tcp_input.c revision 1.398
1 /* $NetBSD: tcp_input.c,v 1.398 2018/03/29 17:12:36 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 1.398 2018/03/29 17:12:36 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 /* always need ip6.h for IP6_EXTHDR_GET */ 204 #include <netinet/ip6.h> 205 #endif 206 207 #include <netinet/tcp.h> 208 #include <netinet/tcp_fsm.h> 209 #include <netinet/tcp_seq.h> 210 #include <netinet/tcp_timer.h> 211 #include <netinet/tcp_var.h> 212 #include <netinet/tcp_private.h> 213 #include <netinet/tcpip.h> 214 #include <netinet/tcp_congctl.h> 215 #include <netinet/tcp_debug.h> 216 217 #ifdef INET6 218 #include "faith.h" 219 #if defined(NFAITH) && NFAITH > 0 220 #include <net/if_faith.h> 221 #endif 222 #endif 223 224 #ifdef IPSEC 225 #include <netipsec/ipsec.h> 226 #include <netipsec/ipsec_var.h> 227 #include <netipsec/key.h> 228 #ifdef INET6 229 #include <netipsec/ipsec6.h> 230 #endif 231 #endif /* IPSEC*/ 232 233 #include <netinet/tcp_vtw.h> 234 235 int tcprexmtthresh = 3; 236 int tcp_log_refused; 237 238 int tcp_do_autorcvbuf = 1; 239 int tcp_autorcvbuf_inc = 16 * 1024; 240 int tcp_autorcvbuf_max = 256 * 1024; 241 int tcp_msl = (TCPTV_MSL / PR_SLOWHZ); 242 243 static int tcp_rst_ppslim_count = 0; 244 static struct timeval tcp_rst_ppslim_last; 245 static int tcp_ackdrop_ppslim_count = 0; 246 static struct timeval tcp_ackdrop_ppslim_last; 247 248 static void syn_cache_timer(void *); 249 250 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 251 252 /* for modulo comparisons of timestamps */ 253 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 254 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 255 256 /* 257 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 258 */ 259 #ifdef INET6 260 static inline void 261 nd6_hint(struct tcpcb *tp) 262 { 263 struct rtentry *rt = NULL; 264 265 if (tp != NULL && tp->t_in6pcb != NULL && tp->t_family == AF_INET6 && 266 (rt = rtcache_validate(&tp->t_in6pcb->in6p_route)) != NULL) 267 nd6_nud_hint(rt); 268 rtcache_unref(rt, &tp->t_in6pcb->in6p_route); 269 } 270 #else 271 static inline void 272 nd6_hint(struct tcpcb *tp) 273 { 274 } 275 #endif 276 277 /* 278 * Compute ACK transmission behavior. Delay the ACK unless 279 * we have already delayed an ACK (must send an ACK every two segments). 280 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 281 * option is enabled. 282 */ 283 static void 284 tcp_setup_ack(struct tcpcb *tp, const struct tcphdr *th) 285 { 286 287 if (tp->t_flags & TF_DELACK || 288 (tcp_ack_on_push && th->th_flags & TH_PUSH)) 289 tp->t_flags |= TF_ACKNOW; 290 else 291 TCP_SET_DELACK(tp); 292 } 293 294 static void 295 icmp_check(struct tcpcb *tp, const struct tcphdr *th, int acked) 296 { 297 298 /* 299 * If we had a pending ICMP message that refers to data that have 300 * just been acknowledged, disregard the recorded ICMP message. 301 */ 302 if ((tp->t_flags & TF_PMTUD_PEND) && 303 SEQ_GT(th->th_ack, tp->t_pmtud_th_seq)) 304 tp->t_flags &= ~TF_PMTUD_PEND; 305 306 /* 307 * Keep track of the largest chunk of data 308 * acknowledged since last PMTU update 309 */ 310 if (tp->t_pmtud_mss_acked < acked) 311 tp->t_pmtud_mss_acked = acked; 312 } 313 314 /* 315 * Convert TCP protocol fields to host order for easier processing. 316 */ 317 static void 318 tcp_fields_to_host(struct tcphdr *th) 319 { 320 321 NTOHL(th->th_seq); 322 NTOHL(th->th_ack); 323 NTOHS(th->th_win); 324 NTOHS(th->th_urp); 325 } 326 327 /* 328 * ... and reverse the above. 329 */ 330 static void 331 tcp_fields_to_net(struct tcphdr *th) 332 { 333 334 HTONL(th->th_seq); 335 HTONL(th->th_ack); 336 HTONS(th->th_win); 337 HTONS(th->th_urp); 338 } 339 340 static void 341 tcp_urp_drop(struct tcphdr *th, int todrop, int *tiflags) 342 { 343 if (th->th_urp > 1) { 344 th->th_urp -= todrop; 345 } else { 346 *tiflags &= ~TH_URG; 347 th->th_urp = 0; 348 } 349 } 350 351 #ifdef TCP_CSUM_COUNTERS 352 #include <sys/device.h> 353 354 extern struct evcnt tcp_hwcsum_ok; 355 extern struct evcnt tcp_hwcsum_bad; 356 extern struct evcnt tcp_hwcsum_data; 357 extern struct evcnt tcp_swcsum; 358 #if defined(INET6) 359 extern struct evcnt tcp6_hwcsum_ok; 360 extern struct evcnt tcp6_hwcsum_bad; 361 extern struct evcnt tcp6_hwcsum_data; 362 extern struct evcnt tcp6_swcsum; 363 #endif /* defined(INET6) */ 364 365 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 366 367 #else 368 369 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 370 371 #endif /* TCP_CSUM_COUNTERS */ 372 373 #ifdef TCP_REASS_COUNTERS 374 #include <sys/device.h> 375 376 extern struct evcnt tcp_reass_; 377 extern struct evcnt tcp_reass_empty; 378 extern struct evcnt tcp_reass_iteration[8]; 379 extern struct evcnt tcp_reass_prependfirst; 380 extern struct evcnt tcp_reass_prepend; 381 extern struct evcnt tcp_reass_insert; 382 extern struct evcnt tcp_reass_inserttail; 383 extern struct evcnt tcp_reass_append; 384 extern struct evcnt tcp_reass_appendtail; 385 extern struct evcnt tcp_reass_overlaptail; 386 extern struct evcnt tcp_reass_overlapfront; 387 extern struct evcnt tcp_reass_segdup; 388 extern struct evcnt tcp_reass_fragdup; 389 390 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 391 392 #else 393 394 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 395 396 #endif /* TCP_REASS_COUNTERS */ 397 398 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *, 399 int); 400 static int tcp_dooptions(struct tcpcb *, const u_char *, int, 401 struct tcphdr *, struct mbuf *, int, struct tcp_opt_info *); 402 403 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 404 #ifdef INET6 405 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 406 #endif 407 408 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 409 410 #if defined(MBUFTRACE) 411 struct mowner tcp_reass_mowner = MOWNER_INIT("tcp", "reass"); 412 #endif /* defined(MBUFTRACE) */ 413 414 static struct pool tcpipqent_pool; 415 416 void 417 tcpipqent_init(void) 418 { 419 420 pool_init(&tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", 421 NULL, IPL_VM); 422 } 423 424 struct ipqent * 425 tcpipqent_alloc(void) 426 { 427 struct ipqent *ipqe; 428 int s; 429 430 s = splvm(); 431 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 432 splx(s); 433 434 return ipqe; 435 } 436 437 void 438 tcpipqent_free(struct ipqent *ipqe) 439 { 440 int s; 441 442 s = splvm(); 443 pool_put(&tcpipqent_pool, ipqe); 444 splx(s); 445 } 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 tcpiphdr 483 * gets munged 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->ipre_mlast, m); 500 TRAVERSE(p->ipre_mlast); 501 m = NULL; 502 tiqe = p; 503 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 504 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 505 goto skip_replacement; 506 } 507 /* 508 * While we're here, if the pkt is completely beyond 509 * anything we have, just insert it at the tail. 510 */ 511 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 512 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 513 goto insert_it; 514 } 515 } 516 517 q = TAILQ_FIRST(&tp->segq); 518 519 if (q != NULL) { 520 /* 521 * If this segment immediately precedes the first out-of-order 522 * block, simply slap the segment in front of it and (mostly) 523 * skip the complicated logic. 524 */ 525 if (pkt_seq + pkt_len == q->ipqe_seq) { 526 q->ipqe_seq = pkt_seq; 527 q->ipqe_len += pkt_len; 528 q->ipqe_flags |= pkt_flags; 529 m_cat(m, q->ipqe_m); 530 q->ipqe_m = m; 531 q->ipre_mlast = m; /* last mbuf may have changed */ 532 TRAVERSE(q->ipre_mlast); 533 tiqe = q; 534 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 535 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 536 goto skip_replacement; 537 } 538 } else { 539 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 540 } 541 542 /* 543 * Find a segment which begins after this one does. 544 */ 545 for (p = NULL; q != NULL; q = nq) { 546 nq = TAILQ_NEXT(q, ipqe_q); 547 #ifdef TCP_REASS_COUNTERS 548 count++; 549 #endif 550 /* 551 * If the received segment is just right after this 552 * fragment, merge the two together and then check 553 * for further overlaps. 554 */ 555 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 556 #ifdef TCPREASS_DEBUG 557 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 558 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 559 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 560 #endif 561 pkt_len += q->ipqe_len; 562 pkt_flags |= q->ipqe_flags; 563 pkt_seq = q->ipqe_seq; 564 m_cat(q->ipre_mlast, m); 565 TRAVERSE(q->ipre_mlast); 566 m = q->ipqe_m; 567 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 568 goto free_ipqe; 569 } 570 /* 571 * If the received segment is completely past this 572 * fragment, we need to go the next fragment. 573 */ 574 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 575 p = q; 576 continue; 577 } 578 /* 579 * If the fragment is past the received segment, 580 * it (or any following) can't be concatenated. 581 */ 582 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 583 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 584 break; 585 } 586 587 /* 588 * We've received all the data in this segment before. 589 * mark it as a duplicate and return. 590 */ 591 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 592 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 593 tcps = TCP_STAT_GETREF(); 594 tcps[TCP_STAT_RCVDUPPACK]++; 595 tcps[TCP_STAT_RCVDUPBYTE] += pkt_len; 596 TCP_STAT_PUTREF(); 597 tcp_new_dsack(tp, pkt_seq, pkt_len); 598 m_freem(m); 599 if (tiqe != NULL) { 600 tcpipqent_free(tiqe); 601 } 602 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 603 goto out; 604 } 605 /* 606 * Received segment completely overlaps this fragment 607 * so we drop the fragment (this keeps the temporal 608 * ordering of segments correct). 609 */ 610 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 611 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 612 rcvpartdupbyte += q->ipqe_len; 613 m_freem(q->ipqe_m); 614 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 615 goto free_ipqe; 616 } 617 /* 618 * RX'ed segment extends past the end of the 619 * fragment. Drop the overlapping bytes. Then 620 * merge the fragment and segment then treat as 621 * a longer received packet. 622 */ 623 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 624 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 625 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 626 #ifdef TCPREASS_DEBUG 627 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 628 tp, overlap, 629 pkt_seq, pkt_seq + pkt_len, pkt_len); 630 #endif 631 m_adj(m, overlap); 632 rcvpartdupbyte += overlap; 633 m_cat(q->ipre_mlast, m); 634 TRAVERSE(q->ipre_mlast); 635 m = q->ipqe_m; 636 pkt_seq = q->ipqe_seq; 637 pkt_len += q->ipqe_len - overlap; 638 rcvoobyte -= overlap; 639 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 640 goto free_ipqe; 641 } 642 /* 643 * RX'ed segment extends past the front of the 644 * fragment. Drop the overlapping bytes on the 645 * received packet. The packet will then be 646 * contatentated with this fragment a bit later. 647 */ 648 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 649 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 650 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 651 #ifdef TCPREASS_DEBUG 652 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 653 tp, overlap, 654 pkt_seq, pkt_seq + pkt_len, pkt_len); 655 #endif 656 m_adj(m, -overlap); 657 pkt_len -= overlap; 658 rcvpartdupbyte += overlap; 659 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 660 rcvoobyte -= overlap; 661 } 662 /* 663 * If the received segment immediates precedes this 664 * fragment then tack the fragment onto this segment 665 * and reinsert the data. 666 */ 667 if (q->ipqe_seq == pkt_seq + pkt_len) { 668 #ifdef TCPREASS_DEBUG 669 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 670 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 671 pkt_seq, pkt_seq + pkt_len, pkt_len); 672 #endif 673 pkt_len += q->ipqe_len; 674 pkt_flags |= q->ipqe_flags; 675 m_cat(m, q->ipqe_m); 676 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 677 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 678 tp->t_segqlen--; 679 KASSERT(tp->t_segqlen >= 0); 680 KASSERT(tp->t_segqlen != 0 || 681 (TAILQ_EMPTY(&tp->segq) && 682 TAILQ_EMPTY(&tp->timeq))); 683 if (tiqe == NULL) { 684 tiqe = q; 685 } else { 686 tcpipqent_free(q); 687 } 688 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 689 break; 690 } 691 /* 692 * If the fragment is before the segment, remember it. 693 * When this loop is terminated, p will contain the 694 * pointer to fragment that is right before the received 695 * segment. 696 */ 697 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 698 p = q; 699 700 continue; 701 702 /* 703 * This is a common operation. It also will allow 704 * to save doing a malloc/free in most instances. 705 */ 706 free_ipqe: 707 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 708 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 709 tp->t_segqlen--; 710 KASSERT(tp->t_segqlen >= 0); 711 KASSERT(tp->t_segqlen != 0 || 712 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 713 if (tiqe == NULL) { 714 tiqe = q; 715 } else { 716 tcpipqent_free(q); 717 } 718 } 719 720 #ifdef TCP_REASS_COUNTERS 721 if (count > 7) 722 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 723 else if (count > 0) 724 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 725 #endif 726 727 insert_it: 728 729 /* 730 * Allocate a new queue entry since the received segment did not 731 * collapse onto any other out-of-order block; thus we are allocating 732 * a new block. If it had collapsed, tiqe would not be NULL and 733 * we would be reusing it. 734 * XXX If we can't, just drop the packet. XXX 735 */ 736 if (tiqe == NULL) { 737 tiqe = tcpipqent_alloc(); 738 if (tiqe == NULL) { 739 TCP_STATINC(TCP_STAT_RCVMEMDROP); 740 m_freem(m); 741 goto out; 742 } 743 } 744 745 /* 746 * Update the counters. 747 */ 748 tp->t_rcvoopack++; 749 tcps = TCP_STAT_GETREF(); 750 tcps[TCP_STAT_RCVOOPACK]++; 751 tcps[TCP_STAT_RCVOOBYTE] += rcvoobyte; 752 if (rcvpartdupbyte) { 753 tcps[TCP_STAT_RCVPARTDUPPACK]++; 754 tcps[TCP_STAT_RCVPARTDUPBYTE] += rcvpartdupbyte; 755 } 756 TCP_STAT_PUTREF(); 757 758 /* 759 * Insert the new fragment queue entry into both queues. 760 */ 761 tiqe->ipqe_m = m; 762 tiqe->ipre_mlast = m; 763 tiqe->ipqe_seq = pkt_seq; 764 tiqe->ipqe_len = pkt_len; 765 tiqe->ipqe_flags = pkt_flags; 766 if (p == NULL) { 767 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 768 #ifdef TCPREASS_DEBUG 769 if (tiqe->ipqe_seq != tp->rcv_nxt) 770 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 771 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 772 #endif 773 } else { 774 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 775 #ifdef TCPREASS_DEBUG 776 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 777 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 778 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 779 #endif 780 } 781 tp->t_segqlen++; 782 783 skip_replacement: 784 785 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 786 787 present: 788 /* 789 * Present data to user, advancing rcv_nxt through 790 * completed sequence space. 791 */ 792 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 793 goto out; 794 q = TAILQ_FIRST(&tp->segq); 795 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 796 goto out; 797 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 798 goto out; 799 800 tp->rcv_nxt += q->ipqe_len; 801 pkt_flags = q->ipqe_flags & TH_FIN; 802 nd6_hint(tp); 803 804 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 805 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 806 tp->t_segqlen--; 807 KASSERT(tp->t_segqlen >= 0); 808 KASSERT(tp->t_segqlen != 0 || 809 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 810 if (so->so_state & SS_CANTRCVMORE) 811 m_freem(q->ipqe_m); 812 else 813 sbappendstream(&so->so_rcv, q->ipqe_m); 814 tcpipqent_free(q); 815 TCP_REASS_UNLOCK(tp); 816 sorwakeup(so); 817 return (pkt_flags); 818 out: 819 TCP_REASS_UNLOCK(tp); 820 return (0); 821 } 822 823 #ifdef INET6 824 int 825 tcp6_input(struct mbuf **mp, int *offp, int proto) 826 { 827 struct mbuf *m = *mp; 828 829 /* 830 * draft-itojun-ipv6-tcp-to-anycast 831 * better place to put this in? 832 */ 833 if (m->m_flags & M_ANYCAST6) { 834 struct ip6_hdr *ip6; 835 if (m->m_len < sizeof(struct ip6_hdr)) { 836 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 837 TCP_STATINC(TCP_STAT_RCVSHORT); 838 return IPPROTO_DONE; 839 } 840 } 841 ip6 = mtod(m, struct ip6_hdr *); 842 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 843 (char *)&ip6->ip6_dst - (char *)ip6); 844 return IPPROTO_DONE; 845 } 846 847 tcp_input(m, *offp, proto); 848 return IPPROTO_DONE; 849 } 850 #endif 851 852 static void 853 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 854 { 855 char src[INET_ADDRSTRLEN]; 856 char dst[INET_ADDRSTRLEN]; 857 858 if (ip) { 859 in_print(src, sizeof(src), &ip->ip_src); 860 in_print(dst, sizeof(dst), &ip->ip_dst); 861 } else { 862 strlcpy(src, "(unknown)", sizeof(src)); 863 strlcpy(dst, "(unknown)", sizeof(dst)); 864 } 865 log(LOG_INFO, 866 "Connection attempt to TCP %s:%d from %s:%d\n", 867 dst, ntohs(th->th_dport), 868 src, ntohs(th->th_sport)); 869 } 870 871 #ifdef INET6 872 static void 873 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 874 { 875 char src[INET6_ADDRSTRLEN]; 876 char dst[INET6_ADDRSTRLEN]; 877 878 if (ip6) { 879 in6_print(src, sizeof(src), &ip6->ip6_src); 880 in6_print(dst, sizeof(dst), &ip6->ip6_dst); 881 } else { 882 strlcpy(src, "(unknown v6)", sizeof(src)); 883 strlcpy(dst, "(unknown v6)", sizeof(dst)); 884 } 885 log(LOG_INFO, 886 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 887 dst, ntohs(th->th_dport), 888 src, ntohs(th->th_sport)); 889 } 890 #endif 891 892 /* 893 * Checksum extended TCP header and data. 894 */ 895 int 896 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, 897 int toff, int off, int tlen) 898 { 899 struct ifnet *rcvif; 900 int s; 901 902 /* 903 * XXX it's better to record and check if this mbuf is 904 * already checked. 905 */ 906 907 rcvif = m_get_rcvif(m, &s); 908 if (__predict_false(rcvif == NULL)) 909 goto badcsum; /* XXX */ 910 911 switch (af) { 912 case AF_INET: 913 switch (m->m_pkthdr.csum_flags & 914 ((rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 915 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 916 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 917 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 918 goto badcsum; 919 920 case M_CSUM_TCPv4|M_CSUM_DATA: { 921 u_int32_t hw_csum = m->m_pkthdr.csum_data; 922 923 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 924 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 925 const struct ip *ip = 926 mtod(m, const struct ip *); 927 928 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 929 ip->ip_dst.s_addr, 930 htons(hw_csum + tlen + off + IPPROTO_TCP)); 931 } 932 if ((hw_csum ^ 0xffff) != 0) 933 goto badcsum; 934 break; 935 } 936 937 case M_CSUM_TCPv4: 938 /* Checksum was okay. */ 939 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 940 break; 941 942 default: 943 /* 944 * Must compute it ourselves. Maybe skip checksum 945 * on loopback interfaces. 946 */ 947 if (__predict_true(!(rcvif->if_flags & IFF_LOOPBACK) || 948 tcp_do_loopback_cksum)) { 949 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 950 if (in4_cksum(m, IPPROTO_TCP, toff, 951 tlen + off) != 0) 952 goto badcsum; 953 } 954 break; 955 } 956 break; 957 958 #ifdef INET6 959 case AF_INET6: 960 switch (m->m_pkthdr.csum_flags & 961 ((rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 962 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 963 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 964 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 965 goto badcsum; 966 967 #if 0 /* notyet */ 968 case M_CSUM_TCPv6|M_CSUM_DATA: 969 #endif 970 971 case M_CSUM_TCPv6: 972 /* Checksum was okay. */ 973 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 974 break; 975 976 default: 977 /* 978 * Must compute it ourselves. Maybe skip checksum 979 * on loopback interfaces. 980 */ 981 if (__predict_true((m->m_flags & M_LOOP) == 0 || 982 tcp_do_loopback_cksum)) { 983 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 984 if (in6_cksum(m, IPPROTO_TCP, toff, 985 tlen + off) != 0) 986 goto badcsum; 987 } 988 } 989 break; 990 #endif /* INET6 */ 991 } 992 m_put_rcvif(rcvif, &s); 993 994 return 0; 995 996 badcsum: 997 m_put_rcvif(rcvif, &s); 998 TCP_STATINC(TCP_STAT_RCVBADSUM); 999 return -1; 1000 } 1001 1002 /* 1003 * When a packet arrives addressed to a vestigial tcpbp, we 1004 * nevertheless have to respond to it per the spec. 1005 * 1006 * This code is duplicated from the one in tcp_input(). 1007 */ 1008 static void tcp_vtw_input(struct tcphdr *th, vestigial_inpcb_t *vp, 1009 struct mbuf *m, int tlen) 1010 { 1011 int tiflags; 1012 int todrop; 1013 uint32_t t_flags = 0; 1014 uint64_t *tcps; 1015 1016 tiflags = th->th_flags; 1017 todrop = vp->rcv_nxt - th->th_seq; 1018 1019 if (todrop > 0) { 1020 if (tiflags & TH_SYN) { 1021 tiflags &= ~TH_SYN; 1022 th->th_seq++; 1023 tcp_urp_drop(th, 1, &tiflags); 1024 todrop--; 1025 } 1026 if (todrop > tlen || 1027 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1028 /* 1029 * Any valid FIN or RST must be to the left of the 1030 * window. At this point the FIN or RST must be a 1031 * duplicate or out of sequence; drop it. 1032 */ 1033 if (tiflags & TH_RST) 1034 goto drop; 1035 tiflags &= ~(TH_FIN|TH_RST); 1036 1037 /* 1038 * Send an ACK to resynchronize and drop any data. 1039 * But keep on processing for RST or ACK. 1040 */ 1041 t_flags |= TF_ACKNOW; 1042 todrop = tlen; 1043 tcps = TCP_STAT_GETREF(); 1044 tcps[TCP_STAT_RCVDUPPACK] += 1; 1045 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 1046 TCP_STAT_PUTREF(); 1047 } else if ((tiflags & TH_RST) && 1048 th->th_seq != vp->rcv_nxt) { 1049 /* 1050 * Test for reset before adjusting the sequence 1051 * number for overlapping data. 1052 */ 1053 goto dropafterack_ratelim; 1054 } else { 1055 tcps = TCP_STAT_GETREF(); 1056 tcps[TCP_STAT_RCVPARTDUPPACK] += 1; 1057 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 1058 TCP_STAT_PUTREF(); 1059 } 1060 1061 // tcp_new_dsack(tp, th->th_seq, todrop); 1062 // hdroptlen += todrop; /*drop from head afterwards*/ 1063 1064 th->th_seq += todrop; 1065 tlen -= todrop; 1066 tcp_urp_drop(th, todrop, &tiflags); 1067 } 1068 1069 /* 1070 * If new data are received on a connection after the 1071 * user processes are gone, then RST the other end. 1072 */ 1073 if (tlen) { 1074 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 1075 goto dropwithreset; 1076 } 1077 1078 /* 1079 * If segment ends after window, drop trailing data 1080 * (and PUSH and FIN); if nothing left, just ACK. 1081 */ 1082 todrop = (th->th_seq + tlen) - (vp->rcv_nxt + vp->rcv_wnd); 1083 1084 if (todrop > 0) { 1085 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 1086 if (todrop >= tlen) { 1087 /* 1088 * The segment actually starts after the window. 1089 * th->th_seq + tlen - vp->rcv_nxt - vp->rcv_wnd >= tlen 1090 * th->th_seq - vp->rcv_nxt - vp->rcv_wnd >= 0 1091 * th->th_seq >= vp->rcv_nxt + vp->rcv_wnd 1092 */ 1093 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 1094 1095 /* 1096 * If a new connection request is received 1097 * while in TIME_WAIT, drop the old connection 1098 * and start over if the sequence numbers 1099 * are above the previous ones. 1100 */ 1101 if ((tiflags & TH_SYN) && 1102 SEQ_GT(th->th_seq, vp->rcv_nxt)) { 1103 /* 1104 * We only support this in the !NOFDREF case, which 1105 * is to say: not here. 1106 */ 1107 goto dropwithreset; 1108 } 1109 1110 /* 1111 * If window is closed can only take segments at 1112 * window edge, and have to drop data and PUSH from 1113 * incoming segments. Continue processing, but 1114 * remember to ack. Otherwise, drop segment 1115 * and (if not RST) ack. 1116 */ 1117 if (vp->rcv_wnd == 0 && th->th_seq == vp->rcv_nxt) { 1118 t_flags |= TF_ACKNOW; 1119 TCP_STATINC(TCP_STAT_RCVWINPROBE); 1120 } else { 1121 goto dropafterack; 1122 } 1123 } else { 1124 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 1125 } 1126 m_adj(m, -todrop); 1127 tlen -= todrop; 1128 tiflags &= ~(TH_PUSH|TH_FIN); 1129 } 1130 1131 if (tiflags & TH_RST) { 1132 if (th->th_seq != vp->rcv_nxt) 1133 goto dropafterack_ratelim; 1134 1135 vtw_del(vp->ctl, vp->vtw); 1136 goto drop; 1137 } 1138 1139 /* 1140 * If the ACK bit is off we drop the segment and return. 1141 */ 1142 if ((tiflags & TH_ACK) == 0) { 1143 if (t_flags & TF_ACKNOW) 1144 goto dropafterack; 1145 goto drop; 1146 } 1147 1148 /* 1149 * In TIME_WAIT state the only thing that should arrive 1150 * is a retransmission of the remote FIN. Acknowledge 1151 * it and restart the finack timer. 1152 */ 1153 vtw_restart(vp); 1154 goto dropafterack; 1155 1156 dropafterack: 1157 /* 1158 * Generate an ACK dropping incoming segment if it occupies 1159 * sequence space, where the ACK reflects our state. 1160 */ 1161 if (tiflags & TH_RST) 1162 goto drop; 1163 goto dropafterack2; 1164 1165 dropafterack_ratelim: 1166 /* 1167 * We may want to rate-limit ACKs against SYN/RST attack. 1168 */ 1169 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 1170 tcp_ackdrop_ppslim) == 0) { 1171 /* XXX stat */ 1172 goto drop; 1173 } 1174 /* ...fall into dropafterack2... */ 1175 1176 dropafterack2: 1177 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, th->th_ack, TH_ACK); 1178 return; 1179 1180 dropwithreset: 1181 /* 1182 * Generate a RST, dropping incoming segment. 1183 * Make ACK acceptable to originator of segment. 1184 */ 1185 if (tiflags & TH_RST) 1186 goto drop; 1187 1188 if (tiflags & TH_ACK) { 1189 tcp_respond(0, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 1190 } else { 1191 if (tiflags & TH_SYN) 1192 ++tlen; 1193 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, (tcp_seq)0, 1194 TH_RST|TH_ACK); 1195 } 1196 return; 1197 drop: 1198 m_freem(m); 1199 } 1200 1201 /* 1202 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 1203 */ 1204 void 1205 tcp_input(struct mbuf *m, ...) 1206 { 1207 struct tcphdr *th; 1208 struct ip *ip; 1209 struct inpcb *inp; 1210 #ifdef INET6 1211 struct ip6_hdr *ip6; 1212 struct in6pcb *in6p; 1213 #endif 1214 u_int8_t *optp = NULL; 1215 int optlen = 0; 1216 int len, tlen, toff, hdroptlen = 0; 1217 struct tcpcb *tp = NULL; 1218 int tiflags; 1219 struct socket *so = NULL; 1220 int todrop, acked, ourfinisacked, needoutput = 0; 1221 bool dupseg; 1222 #ifdef TCP_DEBUG 1223 short ostate = 0; 1224 #endif 1225 u_long tiwin; 1226 struct tcp_opt_info opti; 1227 int off, iphlen; 1228 va_list ap; 1229 int af; /* af on the wire */ 1230 struct mbuf *tcp_saveti = NULL; 1231 uint32_t ts_rtt; 1232 uint8_t iptos; 1233 uint64_t *tcps; 1234 vestigial_inpcb_t vestige; 1235 1236 vestige.valid = 0; 1237 1238 MCLAIM(m, &tcp_rx_mowner); 1239 va_start(ap, m); 1240 toff = va_arg(ap, int); 1241 (void)va_arg(ap, int); /* ignore value, advance ap */ 1242 va_end(ap); 1243 1244 TCP_STATINC(TCP_STAT_RCVTOTAL); 1245 1246 memset(&opti, 0, sizeof(opti)); 1247 opti.ts_present = 0; 1248 opti.maxseg = 0; 1249 1250 /* 1251 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 1252 * 1253 * TCP is, by definition, unicast, so we reject all 1254 * multicast outright. 1255 * 1256 * Note, there are additional src/dst address checks in 1257 * the AF-specific code below. 1258 */ 1259 if (m->m_flags & (M_BCAST|M_MCAST)) { 1260 /* XXX stat */ 1261 goto drop; 1262 } 1263 #ifdef INET6 1264 if (m->m_flags & M_ANYCAST6) { 1265 /* XXX stat */ 1266 goto drop; 1267 } 1268 #endif 1269 1270 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, sizeof(struct tcphdr)); 1271 if (th == NULL) { 1272 TCP_STATINC(TCP_STAT_RCVSHORT); 1273 return; 1274 } 1275 1276 /* 1277 * Get IP and TCP header. 1278 * Note: IP leaves IP header in first mbuf. 1279 */ 1280 ip = mtod(m, struct ip *); 1281 switch (ip->ip_v) { 1282 case 4: 1283 #ifdef INET6 1284 ip6 = NULL; 1285 #endif 1286 af = AF_INET; 1287 iphlen = sizeof(struct ip); 1288 1289 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1290 in_broadcast(ip->ip_dst, m_get_rcvif_NOMPSAFE(m))) 1291 goto drop; 1292 1293 /* We do the checksum after PCB lookup... */ 1294 len = ntohs(ip->ip_len); 1295 tlen = len - toff; 1296 iptos = ip->ip_tos; 1297 break; 1298 #ifdef INET6 1299 case 6: 1300 ip = NULL; 1301 iphlen = sizeof(struct ip6_hdr); 1302 af = AF_INET6; 1303 ip6 = mtod(m, struct ip6_hdr *); 1304 1305 /* 1306 * Be proactive about unspecified IPv6 address in source. 1307 * As we use all-zero to indicate unbounded/unconnected pcb, 1308 * unspecified IPv6 address can be used to confuse us. 1309 * 1310 * Note that packets with unspecified IPv6 destination is 1311 * already dropped in ip6_input. 1312 */ 1313 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1314 /* XXX stat */ 1315 goto drop; 1316 } 1317 1318 /* 1319 * Make sure destination address is not multicast. 1320 * Source address checked in ip6_input(). 1321 */ 1322 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1323 /* XXX stat */ 1324 goto drop; 1325 } 1326 1327 /* We do the checksum after PCB lookup... */ 1328 len = m->m_pkthdr.len; 1329 tlen = len - toff; 1330 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1331 break; 1332 #endif 1333 default: 1334 m_freem(m); 1335 return; 1336 } 1337 1338 /* 1339 * Enforce alignment requirements that are violated in 1340 * some cases, see kern/50766 for details. 1341 */ 1342 if (TCP_HDR_ALIGNED_P(th) == 0) { 1343 m = m_copyup(m, toff + sizeof(struct tcphdr), 0); 1344 if (m == NULL) { 1345 TCP_STATINC(TCP_STAT_RCVSHORT); 1346 return; 1347 } 1348 ip = mtod(m, struct ip *); 1349 #ifdef INET6 1350 ip6 = mtod(m, struct ip6_hdr *); 1351 #endif 1352 th = (struct tcphdr *)(mtod(m, char *) + toff); 1353 } 1354 KASSERT(TCP_HDR_ALIGNED_P(th)); 1355 1356 /* 1357 * Check that TCP offset makes sense, pull out TCP options and 1358 * adjust length. 1359 */ 1360 off = th->th_off << 2; 1361 if (off < sizeof(struct tcphdr) || off > tlen) { 1362 TCP_STATINC(TCP_STAT_RCVBADOFF); 1363 goto drop; 1364 } 1365 tlen -= off; 1366 1367 if (off > sizeof(struct tcphdr)) { 1368 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1369 if (th == NULL) { 1370 TCP_STATINC(TCP_STAT_RCVSHORT); 1371 return; 1372 } 1373 KASSERT(TCP_HDR_ALIGNED_P(th)); 1374 optlen = off - sizeof(struct tcphdr); 1375 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1376 1377 /* 1378 * Do quick retrieval of timestamp options. 1379 * 1380 * If timestamp is the only option and it's formatted as 1381 * recommended in RFC 1323 appendix A, we quickly get the 1382 * values now and don't bother calling tcp_dooptions(), 1383 * etc. 1384 */ 1385 if ((optlen == TCPOLEN_TSTAMP_APPA || 1386 (optlen > TCPOLEN_TSTAMP_APPA && 1387 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1388 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1389 (th->th_flags & TH_SYN) == 0) { 1390 opti.ts_present = 1; 1391 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1392 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1393 optp = NULL; /* we've parsed the options */ 1394 } 1395 } 1396 tiflags = th->th_flags; 1397 1398 /* 1399 * Checksum extended TCP header and data 1400 */ 1401 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1402 goto badcsum; 1403 1404 /* 1405 * Locate pcb for segment. 1406 */ 1407 findpcb: 1408 inp = NULL; 1409 #ifdef INET6 1410 in6p = NULL; 1411 #endif 1412 switch (af) { 1413 case AF_INET: 1414 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1415 ip->ip_dst, th->th_dport, &vestige); 1416 if (inp == NULL && !vestige.valid) { 1417 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1418 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, 1419 th->th_dport); 1420 } 1421 #ifdef INET6 1422 if (inp == NULL && !vestige.valid) { 1423 struct in6_addr s, d; 1424 1425 /* mapped addr case */ 1426 in6_in_2_v4mapin6(&ip->ip_src, &s); 1427 in6_in_2_v4mapin6(&ip->ip_dst, &d); 1428 in6p = in6_pcblookup_connect(&tcbtable, &s, 1429 th->th_sport, &d, th->th_dport, 0, &vestige); 1430 if (in6p == 0 && !vestige.valid) { 1431 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1432 in6p = in6_pcblookup_bind(&tcbtable, &d, 1433 th->th_dport, 0); 1434 } 1435 } 1436 #endif 1437 #ifndef INET6 1438 if (inp == NULL && !vestige.valid) 1439 #else 1440 if (inp == NULL && in6p == NULL && !vestige.valid) 1441 #endif 1442 { 1443 TCP_STATINC(TCP_STAT_NOPORT); 1444 if (tcp_log_refused && 1445 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1446 tcp4_log_refused(ip, th); 1447 } 1448 tcp_fields_to_host(th); 1449 goto dropwithreset_ratelim; 1450 } 1451 #if defined(IPSEC) 1452 if (ipsec_used) { 1453 if (inp && ipsec_in_reject(m, inp)) { 1454 goto drop; 1455 } 1456 #ifdef INET6 1457 else if (in6p && ipsec_in_reject(m, in6p)) { 1458 goto drop; 1459 } 1460 #endif 1461 } 1462 #endif /*IPSEC*/ 1463 break; 1464 #ifdef INET6 1465 case AF_INET6: 1466 { 1467 int faith; 1468 1469 #if defined(NFAITH) && NFAITH > 0 1470 faith = faithprefix(&ip6->ip6_dst); 1471 #else 1472 faith = 0; 1473 #endif 1474 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1475 th->th_sport, &ip6->ip6_dst, th->th_dport, faith, &vestige); 1476 if (!in6p && !vestige.valid) { 1477 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1478 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1479 th->th_dport, faith); 1480 } 1481 if (!in6p && !vestige.valid) { 1482 TCP_STATINC(TCP_STAT_NOPORT); 1483 if (tcp_log_refused && 1484 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1485 tcp6_log_refused(ip6, th); 1486 } 1487 tcp_fields_to_host(th); 1488 goto dropwithreset_ratelim; 1489 } 1490 #if defined(IPSEC) 1491 if (ipsec_used && in6p && ipsec_in_reject(m, in6p)) { 1492 goto drop; 1493 } 1494 #endif 1495 break; 1496 } 1497 #endif 1498 } 1499 1500 tcp_fields_to_host(th); 1501 1502 /* 1503 * If the state is CLOSED (i.e., TCB does not exist) then 1504 * all data in the incoming segment is discarded. 1505 * If the TCB exists but is in CLOSED state, it is embryonic, 1506 * but should either do a listen or a connect soon. 1507 */ 1508 tp = NULL; 1509 so = NULL; 1510 if (inp) { 1511 /* Check the minimum TTL for socket. */ 1512 if (ip->ip_ttl < inp->inp_ip_minttl) 1513 goto drop; 1514 1515 tp = intotcpcb(inp); 1516 so = inp->inp_socket; 1517 } 1518 #ifdef INET6 1519 else if (in6p) { 1520 tp = in6totcpcb(in6p); 1521 so = in6p->in6p_socket; 1522 } 1523 #endif 1524 else if (vestige.valid) { 1525 /* We do not support the resurrection of vtw tcpcps. */ 1526 tcp_vtw_input(th, &vestige, m, tlen); 1527 m = NULL; 1528 goto drop; 1529 } 1530 1531 if (tp == NULL) 1532 goto dropwithreset_ratelim; 1533 if (tp->t_state == TCPS_CLOSED) 1534 goto drop; 1535 1536 KASSERT(so->so_lock == softnet_lock); 1537 KASSERT(solocked(so)); 1538 1539 /* Unscale the window into a 32-bit value. */ 1540 if ((tiflags & TH_SYN) == 0) 1541 tiwin = th->th_win << tp->snd_scale; 1542 else 1543 tiwin = th->th_win; 1544 1545 #ifdef INET6 1546 /* save packet options if user wanted */ 1547 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1548 if (in6p->in6p_options) { 1549 m_freem(in6p->in6p_options); 1550 in6p->in6p_options = NULL; 1551 } 1552 KASSERT(ip6 != NULL); 1553 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1554 } 1555 #endif 1556 1557 if (so->so_options & SO_DEBUG) { 1558 #ifdef TCP_DEBUG 1559 ostate = tp->t_state; 1560 #endif 1561 1562 tcp_saveti = NULL; 1563 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1564 goto nosave; 1565 1566 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1567 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1568 if (tcp_saveti == NULL) 1569 goto nosave; 1570 } else { 1571 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1572 if (tcp_saveti == NULL) 1573 goto nosave; 1574 MCLAIM(m, &tcp_mowner); 1575 tcp_saveti->m_len = iphlen; 1576 m_copydata(m, 0, iphlen, 1577 mtod(tcp_saveti, void *)); 1578 } 1579 1580 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1581 m_freem(tcp_saveti); 1582 tcp_saveti = NULL; 1583 } else { 1584 tcp_saveti->m_len += sizeof(struct tcphdr); 1585 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1586 sizeof(struct tcphdr)); 1587 } 1588 nosave:; 1589 } 1590 1591 if (so->so_options & SO_ACCEPTCONN) { 1592 union syn_cache_sa src; 1593 union syn_cache_sa dst; 1594 1595 KASSERT(tp->t_state == TCPS_LISTEN); 1596 1597 memset(&src, 0, sizeof(src)); 1598 memset(&dst, 0, sizeof(dst)); 1599 switch (af) { 1600 case AF_INET: 1601 src.sin.sin_len = sizeof(struct sockaddr_in); 1602 src.sin.sin_family = AF_INET; 1603 src.sin.sin_addr = ip->ip_src; 1604 src.sin.sin_port = th->th_sport; 1605 1606 dst.sin.sin_len = sizeof(struct sockaddr_in); 1607 dst.sin.sin_family = AF_INET; 1608 dst.sin.sin_addr = ip->ip_dst; 1609 dst.sin.sin_port = th->th_dport; 1610 break; 1611 #ifdef INET6 1612 case AF_INET6: 1613 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1614 src.sin6.sin6_family = AF_INET6; 1615 src.sin6.sin6_addr = ip6->ip6_src; 1616 src.sin6.sin6_port = th->th_sport; 1617 1618 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1619 dst.sin6.sin6_family = AF_INET6; 1620 dst.sin6.sin6_addr = ip6->ip6_dst; 1621 dst.sin6.sin6_port = th->th_dport; 1622 break; 1623 #endif 1624 } 1625 1626 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1627 if (tiflags & TH_RST) { 1628 syn_cache_reset(&src.sa, &dst.sa, th); 1629 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1630 (TH_ACK|TH_SYN)) { 1631 /* 1632 * Received a SYN,ACK. This should never 1633 * happen while we are in LISTEN. Send an RST. 1634 */ 1635 goto badsyn; 1636 } else if (tiflags & TH_ACK) { 1637 so = syn_cache_get(&src.sa, &dst.sa, th, so, m); 1638 if (so == NULL) { 1639 /* 1640 * We don't have a SYN for this ACK; 1641 * send an RST. 1642 */ 1643 goto badsyn; 1644 } else if (so == (struct socket *)(-1)) { 1645 /* 1646 * We were unable to create the 1647 * connection. If the 3-way handshake 1648 * was completed, and RST has been 1649 * sent to the peer. Since the mbuf 1650 * might be in use for the reply, do 1651 * not free it. 1652 */ 1653 m = NULL; 1654 } else { 1655 /* 1656 * We have created a full-blown 1657 * connection. 1658 */ 1659 tp = NULL; 1660 inp = NULL; 1661 #ifdef INET6 1662 in6p = NULL; 1663 #endif 1664 switch (so->so_proto->pr_domain->dom_family) { 1665 case AF_INET: 1666 inp = sotoinpcb(so); 1667 tp = intotcpcb(inp); 1668 break; 1669 #ifdef INET6 1670 case AF_INET6: 1671 in6p = sotoin6pcb(so); 1672 tp = in6totcpcb(in6p); 1673 break; 1674 #endif 1675 } 1676 if (tp == NULL) 1677 goto badsyn; /*XXX*/ 1678 tiwin <<= tp->snd_scale; 1679 goto after_listen; 1680 } 1681 } else { 1682 /* 1683 * None of RST, SYN or ACK was set. 1684 * This is an invalid packet for a 1685 * TCB in LISTEN state. Send a RST. 1686 */ 1687 goto badsyn; 1688 } 1689 } else { 1690 /* 1691 * Received a SYN. 1692 */ 1693 1694 #ifdef INET6 1695 /* 1696 * If deprecated address is forbidden, we do 1697 * not accept SYN to deprecated interface 1698 * address to prevent any new inbound 1699 * connection from getting established. 1700 * When we do not accept SYN, we send a TCP 1701 * RST, with deprecated source address (instead 1702 * of dropping it). We compromise it as it is 1703 * much better for peer to send a RST, and 1704 * RST will be the final packet for the 1705 * exchange. 1706 * 1707 * If we do not forbid deprecated addresses, we 1708 * accept the SYN packet. RFC2462 does not 1709 * suggest dropping SYN in this case. 1710 * If we decipher RFC2462 5.5.4, it says like 1711 * this: 1712 * 1. use of deprecated addr with existing 1713 * communication is okay - "SHOULD continue 1714 * to be used" 1715 * 2. use of it with new communication: 1716 * (2a) "SHOULD NOT be used if alternate 1717 * address with sufficient scope is 1718 * available" 1719 * (2b) nothing mentioned otherwise. 1720 * Here we fall into (2b) case as we have no 1721 * choice in our source address selection - we 1722 * must obey the peer. 1723 * 1724 * The wording in RFC2462 is confusing, and 1725 * there are multiple description text for 1726 * deprecated address handling - worse, they 1727 * are not exactly the same. I believe 5.5.4 1728 * is the best one, so we follow 5.5.4. 1729 */ 1730 if (af == AF_INET6 && !ip6_use_deprecated) { 1731 struct in6_ifaddr *ia6; 1732 int s; 1733 struct ifnet *rcvif = m_get_rcvif(m, &s); 1734 if (rcvif == NULL) 1735 goto dropwithreset; /* XXX */ 1736 if ((ia6 = in6ifa_ifpwithaddr(rcvif, 1737 &ip6->ip6_dst)) && 1738 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1739 tp = NULL; 1740 m_put_rcvif(rcvif, &s); 1741 goto dropwithreset; 1742 } 1743 m_put_rcvif(rcvif, &s); 1744 } 1745 #endif 1746 1747 /* 1748 * LISTEN socket received a SYN from itself? This 1749 * can't possibly be valid; drop the packet. 1750 */ 1751 if (th->th_sport == th->th_dport) { 1752 int eq = 0; 1753 1754 switch (af) { 1755 case AF_INET: 1756 eq = in_hosteq(ip->ip_src, ip->ip_dst); 1757 break; 1758 #ifdef INET6 1759 case AF_INET6: 1760 eq = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, 1761 &ip6->ip6_dst); 1762 break; 1763 #endif 1764 } 1765 if (eq) { 1766 TCP_STATINC(TCP_STAT_BADSYN); 1767 goto drop; 1768 } 1769 } 1770 1771 /* 1772 * SYN looks ok; create compressed TCP 1773 * state for it. 1774 */ 1775 if (so->so_qlen <= so->so_qlimit && 1776 syn_cache_add(&src.sa, &dst.sa, th, toff, 1777 so, m, optp, optlen, &opti)) 1778 m = NULL; 1779 } 1780 1781 goto drop; 1782 } 1783 1784 after_listen: 1785 /* 1786 * From here on, we're dealing with !LISTEN. 1787 */ 1788 KASSERT(tp->t_state != TCPS_LISTEN); 1789 1790 /* 1791 * Segment received on connection. 1792 * Reset idle time and keep-alive timer. 1793 */ 1794 tp->t_rcvtime = tcp_now; 1795 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1796 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1797 1798 /* 1799 * Process options. 1800 */ 1801 #ifdef TCP_SIGNATURE 1802 if (optp || (tp->t_flags & TF_SIGNATURE)) 1803 #else 1804 if (optp) 1805 #endif 1806 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1807 goto drop; 1808 1809 if (TCP_SACK_ENABLED(tp)) { 1810 tcp_del_sackholes(tp, th); 1811 } 1812 1813 if (TCP_ECN_ALLOWED(tp)) { 1814 if (tiflags & TH_CWR) { 1815 tp->t_flags &= ~TF_ECN_SND_ECE; 1816 } 1817 switch (iptos & IPTOS_ECN_MASK) { 1818 case IPTOS_ECN_CE: 1819 tp->t_flags |= TF_ECN_SND_ECE; 1820 TCP_STATINC(TCP_STAT_ECN_CE); 1821 break; 1822 case IPTOS_ECN_ECT0: 1823 TCP_STATINC(TCP_STAT_ECN_ECT); 1824 break; 1825 case IPTOS_ECN_ECT1: 1826 /* XXX: ignore for now -- rpaulo */ 1827 break; 1828 } 1829 /* 1830 * Congestion experienced. 1831 * Ignore if we are already trying to recover. 1832 */ 1833 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1834 tp->t_congctl->cong_exp(tp); 1835 } 1836 1837 if (opti.ts_present && opti.ts_ecr) { 1838 /* 1839 * Calculate the RTT from the returned time stamp and the 1840 * connection's time base. If the time stamp is later than 1841 * the current time, or is extremely old, fall back to non-1323 1842 * RTT calculation. Since ts_rtt is unsigned, we can test both 1843 * at the same time. 1844 * 1845 * Note that ts_rtt is in units of slow ticks (500 1846 * ms). Since most earthbound RTTs are < 500 ms, 1847 * observed values will have large quantization noise. 1848 * Our smoothed RTT is then the fraction of observed 1849 * samples that are 1 tick instead of 0 (times 500 1850 * ms). 1851 * 1852 * ts_rtt is increased by 1 to denote a valid sample, 1853 * with 0 indicating an invalid measurement. This 1854 * extra 1 must be removed when ts_rtt is used, or 1855 * else an an erroneous extra 500 ms will result. 1856 */ 1857 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1858 if (ts_rtt > TCP_PAWS_IDLE) 1859 ts_rtt = 0; 1860 } else { 1861 ts_rtt = 0; 1862 } 1863 1864 /* 1865 * Fast path: check for the two common cases of a uni-directional 1866 * data transfer. If: 1867 * o We are in the ESTABLISHED state, and 1868 * o The packet has no control flags, and 1869 * o The packet is in-sequence, and 1870 * o The window didn't change, and 1871 * o We are not retransmitting 1872 * It's a candidate. 1873 * 1874 * If the length (tlen) is zero and the ack moved forward, we're 1875 * the sender side of the transfer. Just free the data acked and 1876 * wake any higher level process that was blocked waiting for 1877 * space. 1878 * 1879 * If the length is non-zero and the ack didn't move, we're the 1880 * receiver side. If we're getting packets in-order (the reassembly 1881 * queue is empty), add the data to the socket buffer and note 1882 * that we need a delayed ack. 1883 */ 1884 if (tp->t_state == TCPS_ESTABLISHED && 1885 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1886 == TH_ACK && 1887 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1888 th->th_seq == tp->rcv_nxt && 1889 tiwin && tiwin == tp->snd_wnd && 1890 tp->snd_nxt == tp->snd_max) { 1891 1892 /* 1893 * If last ACK falls within this segment's sequence numbers, 1894 * record the timestamp. 1895 * NOTE that the test is modified according to the latest 1896 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 1897 * 1898 * note that we already know 1899 * TSTMP_GEQ(opti.ts_val, tp->ts_recent) 1900 */ 1901 if (opti.ts_present && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1902 tp->ts_recent_age = tcp_now; 1903 tp->ts_recent = opti.ts_val; 1904 } 1905 1906 if (tlen == 0) { 1907 /* Ack prediction. */ 1908 if (SEQ_GT(th->th_ack, tp->snd_una) && 1909 SEQ_LEQ(th->th_ack, tp->snd_max) && 1910 tp->snd_cwnd >= tp->snd_wnd && 1911 tp->t_partialacks < 0) { 1912 /* 1913 * this is a pure ack for outstanding data. 1914 */ 1915 if (ts_rtt) 1916 tcp_xmit_timer(tp, ts_rtt - 1); 1917 else if (tp->t_rtttime && 1918 SEQ_GT(th->th_ack, tp->t_rtseq)) 1919 tcp_xmit_timer(tp, 1920 tcp_now - tp->t_rtttime); 1921 acked = th->th_ack - tp->snd_una; 1922 tcps = TCP_STAT_GETREF(); 1923 tcps[TCP_STAT_PREDACK]++; 1924 tcps[TCP_STAT_RCVACKPACK]++; 1925 tcps[TCP_STAT_RCVACKBYTE] += acked; 1926 TCP_STAT_PUTREF(); 1927 nd6_hint(tp); 1928 1929 if (acked > (tp->t_lastoff - tp->t_inoff)) 1930 tp->t_lastm = NULL; 1931 sbdrop(&so->so_snd, acked); 1932 tp->t_lastoff -= acked; 1933 1934 icmp_check(tp, th, acked); 1935 1936 tp->snd_una = th->th_ack; 1937 tp->snd_fack = tp->snd_una; 1938 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1939 tp->snd_high = tp->snd_una; 1940 m_freem(m); 1941 1942 /* 1943 * If all outstanding data are acked, stop 1944 * retransmit timer, otherwise restart timer 1945 * using current (possibly backed-off) value. 1946 * If process is waiting for space, 1947 * wakeup/selnotify/signal. If data 1948 * are ready to send, let tcp_output 1949 * decide between more output or persist. 1950 */ 1951 if (tp->snd_una == tp->snd_max) 1952 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1953 else if (TCP_TIMER_ISARMED(tp, 1954 TCPT_PERSIST) == 0) 1955 TCP_TIMER_ARM(tp, TCPT_REXMT, 1956 tp->t_rxtcur); 1957 1958 sowwakeup(so); 1959 if (so->so_snd.sb_cc) { 1960 KERNEL_LOCK(1, NULL); 1961 (void)tcp_output(tp); 1962 KERNEL_UNLOCK_ONE(NULL); 1963 } 1964 if (tcp_saveti) 1965 m_freem(tcp_saveti); 1966 return; 1967 } 1968 } else if (th->th_ack == tp->snd_una && 1969 TAILQ_FIRST(&tp->segq) == NULL && 1970 tlen <= sbspace(&so->so_rcv)) { 1971 int newsize = 0; 1972 1973 /* 1974 * this is a pure, in-sequence data packet 1975 * with nothing on the reassembly queue and 1976 * we have enough buffer space to take it. 1977 */ 1978 tp->rcv_nxt += tlen; 1979 tcps = TCP_STAT_GETREF(); 1980 tcps[TCP_STAT_PREDDAT]++; 1981 tcps[TCP_STAT_RCVPACK]++; 1982 tcps[TCP_STAT_RCVBYTE] += tlen; 1983 TCP_STAT_PUTREF(); 1984 nd6_hint(tp); 1985 1986 /* 1987 * Automatic sizing enables the performance of large buffers 1988 * and most of the efficiency of small ones by only allocating 1989 * space when it is needed. 1990 * 1991 * On the receive side the socket buffer memory is only rarely 1992 * used to any significant extent. This allows us to be much 1993 * more aggressive in scaling the receive socket buffer. For 1994 * the case that the buffer space is actually used to a large 1995 * extent and we run out of kernel memory we can simply drop 1996 * the new segments; TCP on the sender will just retransmit it 1997 * later. Setting the buffer size too big may only consume too 1998 * much kernel memory if the application doesn't read() from 1999 * the socket or packet loss or reordering makes use of the 2000 * reassembly queue. 2001 * 2002 * The criteria to step up the receive buffer one notch are: 2003 * 1. the number of bytes received during the time it takes 2004 * one timestamp to be reflected back to us (the RTT); 2005 * 2. received bytes per RTT is within seven eighth of the 2006 * current socket buffer size; 2007 * 3. receive buffer size has not hit maximal automatic size; 2008 * 2009 * This algorithm does one step per RTT at most and only if 2010 * we receive a bulk stream w/o packet losses or reorderings. 2011 * Shrinking the buffer during idle times is not necessary as 2012 * it doesn't consume any memory when idle. 2013 * 2014 * TODO: Only step up if the application is actually serving 2015 * the buffer to better manage the socket buffer resources. 2016 */ 2017 if (tcp_do_autorcvbuf && 2018 opti.ts_ecr && 2019 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 2020 if (opti.ts_ecr > tp->rfbuf_ts && 2021 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 2022 if (tp->rfbuf_cnt > 2023 (so->so_rcv.sb_hiwat / 8 * 7) && 2024 so->so_rcv.sb_hiwat < 2025 tcp_autorcvbuf_max) { 2026 newsize = 2027 min(so->so_rcv.sb_hiwat + 2028 tcp_autorcvbuf_inc, 2029 tcp_autorcvbuf_max); 2030 } 2031 /* Start over with next RTT. */ 2032 tp->rfbuf_ts = 0; 2033 tp->rfbuf_cnt = 0; 2034 } else 2035 tp->rfbuf_cnt += tlen; /* add up */ 2036 } 2037 2038 /* 2039 * Drop TCP, IP headers and TCP options then add data 2040 * to socket buffer. 2041 */ 2042 if (so->so_state & SS_CANTRCVMORE) { 2043 m_freem(m); 2044 } else { 2045 /* 2046 * Set new socket buffer size. 2047 * Give up when limit is reached. 2048 */ 2049 if (newsize) 2050 if (!sbreserve(&so->so_rcv, 2051 newsize, so)) 2052 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 2053 m_adj(m, toff + off); 2054 sbappendstream(&so->so_rcv, m); 2055 } 2056 sorwakeup(so); 2057 tcp_setup_ack(tp, th); 2058 if (tp->t_flags & TF_ACKNOW) { 2059 KERNEL_LOCK(1, NULL); 2060 (void)tcp_output(tp); 2061 KERNEL_UNLOCK_ONE(NULL); 2062 } 2063 if (tcp_saveti) 2064 m_freem(tcp_saveti); 2065 return; 2066 } 2067 } 2068 2069 /* 2070 * Compute mbuf offset to TCP data segment. 2071 */ 2072 hdroptlen = toff + off; 2073 2074 /* 2075 * Calculate amount of space in receive window. Receive window is 2076 * amount of space in rcv queue, but not less than advertised 2077 * window. 2078 */ 2079 { 2080 int win; 2081 win = sbspace(&so->so_rcv); 2082 if (win < 0) 2083 win = 0; 2084 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 2085 } 2086 2087 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 2088 tp->rfbuf_ts = 0; 2089 tp->rfbuf_cnt = 0; 2090 2091 switch (tp->t_state) { 2092 /* 2093 * If the state is SYN_SENT: 2094 * if seg contains an ACK, but not for our SYN, drop the input. 2095 * if seg contains a RST, then drop the connection. 2096 * if seg does not contain SYN, then drop it. 2097 * Otherwise this is an acceptable SYN segment 2098 * initialize tp->rcv_nxt and tp->irs 2099 * if seg contains ack then advance tp->snd_una 2100 * if seg contains a ECE and ECN support is enabled, the stream 2101 * is ECN capable. 2102 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2103 * arrange for segment to be acked (eventually) 2104 * continue processing rest of data/controls, beginning with URG 2105 */ 2106 case TCPS_SYN_SENT: 2107 if ((tiflags & TH_ACK) && 2108 (SEQ_LEQ(th->th_ack, tp->iss) || 2109 SEQ_GT(th->th_ack, tp->snd_max))) 2110 goto dropwithreset; 2111 if (tiflags & TH_RST) { 2112 if (tiflags & TH_ACK) 2113 tp = tcp_drop(tp, ECONNREFUSED); 2114 goto drop; 2115 } 2116 if ((tiflags & TH_SYN) == 0) 2117 goto drop; 2118 if (tiflags & TH_ACK) { 2119 tp->snd_una = th->th_ack; 2120 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2121 tp->snd_nxt = tp->snd_una; 2122 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2123 tp->snd_high = tp->snd_una; 2124 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2125 2126 if ((tiflags & TH_ECE) && tcp_do_ecn) { 2127 tp->t_flags |= TF_ECN_PERMIT; 2128 TCP_STATINC(TCP_STAT_ECN_SHS); 2129 } 2130 } 2131 tp->irs = th->th_seq; 2132 tcp_rcvseqinit(tp); 2133 tp->t_flags |= TF_ACKNOW; 2134 tcp_mss_from_peer(tp, opti.maxseg); 2135 2136 /* 2137 * Initialize the initial congestion window. If we 2138 * had to retransmit the SYN, we must initialize cwnd 2139 * to 1 segment (i.e. the Loss Window). 2140 */ 2141 if (tp->t_flags & TF_SYN_REXMT) 2142 tp->snd_cwnd = tp->t_peermss; 2143 else { 2144 int ss = tcp_init_win; 2145 if (inp != NULL && in_localaddr(inp->inp_faddr)) 2146 ss = tcp_init_win_local; 2147 #ifdef INET6 2148 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 2149 ss = tcp_init_win_local; 2150 #endif 2151 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 2152 } 2153 2154 tcp_rmx_rtt(tp); 2155 if (tiflags & TH_ACK) { 2156 TCP_STATINC(TCP_STAT_CONNECTS); 2157 /* 2158 * move tcp_established before soisconnected 2159 * because upcall handler can drive tcp_output 2160 * functionality. 2161 * XXX we might call soisconnected at the end of 2162 * all processing 2163 */ 2164 tcp_established(tp); 2165 soisconnected(so); 2166 /* Do window scaling on this connection? */ 2167 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2168 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2169 tp->snd_scale = tp->requested_s_scale; 2170 tp->rcv_scale = tp->request_r_scale; 2171 } 2172 TCP_REASS_LOCK(tp); 2173 (void)tcp_reass(tp, NULL, NULL, tlen); 2174 /* 2175 * if we didn't have to retransmit the SYN, 2176 * use its rtt as our initial srtt & rtt var. 2177 */ 2178 if (tp->t_rtttime) 2179 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2180 } else { 2181 tp->t_state = TCPS_SYN_RECEIVED; 2182 } 2183 2184 /* 2185 * Advance th->th_seq to correspond to first data byte. 2186 * If data, trim to stay within window, 2187 * dropping FIN if necessary. 2188 */ 2189 th->th_seq++; 2190 if (tlen > tp->rcv_wnd) { 2191 todrop = tlen - tp->rcv_wnd; 2192 m_adj(m, -todrop); 2193 tlen = tp->rcv_wnd; 2194 tiflags &= ~TH_FIN; 2195 tcps = TCP_STAT_GETREF(); 2196 tcps[TCP_STAT_RCVPACKAFTERWIN]++; 2197 tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop; 2198 TCP_STAT_PUTREF(); 2199 } 2200 tp->snd_wl1 = th->th_seq - 1; 2201 tp->rcv_up = th->th_seq; 2202 goto step6; 2203 2204 /* 2205 * If the state is SYN_RECEIVED: 2206 * If seg contains an ACK, but not for our SYN, drop the input 2207 * and generate an RST. See page 36, rfc793 2208 */ 2209 case TCPS_SYN_RECEIVED: 2210 if ((tiflags & TH_ACK) && 2211 (SEQ_LEQ(th->th_ack, tp->iss) || 2212 SEQ_GT(th->th_ack, tp->snd_max))) 2213 goto dropwithreset; 2214 break; 2215 } 2216 2217 /* 2218 * From here on, we're dealing with !LISTEN and !SYN_SENT. 2219 */ 2220 KASSERT(tp->t_state != TCPS_LISTEN && 2221 tp->t_state != TCPS_SYN_SENT); 2222 2223 /* 2224 * RFC1323 PAWS: if we have a timestamp reply on this segment and 2225 * it's less than ts_recent, drop it. 2226 */ 2227 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2228 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2229 /* Check to see if ts_recent is over 24 days old. */ 2230 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2231 /* 2232 * Invalidate ts_recent. If this segment updates 2233 * ts_recent, the age will be reset later and ts_recent 2234 * will get a valid value. If it does not, setting 2235 * ts_recent to zero will at least satisfy the 2236 * requirement that zero be placed in the timestamp 2237 * echo reply when ts_recent isn't valid. The 2238 * age isn't reset until we get a valid ts_recent 2239 * because we don't want out-of-order segments to be 2240 * dropped when ts_recent is old. 2241 */ 2242 tp->ts_recent = 0; 2243 } else { 2244 tcps = TCP_STAT_GETREF(); 2245 tcps[TCP_STAT_RCVDUPPACK]++; 2246 tcps[TCP_STAT_RCVDUPBYTE] += tlen; 2247 tcps[TCP_STAT_PAWSDROP]++; 2248 TCP_STAT_PUTREF(); 2249 tcp_new_dsack(tp, th->th_seq, tlen); 2250 goto dropafterack; 2251 } 2252 } 2253 2254 /* 2255 * Check that at least some bytes of the segment are within the 2256 * receive window. If segment begins before rcv_nxt, drop leading 2257 * data (and SYN); if nothing left, just ack. 2258 */ 2259 todrop = tp->rcv_nxt - th->th_seq; 2260 dupseg = false; 2261 if (todrop > 0) { 2262 if (tiflags & TH_SYN) { 2263 tiflags &= ~TH_SYN; 2264 th->th_seq++; 2265 tcp_urp_drop(th, 1, &tiflags); 2266 todrop--; 2267 } 2268 if (todrop > tlen || 2269 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2270 /* 2271 * Any valid FIN or RST must be to the left of the 2272 * window. At this point the FIN or RST must be a 2273 * duplicate or out of sequence; drop it. 2274 */ 2275 if (tiflags & TH_RST) 2276 goto drop; 2277 tiflags &= ~(TH_FIN|TH_RST); 2278 2279 /* 2280 * Send an ACK to resynchronize and drop any data. 2281 * But keep on processing for RST or ACK. 2282 */ 2283 tp->t_flags |= TF_ACKNOW; 2284 todrop = tlen; 2285 dupseg = true; 2286 tcps = TCP_STAT_GETREF(); 2287 tcps[TCP_STAT_RCVDUPPACK]++; 2288 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 2289 TCP_STAT_PUTREF(); 2290 } else if ((tiflags & TH_RST) && th->th_seq != tp->rcv_nxt) { 2291 /* 2292 * Test for reset before adjusting the sequence 2293 * number for overlapping data. 2294 */ 2295 goto dropafterack_ratelim; 2296 } else { 2297 tcps = TCP_STAT_GETREF(); 2298 tcps[TCP_STAT_RCVPARTDUPPACK]++; 2299 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 2300 TCP_STAT_PUTREF(); 2301 } 2302 tcp_new_dsack(tp, th->th_seq, todrop); 2303 hdroptlen += todrop; /* drop from head afterwards */ 2304 th->th_seq += todrop; 2305 tlen -= todrop; 2306 tcp_urp_drop(th, todrop, &tiflags); 2307 } 2308 2309 /* 2310 * If new data is received on a connection after the user processes 2311 * are gone, then RST the other end. 2312 */ 2313 if ((so->so_state & SS_NOFDREF) && 2314 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2315 tp = tcp_close(tp); 2316 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 2317 goto dropwithreset; 2318 } 2319 2320 /* 2321 * If the segment ends after the window, drop trailing data (and 2322 * PUSH and FIN); if nothing left, just ACK. 2323 */ 2324 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2325 if (todrop > 0) { 2326 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 2327 if (todrop >= tlen) { 2328 /* 2329 * The segment actually starts after the window. 2330 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2331 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2332 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2333 */ 2334 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 2335 2336 /* 2337 * If a new connection request is received while in 2338 * TIME_WAIT, drop the old connection and start over 2339 * if the sequence numbers are above the previous 2340 * ones. 2341 * 2342 * NOTE: We need to put the header fields back into 2343 * network order. 2344 */ 2345 if ((tiflags & TH_SYN) && 2346 tp->t_state == TCPS_TIME_WAIT && 2347 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2348 tp = tcp_close(tp); 2349 tcp_fields_to_net(th); 2350 m_freem(tcp_saveti); 2351 tcp_saveti = NULL; 2352 goto findpcb; 2353 } 2354 2355 /* 2356 * If window is closed can only take segments at 2357 * window edge, and have to drop data and PUSH from 2358 * incoming segments. Continue processing, but 2359 * remember to ack. Otherwise, drop segment 2360 * and (if not RST) ack. 2361 */ 2362 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2363 KASSERT(todrop == tlen); 2364 tp->t_flags |= TF_ACKNOW; 2365 TCP_STATINC(TCP_STAT_RCVWINPROBE); 2366 } else { 2367 goto dropafterack; 2368 } 2369 } else { 2370 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 2371 } 2372 m_adj(m, -todrop); 2373 tlen -= todrop; 2374 tiflags &= ~(TH_PUSH|TH_FIN); 2375 } 2376 2377 /* 2378 * If last ACK falls within this segment's sequence numbers, 2379 * record the timestamp. 2380 * NOTE: 2381 * 1) That the test incorporates suggestions from the latest 2382 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 2383 * 2) That updating only on newer timestamps interferes with 2384 * our earlier PAWS tests, so this check should be solely 2385 * predicated on the sequence space of this segment. 2386 * 3) That we modify the segment boundary check to be 2387 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2388 * instead of RFC1323's 2389 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2390 * This modified check allows us to overcome RFC1323's 2391 * limitations as described in Stevens TCP/IP Illustrated 2392 * Vol. 2 p.869. In such cases, we can still calculate the 2393 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2394 */ 2395 if (opti.ts_present && 2396 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2397 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2398 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2399 tp->ts_recent_age = tcp_now; 2400 tp->ts_recent = opti.ts_val; 2401 } 2402 2403 /* 2404 * If the RST bit is set examine the state: 2405 * RECEIVED state: 2406 * If passive open, return to LISTEN state. 2407 * If active open, inform user that connection was refused. 2408 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT states: 2409 * Inform user that connection was reset, and close tcb. 2410 * CLOSING, LAST_ACK, TIME_WAIT states: 2411 * Close the tcb. 2412 */ 2413 if (tiflags & TH_RST) { 2414 if (th->th_seq != tp->rcv_nxt) 2415 goto dropafterack_ratelim; 2416 2417 switch (tp->t_state) { 2418 case TCPS_SYN_RECEIVED: 2419 so->so_error = ECONNREFUSED; 2420 goto close; 2421 2422 case TCPS_ESTABLISHED: 2423 case TCPS_FIN_WAIT_1: 2424 case TCPS_FIN_WAIT_2: 2425 case TCPS_CLOSE_WAIT: 2426 so->so_error = ECONNRESET; 2427 close: 2428 tp->t_state = TCPS_CLOSED; 2429 TCP_STATINC(TCP_STAT_DROPS); 2430 tp = tcp_close(tp); 2431 goto drop; 2432 2433 case TCPS_CLOSING: 2434 case TCPS_LAST_ACK: 2435 case TCPS_TIME_WAIT: 2436 tp = tcp_close(tp); 2437 goto drop; 2438 } 2439 } 2440 2441 /* 2442 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2443 * we must be in a synchronized state. RFC791 states (under RST 2444 * generation) that any unacceptable segment (an out-of-order SYN 2445 * qualifies) received in a synchronized state must elicit only an 2446 * empty acknowledgment segment ... and the connection remains in 2447 * the same state. 2448 */ 2449 if (tiflags & TH_SYN) { 2450 if (tp->rcv_nxt == th->th_seq) { 2451 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2452 TH_ACK); 2453 if (tcp_saveti) 2454 m_freem(tcp_saveti); 2455 return; 2456 } 2457 2458 goto dropafterack_ratelim; 2459 } 2460 2461 /* 2462 * If the ACK bit is off we drop the segment and return. 2463 */ 2464 if ((tiflags & TH_ACK) == 0) { 2465 if (tp->t_flags & TF_ACKNOW) 2466 goto dropafterack; 2467 goto drop; 2468 } 2469 2470 /* 2471 * From here on, we're doing ACK processing. 2472 */ 2473 2474 switch (tp->t_state) { 2475 /* 2476 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2477 * ESTABLISHED state and continue processing, otherwise 2478 * send an RST. 2479 */ 2480 case TCPS_SYN_RECEIVED: 2481 if (SEQ_GT(tp->snd_una, th->th_ack) || 2482 SEQ_GT(th->th_ack, tp->snd_max)) 2483 goto dropwithreset; 2484 TCP_STATINC(TCP_STAT_CONNECTS); 2485 soisconnected(so); 2486 tcp_established(tp); 2487 /* Do window scaling? */ 2488 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2489 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2490 tp->snd_scale = tp->requested_s_scale; 2491 tp->rcv_scale = tp->request_r_scale; 2492 } 2493 TCP_REASS_LOCK(tp); 2494 (void)tcp_reass(tp, NULL, NULL, tlen); 2495 tp->snd_wl1 = th->th_seq - 1; 2496 /* FALLTHROUGH */ 2497 2498 /* 2499 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2500 * ACKs. If the ack is in the range 2501 * tp->snd_una < th->th_ack <= tp->snd_max 2502 * then advance tp->snd_una to th->th_ack and drop 2503 * data from the retransmission queue. If this ACK reflects 2504 * more up to date window information we update our window information. 2505 */ 2506 case TCPS_ESTABLISHED: 2507 case TCPS_FIN_WAIT_1: 2508 case TCPS_FIN_WAIT_2: 2509 case TCPS_CLOSE_WAIT: 2510 case TCPS_CLOSING: 2511 case TCPS_LAST_ACK: 2512 case TCPS_TIME_WAIT: 2513 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2514 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2515 TCP_STATINC(TCP_STAT_RCVDUPACK); 2516 /* 2517 * If we have outstanding data (other than 2518 * a window probe), this is a completely 2519 * duplicate ack (ie, window info didn't 2520 * change), the ack is the biggest we've 2521 * seen and we've seen exactly our rexmt 2522 * threshhold of them, assume a packet 2523 * has been dropped and retransmit it. 2524 * Kludge snd_nxt & the congestion 2525 * window so we send only this one 2526 * packet. 2527 */ 2528 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2529 th->th_ack != tp->snd_una) 2530 tp->t_dupacks = 0; 2531 else if (tp->t_partialacks < 0 && 2532 (++tp->t_dupacks == tcprexmtthresh || 2533 TCP_FACK_FASTRECOV(tp))) { 2534 /* 2535 * Do the fast retransmit, and adjust 2536 * congestion control paramenters. 2537 */ 2538 if (tp->t_congctl->fast_retransmit(tp, th)) { 2539 /* False fast retransmit */ 2540 break; 2541 } 2542 goto drop; 2543 } else if (tp->t_dupacks > tcprexmtthresh) { 2544 tp->snd_cwnd += tp->t_segsz; 2545 KERNEL_LOCK(1, NULL); 2546 (void)tcp_output(tp); 2547 KERNEL_UNLOCK_ONE(NULL); 2548 goto drop; 2549 } 2550 } else { 2551 /* 2552 * If the ack appears to be very old, only 2553 * allow data that is in-sequence. This 2554 * makes it somewhat more difficult to insert 2555 * forged data by guessing sequence numbers. 2556 * Sent an ack to try to update the send 2557 * sequence number on the other side. 2558 */ 2559 if (tlen && th->th_seq != tp->rcv_nxt && 2560 SEQ_LT(th->th_ack, 2561 tp->snd_una - tp->max_sndwnd)) 2562 goto dropafterack; 2563 } 2564 break; 2565 } 2566 /* 2567 * If the congestion window was inflated to account 2568 * for the other side's cached packets, retract it. 2569 */ 2570 tp->t_congctl->fast_retransmit_newack(tp, th); 2571 2572 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2573 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2574 goto dropafterack; 2575 } 2576 acked = th->th_ack - tp->snd_una; 2577 tcps = TCP_STAT_GETREF(); 2578 tcps[TCP_STAT_RCVACKPACK]++; 2579 tcps[TCP_STAT_RCVACKBYTE] += acked; 2580 TCP_STAT_PUTREF(); 2581 2582 /* 2583 * If we have a timestamp reply, update smoothed 2584 * round trip time. If no timestamp is present but 2585 * transmit timer is running and timed sequence 2586 * number was acked, update smoothed round trip time. 2587 * Since we now have an rtt measurement, cancel the 2588 * timer backoff (cf., Phil Karn's retransmit alg.). 2589 * Recompute the initial retransmit timer. 2590 */ 2591 if (ts_rtt) 2592 tcp_xmit_timer(tp, ts_rtt - 1); 2593 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2594 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2595 2596 /* 2597 * If all outstanding data is acked, stop retransmit 2598 * timer and remember to restart (more output or persist). 2599 * If there is more data to be acked, restart retransmit 2600 * timer, using current (possibly backed-off) value. 2601 */ 2602 if (th->th_ack == tp->snd_max) { 2603 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2604 needoutput = 1; 2605 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2606 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2607 2608 /* 2609 * New data has been acked, adjust the congestion window. 2610 */ 2611 tp->t_congctl->newack(tp, th); 2612 2613 nd6_hint(tp); 2614 if (acked > so->so_snd.sb_cc) { 2615 tp->snd_wnd -= so->so_snd.sb_cc; 2616 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2617 ourfinisacked = 1; 2618 } else { 2619 if (acked > (tp->t_lastoff - tp->t_inoff)) 2620 tp->t_lastm = NULL; 2621 sbdrop(&so->so_snd, acked); 2622 tp->t_lastoff -= acked; 2623 if (tp->snd_wnd > acked) 2624 tp->snd_wnd -= acked; 2625 else 2626 tp->snd_wnd = 0; 2627 ourfinisacked = 0; 2628 } 2629 sowwakeup(so); 2630 2631 icmp_check(tp, th, acked); 2632 2633 tp->snd_una = th->th_ack; 2634 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2635 tp->snd_fack = tp->snd_una; 2636 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2637 tp->snd_nxt = tp->snd_una; 2638 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2639 tp->snd_high = tp->snd_una; 2640 2641 switch (tp->t_state) { 2642 2643 /* 2644 * In FIN_WAIT_1 STATE in addition to the processing 2645 * for the ESTABLISHED state if our FIN is now acknowledged 2646 * then enter FIN_WAIT_2. 2647 */ 2648 case TCPS_FIN_WAIT_1: 2649 if (ourfinisacked) { 2650 /* 2651 * If we can't receive any more 2652 * data, then closing user can proceed. 2653 * Starting the timer is contrary to the 2654 * specification, but if we don't get a FIN 2655 * we'll hang forever. 2656 */ 2657 if (so->so_state & SS_CANTRCVMORE) { 2658 soisdisconnected(so); 2659 if (tp->t_maxidle > 0) 2660 TCP_TIMER_ARM(tp, TCPT_2MSL, 2661 tp->t_maxidle); 2662 } 2663 tp->t_state = TCPS_FIN_WAIT_2; 2664 } 2665 break; 2666 2667 /* 2668 * In CLOSING STATE in addition to the processing for 2669 * the ESTABLISHED state if the ACK acknowledges our FIN 2670 * then enter the TIME-WAIT state, otherwise ignore 2671 * the segment. 2672 */ 2673 case TCPS_CLOSING: 2674 if (ourfinisacked) { 2675 tp->t_state = TCPS_TIME_WAIT; 2676 tcp_canceltimers(tp); 2677 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2678 soisdisconnected(so); 2679 } 2680 break; 2681 2682 /* 2683 * In LAST_ACK, we may still be waiting for data to drain 2684 * and/or to be acked, as well as for the ack of our FIN. 2685 * If our FIN is now acknowledged, delete the TCB, 2686 * enter the closed state and return. 2687 */ 2688 case TCPS_LAST_ACK: 2689 if (ourfinisacked) { 2690 tp = tcp_close(tp); 2691 goto drop; 2692 } 2693 break; 2694 2695 /* 2696 * In TIME_WAIT state the only thing that should arrive 2697 * is a retransmission of the remote FIN. Acknowledge 2698 * it and restart the finack timer. 2699 */ 2700 case TCPS_TIME_WAIT: 2701 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2702 goto dropafterack; 2703 } 2704 } 2705 2706 step6: 2707 /* 2708 * Update window information. 2709 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2710 */ 2711 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2712 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2713 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2714 /* keep track of pure window updates */ 2715 if (tlen == 0 && 2716 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2717 TCP_STATINC(TCP_STAT_RCVWINUPD); 2718 tp->snd_wnd = tiwin; 2719 tp->snd_wl1 = th->th_seq; 2720 tp->snd_wl2 = th->th_ack; 2721 if (tp->snd_wnd > tp->max_sndwnd) 2722 tp->max_sndwnd = tp->snd_wnd; 2723 needoutput = 1; 2724 } 2725 2726 /* 2727 * Process segments with URG. 2728 */ 2729 if ((tiflags & TH_URG) && th->th_urp && 2730 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2731 /* 2732 * This is a kludge, but if we receive and accept 2733 * random urgent pointers, we'll crash in 2734 * soreceive. It's hard to imagine someone 2735 * actually wanting to send this much urgent data. 2736 */ 2737 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2738 th->th_urp = 0; /* XXX */ 2739 tiflags &= ~TH_URG; /* XXX */ 2740 goto dodata; /* XXX */ 2741 } 2742 2743 /* 2744 * If this segment advances the known urgent pointer, 2745 * then mark the data stream. This should not happen 2746 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2747 * a FIN has been received from the remote side. 2748 * In these states we ignore the URG. 2749 * 2750 * According to RFC961 (Assigned Protocols), 2751 * the urgent pointer points to the last octet 2752 * of urgent data. We continue, however, 2753 * to consider it to indicate the first octet 2754 * of data past the urgent section as the original 2755 * spec states (in one of two places). 2756 */ 2757 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2758 tp->rcv_up = th->th_seq + th->th_urp; 2759 so->so_oobmark = so->so_rcv.sb_cc + 2760 (tp->rcv_up - tp->rcv_nxt) - 1; 2761 if (so->so_oobmark == 0) 2762 so->so_state |= SS_RCVATMARK; 2763 sohasoutofband(so); 2764 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2765 } 2766 2767 /* 2768 * Remove out of band data so doesn't get presented to user. 2769 * This can happen independent of advancing the URG pointer, 2770 * but if two URG's are pending at once, some out-of-band 2771 * data may creep in... ick. 2772 */ 2773 if (th->th_urp <= (u_int16_t)tlen && 2774 (so->so_options & SO_OOBINLINE) == 0) 2775 tcp_pulloutofband(so, th, m, hdroptlen); 2776 } else { 2777 /* 2778 * If no out of band data is expected, 2779 * pull receive urgent pointer along 2780 * with the receive window. 2781 */ 2782 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2783 tp->rcv_up = tp->rcv_nxt; 2784 } 2785 dodata: 2786 2787 /* 2788 * Process the segment text, merging it into the TCP sequencing queue, 2789 * and arranging for acknowledgement of receipt if necessary. 2790 * This process logically involves adjusting tp->rcv_wnd as data 2791 * is presented to the user (this happens in tcp_usrreq.c, 2792 * tcp_rcvd()). If a FIN has already been received on this 2793 * connection then we just ignore the text. 2794 */ 2795 if ((tlen || (tiflags & TH_FIN)) && 2796 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2797 /* 2798 * Insert segment ti into reassembly queue of tcp with 2799 * control block tp. Return TH_FIN if reassembly now includes 2800 * a segment with FIN. The macro form does the common case 2801 * inline (segment is the next to be received on an 2802 * established connection, and the queue is empty), 2803 * avoiding linkage into and removal from the queue and 2804 * repetition of various conversions. 2805 * Set DELACK for segments received in order, but ack 2806 * immediately when segments are out of order 2807 * (so fast retransmit can work). 2808 */ 2809 /* NOTE: this was TCP_REASS() macro, but used only once */ 2810 TCP_REASS_LOCK(tp); 2811 if (th->th_seq == tp->rcv_nxt && 2812 TAILQ_FIRST(&tp->segq) == NULL && 2813 tp->t_state == TCPS_ESTABLISHED) { 2814 tcp_setup_ack(tp, th); 2815 tp->rcv_nxt += tlen; 2816 tiflags = th->th_flags & TH_FIN; 2817 tcps = TCP_STAT_GETREF(); 2818 tcps[TCP_STAT_RCVPACK]++; 2819 tcps[TCP_STAT_RCVBYTE] += tlen; 2820 TCP_STAT_PUTREF(); 2821 nd6_hint(tp); 2822 if (so->so_state & SS_CANTRCVMORE) { 2823 m_freem(m); 2824 } else { 2825 m_adj(m, hdroptlen); 2826 sbappendstream(&(so)->so_rcv, m); 2827 } 2828 TCP_REASS_UNLOCK(tp); 2829 sorwakeup(so); 2830 } else { 2831 m_adj(m, hdroptlen); 2832 tiflags = tcp_reass(tp, th, m, tlen); 2833 tp->t_flags |= TF_ACKNOW; 2834 } 2835 2836 /* 2837 * Note the amount of data that peer has sent into 2838 * our window, in order to estimate the sender's 2839 * buffer size. 2840 */ 2841 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2842 } else { 2843 m_freem(m); 2844 m = NULL; 2845 tiflags &= ~TH_FIN; 2846 } 2847 2848 /* 2849 * If FIN is received ACK the FIN and let the user know 2850 * that the connection is closing. Ignore a FIN received before 2851 * the connection is fully established. 2852 */ 2853 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2854 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2855 socantrcvmore(so); 2856 tp->t_flags |= TF_ACKNOW; 2857 tp->rcv_nxt++; 2858 } 2859 switch (tp->t_state) { 2860 2861 /* 2862 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2863 */ 2864 case TCPS_ESTABLISHED: 2865 tp->t_state = TCPS_CLOSE_WAIT; 2866 break; 2867 2868 /* 2869 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2870 * enter the CLOSING state. 2871 */ 2872 case TCPS_FIN_WAIT_1: 2873 tp->t_state = TCPS_CLOSING; 2874 break; 2875 2876 /* 2877 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2878 * starting the time-wait timer, turning off the other 2879 * standard timers. 2880 */ 2881 case TCPS_FIN_WAIT_2: 2882 tp->t_state = TCPS_TIME_WAIT; 2883 tcp_canceltimers(tp); 2884 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2885 soisdisconnected(so); 2886 break; 2887 2888 /* 2889 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2890 */ 2891 case TCPS_TIME_WAIT: 2892 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2893 break; 2894 } 2895 } 2896 #ifdef TCP_DEBUG 2897 if (so->so_options & SO_DEBUG) 2898 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2899 #endif 2900 2901 /* 2902 * Return any desired output. 2903 */ 2904 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2905 KERNEL_LOCK(1, NULL); 2906 (void)tcp_output(tp); 2907 KERNEL_UNLOCK_ONE(NULL); 2908 } 2909 if (tcp_saveti) 2910 m_freem(tcp_saveti); 2911 2912 if (tp->t_state == TCPS_TIME_WAIT 2913 && (so->so_state & SS_NOFDREF) 2914 && (tp->t_inpcb || af != AF_INET) 2915 && (tp->t_in6pcb || af != AF_INET6) 2916 && ((af == AF_INET ? tcp4_vtw_enable : tcp6_vtw_enable) & 1) != 0 2917 && TAILQ_EMPTY(&tp->segq) 2918 && vtw_add(af, tp)) { 2919 ; 2920 } 2921 return; 2922 2923 badsyn: 2924 /* 2925 * Received a bad SYN. Increment counters and dropwithreset. 2926 */ 2927 TCP_STATINC(TCP_STAT_BADSYN); 2928 tp = NULL; 2929 goto dropwithreset; 2930 2931 dropafterack: 2932 /* 2933 * Generate an ACK dropping incoming segment if it occupies 2934 * sequence space, where the ACK reflects our state. 2935 */ 2936 if (tiflags & TH_RST) 2937 goto drop; 2938 goto dropafterack2; 2939 2940 dropafterack_ratelim: 2941 /* 2942 * We may want to rate-limit ACKs against SYN/RST attack. 2943 */ 2944 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2945 tcp_ackdrop_ppslim) == 0) { 2946 /* XXX stat */ 2947 goto drop; 2948 } 2949 2950 dropafterack2: 2951 m_freem(m); 2952 tp->t_flags |= TF_ACKNOW; 2953 KERNEL_LOCK(1, NULL); 2954 (void)tcp_output(tp); 2955 KERNEL_UNLOCK_ONE(NULL); 2956 if (tcp_saveti) 2957 m_freem(tcp_saveti); 2958 return; 2959 2960 dropwithreset_ratelim: 2961 /* 2962 * We may want to rate-limit RSTs in certain situations, 2963 * particularly if we are sending an RST in response to 2964 * an attempt to connect to or otherwise communicate with 2965 * a port for which we have no socket. 2966 */ 2967 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2968 tcp_rst_ppslim) == 0) { 2969 /* XXX stat */ 2970 goto drop; 2971 } 2972 2973 dropwithreset: 2974 /* 2975 * Generate a RST, dropping incoming segment. 2976 * Make ACK acceptable to originator of segment. 2977 */ 2978 if (tiflags & TH_RST) 2979 goto drop; 2980 if (tiflags & TH_ACK) { 2981 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2982 } else { 2983 if (tiflags & TH_SYN) 2984 tlen++; 2985 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2986 TH_RST|TH_ACK); 2987 } 2988 if (tcp_saveti) 2989 m_freem(tcp_saveti); 2990 return; 2991 2992 badcsum: 2993 drop: 2994 /* 2995 * Drop space held by incoming segment and return. 2996 */ 2997 if (tp) { 2998 if (tp->t_inpcb) 2999 so = tp->t_inpcb->inp_socket; 3000 #ifdef INET6 3001 else if (tp->t_in6pcb) 3002 so = tp->t_in6pcb->in6p_socket; 3003 #endif 3004 else 3005 so = NULL; 3006 #ifdef TCP_DEBUG 3007 if (so && (so->so_options & SO_DEBUG) != 0) 3008 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 3009 #endif 3010 } 3011 if (tcp_saveti) 3012 m_freem(tcp_saveti); 3013 m_freem(m); 3014 return; 3015 } 3016 3017 #ifdef TCP_SIGNATURE 3018 int 3019 tcp_signature_apply(void *fstate, void *data, u_int len) 3020 { 3021 3022 MD5Update(fstate, (u_char *)data, len); 3023 return (0); 3024 } 3025 3026 struct secasvar * 3027 tcp_signature_getsav(struct mbuf *m) 3028 { 3029 struct ip *ip; 3030 struct ip6_hdr *ip6; 3031 3032 ip = mtod(m, struct ip *); 3033 switch (ip->ip_v) { 3034 case 4: 3035 ip = mtod(m, struct ip *); 3036 ip6 = NULL; 3037 break; 3038 case 6: 3039 ip = NULL; 3040 ip6 = mtod(m, struct ip6_hdr *); 3041 break; 3042 default: 3043 return (NULL); 3044 } 3045 3046 #ifdef IPSEC 3047 union sockaddr_union dst; 3048 3049 /* Extract the destination from the IP header in the mbuf. */ 3050 memset(&dst, 0, sizeof(union sockaddr_union)); 3051 if (ip != NULL) { 3052 dst.sa.sa_len = sizeof(struct sockaddr_in); 3053 dst.sa.sa_family = AF_INET; 3054 dst.sin.sin_addr = ip->ip_dst; 3055 } else { 3056 dst.sa.sa_len = sizeof(struct sockaddr_in6); 3057 dst.sa.sa_family = AF_INET6; 3058 dst.sin6.sin6_addr = ip6->ip6_dst; 3059 } 3060 3061 /* 3062 * Look up an SADB entry which matches the address of the peer. 3063 */ 3064 return KEY_LOOKUP_SA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI), 0, 0); 3065 #else 3066 return NULL; 3067 #endif 3068 } 3069 3070 int 3071 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 3072 struct secasvar *sav, char *sig) 3073 { 3074 MD5_CTX ctx; 3075 struct ip *ip; 3076 struct ipovly *ipovly; 3077 #ifdef INET6 3078 struct ip6_hdr *ip6; 3079 struct ip6_hdr_pseudo ip6pseudo; 3080 #endif 3081 struct ippseudo ippseudo; 3082 struct tcphdr th0; 3083 int l, tcphdrlen; 3084 3085 if (sav == NULL) 3086 return (-1); 3087 3088 tcphdrlen = th->th_off * 4; 3089 3090 switch (mtod(m, struct ip *)->ip_v) { 3091 case 4: 3092 MD5Init(&ctx); 3093 ip = mtod(m, struct ip *); 3094 memset(&ippseudo, 0, sizeof(ippseudo)); 3095 ipovly = (struct ipovly *)ip; 3096 ippseudo.ippseudo_src = ipovly->ih_src; 3097 ippseudo.ippseudo_dst = ipovly->ih_dst; 3098 ippseudo.ippseudo_pad = 0; 3099 ippseudo.ippseudo_p = IPPROTO_TCP; 3100 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 3101 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 3102 break; 3103 #if INET6 3104 case 6: 3105 MD5Init(&ctx); 3106 ip6 = mtod(m, struct ip6_hdr *); 3107 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 3108 ip6pseudo.ip6ph_src = ip6->ip6_src; 3109 in6_clearscope(&ip6pseudo.ip6ph_src); 3110 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 3111 in6_clearscope(&ip6pseudo.ip6ph_dst); 3112 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 3113 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 3114 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 3115 break; 3116 #endif 3117 default: 3118 return (-1); 3119 } 3120 3121 th0 = *th; 3122 th0.th_sum = 0; 3123 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 3124 3125 l = m->m_pkthdr.len - thoff - tcphdrlen; 3126 if (l > 0) 3127 m_apply(m, thoff + tcphdrlen, 3128 m->m_pkthdr.len - thoff - tcphdrlen, 3129 tcp_signature_apply, &ctx); 3130 3131 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 3132 MD5Final(sig, &ctx); 3133 3134 return (0); 3135 } 3136 #endif 3137 3138 /* 3139 * Parse and process tcp options. 3140 * 3141 * Returns -1 if this segment should be dropped. (eg. wrong signature) 3142 * Otherwise returns 0. 3143 */ 3144 static int 3145 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, struct tcphdr *th, 3146 struct mbuf *m, int toff, struct tcp_opt_info *oi) 3147 { 3148 u_int16_t mss; 3149 int opt, optlen = 0; 3150 #ifdef TCP_SIGNATURE 3151 void *sigp = NULL; 3152 char sigbuf[TCP_SIGLEN]; 3153 struct secasvar *sav = NULL; 3154 #endif 3155 3156 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 3157 opt = cp[0]; 3158 if (opt == TCPOPT_EOL) 3159 break; 3160 if (opt == TCPOPT_NOP) 3161 optlen = 1; 3162 else { 3163 if (cnt < 2) 3164 break; 3165 optlen = cp[1]; 3166 if (optlen < 2 || optlen > cnt) 3167 break; 3168 } 3169 switch (opt) { 3170 3171 default: 3172 continue; 3173 3174 case TCPOPT_MAXSEG: 3175 if (optlen != TCPOLEN_MAXSEG) 3176 continue; 3177 if (!(th->th_flags & TH_SYN)) 3178 continue; 3179 if (TCPS_HAVERCVDSYN(tp->t_state)) 3180 continue; 3181 memcpy(&mss, cp + 2, sizeof(mss)); 3182 oi->maxseg = ntohs(mss); 3183 break; 3184 3185 case TCPOPT_WINDOW: 3186 if (optlen != TCPOLEN_WINDOW) 3187 continue; 3188 if (!(th->th_flags & TH_SYN)) 3189 continue; 3190 if (TCPS_HAVERCVDSYN(tp->t_state)) 3191 continue; 3192 tp->t_flags |= TF_RCVD_SCALE; 3193 tp->requested_s_scale = cp[2]; 3194 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3195 char buf[INET6_ADDRSTRLEN]; 3196 struct ip *ip = mtod(m, struct ip *); 3197 #ifdef INET6 3198 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 3199 #endif 3200 if (ip) 3201 in_print(buf, sizeof(buf), 3202 &ip->ip_src); 3203 #ifdef INET6 3204 else if (ip6) 3205 in6_print(buf, sizeof(buf), 3206 &ip6->ip6_src); 3207 #endif 3208 else 3209 strlcpy(buf, "(unknown)", sizeof(buf)); 3210 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3211 "assuming %d\n", 3212 tp->requested_s_scale, buf, 3213 TCP_MAX_WINSHIFT); 3214 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3215 } 3216 break; 3217 3218 case TCPOPT_TIMESTAMP: 3219 if (optlen != TCPOLEN_TIMESTAMP) 3220 continue; 3221 oi->ts_present = 1; 3222 memcpy(&oi->ts_val, cp + 2, sizeof(oi->ts_val)); 3223 NTOHL(oi->ts_val); 3224 memcpy(&oi->ts_ecr, cp + 6, sizeof(oi->ts_ecr)); 3225 NTOHL(oi->ts_ecr); 3226 3227 if (!(th->th_flags & TH_SYN)) 3228 continue; 3229 if (TCPS_HAVERCVDSYN(tp->t_state)) 3230 continue; 3231 /* 3232 * A timestamp received in a SYN makes 3233 * it ok to send timestamp requests and replies. 3234 */ 3235 tp->t_flags |= TF_RCVD_TSTMP; 3236 tp->ts_recent = oi->ts_val; 3237 tp->ts_recent_age = tcp_now; 3238 break; 3239 3240 case TCPOPT_SACK_PERMITTED: 3241 if (optlen != TCPOLEN_SACK_PERMITTED) 3242 continue; 3243 if (!(th->th_flags & TH_SYN)) 3244 continue; 3245 if (TCPS_HAVERCVDSYN(tp->t_state)) 3246 continue; 3247 if (tcp_do_sack) { 3248 tp->t_flags |= TF_SACK_PERMIT; 3249 tp->t_flags |= TF_WILL_SACK; 3250 } 3251 break; 3252 3253 case TCPOPT_SACK: 3254 tcp_sack_option(tp, th, cp, optlen); 3255 break; 3256 #ifdef TCP_SIGNATURE 3257 case TCPOPT_SIGNATURE: 3258 if (optlen != TCPOLEN_SIGNATURE) 3259 continue; 3260 if (sigp && memcmp(sigp, cp + 2, TCP_SIGLEN)) 3261 return (-1); 3262 3263 sigp = sigbuf; 3264 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3265 tp->t_flags |= TF_SIGNATURE; 3266 break; 3267 #endif 3268 } 3269 } 3270 3271 #ifndef TCP_SIGNATURE 3272 return 0; 3273 #else 3274 if (tp->t_flags & TF_SIGNATURE) { 3275 sav = tcp_signature_getsav(m); 3276 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3277 return (-1); 3278 } 3279 3280 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) 3281 goto out; 3282 3283 if (sigp) { 3284 char sig[TCP_SIGLEN]; 3285 3286 tcp_fields_to_net(th); 3287 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3288 tcp_fields_to_host(th); 3289 goto out; 3290 } 3291 tcp_fields_to_host(th); 3292 3293 if (memcmp(sig, sigp, TCP_SIGLEN)) { 3294 TCP_STATINC(TCP_STAT_BADSIG); 3295 goto out; 3296 } else 3297 TCP_STATINC(TCP_STAT_GOODSIG); 3298 3299 key_sa_recordxfer(sav, m); 3300 KEY_SA_UNREF(&sav); 3301 } 3302 return 0; 3303 out: 3304 if (sav != NULL) 3305 KEY_SA_UNREF(&sav); 3306 return -1; 3307 #endif 3308 } 3309 3310 /* 3311 * Pull out of band byte out of a segment so 3312 * it doesn't appear in the user's data queue. 3313 * It is still reflected in the segment length for 3314 * sequencing purposes. 3315 */ 3316 void 3317 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3318 struct mbuf *m, int off) 3319 { 3320 int cnt = off + th->th_urp - 1; 3321 3322 while (cnt >= 0) { 3323 if (m->m_len > cnt) { 3324 char *cp = mtod(m, char *) + cnt; 3325 struct tcpcb *tp = sototcpcb(so); 3326 3327 tp->t_iobc = *cp; 3328 tp->t_oobflags |= TCPOOB_HAVEDATA; 3329 memmove(cp, cp + 1, (unsigned)(m->m_len - cnt - 1)); 3330 m->m_len--; 3331 return; 3332 } 3333 cnt -= m->m_len; 3334 m = m->m_next; 3335 if (m == NULL) 3336 break; 3337 } 3338 panic("tcp_pulloutofband"); 3339 } 3340 3341 /* 3342 * Collect new round-trip time estimate 3343 * and update averages and current timeout. 3344 * 3345 * rtt is in units of slow ticks (typically 500 ms) -- essentially the 3346 * difference of two timestamps. 3347 */ 3348 void 3349 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3350 { 3351 int32_t delta; 3352 3353 TCP_STATINC(TCP_STAT_RTTUPDATED); 3354 if (tp->t_srtt != 0) { 3355 /* 3356 * Compute the amount to add to srtt for smoothing, 3357 * *alpha, or 2^(-TCP_RTT_SHIFT). Because 3358 * srtt is stored in 1/32 slow ticks, we conceptually 3359 * shift left 5 bits, subtract srtt to get the 3360 * diference, and then shift right by TCP_RTT_SHIFT 3361 * (3) to obtain 1/8 of the difference. 3362 */ 3363 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3364 /* 3365 * This can never happen, because delta's lowest 3366 * possible value is 1/8 of t_srtt. But if it does, 3367 * set srtt to some reasonable value, here chosen 3368 * as 1/8 tick. 3369 */ 3370 if ((tp->t_srtt += delta) <= 0) 3371 tp->t_srtt = 1 << 2; 3372 /* 3373 * RFC2988 requires that rttvar be updated first. 3374 * This code is compliant because "delta" is the old 3375 * srtt minus the new observation (scaled). 3376 * 3377 * RFC2988 says: 3378 * rttvar = (1-beta) * rttvar + beta * |srtt-observed| 3379 * 3380 * delta is in units of 1/32 ticks, and has then been 3381 * divided by 8. This is equivalent to being in 1/16s 3382 * units and divided by 4. Subtract from it 1/4 of 3383 * the existing rttvar to form the (signed) amount to 3384 * adjust. 3385 */ 3386 if (delta < 0) 3387 delta = -delta; 3388 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3389 /* 3390 * As with srtt, this should never happen. There is 3391 * no support in RFC2988 for this operation. But 1/4s 3392 * as rttvar when faced with something arguably wrong 3393 * is ok. 3394 */ 3395 if ((tp->t_rttvar += delta) <= 0) 3396 tp->t_rttvar = 1 << 2; 3397 3398 /* 3399 * If srtt exceeds .01 second, ensure we use the 'remote' MSL 3400 * Problem is: it doesn't work. Disabled by defaulting 3401 * tcp_rttlocal to 0; see corresponding code in 3402 * tcp_subr that selects local vs remote in a different way. 3403 * 3404 * The static branch prediction hint here should be removed 3405 * when the rtt estimator is fixed and the rtt_enable code 3406 * is turned back on. 3407 */ 3408 if (__predict_false(tcp_rttlocal) && tcp_msl_enable 3409 && tp->t_srtt > tcp_msl_remote_threshold 3410 && tp->t_msl < tcp_msl_remote) { 3411 tp->t_msl = tcp_msl_remote; 3412 } 3413 } else { 3414 /* 3415 * This is the first measurement. Per RFC2988, 2.2, 3416 * set rtt=R and srtt=R/2. 3417 * For srtt, storage representation is 1/32 ticks, 3418 * so shift left by 5. 3419 * For rttvar, storage representation is 1/16 ticks, 3420 * So shift left by 4, but then right by 1 to halve. 3421 */ 3422 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3423 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3424 } 3425 tp->t_rtttime = 0; 3426 tp->t_rxtshift = 0; 3427 3428 /* 3429 * the retransmit should happen at rtt + 4 * rttvar. 3430 * Because of the way we do the smoothing, srtt and rttvar 3431 * will each average +1/2 tick of bias. When we compute 3432 * the retransmit timer, we want 1/2 tick of rounding and 3433 * 1 extra tick because of +-1/2 tick uncertainty in the 3434 * firing of the timer. The bias will give us exactly the 3435 * 1.5 tick we need. But, because the bias is 3436 * statistical, we have to test that we don't drop below 3437 * the minimum feasible timer (which is 2 ticks). 3438 */ 3439 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3440 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3441 3442 /* 3443 * We received an ack for a packet that wasn't retransmitted; 3444 * it is probably safe to discard any error indications we've 3445 * received recently. This isn't quite right, but close enough 3446 * for now (a route might have failed after we sent a segment, 3447 * and the return path might not be symmetrical). 3448 */ 3449 tp->t_softerror = 0; 3450 } 3451 3452 3453 /* 3454 * TCP compressed state engine. Currently used to hold compressed 3455 * state for SYN_RECEIVED. 3456 */ 3457 3458 u_long syn_cache_count; 3459 u_int32_t syn_hash1, syn_hash2; 3460 3461 #define SYN_HASH(sa, sp, dp) \ 3462 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3463 ((u_int32_t)(sp)))^syn_hash2))) 3464 #ifndef INET6 3465 #define SYN_HASHALL(hash, src, dst) \ 3466 do { \ 3467 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3468 ((const struct sockaddr_in *)(src))->sin_port, \ 3469 ((const struct sockaddr_in *)(dst))->sin_port); \ 3470 } while (/*CONSTCOND*/ 0) 3471 #else 3472 #define SYN_HASH6(sa, sp, dp) \ 3473 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3474 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3475 & 0x7fffffff) 3476 3477 #define SYN_HASHALL(hash, src, dst) \ 3478 do { \ 3479 switch ((src)->sa_family) { \ 3480 case AF_INET: \ 3481 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3482 ((const struct sockaddr_in *)(src))->sin_port, \ 3483 ((const struct sockaddr_in *)(dst))->sin_port); \ 3484 break; \ 3485 case AF_INET6: \ 3486 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3487 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3488 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3489 break; \ 3490 default: \ 3491 hash = 0; \ 3492 } \ 3493 } while (/*CONSTCOND*/0) 3494 #endif /* INET6 */ 3495 3496 static struct pool syn_cache_pool; 3497 3498 /* 3499 * We don't estimate RTT with SYNs, so each packet starts with the default 3500 * RTT and each timer step has a fixed timeout value. 3501 */ 3502 static inline void 3503 syn_cache_timer_arm(struct syn_cache *sc) 3504 { 3505 3506 TCPT_RANGESET(sc->sc_rxtcur, 3507 TCPTV_SRTTDFLT * tcp_backoff[sc->sc_rxtshift], TCPTV_MIN, 3508 TCPTV_REXMTMAX); 3509 callout_reset(&sc->sc_timer, 3510 sc->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, sc); 3511 } 3512 3513 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3514 3515 static inline void 3516 syn_cache_rm(struct syn_cache *sc) 3517 { 3518 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3519 sc, sc_bucketq); 3520 sc->sc_tp = NULL; 3521 LIST_REMOVE(sc, sc_tpq); 3522 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3523 callout_stop(&sc->sc_timer); 3524 syn_cache_count--; 3525 } 3526 3527 static inline void 3528 syn_cache_put(struct syn_cache *sc) 3529 { 3530 if (sc->sc_ipopts) 3531 (void) m_free(sc->sc_ipopts); 3532 rtcache_free(&sc->sc_route); 3533 sc->sc_flags |= SCF_DEAD; 3534 if (!callout_invoking(&sc->sc_timer)) 3535 callout_schedule(&(sc)->sc_timer, 1); 3536 } 3537 3538 void 3539 syn_cache_init(void) 3540 { 3541 int i; 3542 3543 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 3544 "synpl", NULL, IPL_SOFTNET); 3545 3546 /* Initialize the hash buckets. */ 3547 for (i = 0; i < tcp_syn_cache_size; i++) 3548 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3549 } 3550 3551 void 3552 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3553 { 3554 struct syn_cache_head *scp; 3555 struct syn_cache *sc2; 3556 int s; 3557 3558 /* 3559 * If there are no entries in the hash table, reinitialize 3560 * the hash secrets. 3561 */ 3562 if (syn_cache_count == 0) { 3563 syn_hash1 = cprng_fast32(); 3564 syn_hash2 = cprng_fast32(); 3565 } 3566 3567 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3568 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3569 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3570 3571 /* 3572 * Make sure that we don't overflow the per-bucket 3573 * limit or the total cache size limit. 3574 */ 3575 s = splsoftnet(); 3576 if (scp->sch_length >= tcp_syn_bucket_limit) { 3577 TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW); 3578 /* 3579 * The bucket is full. Toss the oldest element in the 3580 * bucket. This will be the first entry in the bucket. 3581 */ 3582 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3583 #ifdef DIAGNOSTIC 3584 /* 3585 * This should never happen; we should always find an 3586 * entry in our bucket. 3587 */ 3588 if (sc2 == NULL) 3589 panic("syn_cache_insert: bucketoverflow: impossible"); 3590 #endif 3591 syn_cache_rm(sc2); 3592 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3593 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3594 struct syn_cache_head *scp2, *sce; 3595 3596 TCP_STATINC(TCP_STAT_SC_OVERFLOWED); 3597 /* 3598 * The cache is full. Toss the oldest entry in the 3599 * first non-empty bucket we can find. 3600 * 3601 * XXX We would really like to toss the oldest 3602 * entry in the cache, but we hope that this 3603 * condition doesn't happen very often. 3604 */ 3605 scp2 = scp; 3606 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3607 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3608 for (++scp2; scp2 != scp; scp2++) { 3609 if (scp2 >= sce) 3610 scp2 = &tcp_syn_cache[0]; 3611 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3612 break; 3613 } 3614 #ifdef DIAGNOSTIC 3615 /* 3616 * This should never happen; we should always find a 3617 * non-empty bucket. 3618 */ 3619 if (scp2 == scp) 3620 panic("syn_cache_insert: cacheoverflow: " 3621 "impossible"); 3622 #endif 3623 } 3624 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3625 syn_cache_rm(sc2); 3626 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3627 } 3628 3629 /* 3630 * Initialize the entry's timer. 3631 */ 3632 sc->sc_rxttot = 0; 3633 sc->sc_rxtshift = 0; 3634 syn_cache_timer_arm(sc); 3635 3636 /* Link it from tcpcb entry */ 3637 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3638 3639 /* Put it into the bucket. */ 3640 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3641 scp->sch_length++; 3642 syn_cache_count++; 3643 3644 TCP_STATINC(TCP_STAT_SC_ADDED); 3645 splx(s); 3646 } 3647 3648 /* 3649 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3650 * If we have retransmitted an entry the maximum number of times, expire 3651 * that entry. 3652 */ 3653 static void 3654 syn_cache_timer(void *arg) 3655 { 3656 struct syn_cache *sc = arg; 3657 3658 mutex_enter(softnet_lock); 3659 KERNEL_LOCK(1, NULL); 3660 3661 callout_ack(&sc->sc_timer); 3662 3663 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3664 TCP_STATINC(TCP_STAT_SC_DELAYED_FREE); 3665 goto free; 3666 } 3667 3668 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3669 /* Drop it -- too many retransmissions. */ 3670 goto dropit; 3671 } 3672 3673 /* 3674 * Compute the total amount of time this entry has 3675 * been on a queue. If this entry has been on longer 3676 * than the keep alive timer would allow, expire it. 3677 */ 3678 sc->sc_rxttot += sc->sc_rxtcur; 3679 if (sc->sc_rxttot >= tcp_keepinit) 3680 goto dropit; 3681 3682 TCP_STATINC(TCP_STAT_SC_RETRANSMITTED); 3683 (void)syn_cache_respond(sc); 3684 3685 /* Advance the timer back-off. */ 3686 sc->sc_rxtshift++; 3687 syn_cache_timer_arm(sc); 3688 3689 goto out; 3690 3691 dropit: 3692 TCP_STATINC(TCP_STAT_SC_TIMED_OUT); 3693 syn_cache_rm(sc); 3694 if (sc->sc_ipopts) 3695 (void) m_free(sc->sc_ipopts); 3696 rtcache_free(&sc->sc_route); 3697 3698 free: 3699 callout_destroy(&sc->sc_timer); 3700 pool_put(&syn_cache_pool, sc); 3701 3702 out: 3703 KERNEL_UNLOCK_ONE(NULL); 3704 mutex_exit(softnet_lock); 3705 } 3706 3707 /* 3708 * Remove syn cache created by the specified tcb entry, 3709 * because this does not make sense to keep them 3710 * (if there's no tcb entry, syn cache entry will never be used) 3711 */ 3712 void 3713 syn_cache_cleanup(struct tcpcb *tp) 3714 { 3715 struct syn_cache *sc, *nsc; 3716 int s; 3717 3718 s = splsoftnet(); 3719 3720 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3721 nsc = LIST_NEXT(sc, sc_tpq); 3722 3723 #ifdef DIAGNOSTIC 3724 if (sc->sc_tp != tp) 3725 panic("invalid sc_tp in syn_cache_cleanup"); 3726 #endif 3727 syn_cache_rm(sc); 3728 syn_cache_put(sc); /* calls pool_put but see spl above */ 3729 } 3730 /* just for safety */ 3731 LIST_INIT(&tp->t_sc); 3732 3733 splx(s); 3734 } 3735 3736 /* 3737 * Find an entry in the syn cache. 3738 */ 3739 struct syn_cache * 3740 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3741 struct syn_cache_head **headp) 3742 { 3743 struct syn_cache *sc; 3744 struct syn_cache_head *scp; 3745 u_int32_t hash; 3746 int s; 3747 3748 SYN_HASHALL(hash, src, dst); 3749 3750 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3751 *headp = scp; 3752 s = splsoftnet(); 3753 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3754 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3755 if (sc->sc_hash != hash) 3756 continue; 3757 if (!memcmp(&sc->sc_src, src, src->sa_len) && 3758 !memcmp(&sc->sc_dst, dst, dst->sa_len)) { 3759 splx(s); 3760 return (sc); 3761 } 3762 } 3763 splx(s); 3764 return (NULL); 3765 } 3766 3767 /* 3768 * This function gets called when we receive an ACK for a socket in the 3769 * LISTEN state. We look up the connection in the syn cache, and if it's 3770 * there, we pull it out of the cache and turn it into a full-blown 3771 * connection in the SYN-RECEIVED state. 3772 * 3773 * The return values may not be immediately obvious, and their effects 3774 * can be subtle, so here they are: 3775 * 3776 * NULL SYN was not found in cache; caller should drop the 3777 * packet and send an RST. 3778 * 3779 * -1 We were unable to create the new connection, and are 3780 * aborting it. An ACK,RST is being sent to the peer 3781 * (unless we got screwey sequence numbers; see below), 3782 * because the 3-way handshake has been completed. Caller 3783 * should not free the mbuf, since we may be using it. If 3784 * we are not, we will free it. 3785 * 3786 * Otherwise, the return value is a pointer to the new socket 3787 * associated with the connection. 3788 */ 3789 struct socket * 3790 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3791 struct tcphdr *th, struct socket *so, struct mbuf *m) 3792 { 3793 struct syn_cache *sc; 3794 struct syn_cache_head *scp; 3795 struct inpcb *inp = NULL; 3796 #ifdef INET6 3797 struct in6pcb *in6p = NULL; 3798 #endif 3799 struct tcpcb *tp; 3800 int s; 3801 struct socket *oso; 3802 3803 s = splsoftnet(); 3804 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3805 splx(s); 3806 return NULL; 3807 } 3808 3809 /* 3810 * Verify the sequence and ack numbers. Try getting the correct 3811 * response again. 3812 */ 3813 if ((th->th_ack != sc->sc_iss + 1) || 3814 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3815 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3816 m_freem(m); 3817 (void)syn_cache_respond(sc); 3818 splx(s); 3819 return ((struct socket *)(-1)); 3820 } 3821 3822 /* Remove this cache entry */ 3823 syn_cache_rm(sc); 3824 splx(s); 3825 3826 /* 3827 * Ok, create the full blown connection, and set things up 3828 * as they would have been set up if we had created the 3829 * connection when the SYN arrived. If we can't create 3830 * the connection, abort it. 3831 */ 3832 /* 3833 * inp still has the OLD in_pcb stuff, set the 3834 * v6-related flags on the new guy, too. This is 3835 * done particularly for the case where an AF_INET6 3836 * socket is bound only to a port, and a v4 connection 3837 * comes in on that port. 3838 * we also copy the flowinfo from the original pcb 3839 * to the new one. 3840 */ 3841 oso = so; 3842 so = sonewconn(so, true); 3843 if (so == NULL) 3844 goto resetandabort; 3845 3846 switch (so->so_proto->pr_domain->dom_family) { 3847 case AF_INET: 3848 inp = sotoinpcb(so); 3849 break; 3850 #ifdef INET6 3851 case AF_INET6: 3852 in6p = sotoin6pcb(so); 3853 break; 3854 #endif 3855 } 3856 3857 switch (src->sa_family) { 3858 case AF_INET: 3859 if (inp) { 3860 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3861 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3862 inp->inp_options = ip_srcroute(m); 3863 in_pcbstate(inp, INP_BOUND); 3864 if (inp->inp_options == NULL) { 3865 inp->inp_options = sc->sc_ipopts; 3866 sc->sc_ipopts = NULL; 3867 } 3868 } 3869 #ifdef INET6 3870 else if (in6p) { 3871 /* IPv4 packet to AF_INET6 socket */ 3872 memset(&in6p->in6p_laddr, 0, sizeof(in6p->in6p_laddr)); 3873 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3874 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3875 &in6p->in6p_laddr.s6_addr32[3], 3876 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3877 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3878 in6totcpcb(in6p)->t_family = AF_INET; 3879 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3880 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3881 else 3882 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3883 in6_pcbstate(in6p, IN6P_BOUND); 3884 } 3885 #endif 3886 break; 3887 #ifdef INET6 3888 case AF_INET6: 3889 if (in6p) { 3890 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3891 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3892 in6_pcbstate(in6p, IN6P_BOUND); 3893 } 3894 break; 3895 #endif 3896 } 3897 3898 #ifdef INET6 3899 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3900 struct in6pcb *oin6p = sotoin6pcb(oso); 3901 /* inherit socket options from the listening socket */ 3902 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3903 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3904 m_freem(in6p->in6p_options); 3905 in6p->in6p_options = NULL; 3906 } 3907 ip6_savecontrol(in6p, &in6p->in6p_options, 3908 mtod(m, struct ip6_hdr *), m); 3909 } 3910 #endif 3911 3912 #if defined(IPSEC) 3913 if (ipsec_used) { 3914 /* 3915 * we make a copy of policy, instead of sharing the policy, for 3916 * better behavior in terms of SA lookup and dead SA removal. 3917 */ 3918 if (inp) { 3919 /* copy old policy into new socket's */ 3920 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, 3921 inp->inp_sp)) 3922 printf("tcp_input: could not copy policy\n"); 3923 } 3924 #ifdef INET6 3925 else if (in6p) { 3926 /* copy old policy into new socket's */ 3927 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3928 in6p->in6p_sp)) 3929 printf("tcp_input: could not copy policy\n"); 3930 } 3931 #endif 3932 } 3933 #endif 3934 3935 /* 3936 * Give the new socket our cached route reference. 3937 */ 3938 if (inp) { 3939 rtcache_copy(&inp->inp_route, &sc->sc_route); 3940 rtcache_free(&sc->sc_route); 3941 } 3942 #ifdef INET6 3943 else { 3944 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 3945 rtcache_free(&sc->sc_route); 3946 } 3947 #endif 3948 3949 if (inp) { 3950 struct sockaddr_in sin; 3951 memcpy(&sin, src, src->sa_len); 3952 if (in_pcbconnect(inp, &sin, &lwp0)) { 3953 goto resetandabort; 3954 } 3955 } 3956 #ifdef INET6 3957 else if (in6p) { 3958 struct sockaddr_in6 sin6; 3959 memcpy(&sin6, src, src->sa_len); 3960 if (src->sa_family == AF_INET) { 3961 /* IPv4 packet to AF_INET6 socket */ 3962 in6_sin_2_v4mapsin6((struct sockaddr_in *)src, &sin6); 3963 } 3964 if (in6_pcbconnect(in6p, &sin6, NULL)) { 3965 goto resetandabort; 3966 } 3967 } 3968 #endif 3969 else { 3970 goto resetandabort; 3971 } 3972 3973 if (inp) 3974 tp = intotcpcb(inp); 3975 #ifdef INET6 3976 else if (in6p) 3977 tp = in6totcpcb(in6p); 3978 #endif 3979 else 3980 tp = NULL; 3981 3982 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3983 if (sc->sc_request_r_scale != 15) { 3984 tp->requested_s_scale = sc->sc_requested_s_scale; 3985 tp->request_r_scale = sc->sc_request_r_scale; 3986 tp->snd_scale = sc->sc_requested_s_scale; 3987 tp->rcv_scale = sc->sc_request_r_scale; 3988 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3989 } 3990 if (sc->sc_flags & SCF_TIMESTAMP) 3991 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3992 tp->ts_timebase = sc->sc_timebase; 3993 3994 tp->t_template = tcp_template(tp); 3995 if (tp->t_template == 0) { 3996 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3997 so = NULL; 3998 m_freem(m); 3999 goto abort; 4000 } 4001 4002 tp->iss = sc->sc_iss; 4003 tp->irs = sc->sc_irs; 4004 tcp_sendseqinit(tp); 4005 tcp_rcvseqinit(tp); 4006 tp->t_state = TCPS_SYN_RECEIVED; 4007 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 4008 TCP_STATINC(TCP_STAT_ACCEPTS); 4009 4010 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 4011 tp->t_flags |= TF_WILL_SACK; 4012 4013 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 4014 tp->t_flags |= TF_ECN_PERMIT; 4015 4016 #ifdef TCP_SIGNATURE 4017 if (sc->sc_flags & SCF_SIGNATURE) 4018 tp->t_flags |= TF_SIGNATURE; 4019 #endif 4020 4021 /* Initialize tp->t_ourmss before we deal with the peer's! */ 4022 tp->t_ourmss = sc->sc_ourmaxseg; 4023 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 4024 4025 /* 4026 * Initialize the initial congestion window. If we 4027 * had to retransmit the SYN,ACK, we must initialize cwnd 4028 * to 1 segment (i.e. the Loss Window). 4029 */ 4030 if (sc->sc_rxtshift) 4031 tp->snd_cwnd = tp->t_peermss; 4032 else { 4033 int ss = tcp_init_win; 4034 if (inp != NULL && in_localaddr(inp->inp_faddr)) 4035 ss = tcp_init_win_local; 4036 #ifdef INET6 4037 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 4038 ss = tcp_init_win_local; 4039 #endif 4040 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 4041 } 4042 4043 tcp_rmx_rtt(tp); 4044 tp->snd_wl1 = sc->sc_irs; 4045 tp->rcv_up = sc->sc_irs + 1; 4046 4047 /* 4048 * This is what whould have happened in tcp_output() when 4049 * the SYN,ACK was sent. 4050 */ 4051 tp->snd_up = tp->snd_una; 4052 tp->snd_max = tp->snd_nxt = tp->iss+1; 4053 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 4054 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 4055 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 4056 tp->last_ack_sent = tp->rcv_nxt; 4057 tp->t_partialacks = -1; 4058 tp->t_dupacks = 0; 4059 4060 TCP_STATINC(TCP_STAT_SC_COMPLETED); 4061 s = splsoftnet(); 4062 syn_cache_put(sc); 4063 splx(s); 4064 return so; 4065 4066 resetandabort: 4067 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 4068 abort: 4069 if (so != NULL) { 4070 (void) soqremque(so, 1); 4071 (void) soabort(so); 4072 mutex_enter(softnet_lock); 4073 } 4074 s = splsoftnet(); 4075 syn_cache_put(sc); 4076 splx(s); 4077 TCP_STATINC(TCP_STAT_SC_ABORTED); 4078 return ((struct socket *)(-1)); 4079 } 4080 4081 /* 4082 * This function is called when we get a RST for a 4083 * non-existent connection, so that we can see if the 4084 * connection is in the syn cache. If it is, zap it. 4085 */ 4086 4087 void 4088 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 4089 { 4090 struct syn_cache *sc; 4091 struct syn_cache_head *scp; 4092 int s = splsoftnet(); 4093 4094 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4095 splx(s); 4096 return; 4097 } 4098 if (SEQ_LT(th->th_seq, sc->sc_irs) || 4099 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 4100 splx(s); 4101 return; 4102 } 4103 syn_cache_rm(sc); 4104 TCP_STATINC(TCP_STAT_SC_RESET); 4105 syn_cache_put(sc); /* calls pool_put but see spl above */ 4106 splx(s); 4107 } 4108 4109 void 4110 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 4111 struct tcphdr *th) 4112 { 4113 struct syn_cache *sc; 4114 struct syn_cache_head *scp; 4115 int s; 4116 4117 s = splsoftnet(); 4118 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4119 splx(s); 4120 return; 4121 } 4122 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 4123 if (ntohl(th->th_seq) != sc->sc_iss) { 4124 splx(s); 4125 return; 4126 } 4127 4128 /* 4129 * If we've retransmitted 3 times and this is our second error, 4130 * we remove the entry. Otherwise, we allow it to continue on. 4131 * This prevents us from incorrectly nuking an entry during a 4132 * spurious network outage. 4133 * 4134 * See tcp_notify(). 4135 */ 4136 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 4137 sc->sc_flags |= SCF_UNREACH; 4138 splx(s); 4139 return; 4140 } 4141 4142 syn_cache_rm(sc); 4143 TCP_STATINC(TCP_STAT_SC_UNREACH); 4144 syn_cache_put(sc); /* calls pool_put but see spl above */ 4145 splx(s); 4146 } 4147 4148 /* 4149 * Given a LISTEN socket and an inbound SYN request, add this to the syn 4150 * cache, and send back a segment: 4151 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 4152 * to the source. 4153 * 4154 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 4155 * Doing so would require that we hold onto the data and deliver it 4156 * to the application. However, if we are the target of a SYN-flood 4157 * DoS attack, an attacker could send data which would eventually 4158 * consume all available buffer space if it were ACKed. By not ACKing 4159 * the data, we avoid this DoS scenario. 4160 */ 4161 int 4162 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 4163 unsigned int toff, struct socket *so, struct mbuf *m, u_char *optp, 4164 int optlen, struct tcp_opt_info *oi) 4165 { 4166 struct tcpcb tb, *tp; 4167 long win; 4168 struct syn_cache *sc; 4169 struct syn_cache_head *scp; 4170 struct mbuf *ipopts; 4171 int s; 4172 4173 tp = sototcpcb(so); 4174 4175 /* 4176 * Initialize some local state. 4177 */ 4178 win = sbspace(&so->so_rcv); 4179 if (win > TCP_MAXWIN) 4180 win = TCP_MAXWIN; 4181 4182 #ifdef TCP_SIGNATURE 4183 if (optp || (tp->t_flags & TF_SIGNATURE)) 4184 #else 4185 if (optp) 4186 #endif 4187 { 4188 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4189 #ifdef TCP_SIGNATURE 4190 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4191 #endif 4192 tb.t_state = TCPS_LISTEN; 4193 if (tcp_dooptions(&tb, optp, optlen, th, m, toff, oi) < 0) 4194 return 0; 4195 } else 4196 tb.t_flags = 0; 4197 4198 switch (src->sa_family) { 4199 case AF_INET: 4200 /* Remember the IP options, if any. */ 4201 ipopts = ip_srcroute(m); 4202 break; 4203 default: 4204 ipopts = NULL; 4205 } 4206 4207 /* 4208 * See if we already have an entry for this connection. 4209 * If we do, resend the SYN,ACK. We do not count this 4210 * as a retransmission (XXX though maybe we should). 4211 */ 4212 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4213 TCP_STATINC(TCP_STAT_SC_DUPESYN); 4214 if (ipopts) { 4215 /* 4216 * If we were remembering a previous source route, 4217 * forget it and use the new one we've been given. 4218 */ 4219 if (sc->sc_ipopts) 4220 (void)m_free(sc->sc_ipopts); 4221 sc->sc_ipopts = ipopts; 4222 } 4223 sc->sc_timestamp = tb.ts_recent; 4224 m_freem(m); 4225 if (syn_cache_respond(sc) == 0) { 4226 uint64_t *tcps = TCP_STAT_GETREF(); 4227 tcps[TCP_STAT_SNDACKS]++; 4228 tcps[TCP_STAT_SNDTOTAL]++; 4229 TCP_STAT_PUTREF(); 4230 } 4231 return 1; 4232 } 4233 4234 s = splsoftnet(); 4235 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4236 splx(s); 4237 if (sc == NULL) { 4238 if (ipopts) 4239 (void)m_free(ipopts); 4240 return 0; 4241 } 4242 4243 /* 4244 * Fill in the cache, and put the necessary IP and TCP 4245 * options into the reply. 4246 */ 4247 memset(sc, 0, sizeof(struct syn_cache)); 4248 callout_init(&sc->sc_timer, CALLOUT_MPSAFE); 4249 memcpy(&sc->sc_src, src, src->sa_len); 4250 memcpy(&sc->sc_dst, dst, dst->sa_len); 4251 sc->sc_flags = 0; 4252 sc->sc_ipopts = ipopts; 4253 sc->sc_irs = th->th_seq; 4254 switch (src->sa_family) { 4255 case AF_INET: 4256 { 4257 struct sockaddr_in *srcin = (void *)src; 4258 struct sockaddr_in *dstin = (void *)dst; 4259 4260 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4261 &srcin->sin_addr, dstin->sin_port, 4262 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4263 break; 4264 } 4265 #ifdef INET6 4266 case AF_INET6: 4267 { 4268 struct sockaddr_in6 *srcin6 = (void *)src; 4269 struct sockaddr_in6 *dstin6 = (void *)dst; 4270 4271 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4272 &srcin6->sin6_addr, dstin6->sin6_port, 4273 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4274 break; 4275 } 4276 #endif 4277 } 4278 sc->sc_peermaxseg = oi->maxseg; 4279 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4280 m_get_rcvif_NOMPSAFE(m) : NULL, sc->sc_src.sa.sa_family); 4281 sc->sc_win = win; 4282 sc->sc_timebase = tcp_now - 1; /* see tcp_newtcpcb() */ 4283 sc->sc_timestamp = tb.ts_recent; 4284 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4285 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4286 sc->sc_flags |= SCF_TIMESTAMP; 4287 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4288 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4289 sc->sc_requested_s_scale = tb.requested_s_scale; 4290 sc->sc_request_r_scale = 0; 4291 /* 4292 * Pick the smallest possible scaling factor that 4293 * will still allow us to scale up to sb_max. 4294 * 4295 * We do this because there are broken firewalls that 4296 * will corrupt the window scale option, leading to 4297 * the other endpoint believing that our advertised 4298 * window is unscaled. At scale factors larger than 4299 * 5 the unscaled window will drop below 1500 bytes, 4300 * leading to serious problems when traversing these 4301 * broken firewalls. 4302 * 4303 * With the default sbmax of 256K, a scale factor 4304 * of 3 will be chosen by this algorithm. Those who 4305 * choose a larger sbmax should watch out 4306 * for the compatiblity problems mentioned above. 4307 * 4308 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4309 * or <SYN,ACK>) segment itself is never scaled. 4310 */ 4311 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4312 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4313 sc->sc_request_r_scale++; 4314 } else { 4315 sc->sc_requested_s_scale = 15; 4316 sc->sc_request_r_scale = 15; 4317 } 4318 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4319 sc->sc_flags |= SCF_SACK_PERMIT; 4320 4321 /* 4322 * ECN setup packet received. 4323 */ 4324 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4325 sc->sc_flags |= SCF_ECN_PERMIT; 4326 4327 #ifdef TCP_SIGNATURE 4328 if (tb.t_flags & TF_SIGNATURE) 4329 sc->sc_flags |= SCF_SIGNATURE; 4330 #endif 4331 sc->sc_tp = tp; 4332 m_freem(m); 4333 if (syn_cache_respond(sc) == 0) { 4334 uint64_t *tcps = TCP_STAT_GETREF(); 4335 tcps[TCP_STAT_SNDACKS]++; 4336 tcps[TCP_STAT_SNDTOTAL]++; 4337 TCP_STAT_PUTREF(); 4338 syn_cache_insert(sc, tp); 4339 } else { 4340 s = splsoftnet(); 4341 /* 4342 * syn_cache_put() will try to schedule the timer, so 4343 * we need to initialize it 4344 */ 4345 syn_cache_timer_arm(sc); 4346 syn_cache_put(sc); 4347 splx(s); 4348 TCP_STATINC(TCP_STAT_SC_DROPPED); 4349 } 4350 return 1; 4351 } 4352 4353 /* 4354 * syn_cache_respond: (re)send SYN+ACK. 4355 * 4356 * Returns 0 on success. 4357 */ 4358 4359 int 4360 syn_cache_respond(struct syn_cache *sc) 4361 { 4362 #ifdef INET6 4363 struct rtentry *rt = NULL; 4364 #endif 4365 struct route *ro; 4366 u_int8_t *optp; 4367 int optlen, error; 4368 u_int16_t tlen; 4369 struct ip *ip = NULL; 4370 #ifdef INET6 4371 struct ip6_hdr *ip6 = NULL; 4372 #endif 4373 struct tcpcb *tp; 4374 struct tcphdr *th; 4375 struct mbuf *m; 4376 u_int hlen; 4377 #ifdef TCP_SIGNATURE 4378 struct secasvar *sav = NULL; 4379 u_int8_t *sigp = NULL; 4380 #endif 4381 4382 ro = &sc->sc_route; 4383 switch (sc->sc_src.sa.sa_family) { 4384 case AF_INET: 4385 hlen = sizeof(struct ip); 4386 break; 4387 #ifdef INET6 4388 case AF_INET6: 4389 hlen = sizeof(struct ip6_hdr); 4390 break; 4391 #endif 4392 default: 4393 return EAFNOSUPPORT; 4394 } 4395 4396 /* Worst case scanario, since we don't know the option size yet. */ 4397 tlen = hlen + sizeof(struct tcphdr) + MAX_TCPOPTLEN; 4398 KASSERT(max_linkhdr + tlen <= MCLBYTES); 4399 4400 /* 4401 * Create the IP+TCP header from scratch. 4402 */ 4403 MGETHDR(m, M_DONTWAIT, MT_DATA); 4404 if (m && (max_linkhdr + tlen) > MHLEN) { 4405 MCLGET(m, M_DONTWAIT); 4406 if ((m->m_flags & M_EXT) == 0) { 4407 m_freem(m); 4408 m = NULL; 4409 } 4410 } 4411 if (m == NULL) 4412 return ENOBUFS; 4413 MCLAIM(m, &tcp_tx_mowner); 4414 4415 tp = sc->sc_tp; 4416 4417 /* Fixup the mbuf. */ 4418 m->m_data += max_linkhdr; 4419 m_reset_rcvif(m); 4420 memset(mtod(m, void *), 0, tlen); 4421 4422 switch (sc->sc_src.sa.sa_family) { 4423 case AF_INET: 4424 ip = mtod(m, struct ip *); 4425 ip->ip_v = 4; 4426 ip->ip_dst = sc->sc_src.sin.sin_addr; 4427 ip->ip_src = sc->sc_dst.sin.sin_addr; 4428 ip->ip_p = IPPROTO_TCP; 4429 th = (struct tcphdr *)(ip + 1); 4430 th->th_dport = sc->sc_src.sin.sin_port; 4431 th->th_sport = sc->sc_dst.sin.sin_port; 4432 break; 4433 #ifdef INET6 4434 case AF_INET6: 4435 ip6 = mtod(m, struct ip6_hdr *); 4436 ip6->ip6_vfc = IPV6_VERSION; 4437 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4438 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4439 ip6->ip6_nxt = IPPROTO_TCP; 4440 /* ip6_plen will be updated in ip6_output() */ 4441 th = (struct tcphdr *)(ip6 + 1); 4442 th->th_dport = sc->sc_src.sin6.sin6_port; 4443 th->th_sport = sc->sc_dst.sin6.sin6_port; 4444 break; 4445 #endif 4446 default: 4447 panic("%s: impossible (1)", __func__); 4448 } 4449 4450 th->th_seq = htonl(sc->sc_iss); 4451 th->th_ack = htonl(sc->sc_irs + 1); 4452 th->th_flags = TH_SYN|TH_ACK; 4453 th->th_win = htons(sc->sc_win); 4454 /* th_x2, th_sum, th_urp already 0 from memset */ 4455 4456 /* Tack on the TCP options. */ 4457 optp = (u_int8_t *)(th + 1); 4458 optlen = 0; 4459 *optp++ = TCPOPT_MAXSEG; 4460 *optp++ = TCPOLEN_MAXSEG; 4461 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4462 *optp++ = sc->sc_ourmaxseg & 0xff; 4463 optlen += TCPOLEN_MAXSEG; 4464 4465 if (sc->sc_request_r_scale != 15) { 4466 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4467 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4468 sc->sc_request_r_scale); 4469 optp += TCPOLEN_WINDOW + TCPOLEN_NOP; 4470 optlen += TCPOLEN_WINDOW + TCPOLEN_NOP; 4471 } 4472 4473 if (sc->sc_flags & SCF_SACK_PERMIT) { 4474 /* Let the peer know that we will SACK. */ 4475 *optp++ = TCPOPT_SACK_PERMITTED; 4476 *optp++ = TCPOLEN_SACK_PERMITTED; 4477 optlen += TCPOLEN_SACK_PERMITTED; 4478 } 4479 4480 if (sc->sc_flags & SCF_TIMESTAMP) { 4481 while (optlen % 4 != 2) { 4482 optlen += TCPOLEN_NOP; 4483 *optp++ = TCPOPT_NOP; 4484 } 4485 *optp++ = TCPOPT_TIMESTAMP; 4486 *optp++ = TCPOLEN_TIMESTAMP; 4487 u_int32_t *lp = (u_int32_t *)(optp); 4488 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4489 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4490 *lp = htonl(sc->sc_timestamp); 4491 optp += TCPOLEN_TIMESTAMP - 2; 4492 optlen += TCPOLEN_TIMESTAMP; 4493 } 4494 4495 #ifdef TCP_SIGNATURE 4496 if (sc->sc_flags & SCF_SIGNATURE) { 4497 sav = tcp_signature_getsav(m); 4498 if (sav == NULL) { 4499 m_freem(m); 4500 return EPERM; 4501 } 4502 4503 *optp++ = TCPOPT_SIGNATURE; 4504 *optp++ = TCPOLEN_SIGNATURE; 4505 sigp = optp; 4506 memset(optp, 0, TCP_SIGLEN); 4507 optp += TCP_SIGLEN; 4508 optlen += TCPOLEN_SIGNATURE; 4509 } 4510 #endif 4511 4512 /* 4513 * Terminate and pad TCP options to a 4 byte boundary. 4514 * 4515 * According to RFC793: "The content of the header beyond the 4516 * End-of-Option option must be header padding (i.e., zero)." 4517 * And later: "The padding is composed of zeros." 4518 */ 4519 if (optlen % 4) { 4520 optlen += TCPOLEN_EOL; 4521 *optp++ = TCPOPT_EOL; 4522 } 4523 while (optlen % 4) { 4524 optlen += TCPOLEN_PAD; 4525 *optp++ = TCPOPT_PAD; 4526 } 4527 4528 /* Compute the actual values now that we've added the options. */ 4529 tlen = hlen + sizeof(struct tcphdr) + optlen; 4530 m->m_len = m->m_pkthdr.len = tlen; 4531 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4532 4533 #ifdef TCP_SIGNATURE 4534 if (sav) { 4535 (void)tcp_signature(m, th, hlen, sav, sigp); 4536 key_sa_recordxfer(sav, m); 4537 KEY_SA_UNREF(&sav); 4538 } 4539 #endif 4540 4541 /* 4542 * Send ECN SYN-ACK setup packet. 4543 * Routes can be asymetric, so, even if we receive a packet 4544 * with ECE and CWR set, we must not assume no one will block 4545 * the ECE packet we are about to send. 4546 */ 4547 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4548 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4549 th->th_flags |= TH_ECE; 4550 TCP_STATINC(TCP_STAT_ECN_SHS); 4551 4552 /* 4553 * draft-ietf-tcpm-ecnsyn-00.txt 4554 * 4555 * "[...] a TCP node MAY respond to an ECN-setup 4556 * SYN packet by setting ECT in the responding 4557 * ECN-setup SYN/ACK packet, indicating to routers 4558 * that the SYN/ACK packet is ECN-Capable. 4559 * This allows a congested router along the path 4560 * to mark the packet instead of dropping the 4561 * packet as an indication of congestion." 4562 * 4563 * "[...] There can be a great benefit in setting 4564 * an ECN-capable codepoint in SYN/ACK packets [...] 4565 * Congestion is most likely to occur in 4566 * the server-to-client direction. As a result, 4567 * setting an ECN-capable codepoint in SYN/ACK 4568 * packets can reduce the occurence of three-second 4569 * retransmit timeouts resulting from the drop 4570 * of SYN/ACK packets." 4571 * 4572 * Page 4 and 6, January 2006. 4573 */ 4574 4575 switch (sc->sc_src.sa.sa_family) { 4576 case AF_INET: 4577 ip->ip_tos |= IPTOS_ECN_ECT0; 4578 break; 4579 #ifdef INET6 4580 case AF_INET6: 4581 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4582 break; 4583 #endif 4584 } 4585 TCP_STATINC(TCP_STAT_ECN_ECT); 4586 } 4587 4588 4589 /* 4590 * Compute the packet's checksum. 4591 * 4592 * Fill in some straggling IP bits. Note the stack expects 4593 * ip_len to be in host order, for convenience. 4594 */ 4595 switch (sc->sc_src.sa.sa_family) { 4596 case AF_INET: 4597 ip->ip_len = htons(tlen - hlen); 4598 th->th_sum = 0; 4599 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4600 ip->ip_len = htons(tlen); 4601 ip->ip_ttl = ip_defttl; 4602 /* XXX tos? */ 4603 break; 4604 #ifdef INET6 4605 case AF_INET6: 4606 ip6->ip6_plen = htons(tlen - hlen); 4607 th->th_sum = 0; 4608 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4609 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4610 ip6->ip6_vfc |= IPV6_VERSION; 4611 ip6->ip6_plen = htons(tlen - hlen); 4612 /* ip6_hlim will be initialized afterwards */ 4613 /* XXX flowlabel? */ 4614 break; 4615 #endif 4616 } 4617 4618 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4619 tp = sc->sc_tp; 4620 4621 switch (sc->sc_src.sa.sa_family) { 4622 case AF_INET: 4623 error = ip_output(m, sc->sc_ipopts, ro, 4624 (ip_mtudisc ? IP_MTUDISC : 0), 4625 NULL, tp ? tp->t_inpcb : NULL); 4626 break; 4627 #ifdef INET6 4628 case AF_INET6: 4629 ip6->ip6_hlim = in6_selecthlim(NULL, 4630 (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp : NULL); 4631 rtcache_unref(rt, ro); 4632 4633 error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, 4634 tp ? tp->t_in6pcb : NULL, NULL); 4635 break; 4636 #endif 4637 default: 4638 panic("%s: impossible (2)", __func__); 4639 } 4640 4641 return error; 4642 } 4643
Indexes created Mon Sep 22 13:09:51 GMT 2025