 sbappendstream(&so->so_rcv, m); 2058 } 2059 sorwakeup(so); 2060 tcp_setup_ack(tp, th); 2061 if (tp->t_flags & TF_ACKNOW) { 2062 KERNEL_LOCK(1, NULL); 2063 (void)tcp_output(tp); 2064 KERNEL_UNLOCK_ONE(NULL); 2065 } 2066 if (tcp_saveti) 2067 m_freem(tcp_saveti); 2068 return; 2069 } 2070 } 2071 2072 /* 2073 * Compute mbuf offset to TCP data segment. 2074 */ 2075 hdroptlen = off + thlen; 2076 2077 /* 2078 * Calculate amount of space in receive window. Receive window is 2079 * amount of space in rcv queue, but not less than advertised 2080 * window. 2081 */ 2082 { 2083 int win; 2084 win = sbspace(&so->so_rcv); 2085 if (win < 0) 2086 win = 0; 2087 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 2088 } 2089 2090 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 2091 tp->rfbuf_ts = 0; 2092 tp->rfbuf_cnt = 0; 2093 2094 switch (tp->t_state) { 2095 /* 2096 * If the state is SYN_SENT: 2097 * if seg contains an ACK, but not for our SYN, drop the input. 2098 * if seg contains a RST, then drop the connection. 2099 * if seg does not contain SYN, then drop it. 2100 * Otherwise this is an acceptable SYN segment 2101 * initialize tp->rcv_nxt and tp->irs 2102 * if seg contains ack then advance tp->snd_una 2103 * if seg contains a ECE and ECN support is enabled, the stream 2104 * is ECN capable. 2105 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2106 * arrange for segment to be acked (eventually) 2107 * continue processing rest of data/controls, beginning with URG 2108 */ 2109 case TCPS_SYN_SENT: 2110 if ((tiflags & TH_ACK) && 2111 (SEQ_LEQ(th->th_ack, tp->iss) || 2112 SEQ_GT(th->th_ack, tp->snd_max))) 2113 goto dropwithreset; 2114 if (tiflags & TH_RST) { 2115 if (tiflags & TH_ACK) 2116 tp = tcp_drop(tp, ECONNREFUSED); 2117 goto drop; 2118 } 2119 if ((tiflags & TH_SYN) == 0) 2120 goto drop; 2121 if (tiflags & TH_ACK) { 2122 tp->snd_una = th->th_ack; 2123 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2124 tp->snd_nxt = tp->snd_una; 2125 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2126 tp->snd_high = tp->snd_una; 2127 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2128 2129 if ((tiflags & TH_ECE) && tcp_do_ecn) { 2130 tp->t_flags |= TF_ECN_PERMIT; 2131 TCP_STATINC(TCP_STAT_ECN_SHS); 2132 } 2133 } 2134 tp->irs = th->th_seq; 2135 tcp_rcvseqinit(tp); 2136 tp->t_flags |= TF_ACKNOW; 2137 tcp_mss_from_peer(tp, opti.maxseg); 2138 2139 /* 2140 * Initialize the initial congestion window. If we 2141 * had to retransmit the SYN, we must initialize cwnd 2142 * to 1 segment (i.e. the Loss Window). 2143 */ 2144 if (tp->t_flags & TF_SYN_REXMT) 2145 tp->snd_cwnd = tp->t_peermss; 2146 else { 2147 int ss = tcp_init_win; 2148 if (inp != NULL && in_localaddr(inp->inp_faddr)) 2149 ss = tcp_init_win_local; 2150 #ifdef INET6 2151 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 2152 ss = tcp_init_win_local; 2153 #endif 2154 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 2155 } 2156 2157 tcp_rmx_rtt(tp); 2158 if (tiflags & TH_ACK) { 2159 TCP_STATINC(TCP_STAT_CONNECTS); 2160 /* 2161 * move tcp_established before soisconnected 2162 * because upcall handler can drive tcp_output 2163 * functionality. 2164 * XXX we might call soisconnected at the end of 2165 * all processing 2166 */ 2167 tcp_established(tp); 2168 soisconnected(so); 2169 /* Do window scaling on this connection? */ 2170 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2171 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2172 tp->snd_scale = tp->requested_s_scale; 2173 tp->rcv_scale = tp->request_r_scale; 2174 } 2175 TCP_REASS_LOCK(tp); 2176 (void)tcp_reass(tp, NULL, NULL, tlen); 2177 /* 2178 * if we didn't have to retransmit the SYN, 2179 * use its rtt as our initial srtt & rtt var. 2180 */ 2181 if (tp->t_rtttime) 2182 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2183 } else { 2184 tp->t_state = TCPS_SYN_RECEIVED; 2185 } 2186 2187 /* 2188 * Advance th->th_seq to correspond to first data byte. 2189 * If data, trim to stay within window, 2190 * dropping FIN if necessary. 2191 */ 2192 th->th_seq++; 2193 if (tlen > tp->rcv_wnd) { 2194 todrop = tlen - tp->rcv_wnd; 2195 m_adj(m, -todrop); 2196 tlen = tp->rcv_wnd; 2197 tiflags &= ~TH_FIN; 2198 tcps = TCP_STAT_GETREF(); 2199 tcps[TCP_STAT_RCVPACKAFTERWIN]++; 2200 tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop; 2201 TCP_STAT_PUTREF(); 2202 } 2203 tp->snd_wl1 = th->th_seq - 1; 2204 tp->rcv_up = th->th_seq; 2205 goto step6; 2206 2207 /* 2208 * If the state is SYN_RECEIVED: 2209 * If seg contains an ACK, but not for our SYN, drop the input 2210 * and generate an RST. See page 36, rfc793 2211 */ 2212 case TCPS_SYN_RECEIVED: 2213 if ((tiflags & TH_ACK) && 2214 (SEQ_LEQ(th->th_ack, tp->iss) || 2215 SEQ_GT(th->th_ack, tp->snd_max))) 2216 goto dropwithreset; 2217 break; 2218 } 2219 2220 /* 2221 * From here on, we're dealing with !LISTEN and !SYN_SENT. 2222 */ 2223 KASSERT(tp->t_state != TCPS_LISTEN && 2224 tp->t_state != TCPS_SYN_SENT); 2225 2226 /* 2227 * RFC1323 PAWS: if we have a timestamp reply on this segment and 2228 * it's less than ts_recent, drop it. 2229 */ 2230 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2231 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2232 /* Check to see if ts_recent is over 24 days old. */ 2233 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2234 /* 2235 * Invalidate ts_recent. If this segment updates 2236 * ts_recent, the age will be reset later and ts_recent 2237 * will get a valid value. If it does not, setting 2238 * ts_recent to zero will at least satisfy the 2239 * requirement that zero be placed in the timestamp 2240 * echo reply when ts_recent isn't valid. The 2241 * age isn't reset until we get a valid ts_recent 2242 * because we don't want out-of-order segments to be 2243 * dropped when ts_recent is old. 2244 */ 2245 tp->ts_recent = 0; 2246 } else { 2247 tcps = TCP_STAT_GETREF(); 2248 tcps[TCP_STAT_RCVDUPPACK]++; 2249 tcps[TCP_STAT_RCVDUPBYTE] += tlen; 2250 tcps[TCP_STAT_PAWSDROP]++; 2251 TCP_STAT_PUTREF(); 2252 tcp_new_dsack(tp, th->th_seq, tlen); 2253 goto dropafterack; 2254 } 2255 } 2256 2257 /* 2258 * Check that at least some bytes of the segment are within the 2259 * receive window. If segment begins before rcv_nxt, drop leading 2260 * data (and SYN); if nothing left, just ack. 2261 */ 2262 todrop = tp->rcv_nxt - th->th_seq; 2263 dupseg = false; 2264 if (todrop > 0) { 2265 if (tiflags & TH_SYN) { 2266 tiflags &= ~TH_SYN; 2267 th->th_seq++; 2268 tcp_urp_drop(th, 1, &tiflags); 2269 todrop--; 2270 } 2271 if (todrop > tlen || 2272 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2273 /* 2274 * Any valid FIN or RST must be to the left of the 2275 * window. At this point the FIN or RST must be a 2276 * duplicate or out of sequence; drop it. 2277 */ 2278 if (tiflags & TH_RST) 2279 goto drop; 2280 tiflags &= ~(TH_FIN|TH_RST); 2281 2282 /* 2283 * Send an ACK to resynchronize and drop any data. 2284 * But keep on processing for RST or ACK. 2285 */ 2286 tp->t_flags |= TF_ACKNOW; 2287 todrop = tlen; 2288 dupseg = true; 2289 tcps = TCP_STAT_GETREF(); 2290 tcps[TCP_STAT_RCVDUPPACK]++; 2291 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 2292 TCP_STAT_PUTREF(); 2293 } else if ((tiflags & TH_RST) && th->th_seq != tp->rcv_nxt) { 2294 /* 2295 * Test for reset before adjusting the sequence 2296 * number for overlapping data. 2297 */ 2298 goto dropafterack_ratelim; 2299 } else { 2300 tcps = TCP_STAT_GETREF(); 2301 tcps[TCP_STAT_RCVPARTDUPPACK]++; 2302 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 2303 TCP_STAT_PUTREF(); 2304 } 2305 tcp_new_dsack(tp, th->th_seq, todrop); 2306 hdroptlen += todrop; /* drop from head afterwards (m_adj) */ 2307 th->th_seq += todrop; 2308 tlen -= todrop; 2309 tcp_urp_drop(th, todrop, &tiflags); 2310 } 2311 2312 /* 2313 * If new data is received on a connection after the user processes 2314 * are gone, then RST the other end. 2315 */ 2316 if ((so->so_state & SS_NOFDREF) && 2317 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2318 tp = tcp_close(tp); 2319 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 2320 goto dropwithreset; 2321 } 2322 2323 /* 2324 * If the segment ends after the window, drop trailing data (and 2325 * PUSH and FIN); if nothing left, just ACK. 2326 */ 2327 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2328 if (todrop > 0) { 2329 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 2330 if (todrop >= tlen) { 2331 /* 2332 * The segment actually starts after the window. 2333 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2334 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2335 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2336 */ 2337 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 2338 2339 /* 2340 * If a new connection request is received while in 2341 * TIME_WAIT, drop the old connection and start over 2342 * if the sequence numbers are above the previous 2343 * ones. 2344 * 2345 * NOTE: We need to put the header fields back into 2346 * network order. 2347 */ 2348 if ((tiflags & TH_SYN) && 2349 tp->t_state == TCPS_TIME_WAIT && 2350 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2351 tp = tcp_close(tp); 2352 tcp_fields_to_net(th); 2353 m_freem(tcp_saveti); 2354 tcp_saveti = NULL; 2355 goto findpcb; 2356 } 2357 2358 /* 2359 * If window is closed can only take segments at 2360 * window edge, and have to drop data and PUSH from 2361 * incoming segments. Continue processing, but 2362 * remember to ack. Otherwise, drop segment 2363 * and (if not RST) ack. 2364 */ 2365 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2366 KASSERT(todrop == tlen); 2367 tp->t_flags |= TF_ACKNOW; 2368 TCP_STATINC(TCP_STAT_RCVWINPROBE); 2369 } else { 2370 goto dropafterack; 2371 } 2372 } else { 2373 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 2374 } 2375 m_adj(m, -todrop); 2376 tlen -= todrop; 2377 tiflags &= ~(TH_PUSH|TH_FIN); 2378 } 2379 2380 /* 2381 * If last ACK falls within this segment's sequence numbers, 2382 * record the timestamp. 2383 * NOTE: 2384 * 1) That the test incorporates suggestions from the latest 2385 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2386 * 2) That updating only on newer timestamps interferes with 2387 * our earlier PAWS tests, so this check should be solely 2388 * predicated on the sequence space of this segment. 2389 * 3) That we modify the segment boundary check to be 2390 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2391 * instead of RFC1323's 2392 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2393 * This modified check allows us to overcome RFC1323's 2394 * limitations as described in Stevens TCP/IP Illustrated 2395 * Vol. 2 p.869. In such cases, we can still calculate the 2396 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2397 */ 2398 if (opti.ts_present && 2399 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2400 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2401 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2402 tp->ts_recent_age = tcp_now; 2403 tp->ts_recent = opti.ts_val; 2404 } 2405 2406 /* 2407 * If the RST bit is set examine the state: 2408 * RECEIVED state: 2409 * If passive open, return to LISTEN state. 2410 * If active open, inform user that connection was refused. 2411 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT states: 2412 * Inform user that connection was reset, and close tcb. 2413 * CLOSING, LAST_ACK, TIME_WAIT states: 2414 * Close the tcb. 2415 */ 2416 if (tiflags & TH_RST) { 2417 if (th->th_seq != tp->rcv_nxt) 2418 goto dropafterack_ratelim; 2419 2420 switch (tp->t_state) { 2421 case TCPS_SYN_RECEIVED: 2422 so->so_error = ECONNREFUSED; 2423 goto close; 2424 2425 case TCPS_ESTABLISHED: 2426 case TCPS_FIN_WAIT_1: 2427 case TCPS_FIN_WAIT_2: 2428 case TCPS_CLOSE_WAIT: 2429 so->so_error = ECONNRESET; 2430 close: 2431 tp->t_state = TCPS_CLOSED; 2432 TCP_STATINC(TCP_STAT_DROPS); 2433 tp = tcp_close(tp); 2434 goto drop; 2435 2436 case TCPS_CLOSING: 2437 case TCPS_LAST_ACK: 2438 case TCPS_TIME_WAIT: 2439 tp = tcp_close(tp); 2440 goto drop; 2441 } 2442 } 2443 2444 /* 2445 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2446 * we must be in a synchronized state. RFC793 states (under Reset 2447 * Generation) that any unacceptable segment (an out-of-order SYN 2448 * qualifies) received in a synchronized state must elicit only an 2449 * empty acknowledgment segment ... and the connection remains in 2450 * the same state. 2451 */ 2452 if (tiflags & TH_SYN) { 2453 if (tp->rcv_nxt == th->th_seq) { 2454 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2455 TH_ACK); 2456 if (tcp_saveti) 2457 m_freem(tcp_saveti); 2458 return; 2459 } 2460 2461 goto dropafterack_ratelim; 2462 } 2463 2464 /* 2465 * If the ACK bit is off we drop the segment and return. 2466 */ 2467 if ((tiflags & TH_ACK) == 0) { 2468 if (tp->t_flags & TF_ACKNOW) 2469 goto dropafterack; 2470 goto drop; 2471 } 2472 2473 /* 2474 * From here on, we're doing ACK processing. 2475 */ 2476 2477 switch (tp->t_state) { 2478 /* 2479 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2480 * ESTABLISHED state and continue processing, otherwise 2481 * send an RST. 2482 */ 2483 case TCPS_SYN_RECEIVED: 2484 if (SEQ_GT(tp->snd_una, th->th_ack) || 2485 SEQ_GT(th->th_ack, tp->snd_max)) 2486 goto dropwithreset; 2487 TCP_STATINC(TCP_STAT_CONNECTS); 2488 soisconnected(so); 2489 tcp_established(tp); 2490 /* Do window scaling? */ 2491 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2492 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2493 tp->snd_scale = tp->requested_s_scale; 2494 tp->rcv_scale = tp->request_r_scale; 2495 } 2496 TCP_REASS_LOCK(tp); 2497 (void)tcp_reass(tp, NULL, NULL, tlen); 2498 tp->snd_wl1 = th->th_seq - 1; 2499 /* FALLTHROUGH */ 2500 2501 /* 2502 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2503 * ACKs. If the ack is in the range 2504 * tp->snd_una < th->th_ack <= tp->snd_max 2505 * then advance tp->snd_una to th->th_ack and drop 2506 * data from the retransmission queue. If this ACK reflects 2507 * more up to date window information we update our window information. 2508 */ 2509 case TCPS_ESTABLISHED: 2510 case TCPS_FIN_WAIT_1: 2511 case TCPS_FIN_WAIT_2: 2512 case TCPS_CLOSE_WAIT: 2513 case TCPS_CLOSING: 2514 case TCPS_LAST_ACK: 2515 case TCPS_TIME_WAIT: 2516 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2517 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2518 TCP_STATINC(TCP_STAT_RCVDUPACK); 2519 /* 2520 * If we have outstanding data (other than 2521 * a window probe), this is a completely 2522 * duplicate ack (ie, window info didn't 2523 * change), the ack is the biggest we've 2524 * seen and we've seen exactly our rexmt 2525 * threshold of them, assume a packet 2526 * has been dropped and retransmit it. 2527 * Kludge snd_nxt & the congestion 2528 * window so we send only this one 2529 * packet. 2530 */ 2531 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2532 th->th_ack != tp->snd_una) 2533 tp->t_dupacks = 0; 2534 else if (tp->t_partialacks < 0 && 2535 (++tp->t_dupacks == tcprexmtthresh || 2536 TCP_FACK_FASTRECOV(tp))) { 2537 /* 2538 * Do the fast retransmit, and adjust 2539 * congestion control parameters. 2540 */ 2541 if (tp->t_congctl->fast_retransmit(tp, th)) { 2542 /* False fast retransmit */ 2543 break; 2544 } 2545 goto drop; 2546 } else if (tp->t_dupacks > tcprexmtthresh) { 2547 tp->snd_cwnd += tp->t_segsz; 2548 KERNEL_LOCK(1, NULL); 2549 (void)tcp_output(tp); 2550 KERNEL_UNLOCK_ONE(NULL); 2551 goto drop; 2552 } 2553 } else { 2554 /* 2555 * If the ack appears to be very old, only 2556 * allow data that is in-sequence. This 2557 * makes it somewhat more difficult to insert 2558 * forged data by guessing sequence numbers. 2559 * Sent an ack to try to update the send 2560 * sequence number on the other side. 2561 */ 2562 if (tlen && th->th_seq != tp->rcv_nxt && 2563 SEQ_LT(th->th_ack, 2564 tp->snd_una - tp->max_sndwnd)) 2565 goto dropafterack; 2566 } 2567 break; 2568 } 2569 /* 2570 * If the congestion window was inflated to account 2571 * for the other side's cached packets, retract it. 2572 */ 2573 tp->t_congctl->fast_retransmit_newack(tp, th); 2574 2575 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2576 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2577 goto dropafterack; 2578 } 2579 acked = th->th_ack - tp->snd_una; 2580 tcps = TCP_STAT_GETREF(); 2581 tcps[TCP_STAT_RCVACKPACK]++; 2582 tcps[TCP_STAT_RCVACKBYTE] += acked; 2583 TCP_STAT_PUTREF(); 2584 2585 /* 2586 * If we have a timestamp reply, update smoothed 2587 * round trip time. If no timestamp is present but 2588 * transmit timer is running and timed sequence 2589 * number was acked, update smoothed round trip time. 2590 * Since we now have an rtt measurement, cancel the 2591 * timer backoff (cf., Phil Karn's retransmit alg.). 2592 * Recompute the initial retransmit timer. 2593 */ 2594 if (ts_rtt) 2595 tcp_xmit_timer(tp, ts_rtt - 1); 2596 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2597 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2598 2599 /* 2600 * If all outstanding data is acked, stop retransmit 2601 * timer and remember to restart (more output or persist). 2602 * If there is more data to be acked, restart retransmit 2603 * timer, using current (possibly backed-off) value. 2604 */ 2605 if (th->th_ack == tp->snd_max) { 2606 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2607 needoutput = 1; 2608 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2609 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2610 2611 /* 2612 * New data has been acked, adjust the congestion window. 2613 */ 2614 tp->t_congctl->newack(tp, th); 2615 2616 nd_hint(tp); 2617 if (acked > so->so_snd.sb_cc) { 2618 tp->snd_wnd -= so->so_snd.sb_cc; 2619 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2620 ourfinisacked = 1; 2621 } else { 2622 if (acked > (tp->t_lastoff - tp->t_inoff)) 2623 tp->t_lastm = NULL; 2624 sbdrop(&so->so_snd, acked); 2625 tp->t_lastoff -= acked; 2626 if (tp->snd_wnd > acked) 2627 tp->snd_wnd -= acked; 2628 else 2629 tp->snd_wnd = 0; 2630 ourfinisacked = 0; 2631 } 2632 sowwakeup(so); 2633 2634 icmp_check(tp, th, acked); 2635 2636 tp->snd_una = th->th_ack; 2637 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2638 tp->snd_fack = tp->snd_una; 2639 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2640 tp->snd_nxt = tp->snd_una; 2641 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2642 tp->snd_high = tp->snd_una; 2643 2644 switch (tp->t_state) { 2645 2646 /* 2647 * In FIN_WAIT_1 STATE in addition to the processing 2648 * for the ESTABLISHED state if our FIN is now acknowledged 2649 * then enter FIN_WAIT_2. 2650 */ 2651 case TCPS_FIN_WAIT_1: 2652 if (ourfinisacked) { 2653 /* 2654 * If we can't receive any more 2655 * data, then closing user can proceed. 2656 * Starting the timer is contrary to the 2657 * specification, but if we don't get a FIN 2658 * we'll hang forever. 2659 */ 2660 if (so->so_state & SS_CANTRCVMORE) { 2661 soisdisconnected(so); 2662 if (tp->t_maxidle > 0) 2663 TCP_TIMER_ARM(tp, TCPT_2MSL, 2664 tp->t_maxidle); 2665 } 2666 tp->t_state = TCPS_FIN_WAIT_2; 2667 } 2668 break; 2669 2670 /* 2671 * In CLOSING STATE in addition to the processing for 2672 * the ESTABLISHED state if the ACK acknowledges our FIN 2673 * then enter the TIME-WAIT state, otherwise ignore 2674 * the segment. 2675 */ 2676 case TCPS_CLOSING: 2677 if (ourfinisacked) { 2678 tp->t_state = TCPS_TIME_WAIT; 2679 tcp_canceltimers(tp); 2680 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2681 soisdisconnected(so); 2682 } 2683 break; 2684 2685 /* 2686 * In LAST_ACK, we may still be waiting for data to drain 2687 * and/or to be acked, as well as for the ack of our FIN. 2688 * If our FIN is now acknowledged, delete the TCB, 2689 * enter the closed state and return. 2690 */ 2691 case TCPS_LAST_ACK: 2692 if (ourfinisacked) { 2693 tp = tcp_close(tp); 2694 goto drop; 2695 } 2696 break; 2697 2698 /* 2699 * In TIME_WAIT state the only thing that should arrive 2700 * is a retransmission of the remote FIN. Acknowledge 2701 * it and restart the finack timer. 2702 */ 2703 case TCPS_TIME_WAIT: 2704 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2705 goto dropafterack; 2706 } 2707 } 2708 2709 step6: 2710 /* 2711 * Update window information. 2712 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2713 */ 2714 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2715 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2716 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2717 /* keep track of pure window updates */ 2718 if (tlen == 0 && 2719 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2720 TCP_STATINC(TCP_STAT_RCVWINUPD); 2721 tp->snd_wnd = tiwin; 2722 tp->snd_wl1 = th->th_seq; 2723 tp->snd_wl2 = th->th_ack; 2724 if (tp->snd_wnd > tp->max_sndwnd) 2725 tp->max_sndwnd = tp->snd_wnd; 2726 needoutput = 1; 2727 } 2728 2729 /* 2730 * Process segments with URG. 2731 */ 2732 if ((tiflags & TH_URG) && th->th_urp && 2733 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2734 /* 2735 * This is a kludge, but if we receive and accept 2736 * random urgent pointers, we'll crash in 2737 * soreceive. It's hard to imagine someone 2738 * actually wanting to send this much urgent data. 2739 */ 2740 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2741 th->th_urp = 0; /* XXX */ 2742 tiflags &= ~TH_URG; /* XXX */ 2743 goto dodata; /* XXX */ 2744 } 2745 2746 /* 2747 * If this segment advances the known urgent pointer, 2748 * then mark the data stream. This should not happen 2749 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2750 * a FIN has been received from the remote side. 2751 * In these states we ignore the URG. 2752 * 2753 * According to RFC961 (Assigned Protocols), 2754 * the urgent pointer points to the last octet 2755 * of urgent data. We continue, however, 2756 * to consider it to indicate the first octet 2757 * of data past the urgent section as the original 2758 * spec states (in one of two places). 2759 */ 2760 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2761 tp->rcv_up = th->th_seq + th->th_urp; 2762 so->so_oobmark = so->so_rcv.sb_cc + 2763 (tp->rcv_up - tp->rcv_nxt) - 1; 2764 if (so->so_oobmark == 0) 2765 so->so_state |= SS_RCVATMARK; 2766 sohasoutofband(so); 2767 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2768 } 2769 2770 /* 2771 * Remove out of band data so doesn't get presented to user. 2772 * This can happen independent of advancing the URG pointer, 2773 * but if two URG's are pending at once, some out-of-band 2774 * data may creep in... ick. 2775 */ 2776 if (th->th_urp <= (u_int16_t)tlen && 2777 (so->so_options & SO_OOBINLINE) == 0) 2778 tcp_pulloutofband(so, th, m, hdroptlen); 2779 } else { 2780 /* 2781 * If no out of band data is expected, 2782 * pull receive urgent pointer along 2783 * with the receive window. 2784 */ 2785 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2786 tp->rcv_up = tp->rcv_nxt; 2787 } 2788 dodata: 2789 2790 /* 2791 * Process the segment text, merging it into the TCP sequencing queue, 2792 * and arranging for acknowledgement of receipt if necessary. 2793 * This process logically involves adjusting tp->rcv_wnd as data 2794 * is presented to the user (this happens in tcp_usrreq.c, 2795 * tcp_rcvd()). If a FIN has already been received on this 2796 * connection then we just ignore the text. 2797 */ 2798 if ((tlen || (tiflags & TH_FIN)) && 2799 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2800 /* 2801 * Handle the common case: 2802 * o Segment is the next to be received, and 2803 * o The queue is empty, and 2804 * o The connection is established 2805 * In this case, we avoid calling tcp_reass. 2806 * 2807 * tcp_setup_ack: set DELACK for segments received in order, 2808 * but ack immediately when segments are out of order (so that 2809 * fast retransmit can work). 2810 */ 2811 TCP_REASS_LOCK(tp); 2812 if (th->th_seq == tp->rcv_nxt && 2813 TAILQ_FIRST(&tp->segq) == NULL && 2814 tp->t_state == TCPS_ESTABLISHED) { 2815 tcp_setup_ack(tp, th); 2816 tp->rcv_nxt += tlen; 2817 tiflags = th->th_flags & TH_FIN; 2818 tcps = TCP_STAT_GETREF(); 2819 tcps[TCP_STAT_RCVPACK]++; 2820 tcps[TCP_STAT_RCVBYTE] += tlen; 2821 TCP_STAT_PUTREF(); 2822 nd_hint(tp); 2823 if (so->so_state & SS_CANTRCVMORE) { 2824 m_freem(m); 2825 } else { 2826 m_adj(m, hdroptlen); 2827 sbappendstream(&(so)->so_rcv, m); 2828 } 2829 TCP_REASS_UNLOCK(tp); 2830 sorwakeup(so); 2831 } else { 2832 m_adj(m, hdroptlen); 2833 tiflags = tcp_reass(tp, th, m, tlen); 2834 tp->t_flags |= TF_ACKNOW; 2835 } 2836 2837 /* 2838 * Note the amount of data that peer has sent into 2839 * our window, in order to estimate the sender's 2840 * buffer size. 2841 */ 2842 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2843 } else { 2844 m_freem(m); 2845 m = NULL; 2846 tiflags &= ~TH_FIN; 2847 } 2848 2849 /* 2850 * If FIN is received ACK the FIN and let the user know 2851 * that the connection is closing. Ignore a FIN received before 2852 * the connection is fully established. 2853 */ 2854 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2855 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2856 socantrcvmore(so); 2857 tp->t_flags |= TF_ACKNOW; 2858 tp->rcv_nxt++; 2859 } 2860 switch (tp->t_state) { 2861 2862 /* 2863 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2864 */ 2865 case TCPS_ESTABLISHED: 2866 tp->t_state = TCPS_CLOSE_WAIT; 2867 break; 2868 2869 /* 2870 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2871 * enter the CLOSING state. 2872 */ 2873 case TCPS_FIN_WAIT_1: 2874 tp->t_state = TCPS_CLOSING; 2875 break; 2876 2877 /* 2878 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2879 * starting the time-wait timer, turning off the other 2880 * standard timers. 2881 */ 2882 case TCPS_FIN_WAIT_2: 2883 tp->t_state = TCPS_TIME_WAIT; 2884 tcp_canceltimers(tp); 2885 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2886 soisdisconnected(so); 2887 break; 2888 2889 /* 2890 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2891 */ 2892 case TCPS_TIME_WAIT: 2893 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2894 break; 2895 } 2896 } 2897 #ifdef TCP_DEBUG 2898 if (so->so_options & SO_DEBUG) 2899 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2900 #endif 2901 2902 /* 2903 * Return any desired output. 2904 */ 2905 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2906 KERNEL_LOCK(1, NULL); 2907 (void)tcp_output(tp); 2908 KERNEL_UNLOCK_ONE(NULL); 2909 } 2910 if (tcp_saveti) 2911 m_freem(tcp_saveti); 2912 2913 if (tp->t_state == TCPS_TIME_WAIT 2914 && (so->so_state & SS_NOFDREF) 2915 && (tp->t_inpcb || af != AF_INET) 2916 && (tp->t_in6pcb || af != AF_INET6) 2917 && ((af == AF_INET ? tcp4_vtw_enable : tcp6_vtw_enable) & 1) != 0 2918 && TAILQ_EMPTY(&tp->segq) 2919 && vtw_add(af, tp)) { 2920 ; 2921 } 2922 return; 2923 2924 badsyn: 2925 /* 2926 * Received a bad SYN. Increment counters and dropwithreset. 2927 */ 2928 TCP_STATINC(TCP_STAT_BADSYN); 2929 tp = NULL; 2930 goto dropwithreset; 2931 2932 dropafterack: 2933 /* 2934 * Generate an ACK dropping incoming segment if it occupies 2935 * sequence space, where the ACK reflects our state. 2936 */ 2937 if (tiflags & TH_RST) 2938 goto drop; 2939 goto dropafterack2; 2940 2941 dropafterack_ratelim: 2942 /* 2943 * We may want to rate-limit ACKs against SYN/RST attack. 2944 */ 2945 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2946 tcp_ackdrop_ppslim) == 0) { 2947 /* XXX stat */ 2948 goto drop; 2949 } 2950 2951 dropafterack2: 2952 m_freem(m); 2953 tp->t_flags |= TF_ACKNOW; 2954 KERNEL_LOCK(1, NULL); 2955 (void)tcp_output(tp); 2956 KERNEL_UNLOCK_ONE(NULL); 2957 if (tcp_saveti) 2958 m_freem(tcp_saveti); 2959 return; 2960 2961 dropwithreset_ratelim: 2962 /* 2963 * We may want to rate-limit RSTs in certain situations, 2964 * particularly if we are sending an RST in response to 2965 * an attempt to connect to or otherwise communicate with 2966 * a port for which we have no socket. 2967 */ 2968 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2969 tcp_rst_ppslim) == 0) { 2970 /* XXX stat */ 2971 goto drop; 2972 } 2973 2974 dropwithreset: 2975 /* 2976 * Generate a RST, dropping incoming segment. 2977 * Make ACK acceptable to originator of segment. 2978 */ 2979 if (tiflags & TH_RST) 2980 goto drop; 2981 if (tiflags & TH_ACK) { 2982 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2983 } else { 2984 if (tiflags & TH_SYN) 2985 tlen++; 2986 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2987 TH_RST|TH_ACK); 2988 } 2989 if (tcp_saveti) 2990 m_freem(tcp_saveti); 2991 return; 2992 2993 badcsum: 2994 drop: 2995 /* 2996 * Drop space held by incoming segment and return. 2997 */ 2998 if (tp) { 2999 if (tp->t_inpcb) 3000 so = tp->t_inpcb->inp_socket; 3001 #ifdef INET6 3002 else if (tp->t_in6pcb) 3003 so = tp->t_in6pcb->in6p_socket; 3004 #endif 3005 else 3006 so = NULL; 3007 #ifdef TCP_DEBUG 3008 if (so && (so->so_options & SO_DEBUG) != 0) 3009 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 3010 #endif 3011 } 3012 if (tcp_saveti) 3013 m_freem(tcp_saveti); 3014 m_freem(m); 3015 return; 3016 } 3017 3018 #ifdef TCP_SIGNATURE 3019 int 3020 tcp_signature_apply(void *fstate, void *data, u_int len) 3021 { 3022 3023 MD5Update(fstate, (u_char *)data, len); 3024 return (0); 3025 } 3026 3027 struct secasvar * 3028 tcp_signature_getsav(struct mbuf *m) 3029 { 3030 struct ip *ip; 3031 struct ip6_hdr *ip6; 3032 3033 ip = mtod(m, struct ip *); 3034 switch (ip->ip_v) { 3035 case 4: 3036 ip = mtod(m, struct ip *); 3037 ip6 = NULL; 3038 break; 3039 case 6: 3040 ip = NULL; 3041 ip6 = mtod(m, struct ip6_hdr *); 3042 break; 3043 default: 3044 return (NULL); 3045 } 3046 3047 #ifdef IPSEC 3048 union sockaddr_union dst; 3049 3050 /* Extract the destination from the IP header in the mbuf. */ 3051 memset(&dst, 0, sizeof(union sockaddr_union)); 3052 if (ip != NULL) { 3053 dst.sa.sa_len = sizeof(struct sockaddr_in); 3054 dst.sa.sa_family = AF_INET; 3055 dst.sin.sin_addr = ip->ip_dst; 3056 } else { 3057 dst.sa.sa_len = sizeof(struct sockaddr_in6); 3058 dst.sa.sa_family = AF_INET6; 3059 dst.sin6.sin6_addr = ip6->ip6_dst; 3060 } 3061 3062 /* 3063 * Look up an SADB entry which matches the address of the peer. 3064 */ 3065 return KEY_LOOKUP_SA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI), 0, 0); 3066 #else 3067 return NULL; 3068 #endif 3069 } 3070 3071 int 3072 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 3073 struct secasvar *sav, char *sig) 3074 { 3075 MD5_CTX ctx; 3076 struct ip *ip; 3077 struct ipovly *ipovly; 3078 #ifdef INET6 3079 struct ip6_hdr *ip6; 3080 struct ip6_hdr_pseudo ip6pseudo; 3081 #endif 3082 struct ippseudo ippseudo; 3083 struct tcphdr th0; 3084 int l, tcphdrlen; 3085 3086 if (sav == NULL) 3087 return (-1); 3088 3089 tcphdrlen = th->th_off * 4; 3090 3091 switch (mtod(m, struct ip *)->ip_v) { 3092 case 4: 3093 MD5Init(&ctx); 3094 ip = mtod(m, struct ip *); 3095 memset(&ippseudo, 0, sizeof(ippseudo)); 3096 ipovly = (struct ipovly *)ip; 3097 ippseudo.ippseudo_src = ipovly->ih_src; 3098 ippseudo.ippseudo_dst = ipovly->ih_dst; 3099 ippseudo.ippseudo_pad = 0; 3100 ippseudo.ippseudo_p = IPPROTO_TCP; 3101 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 3102 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 3103 break; 3104 #if INET6 3105 case 6: 3106 MD5Init(&ctx); 3107 ip6 = mtod(m, struct ip6_hdr *); 3108 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 3109 ip6pseudo.ip6ph_src = ip6->ip6_src; 3110 in6_clearscope(&ip6pseudo.ip6ph_src); 3111 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 3112 in6_clearscope(&ip6pseudo.ip6ph_dst); 3113 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 3114 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 3115 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 3116 break; 3117 #endif 3118 default: 3119 return (-1); 3120 } 3121 3122 th0 = *th; 3123 th0.th_sum = 0; 3124 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 3125 3126 l = m->m_pkthdr.len - thoff - tcphdrlen; 3127 if (l > 0) 3128 m_apply(m, thoff + tcphdrlen, 3129 m->m_pkthdr.len - thoff - tcphdrlen, 3130 tcp_signature_apply, &ctx); 3131 3132 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 3133 MD5Final(sig, &ctx); 3134 3135 return (0); 3136 } 3137 #endif 3138 3139 /* 3140 * Parse and process tcp options. 3141 * 3142 * Returns -1 if this segment should be dropped. (eg. wrong signature) 3143 * Otherwise returns 0. 3144 */ 3145 int 3146 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, struct tcphdr *th, 3147 struct mbuf *m, int toff, struct tcp_opt_info *oi) 3148 { 3149 u_int16_t mss; 3150 int opt, optlen = 0; 3151 #ifdef TCP_SIGNATURE 3152 void *sigp = NULL; 3153 char sigbuf[TCP_SIGLEN]; 3154 struct secasvar *sav = NULL; 3155 #endif 3156 3157 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 3158 opt = cp[0]; 3159 if (opt == TCPOPT_EOL) 3160 break; 3161 if (opt == TCPOPT_NOP) 3162 optlen = 1; 3163 else { 3164 if (cnt < 2) 3165 break; 3166 optlen = cp[1]; 3167 if (optlen < 2 || optlen > cnt) 3168 break; 3169 } 3170 switch (opt) { 3171 3172 default: 3173 continue; 3174 3175 case TCPOPT_MAXSEG: 3176 if (optlen != TCPOLEN_MAXSEG) 3177 continue; 3178 if (!(th->th_flags & TH_SYN)) 3179 continue; 3180 if (TCPS_HAVERCVDSYN(tp->t_state)) 3181 continue; 3182 memcpy(&mss, cp + 2, sizeof(mss)); 3183 oi->maxseg = ntohs(mss); 3184 break; 3185 3186 case TCPOPT_WINDOW: 3187 if (optlen != TCPOLEN_WINDOW) 3188 continue; 3189 if (!(th->th_flags & TH_SYN)) 3190 continue; 3191 if (TCPS_HAVERCVDSYN(tp->t_state)) 3192 continue; 3193 tp->t_flags |= TF_RCVD_SCALE; 3194 tp->requested_s_scale = cp[2]; 3195 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3196 char buf[INET6_ADDRSTRLEN]; 3197 struct ip *ip = mtod(m, struct ip *); 3198 #ifdef INET6 3199 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 3200 #endif 3201 3202 switch (ip->ip_v) { 3203 case 4: 3204 in_print(buf, sizeof(buf), 3205 &ip->ip_src); 3206 break; 3207 #ifdef INET6 3208 case 6: 3209 in6_print(buf, sizeof(buf), 3210 &ip6->ip6_src); 3211 break; 3212 #endif 3213 default: 3214 strlcpy(buf, "(unknown)", sizeof(buf)); 3215 break; 3216 } 3217 3218 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3219 "assuming %d\n", 3220 tp->requested_s_scale, buf, 3221 TCP_MAX_WINSHIFT); 3222 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3223 } 3224 break; 3225 3226 case TCPOPT_TIMESTAMP: 3227 if (optlen != TCPOLEN_TIMESTAMP) 3228 continue; 3229 oi->ts_present = 1; 3230 memcpy(&oi->ts_val, cp + 2, sizeof(oi->ts_val)); 3231 NTOHL(oi->ts_val); 3232 memcpy(&oi->ts_ecr, cp + 6, sizeof(oi->ts_ecr)); 3233 NTOHL(oi->ts_ecr); 3234 3235 if (!(th->th_flags & TH_SYN)) 3236 continue; 3237 if (TCPS_HAVERCVDSYN(tp->t_state)) 3238 continue; 3239 /* 3240 * A timestamp received in a SYN makes 3241 * it ok to send timestamp requests and replies. 3242 */ 3243 tp->t_flags |= TF_RCVD_TSTMP; 3244 tp->ts_recent = oi->ts_val; 3245 tp->ts_recent_age = tcp_now; 3246 break; 3247 3248 case TCPOPT_SACK_PERMITTED: 3249 if (optlen != TCPOLEN_SACK_PERMITTED) 3250 continue; 3251 if (!(th->th_flags & TH_SYN)) 3252 continue; 3253 if (TCPS_HAVERCVDSYN(tp->t_state)) 3254 continue; 3255 if (tcp_do_sack) { 3256 tp->t_flags |= TF_SACK_PERMIT; 3257 tp->t_flags |= TF_WILL_SACK; 3258 } 3259 break; 3260 3261 case TCPOPT_SACK: 3262 tcp_sack_option(tp, th, cp, optlen); 3263 break; 3264 #ifdef TCP_SIGNATURE 3265 case TCPOPT_SIGNATURE: 3266 if (optlen != TCPOLEN_SIGNATURE) 3267 continue; 3268 if (sigp && 3269 !consttime_memequal(sigp, cp + 2, TCP_SIGLEN)) 3270 return (-1); 3271 3272 sigp = sigbuf; 3273 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3274 tp->t_flags |= TF_SIGNATURE; 3275 break; 3276 #endif 3277 } 3278 } 3279 3280 #ifndef TCP_SIGNATURE 3281 return 0; 3282 #else 3283 if (tp->t_flags & TF_SIGNATURE) { 3284 sav = tcp_signature_getsav(m); 3285 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3286 return (-1); 3287 } 3288 3289 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) 3290 goto out; 3291 3292 if (sigp) { 3293 char sig[TCP_SIGLEN]; 3294 3295 tcp_fields_to_net(th); 3296 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3297 tcp_fields_to_host(th); 3298 goto out; 3299 } 3300 tcp_fields_to_host(th); 3301 3302 if (!consttime_memequal(sig, sigp, TCP_SIGLEN)) { 3303 TCP_STATINC(TCP_STAT_BADSIG); 3304 goto out; 3305 } else 3306 TCP_STATINC(TCP_STAT_GOODSIG); 3307 3308 key_sa_recordxfer(sav, m); 3309 KEY_SA_UNREF(&sav); 3310 } 3311 return 0; 3312 out: 3313 if (sav != NULL) 3314 KEY_SA_UNREF(&sav); 3315 return -1; 3316 #endif 3317 } 3318 3319 /* 3320 * Pull out of band byte out of a segment so 3321 * it doesn't appear in the user's data queue. 3322 * It is still reflected in the segment length for 3323 * sequencing purposes. 3324 */ 3325 void 3326 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3327 struct mbuf *m, int off) 3328 { 3329 int cnt = off + th->th_urp - 1; 3330 3331 while (cnt >= 0) { 3332 if (m->m_len > cnt) { 3333 char *cp = mtod(m, char *) + cnt; 3334 struct tcpcb *tp = sototcpcb(so); 3335 3336 tp->t_iobc = *cp; 3337 tp->t_oobflags |= TCPOOB_HAVEDATA; 3338 memmove(cp, cp + 1, (unsigned)(m->m_len - cnt - 1)); 3339 m->m_len--; 3340 return; 3341 } 3342 cnt -= m->m_len; 3343 m = m->m_next; 3344 if (m == NULL) 3345 break; 3346 } 3347 panic("tcp_pulloutofband"); 3348 } 3349 3350 /* 3351 * Collect new round-trip time estimate 3352 * and update averages and current timeout. 3353 * 3354 * rtt is in units of slow ticks (typically 500 ms) -- essentially the 3355 * difference of two timestamps. 3356 */ 3357 void 3358 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3359 { 3360 int32_t delta; 3361 3362 TCP_STATINC(TCP_STAT_RTTUPDATED); 3363 if (tp->t_srtt != 0) { 3364 /* 3365 * Compute the amount to add to srtt for smoothing, 3366 * *alpha, or 2^(-TCP_RTT_SHIFT). Because 3367 * srtt is stored in 1/32 slow ticks, we conceptually 3368 * shift left 5 bits, subtract srtt to get the 3369 * difference, and then shift right by TCP_RTT_SHIFT 3370 * (3) to obtain 1/8 of the difference. 3371 */ 3372 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3373 /* 3374 * This can never happen, because delta's lowest 3375 * possible value is 1/8 of t_srtt. But if it does, 3376 * set srtt to some reasonable value, here chosen 3377 * as 1/8 tick. 3378 */ 3379 if ((tp->t_srtt += delta) <= 0) 3380 tp->t_srtt = 1 << 2; 3381 /* 3382 * RFC2988 requires that rttvar be updated first. 3383 * This code is compliant because "delta" is the old 3384 * srtt minus the new observation (scaled). 3385 * 3386 * RFC2988 says: 3387 * rttvar = (1-beta) * rttvar + beta * |srtt-observed| 3388 * 3389 * delta is in units of 1/32 ticks, and has then been 3390 * divided by 8. This is equivalent to being in 1/16s 3391 * units and divided by 4. Subtract from it 1/4 of 3392 * the existing rttvar to form the (signed) amount to 3393 * adjust. 3394 */ 3395 if (delta < 0) 3396 delta = -delta; 3397 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3398 /* 3399 * As with srtt, this should never happen. There is 3400 * no support in RFC2988 for this operation. But 1/4s 3401 * as rttvar when faced with something arguably wrong 3402 * is ok. 3403 */ 3404 if ((tp->t_rttvar += delta) <= 0) 3405 tp->t_rttvar = 1 << 2; 3406 3407 /* 3408 * If srtt exceeds .01 second, ensure we use the 'remote' MSL 3409 * Problem is: it doesn't work. Disabled by defaulting 3410 * tcp_rttlocal to 0; see corresponding code in 3411 * tcp_subr that selects local vs remote in a different way. 3412 * 3413 * The static branch prediction hint here should be removed 3414 * when the rtt estimator is fixed and the rtt_enable code 3415 * is turned back on. 3416 */ 3417 if (__predict_false(tcp_rttlocal) && tcp_msl_enable 3418 && tp->t_srtt > tcp_msl_remote_threshold 3419 && tp->t_msl < tcp_msl_remote) { 3420 tp->t_msl = MIN(tcp_msl_remote, TCP_MAXMSL); 3421 } 3422 } else { 3423 /* 3424 * This is the first measurement. Per RFC2988, 2.2, 3425 * set rtt=R and srtt=R/2. 3426 * For srtt, storage representation is 1/32 ticks, 3427 * so shift left by 5. 3428 * For rttvar, storage representation is 1/16 ticks, 3429 * So shift left by 4, but then right by 1 to halve. 3430 */ 3431 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3432 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3433 } 3434 tp->t_rtttime = 0; 3435 tp->t_rxtshift = 0; 3436 3437 /* 3438 * the retransmit should happen at rtt + 4 * rttvar. 3439 * Because of the way we do the smoothing, srtt and rttvar 3440 * will each average +1/2 tick of bias. When we compute 3441 * the retransmit timer, we want 1/2 tick of rounding and 3442 * 1 extra tick because of +-1/2 tick uncertainty in the 3443 * firing of the timer. The bias will give us exactly the 3444 * 1.5 tick we need. But, because the bias is 3445 * statistical, we have to test that we don't drop below 3446 * the minimum feasible timer (which is 2 ticks). 3447 */ 3448 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3449 uimax(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3450 3451 /* 3452 * We received an ack for a packet that wasn't retransmitted; 3453 * it is probably safe to discard any error indications we've 3454 * received recently. This isn't quite right, but close enough 3455 * for now (a route might have failed after we sent a segment, 3456 * and the return path might not be symmetrical). 3457 */ 3458 tp->t_softerror = 0; 3459 } 3460