tcp_input.c revision 1.386
1 /* $NetBSD: tcp_input.c,v 1.386 2018/03/22 21:19:28 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 1.386 2018/03/22 21:19:28 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 #ifndef INET 191 #include <netinet/in.h> 192 #endif 193 #include <netinet/ip6.h> 194 #include <netinet6/ip6_var.h> 195 #include <netinet6/in6_pcb.h> 196 #include <netinet6/ip6_var.h> 197 #include <netinet6/in6_var.h> 198 #include <netinet/icmp6.h> 199 #include <netinet6/nd6.h> 200 #ifdef TCP_SIGNATURE 201 #include <netinet6/scope6_var.h> 202 #endif 203 #endif 204 205 #ifndef INET6 206 /* always need ip6.h for IP6_EXTHDR_GET */ 207 #include <netinet/ip6.h> 208 #endif 209 210 #include <netinet/tcp.h> 211 #include <netinet/tcp_fsm.h> 212 #include <netinet/tcp_seq.h> 213 #include <netinet/tcp_timer.h> 214 #include <netinet/tcp_var.h> 215 #include <netinet/tcp_private.h> 216 #include <netinet/tcpip.h> 217 #include <netinet/tcp_congctl.h> 218 #include <netinet/tcp_debug.h> 219 220 #ifdef INET6 221 #include "faith.h" 222 #if defined(NFAITH) && NFAITH > 0 223 #include <net/if_faith.h> 224 #endif 225 #endif /* INET6 */ 226 227 #ifdef IPSEC 228 #include <netipsec/ipsec.h> 229 #include <netipsec/ipsec_var.h> 230 #include <netipsec/key.h> 231 #ifdef INET6 232 #include <netipsec/ipsec6.h> 233 #endif 234 #endif /* IPSEC*/ 235 236 #include <netinet/tcp_vtw.h> 237 238 int tcprexmtthresh = 3; 239 int tcp_log_refused; 240 241 int tcp_do_autorcvbuf = 1; 242 int tcp_autorcvbuf_inc = 16 * 1024; 243 int tcp_autorcvbuf_max = 256 * 1024; 244 int tcp_msl = (TCPTV_MSL / PR_SLOWHZ); 245 246 static int tcp_rst_ppslim_count = 0; 247 static struct timeval tcp_rst_ppslim_last; 248 static int tcp_ackdrop_ppslim_count = 0; 249 static struct timeval tcp_ackdrop_ppslim_last; 250 251 static void syn_cache_timer(void *); 252 253 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 254 255 /* for modulo comparisons of timestamps */ 256 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 257 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 258 259 /* 260 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 261 */ 262 #ifdef INET6 263 static inline void 264 nd6_hint(struct tcpcb *tp) 265 { 266 struct rtentry *rt = NULL; 267 268 if (tp != NULL && tp->t_in6pcb != NULL && tp->t_family == AF_INET6 && 269 (rt = rtcache_validate(&tp->t_in6pcb->in6p_route)) != NULL) 270 nd6_nud_hint(rt); 271 rtcache_unref(rt, &tp->t_in6pcb->in6p_route); 272 } 273 #else 274 static inline void 275 nd6_hint(struct tcpcb *tp) 276 { 277 } 278 #endif 279 280 /* 281 * Compute ACK transmission behavior. Delay the ACK unless 282 * we have already delayed an ACK (must send an ACK every two segments). 283 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 284 * option is enabled. 285 */ 286 static void 287 tcp_setup_ack(struct tcpcb *tp, const struct tcphdr *th) 288 { 289 290 if (tp->t_flags & TF_DELACK || 291 (tcp_ack_on_push && th->th_flags & TH_PUSH)) 292 tp->t_flags |= TF_ACKNOW; 293 else 294 TCP_SET_DELACK(tp); 295 } 296 297 static void 298 icmp_check(struct tcpcb *tp, const struct tcphdr *th, int acked) 299 { 300 301 /* 302 * If we had a pending ICMP message that refers to data that have 303 * just been acknowledged, disregard the recorded ICMP message. 304 */ 305 if ((tp->t_flags & TF_PMTUD_PEND) && 306 SEQ_GT(th->th_ack, tp->t_pmtud_th_seq)) 307 tp->t_flags &= ~TF_PMTUD_PEND; 308 309 /* 310 * Keep track of the largest chunk of data 311 * acknowledged since last PMTU update 312 */ 313 if (tp->t_pmtud_mss_acked < acked) 314 tp->t_pmtud_mss_acked = acked; 315 } 316 317 /* 318 * Convert TCP protocol fields to host order for easier processing. 319 */ 320 static void 321 tcp_fields_to_host(struct tcphdr *th) 322 { 323 324 NTOHL(th->th_seq); 325 NTOHL(th->th_ack); 326 NTOHS(th->th_win); 327 NTOHS(th->th_urp); 328 } 329 330 /* 331 * ... and reverse the above. 332 */ 333 static void 334 tcp_fields_to_net(struct tcphdr *th) 335 { 336 337 HTONL(th->th_seq); 338 HTONL(th->th_ack); 339 HTONS(th->th_win); 340 HTONS(th->th_urp); 341 } 342 343 #ifdef TCP_CSUM_COUNTERS 344 #include <sys/device.h> 345 346 #if defined(INET) 347 extern struct evcnt tcp_hwcsum_ok; 348 extern struct evcnt tcp_hwcsum_bad; 349 extern struct evcnt tcp_hwcsum_data; 350 extern struct evcnt tcp_swcsum; 351 #endif /* defined(INET) */ 352 #if defined(INET6) 353 extern struct evcnt tcp6_hwcsum_ok; 354 extern struct evcnt tcp6_hwcsum_bad; 355 extern struct evcnt tcp6_hwcsum_data; 356 extern struct evcnt tcp6_swcsum; 357 #endif /* defined(INET6) */ 358 359 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 360 361 #else 362 363 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 364 365 #endif /* TCP_CSUM_COUNTERS */ 366 367 #ifdef TCP_REASS_COUNTERS 368 #include <sys/device.h> 369 370 extern struct evcnt tcp_reass_; 371 extern struct evcnt tcp_reass_empty; 372 extern struct evcnt tcp_reass_iteration[8]; 373 extern struct evcnt tcp_reass_prependfirst; 374 extern struct evcnt tcp_reass_prepend; 375 extern struct evcnt tcp_reass_insert; 376 extern struct evcnt tcp_reass_inserttail; 377 extern struct evcnt tcp_reass_append; 378 extern struct evcnt tcp_reass_appendtail; 379 extern struct evcnt tcp_reass_overlaptail; 380 extern struct evcnt tcp_reass_overlapfront; 381 extern struct evcnt tcp_reass_segdup; 382 extern struct evcnt tcp_reass_fragdup; 383 384 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 385 386 #else 387 388 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 389 390 #endif /* TCP_REASS_COUNTERS */ 391 392 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *, 393 int); 394 static int tcp_dooptions(struct tcpcb *, const u_char *, int, 395 struct tcphdr *, struct mbuf *, int, struct tcp_opt_info *); 396 397 #ifdef INET 398 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 399 #endif 400 #ifdef INET6 401 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 402 #endif 403 404 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 405 406 #if defined(MBUFTRACE) 407 struct mowner tcp_reass_mowner = MOWNER_INIT("tcp", "reass"); 408 #endif /* defined(MBUFTRACE) */ 409 410 static struct pool tcpipqent_pool; 411 412 void 413 tcpipqent_init(void) 414 { 415 416 pool_init(&tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", 417 NULL, IPL_VM); 418 } 419 420 struct ipqent * 421 tcpipqent_alloc(void) 422 { 423 struct ipqent *ipqe; 424 int s; 425 426 s = splvm(); 427 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 428 splx(s); 429 430 return ipqe; 431 } 432 433 void 434 tcpipqent_free(struct ipqent *ipqe) 435 { 436 int s; 437 438 s = splvm(); 439 pool_put(&tcpipqent_pool, ipqe); 440 splx(s); 441 } 442 443 static int 444 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int tlen) 445 { 446 struct ipqent *p, *q, *nq, *tiqe = NULL; 447 struct socket *so = NULL; 448 int pkt_flags; 449 tcp_seq pkt_seq; 450 unsigned pkt_len; 451 u_long rcvpartdupbyte = 0; 452 u_long rcvoobyte; 453 #ifdef TCP_REASS_COUNTERS 454 u_int count = 0; 455 #endif 456 uint64_t *tcps; 457 458 if (tp->t_inpcb) 459 so = tp->t_inpcb->inp_socket; 460 #ifdef INET6 461 else if (tp->t_in6pcb) 462 so = tp->t_in6pcb->in6p_socket; 463 #endif 464 465 TCP_REASS_LOCK_CHECK(tp); 466 467 /* 468 * Call with th==0 after become established to 469 * force pre-ESTABLISHED data up to user socket. 470 */ 471 if (th == 0) 472 goto present; 473 474 m_claimm(m, &tcp_reass_mowner); 475 476 rcvoobyte = tlen; 477 /* 478 * Copy these to local variables because the tcpiphdr 479 * gets munged while we are collapsing mbufs. 480 */ 481 pkt_seq = th->th_seq; 482 pkt_len = tlen; 483 pkt_flags = th->th_flags; 484 485 TCP_REASS_COUNTER_INCR(&tcp_reass_); 486 487 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 488 /* 489 * When we miss a packet, the vast majority of time we get 490 * packets that follow it in order. So optimize for that. 491 */ 492 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 493 p->ipqe_len += pkt_len; 494 p->ipqe_flags |= pkt_flags; 495 m_cat(p->ipre_mlast, m); 496 TRAVERSE(p->ipre_mlast); 497 m = NULL; 498 tiqe = p; 499 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 500 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 501 goto skip_replacement; 502 } 503 /* 504 * While we're here, if the pkt is completely beyond 505 * anything we have, just insert it at the tail. 506 */ 507 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 508 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 509 goto insert_it; 510 } 511 } 512 513 q = TAILQ_FIRST(&tp->segq); 514 515 if (q != NULL) { 516 /* 517 * If this segment immediately precedes the first out-of-order 518 * block, simply slap the segment in front of it and (mostly) 519 * skip the complicated logic. 520 */ 521 if (pkt_seq + pkt_len == q->ipqe_seq) { 522 q->ipqe_seq = pkt_seq; 523 q->ipqe_len += pkt_len; 524 q->ipqe_flags |= pkt_flags; 525 m_cat(m, q->ipqe_m); 526 q->ipqe_m = m; 527 q->ipre_mlast = m; /* last mbuf may have changed */ 528 TRAVERSE(q->ipre_mlast); 529 tiqe = q; 530 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 531 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 532 goto skip_replacement; 533 } 534 } else { 535 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 536 } 537 538 /* 539 * Find a segment which begins after this one does. 540 */ 541 for (p = NULL; q != NULL; q = nq) { 542 nq = TAILQ_NEXT(q, ipqe_q); 543 #ifdef TCP_REASS_COUNTERS 544 count++; 545 #endif 546 /* 547 * If the received segment is just right after this 548 * fragment, merge the two together and then check 549 * for further overlaps. 550 */ 551 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 552 #ifdef TCPREASS_DEBUG 553 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 554 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 555 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 556 #endif 557 pkt_len += q->ipqe_len; 558 pkt_flags |= q->ipqe_flags; 559 pkt_seq = q->ipqe_seq; 560 m_cat(q->ipre_mlast, m); 561 TRAVERSE(q->ipre_mlast); 562 m = q->ipqe_m; 563 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 564 goto free_ipqe; 565 } 566 /* 567 * If the received segment is completely past this 568 * fragment, we need to go the next fragment. 569 */ 570 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 571 p = q; 572 continue; 573 } 574 /* 575 * If the fragment is past the received segment, 576 * it (or any following) can't be concatenated. 577 */ 578 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 579 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 580 break; 581 } 582 583 /* 584 * We've received all the data in this segment before. 585 * mark it as a duplicate and return. 586 */ 587 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 588 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 589 tcps = TCP_STAT_GETREF(); 590 tcps[TCP_STAT_RCVDUPPACK]++; 591 tcps[TCP_STAT_RCVDUPBYTE] += pkt_len; 592 TCP_STAT_PUTREF(); 593 tcp_new_dsack(tp, pkt_seq, pkt_len); 594 m_freem(m); 595 if (tiqe != NULL) { 596 tcpipqent_free(tiqe); 597 } 598 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 599 goto out; 600 } 601 /* 602 * Received segment completely overlaps this fragment 603 * so we drop the fragment (this keeps the temporal 604 * ordering of segments correct). 605 */ 606 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 607 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 608 rcvpartdupbyte += q->ipqe_len; 609 m_freem(q->ipqe_m); 610 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 611 goto free_ipqe; 612 } 613 /* 614 * RX'ed segment extends past the end of the 615 * fragment. Drop the overlapping bytes. Then 616 * merge the fragment and segment then treat as 617 * a longer received packet. 618 */ 619 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 620 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 621 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 622 #ifdef TCPREASS_DEBUG 623 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 624 tp, overlap, 625 pkt_seq, pkt_seq + pkt_len, pkt_len); 626 #endif 627 m_adj(m, overlap); 628 rcvpartdupbyte += overlap; 629 m_cat(q->ipre_mlast, m); 630 TRAVERSE(q->ipre_mlast); 631 m = q->ipqe_m; 632 pkt_seq = q->ipqe_seq; 633 pkt_len += q->ipqe_len - overlap; 634 rcvoobyte -= overlap; 635 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 636 goto free_ipqe; 637 } 638 /* 639 * RX'ed segment extends past the front of the 640 * fragment. Drop the overlapping bytes on the 641 * received packet. The packet will then be 642 * contatentated with this fragment a bit later. 643 */ 644 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 645 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 646 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 647 #ifdef TCPREASS_DEBUG 648 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 649 tp, overlap, 650 pkt_seq, pkt_seq + pkt_len, pkt_len); 651 #endif 652 m_adj(m, -overlap); 653 pkt_len -= overlap; 654 rcvpartdupbyte += overlap; 655 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 656 rcvoobyte -= overlap; 657 } 658 /* 659 * If the received segment immediates precedes this 660 * fragment then tack the fragment onto this segment 661 * and reinsert the data. 662 */ 663 if (q->ipqe_seq == pkt_seq + pkt_len) { 664 #ifdef TCPREASS_DEBUG 665 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 666 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 667 pkt_seq, pkt_seq + pkt_len, pkt_len); 668 #endif 669 pkt_len += q->ipqe_len; 670 pkt_flags |= q->ipqe_flags; 671 m_cat(m, q->ipqe_m); 672 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 673 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 674 tp->t_segqlen--; 675 KASSERT(tp->t_segqlen >= 0); 676 KASSERT(tp->t_segqlen != 0 || 677 (TAILQ_EMPTY(&tp->segq) && 678 TAILQ_EMPTY(&tp->timeq))); 679 if (tiqe == NULL) { 680 tiqe = q; 681 } else { 682 tcpipqent_free(q); 683 } 684 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 685 break; 686 } 687 /* 688 * If the fragment is before the segment, remember it. 689 * When this loop is terminated, p will contain the 690 * pointer to fragment that is right before the received 691 * segment. 692 */ 693 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 694 p = q; 695 696 continue; 697 698 /* 699 * This is a common operation. It also will allow 700 * to save doing a malloc/free in most instances. 701 */ 702 free_ipqe: 703 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 704 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 705 tp->t_segqlen--; 706 KASSERT(tp->t_segqlen >= 0); 707 KASSERT(tp->t_segqlen != 0 || 708 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 709 if (tiqe == NULL) { 710 tiqe = q; 711 } else { 712 tcpipqent_free(q); 713 } 714 } 715 716 #ifdef TCP_REASS_COUNTERS 717 if (count > 7) 718 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 719 else if (count > 0) 720 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 721 #endif 722 723 insert_it: 724 725 /* 726 * Allocate a new queue entry since the received segment did not 727 * collapse onto any other out-of-order block; thus we are allocating 728 * a new block. If it had collapsed, tiqe would not be NULL and 729 * we would be reusing it. 730 * XXX If we can't, just drop the packet. XXX 731 */ 732 if (tiqe == NULL) { 733 tiqe = tcpipqent_alloc(); 734 if (tiqe == NULL) { 735 TCP_STATINC(TCP_STAT_RCVMEMDROP); 736 m_freem(m); 737 goto out; 738 } 739 } 740 741 /* 742 * Update the counters. 743 */ 744 tp->t_rcvoopack++; 745 tcps = TCP_STAT_GETREF(); 746 tcps[TCP_STAT_RCVOOPACK]++; 747 tcps[TCP_STAT_RCVOOBYTE] += rcvoobyte; 748 if (rcvpartdupbyte) { 749 tcps[TCP_STAT_RCVPARTDUPPACK]++; 750 tcps[TCP_STAT_RCVPARTDUPBYTE] += rcvpartdupbyte; 751 } 752 TCP_STAT_PUTREF(); 753 754 /* 755 * Insert the new fragment queue entry into both queues. 756 */ 757 tiqe->ipqe_m = m; 758 tiqe->ipre_mlast = m; 759 tiqe->ipqe_seq = pkt_seq; 760 tiqe->ipqe_len = pkt_len; 761 tiqe->ipqe_flags = pkt_flags; 762 if (p == NULL) { 763 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 764 #ifdef TCPREASS_DEBUG 765 if (tiqe->ipqe_seq != tp->rcv_nxt) 766 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 767 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 768 #endif 769 } else { 770 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 771 #ifdef TCPREASS_DEBUG 772 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 773 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 774 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 775 #endif 776 } 777 tp->t_segqlen++; 778 779 skip_replacement: 780 781 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 782 783 present: 784 /* 785 * Present data to user, advancing rcv_nxt through 786 * completed sequence space. 787 */ 788 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 789 goto out; 790 q = TAILQ_FIRST(&tp->segq); 791 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 792 goto out; 793 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 794 goto out; 795 796 tp->rcv_nxt += q->ipqe_len; 797 pkt_flags = q->ipqe_flags & TH_FIN; 798 nd6_hint(tp); 799 800 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 801 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 802 tp->t_segqlen--; 803 KASSERT(tp->t_segqlen >= 0); 804 KASSERT(tp->t_segqlen != 0 || 805 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 806 if (so->so_state & SS_CANTRCVMORE) 807 m_freem(q->ipqe_m); 808 else 809 sbappendstream(&so->so_rcv, q->ipqe_m); 810 tcpipqent_free(q); 811 TCP_REASS_UNLOCK(tp); 812 sorwakeup(so); 813 return (pkt_flags); 814 out: 815 TCP_REASS_UNLOCK(tp); 816 return (0); 817 } 818 819 #ifdef INET6 820 int 821 tcp6_input(struct mbuf **mp, int *offp, int proto) 822 { 823 struct mbuf *m = *mp; 824 825 /* 826 * draft-itojun-ipv6-tcp-to-anycast 827 * better place to put this in? 828 */ 829 if (m->m_flags & M_ANYCAST6) { 830 struct ip6_hdr *ip6; 831 if (m->m_len < sizeof(struct ip6_hdr)) { 832 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 833 TCP_STATINC(TCP_STAT_RCVSHORT); 834 return IPPROTO_DONE; 835 } 836 } 837 ip6 = mtod(m, struct ip6_hdr *); 838 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 839 (char *)&ip6->ip6_dst - (char *)ip6); 840 return IPPROTO_DONE; 841 } 842 843 tcp_input(m, *offp, proto); 844 return IPPROTO_DONE; 845 } 846 #endif 847 848 #ifdef INET 849 static void 850 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 851 { 852 char src[INET_ADDRSTRLEN]; 853 char dst[INET_ADDRSTRLEN]; 854 855 if (ip) { 856 in_print(src, sizeof(src), &ip->ip_src); 857 in_print(dst, sizeof(dst), &ip->ip_dst); 858 } 859 else { 860 strlcpy(src, "(unknown)", sizeof(src)); 861 strlcpy(dst, "(unknown)", sizeof(dst)); 862 } 863 log(LOG_INFO, 864 "Connection attempt to TCP %s:%d from %s:%d\n", 865 dst, ntohs(th->th_dport), 866 src, ntohs(th->th_sport)); 867 } 868 #endif 869 870 #ifdef INET6 871 static void 872 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 873 { 874 char src[INET6_ADDRSTRLEN]; 875 char dst[INET6_ADDRSTRLEN]; 876 877 if (ip6) { 878 in6_print(src, sizeof(src), &ip6->ip6_src); 879 in6_print(dst, sizeof(dst), &ip6->ip6_dst); 880 } 881 else { 882 strlcpy(src, "(unknown v6)", sizeof(src)); 883 strlcpy(dst, "(unknown v6)", sizeof(dst)); 884 } 885 log(LOG_INFO, 886 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 887 dst, ntohs(th->th_dport), 888 src, ntohs(th->th_sport)); 889 } 890 #endif 891 892 /* 893 * Checksum extended TCP header and data. 894 */ 895 int 896 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, 897 int toff, int off, int tlen) 898 { 899 struct ifnet *rcvif; 900 int s; 901 902 /* 903 * XXX it's better to record and check if this mbuf is 904 * already checked. 905 */ 906 907 rcvif = m_get_rcvif(m, &s); 908 if (__predict_false(rcvif == NULL)) 909 goto badcsum; /* XXX */ 910 911 switch (af) { 912 #ifdef INET 913 case AF_INET: 914 switch (m->m_pkthdr.csum_flags & 915 ((rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 916 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 917 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 918 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 919 goto badcsum; 920 921 case M_CSUM_TCPv4|M_CSUM_DATA: { 922 u_int32_t hw_csum = m->m_pkthdr.csum_data; 923 924 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 925 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 926 const struct ip *ip = 927 mtod(m, const struct ip *); 928 929 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 930 ip->ip_dst.s_addr, 931 htons(hw_csum + tlen + off + IPPROTO_TCP)); 932 } 933 if ((hw_csum ^ 0xffff) != 0) 934 goto badcsum; 935 break; 936 } 937 938 case M_CSUM_TCPv4: 939 /* Checksum was okay. */ 940 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 941 break; 942 943 default: 944 /* 945 * Must compute it ourselves. Maybe skip checksum 946 * on loopback interfaces. 947 */ 948 if (__predict_true(!(rcvif->if_flags & IFF_LOOPBACK) || 949 tcp_do_loopback_cksum)) { 950 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 951 if (in4_cksum(m, IPPROTO_TCP, toff, 952 tlen + off) != 0) 953 goto badcsum; 954 } 955 break; 956 } 957 break; 958 #endif /* INET4 */ 959 960 #ifdef INET6 961 case AF_INET6: 962 switch (m->m_pkthdr.csum_flags & 963 ((rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 964 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 965 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 966 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 967 goto badcsum; 968 969 #if 0 /* notyet */ 970 case M_CSUM_TCPv6|M_CSUM_DATA: 971 #endif 972 973 case M_CSUM_TCPv6: 974 /* Checksum was okay. */ 975 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 976 break; 977 978 default: 979 /* 980 * Must compute it ourselves. Maybe skip checksum 981 * on loopback interfaces. 982 */ 983 if (__predict_true((m->m_flags & M_LOOP) == 0 || 984 tcp_do_loopback_cksum)) { 985 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 986 if (in6_cksum(m, IPPROTO_TCP, toff, 987 tlen + off) != 0) 988 goto badcsum; 989 } 990 } 991 break; 992 #endif /* INET6 */ 993 } 994 m_put_rcvif(rcvif, &s); 995 996 return 0; 997 998 badcsum: 999 m_put_rcvif(rcvif, &s); 1000 TCP_STATINC(TCP_STAT_RCVBADSUM); 1001 return -1; 1002 } 1003 1004 /* When a packet arrives addressed to a vestigial tcpbp, we 1005 * nevertheless have to respond to it per the spec. 1006 */ 1007 static void tcp_vtw_input(struct tcphdr *th, vestigial_inpcb_t *vp, 1008 struct mbuf *m, int tlen) 1009 { 1010 int tiflags; 1011 int todrop; 1012 uint32_t t_flags = 0; 1013 uint64_t *tcps; 1014 1015 tiflags = th->th_flags; 1016 todrop = vp->rcv_nxt - th->th_seq; 1017 1018 if (todrop > 0) { 1019 if (tiflags & TH_SYN) { 1020 tiflags &= ~TH_SYN; 1021 ++th->th_seq; 1022 if (th->th_urp > 1) 1023 --th->th_urp; 1024 else { 1025 tiflags &= ~TH_URG; 1026 th->th_urp = 0; 1027 } 1028 --todrop; 1029 } 1030 if (todrop > tlen || 1031 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1032 /* 1033 * Any valid FIN or RST must be to the left of the 1034 * window. At this point the FIN or RST must be a 1035 * duplicate or out of sequence; drop it. 1036 */ 1037 if (tiflags & TH_RST) 1038 goto drop; 1039 tiflags &= ~(TH_FIN|TH_RST); 1040 /* 1041 * Send an ACK to resynchronize and drop any data. 1042 * But keep on processing for RST or ACK. 1043 */ 1044 t_flags |= TF_ACKNOW; 1045 todrop = tlen; 1046 tcps = TCP_STAT_GETREF(); 1047 tcps[TCP_STAT_RCVDUPPACK] += 1; 1048 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 1049 TCP_STAT_PUTREF(); 1050 } else if ((tiflags & TH_RST) 1051 && th->th_seq != vp->rcv_nxt) { 1052 /* 1053 * Test for reset before adjusting the sequence 1054 * number for overlapping data. 1055 */ 1056 goto dropafterack_ratelim; 1057 } else { 1058 tcps = TCP_STAT_GETREF(); 1059 tcps[TCP_STAT_RCVPARTDUPPACK] += 1; 1060 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 1061 TCP_STAT_PUTREF(); 1062 } 1063 1064 // tcp_new_dsack(tp, th->th_seq, todrop); 1065 // hdroptlen += todrop; /*drop from head afterwards*/ 1066 1067 th->th_seq += todrop; 1068 tlen -= todrop; 1069 1070 if (th->th_urp > todrop) 1071 th->th_urp -= todrop; 1072 else { 1073 tiflags &= ~TH_URG; 1074 th->th_urp = 0; 1075 } 1076 } 1077 1078 /* 1079 * If new data are received on a connection after the 1080 * user processes are gone, then RST the other end. 1081 */ 1082 if (tlen) { 1083 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 1084 goto dropwithreset; 1085 } 1086 1087 /* 1088 * If segment ends after window, drop trailing data 1089 * (and PUSH and FIN); if nothing left, just ACK. 1090 */ 1091 todrop = (th->th_seq + tlen) - (vp->rcv_nxt+vp->rcv_wnd); 1092 1093 if (todrop > 0) { 1094 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 1095 if (todrop >= tlen) { 1096 /* 1097 * The segment actually starts after the window. 1098 * th->th_seq + tlen - vp->rcv_nxt - vp->rcv_wnd >= tlen 1099 * th->th_seq - vp->rcv_nxt - vp->rcv_wnd >= 0 1100 * th->th_seq >= vp->rcv_nxt + vp->rcv_wnd 1101 */ 1102 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 1103 /* 1104 * If a new connection request is received 1105 * while in TIME_WAIT, drop the old connection 1106 * and start over if the sequence numbers 1107 * are above the previous ones. 1108 */ 1109 if ((tiflags & TH_SYN) 1110 && SEQ_GT(th->th_seq, vp->rcv_nxt)) { 1111 /* We only support this in the !NOFDREF case, which 1112 * is to say: not here. 1113 */ 1114 goto dropwithreset; 1115 } 1116 /* 1117 * If window is closed can only take segments at 1118 * window edge, and have to drop data and PUSH from 1119 * incoming segments. Continue processing, but 1120 * remember to ack. Otherwise, drop segment 1121 * and (if not RST) ack. 1122 */ 1123 if (vp->rcv_wnd == 0 && th->th_seq == vp->rcv_nxt) { 1124 t_flags |= TF_ACKNOW; 1125 TCP_STATINC(TCP_STAT_RCVWINPROBE); 1126 } else 1127 goto dropafterack; 1128 } else 1129 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 1130 m_adj(m, -todrop); 1131 tlen -= todrop; 1132 tiflags &= ~(TH_PUSH|TH_FIN); 1133 } 1134 1135 if (tiflags & TH_RST) { 1136 if (th->th_seq != vp->rcv_nxt) 1137 goto dropafterack_ratelim; 1138 1139 vtw_del(vp->ctl, vp->vtw); 1140 goto drop; 1141 } 1142 1143 /* 1144 * If the ACK bit is off we drop the segment and return. 1145 */ 1146 if ((tiflags & TH_ACK) == 0) { 1147 if (t_flags & TF_ACKNOW) 1148 goto dropafterack; 1149 else 1150 goto drop; 1151 } 1152 1153 /* 1154 * In TIME_WAIT state the only thing that should arrive 1155 * is a retransmission of the remote FIN. Acknowledge 1156 * it and restart the finack timer. 1157 */ 1158 vtw_restart(vp); 1159 goto dropafterack; 1160 1161 dropafterack: 1162 /* 1163 * Generate an ACK dropping incoming segment if it occupies 1164 * sequence space, where the ACK reflects our state. 1165 */ 1166 if (tiflags & TH_RST) 1167 goto drop; 1168 goto dropafterack2; 1169 1170 dropafterack_ratelim: 1171 /* 1172 * We may want to rate-limit ACKs against SYN/RST attack. 1173 */ 1174 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 1175 tcp_ackdrop_ppslim) == 0) { 1176 /* XXX stat */ 1177 goto drop; 1178 } 1179 /* ...fall into dropafterack2... */ 1180 1181 dropafterack2: 1182 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, th->th_ack, 1183 TH_ACK); 1184 return; 1185 1186 dropwithreset: 1187 /* 1188 * Generate a RST, dropping incoming segment. 1189 * Make ACK acceptable to originator of segment. 1190 */ 1191 if (tiflags & TH_RST) 1192 goto drop; 1193 1194 if (tiflags & TH_ACK) 1195 tcp_respond(0, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 1196 else { 1197 if (tiflags & TH_SYN) 1198 ++tlen; 1199 (void)tcp_respond(0, m, m, th, th->th_seq + tlen, (tcp_seq)0, 1200 TH_RST|TH_ACK); 1201 } 1202 return; 1203 drop: 1204 m_freem(m); 1205 } 1206 1207 /* 1208 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 1209 */ 1210 void 1211 tcp_input(struct mbuf *m, ...) 1212 { 1213 struct tcphdr *th; 1214 struct ip *ip; 1215 struct inpcb *inp; 1216 #ifdef INET6 1217 struct ip6_hdr *ip6; 1218 struct in6pcb *in6p; 1219 #endif 1220 u_int8_t *optp = NULL; 1221 int optlen = 0; 1222 int len, tlen, toff, hdroptlen = 0; 1223 struct tcpcb *tp = NULL; 1224 int tiflags; 1225 struct socket *so = NULL; 1226 int todrop, acked, ourfinisacked, needoutput = 0; 1227 bool dupseg; 1228 #ifdef TCP_DEBUG 1229 short ostate = 0; 1230 #endif 1231 u_long tiwin; 1232 struct tcp_opt_info opti; 1233 int off, iphlen; 1234 va_list ap; 1235 int af; /* af on the wire */ 1236 struct mbuf *tcp_saveti = NULL; 1237 uint32_t ts_rtt; 1238 uint8_t iptos; 1239 uint64_t *tcps; 1240 vestigial_inpcb_t vestige; 1241 1242 vestige.valid = 0; 1243 1244 MCLAIM(m, &tcp_rx_mowner); 1245 va_start(ap, m); 1246 toff = va_arg(ap, int); 1247 (void)va_arg(ap, int); /* ignore value, advance ap */ 1248 va_end(ap); 1249 1250 TCP_STATINC(TCP_STAT_RCVTOTAL); 1251 1252 memset(&opti, 0, sizeof(opti)); 1253 opti.ts_present = 0; 1254 opti.maxseg = 0; 1255 1256 /* 1257 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 1258 * 1259 * TCP is, by definition, unicast, so we reject all 1260 * multicast outright. 1261 * 1262 * Note, there are additional src/dst address checks in 1263 * the AF-specific code below. 1264 */ 1265 if (m->m_flags & (M_BCAST|M_MCAST)) { 1266 /* XXX stat */ 1267 goto drop; 1268 } 1269 #ifdef INET6 1270 if (m->m_flags & M_ANYCAST6) { 1271 /* XXX stat */ 1272 goto drop; 1273 } 1274 #endif 1275 1276 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, sizeof(struct tcphdr)); 1277 if (th == NULL) { 1278 TCP_STATINC(TCP_STAT_RCVSHORT); 1279 return; 1280 } 1281 1282 /* 1283 * Get IP and TCP header. 1284 * Note: IP leaves IP header in first mbuf. 1285 */ 1286 ip = mtod(m, struct ip *); 1287 switch (ip->ip_v) { 1288 #ifdef INET 1289 case 4: 1290 #ifdef INET6 1291 ip6 = NULL; 1292 #endif 1293 af = AF_INET; 1294 iphlen = sizeof(struct ip); 1295 1296 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1297 in_broadcast(ip->ip_dst, m_get_rcvif_NOMPSAFE(m))) 1298 goto drop; 1299 1300 /* We do the checksum after PCB lookup... */ 1301 len = ntohs(ip->ip_len); 1302 tlen = len - toff; 1303 iptos = ip->ip_tos; 1304 break; 1305 #endif 1306 #ifdef INET6 1307 case 6: 1308 ip = NULL; 1309 iphlen = sizeof(struct ip6_hdr); 1310 af = AF_INET6; 1311 ip6 = mtod(m, struct ip6_hdr *); 1312 1313 /* 1314 * Be proactive about unspecified IPv6 address in source. 1315 * As we use all-zero to indicate unbounded/unconnected pcb, 1316 * unspecified IPv6 address can be used to confuse us. 1317 * 1318 * Note that packets with unspecified IPv6 destination is 1319 * already dropped in ip6_input. 1320 */ 1321 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1322 /* XXX stat */ 1323 goto drop; 1324 } 1325 1326 /* 1327 * Make sure destination address is not multicast. 1328 * Source address checked in ip6_input(). 1329 */ 1330 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1331 /* XXX stat */ 1332 goto drop; 1333 } 1334 1335 /* We do the checksum after PCB lookup... */ 1336 len = m->m_pkthdr.len; 1337 tlen = len - toff; 1338 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1339 break; 1340 #endif 1341 default: 1342 m_freem(m); 1343 return; 1344 } 1345 1346 /* 1347 * Enforce alignment requirements that are violated in 1348 * some cases, see kern/50766 for details. 1349 */ 1350 if (TCP_HDR_ALIGNED_P(th) == 0) { 1351 m = m_copyup(m, toff + sizeof(struct tcphdr), 0); 1352 if (m == NULL) { 1353 TCP_STATINC(TCP_STAT_RCVSHORT); 1354 return; 1355 } 1356 ip = mtod(m, struct ip *); 1357 #ifdef INET6 1358 ip6 = mtod(m, struct ip6_hdr *); 1359 #endif 1360 th = (struct tcphdr *)(mtod(m, char *) + toff); 1361 } 1362 KASSERT(TCP_HDR_ALIGNED_P(th)); 1363 1364 /* 1365 * Check that TCP offset makes sense, pull out TCP options and 1366 * adjust length. 1367 */ 1368 off = th->th_off << 2; 1369 if (off < sizeof(struct tcphdr) || off > tlen) { 1370 TCP_STATINC(TCP_STAT_RCVBADOFF); 1371 goto drop; 1372 } 1373 tlen -= off; 1374 1375 if (off > sizeof(struct tcphdr)) { 1376 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1377 if (th == NULL) { 1378 TCP_STATINC(TCP_STAT_RCVSHORT); 1379 return; 1380 } 1381 KASSERT(TCP_HDR_ALIGNED_P(th)); 1382 optlen = off - sizeof(struct tcphdr); 1383 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1384 /* 1385 * Do quick retrieval of timestamp options ("options 1386 * prediction?"). If timestamp is the only option and it's 1387 * formatted as recommended in RFC 1323 appendix A, we 1388 * quickly get the values now and not bother calling 1389 * tcp_dooptions(), etc. 1390 */ 1391 if ((optlen == TCPOLEN_TSTAMP_APPA || 1392 (optlen > TCPOLEN_TSTAMP_APPA && 1393 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1394 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1395 (th->th_flags & TH_SYN) == 0) { 1396 opti.ts_present = 1; 1397 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1398 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1399 optp = NULL; /* we've parsed the options */ 1400 } 1401 } 1402 tiflags = th->th_flags; 1403 1404 /* 1405 * Checksum extended TCP header and data 1406 */ 1407 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1408 goto badcsum; 1409 1410 /* 1411 * Locate pcb for segment. 1412 */ 1413 findpcb: 1414 inp = NULL; 1415 #ifdef INET6 1416 in6p = NULL; 1417 #endif 1418 switch (af) { 1419 #ifdef INET 1420 case AF_INET: 1421 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1422 ip->ip_dst, th->th_dport, &vestige); 1423 if (inp == NULL && !vestige.valid) { 1424 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1425 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, 1426 th->th_dport); 1427 } 1428 #ifdef INET6 1429 if (inp == NULL && !vestige.valid) { 1430 struct in6_addr s, d; 1431 1432 /* mapped addr case */ 1433 in6_in_2_v4mapin6(&ip->ip_src, &s); 1434 in6_in_2_v4mapin6(&ip->ip_dst, &d); 1435 in6p = in6_pcblookup_connect(&tcbtable, &s, 1436 th->th_sport, &d, th->th_dport, 0, &vestige); 1437 if (in6p == 0 && !vestige.valid) { 1438 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1439 in6p = in6_pcblookup_bind(&tcbtable, &d, 1440 th->th_dport, 0); 1441 } 1442 } 1443 #endif 1444 #ifndef INET6 1445 if (inp == NULL && !vestige.valid) 1446 #else 1447 if (inp == NULL && in6p == NULL && !vestige.valid) 1448 #endif 1449 { 1450 TCP_STATINC(TCP_STAT_NOPORT); 1451 if (tcp_log_refused && 1452 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1453 tcp4_log_refused(ip, th); 1454 } 1455 tcp_fields_to_host(th); 1456 goto dropwithreset_ratelim; 1457 } 1458 #if defined(IPSEC) 1459 if (ipsec_used) { 1460 if (inp && ipsec_in_reject(m, inp)) { 1461 goto drop; 1462 } 1463 #ifdef INET6 1464 else if (in6p && ipsec_in_reject(m, in6p)) { 1465 goto drop; 1466 } 1467 #endif 1468 } 1469 #endif /*IPSEC*/ 1470 break; 1471 #endif /*INET*/ 1472 #ifdef INET6 1473 case AF_INET6: 1474 { 1475 int faith; 1476 1477 #if defined(NFAITH) && NFAITH > 0 1478 faith = faithprefix(&ip6->ip6_dst); 1479 #else 1480 faith = 0; 1481 #endif 1482 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1483 th->th_sport, &ip6->ip6_dst, th->th_dport, faith, &vestige); 1484 if (!in6p && !vestige.valid) { 1485 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1486 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1487 th->th_dport, faith); 1488 } 1489 if (!in6p && !vestige.valid) { 1490 TCP_STATINC(TCP_STAT_NOPORT); 1491 if (tcp_log_refused && 1492 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1493 tcp6_log_refused(ip6, th); 1494 } 1495 tcp_fields_to_host(th); 1496 goto dropwithreset_ratelim; 1497 } 1498 #if defined(IPSEC) 1499 if (ipsec_used && in6p && ipsec_in_reject(m, in6p)) { 1500 goto drop; 1501 } 1502 #endif 1503 break; 1504 } 1505 #endif 1506 } 1507 1508 tcp_fields_to_host(th); 1509 1510 /* 1511 * If the state is CLOSED (i.e., TCB does not exist) then 1512 * all data in the incoming segment is discarded. 1513 * If the TCB exists but is in CLOSED state, it is embryonic, 1514 * but should either do a listen or a connect soon. 1515 */ 1516 tp = NULL; 1517 so = NULL; 1518 if (inp) { 1519 /* Check the minimum TTL for socket. */ 1520 if (ip->ip_ttl < inp->inp_ip_minttl) 1521 goto drop; 1522 1523 tp = intotcpcb(inp); 1524 so = inp->inp_socket; 1525 } 1526 #ifdef INET6 1527 else if (in6p) { 1528 tp = in6totcpcb(in6p); 1529 so = in6p->in6p_socket; 1530 } 1531 #endif 1532 else if (vestige.valid) { 1533 /* We do not support the resurrection of vtw tcpcps. */ 1534 tcp_vtw_input(th, &vestige, m, tlen); 1535 m = NULL; 1536 goto drop; 1537 } 1538 1539 if (tp == NULL) 1540 goto dropwithreset_ratelim; 1541 if (tp->t_state == TCPS_CLOSED) 1542 goto drop; 1543 1544 KASSERT(so->so_lock == softnet_lock); 1545 KASSERT(solocked(so)); 1546 1547 /* Unscale the window into a 32-bit value. */ 1548 if ((tiflags & TH_SYN) == 0) 1549 tiwin = th->th_win << tp->snd_scale; 1550 else 1551 tiwin = th->th_win; 1552 1553 #ifdef INET6 1554 /* save packet options if user wanted */ 1555 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1556 if (in6p->in6p_options) { 1557 m_freem(in6p->in6p_options); 1558 in6p->in6p_options = NULL; 1559 } 1560 KASSERT(ip6 != NULL); 1561 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1562 } 1563 #endif 1564 1565 if (so->so_options & SO_DEBUG) { 1566 #ifdef TCP_DEBUG 1567 ostate = tp->t_state; 1568 #endif 1569 1570 tcp_saveti = NULL; 1571 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1572 goto nosave; 1573 1574 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1575 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1576 if (tcp_saveti == NULL) 1577 goto nosave; 1578 } else { 1579 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1580 if (tcp_saveti == NULL) 1581 goto nosave; 1582 MCLAIM(m, &tcp_mowner); 1583 tcp_saveti->m_len = iphlen; 1584 m_copydata(m, 0, iphlen, 1585 mtod(tcp_saveti, void *)); 1586 } 1587 1588 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1589 m_freem(tcp_saveti); 1590 tcp_saveti = NULL; 1591 } else { 1592 tcp_saveti->m_len += sizeof(struct tcphdr); 1593 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1594 sizeof(struct tcphdr)); 1595 } 1596 nosave:; 1597 } 1598 1599 if (so->so_options & SO_ACCEPTCONN) { 1600 union syn_cache_sa src; 1601 union syn_cache_sa dst; 1602 1603 KASSERT(tp->t_state == TCPS_LISTEN); 1604 1605 memset(&src, 0, sizeof(src)); 1606 memset(&dst, 0, sizeof(dst)); 1607 switch (af) { 1608 #ifdef INET 1609 case AF_INET: 1610 src.sin.sin_len = sizeof(struct sockaddr_in); 1611 src.sin.sin_family = AF_INET; 1612 src.sin.sin_addr = ip->ip_src; 1613 src.sin.sin_port = th->th_sport; 1614 1615 dst.sin.sin_len = sizeof(struct sockaddr_in); 1616 dst.sin.sin_family = AF_INET; 1617 dst.sin.sin_addr = ip->ip_dst; 1618 dst.sin.sin_port = th->th_dport; 1619 break; 1620 #endif 1621 #ifdef INET6 1622 case AF_INET6: 1623 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1624 src.sin6.sin6_family = AF_INET6; 1625 src.sin6.sin6_addr = ip6->ip6_src; 1626 src.sin6.sin6_port = th->th_sport; 1627 1628 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1629 dst.sin6.sin6_family = AF_INET6; 1630 dst.sin6.sin6_addr = ip6->ip6_dst; 1631 dst.sin6.sin6_port = th->th_dport; 1632 break; 1633 #endif 1634 } 1635 1636 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1637 if (tiflags & TH_RST) { 1638 syn_cache_reset(&src.sa, &dst.sa, th); 1639 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1640 (TH_ACK|TH_SYN)) { 1641 /* 1642 * Received a SYN,ACK. This should never 1643 * happen while we are in LISTEN. Send an RST. 1644 */ 1645 goto badsyn; 1646 } else if (tiflags & TH_ACK) { 1647 so = syn_cache_get(&src.sa, &dst.sa, 1648 th, toff, tlen, 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 #ifdef INET 1677 case AF_INET: 1678 inp = sotoinpcb(so); 1679 tp = intotcpcb(inp); 1680 break; 1681 #endif 1682 #ifdef INET6 1683 case AF_INET6: 1684 in6p = sotoin6pcb(so); 1685 tp = in6totcpcb(in6p); 1686 break; 1687 #endif 1688 } 1689 if (tp == NULL) 1690 goto badsyn; /*XXX*/ 1691 tiwin <<= tp->snd_scale; 1692 goto after_listen; 1693 } 1694 } else { 1695 /* 1696 * None of RST, SYN or ACK was set. 1697 * This is an invalid packet for a 1698 * TCB in LISTEN state. Send a RST. 1699 */ 1700 goto badsyn; 1701 } 1702 } else { 1703 /* 1704 * Received a SYN. 1705 */ 1706 1707 #ifdef INET6 1708 /* 1709 * If deprecated address is forbidden, we do 1710 * not accept SYN to deprecated interface 1711 * address to prevent any new inbound 1712 * connection from getting established. 1713 * When we do not accept SYN, we send a TCP 1714 * RST, with deprecated source address (instead 1715 * of dropping it). We compromise it as it is 1716 * much better for peer to send a RST, and 1717 * RST will be the final packet for the 1718 * exchange. 1719 * 1720 * If we do not forbid deprecated addresses, we 1721 * accept the SYN packet. RFC2462 does not 1722 * suggest dropping SYN in this case. 1723 * If we decipher RFC2462 5.5.4, it says like 1724 * this: 1725 * 1. use of deprecated addr with existing 1726 * communication is okay - "SHOULD continue 1727 * to be used" 1728 * 2. use of it with new communication: 1729 * (2a) "SHOULD NOT be used if alternate 1730 * address with sufficient scope is 1731 * available" 1732 * (2b) nothing mentioned otherwise. 1733 * Here we fall into (2b) case as we have no 1734 * choice in our source address selection - we 1735 * must obey the peer. 1736 * 1737 * The wording in RFC2462 is confusing, and 1738 * there are multiple description text for 1739 * deprecated address handling - worse, they 1740 * are not exactly the same. I believe 5.5.4 1741 * is the best one, so we follow 5.5.4. 1742 */ 1743 if (af == AF_INET6 && !ip6_use_deprecated) { 1744 struct in6_ifaddr *ia6; 1745 int s; 1746 struct ifnet *rcvif = m_get_rcvif(m, &s); 1747 if (rcvif == NULL) 1748 goto dropwithreset; /* XXX */ 1749 if ((ia6 = in6ifa_ifpwithaddr(rcvif, 1750 &ip6->ip6_dst)) && 1751 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1752 tp = NULL; 1753 m_put_rcvif(rcvif, &s); 1754 goto dropwithreset; 1755 } 1756 m_put_rcvif(rcvif, &s); 1757 } 1758 #endif 1759 1760 /* 1761 * LISTEN socket received a SYN from itself? This 1762 * can't possibly be valid; drop the packet. 1763 */ 1764 if (th->th_sport == th->th_dport) { 1765 int eq = 0; 1766 1767 switch (af) { 1768 #ifdef INET 1769 case AF_INET: 1770 eq = in_hosteq(ip->ip_src, ip->ip_dst); 1771 break; 1772 #endif 1773 #ifdef INET6 1774 case AF_INET6: 1775 eq = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, 1776 &ip6->ip6_dst); 1777 break; 1778 #endif 1779 } 1780 if (eq) { 1781 TCP_STATINC(TCP_STAT_BADSYN); 1782 goto drop; 1783 } 1784 } 1785 1786 /* 1787 * SYN looks ok; create compressed TCP 1788 * state for it. 1789 */ 1790 if (so->so_qlen <= so->so_qlimit && 1791 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1792 so, m, optp, optlen, &opti)) 1793 m = NULL; 1794 } 1795 1796 goto drop; 1797 } 1798 1799 after_listen: 1800 /* 1801 * From here on, we're dealing with !LISTEN. 1802 */ 1803 KASSERT(tp->t_state != TCPS_LISTEN); 1804 1805 /* 1806 * Segment received on connection. 1807 * Reset idle time and keep-alive timer. 1808 */ 1809 tp->t_rcvtime = tcp_now; 1810 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1811 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1812 1813 /* 1814 * Process options. 1815 */ 1816 #ifdef TCP_SIGNATURE 1817 if (optp || (tp->t_flags & TF_SIGNATURE)) 1818 #else 1819 if (optp) 1820 #endif 1821 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1822 goto drop; 1823 1824 if (TCP_SACK_ENABLED(tp)) { 1825 tcp_del_sackholes(tp, th); 1826 } 1827 1828 if (TCP_ECN_ALLOWED(tp)) { 1829 if (tiflags & TH_CWR) { 1830 tp->t_flags &= ~TF_ECN_SND_ECE; 1831 } 1832 switch (iptos & IPTOS_ECN_MASK) { 1833 case IPTOS_ECN_CE: 1834 tp->t_flags |= TF_ECN_SND_ECE; 1835 TCP_STATINC(TCP_STAT_ECN_CE); 1836 break; 1837 case IPTOS_ECN_ECT0: 1838 TCP_STATINC(TCP_STAT_ECN_ECT); 1839 break; 1840 case IPTOS_ECN_ECT1: 1841 /* XXX: ignore for now -- rpaulo */ 1842 break; 1843 } 1844 /* 1845 * Congestion experienced. 1846 * Ignore if we are already trying to recover. 1847 */ 1848 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1849 tp->t_congctl->cong_exp(tp); 1850 } 1851 1852 if (opti.ts_present && opti.ts_ecr) { 1853 /* 1854 * Calculate the RTT from the returned time stamp and the 1855 * connection's time base. If the time stamp is later than 1856 * the current time, or is extremely old, fall back to non-1323 1857 * RTT calculation. Since ts_rtt is unsigned, we can test both 1858 * at the same time. 1859 * 1860 * Note that ts_rtt is in units of slow ticks (500 1861 * ms). Since most earthbound RTTs are < 500 ms, 1862 * observed values will have large quantization noise. 1863 * Our smoothed RTT is then the fraction of observed 1864 * samples that are 1 tick instead of 0 (times 500 1865 * ms). 1866 * 1867 * ts_rtt is increased by 1 to denote a valid sample, 1868 * with 0 indicating an invalid measurement. This 1869 * extra 1 must be removed when ts_rtt is used, or 1870 * else an an erroneous extra 500 ms will result. 1871 */ 1872 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1873 if (ts_rtt > TCP_PAWS_IDLE) 1874 ts_rtt = 0; 1875 } else { 1876 ts_rtt = 0; 1877 } 1878 1879 /* 1880 * Header prediction: check for the two common cases 1881 * of a uni-directional data xfer. If the packet has 1882 * no control flags, is in-sequence, the window didn't 1883 * change and we're not retransmitting, it's a 1884 * candidate. If the length is zero and the ack moved 1885 * forward, we're the sender side of the xfer. Just 1886 * free the data acked & wake any higher level process 1887 * that was blocked waiting for space. If the length 1888 * is non-zero and the ack didn't move, we're the 1889 * receiver side. If we're getting packets in-order 1890 * (the reassembly queue is empty), add the data to 1891 * the socket buffer and note that we need a delayed ack. 1892 */ 1893 if (tp->t_state == TCPS_ESTABLISHED && 1894 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1895 == TH_ACK && 1896 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1897 th->th_seq == tp->rcv_nxt && 1898 tiwin && tiwin == tp->snd_wnd && 1899 tp->snd_nxt == tp->snd_max) { 1900 1901 /* 1902 * If last ACK falls within this segment's sequence numbers, 1903 * record the timestamp. 1904 * NOTE that the test is modified according to the latest 1905 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 1906 * 1907 * note that we already know 1908 * TSTMP_GEQ(opti.ts_val, tp->ts_recent) 1909 */ 1910 if (opti.ts_present && 1911 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1912 tp->ts_recent_age = tcp_now; 1913 tp->ts_recent = opti.ts_val; 1914 } 1915 1916 if (tlen == 0) { 1917 /* Ack prediction. */ 1918 if (SEQ_GT(th->th_ack, tp->snd_una) && 1919 SEQ_LEQ(th->th_ack, tp->snd_max) && 1920 tp->snd_cwnd >= tp->snd_wnd && 1921 tp->t_partialacks < 0) { 1922 /* 1923 * this is a pure ack for outstanding data. 1924 */ 1925 if (ts_rtt) 1926 tcp_xmit_timer(tp, ts_rtt - 1); 1927 else if (tp->t_rtttime && 1928 SEQ_GT(th->th_ack, tp->t_rtseq)) 1929 tcp_xmit_timer(tp, 1930 tcp_now - tp->t_rtttime); 1931 acked = th->th_ack - tp->snd_una; 1932 tcps = TCP_STAT_GETREF(); 1933 tcps[TCP_STAT_PREDACK]++; 1934 tcps[TCP_STAT_RCVACKPACK]++; 1935 tcps[TCP_STAT_RCVACKBYTE] += acked; 1936 TCP_STAT_PUTREF(); 1937 nd6_hint(tp); 1938 1939 if (acked > (tp->t_lastoff - tp->t_inoff)) 1940 tp->t_lastm = NULL; 1941 sbdrop(&so->so_snd, acked); 1942 tp->t_lastoff -= acked; 1943 1944 icmp_check(tp, th, acked); 1945 1946 tp->snd_una = th->th_ack; 1947 tp->snd_fack = tp->snd_una; 1948 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1949 tp->snd_high = tp->snd_una; 1950 m_freem(m); 1951 1952 /* 1953 * If all outstanding data are acked, stop 1954 * retransmit timer, otherwise restart timer 1955 * using current (possibly backed-off) value. 1956 * If process is waiting for space, 1957 * wakeup/selnotify/signal. If data 1958 * are ready to send, let tcp_output 1959 * decide between more output or persist. 1960 */ 1961 if (tp->snd_una == tp->snd_max) 1962 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1963 else if (TCP_TIMER_ISARMED(tp, 1964 TCPT_PERSIST) == 0) 1965 TCP_TIMER_ARM(tp, TCPT_REXMT, 1966 tp->t_rxtcur); 1967 1968 sowwakeup(so); 1969 if (so->so_snd.sb_cc) { 1970 KERNEL_LOCK(1, NULL); 1971 (void) tcp_output(tp); 1972 KERNEL_UNLOCK_ONE(NULL); 1973 } 1974 if (tcp_saveti) 1975 m_freem(tcp_saveti); 1976 return; 1977 } 1978 } else if (th->th_ack == tp->snd_una && 1979 TAILQ_FIRST(&tp->segq) == NULL && 1980 tlen <= sbspace(&so->so_rcv)) { 1981 int newsize = 0; /* automatic sockbuf scaling */ 1982 1983 /* 1984 * this is a pure, in-sequence data packet 1985 * with nothing on the reassembly queue and 1986 * we have enough buffer space to take it. 1987 */ 1988 tp->rcv_nxt += tlen; 1989 tcps = TCP_STAT_GETREF(); 1990 tcps[TCP_STAT_PREDDAT]++; 1991 tcps[TCP_STAT_RCVPACK]++; 1992 tcps[TCP_STAT_RCVBYTE] += tlen; 1993 TCP_STAT_PUTREF(); 1994 nd6_hint(tp); 1995 1996 /* 1997 * Automatic sizing enables the performance of large buffers 1998 * and most of the efficiency of small ones by only allocating 1999 * space when it is needed. 2000 * 2001 * On the receive side the socket buffer memory is only rarely 2002 * used to any significant extent. This allows us to be much 2003 * more aggressive in scaling the receive socket buffer. For 2004 * the case that the buffer space is actually used to a large 2005 * extent and we run out of kernel memory we can simply drop 2006 * the new segments; TCP on the sender will just retransmit it 2007 * later. Setting the buffer size too big may only consume too 2008 * much kernel memory if the application doesn't read() from 2009 * the socket or packet loss or reordering makes use of the 2010 * reassembly queue. 2011 * 2012 * The criteria to step up the receive buffer one notch are: 2013 * 1. the number of bytes received during the time it takes 2014 * one timestamp to be reflected back to us (the RTT); 2015 * 2. received bytes per RTT is within seven eighth of the 2016 * current socket buffer size; 2017 * 3. receive buffer size has not hit maximal automatic size; 2018 * 2019 * This algorithm does one step per RTT at most and only if 2020 * we receive a bulk stream w/o packet losses or reorderings. 2021 * Shrinking the buffer during idle times is not necessary as 2022 * it doesn't consume any memory when idle. 2023 * 2024 * TODO: Only step up if the application is actually serving 2025 * the buffer to better manage the socket buffer resources. 2026 */ 2027 if (tcp_do_autorcvbuf && 2028 opti.ts_ecr && 2029 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 2030 if (opti.ts_ecr > tp->rfbuf_ts && 2031 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 2032 if (tp->rfbuf_cnt > 2033 (so->so_rcv.sb_hiwat / 8 * 7) && 2034 so->so_rcv.sb_hiwat < 2035 tcp_autorcvbuf_max) { 2036 newsize = 2037 min(so->so_rcv.sb_hiwat + 2038 tcp_autorcvbuf_inc, 2039 tcp_autorcvbuf_max); 2040 } 2041 /* Start over with next RTT. */ 2042 tp->rfbuf_ts = 0; 2043 tp->rfbuf_cnt = 0; 2044 } else 2045 tp->rfbuf_cnt += tlen; /* add up */ 2046 } 2047 2048 /* 2049 * Drop TCP, IP headers and TCP options then add data 2050 * to socket buffer. 2051 */ 2052 if (so->so_state & SS_CANTRCVMORE) 2053 m_freem(m); 2054 else { 2055 /* 2056 * Set new socket buffer size. 2057 * Give up when limit is reached. 2058 */ 2059 if (newsize) 2060 if (!sbreserve(&so->so_rcv, 2061 newsize, so)) 2062 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 2063 m_adj(m, toff + off); 2064 sbappendstream(&so->so_rcv, m); 2065 } 2066 sorwakeup(so); 2067 tcp_setup_ack(tp, th); 2068 if (tp->t_flags & TF_ACKNOW) { 2069 KERNEL_LOCK(1, NULL); 2070 (void) tcp_output(tp); 2071 KERNEL_UNLOCK_ONE(NULL); 2072 } 2073 if (tcp_saveti) 2074 m_freem(tcp_saveti); 2075 return; 2076 } 2077 } 2078 2079 /* 2080 * Compute mbuf offset to TCP data segment. 2081 */ 2082 hdroptlen = toff + off; 2083 2084 /* 2085 * Calculate amount of space in receive window. Receive window is 2086 * amount of space in rcv queue, but not less than advertised 2087 * window. 2088 */ 2089 { 2090 int win; 2091 win = sbspace(&so->so_rcv); 2092 if (win < 0) 2093 win = 0; 2094 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 2095 } 2096 2097 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 2098 tp->rfbuf_ts = 0; 2099 tp->rfbuf_cnt = 0; 2100 2101 switch (tp->t_state) { 2102 /* 2103 * If the state is SYN_SENT: 2104 * if seg contains an ACK, but not for our SYN, drop the input. 2105 * if seg contains a RST, then drop the connection. 2106 * if seg does not contain SYN, then drop it. 2107 * Otherwise this is an acceptable SYN segment 2108 * initialize tp->rcv_nxt and tp->irs 2109 * if seg contains ack then advance tp->snd_una 2110 * if seg contains a ECE and ECN support is enabled, the stream 2111 * is ECN capable. 2112 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2113 * arrange for segment to be acked (eventually) 2114 * continue processing rest of data/controls, beginning with URG 2115 */ 2116 case TCPS_SYN_SENT: 2117 if ((tiflags & TH_ACK) && 2118 (SEQ_LEQ(th->th_ack, tp->iss) || 2119 SEQ_GT(th->th_ack, tp->snd_max))) 2120 goto dropwithreset; 2121 if (tiflags & TH_RST) { 2122 if (tiflags & TH_ACK) 2123 tp = tcp_drop(tp, ECONNREFUSED); 2124 goto drop; 2125 } 2126 if ((tiflags & TH_SYN) == 0) 2127 goto drop; 2128 if (tiflags & TH_ACK) { 2129 tp->snd_una = th->th_ack; 2130 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2131 tp->snd_nxt = tp->snd_una; 2132 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2133 tp->snd_high = tp->snd_una; 2134 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2135 2136 if ((tiflags & TH_ECE) && tcp_do_ecn) { 2137 tp->t_flags |= TF_ECN_PERMIT; 2138 TCP_STATINC(TCP_STAT_ECN_SHS); 2139 } 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 #ifdef INET 2157 if (inp != NULL && in_localaddr(inp->inp_faddr)) 2158 ss = tcp_init_win_local; 2159 #endif 2160 #ifdef INET6 2161 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 2162 ss = tcp_init_win_local; 2163 #endif 2164 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 2165 } 2166 2167 tcp_rmx_rtt(tp); 2168 if (tiflags & TH_ACK) { 2169 TCP_STATINC(TCP_STAT_CONNECTS); 2170 /* 2171 * move tcp_established before soisconnected 2172 * because upcall handler can drive tcp_output 2173 * functionality. 2174 * XXX we might call soisconnected at the end of 2175 * all processing 2176 */ 2177 tcp_established(tp); 2178 soisconnected(so); 2179 /* Do window scaling on this connection? */ 2180 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2181 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2182 tp->snd_scale = tp->requested_s_scale; 2183 tp->rcv_scale = tp->request_r_scale; 2184 } 2185 TCP_REASS_LOCK(tp); 2186 (void)tcp_reass(tp, NULL, NULL, tlen); 2187 /* 2188 * if we didn't have to retransmit the SYN, 2189 * use its rtt as our initial srtt & rtt var. 2190 */ 2191 if (tp->t_rtttime) 2192 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2193 } else 2194 tp->t_state = TCPS_SYN_RECEIVED; 2195 2196 /* 2197 * Advance th->th_seq to correspond to first data byte. 2198 * If data, trim to stay within window, 2199 * dropping FIN if necessary. 2200 */ 2201 th->th_seq++; 2202 if (tlen > tp->rcv_wnd) { 2203 todrop = tlen - tp->rcv_wnd; 2204 m_adj(m, -todrop); 2205 tlen = tp->rcv_wnd; 2206 tiflags &= ~TH_FIN; 2207 tcps = TCP_STAT_GETREF(); 2208 tcps[TCP_STAT_RCVPACKAFTERWIN]++; 2209 tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop; 2210 TCP_STAT_PUTREF(); 2211 } 2212 tp->snd_wl1 = th->th_seq - 1; 2213 tp->rcv_up = th->th_seq; 2214 goto step6; 2215 2216 /* 2217 * If the state is SYN_RECEIVED: 2218 * If seg contains an ACK, but not for our SYN, drop the input 2219 * and generate an RST. See page 36, rfc793 2220 */ 2221 case TCPS_SYN_RECEIVED: 2222 if ((tiflags & TH_ACK) && 2223 (SEQ_LEQ(th->th_ack, tp->iss) || 2224 SEQ_GT(th->th_ack, tp->snd_max))) 2225 goto dropwithreset; 2226 break; 2227 } 2228 2229 /* 2230 * From here on, we're dealing with !LISTEN and !SYN_SENT. 2231 */ 2232 KASSERT(tp->t_state != TCPS_LISTEN && 2233 tp->t_state != TCPS_SYN_SENT); 2234 2235 /* 2236 * First check timestamp, if present. 2237 * Then check that at least some bytes of segment are within 2238 * receive window. If segment begins before rcv_nxt, 2239 * drop leading data (and SYN); if nothing left, just ack. 2240 * 2241 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2242 * and it's less than ts_recent, drop it. 2243 */ 2244 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2245 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2246 /* Check to see if ts_recent is over 24 days old. */ 2247 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2248 /* 2249 * Invalidate ts_recent. If this segment updates 2250 * ts_recent, the age will be reset later and ts_recent 2251 * will get a valid value. If it does not, setting 2252 * ts_recent to zero will at least satisfy the 2253 * requirement that zero be placed in the timestamp 2254 * echo reply when ts_recent isn't valid. The 2255 * age isn't reset until we get a valid ts_recent 2256 * because we don't want out-of-order segments to be 2257 * dropped when ts_recent is old. 2258 */ 2259 tp->ts_recent = 0; 2260 } else { 2261 tcps = TCP_STAT_GETREF(); 2262 tcps[TCP_STAT_RCVDUPPACK]++; 2263 tcps[TCP_STAT_RCVDUPBYTE] += tlen; 2264 tcps[TCP_STAT_PAWSDROP]++; 2265 TCP_STAT_PUTREF(); 2266 tcp_new_dsack(tp, th->th_seq, tlen); 2267 goto dropafterack; 2268 } 2269 } 2270 2271 todrop = tp->rcv_nxt - th->th_seq; 2272 dupseg = false; 2273 if (todrop > 0) { 2274 if (tiflags & TH_SYN) { 2275 tiflags &= ~TH_SYN; 2276 th->th_seq++; 2277 if (th->th_urp > 1) 2278 th->th_urp--; 2279 else { 2280 tiflags &= ~TH_URG; 2281 th->th_urp = 0; 2282 } 2283 todrop--; 2284 } 2285 if (todrop > tlen || 2286 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2287 /* 2288 * Any valid FIN or RST must be to the left of the 2289 * window. At this point the FIN or RST must be a 2290 * duplicate or out of sequence; drop it. 2291 */ 2292 if (tiflags & TH_RST) 2293 goto drop; 2294 tiflags &= ~(TH_FIN|TH_RST); 2295 /* 2296 * Send an ACK to resynchronize and drop any data. 2297 * But keep on processing for RST or ACK. 2298 */ 2299 tp->t_flags |= TF_ACKNOW; 2300 todrop = tlen; 2301 dupseg = true; 2302 tcps = TCP_STAT_GETREF(); 2303 tcps[TCP_STAT_RCVDUPPACK]++; 2304 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 2305 TCP_STAT_PUTREF(); 2306 } else if ((tiflags & TH_RST) && 2307 th->th_seq != tp->rcv_nxt) { 2308 /* 2309 * Test for reset before adjusting the sequence 2310 * number for overlapping data. 2311 */ 2312 goto dropafterack_ratelim; 2313 } else { 2314 tcps = TCP_STAT_GETREF(); 2315 tcps[TCP_STAT_RCVPARTDUPPACK]++; 2316 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 2317 TCP_STAT_PUTREF(); 2318 } 2319 tcp_new_dsack(tp, th->th_seq, todrop); 2320 hdroptlen += todrop; /*drop from head afterwards*/ 2321 th->th_seq += todrop; 2322 tlen -= todrop; 2323 if (th->th_urp > todrop) 2324 th->th_urp -= todrop; 2325 else { 2326 tiflags &= ~TH_URG; 2327 th->th_urp = 0; 2328 } 2329 } 2330 2331 /* 2332 * If new data are received on a connection after the 2333 * user processes are gone, then RST the other end. 2334 */ 2335 if ((so->so_state & SS_NOFDREF) && 2336 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2337 tp = tcp_close(tp); 2338 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 2339 goto dropwithreset; 2340 } 2341 2342 /* 2343 * If segment ends after window, drop trailing data 2344 * (and PUSH and FIN); if nothing left, just ACK. 2345 */ 2346 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 2347 if (todrop > 0) { 2348 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 2349 if (todrop >= tlen) { 2350 /* 2351 * The segment actually starts after the window. 2352 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2353 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2354 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2355 */ 2356 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 2357 /* 2358 * If a new connection request is received 2359 * while in TIME_WAIT, drop the old connection 2360 * and start over if the sequence numbers 2361 * are above the previous ones. 2362 * 2363 * NOTE: We will checksum the packet again, and 2364 * so we need to put the header fields back into 2365 * network order! 2366 * XXX This kind of sucks, but we don't expect 2367 * XXX this to happen very often, so maybe it 2368 * XXX doesn't matter so much. 2369 */ 2370 if (tiflags & TH_SYN && 2371 tp->t_state == TCPS_TIME_WAIT && 2372 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2373 tp = tcp_close(tp); 2374 tcp_fields_to_net(th); 2375 goto findpcb; 2376 } 2377 /* 2378 * If window is closed can only take segments at 2379 * window edge, and have to drop data and PUSH from 2380 * incoming segments. Continue processing, but 2381 * remember to ack. Otherwise, drop segment 2382 * and (if not RST) ack. 2383 */ 2384 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2385 tp->t_flags |= TF_ACKNOW; 2386 TCP_STATINC(TCP_STAT_RCVWINPROBE); 2387 } else 2388 goto dropafterack; 2389 } else 2390 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 2391 m_adj(m, -todrop); 2392 tlen -= todrop; 2393 tiflags &= ~(TH_PUSH|TH_FIN); 2394 } 2395 2396 /* 2397 * If last ACK falls within this segment's sequence numbers, 2398 * record the timestamp. 2399 * NOTE: 2400 * 1) That the test incorporates suggestions from the latest 2401 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 2402 * 2) That updating only on newer timestamps interferes with 2403 * our earlier PAWS tests, so this check should be solely 2404 * predicated on the sequence space of this segment. 2405 * 3) That we modify the segment boundary check to be 2406 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2407 * instead of RFC1323's 2408 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2409 * This modified check allows us to overcome RFC1323's 2410 * limitations as described in Stevens TCP/IP Illustrated 2411 * Vol. 2 p.869. In such cases, we can still calculate the 2412 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2413 */ 2414 if (opti.ts_present && 2415 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2416 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2417 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2418 tp->ts_recent_age = tcp_now; 2419 tp->ts_recent = opti.ts_val; 2420 } 2421 2422 /* 2423 * If the RST bit is set examine the state: 2424 * RECEIVED state: 2425 * If passive open, return to LISTEN state. 2426 * If active open, inform user that connection was refused. 2427 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT states: 2428 * Inform user that connection was reset, and close tcb. 2429 * CLOSING, LAST_ACK, TIME_WAIT states: 2430 * Close the tcb. 2431 */ 2432 if (tiflags & TH_RST) { 2433 if (th->th_seq != tp->rcv_nxt) 2434 goto dropafterack_ratelim; 2435 2436 switch (tp->t_state) { 2437 case TCPS_SYN_RECEIVED: 2438 so->so_error = ECONNREFUSED; 2439 goto close; 2440 2441 case TCPS_ESTABLISHED: 2442 case TCPS_FIN_WAIT_1: 2443 case TCPS_FIN_WAIT_2: 2444 case TCPS_CLOSE_WAIT: 2445 so->so_error = ECONNRESET; 2446 close: 2447 tp->t_state = TCPS_CLOSED; 2448 TCP_STATINC(TCP_STAT_DROPS); 2449 tp = tcp_close(tp); 2450 goto drop; 2451 2452 case TCPS_CLOSING: 2453 case TCPS_LAST_ACK: 2454 case TCPS_TIME_WAIT: 2455 tp = tcp_close(tp); 2456 goto drop; 2457 } 2458 } 2459 2460 /* 2461 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2462 * we must be in a synchronized state. RFC791 states (under RST 2463 * generation) that any unacceptable segment (an out-of-order SYN 2464 * qualifies) received in a synchronized state must elicit only an 2465 * empty acknowledgment segment ... and the connection remains in 2466 * the same state. 2467 */ 2468 if (tiflags & TH_SYN) { 2469 if (tp->rcv_nxt == th->th_seq) { 2470 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2471 TH_ACK); 2472 if (tcp_saveti) 2473 m_freem(tcp_saveti); 2474 return; 2475 } 2476 2477 goto dropafterack_ratelim; 2478 } 2479 2480 /* 2481 * If the ACK bit is off we drop the segment and return. 2482 */ 2483 if ((tiflags & TH_ACK) == 0) { 2484 if (tp->t_flags & TF_ACKNOW) 2485 goto dropafterack; 2486 else 2487 goto drop; 2488 } 2489 2490 /* 2491 * Ack processing. 2492 */ 2493 switch (tp->t_state) { 2494 2495 /* 2496 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2497 * ESTABLISHED state and continue processing, otherwise 2498 * send an RST. 2499 */ 2500 case TCPS_SYN_RECEIVED: 2501 if (SEQ_GT(tp->snd_una, th->th_ack) || 2502 SEQ_GT(th->th_ack, tp->snd_max)) 2503 goto dropwithreset; 2504 TCP_STATINC(TCP_STAT_CONNECTS); 2505 soisconnected(so); 2506 tcp_established(tp); 2507 /* Do window scaling? */ 2508 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2509 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2510 tp->snd_scale = tp->requested_s_scale; 2511 tp->rcv_scale = tp->request_r_scale; 2512 } 2513 TCP_REASS_LOCK(tp); 2514 (void)tcp_reass(tp, NULL, NULL, tlen); 2515 tp->snd_wl1 = th->th_seq - 1; 2516 /* fall into ... */ 2517 2518 /* 2519 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2520 * ACKs. If the ack is in the range 2521 * tp->snd_una < th->th_ack <= tp->snd_max 2522 * then advance tp->snd_una to th->th_ack and drop 2523 * data from the retransmission queue. If this ACK reflects 2524 * more up to date window information we update our window information. 2525 */ 2526 case TCPS_ESTABLISHED: 2527 case TCPS_FIN_WAIT_1: 2528 case TCPS_FIN_WAIT_2: 2529 case TCPS_CLOSE_WAIT: 2530 case TCPS_CLOSING: 2531 case TCPS_LAST_ACK: 2532 case TCPS_TIME_WAIT: 2533 2534 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2535 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2536 TCP_STATINC(TCP_STAT_RCVDUPACK); 2537 /* 2538 * If we have outstanding data (other than 2539 * a window probe), this is a completely 2540 * duplicate ack (ie, window info didn't 2541 * change), the ack is the biggest we've 2542 * seen and we've seen exactly our rexmt 2543 * threshhold of them, assume a packet 2544 * has been dropped and retransmit it. 2545 * Kludge snd_nxt & the congestion 2546 * window so we send only this one 2547 * packet. 2548 */ 2549 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2550 th->th_ack != tp->snd_una) 2551 tp->t_dupacks = 0; 2552 else if (tp->t_partialacks < 0 && 2553 (++tp->t_dupacks == tcprexmtthresh || 2554 TCP_FACK_FASTRECOV(tp))) { 2555 /* 2556 * Do the fast retransmit, and adjust 2557 * congestion control paramenters. 2558 */ 2559 if (tp->t_congctl->fast_retransmit(tp, th)) { 2560 /* False fast retransmit */ 2561 break; 2562 } else 2563 goto drop; 2564 } else if (tp->t_dupacks > tcprexmtthresh) { 2565 tp->snd_cwnd += tp->t_segsz; 2566 KERNEL_LOCK(1, NULL); 2567 (void) tcp_output(tp); 2568 KERNEL_UNLOCK_ONE(NULL); 2569 goto drop; 2570 } 2571 } else { 2572 /* 2573 * If the ack appears to be very old, only 2574 * allow data that is in-sequence. This 2575 * makes it somewhat more difficult to insert 2576 * forged data by guessing sequence numbers. 2577 * Sent an ack to try to update the send 2578 * sequence number on the other side. 2579 */ 2580 if (tlen && th->th_seq != tp->rcv_nxt && 2581 SEQ_LT(th->th_ack, 2582 tp->snd_una - tp->max_sndwnd)) 2583 goto dropafterack; 2584 } 2585 break; 2586 } 2587 /* 2588 * If the congestion window was inflated to account 2589 * for the other side's cached packets, retract it. 2590 */ 2591 tp->t_congctl->fast_retransmit_newack(tp, th); 2592 2593 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2594 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2595 goto dropafterack; 2596 } 2597 acked = th->th_ack - tp->snd_una; 2598 tcps = TCP_STAT_GETREF(); 2599 tcps[TCP_STAT_RCVACKPACK]++; 2600 tcps[TCP_STAT_RCVACKBYTE] += acked; 2601 TCP_STAT_PUTREF(); 2602 2603 /* 2604 * If we have a timestamp reply, update smoothed 2605 * round trip time. If no timestamp is present but 2606 * transmit timer is running and timed sequence 2607 * number was acked, update smoothed round trip time. 2608 * Since we now have an rtt measurement, cancel the 2609 * timer backoff (cf., Phil Karn's retransmit alg.). 2610 * Recompute the initial retransmit timer. 2611 */ 2612 if (ts_rtt) 2613 tcp_xmit_timer(tp, ts_rtt - 1); 2614 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2615 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2616 2617 /* 2618 * If all outstanding data is acked, stop retransmit 2619 * timer and remember to restart (more output or persist). 2620 * If there is more data to be acked, restart retransmit 2621 * timer, using current (possibly backed-off) value. 2622 */ 2623 if (th->th_ack == tp->snd_max) { 2624 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2625 needoutput = 1; 2626 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2627 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2628 2629 /* 2630 * New data has been acked, adjust the congestion window. 2631 */ 2632 tp->t_congctl->newack(tp, th); 2633 2634 nd6_hint(tp); 2635 if (acked > so->so_snd.sb_cc) { 2636 tp->snd_wnd -= so->so_snd.sb_cc; 2637 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2638 ourfinisacked = 1; 2639 } else { 2640 if (acked > (tp->t_lastoff - tp->t_inoff)) 2641 tp->t_lastm = NULL; 2642 sbdrop(&so->so_snd, acked); 2643 tp->t_lastoff -= acked; 2644 if (tp->snd_wnd > acked) 2645 tp->snd_wnd -= acked; 2646 else 2647 tp->snd_wnd = 0; 2648 ourfinisacked = 0; 2649 } 2650 sowwakeup(so); 2651 2652 icmp_check(tp, th, acked); 2653 2654 tp->snd_una = th->th_ack; 2655 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2656 tp->snd_fack = tp->snd_una; 2657 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2658 tp->snd_nxt = tp->snd_una; 2659 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2660 tp->snd_high = tp->snd_una; 2661 2662 switch (tp->t_state) { 2663 2664 /* 2665 * In FIN_WAIT_1 STATE in addition to the processing 2666 * for the ESTABLISHED state if our FIN is now acknowledged 2667 * then enter FIN_WAIT_2. 2668 */ 2669 case TCPS_FIN_WAIT_1: 2670 if (ourfinisacked) { 2671 /* 2672 * If we can't receive any more 2673 * data, then closing user can proceed. 2674 * Starting the timer is contrary to the 2675 * specification, but if we don't get a FIN 2676 * we'll hang forever. 2677 */ 2678 if (so->so_state & SS_CANTRCVMORE) { 2679 soisdisconnected(so); 2680 if (tp->t_maxidle > 0) 2681 TCP_TIMER_ARM(tp, TCPT_2MSL, 2682 tp->t_maxidle); 2683 } 2684 tp->t_state = TCPS_FIN_WAIT_2; 2685 } 2686 break; 2687 2688 /* 2689 * In CLOSING STATE in addition to the processing for 2690 * the ESTABLISHED state if the ACK acknowledges our FIN 2691 * then enter the TIME-WAIT state, otherwise ignore 2692 * the segment. 2693 */ 2694 case TCPS_CLOSING: 2695 if (ourfinisacked) { 2696 tp->t_state = TCPS_TIME_WAIT; 2697 tcp_canceltimers(tp); 2698 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2699 soisdisconnected(so); 2700 } 2701 break; 2702 2703 /* 2704 * In LAST_ACK, we may still be waiting for data to drain 2705 * and/or to be acked, as well as for the ack of our FIN. 2706 * If our FIN is now acknowledged, delete the TCB, 2707 * enter the closed state and return. 2708 */ 2709 case TCPS_LAST_ACK: 2710 if (ourfinisacked) { 2711 tp = tcp_close(tp); 2712 goto drop; 2713 } 2714 break; 2715 2716 /* 2717 * In TIME_WAIT state the only thing that should arrive 2718 * is a retransmission of the remote FIN. Acknowledge 2719 * it and restart the finack timer. 2720 */ 2721 case TCPS_TIME_WAIT: 2722 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2723 goto dropafterack; 2724 } 2725 } 2726 2727 step6: 2728 /* 2729 * Update window information. 2730 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2731 */ 2732 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2733 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2734 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2735 /* keep track of pure window updates */ 2736 if (tlen == 0 && 2737 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2738 TCP_STATINC(TCP_STAT_RCVWINUPD); 2739 tp->snd_wnd = tiwin; 2740 tp->snd_wl1 = th->th_seq; 2741 tp->snd_wl2 = th->th_ack; 2742 if (tp->snd_wnd > tp->max_sndwnd) 2743 tp->max_sndwnd = tp->snd_wnd; 2744 needoutput = 1; 2745 } 2746 2747 /* 2748 * Process segments with URG. 2749 */ 2750 if ((tiflags & TH_URG) && th->th_urp && 2751 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2752 /* 2753 * This is a kludge, but if we receive and accept 2754 * random urgent pointers, we'll crash in 2755 * soreceive. It's hard to imagine someone 2756 * actually wanting to send this much urgent data. 2757 */ 2758 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2759 th->th_urp = 0; /* XXX */ 2760 tiflags &= ~TH_URG; /* XXX */ 2761 goto dodata; /* XXX */ 2762 } 2763 /* 2764 * If this segment advances the known urgent pointer, 2765 * then mark the data stream. This should not happen 2766 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2767 * a FIN has been received from the remote side. 2768 * In these states we ignore the URG. 2769 * 2770 * According to RFC961 (Assigned Protocols), 2771 * the urgent pointer points to the last octet 2772 * of urgent data. We continue, however, 2773 * to consider it to indicate the first octet 2774 * of data past the urgent section as the original 2775 * spec states (in one of two places). 2776 */ 2777 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2778 tp->rcv_up = th->th_seq + th->th_urp; 2779 so->so_oobmark = so->so_rcv.sb_cc + 2780 (tp->rcv_up - tp->rcv_nxt) - 1; 2781 if (so->so_oobmark == 0) 2782 so->so_state |= SS_RCVATMARK; 2783 sohasoutofband(so); 2784 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2785 } 2786 /* 2787 * Remove out of band data so doesn't get presented to user. 2788 * This can happen independent of advancing the URG pointer, 2789 * but if two URG's are pending at once, some out-of-band 2790 * data may creep in... ick. 2791 */ 2792 if (th->th_urp <= (u_int16_t)tlen && 2793 (so->so_options & SO_OOBINLINE) == 0) 2794 tcp_pulloutofband(so, th, m, hdroptlen); 2795 } else 2796 /* 2797 * If no out of band data is expected, 2798 * pull receive urgent pointer along 2799 * with the receive window. 2800 */ 2801 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2802 tp->rcv_up = tp->rcv_nxt; 2803 dodata: /* XXX */ 2804 2805 /* 2806 * Process the segment text, merging it into the TCP sequencing queue, 2807 * and arranging for acknowledgement of receipt if necessary. 2808 * This process logically involves adjusting tp->rcv_wnd as data 2809 * is presented to the user (this happens in tcp_usrreq.c, 2810 * tcp_rcvd()). If a FIN has already been received on this 2811 * connection then we just ignore the text. 2812 */ 2813 if ((tlen || (tiflags & TH_FIN)) && 2814 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2815 /* 2816 * Insert segment ti into reassembly queue of tcp with 2817 * control block tp. Return TH_FIN if reassembly now includes 2818 * a segment with FIN. The macro form does the common case 2819 * inline (segment is the next to be received on an 2820 * established connection, and the queue is empty), 2821 * avoiding linkage into and removal from the queue and 2822 * repetition of various conversions. 2823 * Set DELACK for segments received in order, but ack 2824 * immediately when segments are out of order 2825 * (so fast retransmit can work). 2826 */ 2827 /* NOTE: this was TCP_REASS() macro, but used only once */ 2828 TCP_REASS_LOCK(tp); 2829 if (th->th_seq == tp->rcv_nxt && 2830 TAILQ_FIRST(&tp->segq) == NULL && 2831 tp->t_state == TCPS_ESTABLISHED) { 2832 tcp_setup_ack(tp, th); 2833 tp->rcv_nxt += tlen; 2834 tiflags = th->th_flags & TH_FIN; 2835 tcps = TCP_STAT_GETREF(); 2836 tcps[TCP_STAT_RCVPACK]++; 2837 tcps[TCP_STAT_RCVBYTE] += tlen; 2838 TCP_STAT_PUTREF(); 2839 nd6_hint(tp); 2840 if (so->so_state & SS_CANTRCVMORE) 2841 m_freem(m); 2842 else { 2843 m_adj(m, hdroptlen); 2844 sbappendstream(&(so)->so_rcv, m); 2845 } 2846 TCP_REASS_UNLOCK(tp); 2847 sorwakeup(so); 2848 } else { 2849 m_adj(m, hdroptlen); 2850 tiflags = tcp_reass(tp, th, m, tlen); 2851 tp->t_flags |= TF_ACKNOW; 2852 } 2853 2854 /* 2855 * Note the amount of data that peer has sent into 2856 * our window, in order to estimate the sender's 2857 * buffer size. 2858 */ 2859 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2860 } else { 2861 m_freem(m); 2862 m = NULL; 2863 tiflags &= ~TH_FIN; 2864 } 2865 2866 /* 2867 * If FIN is received ACK the FIN and let the user know 2868 * that the connection is closing. Ignore a FIN received before 2869 * the connection is fully established. 2870 */ 2871 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2872 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2873 socantrcvmore(so); 2874 tp->t_flags |= TF_ACKNOW; 2875 tp->rcv_nxt++; 2876 } 2877 switch (tp->t_state) { 2878 2879 /* 2880 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2881 */ 2882 case TCPS_ESTABLISHED: 2883 tp->t_state = TCPS_CLOSE_WAIT; 2884 break; 2885 2886 /* 2887 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2888 * enter the CLOSING state. 2889 */ 2890 case TCPS_FIN_WAIT_1: 2891 tp->t_state = TCPS_CLOSING; 2892 break; 2893 2894 /* 2895 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2896 * starting the time-wait timer, turning off the other 2897 * standard timers. 2898 */ 2899 case TCPS_FIN_WAIT_2: 2900 tp->t_state = TCPS_TIME_WAIT; 2901 tcp_canceltimers(tp); 2902 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2903 soisdisconnected(so); 2904 break; 2905 2906 /* 2907 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2908 */ 2909 case TCPS_TIME_WAIT: 2910 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2911 break; 2912 } 2913 } 2914 #ifdef TCP_DEBUG 2915 if (so->so_options & SO_DEBUG) 2916 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2917 #endif 2918 2919 /* 2920 * Return any desired output. 2921 */ 2922 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2923 KERNEL_LOCK(1, NULL); 2924 (void) tcp_output(tp); 2925 KERNEL_UNLOCK_ONE(NULL); 2926 } 2927 if (tcp_saveti) 2928 m_freem(tcp_saveti); 2929 2930 if (tp->t_state == TCPS_TIME_WAIT 2931 && (so->so_state & SS_NOFDREF) 2932 && (tp->t_inpcb || af != AF_INET) 2933 && (tp->t_in6pcb || af != AF_INET6) 2934 && ((af == AF_INET ? tcp4_vtw_enable : tcp6_vtw_enable) & 1) != 0 2935 && TAILQ_EMPTY(&tp->segq) 2936 && vtw_add(af, tp)) { 2937 ; 2938 } 2939 return; 2940 2941 badsyn: 2942 /* 2943 * Received a bad SYN. Increment counters and dropwithreset. 2944 */ 2945 TCP_STATINC(TCP_STAT_BADSYN); 2946 tp = NULL; 2947 goto dropwithreset; 2948 2949 dropafterack: 2950 /* 2951 * Generate an ACK dropping incoming segment if it occupies 2952 * sequence space, where the ACK reflects our state. 2953 */ 2954 if (tiflags & TH_RST) 2955 goto drop; 2956 goto dropafterack2; 2957 2958 dropafterack_ratelim: 2959 /* 2960 * We may want to rate-limit ACKs against SYN/RST attack. 2961 */ 2962 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2963 tcp_ackdrop_ppslim) == 0) { 2964 /* XXX stat */ 2965 goto drop; 2966 } 2967 /* ...fall into dropafterack2... */ 2968 2969 dropafterack2: 2970 m_freem(m); 2971 tp->t_flags |= TF_ACKNOW; 2972 KERNEL_LOCK(1, NULL); 2973 (void) tcp_output(tp); 2974 KERNEL_UNLOCK_ONE(NULL); 2975 if (tcp_saveti) 2976 m_freem(tcp_saveti); 2977 return; 2978 2979 dropwithreset_ratelim: 2980 /* 2981 * We may want to rate-limit RSTs in certain situations, 2982 * particularly if we are sending an RST in response to 2983 * an attempt to connect to or otherwise communicate with 2984 * a port for which we have no socket. 2985 */ 2986 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2987 tcp_rst_ppslim) == 0) { 2988 /* XXX stat */ 2989 goto drop; 2990 } 2991 /* ...fall into dropwithreset... */ 2992 2993 dropwithreset: 2994 /* 2995 * Generate a RST, dropping incoming segment. 2996 * Make ACK acceptable to originator of segment. 2997 */ 2998 if (tiflags & TH_RST) 2999 goto drop; 3000 if (tiflags & TH_ACK) { 3001 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3002 } else { 3003 if (tiflags & TH_SYN) 3004 tlen++; 3005 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 3006 TH_RST|TH_ACK); 3007 } 3008 if (tcp_saveti) 3009 m_freem(tcp_saveti); 3010 return; 3011 3012 badcsum: 3013 drop: 3014 /* 3015 * Drop space held by incoming segment and return. 3016 */ 3017 if (tp) { 3018 if (tp->t_inpcb) 3019 so = tp->t_inpcb->inp_socket; 3020 #ifdef INET6 3021 else if (tp->t_in6pcb) 3022 so = tp->t_in6pcb->in6p_socket; 3023 #endif 3024 else 3025 so = NULL; 3026 #ifdef TCP_DEBUG 3027 if (so && (so->so_options & SO_DEBUG) != 0) 3028 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 3029 #endif 3030 } 3031 if (tcp_saveti) 3032 m_freem(tcp_saveti); 3033 m_freem(m); 3034 return; 3035 } 3036 3037 #ifdef TCP_SIGNATURE 3038 int 3039 tcp_signature_apply(void *fstate, void *data, u_int len) 3040 { 3041 3042 MD5Update(fstate, (u_char *)data, len); 3043 return (0); 3044 } 3045 3046 struct secasvar * 3047 tcp_signature_getsav(struct mbuf *m) 3048 { 3049 struct ip *ip; 3050 struct ip6_hdr *ip6; 3051 3052 ip = mtod(m, struct ip *); 3053 switch (ip->ip_v) { 3054 case 4: 3055 ip = mtod(m, struct ip *); 3056 ip6 = NULL; 3057 break; 3058 case 6: 3059 ip = NULL; 3060 ip6 = mtod(m, struct ip6_hdr *); 3061 break; 3062 default: 3063 return (NULL); 3064 } 3065 3066 #ifdef IPSEC 3067 union sockaddr_union dst; 3068 3069 /* Extract the destination from the IP header in the mbuf. */ 3070 memset(&dst, 0, sizeof(union sockaddr_union)); 3071 if (ip != NULL) { 3072 dst.sa.sa_len = sizeof(struct sockaddr_in); 3073 dst.sa.sa_family = AF_INET; 3074 dst.sin.sin_addr = ip->ip_dst; 3075 } else { 3076 dst.sa.sa_len = sizeof(struct sockaddr_in6); 3077 dst.sa.sa_family = AF_INET6; 3078 dst.sin6.sin6_addr = ip6->ip6_dst; 3079 } 3080 3081 /* 3082 * Look up an SADB entry which matches the address of the peer. 3083 */ 3084 return KEY_LOOKUP_SA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI), 0, 0); 3085 #else 3086 return NULL; 3087 #endif 3088 } 3089 3090 int 3091 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 3092 struct secasvar *sav, char *sig) 3093 { 3094 MD5_CTX ctx; 3095 struct ip *ip; 3096 struct ipovly *ipovly; 3097 #ifdef INET6 3098 struct ip6_hdr *ip6; 3099 struct ip6_hdr_pseudo ip6pseudo; 3100 #endif /* INET6 */ 3101 struct ippseudo ippseudo; 3102 struct tcphdr th0; 3103 int l, tcphdrlen; 3104 3105 if (sav == NULL) 3106 return (-1); 3107 3108 tcphdrlen = th->th_off * 4; 3109 3110 switch (mtod(m, struct ip *)->ip_v) { 3111 case 4: 3112 MD5Init(&ctx); 3113 ip = mtod(m, struct ip *); 3114 memset(&ippseudo, 0, sizeof(ippseudo)); 3115 ipovly = (struct ipovly *)ip; 3116 ippseudo.ippseudo_src = ipovly->ih_src; 3117 ippseudo.ippseudo_dst = ipovly->ih_dst; 3118 ippseudo.ippseudo_pad = 0; 3119 ippseudo.ippseudo_p = IPPROTO_TCP; 3120 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 3121 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 3122 break; 3123 #if INET6 3124 case 6: 3125 MD5Init(&ctx); 3126 ip6 = mtod(m, struct ip6_hdr *); 3127 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 3128 ip6pseudo.ip6ph_src = ip6->ip6_src; 3129 in6_clearscope(&ip6pseudo.ip6ph_src); 3130 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 3131 in6_clearscope(&ip6pseudo.ip6ph_dst); 3132 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 3133 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 3134 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 3135 break; 3136 #endif /* INET6 */ 3137 default: 3138 return (-1); 3139 } 3140 3141 th0 = *th; 3142 th0.th_sum = 0; 3143 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 3144 3145 l = m->m_pkthdr.len - thoff - tcphdrlen; 3146 if (l > 0) 3147 m_apply(m, thoff + tcphdrlen, 3148 m->m_pkthdr.len - thoff - tcphdrlen, 3149 tcp_signature_apply, &ctx); 3150 3151 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 3152 MD5Final(sig, &ctx); 3153 3154 return (0); 3155 } 3156 #endif 3157 3158 /* 3159 * tcp_dooptions: parse and process tcp options. 3160 * 3161 * returns -1 if this segment should be dropped. (eg. wrong signature) 3162 * otherwise returns 0. 3163 */ 3164 3165 static int 3166 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, struct tcphdr *th, 3167 struct mbuf *m, int toff, struct tcp_opt_info *oi) 3168 { 3169 u_int16_t mss; 3170 int opt, optlen = 0; 3171 #ifdef TCP_SIGNATURE 3172 void *sigp = NULL; 3173 char sigbuf[TCP_SIGLEN]; 3174 struct secasvar *sav = NULL; 3175 #endif 3176 3177 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 3178 opt = cp[0]; 3179 if (opt == TCPOPT_EOL) 3180 break; 3181 if (opt == TCPOPT_NOP) 3182 optlen = 1; 3183 else { 3184 if (cnt < 2) 3185 break; 3186 optlen = cp[1]; 3187 if (optlen < 2 || optlen > cnt) 3188 break; 3189 } 3190 switch (opt) { 3191 3192 default: 3193 continue; 3194 3195 case TCPOPT_MAXSEG: 3196 if (optlen != TCPOLEN_MAXSEG) 3197 continue; 3198 if (!(th->th_flags & TH_SYN)) 3199 continue; 3200 if (TCPS_HAVERCVDSYN(tp->t_state)) 3201 continue; 3202 bcopy(cp + 2, &mss, sizeof(mss)); 3203 oi->maxseg = ntohs(mss); 3204 break; 3205 3206 case TCPOPT_WINDOW: 3207 if (optlen != TCPOLEN_WINDOW) 3208 continue; 3209 if (!(th->th_flags & TH_SYN)) 3210 continue; 3211 if (TCPS_HAVERCVDSYN(tp->t_state)) 3212 continue; 3213 tp->t_flags |= TF_RCVD_SCALE; 3214 tp->requested_s_scale = cp[2]; 3215 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3216 char buf[INET6_ADDRSTRLEN]; 3217 struct ip *ip = mtod(m, struct ip *); 3218 #ifdef INET6 3219 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 3220 #endif 3221 if (ip) 3222 in_print(buf, sizeof(buf), 3223 &ip->ip_src); 3224 #ifdef INET6 3225 else if (ip6) 3226 in6_print(buf, sizeof(buf), 3227 &ip6->ip6_src); 3228 #endif 3229 else 3230 strlcpy(buf, "(unknown)", sizeof(buf)); 3231 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3232 "assuming %d\n", 3233 tp->requested_s_scale, buf, 3234 TCP_MAX_WINSHIFT); 3235 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3236 } 3237 break; 3238 3239 case TCPOPT_TIMESTAMP: 3240 if (optlen != TCPOLEN_TIMESTAMP) 3241 continue; 3242 oi->ts_present = 1; 3243 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 3244 NTOHL(oi->ts_val); 3245 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 3246 NTOHL(oi->ts_ecr); 3247 3248 if (!(th->th_flags & TH_SYN)) 3249 continue; 3250 if (TCPS_HAVERCVDSYN(tp->t_state)) 3251 continue; 3252 /* 3253 * A timestamp received in a SYN makes 3254 * it ok to send timestamp requests and replies. 3255 */ 3256 tp->t_flags |= TF_RCVD_TSTMP; 3257 tp->ts_recent = oi->ts_val; 3258 tp->ts_recent_age = tcp_now; 3259 break; 3260 3261 case TCPOPT_SACK_PERMITTED: 3262 if (optlen != TCPOLEN_SACK_PERMITTED) 3263 continue; 3264 if (!(th->th_flags & TH_SYN)) 3265 continue; 3266 if (TCPS_HAVERCVDSYN(tp->t_state)) 3267 continue; 3268 if (tcp_do_sack) { 3269 tp->t_flags |= TF_SACK_PERMIT; 3270 tp->t_flags |= TF_WILL_SACK; 3271 } 3272 break; 3273 3274 case TCPOPT_SACK: 3275 tcp_sack_option(tp, th, cp, optlen); 3276 break; 3277 #ifdef TCP_SIGNATURE 3278 case TCPOPT_SIGNATURE: 3279 if (optlen != TCPOLEN_SIGNATURE) 3280 continue; 3281 if (sigp && memcmp(sigp, cp + 2, TCP_SIGLEN)) 3282 return (-1); 3283 3284 sigp = sigbuf; 3285 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3286 tp->t_flags |= TF_SIGNATURE; 3287 break; 3288 #endif 3289 } 3290 } 3291 3292 #ifndef TCP_SIGNATURE 3293 return 0; 3294 #else 3295 if (tp->t_flags & TF_SIGNATURE) { 3296 sav = tcp_signature_getsav(m); 3297 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3298 return (-1); 3299 } 3300 3301 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) 3302 goto out; 3303 3304 if (sigp) { 3305 char sig[TCP_SIGLEN]; 3306 3307 tcp_fields_to_net(th); 3308 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3309 tcp_fields_to_host(th); 3310 goto out; 3311 } 3312 tcp_fields_to_host(th); 3313 3314 if (memcmp(sig, sigp, TCP_SIGLEN)) { 3315 TCP_STATINC(TCP_STAT_BADSIG); 3316 goto out; 3317 } else 3318 TCP_STATINC(TCP_STAT_GOODSIG); 3319 3320 key_sa_recordxfer(sav, m); 3321 KEY_SA_UNREF(&sav); 3322 } 3323 return 0; 3324 out: 3325 if (sav != NULL) 3326 KEY_SA_UNREF(&sav); 3327 return -1; 3328 #endif 3329 } 3330 3331 /* 3332 * Pull out of band byte out of a segment so 3333 * it doesn't appear in the user's data queue. 3334 * It is still reflected in the segment length for 3335 * sequencing purposes. 3336 */ 3337 void 3338 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3339 struct mbuf *m, int off) 3340 { 3341 int cnt = off + th->th_urp - 1; 3342 3343 while (cnt >= 0) { 3344 if (m->m_len > cnt) { 3345 char *cp = mtod(m, char *) + cnt; 3346 struct tcpcb *tp = sototcpcb(so); 3347 3348 tp->t_iobc = *cp; 3349 tp->t_oobflags |= TCPOOB_HAVEDATA; 3350 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3351 m->m_len--; 3352 return; 3353 } 3354 cnt -= m->m_len; 3355 m = m->m_next; 3356 if (m == 0) 3357 break; 3358 } 3359 panic("tcp_pulloutofband"); 3360 } 3361 3362 /* 3363 * Collect new round-trip time estimate 3364 * and update averages and current timeout. 3365 * 3366 * rtt is in units of slow ticks (typically 500 ms) -- essentially the 3367 * difference of two timestamps. 3368 */ 3369 void 3370 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3371 { 3372 int32_t delta; 3373 3374 TCP_STATINC(TCP_STAT_RTTUPDATED); 3375 if (tp->t_srtt != 0) { 3376 /* 3377 * Compute the amount to add to srtt for smoothing, 3378 * *alpha, or 2^(-TCP_RTT_SHIFT). Because 3379 * srtt is stored in 1/32 slow ticks, we conceptually 3380 * shift left 5 bits, subtract srtt to get the 3381 * diference, and then shift right by TCP_RTT_SHIFT 3382 * (3) to obtain 1/8 of the difference. 3383 */ 3384 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3385 /* 3386 * This can never happen, because delta's lowest 3387 * possible value is 1/8 of t_srtt. But if it does, 3388 * set srtt to some reasonable value, here chosen 3389 * as 1/8 tick. 3390 */ 3391 if ((tp->t_srtt += delta) <= 0) 3392 tp->t_srtt = 1 << 2; 3393 /* 3394 * RFC2988 requires that rttvar be updated first. 3395 * This code is compliant because "delta" is the old 3396 * srtt minus the new observation (scaled). 3397 * 3398 * RFC2988 says: 3399 * rttvar = (1-beta) * rttvar + beta * |srtt-observed| 3400 * 3401 * delta is in units of 1/32 ticks, and has then been 3402 * divided by 8. This is equivalent to being in 1/16s 3403 * units and divided by 4. Subtract from it 1/4 of 3404 * the existing rttvar to form the (signed) amount to 3405 * adjust. 3406 */ 3407 if (delta < 0) 3408 delta = -delta; 3409 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3410 /* 3411 * As with srtt, this should never happen. There is 3412 * no support in RFC2988 for this operation. But 1/4s 3413 * as rttvar when faced with something arguably wrong 3414 * is ok. 3415 */ 3416 if ((tp->t_rttvar += delta) <= 0) 3417 tp->t_rttvar = 1 << 2; 3418 3419 /* 3420 * If srtt exceeds .01 second, ensure we use the 'remote' MSL 3421 * Problem is: it doesn't work. Disabled by defaulting 3422 * tcp_rttlocal to 0; see corresponding code in 3423 * tcp_subr that selects local vs remote in a different way. 3424 * 3425 * The static branch prediction hint here should be removed 3426 * when the rtt estimator is fixed and the rtt_enable code 3427 * is turned back on. 3428 */ 3429 if (__predict_false(tcp_rttlocal) && tcp_msl_enable 3430 && tp->t_srtt > tcp_msl_remote_threshold 3431 && tp->t_msl < tcp_msl_remote) { 3432 tp->t_msl = tcp_msl_remote; 3433 } 3434 } else { 3435 /* 3436 * This is the first measurement. Per RFC2988, 2.2, 3437 * set rtt=R and srtt=R/2. 3438 * For srtt, storage representation is 1/32 ticks, 3439 * so shift left by 5. 3440 * For rttvar, storage representation is 1/16 ticks, 3441 * So shift left by 4, but then right by 1 to halve. 3442 */ 3443 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3444 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3445 } 3446 tp->t_rtttime = 0; 3447 tp->t_rxtshift = 0; 3448 3449 /* 3450 * the retransmit should happen at rtt + 4 * rttvar. 3451 * Because of the way we do the smoothing, srtt and rttvar 3452 * will each average +1/2 tick of bias. When we compute 3453 * the retransmit timer, we want 1/2 tick of rounding and 3454 * 1 extra tick because of +-1/2 tick uncertainty in the 3455 * firing of the timer. The bias will give us exactly the 3456 * 1.5 tick we need. But, because the bias is 3457 * statistical, we have to test that we don't drop below 3458 * the minimum feasible timer (which is 2 ticks). 3459 */ 3460 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3461 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3462 3463 /* 3464 * We received an ack for a packet that wasn't retransmitted; 3465 * it is probably safe to discard any error indications we've 3466 * received recently. This isn't quite right, but close enough 3467 * for now (a route might have failed after we sent a segment, 3468 * and the return path might not be symmetrical). 3469 */ 3470 tp->t_softerror = 0; 3471 } 3472 3473 3474 /* 3475 * TCP compressed state engine. Currently used to hold compressed 3476 * state for SYN_RECEIVED. 3477 */ 3478 3479 u_long syn_cache_count; 3480 u_int32_t syn_hash1, syn_hash2; 3481 3482 #define SYN_HASH(sa, sp, dp) \ 3483 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3484 ((u_int32_t)(sp)))^syn_hash2))) 3485 #ifndef INET6 3486 #define SYN_HASHALL(hash, src, dst) \ 3487 do { \ 3488 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3489 ((const struct sockaddr_in *)(src))->sin_port, \ 3490 ((const struct sockaddr_in *)(dst))->sin_port); \ 3491 } while (/*CONSTCOND*/ 0) 3492 #else 3493 #define SYN_HASH6(sa, sp, dp) \ 3494 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3495 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3496 & 0x7fffffff) 3497 3498 #define SYN_HASHALL(hash, src, dst) \ 3499 do { \ 3500 switch ((src)->sa_family) { \ 3501 case AF_INET: \ 3502 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3503 ((const struct sockaddr_in *)(src))->sin_port, \ 3504 ((const struct sockaddr_in *)(dst))->sin_port); \ 3505 break; \ 3506 case AF_INET6: \ 3507 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3508 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3509 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3510 break; \ 3511 default: \ 3512 hash = 0; \ 3513 } \ 3514 } while (/*CONSTCOND*/0) 3515 #endif /* INET6 */ 3516 3517 static struct pool syn_cache_pool; 3518 3519 /* 3520 * We don't estimate RTT with SYNs, so each packet starts with the default 3521 * RTT and each timer step has a fixed timeout value. 3522 */ 3523 static inline void 3524 syn_cache_timer_arm(struct syn_cache *sc) 3525 { 3526 3527 TCPT_RANGESET(sc->sc_rxtcur, 3528 TCPTV_SRTTDFLT * tcp_backoff[sc->sc_rxtshift], TCPTV_MIN, 3529 TCPTV_REXMTMAX); 3530 callout_reset(&sc->sc_timer, 3531 sc->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, sc); 3532 } 3533 3534 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3535 3536 static inline void 3537 syn_cache_rm(struct syn_cache *sc) 3538 { 3539 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3540 sc, sc_bucketq); 3541 sc->sc_tp = NULL; 3542 LIST_REMOVE(sc, sc_tpq); 3543 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3544 callout_stop(&sc->sc_timer); 3545 syn_cache_count--; 3546 } 3547 3548 static inline void 3549 syn_cache_put(struct syn_cache *sc) 3550 { 3551 if (sc->sc_ipopts) 3552 (void) m_free(sc->sc_ipopts); 3553 rtcache_free(&sc->sc_route); 3554 sc->sc_flags |= SCF_DEAD; 3555 if (!callout_invoking(&sc->sc_timer)) 3556 callout_schedule(&(sc)->sc_timer, 1); 3557 } 3558 3559 void 3560 syn_cache_init(void) 3561 { 3562 int i; 3563 3564 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 3565 "synpl", NULL, IPL_SOFTNET); 3566 3567 /* Initialize the hash buckets. */ 3568 for (i = 0; i < tcp_syn_cache_size; i++) 3569 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3570 } 3571 3572 void 3573 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3574 { 3575 struct syn_cache_head *scp; 3576 struct syn_cache *sc2; 3577 int s; 3578 3579 /* 3580 * If there are no entries in the hash table, reinitialize 3581 * the hash secrets. 3582 */ 3583 if (syn_cache_count == 0) { 3584 syn_hash1 = cprng_fast32(); 3585 syn_hash2 = cprng_fast32(); 3586 } 3587 3588 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3589 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3590 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3591 3592 /* 3593 * Make sure that we don't overflow the per-bucket 3594 * limit or the total cache size limit. 3595 */ 3596 s = splsoftnet(); 3597 if (scp->sch_length >= tcp_syn_bucket_limit) { 3598 TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW); 3599 /* 3600 * The bucket is full. Toss the oldest element in the 3601 * bucket. This will be the first entry in the bucket. 3602 */ 3603 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3604 #ifdef DIAGNOSTIC 3605 /* 3606 * This should never happen; we should always find an 3607 * entry in our bucket. 3608 */ 3609 if (sc2 == NULL) 3610 panic("syn_cache_insert: bucketoverflow: impossible"); 3611 #endif 3612 syn_cache_rm(sc2); 3613 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3614 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3615 struct syn_cache_head *scp2, *sce; 3616 3617 TCP_STATINC(TCP_STAT_SC_OVERFLOWED); 3618 /* 3619 * The cache is full. Toss the oldest entry in the 3620 * first non-empty bucket we can find. 3621 * 3622 * XXX We would really like to toss the oldest 3623 * entry in the cache, but we hope that this 3624 * condition doesn't happen very often. 3625 */ 3626 scp2 = scp; 3627 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3628 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3629 for (++scp2; scp2 != scp; scp2++) { 3630 if (scp2 >= sce) 3631 scp2 = &tcp_syn_cache[0]; 3632 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3633 break; 3634 } 3635 #ifdef DIAGNOSTIC 3636 /* 3637 * This should never happen; we should always find a 3638 * non-empty bucket. 3639 */ 3640 if (scp2 == scp) 3641 panic("syn_cache_insert: cacheoverflow: " 3642 "impossible"); 3643 #endif 3644 } 3645 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3646 syn_cache_rm(sc2); 3647 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3648 } 3649 3650 /* 3651 * Initialize the entry's timer. 3652 */ 3653 sc->sc_rxttot = 0; 3654 sc->sc_rxtshift = 0; 3655 syn_cache_timer_arm(sc); 3656 3657 /* Link it from tcpcb entry */ 3658 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3659 3660 /* Put it into the bucket. */ 3661 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3662 scp->sch_length++; 3663 syn_cache_count++; 3664 3665 TCP_STATINC(TCP_STAT_SC_ADDED); 3666 splx(s); 3667 } 3668 3669 /* 3670 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3671 * If we have retransmitted an entry the maximum number of times, expire 3672 * that entry. 3673 */ 3674 static void 3675 syn_cache_timer(void *arg) 3676 { 3677 struct syn_cache *sc = arg; 3678 3679 mutex_enter(softnet_lock); 3680 KERNEL_LOCK(1, NULL); 3681 3682 callout_ack(&sc->sc_timer); 3683 3684 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3685 TCP_STATINC(TCP_STAT_SC_DELAYED_FREE); 3686 goto free; 3687 } 3688 3689 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3690 /* Drop it -- too many retransmissions. */ 3691 goto dropit; 3692 } 3693 3694 /* 3695 * Compute the total amount of time this entry has 3696 * been on a queue. If this entry has been on longer 3697 * than the keep alive timer would allow, expire it. 3698 */ 3699 sc->sc_rxttot += sc->sc_rxtcur; 3700 if (sc->sc_rxttot >= tcp_keepinit) 3701 goto dropit; 3702 3703 TCP_STATINC(TCP_STAT_SC_RETRANSMITTED); 3704 (void)syn_cache_respond(sc); 3705 3706 /* Advance the timer back-off. */ 3707 sc->sc_rxtshift++; 3708 syn_cache_timer_arm(sc); 3709 3710 goto out; 3711 3712 dropit: 3713 TCP_STATINC(TCP_STAT_SC_TIMED_OUT); 3714 syn_cache_rm(sc); 3715 if (sc->sc_ipopts) 3716 (void) m_free(sc->sc_ipopts); 3717 rtcache_free(&sc->sc_route); 3718 3719 free: 3720 callout_destroy(&sc->sc_timer); 3721 pool_put(&syn_cache_pool, sc); 3722 3723 out: 3724 KERNEL_UNLOCK_ONE(NULL); 3725 mutex_exit(softnet_lock); 3726 } 3727 3728 /* 3729 * Remove syn cache created by the specified tcb entry, 3730 * because this does not make sense to keep them 3731 * (if there's no tcb entry, syn cache entry will never be used) 3732 */ 3733 void 3734 syn_cache_cleanup(struct tcpcb *tp) 3735 { 3736 struct syn_cache *sc, *nsc; 3737 int s; 3738 3739 s = splsoftnet(); 3740 3741 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3742 nsc = LIST_NEXT(sc, sc_tpq); 3743 3744 #ifdef DIAGNOSTIC 3745 if (sc->sc_tp != tp) 3746 panic("invalid sc_tp in syn_cache_cleanup"); 3747 #endif 3748 syn_cache_rm(sc); 3749 syn_cache_put(sc); /* calls pool_put but see spl above */ 3750 } 3751 /* just for safety */ 3752 LIST_INIT(&tp->t_sc); 3753 3754 splx(s); 3755 } 3756 3757 /* 3758 * Find an entry in the syn cache. 3759 */ 3760 struct syn_cache * 3761 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3762 struct syn_cache_head **headp) 3763 { 3764 struct syn_cache *sc; 3765 struct syn_cache_head *scp; 3766 u_int32_t hash; 3767 int s; 3768 3769 SYN_HASHALL(hash, src, dst); 3770 3771 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3772 *headp = scp; 3773 s = splsoftnet(); 3774 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3775 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3776 if (sc->sc_hash != hash) 3777 continue; 3778 if (!memcmp(&sc->sc_src, src, src->sa_len) && 3779 !memcmp(&sc->sc_dst, dst, dst->sa_len)) { 3780 splx(s); 3781 return (sc); 3782 } 3783 } 3784 splx(s); 3785 return (NULL); 3786 } 3787 3788 /* 3789 * This function gets called when we receive an ACK for a 3790 * socket in the LISTEN state. We look up the connection 3791 * in the syn cache, and if its there, we pull it out of 3792 * the cache and turn it into a full-blown connection in 3793 * the SYN-RECEIVED state. 3794 * 3795 * The return values may not be immediately obvious, and their effects 3796 * can be subtle, so here they are: 3797 * 3798 * NULL SYN was not found in cache; caller should drop the 3799 * packet and send an RST. 3800 * 3801 * -1 We were unable to create the new connection, and are 3802 * aborting it. An ACK,RST is being sent to the peer 3803 * (unless we got screwey sequence numbners; see below), 3804 * because the 3-way handshake has been completed. Caller 3805 * should not free the mbuf, since we may be using it. If 3806 * we are not, we will free it. 3807 * 3808 * Otherwise, the return value is a pointer to the new socket 3809 * associated with the connection. 3810 */ 3811 struct socket * 3812 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3813 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3814 struct socket *so, struct mbuf *m) 3815 { 3816 struct syn_cache *sc; 3817 struct syn_cache_head *scp; 3818 struct inpcb *inp = NULL; 3819 #ifdef INET6 3820 struct in6pcb *in6p = NULL; 3821 #endif 3822 struct tcpcb *tp = 0; 3823 int s; 3824 struct socket *oso; 3825 3826 s = splsoftnet(); 3827 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3828 splx(s); 3829 return (NULL); 3830 } 3831 3832 /* 3833 * Verify the sequence and ack numbers. Try getting the correct 3834 * response again. 3835 */ 3836 if ((th->th_ack != sc->sc_iss + 1) || 3837 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3838 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3839 m_freem(m); 3840 (void)syn_cache_respond(sc); 3841 splx(s); 3842 return ((struct socket *)(-1)); 3843 } 3844 3845 /* Remove this cache entry */ 3846 syn_cache_rm(sc); 3847 splx(s); 3848 3849 /* 3850 * Ok, create the full blown connection, and set things up 3851 * as they would have been set up if we had created the 3852 * connection when the SYN arrived. If we can't create 3853 * the connection, abort it. 3854 */ 3855 /* 3856 * inp still has the OLD in_pcb stuff, set the 3857 * v6-related flags on the new guy, too. This is 3858 * done particularly for the case where an AF_INET6 3859 * socket is bound only to a port, and a v4 connection 3860 * comes in on that port. 3861 * we also copy the flowinfo from the original pcb 3862 * to the new one. 3863 */ 3864 oso = so; 3865 so = sonewconn(so, true); 3866 if (so == NULL) 3867 goto resetandabort; 3868 3869 switch (so->so_proto->pr_domain->dom_family) { 3870 #ifdef INET 3871 case AF_INET: 3872 inp = sotoinpcb(so); 3873 break; 3874 #endif 3875 #ifdef INET6 3876 case AF_INET6: 3877 in6p = sotoin6pcb(so); 3878 break; 3879 #endif 3880 } 3881 switch (src->sa_family) { 3882 #ifdef INET 3883 case AF_INET: 3884 if (inp) { 3885 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3886 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3887 inp->inp_options = ip_srcroute(m); 3888 in_pcbstate(inp, INP_BOUND); 3889 if (inp->inp_options == NULL) { 3890 inp->inp_options = sc->sc_ipopts; 3891 sc->sc_ipopts = NULL; 3892 } 3893 } 3894 #ifdef INET6 3895 else if (in6p) { 3896 /* IPv4 packet to AF_INET6 socket */ 3897 memset(&in6p->in6p_laddr, 0, sizeof(in6p->in6p_laddr)); 3898 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3899 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3900 &in6p->in6p_laddr.s6_addr32[3], 3901 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3902 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3903 in6totcpcb(in6p)->t_family = AF_INET; 3904 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3905 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3906 else 3907 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3908 in6_pcbstate(in6p, IN6P_BOUND); 3909 } 3910 #endif 3911 break; 3912 #endif 3913 #ifdef INET6 3914 case AF_INET6: 3915 if (in6p) { 3916 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3917 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3918 in6_pcbstate(in6p, IN6P_BOUND); 3919 } 3920 break; 3921 #endif 3922 } 3923 #ifdef INET6 3924 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3925 struct in6pcb *oin6p = sotoin6pcb(oso); 3926 /* inherit socket options from the listening socket */ 3927 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3928 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3929 m_freem(in6p->in6p_options); 3930 in6p->in6p_options = 0; 3931 } 3932 ip6_savecontrol(in6p, &in6p->in6p_options, 3933 mtod(m, struct ip6_hdr *), m); 3934 } 3935 #endif 3936 3937 #if defined(IPSEC) 3938 if (ipsec_used) { 3939 /* 3940 * we make a copy of policy, instead of sharing the policy, for 3941 * better behavior in terms of SA lookup and dead SA removal. 3942 */ 3943 if (inp) { 3944 /* copy old policy into new socket's */ 3945 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, 3946 inp->inp_sp)) 3947 printf("tcp_input: could not copy policy\n"); 3948 } 3949 #ifdef INET6 3950 else if (in6p) { 3951 /* copy old policy into new socket's */ 3952 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3953 in6p->in6p_sp)) 3954 printf("tcp_input: could not copy policy\n"); 3955 } 3956 #endif 3957 } 3958 #endif 3959 3960 /* 3961 * Give the new socket our cached route reference. 3962 */ 3963 if (inp) { 3964 rtcache_copy(&inp->inp_route, &sc->sc_route); 3965 rtcache_free(&sc->sc_route); 3966 } 3967 #ifdef INET6 3968 else { 3969 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 3970 rtcache_free(&sc->sc_route); 3971 } 3972 #endif 3973 3974 if (inp) { 3975 struct sockaddr_in sin; 3976 memcpy(&sin, src, src->sa_len); 3977 if (in_pcbconnect(inp, &sin, &lwp0)) { 3978 goto resetandabort; 3979 } 3980 } 3981 #ifdef INET6 3982 else if (in6p) { 3983 struct sockaddr_in6 sin6; 3984 memcpy(&sin6, src, src->sa_len); 3985 if (src->sa_family == AF_INET) { 3986 /* IPv4 packet to AF_INET6 socket */ 3987 in6_sin_2_v4mapsin6((struct sockaddr_in *)src, &sin6); 3988 } 3989 if (in6_pcbconnect(in6p, &sin6, NULL)) { 3990 goto resetandabort; 3991 } 3992 } 3993 #endif 3994 else { 3995 goto resetandabort; 3996 } 3997 3998 if (inp) 3999 tp = intotcpcb(inp); 4000 #ifdef INET6 4001 else if (in6p) 4002 tp = in6totcpcb(in6p); 4003 #endif 4004 else 4005 tp = NULL; 4006 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 4007 if (sc->sc_request_r_scale != 15) { 4008 tp->requested_s_scale = sc->sc_requested_s_scale; 4009 tp->request_r_scale = sc->sc_request_r_scale; 4010 tp->snd_scale = sc->sc_requested_s_scale; 4011 tp->rcv_scale = sc->sc_request_r_scale; 4012 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 4013 } 4014 if (sc->sc_flags & SCF_TIMESTAMP) 4015 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 4016 tp->ts_timebase = sc->sc_timebase; 4017 4018 tp->t_template = tcp_template(tp); 4019 if (tp->t_template == 0) { 4020 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 4021 so = NULL; 4022 m_freem(m); 4023 goto abort; 4024 } 4025 4026 tp->iss = sc->sc_iss; 4027 tp->irs = sc->sc_irs; 4028 tcp_sendseqinit(tp); 4029 tcp_rcvseqinit(tp); 4030 tp->t_state = TCPS_SYN_RECEIVED; 4031 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 4032 TCP_STATINC(TCP_STAT_ACCEPTS); 4033 4034 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 4035 tp->t_flags |= TF_WILL_SACK; 4036 4037 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 4038 tp->t_flags |= TF_ECN_PERMIT; 4039 4040 #ifdef TCP_SIGNATURE 4041 if (sc->sc_flags & SCF_SIGNATURE) 4042 tp->t_flags |= TF_SIGNATURE; 4043 #endif 4044 4045 /* Initialize tp->t_ourmss before we deal with the peer's! */ 4046 tp->t_ourmss = sc->sc_ourmaxseg; 4047 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 4048 4049 /* 4050 * Initialize the initial congestion window. If we 4051 * had to retransmit the SYN,ACK, we must initialize cwnd 4052 * to 1 segment (i.e. the Loss Window). 4053 */ 4054 if (sc->sc_rxtshift) 4055 tp->snd_cwnd = tp->t_peermss; 4056 else { 4057 int ss = tcp_init_win; 4058 #ifdef INET 4059 if (inp != NULL && in_localaddr(inp->inp_faddr)) 4060 ss = tcp_init_win_local; 4061 #endif 4062 #ifdef INET6 4063 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 4064 ss = tcp_init_win_local; 4065 #endif 4066 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 4067 } 4068 4069 tcp_rmx_rtt(tp); 4070 tp->snd_wl1 = sc->sc_irs; 4071 tp->rcv_up = sc->sc_irs + 1; 4072 4073 /* 4074 * This is what whould have happened in tcp_output() when 4075 * the SYN,ACK was sent. 4076 */ 4077 tp->snd_up = tp->snd_una; 4078 tp->snd_max = tp->snd_nxt = tp->iss+1; 4079 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 4080 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 4081 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 4082 tp->last_ack_sent = tp->rcv_nxt; 4083 tp->t_partialacks = -1; 4084 tp->t_dupacks = 0; 4085 4086 TCP_STATINC(TCP_STAT_SC_COMPLETED); 4087 s = splsoftnet(); 4088 syn_cache_put(sc); 4089 splx(s); 4090 return (so); 4091 4092 resetandabort: 4093 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 4094 abort: 4095 if (so != NULL) { 4096 (void) soqremque(so, 1); 4097 (void) soabort(so); 4098 mutex_enter(softnet_lock); 4099 } 4100 s = splsoftnet(); 4101 syn_cache_put(sc); 4102 splx(s); 4103 TCP_STATINC(TCP_STAT_SC_ABORTED); 4104 return ((struct socket *)(-1)); 4105 } 4106 4107 /* 4108 * This function is called when we get a RST for a 4109 * non-existent connection, so that we can see if the 4110 * connection is in the syn cache. If it is, zap it. 4111 */ 4112 4113 void 4114 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 4115 { 4116 struct syn_cache *sc; 4117 struct syn_cache_head *scp; 4118 int s = splsoftnet(); 4119 4120 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4121 splx(s); 4122 return; 4123 } 4124 if (SEQ_LT(th->th_seq, sc->sc_irs) || 4125 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 4126 splx(s); 4127 return; 4128 } 4129 syn_cache_rm(sc); 4130 TCP_STATINC(TCP_STAT_SC_RESET); 4131 syn_cache_put(sc); /* calls pool_put but see spl above */ 4132 splx(s); 4133 } 4134 4135 void 4136 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 4137 struct tcphdr *th) 4138 { 4139 struct syn_cache *sc; 4140 struct syn_cache_head *scp; 4141 int s; 4142 4143 s = splsoftnet(); 4144 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4145 splx(s); 4146 return; 4147 } 4148 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 4149 if (ntohl (th->th_seq) != sc->sc_iss) { 4150 splx(s); 4151 return; 4152 } 4153 4154 /* 4155 * If we've retransmitted 3 times and this is our second error, 4156 * we remove the entry. Otherwise, we allow it to continue on. 4157 * This prevents us from incorrectly nuking an entry during a 4158 * spurious network outage. 4159 * 4160 * See tcp_notify(). 4161 */ 4162 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 4163 sc->sc_flags |= SCF_UNREACH; 4164 splx(s); 4165 return; 4166 } 4167 4168 syn_cache_rm(sc); 4169 TCP_STATINC(TCP_STAT_SC_UNREACH); 4170 syn_cache_put(sc); /* calls pool_put but see spl above */ 4171 splx(s); 4172 } 4173 4174 /* 4175 * Given a LISTEN socket and an inbound SYN request, add 4176 * this to the syn cache, and send back a segment: 4177 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 4178 * to the source. 4179 * 4180 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 4181 * Doing so would require that we hold onto the data and deliver it 4182 * to the application. However, if we are the target of a SYN-flood 4183 * DoS attack, an attacker could send data which would eventually 4184 * consume all available buffer space if it were ACKed. By not ACKing 4185 * the data, we avoid this DoS scenario. 4186 */ 4187 int 4188 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 4189 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 4190 int optlen, struct tcp_opt_info *oi) 4191 { 4192 struct tcpcb tb, *tp; 4193 long win; 4194 struct syn_cache *sc; 4195 struct syn_cache_head *scp; 4196 struct mbuf *ipopts; 4197 struct tcp_opt_info opti; 4198 int s; 4199 4200 tp = sototcpcb(so); 4201 4202 memset(&opti, 0, sizeof(opti)); 4203 4204 /* 4205 * Initialize some local state. 4206 */ 4207 win = sbspace(&so->so_rcv); 4208 if (win > TCP_MAXWIN) 4209 win = TCP_MAXWIN; 4210 4211 #ifdef TCP_SIGNATURE 4212 if (optp || (tp->t_flags & TF_SIGNATURE)) 4213 #else 4214 if (optp) 4215 #endif 4216 { 4217 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4218 #ifdef TCP_SIGNATURE 4219 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4220 #endif 4221 tb.t_state = TCPS_LISTEN; 4222 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 4223 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 4224 return 0; 4225 } else 4226 tb.t_flags = 0; 4227 4228 switch (src->sa_family) { 4229 #ifdef INET 4230 case AF_INET: 4231 /* 4232 * Remember the IP options, if any. 4233 */ 4234 ipopts = ip_srcroute(m); 4235 break; 4236 #endif 4237 default: 4238 ipopts = NULL; 4239 } 4240 4241 /* 4242 * See if we already have an entry for this connection. 4243 * If we do, resend the SYN,ACK. We do not count this 4244 * as a retransmission (XXX though maybe we should). 4245 */ 4246 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4247 TCP_STATINC(TCP_STAT_SC_DUPESYN); 4248 if (ipopts) { 4249 /* 4250 * If we were remembering a previous source route, 4251 * forget it and use the new one we've been given. 4252 */ 4253 if (sc->sc_ipopts) 4254 (void)m_free(sc->sc_ipopts); 4255 sc->sc_ipopts = ipopts; 4256 } 4257 sc->sc_timestamp = tb.ts_recent; 4258 m_freem(m); 4259 if (syn_cache_respond(sc) == 0) { 4260 uint64_t *tcps = TCP_STAT_GETREF(); 4261 tcps[TCP_STAT_SNDACKS]++; 4262 tcps[TCP_STAT_SNDTOTAL]++; 4263 TCP_STAT_PUTREF(); 4264 } 4265 return 1; 4266 } 4267 4268 s = splsoftnet(); 4269 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4270 splx(s); 4271 if (sc == NULL) { 4272 if (ipopts) 4273 (void)m_free(ipopts); 4274 return 0; 4275 } 4276 4277 /* 4278 * Fill in the cache, and put the necessary IP and TCP 4279 * options into the reply. 4280 */ 4281 memset(sc, 0, sizeof(struct syn_cache)); 4282 callout_init(&sc->sc_timer, CALLOUT_MPSAFE); 4283 bcopy(src, &sc->sc_src, src->sa_len); 4284 bcopy(dst, &sc->sc_dst, dst->sa_len); 4285 sc->sc_flags = 0; 4286 sc->sc_ipopts = ipopts; 4287 sc->sc_irs = th->th_seq; 4288 switch (src->sa_family) { 4289 #ifdef INET 4290 case AF_INET: 4291 { 4292 struct sockaddr_in *srcin = (void *)src; 4293 struct sockaddr_in *dstin = (void *)dst; 4294 4295 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4296 &srcin->sin_addr, dstin->sin_port, 4297 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4298 break; 4299 } 4300 #endif /* INET */ 4301 #ifdef INET6 4302 case AF_INET6: 4303 { 4304 struct sockaddr_in6 *srcin6 = (void *)src; 4305 struct sockaddr_in6 *dstin6 = (void *)dst; 4306 4307 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4308 &srcin6->sin6_addr, dstin6->sin6_port, 4309 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4310 break; 4311 } 4312 #endif /* INET6 */ 4313 } 4314 sc->sc_peermaxseg = oi->maxseg; 4315 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4316 m_get_rcvif_NOMPSAFE(m) : NULL, sc->sc_src.sa.sa_family); 4317 sc->sc_win = win; 4318 sc->sc_timebase = tcp_now - 1; /* see tcp_newtcpcb() */ 4319 sc->sc_timestamp = tb.ts_recent; 4320 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4321 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4322 sc->sc_flags |= SCF_TIMESTAMP; 4323 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4324 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4325 sc->sc_requested_s_scale = tb.requested_s_scale; 4326 sc->sc_request_r_scale = 0; 4327 /* 4328 * Pick the smallest possible scaling factor that 4329 * will still allow us to scale up to sb_max. 4330 * 4331 * We do this because there are broken firewalls that 4332 * will corrupt the window scale option, leading to 4333 * the other endpoint believing that our advertised 4334 * window is unscaled. At scale factors larger than 4335 * 5 the unscaled window will drop below 1500 bytes, 4336 * leading to serious problems when traversing these 4337 * broken firewalls. 4338 * 4339 * With the default sbmax of 256K, a scale factor 4340 * of 3 will be chosen by this algorithm. Those who 4341 * choose a larger sbmax should watch out 4342 * for the compatiblity problems mentioned above. 4343 * 4344 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4345 * or <SYN,ACK>) segment itself is never scaled. 4346 */ 4347 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4348 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4349 sc->sc_request_r_scale++; 4350 } else { 4351 sc->sc_requested_s_scale = 15; 4352 sc->sc_request_r_scale = 15; 4353 } 4354 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4355 sc->sc_flags |= SCF_SACK_PERMIT; 4356 4357 /* 4358 * ECN setup packet received. 4359 */ 4360 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4361 sc->sc_flags |= SCF_ECN_PERMIT; 4362 4363 #ifdef TCP_SIGNATURE 4364 if (tb.t_flags & TF_SIGNATURE) 4365 sc->sc_flags |= SCF_SIGNATURE; 4366 #endif 4367 sc->sc_tp = tp; 4368 m_freem(m); 4369 if (syn_cache_respond(sc) == 0) { 4370 uint64_t *tcps = TCP_STAT_GETREF(); 4371 tcps[TCP_STAT_SNDACKS]++; 4372 tcps[TCP_STAT_SNDTOTAL]++; 4373 TCP_STAT_PUTREF(); 4374 syn_cache_insert(sc, tp); 4375 } else { 4376 s = splsoftnet(); 4377 /* 4378 * syn_cache_put() will try to schedule the timer, so 4379 * we need to initialize it 4380 */ 4381 syn_cache_timer_arm(sc); 4382 syn_cache_put(sc); 4383 splx(s); 4384 TCP_STATINC(TCP_STAT_SC_DROPPED); 4385 } 4386 return 1; 4387 } 4388 4389 /* 4390 * syn_cache_respond: (re)send SYN+ACK. 4391 * 4392 * returns 0 on success. otherwise returns an errno, typically ENOBUFS. 4393 */ 4394 4395 int 4396 syn_cache_respond(struct syn_cache *sc) 4397 { 4398 #ifdef INET6 4399 struct rtentry *rt = NULL; 4400 #endif 4401 struct route *ro; 4402 u_int8_t *optp; 4403 int optlen, error; 4404 u_int16_t tlen; 4405 struct ip *ip = NULL; 4406 #ifdef INET6 4407 struct ip6_hdr *ip6 = NULL; 4408 #endif 4409 struct tcpcb *tp = NULL; 4410 struct tcphdr *th; 4411 struct mbuf *m; 4412 u_int hlen; 4413 #ifdef TCP_SIGNATURE 4414 struct secasvar *sav = NULL; 4415 u_int8_t *sigp = NULL; 4416 #endif 4417 4418 ro = &sc->sc_route; 4419 switch (sc->sc_src.sa.sa_family) { 4420 case AF_INET: 4421 hlen = sizeof(struct ip); 4422 break; 4423 #ifdef INET6 4424 case AF_INET6: 4425 hlen = sizeof(struct ip6_hdr); 4426 break; 4427 #endif 4428 default: 4429 return (EAFNOSUPPORT); 4430 } 4431 4432 /* worst case scanario, since we don't know the option size yet */ 4433 tlen = hlen + sizeof(struct tcphdr) + MAX_TCPOPTLEN; 4434 4435 /* 4436 * Create the IP+TCP header from scratch. 4437 */ 4438 #ifdef DIAGNOSTIC 4439 if (max_linkhdr + tlen > MCLBYTES) 4440 return ENOBUFS; 4441 #endif 4442 4443 MGETHDR(m, M_DONTWAIT, MT_DATA); 4444 if (m && (max_linkhdr + tlen) > MHLEN) { 4445 MCLGET(m, M_DONTWAIT); 4446 if ((m->m_flags & M_EXT) == 0) { 4447 m_freem(m); 4448 m = NULL; 4449 } 4450 } 4451 if (m == NULL) 4452 return ENOBUFS; 4453 MCLAIM(m, &tcp_tx_mowner); 4454 4455 /* Fixup the mbuf. */ 4456 m->m_data += max_linkhdr; 4457 if (sc->sc_tp) 4458 tp = sc->sc_tp; 4459 m_reset_rcvif(m); 4460 memset(mtod(m, u_char *), 0, tlen); 4461 4462 switch (sc->sc_src.sa.sa_family) { 4463 case AF_INET: 4464 ip = mtod(m, struct ip *); 4465 ip->ip_v = 4; 4466 ip->ip_dst = sc->sc_src.sin.sin_addr; 4467 ip->ip_src = sc->sc_dst.sin.sin_addr; 4468 ip->ip_p = IPPROTO_TCP; 4469 th = (struct tcphdr *)(ip + 1); 4470 th->th_dport = sc->sc_src.sin.sin_port; 4471 th->th_sport = sc->sc_dst.sin.sin_port; 4472 break; 4473 #ifdef INET6 4474 case AF_INET6: 4475 ip6 = mtod(m, struct ip6_hdr *); 4476 ip6->ip6_vfc = IPV6_VERSION; 4477 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4478 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4479 ip6->ip6_nxt = IPPROTO_TCP; 4480 /* ip6_plen will be updated in ip6_output() */ 4481 th = (struct tcphdr *)(ip6 + 1); 4482 th->th_dport = sc->sc_src.sin6.sin6_port; 4483 th->th_sport = sc->sc_dst.sin6.sin6_port; 4484 break; 4485 #endif 4486 default: 4487 return ENOBUFS; 4488 } 4489 4490 th->th_seq = htonl(sc->sc_iss); 4491 th->th_ack = htonl(sc->sc_irs + 1); 4492 th->th_flags = TH_SYN|TH_ACK; 4493 th->th_win = htons(sc->sc_win); 4494 /* th_x2, th_sum, th_urp already 0 from memset */ 4495 4496 /* Tack on the TCP options. */ 4497 optp = (u_int8_t *)(th + 1); 4498 optlen = 0; 4499 *optp++ = TCPOPT_MAXSEG; 4500 *optp++ = TCPOLEN_MAXSEG; 4501 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4502 *optp++ = sc->sc_ourmaxseg & 0xff; 4503 optlen += TCPOLEN_MAXSEG; 4504 4505 if (sc->sc_request_r_scale != 15) { 4506 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4507 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4508 sc->sc_request_r_scale); 4509 optp += TCPOLEN_WINDOW + TCPOLEN_NOP; 4510 optlen += TCPOLEN_WINDOW + TCPOLEN_NOP; 4511 } 4512 4513 if (sc->sc_flags & SCF_SACK_PERMIT) { 4514 /* Let the peer know that we will SACK. */ 4515 *optp++ = TCPOPT_SACK_PERMITTED; 4516 *optp++ = TCPOLEN_SACK_PERMITTED; 4517 optlen += TCPOLEN_SACK_PERMITTED; 4518 } 4519 4520 if (sc->sc_flags & SCF_TIMESTAMP) { 4521 while (optlen % 4 != 2) { 4522 optlen += TCPOLEN_NOP; 4523 *optp++ = TCPOPT_NOP; 4524 } 4525 *optp++ = TCPOPT_TIMESTAMP; 4526 *optp++ = TCPOLEN_TIMESTAMP; 4527 u_int32_t *lp = (u_int32_t *)(optp); 4528 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4529 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4530 *lp = htonl(sc->sc_timestamp); 4531 optp += TCPOLEN_TIMESTAMP - 2; 4532 optlen += TCPOLEN_TIMESTAMP; 4533 } 4534 4535 #ifdef TCP_SIGNATURE 4536 if (sc->sc_flags & SCF_SIGNATURE) { 4537 sav = tcp_signature_getsav(m); 4538 if (sav == NULL) { 4539 if (m) 4540 m_freem(m); 4541 return (EPERM); 4542 } 4543 4544 *optp++ = TCPOPT_SIGNATURE; 4545 *optp++ = TCPOLEN_SIGNATURE; 4546 sigp = optp; 4547 memset(optp, 0, TCP_SIGLEN); 4548 optp += TCP_SIGLEN; 4549 optlen += TCPOLEN_SIGNATURE; 4550 4551 } 4552 #endif 4553 /* Terminate and pad TCP options to a 4 byte boundary. */ 4554 if (optlen % 4) { 4555 optlen += TCPOLEN_EOL; 4556 *optp++ = TCPOPT_EOL; 4557 } 4558 /* 4559 * According to RFC 793 (STD0007): 4560 * "The content of the header beyond the End-of-Option option 4561 * must be header padding (i.e., zero)." 4562 * and later: "The padding is composed of zeros." 4563 */ 4564 while (optlen % 4) { 4565 optlen += TCPOLEN_PAD; 4566 *optp++ = TCPOPT_PAD; 4567 } 4568 4569 /* compute the actual values now that we've added the options */ 4570 tlen = hlen + sizeof(struct tcphdr) + optlen; 4571 m->m_len = m->m_pkthdr.len = tlen; 4572 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4573 4574 #ifdef TCP_SIGNATURE 4575 if (sav) { 4576 (void)tcp_signature(m, th, hlen, sav, sigp); 4577 key_sa_recordxfer(sav, m); 4578 KEY_SA_UNREF(&sav); 4579 } 4580 #endif 4581 4582 /* 4583 * Send ECN SYN-ACK setup packet. 4584 * Routes can be asymetric, so, even if we receive a packet 4585 * with ECE and CWR set, we must not assume no one will block 4586 * the ECE packet we are about to send. 4587 */ 4588 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4589 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4590 th->th_flags |= TH_ECE; 4591 TCP_STATINC(TCP_STAT_ECN_SHS); 4592 4593 /* 4594 * draft-ietf-tcpm-ecnsyn-00.txt 4595 * 4596 * "[...] a TCP node MAY respond to an ECN-setup 4597 * SYN packet by setting ECT in the responding 4598 * ECN-setup SYN/ACK packet, indicating to routers 4599 * that the SYN/ACK packet is ECN-Capable. 4600 * This allows a congested router along the path 4601 * to mark the packet instead of dropping the 4602 * packet as an indication of congestion." 4603 * 4604 * "[...] There can be a great benefit in setting 4605 * an ECN-capable codepoint in SYN/ACK packets [...] 4606 * Congestion is most likely to occur in 4607 * the server-to-client direction. As a result, 4608 * setting an ECN-capable codepoint in SYN/ACK 4609 * packets can reduce the occurence of three-second 4610 * retransmit timeouts resulting from the drop 4611 * of SYN/ACK packets." 4612 * 4613 * Page 4 and 6, January 2006. 4614 */ 4615 4616 switch (sc->sc_src.sa.sa_family) { 4617 #ifdef INET 4618 case AF_INET: 4619 ip->ip_tos |= IPTOS_ECN_ECT0; 4620 break; 4621 #endif 4622 #ifdef INET6 4623 case AF_INET6: 4624 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4625 break; 4626 #endif 4627 } 4628 TCP_STATINC(TCP_STAT_ECN_ECT); 4629 } 4630 4631 4632 /* Compute the packet's checksum. */ 4633 switch (sc->sc_src.sa.sa_family) { 4634 case AF_INET: 4635 ip->ip_len = htons(tlen - hlen); 4636 th->th_sum = 0; 4637 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4638 break; 4639 #ifdef INET6 4640 case AF_INET6: 4641 ip6->ip6_plen = htons(tlen - hlen); 4642 th->th_sum = 0; 4643 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4644 break; 4645 #endif 4646 } 4647 4648 /* 4649 * Fill in some straggling IP bits. Note the stack expects 4650 * ip_len to be in host order, for convenience. 4651 */ 4652 switch (sc->sc_src.sa.sa_family) { 4653 #ifdef INET 4654 case AF_INET: 4655 ip->ip_len = htons(tlen); 4656 ip->ip_ttl = ip_defttl; 4657 /* XXX tos? */ 4658 break; 4659 #endif 4660 #ifdef INET6 4661 case AF_INET6: 4662 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4663 ip6->ip6_vfc |= IPV6_VERSION; 4664 ip6->ip6_plen = htons(tlen - hlen); 4665 /* ip6_hlim will be initialized afterwards */ 4666 /* XXX flowlabel? */ 4667 break; 4668 #endif 4669 } 4670 4671 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4672 tp = sc->sc_tp; 4673 4674 switch (sc->sc_src.sa.sa_family) { 4675 #ifdef INET 4676 case AF_INET: 4677 error = ip_output(m, sc->sc_ipopts, ro, 4678 (ip_mtudisc ? IP_MTUDISC : 0), 4679 NULL, tp ? tp->t_inpcb : NULL); 4680 break; 4681 #endif 4682 #ifdef INET6 4683 case AF_INET6: 4684 ip6->ip6_hlim = in6_selecthlim(NULL, 4685 (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp : NULL); 4686 rtcache_unref(rt, ro); 4687 4688 error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, 4689 tp ? tp->t_in6pcb : NULL, NULL); 4690 break; 4691 #endif 4692 default: 4693 error = EAFNOSUPPORT; 4694 break; 4695 } 4696 return (error); 4697 } 4698
Indexes created Sun Sep 21 20:09:37 GMT 2025