tcp_input.c revision 1.399
1 /* $NetBSD: tcp_input.c,v 1.399 2018/03/29 17:46:17 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 1.399 2018/03/29 17:46:17 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 /* 448 * Insert segment ti into reassembly queue of tcp with 449 * control block tp. Return TH_FIN if reassembly now includes 450 * a segment with FIN. 451 */ 452 static int 453 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int tlen) 454 { 455 struct ipqent *p, *q, *nq, *tiqe = NULL; 456 struct socket *so = NULL; 457 int pkt_flags; 458 tcp_seq pkt_seq; 459 unsigned pkt_len; 460 u_long rcvpartdupbyte = 0; 461 u_long rcvoobyte; 462 #ifdef TCP_REASS_COUNTERS 463 u_int count = 0; 464 #endif 465 uint64_t *tcps; 466 467 if (tp->t_inpcb) 468 so = tp->t_inpcb->inp_socket; 469 #ifdef INET6 470 else if (tp->t_in6pcb) 471 so = tp->t_in6pcb->in6p_socket; 472 #endif 473 474 TCP_REASS_LOCK_CHECK(tp); 475 476 /* 477 * Call with th==NULL after become established to 478 * force pre-ESTABLISHED data up to user socket. 479 */ 480 if (th == NULL) 481 goto present; 482 483 m_claimm(m, &tcp_reass_mowner); 484 485 rcvoobyte = tlen; 486 /* 487 * Copy these to local variables because the TCP header gets munged 488 * while we are collapsing mbufs. 489 */ 490 pkt_seq = th->th_seq; 491 pkt_len = tlen; 492 pkt_flags = th->th_flags; 493 494 TCP_REASS_COUNTER_INCR(&tcp_reass_); 495 496 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 497 /* 498 * When we miss a packet, the vast majority of time we get 499 * packets that follow it in order. So optimize for that. 500 */ 501 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 502 p->ipqe_len += pkt_len; 503 p->ipqe_flags |= pkt_flags; 504 m_cat(p->ipre_mlast, m); 505 TRAVERSE(p->ipre_mlast); 506 m = NULL; 507 tiqe = p; 508 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 509 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 510 goto skip_replacement; 511 } 512 /* 513 * While we're here, if the pkt is completely beyond 514 * anything we have, just insert it at the tail. 515 */ 516 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 517 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 518 goto insert_it; 519 } 520 } 521 522 q = TAILQ_FIRST(&tp->segq); 523 524 if (q != NULL) { 525 /* 526 * If this segment immediately precedes the first out-of-order 527 * block, simply slap the segment in front of it and (mostly) 528 * skip the complicated logic. 529 */ 530 if (pkt_seq + pkt_len == q->ipqe_seq) { 531 q->ipqe_seq = pkt_seq; 532 q->ipqe_len += pkt_len; 533 q->ipqe_flags |= pkt_flags; 534 m_cat(m, q->ipqe_m); 535 q->ipqe_m = m; 536 q->ipre_mlast = m; /* last mbuf may have changed */ 537 TRAVERSE(q->ipre_mlast); 538 tiqe = q; 539 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 540 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 541 goto skip_replacement; 542 } 543 } else { 544 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 545 } 546 547 /* 548 * Find a segment which begins after this one does. 549 */ 550 for (p = NULL; q != NULL; q = nq) { 551 nq = TAILQ_NEXT(q, ipqe_q); 552 #ifdef TCP_REASS_COUNTERS 553 count++; 554 #endif 555 556 /* 557 * If the received segment is just right after this 558 * fragment, merge the two together and then check 559 * for further overlaps. 560 */ 561 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 562 #ifdef TCPREASS_DEBUG 563 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 564 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 565 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 566 #endif 567 pkt_len += q->ipqe_len; 568 pkt_flags |= q->ipqe_flags; 569 pkt_seq = q->ipqe_seq; 570 m_cat(q->ipre_mlast, m); 571 TRAVERSE(q->ipre_mlast); 572 m = q->ipqe_m; 573 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 574 goto free_ipqe; 575 } 576 577 /* 578 * If the received segment is completely past this 579 * fragment, we need to go to the next fragment. 580 */ 581 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 582 p = q; 583 continue; 584 } 585 586 /* 587 * If the fragment is past the received segment, 588 * it (or any following) can't be concatenated. 589 */ 590 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 591 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 592 break; 593 } 594 595 /* 596 * We've received all the data in this segment before. 597 * Mark it as a duplicate and return. 598 */ 599 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 600 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 601 tcps = TCP_STAT_GETREF(); 602 tcps[TCP_STAT_RCVDUPPACK]++; 603 tcps[TCP_STAT_RCVDUPBYTE] += pkt_len; 604 TCP_STAT_PUTREF(); 605 tcp_new_dsack(tp, pkt_seq, pkt_len); 606 m_freem(m); 607 if (tiqe != NULL) { 608 tcpipqent_free(tiqe); 609 } 610 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 611 goto out; 612 } 613 614 /* 615 * Received segment completely overlaps this fragment 616 * so we drop the fragment (this keeps the temporal 617 * ordering of segments correct). 618 */ 619 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 620 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 621 rcvpartdupbyte += q->ipqe_len; 622 m_freem(q->ipqe_m); 623 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 624 goto free_ipqe; 625 } 626 627 /* 628 * Received segment extends past the end of the fragment. 629 * Drop the overlapping bytes, merge the fragment and 630 * segment, and treat as a longer received packet. 631 */ 632 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 633 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 634 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 635 #ifdef TCPREASS_DEBUG 636 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 637 tp, overlap, 638 pkt_seq, pkt_seq + pkt_len, pkt_len); 639 #endif 640 m_adj(m, overlap); 641 rcvpartdupbyte += overlap; 642 m_cat(q->ipre_mlast, m); 643 TRAVERSE(q->ipre_mlast); 644 m = q->ipqe_m; 645 pkt_seq = q->ipqe_seq; 646 pkt_len += q->ipqe_len - overlap; 647 rcvoobyte -= overlap; 648 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 649 goto free_ipqe; 650 } 651 652 /* 653 * Received segment extends past the front of the fragment. 654 * Drop the overlapping bytes on the received packet. The 655 * packet will then be concatenated with this fragment a 656 * bit later. 657 */ 658 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 659 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 660 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 661 #ifdef TCPREASS_DEBUG 662 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 663 tp, overlap, 664 pkt_seq, pkt_seq + pkt_len, pkt_len); 665 #endif 666 m_adj(m, -overlap); 667 pkt_len -= overlap; 668 rcvpartdupbyte += overlap; 669 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 670 rcvoobyte -= overlap; 671 } 672 673 /* 674 * If the received segment immediately precedes this 675 * fragment then tack the fragment onto this segment 676 * and reinsert the data. 677 */ 678 if (q->ipqe_seq == pkt_seq + pkt_len) { 679 #ifdef TCPREASS_DEBUG 680 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 681 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 682 pkt_seq, pkt_seq + pkt_len, pkt_len); 683 #endif 684 pkt_len += q->ipqe_len; 685 pkt_flags |= q->ipqe_flags; 686 m_cat(m, q->ipqe_m); 687 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 688 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 689 tp->t_segqlen--; 690 KASSERT(tp->t_segqlen >= 0); 691 KASSERT(tp->t_segqlen != 0 || 692 (TAILQ_EMPTY(&tp->segq) && 693 TAILQ_EMPTY(&tp->timeq))); 694 if (tiqe == NULL) { 695 tiqe = q; 696 } else { 697 tcpipqent_free(q); 698 } 699 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 700 break; 701 } 702 703 /* 704 * If the fragment is before the segment, remember it. 705 * When this loop is terminated, p will contain the 706 * pointer to the fragment that is right before the 707 * received segment. 708 */ 709 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 710 p = q; 711 712 continue; 713 714 /* 715 * This is a common operation. It also will allow 716 * to save doing a malloc/free in most instances. 717 */ 718 free_ipqe: 719 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 720 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 721 tp->t_segqlen--; 722 KASSERT(tp->t_segqlen >= 0); 723 KASSERT(tp->t_segqlen != 0 || 724 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 725 if (tiqe == NULL) { 726 tiqe = q; 727 } else { 728 tcpipqent_free(q); 729 } 730 } 731 732 #ifdef TCP_REASS_COUNTERS 733 if (count > 7) 734 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 735 else if (count > 0) 736 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 737 #endif 738 739 insert_it: 740 /* 741 * Allocate a new queue entry (block) since the received segment 742 * did not collapse onto any other out-of-order block. If it had 743 * collapsed, tiqe would not be NULL and we would be reusing it. 744 * 745 * If the allocation fails, drop the packet. 746 */ 747 if (tiqe == NULL) { 748 tiqe = tcpipqent_alloc(); 749 if (tiqe == NULL) { 750 TCP_STATINC(TCP_STAT_RCVMEMDROP); 751 m_freem(m); 752 goto out; 753 } 754 } 755 756 /* 757 * Update the counters. 758 */ 759 tp->t_rcvoopack++; 760 tcps = TCP_STAT_GETREF(); 761 tcps[TCP_STAT_RCVOOPACK]++; 762 tcps[TCP_STAT_RCVOOBYTE] += rcvoobyte; 763 if (rcvpartdupbyte) { 764 tcps[TCP_STAT_RCVPARTDUPPACK]++; 765 tcps[TCP_STAT_RCVPARTDUPBYTE] += rcvpartdupbyte; 766 } 767 TCP_STAT_PUTREF(); 768 769 /* 770 * Insert the new fragment queue entry into both queues. 771 */ 772 tiqe->ipqe_m = m; 773 tiqe->ipre_mlast = m; 774 tiqe->ipqe_seq = pkt_seq; 775 tiqe->ipqe_len = pkt_len; 776 tiqe->ipqe_flags = pkt_flags; 777 if (p == NULL) { 778 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 779 #ifdef TCPREASS_DEBUG 780 if (tiqe->ipqe_seq != tp->rcv_nxt) 781 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 782 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 783 #endif 784 } else { 785 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 786 #ifdef TCPREASS_DEBUG 787 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 788 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 789 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 790 #endif 791 } 792 tp->t_segqlen++; 793 794 skip_replacement: 795 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 796 797 present: 798 /* 799 * Present data to user, advancing rcv_nxt through 800 * completed sequence space. 801 */ 802 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 803 goto out; 804 q = TAILQ_FIRST(&tp->segq); 805 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 806 goto out; 807 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 808 goto out; 809 810 tp->rcv_nxt += q->ipqe_len; 811 pkt_flags = q->ipqe_flags & TH_FIN; 812 nd6_hint(tp); 813 814 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 815 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 816 tp->t_segqlen--; 817 KASSERT(tp->t_segqlen >= 0); 818 KASSERT(tp->t_segqlen != 0 || 819 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 820 if (so->so_state & SS_CANTRCVMORE) 821 m_freem(q->ipqe_m); 822 else 823 sbappendstream(&so->so_rcv, q->ipqe_m); 824 tcpipqent_free(q); 825 TCP_REASS_UNLOCK(tp); 826 sorwakeup(so); 827 return pkt_flags; 828 829 out: 830 TCP_REASS_UNLOCK(tp); 831 return 0; 832 } 833 834 #ifdef INET6 835 int 836 tcp6_input(struct mbuf **mp, int *offp, int proto) 837 { 838 struct mbuf *m = *mp; 839 840 /* 841 * draft-itojun-ipv6-tcp-to-anycast 842 * better place to put this in? 843 */ 844 if (m->m_flags & M_ANYCAST6) { 845 struct ip6_hdr *ip6; 846 if (m->m_len < sizeof(struct ip6_hdr)) { 847 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 848 TCP_STATINC(TCP_STAT_RCVSHORT); 849 return IPPROTO_DONE; 850 } 851 } 852 ip6 = mtod(m, struct ip6_hdr *); 853 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 854 (char *)&ip6->ip6_dst - (char *)ip6); 855 return IPPROTO_DONE; 856 } 857 858 tcp_input(m, *offp, proto); 859 return IPPROTO_DONE; 860 } 861 #endif 862 863 static void 864 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 865 { 866 char src[INET_ADDRSTRLEN]; 867 char dst[INET_ADDRSTRLEN]; 868 869 if (ip) { 870 in_print(src, sizeof(src), &ip->ip_src); 871 in_print(dst, sizeof(dst), &ip->ip_dst); 872 } else { 873 strlcpy(src, "(unknown)", sizeof(src)); 874 strlcpy(dst, "(unknown)", sizeof(dst)); 875 } 876 log(LOG_INFO, 877 "Connection attempt to TCP %s:%d from %s:%d\n", 878 dst, ntohs(th->th_dport), 879 src, ntohs(th->th_sport)); 880 } 881 882 #ifdef INET6 883 static void 884 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 885 { 886 char src[INET6_ADDRSTRLEN]; 887 char dst[INET6_ADDRSTRLEN]; 888 889 if (ip6) { 890 in6_print(src, sizeof(src), &ip6->ip6_src); 891 in6_print(dst, sizeof(dst), &ip6->ip6_dst); 892 } else { 893 strlcpy(src, "(unknown v6)", sizeof(src)); 894 strlcpy(dst, "(unknown v6)", sizeof(dst)); 895 } 896 log(LOG_INFO, 897 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 898 dst, ntohs(th->th_dport), 899 src, ntohs(th->th_sport)); 900 } 901 #endif 902 903 /* 904 * Checksum extended TCP header and data. 905 */ 906 int 907 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, 908 int toff, int off, int tlen) 909 { 910 struct ifnet *rcvif; 911 int s; 912 913 /* 914 * XXX it's better to record and check if this mbuf is 915 * already checked. 916 */ 917 918 rcvif = m_get_rcvif(m, &s); 919 if (__predict_false(rcvif == NULL)) 920 goto badcsum; /* XXX */ 921 922 switch (af) { 923 case AF_INET: 924 switch (m->m_pkthdr.csum_flags & 925 ((rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 926 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 927 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 928 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 929 goto badcsum; 930 931 case M_CSUM_TCPv4|M_CSUM_DATA: { 932 u_int32_t hw_csum = m->m_pkthdr.csum_data; 933 934 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 935 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 936 const struct ip *ip = 937 mtod(m, const struct ip *); 938 939 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 940 ip->ip_dst.s_addr, 941 htons(hw_csum + tlen + off + IPPROTO_TCP)); 942 } 943 if ((hw_csum ^ 0xffff) != 0) 944 goto badcsum; 945 break; 946 } 947 948 case M_CSUM_TCPv4: 949 /* Checksum was okay. */ 950 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 951 break; 952 953 default: 954 /* 955 * Must compute it ourselves. Maybe skip checksum 956 * on loopback interfaces. 957 */ 958 if (__predict_true(!(rcvif->if_flags & IFF_LOOPBACK) || 959 tcp_do_loopback_cksum)) { 960 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 961 if (in4_cksum(m, IPPROTO_TCP, toff, 962 tlen + off) != 0) 963 goto badcsum; 964 } 965 break; 966 } 967 break; 968 969 #ifdef INET6 970 case AF_INET6: 971 switch (m->m_pkthdr.csum_flags & 972 ((rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 973 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 974 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 975 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 976 goto badcsum; 977 978 #if 0 /* notyet */ 979 case M_CSUM_TCPv6|M_CSUM_DATA: 980 #endif 981 982 case M_CSUM_TCPv6: 983 /* Checksum was okay. */ 984 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 985 break; 986 987 default: 988 /* 989 * Must compute it ourselves. Maybe skip checksum 990 * on loopback interfaces. 991 */ 992 if (__predict_true((m->m_flags & M_LOOP) == 0 || 993 tcp_do_loopback_cksum)) { 994 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 995 if (in6_cksum(m, IPPROTO_TCP, toff, 996 tlen + off) != 0) 997 goto badcsum; 998 } 999 } 1000 break; 1001 #endif /* INET6 */ 1002 } 1003 m_put_rcvif(rcvif, &s); 1004 1005 return 0; 1006 1007 badcsum: 1008 m_put_rcvif(rcvif, &s); 1009 TCP_STATINC(TCP_STAT_RCVBADSUM); 1010 return -1; 1011 } 1012 1013 /* 1014 * When a packet arrives addressed to a vestigial tcpbp, we 1015 * nevertheless have to respond to it per the spec. 1016 * 1017 * This code is duplicated from the one in tcp_input(). 1018 */ 1019 static void tcp_vtw_input(struct tcphdr *th, vestigial_inpcb_t *vp, 1020 struct mbuf *m, int tlen) 1021 { 1022 int tiflags; 1023 int todrop; 1024 uint32_t t_flags = 0; 1025 uint64_t *tcps; 1026 1027 tiflags = th->th_flags; 1028 todrop = vp->rcv_nxt - th->th_seq; 1029 1030 if (todrop > 0) { 1031 if (tiflags & TH_SYN) { 1032 tiflags &= ~TH_SYN; 1033 th->th_seq++; 1034 tcp_urp_drop(th, 1, &tiflags); 1035 todrop--; 1036 } 1037 if (todrop > tlen || 1038 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1039 /* 1040 * Any valid FIN or RST must be to the left of the 1041 * window. At this point the FIN or RST must be a 1042 * duplicate or out of sequence; drop it. 1043 */ 1044 if (tiflags & TH_RST) 1045 goto drop; 1046 tiflags &= ~(TH_FIN|TH_RST); 1047 1048 /* 1049 * Send an ACK to resynchronize and drop any data. 1050 * But keep on processing for RST or ACK. 1051 */ 1052 t_flags |= TF_ACKNOW; 1053 todrop = tlen; 1054 tcps = TCP_STAT_GETREF(); 1055 tcps[TCP_STAT_RCVDUPPACK] += 1; 1056 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 1057 TCP_STAT_PUTREF(); 1058 } else if ((tiflags & TH_RST) && 1059 th->th_seq != vp->rcv_nxt) { 1060 /* 1061 * Test for reset before adjusting the sequence 1062 * number for overlapping data. 1063 */ 1064 goto dropafterack_ratelim; 1065 } else { 1066 tcps = TCP_STAT_GETREF(); 1067 tcps[TCP_STAT_RCVPARTDUPPACK] += 1; 1068 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 1069 TCP_STAT_PUTREF(); 1070 } 1071 1072 // tcp_new_dsack(tp, th->th_seq, todrop); 1073 // hdroptlen += todrop; /*drop from head afterwards*/ 1074 1075 th->th_seq += todrop; 1076 tlen -= todrop; 1077 tcp_urp_drop(th, todrop, &tiflags); 1078 } 1079 1080 /* 1081 * If new data are received on a connection after the 1082 * user processes are gone, then RST the other end. 1083 */ 1084 if (tlen) { 1085 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 1086 goto dropwithreset; 1087 } 1088 1089 /* 1090 * If segment ends after window, drop trailing data 1091 * (and PUSH and FIN); if nothing left, just ACK. 1092 */ 1093 todrop = (th->th_seq + tlen) - (vp->rcv_nxt + vp->rcv_wnd); 1094 1095 if (todrop > 0) { 1096 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 1097 if (todrop >= tlen) { 1098 /* 1099 * The segment actually starts after the window. 1100 * th->th_seq + tlen - vp->rcv_nxt - vp->rcv_wnd >= tlen 1101 * th->th_seq - vp->rcv_nxt - vp->rcv_wnd >= 0 1102 * th->th_seq >= vp->rcv_nxt + vp->rcv_wnd 1103 */ 1104 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 1105 1106 /* 1107 * If a new connection request is received 1108 * while in TIME_WAIT, drop the old connection 1109 * and start over if the sequence numbers 1110 * are above the previous ones. 1111 */ 1112 if ((tiflags & TH_SYN) && 1113 SEQ_GT(th->th_seq, vp->rcv_nxt)) { 1114 /* 1115 * We only support this in the !NOFDREF case, which 1116 * is to say: not here. 1117 */ 1118 goto dropwithreset; 1119 } 1120 1121 /* 1122 * If window is closed can only take segments at 1123 * window edge, and have to drop data and PUSH from 1124 * incoming segments. Continue processing, but 1125 * remember to ack. Otherwise, drop segment 1126 * and (if not RST) ack. 1127 */ 1128 if (vp->rcv_wnd == 0 && th->th_seq == vp->rcv_nxt) { 1129 t_flags |= TF_ACKNOW; 1130 TCP_STATINC(TCP_STAT_RCVWINPROBE); 1131 } else { 1132 goto dropafterack; 1133 } 1134 } else { 1135 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 1136 } 1137 m_adj(m, -todrop); 1138 tlen -= todrop; 1139 tiflags &= ~(TH_PUSH|TH_FIN); 1140 } 1141 1142 if (tiflags & TH_RST) { 1143 if (th->th_seq != vp->rcv_nxt) 1144 goto dropafterack_ratelim; 1145 1146 vtw_del(vp->ctl, vp->vtw); 1147 goto drop; 1148 } 1149 1150 /* 1151 * If the ACK bit is off we drop the segment and return. 1152 */ 1153 if ((tiflags & TH_ACK) == 0) { 1154 if (t_flags & TF_ACKNOW) 1155 goto dropafterack; 1156 goto drop; 1157 } 1158 1159 /* 1160 * In TIME_WAIT state the only thing that should arrive 1161 * is a retransmission of the remote FIN. Acknowledge 1162 * it and restart the finack timer. 1163 */ 1164 vtw_restart(vp); 1165 goto dropafterack; 1166 1167 dropafterack: 1168 /* 1169 * Generate an ACK dropping incoming segment if it occupies 1170 * sequence space, where the ACK reflects our state. 1171 */ 1172 if (tiflags & TH_RST) 1173 goto drop; 1174 goto dropafterack2; 1175 1176 dropafterack_ratelim: 1177 /* 1178 * We may want to rate-limit ACKs against SYN/RST attack. 1179 */ 1180 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 1181 tcp_ackdrop_ppslim) == 0) { 1182 /* XXX stat */ 1183 goto drop; 1184 } 1185 /* ...fall into dropafterack2... */ 1186 1187 dropafterack2: 1188 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, th->th_ack, TH_ACK); 1189 return; 1190 1191 dropwithreset: 1192 /* 1193 * Generate a RST, dropping incoming segment. 1194 * Make ACK acceptable to originator of segment. 1195 */ 1196 if (tiflags & TH_RST) 1197 goto drop; 1198 1199 if (tiflags & TH_ACK) { 1200 tcp_respond(0, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 1201 } else { 1202 if (tiflags & TH_SYN) 1203 ++tlen; 1204 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, (tcp_seq)0, 1205 TH_RST|TH_ACK); 1206 } 1207 return; 1208 drop: 1209 m_freem(m); 1210 } 1211 1212 /* 1213 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 1214 */ 1215 void 1216 tcp_input(struct mbuf *m, ...) 1217 { 1218 struct tcphdr *th; 1219 struct ip *ip; 1220 struct inpcb *inp; 1221 #ifdef INET6 1222 struct ip6_hdr *ip6; 1223 struct in6pcb *in6p; 1224 #endif 1225 u_int8_t *optp = NULL; 1226 int optlen = 0; 1227 int len, tlen, toff, hdroptlen = 0; 1228 struct tcpcb *tp = NULL; 1229 int tiflags; 1230 struct socket *so = NULL; 1231 int todrop, acked, ourfinisacked, needoutput = 0; 1232 bool dupseg; 1233 #ifdef TCP_DEBUG 1234 short ostate = 0; 1235 #endif 1236 u_long tiwin; 1237 struct tcp_opt_info opti; 1238 int off, iphlen; 1239 va_list ap; 1240 int af; /* af on the wire */ 1241 struct mbuf *tcp_saveti = NULL; 1242 uint32_t ts_rtt; 1243 uint8_t iptos; 1244 uint64_t *tcps; 1245 vestigial_inpcb_t vestige; 1246 1247 vestige.valid = 0; 1248 1249 MCLAIM(m, &tcp_rx_mowner); 1250 va_start(ap, m); 1251 toff = va_arg(ap, int); 1252 (void)va_arg(ap, int); /* ignore value, advance ap */ 1253 va_end(ap); 1254 1255 TCP_STATINC(TCP_STAT_RCVTOTAL); 1256 1257 memset(&opti, 0, sizeof(opti)); 1258 opti.ts_present = 0; 1259 opti.maxseg = 0; 1260 1261 /* 1262 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 1263 * 1264 * TCP is, by definition, unicast, so we reject all 1265 * multicast outright. 1266 * 1267 * Note, there are additional src/dst address checks in 1268 * the AF-specific code below. 1269 */ 1270 if (m->m_flags & (M_BCAST|M_MCAST)) { 1271 /* XXX stat */ 1272 goto drop; 1273 } 1274 #ifdef INET6 1275 if (m->m_flags & M_ANYCAST6) { 1276 /* XXX stat */ 1277 goto drop; 1278 } 1279 #endif 1280 1281 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, sizeof(struct tcphdr)); 1282 if (th == NULL) { 1283 TCP_STATINC(TCP_STAT_RCVSHORT); 1284 return; 1285 } 1286 1287 /* 1288 * Get IP and TCP header. 1289 * Note: IP leaves IP header in first mbuf. 1290 */ 1291 ip = mtod(m, struct ip *); 1292 switch (ip->ip_v) { 1293 case 4: 1294 #ifdef INET6 1295 ip6 = NULL; 1296 #endif 1297 af = AF_INET; 1298 iphlen = sizeof(struct ip); 1299 1300 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1301 in_broadcast(ip->ip_dst, m_get_rcvif_NOMPSAFE(m))) 1302 goto drop; 1303 1304 /* We do the checksum after PCB lookup... */ 1305 len = ntohs(ip->ip_len); 1306 tlen = len - toff; 1307 iptos = ip->ip_tos; 1308 break; 1309 #ifdef INET6 1310 case 6: 1311 ip = NULL; 1312 iphlen = sizeof(struct ip6_hdr); 1313 af = AF_INET6; 1314 ip6 = mtod(m, struct ip6_hdr *); 1315 1316 /* 1317 * Be proactive about unspecified IPv6 address in source. 1318 * As we use all-zero to indicate unbounded/unconnected pcb, 1319 * unspecified IPv6 address can be used to confuse us. 1320 * 1321 * Note that packets with unspecified IPv6 destination is 1322 * already dropped in ip6_input. 1323 */ 1324 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1325 /* XXX stat */ 1326 goto drop; 1327 } 1328 1329 /* 1330 * Make sure destination address is not multicast. 1331 * Source address checked in ip6_input(). 1332 */ 1333 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1334 /* XXX stat */ 1335 goto drop; 1336 } 1337 1338 /* We do the checksum after PCB lookup... */ 1339 len = m->m_pkthdr.len; 1340 tlen = len - toff; 1341 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1342 break; 1343 #endif 1344 default: 1345 m_freem(m); 1346 return; 1347 } 1348 1349 /* 1350 * Enforce alignment requirements that are violated in 1351 * some cases, see kern/50766 for details. 1352 */ 1353 if (TCP_HDR_ALIGNED_P(th) == 0) { 1354 m = m_copyup(m, toff + sizeof(struct tcphdr), 0); 1355 if (m == NULL) { 1356 TCP_STATINC(TCP_STAT_RCVSHORT); 1357 return; 1358 } 1359 ip = mtod(m, struct ip *); 1360 #ifdef INET6 1361 ip6 = mtod(m, struct ip6_hdr *); 1362 #endif 1363 th = (struct tcphdr *)(mtod(m, char *) + toff); 1364 } 1365 KASSERT(TCP_HDR_ALIGNED_P(th)); 1366 1367 /* 1368 * Check that TCP offset makes sense, pull out TCP options and 1369 * adjust length. 1370 */ 1371 off = th->th_off << 2; 1372 if (off < sizeof(struct tcphdr) || off > tlen) { 1373 TCP_STATINC(TCP_STAT_RCVBADOFF); 1374 goto drop; 1375 } 1376 tlen -= off; 1377 1378 if (off > sizeof(struct tcphdr)) { 1379 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1380 if (th == NULL) { 1381 TCP_STATINC(TCP_STAT_RCVSHORT); 1382 return; 1383 } 1384 KASSERT(TCP_HDR_ALIGNED_P(th)); 1385 optlen = off - sizeof(struct tcphdr); 1386 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1387 1388 /* 1389 * Do quick retrieval of timestamp options. 1390 * 1391 * If timestamp is the only option and it's formatted as 1392 * recommended in RFC 1323 appendix A, we quickly get the 1393 * values now and don't bother calling tcp_dooptions(), 1394 * etc. 1395 */ 1396 if ((optlen == TCPOLEN_TSTAMP_APPA || 1397 (optlen > TCPOLEN_TSTAMP_APPA && 1398 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1399 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1400 (th->th_flags & TH_SYN) == 0) { 1401 opti.ts_present = 1; 1402 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1403 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1404 optp = NULL; /* we've parsed the options */ 1405 } 1406 } 1407 tiflags = th->th_flags; 1408 1409 /* 1410 * Checksum extended TCP header and data 1411 */ 1412 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1413 goto badcsum; 1414 1415 /* 1416 * Locate pcb for segment. 1417 */ 1418 findpcb: 1419 inp = NULL; 1420 #ifdef INET6 1421 in6p = NULL; 1422 #endif 1423 switch (af) { 1424 case AF_INET: 1425 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1426 ip->ip_dst, th->th_dport, &vestige); 1427 if (inp == NULL && !vestige.valid) { 1428 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1429 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, 1430 th->th_dport); 1431 } 1432 #ifdef INET6 1433 if (inp == NULL && !vestige.valid) { 1434 struct in6_addr s, d; 1435 1436 /* mapped addr case */ 1437 in6_in_2_v4mapin6(&ip->ip_src, &s); 1438 in6_in_2_v4mapin6(&ip->ip_dst, &d); 1439 in6p = in6_pcblookup_connect(&tcbtable, &s, 1440 th->th_sport, &d, th->th_dport, 0, &vestige); 1441 if (in6p == 0 && !vestige.valid) { 1442 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1443 in6p = in6_pcblookup_bind(&tcbtable, &d, 1444 th->th_dport, 0); 1445 } 1446 } 1447 #endif 1448 #ifndef INET6 1449 if (inp == NULL && !vestige.valid) 1450 #else 1451 if (inp == NULL && in6p == NULL && !vestige.valid) 1452 #endif 1453 { 1454 TCP_STATINC(TCP_STAT_NOPORT); 1455 if (tcp_log_refused && 1456 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1457 tcp4_log_refused(ip, th); 1458 } 1459 tcp_fields_to_host(th); 1460 goto dropwithreset_ratelim; 1461 } 1462 #if defined(IPSEC) 1463 if (ipsec_used) { 1464 if (inp && ipsec_in_reject(m, inp)) { 1465 goto drop; 1466 } 1467 #ifdef INET6 1468 else if (in6p && ipsec_in_reject(m, in6p)) { 1469 goto drop; 1470 } 1471 #endif 1472 } 1473 #endif /*IPSEC*/ 1474 break; 1475 #ifdef INET6 1476 case AF_INET6: 1477 { 1478 int faith; 1479 1480 #if defined(NFAITH) && NFAITH > 0 1481 faith = faithprefix(&ip6->ip6_dst); 1482 #else 1483 faith = 0; 1484 #endif 1485 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1486 th->th_sport, &ip6->ip6_dst, th->th_dport, faith, &vestige); 1487 if (!in6p && !vestige.valid) { 1488 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1489 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1490 th->th_dport, faith); 1491 } 1492 if (!in6p && !vestige.valid) { 1493 TCP_STATINC(TCP_STAT_NOPORT); 1494 if (tcp_log_refused && 1495 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1496 tcp6_log_refused(ip6, th); 1497 } 1498 tcp_fields_to_host(th); 1499 goto dropwithreset_ratelim; 1500 } 1501 #if defined(IPSEC) 1502 if (ipsec_used && in6p && ipsec_in_reject(m, in6p)) { 1503 goto drop; 1504 } 1505 #endif 1506 break; 1507 } 1508 #endif 1509 } 1510 1511 tcp_fields_to_host(th); 1512 1513 /* 1514 * If the state is CLOSED (i.e., TCB does not exist) then 1515 * all data in the incoming segment is discarded. 1516 * If the TCB exists but is in CLOSED state, it is embryonic, 1517 * but should either do a listen or a connect soon. 1518 */ 1519 tp = NULL; 1520 so = NULL; 1521 if (inp) { 1522 /* Check the minimum TTL for socket. */ 1523 if (ip->ip_ttl < inp->inp_ip_minttl) 1524 goto drop; 1525 1526 tp = intotcpcb(inp); 1527 so = inp->inp_socket; 1528 } 1529 #ifdef INET6 1530 else if (in6p) { 1531 tp = in6totcpcb(in6p); 1532 so = in6p->in6p_socket; 1533 } 1534 #endif 1535 else if (vestige.valid) { 1536 /* We do not support the resurrection of vtw tcpcps. */ 1537 tcp_vtw_input(th, &vestige, m, tlen); 1538 m = NULL; 1539 goto drop; 1540 } 1541 1542 if (tp == NULL) 1543 goto dropwithreset_ratelim; 1544 if (tp->t_state == TCPS_CLOSED) 1545 goto drop; 1546 1547 KASSERT(so->so_lock == softnet_lock); 1548 KASSERT(solocked(so)); 1549 1550 /* Unscale the window into a 32-bit value. */ 1551 if ((tiflags & TH_SYN) == 0) 1552 tiwin = th->th_win << tp->snd_scale; 1553 else 1554 tiwin = th->th_win; 1555 1556 #ifdef INET6 1557 /* save packet options if user wanted */ 1558 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1559 if (in6p->in6p_options) { 1560 m_freem(in6p->in6p_options); 1561 in6p->in6p_options = NULL; 1562 } 1563 KASSERT(ip6 != NULL); 1564 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1565 } 1566 #endif 1567 1568 if (so->so_options & SO_DEBUG) { 1569 #ifdef TCP_DEBUG 1570 ostate = tp->t_state; 1571 #endif 1572 1573 tcp_saveti = NULL; 1574 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1575 goto nosave; 1576 1577 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1578 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1579 if (tcp_saveti == NULL) 1580 goto nosave; 1581 } else { 1582 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1583 if (tcp_saveti == NULL) 1584 goto nosave; 1585 MCLAIM(m, &tcp_mowner); 1586 tcp_saveti->m_len = iphlen; 1587 m_copydata(m, 0, iphlen, 1588 mtod(tcp_saveti, void *)); 1589 } 1590 1591 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1592 m_freem(tcp_saveti); 1593 tcp_saveti = NULL; 1594 } else { 1595 tcp_saveti->m_len += sizeof(struct tcphdr); 1596 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1597 sizeof(struct tcphdr)); 1598 } 1599 nosave:; 1600 } 1601 1602 if (so->so_options & SO_ACCEPTCONN) { 1603 union syn_cache_sa src; 1604 union syn_cache_sa dst; 1605 1606 KASSERT(tp->t_state == TCPS_LISTEN); 1607 1608 memset(&src, 0, sizeof(src)); 1609 memset(&dst, 0, sizeof(dst)); 1610 switch (af) { 1611 case AF_INET: 1612 src.sin.sin_len = sizeof(struct sockaddr_in); 1613 src.sin.sin_family = AF_INET; 1614 src.sin.sin_addr = ip->ip_src; 1615 src.sin.sin_port = th->th_sport; 1616 1617 dst.sin.sin_len = sizeof(struct sockaddr_in); 1618 dst.sin.sin_family = AF_INET; 1619 dst.sin.sin_addr = ip->ip_dst; 1620 dst.sin.sin_port = th->th_dport; 1621 break; 1622 #ifdef INET6 1623 case AF_INET6: 1624 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1625 src.sin6.sin6_family = AF_INET6; 1626 src.sin6.sin6_addr = ip6->ip6_src; 1627 src.sin6.sin6_port = th->th_sport; 1628 1629 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1630 dst.sin6.sin6_family = AF_INET6; 1631 dst.sin6.sin6_addr = ip6->ip6_dst; 1632 dst.sin6.sin6_port = th->th_dport; 1633 break; 1634 #endif 1635 } 1636 1637 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1638 if (tiflags & TH_RST) { 1639 syn_cache_reset(&src.sa, &dst.sa, th); 1640 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1641 (TH_ACK|TH_SYN)) { 1642 /* 1643 * Received a SYN,ACK. This should never 1644 * happen while we are in LISTEN. Send an RST. 1645 */ 1646 goto badsyn; 1647 } else if (tiflags & TH_ACK) { 1648 so = syn_cache_get(&src.sa, &dst.sa, th, so, m); 1649 if (so == NULL) { 1650 /* 1651 * We don't have a SYN for this ACK; 1652 * send an RST. 1653 */ 1654 goto badsyn; 1655 } else if (so == (struct socket *)(-1)) { 1656 /* 1657 * We were unable to create the 1658 * connection. If the 3-way handshake 1659 * was completed, and RST has been 1660 * sent to the peer. Since the mbuf 1661 * might be in use for the reply, do 1662 * not free it. 1663 */ 1664 m = NULL; 1665 } else { 1666 /* 1667 * We have created a full-blown 1668 * connection. 1669 */ 1670 tp = NULL; 1671 inp = NULL; 1672 #ifdef INET6 1673 in6p = NULL; 1674 #endif 1675 switch (so->so_proto->pr_domain->dom_family) { 1676 case AF_INET: 1677 inp = sotoinpcb(so); 1678 tp = intotcpcb(inp); 1679 break; 1680 #ifdef INET6 1681 case AF_INET6: 1682 in6p = sotoin6pcb(so); 1683 tp = in6totcpcb(in6p); 1684 break; 1685 #endif 1686 } 1687 if (tp == NULL) 1688 goto badsyn; /*XXX*/ 1689 tiwin <<= tp->snd_scale; 1690 goto after_listen; 1691 } 1692 } else { 1693 /* 1694 * None of RST, SYN or ACK was set. 1695 * This is an invalid packet for a 1696 * TCB in LISTEN state. Send a RST. 1697 */ 1698 goto badsyn; 1699 } 1700 } else { 1701 /* 1702 * Received a SYN. 1703 */ 1704 1705 #ifdef INET6 1706 /* 1707 * If deprecated address is forbidden, we do 1708 * not accept SYN to deprecated interface 1709 * address to prevent any new inbound 1710 * connection from getting established. 1711 * When we do not accept SYN, we send a TCP 1712 * RST, with deprecated source address (instead 1713 * of dropping it). We compromise it as it is 1714 * much better for peer to send a RST, and 1715 * RST will be the final packet for the 1716 * exchange. 1717 * 1718 * If we do not forbid deprecated addresses, we 1719 * accept the SYN packet. RFC2462 does not 1720 * suggest dropping SYN in this case. 1721 * If we decipher RFC2462 5.5.4, it says like 1722 * this: 1723 * 1. use of deprecated addr with existing 1724 * communication is okay - "SHOULD continue 1725 * to be used" 1726 * 2. use of it with new communication: 1727 * (2a) "SHOULD NOT be used if alternate 1728 * address with sufficient scope is 1729 * available" 1730 * (2b) nothing mentioned otherwise. 1731 * Here we fall into (2b) case as we have no 1732 * choice in our source address selection - we 1733 * must obey the peer. 1734 * 1735 * The wording in RFC2462 is confusing, and 1736 * there are multiple description text for 1737 * deprecated address handling - worse, they 1738 * are not exactly the same. I believe 5.5.4 1739 * is the best one, so we follow 5.5.4. 1740 */ 1741 if (af == AF_INET6 && !ip6_use_deprecated) { 1742 struct in6_ifaddr *ia6; 1743 int s; 1744 struct ifnet *rcvif = m_get_rcvif(m, &s); 1745 if (rcvif == NULL) 1746 goto dropwithreset; /* XXX */ 1747 if ((ia6 = in6ifa_ifpwithaddr(rcvif, 1748 &ip6->ip6_dst)) && 1749 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1750 tp = NULL; 1751 m_put_rcvif(rcvif, &s); 1752 goto dropwithreset; 1753 } 1754 m_put_rcvif(rcvif, &s); 1755 } 1756 #endif 1757 1758 /* 1759 * LISTEN socket received a SYN from itself? This 1760 * can't possibly be valid; drop the packet. 1761 */ 1762 if (th->th_sport == th->th_dport) { 1763 int eq = 0; 1764 1765 switch (af) { 1766 case AF_INET: 1767 eq = in_hosteq(ip->ip_src, ip->ip_dst); 1768 break; 1769 #ifdef INET6 1770 case AF_INET6: 1771 eq = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, 1772 &ip6->ip6_dst); 1773 break; 1774 #endif 1775 } 1776 if (eq) { 1777 TCP_STATINC(TCP_STAT_BADSYN); 1778 goto drop; 1779 } 1780 } 1781 1782 /* 1783 * SYN looks ok; create compressed TCP 1784 * state for it. 1785 */ 1786 if (so->so_qlen <= so->so_qlimit && 1787 syn_cache_add(&src.sa, &dst.sa, th, toff, 1788 so, m, optp, optlen, &opti)) 1789 m = NULL; 1790 } 1791 1792 goto drop; 1793 } 1794 1795 after_listen: 1796 /* 1797 * From here on, we're dealing with !LISTEN. 1798 */ 1799 KASSERT(tp->t_state != TCPS_LISTEN); 1800 1801 /* 1802 * Segment received on connection. 1803 * Reset idle time and keep-alive timer. 1804 */ 1805 tp->t_rcvtime = tcp_now; 1806 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1807 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1808 1809 /* 1810 * Process options. 1811 */ 1812 #ifdef TCP_SIGNATURE 1813 if (optp || (tp->t_flags & TF_SIGNATURE)) 1814 #else 1815 if (optp) 1816 #endif 1817 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1818 goto drop; 1819 1820 if (TCP_SACK_ENABLED(tp)) { 1821 tcp_del_sackholes(tp, th); 1822 } 1823 1824 if (TCP_ECN_ALLOWED(tp)) { 1825 if (tiflags & TH_CWR) { 1826 tp->t_flags &= ~TF_ECN_SND_ECE; 1827 } 1828 switch (iptos & IPTOS_ECN_MASK) { 1829 case IPTOS_ECN_CE: 1830 tp->t_flags |= TF_ECN_SND_ECE; 1831 TCP_STATINC(TCP_STAT_ECN_CE); 1832 break; 1833 case IPTOS_ECN_ECT0: 1834 TCP_STATINC(TCP_STAT_ECN_ECT); 1835 break; 1836 case IPTOS_ECN_ECT1: 1837 /* XXX: ignore for now -- rpaulo */ 1838 break; 1839 } 1840 /* 1841 * Congestion experienced. 1842 * Ignore if we are already trying to recover. 1843 */ 1844 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1845 tp->t_congctl->cong_exp(tp); 1846 } 1847 1848 if (opti.ts_present && opti.ts_ecr) { 1849 /* 1850 * Calculate the RTT from the returned time stamp and the 1851 * connection's time base. If the time stamp is later than 1852 * the current time, or is extremely old, fall back to non-1323 1853 * RTT calculation. Since ts_rtt is unsigned, we can test both 1854 * at the same time. 1855 * 1856 * Note that ts_rtt is in units of slow ticks (500 1857 * ms). Since most earthbound RTTs are < 500 ms, 1858 * observed values will have large quantization noise. 1859 * Our smoothed RTT is then the fraction of observed 1860 * samples that are 1 tick instead of 0 (times 500 1861 * ms). 1862 * 1863 * ts_rtt is increased by 1 to denote a valid sample, 1864 * with 0 indicating an invalid measurement. This 1865 * extra 1 must be removed when ts_rtt is used, or 1866 * else an an erroneous extra 500 ms will result. 1867 */ 1868 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1869 if (ts_rtt > TCP_PAWS_IDLE) 1870 ts_rtt = 0; 1871 } else { 1872 ts_rtt = 0; 1873 } 1874 1875 /* 1876 * Fast path: check for the two common cases of a uni-directional 1877 * data transfer. If: 1878 * o We are in the ESTABLISHED state, and 1879 * o The packet has no control flags, and 1880 * o The packet is in-sequence, and 1881 * o The window didn't change, and 1882 * o We are not retransmitting 1883 * It's a candidate. 1884 * 1885 * If the length (tlen) is zero and the ack moved forward, we're 1886 * the sender side of the transfer. Just free the data acked and 1887 * wake any higher level process that was blocked waiting for 1888 * space. 1889 * 1890 * If the length is non-zero and the ack didn't move, we're the 1891 * receiver side. If we're getting packets in-order (the reassembly 1892 * queue is empty), add the data to the socket buffer and note 1893 * that we need a delayed ack. 1894 */ 1895 if (tp->t_state == TCPS_ESTABLISHED && 1896 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1897 == TH_ACK && 1898 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1899 th->th_seq == tp->rcv_nxt && 1900 tiwin && tiwin == tp->snd_wnd && 1901 tp->snd_nxt == tp->snd_max) { 1902 1903 /* 1904 * If last ACK falls within this segment's sequence numbers, 1905 * record the timestamp. 1906 * NOTE that the test is modified according to the latest 1907 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 1908 * 1909 * note that we already know 1910 * TSTMP_GEQ(opti.ts_val, tp->ts_recent) 1911 */ 1912 if (opti.ts_present && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1913 tp->ts_recent_age = tcp_now; 1914 tp->ts_recent = opti.ts_val; 1915 } 1916 1917 if (tlen == 0) { 1918 /* Ack prediction. */ 1919 if (SEQ_GT(th->th_ack, tp->snd_una) && 1920 SEQ_LEQ(th->th_ack, tp->snd_max) && 1921 tp->snd_cwnd >= tp->snd_wnd && 1922 tp->t_partialacks < 0) { 1923 /* 1924 * this is a pure ack for outstanding data. 1925 */ 1926 if (ts_rtt) 1927 tcp_xmit_timer(tp, ts_rtt - 1); 1928 else if (tp->t_rtttime && 1929 SEQ_GT(th->th_ack, tp->t_rtseq)) 1930 tcp_xmit_timer(tp, 1931 tcp_now - tp->t_rtttime); 1932 acked = th->th_ack - tp->snd_una; 1933 tcps = TCP_STAT_GETREF(); 1934 tcps[TCP_STAT_PREDACK]++; 1935 tcps[TCP_STAT_RCVACKPACK]++; 1936 tcps[TCP_STAT_RCVACKBYTE] += acked; 1937 TCP_STAT_PUTREF(); 1938 nd6_hint(tp); 1939 1940 if (acked > (tp->t_lastoff - tp->t_inoff)) 1941 tp->t_lastm = NULL; 1942 sbdrop(&so->so_snd, acked); 1943 tp->t_lastoff -= acked; 1944 1945 icmp_check(tp, th, acked); 1946 1947 tp->snd_una = th->th_ack; 1948 tp->snd_fack = tp->snd_una; 1949 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1950 tp->snd_high = tp->snd_una; 1951 m_freem(m); 1952 1953 /* 1954 * If all outstanding data are acked, stop 1955 * retransmit timer, otherwise restart timer 1956 * using current (possibly backed-off) value. 1957 * If process is waiting for space, 1958 * wakeup/selnotify/signal. If data 1959 * are ready to send, let tcp_output 1960 * decide between more output or persist. 1961 */ 1962 if (tp->snd_una == tp->snd_max) 1963 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1964 else if (TCP_TIMER_ISARMED(tp, 1965 TCPT_PERSIST) == 0) 1966 TCP_TIMER_ARM(tp, TCPT_REXMT, 1967 tp->t_rxtcur); 1968 1969 sowwakeup(so); 1970 if (so->so_snd.sb_cc) { 1971 KERNEL_LOCK(1, NULL); 1972 (void)tcp_output(tp); 1973 KERNEL_UNLOCK_ONE(NULL); 1974 } 1975 if (tcp_saveti) 1976 m_freem(tcp_saveti); 1977 return; 1978 } 1979 } else if (th->th_ack == tp->snd_una && 1980 TAILQ_FIRST(&tp->segq) == NULL && 1981 tlen <= sbspace(&so->so_rcv)) { 1982 int newsize = 0; 1983 1984 /* 1985 * this is a pure, in-sequence data packet 1986 * with nothing on the reassembly queue and 1987 * we have enough buffer space to take it. 1988 */ 1989 tp->rcv_nxt += tlen; 1990 tcps = TCP_STAT_GETREF(); 1991 tcps[TCP_STAT_PREDDAT]++; 1992 tcps[TCP_STAT_RCVPACK]++; 1993 tcps[TCP_STAT_RCVBYTE] += tlen; 1994 TCP_STAT_PUTREF(); 1995 nd6_hint(tp); 1996 1997 /* 1998 * Automatic sizing enables the performance of large buffers 1999 * and most of the efficiency of small ones by only allocating 2000 * space when it is needed. 2001 * 2002 * On the receive side the socket buffer memory is only rarely 2003 * used to any significant extent. This allows us to be much 2004 * more aggressive in scaling the receive socket buffer. For 2005 * the case that the buffer space is actually used to a large 2006 * extent and we run out of kernel memory we can simply drop 2007 * the new segments; TCP on the sender will just retransmit it 2008 * later. Setting the buffer size too big may only consume too 2009 * much kernel memory if the application doesn't read() from 2010 * the socket or packet loss or reordering makes use of the 2011 * reassembly queue. 2012 * 2013 * The criteria to step up the receive buffer one notch are: 2014 * 1. the number of bytes received during the time it takes 2015 * one timestamp to be reflected back to us (the RTT); 2016 * 2. received bytes per RTT is within seven eighth of the 2017 * current socket buffer size; 2018 * 3. receive buffer size has not hit maximal automatic size; 2019 * 2020 * This algorithm does one step per RTT at most and only if 2021 * we receive a bulk stream w/o packet losses or reorderings. 2022 * Shrinking the buffer during idle times is not necessary as 2023 * it doesn't consume any memory when idle. 2024 * 2025 * TODO: Only step up if the application is actually serving 2026 * the buffer to better manage the socket buffer resources. 2027 */ 2028 if (tcp_do_autorcvbuf && 2029 opti.ts_ecr && 2030 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 2031 if (opti.ts_ecr > tp->rfbuf_ts && 2032 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 2033 if (tp->rfbuf_cnt > 2034 (so->so_rcv.sb_hiwat / 8 * 7) && 2035 so->so_rcv.sb_hiwat < 2036 tcp_autorcvbuf_max) { 2037 newsize = 2038 min(so->so_rcv.sb_hiwat + 2039 tcp_autorcvbuf_inc, 2040 tcp_autorcvbuf_max); 2041 } 2042 /* Start over with next RTT. */ 2043 tp->rfbuf_ts = 0; 2044 tp->rfbuf_cnt = 0; 2045 } else 2046 tp->rfbuf_cnt += tlen; /* add up */ 2047 } 2048 2049 /* 2050 * Drop TCP, IP headers and TCP options then add data 2051 * to socket buffer. 2052 */ 2053 if (so->so_state & SS_CANTRCVMORE) { 2054 m_freem(m); 2055 } else { 2056 /* 2057 * Set new socket buffer size. 2058 * Give up when limit is reached. 2059 */ 2060 if (newsize) 2061 if (!sbreserve(&so->so_rcv, 2062 newsize, so)) 2063 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 2064 m_adj(m, toff + off); 2065 sbappendstream(&so->so_rcv, m); 2066 } 2067 sorwakeup(so); 2068 tcp_setup_ack(tp, th); 2069 if (tp->t_flags & TF_ACKNOW) { 2070 KERNEL_LOCK(1, NULL); 2071 (void)tcp_output(tp); 2072 KERNEL_UNLOCK_ONE(NULL); 2073 } 2074 if (tcp_saveti) 2075 m_freem(tcp_saveti); 2076 return; 2077 } 2078 } 2079 2080 /* 2081 * Compute mbuf offset to TCP data segment. 2082 */ 2083 hdroptlen = toff + off; 2084 2085 /* 2086 * Calculate amount of space in receive window. Receive window is 2087 * amount of space in rcv queue, but not less than advertised 2088 * window. 2089 */ 2090 { 2091 int win; 2092 win = sbspace(&so->so_rcv); 2093 if (win < 0) 2094 win = 0; 2095 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 2096 } 2097 2098 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 2099 tp->rfbuf_ts = 0; 2100 tp->rfbuf_cnt = 0; 2101 2102 switch (tp->t_state) { 2103 /* 2104 * If the state is SYN_SENT: 2105 * if seg contains an ACK, but not for our SYN, drop the input. 2106 * if seg contains a RST, then drop the connection. 2107 * if seg does not contain SYN, then drop it. 2108 * Otherwise this is an acceptable SYN segment 2109 * initialize tp->rcv_nxt and tp->irs 2110 * if seg contains ack then advance tp->snd_una 2111 * if seg contains a ECE and ECN support is enabled, the stream 2112 * is ECN capable. 2113 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2114 * arrange for segment to be acked (eventually) 2115 * continue processing rest of data/controls, beginning with URG 2116 */ 2117 case TCPS_SYN_SENT: 2118 if ((tiflags & TH_ACK) && 2119 (SEQ_LEQ(th->th_ack, tp->iss) || 2120 SEQ_GT(th->th_ack, tp->snd_max))) 2121 goto dropwithreset; 2122 if (tiflags & TH_RST) { 2123 if (tiflags & TH_ACK) 2124 tp = tcp_drop(tp, ECONNREFUSED); 2125 goto drop; 2126 } 2127 if ((tiflags & TH_SYN) == 0) 2128 goto drop; 2129 if (tiflags & TH_ACK) { 2130 tp->snd_una = th->th_ack; 2131 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2132 tp->snd_nxt = tp->snd_una; 2133 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2134 tp->snd_high = tp->snd_una; 2135 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2136 2137 if ((tiflags & TH_ECE) && tcp_do_ecn) { 2138 tp->t_flags |= TF_ECN_PERMIT; 2139 TCP_STATINC(TCP_STAT_ECN_SHS); 2140 } 2141 } 2142 tp->irs = th->th_seq; 2143 tcp_rcvseqinit(tp); 2144 tp->t_flags |= TF_ACKNOW; 2145 tcp_mss_from_peer(tp, opti.maxseg); 2146 2147 /* 2148 * Initialize the initial congestion window. If we 2149 * had to retransmit the SYN, we must initialize cwnd 2150 * to 1 segment (i.e. the Loss Window). 2151 */ 2152 if (tp->t_flags & TF_SYN_REXMT) 2153 tp->snd_cwnd = tp->t_peermss; 2154 else { 2155 int ss = tcp_init_win; 2156 if (inp != NULL && in_localaddr(inp->inp_faddr)) 2157 ss = tcp_init_win_local; 2158 #ifdef INET6 2159 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 2160 ss = tcp_init_win_local; 2161 #endif 2162 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 2163 } 2164 2165 tcp_rmx_rtt(tp); 2166 if (tiflags & TH_ACK) { 2167 TCP_STATINC(TCP_STAT_CONNECTS); 2168 /* 2169 * move tcp_established before soisconnected 2170 * because upcall handler can drive tcp_output 2171 * functionality. 2172 * XXX we might call soisconnected at the end of 2173 * all processing 2174 */ 2175 tcp_established(tp); 2176 soisconnected(so); 2177 /* Do window scaling on this connection? */ 2178 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2179 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2180 tp->snd_scale = tp->requested_s_scale; 2181 tp->rcv_scale = tp->request_r_scale; 2182 } 2183 TCP_REASS_LOCK(tp); 2184 (void)tcp_reass(tp, NULL, NULL, tlen); 2185 /* 2186 * if we didn't have to retransmit the SYN, 2187 * use its rtt as our initial srtt & rtt var. 2188 */ 2189 if (tp->t_rtttime) 2190 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2191 } else { 2192 tp->t_state = TCPS_SYN_RECEIVED; 2193 } 2194 2195 /* 2196 * Advance th->th_seq to correspond to first data byte. 2197 * If data, trim to stay within window, 2198 * dropping FIN if necessary. 2199 */ 2200 th->th_seq++; 2201 if (tlen > tp->rcv_wnd) { 2202 todrop = tlen - tp->rcv_wnd; 2203 m_adj(m, -todrop); 2204 tlen = tp->rcv_wnd; 2205 tiflags &= ~TH_FIN; 2206 tcps = TCP_STAT_GETREF(); 2207 tcps[TCP_STAT_RCVPACKAFTERWIN]++; 2208 tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop; 2209 TCP_STAT_PUTREF(); 2210 } 2211 tp->snd_wl1 = th->th_seq - 1; 2212 tp->rcv_up = th->th_seq; 2213 goto step6; 2214 2215 /* 2216 * If the state is SYN_RECEIVED: 2217 * If seg contains an ACK, but not for our SYN, drop the input 2218 * and generate an RST. See page 36, rfc793 2219 */ 2220 case TCPS_SYN_RECEIVED: 2221 if ((tiflags & TH_ACK) && 2222 (SEQ_LEQ(th->th_ack, tp->iss) || 2223 SEQ_GT(th->th_ack, tp->snd_max))) 2224 goto dropwithreset; 2225 break; 2226 } 2227 2228 /* 2229 * From here on, we're dealing with !LISTEN and !SYN_SENT. 2230 */ 2231 KASSERT(tp->t_state != TCPS_LISTEN && 2232 tp->t_state != TCPS_SYN_SENT); 2233 2234 /* 2235 * RFC1323 PAWS: if we have a timestamp reply on this segment and 2236 * it's less than ts_recent, drop it. 2237 */ 2238 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2239 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2240 /* Check to see if ts_recent is over 24 days old. */ 2241 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2242 /* 2243 * Invalidate ts_recent. If this segment updates 2244 * ts_recent, the age will be reset later and ts_recent 2245 * will get a valid value. If it does not, setting 2246 * ts_recent to zero will at least satisfy the 2247 * requirement that zero be placed in the timestamp 2248 * echo reply when ts_recent isn't valid. The 2249 * age isn't reset until we get a valid ts_recent 2250 * because we don't want out-of-order segments to be 2251 * dropped when ts_recent is old. 2252 */ 2253 tp->ts_recent = 0; 2254 } else { 2255 tcps = TCP_STAT_GETREF(); 2256 tcps[TCP_STAT_RCVDUPPACK]++; 2257 tcps[TCP_STAT_RCVDUPBYTE] += tlen; 2258 tcps[TCP_STAT_PAWSDROP]++; 2259 TCP_STAT_PUTREF(); 2260 tcp_new_dsack(tp, th->th_seq, tlen); 2261 goto dropafterack; 2262 } 2263 } 2264 2265 /* 2266 * Check that at least some bytes of the segment are within the 2267 * receive window. If segment begins before rcv_nxt, drop leading 2268 * data (and SYN); if nothing left, just ack. 2269 */ 2270 todrop = tp->rcv_nxt - th->th_seq; 2271 dupseg = false; 2272 if (todrop > 0) { 2273 if (tiflags & TH_SYN) { 2274 tiflags &= ~TH_SYN; 2275 th->th_seq++; 2276 tcp_urp_drop(th, 1, &tiflags); 2277 todrop--; 2278 } 2279 if (todrop > tlen || 2280 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2281 /* 2282 * Any valid FIN or RST must be to the left of the 2283 * window. At this point the FIN or RST must be a 2284 * duplicate or out of sequence; drop it. 2285 */ 2286 if (tiflags & TH_RST) 2287 goto drop; 2288 tiflags &= ~(TH_FIN|TH_RST); 2289 2290 /* 2291 * Send an ACK to resynchronize and drop any data. 2292 * But keep on processing for RST or ACK. 2293 */ 2294 tp->t_flags |= TF_ACKNOW; 2295 todrop = tlen; 2296 dupseg = true; 2297 tcps = TCP_STAT_GETREF(); 2298 tcps[TCP_STAT_RCVDUPPACK]++; 2299 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 2300 TCP_STAT_PUTREF(); 2301 } else if ((tiflags & TH_RST) && th->th_seq != tp->rcv_nxt) { 2302 /* 2303 * Test for reset before adjusting the sequence 2304 * number for overlapping data. 2305 */ 2306 goto dropafterack_ratelim; 2307 } else { 2308 tcps = TCP_STAT_GETREF(); 2309 tcps[TCP_STAT_RCVPARTDUPPACK]++; 2310 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 2311 TCP_STAT_PUTREF(); 2312 } 2313 tcp_new_dsack(tp, th->th_seq, todrop); 2314 hdroptlen += todrop; /* drop from head afterwards (m_adj) */ 2315 th->th_seq += todrop; 2316 tlen -= todrop; 2317 tcp_urp_drop(th, todrop, &tiflags); 2318 } 2319 2320 /* 2321 * If new data is received on a connection after the user processes 2322 * are gone, then RST the other end. 2323 */ 2324 if ((so->so_state & SS_NOFDREF) && 2325 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2326 tp = tcp_close(tp); 2327 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 2328 goto dropwithreset; 2329 } 2330 2331 /* 2332 * If the segment ends after the window, drop trailing data (and 2333 * PUSH and FIN); if nothing left, just ACK. 2334 */ 2335 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2336 if (todrop > 0) { 2337 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 2338 if (todrop >= tlen) { 2339 /* 2340 * The segment actually starts after the window. 2341 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2342 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2343 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2344 */ 2345 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 2346 2347 /* 2348 * If a new connection request is received while in 2349 * TIME_WAIT, drop the old connection and start over 2350 * if the sequence numbers are above the previous 2351 * ones. 2352 * 2353 * NOTE: We need to put the header fields back into 2354 * network order. 2355 */ 2356 if ((tiflags & TH_SYN) && 2357 tp->t_state == TCPS_TIME_WAIT && 2358 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2359 tp = tcp_close(tp); 2360 tcp_fields_to_net(th); 2361 m_freem(tcp_saveti); 2362 tcp_saveti = NULL; 2363 goto findpcb; 2364 } 2365 2366 /* 2367 * If window is closed can only take segments at 2368 * window edge, and have to drop data and PUSH from 2369 * incoming segments. Continue processing, but 2370 * remember to ack. Otherwise, drop segment 2371 * and (if not RST) ack. 2372 */ 2373 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2374 KASSERT(todrop == tlen); 2375 tp->t_flags |= TF_ACKNOW; 2376 TCP_STATINC(TCP_STAT_RCVWINPROBE); 2377 } else { 2378 goto dropafterack; 2379 } 2380 } else { 2381 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 2382 } 2383 m_adj(m, -todrop); 2384 tlen -= todrop; 2385 tiflags &= ~(TH_PUSH|TH_FIN); 2386 } 2387 2388 /* 2389 * If last ACK falls within this segment's sequence numbers, 2390 * record the timestamp. 2391 * NOTE: 2392 * 1) That the test incorporates suggestions from the latest 2393 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 2394 * 2) That updating only on newer timestamps interferes with 2395 * our earlier PAWS tests, so this check should be solely 2396 * predicated on the sequence space of this segment. 2397 * 3) That we modify the segment boundary check to be 2398 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2399 * instead of RFC1323's 2400 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2401 * This modified check allows us to overcome RFC1323's 2402 * limitations as described in Stevens TCP/IP Illustrated 2403 * Vol. 2 p.869. In such cases, we can still calculate the 2404 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2405 */ 2406 if (opti.ts_present && 2407 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2408 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2409 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2410 tp->ts_recent_age = tcp_now; 2411 tp->ts_recent = opti.ts_val; 2412 } 2413 2414 /* 2415 * If the RST bit is set examine the state: 2416 * RECEIVED state: 2417 * If passive open, return to LISTEN state. 2418 * If active open, inform user that connection was refused. 2419 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT states: 2420 * Inform user that connection was reset, and close tcb. 2421 * CLOSING, LAST_ACK, TIME_WAIT states: 2422 * Close the tcb. 2423 */ 2424 if (tiflags & TH_RST) { 2425 if (th->th_seq != tp->rcv_nxt) 2426 goto dropafterack_ratelim; 2427 2428 switch (tp->t_state) { 2429 case TCPS_SYN_RECEIVED: 2430 so->so_error = ECONNREFUSED; 2431 goto close; 2432 2433 case TCPS_ESTABLISHED: 2434 case TCPS_FIN_WAIT_1: 2435 case TCPS_FIN_WAIT_2: 2436 case TCPS_CLOSE_WAIT: 2437 so->so_error = ECONNRESET; 2438 close: 2439 tp->t_state = TCPS_CLOSED; 2440 TCP_STATINC(TCP_STAT_DROPS); 2441 tp = tcp_close(tp); 2442 goto drop; 2443 2444 case TCPS_CLOSING: 2445 case TCPS_LAST_ACK: 2446 case TCPS_TIME_WAIT: 2447 tp = tcp_close(tp); 2448 goto drop; 2449 } 2450 } 2451 2452 /* 2453 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2454 * we must be in a synchronized state. RFC791 states (under RST 2455 * generation) that any unacceptable segment (an out-of-order SYN 2456 * qualifies) received in a synchronized state must elicit only an 2457 * empty acknowledgment segment ... and the connection remains in 2458 * the same state. 2459 */ 2460 if (tiflags & TH_SYN) { 2461 if (tp->rcv_nxt == th->th_seq) { 2462 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2463 TH_ACK); 2464 if (tcp_saveti) 2465 m_freem(tcp_saveti); 2466 return; 2467 } 2468 2469 goto dropafterack_ratelim; 2470 } 2471 2472 /* 2473 * If the ACK bit is off we drop the segment and return. 2474 */ 2475 if ((tiflags & TH_ACK) == 0) { 2476 if (tp->t_flags & TF_ACKNOW) 2477 goto dropafterack; 2478 goto drop; 2479 } 2480 2481 /* 2482 * From here on, we're doing ACK processing. 2483 */ 2484 2485 switch (tp->t_state) { 2486 /* 2487 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2488 * ESTABLISHED state and continue processing, otherwise 2489 * send an RST. 2490 */ 2491 case TCPS_SYN_RECEIVED: 2492 if (SEQ_GT(tp->snd_una, th->th_ack) || 2493 SEQ_GT(th->th_ack, tp->snd_max)) 2494 goto dropwithreset; 2495 TCP_STATINC(TCP_STAT_CONNECTS); 2496 soisconnected(so); 2497 tcp_established(tp); 2498 /* Do window scaling? */ 2499 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2500 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2501 tp->snd_scale = tp->requested_s_scale; 2502 tp->rcv_scale = tp->request_r_scale; 2503 } 2504 TCP_REASS_LOCK(tp); 2505 (void)tcp_reass(tp, NULL, NULL, tlen); 2506 tp->snd_wl1 = th->th_seq - 1; 2507 /* FALLTHROUGH */ 2508 2509 /* 2510 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2511 * ACKs. If the ack is in the range 2512 * tp->snd_una < th->th_ack <= tp->snd_max 2513 * then advance tp->snd_una to th->th_ack and drop 2514 * data from the retransmission queue. If this ACK reflects 2515 * more up to date window information we update our window information. 2516 */ 2517 case TCPS_ESTABLISHED: 2518 case TCPS_FIN_WAIT_1: 2519 case TCPS_FIN_WAIT_2: 2520 case TCPS_CLOSE_WAIT: 2521 case TCPS_CLOSING: 2522 case TCPS_LAST_ACK: 2523 case TCPS_TIME_WAIT: 2524 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2525 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2526 TCP_STATINC(TCP_STAT_RCVDUPACK); 2527 /* 2528 * If we have outstanding data (other than 2529 * a window probe), this is a completely 2530 * duplicate ack (ie, window info didn't 2531 * change), the ack is the biggest we've 2532 * seen and we've seen exactly our rexmt 2533 * threshhold of them, assume a packet 2534 * has been dropped and retransmit it. 2535 * Kludge snd_nxt & the congestion 2536 * window so we send only this one 2537 * packet. 2538 */ 2539 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2540 th->th_ack != tp->snd_una) 2541 tp->t_dupacks = 0; 2542 else if (tp->t_partialacks < 0 && 2543 (++tp->t_dupacks == tcprexmtthresh || 2544 TCP_FACK_FASTRECOV(tp))) { 2545 /* 2546 * Do the fast retransmit, and adjust 2547 * congestion control paramenters. 2548 */ 2549 if (tp->t_congctl->fast_retransmit(tp, th)) { 2550 /* False fast retransmit */ 2551 break; 2552 } 2553 goto drop; 2554 } else if (tp->t_dupacks > tcprexmtthresh) { 2555 tp->snd_cwnd += tp->t_segsz; 2556 KERNEL_LOCK(1, NULL); 2557 (void)tcp_output(tp); 2558 KERNEL_UNLOCK_ONE(NULL); 2559 goto drop; 2560 } 2561 } else { 2562 /* 2563 * If the ack appears to be very old, only 2564 * allow data that is in-sequence. This 2565 * makes it somewhat more difficult to insert 2566 * forged data by guessing sequence numbers. 2567 * Sent an ack to try to update the send 2568 * sequence number on the other side. 2569 */ 2570 if (tlen && th->th_seq != tp->rcv_nxt && 2571 SEQ_LT(th->th_ack, 2572 tp->snd_una - tp->max_sndwnd)) 2573 goto dropafterack; 2574 } 2575 break; 2576 } 2577 /* 2578 * If the congestion window was inflated to account 2579 * for the other side's cached packets, retract it. 2580 */ 2581 tp->t_congctl->fast_retransmit_newack(tp, th); 2582 2583 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2584 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2585 goto dropafterack; 2586 } 2587 acked = th->th_ack - tp->snd_una; 2588 tcps = TCP_STAT_GETREF(); 2589 tcps[TCP_STAT_RCVACKPACK]++; 2590 tcps[TCP_STAT_RCVACKBYTE] += acked; 2591 TCP_STAT_PUTREF(); 2592 2593 /* 2594 * If we have a timestamp reply, update smoothed 2595 * round trip time. If no timestamp is present but 2596 * transmit timer is running and timed sequence 2597 * number was acked, update smoothed round trip time. 2598 * Since we now have an rtt measurement, cancel the 2599 * timer backoff (cf., Phil Karn's retransmit alg.). 2600 * Recompute the initial retransmit timer. 2601 */ 2602 if (ts_rtt) 2603 tcp_xmit_timer(tp, ts_rtt - 1); 2604 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2605 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2606 2607 /* 2608 * If all outstanding data is acked, stop retransmit 2609 * timer and remember to restart (more output or persist). 2610 * If there is more data to be acked, restart retransmit 2611 * timer, using current (possibly backed-off) value. 2612 */ 2613 if (th->th_ack == tp->snd_max) { 2614 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2615 needoutput = 1; 2616 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2617 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2618 2619 /* 2620 * New data has been acked, adjust the congestion window. 2621 */ 2622 tp->t_congctl->newack(tp, th); 2623 2624 nd6_hint(tp); 2625 if (acked > so->so_snd.sb_cc) { 2626 tp->snd_wnd -= so->so_snd.sb_cc; 2627 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2628 ourfinisacked = 1; 2629 } else { 2630 if (acked > (tp->t_lastoff - tp->t_inoff)) 2631 tp->t_lastm = NULL; 2632 sbdrop(&so->so_snd, acked); 2633 tp->t_lastoff -= acked; 2634 if (tp->snd_wnd > acked) 2635 tp->snd_wnd -= acked; 2636 else 2637 tp->snd_wnd = 0; 2638 ourfinisacked = 0; 2639 } 2640 sowwakeup(so); 2641 2642 icmp_check(tp, th, acked); 2643 2644 tp->snd_una = th->th_ack; 2645 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2646 tp->snd_fack = tp->snd_una; 2647 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2648 tp->snd_nxt = tp->snd_una; 2649 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2650 tp->snd_high = tp->snd_una; 2651 2652 switch (tp->t_state) { 2653 2654 /* 2655 * In FIN_WAIT_1 STATE in addition to the processing 2656 * for the ESTABLISHED state if our FIN is now acknowledged 2657 * then enter FIN_WAIT_2. 2658 */ 2659 case TCPS_FIN_WAIT_1: 2660 if (ourfinisacked) { 2661 /* 2662 * If we can't receive any more 2663 * data, then closing user can proceed. 2664 * Starting the timer is contrary to the 2665 * specification, but if we don't get a FIN 2666 * we'll hang forever. 2667 */ 2668 if (so->so_state & SS_CANTRCVMORE) { 2669 soisdisconnected(so); 2670 if (tp->t_maxidle > 0) 2671 TCP_TIMER_ARM(tp, TCPT_2MSL, 2672 tp->t_maxidle); 2673 } 2674 tp->t_state = TCPS_FIN_WAIT_2; 2675 } 2676 break; 2677 2678 /* 2679 * In CLOSING STATE in addition to the processing for 2680 * the ESTABLISHED state if the ACK acknowledges our FIN 2681 * then enter the TIME-WAIT state, otherwise ignore 2682 * the segment. 2683 */ 2684 case TCPS_CLOSING: 2685 if (ourfinisacked) { 2686 tp->t_state = TCPS_TIME_WAIT; 2687 tcp_canceltimers(tp); 2688 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2689 soisdisconnected(so); 2690 } 2691 break; 2692 2693 /* 2694 * In LAST_ACK, we may still be waiting for data to drain 2695 * and/or to be acked, as well as for the ack of our FIN. 2696 * If our FIN is now acknowledged, delete the TCB, 2697 * enter the closed state and return. 2698 */ 2699 case TCPS_LAST_ACK: 2700 if (ourfinisacked) { 2701 tp = tcp_close(tp); 2702 goto drop; 2703 } 2704 break; 2705 2706 /* 2707 * In TIME_WAIT state the only thing that should arrive 2708 * is a retransmission of the remote FIN. Acknowledge 2709 * it and restart the finack timer. 2710 */ 2711 case TCPS_TIME_WAIT: 2712 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2713 goto dropafterack; 2714 } 2715 } 2716 2717 step6: 2718 /* 2719 * Update window information. 2720 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2721 */ 2722 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2723 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2724 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2725 /* keep track of pure window updates */ 2726 if (tlen == 0 && 2727 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2728 TCP_STATINC(TCP_STAT_RCVWINUPD); 2729 tp->snd_wnd = tiwin; 2730 tp->snd_wl1 = th->th_seq; 2731 tp->snd_wl2 = th->th_ack; 2732 if (tp->snd_wnd > tp->max_sndwnd) 2733 tp->max_sndwnd = tp->snd_wnd; 2734 needoutput = 1; 2735 } 2736 2737 /* 2738 * Process segments with URG. 2739 */ 2740 if ((tiflags & TH_URG) && th->th_urp && 2741 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2742 /* 2743 * This is a kludge, but if we receive and accept 2744 * random urgent pointers, we'll crash in 2745 * soreceive. It's hard to imagine someone 2746 * actually wanting to send this much urgent data. 2747 */ 2748 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2749 th->th_urp = 0; /* XXX */ 2750 tiflags &= ~TH_URG; /* XXX */ 2751 goto dodata; /* XXX */ 2752 } 2753 2754 /* 2755 * If this segment advances the known urgent pointer, 2756 * then mark the data stream. This should not happen 2757 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2758 * a FIN has been received from the remote side. 2759 * In these states we ignore the URG. 2760 * 2761 * According to RFC961 (Assigned Protocols), 2762 * the urgent pointer points to the last octet 2763 * of urgent data. We continue, however, 2764 * to consider it to indicate the first octet 2765 * of data past the urgent section as the original 2766 * spec states (in one of two places). 2767 */ 2768 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2769 tp->rcv_up = th->th_seq + th->th_urp; 2770 so->so_oobmark = so->so_rcv.sb_cc + 2771 (tp->rcv_up - tp->rcv_nxt) - 1; 2772 if (so->so_oobmark == 0) 2773 so->so_state |= SS_RCVATMARK; 2774 sohasoutofband(so); 2775 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2776 } 2777 2778 /* 2779 * Remove out of band data so doesn't get presented to user. 2780 * This can happen independent of advancing the URG pointer, 2781 * but if two URG's are pending at once, some out-of-band 2782 * data may creep in... ick. 2783 */ 2784 if (th->th_urp <= (u_int16_t)tlen && 2785 (so->so_options & SO_OOBINLINE) == 0) 2786 tcp_pulloutofband(so, th, m, hdroptlen); 2787 } else { 2788 /* 2789 * If no out of band data is expected, 2790 * pull receive urgent pointer along 2791 * with the receive window. 2792 */ 2793 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2794 tp->rcv_up = tp->rcv_nxt; 2795 } 2796 dodata: 2797 2798 /* 2799 * Process the segment text, merging it into the TCP sequencing queue, 2800 * and arranging for acknowledgement of receipt if necessary. 2801 * This process logically involves adjusting tp->rcv_wnd as data 2802 * is presented to the user (this happens in tcp_usrreq.c, 2803 * tcp_rcvd()). If a FIN has already been received on this 2804 * connection then we just ignore the text. 2805 */ 2806 if ((tlen || (tiflags & TH_FIN)) && 2807 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2808 /* 2809 * Handle the common case: 2810 * o Segment is the next to be received, and 2811 * o The queue is empty, and 2812 * o The connection is established 2813 * In this case, we avoid calling tcp_reass. 2814 * 2815 * tcp_setup_ack: set DELACK for segments received in order, 2816 * but ack immediately when segments are out of order (so that 2817 * fast retransmit can work). 2818 */ 2819 TCP_REASS_LOCK(tp); 2820 if (th->th_seq == tp->rcv_nxt && 2821 TAILQ_FIRST(&tp->segq) == NULL && 2822 tp->t_state == TCPS_ESTABLISHED) { 2823 tcp_setup_ack(tp, th); 2824 tp->rcv_nxt += tlen; 2825 tiflags = th->th_flags & TH_FIN; 2826 tcps = TCP_STAT_GETREF(); 2827 tcps[TCP_STAT_RCVPACK]++; 2828 tcps[TCP_STAT_RCVBYTE] += tlen; 2829 TCP_STAT_PUTREF(); 2830 nd6_hint(tp); 2831 if (so->so_state & SS_CANTRCVMORE) { 2832 m_freem(m); 2833 } else { 2834 m_adj(m, hdroptlen); 2835 sbappendstream(&(so)->so_rcv, m); 2836 } 2837 TCP_REASS_UNLOCK(tp); 2838 sorwakeup(so); 2839 } else { 2840 m_adj(m, hdroptlen); 2841 tiflags = tcp_reass(tp, th, m, tlen); 2842 tp->t_flags |= TF_ACKNOW; 2843 } 2844 2845 /* 2846 * Note the amount of data that peer has sent into 2847 * our window, in order to estimate the sender's 2848 * buffer size. 2849 */ 2850 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2851 } else { 2852 m_freem(m); 2853 m = NULL; 2854 tiflags &= ~TH_FIN; 2855 } 2856 2857 /* 2858 * If FIN is received ACK the FIN and let the user know 2859 * that the connection is closing. Ignore a FIN received before 2860 * the connection is fully established. 2861 */ 2862 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2863 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2864 socantrcvmore(so); 2865 tp->t_flags |= TF_ACKNOW; 2866 tp->rcv_nxt++; 2867 } 2868 switch (tp->t_state) { 2869 2870 /* 2871 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2872 */ 2873 case TCPS_ESTABLISHED: 2874 tp->t_state = TCPS_CLOSE_WAIT; 2875 break; 2876 2877 /* 2878 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2879 * enter the CLOSING state. 2880 */ 2881 case TCPS_FIN_WAIT_1: 2882 tp->t_state = TCPS_CLOSING; 2883 break; 2884 2885 /* 2886 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2887 * starting the time-wait timer, turning off the other 2888 * standard timers. 2889 */ 2890 case TCPS_FIN_WAIT_2: 2891 tp->t_state = TCPS_TIME_WAIT; 2892 tcp_canceltimers(tp); 2893 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2894 soisdisconnected(so); 2895 break; 2896 2897 /* 2898 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2899 */ 2900 case TCPS_TIME_WAIT: 2901 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2902 break; 2903 } 2904 } 2905 #ifdef TCP_DEBUG 2906 if (so->so_options & SO_DEBUG) 2907 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2908 #endif 2909 2910 /* 2911 * Return any desired output. 2912 */ 2913 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2914 KERNEL_LOCK(1, NULL); 2915 (void)tcp_output(tp); 2916 KERNEL_UNLOCK_ONE(NULL); 2917 } 2918 if (tcp_saveti) 2919 m_freem(tcp_saveti); 2920 2921 if (tp->t_state == TCPS_TIME_WAIT 2922 && (so->so_state & SS_NOFDREF) 2923 && (tp->t_inpcb || af != AF_INET) 2924 && (tp->t_in6pcb || af != AF_INET6) 2925 && ((af == AF_INET ? tcp4_vtw_enable : tcp6_vtw_enable) & 1) != 0 2926 && TAILQ_EMPTY(&tp->segq) 2927 && vtw_add(af, tp)) { 2928 ; 2929 } 2930 return; 2931 2932 badsyn: 2933 /* 2934 * Received a bad SYN. Increment counters and dropwithreset. 2935 */ 2936 TCP_STATINC(TCP_STAT_BADSYN); 2937 tp = NULL; 2938 goto dropwithreset; 2939 2940 dropafterack: 2941 /* 2942 * Generate an ACK dropping incoming segment if it occupies 2943 * sequence space, where the ACK reflects our state. 2944 */ 2945 if (tiflags & TH_RST) 2946 goto drop; 2947 goto dropafterack2; 2948 2949 dropafterack_ratelim: 2950 /* 2951 * We may want to rate-limit ACKs against SYN/RST attack. 2952 */ 2953 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2954 tcp_ackdrop_ppslim) == 0) { 2955 /* XXX stat */ 2956 goto drop; 2957 } 2958 2959 dropafterack2: 2960 m_freem(m); 2961 tp->t_flags |= TF_ACKNOW; 2962 KERNEL_LOCK(1, NULL); 2963 (void)tcp_output(tp); 2964 KERNEL_UNLOCK_ONE(NULL); 2965 if (tcp_saveti) 2966 m_freem(tcp_saveti); 2967 return; 2968 2969 dropwithreset_ratelim: 2970 /* 2971 * We may want to rate-limit RSTs in certain situations, 2972 * particularly if we are sending an RST in response to 2973 * an attempt to connect to or otherwise communicate with 2974 * a port for which we have no socket. 2975 */ 2976 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2977 tcp_rst_ppslim) == 0) { 2978 /* XXX stat */ 2979 goto drop; 2980 } 2981 2982 dropwithreset: 2983 /* 2984 * Generate a RST, dropping incoming segment. 2985 * Make ACK acceptable to originator of segment. 2986 */ 2987 if (tiflags & TH_RST) 2988 goto drop; 2989 if (tiflags & TH_ACK) { 2990 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2991 } else { 2992 if (tiflags & TH_SYN) 2993 tlen++; 2994 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2995 TH_RST|TH_ACK); 2996 } 2997 if (tcp_saveti) 2998 m_freem(tcp_saveti); 2999 return; 3000 3001 badcsum: 3002 drop: 3003 /* 3004 * Drop space held by incoming segment and return. 3005 */ 3006 if (tp) { 3007 if (tp->t_inpcb) 3008 so = tp->t_inpcb->inp_socket; 3009 #ifdef INET6 3010 else if (tp->t_in6pcb) 3011 so = tp->t_in6pcb->in6p_socket; 3012 #endif 3013 else 3014 so = NULL; 3015 #ifdef TCP_DEBUG 3016 if (so && (so->so_options & SO_DEBUG) != 0) 3017 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 3018 #endif 3019 } 3020 if (tcp_saveti) 3021 m_freem(tcp_saveti); 3022 m_freem(m); 3023 return; 3024 } 3025 3026 #ifdef TCP_SIGNATURE 3027 int 3028 tcp_signature_apply(void *fstate, void *data, u_int len) 3029 { 3030 3031 MD5Update(fstate, (u_char *)data, len); 3032 return (0); 3033 } 3034 3035 struct secasvar * 3036 tcp_signature_getsav(struct mbuf *m) 3037 { 3038 struct ip *ip; 3039 struct ip6_hdr *ip6; 3040 3041 ip = mtod(m, struct ip *); 3042 switch (ip->ip_v) { 3043 case 4: 3044 ip = mtod(m, struct ip *); 3045 ip6 = NULL; 3046 break; 3047 case 6: 3048 ip = NULL; 3049 ip6 = mtod(m, struct ip6_hdr *); 3050 break; 3051 default: 3052 return (NULL); 3053 } 3054 3055 #ifdef IPSEC 3056 union sockaddr_union dst; 3057 3058 /* Extract the destination from the IP header in the mbuf. */ 3059 memset(&dst, 0, sizeof(union sockaddr_union)); 3060 if (ip != NULL) { 3061 dst.sa.sa_len = sizeof(struct sockaddr_in); 3062 dst.sa.sa_family = AF_INET; 3063 dst.sin.sin_addr = ip->ip_dst; 3064 } else { 3065 dst.sa.sa_len = sizeof(struct sockaddr_in6); 3066 dst.sa.sa_family = AF_INET6; 3067 dst.sin6.sin6_addr = ip6->ip6_dst; 3068 } 3069 3070 /* 3071 * Look up an SADB entry which matches the address of the peer. 3072 */ 3073 return KEY_LOOKUP_SA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI), 0, 0); 3074 #else 3075 return NULL; 3076 #endif 3077 } 3078 3079 int 3080 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 3081 struct secasvar *sav, char *sig) 3082 { 3083 MD5_CTX ctx; 3084 struct ip *ip; 3085 struct ipovly *ipovly; 3086 #ifdef INET6 3087 struct ip6_hdr *ip6; 3088 struct ip6_hdr_pseudo ip6pseudo; 3089 #endif 3090 struct ippseudo ippseudo; 3091 struct tcphdr th0; 3092 int l, tcphdrlen; 3093 3094 if (sav == NULL) 3095 return (-1); 3096 3097 tcphdrlen = th->th_off * 4; 3098 3099 switch (mtod(m, struct ip *)->ip_v) { 3100 case 4: 3101 MD5Init(&ctx); 3102 ip = mtod(m, struct ip *); 3103 memset(&ippseudo, 0, sizeof(ippseudo)); 3104 ipovly = (struct ipovly *)ip; 3105 ippseudo.ippseudo_src = ipovly->ih_src; 3106 ippseudo.ippseudo_dst = ipovly->ih_dst; 3107 ippseudo.ippseudo_pad = 0; 3108 ippseudo.ippseudo_p = IPPROTO_TCP; 3109 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 3110 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 3111 break; 3112 #if INET6 3113 case 6: 3114 MD5Init(&ctx); 3115 ip6 = mtod(m, struct ip6_hdr *); 3116 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 3117 ip6pseudo.ip6ph_src = ip6->ip6_src; 3118 in6_clearscope(&ip6pseudo.ip6ph_src); 3119 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 3120 in6_clearscope(&ip6pseudo.ip6ph_dst); 3121 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 3122 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 3123 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 3124 break; 3125 #endif 3126 default: 3127 return (-1); 3128 } 3129 3130 th0 = *th; 3131 th0.th_sum = 0; 3132 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 3133 3134 l = m->m_pkthdr.len - thoff - tcphdrlen; 3135 if (l > 0) 3136 m_apply(m, thoff + tcphdrlen, 3137 m->m_pkthdr.len - thoff - tcphdrlen, 3138 tcp_signature_apply, &ctx); 3139 3140 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 3141 MD5Final(sig, &ctx); 3142 3143 return (0); 3144 } 3145 #endif 3146 3147 /* 3148 * Parse and process tcp options. 3149 * 3150 * Returns -1 if this segment should be dropped. (eg. wrong signature) 3151 * Otherwise returns 0. 3152 */ 3153 static int 3154 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, struct tcphdr *th, 3155 struct mbuf *m, int toff, struct tcp_opt_info *oi) 3156 { 3157 u_int16_t mss; 3158 int opt, optlen = 0; 3159 #ifdef TCP_SIGNATURE 3160 void *sigp = NULL; 3161 char sigbuf[TCP_SIGLEN]; 3162 struct secasvar *sav = NULL; 3163 #endif 3164 3165 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 3166 opt = cp[0]; 3167 if (opt == TCPOPT_EOL) 3168 break; 3169 if (opt == TCPOPT_NOP) 3170 optlen = 1; 3171 else { 3172 if (cnt < 2) 3173 break; 3174 optlen = cp[1]; 3175 if (optlen < 2 || optlen > cnt) 3176 break; 3177 } 3178 switch (opt) { 3179 3180 default: 3181 continue; 3182 3183 case TCPOPT_MAXSEG: 3184 if (optlen != TCPOLEN_MAXSEG) 3185 continue; 3186 if (!(th->th_flags & TH_SYN)) 3187 continue; 3188 if (TCPS_HAVERCVDSYN(tp->t_state)) 3189 continue; 3190 memcpy(&mss, cp + 2, sizeof(mss)); 3191 oi->maxseg = ntohs(mss); 3192 break; 3193 3194 case TCPOPT_WINDOW: 3195 if (optlen != TCPOLEN_WINDOW) 3196 continue; 3197 if (!(th->th_flags & TH_SYN)) 3198 continue; 3199 if (TCPS_HAVERCVDSYN(tp->t_state)) 3200 continue; 3201 tp->t_flags |= TF_RCVD_SCALE; 3202 tp->requested_s_scale = cp[2]; 3203 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3204 char buf[INET6_ADDRSTRLEN]; 3205 struct ip *ip = mtod(m, struct ip *); 3206 #ifdef INET6 3207 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 3208 #endif 3209 if (ip) 3210 in_print(buf, sizeof(buf), 3211 &ip->ip_src); 3212 #ifdef INET6 3213 else if (ip6) 3214 in6_print(buf, sizeof(buf), 3215 &ip6->ip6_src); 3216 #endif 3217 else 3218 strlcpy(buf, "(unknown)", sizeof(buf)); 3219 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3220 "assuming %d\n", 3221 tp->requested_s_scale, buf, 3222 TCP_MAX_WINSHIFT); 3223 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3224 } 3225 break; 3226 3227 case TCPOPT_TIMESTAMP: 3228 if (optlen != TCPOLEN_TIMESTAMP) 3229 continue; 3230 oi->ts_present = 1; 3231 memcpy(&oi->ts_val, cp + 2, sizeof(oi->ts_val)); 3232 NTOHL(oi->ts_val); 3233 memcpy(&oi->ts_ecr, cp + 6, sizeof(oi->ts_ecr)); 3234 NTOHL(oi->ts_ecr); 3235 3236 if (!(th->th_flags & TH_SYN)) 3237 continue; 3238 if (TCPS_HAVERCVDSYN(tp->t_state)) 3239 continue; 3240 /* 3241 * A timestamp received in a SYN makes 3242 * it ok to send timestamp requests and replies. 3243 */ 3244 tp->t_flags |= TF_RCVD_TSTMP; 3245 tp->ts_recent = oi->ts_val; 3246 tp->ts_recent_age = tcp_now; 3247 break; 3248 3249 case TCPOPT_SACK_PERMITTED: 3250 if (optlen != TCPOLEN_SACK_PERMITTED) 3251 continue; 3252 if (!(th->th_flags & TH_SYN)) 3253 continue; 3254 if (TCPS_HAVERCVDSYN(tp->t_state)) 3255 continue; 3256 if (tcp_do_sack) { 3257 tp->t_flags |= TF_SACK_PERMIT; 3258 tp->t_flags |= TF_WILL_SACK; 3259 } 3260 break; 3261 3262 case TCPOPT_SACK: 3263 tcp_sack_option(tp, th, cp, optlen); 3264 break; 3265 #ifdef TCP_SIGNATURE 3266 case TCPOPT_SIGNATURE: 3267 if (optlen != TCPOLEN_SIGNATURE) 3268 continue; 3269 if (sigp && memcmp(sigp, cp + 2, TCP_SIGLEN)) 3270 return (-1); 3271 3272 sigp = sigbuf; 3273 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3274 tp->t_flags |= TF_SIGNATURE; 3275 break; 3276 #endif 3277 } 3278 } 3279 3280 #ifndef TCP_SIGNATURE 3281 return 0; 3282 #else 3283 if (tp->t_flags & TF_SIGNATURE) { 3284 sav = tcp_signature_getsav(m); 3285 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3286 return (-1); 3287 } 3288 3289 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) 3290 goto out; 3291 3292 if (sigp) { 3293 char sig[TCP_SIGLEN]; 3294 3295 tcp_fields_to_net(th); 3296 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3297 tcp_fields_to_host(th); 3298 goto out; 3299 } 3300 tcp_fields_to_host(th); 3301 3302 if (memcmp(sig, sigp, TCP_SIGLEN)) { 3303 TCP_STATINC(TCP_STAT_BADSIG); 3304 goto out; 3305 } else 3306 TCP_STATINC(TCP_STAT_GOODSIG); 3307 3308 key_sa_recordxfer(sav, m); 3309 KEY_SA_UNREF(&sav); 3310 } 3311 return 0; 3312 out: 3313 if (sav != NULL) 3314 KEY_SA_UNREF(&sav); 3315 return -1; 3316 #endif 3317 } 3318 3319 /* 3320 * Pull out of band byte out of a segment so 3321 * it doesn't appear in the user's data queue. 3322 * It is still reflected in the segment length for 3323 * sequencing purposes. 3324 */ 3325 void 3326 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3327 struct mbuf *m, int off) 3328 { 3329 int cnt = off + th->th_urp - 1; 3330 3331 while (cnt >= 0) { 3332 if (m->m_len > cnt) { 3333 char *cp = mtod(m, char *) + cnt; 3334 struct tcpcb *tp = sototcpcb(so); 3335 3336 tp->t_iobc = *cp; 3337 tp->t_oobflags |= TCPOOB_HAVEDATA; 3338 memmove(cp, cp + 1, (unsigned)(m->m_len - cnt - 1)); 3339 m->m_len--; 3340 return; 3341 } 3342 cnt -= m->m_len; 3343 m = m->m_next; 3344 if (m == NULL) 3345 break; 3346 } 3347 panic("tcp_pulloutofband"); 3348 } 3349 3350 /* 3351 * Collect new round-trip time estimate 3352 * and update averages and current timeout. 3353 * 3354 * rtt is in units of slow ticks (typically 500 ms) -- essentially the 3355 * difference of two timestamps. 3356 */ 3357 void 3358 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3359 { 3360 int32_t delta; 3361 3362 TCP_STATINC(TCP_STAT_RTTUPDATED); 3363 if (tp->t_srtt != 0) { 3364 /* 3365 * Compute the amount to add to srtt for smoothing, 3366 * *alpha, or 2^(-TCP_RTT_SHIFT). Because 3367 * srtt is stored in 1/32 slow ticks, we conceptually 3368 * shift left 5 bits, subtract srtt to get the 3369 * diference, and then shift right by TCP_RTT_SHIFT 3370 * (3) to obtain 1/8 of the difference. 3371 */ 3372 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3373 /* 3374 * This can never happen, because delta's lowest 3375 * possible value is 1/8 of t_srtt. But if it does, 3376 * set srtt to some reasonable value, here chosen 3377 * as 1/8 tick. 3378 */ 3379 if ((tp->t_srtt += delta) <= 0) 3380 tp->t_srtt = 1 << 2; 3381 /* 3382 * RFC2988 requires that rttvar be updated first. 3383 * This code is compliant because "delta" is the old 3384 * srtt minus the new observation (scaled). 3385 * 3386 * RFC2988 says: 3387 * rttvar = (1-beta) * rttvar + beta * |srtt-observed| 3388 * 3389 * delta is in units of 1/32 ticks, and has then been 3390 * divided by 8. This is equivalent to being in 1/16s 3391 * units and divided by 4. Subtract from it 1/4 of 3392 * the existing rttvar to form the (signed) amount to 3393 * adjust. 3394 */ 3395 if (delta < 0) 3396 delta = -delta; 3397 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3398 /* 3399 * As with srtt, this should never happen. There is 3400 * no support in RFC2988 for this operation. But 1/4s 3401 * as rttvar when faced with something arguably wrong 3402 * is ok. 3403 */ 3404 if ((tp->t_rttvar += delta) <= 0) 3405 tp->t_rttvar = 1 << 2; 3406 3407 /* 3408 * If srtt exceeds .01 second, ensure we use the 'remote' MSL 3409 * Problem is: it doesn't work. Disabled by defaulting 3410 * tcp_rttlocal to 0; see corresponding code in 3411 * tcp_subr that selects local vs remote in a different way. 3412 * 3413 * The static branch prediction hint here should be removed 3414 * when the rtt estimator is fixed and the rtt_enable code 3415 * is turned back on. 3416 */ 3417 if (__predict_false(tcp_rttlocal) && tcp_msl_enable 3418 && tp->t_srtt > tcp_msl_remote_threshold 3419 && tp->t_msl < tcp_msl_remote) { 3420 tp->t_msl = tcp_msl_remote; 3421 } 3422 } else { 3423 /* 3424 * This is the first measurement. Per RFC2988, 2.2, 3425 * set rtt=R and srtt=R/2. 3426 * For srtt, storage representation is 1/32 ticks, 3427 * so shift left by 5. 3428 * For rttvar, storage representation is 1/16 ticks, 3429 * So shift left by 4, but then right by 1 to halve. 3430 */ 3431 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3432 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3433 } 3434 tp->t_rtttime = 0; 3435 tp->t_rxtshift = 0; 3436 3437 /* 3438 * the retransmit should happen at rtt + 4 * rttvar. 3439 * Because of the way we do the smoothing, srtt and rttvar 3440 * will each average +1/2 tick of bias. When we compute 3441 * the retransmit timer, we want 1/2 tick of rounding and 3442 * 1 extra tick because of +-1/2 tick uncertainty in the 3443 * firing of the timer. The bias will give us exactly the 3444 * 1.5 tick we need. But, because the bias is 3445 * statistical, we have to test that we don't drop below 3446 * the minimum feasible timer (which is 2 ticks). 3447 */ 3448 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3449 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3450 3451 /* 3452 * We received an ack for a packet that wasn't retransmitted; 3453 * it is probably safe to discard any error indications we've 3454 * received recently. This isn't quite right, but close enough 3455 * for now (a route might have failed after we sent a segment, 3456 * and the return path might not be symmetrical). 3457 */ 3458 tp->t_softerror = 0; 3459 } 3460 3461 3462 /* 3463 * TCP compressed state engine. Currently used to hold compressed 3464 * state for SYN_RECEIVED. 3465 */ 3466 3467 u_long syn_cache_count; 3468 u_int32_t syn_hash1, syn_hash2; 3469 3470 #define SYN_HASH(sa, sp, dp) \ 3471 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3472 ((u_int32_t)(sp)))^syn_hash2))) 3473 #ifndef INET6 3474 #define SYN_HASHALL(hash, src, dst) \ 3475 do { \ 3476 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3477 ((const struct sockaddr_in *)(src))->sin_port, \ 3478 ((const struct sockaddr_in *)(dst))->sin_port); \ 3479 } while (/*CONSTCOND*/ 0) 3480 #else 3481 #define SYN_HASH6(sa, sp, dp) \ 3482 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3483 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3484 & 0x7fffffff) 3485 3486 #define SYN_HASHALL(hash, src, dst) \ 3487 do { \ 3488 switch ((src)->sa_family) { \ 3489 case AF_INET: \ 3490 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3491 ((const struct sockaddr_in *)(src))->sin_port, \ 3492 ((const struct sockaddr_in *)(dst))->sin_port); \ 3493 break; \ 3494 case AF_INET6: \ 3495 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3496 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3497 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3498 break; \ 3499 default: \ 3500 hash = 0; \ 3501 } \ 3502 } while (/*CONSTCOND*/0) 3503 #endif /* INET6 */ 3504 3505 static struct pool syn_cache_pool; 3506 3507 /* 3508 * We don't estimate RTT with SYNs, so each packet starts with the default 3509 * RTT and each timer step has a fixed timeout value. 3510 */ 3511 static inline void 3512 syn_cache_timer_arm(struct syn_cache *sc) 3513 { 3514 3515 TCPT_RANGESET(sc->sc_rxtcur, 3516 TCPTV_SRTTDFLT * tcp_backoff[sc->sc_rxtshift], TCPTV_MIN, 3517 TCPTV_REXMTMAX); 3518 callout_reset(&sc->sc_timer, 3519 sc->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, sc); 3520 } 3521 3522 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3523 3524 static inline void 3525 syn_cache_rm(struct syn_cache *sc) 3526 { 3527 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3528 sc, sc_bucketq); 3529 sc->sc_tp = NULL; 3530 LIST_REMOVE(sc, sc_tpq); 3531 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3532 callout_stop(&sc->sc_timer); 3533 syn_cache_count--; 3534 } 3535 3536 static inline void 3537 syn_cache_put(struct syn_cache *sc) 3538 { 3539 if (sc->sc_ipopts) 3540 (void) m_free(sc->sc_ipopts); 3541 rtcache_free(&sc->sc_route); 3542 sc->sc_flags |= SCF_DEAD; 3543 if (!callout_invoking(&sc->sc_timer)) 3544 callout_schedule(&(sc)->sc_timer, 1); 3545 } 3546 3547 void 3548 syn_cache_init(void) 3549 { 3550 int i; 3551 3552 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 3553 "synpl", NULL, IPL_SOFTNET); 3554 3555 /* Initialize the hash buckets. */ 3556 for (i = 0; i < tcp_syn_cache_size; i++) 3557 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3558 } 3559 3560 void 3561 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3562 { 3563 struct syn_cache_head *scp; 3564 struct syn_cache *sc2; 3565 int s; 3566 3567 /* 3568 * If there are no entries in the hash table, reinitialize 3569 * the hash secrets. 3570 */ 3571 if (syn_cache_count == 0) { 3572 syn_hash1 = cprng_fast32(); 3573 syn_hash2 = cprng_fast32(); 3574 } 3575 3576 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3577 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3578 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3579 3580 /* 3581 * Make sure that we don't overflow the per-bucket 3582 * limit or the total cache size limit. 3583 */ 3584 s = splsoftnet(); 3585 if (scp->sch_length >= tcp_syn_bucket_limit) { 3586 TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW); 3587 /* 3588 * The bucket is full. Toss the oldest element in the 3589 * bucket. This will be the first entry in the bucket. 3590 */ 3591 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3592 #ifdef DIAGNOSTIC 3593 /* 3594 * This should never happen; we should always find an 3595 * entry in our bucket. 3596 */ 3597 if (sc2 == NULL) 3598 panic("syn_cache_insert: bucketoverflow: impossible"); 3599 #endif 3600 syn_cache_rm(sc2); 3601 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3602 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3603 struct syn_cache_head *scp2, *sce; 3604 3605 TCP_STATINC(TCP_STAT_SC_OVERFLOWED); 3606 /* 3607 * The cache is full. Toss the oldest entry in the 3608 * first non-empty bucket we can find. 3609 * 3610 * XXX We would really like to toss the oldest 3611 * entry in the cache, but we hope that this 3612 * condition doesn't happen very often. 3613 */ 3614 scp2 = scp; 3615 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3616 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3617 for (++scp2; scp2 != scp; scp2++) { 3618 if (scp2 >= sce) 3619 scp2 = &tcp_syn_cache[0]; 3620 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3621 break; 3622 } 3623 #ifdef DIAGNOSTIC 3624 /* 3625 * This should never happen; we should always find a 3626 * non-empty bucket. 3627 */ 3628 if (scp2 == scp) 3629 panic("syn_cache_insert: cacheoverflow: " 3630 "impossible"); 3631 #endif 3632 } 3633 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3634 syn_cache_rm(sc2); 3635 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3636 } 3637 3638 /* 3639 * Initialize the entry's timer. 3640 */ 3641 sc->sc_rxttot = 0; 3642 sc->sc_rxtshift = 0; 3643 syn_cache_timer_arm(sc); 3644 3645 /* Link it from tcpcb entry */ 3646 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3647 3648 /* Put it into the bucket. */ 3649 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3650 scp->sch_length++; 3651 syn_cache_count++; 3652 3653 TCP_STATINC(TCP_STAT_SC_ADDED); 3654 splx(s); 3655 } 3656 3657 /* 3658 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3659 * If we have retransmitted an entry the maximum number of times, expire 3660 * that entry. 3661 */ 3662 static void 3663 syn_cache_timer(void *arg) 3664 { 3665 struct syn_cache *sc = arg; 3666 3667 mutex_enter(softnet_lock); 3668 KERNEL_LOCK(1, NULL); 3669 3670 callout_ack(&sc->sc_timer); 3671 3672 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3673 TCP_STATINC(TCP_STAT_SC_DELAYED_FREE); 3674 goto free; 3675 } 3676 3677 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3678 /* Drop it -- too many retransmissions. */ 3679 goto dropit; 3680 } 3681 3682 /* 3683 * Compute the total amount of time this entry has 3684 * been on a queue. If this entry has been on longer 3685 * than the keep alive timer would allow, expire it. 3686 */ 3687 sc->sc_rxttot += sc->sc_rxtcur; 3688 if (sc->sc_rxttot >= tcp_keepinit) 3689 goto dropit; 3690 3691 TCP_STATINC(TCP_STAT_SC_RETRANSMITTED); 3692 (void)syn_cache_respond(sc); 3693 3694 /* Advance the timer back-off. */ 3695 sc->sc_rxtshift++; 3696 syn_cache_timer_arm(sc); 3697 3698 goto out; 3699 3700 dropit: 3701 TCP_STATINC(TCP_STAT_SC_TIMED_OUT); 3702 syn_cache_rm(sc); 3703 if (sc->sc_ipopts) 3704 (void) m_free(sc->sc_ipopts); 3705 rtcache_free(&sc->sc_route); 3706 3707 free: 3708 callout_destroy(&sc->sc_timer); 3709 pool_put(&syn_cache_pool, sc); 3710 3711 out: 3712 KERNEL_UNLOCK_ONE(NULL); 3713 mutex_exit(softnet_lock); 3714 } 3715 3716 /* 3717 * Remove syn cache created by the specified tcb entry, 3718 * because this does not make sense to keep them 3719 * (if there's no tcb entry, syn cache entry will never be used) 3720 */ 3721 void 3722 syn_cache_cleanup(struct tcpcb *tp) 3723 { 3724 struct syn_cache *sc, *nsc; 3725 int s; 3726 3727 s = splsoftnet(); 3728 3729 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3730 nsc = LIST_NEXT(sc, sc_tpq); 3731 3732 #ifdef DIAGNOSTIC 3733 if (sc->sc_tp != tp) 3734 panic("invalid sc_tp in syn_cache_cleanup"); 3735 #endif 3736 syn_cache_rm(sc); 3737 syn_cache_put(sc); /* calls pool_put but see spl above */ 3738 } 3739 /* just for safety */ 3740 LIST_INIT(&tp->t_sc); 3741 3742 splx(s); 3743 } 3744 3745 /* 3746 * Find an entry in the syn cache. 3747 */ 3748 struct syn_cache * 3749 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3750 struct syn_cache_head **headp) 3751 { 3752 struct syn_cache *sc; 3753 struct syn_cache_head *scp; 3754 u_int32_t hash; 3755 int s; 3756 3757 SYN_HASHALL(hash, src, dst); 3758 3759 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3760 *headp = scp; 3761 s = splsoftnet(); 3762 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3763 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3764 if (sc->sc_hash != hash) 3765 continue; 3766 if (!memcmp(&sc->sc_src, src, src->sa_len) && 3767 !memcmp(&sc->sc_dst, dst, dst->sa_len)) { 3768 splx(s); 3769 return (sc); 3770 } 3771 } 3772 splx(s); 3773 return (NULL); 3774 } 3775 3776 /* 3777 * This function gets called when we receive an ACK for a socket in the 3778 * LISTEN state. We look up the connection in the syn cache, and if it's 3779 * there, we pull it out of the cache and turn it into a full-blown 3780 * connection in the SYN-RECEIVED state. 3781 * 3782 * The return values may not be immediately obvious, and their effects 3783 * can be subtle, so here they are: 3784 * 3785 * NULL SYN was not found in cache; caller should drop the 3786 * packet and send an RST. 3787 * 3788 * -1 We were unable to create the new connection, and are 3789 * aborting it. An ACK,RST is being sent to the peer 3790 * (unless we got screwey sequence numbers; see below), 3791 * because the 3-way handshake has been completed. Caller 3792 * should not free the mbuf, since we may be using it. If 3793 * we are not, we will free it. 3794 * 3795 * Otherwise, the return value is a pointer to the new socket 3796 * associated with the connection. 3797 */ 3798 struct socket * 3799 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3800 struct tcphdr *th, struct socket *so, struct mbuf *m) 3801 { 3802 struct syn_cache *sc; 3803 struct syn_cache_head *scp; 3804 struct inpcb *inp = NULL; 3805 #ifdef INET6 3806 struct in6pcb *in6p = NULL; 3807 #endif 3808 struct tcpcb *tp; 3809 int s; 3810 struct socket *oso; 3811 3812 s = splsoftnet(); 3813 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3814 splx(s); 3815 return NULL; 3816 } 3817 3818 /* 3819 * Verify the sequence and ack numbers. Try getting the correct 3820 * response again. 3821 */ 3822 if ((th->th_ack != sc->sc_iss + 1) || 3823 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3824 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3825 m_freem(m); 3826 (void)syn_cache_respond(sc); 3827 splx(s); 3828 return ((struct socket *)(-1)); 3829 } 3830 3831 /* Remove this cache entry */ 3832 syn_cache_rm(sc); 3833 splx(s); 3834 3835 /* 3836 * Ok, create the full blown connection, and set things up 3837 * as they would have been set up if we had created the 3838 * connection when the SYN arrived. If we can't create 3839 * the connection, abort it. 3840 */ 3841 /* 3842 * inp still has the OLD in_pcb stuff, set the 3843 * v6-related flags on the new guy, too. This is 3844 * done particularly for the case where an AF_INET6 3845 * socket is bound only to a port, and a v4 connection 3846 * comes in on that port. 3847 * we also copy the flowinfo from the original pcb 3848 * to the new one. 3849 */ 3850 oso = so; 3851 so = sonewconn(so, true); 3852 if (so == NULL) 3853 goto resetandabort; 3854 3855 switch (so->so_proto->pr_domain->dom_family) { 3856 case AF_INET: 3857 inp = sotoinpcb(so); 3858 break; 3859 #ifdef INET6 3860 case AF_INET6: 3861 in6p = sotoin6pcb(so); 3862 break; 3863 #endif 3864 } 3865 3866 switch (src->sa_family) { 3867 case AF_INET: 3868 if (inp) { 3869 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3870 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3871 inp->inp_options = ip_srcroute(m); 3872 in_pcbstate(inp, INP_BOUND); 3873 if (inp->inp_options == NULL) { 3874 inp->inp_options = sc->sc_ipopts; 3875 sc->sc_ipopts = NULL; 3876 } 3877 } 3878 #ifdef INET6 3879 else if (in6p) { 3880 /* IPv4 packet to AF_INET6 socket */ 3881 memset(&in6p->in6p_laddr, 0, sizeof(in6p->in6p_laddr)); 3882 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3883 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3884 &in6p->in6p_laddr.s6_addr32[3], 3885 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3886 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3887 in6totcpcb(in6p)->t_family = AF_INET; 3888 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3889 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3890 else 3891 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3892 in6_pcbstate(in6p, IN6P_BOUND); 3893 } 3894 #endif 3895 break; 3896 #ifdef INET6 3897 case AF_INET6: 3898 if (in6p) { 3899 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3900 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3901 in6_pcbstate(in6p, IN6P_BOUND); 3902 } 3903 break; 3904 #endif 3905 } 3906 3907 #ifdef INET6 3908 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3909 struct in6pcb *oin6p = sotoin6pcb(oso); 3910 /* inherit socket options from the listening socket */ 3911 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3912 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3913 m_freem(in6p->in6p_options); 3914 in6p->in6p_options = NULL; 3915 } 3916 ip6_savecontrol(in6p, &in6p->in6p_options, 3917 mtod(m, struct ip6_hdr *), m); 3918 } 3919 #endif 3920 3921 #if defined(IPSEC) 3922 if (ipsec_used) { 3923 /* 3924 * we make a copy of policy, instead of sharing the policy, for 3925 * better behavior in terms of SA lookup and dead SA removal. 3926 */ 3927 if (inp) { 3928 /* copy old policy into new socket's */ 3929 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, 3930 inp->inp_sp)) 3931 printf("tcp_input: could not copy policy\n"); 3932 } 3933 #ifdef INET6 3934 else if (in6p) { 3935 /* copy old policy into new socket's */ 3936 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3937 in6p->in6p_sp)) 3938 printf("tcp_input: could not copy policy\n"); 3939 } 3940 #endif 3941 } 3942 #endif 3943 3944 /* 3945 * Give the new socket our cached route reference. 3946 */ 3947 if (inp) { 3948 rtcache_copy(&inp->inp_route, &sc->sc_route); 3949 rtcache_free(&sc->sc_route); 3950 } 3951 #ifdef INET6 3952 else { 3953 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 3954 rtcache_free(&sc->sc_route); 3955 } 3956 #endif 3957 3958 if (inp) { 3959 struct sockaddr_in sin; 3960 memcpy(&sin, src, src->sa_len); 3961 if (in_pcbconnect(inp, &sin, &lwp0)) { 3962 goto resetandabort; 3963 } 3964 } 3965 #ifdef INET6 3966 else if (in6p) { 3967 struct sockaddr_in6 sin6; 3968 memcpy(&sin6, src, src->sa_len); 3969 if (src->sa_family == AF_INET) { 3970 /* IPv4 packet to AF_INET6 socket */ 3971 in6_sin_2_v4mapsin6((struct sockaddr_in *)src, &sin6); 3972 } 3973 if (in6_pcbconnect(in6p, &sin6, NULL)) { 3974 goto resetandabort; 3975 } 3976 } 3977 #endif 3978 else { 3979 goto resetandabort; 3980 } 3981 3982 if (inp) 3983 tp = intotcpcb(inp); 3984 #ifdef INET6 3985 else if (in6p) 3986 tp = in6totcpcb(in6p); 3987 #endif 3988 else 3989 tp = NULL; 3990 3991 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3992 if (sc->sc_request_r_scale != 15) { 3993 tp->requested_s_scale = sc->sc_requested_s_scale; 3994 tp->request_r_scale = sc->sc_request_r_scale; 3995 tp->snd_scale = sc->sc_requested_s_scale; 3996 tp->rcv_scale = sc->sc_request_r_scale; 3997 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3998 } 3999 if (sc->sc_flags & SCF_TIMESTAMP) 4000 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 4001 tp->ts_timebase = sc->sc_timebase; 4002 4003 tp->t_template = tcp_template(tp); 4004 if (tp->t_template == 0) { 4005 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 4006 so = NULL; 4007 m_freem(m); 4008 goto abort; 4009 } 4010 4011 tp->iss = sc->sc_iss; 4012 tp->irs = sc->sc_irs; 4013 tcp_sendseqinit(tp); 4014 tcp_rcvseqinit(tp); 4015 tp->t_state = TCPS_SYN_RECEIVED; 4016 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 4017 TCP_STATINC(TCP_STAT_ACCEPTS); 4018 4019 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 4020 tp->t_flags |= TF_WILL_SACK; 4021 4022 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 4023 tp->t_flags |= TF_ECN_PERMIT; 4024 4025 #ifdef TCP_SIGNATURE 4026 if (sc->sc_flags & SCF_SIGNATURE) 4027 tp->t_flags |= TF_SIGNATURE; 4028 #endif 4029 4030 /* Initialize tp->t_ourmss before we deal with the peer's! */ 4031 tp->t_ourmss = sc->sc_ourmaxseg; 4032 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 4033 4034 /* 4035 * Initialize the initial congestion window. If we 4036 * had to retransmit the SYN,ACK, we must initialize cwnd 4037 * to 1 segment (i.e. the Loss Window). 4038 */ 4039 if (sc->sc_rxtshift) 4040 tp->snd_cwnd = tp->t_peermss; 4041 else { 4042 int ss = tcp_init_win; 4043 if (inp != NULL && in_localaddr(inp->inp_faddr)) 4044 ss = tcp_init_win_local; 4045 #ifdef INET6 4046 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 4047 ss = tcp_init_win_local; 4048 #endif 4049 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 4050 } 4051 4052 tcp_rmx_rtt(tp); 4053 tp->snd_wl1 = sc->sc_irs; 4054 tp->rcv_up = sc->sc_irs + 1; 4055 4056 /* 4057 * This is what whould have happened in tcp_output() when 4058 * the SYN,ACK was sent. 4059 */ 4060 tp->snd_up = tp->snd_una; 4061 tp->snd_max = tp->snd_nxt = tp->iss+1; 4062 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 4063 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 4064 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 4065 tp->last_ack_sent = tp->rcv_nxt; 4066 tp->t_partialacks = -1; 4067 tp->t_dupacks = 0; 4068 4069 TCP_STATINC(TCP_STAT_SC_COMPLETED); 4070 s = splsoftnet(); 4071 syn_cache_put(sc); 4072 splx(s); 4073 return so; 4074 4075 resetandabort: 4076 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 4077 abort: 4078 if (so != NULL) { 4079 (void) soqremque(so, 1); 4080 (void) soabort(so); 4081 mutex_enter(softnet_lock); 4082 } 4083 s = splsoftnet(); 4084 syn_cache_put(sc); 4085 splx(s); 4086 TCP_STATINC(TCP_STAT_SC_ABORTED); 4087 return ((struct socket *)(-1)); 4088 } 4089 4090 /* 4091 * This function is called when we get a RST for a 4092 * non-existent connection, so that we can see if the 4093 * connection is in the syn cache. If it is, zap it. 4094 */ 4095 4096 void 4097 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 4098 { 4099 struct syn_cache *sc; 4100 struct syn_cache_head *scp; 4101 int s = splsoftnet(); 4102 4103 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4104 splx(s); 4105 return; 4106 } 4107 if (SEQ_LT(th->th_seq, sc->sc_irs) || 4108 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 4109 splx(s); 4110 return; 4111 } 4112 syn_cache_rm(sc); 4113 TCP_STATINC(TCP_STAT_SC_RESET); 4114 syn_cache_put(sc); /* calls pool_put but see spl above */ 4115 splx(s); 4116 } 4117 4118 void 4119 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 4120 struct tcphdr *th) 4121 { 4122 struct syn_cache *sc; 4123 struct syn_cache_head *scp; 4124 int s; 4125 4126 s = splsoftnet(); 4127 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4128 splx(s); 4129 return; 4130 } 4131 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 4132 if (ntohl(th->th_seq) != sc->sc_iss) { 4133 splx(s); 4134 return; 4135 } 4136 4137 /* 4138 * If we've retransmitted 3 times and this is our second error, 4139 * we remove the entry. Otherwise, we allow it to continue on. 4140 * This prevents us from incorrectly nuking an entry during a 4141 * spurious network outage. 4142 * 4143 * See tcp_notify(). 4144 */ 4145 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 4146 sc->sc_flags |= SCF_UNREACH; 4147 splx(s); 4148 return; 4149 } 4150 4151 syn_cache_rm(sc); 4152 TCP_STATINC(TCP_STAT_SC_UNREACH); 4153 syn_cache_put(sc); /* calls pool_put but see spl above */ 4154 splx(s); 4155 } 4156 4157 /* 4158 * Given a LISTEN socket and an inbound SYN request, add this to the syn 4159 * cache, and send back a segment: 4160 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 4161 * to the source. 4162 * 4163 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 4164 * Doing so would require that we hold onto the data and deliver it 4165 * to the application. However, if we are the target of a SYN-flood 4166 * DoS attack, an attacker could send data which would eventually 4167 * consume all available buffer space if it were ACKed. By not ACKing 4168 * the data, we avoid this DoS scenario. 4169 */ 4170 int 4171 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 4172 unsigned int toff, struct socket *so, struct mbuf *m, u_char *optp, 4173 int optlen, struct tcp_opt_info *oi) 4174 { 4175 struct tcpcb tb, *tp; 4176 long win; 4177 struct syn_cache *sc; 4178 struct syn_cache_head *scp; 4179 struct mbuf *ipopts; 4180 int s; 4181 4182 tp = sototcpcb(so); 4183 4184 /* 4185 * Initialize some local state. 4186 */ 4187 win = sbspace(&so->so_rcv); 4188 if (win > TCP_MAXWIN) 4189 win = TCP_MAXWIN; 4190 4191 #ifdef TCP_SIGNATURE 4192 if (optp || (tp->t_flags & TF_SIGNATURE)) 4193 #else 4194 if (optp) 4195 #endif 4196 { 4197 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4198 #ifdef TCP_SIGNATURE 4199 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4200 #endif 4201 tb.t_state = TCPS_LISTEN; 4202 if (tcp_dooptions(&tb, optp, optlen, th, m, toff, oi) < 0) 4203 return 0; 4204 } else 4205 tb.t_flags = 0; 4206 4207 switch (src->sa_family) { 4208 case AF_INET: 4209 /* Remember the IP options, if any. */ 4210 ipopts = ip_srcroute(m); 4211 break; 4212 default: 4213 ipopts = NULL; 4214 } 4215 4216 /* 4217 * See if we already have an entry for this connection. 4218 * If we do, resend the SYN,ACK. We do not count this 4219 * as a retransmission (XXX though maybe we should). 4220 */ 4221 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4222 TCP_STATINC(TCP_STAT_SC_DUPESYN); 4223 if (ipopts) { 4224 /* 4225 * If we were remembering a previous source route, 4226 * forget it and use the new one we've been given. 4227 */ 4228 if (sc->sc_ipopts) 4229 (void)m_free(sc->sc_ipopts); 4230 sc->sc_ipopts = ipopts; 4231 } 4232 sc->sc_timestamp = tb.ts_recent; 4233 m_freem(m); 4234 if (syn_cache_respond(sc) == 0) { 4235 uint64_t *tcps = TCP_STAT_GETREF(); 4236 tcps[TCP_STAT_SNDACKS]++; 4237 tcps[TCP_STAT_SNDTOTAL]++; 4238 TCP_STAT_PUTREF(); 4239 } 4240 return 1; 4241 } 4242 4243 s = splsoftnet(); 4244 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4245 splx(s); 4246 if (sc == NULL) { 4247 if (ipopts) 4248 (void)m_free(ipopts); 4249 return 0; 4250 } 4251 4252 /* 4253 * Fill in the cache, and put the necessary IP and TCP 4254 * options into the reply. 4255 */ 4256 memset(sc, 0, sizeof(struct syn_cache)); 4257 callout_init(&sc->sc_timer, CALLOUT_MPSAFE); 4258 memcpy(&sc->sc_src, src, src->sa_len); 4259 memcpy(&sc->sc_dst, dst, dst->sa_len); 4260 sc->sc_flags = 0; 4261 sc->sc_ipopts = ipopts; 4262 sc->sc_irs = th->th_seq; 4263 switch (src->sa_family) { 4264 case AF_INET: 4265 { 4266 struct sockaddr_in *srcin = (void *)src; 4267 struct sockaddr_in *dstin = (void *)dst; 4268 4269 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4270 &srcin->sin_addr, dstin->sin_port, 4271 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4272 break; 4273 } 4274 #ifdef INET6 4275 case AF_INET6: 4276 { 4277 struct sockaddr_in6 *srcin6 = (void *)src; 4278 struct sockaddr_in6 *dstin6 = (void *)dst; 4279 4280 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4281 &srcin6->sin6_addr, dstin6->sin6_port, 4282 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4283 break; 4284 } 4285 #endif 4286 } 4287 sc->sc_peermaxseg = oi->maxseg; 4288 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4289 m_get_rcvif_NOMPSAFE(m) : NULL, sc->sc_src.sa.sa_family); 4290 sc->sc_win = win; 4291 sc->sc_timebase = tcp_now - 1; /* see tcp_newtcpcb() */ 4292 sc->sc_timestamp = tb.ts_recent; 4293 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4294 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4295 sc->sc_flags |= SCF_TIMESTAMP; 4296 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4297 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4298 sc->sc_requested_s_scale = tb.requested_s_scale; 4299 sc->sc_request_r_scale = 0; 4300 /* 4301 * Pick the smallest possible scaling factor that 4302 * will still allow us to scale up to sb_max. 4303 * 4304 * We do this because there are broken firewalls that 4305 * will corrupt the window scale option, leading to 4306 * the other endpoint believing that our advertised 4307 * window is unscaled. At scale factors larger than 4308 * 5 the unscaled window will drop below 1500 bytes, 4309 * leading to serious problems when traversing these 4310 * broken firewalls. 4311 * 4312 * With the default sbmax of 256K, a scale factor 4313 * of 3 will be chosen by this algorithm. Those who 4314 * choose a larger sbmax should watch out 4315 * for the compatiblity problems mentioned above. 4316 * 4317 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4318 * or <SYN,ACK>) segment itself is never scaled. 4319 */ 4320 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4321 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4322 sc->sc_request_r_scale++; 4323 } else { 4324 sc->sc_requested_s_scale = 15; 4325 sc->sc_request_r_scale = 15; 4326 } 4327 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4328 sc->sc_flags |= SCF_SACK_PERMIT; 4329 4330 /* 4331 * ECN setup packet received. 4332 */ 4333 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4334 sc->sc_flags |= SCF_ECN_PERMIT; 4335 4336 #ifdef TCP_SIGNATURE 4337 if (tb.t_flags & TF_SIGNATURE) 4338 sc->sc_flags |= SCF_SIGNATURE; 4339 #endif 4340 sc->sc_tp = tp; 4341 m_freem(m); 4342 if (syn_cache_respond(sc) == 0) { 4343 uint64_t *tcps = TCP_STAT_GETREF(); 4344 tcps[TCP_STAT_SNDACKS]++; 4345 tcps[TCP_STAT_SNDTOTAL]++; 4346 TCP_STAT_PUTREF(); 4347 syn_cache_insert(sc, tp); 4348 } else { 4349 s = splsoftnet(); 4350 /* 4351 * syn_cache_put() will try to schedule the timer, so 4352 * we need to initialize it 4353 */ 4354 syn_cache_timer_arm(sc); 4355 syn_cache_put(sc); 4356 splx(s); 4357 TCP_STATINC(TCP_STAT_SC_DROPPED); 4358 } 4359 return 1; 4360 } 4361 4362 /* 4363 * syn_cache_respond: (re)send SYN+ACK. 4364 * 4365 * Returns 0 on success. 4366 */ 4367 4368 int 4369 syn_cache_respond(struct syn_cache *sc) 4370 { 4371 #ifdef INET6 4372 struct rtentry *rt = NULL; 4373 #endif 4374 struct route *ro; 4375 u_int8_t *optp; 4376 int optlen, error; 4377 u_int16_t tlen; 4378 struct ip *ip = NULL; 4379 #ifdef INET6 4380 struct ip6_hdr *ip6 = NULL; 4381 #endif 4382 struct tcpcb *tp; 4383 struct tcphdr *th; 4384 struct mbuf *m; 4385 u_int hlen; 4386 #ifdef TCP_SIGNATURE 4387 struct secasvar *sav = NULL; 4388 u_int8_t *sigp = NULL; 4389 #endif 4390 4391 ro = &sc->sc_route; 4392 switch (sc->sc_src.sa.sa_family) { 4393 case AF_INET: 4394 hlen = sizeof(struct ip); 4395 break; 4396 #ifdef INET6 4397 case AF_INET6: 4398 hlen = sizeof(struct ip6_hdr); 4399 break; 4400 #endif 4401 default: 4402 return EAFNOSUPPORT; 4403 } 4404 4405 /* Worst case scanario, since we don't know the option size yet. */ 4406 tlen = hlen + sizeof(struct tcphdr) + MAX_TCPOPTLEN; 4407 KASSERT(max_linkhdr + tlen <= MCLBYTES); 4408 4409 /* 4410 * Create the IP+TCP header from scratch. 4411 */ 4412 MGETHDR(m, M_DONTWAIT, MT_DATA); 4413 if (m && (max_linkhdr + tlen) > MHLEN) { 4414 MCLGET(m, M_DONTWAIT); 4415 if ((m->m_flags & M_EXT) == 0) { 4416 m_freem(m); 4417 m = NULL; 4418 } 4419 } 4420 if (m == NULL) 4421 return ENOBUFS; 4422 MCLAIM(m, &tcp_tx_mowner); 4423 4424 tp = sc->sc_tp; 4425 4426 /* Fixup the mbuf. */ 4427 m->m_data += max_linkhdr; 4428 m_reset_rcvif(m); 4429 memset(mtod(m, void *), 0, tlen); 4430 4431 switch (sc->sc_src.sa.sa_family) { 4432 case AF_INET: 4433 ip = mtod(m, struct ip *); 4434 ip->ip_v = 4; 4435 ip->ip_dst = sc->sc_src.sin.sin_addr; 4436 ip->ip_src = sc->sc_dst.sin.sin_addr; 4437 ip->ip_p = IPPROTO_TCP; 4438 th = (struct tcphdr *)(ip + 1); 4439 th->th_dport = sc->sc_src.sin.sin_port; 4440 th->th_sport = sc->sc_dst.sin.sin_port; 4441 break; 4442 #ifdef INET6 4443 case AF_INET6: 4444 ip6 = mtod(m, struct ip6_hdr *); 4445 ip6->ip6_vfc = IPV6_VERSION; 4446 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4447 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4448 ip6->ip6_nxt = IPPROTO_TCP; 4449 /* ip6_plen will be updated in ip6_output() */ 4450 th = (struct tcphdr *)(ip6 + 1); 4451 th->th_dport = sc->sc_src.sin6.sin6_port; 4452 th->th_sport = sc->sc_dst.sin6.sin6_port; 4453 break; 4454 #endif 4455 default: 4456 panic("%s: impossible (1)", __func__); 4457 } 4458 4459 th->th_seq = htonl(sc->sc_iss); 4460 th->th_ack = htonl(sc->sc_irs + 1); 4461 th->th_flags = TH_SYN|TH_ACK; 4462 th->th_win = htons(sc->sc_win); 4463 /* th_x2, th_sum, th_urp already 0 from memset */ 4464 4465 /* Tack on the TCP options. */ 4466 optp = (u_int8_t *)(th + 1); 4467 optlen = 0; 4468 *optp++ = TCPOPT_MAXSEG; 4469 *optp++ = TCPOLEN_MAXSEG; 4470 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4471 *optp++ = sc->sc_ourmaxseg & 0xff; 4472 optlen += TCPOLEN_MAXSEG; 4473 4474 if (sc->sc_request_r_scale != 15) { 4475 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4476 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4477 sc->sc_request_r_scale); 4478 optp += TCPOLEN_WINDOW + TCPOLEN_NOP; 4479 optlen += TCPOLEN_WINDOW + TCPOLEN_NOP; 4480 } 4481 4482 if (sc->sc_flags & SCF_SACK_PERMIT) { 4483 /* Let the peer know that we will SACK. */ 4484 *optp++ = TCPOPT_SACK_PERMITTED; 4485 *optp++ = TCPOLEN_SACK_PERMITTED; 4486 optlen += TCPOLEN_SACK_PERMITTED; 4487 } 4488 4489 if (sc->sc_flags & SCF_TIMESTAMP) { 4490 while (optlen % 4 != 2) { 4491 optlen += TCPOLEN_NOP; 4492 *optp++ = TCPOPT_NOP; 4493 } 4494 *optp++ = TCPOPT_TIMESTAMP; 4495 *optp++ = TCPOLEN_TIMESTAMP; 4496 u_int32_t *lp = (u_int32_t *)(optp); 4497 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4498 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4499 *lp = htonl(sc->sc_timestamp); 4500 optp += TCPOLEN_TIMESTAMP - 2; 4501 optlen += TCPOLEN_TIMESTAMP; 4502 } 4503 4504 #ifdef TCP_SIGNATURE 4505 if (sc->sc_flags & SCF_SIGNATURE) { 4506 sav = tcp_signature_getsav(m); 4507 if (sav == NULL) { 4508 m_freem(m); 4509 return EPERM; 4510 } 4511 4512 *optp++ = TCPOPT_SIGNATURE; 4513 *optp++ = TCPOLEN_SIGNATURE; 4514 sigp = optp; 4515 memset(optp, 0, TCP_SIGLEN); 4516 optp += TCP_SIGLEN; 4517 optlen += TCPOLEN_SIGNATURE; 4518 } 4519 #endif 4520 4521 /* 4522 * Terminate and pad TCP options to a 4 byte boundary. 4523 * 4524 * According to RFC793: "The content of the header beyond the 4525 * End-of-Option option must be header padding (i.e., zero)." 4526 * And later: "The padding is composed of zeros." 4527 */ 4528 if (optlen % 4) { 4529 optlen += TCPOLEN_EOL; 4530 *optp++ = TCPOPT_EOL; 4531 } 4532 while (optlen % 4) { 4533 optlen += TCPOLEN_PAD; 4534 *optp++ = TCPOPT_PAD; 4535 } 4536 4537 /* Compute the actual values now that we've added the options. */ 4538 tlen = hlen + sizeof(struct tcphdr) + optlen; 4539 m->m_len = m->m_pkthdr.len = tlen; 4540 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4541 4542 #ifdef TCP_SIGNATURE 4543 if (sav) { 4544 (void)tcp_signature(m, th, hlen, sav, sigp); 4545 key_sa_recordxfer(sav, m); 4546 KEY_SA_UNREF(&sav); 4547 } 4548 #endif 4549 4550 /* 4551 * Send ECN SYN-ACK setup packet. 4552 * Routes can be asymetric, so, even if we receive a packet 4553 * with ECE and CWR set, we must not assume no one will block 4554 * the ECE packet we are about to send. 4555 */ 4556 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4557 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4558 th->th_flags |= TH_ECE; 4559 TCP_STATINC(TCP_STAT_ECN_SHS); 4560 4561 /* 4562 * draft-ietf-tcpm-ecnsyn-00.txt 4563 * 4564 * "[...] a TCP node MAY respond to an ECN-setup 4565 * SYN packet by setting ECT in the responding 4566 * ECN-setup SYN/ACK packet, indicating to routers 4567 * that the SYN/ACK packet is ECN-Capable. 4568 * This allows a congested router along the path 4569 * to mark the packet instead of dropping the 4570 * packet as an indication of congestion." 4571 * 4572 * "[...] There can be a great benefit in setting 4573 * an ECN-capable codepoint in SYN/ACK packets [...] 4574 * Congestion is most likely to occur in 4575 * the server-to-client direction. As a result, 4576 * setting an ECN-capable codepoint in SYN/ACK 4577 * packets can reduce the occurence of three-second 4578 * retransmit timeouts resulting from the drop 4579 * of SYN/ACK packets." 4580 * 4581 * Page 4 and 6, January 2006. 4582 */ 4583 4584 switch (sc->sc_src.sa.sa_family) { 4585 case AF_INET: 4586 ip->ip_tos |= IPTOS_ECN_ECT0; 4587 break; 4588 #ifdef INET6 4589 case AF_INET6: 4590 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4591 break; 4592 #endif 4593 } 4594 TCP_STATINC(TCP_STAT_ECN_ECT); 4595 } 4596 4597 4598 /* 4599 * Compute the packet's checksum. 4600 * 4601 * Fill in some straggling IP bits. Note the stack expects 4602 * ip_len to be in host order, for convenience. 4603 */ 4604 switch (sc->sc_src.sa.sa_family) { 4605 case AF_INET: 4606 ip->ip_len = htons(tlen - hlen); 4607 th->th_sum = 0; 4608 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4609 ip->ip_len = htons(tlen); 4610 ip->ip_ttl = ip_defttl; 4611 /* XXX tos? */ 4612 break; 4613 #ifdef INET6 4614 case AF_INET6: 4615 ip6->ip6_plen = htons(tlen - hlen); 4616 th->th_sum = 0; 4617 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4618 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4619 ip6->ip6_vfc |= IPV6_VERSION; 4620 ip6->ip6_plen = htons(tlen - hlen); 4621 /* ip6_hlim will be initialized afterwards */ 4622 /* XXX flowlabel? */ 4623 break; 4624 #endif 4625 } 4626 4627 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4628 tp = sc->sc_tp; 4629 4630 switch (sc->sc_src.sa.sa_family) { 4631 case AF_INET: 4632 error = ip_output(m, sc->sc_ipopts, ro, 4633 (ip_mtudisc ? IP_MTUDISC : 0), 4634 NULL, tp ? tp->t_inpcb : NULL); 4635 break; 4636 #ifdef INET6 4637 case AF_INET6: 4638 ip6->ip6_hlim = in6_selecthlim(NULL, 4639 (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp : NULL); 4640 rtcache_unref(rt, ro); 4641 4642 error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, 4643 tp ? tp->t_in6pcb : NULL, NULL); 4644 break; 4645 #endif 4646 default: 4647 panic("%s: impossible (2)", __func__); 4648 } 4649 4650 return error; 4651 } 4652
Indexes created Wed Sep 24 05:09:52 GMT 2025