1 /* $NetBSD: tcp_input.c,v 1.408 2018/05/18 18:58:51 maxv 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 /* 141 * TODO list for SYN cache stuff: 142 * 143 * Find room for a "state" field, which is needed to keep a 144 * compressed state for TIME_WAIT TCBs. It's been noted already 145 * that this is fairly important for very high-volume web and 146 * mail servers, which use a large number of short-lived 147 * connections. 148 */ 149 150 #include <sys/cdefs.h> 151 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.408 2018/05/18 18:58:51 maxv Exp $"); 152 153 #ifdef _KERNEL_OPT 154 #include "opt_inet.h" 155 #include "opt_ipsec.h" 156 #include "opt_inet_csum.h" 157 #include "opt_tcp_debug.h" 158 #endif 159 160 #include <sys/param.h> 161 #include <sys/systm.h> 162 #include <sys/malloc.h> 163 #include <sys/mbuf.h> 164 #include <sys/protosw.h> 165 #include <sys/socket.h> 166 #include <sys/socketvar.h> 167 #include <sys/errno.h> 168 #include <sys/syslog.h> 169 #include <sys/pool.h> 170 #include <sys/domain.h> 171 #include <sys/kernel.h> 172 #ifdef TCP_SIGNATURE 173 #include <sys/md5.h> 174 #endif 175 #include <sys/lwp.h> /* for lwp0 */ 176 #include <sys/cprng.h> 177 178 #include <net/if.h> 179 #include <net/if_types.h> 180 181 #include <netinet/in.h> 182 #include <netinet/in_systm.h> 183 #include <netinet/ip.h> 184 #include <netinet/in_pcb.h> 185 #include <netinet/in_var.h> 186 #include <netinet/ip_var.h> 187 #include <netinet/in_offload.h> 188 189 #ifdef INET6 190 #include <netinet/ip6.h> 191 #include <netinet6/ip6_var.h> 192 #include <netinet6/in6_pcb.h> 193 #include <netinet6/ip6_var.h> 194 #include <netinet6/in6_var.h> 195 #include <netinet/icmp6.h> 196 #include <netinet6/nd6.h> 197 #ifdef TCP_SIGNATURE 198 #include <netinet6/scope6_var.h> 199 #endif 200 #endif 201 202 #ifndef INET6 203 #include <netinet/ip6.h> 204 #endif 205 206 #include <netinet/tcp.h> 207 #include <netinet/tcp_fsm.h> 208 #include <netinet/tcp_seq.h> 209 #include <netinet/tcp_timer.h> 210 #include <netinet/tcp_var.h> 211 #include <netinet/tcp_private.h> 212 #include <netinet/tcp_congctl.h> 213 #include <netinet/tcp_debug.h> 214 215 #ifdef INET6 216 #include "faith.h" 217 #if defined(NFAITH) && NFAITH > 0 218 #include <net/if_faith.h> 219 #endif 220 #endif 221 222 #ifdef IPSEC 223 #include <netipsec/ipsec.h> 224 #include <netipsec/key.h> 225 #ifdef INET6 226 #include <netipsec/ipsec6.h> 227 #endif 228 #endif /* IPSEC*/ 229 230 #include <netinet/tcp_vtw.h> 231 232 int tcprexmtthresh = 3; 233 int tcp_log_refused; 234 235 int tcp_do_autorcvbuf = 1; 236 int tcp_autorcvbuf_inc = 16 * 1024; 237 int tcp_autorcvbuf_max = 256 * 1024; 238 int tcp_msl = (TCPTV_MSL / PR_SLOWHZ); 239 240 static int tcp_rst_ppslim_count = 0; 241 static struct timeval tcp_rst_ppslim_last; 242 static int tcp_ackdrop_ppslim_count = 0; 243 static struct timeval tcp_ackdrop_ppslim_last; 244 245 static void syn_cache_timer(void *); 246 247 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 248 249 /* for modulo comparisons of timestamps */ 250 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 251 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 252 253 /* 254 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 255 */ 256 #ifdef INET6 257 static inline void 258 nd6_hint(struct tcpcb *tp) 259 { 260 struct rtentry *rt = NULL; 261 262 if (tp != NULL && tp->t_in6pcb != NULL && tp->t_family == AF_INET6 && 263 (rt = rtcache_validate(&tp->t_in6pcb->in6p_route)) != NULL) 264 nd6_nud_hint(rt); 265 rtcache_unref(rt, &tp->t_in6pcb->in6p_route); 266 } 267 #else 268 static inline void 269 nd6_hint(struct tcpcb *tp) 270 { 271 } 272 #endif 273 274 /* 275 * Compute ACK transmission behavior. Delay the ACK unless 276 * we have already delayed an ACK (must send an ACK every two segments). 277 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 278 * option is enabled. 279 */ 280 static void 281 tcp_setup_ack(struct tcpcb *tp, const struct tcphdr *th) 282 { 283 284 if (tp->t_flags & TF_DELACK || 285 (tcp_ack_on_push && th->th_flags & TH_PUSH)) 286 tp->t_flags |= TF_ACKNOW; 287 else 288 TCP_SET_DELACK(tp); 289 } 290 291 static void 292 icmp_check(struct tcpcb *tp, const struct tcphdr *th, int acked) 293 { 294 295 /* 296 * If we had a pending ICMP message that refers to data that have 297 * just been acknowledged, disregard the recorded ICMP message. 298 */ 299 if ((tp->t_flags & TF_PMTUD_PEND) && 300 SEQ_GT(th->th_ack, tp->t_pmtud_th_seq)) 301 tp->t_flags &= ~TF_PMTUD_PEND; 302 303 /* 304 * Keep track of the largest chunk of data 305 * acknowledged since last PMTU update 306 */ 307 if (tp->t_pmtud_mss_acked < acked) 308 tp->t_pmtud_mss_acked = acked; 309 } 310 311 /* 312 * Convert TCP protocol fields to host order for easier processing. 313 */ 314 static void 315 tcp_fields_to_host(struct tcphdr *th) 316 { 317 318 NTOHL(th->th_seq); 319 NTOHL(th->th_ack); 320 NTOHS(th->th_win); 321 NTOHS(th->th_urp); 322 } 323 324 /* 325 * ... and reverse the above. 326 */ 327 static void 328 tcp_fields_to_net(struct tcphdr *th) 329 { 330 331 HTONL(th->th_seq); 332 HTONL(th->th_ack); 333 HTONS(th->th_win); 334 HTONS(th->th_urp); 335 } 336 337 static void 338 tcp_urp_drop(struct tcphdr *th, int todrop, int *tiflags) 339 { 340 if (th->th_urp > todrop) { 341 th->th_urp -= todrop; 342 } else { 343 *tiflags &= ~TH_URG; 344 th->th_urp = 0; 345 } 346 } 347 348 #ifdef TCP_CSUM_COUNTERS 349 #include <sys/device.h> 350 351 extern struct evcnt tcp_hwcsum_ok; 352 extern struct evcnt tcp_hwcsum_bad; 353 extern struct evcnt tcp_hwcsum_data; 354 extern struct evcnt tcp_swcsum; 355 #if defined(INET6) 356 extern struct evcnt tcp6_hwcsum_ok; 357 extern struct evcnt tcp6_hwcsum_bad; 358 extern struct evcnt tcp6_hwcsum_data; 359 extern struct evcnt tcp6_swcsum; 360 #endif /* defined(INET6) */ 361 362 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 363 364 #else 365 366 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 367 368 #endif /* TCP_CSUM_COUNTERS */ 369 370 #ifdef TCP_REASS_COUNTERS 371 #include <sys/device.h> 372 373 extern struct evcnt tcp_reass_; 374 extern struct evcnt tcp_reass_empty; 375 extern struct evcnt tcp_reass_iteration[8]; 376 extern struct evcnt tcp_reass_prependfirst; 377 extern struct evcnt tcp_reass_prepend; 378 extern struct evcnt tcp_reass_insert; 379 extern struct evcnt tcp_reass_inserttail; 380 extern struct evcnt tcp_reass_append; 381 extern struct evcnt tcp_reass_appendtail; 382 extern struct evcnt tcp_reass_overlaptail; 383 extern struct evcnt tcp_reass_overlapfront; 384 extern struct evcnt tcp_reass_segdup; 385 extern struct evcnt tcp_reass_fragdup; 386 387 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 388 389 #else 390 391 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 392 393 #endif /* TCP_REASS_COUNTERS */ 394 395 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *, 396 int); 397 static int tcp_dooptions(struct tcpcb *, const u_char *, int, 398 struct tcphdr *, struct mbuf *, int, struct tcp_opt_info *); 399 400 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 401 #ifdef INET6 402 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 403 #endif 404 405 #if defined(MBUFTRACE) 406 struct mowner tcp_reass_mowner = MOWNER_INIT("tcp", "reass"); 407 #endif /* defined(MBUFTRACE) */ 408 409 static struct pool tcpipqent_pool; 410 411 void 412 tcpipqent_init(void) 413 { 414 415 pool_init(&tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", 416 NULL, IPL_VM); 417 } 418 419 struct ipqent * 420 tcpipqent_alloc(void) 421 { 422 struct ipqent *ipqe; 423 int s; 424 425 s = splvm(); 426 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 427 splx(s); 428 429 return ipqe; 430 } 431 432 void 433 tcpipqent_free(struct ipqent *ipqe) 434 { 435 int s; 436 437 s = splvm(); 438 pool_put(&tcpipqent_pool, ipqe); 439 splx(s); 440 } 441 442 /* 443 * Insert segment ti into reassembly queue of tcp with 444 * control block tp. Return TH_FIN if reassembly now includes 445 * a segment with FIN. 446 */ 447 static int 448 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int tlen) 449 { 450 struct ipqent *p, *q, *nq, *tiqe = NULL; 451 struct socket *so = NULL; 452 int pkt_flags; 453 tcp_seq pkt_seq; 454 unsigned pkt_len; 455 u_long rcvpartdupbyte = 0; 456 u_long rcvoobyte; 457 #ifdef TCP_REASS_COUNTERS 458 u_int count = 0; 459 #endif 460 uint64_t *tcps; 461 462 if (tp->t_inpcb) 463 so = tp->t_inpcb->inp_socket; 464 #ifdef INET6 465 else if (tp->t_in6pcb) 466 so = tp->t_in6pcb->in6p_socket; 467 #endif 468 469 TCP_REASS_LOCK_CHECK(tp); 470 471 /* 472 * Call with th==NULL after become established to 473 * force pre-ESTABLISHED data up to user socket. 474 */ 475 if (th == NULL) 476 goto present; 477 478 m_claimm(m, &tcp_reass_mowner); 479 480 rcvoobyte = tlen; 481 /* 482 * Copy these to local variables because the TCP header gets munged 483 * while we are collapsing mbufs. 484 */ 485 pkt_seq = th->th_seq; 486 pkt_len = tlen; 487 pkt_flags = th->th_flags; 488 489 TCP_REASS_COUNTER_INCR(&tcp_reass_); 490 491 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 492 /* 493 * When we miss a packet, the vast majority of time we get 494 * packets that follow it in order. So optimize for that. 495 */ 496 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 497 p->ipqe_len += pkt_len; 498 p->ipqe_flags |= pkt_flags; 499 m_cat(p->ipqe_m, m); 500 m = NULL; 501 tiqe = p; 502 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 503 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 504 goto skip_replacement; 505 } 506 /* 507 * While we're here, if the pkt is completely beyond 508 * anything we have, just insert it at the tail. 509 */ 510 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 511 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 512 goto insert_it; 513 } 514 } 515 516 q = TAILQ_FIRST(&tp->segq); 517 518 if (q != NULL) { 519 /* 520 * If this segment immediately precedes the first out-of-order 521 * block, simply slap the segment in front of it and (mostly) 522 * skip the complicated logic. 523 */ 524 if (pkt_seq + pkt_len == q->ipqe_seq) { 525 q->ipqe_seq = pkt_seq; 526 q->ipqe_len += pkt_len; 527 q->ipqe_flags |= pkt_flags; 528 m_cat(m, q->ipqe_m); 529 q->ipqe_m = m; 530 tiqe = q; 531 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 532 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 533 goto skip_replacement; 534 } 535 } else { 536 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 537 } 538 539 /* 540 * Find a segment which begins after this one does. 541 */ 542 for (p = NULL; q != NULL; q = nq) { 543 nq = TAILQ_NEXT(q, ipqe_q); 544 #ifdef TCP_REASS_COUNTERS 545 count++; 546 #endif 547 548 /* 549 * If the received segment is just right after this 550 * fragment, merge the two together and then check 551 * for further overlaps. 552 */ 553 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 554 pkt_len += q->ipqe_len; 555 pkt_flags |= q->ipqe_flags; 556 pkt_seq = q->ipqe_seq; 557 m_cat(q->ipqe_m, m); 558 m = q->ipqe_m; 559 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 560 goto free_ipqe; 561 } 562 563 /* 564 * If the received segment is completely past this 565 * fragment, we need to go to the next fragment. 566 */ 567 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 568 p = q; 569 continue; 570 } 571 572 /* 573 * If the fragment is past the received segment, 574 * it (or any following) can't be concatenated. 575 */ 576 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 577 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 578 break; 579 } 580 581 /* 582 * We've received all the data in this segment before. 583 * Mark it as a duplicate and return. 584 */ 585 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 586 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 587 tcps = TCP_STAT_GETREF(); 588 tcps[TCP_STAT_RCVDUPPACK]++; 589 tcps[TCP_STAT_RCVDUPBYTE] += pkt_len; 590 TCP_STAT_PUTREF(); 591 tcp_new_dsack(tp, pkt_seq, pkt_len); 592 m_freem(m); 593 if (tiqe != NULL) { 594 tcpipqent_free(tiqe); 595 } 596 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 597 goto out; 598 } 599 600 /* 601 * Received segment completely overlaps this fragment 602 * so we drop the fragment (this keeps the temporal 603 * ordering of segments correct). 604 */ 605 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 606 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 607 rcvpartdupbyte += q->ipqe_len; 608 m_freem(q->ipqe_m); 609 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 610 goto free_ipqe; 611 } 612 613 /* 614 * Received segment extends past the end of the fragment. 615 * Drop the overlapping bytes, merge the fragment and 616 * segment, and treat as a longer received packet. 617 */ 618 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 619 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 620 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 621 m_adj(m, overlap); 622 rcvpartdupbyte += overlap; 623 m_cat(q->ipqe_m, m); 624 m = q->ipqe_m; 625 pkt_seq = q->ipqe_seq; 626 pkt_len += q->ipqe_len - overlap; 627 rcvoobyte -= overlap; 628 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 629 goto free_ipqe; 630 } 631 632 /* 633 * Received segment extends past the front of the fragment. 634 * Drop the overlapping bytes on the received packet. The 635 * packet will then be concatenated with this fragment a 636 * bit later. 637 */ 638 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 639 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 640 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 641 m_adj(m, -overlap); 642 pkt_len -= overlap; 643 rcvpartdupbyte += overlap; 644 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 645 rcvoobyte -= overlap; 646 } 647 648 /* 649 * If the received segment immediately precedes this 650 * fragment then tack the fragment onto this segment 651 * and reinsert the data. 652 */ 653 if (q->ipqe_seq == pkt_seq + pkt_len) { 654 pkt_len += q->ipqe_len; 655 pkt_flags |= q->ipqe_flags; 656 m_cat(m, q->ipqe_m); 657 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 658 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 659 tp->t_segqlen--; 660 KASSERT(tp->t_segqlen >= 0); 661 KASSERT(tp->t_segqlen != 0 || 662 (TAILQ_EMPTY(&tp->segq) && 663 TAILQ_EMPTY(&tp->timeq))); 664 if (tiqe == NULL) { 665 tiqe = q; 666 } else { 667 tcpipqent_free(q); 668 } 669 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 670 break; 671 } 672 673 /* 674 * If the fragment is before the segment, remember it. 675 * When this loop is terminated, p will contain the 676 * pointer to the fragment that is right before the 677 * received segment. 678 */ 679 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 680 p = q; 681 682 continue; 683 684 /* 685 * This is a common operation. It also will allow 686 * to save doing a malloc/free in most instances. 687 */ 688 free_ipqe: 689 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 690 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 691 tp->t_segqlen--; 692 KASSERT(tp->t_segqlen >= 0); 693 KASSERT(tp->t_segqlen != 0 || 694 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 695 if (tiqe == NULL) { 696 tiqe = q; 697 } else { 698 tcpipqent_free(q); 699 } 700 } 701 702 #ifdef TCP_REASS_COUNTERS 703 if (count > 7) 704 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 705 else if (count > 0) 706 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 707 #endif 708 709 insert_it: 710 /* 711 * Allocate a new queue entry (block) since the received segment 712 * did not collapse onto any other out-of-order block. If it had 713 * collapsed, tiqe would not be NULL and we would be reusing it. 714 * 715 * If the allocation fails, drop the packet. 716 */ 717 if (tiqe == NULL) { 718 tiqe = tcpipqent_alloc(); 719 if (tiqe == NULL) { 720 TCP_STATINC(TCP_STAT_RCVMEMDROP); 721 m_freem(m); 722 goto out; 723 } 724 } 725 726 /* 727 * Update the counters. 728 */ 729 tp->t_rcvoopack++; 730 tcps = TCP_STAT_GETREF(); 731 tcps[TCP_STAT_RCVOOPACK]++; 732 tcps[TCP_STAT_RCVOOBYTE] += rcvoobyte; 733 if (rcvpartdupbyte) { 734 tcps[TCP_STAT_RCVPARTDUPPACK]++; 735 tcps[TCP_STAT_RCVPARTDUPBYTE] += rcvpartdupbyte; 736 } 737 TCP_STAT_PUTREF(); 738 739 /* 740 * Insert the new fragment queue entry into both queues. 741 */ 742 tiqe->ipqe_m = m; 743 tiqe->ipqe_seq = pkt_seq; 744 tiqe->ipqe_len = pkt_len; 745 tiqe->ipqe_flags = pkt_flags; 746 if (p == NULL) { 747 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 748 } else { 749 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 750 } 751 tp->t_segqlen++; 752 753 skip_replacement: 754 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 755 756 present: 757 /* 758 * Present data to user, advancing rcv_nxt through 759 * completed sequence space. 760 */ 761 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 762 goto out; 763 q = TAILQ_FIRST(&tp->segq); 764 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 765 goto out; 766 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 767 goto out; 768 769 tp->rcv_nxt += q->ipqe_len; 770 pkt_flags = q->ipqe_flags & TH_FIN; 771 nd6_hint(tp); 772 773 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 774 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 775 tp->t_segqlen--; 776 KASSERT(tp->t_segqlen >= 0); 777 KASSERT(tp->t_segqlen != 0 || 778 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 779 if (so->so_state & SS_CANTRCVMORE) 780 m_freem(q->ipqe_m); 781 else 782 sbappendstream(&so->so_rcv, q->ipqe_m); 783 tcpipqent_free(q); 784 TCP_REASS_UNLOCK(tp); 785 sorwakeup(so); 786 return pkt_flags; 787 788 out: 789 TCP_REASS_UNLOCK(tp); 790 return 0; 791 } 792 793 #ifdef INET6 794 int 795 tcp6_input(struct mbuf **mp, int *offp, int proto) 796 { 797 struct mbuf *m = *mp; 798 799 /* 800 * draft-itojun-ipv6-tcp-to-anycast 801 * better place to put this in? 802 */ 803 if (m->m_flags & M_ANYCAST6) { 804 struct ip6_hdr *ip6; 805 if (m->m_len < sizeof(struct ip6_hdr)) { 806 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 807 TCP_STATINC(TCP_STAT_RCVSHORT); 808 return IPPROTO_DONE; 809 } 810 } 811 ip6 = mtod(m, struct ip6_hdr *); 812 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 813 (char *)&ip6->ip6_dst - (char *)ip6); 814 return IPPROTO_DONE; 815 } 816 817 tcp_input(m, *offp, proto); 818 return IPPROTO_DONE; 819 } 820 #endif 821 822 static void 823 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 824 { 825 char src[INET_ADDRSTRLEN]; 826 char dst[INET_ADDRSTRLEN]; 827 828 if (ip) { 829 in_print(src, sizeof(src), &ip->ip_src); 830 in_print(dst, sizeof(dst), &ip->ip_dst); 831 } else { 832 strlcpy(src, "(unknown)", sizeof(src)); 833 strlcpy(dst, "(unknown)", sizeof(dst)); 834 } 835 log(LOG_INFO, 836 "Connection attempt to TCP %s:%d from %s:%d\n", 837 dst, ntohs(th->th_dport), 838 src, ntohs(th->th_sport)); 839 } 840 841 #ifdef INET6 842 static void 843 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 844 { 845 char src[INET6_ADDRSTRLEN]; 846 char dst[INET6_ADDRSTRLEN]; 847 848 if (ip6) { 849 in6_print(src, sizeof(src), &ip6->ip6_src); 850 in6_print(dst, sizeof(dst), &ip6->ip6_dst); 851 } else { 852 strlcpy(src, "(unknown v6)", sizeof(src)); 853 strlcpy(dst, "(unknown v6)", sizeof(dst)); 854 } 855 log(LOG_INFO, 856 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 857 dst, ntohs(th->th_dport), 858 src, ntohs(th->th_sport)); 859 } 860 #endif 861 862 /* 863 * Checksum extended TCP header and data. 864 */ 865 int 866 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, 867 int toff, int off, int tlen) 868 { 869 struct ifnet *rcvif; 870 int s; 871 872 /* 873 * XXX it's better to record and check if this mbuf is 874 * already checked. 875 */ 876 877 rcvif = m_get_rcvif(m, &s); 878 if (__predict_false(rcvif == NULL)) 879 goto badcsum; /* XXX */ 880 881 switch (af) { 882 case AF_INET: 883 switch (m->m_pkthdr.csum_flags & 884 ((rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 885 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 886 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 887 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 888 goto badcsum; 889 890 case M_CSUM_TCPv4|M_CSUM_DATA: { 891 u_int32_t hw_csum = m->m_pkthdr.csum_data; 892 893 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 894 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 895 const struct ip *ip = 896 mtod(m, const struct ip *); 897 898 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 899 ip->ip_dst.s_addr, 900 htons(hw_csum + tlen + off + IPPROTO_TCP)); 901 } 902 if ((hw_csum ^ 0xffff) != 0) 903 goto badcsum; 904 break; 905 } 906 907 case M_CSUM_TCPv4: 908 /* Checksum was okay. */ 909 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 910 break; 911 912 default: 913 /* 914 * Must compute it ourselves. Maybe skip checksum 915 * on loopback interfaces. 916 */ 917 if (__predict_true(!(rcvif->if_flags & IFF_LOOPBACK) || 918 tcp_do_loopback_cksum)) { 919 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 920 if (in4_cksum(m, IPPROTO_TCP, toff, 921 tlen + off) != 0) 922 goto badcsum; 923 } 924 break; 925 } 926 break; 927 928 #ifdef INET6 929 case AF_INET6: 930 switch (m->m_pkthdr.csum_flags & 931 ((rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 932 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 933 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 934 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 935 goto badcsum; 936 937 #if 0 /* notyet */ 938 case M_CSUM_TCPv6|M_CSUM_DATA: 939 #endif 940 941 case M_CSUM_TCPv6: 942 /* Checksum was okay. */ 943 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 944 break; 945 946 default: 947 /* 948 * Must compute it ourselves. Maybe skip checksum 949 * on loopback interfaces. 950 */ 951 if (__predict_true((m->m_flags & M_LOOP) == 0 || 952 tcp_do_loopback_cksum)) { 953 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 954 if (in6_cksum(m, IPPROTO_TCP, toff, 955 tlen + off) != 0) 956 goto badcsum; 957 } 958 } 959 break; 960 #endif /* INET6 */ 961 } 962 m_put_rcvif(rcvif, &s); 963 964 return 0; 965 966 badcsum: 967 m_put_rcvif(rcvif, &s); 968 TCP_STATINC(TCP_STAT_RCVBADSUM); 969 return -1; 970 } 971 972 /* 973 * When a packet arrives addressed to a vestigial tcpbp, we 974 * nevertheless have to respond to it per the spec. 975 * 976 * This code is duplicated from the one in tcp_input(). 977 */ 978 static void tcp_vtw_input(struct tcphdr *th, vestigial_inpcb_t *vp, 979 struct mbuf *m, int tlen) 980 { 981 int tiflags; 982 int todrop; 983 uint32_t t_flags = 0; 984 uint64_t *tcps; 985 986 tiflags = th->th_flags; 987 todrop = vp->rcv_nxt - th->th_seq; 988 989 if (todrop > 0) { 990 if (tiflags & TH_SYN) { 991 tiflags &= ~TH_SYN; 992 th->th_seq++; 993 tcp_urp_drop(th, 1, &tiflags); 994 todrop--; 995 } 996 if (todrop > tlen || 997 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 998 /* 999 * Any valid FIN or RST must be to the left of the 1000 * window. At this point the FIN or RST must be a 1001 * duplicate or out of sequence; drop it. 1002 */ 1003 if (tiflags & TH_RST) 1004 goto drop; 1005 tiflags &= ~(TH_FIN|TH_RST); 1006 1007 /* 1008 * Send an ACK to resynchronize and drop any data. 1009 * But keep on processing for RST or ACK. 1010 */ 1011 t_flags |= TF_ACKNOW; 1012 todrop = tlen; 1013 tcps = TCP_STAT_GETREF(); 1014 tcps[TCP_STAT_RCVDUPPACK] += 1; 1015 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 1016 TCP_STAT_PUTREF(); 1017 } else if ((tiflags & TH_RST) && 1018 th->th_seq != vp->rcv_nxt) { 1019 /* 1020 * Test for reset before adjusting the sequence 1021 * number for overlapping data. 1022 */ 1023 goto dropafterack_ratelim; 1024 } else { 1025 tcps = TCP_STAT_GETREF(); 1026 tcps[TCP_STAT_RCVPARTDUPPACK] += 1; 1027 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 1028 TCP_STAT_PUTREF(); 1029 } 1030 1031 // tcp_new_dsack(tp, th->th_seq, todrop); 1032 // hdroptlen += todrop; /*drop from head afterwards*/ 1033 1034 th->th_seq += todrop; 1035 tlen -= todrop; 1036 tcp_urp_drop(th, todrop, &tiflags); 1037 } 1038 1039 /* 1040 * If new data are received on a connection after the 1041 * user processes are gone, then RST the other end. 1042 */ 1043 if (tlen) { 1044 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 1045 goto dropwithreset; 1046 } 1047 1048 /* 1049 * If segment ends after window, drop trailing data 1050 * (and PUSH and FIN); if nothing left, just ACK. 1051 */ 1052 todrop = (th->th_seq + tlen) - (vp->rcv_nxt + vp->rcv_wnd); 1053 1054 if (todrop > 0) { 1055 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 1056 if (todrop >= tlen) { 1057 /* 1058 * The segment actually starts after the window. 1059 * th->th_seq + tlen - vp->rcv_nxt - vp->rcv_wnd >= tlen 1060 * th->th_seq - vp->rcv_nxt - vp->rcv_wnd >= 0 1061 * th->th_seq >= vp->rcv_nxt + vp->rcv_wnd 1062 */ 1063 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 1064 1065 /* 1066 * If a new connection request is received 1067 * while in TIME_WAIT, drop the old connection 1068 * and start over if the sequence numbers 1069 * are above the previous ones. 1070 */ 1071 if ((tiflags & TH_SYN) && 1072 SEQ_GT(th->th_seq, vp->rcv_nxt)) { 1073 /* 1074 * We only support this in the !NOFDREF case, which 1075 * is to say: not here. 1076 */ 1077 goto dropwithreset; 1078 } 1079 1080 /* 1081 * If window is closed can only take segments at 1082 * window edge, and have to drop data and PUSH from 1083 * incoming segments. Continue processing, but 1084 * remember to ack. Otherwise, drop segment 1085 * and (if not RST) ack. 1086 */ 1087 if (vp->rcv_wnd == 0 && th->th_seq == vp->rcv_nxt) { 1088 t_flags |= TF_ACKNOW; 1089 TCP_STATINC(TCP_STAT_RCVWINPROBE); 1090 } else { 1091 goto dropafterack; 1092 } 1093 } else { 1094 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 1095 } 1096 m_adj(m, -todrop); 1097 tlen -= todrop; 1098 tiflags &= ~(TH_PUSH|TH_FIN); 1099 } 1100 1101 if (tiflags & TH_RST) { 1102 if (th->th_seq != vp->rcv_nxt) 1103 goto dropafterack_ratelim; 1104 1105 vtw_del(vp->ctl, vp->vtw); 1106 goto drop; 1107 } 1108 1109 /* 1110 * If the ACK bit is off we drop the segment and return. 1111 */ 1112 if ((tiflags & TH_ACK) == 0) { 1113 if (t_flags & TF_ACKNOW) 1114 goto dropafterack; 1115 goto drop; 1116 } 1117 1118 /* 1119 * In TIME_WAIT state the only thing that should arrive 1120 * is a retransmission of the remote FIN. Acknowledge 1121 * it and restart the finack timer. 1122 */ 1123 vtw_restart(vp); 1124 goto dropafterack; 1125 1126 dropafterack: 1127 /* 1128 * Generate an ACK dropping incoming segment if it occupies 1129 * sequence space, where the ACK reflects our state. 1130 */ 1131 if (tiflags & TH_RST) 1132 goto drop; 1133 goto dropafterack2; 1134 1135 dropafterack_ratelim: 1136 /* 1137 * We may want to rate-limit ACKs against SYN/RST attack. 1138 */ 1139 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 1140 tcp_ackdrop_ppslim) == 0) { 1141 /* XXX stat */ 1142 goto drop; 1143 } 1144 /* ...fall into dropafterack2... */ 1145 1146 dropafterack2: 1147 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, th->th_ack, TH_ACK); 1148 return; 1149 1150 dropwithreset: 1151 /* 1152 * Generate a RST, dropping incoming segment. 1153 * Make ACK acceptable to originator of segment. 1154 */ 1155 if (tiflags & TH_RST) 1156 goto drop; 1157 1158 if (tiflags & TH_ACK) { 1159 tcp_respond(0, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 1160 } else { 1161 if (tiflags & TH_SYN) 1162 ++tlen; 1163 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, (tcp_seq)0, 1164 TH_RST|TH_ACK); 1165 } 1166 return; 1167 drop: 1168 m_freem(m); 1169 } 1170 1171 /* 1172 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 1173 */ 1174 void 1175 tcp_input(struct mbuf *m, ...) 1176 { 1177 struct tcphdr *th; 1178 struct ip *ip; 1179 struct inpcb *inp; 1180 #ifdef INET6 1181 struct ip6_hdr *ip6; 1182 struct in6pcb *in6p; 1183 #endif 1184 u_int8_t *optp = NULL; 1185 int optlen = 0; 1186 int len, tlen, toff, hdroptlen = 0; 1187 struct tcpcb *tp = NULL; 1188 int tiflags; 1189 struct socket *so = NULL; 1190 int todrop, acked, ourfinisacked, needoutput = 0; 1191 bool dupseg; 1192 #ifdef TCP_DEBUG 1193 short ostate = 0; 1194 #endif 1195 u_long tiwin; 1196 struct tcp_opt_info opti; 1197 int off, iphlen; 1198 va_list ap; 1199 int af; /* af on the wire */ 1200 struct mbuf *tcp_saveti = NULL; 1201 uint32_t ts_rtt; 1202 uint8_t iptos; 1203 uint64_t *tcps; 1204 vestigial_inpcb_t vestige; 1205 1206 vestige.valid = 0; 1207 1208 MCLAIM(m, &tcp_rx_mowner); 1209 va_start(ap, m); 1210 toff = va_arg(ap, int); 1211 (void)va_arg(ap, int); /* ignore value, advance ap */ 1212 va_end(ap); 1213 1214 TCP_STATINC(TCP_STAT_RCVTOTAL); 1215 1216 memset(&opti, 0, sizeof(opti)); 1217 opti.ts_present = 0; 1218 opti.maxseg = 0; 1219 1220 /* 1221 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 1222 * 1223 * TCP is, by definition, unicast, so we reject all 1224 * multicast outright. 1225 * 1226 * Note, there are additional src/dst address checks in 1227 * the AF-specific code below. 1228 */ 1229 if (m->m_flags & (M_BCAST|M_MCAST)) { 1230 /* XXX stat */ 1231 goto drop; 1232 } 1233 #ifdef INET6 1234 if (m->m_flags & M_ANYCAST6) { 1235 /* XXX stat */ 1236 goto drop; 1237 } 1238 #endif 1239 1240 M_REGION_GET(th, struct tcphdr *, m, toff, sizeof(struct tcphdr)); 1241 if (th == NULL) { 1242 TCP_STATINC(TCP_STAT_RCVSHORT); 1243 return; 1244 } 1245 1246 /* 1247 * Get IP and TCP header. 1248 * Note: IP leaves IP header in first mbuf. 1249 */ 1250 ip = mtod(m, struct ip *); 1251 switch (ip->ip_v) { 1252 case 4: 1253 #ifdef INET6 1254 ip6 = NULL; 1255 #endif 1256 af = AF_INET; 1257 iphlen = sizeof(struct ip); 1258 1259 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1260 in_broadcast(ip->ip_dst, m_get_rcvif_NOMPSAFE(m))) 1261 goto drop; 1262 1263 /* We do the checksum after PCB lookup... */ 1264 len = ntohs(ip->ip_len); 1265 tlen = len - toff; 1266 iptos = ip->ip_tos; 1267 break; 1268 #ifdef INET6 1269 case 6: 1270 ip = NULL; 1271 iphlen = sizeof(struct ip6_hdr); 1272 af = AF_INET6; 1273 ip6 = mtod(m, struct ip6_hdr *); 1274 1275 /* 1276 * Be proactive about unspecified IPv6 address in source. 1277 * As we use all-zero to indicate unbounded/unconnected pcb, 1278 * unspecified IPv6 address can be used to confuse us. 1279 * 1280 * Note that packets with unspecified IPv6 destination is 1281 * already dropped in ip6_input. 1282 */ 1283 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1284 /* XXX stat */ 1285 goto drop; 1286 } 1287 1288 /* 1289 * Make sure destination address is not multicast. 1290 * Source address checked in ip6_input(). 1291 */ 1292 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1293 /* XXX stat */ 1294 goto drop; 1295 } 1296 1297 /* We do the checksum after PCB lookup... */ 1298 len = m->m_pkthdr.len; 1299 tlen = len - toff; 1300 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1301 break; 1302 #endif 1303 default: 1304 m_freem(m); 1305 return; 1306 } 1307 1308 /* 1309 * Enforce alignment requirements that are violated in 1310 * some cases, see kern/50766 for details. 1311 */ 1312 if (TCP_HDR_ALIGNED_P(th) == 0) { 1313 m = m_copyup(m, toff + sizeof(struct tcphdr), 0); 1314 if (m == NULL) { 1315 TCP_STATINC(TCP_STAT_RCVSHORT); 1316 return; 1317 } 1318 ip = mtod(m, struct ip *); 1319 #ifdef INET6 1320 ip6 = mtod(m, struct ip6_hdr *); 1321 #endif 1322 th = (struct tcphdr *)(mtod(m, char *) + toff); 1323 } 1324 KASSERT(TCP_HDR_ALIGNED_P(th)); 1325 1326 /* 1327 * Check that TCP offset makes sense, pull out TCP options and 1328 * adjust length. 1329 */ 1330 off = th->th_off << 2; 1331 if (off < sizeof(struct tcphdr) || off > tlen) { 1332 TCP_STATINC(TCP_STAT_RCVBADOFF); 1333 goto drop; 1334 } 1335 tlen -= off; 1336 1337 if (off > sizeof(struct tcphdr)) { 1338 M_REGION_GET(th, struct tcphdr *, m, toff, off); 1339 if (th == NULL) { 1340 TCP_STATINC(TCP_STAT_RCVSHORT); 1341 return; 1342 } 1343 KASSERT(TCP_HDR_ALIGNED_P(th)); 1344 optlen = off - sizeof(struct tcphdr); 1345 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1346 1347 /* 1348 * Do quick retrieval of timestamp options. 1349 * 1350 * If timestamp is the only option and it's formatted as 1351 * recommended in RFC 1323 appendix A, we quickly get the 1352 * values now and don't bother calling tcp_dooptions(), 1353 * etc. 1354 */ 1355 if ((optlen == TCPOLEN_TSTAMP_APPA || 1356 (optlen > TCPOLEN_TSTAMP_APPA && 1357 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1358 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1359 (th->th_flags & TH_SYN) == 0) { 1360 opti.ts_present = 1; 1361 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1362 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1363 optp = NULL; /* we've parsed the options */ 1364 } 1365 } 1366 tiflags = th->th_flags; 1367 1368 /* 1369 * Checksum extended TCP header and data 1370 */ 1371 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1372 goto badcsum; 1373 1374 /* 1375 * Locate pcb for segment. 1376 */ 1377 findpcb: 1378 inp = NULL; 1379 #ifdef INET6 1380 in6p = NULL; 1381 #endif 1382 switch (af) { 1383 case AF_INET: 1384 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1385 ip->ip_dst, th->th_dport, &vestige); 1386 if (inp == NULL && !vestige.valid) { 1387 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1388 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, 1389 th->th_dport); 1390 } 1391 #ifdef INET6 1392 if (inp == NULL && !vestige.valid) { 1393 struct in6_addr s, d; 1394 1395 /* mapped addr case */ 1396 in6_in_2_v4mapin6(&ip->ip_src, &s); 1397 in6_in_2_v4mapin6(&ip->ip_dst, &d); 1398 in6p = in6_pcblookup_connect(&tcbtable, &s, 1399 th->th_sport, &d, th->th_dport, 0, &vestige); 1400 if (in6p == 0 && !vestige.valid) { 1401 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1402 in6p = in6_pcblookup_bind(&tcbtable, &d, 1403 th->th_dport, 0); 1404 } 1405 } 1406 #endif 1407 #ifndef INET6 1408 if (inp == NULL && !vestige.valid) 1409 #else 1410 if (inp == NULL && in6p == NULL && !vestige.valid) 1411 #endif 1412 { 1413 TCP_STATINC(TCP_STAT_NOPORT); 1414 if (tcp_log_refused && 1415 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1416 tcp4_log_refused(ip, th); 1417 } 1418 tcp_fields_to_host(th); 1419 goto dropwithreset_ratelim; 1420 } 1421 #if defined(IPSEC) 1422 if (ipsec_used) { 1423 if (inp && ipsec_in_reject(m, inp)) { 1424 goto drop; 1425 } 1426 #ifdef INET6 1427 else if (in6p && ipsec_in_reject(m, in6p)) { 1428 goto drop; 1429 } 1430 #endif 1431 } 1432 #endif /*IPSEC*/ 1433 break; 1434 #ifdef INET6 1435 case AF_INET6: 1436 { 1437 int faith; 1438 1439 #if defined(NFAITH) && NFAITH > 0 1440 faith = faithprefix(&ip6->ip6_dst); 1441 #else 1442 faith = 0; 1443 #endif 1444 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1445 th->th_sport, &ip6->ip6_dst, th->th_dport, faith, &vestige); 1446 if (!in6p && !vestige.valid) { 1447 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1448 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1449 th->th_dport, faith); 1450 } 1451 if (!in6p && !vestige.valid) { 1452 TCP_STATINC(TCP_STAT_NOPORT); 1453 if (tcp_log_refused && 1454 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1455 tcp6_log_refused(ip6, th); 1456 } 1457 tcp_fields_to_host(th); 1458 goto dropwithreset_ratelim; 1459 } 1460 #if defined(IPSEC) 1461 if (ipsec_used && in6p && ipsec_in_reject(m, in6p)) { 1462 goto drop; 1463 } 1464 #endif 1465 break; 1466 } 1467 #endif 1468 } 1469 1470 tcp_fields_to_host(th); 1471 1472 /* 1473 * If the state is CLOSED (i.e., TCB does not exist) then 1474 * all data in the incoming segment is discarded. 1475 * If the TCB exists but is in CLOSED state, it is embryonic, 1476 * but should either do a listen or a connect soon. 1477 */ 1478 tp = NULL; 1479 so = NULL; 1480 if (inp) { 1481 /* Check the minimum TTL for socket. */ 1482 if (ip->ip_ttl < inp->inp_ip_minttl) 1483 goto drop; 1484 1485 tp = intotcpcb(inp); 1486 so = inp->inp_socket; 1487 } 1488 #ifdef INET6 1489 else if (in6p) { 1490 tp = in6totcpcb(in6p); 1491 so = in6p->in6p_socket; 1492 } 1493 #endif 1494 else if (vestige.valid) { 1495 /* We do not support the resurrection of vtw tcpcps. */ 1496 tcp_vtw_input(th, &vestige, m, tlen); 1497 m = NULL; 1498 goto drop; 1499 } 1500 1501 if (tp == NULL) 1502 goto dropwithreset_ratelim; 1503 if (tp->t_state == TCPS_CLOSED) 1504 goto drop; 1505 1506 KASSERT(so->so_lock == softnet_lock); 1507 KASSERT(solocked(so)); 1508 1509 /* Unscale the window into a 32-bit value. */ 1510 if ((tiflags & TH_SYN) == 0) 1511 tiwin = th->th_win << tp->snd_scale; 1512 else 1513 tiwin = th->th_win; 1514 1515 #ifdef INET6 1516 /* save packet options if user wanted */ 1517 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1518 if (in6p->in6p_options) { 1519 m_freem(in6p->in6p_options); 1520 in6p->in6p_options = NULL; 1521 } 1522 KASSERT(ip6 != NULL); 1523 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1524 } 1525 #endif 1526 1527 if (so->so_options & SO_DEBUG) { 1528 #ifdef TCP_DEBUG 1529 ostate = tp->t_state; 1530 #endif 1531 1532 tcp_saveti = NULL; 1533 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1534 goto nosave; 1535 1536 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1537 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1538 if (tcp_saveti == NULL) 1539 goto nosave; 1540 } else { 1541 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1542 if (tcp_saveti == NULL) 1543 goto nosave; 1544 MCLAIM(m, &tcp_mowner); 1545 tcp_saveti->m_len = iphlen; 1546 m_copydata(m, 0, iphlen, 1547 mtod(tcp_saveti, void *)); 1548 } 1549 1550 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1551 m_freem(tcp_saveti); 1552 tcp_saveti = NULL; 1553 } else { 1554 tcp_saveti->m_len += sizeof(struct tcphdr); 1555 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1556 sizeof(struct tcphdr)); 1557 } 1558 nosave:; 1559 } 1560 1561 if (so->so_options & SO_ACCEPTCONN) { 1562 union syn_cache_sa src; 1563 union syn_cache_sa dst; 1564 1565 KASSERT(tp->t_state == TCPS_LISTEN); 1566 1567 memset(&src, 0, sizeof(src)); 1568 memset(&dst, 0, sizeof(dst)); 1569 switch (af) { 1570 case AF_INET: 1571 src.sin.sin_len = sizeof(struct sockaddr_in); 1572 src.sin.sin_family = AF_INET; 1573 src.sin.sin_addr = ip->ip_src; 1574 src.sin.sin_port = th->th_sport; 1575 1576 dst.sin.sin_len = sizeof(struct sockaddr_in); 1577 dst.sin.sin_family = AF_INET; 1578 dst.sin.sin_addr = ip->ip_dst; 1579 dst.sin.sin_port = th->th_dport; 1580 break; 1581 #ifdef INET6 1582 case AF_INET6: 1583 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1584 src.sin6.sin6_family = AF_INET6; 1585 src.sin6.sin6_addr = ip6->ip6_src; 1586 src.sin6.sin6_port = th->th_sport; 1587 1588 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1589 dst.sin6.sin6_family = AF_INET6; 1590 dst.sin6.sin6_addr = ip6->ip6_dst; 1591 dst.sin6.sin6_port = th->th_dport; 1592 break; 1593 #endif 1594 } 1595 1596 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1597 if (tiflags & TH_RST) { 1598 syn_cache_reset(&src.sa, &dst.sa, th); 1599 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1600 (TH_ACK|TH_SYN)) { 1601 /* 1602 * Received a SYN,ACK. This should never 1603 * happen while we are in LISTEN. Send an RST. 1604 */ 1605 goto badsyn; 1606 } else if (tiflags & TH_ACK) { 1607 so = syn_cache_get(&src.sa, &dst.sa, th, so, m); 1608 if (so == NULL) { 1609 /* 1610 * We don't have a SYN for this ACK; 1611 * send an RST. 1612 */ 1613 goto badsyn; 1614 } else if (so == (struct socket *)(-1)) { 1615 /* 1616 * We were unable to create the 1617 * connection. If the 3-way handshake 1618 * was completed, and RST has been 1619 * sent to the peer. Since the mbuf 1620 * might be in use for the reply, do 1621 * not free it. 1622 */ 1623 m = NULL; 1624 } else { 1625 /* 1626 * We have created a full-blown 1627 * connection. 1628 */ 1629 tp = NULL; 1630 inp = NULL; 1631 #ifdef INET6 1632 in6p = NULL; 1633 #endif 1634 switch (so->so_proto->pr_domain->dom_family) { 1635 case AF_INET: 1636 inp = sotoinpcb(so); 1637 tp = intotcpcb(inp); 1638 break; 1639 #ifdef INET6 1640 case AF_INET6: 1641 in6p = sotoin6pcb(so); 1642 tp = in6totcpcb(in6p); 1643 break; 1644 #endif 1645 } 1646 if (tp == NULL) 1647 goto badsyn; /*XXX*/ 1648 tiwin <<= tp->snd_scale; 1649 goto after_listen; 1650 } 1651 } else { 1652 /* 1653 * None of RST, SYN or ACK was set. 1654 * This is an invalid packet for a 1655 * TCB in LISTEN state. Send a RST. 1656 */ 1657 goto badsyn; 1658 } 1659 } else { 1660 /* 1661 * Received a SYN. 1662 */ 1663 1664 #ifdef INET6 1665 /* 1666 * If deprecated address is forbidden, we do 1667 * not accept SYN to deprecated interface 1668 * address to prevent any new inbound 1669 * connection from getting established. 1670 * When we do not accept SYN, we send a TCP 1671 * RST, with deprecated source address (instead 1672 * of dropping it). We compromise it as it is 1673 * much better for peer to send a RST, and 1674 * RST will be the final packet for the 1675 * exchange. 1676 * 1677 * If we do not forbid deprecated addresses, we 1678 * accept the SYN packet. RFC2462 does not 1679 * suggest dropping SYN in this case. 1680 * If we decipher RFC2462 5.5.4, it says like 1681 * this: 1682 * 1. use of deprecated addr with existing 1683 * communication is okay - "SHOULD continue 1684 * to be used" 1685 * 2. use of it with new communication: 1686 * (2a) "SHOULD NOT be used if alternate 1687 * address with sufficient scope is 1688 * available" 1689 * (2b) nothing mentioned otherwise. 1690 * Here we fall into (2b) case as we have no 1691 * choice in our source address selection - we 1692 * must obey the peer. 1693 * 1694 * The wording in RFC2462 is confusing, and 1695 * there are multiple description text for 1696 * deprecated address handling - worse, they 1697 * are not exactly the same. I believe 5.5.4 1698 * is the best one, so we follow 5.5.4. 1699 */ 1700 if (af == AF_INET6 && !ip6_use_deprecated) { 1701 struct in6_ifaddr *ia6; 1702 int s; 1703 struct ifnet *rcvif = m_get_rcvif(m, &s); 1704 if (rcvif == NULL) 1705 goto dropwithreset; /* XXX */ 1706 if ((ia6 = in6ifa_ifpwithaddr(rcvif, 1707 &ip6->ip6_dst)) && 1708 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1709 tp = NULL; 1710 m_put_rcvif(rcvif, &s); 1711 goto dropwithreset; 1712 } 1713 m_put_rcvif(rcvif, &s); 1714 } 1715 #endif 1716 1717 /* 1718 * LISTEN socket received a SYN from itself? This 1719 * can't possibly be valid; drop the packet. 1720 */ 1721 if (th->th_sport == th->th_dport) { 1722 int eq = 0; 1723 1724 switch (af) { 1725 case AF_INET: 1726 eq = in_hosteq(ip->ip_src, ip->ip_dst); 1727 break; 1728 #ifdef INET6 1729 case AF_INET6: 1730 eq = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, 1731 &ip6->ip6_dst); 1732 break; 1733 #endif 1734 } 1735 if (eq) { 1736 TCP_STATINC(TCP_STAT_BADSYN); 1737 goto drop; 1738 } 1739 } 1740 1741 /* 1742 * SYN looks ok; create compressed TCP 1743 * state for it. 1744 */ 1745 if (so->so_qlen <= so->so_qlimit && 1746 syn_cache_add(&src.sa, &dst.sa, th, toff, 1747 so, m, optp, optlen, &opti)) 1748 m = NULL; 1749 } 1750 1751 goto drop; 1752 } 1753 1754 after_listen: 1755 /* 1756 * From here on, we're dealing with !LISTEN. 1757 */ 1758 KASSERT(tp->t_state != TCPS_LISTEN); 1759 1760 /* 1761 * Segment received on connection. 1762 * Reset idle time and keep-alive timer. 1763 */ 1764 tp->t_rcvtime = tcp_now; 1765 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1766 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1767 1768 /* 1769 * Process options. 1770 */ 1771 #ifdef TCP_SIGNATURE 1772 if (optp || (tp->t_flags & TF_SIGNATURE)) 1773 #else 1774 if (optp) 1775 #endif 1776 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1777 goto drop; 1778 1779 if (TCP_SACK_ENABLED(tp)) { 1780 tcp_del_sackholes(tp, th); 1781 } 1782 1783 if (TCP_ECN_ALLOWED(tp)) { 1784 if (tiflags & TH_CWR) { 1785 tp->t_flags &= ~TF_ECN_SND_ECE; 1786 } 1787 switch (iptos & IPTOS_ECN_MASK) { 1788 case IPTOS_ECN_CE: 1789 tp->t_flags |= TF_ECN_SND_ECE; 1790 TCP_STATINC(TCP_STAT_ECN_CE); 1791 break; 1792 case IPTOS_ECN_ECT0: 1793 TCP_STATINC(TCP_STAT_ECN_ECT); 1794 break; 1795 case IPTOS_ECN_ECT1: 1796 /* XXX: ignore for now -- rpaulo */ 1797 break; 1798 } 1799 /* 1800 * Congestion experienced. 1801 * Ignore if we are already trying to recover. 1802 */ 1803 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1804 tp->t_congctl->cong_exp(tp); 1805 } 1806 1807 if (opti.ts_present && opti.ts_ecr) { 1808 /* 1809 * Calculate the RTT from the returned time stamp and the 1810 * connection's time base. If the time stamp is later than 1811 * the current time, or is extremely old, fall back to non-1323 1812 * RTT calculation. Since ts_rtt is unsigned, we can test both 1813 * at the same time. 1814 * 1815 * Note that ts_rtt is in units of slow ticks (500 1816 * ms). Since most earthbound RTTs are < 500 ms, 1817 * observed values will have large quantization noise. 1818 * Our smoothed RTT is then the fraction of observed 1819 * samples that are 1 tick instead of 0 (times 500 1820 * ms). 1821 * 1822 * ts_rtt is increased by 1 to denote a valid sample, 1823 * with 0 indicating an invalid measurement. This 1824 * extra 1 must be removed when ts_rtt is used, or 1825 * else an an erroneous extra 500 ms will result. 1826 */ 1827 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1828 if (ts_rtt > TCP_PAWS_IDLE) 1829 ts_rtt = 0; 1830 } else { 1831 ts_rtt = 0; 1832 } 1833 1834 /* 1835 * Fast path: check for the two common cases of a uni-directional 1836 * data transfer. If: 1837 * o We are in the ESTABLISHED state, and 1838 * o The packet has no control flags, and 1839 * o The packet is in-sequence, and 1840 * o The window didn't change, and 1841 * o We are not retransmitting 1842 * It's a candidate. 1843 * 1844 * If the length (tlen) is zero and the ack moved forward, we're 1845 * the sender side of the transfer. Just free the data acked and 1846 * wake any higher level process that was blocked waiting for 1847 * space. 1848 * 1849 * If the length is non-zero and the ack didn't move, we're the 1850 * receiver side. If we're getting packets in-order (the reassembly 1851 * queue is empty), add the data to the socket buffer and note 1852 * that we need a delayed ack. 1853 */ 1854 if (tp->t_state == TCPS_ESTABLISHED && 1855 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1856 == TH_ACK && 1857 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1858 th->th_seq == tp->rcv_nxt && 1859 tiwin && tiwin == tp->snd_wnd && 1860 tp->snd_nxt == tp->snd_max) { 1861 1862 /* 1863 * If last ACK falls within this segment's sequence numbers, 1864 * record the timestamp. 1865 * NOTE that the test is modified according to the latest 1866 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1867 * 1868 * note that we already know 1869 * TSTMP_GEQ(opti.ts_val, tp->ts_recent) 1870 */ 1871 if (opti.ts_present && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1872 tp->ts_recent_age = tcp_now; 1873 tp->ts_recent = opti.ts_val; 1874 } 1875 1876 if (tlen == 0) { 1877 /* Ack prediction. */ 1878 if (SEQ_GT(th->th_ack, tp->snd_una) && 1879 SEQ_LEQ(th->th_ack, tp->snd_max) && 1880 tp->snd_cwnd >= tp->snd_wnd && 1881 tp->t_partialacks < 0) { 1882 /* 1883 * this is a pure ack for outstanding data. 1884 */ 1885 if (ts_rtt) 1886 tcp_xmit_timer(tp, ts_rtt - 1); 1887 else if (tp->t_rtttime && 1888 SEQ_GT(th->th_ack, tp->t_rtseq)) 1889 tcp_xmit_timer(tp, 1890 tcp_now - tp->t_rtttime); 1891 acked = th->th_ack - tp->snd_una; 1892 tcps = TCP_STAT_GETREF(); 1893 tcps[TCP_STAT_PREDACK]++; 1894 tcps[TCP_STAT_RCVACKPACK]++; 1895 tcps[TCP_STAT_RCVACKBYTE] += acked; 1896 TCP_STAT_PUTREF(); 1897 nd6_hint(tp); 1898 1899 if (acked > (tp->t_lastoff - tp->t_inoff)) 1900 tp->t_lastm = NULL; 1901 sbdrop(&so->so_snd, acked); 1902 tp->t_lastoff -= acked; 1903 1904 icmp_check(tp, th, acked); 1905 1906 tp->snd_una = th->th_ack; 1907 tp->snd_fack = tp->snd_una; 1908 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1909 tp->snd_high = tp->snd_una; 1910 m_freem(m); 1911 1912 /* 1913 * If all outstanding data are acked, stop 1914 * retransmit timer, otherwise restart timer 1915 * using current (possibly backed-off) value. 1916 * If process is waiting for space, 1917 * wakeup/selnotify/signal. If data 1918 * are ready to send, let tcp_output 1919 * decide between more output or persist. 1920 */ 1921 if (tp->snd_una == tp->snd_max) 1922 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1923 else if (TCP_TIMER_ISARMED(tp, 1924 TCPT_PERSIST) == 0) 1925 TCP_TIMER_ARM(tp, TCPT_REXMT, 1926 tp->t_rxtcur); 1927 1928 sowwakeup(so); 1929 if (so->so_snd.sb_cc) { 1930 KERNEL_LOCK(1, NULL); 1931 (void)tcp_output(tp); 1932 KERNEL_UNLOCK_ONE(NULL); 1933 } 1934 if (tcp_saveti) 1935 m_freem(tcp_saveti); 1936 return; 1937 } 1938 } else if (th->th_ack == tp->snd_una && 1939 TAILQ_FIRST(&tp->segq) == NULL && 1940 tlen <= sbspace(&so->so_rcv)) { 1941 int newsize = 0; 1942 1943 /* 1944 * this is a pure, in-sequence data packet 1945 * with nothing on the reassembly queue and 1946 * we have enough buffer space to take it. 1947 */ 1948 tp->rcv_nxt += tlen; 1949 tcps = TCP_STAT_GETREF(); 1950 tcps[TCP_STAT_PREDDAT]++; 1951 tcps[TCP_STAT_RCVPACK]++; 1952 tcps[TCP_STAT_RCVBYTE] += tlen; 1953 TCP_STAT_PUTREF(); 1954 nd6_hint(tp); 1955 1956 /* 1957 * Automatic sizing enables the performance of large buffers 1958 * and most of the efficiency of small ones by only allocating 1959 * space when it is needed. 1960 * 1961 * On the receive side the socket buffer memory is only rarely 1962 * used to any significant extent. This allows us to be much 1963 * more aggressive in scaling the receive socket buffer. For 1964 * the case that the buffer space is actually used to a large 1965 * extent and we run out of kernel memory we can simply drop 1966 * the new segments; TCP on the sender will just retransmit it 1967 * later. Setting the buffer size too big may only consume too 1968 * much kernel memory if the application doesn't read() from 1969 * the socket or packet loss or reordering makes use of the 1970 * reassembly queue. 1971 * 1972 * The criteria to step up the receive buffer one notch are: 1973 * 1. the number of bytes received during the time it takes 1974 * one timestamp to be reflected back to us (the RTT); 1975 * 2. received bytes per RTT is within seven eighth of the 1976 * current socket buffer size; 1977 * 3. receive buffer size has not hit maximal automatic size; 1978 * 1979 * This algorithm does one step per RTT at most and only if 1980 * we receive a bulk stream w/o packet losses or reorderings. 1981 * Shrinking the buffer during idle times is not necessary as 1982 * it doesn't consume any memory when idle. 1983 * 1984 * TODO: Only step up if the application is actually serving 1985 * the buffer to better manage the socket buffer resources. 1986 */ 1987 if (tcp_do_autorcvbuf && 1988 opti.ts_ecr && 1989 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1990 if (opti.ts_ecr > tp->rfbuf_ts && 1991 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 1992 if (tp->rfbuf_cnt > 1993 (so->so_rcv.sb_hiwat / 8 * 7) && 1994 so->so_rcv.sb_hiwat < 1995 tcp_autorcvbuf_max) { 1996 newsize = 1997 min(so->so_rcv.sb_hiwat + 1998 tcp_autorcvbuf_inc, 1999 tcp_autorcvbuf_max); 2000 } 2001 /* Start over with next RTT. */ 2002 tp->rfbuf_ts = 0; 2003 tp->rfbuf_cnt = 0; 2004 } else 2005 tp->rfbuf_cnt += tlen; /* add up */ 2006 } 2007 2008 /* 2009 * Drop TCP, IP headers and TCP options then add data 2010 * to socket buffer. 2011 */ 2012 if (so->so_state & SS_CANTRCVMORE) { 2013 m_freem(m); 2014 } else { 2015 /* 2016 * Set new socket buffer size. 2017 * Give up when limit is reached. 2018 */ 2019 if (newsize) 2020 if (!sbreserve(&so->so_rcv, 2021 newsize, so)) 2022 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 2023 m_adj(m, toff + off); 2024 sbappendstream(&so->so_rcv, m); 2025 } 2026 sorwakeup(so); 2027 tcp_setup_ack(tp, th); 2028 if (tp->t_flags & TF_ACKNOW) { 2029 KERNEL_LOCK(1, NULL); 2030 (void)tcp_output(tp); 2031 KERNEL_UNLOCK_ONE(NULL); 2032 } 2033 if (tcp_saveti) 2034 m_freem(tcp_saveti); 2035 return; 2036 } 2037 } 2038 2039 /* 2040 * Compute mbuf offset to TCP data segment. 2041 */ 2042 hdroptlen = toff + off; 2043 2044 /* 2045 * Calculate amount of space in receive window. Receive window is 2046 * amount of space in rcv queue, but not less than advertised 2047 * window. 2048 */ 2049 { 2050 int win; 2051 win = sbspace(&so->so_rcv); 2052 if (win < 0) 2053 win = 0; 2054 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 2055 } 2056 2057 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 2058 tp->rfbuf_ts = 0; 2059 tp->rfbuf_cnt = 0; 2060 2061 switch (tp->t_state) { 2062 /* 2063 * If the state is SYN_SENT: 2064 * if seg contains an ACK, but not for our SYN, drop the input. 2065 * if seg contains a RST, then drop the connection. 2066 * if seg does not contain SYN, then drop it. 2067 * Otherwise this is an acceptable SYN segment 2068 * initialize tp->rcv_nxt and tp->irs 2069 * if seg contains ack then advance tp->snd_una 2070 * if seg contains a ECE and ECN support is enabled, the stream 2071 * is ECN capable. 2072 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2073 * arrange for segment to be acked (eventually) 2074 * continue processing rest of data/controls, beginning with URG 2075 */ 2076 case TCPS_SYN_SENT: 2077 if ((tiflags & TH_ACK) && 2078 (SEQ_LEQ(th->th_ack, tp->iss) || 2079 SEQ_GT(th->th_ack, tp->snd_max))) 2080 goto dropwithreset; 2081 if (tiflags & TH_RST) { 2082 if (tiflags & TH_ACK) 2083 tp = tcp_drop(tp, ECONNREFUSED); 2084 goto drop; 2085 } 2086 if ((tiflags & TH_SYN) == 0) 2087 goto drop; 2088 if (tiflags & TH_ACK) { 2089 tp->snd_una = th->th_ack; 2090 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2091 tp->snd_nxt = tp->snd_una; 2092 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2093 tp->snd_high = tp->snd_una; 2094 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2095 2096 if ((tiflags & TH_ECE) && tcp_do_ecn) { 2097 tp->t_flags |= TF_ECN_PERMIT; 2098 TCP_STATINC(TCP_STAT_ECN_SHS); 2099 } 2100 } 2101 tp->irs = th->th_seq; 2102 tcp_rcvseqinit(tp); 2103 tp->t_flags |= TF_ACKNOW; 2104 tcp_mss_from_peer(tp, opti.maxseg); 2105 2106 /* 2107 * Initialize the initial congestion window. If we 2108 * had to retransmit the SYN, we must initialize cwnd 2109 * to 1 segment (i.e. the Loss Window). 2110 */ 2111 if (tp->t_flags & TF_SYN_REXMT) 2112 tp->snd_cwnd = tp->t_peermss; 2113 else { 2114 int ss = tcp_init_win; 2115 if (inp != NULL && in_localaddr(inp->inp_faddr)) 2116 ss = tcp_init_win_local; 2117 #ifdef INET6 2118 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 2119 ss = tcp_init_win_local; 2120 #endif 2121 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 2122 } 2123 2124 tcp_rmx_rtt(tp); 2125 if (tiflags & TH_ACK) { 2126 TCP_STATINC(TCP_STAT_CONNECTS); 2127 /* 2128 * move tcp_established before soisconnected 2129 * because upcall handler can drive tcp_output 2130 * functionality. 2131 * XXX we might call soisconnected at the end of 2132 * all processing 2133 */ 2134 tcp_established(tp); 2135 soisconnected(so); 2136 /* Do window scaling on this connection? */ 2137 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2138 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2139 tp->snd_scale = tp->requested_s_scale; 2140 tp->rcv_scale = tp->request_r_scale; 2141 } 2142 TCP_REASS_LOCK(tp); 2143 (void)tcp_reass(tp, NULL, NULL, tlen); 2144 /* 2145 * if we didn't have to retransmit the SYN, 2146 * use its rtt as our initial srtt & rtt var. 2147 */ 2148 if (tp->t_rtttime) 2149 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2150 } else { 2151 tp->t_state = TCPS_SYN_RECEIVED; 2152 } 2153 2154 /* 2155 * Advance th->th_seq to correspond to first data byte. 2156 * If data, trim to stay within window, 2157 * dropping FIN if necessary. 2158 */ 2159 th->th_seq++; 2160 if (tlen > tp->rcv_wnd) { 2161 todrop = tlen - tp->rcv_wnd; 2162 m_adj(m, -todrop); 2163 tlen = tp->rcv_wnd; 2164 tiflags &= ~TH_FIN; 2165 tcps = TCP_STAT_GETREF(); 2166 tcps[TCP_STAT_RCVPACKAFTERWIN]++; 2167 tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop; 2168 TCP_STAT_PUTREF(); 2169 } 2170 tp->snd_wl1 = th->th_seq - 1; 2171 tp->rcv_up = th->th_seq; 2172 goto step6; 2173 2174 /* 2175 * If the state is SYN_RECEIVED: 2176 * If seg contains an ACK, but not for our SYN, drop the input 2177 * and generate an RST. See page 36, rfc793 2178 */ 2179 case TCPS_SYN_RECEIVED: 2180 if ((tiflags & TH_ACK) && 2181 (SEQ_LEQ(th->th_ack, tp->iss) || 2182 SEQ_GT(th->th_ack, tp->snd_max))) 2183 goto dropwithreset; 2184 break; 2185 } 2186 2187 /* 2188 * From here on, we're dealing with !LISTEN and !SYN_SENT. 2189 */ 2190 KASSERT(tp->t_state != TCPS_LISTEN && 2191 tp->t_state != TCPS_SYN_SENT); 2192 2193 /* 2194 * RFC1323 PAWS: if we have a timestamp reply on this segment and 2195 * it's less than ts_recent, drop it. 2196 */ 2197 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2198 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2199 /* Check to see if ts_recent is over 24 days old. */ 2200 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2201 /* 2202 * Invalidate ts_recent. If this segment updates 2203 * ts_recent, the age will be reset later and ts_recent 2204 * will get a valid value. If it does not, setting 2205 * ts_recent to zero will at least satisfy the 2206 * requirement that zero be placed in the timestamp 2207 * echo reply when ts_recent isn't valid. The 2208 * age isn't reset until we get a valid ts_recent 2209 * because we don't want out-of-order segments to be 2210 * dropped when ts_recent is old. 2211 */ 2212 tp->ts_recent = 0; 2213 } else { 2214 tcps = TCP_STAT_GETREF(); 2215 tcps[TCP_STAT_RCVDUPPACK]++; 2216 tcps[TCP_STAT_RCVDUPBYTE] += tlen; 2217 tcps[TCP_STAT_PAWSDROP]++; 2218 TCP_STAT_PUTREF(); 2219 tcp_new_dsack(tp, th->th_seq, tlen); 2220 goto dropafterack; 2221 } 2222 } 2223 2224 /* 2225 * Check that at least some bytes of the segment are within the 2226 * receive window. If segment begins before rcv_nxt, drop leading 2227 * data (and SYN); if nothing left, just ack. 2228 */ 2229 todrop = tp->rcv_nxt - th->th_seq; 2230 dupseg = false; 2231 if (todrop > 0) { 2232 if (tiflags & TH_SYN) { 2233 tiflags &= ~TH_SYN; 2234 th->th_seq++; 2235 tcp_urp_drop(th, 1, &tiflags); 2236 todrop--; 2237 } 2238 if (todrop > tlen || 2239 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2240 /* 2241 * Any valid FIN or RST must be to the left of the 2242 * window. At this point the FIN or RST must be a 2243 * duplicate or out of sequence; drop it. 2244 */ 2245 if (tiflags & TH_RST) 2246 goto drop; 2247 tiflags &= ~(TH_FIN|TH_RST); 2248 2249 /* 2250 * Send an ACK to resynchronize and drop any data. 2251 * But keep on processing for RST or ACK. 2252 */ 2253 tp->t_flags |= TF_ACKNOW; 2254 todrop = tlen; 2255 dupseg = true; 2256 tcps = TCP_STAT_GETREF(); 2257 tcps[TCP_STAT_RCVDUPPACK]++; 2258 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 2259 TCP_STAT_PUTREF(); 2260 } else if ((tiflags & TH_RST) && th->th_seq != tp->rcv_nxt) { 2261 /* 2262 * Test for reset before adjusting the sequence 2263 * number for overlapping data. 2264 */ 2265 goto dropafterack_ratelim; 2266 } else { 2267 tcps = TCP_STAT_GETREF(); 2268 tcps[TCP_STAT_RCVPARTDUPPACK]++; 2269 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 2270 TCP_STAT_PUTREF(); 2271 } 2272 tcp_new_dsack(tp, th->th_seq, todrop); 2273 hdroptlen += todrop; /* drop from head afterwards (m_adj) */ 2274 th->th_seq += todrop; 2275 tlen -= todrop; 2276 tcp_urp_drop(th, todrop, &tiflags); 2277 } 2278 2279 /* 2280 * If new data is received on a connection after the user processes 2281 * are gone, then RST the other end. 2282 */ 2283 if ((so->so_state & SS_NOFDREF) && 2284 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2285 tp = tcp_close(tp); 2286 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 2287 goto dropwithreset; 2288 } 2289 2290 /* 2291 * If the segment ends after the window, drop trailing data (and 2292 * PUSH and FIN); if nothing left, just ACK. 2293 */ 2294 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2295 if (todrop > 0) { 2296 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 2297 if (todrop >= tlen) { 2298 /* 2299 * The segment actually starts after the window. 2300 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2301 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2302 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2303 */ 2304 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 2305 2306 /* 2307 * If a new connection request is received while in 2308 * TIME_WAIT, drop the old connection and start over 2309 * if the sequence numbers are above the previous 2310 * ones. 2311 * 2312 * NOTE: We need to put the header fields back into 2313 * network order. 2314 */ 2315 if ((tiflags & TH_SYN) && 2316 tp->t_state == TCPS_TIME_WAIT && 2317 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2318 tp = tcp_close(tp); 2319 tcp_fields_to_net(th); 2320 m_freem(tcp_saveti); 2321 tcp_saveti = NULL; 2322 goto findpcb; 2323 } 2324 2325 /* 2326 * If window is closed can only take segments at 2327 * window edge, and have to drop data and PUSH from 2328 * incoming segments. Continue processing, but 2329 * remember to ack. Otherwise, drop segment 2330 * and (if not RST) ack. 2331 */ 2332 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2333 KASSERT(todrop == tlen); 2334 tp->t_flags |= TF_ACKNOW; 2335 TCP_STATINC(TCP_STAT_RCVWINPROBE); 2336 } else { 2337 goto dropafterack; 2338 } 2339 } else { 2340 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 2341 } 2342 m_adj(m, -todrop); 2343 tlen -= todrop; 2344 tiflags &= ~(TH_PUSH|TH_FIN); 2345 } 2346 2347 /* 2348 * If last ACK falls within this segment's sequence numbers, 2349 * record the timestamp. 2350 * NOTE: 2351 * 1) That the test incorporates suggestions from the latest 2352 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2353 * 2) That updating only on newer timestamps interferes with 2354 * our earlier PAWS tests, so this check should be solely 2355 * predicated on the sequence space of this segment. 2356 * 3) That we modify the segment boundary check to be 2357 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2358 * instead of RFC1323's 2359 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2360 * This modified check allows us to overcome RFC1323's 2361 * limitations as described in Stevens TCP/IP Illustrated 2362 * Vol. 2 p.869. In such cases, we can still calculate the 2363 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2364 */ 2365 if (opti.ts_present && 2366 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2367 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2368 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2369 tp->ts_recent_age = tcp_now; 2370 tp->ts_recent = opti.ts_val; 2371 } 2372 2373 /* 2374 * If the RST bit is set examine the state: 2375 * RECEIVED state: 2376 * If passive open, return to LISTEN state. 2377 * If active open, inform user that connection was refused. 2378 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT states: 2379 * Inform user that connection was reset, and close tcb. 2380 * CLOSING, LAST_ACK, TIME_WAIT states: 2381 * Close the tcb. 2382 */ 2383 if (tiflags & TH_RST) { 2384 if (th->th_seq != tp->rcv_nxt) 2385 goto dropafterack_ratelim; 2386 2387 switch (tp->t_state) { 2388 case TCPS_SYN_RECEIVED: 2389 so->so_error = ECONNREFUSED; 2390 goto close; 2391 2392 case TCPS_ESTABLISHED: 2393 case TCPS_FIN_WAIT_1: 2394 case TCPS_FIN_WAIT_2: 2395 case TCPS_CLOSE_WAIT: 2396 so->so_error = ECONNRESET; 2397 close: 2398 tp->t_state = TCPS_CLOSED; 2399 TCP_STATINC(TCP_STAT_DROPS); 2400 tp = tcp_close(tp); 2401 goto drop; 2402 2403 case TCPS_CLOSING: 2404 case TCPS_LAST_ACK: 2405 case TCPS_TIME_WAIT: 2406 tp = tcp_close(tp); 2407 goto drop; 2408 } 2409 } 2410 2411 /* 2412 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2413 * we must be in a synchronized state. RFC791 states (under RST 2414 * generation) that any unacceptable segment (an out-of-order SYN 2415 * qualifies) received in a synchronized state must elicit only an 2416 * empty acknowledgment segment ... and the connection remains in 2417 * the same state. 2418 */ 2419 if (tiflags & TH_SYN) { 2420 if (tp->rcv_nxt == th->th_seq) { 2421 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2422 TH_ACK); 2423 if (tcp_saveti) 2424 m_freem(tcp_saveti); 2425 return; 2426 } 2427 2428 goto dropafterack_ratelim; 2429 } 2430 2431 /* 2432 * If the ACK bit is off we drop the segment and return. 2433 */ 2434 if ((tiflags & TH_ACK) == 0) { 2435 if (tp->t_flags & TF_ACKNOW) 2436 goto dropafterack; 2437 goto drop; 2438 } 2439 2440 /* 2441 * From here on, we're doing ACK processing. 2442 */ 2443 2444 switch (tp->t_state) { 2445 /* 2446 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2447 * ESTABLISHED state and continue processing, otherwise 2448 * send an RST. 2449 */ 2450 case TCPS_SYN_RECEIVED: 2451 if (SEQ_GT(tp->snd_una, th->th_ack) || 2452 SEQ_GT(th->th_ack, tp->snd_max)) 2453 goto dropwithreset; 2454 TCP_STATINC(TCP_STAT_CONNECTS); 2455 soisconnected(so); 2456 tcp_established(tp); 2457 /* Do window scaling? */ 2458 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2459 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2460 tp->snd_scale = tp->requested_s_scale; 2461 tp->rcv_scale = tp->request_r_scale; 2462 } 2463 TCP_REASS_LOCK(tp); 2464 (void)tcp_reass(tp, NULL, NULL, tlen); 2465 tp->snd_wl1 = th->th_seq - 1; 2466 /* FALLTHROUGH */ 2467 2468 /* 2469 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2470 * ACKs. If the ack is in the range 2471 * tp->snd_una < th->th_ack <= tp->snd_max 2472 * then advance tp->snd_una to th->th_ack and drop 2473 * data from the retransmission queue. If this ACK reflects 2474 * more up to date window information we update our window information. 2475 */ 2476 case TCPS_ESTABLISHED: 2477 case TCPS_FIN_WAIT_1: 2478 case TCPS_FIN_WAIT_2: 2479 case TCPS_CLOSE_WAIT: 2480 case TCPS_CLOSING: 2481 case TCPS_LAST_ACK: 2482 case TCPS_TIME_WAIT: 2483 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2484 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2485 TCP_STATINC(TCP_STAT_RCVDUPACK); 2486 /* 2487 * If we have outstanding data (other than 2488 * a window probe), this is a completely 2489 * duplicate ack (ie, window info didn't 2490 * change), the ack is the biggest we've 2491 * seen and we've seen exactly our rexmt 2492 * threshhold of them, assume a packet 2493 * has been dropped and retransmit it. 2494 * Kludge snd_nxt & the congestion 2495 * window so we send only this one 2496 * packet. 2497 */ 2498 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2499 th->th_ack != tp->snd_una) 2500 tp->t_dupacks = 0; 2501 else if (tp->t_partialacks < 0 && 2502 (++tp->t_dupacks == tcprexmtthresh || 2503 TCP_FACK_FASTRECOV(tp))) { 2504 /* 2505 * Do the fast retransmit, and adjust 2506 * congestion control paramenters. 2507 */ 2508 if (tp->t_congctl->fast_retransmit(tp, th)) { 2509 /* False fast retransmit */ 2510 break; 2511 } 2512 goto drop; 2513 } else if (tp->t_dupacks > tcprexmtthresh) { 2514 tp->snd_cwnd += tp->t_segsz; 2515 KERNEL_LOCK(1, NULL); 2516 (void)tcp_output(tp); 2517 KERNEL_UNLOCK_ONE(NULL); 2518 goto drop; 2519 } 2520 } else { 2521 /* 2522 * If the ack appears to be very old, only 2523 * allow data that is in-sequence. This 2524 * makes it somewhat more difficult to insert 2525 * forged data by guessing sequence numbers. 2526 * Sent an ack to try to update the send 2527 * sequence number on the other side. 2528 */ 2529 if (tlen && th->th_seq != tp->rcv_nxt && 2530 SEQ_LT(th->th_ack, 2531 tp->snd_una - tp->max_sndwnd)) 2532 goto dropafterack; 2533 } 2534 break; 2535 } 2536 /* 2537 * If the congestion window was inflated to account 2538 * for the other side's cached packets, retract it. 2539 */ 2540 tp->t_congctl->fast_retransmit_newack(tp, th); 2541 2542 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2543 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2544 goto dropafterack; 2545 } 2546 acked = th->th_ack - tp->snd_una; 2547 tcps = TCP_STAT_GETREF(); 2548 tcps[TCP_STAT_RCVACKPACK]++; 2549 tcps[TCP_STAT_RCVACKBYTE] += acked; 2550 TCP_STAT_PUTREF(); 2551 2552 /* 2553 * If we have a timestamp reply, update smoothed 2554 * round trip time. If no timestamp is present but 2555 * transmit timer is running and timed sequence 2556 * number was acked, update smoothed round trip time. 2557 * Since we now have an rtt measurement, cancel the 2558 * timer backoff (cf., Phil Karn's retransmit alg.). 2559 * Recompute the initial retransmit timer. 2560 */ 2561 if (ts_rtt) 2562 tcp_xmit_timer(tp, ts_rtt - 1); 2563 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2564 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2565 2566 /* 2567 * If all outstanding data is acked, stop retransmit 2568 * timer and remember to restart (more output or persist). 2569 * If there is more data to be acked, restart retransmit 2570 * timer, using current (possibly backed-off) value. 2571 */ 2572 if (th->th_ack == tp->snd_max) { 2573 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2574 needoutput = 1; 2575 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2576 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2577 2578 /* 2579 * New data has been acked, adjust the congestion window. 2580 */ 2581 tp->t_congctl->newack(tp, th); 2582 2583 nd6_hint(tp); 2584 if (acked > so->so_snd.sb_cc) { 2585 tp->snd_wnd -= so->so_snd.sb_cc; 2586 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2587 ourfinisacked = 1; 2588 } else { 2589 if (acked > (tp->t_lastoff - tp->t_inoff)) 2590 tp->t_lastm = NULL; 2591 sbdrop(&so->so_snd, acked); 2592 tp->t_lastoff -= acked; 2593 if (tp->snd_wnd > acked) 2594 tp->snd_wnd -= acked; 2595 else 2596 tp->snd_wnd = 0; 2597 ourfinisacked = 0; 2598 } 2599 sowwakeup(so); 2600 2601 icmp_check(tp, th, acked); 2602 2603 tp->snd_una = th->th_ack; 2604 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2605 tp->snd_fack = tp->snd_una; 2606 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2607 tp->snd_nxt = tp->snd_una; 2608 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2609 tp->snd_high = tp->snd_una; 2610 2611 switch (tp->t_state) { 2612 2613 /* 2614 * In FIN_WAIT_1 STATE in addition to the processing 2615 * for the ESTABLISHED state if our FIN is now acknowledged 2616 * then enter FIN_WAIT_2. 2617 */ 2618 case TCPS_FIN_WAIT_1: 2619 if (ourfinisacked) { 2620 /* 2621 * If we can't receive any more 2622 * data, then closing user can proceed. 2623 * Starting the timer is contrary to the 2624 * specification, but if we don't get a FIN 2625 * we'll hang forever. 2626 */ 2627 if (so->so_state & SS_CANTRCVMORE) { 2628 soisdisconnected(so); 2629 if (tp->t_maxidle > 0) 2630 TCP_TIMER_ARM(tp, TCPT_2MSL, 2631 tp->t_maxidle); 2632 } 2633 tp->t_state = TCPS_FIN_WAIT_2; 2634 } 2635 break; 2636 2637 /* 2638 * In CLOSING STATE in addition to the processing for 2639 * the ESTABLISHED state if the ACK acknowledges our FIN 2640 * then enter the TIME-WAIT state, otherwise ignore 2641 * the segment. 2642 */ 2643 case TCPS_CLOSING: 2644 if (ourfinisacked) { 2645 tp->t_state = TCPS_TIME_WAIT; 2646 tcp_canceltimers(tp); 2647 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2648 soisdisconnected(so); 2649 } 2650 break; 2651 2652 /* 2653 * In LAST_ACK, we may still be waiting for data to drain 2654 * and/or to be acked, as well as for the ack of our FIN. 2655 * If our FIN is now acknowledged, delete the TCB, 2656 * enter the closed state and return. 2657 */ 2658 case TCPS_LAST_ACK: 2659 if (ourfinisacked) { 2660 tp = tcp_close(tp); 2661 goto drop; 2662 } 2663 break; 2664 2665 /* 2666 * In TIME_WAIT state the only thing that should arrive 2667 * is a retransmission of the remote FIN. Acknowledge 2668 * it and restart the finack timer. 2669 */ 2670 case TCPS_TIME_WAIT: 2671 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2672 goto dropafterack; 2673 } 2674 } 2675 2676 step6: 2677 /* 2678 * Update window information. 2679 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2680 */ 2681 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2682 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2683 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2684 /* keep track of pure window updates */ 2685 if (tlen == 0 && 2686 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2687 TCP_STATINC(TCP_STAT_RCVWINUPD); 2688 tp->snd_wnd = tiwin; 2689 tp->snd_wl1 = th->th_seq; 2690 tp->snd_wl2 = th->th_ack; 2691 if (tp->snd_wnd > tp->max_sndwnd) 2692 tp->max_sndwnd = tp->snd_wnd; 2693 needoutput = 1; 2694 } 2695 2696 /* 2697 * Process segments with URG. 2698 */ 2699 if ((tiflags & TH_URG) && th->th_urp && 2700 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2701 /* 2702 * This is a kludge, but if we receive and accept 2703 * random urgent pointers, we'll crash in 2704 * soreceive. It's hard to imagine someone 2705 * actually wanting to send this much urgent data. 2706 */ 2707 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2708 th->th_urp = 0; /* XXX */ 2709 tiflags &= ~TH_URG; /* XXX */ 2710 goto dodata; /* XXX */ 2711 } 2712 2713 /* 2714 * If this segment advances the known urgent pointer, 2715 * then mark the data stream. This should not happen 2716 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2717 * a FIN has been received from the remote side. 2718 * In these states we ignore the URG. 2719 * 2720 * According to RFC961 (Assigned Protocols), 2721 * the urgent pointer points to the last octet 2722 * of urgent data. We continue, however, 2723 * to consider it to indicate the first octet 2724 * of data past the urgent section as the original 2725 * spec states (in one of two places). 2726 */ 2727 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2728 tp->rcv_up = th->th_seq + th->th_urp; 2729 so->so_oobmark = so->so_rcv.sb_cc + 2730 (tp->rcv_up - tp->rcv_nxt) - 1; 2731 if (so->so_oobmark == 0) 2732 so->so_state |= SS_RCVATMARK; 2733 sohasoutofband(so); 2734 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2735 } 2736 2737 /* 2738 * Remove out of band data so doesn't get presented to user. 2739 * This can happen independent of advancing the URG pointer, 2740 * but if two URG's are pending at once, some out-of-band 2741 * data may creep in... ick. 2742 */ 2743 if (th->th_urp <= (u_int16_t)tlen && 2744 (so->so_options & SO_OOBINLINE) == 0) 2745 tcp_pulloutofband(so, th, m, hdroptlen); 2746 } else { 2747 /* 2748 * If no out of band data is expected, 2749 * pull receive urgent pointer along 2750 * with the receive window. 2751 */ 2752 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2753 tp->rcv_up = tp->rcv_nxt; 2754 } 2755 dodata: 2756 2757 /* 2758 * Process the segment text, merging it into the TCP sequencing queue, 2759 * and arranging for acknowledgement of receipt if necessary. 2760 * This process logically involves adjusting tp->rcv_wnd as data 2761 * is presented to the user (this happens in tcp_usrreq.c, 2762 * tcp_rcvd()). If a FIN has already been received on this 2763 * connection then we just ignore the text. 2764 */ 2765 if ((tlen || (tiflags & TH_FIN)) && 2766 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2767 /* 2768 * Handle the common case: 2769 * o Segment is the next to be received, and 2770 * o The queue is empty, and 2771 * o The connection is established 2772 * In this case, we avoid calling tcp_reass. 2773 * 2774 * tcp_setup_ack: set DELACK for segments received in order, 2775 * but ack immediately when segments are out of order (so that 2776 * fast retransmit can work). 2777 */ 2778 TCP_REASS_LOCK(tp); 2779 if (th->th_seq == tp->rcv_nxt && 2780 TAILQ_FIRST(&tp->segq) == NULL && 2781 tp->t_state == TCPS_ESTABLISHED) { 2782 tcp_setup_ack(tp, th); 2783 tp->rcv_nxt += tlen; 2784 tiflags = th->th_flags & TH_FIN; 2785 tcps = TCP_STAT_GETREF(); 2786 tcps[TCP_STAT_RCVPACK]++; 2787 tcps[TCP_STAT_RCVBYTE] += tlen; 2788 TCP_STAT_PUTREF(); 2789 nd6_hint(tp); 2790 if (so->so_state & SS_CANTRCVMORE) { 2791 m_freem(m); 2792 } else { 2793 m_adj(m, hdroptlen); 2794 sbappendstream(&(so)->so_rcv, m); 2795 } 2796 TCP_REASS_UNLOCK(tp); 2797 sorwakeup(so); 2798 } else { 2799 m_adj(m, hdroptlen); 2800 tiflags = tcp_reass(tp, th, m, tlen); 2801 tp->t_flags |= TF_ACKNOW; 2802 } 2803 2804 /* 2805 * Note the amount of data that peer has sent into 2806 * our window, in order to estimate the sender's 2807 * buffer size. 2808 */ 2809 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2810 } else { 2811 m_freem(m); 2812 m = NULL; 2813 tiflags &= ~TH_FIN; 2814 } 2815 2816 /* 2817 * If FIN is received ACK the FIN and let the user know 2818 * that the connection is closing. Ignore a FIN received before 2819 * the connection is fully established. 2820 */ 2821 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2822 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2823 socantrcvmore(so); 2824 tp->t_flags |= TF_ACKNOW; 2825 tp->rcv_nxt++; 2826 } 2827 switch (tp->t_state) { 2828 2829 /* 2830 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2831 */ 2832 case TCPS_ESTABLISHED: 2833 tp->t_state = TCPS_CLOSE_WAIT; 2834 break; 2835 2836 /* 2837 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2838 * enter the CLOSING state. 2839 */ 2840 case TCPS_FIN_WAIT_1: 2841 tp->t_state = TCPS_CLOSING; 2842 break; 2843 2844 /* 2845 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2846 * starting the time-wait timer, turning off the other 2847 * standard timers. 2848 */ 2849 case TCPS_FIN_WAIT_2: 2850 tp->t_state = TCPS_TIME_WAIT; 2851 tcp_canceltimers(tp); 2852 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2853 soisdisconnected(so); 2854 break; 2855 2856 /* 2857 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2858 */ 2859 case TCPS_TIME_WAIT: 2860 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2861 break; 2862 } 2863 } 2864 #ifdef TCP_DEBUG 2865 if (so->so_options & SO_DEBUG) 2866 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2867 #endif 2868 2869 /* 2870 * Return any desired output. 2871 */ 2872 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2873 KERNEL_LOCK(1, NULL); 2874 (void)tcp_output(tp); 2875 KERNEL_UNLOCK_ONE(NULL); 2876 } 2877 if (tcp_saveti) 2878 m_freem(tcp_saveti); 2879 2880 if (tp->t_state == TCPS_TIME_WAIT 2881 && (so->so_state & SS_NOFDREF) 2882 && (tp->t_inpcb || af != AF_INET) 2883 && (tp->t_in6pcb || af != AF_INET6) 2884 && ((af == AF_INET ? tcp4_vtw_enable : tcp6_vtw_enable) & 1) != 0 2885 && TAILQ_EMPTY(&tp->segq) 2886 && vtw_add(af, tp)) { 2887 ; 2888 } 2889 return; 2890 2891 badsyn: 2892 /* 2893 * Received a bad SYN. Increment counters and dropwithreset. 2894 */ 2895 TCP_STATINC(TCP_STAT_BADSYN); 2896 tp = NULL; 2897 goto dropwithreset; 2898 2899 dropafterack: 2900 /* 2901 * Generate an ACK dropping incoming segment if it occupies 2902 * sequence space, where the ACK reflects our state. 2903 */ 2904 if (tiflags & TH_RST) 2905 goto drop; 2906 goto dropafterack2; 2907 2908 dropafterack_ratelim: 2909 /* 2910 * We may want to rate-limit ACKs against SYN/RST attack. 2911 */ 2912 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2913 tcp_ackdrop_ppslim) == 0) { 2914 /* XXX stat */ 2915 goto drop; 2916 } 2917 2918 dropafterack2: 2919 m_freem(m); 2920 tp->t_flags |= TF_ACKNOW; 2921 KERNEL_LOCK(1, NULL); 2922 (void)tcp_output(tp); 2923 KERNEL_UNLOCK_ONE(NULL); 2924 if (tcp_saveti) 2925 m_freem(tcp_saveti); 2926 return; 2927 2928 dropwithreset_ratelim: 2929 /* 2930 * We may want to rate-limit RSTs in certain situations, 2931 * particularly if we are sending an RST in response to 2932 * an attempt to connect to or otherwise communicate with 2933 * a port for which we have no socket. 2934 */ 2935 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2936 tcp_rst_ppslim) == 0) { 2937 /* XXX stat */ 2938 goto drop; 2939 } 2940 2941 dropwithreset: 2942 /* 2943 * Generate a RST, dropping incoming segment. 2944 * Make ACK acceptable to originator of segment. 2945 */ 2946 if (tiflags & TH_RST) 2947 goto drop; 2948 if (tiflags & TH_ACK) { 2949 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2950 } else { 2951 if (tiflags & TH_SYN) 2952 tlen++; 2953 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2954 TH_RST|TH_ACK); 2955 } 2956 if (tcp_saveti) 2957 m_freem(tcp_saveti); 2958 return; 2959 2960 badcsum: 2961 drop: 2962 /* 2963 * Drop space held by incoming segment and return. 2964 */ 2965 if (tp) { 2966 if (tp->t_inpcb) 2967 so = tp->t_inpcb->inp_socket; 2968 #ifdef INET6 2969 else if (tp->t_in6pcb) 2970 so = tp->t_in6pcb->in6p_socket; 2971 #endif 2972 else 2973 so = NULL; 2974 #ifdef TCP_DEBUG 2975 if (so && (so->so_options & SO_DEBUG) != 0) 2976 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2977 #endif 2978 } 2979 if (tcp_saveti) 2980 m_freem(tcp_saveti); 2981 m_freem(m); 2982 return; 2983 } 2984 2985 #ifdef TCP_SIGNATURE 2986 int 2987 tcp_signature_apply(void *fstate, void *data, u_int len) 2988 { 2989 2990 MD5Update(fstate, (u_char *)data, len); 2991 return (0); 2992 } 2993 2994 struct secasvar * 2995 tcp_signature_getsav(struct mbuf *m) 2996 { 2997 struct ip *ip; 2998 struct ip6_hdr *ip6; 2999 3000 ip = mtod(m, struct ip *); 3001 switch (ip->ip_v) { 3002 case 4: 3003 ip = mtod(m, struct ip *); 3004 ip6 = NULL; 3005 break; 3006 case 6: 3007 ip = NULL; 3008 ip6 = mtod(m, struct ip6_hdr *); 3009 break; 3010 default: 3011 return (NULL); 3012 } 3013 3014 #ifdef IPSEC 3015 union sockaddr_union dst; 3016 3017 /* Extract the destination from the IP header in the mbuf. */ 3018 memset(&dst, 0, sizeof(union sockaddr_union)); 3019 if (ip != NULL) { 3020 dst.sa.sa_len = sizeof(struct sockaddr_in); 3021 dst.sa.sa_family = AF_INET; 3022 dst.sin.sin_addr = ip->ip_dst; 3023 } else { 3024 dst.sa.sa_len = sizeof(struct sockaddr_in6); 3025 dst.sa.sa_family = AF_INET6; 3026 dst.sin6.sin6_addr = ip6->ip6_dst; 3027 } 3028 3029 /* 3030 * Look up an SADB entry which matches the address of the peer. 3031 */ 3032 return KEY_LOOKUP_SA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI), 0, 0); 3033 #else 3034 return NULL; 3035 #endif 3036 } 3037 3038 int 3039 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 3040 struct secasvar *sav, char *sig) 3041 { 3042 MD5_CTX ctx; 3043 struct ip *ip; 3044 struct ipovly *ipovly; 3045 #ifdef INET6 3046 struct ip6_hdr *ip6; 3047 struct ip6_hdr_pseudo ip6pseudo; 3048 #endif 3049 struct ippseudo ippseudo; 3050 struct tcphdr th0; 3051 int l, tcphdrlen; 3052 3053 if (sav == NULL) 3054 return (-1); 3055 3056 tcphdrlen = th->th_off * 4; 3057 3058 switch (mtod(m, struct ip *)->ip_v) { 3059 case 4: 3060 MD5Init(&ctx); 3061 ip = mtod(m, struct ip *); 3062 memset(&ippseudo, 0, sizeof(ippseudo)); 3063 ipovly = (struct ipovly *)ip; 3064 ippseudo.ippseudo_src = ipovly->ih_src; 3065 ippseudo.ippseudo_dst = ipovly->ih_dst; 3066 ippseudo.ippseudo_pad = 0; 3067 ippseudo.ippseudo_p = IPPROTO_TCP; 3068 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 3069 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 3070 break; 3071 #if INET6 3072 case 6: 3073 MD5Init(&ctx); 3074 ip6 = mtod(m, struct ip6_hdr *); 3075 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 3076 ip6pseudo.ip6ph_src = ip6->ip6_src; 3077 in6_clearscope(&ip6pseudo.ip6ph_src); 3078 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 3079 in6_clearscope(&ip6pseudo.ip6ph_dst); 3080 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 3081 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 3082 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 3083 break; 3084 #endif 3085 default: 3086 return (-1); 3087 } 3088 3089 th0 = *th; 3090 th0.th_sum = 0; 3091 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 3092 3093 l = m->m_pkthdr.len - thoff - tcphdrlen; 3094 if (l > 0) 3095 m_apply(m, thoff + tcphdrlen, 3096 m->m_pkthdr.len - thoff - tcphdrlen, 3097 tcp_signature_apply, &ctx); 3098 3099 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 3100 MD5Final(sig, &ctx); 3101 3102 return (0); 3103 } 3104 #endif 3105 3106 /* 3107 * Parse and process tcp options. 3108 * 3109 * Returns -1 if this segment should be dropped. (eg. wrong signature) 3110 * Otherwise returns 0. 3111 */ 3112 static int 3113 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, struct tcphdr *th, 3114 struct mbuf *m, int toff, struct tcp_opt_info *oi) 3115 { 3116 u_int16_t mss; 3117 int opt, optlen = 0; 3118 #ifdef TCP_SIGNATURE 3119 void *sigp = NULL; 3120 char sigbuf[TCP_SIGLEN]; 3121 struct secasvar *sav = NULL; 3122 #endif 3123 3124 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 3125 opt = cp[0]; 3126 if (opt == TCPOPT_EOL) 3127 break; 3128 if (opt == TCPOPT_NOP) 3129 optlen = 1; 3130 else { 3131 if (cnt < 2) 3132 break; 3133 optlen = cp[1]; 3134 if (optlen < 2 || optlen > cnt) 3135 break; 3136 } 3137 switch (opt) { 3138 3139 default: 3140 continue; 3141 3142 case TCPOPT_MAXSEG: 3143 if (optlen != TCPOLEN_MAXSEG) 3144 continue; 3145 if (!(th->th_flags & TH_SYN)) 3146 continue; 3147 if (TCPS_HAVERCVDSYN(tp->t_state)) 3148 continue; 3149 memcpy(&mss, cp + 2, sizeof(mss)); 3150 oi->maxseg = ntohs(mss); 3151 break; 3152 3153 case TCPOPT_WINDOW: 3154 if (optlen != TCPOLEN_WINDOW) 3155 continue; 3156 if (!(th->th_flags & TH_SYN)) 3157 continue; 3158 if (TCPS_HAVERCVDSYN(tp->t_state)) 3159 continue; 3160 tp->t_flags |= TF_RCVD_SCALE; 3161 tp->requested_s_scale = cp[2]; 3162 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3163 char buf[INET6_ADDRSTRLEN]; 3164 struct ip *ip = mtod(m, struct ip *); 3165 #ifdef INET6 3166 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 3167 #endif 3168 3169 switch (ip->ip_v) { 3170 case 4: 3171 in_print(buf, sizeof(buf), 3172 &ip->ip_src); 3173 break; 3174 #ifdef INET6 3175 case 6: 3176 in6_print(buf, sizeof(buf), 3177 &ip6->ip6_src); 3178 break; 3179 #endif 3180 default: 3181 strlcpy(buf, "(unknown)", sizeof(buf)); 3182 break; 3183 } 3184 3185 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3186 "assuming %d\n", 3187 tp->requested_s_scale, buf, 3188 TCP_MAX_WINSHIFT); 3189 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3190 } 3191 break; 3192 3193 case TCPOPT_TIMESTAMP: 3194 if (optlen != TCPOLEN_TIMESTAMP) 3195 continue; 3196 oi->ts_present = 1; 3197 memcpy(&oi->ts_val, cp + 2, sizeof(oi->ts_val)); 3198 NTOHL(oi->ts_val); 3199 memcpy(&oi->ts_ecr, cp + 6, sizeof(oi->ts_ecr)); 3200 NTOHL(oi->ts_ecr); 3201 3202 if (!(th->th_flags & TH_SYN)) 3203 continue; 3204 if (TCPS_HAVERCVDSYN(tp->t_state)) 3205 continue; 3206 /* 3207 * A timestamp received in a SYN makes 3208 * it ok to send timestamp requests and replies. 3209 */ 3210 tp->t_flags |= TF_RCVD_TSTMP; 3211 tp->ts_recent = oi->ts_val; 3212 tp->ts_recent_age = tcp_now; 3213 break; 3214 3215 case TCPOPT_SACK_PERMITTED: 3216 if (optlen != TCPOLEN_SACK_PERMITTED) 3217 continue; 3218 if (!(th->th_flags & TH_SYN)) 3219 continue; 3220 if (TCPS_HAVERCVDSYN(tp->t_state)) 3221 continue; 3222 if (tcp_do_sack) { 3223 tp->t_flags |= TF_SACK_PERMIT; 3224 tp->t_flags |= TF_WILL_SACK; 3225 } 3226 break; 3227 3228 case TCPOPT_SACK: 3229 tcp_sack_option(tp, th, cp, optlen); 3230 break; 3231 #ifdef TCP_SIGNATURE 3232 case TCPOPT_SIGNATURE: 3233 if (optlen != TCPOLEN_SIGNATURE) 3234 continue; 3235 if (sigp && 3236 !consttime_memequal(sigp, cp + 2, TCP_SIGLEN)) 3237 return (-1); 3238 3239 sigp = sigbuf; 3240 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3241 tp->t_flags |= TF_SIGNATURE; 3242 break; 3243 #endif 3244 } 3245 } 3246 3247 #ifndef TCP_SIGNATURE 3248 return 0; 3249 #else 3250 if (tp->t_flags & TF_SIGNATURE) { 3251 sav = tcp_signature_getsav(m); 3252 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3253 return (-1); 3254 } 3255 3256 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) 3257 goto out; 3258 3259 if (sigp) { 3260 char sig[TCP_SIGLEN]; 3261 3262 tcp_fields_to_net(th); 3263 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3264 tcp_fields_to_host(th); 3265 goto out; 3266 } 3267 tcp_fields_to_host(th); 3268 3269 if (!consttime_memequal(sig, sigp, TCP_SIGLEN)) { 3270 TCP_STATINC(TCP_STAT_BADSIG); 3271 goto out; 3272 } else 3273 TCP_STATINC(TCP_STAT_GOODSIG); 3274 3275 key_sa_recordxfer(sav, m); 3276 KEY_SA_UNREF(&sav); 3277 } 3278 return 0; 3279 out: 3280 if (sav != NULL) 3281 KEY_SA_UNREF(&sav); 3282 return -1; 3283 #endif 3284 } 3285 3286 /* 3287 * Pull out of band byte out of a segment so 3288 * it doesn't appear in the user's data queue. 3289 * It is still reflected in the segment length for 3290 * sequencing purposes. 3291 */ 3292 void 3293 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3294 struct mbuf *m, int off) 3295 { 3296 int cnt = off + th->th_urp - 1; 3297 3298 while (cnt >= 0) { 3299 if (m->m_len > cnt) { 3300 char *cp = mtod(m, char *) + cnt; 3301 struct tcpcb *tp = sototcpcb(so); 3302 3303 tp->t_iobc = *cp; 3304 tp->t_oobflags |= TCPOOB_HAVEDATA; 3305 memmove(cp, cp + 1, (unsigned)(m->m_len - cnt - 1)); 3306 m->m_len--; 3307 return; 3308 } 3309 cnt -= m->m_len; 3310 m = m->m_next; 3311 if (m == NULL) 3312 break; 3313 } 3314 panic("tcp_pulloutofband"); 3315 } 3316 3317 /* 3318 * Collect new round-trip time estimate 3319 * and update averages and current timeout. 3320 * 3321 * rtt is in units of slow ticks (typically 500 ms) -- essentially the 3322 * difference of two timestamps. 3323 */ 3324 void 3325 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3326 { 3327 int32_t delta; 3328 3329 TCP_STATINC(TCP_STAT_RTTUPDATED); 3330 if (tp->t_srtt != 0) { 3331 /* 3332 * Compute the amount to add to srtt for smoothing, 3333 * *alpha, or 2^(-TCP_RTT_SHIFT). Because 3334 * srtt is stored in 1/32 slow ticks, we conceptually 3335 * shift left 5 bits, subtract srtt to get the 3336 * diference, and then shift right by TCP_RTT_SHIFT 3337 * (3) to obtain 1/8 of the difference. 3338 */ 3339 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3340 /* 3341 * This can never happen, because delta's lowest 3342 * possible value is 1/8 of t_srtt. But if it does, 3343 * set srtt to some reasonable value, here chosen 3344 * as 1/8 tick. 3345 */ 3346 if ((tp->t_srtt += delta) <= 0) 3347 tp->t_srtt = 1 << 2; 3348 /* 3349 * RFC2988 requires that rttvar be updated first. 3350 * This code is compliant because "delta" is the old 3351 * srtt minus the new observation (scaled). 3352 * 3353 * RFC2988 says: 3354 * rttvar = (1-beta) * rttvar + beta * |srtt-observed| 3355 * 3356 * delta is in units of 1/32 ticks, and has then been 3357 * divided by 8. This is equivalent to being in 1/16s 3358 * units and divided by 4. Subtract from it 1/4 of 3359 * the existing rttvar to form the (signed) amount to 3360 * adjust. 3361 */ 3362 if (delta < 0) 3363 delta = -delta; 3364 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3365 /* 3366 * As with srtt, this should never happen. There is 3367 * no support in RFC2988 for this operation. But 1/4s 3368 * as rttvar when faced with something arguably wrong 3369 * is ok. 3370 */ 3371 if ((tp->t_rttvar += delta) <= 0) 3372 tp->t_rttvar = 1 << 2; 3373 3374 /* 3375 * If srtt exceeds .01 second, ensure we use the 'remote' MSL 3376 * Problem is: it doesn't work. Disabled by defaulting 3377 * tcp_rttlocal to 0; see corresponding code in 3378 * tcp_subr that selects local vs remote in a different way. 3379 * 3380 * The static branch prediction hint here should be removed 3381 * when the rtt estimator is fixed and the rtt_enable code 3382 * is turned back on. 3383 */ 3384 if (__predict_false(tcp_rttlocal) && tcp_msl_enable 3385 && tp->t_srtt > tcp_msl_remote_threshold 3386 && tp->t_msl < tcp_msl_remote) { 3387 tp->t_msl = tcp_msl_remote; 3388 } 3389 } else { 3390 /* 3391 * This is the first measurement. Per RFC2988, 2.2, 3392 * set rtt=R and srtt=R/2. 3393 * For srtt, storage representation is 1/32 ticks, 3394 * so shift left by 5. 3395 * For rttvar, storage representation is 1/16 ticks, 3396 * So shift left by 4, but then right by 1 to halve. 3397 */ 3398 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3399 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3400 } 3401 tp->t_rtttime = 0; 3402 tp->t_rxtshift = 0; 3403 3404 /* 3405 * the retransmit should happen at rtt + 4 * rttvar. 3406 * Because of the way we do the smoothing, srtt and rttvar 3407 * will each average +1/2 tick of bias. When we compute 3408 * the retransmit timer, we want 1/2 tick of rounding and 3409 * 1 extra tick because of +-1/2 tick uncertainty in the 3410 * firing of the timer. The bias will give us exactly the 3411 * 1.5 tick we need. But, because the bias is 3412 * statistical, we have to test that we don't drop below 3413 * the minimum feasible timer (which is 2 ticks). 3414 */ 3415 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3416 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3417 3418 /* 3419 * We received an ack for a packet that wasn't retransmitted; 3420 * it is probably safe to discard any error indications we've 3421 * received recently. This isn't quite right, but close enough 3422 * for now (a route might have failed after we sent a segment, 3423 * and the return path might not be symmetrical). 3424 */ 3425 tp->t_softerror = 0; 3426 } 3427 3428 3429 /* 3430 * TCP compressed state engine. Currently used to hold compressed 3431 * state for SYN_RECEIVED. 3432 */ 3433 3434 u_long syn_cache_count; 3435 u_int32_t syn_hash1, syn_hash2; 3436 3437 #define SYN_HASH(sa, sp, dp) \ 3438 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3439 ((u_int32_t)(sp)))^syn_hash2))) 3440 #ifndef INET6 3441 #define SYN_HASHALL(hash, src, dst) \ 3442 do { \ 3443 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3444 ((const struct sockaddr_in *)(src))->sin_port, \ 3445 ((const struct sockaddr_in *)(dst))->sin_port); \ 3446 } while (/*CONSTCOND*/ 0) 3447 #else 3448 #define SYN_HASH6(sa, sp, dp) \ 3449 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3450 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3451 & 0x7fffffff) 3452 3453 #define SYN_HASHALL(hash, src, dst) \ 3454 do { \ 3455 switch ((src)->sa_family) { \ 3456 case AF_INET: \ 3457 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3458 ((const struct sockaddr_in *)(src))->sin_port, \ 3459 ((const struct sockaddr_in *)(dst))->sin_port); \ 3460 break; \ 3461 case AF_INET6: \ 3462 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3463 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3464 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3465 break; \ 3466 default: \ 3467 hash = 0; \ 3468 } \ 3469 } while (/*CONSTCOND*/0) 3470 #endif /* INET6 */ 3471 3472 static struct pool syn_cache_pool; 3473 3474 /* 3475 * We don't estimate RTT with SYNs, so each packet starts with the default 3476 * RTT and each timer step has a fixed timeout value. 3477 */ 3478 static inline void 3479 syn_cache_timer_arm(struct syn_cache *sc) 3480 { 3481 3482 TCPT_RANGESET(sc->sc_rxtcur, 3483 TCPTV_SRTTDFLT * tcp_backoff[sc->sc_rxtshift], TCPTV_MIN, 3484 TCPTV_REXMTMAX); 3485 callout_reset(&sc->sc_timer, 3486 sc->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, sc); 3487 } 3488 3489 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3490 3491 static inline void 3492 syn_cache_rm(struct syn_cache *sc) 3493 { 3494 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3495 sc, sc_bucketq); 3496 sc->sc_tp = NULL; 3497 LIST_REMOVE(sc, sc_tpq); 3498 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3499 callout_stop(&sc->sc_timer); 3500 syn_cache_count--; 3501 } 3502 3503 static inline void 3504 syn_cache_put(struct syn_cache *sc) 3505 { 3506 if (sc->sc_ipopts) 3507 (void) m_free(sc->sc_ipopts); 3508 rtcache_free(&sc->sc_route); 3509 sc->sc_flags |= SCF_DEAD; 3510 if (!callout_invoking(&sc->sc_timer)) 3511 callout_schedule(&(sc)->sc_timer, 1); 3512 } 3513 3514 void 3515 syn_cache_init(void) 3516 { 3517 int i; 3518 3519 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 3520 "synpl", NULL, IPL_SOFTNET); 3521 3522 /* Initialize the hash buckets. */ 3523 for (i = 0; i < tcp_syn_cache_size; i++) 3524 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3525 } 3526 3527 void 3528 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3529 { 3530 struct syn_cache_head *scp; 3531 struct syn_cache *sc2; 3532 int s; 3533 3534 /* 3535 * If there are no entries in the hash table, reinitialize 3536 * the hash secrets. 3537 */ 3538 if (syn_cache_count == 0) { 3539 syn_hash1 = cprng_fast32(); 3540 syn_hash2 = cprng_fast32(); 3541 } 3542 3543 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3544 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3545 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3546 3547 /* 3548 * Make sure that we don't overflow the per-bucket 3549 * limit or the total cache size limit. 3550 */ 3551 s = splsoftnet(); 3552 if (scp->sch_length >= tcp_syn_bucket_limit) { 3553 TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW); 3554 /* 3555 * The bucket is full. Toss the oldest element in the 3556 * bucket. This will be the first entry in the bucket. 3557 */ 3558 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3559 #ifdef DIAGNOSTIC 3560 /* 3561 * This should never happen; we should always find an 3562 * entry in our bucket. 3563 */ 3564 if (sc2 == NULL) 3565 panic("syn_cache_insert: bucketoverflow: impossible"); 3566 #endif 3567 syn_cache_rm(sc2); 3568 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3569 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3570 struct syn_cache_head *scp2, *sce; 3571 3572 TCP_STATINC(TCP_STAT_SC_OVERFLOWED); 3573 /* 3574 * The cache is full. Toss the oldest entry in the 3575 * first non-empty bucket we can find. 3576 * 3577 * XXX We would really like to toss the oldest 3578 * entry in the cache, but we hope that this 3579 * condition doesn't happen very often. 3580 */ 3581 scp2 = scp; 3582 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3583 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3584 for (++scp2; scp2 != scp; scp2++) { 3585 if (scp2 >= sce) 3586 scp2 = &tcp_syn_cache[0]; 3587 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3588 break; 3589 } 3590 #ifdef DIAGNOSTIC 3591 /* 3592 * This should never happen; we should always find a 3593 * non-empty bucket. 3594 */ 3595 if (scp2 == scp) 3596 panic("syn_cache_insert: cacheoverflow: " 3597 "impossible"); 3598 #endif 3599 } 3600 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3601 syn_cache_rm(sc2); 3602 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3603 } 3604 3605 /* 3606 * Initialize the entry's timer. 3607 */ 3608 sc->sc_rxttot = 0; 3609 sc->sc_rxtshift = 0; 3610 syn_cache_timer_arm(sc); 3611 3612 /* Link it from tcpcb entry */ 3613 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3614 3615 /* Put it into the bucket. */ 3616 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3617 scp->sch_length++; 3618 syn_cache_count++; 3619 3620 TCP_STATINC(TCP_STAT_SC_ADDED); 3621 splx(s); 3622 } 3623 3624 /* 3625 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3626 * If we have retransmitted an entry the maximum number of times, expire 3627 * that entry. 3628 */ 3629 static void 3630 syn_cache_timer(void *arg) 3631 { 3632 struct syn_cache *sc = arg; 3633 3634 mutex_enter(softnet_lock); 3635 KERNEL_LOCK(1, NULL); 3636 3637 callout_ack(&sc->sc_timer); 3638 3639 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3640 TCP_STATINC(TCP_STAT_SC_DELAYED_FREE); 3641 goto free; 3642 } 3643 3644 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3645 /* Drop it -- too many retransmissions. */ 3646 goto dropit; 3647 } 3648 3649 /* 3650 * Compute the total amount of time this entry has 3651 * been on a queue. If this entry has been on longer 3652 * than the keep alive timer would allow, expire it. 3653 */ 3654 sc->sc_rxttot += sc->sc_rxtcur; 3655 if (sc->sc_rxttot >= tcp_keepinit) 3656 goto dropit; 3657 3658 TCP_STATINC(TCP_STAT_SC_RETRANSMITTED); 3659 (void)syn_cache_respond(sc); 3660 3661 /* Advance the timer back-off. */ 3662 sc->sc_rxtshift++; 3663 syn_cache_timer_arm(sc); 3664 3665 goto out; 3666 3667 dropit: 3668 TCP_STATINC(TCP_STAT_SC_TIMED_OUT); 3669 syn_cache_rm(sc); 3670 if (sc->sc_ipopts) 3671 (void) m_free(sc->sc_ipopts); 3672 rtcache_free(&sc->sc_route); 3673 3674 free: 3675 callout_destroy(&sc->sc_timer); 3676 pool_put(&syn_cache_pool, sc); 3677 3678 out: 3679 KERNEL_UNLOCK_ONE(NULL); 3680 mutex_exit(softnet_lock); 3681 } 3682 3683 /* 3684 * Remove syn cache created by the specified tcb entry, 3685 * because this does not make sense to keep them 3686 * (if there's no tcb entry, syn cache entry will never be used) 3687 */ 3688 void 3689 syn_cache_cleanup(struct tcpcb *tp) 3690 { 3691 struct syn_cache *sc, *nsc; 3692 int s; 3693 3694 s = splsoftnet(); 3695 3696 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3697 nsc = LIST_NEXT(sc, sc_tpq); 3698 3699 #ifdef DIAGNOSTIC 3700 if (sc->sc_tp != tp) 3701 panic("invalid sc_tp in syn_cache_cleanup"); 3702 #endif 3703 syn_cache_rm(sc); 3704 syn_cache_put(sc); /* calls pool_put but see spl above */ 3705 } 3706 /* just for safety */ 3707 LIST_INIT(&tp->t_sc); 3708 3709 splx(s); 3710 } 3711 3712 /* 3713 * Find an entry in the syn cache. 3714 */ 3715 struct syn_cache * 3716 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3717 struct syn_cache_head **headp) 3718 { 3719 struct syn_cache *sc; 3720 struct syn_cache_head *scp; 3721 u_int32_t hash; 3722 int s; 3723 3724 SYN_HASHALL(hash, src, dst); 3725 3726 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3727 *headp = scp; 3728 s = splsoftnet(); 3729 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3730 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3731 if (sc->sc_hash != hash) 3732 continue; 3733 if (!memcmp(&sc->sc_src, src, src->sa_len) && 3734 !memcmp(&sc->sc_dst, dst, dst->sa_len)) { 3735 splx(s); 3736 return (sc); 3737 } 3738 } 3739 splx(s); 3740 return (NULL); 3741 } 3742 3743 /* 3744 * This function gets called when we receive an ACK for a socket in the 3745 * LISTEN state. We look up the connection in the syn cache, and if it's 3746 * there, we pull it out of the cache and turn it into a full-blown 3747 * connection in the SYN-RECEIVED state. 3748 * 3749 * The return values may not be immediately obvious, and their effects 3750 * can be subtle, so here they are: 3751 * 3752 * NULL SYN was not found in cache; caller should drop the 3753 * packet and send an RST. 3754 * 3755 * -1 We were unable to create the new connection, and are 3756 * aborting it. An ACK,RST is being sent to the peer 3757 * (unless we got screwey sequence numbers; see below), 3758 * because the 3-way handshake has been completed. Caller 3759 * should not free the mbuf, since we may be using it. If 3760 * we are not, we will free it. 3761 * 3762 * Otherwise, the return value is a pointer to the new socket 3763 * associated with the connection. 3764 */ 3765 struct socket * 3766 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3767 struct tcphdr *th, struct socket *so, struct mbuf *m) 3768 { 3769 struct syn_cache *sc; 3770 struct syn_cache_head *scp; 3771 struct inpcb *inp = NULL; 3772 #ifdef INET6 3773 struct in6pcb *in6p = NULL; 3774 #endif 3775 struct tcpcb *tp; 3776 int s; 3777 struct socket *oso; 3778 3779 s = splsoftnet(); 3780 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3781 splx(s); 3782 return NULL; 3783 } 3784 3785 /* 3786 * Verify the sequence and ack numbers. Try getting the correct 3787 * response again. 3788 */ 3789 if ((th->th_ack != sc->sc_iss + 1) || 3790 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3791 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3792 m_freem(m); 3793 (void)syn_cache_respond(sc); 3794 splx(s); 3795 return ((struct socket *)(-1)); 3796 } 3797 3798 /* Remove this cache entry */ 3799 syn_cache_rm(sc); 3800 splx(s); 3801 3802 /* 3803 * Ok, create the full blown connection, and set things up 3804 * as they would have been set up if we had created the 3805 * connection when the SYN arrived. If we can't create 3806 * the connection, abort it. 3807 */ 3808 /* 3809 * inp still has the OLD in_pcb stuff, set the 3810 * v6-related flags on the new guy, too. This is 3811 * done particularly for the case where an AF_INET6 3812 * socket is bound only to a port, and a v4 connection 3813 * comes in on that port. 3814 * we also copy the flowinfo from the original pcb 3815 * to the new one. 3816 */ 3817 oso = so; 3818 so = sonewconn(so, true); 3819 if (so == NULL) 3820 goto resetandabort; 3821 3822 switch (so->so_proto->pr_domain->dom_family) { 3823 case AF_INET: 3824 inp = sotoinpcb(so); 3825 break; 3826 #ifdef INET6 3827 case AF_INET6: 3828 in6p = sotoin6pcb(so); 3829 break; 3830 #endif 3831 } 3832 3833 switch (src->sa_family) { 3834 case AF_INET: 3835 if (inp) { 3836 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3837 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3838 inp->inp_options = ip_srcroute(m); 3839 in_pcbstate(inp, INP_BOUND); 3840 if (inp->inp_options == NULL) { 3841 inp->inp_options = sc->sc_ipopts; 3842 sc->sc_ipopts = NULL; 3843 } 3844 } 3845 #ifdef INET6 3846 else if (in6p) { 3847 /* IPv4 packet to AF_INET6 socket */ 3848 memset(&in6p->in6p_laddr, 0, sizeof(in6p->in6p_laddr)); 3849 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3850 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3851 &in6p->in6p_laddr.s6_addr32[3], 3852 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3853 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3854 in6totcpcb(in6p)->t_family = AF_INET; 3855 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3856 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3857 else 3858 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3859 in6_pcbstate(in6p, IN6P_BOUND); 3860 } 3861 #endif 3862 break; 3863 #ifdef INET6 3864 case AF_INET6: 3865 if (in6p) { 3866 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3867 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3868 in6_pcbstate(in6p, IN6P_BOUND); 3869 } 3870 break; 3871 #endif 3872 } 3873 3874 #ifdef INET6 3875 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3876 struct in6pcb *oin6p = sotoin6pcb(oso); 3877 /* inherit socket options from the listening socket */ 3878 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3879 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3880 m_freem(in6p->in6p_options); 3881 in6p->in6p_options = NULL; 3882 } 3883 ip6_savecontrol(in6p, &in6p->in6p_options, 3884 mtod(m, struct ip6_hdr *), m); 3885 } 3886 #endif 3887 3888 /* 3889 * Give the new socket our cached route reference. 3890 */ 3891 if (inp) { 3892 rtcache_copy(&inp->inp_route, &sc->sc_route); 3893 rtcache_free(&sc->sc_route); 3894 } 3895 #ifdef INET6 3896 else { 3897 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 3898 rtcache_free(&sc->sc_route); 3899 } 3900 #endif 3901 3902 if (inp) { 3903 struct sockaddr_in sin; 3904 memcpy(&sin, src, src->sa_len); 3905 if (in_pcbconnect(inp, &sin, &lwp0)) { 3906 goto resetandabort; 3907 } 3908 } 3909 #ifdef INET6 3910 else if (in6p) { 3911 struct sockaddr_in6 sin6; 3912 memcpy(&sin6, src, src->sa_len); 3913 if (src->sa_family == AF_INET) { 3914 /* IPv4 packet to AF_INET6 socket */ 3915 in6_sin_2_v4mapsin6((struct sockaddr_in *)src, &sin6); 3916 } 3917 if (in6_pcbconnect(in6p, &sin6, NULL)) { 3918 goto resetandabort; 3919 } 3920 } 3921 #endif 3922 else { 3923 goto resetandabort; 3924 } 3925 3926 if (inp) 3927 tp = intotcpcb(inp); 3928 #ifdef INET6 3929 else if (in6p) 3930 tp = in6totcpcb(in6p); 3931 #endif 3932 else 3933 tp = NULL; 3934 3935 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3936 if (sc->sc_request_r_scale != 15) { 3937 tp->requested_s_scale = sc->sc_requested_s_scale; 3938 tp->request_r_scale = sc->sc_request_r_scale; 3939 tp->snd_scale = sc->sc_requested_s_scale; 3940 tp->rcv_scale = sc->sc_request_r_scale; 3941 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3942 } 3943 if (sc->sc_flags & SCF_TIMESTAMP) 3944 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3945 tp->ts_timebase = sc->sc_timebase; 3946 3947 tp->t_template = tcp_template(tp); 3948 if (tp->t_template == 0) { 3949 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3950 so = NULL; 3951 m_freem(m); 3952 goto abort; 3953 } 3954 3955 tp->iss = sc->sc_iss; 3956 tp->irs = sc->sc_irs; 3957 tcp_sendseqinit(tp); 3958 tcp_rcvseqinit(tp); 3959 tp->t_state = TCPS_SYN_RECEIVED; 3960 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 3961 TCP_STATINC(TCP_STAT_ACCEPTS); 3962 3963 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 3964 tp->t_flags |= TF_WILL_SACK; 3965 3966 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 3967 tp->t_flags |= TF_ECN_PERMIT; 3968 3969 #ifdef TCP_SIGNATURE 3970 if (sc->sc_flags & SCF_SIGNATURE) 3971 tp->t_flags |= TF_SIGNATURE; 3972 #endif 3973 3974 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3975 tp->t_ourmss = sc->sc_ourmaxseg; 3976 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3977 3978 /* 3979 * Initialize the initial congestion window. If we 3980 * had to retransmit the SYN,ACK, we must initialize cwnd 3981 * to 1 segment (i.e. the Loss Window). 3982 */ 3983 if (sc->sc_rxtshift) 3984 tp->snd_cwnd = tp->t_peermss; 3985 else { 3986 int ss = tcp_init_win; 3987 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3988 ss = tcp_init_win_local; 3989 #ifdef INET6 3990 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3991 ss = tcp_init_win_local; 3992 #endif 3993 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3994 } 3995 3996 tcp_rmx_rtt(tp); 3997 tp->snd_wl1 = sc->sc_irs; 3998 tp->rcv_up = sc->sc_irs + 1; 3999 4000 /* 4001 * This is what whould have happened in tcp_output() when 4002 * the SYN,ACK was sent. 4003 */ 4004 tp->snd_up = tp->snd_una; 4005 tp->snd_max = tp->snd_nxt = tp->iss+1; 4006 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 4007 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 4008 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 4009 tp->last_ack_sent = tp->rcv_nxt; 4010 tp->t_partialacks = -1; 4011 tp->t_dupacks = 0; 4012 4013 TCP_STATINC(TCP_STAT_SC_COMPLETED); 4014 s = splsoftnet(); 4015 syn_cache_put(sc); 4016 splx(s); 4017 return so; 4018 4019 resetandabort: 4020 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 4021 abort: 4022 if (so != NULL) { 4023 (void) soqremque(so, 1); 4024 (void) soabort(so); 4025 mutex_enter(softnet_lock); 4026 } 4027 s = splsoftnet(); 4028 syn_cache_put(sc); 4029 splx(s); 4030 TCP_STATINC(TCP_STAT_SC_ABORTED); 4031 return ((struct socket *)(-1)); 4032 } 4033 4034 /* 4035 * This function is called when we get a RST for a 4036 * non-existent connection, so that we can see if the 4037 * connection is in the syn cache. If it is, zap it. 4038 */ 4039 4040 void 4041 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 4042 { 4043 struct syn_cache *sc; 4044 struct syn_cache_head *scp; 4045 int s = splsoftnet(); 4046 4047 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4048 splx(s); 4049 return; 4050 } 4051 if (SEQ_LT(th->th_seq, sc->sc_irs) || 4052 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 4053 splx(s); 4054 return; 4055 } 4056 syn_cache_rm(sc); 4057 TCP_STATINC(TCP_STAT_SC_RESET); 4058 syn_cache_put(sc); /* calls pool_put but see spl above */ 4059 splx(s); 4060 } 4061 4062 void 4063 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 4064 struct tcphdr *th) 4065 { 4066 struct syn_cache *sc; 4067 struct syn_cache_head *scp; 4068 int s; 4069 4070 s = splsoftnet(); 4071 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4072 splx(s); 4073 return; 4074 } 4075 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 4076 if (ntohl(th->th_seq) != sc->sc_iss) { 4077 splx(s); 4078 return; 4079 } 4080 4081 /* 4082 * If we've retransmitted 3 times and this is our second error, 4083 * we remove the entry. Otherwise, we allow it to continue on. 4084 * This prevents us from incorrectly nuking an entry during a 4085 * spurious network outage. 4086 * 4087 * See tcp_notify(). 4088 */ 4089 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 4090 sc->sc_flags |= SCF_UNREACH; 4091 splx(s); 4092 return; 4093 } 4094 4095 syn_cache_rm(sc); 4096 TCP_STATINC(TCP_STAT_SC_UNREACH); 4097 syn_cache_put(sc); /* calls pool_put but see spl above */ 4098 splx(s); 4099 } 4100 4101 /* 4102 * Given a LISTEN socket and an inbound SYN request, add this to the syn 4103 * cache, and send back a segment: 4104 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 4105 * to the source. 4106 * 4107 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 4108 * Doing so would require that we hold onto the data and deliver it 4109 * to the application. However, if we are the target of a SYN-flood 4110 * DoS attack, an attacker could send data which would eventually 4111 * consume all available buffer space if it were ACKed. By not ACKing 4112 * the data, we avoid this DoS scenario. 4113 */ 4114 int 4115 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 4116 unsigned int toff, struct socket *so, struct mbuf *m, u_char *optp, 4117 int optlen, struct tcp_opt_info *oi) 4118 { 4119 struct tcpcb tb, *tp; 4120 long win; 4121 struct syn_cache *sc; 4122 struct syn_cache_head *scp; 4123 struct mbuf *ipopts; 4124 int s; 4125 4126 tp = sototcpcb(so); 4127 4128 /* 4129 * Initialize some local state. 4130 */ 4131 win = sbspace(&so->so_rcv); 4132 if (win > TCP_MAXWIN) 4133 win = TCP_MAXWIN; 4134 4135 #ifdef TCP_SIGNATURE 4136 if (optp || (tp->t_flags & TF_SIGNATURE)) 4137 #else 4138 if (optp) 4139 #endif 4140 { 4141 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4142 #ifdef TCP_SIGNATURE 4143 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4144 #endif 4145 tb.t_state = TCPS_LISTEN; 4146 if (tcp_dooptions(&tb, optp, optlen, th, m, toff, oi) < 0) 4147 return 0; 4148 } else 4149 tb.t_flags = 0; 4150 4151 switch (src->sa_family) { 4152 case AF_INET: 4153 /* Remember the IP options, if any. */ 4154 ipopts = ip_srcroute(m); 4155 break; 4156 default: 4157 ipopts = NULL; 4158 } 4159 4160 /* 4161 * See if we already have an entry for this connection. 4162 * If we do, resend the SYN,ACK. We do not count this 4163 * as a retransmission (XXX though maybe we should). 4164 */ 4165 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4166 TCP_STATINC(TCP_STAT_SC_DUPESYN); 4167 if (ipopts) { 4168 /* 4169 * If we were remembering a previous source route, 4170 * forget it and use the new one we've been given. 4171 */ 4172 if (sc->sc_ipopts) 4173 (void)m_free(sc->sc_ipopts); 4174 sc->sc_ipopts = ipopts; 4175 } 4176 sc->sc_timestamp = tb.ts_recent; 4177 m_freem(m); 4178 if (syn_cache_respond(sc) == 0) { 4179 uint64_t *tcps = TCP_STAT_GETREF(); 4180 tcps[TCP_STAT_SNDACKS]++; 4181 tcps[TCP_STAT_SNDTOTAL]++; 4182 TCP_STAT_PUTREF(); 4183 } 4184 return 1; 4185 } 4186 4187 s = splsoftnet(); 4188 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4189 splx(s); 4190 if (sc == NULL) { 4191 if (ipopts) 4192 (void)m_free(ipopts); 4193 return 0; 4194 } 4195 4196 /* 4197 * Fill in the cache, and put the necessary IP and TCP 4198 * options into the reply. 4199 */ 4200 memset(sc, 0, sizeof(struct syn_cache)); 4201 callout_init(&sc->sc_timer, CALLOUT_MPSAFE); 4202 memcpy(&sc->sc_src, src, src->sa_len); 4203 memcpy(&sc->sc_dst, dst, dst->sa_len); 4204 sc->sc_flags = 0; 4205 sc->sc_ipopts = ipopts; 4206 sc->sc_irs = th->th_seq; 4207 switch (src->sa_family) { 4208 case AF_INET: 4209 { 4210 struct sockaddr_in *srcin = (void *)src; 4211 struct sockaddr_in *dstin = (void *)dst; 4212 4213 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4214 &srcin->sin_addr, dstin->sin_port, 4215 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4216 break; 4217 } 4218 #ifdef INET6 4219 case AF_INET6: 4220 { 4221 struct sockaddr_in6 *srcin6 = (void *)src; 4222 struct sockaddr_in6 *dstin6 = (void *)dst; 4223 4224 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4225 &srcin6->sin6_addr, dstin6->sin6_port, 4226 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4227 break; 4228 } 4229 #endif 4230 } 4231 sc->sc_peermaxseg = oi->maxseg; 4232 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4233 m_get_rcvif_NOMPSAFE(m) : NULL, sc->sc_src.sa.sa_family); 4234 sc->sc_win = win; 4235 sc->sc_timebase = tcp_now - 1; /* see tcp_newtcpcb() */ 4236 sc->sc_timestamp = tb.ts_recent; 4237 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4238 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4239 sc->sc_flags |= SCF_TIMESTAMP; 4240 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4241 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4242 sc->sc_requested_s_scale = tb.requested_s_scale; 4243 sc->sc_request_r_scale = 0; 4244 /* 4245 * Pick the smallest possible scaling factor that 4246 * will still allow us to scale up to sb_max. 4247 * 4248 * We do this because there are broken firewalls that 4249 * will corrupt the window scale option, leading to 4250 * the other endpoint believing that our advertised 4251 * window is unscaled. At scale factors larger than 4252 * 5 the unscaled window will drop below 1500 bytes, 4253 * leading to serious problems when traversing these 4254 * broken firewalls. 4255 * 4256 * With the default sbmax of 256K, a scale factor 4257 * of 3 will be chosen by this algorithm. Those who 4258 * choose a larger sbmax should watch out 4259 * for the compatiblity problems mentioned above. 4260 * 4261 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4262 * or <SYN,ACK>) segment itself is never scaled. 4263 */ 4264 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4265 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4266 sc->sc_request_r_scale++; 4267 } else { 4268 sc->sc_requested_s_scale = 15; 4269 sc->sc_request_r_scale = 15; 4270 } 4271 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4272 sc->sc_flags |= SCF_SACK_PERMIT; 4273 4274 /* 4275 * ECN setup packet received. 4276 */ 4277 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4278 sc->sc_flags |= SCF_ECN_PERMIT; 4279 4280 #ifdef TCP_SIGNATURE 4281 if (tb.t_flags & TF_SIGNATURE) 4282 sc->sc_flags |= SCF_SIGNATURE; 4283 #endif 4284 sc->sc_tp = tp; 4285 m_freem(m); 4286 if (syn_cache_respond(sc) == 0) { 4287 uint64_t *tcps = TCP_STAT_GETREF(); 4288 tcps[TCP_STAT_SNDACKS]++; 4289 tcps[TCP_STAT_SNDTOTAL]++; 4290 TCP_STAT_PUTREF(); 4291 syn_cache_insert(sc, tp); 4292 } else { 4293 s = splsoftnet(); 4294 /* 4295 * syn_cache_put() will try to schedule the timer, so 4296 * we need to initialize it 4297 */ 4298 syn_cache_timer_arm(sc); 4299 syn_cache_put(sc); 4300 splx(s); 4301 TCP_STATINC(TCP_STAT_SC_DROPPED); 4302 } 4303 return 1; 4304 } 4305 4306 /* 4307 * syn_cache_respond: (re)send SYN+ACK. 4308 * 4309 * Returns 0 on success. 4310 */ 4311 4312 int 4313 syn_cache_respond(struct syn_cache *sc) 4314 { 4315 #ifdef INET6 4316 struct rtentry *rt = NULL; 4317 #endif 4318 struct route *ro; 4319 u_int8_t *optp; 4320 int optlen, error; 4321 u_int16_t tlen; 4322 struct ip *ip = NULL; 4323 #ifdef INET6 4324 struct ip6_hdr *ip6 = NULL; 4325 #endif 4326 struct tcpcb *tp; 4327 struct tcphdr *th; 4328 struct mbuf *m; 4329 u_int hlen; 4330 #ifdef TCP_SIGNATURE 4331 struct secasvar *sav = NULL; 4332 u_int8_t *sigp = NULL; 4333 #endif 4334 4335 ro = &sc->sc_route; 4336 switch (sc->sc_src.sa.sa_family) { 4337 case AF_INET: 4338 hlen = sizeof(struct ip); 4339 break; 4340 #ifdef INET6 4341 case AF_INET6: 4342 hlen = sizeof(struct ip6_hdr); 4343 break; 4344 #endif 4345 default: 4346 return EAFNOSUPPORT; 4347 } 4348 4349 /* Worst case scanario, since we don't know the option size yet. */ 4350 tlen = hlen + sizeof(struct tcphdr) + MAX_TCPOPTLEN; 4351 KASSERT(max_linkhdr + tlen <= MCLBYTES); 4352 4353 /* 4354 * Create the IP+TCP header from scratch. 4355 */ 4356 MGETHDR(m, M_DONTWAIT, MT_DATA); 4357 if (m && (max_linkhdr + tlen) > MHLEN) { 4358 MCLGET(m, M_DONTWAIT); 4359 if ((m->m_flags & M_EXT) == 0) { 4360 m_freem(m); 4361 m = NULL; 4362 } 4363 } 4364 if (m == NULL) 4365 return ENOBUFS; 4366 MCLAIM(m, &tcp_tx_mowner); 4367 4368 tp = sc->sc_tp; 4369 4370 /* Fixup the mbuf. */ 4371 m->m_data += max_linkhdr; 4372 m_reset_rcvif(m); 4373 memset(mtod(m, void *), 0, tlen); 4374 4375 switch (sc->sc_src.sa.sa_family) { 4376 case AF_INET: 4377 ip = mtod(m, struct ip *); 4378 ip->ip_v = 4; 4379 ip->ip_dst = sc->sc_src.sin.sin_addr; 4380 ip->ip_src = sc->sc_dst.sin.sin_addr; 4381 ip->ip_p = IPPROTO_TCP; 4382 th = (struct tcphdr *)(ip + 1); 4383 th->th_dport = sc->sc_src.sin.sin_port; 4384 th->th_sport = sc->sc_dst.sin.sin_port; 4385 break; 4386 #ifdef INET6 4387 case AF_INET6: 4388 ip6 = mtod(m, struct ip6_hdr *); 4389 ip6->ip6_vfc = IPV6_VERSION; 4390 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4391 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4392 ip6->ip6_nxt = IPPROTO_TCP; 4393 /* ip6_plen will be updated in ip6_output() */ 4394 th = (struct tcphdr *)(ip6 + 1); 4395 th->th_dport = sc->sc_src.sin6.sin6_port; 4396 th->th_sport = sc->sc_dst.sin6.sin6_port; 4397 break; 4398 #endif 4399 default: 4400 panic("%s: impossible (1)", __func__); 4401 } 4402 4403 th->th_seq = htonl(sc->sc_iss); 4404 th->th_ack = htonl(sc->sc_irs + 1); 4405 th->th_flags = TH_SYN|TH_ACK; 4406 th->th_win = htons(sc->sc_win); 4407 /* th_x2, th_sum, th_urp already 0 from memset */ 4408 4409 /* Tack on the TCP options. */ 4410 optp = (u_int8_t *)(th + 1); 4411 optlen = 0; 4412 *optp++ = TCPOPT_MAXSEG; 4413 *optp++ = TCPOLEN_MAXSEG; 4414 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4415 *optp++ = sc->sc_ourmaxseg & 0xff; 4416 optlen += TCPOLEN_MAXSEG; 4417 4418 if (sc->sc_request_r_scale != 15) { 4419 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4420 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4421 sc->sc_request_r_scale); 4422 optp += TCPOLEN_WINDOW + TCPOLEN_NOP; 4423 optlen += TCPOLEN_WINDOW + TCPOLEN_NOP; 4424 } 4425 4426 if (sc->sc_flags & SCF_SACK_PERMIT) { 4427 /* Let the peer know that we will SACK. */ 4428 *optp++ = TCPOPT_SACK_PERMITTED; 4429 *optp++ = TCPOLEN_SACK_PERMITTED; 4430 optlen += TCPOLEN_SACK_PERMITTED; 4431 } 4432 4433 if (sc->sc_flags & SCF_TIMESTAMP) { 4434 while (optlen % 4 != 2) { 4435 optlen += TCPOLEN_NOP; 4436 *optp++ = TCPOPT_NOP; 4437 } 4438 *optp++ = TCPOPT_TIMESTAMP; 4439 *optp++ = TCPOLEN_TIMESTAMP; 4440 u_int32_t *lp = (u_int32_t *)(optp); 4441 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4442 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4443 *lp = htonl(sc->sc_timestamp); 4444 optp += TCPOLEN_TIMESTAMP - 2; 4445 optlen += TCPOLEN_TIMESTAMP; 4446 } 4447 4448 #ifdef TCP_SIGNATURE 4449 if (sc->sc_flags & SCF_SIGNATURE) { 4450 sav = tcp_signature_getsav(m); 4451 if (sav == NULL) { 4452 m_freem(m); 4453 return EPERM; 4454 } 4455 4456 *optp++ = TCPOPT_SIGNATURE; 4457 *optp++ = TCPOLEN_SIGNATURE; 4458 sigp = optp; 4459 memset(optp, 0, TCP_SIGLEN); 4460 optp += TCP_SIGLEN; 4461 optlen += TCPOLEN_SIGNATURE; 4462 } 4463 #endif 4464 4465 /* 4466 * Terminate and pad TCP options to a 4 byte boundary. 4467 * 4468 * According to RFC793: "The content of the header beyond the 4469 * End-of-Option option must be header padding (i.e., zero)." 4470 * And later: "The padding is composed of zeros." 4471 */ 4472 if (optlen % 4) { 4473 optlen += TCPOLEN_EOL; 4474 *optp++ = TCPOPT_EOL; 4475 } 4476 while (optlen % 4) { 4477 optlen += TCPOLEN_PAD; 4478 *optp++ = TCPOPT_PAD; 4479 } 4480 4481 /* Compute the actual values now that we've added the options. */ 4482 tlen = hlen + sizeof(struct tcphdr) + optlen; 4483 m->m_len = m->m_pkthdr.len = tlen; 4484 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4485 4486 #ifdef TCP_SIGNATURE 4487 if (sav) { 4488 (void)tcp_signature(m, th, hlen, sav, sigp); 4489 key_sa_recordxfer(sav, m); 4490 KEY_SA_UNREF(&sav); 4491 } 4492 #endif 4493 4494 /* 4495 * Send ECN SYN-ACK setup packet. 4496 * Routes can be asymetric, so, even if we receive a packet 4497 * with ECE and CWR set, we must not assume no one will block 4498 * the ECE packet we are about to send. 4499 */ 4500 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4501 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4502 th->th_flags |= TH_ECE; 4503 TCP_STATINC(TCP_STAT_ECN_SHS); 4504 4505 /* 4506 * draft-ietf-tcpm-ecnsyn-00.txt 4507 * 4508 * "[...] a TCP node MAY respond to an ECN-setup 4509 * SYN packet by setting ECT in the responding 4510 * ECN-setup SYN/ACK packet, indicating to routers 4511 * that the SYN/ACK packet is ECN-Capable. 4512 * This allows a congested router along the path 4513 * to mark the packet instead of dropping the 4514 * packet as an indication of congestion." 4515 * 4516 * "[...] There can be a great benefit in setting 4517 * an ECN-capable codepoint in SYN/ACK packets [...] 4518 * Congestion is most likely to occur in 4519 * the server-to-client direction. As a result, 4520 * setting an ECN-capable codepoint in SYN/ACK 4521 * packets can reduce the occurence of three-second 4522 * retransmit timeouts resulting from the drop 4523 * of SYN/ACK packets." 4524 * 4525 * Page 4 and 6, January 2006. 4526 */ 4527 4528 switch (sc->sc_src.sa.sa_family) { 4529 case AF_INET: 4530 ip->ip_tos |= IPTOS_ECN_ECT0; 4531 break; 4532 #ifdef INET6 4533 case AF_INET6: 4534 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4535 break; 4536 #endif 4537 } 4538 TCP_STATINC(TCP_STAT_ECN_ECT); 4539 } 4540 4541 4542 /* 4543 * Compute the packet's checksum. 4544 * 4545 * Fill in some straggling IP bits. Note the stack expects 4546 * ip_len to be in host order, for convenience. 4547 */ 4548 switch (sc->sc_src.sa.sa_family) { 4549 case AF_INET: 4550 ip->ip_len = htons(tlen - hlen); 4551 th->th_sum = 0; 4552 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4553 ip->ip_len = htons(tlen); 4554 ip->ip_ttl = ip_defttl; 4555 /* XXX tos? */ 4556 break; 4557 #ifdef INET6 4558 case AF_INET6: 4559 ip6->ip6_plen = htons(tlen - hlen); 4560 th->th_sum = 0; 4561 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4562 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4563 ip6->ip6_vfc |= IPV6_VERSION; 4564 ip6->ip6_plen = htons(tlen - hlen); 4565 /* ip6_hlim will be initialized afterwards */ 4566 /* XXX flowlabel? */ 4567 break; 4568 #endif 4569 } 4570 4571 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4572 tp = sc->sc_tp; 4573 4574 switch (sc->sc_src.sa.sa_family) { 4575 case AF_INET: 4576 error = ip_output(m, sc->sc_ipopts, ro, 4577 (ip_mtudisc ? IP_MTUDISC : 0), 4578 NULL, tp ? tp->t_inpcb : NULL); 4579 break; 4580 #ifdef INET6 4581 case AF_INET6: 4582 ip6->ip6_hlim = in6_selecthlim(NULL, 4583 (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp : NULL); 4584 rtcache_unref(rt, ro); 4585 4586 error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, 4587 tp ? tp->t_in6pcb : NULL, NULL); 4588 break; 4589 #endif 4590 default: 4591 panic("%s: impossible (2)", __func__); 4592 } 4593 4594 return error; 4595 } 4596