tcp_input.c revision 1.393
1 /* $NetBSD: tcp_input.c,v 1.393 2018/03/28 14:43:55 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 1.393 2018/03/28 14:43:55 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 else 1146 goto drop; 1147 } 1148 1149 /* 1150 * In TIME_WAIT state the only thing that should arrive 1151 * is a retransmission of the remote FIN. Acknowledge 1152 * it and restart the finack timer. 1153 */ 1154 vtw_restart(vp); 1155 goto dropafterack; 1156 1157 dropafterack: 1158 /* 1159 * Generate an ACK dropping incoming segment if it occupies 1160 * sequence space, where the ACK reflects our state. 1161 */ 1162 if (tiflags & TH_RST) 1163 goto drop; 1164 goto dropafterack2; 1165 1166 dropafterack_ratelim: 1167 /* 1168 * We may want to rate-limit ACKs against SYN/RST attack. 1169 */ 1170 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 1171 tcp_ackdrop_ppslim) == 0) { 1172 /* XXX stat */ 1173 goto drop; 1174 } 1175 /* ...fall into dropafterack2... */ 1176 1177 dropafterack2: 1178 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, th->th_ack, TH_ACK); 1179 return; 1180 1181 dropwithreset: 1182 /* 1183 * Generate a RST, dropping incoming segment. 1184 * Make ACK acceptable to originator of segment. 1185 */ 1186 if (tiflags & TH_RST) 1187 goto drop; 1188 1189 if (tiflags & TH_ACK) { 1190 tcp_respond(0, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 1191 } else { 1192 if (tiflags & TH_SYN) 1193 ++tlen; 1194 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, (tcp_seq)0, 1195 TH_RST|TH_ACK); 1196 } 1197 return; 1198 drop: 1199 m_freem(m); 1200 } 1201 1202 /* 1203 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 1204 */ 1205 void 1206 tcp_input(struct mbuf *m, ...) 1207 { 1208 struct tcphdr *th; 1209 struct ip *ip; 1210 struct inpcb *inp; 1211 #ifdef INET6 1212 struct ip6_hdr *ip6; 1213 struct in6pcb *in6p; 1214 #endif 1215 u_int8_t *optp = NULL; 1216 int optlen = 0; 1217 int len, tlen, toff, hdroptlen = 0; 1218 struct tcpcb *tp = NULL; 1219 int tiflags; 1220 struct socket *so = NULL; 1221 int todrop, acked, ourfinisacked, needoutput = 0; 1222 bool dupseg; 1223 #ifdef TCP_DEBUG 1224 short ostate = 0; 1225 #endif 1226 u_long tiwin; 1227 struct tcp_opt_info opti; 1228 int off, iphlen; 1229 va_list ap; 1230 int af; /* af on the wire */ 1231 struct mbuf *tcp_saveti = NULL; 1232 uint32_t ts_rtt; 1233 uint8_t iptos; 1234 uint64_t *tcps; 1235 vestigial_inpcb_t vestige; 1236 1237 vestige.valid = 0; 1238 1239 MCLAIM(m, &tcp_rx_mowner); 1240 va_start(ap, m); 1241 toff = va_arg(ap, int); 1242 (void)va_arg(ap, int); /* ignore value, advance ap */ 1243 va_end(ap); 1244 1245 TCP_STATINC(TCP_STAT_RCVTOTAL); 1246 1247 memset(&opti, 0, sizeof(opti)); 1248 opti.ts_present = 0; 1249 opti.maxseg = 0; 1250 1251 /* 1252 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 1253 * 1254 * TCP is, by definition, unicast, so we reject all 1255 * multicast outright. 1256 * 1257 * Note, there are additional src/dst address checks in 1258 * the AF-specific code below. 1259 */ 1260 if (m->m_flags & (M_BCAST|M_MCAST)) { 1261 /* XXX stat */ 1262 goto drop; 1263 } 1264 #ifdef INET6 1265 if (m->m_flags & M_ANYCAST6) { 1266 /* XXX stat */ 1267 goto drop; 1268 } 1269 #endif 1270 1271 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, sizeof(struct tcphdr)); 1272 if (th == NULL) { 1273 TCP_STATINC(TCP_STAT_RCVSHORT); 1274 return; 1275 } 1276 1277 /* 1278 * Get IP and TCP header. 1279 * Note: IP leaves IP header in first mbuf. 1280 */ 1281 ip = mtod(m, struct ip *); 1282 switch (ip->ip_v) { 1283 case 4: 1284 #ifdef INET6 1285 ip6 = NULL; 1286 #endif 1287 af = AF_INET; 1288 iphlen = sizeof(struct ip); 1289 1290 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1291 in_broadcast(ip->ip_dst, m_get_rcvif_NOMPSAFE(m))) 1292 goto drop; 1293 1294 /* We do the checksum after PCB lookup... */ 1295 len = ntohs(ip->ip_len); 1296 tlen = len - toff; 1297 iptos = ip->ip_tos; 1298 break; 1299 #ifdef INET6 1300 case 6: 1301 ip = NULL; 1302 iphlen = sizeof(struct ip6_hdr); 1303 af = AF_INET6; 1304 ip6 = mtod(m, struct ip6_hdr *); 1305 1306 /* 1307 * Be proactive about unspecified IPv6 address in source. 1308 * As we use all-zero to indicate unbounded/unconnected pcb, 1309 * unspecified IPv6 address can be used to confuse us. 1310 * 1311 * Note that packets with unspecified IPv6 destination is 1312 * already dropped in ip6_input. 1313 */ 1314 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1315 /* XXX stat */ 1316 goto drop; 1317 } 1318 1319 /* 1320 * Make sure destination address is not multicast. 1321 * Source address checked in ip6_input(). 1322 */ 1323 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1324 /* XXX stat */ 1325 goto drop; 1326 } 1327 1328 /* We do the checksum after PCB lookup... */ 1329 len = m->m_pkthdr.len; 1330 tlen = len - toff; 1331 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1332 break; 1333 #endif 1334 default: 1335 m_freem(m); 1336 return; 1337 } 1338 1339 /* 1340 * Enforce alignment requirements that are violated in 1341 * some cases, see kern/50766 for details. 1342 */ 1343 if (TCP_HDR_ALIGNED_P(th) == 0) { 1344 m = m_copyup(m, toff + sizeof(struct tcphdr), 0); 1345 if (m == NULL) { 1346 TCP_STATINC(TCP_STAT_RCVSHORT); 1347 return; 1348 } 1349 ip = mtod(m, struct ip *); 1350 #ifdef INET6 1351 ip6 = mtod(m, struct ip6_hdr *); 1352 #endif 1353 th = (struct tcphdr *)(mtod(m, char *) + toff); 1354 } 1355 KASSERT(TCP_HDR_ALIGNED_P(th)); 1356 1357 /* 1358 * Check that TCP offset makes sense, pull out TCP options and 1359 * adjust length. 1360 */ 1361 off = th->th_off << 2; 1362 if (off < sizeof(struct tcphdr) || off > tlen) { 1363 TCP_STATINC(TCP_STAT_RCVBADOFF); 1364 goto drop; 1365 } 1366 tlen -= off; 1367 1368 if (off > sizeof(struct tcphdr)) { 1369 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1370 if (th == NULL) { 1371 TCP_STATINC(TCP_STAT_RCVSHORT); 1372 return; 1373 } 1374 KASSERT(TCP_HDR_ALIGNED_P(th)); 1375 optlen = off - sizeof(struct tcphdr); 1376 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1377 1378 /* 1379 * Do quick retrieval of timestamp options. 1380 * 1381 * If timestamp is the only option and it's formatted as 1382 * recommended in RFC 1323 appendix A, we quickly get the 1383 * values now and don't bother calling tcp_dooptions(), 1384 * etc. 1385 */ 1386 if ((optlen == TCPOLEN_TSTAMP_APPA || 1387 (optlen > TCPOLEN_TSTAMP_APPA && 1388 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1389 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1390 (th->th_flags & TH_SYN) == 0) { 1391 opti.ts_present = 1; 1392 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1393 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1394 optp = NULL; /* we've parsed the options */ 1395 } 1396 } 1397 tiflags = th->th_flags; 1398 1399 /* 1400 * Checksum extended TCP header and data 1401 */ 1402 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1403 goto badcsum; 1404 1405 /* 1406 * Locate pcb for segment. 1407 */ 1408 findpcb: 1409 inp = NULL; 1410 #ifdef INET6 1411 in6p = NULL; 1412 #endif 1413 switch (af) { 1414 case AF_INET: 1415 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1416 ip->ip_dst, th->th_dport, &vestige); 1417 if (inp == NULL && !vestige.valid) { 1418 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1419 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, 1420 th->th_dport); 1421 } 1422 #ifdef INET6 1423 if (inp == NULL && !vestige.valid) { 1424 struct in6_addr s, d; 1425 1426 /* mapped addr case */ 1427 in6_in_2_v4mapin6(&ip->ip_src, &s); 1428 in6_in_2_v4mapin6(&ip->ip_dst, &d); 1429 in6p = in6_pcblookup_connect(&tcbtable, &s, 1430 th->th_sport, &d, th->th_dport, 0, &vestige); 1431 if (in6p == 0 && !vestige.valid) { 1432 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1433 in6p = in6_pcblookup_bind(&tcbtable, &d, 1434 th->th_dport, 0); 1435 } 1436 } 1437 #endif 1438 #ifndef INET6 1439 if (inp == NULL && !vestige.valid) 1440 #else 1441 if (inp == NULL && in6p == NULL && !vestige.valid) 1442 #endif 1443 { 1444 TCP_STATINC(TCP_STAT_NOPORT); 1445 if (tcp_log_refused && 1446 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1447 tcp4_log_refused(ip, th); 1448 } 1449 tcp_fields_to_host(th); 1450 goto dropwithreset_ratelim; 1451 } 1452 #if defined(IPSEC) 1453 if (ipsec_used) { 1454 if (inp && ipsec_in_reject(m, inp)) { 1455 goto drop; 1456 } 1457 #ifdef INET6 1458 else if (in6p && ipsec_in_reject(m, in6p)) { 1459 goto drop; 1460 } 1461 #endif 1462 } 1463 #endif /*IPSEC*/ 1464 break; 1465 #ifdef INET6 1466 case AF_INET6: 1467 { 1468 int faith; 1469 1470 #if defined(NFAITH) && NFAITH > 0 1471 faith = faithprefix(&ip6->ip6_dst); 1472 #else 1473 faith = 0; 1474 #endif 1475 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1476 th->th_sport, &ip6->ip6_dst, th->th_dport, faith, &vestige); 1477 if (!in6p && !vestige.valid) { 1478 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1479 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1480 th->th_dport, faith); 1481 } 1482 if (!in6p && !vestige.valid) { 1483 TCP_STATINC(TCP_STAT_NOPORT); 1484 if (tcp_log_refused && 1485 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1486 tcp6_log_refused(ip6, th); 1487 } 1488 tcp_fields_to_host(th); 1489 goto dropwithreset_ratelim; 1490 } 1491 #if defined(IPSEC) 1492 if (ipsec_used && in6p && ipsec_in_reject(m, in6p)) { 1493 goto drop; 1494 } 1495 #endif 1496 break; 1497 } 1498 #endif 1499 } 1500 1501 tcp_fields_to_host(th); 1502 1503 /* 1504 * If the state is CLOSED (i.e., TCB does not exist) then 1505 * all data in the incoming segment is discarded. 1506 * If the TCB exists but is in CLOSED state, it is embryonic, 1507 * but should either do a listen or a connect soon. 1508 */ 1509 tp = NULL; 1510 so = NULL; 1511 if (inp) { 1512 /* Check the minimum TTL for socket. */ 1513 if (ip->ip_ttl < inp->inp_ip_minttl) 1514 goto drop; 1515 1516 tp = intotcpcb(inp); 1517 so = inp->inp_socket; 1518 } 1519 #ifdef INET6 1520 else if (in6p) { 1521 tp = in6totcpcb(in6p); 1522 so = in6p->in6p_socket; 1523 } 1524 #endif 1525 else if (vestige.valid) { 1526 /* We do not support the resurrection of vtw tcpcps. */ 1527 tcp_vtw_input(th, &vestige, m, tlen); 1528 m = NULL; 1529 goto drop; 1530 } 1531 1532 if (tp == NULL) 1533 goto dropwithreset_ratelim; 1534 if (tp->t_state == TCPS_CLOSED) 1535 goto drop; 1536 1537 KASSERT(so->so_lock == softnet_lock); 1538 KASSERT(solocked(so)); 1539 1540 /* Unscale the window into a 32-bit value. */ 1541 if ((tiflags & TH_SYN) == 0) 1542 tiwin = th->th_win << tp->snd_scale; 1543 else 1544 tiwin = th->th_win; 1545 1546 #ifdef INET6 1547 /* save packet options if user wanted */ 1548 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1549 if (in6p->in6p_options) { 1550 m_freem(in6p->in6p_options); 1551 in6p->in6p_options = NULL; 1552 } 1553 KASSERT(ip6 != NULL); 1554 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1555 } 1556 #endif 1557 1558 if (so->so_options & SO_DEBUG) { 1559 #ifdef TCP_DEBUG 1560 ostate = tp->t_state; 1561 #endif 1562 1563 tcp_saveti = NULL; 1564 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1565 goto nosave; 1566 1567 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1568 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1569 if (tcp_saveti == NULL) 1570 goto nosave; 1571 } else { 1572 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1573 if (tcp_saveti == NULL) 1574 goto nosave; 1575 MCLAIM(m, &tcp_mowner); 1576 tcp_saveti->m_len = iphlen; 1577 m_copydata(m, 0, iphlen, 1578 mtod(tcp_saveti, void *)); 1579 } 1580 1581 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1582 m_freem(tcp_saveti); 1583 tcp_saveti = NULL; 1584 } else { 1585 tcp_saveti->m_len += sizeof(struct tcphdr); 1586 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1587 sizeof(struct tcphdr)); 1588 } 1589 nosave:; 1590 } 1591 1592 if (so->so_options & SO_ACCEPTCONN) { 1593 union syn_cache_sa src; 1594 union syn_cache_sa dst; 1595 1596 KASSERT(tp->t_state == TCPS_LISTEN); 1597 1598 memset(&src, 0, sizeof(src)); 1599 memset(&dst, 0, sizeof(dst)); 1600 switch (af) { 1601 case AF_INET: 1602 src.sin.sin_len = sizeof(struct sockaddr_in); 1603 src.sin.sin_family = AF_INET; 1604 src.sin.sin_addr = ip->ip_src; 1605 src.sin.sin_port = th->th_sport; 1606 1607 dst.sin.sin_len = sizeof(struct sockaddr_in); 1608 dst.sin.sin_family = AF_INET; 1609 dst.sin.sin_addr = ip->ip_dst; 1610 dst.sin.sin_port = th->th_dport; 1611 break; 1612 #ifdef INET6 1613 case AF_INET6: 1614 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1615 src.sin6.sin6_family = AF_INET6; 1616 src.sin6.sin6_addr = ip6->ip6_src; 1617 src.sin6.sin6_port = th->th_sport; 1618 1619 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1620 dst.sin6.sin6_family = AF_INET6; 1621 dst.sin6.sin6_addr = ip6->ip6_dst; 1622 dst.sin6.sin6_port = th->th_dport; 1623 break; 1624 #endif 1625 } 1626 1627 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1628 if (tiflags & TH_RST) { 1629 syn_cache_reset(&src.sa, &dst.sa, th); 1630 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1631 (TH_ACK|TH_SYN)) { 1632 /* 1633 * Received a SYN,ACK. This should never 1634 * happen while we are in LISTEN. Send an RST. 1635 */ 1636 goto badsyn; 1637 } else if (tiflags & TH_ACK) { 1638 so = syn_cache_get(&src.sa, &dst.sa, th, so, m); 1639 if (so == NULL) { 1640 /* 1641 * We don't have a SYN for this ACK; 1642 * send an RST. 1643 */ 1644 goto badsyn; 1645 } else if (so == (struct socket *)(-1)) { 1646 /* 1647 * We were unable to create the 1648 * connection. If the 3-way handshake 1649 * was completed, and RST has been 1650 * sent to the peer. Since the mbuf 1651 * might be in use for the reply, do 1652 * not free it. 1653 */ 1654 m = NULL; 1655 } else { 1656 /* 1657 * We have created a full-blown 1658 * connection. 1659 */ 1660 tp = NULL; 1661 inp = NULL; 1662 #ifdef INET6 1663 in6p = NULL; 1664 #endif 1665 switch (so->so_proto->pr_domain->dom_family) { 1666 case AF_INET: 1667 inp = sotoinpcb(so); 1668 tp = intotcpcb(inp); 1669 break; 1670 #ifdef INET6 1671 case AF_INET6: 1672 in6p = sotoin6pcb(so); 1673 tp = in6totcpcb(in6p); 1674 break; 1675 #endif 1676 } 1677 if (tp == NULL) 1678 goto badsyn; /*XXX*/ 1679 tiwin <<= tp->snd_scale; 1680 goto after_listen; 1681 } 1682 } else { 1683 /* 1684 * None of RST, SYN or ACK was set. 1685 * This is an invalid packet for a 1686 * TCB in LISTEN state. Send a RST. 1687 */ 1688 goto badsyn; 1689 } 1690 } else { 1691 /* 1692 * Received a SYN. 1693 */ 1694 1695 #ifdef INET6 1696 /* 1697 * If deprecated address is forbidden, we do 1698 * not accept SYN to deprecated interface 1699 * address to prevent any new inbound 1700 * connection from getting established. 1701 * When we do not accept SYN, we send a TCP 1702 * RST, with deprecated source address (instead 1703 * of dropping it). We compromise it as it is 1704 * much better for peer to send a RST, and 1705 * RST will be the final packet for the 1706 * exchange. 1707 * 1708 * If we do not forbid deprecated addresses, we 1709 * accept the SYN packet. RFC2462 does not 1710 * suggest dropping SYN in this case. 1711 * If we decipher RFC2462 5.5.4, it says like 1712 * this: 1713 * 1. use of deprecated addr with existing 1714 * communication is okay - "SHOULD continue 1715 * to be used" 1716 * 2. use of it with new communication: 1717 * (2a) "SHOULD NOT be used if alternate 1718 * address with sufficient scope is 1719 * available" 1720 * (2b) nothing mentioned otherwise. 1721 * Here we fall into (2b) case as we have no 1722 * choice in our source address selection - we 1723 * must obey the peer. 1724 * 1725 * The wording in RFC2462 is confusing, and 1726 * there are multiple description text for 1727 * deprecated address handling - worse, they 1728 * are not exactly the same. I believe 5.5.4 1729 * is the best one, so we follow 5.5.4. 1730 */ 1731 if (af == AF_INET6 && !ip6_use_deprecated) { 1732 struct in6_ifaddr *ia6; 1733 int s; 1734 struct ifnet *rcvif = m_get_rcvif(m, &s); 1735 if (rcvif == NULL) 1736 goto dropwithreset; /* XXX */ 1737 if ((ia6 = in6ifa_ifpwithaddr(rcvif, 1738 &ip6->ip6_dst)) && 1739 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1740 tp = NULL; 1741 m_put_rcvif(rcvif, &s); 1742 goto dropwithreset; 1743 } 1744 m_put_rcvif(rcvif, &s); 1745 } 1746 #endif 1747 1748 /* 1749 * LISTEN socket received a SYN from itself? This 1750 * can't possibly be valid; drop the packet. 1751 */ 1752 if (th->th_sport == th->th_dport) { 1753 int eq = 0; 1754 1755 switch (af) { 1756 case AF_INET: 1757 eq = in_hosteq(ip->ip_src, ip->ip_dst); 1758 break; 1759 #ifdef INET6 1760 case AF_INET6: 1761 eq = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, 1762 &ip6->ip6_dst); 1763 break; 1764 #endif 1765 } 1766 if (eq) { 1767 TCP_STATINC(TCP_STAT_BADSYN); 1768 goto drop; 1769 } 1770 } 1771 1772 /* 1773 * SYN looks ok; create compressed TCP 1774 * state for it. 1775 */ 1776 if (so->so_qlen <= so->so_qlimit && 1777 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1778 so, m, optp, optlen, &opti)) 1779 m = NULL; 1780 } 1781 1782 goto drop; 1783 } 1784 1785 after_listen: 1786 /* 1787 * From here on, we're dealing with !LISTEN. 1788 */ 1789 KASSERT(tp->t_state != TCPS_LISTEN); 1790 1791 /* 1792 * Segment received on connection. 1793 * Reset idle time and keep-alive timer. 1794 */ 1795 tp->t_rcvtime = tcp_now; 1796 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1797 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1798 1799 /* 1800 * Process options. 1801 */ 1802 #ifdef TCP_SIGNATURE 1803 if (optp || (tp->t_flags & TF_SIGNATURE)) 1804 #else 1805 if (optp) 1806 #endif 1807 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1808 goto drop; 1809 1810 if (TCP_SACK_ENABLED(tp)) { 1811 tcp_del_sackholes(tp, th); 1812 } 1813 1814 if (TCP_ECN_ALLOWED(tp)) { 1815 if (tiflags & TH_CWR) { 1816 tp->t_flags &= ~TF_ECN_SND_ECE; 1817 } 1818 switch (iptos & IPTOS_ECN_MASK) { 1819 case IPTOS_ECN_CE: 1820 tp->t_flags |= TF_ECN_SND_ECE; 1821 TCP_STATINC(TCP_STAT_ECN_CE); 1822 break; 1823 case IPTOS_ECN_ECT0: 1824 TCP_STATINC(TCP_STAT_ECN_ECT); 1825 break; 1826 case IPTOS_ECN_ECT1: 1827 /* XXX: ignore for now -- rpaulo */ 1828 break; 1829 } 1830 /* 1831 * Congestion experienced. 1832 * Ignore if we are already trying to recover. 1833 */ 1834 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1835 tp->t_congctl->cong_exp(tp); 1836 } 1837 1838 if (opti.ts_present && opti.ts_ecr) { 1839 /* 1840 * Calculate the RTT from the returned time stamp and the 1841 * connection's time base. If the time stamp is later than 1842 * the current time, or is extremely old, fall back to non-1323 1843 * RTT calculation. Since ts_rtt is unsigned, we can test both 1844 * at the same time. 1845 * 1846 * Note that ts_rtt is in units of slow ticks (500 1847 * ms). Since most earthbound RTTs are < 500 ms, 1848 * observed values will have large quantization noise. 1849 * Our smoothed RTT is then the fraction of observed 1850 * samples that are 1 tick instead of 0 (times 500 1851 * ms). 1852 * 1853 * ts_rtt is increased by 1 to denote a valid sample, 1854 * with 0 indicating an invalid measurement. This 1855 * extra 1 must be removed when ts_rtt is used, or 1856 * else an an erroneous extra 500 ms will result. 1857 */ 1858 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1859 if (ts_rtt > TCP_PAWS_IDLE) 1860 ts_rtt = 0; 1861 } else { 1862 ts_rtt = 0; 1863 } 1864 1865 /* 1866 * Fast path: check for the two common cases of a uni-directional 1867 * data transfer. If: 1868 * o We are in the ESTABLISHED state, and 1869 * o The packet has no control flags, and 1870 * o The packet is in-sequence, and 1871 * o The window didn't change, and 1872 * o We are not retransmitting 1873 * It's a candidate. 1874 * 1875 * If the length (tlen) is zero and the ack moved forward, we're 1876 * the sender side of the transfer. Just free the data acked and 1877 * wake any higher level process that was blocked waiting for 1878 * space. 1879 * 1880 * If the length is non-zero and the ack didn't move, we're the 1881 * receiver side. If we're getting packets in-order (the reassembly 1882 * queue is empty), add the data to the socket buffer and note 1883 * that we need a delayed ack. 1884 */ 1885 if (tp->t_state == TCPS_ESTABLISHED && 1886 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1887 == TH_ACK && 1888 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1889 th->th_seq == tp->rcv_nxt && 1890 tiwin && tiwin == tp->snd_wnd && 1891 tp->snd_nxt == tp->snd_max) { 1892 1893 /* 1894 * If last ACK falls within this segment's sequence numbers, 1895 * record the timestamp. 1896 * NOTE that the test is modified according to the latest 1897 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 1898 * 1899 * note that we already know 1900 * TSTMP_GEQ(opti.ts_val, tp->ts_recent) 1901 */ 1902 if (opti.ts_present && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1903 tp->ts_recent_age = tcp_now; 1904 tp->ts_recent = opti.ts_val; 1905 } 1906 1907 if (tlen == 0) { 1908 /* Ack prediction. */ 1909 if (SEQ_GT(th->th_ack, tp->snd_una) && 1910 SEQ_LEQ(th->th_ack, tp->snd_max) && 1911 tp->snd_cwnd >= tp->snd_wnd && 1912 tp->t_partialacks < 0) { 1913 /* 1914 * this is a pure ack for outstanding data. 1915 */ 1916 if (ts_rtt) 1917 tcp_xmit_timer(tp, ts_rtt - 1); 1918 else if (tp->t_rtttime && 1919 SEQ_GT(th->th_ack, tp->t_rtseq)) 1920 tcp_xmit_timer(tp, 1921 tcp_now - tp->t_rtttime); 1922 acked = th->th_ack - tp->snd_una; 1923 tcps = TCP_STAT_GETREF(); 1924 tcps[TCP_STAT_PREDACK]++; 1925 tcps[TCP_STAT_RCVACKPACK]++; 1926 tcps[TCP_STAT_RCVACKBYTE] += acked; 1927 TCP_STAT_PUTREF(); 1928 nd6_hint(tp); 1929 1930 if (acked > (tp->t_lastoff - tp->t_inoff)) 1931 tp->t_lastm = NULL; 1932 sbdrop(&so->so_snd, acked); 1933 tp->t_lastoff -= acked; 1934 1935 icmp_check(tp, th, acked); 1936 1937 tp->snd_una = th->th_ack; 1938 tp->snd_fack = tp->snd_una; 1939 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1940 tp->snd_high = tp->snd_una; 1941 m_freem(m); 1942 1943 /* 1944 * If all outstanding data are acked, stop 1945 * retransmit timer, otherwise restart timer 1946 * using current (possibly backed-off) value. 1947 * If process is waiting for space, 1948 * wakeup/selnotify/signal. If data 1949 * are ready to send, let tcp_output 1950 * decide between more output or persist. 1951 */ 1952 if (tp->snd_una == tp->snd_max) 1953 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1954 else if (TCP_TIMER_ISARMED(tp, 1955 TCPT_PERSIST) == 0) 1956 TCP_TIMER_ARM(tp, TCPT_REXMT, 1957 tp->t_rxtcur); 1958 1959 sowwakeup(so); 1960 if (so->so_snd.sb_cc) { 1961 KERNEL_LOCK(1, NULL); 1962 (void)tcp_output(tp); 1963 KERNEL_UNLOCK_ONE(NULL); 1964 } 1965 if (tcp_saveti) 1966 m_freem(tcp_saveti); 1967 return; 1968 } 1969 } else if (th->th_ack == tp->snd_una && 1970 TAILQ_FIRST(&tp->segq) == NULL && 1971 tlen <= sbspace(&so->so_rcv)) { 1972 int newsize = 0; 1973 1974 /* 1975 * this is a pure, in-sequence data packet 1976 * with nothing on the reassembly queue and 1977 * we have enough buffer space to take it. 1978 */ 1979 tp->rcv_nxt += tlen; 1980 tcps = TCP_STAT_GETREF(); 1981 tcps[TCP_STAT_PREDDAT]++; 1982 tcps[TCP_STAT_RCVPACK]++; 1983 tcps[TCP_STAT_RCVBYTE] += tlen; 1984 TCP_STAT_PUTREF(); 1985 nd6_hint(tp); 1986 1987 /* 1988 * Automatic sizing enables the performance of large buffers 1989 * and most of the efficiency of small ones by only allocating 1990 * space when it is needed. 1991 * 1992 * On the receive side the socket buffer memory is only rarely 1993 * used to any significant extent. This allows us to be much 1994 * more aggressive in scaling the receive socket buffer. For 1995 * the case that the buffer space is actually used to a large 1996 * extent and we run out of kernel memory we can simply drop 1997 * the new segments; TCP on the sender will just retransmit it 1998 * later. Setting the buffer size too big may only consume too 1999 * much kernel memory if the application doesn't read() from 2000 * the socket or packet loss or reordering makes use of the 2001 * reassembly queue. 2002 * 2003 * The criteria to step up the receive buffer one notch are: 2004 * 1. the number of bytes received during the time it takes 2005 * one timestamp to be reflected back to us (the RTT); 2006 * 2. received bytes per RTT is within seven eighth of the 2007 * current socket buffer size; 2008 * 3. receive buffer size has not hit maximal automatic size; 2009 * 2010 * This algorithm does one step per RTT at most and only if 2011 * we receive a bulk stream w/o packet losses or reorderings. 2012 * Shrinking the buffer during idle times is not necessary as 2013 * it doesn't consume any memory when idle. 2014 * 2015 * TODO: Only step up if the application is actually serving 2016 * the buffer to better manage the socket buffer resources. 2017 */ 2018 if (tcp_do_autorcvbuf && 2019 opti.ts_ecr && 2020 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 2021 if (opti.ts_ecr > tp->rfbuf_ts && 2022 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 2023 if (tp->rfbuf_cnt > 2024 (so->so_rcv.sb_hiwat / 8 * 7) && 2025 so->so_rcv.sb_hiwat < 2026 tcp_autorcvbuf_max) { 2027 newsize = 2028 min(so->so_rcv.sb_hiwat + 2029 tcp_autorcvbuf_inc, 2030 tcp_autorcvbuf_max); 2031 } 2032 /* Start over with next RTT. */ 2033 tp->rfbuf_ts = 0; 2034 tp->rfbuf_cnt = 0; 2035 } else 2036 tp->rfbuf_cnt += tlen; /* add up */ 2037 } 2038 2039 /* 2040 * Drop TCP, IP headers and TCP options then add data 2041 * to socket buffer. 2042 */ 2043 if (so->so_state & SS_CANTRCVMORE) { 2044 m_freem(m); 2045 } else { 2046 /* 2047 * Set new socket buffer size. 2048 * Give up when limit is reached. 2049 */ 2050 if (newsize) 2051 if (!sbreserve(&so->so_rcv, 2052 newsize, so)) 2053 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 2054 m_adj(m, toff + off); 2055 sbappendstream(&so->so_rcv, m); 2056 } 2057 sorwakeup(so); 2058 tcp_setup_ack(tp, th); 2059 if (tp->t_flags & TF_ACKNOW) { 2060 KERNEL_LOCK(1, NULL); 2061 (void)tcp_output(tp); 2062 KERNEL_UNLOCK_ONE(NULL); 2063 } 2064 if (tcp_saveti) 2065 m_freem(tcp_saveti); 2066 return; 2067 } 2068 } 2069 2070 /* 2071 * Compute mbuf offset to TCP data segment. 2072 */ 2073 hdroptlen = toff + off; 2074 2075 /* 2076 * Calculate amount of space in receive window. Receive window is 2077 * amount of space in rcv queue, but not less than advertised 2078 * window. 2079 */ 2080 { 2081 int win; 2082 win = sbspace(&so->so_rcv); 2083 if (win < 0) 2084 win = 0; 2085 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 2086 } 2087 2088 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 2089 tp->rfbuf_ts = 0; 2090 tp->rfbuf_cnt = 0; 2091 2092 switch (tp->t_state) { 2093 /* 2094 * If the state is SYN_SENT: 2095 * if seg contains an ACK, but not for our SYN, drop the input. 2096 * if seg contains a RST, then drop the connection. 2097 * if seg does not contain SYN, then drop it. 2098 * Otherwise this is an acceptable SYN segment 2099 * initialize tp->rcv_nxt and tp->irs 2100 * if seg contains ack then advance tp->snd_una 2101 * if seg contains a ECE and ECN support is enabled, the stream 2102 * is ECN capable. 2103 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2104 * arrange for segment to be acked (eventually) 2105 * continue processing rest of data/controls, beginning with URG 2106 */ 2107 case TCPS_SYN_SENT: 2108 if ((tiflags & TH_ACK) && 2109 (SEQ_LEQ(th->th_ack, tp->iss) || 2110 SEQ_GT(th->th_ack, tp->snd_max))) 2111 goto dropwithreset; 2112 if (tiflags & TH_RST) { 2113 if (tiflags & TH_ACK) 2114 tp = tcp_drop(tp, ECONNREFUSED); 2115 goto drop; 2116 } 2117 if ((tiflags & TH_SYN) == 0) 2118 goto drop; 2119 if (tiflags & TH_ACK) { 2120 tp->snd_una = th->th_ack; 2121 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2122 tp->snd_nxt = tp->snd_una; 2123 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2124 tp->snd_high = tp->snd_una; 2125 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2126 2127 if ((tiflags & TH_ECE) && tcp_do_ecn) { 2128 tp->t_flags |= TF_ECN_PERMIT; 2129 TCP_STATINC(TCP_STAT_ECN_SHS); 2130 } 2131 } 2132 tp->irs = th->th_seq; 2133 tcp_rcvseqinit(tp); 2134 tp->t_flags |= TF_ACKNOW; 2135 tcp_mss_from_peer(tp, opti.maxseg); 2136 2137 /* 2138 * Initialize the initial congestion window. If we 2139 * had to retransmit the SYN, we must initialize cwnd 2140 * to 1 segment (i.e. the Loss Window). 2141 */ 2142 if (tp->t_flags & TF_SYN_REXMT) 2143 tp->snd_cwnd = tp->t_peermss; 2144 else { 2145 int ss = tcp_init_win; 2146 if (inp != NULL && in_localaddr(inp->inp_faddr)) 2147 ss = tcp_init_win_local; 2148 #ifdef INET6 2149 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 2150 ss = tcp_init_win_local; 2151 #endif 2152 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 2153 } 2154 2155 tcp_rmx_rtt(tp); 2156 if (tiflags & TH_ACK) { 2157 TCP_STATINC(TCP_STAT_CONNECTS); 2158 /* 2159 * move tcp_established before soisconnected 2160 * because upcall handler can drive tcp_output 2161 * functionality. 2162 * XXX we might call soisconnected at the end of 2163 * all processing 2164 */ 2165 tcp_established(tp); 2166 soisconnected(so); 2167 /* Do window scaling on this connection? */ 2168 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2169 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2170 tp->snd_scale = tp->requested_s_scale; 2171 tp->rcv_scale = tp->request_r_scale; 2172 } 2173 TCP_REASS_LOCK(tp); 2174 (void)tcp_reass(tp, NULL, NULL, tlen); 2175 /* 2176 * if we didn't have to retransmit the SYN, 2177 * use its rtt as our initial srtt & rtt var. 2178 */ 2179 if (tp->t_rtttime) 2180 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2181 } else { 2182 tp->t_state = TCPS_SYN_RECEIVED; 2183 } 2184 2185 /* 2186 * Advance th->th_seq to correspond to first data byte. 2187 * If data, trim to stay within window, 2188 * dropping FIN if necessary. 2189 */ 2190 th->th_seq++; 2191 if (tlen > tp->rcv_wnd) { 2192 todrop = tlen - tp->rcv_wnd; 2193 m_adj(m, -todrop); 2194 tlen = tp->rcv_wnd; 2195 tiflags &= ~TH_FIN; 2196 tcps = TCP_STAT_GETREF(); 2197 tcps[TCP_STAT_RCVPACKAFTERWIN]++; 2198 tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop; 2199 TCP_STAT_PUTREF(); 2200 } 2201 tp->snd_wl1 = th->th_seq - 1; 2202 tp->rcv_up = th->th_seq; 2203 goto step6; 2204 2205 /* 2206 * If the state is SYN_RECEIVED: 2207 * If seg contains an ACK, but not for our SYN, drop the input 2208 * and generate an RST. See page 36, rfc793 2209 */ 2210 case TCPS_SYN_RECEIVED: 2211 if ((tiflags & TH_ACK) && 2212 (SEQ_LEQ(th->th_ack, tp->iss) || 2213 SEQ_GT(th->th_ack, tp->snd_max))) 2214 goto dropwithreset; 2215 break; 2216 } 2217 2218 /* 2219 * From here on, we're dealing with !LISTEN and !SYN_SENT. 2220 */ 2221 KASSERT(tp->t_state != TCPS_LISTEN && 2222 tp->t_state != TCPS_SYN_SENT); 2223 2224 /* 2225 * RFC1323 PAWS: if we have a timestamp reply on this segment and 2226 * it's less than ts_recent, drop it. 2227 */ 2228 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2229 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2230 /* Check to see if ts_recent is over 24 days old. */ 2231 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2232 /* 2233 * Invalidate ts_recent. If this segment updates 2234 * ts_recent, the age will be reset later and ts_recent 2235 * will get a valid value. If it does not, setting 2236 * ts_recent to zero will at least satisfy the 2237 * requirement that zero be placed in the timestamp 2238 * echo reply when ts_recent isn't valid. The 2239 * age isn't reset until we get a valid ts_recent 2240 * because we don't want out-of-order segments to be 2241 * dropped when ts_recent is old. 2242 */ 2243 tp->ts_recent = 0; 2244 } else { 2245 tcps = TCP_STAT_GETREF(); 2246 tcps[TCP_STAT_RCVDUPPACK]++; 2247 tcps[TCP_STAT_RCVDUPBYTE] += tlen; 2248 tcps[TCP_STAT_PAWSDROP]++; 2249 TCP_STAT_PUTREF(); 2250 tcp_new_dsack(tp, th->th_seq, tlen); 2251 goto dropafterack; 2252 } 2253 } 2254 2255 /* 2256 * Check that at least some bytes of the segment are within the 2257 * receive window. If segment begins before rcv_nxt, drop leading 2258 * data (and SYN); if nothing left, just ack. 2259 */ 2260 todrop = tp->rcv_nxt - th->th_seq; 2261 dupseg = false; 2262 if (todrop > 0) { 2263 if (tiflags & TH_SYN) { 2264 tiflags &= ~TH_SYN; 2265 th->th_seq++; 2266 tcp_urp_drop(th, 1, &tiflags); 2267 todrop--; 2268 } 2269 if (todrop > tlen || 2270 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2271 /* 2272 * Any valid FIN or RST must be to the left of the 2273 * window. At this point the FIN or RST must be a 2274 * duplicate or out of sequence; drop it. 2275 */ 2276 if (tiflags & TH_RST) 2277 goto drop; 2278 tiflags &= ~(TH_FIN|TH_RST); 2279 2280 /* 2281 * Send an ACK to resynchronize and drop any data. 2282 * But keep on processing for RST or ACK. 2283 */ 2284 tp->t_flags |= TF_ACKNOW; 2285 todrop = tlen; 2286 dupseg = true; 2287 tcps = TCP_STAT_GETREF(); 2288 tcps[TCP_STAT_RCVDUPPACK]++; 2289 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 2290 TCP_STAT_PUTREF(); 2291 } else if ((tiflags & TH_RST) && th->th_seq != tp->rcv_nxt) { 2292 /* 2293 * Test for reset before adjusting the sequence 2294 * number for overlapping data. 2295 */ 2296 goto dropafterack_ratelim; 2297 } else { 2298 tcps = TCP_STAT_GETREF(); 2299 tcps[TCP_STAT_RCVPARTDUPPACK]++; 2300 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 2301 TCP_STAT_PUTREF(); 2302 } 2303 tcp_new_dsack(tp, th->th_seq, todrop); 2304 hdroptlen += todrop; /* drop from head afterwards */ 2305 th->th_seq += todrop; 2306 tlen -= todrop; 2307 tcp_urp_drop(th, todrop, &tiflags); 2308 } 2309 2310 /* 2311 * If new data is received on a connection after the user processes 2312 * are gone, then RST the other end. 2313 */ 2314 if ((so->so_state & SS_NOFDREF) && 2315 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2316 tp = tcp_close(tp); 2317 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 2318 goto dropwithreset; 2319 } 2320 2321 /* 2322 * If the segment ends after the window, drop trailing data (and 2323 * PUSH and FIN); if nothing left, just ACK. 2324 */ 2325 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2326 if (todrop > 0) { 2327 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 2328 if (todrop >= tlen) { 2329 /* 2330 * The segment actually starts after the window. 2331 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2332 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2333 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2334 */ 2335 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 2336 2337 /* 2338 * If a new connection request is received while in 2339 * TIME_WAIT, drop the old connection and start over 2340 * if the sequence numbers are above the previous 2341 * ones. 2342 * 2343 * NOTE: We need to put the header fields back into 2344 * network order. 2345 */ 2346 if ((tiflags & TH_SYN) && 2347 tp->t_state == TCPS_TIME_WAIT && 2348 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2349 tp = tcp_close(tp); 2350 tcp_fields_to_net(th); 2351 goto findpcb; 2352 } 2353 2354 /* 2355 * If window is closed can only take segments at 2356 * window edge, and have to drop data and PUSH from 2357 * incoming segments. Continue processing, but 2358 * remember to ack. Otherwise, drop segment 2359 * and (if not RST) ack. 2360 */ 2361 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2362 tp->t_flags |= TF_ACKNOW; 2363 TCP_STATINC(TCP_STAT_RCVWINPROBE); 2364 } else { 2365 goto dropafterack; 2366 } 2367 } else { 2368 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 2369 } 2370 m_adj(m, -todrop); 2371 tlen -= todrop; 2372 tiflags &= ~(TH_PUSH|TH_FIN); 2373 } 2374 2375 /* 2376 * If last ACK falls within this segment's sequence numbers, 2377 * record the timestamp. 2378 * NOTE: 2379 * 1) That the test incorporates suggestions from the latest 2380 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 2381 * 2) That updating only on newer timestamps interferes with 2382 * our earlier PAWS tests, so this check should be solely 2383 * predicated on the sequence space of this segment. 2384 * 3) That we modify the segment boundary check to be 2385 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2386 * instead of RFC1323's 2387 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2388 * This modified check allows us to overcome RFC1323's 2389 * limitations as described in Stevens TCP/IP Illustrated 2390 * Vol. 2 p.869. In such cases, we can still calculate the 2391 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2392 */ 2393 if (opti.ts_present && 2394 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2395 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2396 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2397 tp->ts_recent_age = tcp_now; 2398 tp->ts_recent = opti.ts_val; 2399 } 2400 2401 /* 2402 * If the RST bit is set examine the state: 2403 * RECEIVED state: 2404 * If passive open, return to LISTEN state. 2405 * If active open, inform user that connection was refused. 2406 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT states: 2407 * Inform user that connection was reset, and close tcb. 2408 * CLOSING, LAST_ACK, TIME_WAIT states: 2409 * Close the tcb. 2410 */ 2411 if (tiflags & TH_RST) { 2412 if (th->th_seq != tp->rcv_nxt) 2413 goto dropafterack_ratelim; 2414 2415 switch (tp->t_state) { 2416 case TCPS_SYN_RECEIVED: 2417 so->so_error = ECONNREFUSED; 2418 goto close; 2419 2420 case TCPS_ESTABLISHED: 2421 case TCPS_FIN_WAIT_1: 2422 case TCPS_FIN_WAIT_2: 2423 case TCPS_CLOSE_WAIT: 2424 so->so_error = ECONNRESET; 2425 close: 2426 tp->t_state = TCPS_CLOSED; 2427 TCP_STATINC(TCP_STAT_DROPS); 2428 tp = tcp_close(tp); 2429 goto drop; 2430 2431 case TCPS_CLOSING: 2432 case TCPS_LAST_ACK: 2433 case TCPS_TIME_WAIT: 2434 tp = tcp_close(tp); 2435 goto drop; 2436 } 2437 } 2438 2439 /* 2440 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2441 * we must be in a synchronized state. RFC791 states (under RST 2442 * generation) that any unacceptable segment (an out-of-order SYN 2443 * qualifies) received in a synchronized state must elicit only an 2444 * empty acknowledgment segment ... and the connection remains in 2445 * the same state. 2446 */ 2447 if (tiflags & TH_SYN) { 2448 if (tp->rcv_nxt == th->th_seq) { 2449 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2450 TH_ACK); 2451 if (tcp_saveti) 2452 m_freem(tcp_saveti); 2453 return; 2454 } 2455 2456 goto dropafterack_ratelim; 2457 } 2458 2459 /* 2460 * If the ACK bit is off we drop the segment and return. 2461 */ 2462 if ((tiflags & TH_ACK) == 0) { 2463 if (tp->t_flags & TF_ACKNOW) 2464 goto dropafterack; 2465 else 2466 goto drop; 2467 } 2468 2469 /* 2470 * From here on, we're doing ACK processing. 2471 */ 2472 2473 switch (tp->t_state) { 2474 /* 2475 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2476 * ESTABLISHED state and continue processing, otherwise 2477 * send an RST. 2478 */ 2479 case TCPS_SYN_RECEIVED: 2480 if (SEQ_GT(tp->snd_una, th->th_ack) || 2481 SEQ_GT(th->th_ack, tp->snd_max)) 2482 goto dropwithreset; 2483 TCP_STATINC(TCP_STAT_CONNECTS); 2484 soisconnected(so); 2485 tcp_established(tp); 2486 /* Do window scaling? */ 2487 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2488 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2489 tp->snd_scale = tp->requested_s_scale; 2490 tp->rcv_scale = tp->request_r_scale; 2491 } 2492 TCP_REASS_LOCK(tp); 2493 (void)tcp_reass(tp, NULL, NULL, tlen); 2494 tp->snd_wl1 = th->th_seq - 1; 2495 /* FALLTHROUGH */ 2496 2497 /* 2498 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2499 * ACKs. If the ack is in the range 2500 * tp->snd_una < th->th_ack <= tp->snd_max 2501 * then advance tp->snd_una to th->th_ack and drop 2502 * data from the retransmission queue. If this ACK reflects 2503 * more up to date window information we update our window information. 2504 */ 2505 case TCPS_ESTABLISHED: 2506 case TCPS_FIN_WAIT_1: 2507 case TCPS_FIN_WAIT_2: 2508 case TCPS_CLOSE_WAIT: 2509 case TCPS_CLOSING: 2510 case TCPS_LAST_ACK: 2511 case TCPS_TIME_WAIT: 2512 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2513 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2514 TCP_STATINC(TCP_STAT_RCVDUPACK); 2515 /* 2516 * If we have outstanding data (other than 2517 * a window probe), this is a completely 2518 * duplicate ack (ie, window info didn't 2519 * change), the ack is the biggest we've 2520 * seen and we've seen exactly our rexmt 2521 * threshhold of them, assume a packet 2522 * has been dropped and retransmit it. 2523 * Kludge snd_nxt & the congestion 2524 * window so we send only this one 2525 * packet. 2526 */ 2527 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2528 th->th_ack != tp->snd_una) 2529 tp->t_dupacks = 0; 2530 else if (tp->t_partialacks < 0 && 2531 (++tp->t_dupacks == tcprexmtthresh || 2532 TCP_FACK_FASTRECOV(tp))) { 2533 /* 2534 * Do the fast retransmit, and adjust 2535 * congestion control paramenters. 2536 */ 2537 if (tp->t_congctl->fast_retransmit(tp, th)) { 2538 /* False fast retransmit */ 2539 break; 2540 } else 2541 goto drop; 2542 } else if (tp->t_dupacks > tcprexmtthresh) { 2543 tp->snd_cwnd += tp->t_segsz; 2544 KERNEL_LOCK(1, NULL); 2545 (void)tcp_output(tp); 2546 KERNEL_UNLOCK_ONE(NULL); 2547 goto drop; 2548 } 2549 } else { 2550 /* 2551 * If the ack appears to be very old, only 2552 * allow data that is in-sequence. This 2553 * makes it somewhat more difficult to insert 2554 * forged data by guessing sequence numbers. 2555 * Sent an ack to try to update the send 2556 * sequence number on the other side. 2557 */ 2558 if (tlen && th->th_seq != tp->rcv_nxt && 2559 SEQ_LT(th->th_ack, 2560 tp->snd_una - tp->max_sndwnd)) 2561 goto dropafterack; 2562 } 2563 break; 2564 } 2565 /* 2566 * If the congestion window was inflated to account 2567 * for the other side's cached packets, retract it. 2568 */ 2569 tp->t_congctl->fast_retransmit_newack(tp, th); 2570 2571 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2572 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2573 goto dropafterack; 2574 } 2575 acked = th->th_ack - tp->snd_una; 2576 tcps = TCP_STAT_GETREF(); 2577 tcps[TCP_STAT_RCVACKPACK]++; 2578 tcps[TCP_STAT_RCVACKBYTE] += acked; 2579 TCP_STAT_PUTREF(); 2580 2581 /* 2582 * If we have a timestamp reply, update smoothed 2583 * round trip time. If no timestamp is present but 2584 * transmit timer is running and timed sequence 2585 * number was acked, update smoothed round trip time. 2586 * Since we now have an rtt measurement, cancel the 2587 * timer backoff (cf., Phil Karn's retransmit alg.). 2588 * Recompute the initial retransmit timer. 2589 */ 2590 if (ts_rtt) 2591 tcp_xmit_timer(tp, ts_rtt - 1); 2592 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2593 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2594 2595 /* 2596 * If all outstanding data is acked, stop retransmit 2597 * timer and remember to restart (more output or persist). 2598 * If there is more data to be acked, restart retransmit 2599 * timer, using current (possibly backed-off) value. 2600 */ 2601 if (th->th_ack == tp->snd_max) { 2602 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2603 needoutput = 1; 2604 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2605 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2606 2607 /* 2608 * New data has been acked, adjust the congestion window. 2609 */ 2610 tp->t_congctl->newack(tp, th); 2611 2612 nd6_hint(tp); 2613 if (acked > so->so_snd.sb_cc) { 2614 tp->snd_wnd -= so->so_snd.sb_cc; 2615 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2616 ourfinisacked = 1; 2617 } else { 2618 if (acked > (tp->t_lastoff - tp->t_inoff)) 2619 tp->t_lastm = NULL; 2620 sbdrop(&so->so_snd, acked); 2621 tp->t_lastoff -= acked; 2622 if (tp->snd_wnd > acked) 2623 tp->snd_wnd -= acked; 2624 else 2625 tp->snd_wnd = 0; 2626 ourfinisacked = 0; 2627 } 2628 sowwakeup(so); 2629 2630 icmp_check(tp, th, acked); 2631 2632 tp->snd_una = th->th_ack; 2633 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2634 tp->snd_fack = tp->snd_una; 2635 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2636 tp->snd_nxt = tp->snd_una; 2637 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2638 tp->snd_high = tp->snd_una; 2639 2640 switch (tp->t_state) { 2641 2642 /* 2643 * In FIN_WAIT_1 STATE in addition to the processing 2644 * for the ESTABLISHED state if our FIN is now acknowledged 2645 * then enter FIN_WAIT_2. 2646 */ 2647 case TCPS_FIN_WAIT_1: 2648 if (ourfinisacked) { 2649 /* 2650 * If we can't receive any more 2651 * data, then closing user can proceed. 2652 * Starting the timer is contrary to the 2653 * specification, but if we don't get a FIN 2654 * we'll hang forever. 2655 */ 2656 if (so->so_state & SS_CANTRCVMORE) { 2657 soisdisconnected(so); 2658 if (tp->t_maxidle > 0) 2659 TCP_TIMER_ARM(tp, TCPT_2MSL, 2660 tp->t_maxidle); 2661 } 2662 tp->t_state = TCPS_FIN_WAIT_2; 2663 } 2664 break; 2665 2666 /* 2667 * In CLOSING STATE in addition to the processing for 2668 * the ESTABLISHED state if the ACK acknowledges our FIN 2669 * then enter the TIME-WAIT state, otherwise ignore 2670 * the segment. 2671 */ 2672 case TCPS_CLOSING: 2673 if (ourfinisacked) { 2674 tp->t_state = TCPS_TIME_WAIT; 2675 tcp_canceltimers(tp); 2676 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2677 soisdisconnected(so); 2678 } 2679 break; 2680 2681 /* 2682 * In LAST_ACK, we may still be waiting for data to drain 2683 * and/or to be acked, as well as for the ack of our FIN. 2684 * If our FIN is now acknowledged, delete the TCB, 2685 * enter the closed state and return. 2686 */ 2687 case TCPS_LAST_ACK: 2688 if (ourfinisacked) { 2689 tp = tcp_close(tp); 2690 goto drop; 2691 } 2692 break; 2693 2694 /* 2695 * In TIME_WAIT state the only thing that should arrive 2696 * is a retransmission of the remote FIN. Acknowledge 2697 * it and restart the finack timer. 2698 */ 2699 case TCPS_TIME_WAIT: 2700 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2701 goto dropafterack; 2702 } 2703 } 2704 2705 step6: 2706 /* 2707 * Update window information. 2708 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2709 */ 2710 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2711 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2712 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2713 /* keep track of pure window updates */ 2714 if (tlen == 0 && 2715 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2716 TCP_STATINC(TCP_STAT_RCVWINUPD); 2717 tp->snd_wnd = tiwin; 2718 tp->snd_wl1 = th->th_seq; 2719 tp->snd_wl2 = th->th_ack; 2720 if (tp->snd_wnd > tp->max_sndwnd) 2721 tp->max_sndwnd = tp->snd_wnd; 2722 needoutput = 1; 2723 } 2724 2725 /* 2726 * Process segments with URG. 2727 */ 2728 if ((tiflags & TH_URG) && th->th_urp && 2729 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2730 /* 2731 * This is a kludge, but if we receive and accept 2732 * random urgent pointers, we'll crash in 2733 * soreceive. It's hard to imagine someone 2734 * actually wanting to send this much urgent data. 2735 */ 2736 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2737 th->th_urp = 0; /* XXX */ 2738 tiflags &= ~TH_URG; /* XXX */ 2739 goto dodata; /* XXX */ 2740 } 2741 2742 /* 2743 * If this segment advances the known urgent pointer, 2744 * then mark the data stream. This should not happen 2745 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2746 * a FIN has been received from the remote side. 2747 * In these states we ignore the URG. 2748 * 2749 * According to RFC961 (Assigned Protocols), 2750 * the urgent pointer points to the last octet 2751 * of urgent data. We continue, however, 2752 * to consider it to indicate the first octet 2753 * of data past the urgent section as the original 2754 * spec states (in one of two places). 2755 */ 2756 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2757 tp->rcv_up = th->th_seq + th->th_urp; 2758 so->so_oobmark = so->so_rcv.sb_cc + 2759 (tp->rcv_up - tp->rcv_nxt) - 1; 2760 if (so->so_oobmark == 0) 2761 so->so_state |= SS_RCVATMARK; 2762 sohasoutofband(so); 2763 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2764 } 2765 2766 /* 2767 * Remove out of band data so doesn't get presented to user. 2768 * This can happen independent of advancing the URG pointer, 2769 * but if two URG's are pending at once, some out-of-band 2770 * data may creep in... ick. 2771 */ 2772 if (th->th_urp <= (u_int16_t)tlen && 2773 (so->so_options & SO_OOBINLINE) == 0) 2774 tcp_pulloutofband(so, th, m, hdroptlen); 2775 } else { 2776 /* 2777 * If no out of band data is expected, 2778 * pull receive urgent pointer along 2779 * with the receive window. 2780 */ 2781 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2782 tp->rcv_up = tp->rcv_nxt; 2783 } 2784 dodata: 2785 2786 /* 2787 * Process the segment text, merging it into the TCP sequencing queue, 2788 * and arranging for acknowledgement of receipt if necessary. 2789 * This process logically involves adjusting tp->rcv_wnd as data 2790 * is presented to the user (this happens in tcp_usrreq.c, 2791 * tcp_rcvd()). If a FIN has already been received on this 2792 * connection then we just ignore the text. 2793 */ 2794 if ((tlen || (tiflags & TH_FIN)) && 2795 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2796 /* 2797 * Insert segment ti into reassembly queue of tcp with 2798 * control block tp. Return TH_FIN if reassembly now includes 2799 * a segment with FIN. The macro form does the common case 2800 * inline (segment is the next to be received on an 2801 * established connection, and the queue is empty), 2802 * avoiding linkage into and removal from the queue and 2803 * repetition of various conversions. 2804 * Set DELACK for segments received in order, but ack 2805 * immediately when segments are out of order 2806 * (so fast retransmit can work). 2807 */ 2808 /* NOTE: this was TCP_REASS() macro, but used only once */ 2809 TCP_REASS_LOCK(tp); 2810 if (th->th_seq == tp->rcv_nxt && 2811 TAILQ_FIRST(&tp->segq) == NULL && 2812 tp->t_state == TCPS_ESTABLISHED) { 2813 tcp_setup_ack(tp, th); 2814 tp->rcv_nxt += tlen; 2815 tiflags = th->th_flags & TH_FIN; 2816 tcps = TCP_STAT_GETREF(); 2817 tcps[TCP_STAT_RCVPACK]++; 2818 tcps[TCP_STAT_RCVBYTE] += tlen; 2819 TCP_STAT_PUTREF(); 2820 nd6_hint(tp); 2821 if (so->so_state & SS_CANTRCVMORE) { 2822 m_freem(m); 2823 } else { 2824 m_adj(m, hdroptlen); 2825 sbappendstream(&(so)->so_rcv, m); 2826 } 2827 TCP_REASS_UNLOCK(tp); 2828 sorwakeup(so); 2829 } else { 2830 m_adj(m, hdroptlen); 2831 tiflags = tcp_reass(tp, th, m, tlen); 2832 tp->t_flags |= TF_ACKNOW; 2833 } 2834 2835 /* 2836 * Note the amount of data that peer has sent into 2837 * our window, in order to estimate the sender's 2838 * buffer size. 2839 */ 2840 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2841 } else { 2842 m_freem(m); 2843 m = NULL; 2844 tiflags &= ~TH_FIN; 2845 } 2846 2847 /* 2848 * If FIN is received ACK the FIN and let the user know 2849 * that the connection is closing. Ignore a FIN received before 2850 * the connection is fully established. 2851 */ 2852 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2853 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2854 socantrcvmore(so); 2855 tp->t_flags |= TF_ACKNOW; 2856 tp->rcv_nxt++; 2857 } 2858 switch (tp->t_state) { 2859 2860 /* 2861 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2862 */ 2863 case TCPS_ESTABLISHED: 2864 tp->t_state = TCPS_CLOSE_WAIT; 2865 break; 2866 2867 /* 2868 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2869 * enter the CLOSING state. 2870 */ 2871 case TCPS_FIN_WAIT_1: 2872 tp->t_state = TCPS_CLOSING; 2873 break; 2874 2875 /* 2876 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2877 * starting the time-wait timer, turning off the other 2878 * standard timers. 2879 */ 2880 case TCPS_FIN_WAIT_2: 2881 tp->t_state = TCPS_TIME_WAIT; 2882 tcp_canceltimers(tp); 2883 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2884 soisdisconnected(so); 2885 break; 2886 2887 /* 2888 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2889 */ 2890 case TCPS_TIME_WAIT: 2891 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2892 break; 2893 } 2894 } 2895 #ifdef TCP_DEBUG 2896 if (so->so_options & SO_DEBUG) 2897 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2898 #endif 2899 2900 /* 2901 * Return any desired output. 2902 */ 2903 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2904 KERNEL_LOCK(1, NULL); 2905 (void)tcp_output(tp); 2906 KERNEL_UNLOCK_ONE(NULL); 2907 } 2908 if (tcp_saveti) 2909 m_freem(tcp_saveti); 2910 2911 if (tp->t_state == TCPS_TIME_WAIT 2912 && (so->so_state & SS_NOFDREF) 2913 && (tp->t_inpcb || af != AF_INET) 2914 && (tp->t_in6pcb || af != AF_INET6) 2915 && ((af == AF_INET ? tcp4_vtw_enable : tcp6_vtw_enable) & 1) != 0 2916 && TAILQ_EMPTY(&tp->segq) 2917 && vtw_add(af, tp)) { 2918 ; 2919 } 2920 return; 2921 2922 badsyn: 2923 /* 2924 * Received a bad SYN. Increment counters and dropwithreset. 2925 */ 2926 TCP_STATINC(TCP_STAT_BADSYN); 2927 tp = NULL; 2928 goto dropwithreset; 2929 2930 dropafterack: 2931 /* 2932 * Generate an ACK dropping incoming segment if it occupies 2933 * sequence space, where the ACK reflects our state. 2934 */ 2935 if (tiflags & TH_RST) 2936 goto drop; 2937 goto dropafterack2; 2938 2939 dropafterack_ratelim: 2940 /* 2941 * We may want to rate-limit ACKs against SYN/RST attack. 2942 */ 2943 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2944 tcp_ackdrop_ppslim) == 0) { 2945 /* XXX stat */ 2946 goto drop; 2947 } 2948 2949 dropafterack2: 2950 m_freem(m); 2951 tp->t_flags |= TF_ACKNOW; 2952 KERNEL_LOCK(1, NULL); 2953 (void)tcp_output(tp); 2954 KERNEL_UNLOCK_ONE(NULL); 2955 if (tcp_saveti) 2956 m_freem(tcp_saveti); 2957 return; 2958 2959 dropwithreset_ratelim: 2960 /* 2961 * We may want to rate-limit RSTs in certain situations, 2962 * particularly if we are sending an RST in response to 2963 * an attempt to connect to or otherwise communicate with 2964 * a port for which we have no socket. 2965 */ 2966 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2967 tcp_rst_ppslim) == 0) { 2968 /* XXX stat */ 2969 goto drop; 2970 } 2971 2972 dropwithreset: 2973 /* 2974 * Generate a RST, dropping incoming segment. 2975 * Make ACK acceptable to originator of segment. 2976 */ 2977 if (tiflags & TH_RST) 2978 goto drop; 2979 if (tiflags & TH_ACK) { 2980 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2981 } else { 2982 if (tiflags & TH_SYN) 2983 tlen++; 2984 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2985 TH_RST|TH_ACK); 2986 } 2987 if (tcp_saveti) 2988 m_freem(tcp_saveti); 2989 return; 2990 2991 badcsum: 2992 drop: 2993 /* 2994 * Drop space held by incoming segment and return. 2995 */ 2996 if (tp) { 2997 if (tp->t_inpcb) 2998 so = tp->t_inpcb->inp_socket; 2999 #ifdef INET6 3000 else if (tp->t_in6pcb) 3001 so = tp->t_in6pcb->in6p_socket; 3002 #endif 3003 else 3004 so = NULL; 3005 #ifdef TCP_DEBUG 3006 if (so && (so->so_options & SO_DEBUG) != 0) 3007 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 3008 #endif 3009 } 3010 if (tcp_saveti) 3011 m_freem(tcp_saveti); 3012 m_freem(m); 3013 return; 3014 } 3015 3016 #ifdef TCP_SIGNATURE 3017 int 3018 tcp_signature_apply(void *fstate, void *data, u_int len) 3019 { 3020 3021 MD5Update(fstate, (u_char *)data, len); 3022 return (0); 3023 } 3024 3025 struct secasvar * 3026 tcp_signature_getsav(struct mbuf *m) 3027 { 3028 struct ip *ip; 3029 struct ip6_hdr *ip6; 3030 3031 ip = mtod(m, struct ip *); 3032 switch (ip->ip_v) { 3033 case 4: 3034 ip = mtod(m, struct ip *); 3035 ip6 = NULL; 3036 break; 3037 case 6: 3038 ip = NULL; 3039 ip6 = mtod(m, struct ip6_hdr *); 3040 break; 3041 default: 3042 return (NULL); 3043 } 3044 3045 #ifdef IPSEC 3046 union sockaddr_union dst; 3047 3048 /* Extract the destination from the IP header in the mbuf. */ 3049 memset(&dst, 0, sizeof(union sockaddr_union)); 3050 if (ip != NULL) { 3051 dst.sa.sa_len = sizeof(struct sockaddr_in); 3052 dst.sa.sa_family = AF_INET; 3053 dst.sin.sin_addr = ip->ip_dst; 3054 } else { 3055 dst.sa.sa_len = sizeof(struct sockaddr_in6); 3056 dst.sa.sa_family = AF_INET6; 3057 dst.sin6.sin6_addr = ip6->ip6_dst; 3058 } 3059 3060 /* 3061 * Look up an SADB entry which matches the address of the peer. 3062 */ 3063 return KEY_LOOKUP_SA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI), 0, 0); 3064 #else 3065 return NULL; 3066 #endif 3067 } 3068 3069 int 3070 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 3071 struct secasvar *sav, char *sig) 3072 { 3073 MD5_CTX ctx; 3074 struct ip *ip; 3075 struct ipovly *ipovly; 3076 #ifdef INET6 3077 struct ip6_hdr *ip6; 3078 struct ip6_hdr_pseudo ip6pseudo; 3079 #endif 3080 struct ippseudo ippseudo; 3081 struct tcphdr th0; 3082 int l, tcphdrlen; 3083 3084 if (sav == NULL) 3085 return (-1); 3086 3087 tcphdrlen = th->th_off * 4; 3088 3089 switch (mtod(m, struct ip *)->ip_v) { 3090 case 4: 3091 MD5Init(&ctx); 3092 ip = mtod(m, struct ip *); 3093 memset(&ippseudo, 0, sizeof(ippseudo)); 3094 ipovly = (struct ipovly *)ip; 3095 ippseudo.ippseudo_src = ipovly->ih_src; 3096 ippseudo.ippseudo_dst = ipovly->ih_dst; 3097 ippseudo.ippseudo_pad = 0; 3098 ippseudo.ippseudo_p = IPPROTO_TCP; 3099 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 3100 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 3101 break; 3102 #if INET6 3103 case 6: 3104 MD5Init(&ctx); 3105 ip6 = mtod(m, struct ip6_hdr *); 3106 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 3107 ip6pseudo.ip6ph_src = ip6->ip6_src; 3108 in6_clearscope(&ip6pseudo.ip6ph_src); 3109 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 3110 in6_clearscope(&ip6pseudo.ip6ph_dst); 3111 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 3112 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 3113 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 3114 break; 3115 #endif 3116 default: 3117 return (-1); 3118 } 3119 3120 th0 = *th; 3121 th0.th_sum = 0; 3122 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 3123 3124 l = m->m_pkthdr.len - thoff - tcphdrlen; 3125 if (l > 0) 3126 m_apply(m, thoff + tcphdrlen, 3127 m->m_pkthdr.len - thoff - tcphdrlen, 3128 tcp_signature_apply, &ctx); 3129 3130 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 3131 MD5Final(sig, &ctx); 3132 3133 return (0); 3134 } 3135 #endif 3136 3137 /* 3138 * Parse and process tcp options. 3139 * 3140 * Returns -1 if this segment should be dropped. (eg. wrong signature) 3141 * Otherwise returns 0. 3142 */ 3143 static int 3144 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, struct tcphdr *th, 3145 struct mbuf *m, int toff, struct tcp_opt_info *oi) 3146 { 3147 u_int16_t mss; 3148 int opt, optlen = 0; 3149 #ifdef TCP_SIGNATURE 3150 void *sigp = NULL; 3151 char sigbuf[TCP_SIGLEN]; 3152 struct secasvar *sav = NULL; 3153 #endif 3154 3155 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 3156 opt = cp[0]; 3157 if (opt == TCPOPT_EOL) 3158 break; 3159 if (opt == TCPOPT_NOP) 3160 optlen = 1; 3161 else { 3162 if (cnt < 2) 3163 break; 3164 optlen = cp[1]; 3165 if (optlen < 2 || optlen > cnt) 3166 break; 3167 } 3168 switch (opt) { 3169 3170 default: 3171 continue; 3172 3173 case TCPOPT_MAXSEG: 3174 if (optlen != TCPOLEN_MAXSEG) 3175 continue; 3176 if (!(th->th_flags & TH_SYN)) 3177 continue; 3178 if (TCPS_HAVERCVDSYN(tp->t_state)) 3179 continue; 3180 memcpy(&mss, cp + 2, sizeof(mss)); 3181 oi->maxseg = ntohs(mss); 3182 break; 3183 3184 case TCPOPT_WINDOW: 3185 if (optlen != TCPOLEN_WINDOW) 3186 continue; 3187 if (!(th->th_flags & TH_SYN)) 3188 continue; 3189 if (TCPS_HAVERCVDSYN(tp->t_state)) 3190 continue; 3191 tp->t_flags |= TF_RCVD_SCALE; 3192 tp->requested_s_scale = cp[2]; 3193 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3194 char buf[INET6_ADDRSTRLEN]; 3195 struct ip *ip = mtod(m, struct ip *); 3196 #ifdef INET6 3197 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 3198 #endif 3199 if (ip) 3200 in_print(buf, sizeof(buf), 3201 &ip->ip_src); 3202 #ifdef INET6 3203 else if (ip6) 3204 in6_print(buf, sizeof(buf), 3205 &ip6->ip6_src); 3206 #endif 3207 else 3208 strlcpy(buf, "(unknown)", sizeof(buf)); 3209 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3210 "assuming %d\n", 3211 tp->requested_s_scale, buf, 3212 TCP_MAX_WINSHIFT); 3213 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3214 } 3215 break; 3216 3217 case TCPOPT_TIMESTAMP: 3218 if (optlen != TCPOLEN_TIMESTAMP) 3219 continue; 3220 oi->ts_present = 1; 3221 memcpy(&oi->ts_val, cp + 2, sizeof(oi->ts_val)); 3222 NTOHL(oi->ts_val); 3223 memcpy(&oi->ts_ecr, cp + 6, sizeof(oi->ts_ecr)); 3224 NTOHL(oi->ts_ecr); 3225 3226 if (!(th->th_flags & TH_SYN)) 3227 continue; 3228 if (TCPS_HAVERCVDSYN(tp->t_state)) 3229 continue; 3230 /* 3231 * A timestamp received in a SYN makes 3232 * it ok to send timestamp requests and replies. 3233 */ 3234 tp->t_flags |= TF_RCVD_TSTMP; 3235 tp->ts_recent = oi->ts_val; 3236 tp->ts_recent_age = tcp_now; 3237 break; 3238 3239 case TCPOPT_SACK_PERMITTED: 3240 if (optlen != TCPOLEN_SACK_PERMITTED) 3241 continue; 3242 if (!(th->th_flags & TH_SYN)) 3243 continue; 3244 if (TCPS_HAVERCVDSYN(tp->t_state)) 3245 continue; 3246 if (tcp_do_sack) { 3247 tp->t_flags |= TF_SACK_PERMIT; 3248 tp->t_flags |= TF_WILL_SACK; 3249 } 3250 break; 3251 3252 case TCPOPT_SACK: 3253 tcp_sack_option(tp, th, cp, optlen); 3254 break; 3255 #ifdef TCP_SIGNATURE 3256 case TCPOPT_SIGNATURE: 3257 if (optlen != TCPOLEN_SIGNATURE) 3258 continue; 3259 if (sigp && memcmp(sigp, cp + 2, TCP_SIGLEN)) 3260 return (-1); 3261 3262 sigp = sigbuf; 3263 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3264 tp->t_flags |= TF_SIGNATURE; 3265 break; 3266 #endif 3267 } 3268 } 3269 3270 #ifndef TCP_SIGNATURE 3271 return 0; 3272 #else 3273 if (tp->t_flags & TF_SIGNATURE) { 3274 sav = tcp_signature_getsav(m); 3275 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3276 return (-1); 3277 } 3278 3279 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) 3280 goto out; 3281 3282 if (sigp) { 3283 char sig[TCP_SIGLEN]; 3284 3285 tcp_fields_to_net(th); 3286 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3287 tcp_fields_to_host(th); 3288 goto out; 3289 } 3290 tcp_fields_to_host(th); 3291 3292 if (memcmp(sig, sigp, TCP_SIGLEN)) { 3293 TCP_STATINC(TCP_STAT_BADSIG); 3294 goto out; 3295 } else 3296 TCP_STATINC(TCP_STAT_GOODSIG); 3297 3298 key_sa_recordxfer(sav, m); 3299 KEY_SA_UNREF(&sav); 3300 } 3301 return 0; 3302 out: 3303 if (sav != NULL) 3304 KEY_SA_UNREF(&sav); 3305 return -1; 3306 #endif 3307 } 3308 3309 /* 3310 * Pull out of band byte out of a segment so 3311 * it doesn't appear in the user's data queue. 3312 * It is still reflected in the segment length for 3313 * sequencing purposes. 3314 */ 3315 void 3316 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3317 struct mbuf *m, int off) 3318 { 3319 int cnt = off + th->th_urp - 1; 3320 3321 while (cnt >= 0) { 3322 if (m->m_len > cnt) { 3323 char *cp = mtod(m, char *) + cnt; 3324 struct tcpcb *tp = sototcpcb(so); 3325 3326 tp->t_iobc = *cp; 3327 tp->t_oobflags |= TCPOOB_HAVEDATA; 3328 memmove(cp, cp + 1, (unsigned)(m->m_len - cnt - 1)); 3329 m->m_len--; 3330 return; 3331 } 3332 cnt -= m->m_len; 3333 m = m->m_next; 3334 if (m == NULL) 3335 break; 3336 } 3337 panic("tcp_pulloutofband"); 3338 } 3339 3340 /* 3341 * Collect new round-trip time estimate 3342 * and update averages and current timeout. 3343 * 3344 * rtt is in units of slow ticks (typically 500 ms) -- essentially the 3345 * difference of two timestamps. 3346 */ 3347 void 3348 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3349 { 3350 int32_t delta; 3351 3352 TCP_STATINC(TCP_STAT_RTTUPDATED); 3353 if (tp->t_srtt != 0) { 3354 /* 3355 * Compute the amount to add to srtt for smoothing, 3356 * *alpha, or 2^(-TCP_RTT_SHIFT). Because 3357 * srtt is stored in 1/32 slow ticks, we conceptually 3358 * shift left 5 bits, subtract srtt to get the 3359 * diference, and then shift right by TCP_RTT_SHIFT 3360 * (3) to obtain 1/8 of the difference. 3361 */ 3362 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3363 /* 3364 * This can never happen, because delta's lowest 3365 * possible value is 1/8 of t_srtt. But if it does, 3366 * set srtt to some reasonable value, here chosen 3367 * as 1/8 tick. 3368 */ 3369 if ((tp->t_srtt += delta) <= 0) 3370 tp->t_srtt = 1 << 2; 3371 /* 3372 * RFC2988 requires that rttvar be updated first. 3373 * This code is compliant because "delta" is the old 3374 * srtt minus the new observation (scaled). 3375 * 3376 * RFC2988 says: 3377 * rttvar = (1-beta) * rttvar + beta * |srtt-observed| 3378 * 3379 * delta is in units of 1/32 ticks, and has then been 3380 * divided by 8. This is equivalent to being in 1/16s 3381 * units and divided by 4. Subtract from it 1/4 of 3382 * the existing rttvar to form the (signed) amount to 3383 * adjust. 3384 */ 3385 if (delta < 0) 3386 delta = -delta; 3387 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3388 /* 3389 * As with srtt, this should never happen. There is 3390 * no support in RFC2988 for this operation. But 1/4s 3391 * as rttvar when faced with something arguably wrong 3392 * is ok. 3393 */ 3394 if ((tp->t_rttvar += delta) <= 0) 3395 tp->t_rttvar = 1 << 2; 3396 3397 /* 3398 * If srtt exceeds .01 second, ensure we use the 'remote' MSL 3399 * Problem is: it doesn't work. Disabled by defaulting 3400 * tcp_rttlocal to 0; see corresponding code in 3401 * tcp_subr that selects local vs remote in a different way. 3402 * 3403 * The static branch prediction hint here should be removed 3404 * when the rtt estimator is fixed and the rtt_enable code 3405 * is turned back on. 3406 */ 3407 if (__predict_false(tcp_rttlocal) && tcp_msl_enable 3408 && tp->t_srtt > tcp_msl_remote_threshold 3409 && tp->t_msl < tcp_msl_remote) { 3410 tp->t_msl = tcp_msl_remote; 3411 } 3412 } else { 3413 /* 3414 * This is the first measurement. Per RFC2988, 2.2, 3415 * set rtt=R and srtt=R/2. 3416 * For srtt, storage representation is 1/32 ticks, 3417 * so shift left by 5. 3418 * For rttvar, storage representation is 1/16 ticks, 3419 * So shift left by 4, but then right by 1 to halve. 3420 */ 3421 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3422 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3423 } 3424 tp->t_rtttime = 0; 3425 tp->t_rxtshift = 0; 3426 3427 /* 3428 * the retransmit should happen at rtt + 4 * rttvar. 3429 * Because of the way we do the smoothing, srtt and rttvar 3430 * will each average +1/2 tick of bias. When we compute 3431 * the retransmit timer, we want 1/2 tick of rounding and 3432 * 1 extra tick because of +-1/2 tick uncertainty in the 3433 * firing of the timer. The bias will give us exactly the 3434 * 1.5 tick we need. But, because the bias is 3435 * statistical, we have to test that we don't drop below 3436 * the minimum feasible timer (which is 2 ticks). 3437 */ 3438 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3439 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3440 3441 /* 3442 * We received an ack for a packet that wasn't retransmitted; 3443 * it is probably safe to discard any error indications we've 3444 * received recently. This isn't quite right, but close enough 3445 * for now (a route might have failed after we sent a segment, 3446 * and the return path might not be symmetrical). 3447 */ 3448 tp->t_softerror = 0; 3449 } 3450 3451 3452 /* 3453 * TCP compressed state engine. Currently used to hold compressed 3454 * state for SYN_RECEIVED. 3455 */ 3456 3457 u_long syn_cache_count; 3458 u_int32_t syn_hash1, syn_hash2; 3459 3460 #define SYN_HASH(sa, sp, dp) \ 3461 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3462 ((u_int32_t)(sp)))^syn_hash2))) 3463 #ifndef INET6 3464 #define SYN_HASHALL(hash, src, dst) \ 3465 do { \ 3466 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3467 ((const struct sockaddr_in *)(src))->sin_port, \ 3468 ((const struct sockaddr_in *)(dst))->sin_port); \ 3469 } while (/*CONSTCOND*/ 0) 3470 #else 3471 #define SYN_HASH6(sa, sp, dp) \ 3472 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3473 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3474 & 0x7fffffff) 3475 3476 #define SYN_HASHALL(hash, src, dst) \ 3477 do { \ 3478 switch ((src)->sa_family) { \ 3479 case AF_INET: \ 3480 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3481 ((const struct sockaddr_in *)(src))->sin_port, \ 3482 ((const struct sockaddr_in *)(dst))->sin_port); \ 3483 break; \ 3484 case AF_INET6: \ 3485 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3486 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3487 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3488 break; \ 3489 default: \ 3490 hash = 0; \ 3491 } \ 3492 } while (/*CONSTCOND*/0) 3493 #endif /* INET6 */ 3494 3495 static struct pool syn_cache_pool; 3496 3497 /* 3498 * We don't estimate RTT with SYNs, so each packet starts with the default 3499 * RTT and each timer step has a fixed timeout value. 3500 */ 3501 static inline void 3502 syn_cache_timer_arm(struct syn_cache *sc) 3503 { 3504 3505 TCPT_RANGESET(sc->sc_rxtcur, 3506 TCPTV_SRTTDFLT * tcp_backoff[sc->sc_rxtshift], TCPTV_MIN, 3507 TCPTV_REXMTMAX); 3508 callout_reset(&sc->sc_timer, 3509 sc->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, sc); 3510 } 3511 3512 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3513 3514 static inline void 3515 syn_cache_rm(struct syn_cache *sc) 3516 { 3517 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3518 sc, sc_bucketq); 3519 sc->sc_tp = NULL; 3520 LIST_REMOVE(sc, sc_tpq); 3521 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3522 callout_stop(&sc->sc_timer); 3523 syn_cache_count--; 3524 } 3525 3526 static inline void 3527 syn_cache_put(struct syn_cache *sc) 3528 { 3529 if (sc->sc_ipopts) 3530 (void) m_free(sc->sc_ipopts); 3531 rtcache_free(&sc->sc_route); 3532 sc->sc_flags |= SCF_DEAD; 3533 if (!callout_invoking(&sc->sc_timer)) 3534 callout_schedule(&(sc)->sc_timer, 1); 3535 } 3536 3537 void 3538 syn_cache_init(void) 3539 { 3540 int i; 3541 3542 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 3543 "synpl", NULL, IPL_SOFTNET); 3544 3545 /* Initialize the hash buckets. */ 3546 for (i = 0; i < tcp_syn_cache_size; i++) 3547 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3548 } 3549 3550 void 3551 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3552 { 3553 struct syn_cache_head *scp; 3554 struct syn_cache *sc2; 3555 int s; 3556 3557 /* 3558 * If there are no entries in the hash table, reinitialize 3559 * the hash secrets. 3560 */ 3561 if (syn_cache_count == 0) { 3562 syn_hash1 = cprng_fast32(); 3563 syn_hash2 = cprng_fast32(); 3564 } 3565 3566 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3567 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3568 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3569 3570 /* 3571 * Make sure that we don't overflow the per-bucket 3572 * limit or the total cache size limit. 3573 */ 3574 s = splsoftnet(); 3575 if (scp->sch_length >= tcp_syn_bucket_limit) { 3576 TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW); 3577 /* 3578 * The bucket is full. Toss the oldest element in the 3579 * bucket. This will be the first entry in the bucket. 3580 */ 3581 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3582 #ifdef DIAGNOSTIC 3583 /* 3584 * This should never happen; we should always find an 3585 * entry in our bucket. 3586 */ 3587 if (sc2 == NULL) 3588 panic("syn_cache_insert: bucketoverflow: impossible"); 3589 #endif 3590 syn_cache_rm(sc2); 3591 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3592 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3593 struct syn_cache_head *scp2, *sce; 3594 3595 TCP_STATINC(TCP_STAT_SC_OVERFLOWED); 3596 /* 3597 * The cache is full. Toss the oldest entry in the 3598 * first non-empty bucket we can find. 3599 * 3600 * XXX We would really like to toss the oldest 3601 * entry in the cache, but we hope that this 3602 * condition doesn't happen very often. 3603 */ 3604 scp2 = scp; 3605 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3606 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3607 for (++scp2; scp2 != scp; scp2++) { 3608 if (scp2 >= sce) 3609 scp2 = &tcp_syn_cache[0]; 3610 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3611 break; 3612 } 3613 #ifdef DIAGNOSTIC 3614 /* 3615 * This should never happen; we should always find a 3616 * non-empty bucket. 3617 */ 3618 if (scp2 == scp) 3619 panic("syn_cache_insert: cacheoverflow: " 3620 "impossible"); 3621 #endif 3622 } 3623 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3624 syn_cache_rm(sc2); 3625 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3626 } 3627 3628 /* 3629 * Initialize the entry's timer. 3630 */ 3631 sc->sc_rxttot = 0; 3632 sc->sc_rxtshift = 0; 3633 syn_cache_timer_arm(sc); 3634 3635 /* Link it from tcpcb entry */ 3636 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3637 3638 /* Put it into the bucket. */ 3639 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3640 scp->sch_length++; 3641 syn_cache_count++; 3642 3643 TCP_STATINC(TCP_STAT_SC_ADDED); 3644 splx(s); 3645 } 3646 3647 /* 3648 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3649 * If we have retransmitted an entry the maximum number of times, expire 3650 * that entry. 3651 */ 3652 static void 3653 syn_cache_timer(void *arg) 3654 { 3655 struct syn_cache *sc = arg; 3656 3657 mutex_enter(softnet_lock); 3658 KERNEL_LOCK(1, NULL); 3659 3660 callout_ack(&sc->sc_timer); 3661 3662 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3663 TCP_STATINC(TCP_STAT_SC_DELAYED_FREE); 3664 goto free; 3665 } 3666 3667 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3668 /* Drop it -- too many retransmissions. */ 3669 goto dropit; 3670 } 3671 3672 /* 3673 * Compute the total amount of time this entry has 3674 * been on a queue. If this entry has been on longer 3675 * than the keep alive timer would allow, expire it. 3676 */ 3677 sc->sc_rxttot += sc->sc_rxtcur; 3678 if (sc->sc_rxttot >= tcp_keepinit) 3679 goto dropit; 3680 3681 TCP_STATINC(TCP_STAT_SC_RETRANSMITTED); 3682 (void)syn_cache_respond(sc); 3683 3684 /* Advance the timer back-off. */ 3685 sc->sc_rxtshift++; 3686 syn_cache_timer_arm(sc); 3687 3688 goto out; 3689 3690 dropit: 3691 TCP_STATINC(TCP_STAT_SC_TIMED_OUT); 3692 syn_cache_rm(sc); 3693 if (sc->sc_ipopts) 3694 (void) m_free(sc->sc_ipopts); 3695 rtcache_free(&sc->sc_route); 3696 3697 free: 3698 callout_destroy(&sc->sc_timer); 3699 pool_put(&syn_cache_pool, sc); 3700 3701 out: 3702 KERNEL_UNLOCK_ONE(NULL); 3703 mutex_exit(softnet_lock); 3704 } 3705 3706 /* 3707 * Remove syn cache created by the specified tcb entry, 3708 * because this does not make sense to keep them 3709 * (if there's no tcb entry, syn cache entry will never be used) 3710 */ 3711 void 3712 syn_cache_cleanup(struct tcpcb *tp) 3713 { 3714 struct syn_cache *sc, *nsc; 3715 int s; 3716 3717 s = splsoftnet(); 3718 3719 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3720 nsc = LIST_NEXT(sc, sc_tpq); 3721 3722 #ifdef DIAGNOSTIC 3723 if (sc->sc_tp != tp) 3724 panic("invalid sc_tp in syn_cache_cleanup"); 3725 #endif 3726 syn_cache_rm(sc); 3727 syn_cache_put(sc); /* calls pool_put but see spl above */ 3728 } 3729 /* just for safety */ 3730 LIST_INIT(&tp->t_sc); 3731 3732 splx(s); 3733 } 3734 3735 /* 3736 * Find an entry in the syn cache. 3737 */ 3738 struct syn_cache * 3739 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3740 struct syn_cache_head **headp) 3741 { 3742 struct syn_cache *sc; 3743 struct syn_cache_head *scp; 3744 u_int32_t hash; 3745 int s; 3746 3747 SYN_HASHALL(hash, src, dst); 3748 3749 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3750 *headp = scp; 3751 s = splsoftnet(); 3752 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3753 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3754 if (sc->sc_hash != hash) 3755 continue; 3756 if (!memcmp(&sc->sc_src, src, src->sa_len) && 3757 !memcmp(&sc->sc_dst, dst, dst->sa_len)) { 3758 splx(s); 3759 return (sc); 3760 } 3761 } 3762 splx(s); 3763 return (NULL); 3764 } 3765 3766 /* 3767 * This function gets called when we receive an ACK for a socket in the 3768 * LISTEN state. We look up the connection in the syn cache, and if it's 3769 * there, we pull it out of the cache and turn it into a full-blown 3770 * connection in the SYN-RECEIVED state. 3771 * 3772 * The return values may not be immediately obvious, and their effects 3773 * can be subtle, so here they are: 3774 * 3775 * NULL SYN was not found in cache; caller should drop the 3776 * packet and send an RST. 3777 * 3778 * -1 We were unable to create the new connection, and are 3779 * aborting it. An ACK,RST is being sent to the peer 3780 * (unless we got screwey sequence numbers; see below), 3781 * because the 3-way handshake has been completed. Caller 3782 * should not free the mbuf, since we may be using it. If 3783 * we are not, we will free it. 3784 * 3785 * Otherwise, the return value is a pointer to the new socket 3786 * associated with the connection. 3787 */ 3788 struct socket * 3789 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3790 struct tcphdr *th, struct socket *so, struct mbuf *m) 3791 { 3792 struct syn_cache *sc; 3793 struct syn_cache_head *scp; 3794 struct inpcb *inp = NULL; 3795 #ifdef INET6 3796 struct in6pcb *in6p = NULL; 3797 #endif 3798 struct tcpcb *tp; 3799 int s; 3800 struct socket *oso; 3801 3802 s = splsoftnet(); 3803 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3804 splx(s); 3805 return NULL; 3806 } 3807 3808 /* 3809 * Verify the sequence and ack numbers. Try getting the correct 3810 * response again. 3811 */ 3812 if ((th->th_ack != sc->sc_iss + 1) || 3813 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3814 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3815 m_freem(m); 3816 (void)syn_cache_respond(sc); 3817 splx(s); 3818 return ((struct socket *)(-1)); 3819 } 3820 3821 /* Remove this cache entry */ 3822 syn_cache_rm(sc); 3823 splx(s); 3824 3825 /* 3826 * Ok, create the full blown connection, and set things up 3827 * as they would have been set up if we had created the 3828 * connection when the SYN arrived. If we can't create 3829 * the connection, abort it. 3830 */ 3831 /* 3832 * inp still has the OLD in_pcb stuff, set the 3833 * v6-related flags on the new guy, too. This is 3834 * done particularly for the case where an AF_INET6 3835 * socket is bound only to a port, and a v4 connection 3836 * comes in on that port. 3837 * we also copy the flowinfo from the original pcb 3838 * to the new one. 3839 */ 3840 oso = so; 3841 so = sonewconn(so, true); 3842 if (so == NULL) 3843 goto resetandabort; 3844 3845 switch (so->so_proto->pr_domain->dom_family) { 3846 case AF_INET: 3847 inp = sotoinpcb(so); 3848 break; 3849 #ifdef INET6 3850 case AF_INET6: 3851 in6p = sotoin6pcb(so); 3852 break; 3853 #endif 3854 } 3855 3856 switch (src->sa_family) { 3857 case AF_INET: 3858 if (inp) { 3859 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3860 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3861 inp->inp_options = ip_srcroute(m); 3862 in_pcbstate(inp, INP_BOUND); 3863 if (inp->inp_options == NULL) { 3864 inp->inp_options = sc->sc_ipopts; 3865 sc->sc_ipopts = NULL; 3866 } 3867 } 3868 #ifdef INET6 3869 else if (in6p) { 3870 /* IPv4 packet to AF_INET6 socket */ 3871 memset(&in6p->in6p_laddr, 0, sizeof(in6p->in6p_laddr)); 3872 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3873 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3874 &in6p->in6p_laddr.s6_addr32[3], 3875 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3876 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3877 in6totcpcb(in6p)->t_family = AF_INET; 3878 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3879 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3880 else 3881 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3882 in6_pcbstate(in6p, IN6P_BOUND); 3883 } 3884 #endif 3885 break; 3886 #ifdef INET6 3887 case AF_INET6: 3888 if (in6p) { 3889 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3890 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3891 in6_pcbstate(in6p, IN6P_BOUND); 3892 } 3893 break; 3894 #endif 3895 } 3896 3897 #ifdef INET6 3898 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3899 struct in6pcb *oin6p = sotoin6pcb(oso); 3900 /* inherit socket options from the listening socket */ 3901 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3902 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3903 m_freem(in6p->in6p_options); 3904 in6p->in6p_options = NULL; 3905 } 3906 ip6_savecontrol(in6p, &in6p->in6p_options, 3907 mtod(m, struct ip6_hdr *), m); 3908 } 3909 #endif 3910 3911 #if defined(IPSEC) 3912 if (ipsec_used) { 3913 /* 3914 * we make a copy of policy, instead of sharing the policy, for 3915 * better behavior in terms of SA lookup and dead SA removal. 3916 */ 3917 if (inp) { 3918 /* copy old policy into new socket's */ 3919 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, 3920 inp->inp_sp)) 3921 printf("tcp_input: could not copy policy\n"); 3922 } 3923 #ifdef INET6 3924 else if (in6p) { 3925 /* copy old policy into new socket's */ 3926 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3927 in6p->in6p_sp)) 3928 printf("tcp_input: could not copy policy\n"); 3929 } 3930 #endif 3931 } 3932 #endif 3933 3934 /* 3935 * Give the new socket our cached route reference. 3936 */ 3937 if (inp) { 3938 rtcache_copy(&inp->inp_route, &sc->sc_route); 3939 rtcache_free(&sc->sc_route); 3940 } 3941 #ifdef INET6 3942 else { 3943 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 3944 rtcache_free(&sc->sc_route); 3945 } 3946 #endif 3947 3948 if (inp) { 3949 struct sockaddr_in sin; 3950 memcpy(&sin, src, src->sa_len); 3951 if (in_pcbconnect(inp, &sin, &lwp0)) { 3952 goto resetandabort; 3953 } 3954 } 3955 #ifdef INET6 3956 else if (in6p) { 3957 struct sockaddr_in6 sin6; 3958 memcpy(&sin6, src, src->sa_len); 3959 if (src->sa_family == AF_INET) { 3960 /* IPv4 packet to AF_INET6 socket */ 3961 in6_sin_2_v4mapsin6((struct sockaddr_in *)src, &sin6); 3962 } 3963 if (in6_pcbconnect(in6p, &sin6, NULL)) { 3964 goto resetandabort; 3965 } 3966 } 3967 #endif 3968 else { 3969 goto resetandabort; 3970 } 3971 3972 if (inp) 3973 tp = intotcpcb(inp); 3974 #ifdef INET6 3975 else if (in6p) 3976 tp = in6totcpcb(in6p); 3977 #endif 3978 else 3979 tp = NULL; 3980 3981 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3982 if (sc->sc_request_r_scale != 15) { 3983 tp->requested_s_scale = sc->sc_requested_s_scale; 3984 tp->request_r_scale = sc->sc_request_r_scale; 3985 tp->snd_scale = sc->sc_requested_s_scale; 3986 tp->rcv_scale = sc->sc_request_r_scale; 3987 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3988 } 3989 if (sc->sc_flags & SCF_TIMESTAMP) 3990 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3991 tp->ts_timebase = sc->sc_timebase; 3992 3993 tp->t_template = tcp_template(tp); 3994 if (tp->t_template == 0) { 3995 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3996 so = NULL; 3997 m_freem(m); 3998 goto abort; 3999 } 4000 4001 tp->iss = sc->sc_iss; 4002 tp->irs = sc->sc_irs; 4003 tcp_sendseqinit(tp); 4004 tcp_rcvseqinit(tp); 4005 tp->t_state = TCPS_SYN_RECEIVED; 4006 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 4007 TCP_STATINC(TCP_STAT_ACCEPTS); 4008 4009 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 4010 tp->t_flags |= TF_WILL_SACK; 4011 4012 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 4013 tp->t_flags |= TF_ECN_PERMIT; 4014 4015 #ifdef TCP_SIGNATURE 4016 if (sc->sc_flags & SCF_SIGNATURE) 4017 tp->t_flags |= TF_SIGNATURE; 4018 #endif 4019 4020 /* Initialize tp->t_ourmss before we deal with the peer's! */ 4021 tp->t_ourmss = sc->sc_ourmaxseg; 4022 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 4023 4024 /* 4025 * Initialize the initial congestion window. If we 4026 * had to retransmit the SYN,ACK, we must initialize cwnd 4027 * to 1 segment (i.e. the Loss Window). 4028 */ 4029 if (sc->sc_rxtshift) 4030 tp->snd_cwnd = tp->t_peermss; 4031 else { 4032 int ss = tcp_init_win; 4033 if (inp != NULL && in_localaddr(inp->inp_faddr)) 4034 ss = tcp_init_win_local; 4035 #ifdef INET6 4036 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 4037 ss = tcp_init_win_local; 4038 #endif 4039 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 4040 } 4041 4042 tcp_rmx_rtt(tp); 4043 tp->snd_wl1 = sc->sc_irs; 4044 tp->rcv_up = sc->sc_irs + 1; 4045 4046 /* 4047 * This is what whould have happened in tcp_output() when 4048 * the SYN,ACK was sent. 4049 */ 4050 tp->snd_up = tp->snd_una; 4051 tp->snd_max = tp->snd_nxt = tp->iss+1; 4052 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 4053 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 4054 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 4055 tp->last_ack_sent = tp->rcv_nxt; 4056 tp->t_partialacks = -1; 4057 tp->t_dupacks = 0; 4058 4059 TCP_STATINC(TCP_STAT_SC_COMPLETED); 4060 s = splsoftnet(); 4061 syn_cache_put(sc); 4062 splx(s); 4063 return so; 4064 4065 resetandabort: 4066 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 4067 abort: 4068 if (so != NULL) { 4069 (void) soqremque(so, 1); 4070 (void) soabort(so); 4071 mutex_enter(softnet_lock); 4072 } 4073 s = splsoftnet(); 4074 syn_cache_put(sc); 4075 splx(s); 4076 TCP_STATINC(TCP_STAT_SC_ABORTED); 4077 return ((struct socket *)(-1)); 4078 } 4079 4080 /* 4081 * This function is called when we get a RST for a 4082 * non-existent connection, so that we can see if the 4083 * connection is in the syn cache. If it is, zap it. 4084 */ 4085 4086 void 4087 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 4088 { 4089 struct syn_cache *sc; 4090 struct syn_cache_head *scp; 4091 int s = splsoftnet(); 4092 4093 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4094 splx(s); 4095 return; 4096 } 4097 if (SEQ_LT(th->th_seq, sc->sc_irs) || 4098 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 4099 splx(s); 4100 return; 4101 } 4102 syn_cache_rm(sc); 4103 TCP_STATINC(TCP_STAT_SC_RESET); 4104 syn_cache_put(sc); /* calls pool_put but see spl above */ 4105 splx(s); 4106 } 4107 4108 void 4109 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 4110 struct tcphdr *th) 4111 { 4112 struct syn_cache *sc; 4113 struct syn_cache_head *scp; 4114 int s; 4115 4116 s = splsoftnet(); 4117 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4118 splx(s); 4119 return; 4120 } 4121 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 4122 if (ntohl(th->th_seq) != sc->sc_iss) { 4123 splx(s); 4124 return; 4125 } 4126 4127 /* 4128 * If we've retransmitted 3 times and this is our second error, 4129 * we remove the entry. Otherwise, we allow it to continue on. 4130 * This prevents us from incorrectly nuking an entry during a 4131 * spurious network outage. 4132 * 4133 * See tcp_notify(). 4134 */ 4135 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 4136 sc->sc_flags |= SCF_UNREACH; 4137 splx(s); 4138 return; 4139 } 4140 4141 syn_cache_rm(sc); 4142 TCP_STATINC(TCP_STAT_SC_UNREACH); 4143 syn_cache_put(sc); /* calls pool_put but see spl above */ 4144 splx(s); 4145 } 4146 4147 /* 4148 * Given a LISTEN socket and an inbound SYN request, add this to the syn 4149 * cache, and send back a segment: 4150 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 4151 * to the source. 4152 * 4153 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 4154 * Doing so would require that we hold onto the data and deliver it 4155 * to the application. However, if we are the target of a SYN-flood 4156 * DoS attack, an attacker could send data which would eventually 4157 * consume all available buffer space if it were ACKed. By not ACKing 4158 * the data, we avoid this DoS scenario. 4159 */ 4160 int 4161 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 4162 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 4163 int optlen, struct tcp_opt_info *oi) 4164 { 4165 struct tcpcb tb, *tp; 4166 long win; 4167 struct syn_cache *sc; 4168 struct syn_cache_head *scp; 4169 struct mbuf *ipopts; 4170 int s; 4171 4172 tp = sototcpcb(so); 4173 4174 /* 4175 * Initialize some local state. 4176 */ 4177 win = sbspace(&so->so_rcv); 4178 if (win > TCP_MAXWIN) 4179 win = TCP_MAXWIN; 4180 4181 #ifdef TCP_SIGNATURE 4182 if (optp || (tp->t_flags & TF_SIGNATURE)) 4183 #else 4184 if (optp) 4185 #endif 4186 { 4187 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4188 #ifdef TCP_SIGNATURE 4189 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4190 #endif 4191 tb.t_state = TCPS_LISTEN; 4192 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 4193 sizeof(struct tcphdr) - optlen - hlen, 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 10:09:38 GMT 2025