tcp_input.c revision 1.383
1 /* $NetBSD: tcp_input.c,v 1.383 2018/03/01 06:08:43 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 1.383 2018/03/01 06:08:43 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 /*IPSEC*/ 1503 break; 1504 } 1505 #endif 1506 } 1507 1508 /* 1509 * If the state is CLOSED (i.e., TCB does not exist) then 1510 * all data in the incoming segment is discarded. 1511 * If the TCB exists but is in CLOSED state, it is embryonic, 1512 * but should either do a listen or a connect soon. 1513 */ 1514 tp = NULL; 1515 so = NULL; 1516 if (inp) { 1517 /* Check the minimum TTL for socket. */ 1518 if (ip->ip_ttl < inp->inp_ip_minttl) 1519 goto drop; 1520 1521 tp = intotcpcb(inp); 1522 so = inp->inp_socket; 1523 } 1524 #ifdef INET6 1525 else if (in6p) { 1526 tp = in6totcpcb(in6p); 1527 so = in6p->in6p_socket; 1528 } 1529 #endif 1530 else if (vestige.valid) { 1531 /* We do not support the resurrection of vtw tcpcps. */ 1532 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1533 goto badcsum; 1534 1535 tcp_fields_to_host(th); 1536 tcp_vtw_input(th, &vestige, m, tlen); 1537 m = NULL; 1538 goto drop; 1539 } 1540 1541 if (tp == NULL) { 1542 tcp_fields_to_host(th); 1543 goto dropwithreset_ratelim; 1544 } 1545 if (tp->t_state == TCPS_CLOSED) 1546 goto drop; 1547 1548 KASSERT(so->so_lock == softnet_lock); 1549 KASSERT(solocked(so)); 1550 1551 tcp_fields_to_host(th); 1552 1553 /* Unscale the window into a 32-bit value. */ 1554 if ((tiflags & TH_SYN) == 0) 1555 tiwin = th->th_win << tp->snd_scale; 1556 else 1557 tiwin = th->th_win; 1558 1559 #ifdef INET6 1560 /* save packet options if user wanted */ 1561 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1562 if (in6p->in6p_options) { 1563 m_freem(in6p->in6p_options); 1564 in6p->in6p_options = NULL; 1565 } 1566 KASSERT(ip6 != NULL); 1567 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1568 } 1569 #endif 1570 1571 if (so->so_options & SO_DEBUG) { 1572 #ifdef TCP_DEBUG 1573 ostate = tp->t_state; 1574 #endif 1575 1576 tcp_saveti = NULL; 1577 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1578 goto nosave; 1579 1580 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1581 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1582 if (tcp_saveti == NULL) 1583 goto nosave; 1584 } else { 1585 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1586 if (tcp_saveti == NULL) 1587 goto nosave; 1588 MCLAIM(m, &tcp_mowner); 1589 tcp_saveti->m_len = iphlen; 1590 m_copydata(m, 0, iphlen, 1591 mtod(tcp_saveti, void *)); 1592 } 1593 1594 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1595 m_freem(tcp_saveti); 1596 tcp_saveti = NULL; 1597 } else { 1598 tcp_saveti->m_len += sizeof(struct tcphdr); 1599 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1600 sizeof(struct tcphdr)); 1601 } 1602 nosave:; 1603 } 1604 1605 if (so->so_options & SO_ACCEPTCONN) { 1606 union syn_cache_sa src; 1607 union syn_cache_sa dst; 1608 1609 KASSERT(tp->t_state == TCPS_LISTEN); 1610 1611 memset(&src, 0, sizeof(src)); 1612 memset(&dst, 0, sizeof(dst)); 1613 switch (af) { 1614 #ifdef INET 1615 case AF_INET: 1616 src.sin.sin_len = sizeof(struct sockaddr_in); 1617 src.sin.sin_family = AF_INET; 1618 src.sin.sin_addr = ip->ip_src; 1619 src.sin.sin_port = th->th_sport; 1620 1621 dst.sin.sin_len = sizeof(struct sockaddr_in); 1622 dst.sin.sin_family = AF_INET; 1623 dst.sin.sin_addr = ip->ip_dst; 1624 dst.sin.sin_port = th->th_dport; 1625 break; 1626 #endif 1627 #ifdef INET6 1628 case AF_INET6: 1629 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1630 src.sin6.sin6_family = AF_INET6; 1631 src.sin6.sin6_addr = ip6->ip6_src; 1632 src.sin6.sin6_port = th->th_sport; 1633 1634 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1635 dst.sin6.sin6_family = AF_INET6; 1636 dst.sin6.sin6_addr = ip6->ip6_dst; 1637 dst.sin6.sin6_port = th->th_dport; 1638 break; 1639 #endif /* INET6 */ 1640 } 1641 1642 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1643 if (tiflags & TH_RST) { 1644 syn_cache_reset(&src.sa, &dst.sa, th); 1645 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1646 (TH_ACK|TH_SYN)) { 1647 /* 1648 * Received a SYN,ACK. This should never 1649 * happen while we are in LISTEN. Send an RST. 1650 */ 1651 goto badsyn; 1652 } else if (tiflags & TH_ACK) { 1653 so = syn_cache_get(&src.sa, &dst.sa, 1654 th, toff, tlen, so, m); 1655 if (so == NULL) { 1656 /* 1657 * We don't have a SYN for this ACK; 1658 * send an RST. 1659 */ 1660 goto badsyn; 1661 } else if (so == (struct socket *)(-1)) { 1662 /* 1663 * We were unable to create the 1664 * connection. If the 3-way handshake 1665 * was completed, and RST has been 1666 * sent to the peer. Since the mbuf 1667 * might be in use for the reply, do 1668 * not free it. 1669 */ 1670 m = NULL; 1671 } else { 1672 /* 1673 * We have created a full-blown 1674 * connection. 1675 */ 1676 tp = NULL; 1677 inp = NULL; 1678 #ifdef INET6 1679 in6p = NULL; 1680 #endif 1681 switch (so->so_proto->pr_domain->dom_family) { 1682 #ifdef INET 1683 case AF_INET: 1684 inp = sotoinpcb(so); 1685 tp = intotcpcb(inp); 1686 break; 1687 #endif 1688 #ifdef INET6 1689 case AF_INET6: 1690 in6p = sotoin6pcb(so); 1691 tp = in6totcpcb(in6p); 1692 break; 1693 #endif 1694 } 1695 if (tp == NULL) 1696 goto badsyn; /*XXX*/ 1697 tiwin <<= tp->snd_scale; 1698 goto after_listen; 1699 } 1700 } else { 1701 /* 1702 * None of RST, SYN or ACK was set. 1703 * This is an invalid packet for a 1704 * TCB in LISTEN state. Send a RST. 1705 */ 1706 goto badsyn; 1707 } 1708 } else { 1709 /* 1710 * Received a SYN. 1711 */ 1712 1713 #ifdef INET6 1714 /* 1715 * If deprecated address is forbidden, we do 1716 * not accept SYN to deprecated interface 1717 * address to prevent any new inbound 1718 * connection from getting established. 1719 * When we do not accept SYN, we send a TCP 1720 * RST, with deprecated source address (instead 1721 * of dropping it). We compromise it as it is 1722 * much better for peer to send a RST, and 1723 * RST will be the final packet for the 1724 * exchange. 1725 * 1726 * If we do not forbid deprecated addresses, we 1727 * accept the SYN packet. RFC2462 does not 1728 * suggest dropping SYN in this case. 1729 * If we decipher RFC2462 5.5.4, it says like 1730 * this: 1731 * 1. use of deprecated addr with existing 1732 * communication is okay - "SHOULD continue 1733 * to be used" 1734 * 2. use of it with new communication: 1735 * (2a) "SHOULD NOT be used if alternate 1736 * address with sufficient scope is 1737 * available" 1738 * (2b) nothing mentioned otherwise. 1739 * Here we fall into (2b) case as we have no 1740 * choice in our source address selection - we 1741 * must obey the peer. 1742 * 1743 * The wording in RFC2462 is confusing, and 1744 * there are multiple description text for 1745 * deprecated address handling - worse, they 1746 * are not exactly the same. I believe 5.5.4 1747 * is the best one, so we follow 5.5.4. 1748 */ 1749 if (af == AF_INET6 && !ip6_use_deprecated) { 1750 struct in6_ifaddr *ia6; 1751 int s; 1752 struct ifnet *rcvif = m_get_rcvif(m, &s); 1753 if (rcvif == NULL) 1754 goto dropwithreset; /* XXX */ 1755 if ((ia6 = in6ifa_ifpwithaddr(rcvif, 1756 &ip6->ip6_dst)) && 1757 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1758 tp = NULL; 1759 m_put_rcvif(rcvif, &s); 1760 goto dropwithreset; 1761 } 1762 m_put_rcvif(rcvif, &s); 1763 } 1764 #endif 1765 1766 /* 1767 * LISTEN socket received a SYN from itself? This 1768 * can't possibly be valid; drop the packet. 1769 */ 1770 if (th->th_sport == th->th_dport) { 1771 int eq = 0; 1772 1773 switch (af) { 1774 #ifdef INET 1775 case AF_INET: 1776 eq = in_hosteq(ip->ip_src, ip->ip_dst); 1777 break; 1778 #endif 1779 #ifdef INET6 1780 case AF_INET6: 1781 eq = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, 1782 &ip6->ip6_dst); 1783 break; 1784 #endif 1785 } 1786 if (eq) { 1787 TCP_STATINC(TCP_STAT_BADSYN); 1788 goto drop; 1789 } 1790 } 1791 1792 /* 1793 * SYN looks ok; create compressed TCP 1794 * state for it. 1795 */ 1796 if (so->so_qlen <= so->so_qlimit && 1797 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1798 so, m, optp, optlen, &opti)) 1799 m = NULL; 1800 } 1801 1802 goto drop; 1803 } 1804 1805 after_listen: 1806 /* 1807 * Should not happen now that all embryonic connections 1808 * are handled with compressed state. 1809 */ 1810 KASSERT(tp->t_state != TCPS_LISTEN); 1811 1812 /* 1813 * Segment received on connection. 1814 * Reset idle time and keep-alive timer. 1815 */ 1816 tp->t_rcvtime = tcp_now; 1817 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1818 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1819 1820 /* 1821 * Process options. 1822 */ 1823 #ifdef TCP_SIGNATURE 1824 if (optp || (tp->t_flags & TF_SIGNATURE)) 1825 #else 1826 if (optp) 1827 #endif 1828 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1829 goto drop; 1830 1831 if (TCP_SACK_ENABLED(tp)) { 1832 tcp_del_sackholes(tp, th); 1833 } 1834 1835 if (TCP_ECN_ALLOWED(tp)) { 1836 if (tiflags & TH_CWR) { 1837 tp->t_flags &= ~TF_ECN_SND_ECE; 1838 } 1839 switch (iptos & IPTOS_ECN_MASK) { 1840 case IPTOS_ECN_CE: 1841 tp->t_flags |= TF_ECN_SND_ECE; 1842 TCP_STATINC(TCP_STAT_ECN_CE); 1843 break; 1844 case IPTOS_ECN_ECT0: 1845 TCP_STATINC(TCP_STAT_ECN_ECT); 1846 break; 1847 case IPTOS_ECN_ECT1: 1848 /* XXX: ignore for now -- rpaulo */ 1849 break; 1850 } 1851 /* 1852 * Congestion experienced. 1853 * Ignore if we are already trying to recover. 1854 */ 1855 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1856 tp->t_congctl->cong_exp(tp); 1857 } 1858 1859 if (opti.ts_present && opti.ts_ecr) { 1860 /* 1861 * Calculate the RTT from the returned time stamp and the 1862 * connection's time base. If the time stamp is later than 1863 * the current time, or is extremely old, fall back to non-1323 1864 * RTT calculation. Since ts_rtt is unsigned, we can test both 1865 * at the same time. 1866 * 1867 * Note that ts_rtt is in units of slow ticks (500 1868 * ms). Since most earthbound RTTs are < 500 ms, 1869 * observed values will have large quantization noise. 1870 * Our smoothed RTT is then the fraction of observed 1871 * samples that are 1 tick instead of 0 (times 500 1872 * ms). 1873 * 1874 * ts_rtt is increased by 1 to denote a valid sample, 1875 * with 0 indicating an invalid measurement. This 1876 * extra 1 must be removed when ts_rtt is used, or 1877 * else an an erroneous extra 500 ms will result. 1878 */ 1879 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1880 if (ts_rtt > TCP_PAWS_IDLE) 1881 ts_rtt = 0; 1882 } else { 1883 ts_rtt = 0; 1884 } 1885 1886 /* 1887 * Header prediction: check for the two common cases 1888 * of a uni-directional data xfer. If the packet has 1889 * no control flags, is in-sequence, the window didn't 1890 * change and we're not retransmitting, it's a 1891 * candidate. If the length is zero and the ack moved 1892 * forward, we're the sender side of the xfer. Just 1893 * free the data acked & wake any higher level process 1894 * that was blocked waiting for space. If the length 1895 * is non-zero and the ack didn't move, we're the 1896 * receiver side. If we're getting packets in-order 1897 * (the reassembly queue is empty), add the data to 1898 * the socket buffer and note that we need a delayed ack. 1899 */ 1900 if (tp->t_state == TCPS_ESTABLISHED && 1901 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1902 == TH_ACK && 1903 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1904 th->th_seq == tp->rcv_nxt && 1905 tiwin && tiwin == tp->snd_wnd && 1906 tp->snd_nxt == tp->snd_max) { 1907 1908 /* 1909 * If last ACK falls within this segment's sequence numbers, 1910 * record the timestamp. 1911 * NOTE that the test is modified according to the latest 1912 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 1913 * 1914 * note that we already know 1915 * TSTMP_GEQ(opti.ts_val, tp->ts_recent) 1916 */ 1917 if (opti.ts_present && 1918 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1919 tp->ts_recent_age = tcp_now; 1920 tp->ts_recent = opti.ts_val; 1921 } 1922 1923 if (tlen == 0) { 1924 /* Ack prediction. */ 1925 if (SEQ_GT(th->th_ack, tp->snd_una) && 1926 SEQ_LEQ(th->th_ack, tp->snd_max) && 1927 tp->snd_cwnd >= tp->snd_wnd && 1928 tp->t_partialacks < 0) { 1929 /* 1930 * this is a pure ack for outstanding data. 1931 */ 1932 if (ts_rtt) 1933 tcp_xmit_timer(tp, ts_rtt - 1); 1934 else if (tp->t_rtttime && 1935 SEQ_GT(th->th_ack, tp->t_rtseq)) 1936 tcp_xmit_timer(tp, 1937 tcp_now - tp->t_rtttime); 1938 acked = th->th_ack - tp->snd_una; 1939 tcps = TCP_STAT_GETREF(); 1940 tcps[TCP_STAT_PREDACK]++; 1941 tcps[TCP_STAT_RCVACKPACK]++; 1942 tcps[TCP_STAT_RCVACKBYTE] += acked; 1943 TCP_STAT_PUTREF(); 1944 nd6_hint(tp); 1945 1946 if (acked > (tp->t_lastoff - tp->t_inoff)) 1947 tp->t_lastm = NULL; 1948 sbdrop(&so->so_snd, acked); 1949 tp->t_lastoff -= acked; 1950 1951 icmp_check(tp, th, acked); 1952 1953 tp->snd_una = th->th_ack; 1954 tp->snd_fack = tp->snd_una; 1955 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1956 tp->snd_high = tp->snd_una; 1957 m_freem(m); 1958 1959 /* 1960 * If all outstanding data are acked, stop 1961 * retransmit timer, otherwise restart timer 1962 * using current (possibly backed-off) value. 1963 * If process is waiting for space, 1964 * wakeup/selnotify/signal. If data 1965 * are ready to send, let tcp_output 1966 * decide between more output or persist. 1967 */ 1968 if (tp->snd_una == tp->snd_max) 1969 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1970 else if (TCP_TIMER_ISARMED(tp, 1971 TCPT_PERSIST) == 0) 1972 TCP_TIMER_ARM(tp, TCPT_REXMT, 1973 tp->t_rxtcur); 1974 1975 sowwakeup(so); 1976 if (so->so_snd.sb_cc) { 1977 KERNEL_LOCK(1, NULL); 1978 (void) tcp_output(tp); 1979 KERNEL_UNLOCK_ONE(NULL); 1980 } 1981 if (tcp_saveti) 1982 m_freem(tcp_saveti); 1983 return; 1984 } 1985 } else if (th->th_ack == tp->snd_una && 1986 TAILQ_FIRST(&tp->segq) == NULL && 1987 tlen <= sbspace(&so->so_rcv)) { 1988 int newsize = 0; /* automatic sockbuf scaling */ 1989 1990 /* 1991 * this is a pure, in-sequence data packet 1992 * with nothing on the reassembly queue and 1993 * we have enough buffer space to take it. 1994 */ 1995 tp->rcv_nxt += tlen; 1996 tcps = TCP_STAT_GETREF(); 1997 tcps[TCP_STAT_PREDDAT]++; 1998 tcps[TCP_STAT_RCVPACK]++; 1999 tcps[TCP_STAT_RCVBYTE] += tlen; 2000 TCP_STAT_PUTREF(); 2001 nd6_hint(tp); 2002 2003 /* 2004 * Automatic sizing enables the performance of large buffers 2005 * and most of the efficiency of small ones by only allocating 2006 * space when it is needed. 2007 * 2008 * On the receive side the socket buffer memory is only rarely 2009 * used to any significant extent. This allows us to be much 2010 * more aggressive in scaling the receive socket buffer. For 2011 * the case that the buffer space is actually used to a large 2012 * extent and we run out of kernel memory we can simply drop 2013 * the new segments; TCP on the sender will just retransmit it 2014 * later. Setting the buffer size too big may only consume too 2015 * much kernel memory if the application doesn't read() from 2016 * the socket or packet loss or reordering makes use of the 2017 * reassembly queue. 2018 * 2019 * The criteria to step up the receive buffer one notch are: 2020 * 1. the number of bytes received during the time it takes 2021 * one timestamp to be reflected back to us (the RTT); 2022 * 2. received bytes per RTT is within seven eighth of the 2023 * current socket buffer size; 2024 * 3. receive buffer size has not hit maximal automatic size; 2025 * 2026 * This algorithm does one step per RTT at most and only if 2027 * we receive a bulk stream w/o packet losses or reorderings. 2028 * Shrinking the buffer during idle times is not necessary as 2029 * it doesn't consume any memory when idle. 2030 * 2031 * TODO: Only step up if the application is actually serving 2032 * the buffer to better manage the socket buffer resources. 2033 */ 2034 if (tcp_do_autorcvbuf && 2035 opti.ts_ecr && 2036 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 2037 if (opti.ts_ecr > tp->rfbuf_ts && 2038 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 2039 if (tp->rfbuf_cnt > 2040 (so->so_rcv.sb_hiwat / 8 * 7) && 2041 so->so_rcv.sb_hiwat < 2042 tcp_autorcvbuf_max) { 2043 newsize = 2044 min(so->so_rcv.sb_hiwat + 2045 tcp_autorcvbuf_inc, 2046 tcp_autorcvbuf_max); 2047 } 2048 /* Start over with next RTT. */ 2049 tp->rfbuf_ts = 0; 2050 tp->rfbuf_cnt = 0; 2051 } else 2052 tp->rfbuf_cnt += tlen; /* add up */ 2053 } 2054 2055 /* 2056 * Drop TCP, IP headers and TCP options then add data 2057 * to socket buffer. 2058 */ 2059 if (so->so_state & SS_CANTRCVMORE) 2060 m_freem(m); 2061 else { 2062 /* 2063 * Set new socket buffer size. 2064 * Give up when limit is reached. 2065 */ 2066 if (newsize) 2067 if (!sbreserve(&so->so_rcv, 2068 newsize, so)) 2069 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 2070 m_adj(m, toff + off); 2071 sbappendstream(&so->so_rcv, m); 2072 } 2073 sorwakeup(so); 2074 tcp_setup_ack(tp, th); 2075 if (tp->t_flags & TF_ACKNOW) { 2076 KERNEL_LOCK(1, NULL); 2077 (void) tcp_output(tp); 2078 KERNEL_UNLOCK_ONE(NULL); 2079 } 2080 if (tcp_saveti) 2081 m_freem(tcp_saveti); 2082 return; 2083 } 2084 } 2085 2086 /* 2087 * Compute mbuf offset to TCP data segment. 2088 */ 2089 hdroptlen = toff + off; 2090 2091 /* 2092 * Calculate amount of space in receive window, 2093 * and then do TCP input processing. 2094 * Receive window is amount of space in rcv queue, 2095 * but not less than advertised window. 2096 */ 2097 { int win; 2098 2099 win = sbspace(&so->so_rcv); 2100 if (win < 0) 2101 win = 0; 2102 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 2103 } 2104 2105 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 2106 tp->rfbuf_ts = 0; 2107 tp->rfbuf_cnt = 0; 2108 2109 switch (tp->t_state) { 2110 /* 2111 * If the state is SYN_SENT: 2112 * if seg contains an ACK, but not for our SYN, drop the input. 2113 * if seg contains a RST, then drop the connection. 2114 * if seg does not contain SYN, then drop it. 2115 * Otherwise this is an acceptable SYN segment 2116 * initialize tp->rcv_nxt and tp->irs 2117 * if seg contains ack then advance tp->snd_una 2118 * if seg contains a ECE and ECN support is enabled, the stream 2119 * is ECN capable. 2120 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2121 * arrange for segment to be acked (eventually) 2122 * continue processing rest of data/controls, beginning with URG 2123 */ 2124 case TCPS_SYN_SENT: 2125 if ((tiflags & TH_ACK) && 2126 (SEQ_LEQ(th->th_ack, tp->iss) || 2127 SEQ_GT(th->th_ack, tp->snd_max))) 2128 goto dropwithreset; 2129 if (tiflags & TH_RST) { 2130 if (tiflags & TH_ACK) 2131 tp = tcp_drop(tp, ECONNREFUSED); 2132 goto drop; 2133 } 2134 if ((tiflags & TH_SYN) == 0) 2135 goto drop; 2136 if (tiflags & TH_ACK) { 2137 tp->snd_una = th->th_ack; 2138 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2139 tp->snd_nxt = tp->snd_una; 2140 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2141 tp->snd_high = tp->snd_una; 2142 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2143 2144 if ((tiflags & TH_ECE) && tcp_do_ecn) { 2145 tp->t_flags |= TF_ECN_PERMIT; 2146 TCP_STATINC(TCP_STAT_ECN_SHS); 2147 } 2148 2149 } 2150 tp->irs = th->th_seq; 2151 tcp_rcvseqinit(tp); 2152 tp->t_flags |= TF_ACKNOW; 2153 tcp_mss_from_peer(tp, opti.maxseg); 2154 2155 /* 2156 * Initialize the initial congestion window. If we 2157 * had to retransmit the SYN, we must initialize cwnd 2158 * to 1 segment (i.e. the Loss Window). 2159 */ 2160 if (tp->t_flags & TF_SYN_REXMT) 2161 tp->snd_cwnd = tp->t_peermss; 2162 else { 2163 int ss = tcp_init_win; 2164 #ifdef INET 2165 if (inp != NULL && in_localaddr(inp->inp_faddr)) 2166 ss = tcp_init_win_local; 2167 #endif 2168 #ifdef INET6 2169 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 2170 ss = tcp_init_win_local; 2171 #endif 2172 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 2173 } 2174 2175 tcp_rmx_rtt(tp); 2176 if (tiflags & TH_ACK) { 2177 TCP_STATINC(TCP_STAT_CONNECTS); 2178 /* 2179 * move tcp_established before soisconnected 2180 * because upcall handler can drive tcp_output 2181 * functionality. 2182 * XXX we might call soisconnected at the end of 2183 * all processing 2184 */ 2185 tcp_established(tp); 2186 soisconnected(so); 2187 /* Do window scaling on this connection? */ 2188 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2189 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2190 tp->snd_scale = tp->requested_s_scale; 2191 tp->rcv_scale = tp->request_r_scale; 2192 } 2193 TCP_REASS_LOCK(tp); 2194 (void) tcp_reass(tp, NULL, NULL, &tlen); 2195 /* 2196 * if we didn't have to retransmit the SYN, 2197 * use its rtt as our initial srtt & rtt var. 2198 */ 2199 if (tp->t_rtttime) 2200 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2201 } else 2202 tp->t_state = TCPS_SYN_RECEIVED; 2203 2204 /* 2205 * Advance th->th_seq to correspond to first data byte. 2206 * If data, trim to stay within window, 2207 * dropping FIN if necessary. 2208 */ 2209 th->th_seq++; 2210 if (tlen > tp->rcv_wnd) { 2211 todrop = tlen - tp->rcv_wnd; 2212 m_adj(m, -todrop); 2213 tlen = tp->rcv_wnd; 2214 tiflags &= ~TH_FIN; 2215 tcps = TCP_STAT_GETREF(); 2216 tcps[TCP_STAT_RCVPACKAFTERWIN]++; 2217 tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop; 2218 TCP_STAT_PUTREF(); 2219 } 2220 tp->snd_wl1 = th->th_seq - 1; 2221 tp->rcv_up = th->th_seq; 2222 goto step6; 2223 2224 /* 2225 * If the state is SYN_RECEIVED: 2226 * If seg contains an ACK, but not for our SYN, drop the input 2227 * and generate an RST. See page 36, rfc793 2228 */ 2229 case TCPS_SYN_RECEIVED: 2230 if ((tiflags & TH_ACK) && 2231 (SEQ_LEQ(th->th_ack, tp->iss) || 2232 SEQ_GT(th->th_ack, tp->snd_max))) 2233 goto dropwithreset; 2234 break; 2235 } 2236 2237 /* 2238 * States other than LISTEN or SYN_SENT. 2239 * First check timestamp, if present. 2240 * Then check that at least some bytes of segment are within 2241 * receive window. If segment begins before rcv_nxt, 2242 * drop leading data (and SYN); if nothing left, just ack. 2243 * 2244 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2245 * and it's less than ts_recent, drop it. 2246 */ 2247 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2248 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2249 2250 /* Check to see if ts_recent is over 24 days old. */ 2251 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2252 /* 2253 * Invalidate ts_recent. If this segment updates 2254 * ts_recent, the age will be reset later and ts_recent 2255 * will get a valid value. If it does not, setting 2256 * ts_recent to zero will at least satisfy the 2257 * requirement that zero be placed in the timestamp 2258 * echo reply when ts_recent isn't valid. The 2259 * age isn't reset until we get a valid ts_recent 2260 * because we don't want out-of-order segments to be 2261 * dropped when ts_recent is old. 2262 */ 2263 tp->ts_recent = 0; 2264 } else { 2265 tcps = TCP_STAT_GETREF(); 2266 tcps[TCP_STAT_RCVDUPPACK]++; 2267 tcps[TCP_STAT_RCVDUPBYTE] += tlen; 2268 tcps[TCP_STAT_PAWSDROP]++; 2269 TCP_STAT_PUTREF(); 2270 tcp_new_dsack(tp, th->th_seq, tlen); 2271 goto dropafterack; 2272 } 2273 } 2274 2275 todrop = tp->rcv_nxt - th->th_seq; 2276 dupseg = false; 2277 if (todrop > 0) { 2278 if (tiflags & TH_SYN) { 2279 tiflags &= ~TH_SYN; 2280 th->th_seq++; 2281 if (th->th_urp > 1) 2282 th->th_urp--; 2283 else { 2284 tiflags &= ~TH_URG; 2285 th->th_urp = 0; 2286 } 2287 todrop--; 2288 } 2289 if (todrop > tlen || 2290 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2291 /* 2292 * Any valid FIN or RST must be to the left of the 2293 * window. At this point the FIN or RST must be a 2294 * duplicate or out of sequence; drop it. 2295 */ 2296 if (tiflags & TH_RST) 2297 goto drop; 2298 tiflags &= ~(TH_FIN|TH_RST); 2299 /* 2300 * Send an ACK to resynchronize and drop any data. 2301 * But keep on processing for RST or ACK. 2302 */ 2303 tp->t_flags |= TF_ACKNOW; 2304 todrop = tlen; 2305 dupseg = true; 2306 tcps = TCP_STAT_GETREF(); 2307 tcps[TCP_STAT_RCVDUPPACK]++; 2308 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 2309 TCP_STAT_PUTREF(); 2310 } else if ((tiflags & TH_RST) && 2311 th->th_seq != tp->rcv_nxt) { 2312 /* 2313 * Test for reset before adjusting the sequence 2314 * number for overlapping data. 2315 */ 2316 goto dropafterack_ratelim; 2317 } else { 2318 tcps = TCP_STAT_GETREF(); 2319 tcps[TCP_STAT_RCVPARTDUPPACK]++; 2320 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 2321 TCP_STAT_PUTREF(); 2322 } 2323 tcp_new_dsack(tp, th->th_seq, todrop); 2324 hdroptlen += todrop; /*drop from head afterwards*/ 2325 th->th_seq += todrop; 2326 tlen -= todrop; 2327 if (th->th_urp > todrop) 2328 th->th_urp -= todrop; 2329 else { 2330 tiflags &= ~TH_URG; 2331 th->th_urp = 0; 2332 } 2333 } 2334 2335 /* 2336 * If new data are received on a connection after the 2337 * user processes are gone, then RST the other end. 2338 */ 2339 if ((so->so_state & SS_NOFDREF) && 2340 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2341 tp = tcp_close(tp); 2342 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 2343 goto dropwithreset; 2344 } 2345 2346 /* 2347 * If segment ends after window, drop trailing data 2348 * (and PUSH and FIN); if nothing left, just ACK. 2349 */ 2350 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 2351 if (todrop > 0) { 2352 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 2353 if (todrop >= tlen) { 2354 /* 2355 * The segment actually starts after the window. 2356 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2357 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2358 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2359 */ 2360 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 2361 /* 2362 * If a new connection request is received 2363 * while in TIME_WAIT, drop the old connection 2364 * and start over if the sequence numbers 2365 * are above the previous ones. 2366 * 2367 * NOTE: We will checksum the packet again, and 2368 * so we need to put the header fields back into 2369 * network order! 2370 * XXX This kind of sucks, but we don't expect 2371 * XXX this to happen very often, so maybe it 2372 * XXX doesn't matter so much. 2373 */ 2374 if (tiflags & TH_SYN && 2375 tp->t_state == TCPS_TIME_WAIT && 2376 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2377 tp = tcp_close(tp); 2378 tcp_fields_to_net(th); 2379 goto findpcb; 2380 } 2381 /* 2382 * If window is closed can only take segments at 2383 * window edge, and have to drop data and PUSH from 2384 * incoming segments. Continue processing, but 2385 * remember to ack. Otherwise, drop segment 2386 * and (if not RST) ack. 2387 */ 2388 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2389 tp->t_flags |= TF_ACKNOW; 2390 TCP_STATINC(TCP_STAT_RCVWINPROBE); 2391 } else 2392 goto dropafterack; 2393 } else 2394 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 2395 m_adj(m, -todrop); 2396 tlen -= todrop; 2397 tiflags &= ~(TH_PUSH|TH_FIN); 2398 } 2399 2400 /* 2401 * If last ACK falls within this segment's sequence numbers, 2402 * record the timestamp. 2403 * NOTE: 2404 * 1) That the test incorporates suggestions from the latest 2405 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 2406 * 2) That updating only on newer timestamps interferes with 2407 * our earlier PAWS tests, so this check should be solely 2408 * predicated on the sequence space of this segment. 2409 * 3) That we modify the segment boundary check to be 2410 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2411 * instead of RFC1323's 2412 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2413 * This modified check allows us to overcome RFC1323's 2414 * limitations as described in Stevens TCP/IP Illustrated 2415 * Vol. 2 p.869. In such cases, we can still calculate the 2416 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2417 */ 2418 if (opti.ts_present && 2419 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2420 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2421 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2422 tp->ts_recent_age = tcp_now; 2423 tp->ts_recent = opti.ts_val; 2424 } 2425 2426 /* 2427 * If the RST bit is set examine the state: 2428 * SYN_RECEIVED STATE: 2429 * If passive open, return to LISTEN state. 2430 * If active open, inform user that connection was refused. 2431 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 2432 * Inform user that connection was reset, and close tcb. 2433 * CLOSING, LAST_ACK, TIME_WAIT STATES 2434 * Close the tcb. 2435 */ 2436 if (tiflags & TH_RST) { 2437 if (th->th_seq != tp->rcv_nxt) 2438 goto dropafterack_ratelim; 2439 2440 switch (tp->t_state) { 2441 case TCPS_SYN_RECEIVED: 2442 so->so_error = ECONNREFUSED; 2443 goto close; 2444 2445 case TCPS_ESTABLISHED: 2446 case TCPS_FIN_WAIT_1: 2447 case TCPS_FIN_WAIT_2: 2448 case TCPS_CLOSE_WAIT: 2449 so->so_error = ECONNRESET; 2450 close: 2451 tp->t_state = TCPS_CLOSED; 2452 TCP_STATINC(TCP_STAT_DROPS); 2453 tp = tcp_close(tp); 2454 goto drop; 2455 2456 case TCPS_CLOSING: 2457 case TCPS_LAST_ACK: 2458 case TCPS_TIME_WAIT: 2459 tp = tcp_close(tp); 2460 goto drop; 2461 } 2462 } 2463 2464 /* 2465 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2466 * we must be in a synchronized state. RFC791 states (under RST 2467 * generation) that any unacceptable segment (an out-of-order SYN 2468 * qualifies) received in a synchronized state must elicit only an 2469 * empty acknowledgment segment ... and the connection remains in 2470 * the same state. 2471 */ 2472 if (tiflags & TH_SYN) { 2473 if (tp->rcv_nxt == th->th_seq) { 2474 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2475 TH_ACK); 2476 if (tcp_saveti) 2477 m_freem(tcp_saveti); 2478 return; 2479 } 2480 2481 goto dropafterack_ratelim; 2482 } 2483 2484 /* 2485 * If the ACK bit is off we drop the segment and return. 2486 */ 2487 if ((tiflags & TH_ACK) == 0) { 2488 if (tp->t_flags & TF_ACKNOW) 2489 goto dropafterack; 2490 else 2491 goto drop; 2492 } 2493 2494 /* 2495 * Ack processing. 2496 */ 2497 switch (tp->t_state) { 2498 2499 /* 2500 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2501 * ESTABLISHED state and continue processing, otherwise 2502 * send an RST. 2503 */ 2504 case TCPS_SYN_RECEIVED: 2505 if (SEQ_GT(tp->snd_una, th->th_ack) || 2506 SEQ_GT(th->th_ack, tp->snd_max)) 2507 goto dropwithreset; 2508 TCP_STATINC(TCP_STAT_CONNECTS); 2509 soisconnected(so); 2510 tcp_established(tp); 2511 /* Do window scaling? */ 2512 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2513 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2514 tp->snd_scale = tp->requested_s_scale; 2515 tp->rcv_scale = tp->request_r_scale; 2516 } 2517 TCP_REASS_LOCK(tp); 2518 (void) tcp_reass(tp, NULL, NULL, &tlen); 2519 tp->snd_wl1 = th->th_seq - 1; 2520 /* fall into ... */ 2521 2522 /* 2523 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2524 * ACKs. If the ack is in the range 2525 * tp->snd_una < th->th_ack <= tp->snd_max 2526 * then advance tp->snd_una to th->th_ack and drop 2527 * data from the retransmission queue. If this ACK reflects 2528 * more up to date window information we update our window information. 2529 */ 2530 case TCPS_ESTABLISHED: 2531 case TCPS_FIN_WAIT_1: 2532 case TCPS_FIN_WAIT_2: 2533 case TCPS_CLOSE_WAIT: 2534 case TCPS_CLOSING: 2535 case TCPS_LAST_ACK: 2536 case TCPS_TIME_WAIT: 2537 2538 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2539 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2540 TCP_STATINC(TCP_STAT_RCVDUPACK); 2541 /* 2542 * If we have outstanding data (other than 2543 * a window probe), this is a completely 2544 * duplicate ack (ie, window info didn't 2545 * change), the ack is the biggest we've 2546 * seen and we've seen exactly our rexmt 2547 * threshhold of them, assume a packet 2548 * has been dropped and retransmit it. 2549 * Kludge snd_nxt & the congestion 2550 * window so we send only this one 2551 * packet. 2552 */ 2553 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2554 th->th_ack != tp->snd_una) 2555 tp->t_dupacks = 0; 2556 else if (tp->t_partialacks < 0 && 2557 (++tp->t_dupacks == tcprexmtthresh || 2558 TCP_FACK_FASTRECOV(tp))) { 2559 /* 2560 * Do the fast retransmit, and adjust 2561 * congestion control paramenters. 2562 */ 2563 if (tp->t_congctl->fast_retransmit(tp, th)) { 2564 /* False fast retransmit */ 2565 break; 2566 } else 2567 goto drop; 2568 } else if (tp->t_dupacks > tcprexmtthresh) { 2569 tp->snd_cwnd += tp->t_segsz; 2570 KERNEL_LOCK(1, NULL); 2571 (void) tcp_output(tp); 2572 KERNEL_UNLOCK_ONE(NULL); 2573 goto drop; 2574 } 2575 } else { 2576 /* 2577 * If the ack appears to be very old, only 2578 * allow data that is in-sequence. This 2579 * makes it somewhat more difficult to insert 2580 * forged data by guessing sequence numbers. 2581 * Sent an ack to try to update the send 2582 * sequence number on the other side. 2583 */ 2584 if (tlen && th->th_seq != tp->rcv_nxt && 2585 SEQ_LT(th->th_ack, 2586 tp->snd_una - tp->max_sndwnd)) 2587 goto dropafterack; 2588 } 2589 break; 2590 } 2591 /* 2592 * If the congestion window was inflated to account 2593 * for the other side's cached packets, retract it. 2594 */ 2595 tp->t_congctl->fast_retransmit_newack(tp, th); 2596 2597 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2598 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2599 goto dropafterack; 2600 } 2601 acked = th->th_ack - tp->snd_una; 2602 tcps = TCP_STAT_GETREF(); 2603 tcps[TCP_STAT_RCVACKPACK]++; 2604 tcps[TCP_STAT_RCVACKBYTE] += acked; 2605 TCP_STAT_PUTREF(); 2606 2607 /* 2608 * If we have a timestamp reply, update smoothed 2609 * round trip time. If no timestamp is present but 2610 * transmit timer is running and timed sequence 2611 * number was acked, update smoothed round trip time. 2612 * Since we now have an rtt measurement, cancel the 2613 * timer backoff (cf., Phil Karn's retransmit alg.). 2614 * Recompute the initial retransmit timer. 2615 */ 2616 if (ts_rtt) 2617 tcp_xmit_timer(tp, ts_rtt - 1); 2618 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2619 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2620 2621 /* 2622 * If all outstanding data is acked, stop retransmit 2623 * timer and remember to restart (more output or persist). 2624 * If there is more data to be acked, restart retransmit 2625 * timer, using current (possibly backed-off) value. 2626 */ 2627 if (th->th_ack == tp->snd_max) { 2628 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2629 needoutput = 1; 2630 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2631 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2632 2633 /* 2634 * New data has been acked, adjust the congestion window. 2635 */ 2636 tp->t_congctl->newack(tp, th); 2637 2638 nd6_hint(tp); 2639 if (acked > so->so_snd.sb_cc) { 2640 tp->snd_wnd -= so->so_snd.sb_cc; 2641 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2642 ourfinisacked = 1; 2643 } else { 2644 if (acked > (tp->t_lastoff - tp->t_inoff)) 2645 tp->t_lastm = NULL; 2646 sbdrop(&so->so_snd, acked); 2647 tp->t_lastoff -= acked; 2648 if (tp->snd_wnd > acked) 2649 tp->snd_wnd -= acked; 2650 else 2651 tp->snd_wnd = 0; 2652 ourfinisacked = 0; 2653 } 2654 sowwakeup(so); 2655 2656 icmp_check(tp, th, acked); 2657 2658 tp->snd_una = th->th_ack; 2659 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2660 tp->snd_fack = tp->snd_una; 2661 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2662 tp->snd_nxt = tp->snd_una; 2663 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2664 tp->snd_high = tp->snd_una; 2665 2666 switch (tp->t_state) { 2667 2668 /* 2669 * In FIN_WAIT_1 STATE in addition to the processing 2670 * for the ESTABLISHED state if our FIN is now acknowledged 2671 * then enter FIN_WAIT_2. 2672 */ 2673 case TCPS_FIN_WAIT_1: 2674 if (ourfinisacked) { 2675 /* 2676 * If we can't receive any more 2677 * data, then closing user can proceed. 2678 * Starting the timer is contrary to the 2679 * specification, but if we don't get a FIN 2680 * we'll hang forever. 2681 */ 2682 if (so->so_state & SS_CANTRCVMORE) { 2683 soisdisconnected(so); 2684 if (tp->t_maxidle > 0) 2685 TCP_TIMER_ARM(tp, TCPT_2MSL, 2686 tp->t_maxidle); 2687 } 2688 tp->t_state = TCPS_FIN_WAIT_2; 2689 } 2690 break; 2691 2692 /* 2693 * In CLOSING STATE in addition to the processing for 2694 * the ESTABLISHED state if the ACK acknowledges our FIN 2695 * then enter the TIME-WAIT state, otherwise ignore 2696 * the segment. 2697 */ 2698 case TCPS_CLOSING: 2699 if (ourfinisacked) { 2700 tp->t_state = TCPS_TIME_WAIT; 2701 tcp_canceltimers(tp); 2702 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2703 soisdisconnected(so); 2704 } 2705 break; 2706 2707 /* 2708 * In LAST_ACK, we may still be waiting for data to drain 2709 * and/or to be acked, as well as for the ack of our FIN. 2710 * If our FIN is now acknowledged, delete the TCB, 2711 * enter the closed state and return. 2712 */ 2713 case TCPS_LAST_ACK: 2714 if (ourfinisacked) { 2715 tp = tcp_close(tp); 2716 goto drop; 2717 } 2718 break; 2719 2720 /* 2721 * In TIME_WAIT state the only thing that should arrive 2722 * is a retransmission of the remote FIN. Acknowledge 2723 * it and restart the finack timer. 2724 */ 2725 case TCPS_TIME_WAIT: 2726 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2727 goto dropafterack; 2728 } 2729 } 2730 2731 step6: 2732 /* 2733 * Update window information. 2734 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2735 */ 2736 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2737 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2738 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2739 /* keep track of pure window updates */ 2740 if (tlen == 0 && 2741 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2742 TCP_STATINC(TCP_STAT_RCVWINUPD); 2743 tp->snd_wnd = tiwin; 2744 tp->snd_wl1 = th->th_seq; 2745 tp->snd_wl2 = th->th_ack; 2746 if (tp->snd_wnd > tp->max_sndwnd) 2747 tp->max_sndwnd = tp->snd_wnd; 2748 needoutput = 1; 2749 } 2750 2751 /* 2752 * Process segments with URG. 2753 */ 2754 if ((tiflags & TH_URG) && th->th_urp && 2755 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2756 /* 2757 * This is a kludge, but if we receive and accept 2758 * random urgent pointers, we'll crash in 2759 * soreceive. It's hard to imagine someone 2760 * actually wanting to send this much urgent data. 2761 */ 2762 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2763 th->th_urp = 0; /* XXX */ 2764 tiflags &= ~TH_URG; /* XXX */ 2765 goto dodata; /* XXX */ 2766 } 2767 /* 2768 * If this segment advances the known urgent pointer, 2769 * then mark the data stream. This should not happen 2770 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2771 * a FIN has been received from the remote side. 2772 * In these states we ignore the URG. 2773 * 2774 * According to RFC961 (Assigned Protocols), 2775 * the urgent pointer points to the last octet 2776 * of urgent data. We continue, however, 2777 * to consider it to indicate the first octet 2778 * of data past the urgent section as the original 2779 * spec states (in one of two places). 2780 */ 2781 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2782 tp->rcv_up = th->th_seq + th->th_urp; 2783 so->so_oobmark = so->so_rcv.sb_cc + 2784 (tp->rcv_up - tp->rcv_nxt) - 1; 2785 if (so->so_oobmark == 0) 2786 so->so_state |= SS_RCVATMARK; 2787 sohasoutofband(so); 2788 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2789 } 2790 /* 2791 * Remove out of band data so doesn't get presented to user. 2792 * This can happen independent of advancing the URG pointer, 2793 * but if two URG's are pending at once, some out-of-band 2794 * data may creep in... ick. 2795 */ 2796 if (th->th_urp <= (u_int16_t) tlen 2797 #ifdef SO_OOBINLINE 2798 && (so->so_options & SO_OOBINLINE) == 0 2799 #endif 2800 ) 2801 tcp_pulloutofband(so, th, m, hdroptlen); 2802 } else 2803 /* 2804 * If no out of band data is expected, 2805 * pull receive urgent pointer along 2806 * with the receive window. 2807 */ 2808 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2809 tp->rcv_up = tp->rcv_nxt; 2810 dodata: /* XXX */ 2811 2812 /* 2813 * Process the segment text, merging it into the TCP sequencing queue, 2814 * and arranging for acknowledgement of receipt if necessary. 2815 * This process logically involves adjusting tp->rcv_wnd as data 2816 * is presented to the user (this happens in tcp_usrreq.c, 2817 * tcp_rcvd()). If a FIN has already been received on this 2818 * connection then we just ignore the text. 2819 */ 2820 if ((tlen || (tiflags & TH_FIN)) && 2821 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2822 /* 2823 * Insert segment ti into reassembly queue of tcp with 2824 * control block tp. Return TH_FIN if reassembly now includes 2825 * a segment with FIN. The macro form does the common case 2826 * inline (segment is the next to be received on an 2827 * established connection, and the queue is empty), 2828 * avoiding linkage into and removal from the queue and 2829 * repetition of various conversions. 2830 * Set DELACK for segments received in order, but ack 2831 * immediately when segments are out of order 2832 * (so fast retransmit can work). 2833 */ 2834 /* NOTE: this was TCP_REASS() macro, but used only once */ 2835 TCP_REASS_LOCK(tp); 2836 if (th->th_seq == tp->rcv_nxt && 2837 TAILQ_FIRST(&tp->segq) == NULL && 2838 tp->t_state == TCPS_ESTABLISHED) { 2839 tcp_setup_ack(tp, th); 2840 tp->rcv_nxt += tlen; 2841 tiflags = th->th_flags & TH_FIN; 2842 tcps = TCP_STAT_GETREF(); 2843 tcps[TCP_STAT_RCVPACK]++; 2844 tcps[TCP_STAT_RCVBYTE] += tlen; 2845 TCP_STAT_PUTREF(); 2846 nd6_hint(tp); 2847 if (so->so_state & SS_CANTRCVMORE) 2848 m_freem(m); 2849 else { 2850 m_adj(m, hdroptlen); 2851 sbappendstream(&(so)->so_rcv, m); 2852 } 2853 TCP_REASS_UNLOCK(tp); 2854 sorwakeup(so); 2855 } else { 2856 m_adj(m, hdroptlen); 2857 tiflags = tcp_reass(tp, th, m, &tlen); 2858 tp->t_flags |= TF_ACKNOW; 2859 } 2860 2861 /* 2862 * Note the amount of data that peer has sent into 2863 * our window, in order to estimate the sender's 2864 * buffer size. 2865 */ 2866 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2867 } else { 2868 m_freem(m); 2869 m = NULL; 2870 tiflags &= ~TH_FIN; 2871 } 2872 2873 /* 2874 * If FIN is received ACK the FIN and let the user know 2875 * that the connection is closing. Ignore a FIN received before 2876 * the connection is fully established. 2877 */ 2878 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2879 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2880 socantrcvmore(so); 2881 tp->t_flags |= TF_ACKNOW; 2882 tp->rcv_nxt++; 2883 } 2884 switch (tp->t_state) { 2885 2886 /* 2887 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2888 */ 2889 case TCPS_ESTABLISHED: 2890 tp->t_state = TCPS_CLOSE_WAIT; 2891 break; 2892 2893 /* 2894 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2895 * enter the CLOSING state. 2896 */ 2897 case TCPS_FIN_WAIT_1: 2898 tp->t_state = TCPS_CLOSING; 2899 break; 2900 2901 /* 2902 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2903 * starting the time-wait timer, turning off the other 2904 * standard timers. 2905 */ 2906 case TCPS_FIN_WAIT_2: 2907 tp->t_state = TCPS_TIME_WAIT; 2908 tcp_canceltimers(tp); 2909 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2910 soisdisconnected(so); 2911 break; 2912 2913 /* 2914 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2915 */ 2916 case TCPS_TIME_WAIT: 2917 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2918 break; 2919 } 2920 } 2921 #ifdef TCP_DEBUG 2922 if (so->so_options & SO_DEBUG) 2923 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2924 #endif 2925 2926 /* 2927 * Return any desired output. 2928 */ 2929 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2930 KERNEL_LOCK(1, NULL); 2931 (void) tcp_output(tp); 2932 KERNEL_UNLOCK_ONE(NULL); 2933 } 2934 if (tcp_saveti) 2935 m_freem(tcp_saveti); 2936 2937 if (tp->t_state == TCPS_TIME_WAIT 2938 && (so->so_state & SS_NOFDREF) 2939 && (tp->t_inpcb || af != AF_INET) 2940 && (tp->t_in6pcb || af != AF_INET6) 2941 && ((af == AF_INET ? tcp4_vtw_enable : tcp6_vtw_enable) & 1) != 0 2942 && TAILQ_EMPTY(&tp->segq) 2943 && vtw_add(af, tp)) { 2944 ; 2945 } 2946 return; 2947 2948 badsyn: 2949 /* 2950 * Received a bad SYN. Increment counters and dropwithreset. 2951 */ 2952 TCP_STATINC(TCP_STAT_BADSYN); 2953 tp = NULL; 2954 goto dropwithreset; 2955 2956 dropafterack: 2957 /* 2958 * Generate an ACK dropping incoming segment if it occupies 2959 * sequence space, where the ACK reflects our state. 2960 */ 2961 if (tiflags & TH_RST) 2962 goto drop; 2963 goto dropafterack2; 2964 2965 dropafterack_ratelim: 2966 /* 2967 * We may want to rate-limit ACKs against SYN/RST attack. 2968 */ 2969 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2970 tcp_ackdrop_ppslim) == 0) { 2971 /* XXX stat */ 2972 goto drop; 2973 } 2974 /* ...fall into dropafterack2... */ 2975 2976 dropafterack2: 2977 m_freem(m); 2978 tp->t_flags |= TF_ACKNOW; 2979 KERNEL_LOCK(1, NULL); 2980 (void) tcp_output(tp); 2981 KERNEL_UNLOCK_ONE(NULL); 2982 if (tcp_saveti) 2983 m_freem(tcp_saveti); 2984 return; 2985 2986 dropwithreset_ratelim: 2987 /* 2988 * We may want to rate-limit RSTs in certain situations, 2989 * particularly if we are sending an RST in response to 2990 * an attempt to connect to or otherwise communicate with 2991 * a port for which we have no socket. 2992 */ 2993 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2994 tcp_rst_ppslim) == 0) { 2995 /* XXX stat */ 2996 goto drop; 2997 } 2998 /* ...fall into dropwithreset... */ 2999 3000 dropwithreset: 3001 /* 3002 * Generate a RST, dropping incoming segment. 3003 * Make ACK acceptable to originator of segment. 3004 */ 3005 if (tiflags & TH_RST) 3006 goto drop; 3007 if (tiflags & TH_ACK) { 3008 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3009 } else { 3010 if (tiflags & TH_SYN) 3011 tlen++; 3012 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 3013 TH_RST|TH_ACK); 3014 } 3015 if (tcp_saveti) 3016 m_freem(tcp_saveti); 3017 return; 3018 3019 badcsum: 3020 drop: 3021 /* 3022 * Drop space held by incoming segment and return. 3023 */ 3024 if (tp) { 3025 if (tp->t_inpcb) 3026 so = tp->t_inpcb->inp_socket; 3027 #ifdef INET6 3028 else if (tp->t_in6pcb) 3029 so = tp->t_in6pcb->in6p_socket; 3030 #endif 3031 else 3032 so = NULL; 3033 #ifdef TCP_DEBUG 3034 if (so && (so->so_options & SO_DEBUG) != 0) 3035 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 3036 #endif 3037 } 3038 if (tcp_saveti) 3039 m_freem(tcp_saveti); 3040 m_freem(m); 3041 return; 3042 } 3043 3044 #ifdef TCP_SIGNATURE 3045 int 3046 tcp_signature_apply(void *fstate, void *data, u_int len) 3047 { 3048 3049 MD5Update(fstate, (u_char *)data, len); 3050 return (0); 3051 } 3052 3053 struct secasvar * 3054 tcp_signature_getsav(struct mbuf *m) 3055 { 3056 struct ip *ip; 3057 struct ip6_hdr *ip6; 3058 3059 ip = mtod(m, struct ip *); 3060 switch (ip->ip_v) { 3061 case 4: 3062 ip = mtod(m, struct ip *); 3063 ip6 = NULL; 3064 break; 3065 case 6: 3066 ip = NULL; 3067 ip6 = mtod(m, struct ip6_hdr *); 3068 break; 3069 default: 3070 return (NULL); 3071 } 3072 3073 #ifdef IPSEC 3074 union sockaddr_union dst; 3075 3076 /* Extract the destination from the IP header in the mbuf. */ 3077 memset(&dst, 0, sizeof(union sockaddr_union)); 3078 if (ip != NULL) { 3079 dst.sa.sa_len = sizeof(struct sockaddr_in); 3080 dst.sa.sa_family = AF_INET; 3081 dst.sin.sin_addr = ip->ip_dst; 3082 } else { 3083 dst.sa.sa_len = sizeof(struct sockaddr_in6); 3084 dst.sa.sa_family = AF_INET6; 3085 dst.sin6.sin6_addr = ip6->ip6_dst; 3086 } 3087 3088 /* 3089 * Look up an SADB entry which matches the address of the peer. 3090 */ 3091 return KEY_LOOKUP_SA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI), 0, 0); 3092 #else 3093 return NULL; 3094 #endif 3095 } 3096 3097 int 3098 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 3099 struct secasvar *sav, char *sig) 3100 { 3101 MD5_CTX ctx; 3102 struct ip *ip; 3103 struct ipovly *ipovly; 3104 #ifdef INET6 3105 struct ip6_hdr *ip6; 3106 struct ip6_hdr_pseudo ip6pseudo; 3107 #endif /* INET6 */ 3108 struct ippseudo ippseudo; 3109 struct tcphdr th0; 3110 int l, tcphdrlen; 3111 3112 if (sav == NULL) 3113 return (-1); 3114 3115 tcphdrlen = th->th_off * 4; 3116 3117 switch (mtod(m, struct ip *)->ip_v) { 3118 case 4: 3119 MD5Init(&ctx); 3120 ip = mtod(m, struct ip *); 3121 memset(&ippseudo, 0, sizeof(ippseudo)); 3122 ipovly = (struct ipovly *)ip; 3123 ippseudo.ippseudo_src = ipovly->ih_src; 3124 ippseudo.ippseudo_dst = ipovly->ih_dst; 3125 ippseudo.ippseudo_pad = 0; 3126 ippseudo.ippseudo_p = IPPROTO_TCP; 3127 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 3128 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 3129 break; 3130 #if INET6 3131 case 6: 3132 MD5Init(&ctx); 3133 ip6 = mtod(m, struct ip6_hdr *); 3134 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 3135 ip6pseudo.ip6ph_src = ip6->ip6_src; 3136 in6_clearscope(&ip6pseudo.ip6ph_src); 3137 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 3138 in6_clearscope(&ip6pseudo.ip6ph_dst); 3139 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 3140 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 3141 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 3142 break; 3143 #endif /* INET6 */ 3144 default: 3145 return (-1); 3146 } 3147 3148 th0 = *th; 3149 th0.th_sum = 0; 3150 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 3151 3152 l = m->m_pkthdr.len - thoff - tcphdrlen; 3153 if (l > 0) 3154 m_apply(m, thoff + tcphdrlen, 3155 m->m_pkthdr.len - thoff - tcphdrlen, 3156 tcp_signature_apply, &ctx); 3157 3158 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 3159 MD5Final(sig, &ctx); 3160 3161 return (0); 3162 } 3163 #endif 3164 3165 /* 3166 * tcp_dooptions: parse and process tcp options. 3167 * 3168 * returns -1 if this segment should be dropped. (eg. wrong signature) 3169 * otherwise returns 0. 3170 */ 3171 3172 static int 3173 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, struct tcphdr *th, 3174 struct mbuf *m, int toff, struct tcp_opt_info *oi) 3175 { 3176 u_int16_t mss; 3177 int opt, optlen = 0; 3178 #ifdef TCP_SIGNATURE 3179 void *sigp = NULL; 3180 char sigbuf[TCP_SIGLEN]; 3181 struct secasvar *sav = NULL; 3182 #endif 3183 3184 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 3185 opt = cp[0]; 3186 if (opt == TCPOPT_EOL) 3187 break; 3188 if (opt == TCPOPT_NOP) 3189 optlen = 1; 3190 else { 3191 if (cnt < 2) 3192 break; 3193 optlen = cp[1]; 3194 if (optlen < 2 || optlen > cnt) 3195 break; 3196 } 3197 switch (opt) { 3198 3199 default: 3200 continue; 3201 3202 case TCPOPT_MAXSEG: 3203 if (optlen != TCPOLEN_MAXSEG) 3204 continue; 3205 if (!(th->th_flags & TH_SYN)) 3206 continue; 3207 if (TCPS_HAVERCVDSYN(tp->t_state)) 3208 continue; 3209 bcopy(cp + 2, &mss, sizeof(mss)); 3210 oi->maxseg = ntohs(mss); 3211 break; 3212 3213 case TCPOPT_WINDOW: 3214 if (optlen != TCPOLEN_WINDOW) 3215 continue; 3216 if (!(th->th_flags & TH_SYN)) 3217 continue; 3218 if (TCPS_HAVERCVDSYN(tp->t_state)) 3219 continue; 3220 tp->t_flags |= TF_RCVD_SCALE; 3221 tp->requested_s_scale = cp[2]; 3222 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3223 char buf[INET6_ADDRSTRLEN]; 3224 struct ip *ip = mtod(m, struct ip *); 3225 #ifdef INET6 3226 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 3227 #endif 3228 if (ip) 3229 in_print(buf, sizeof(buf), 3230 &ip->ip_src); 3231 #ifdef INET6 3232 else if (ip6) 3233 in6_print(buf, sizeof(buf), 3234 &ip6->ip6_src); 3235 #endif 3236 else 3237 strlcpy(buf, "(unknown)", sizeof(buf)); 3238 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3239 "assuming %d\n", 3240 tp->requested_s_scale, buf, 3241 TCP_MAX_WINSHIFT); 3242 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3243 } 3244 break; 3245 3246 case TCPOPT_TIMESTAMP: 3247 if (optlen != TCPOLEN_TIMESTAMP) 3248 continue; 3249 oi->ts_present = 1; 3250 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 3251 NTOHL(oi->ts_val); 3252 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 3253 NTOHL(oi->ts_ecr); 3254 3255 if (!(th->th_flags & TH_SYN)) 3256 continue; 3257 if (TCPS_HAVERCVDSYN(tp->t_state)) 3258 continue; 3259 /* 3260 * A timestamp received in a SYN makes 3261 * it ok to send timestamp requests and replies. 3262 */ 3263 tp->t_flags |= TF_RCVD_TSTMP; 3264 tp->ts_recent = oi->ts_val; 3265 tp->ts_recent_age = tcp_now; 3266 break; 3267 3268 case TCPOPT_SACK_PERMITTED: 3269 if (optlen != TCPOLEN_SACK_PERMITTED) 3270 continue; 3271 if (!(th->th_flags & TH_SYN)) 3272 continue; 3273 if (TCPS_HAVERCVDSYN(tp->t_state)) 3274 continue; 3275 if (tcp_do_sack) { 3276 tp->t_flags |= TF_SACK_PERMIT; 3277 tp->t_flags |= TF_WILL_SACK; 3278 } 3279 break; 3280 3281 case TCPOPT_SACK: 3282 tcp_sack_option(tp, th, cp, optlen); 3283 break; 3284 #ifdef TCP_SIGNATURE 3285 case TCPOPT_SIGNATURE: 3286 if (optlen != TCPOLEN_SIGNATURE) 3287 continue; 3288 if (sigp && memcmp(sigp, cp + 2, TCP_SIGLEN)) 3289 return (-1); 3290 3291 sigp = sigbuf; 3292 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3293 tp->t_flags |= TF_SIGNATURE; 3294 break; 3295 #endif 3296 } 3297 } 3298 3299 #ifndef TCP_SIGNATURE 3300 return 0; 3301 #else 3302 if (tp->t_flags & TF_SIGNATURE) { 3303 sav = tcp_signature_getsav(m); 3304 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3305 return (-1); 3306 } 3307 3308 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) 3309 goto out; 3310 3311 if (sigp) { 3312 char sig[TCP_SIGLEN]; 3313 3314 tcp_fields_to_net(th); 3315 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3316 tcp_fields_to_host(th); 3317 goto out; 3318 } 3319 tcp_fields_to_host(th); 3320 3321 if (memcmp(sig, sigp, TCP_SIGLEN)) { 3322 TCP_STATINC(TCP_STAT_BADSIG); 3323 goto out; 3324 } else 3325 TCP_STATINC(TCP_STAT_GOODSIG); 3326 3327 key_sa_recordxfer(sav, m); 3328 KEY_SA_UNREF(&sav); 3329 } 3330 return 0; 3331 out: 3332 if (sav != NULL) 3333 KEY_SA_UNREF(&sav); 3334 return -1; 3335 #endif 3336 } 3337 3338 /* 3339 * Pull out of band byte out of a segment so 3340 * it doesn't appear in the user's data queue. 3341 * It is still reflected in the segment length for 3342 * sequencing purposes. 3343 */ 3344 void 3345 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3346 struct mbuf *m, int off) 3347 { 3348 int cnt = off + th->th_urp - 1; 3349 3350 while (cnt >= 0) { 3351 if (m->m_len > cnt) { 3352 char *cp = mtod(m, char *) + cnt; 3353 struct tcpcb *tp = sototcpcb(so); 3354 3355 tp->t_iobc = *cp; 3356 tp->t_oobflags |= TCPOOB_HAVEDATA; 3357 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3358 m->m_len--; 3359 return; 3360 } 3361 cnt -= m->m_len; 3362 m = m->m_next; 3363 if (m == 0) 3364 break; 3365 } 3366 panic("tcp_pulloutofband"); 3367 } 3368 3369 /* 3370 * Collect new round-trip time estimate 3371 * and update averages and current timeout. 3372 * 3373 * rtt is in units of slow ticks (typically 500 ms) -- essentially the 3374 * difference of two timestamps. 3375 */ 3376 void 3377 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3378 { 3379 int32_t delta; 3380 3381 TCP_STATINC(TCP_STAT_RTTUPDATED); 3382 if (tp->t_srtt != 0) { 3383 /* 3384 * Compute the amount to add to srtt for smoothing, 3385 * *alpha, or 2^(-TCP_RTT_SHIFT). Because 3386 * srtt is stored in 1/32 slow ticks, we conceptually 3387 * shift left 5 bits, subtract srtt to get the 3388 * diference, and then shift right by TCP_RTT_SHIFT 3389 * (3) to obtain 1/8 of the difference. 3390 */ 3391 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3392 /* 3393 * This can never happen, because delta's lowest 3394 * possible value is 1/8 of t_srtt. But if it does, 3395 * set srtt to some reasonable value, here chosen 3396 * as 1/8 tick. 3397 */ 3398 if ((tp->t_srtt += delta) <= 0) 3399 tp->t_srtt = 1 << 2; 3400 /* 3401 * RFC2988 requires that rttvar be updated first. 3402 * This code is compliant because "delta" is the old 3403 * srtt minus the new observation (scaled). 3404 * 3405 * RFC2988 says: 3406 * rttvar = (1-beta) * rttvar + beta * |srtt-observed| 3407 * 3408 * delta is in units of 1/32 ticks, and has then been 3409 * divided by 8. This is equivalent to being in 1/16s 3410 * units and divided by 4. Subtract from it 1/4 of 3411 * the existing rttvar to form the (signed) amount to 3412 * adjust. 3413 */ 3414 if (delta < 0) 3415 delta = -delta; 3416 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3417 /* 3418 * As with srtt, this should never happen. There is 3419 * no support in RFC2988 for this operation. But 1/4s 3420 * as rttvar when faced with something arguably wrong 3421 * is ok. 3422 */ 3423 if ((tp->t_rttvar += delta) <= 0) 3424 tp->t_rttvar = 1 << 2; 3425 3426 /* 3427 * If srtt exceeds .01 second, ensure we use the 'remote' MSL 3428 * Problem is: it doesn't work. Disabled by defaulting 3429 * tcp_rttlocal to 0; see corresponding code in 3430 * tcp_subr that selects local vs remote in a different way. 3431 * 3432 * The static branch prediction hint here should be removed 3433 * when the rtt estimator is fixed and the rtt_enable code 3434 * is turned back on. 3435 */ 3436 if (__predict_false(tcp_rttlocal) && tcp_msl_enable 3437 && tp->t_srtt > tcp_msl_remote_threshold 3438 && tp->t_msl < tcp_msl_remote) { 3439 tp->t_msl = tcp_msl_remote; 3440 } 3441 } else { 3442 /* 3443 * This is the first measurement. Per RFC2988, 2.2, 3444 * set rtt=R and srtt=R/2. 3445 * For srtt, storage representation is 1/32 ticks, 3446 * so shift left by 5. 3447 * For rttvar, storage representation is 1/16 ticks, 3448 * So shift left by 4, but then right by 1 to halve. 3449 */ 3450 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3451 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3452 } 3453 tp->t_rtttime = 0; 3454 tp->t_rxtshift = 0; 3455 3456 /* 3457 * the retransmit should happen at rtt + 4 * rttvar. 3458 * Because of the way we do the smoothing, srtt and rttvar 3459 * will each average +1/2 tick of bias. When we compute 3460 * the retransmit timer, we want 1/2 tick of rounding and 3461 * 1 extra tick because of +-1/2 tick uncertainty in the 3462 * firing of the timer. The bias will give us exactly the 3463 * 1.5 tick we need. But, because the bias is 3464 * statistical, we have to test that we don't drop below 3465 * the minimum feasible timer (which is 2 ticks). 3466 */ 3467 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3468 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3469 3470 /* 3471 * We received an ack for a packet that wasn't retransmitted; 3472 * it is probably safe to discard any error indications we've 3473 * received recently. This isn't quite right, but close enough 3474 * for now (a route might have failed after we sent a segment, 3475 * and the return path might not be symmetrical). 3476 */ 3477 tp->t_softerror = 0; 3478 } 3479 3480 3481 /* 3482 * TCP compressed state engine. Currently used to hold compressed 3483 * state for SYN_RECEIVED. 3484 */ 3485 3486 u_long syn_cache_count; 3487 u_int32_t syn_hash1, syn_hash2; 3488 3489 #define SYN_HASH(sa, sp, dp) \ 3490 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3491 ((u_int32_t)(sp)))^syn_hash2))) 3492 #ifndef INET6 3493 #define SYN_HASHALL(hash, src, dst) \ 3494 do { \ 3495 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3496 ((const struct sockaddr_in *)(src))->sin_port, \ 3497 ((const struct sockaddr_in *)(dst))->sin_port); \ 3498 } while (/*CONSTCOND*/ 0) 3499 #else 3500 #define SYN_HASH6(sa, sp, dp) \ 3501 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3502 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3503 & 0x7fffffff) 3504 3505 #define SYN_HASHALL(hash, src, dst) \ 3506 do { \ 3507 switch ((src)->sa_family) { \ 3508 case AF_INET: \ 3509 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3510 ((const struct sockaddr_in *)(src))->sin_port, \ 3511 ((const struct sockaddr_in *)(dst))->sin_port); \ 3512 break; \ 3513 case AF_INET6: \ 3514 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3515 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3516 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3517 break; \ 3518 default: \ 3519 hash = 0; \ 3520 } \ 3521 } while (/*CONSTCOND*/0) 3522 #endif /* INET6 */ 3523 3524 static struct pool syn_cache_pool; 3525 3526 /* 3527 * We don't estimate RTT with SYNs, so each packet starts with the default 3528 * RTT and each timer step has a fixed timeout value. 3529 */ 3530 static inline void 3531 syn_cache_timer_arm(struct syn_cache *sc) 3532 { 3533 3534 TCPT_RANGESET(sc->sc_rxtcur, 3535 TCPTV_SRTTDFLT * tcp_backoff[sc->sc_rxtshift], TCPTV_MIN, 3536 TCPTV_REXMTMAX); 3537 callout_reset(&sc->sc_timer, 3538 sc->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, sc); 3539 } 3540 3541 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3542 3543 static inline void 3544 syn_cache_rm(struct syn_cache *sc) 3545 { 3546 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3547 sc, sc_bucketq); 3548 sc->sc_tp = NULL; 3549 LIST_REMOVE(sc, sc_tpq); 3550 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3551 callout_stop(&sc->sc_timer); 3552 syn_cache_count--; 3553 } 3554 3555 static inline void 3556 syn_cache_put(struct syn_cache *sc) 3557 { 3558 if (sc->sc_ipopts) 3559 (void) m_free(sc->sc_ipopts); 3560 rtcache_free(&sc->sc_route); 3561 sc->sc_flags |= SCF_DEAD; 3562 if (!callout_invoking(&sc->sc_timer)) 3563 callout_schedule(&(sc)->sc_timer, 1); 3564 } 3565 3566 void 3567 syn_cache_init(void) 3568 { 3569 int i; 3570 3571 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 3572 "synpl", NULL, IPL_SOFTNET); 3573 3574 /* Initialize the hash buckets. */ 3575 for (i = 0; i < tcp_syn_cache_size; i++) 3576 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3577 } 3578 3579 void 3580 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3581 { 3582 struct syn_cache_head *scp; 3583 struct syn_cache *sc2; 3584 int s; 3585 3586 /* 3587 * If there are no entries in the hash table, reinitialize 3588 * the hash secrets. 3589 */ 3590 if (syn_cache_count == 0) { 3591 syn_hash1 = cprng_fast32(); 3592 syn_hash2 = cprng_fast32(); 3593 } 3594 3595 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3596 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3597 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3598 3599 /* 3600 * Make sure that we don't overflow the per-bucket 3601 * limit or the total cache size limit. 3602 */ 3603 s = splsoftnet(); 3604 if (scp->sch_length >= tcp_syn_bucket_limit) { 3605 TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW); 3606 /* 3607 * The bucket is full. Toss the oldest element in the 3608 * bucket. This will be the first entry in the bucket. 3609 */ 3610 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3611 #ifdef DIAGNOSTIC 3612 /* 3613 * This should never happen; we should always find an 3614 * entry in our bucket. 3615 */ 3616 if (sc2 == NULL) 3617 panic("syn_cache_insert: bucketoverflow: impossible"); 3618 #endif 3619 syn_cache_rm(sc2); 3620 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3621 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3622 struct syn_cache_head *scp2, *sce; 3623 3624 TCP_STATINC(TCP_STAT_SC_OVERFLOWED); 3625 /* 3626 * The cache is full. Toss the oldest entry in the 3627 * first non-empty bucket we can find. 3628 * 3629 * XXX We would really like to toss the oldest 3630 * entry in the cache, but we hope that this 3631 * condition doesn't happen very often. 3632 */ 3633 scp2 = scp; 3634 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3635 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3636 for (++scp2; scp2 != scp; scp2++) { 3637 if (scp2 >= sce) 3638 scp2 = &tcp_syn_cache[0]; 3639 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3640 break; 3641 } 3642 #ifdef DIAGNOSTIC 3643 /* 3644 * This should never happen; we should always find a 3645 * non-empty bucket. 3646 */ 3647 if (scp2 == scp) 3648 panic("syn_cache_insert: cacheoverflow: " 3649 "impossible"); 3650 #endif 3651 } 3652 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3653 syn_cache_rm(sc2); 3654 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3655 } 3656 3657 /* 3658 * Initialize the entry's timer. 3659 */ 3660 sc->sc_rxttot = 0; 3661 sc->sc_rxtshift = 0; 3662 syn_cache_timer_arm(sc); 3663 3664 /* Link it from tcpcb entry */ 3665 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3666 3667 /* Put it into the bucket. */ 3668 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3669 scp->sch_length++; 3670 syn_cache_count++; 3671 3672 TCP_STATINC(TCP_STAT_SC_ADDED); 3673 splx(s); 3674 } 3675 3676 /* 3677 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3678 * If we have retransmitted an entry the maximum number of times, expire 3679 * that entry. 3680 */ 3681 static void 3682 syn_cache_timer(void *arg) 3683 { 3684 struct syn_cache *sc = arg; 3685 3686 mutex_enter(softnet_lock); 3687 KERNEL_LOCK(1, NULL); 3688 3689 callout_ack(&sc->sc_timer); 3690 3691 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3692 TCP_STATINC(TCP_STAT_SC_DELAYED_FREE); 3693 goto free; 3694 } 3695 3696 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3697 /* Drop it -- too many retransmissions. */ 3698 goto dropit; 3699 } 3700 3701 /* 3702 * Compute the total amount of time this entry has 3703 * been on a queue. If this entry has been on longer 3704 * than the keep alive timer would allow, expire it. 3705 */ 3706 sc->sc_rxttot += sc->sc_rxtcur; 3707 if (sc->sc_rxttot >= tcp_keepinit) 3708 goto dropit; 3709 3710 TCP_STATINC(TCP_STAT_SC_RETRANSMITTED); 3711 (void)syn_cache_respond(sc); 3712 3713 /* Advance the timer back-off. */ 3714 sc->sc_rxtshift++; 3715 syn_cache_timer_arm(sc); 3716 3717 goto out; 3718 3719 dropit: 3720 TCP_STATINC(TCP_STAT_SC_TIMED_OUT); 3721 syn_cache_rm(sc); 3722 if (sc->sc_ipopts) 3723 (void) m_free(sc->sc_ipopts); 3724 rtcache_free(&sc->sc_route); 3725 3726 free: 3727 callout_destroy(&sc->sc_timer); 3728 pool_put(&syn_cache_pool, sc); 3729 3730 out: 3731 KERNEL_UNLOCK_ONE(NULL); 3732 mutex_exit(softnet_lock); 3733 } 3734 3735 /* 3736 * Remove syn cache created by the specified tcb entry, 3737 * because this does not make sense to keep them 3738 * (if there's no tcb entry, syn cache entry will never be used) 3739 */ 3740 void 3741 syn_cache_cleanup(struct tcpcb *tp) 3742 { 3743 struct syn_cache *sc, *nsc; 3744 int s; 3745 3746 s = splsoftnet(); 3747 3748 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3749 nsc = LIST_NEXT(sc, sc_tpq); 3750 3751 #ifdef DIAGNOSTIC 3752 if (sc->sc_tp != tp) 3753 panic("invalid sc_tp in syn_cache_cleanup"); 3754 #endif 3755 syn_cache_rm(sc); 3756 syn_cache_put(sc); /* calls pool_put but see spl above */ 3757 } 3758 /* just for safety */ 3759 LIST_INIT(&tp->t_sc); 3760 3761 splx(s); 3762 } 3763 3764 /* 3765 * Find an entry in the syn cache. 3766 */ 3767 struct syn_cache * 3768 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3769 struct syn_cache_head **headp) 3770 { 3771 struct syn_cache *sc; 3772 struct syn_cache_head *scp; 3773 u_int32_t hash; 3774 int s; 3775 3776 SYN_HASHALL(hash, src, dst); 3777 3778 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3779 *headp = scp; 3780 s = splsoftnet(); 3781 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3782 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3783 if (sc->sc_hash != hash) 3784 continue; 3785 if (!memcmp(&sc->sc_src, src, src->sa_len) && 3786 !memcmp(&sc->sc_dst, dst, dst->sa_len)) { 3787 splx(s); 3788 return (sc); 3789 } 3790 } 3791 splx(s); 3792 return (NULL); 3793 } 3794 3795 /* 3796 * This function gets called when we receive an ACK for a 3797 * socket in the LISTEN state. We look up the connection 3798 * in the syn cache, and if its there, we pull it out of 3799 * the cache and turn it into a full-blown connection in 3800 * the SYN-RECEIVED state. 3801 * 3802 * The return values may not be immediately obvious, and their effects 3803 * can be subtle, so here they are: 3804 * 3805 * NULL SYN was not found in cache; caller should drop the 3806 * packet and send an RST. 3807 * 3808 * -1 We were unable to create the new connection, and are 3809 * aborting it. An ACK,RST is being sent to the peer 3810 * (unless we got screwey sequence numbners; see below), 3811 * because the 3-way handshake has been completed. Caller 3812 * should not free the mbuf, since we may be using it. If 3813 * we are not, we will free it. 3814 * 3815 * Otherwise, the return value is a pointer to the new socket 3816 * associated with the connection. 3817 */ 3818 struct socket * 3819 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3820 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3821 struct socket *so, struct mbuf *m) 3822 { 3823 struct syn_cache *sc; 3824 struct syn_cache_head *scp; 3825 struct inpcb *inp = NULL; 3826 #ifdef INET6 3827 struct in6pcb *in6p = NULL; 3828 #endif 3829 struct tcpcb *tp = 0; 3830 int s; 3831 struct socket *oso; 3832 3833 s = splsoftnet(); 3834 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3835 splx(s); 3836 return (NULL); 3837 } 3838 3839 /* 3840 * Verify the sequence and ack numbers. Try getting the correct 3841 * response again. 3842 */ 3843 if ((th->th_ack != sc->sc_iss + 1) || 3844 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3845 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3846 m_freem(m); 3847 (void)syn_cache_respond(sc); 3848 splx(s); 3849 return ((struct socket *)(-1)); 3850 } 3851 3852 /* Remove this cache entry */ 3853 syn_cache_rm(sc); 3854 splx(s); 3855 3856 /* 3857 * Ok, create the full blown connection, and set things up 3858 * as they would have been set up if we had created the 3859 * connection when the SYN arrived. If we can't create 3860 * the connection, abort it. 3861 */ 3862 /* 3863 * inp still has the OLD in_pcb stuff, set the 3864 * v6-related flags on the new guy, too. This is 3865 * done particularly for the case where an AF_INET6 3866 * socket is bound only to a port, and a v4 connection 3867 * comes in on that port. 3868 * we also copy the flowinfo from the original pcb 3869 * to the new one. 3870 */ 3871 oso = so; 3872 so = sonewconn(so, true); 3873 if (so == NULL) 3874 goto resetandabort; 3875 3876 switch (so->so_proto->pr_domain->dom_family) { 3877 #ifdef INET 3878 case AF_INET: 3879 inp = sotoinpcb(so); 3880 break; 3881 #endif 3882 #ifdef INET6 3883 case AF_INET6: 3884 in6p = sotoin6pcb(so); 3885 break; 3886 #endif 3887 } 3888 switch (src->sa_family) { 3889 #ifdef INET 3890 case AF_INET: 3891 if (inp) { 3892 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3893 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3894 inp->inp_options = ip_srcroute(m); 3895 in_pcbstate(inp, INP_BOUND); 3896 if (inp->inp_options == NULL) { 3897 inp->inp_options = sc->sc_ipopts; 3898 sc->sc_ipopts = NULL; 3899 } 3900 } 3901 #ifdef INET6 3902 else if (in6p) { 3903 /* IPv4 packet to AF_INET6 socket */ 3904 memset(&in6p->in6p_laddr, 0, sizeof(in6p->in6p_laddr)); 3905 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3906 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3907 &in6p->in6p_laddr.s6_addr32[3], 3908 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3909 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3910 in6totcpcb(in6p)->t_family = AF_INET; 3911 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3912 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3913 else 3914 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3915 in6_pcbstate(in6p, IN6P_BOUND); 3916 } 3917 #endif 3918 break; 3919 #endif 3920 #ifdef INET6 3921 case AF_INET6: 3922 if (in6p) { 3923 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3924 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3925 in6_pcbstate(in6p, IN6P_BOUND); 3926 } 3927 break; 3928 #endif 3929 } 3930 #ifdef INET6 3931 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3932 struct in6pcb *oin6p = sotoin6pcb(oso); 3933 /* inherit socket options from the listening socket */ 3934 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3935 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3936 m_freem(in6p->in6p_options); 3937 in6p->in6p_options = 0; 3938 } 3939 ip6_savecontrol(in6p, &in6p->in6p_options, 3940 mtod(m, struct ip6_hdr *), m); 3941 } 3942 #endif 3943 3944 #if defined(IPSEC) 3945 if (ipsec_used) { 3946 /* 3947 * we make a copy of policy, instead of sharing the policy, for 3948 * better behavior in terms of SA lookup and dead SA removal. 3949 */ 3950 if (inp) { 3951 /* copy old policy into new socket's */ 3952 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, 3953 inp->inp_sp)) 3954 printf("tcp_input: could not copy policy\n"); 3955 } 3956 #ifdef INET6 3957 else if (in6p) { 3958 /* copy old policy into new socket's */ 3959 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3960 in6p->in6p_sp)) 3961 printf("tcp_input: could not copy policy\n"); 3962 } 3963 #endif 3964 } 3965 #endif 3966 3967 /* 3968 * Give the new socket our cached route reference. 3969 */ 3970 if (inp) { 3971 rtcache_copy(&inp->inp_route, &sc->sc_route); 3972 rtcache_free(&sc->sc_route); 3973 } 3974 #ifdef INET6 3975 else { 3976 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 3977 rtcache_free(&sc->sc_route); 3978 } 3979 #endif 3980 3981 if (inp) { 3982 struct sockaddr_in sin; 3983 memcpy(&sin, src, src->sa_len); 3984 if (in_pcbconnect(inp, &sin, &lwp0)) { 3985 goto resetandabort; 3986 } 3987 } 3988 #ifdef INET6 3989 else if (in6p) { 3990 struct sockaddr_in6 sin6; 3991 memcpy(&sin6, src, src->sa_len); 3992 if (src->sa_family == AF_INET) { 3993 /* IPv4 packet to AF_INET6 socket */ 3994 in6_sin_2_v4mapsin6((struct sockaddr_in *)src, &sin6); 3995 } 3996 if (in6_pcbconnect(in6p, &sin6, NULL)) { 3997 goto resetandabort; 3998 } 3999 } 4000 #endif 4001 else { 4002 goto resetandabort; 4003 } 4004 4005 if (inp) 4006 tp = intotcpcb(inp); 4007 #ifdef INET6 4008 else if (in6p) 4009 tp = in6totcpcb(in6p); 4010 #endif 4011 else 4012 tp = NULL; 4013 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 4014 if (sc->sc_request_r_scale != 15) { 4015 tp->requested_s_scale = sc->sc_requested_s_scale; 4016 tp->request_r_scale = sc->sc_request_r_scale; 4017 tp->snd_scale = sc->sc_requested_s_scale; 4018 tp->rcv_scale = sc->sc_request_r_scale; 4019 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 4020 } 4021 if (sc->sc_flags & SCF_TIMESTAMP) 4022 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 4023 tp->ts_timebase = sc->sc_timebase; 4024 4025 tp->t_template = tcp_template(tp); 4026 if (tp->t_template == 0) { 4027 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 4028 so = NULL; 4029 m_freem(m); 4030 goto abort; 4031 } 4032 4033 tp->iss = sc->sc_iss; 4034 tp->irs = sc->sc_irs; 4035 tcp_sendseqinit(tp); 4036 tcp_rcvseqinit(tp); 4037 tp->t_state = TCPS_SYN_RECEIVED; 4038 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 4039 TCP_STATINC(TCP_STAT_ACCEPTS); 4040 4041 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 4042 tp->t_flags |= TF_WILL_SACK; 4043 4044 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 4045 tp->t_flags |= TF_ECN_PERMIT; 4046 4047 #ifdef TCP_SIGNATURE 4048 if (sc->sc_flags & SCF_SIGNATURE) 4049 tp->t_flags |= TF_SIGNATURE; 4050 #endif 4051 4052 /* Initialize tp->t_ourmss before we deal with the peer's! */ 4053 tp->t_ourmss = sc->sc_ourmaxseg; 4054 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 4055 4056 /* 4057 * Initialize the initial congestion window. If we 4058 * had to retransmit the SYN,ACK, we must initialize cwnd 4059 * to 1 segment (i.e. the Loss Window). 4060 */ 4061 if (sc->sc_rxtshift) 4062 tp->snd_cwnd = tp->t_peermss; 4063 else { 4064 int ss = tcp_init_win; 4065 #ifdef INET 4066 if (inp != NULL && in_localaddr(inp->inp_faddr)) 4067 ss = tcp_init_win_local; 4068 #endif 4069 #ifdef INET6 4070 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 4071 ss = tcp_init_win_local; 4072 #endif 4073 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 4074 } 4075 4076 tcp_rmx_rtt(tp); 4077 tp->snd_wl1 = sc->sc_irs; 4078 tp->rcv_up = sc->sc_irs + 1; 4079 4080 /* 4081 * This is what whould have happened in tcp_output() when 4082 * the SYN,ACK was sent. 4083 */ 4084 tp->snd_up = tp->snd_una; 4085 tp->snd_max = tp->snd_nxt = tp->iss+1; 4086 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 4087 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 4088 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 4089 tp->last_ack_sent = tp->rcv_nxt; 4090 tp->t_partialacks = -1; 4091 tp->t_dupacks = 0; 4092 4093 TCP_STATINC(TCP_STAT_SC_COMPLETED); 4094 s = splsoftnet(); 4095 syn_cache_put(sc); 4096 splx(s); 4097 return (so); 4098 4099 resetandabort: 4100 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 4101 abort: 4102 if (so != NULL) { 4103 (void) soqremque(so, 1); 4104 (void) soabort(so); 4105 mutex_enter(softnet_lock); 4106 } 4107 s = splsoftnet(); 4108 syn_cache_put(sc); 4109 splx(s); 4110 TCP_STATINC(TCP_STAT_SC_ABORTED); 4111 return ((struct socket *)(-1)); 4112 } 4113 4114 /* 4115 * This function is called when we get a RST for a 4116 * non-existent connection, so that we can see if the 4117 * connection is in the syn cache. If it is, zap it. 4118 */ 4119 4120 void 4121 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 4122 { 4123 struct syn_cache *sc; 4124 struct syn_cache_head *scp; 4125 int s = splsoftnet(); 4126 4127 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4128 splx(s); 4129 return; 4130 } 4131 if (SEQ_LT(th->th_seq, sc->sc_irs) || 4132 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 4133 splx(s); 4134 return; 4135 } 4136 syn_cache_rm(sc); 4137 TCP_STATINC(TCP_STAT_SC_RESET); 4138 syn_cache_put(sc); /* calls pool_put but see spl above */ 4139 splx(s); 4140 } 4141 4142 void 4143 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 4144 struct tcphdr *th) 4145 { 4146 struct syn_cache *sc; 4147 struct syn_cache_head *scp; 4148 int s; 4149 4150 s = splsoftnet(); 4151 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4152 splx(s); 4153 return; 4154 } 4155 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 4156 if (ntohl (th->th_seq) != sc->sc_iss) { 4157 splx(s); 4158 return; 4159 } 4160 4161 /* 4162 * If we've retransmitted 3 times and this is our second error, 4163 * we remove the entry. Otherwise, we allow it to continue on. 4164 * This prevents us from incorrectly nuking an entry during a 4165 * spurious network outage. 4166 * 4167 * See tcp_notify(). 4168 */ 4169 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 4170 sc->sc_flags |= SCF_UNREACH; 4171 splx(s); 4172 return; 4173 } 4174 4175 syn_cache_rm(sc); 4176 TCP_STATINC(TCP_STAT_SC_UNREACH); 4177 syn_cache_put(sc); /* calls pool_put but see spl above */ 4178 splx(s); 4179 } 4180 4181 /* 4182 * Given a LISTEN socket and an inbound SYN request, add 4183 * this to the syn cache, and send back a segment: 4184 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 4185 * to the source. 4186 * 4187 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 4188 * Doing so would require that we hold onto the data and deliver it 4189 * to the application. However, if we are the target of a SYN-flood 4190 * DoS attack, an attacker could send data which would eventually 4191 * consume all available buffer space if it were ACKed. By not ACKing 4192 * the data, we avoid this DoS scenario. 4193 */ 4194 int 4195 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 4196 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 4197 int optlen, struct tcp_opt_info *oi) 4198 { 4199 struct tcpcb tb, *tp; 4200 long win; 4201 struct syn_cache *sc; 4202 struct syn_cache_head *scp; 4203 struct mbuf *ipopts; 4204 struct tcp_opt_info opti; 4205 int s; 4206 4207 tp = sototcpcb(so); 4208 4209 memset(&opti, 0, sizeof(opti)); 4210 4211 /* 4212 * Initialize some local state. 4213 */ 4214 win = sbspace(&so->so_rcv); 4215 if (win > TCP_MAXWIN) 4216 win = TCP_MAXWIN; 4217 4218 #ifdef TCP_SIGNATURE 4219 if (optp || (tp->t_flags & TF_SIGNATURE)) 4220 #else 4221 if (optp) 4222 #endif 4223 { 4224 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4225 #ifdef TCP_SIGNATURE 4226 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4227 #endif 4228 tb.t_state = TCPS_LISTEN; 4229 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 4230 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 4231 return 0; 4232 } else 4233 tb.t_flags = 0; 4234 4235 switch (src->sa_family) { 4236 #ifdef INET 4237 case AF_INET: 4238 /* 4239 * Remember the IP options, if any. 4240 */ 4241 ipopts = ip_srcroute(m); 4242 break; 4243 #endif 4244 default: 4245 ipopts = NULL; 4246 } 4247 4248 /* 4249 * See if we already have an entry for this connection. 4250 * If we do, resend the SYN,ACK. We do not count this 4251 * as a retransmission (XXX though maybe we should). 4252 */ 4253 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4254 TCP_STATINC(TCP_STAT_SC_DUPESYN); 4255 if (ipopts) { 4256 /* 4257 * If we were remembering a previous source route, 4258 * forget it and use the new one we've been given. 4259 */ 4260 if (sc->sc_ipopts) 4261 (void)m_free(sc->sc_ipopts); 4262 sc->sc_ipopts = ipopts; 4263 } 4264 sc->sc_timestamp = tb.ts_recent; 4265 m_freem(m); 4266 if (syn_cache_respond(sc) == 0) { 4267 uint64_t *tcps = TCP_STAT_GETREF(); 4268 tcps[TCP_STAT_SNDACKS]++; 4269 tcps[TCP_STAT_SNDTOTAL]++; 4270 TCP_STAT_PUTREF(); 4271 } 4272 return 1; 4273 } 4274 4275 s = splsoftnet(); 4276 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4277 splx(s); 4278 if (sc == NULL) { 4279 if (ipopts) 4280 (void)m_free(ipopts); 4281 return 0; 4282 } 4283 4284 /* 4285 * Fill in the cache, and put the necessary IP and TCP 4286 * options into the reply. 4287 */ 4288 memset(sc, 0, sizeof(struct syn_cache)); 4289 callout_init(&sc->sc_timer, CALLOUT_MPSAFE); 4290 bcopy(src, &sc->sc_src, src->sa_len); 4291 bcopy(dst, &sc->sc_dst, dst->sa_len); 4292 sc->sc_flags = 0; 4293 sc->sc_ipopts = ipopts; 4294 sc->sc_irs = th->th_seq; 4295 switch (src->sa_family) { 4296 #ifdef INET 4297 case AF_INET: 4298 { 4299 struct sockaddr_in *srcin = (void *)src; 4300 struct sockaddr_in *dstin = (void *)dst; 4301 4302 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4303 &srcin->sin_addr, dstin->sin_port, 4304 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4305 break; 4306 } 4307 #endif /* INET */ 4308 #ifdef INET6 4309 case AF_INET6: 4310 { 4311 struct sockaddr_in6 *srcin6 = (void *)src; 4312 struct sockaddr_in6 *dstin6 = (void *)dst; 4313 4314 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4315 &srcin6->sin6_addr, dstin6->sin6_port, 4316 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4317 break; 4318 } 4319 #endif /* INET6 */ 4320 } 4321 sc->sc_peermaxseg = oi->maxseg; 4322 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4323 m_get_rcvif_NOMPSAFE(m) : NULL, sc->sc_src.sa.sa_family); 4324 sc->sc_win = win; 4325 sc->sc_timebase = tcp_now - 1; /* see tcp_newtcpcb() */ 4326 sc->sc_timestamp = tb.ts_recent; 4327 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4328 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4329 sc->sc_flags |= SCF_TIMESTAMP; 4330 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4331 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4332 sc->sc_requested_s_scale = tb.requested_s_scale; 4333 sc->sc_request_r_scale = 0; 4334 /* 4335 * Pick the smallest possible scaling factor that 4336 * will still allow us to scale up to sb_max. 4337 * 4338 * We do this because there are broken firewalls that 4339 * will corrupt the window scale option, leading to 4340 * the other endpoint believing that our advertised 4341 * window is unscaled. At scale factors larger than 4342 * 5 the unscaled window will drop below 1500 bytes, 4343 * leading to serious problems when traversing these 4344 * broken firewalls. 4345 * 4346 * With the default sbmax of 256K, a scale factor 4347 * of 3 will be chosen by this algorithm. Those who 4348 * choose a larger sbmax should watch out 4349 * for the compatiblity problems mentioned above. 4350 * 4351 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4352 * or <SYN,ACK>) segment itself is never scaled. 4353 */ 4354 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4355 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4356 sc->sc_request_r_scale++; 4357 } else { 4358 sc->sc_requested_s_scale = 15; 4359 sc->sc_request_r_scale = 15; 4360 } 4361 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4362 sc->sc_flags |= SCF_SACK_PERMIT; 4363 4364 /* 4365 * ECN setup packet received. 4366 */ 4367 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4368 sc->sc_flags |= SCF_ECN_PERMIT; 4369 4370 #ifdef TCP_SIGNATURE 4371 if (tb.t_flags & TF_SIGNATURE) 4372 sc->sc_flags |= SCF_SIGNATURE; 4373 #endif 4374 sc->sc_tp = tp; 4375 m_freem(m); 4376 if (syn_cache_respond(sc) == 0) { 4377 uint64_t *tcps = TCP_STAT_GETREF(); 4378 tcps[TCP_STAT_SNDACKS]++; 4379 tcps[TCP_STAT_SNDTOTAL]++; 4380 TCP_STAT_PUTREF(); 4381 syn_cache_insert(sc, tp); 4382 } else { 4383 s = splsoftnet(); 4384 /* 4385 * syn_cache_put() will try to schedule the timer, so 4386 * we need to initialize it 4387 */ 4388 syn_cache_timer_arm(sc); 4389 syn_cache_put(sc); 4390 splx(s); 4391 TCP_STATINC(TCP_STAT_SC_DROPPED); 4392 } 4393 return 1; 4394 } 4395 4396 /* 4397 * syn_cache_respond: (re)send SYN+ACK. 4398 * 4399 * returns 0 on success. otherwise returns an errno, typically ENOBUFS. 4400 */ 4401 4402 int 4403 syn_cache_respond(struct syn_cache *sc) 4404 { 4405 #ifdef INET6 4406 struct rtentry *rt = NULL; 4407 #endif 4408 struct route *ro; 4409 u_int8_t *optp; 4410 int optlen, error; 4411 u_int16_t tlen; 4412 struct ip *ip = NULL; 4413 #ifdef INET6 4414 struct ip6_hdr *ip6 = NULL; 4415 #endif 4416 struct tcpcb *tp = NULL; 4417 struct tcphdr *th; 4418 struct mbuf *m; 4419 u_int hlen; 4420 #ifdef TCP_SIGNATURE 4421 struct secasvar *sav = NULL; 4422 u_int8_t *sigp = NULL; 4423 #endif 4424 4425 ro = &sc->sc_route; 4426 switch (sc->sc_src.sa.sa_family) { 4427 case AF_INET: 4428 hlen = sizeof(struct ip); 4429 break; 4430 #ifdef INET6 4431 case AF_INET6: 4432 hlen = sizeof(struct ip6_hdr); 4433 break; 4434 #endif 4435 default: 4436 return (EAFNOSUPPORT); 4437 } 4438 4439 /* worst case scanario, since we don't know the option size yet */ 4440 tlen = hlen + sizeof(struct tcphdr) + MAX_TCPOPTLEN; 4441 4442 /* 4443 * Create the IP+TCP header from scratch. 4444 */ 4445 #ifdef DIAGNOSTIC 4446 if (max_linkhdr + tlen > MCLBYTES) 4447 return ENOBUFS; 4448 #endif 4449 4450 MGETHDR(m, M_DONTWAIT, MT_DATA); 4451 if (m && (max_linkhdr + tlen) > MHLEN) { 4452 MCLGET(m, M_DONTWAIT); 4453 if ((m->m_flags & M_EXT) == 0) { 4454 m_freem(m); 4455 m = NULL; 4456 } 4457 } 4458 if (m == NULL) 4459 return ENOBUFS; 4460 MCLAIM(m, &tcp_tx_mowner); 4461 4462 /* Fixup the mbuf. */ 4463 m->m_data += max_linkhdr; 4464 if (sc->sc_tp) 4465 tp = sc->sc_tp; 4466 m_reset_rcvif(m); 4467 memset(mtod(m, u_char *), 0, tlen); 4468 4469 switch (sc->sc_src.sa.sa_family) { 4470 case AF_INET: 4471 ip = mtod(m, struct ip *); 4472 ip->ip_v = 4; 4473 ip->ip_dst = sc->sc_src.sin.sin_addr; 4474 ip->ip_src = sc->sc_dst.sin.sin_addr; 4475 ip->ip_p = IPPROTO_TCP; 4476 th = (struct tcphdr *)(ip + 1); 4477 th->th_dport = sc->sc_src.sin.sin_port; 4478 th->th_sport = sc->sc_dst.sin.sin_port; 4479 break; 4480 #ifdef INET6 4481 case AF_INET6: 4482 ip6 = mtod(m, struct ip6_hdr *); 4483 ip6->ip6_vfc = IPV6_VERSION; 4484 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4485 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4486 ip6->ip6_nxt = IPPROTO_TCP; 4487 /* ip6_plen will be updated in ip6_output() */ 4488 th = (struct tcphdr *)(ip6 + 1); 4489 th->th_dport = sc->sc_src.sin6.sin6_port; 4490 th->th_sport = sc->sc_dst.sin6.sin6_port; 4491 break; 4492 #endif 4493 default: 4494 return ENOBUFS; 4495 } 4496 4497 th->th_seq = htonl(sc->sc_iss); 4498 th->th_ack = htonl(sc->sc_irs + 1); 4499 th->th_flags = TH_SYN|TH_ACK; 4500 th->th_win = htons(sc->sc_win); 4501 /* th_x2, th_sum, th_urp already 0 from memset */ 4502 4503 /* Tack on the TCP options. */ 4504 optp = (u_int8_t *)(th + 1); 4505 optlen = 0; 4506 *optp++ = TCPOPT_MAXSEG; 4507 *optp++ = TCPOLEN_MAXSEG; 4508 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4509 *optp++ = sc->sc_ourmaxseg & 0xff; 4510 optlen += TCPOLEN_MAXSEG; 4511 4512 if (sc->sc_request_r_scale != 15) { 4513 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4514 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4515 sc->sc_request_r_scale); 4516 optp += TCPOLEN_WINDOW + TCPOLEN_NOP; 4517 optlen += TCPOLEN_WINDOW + TCPOLEN_NOP; 4518 } 4519 4520 if (sc->sc_flags & SCF_SACK_PERMIT) { 4521 /* Let the peer know that we will SACK. */ 4522 *optp++ = TCPOPT_SACK_PERMITTED; 4523 *optp++ = TCPOLEN_SACK_PERMITTED; 4524 optlen += TCPOLEN_SACK_PERMITTED; 4525 } 4526 4527 if (sc->sc_flags & SCF_TIMESTAMP) { 4528 while (optlen % 4 != 2) { 4529 optlen += TCPOLEN_NOP; 4530 *optp++ = TCPOPT_NOP; 4531 } 4532 *optp++ = TCPOPT_TIMESTAMP; 4533 *optp++ = TCPOLEN_TIMESTAMP; 4534 u_int32_t *lp = (u_int32_t *)(optp); 4535 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4536 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4537 *lp = htonl(sc->sc_timestamp); 4538 optp += TCPOLEN_TIMESTAMP - 2; 4539 optlen += TCPOLEN_TIMESTAMP; 4540 } 4541 4542 #ifdef TCP_SIGNATURE 4543 if (sc->sc_flags & SCF_SIGNATURE) { 4544 sav = tcp_signature_getsav(m); 4545 if (sav == NULL) { 4546 if (m) 4547 m_freem(m); 4548 return (EPERM); 4549 } 4550 4551 *optp++ = TCPOPT_SIGNATURE; 4552 *optp++ = TCPOLEN_SIGNATURE; 4553 sigp = optp; 4554 memset(optp, 0, TCP_SIGLEN); 4555 optp += TCP_SIGLEN; 4556 optlen += TCPOLEN_SIGNATURE; 4557 4558 } 4559 #endif 4560 /* Terminate and pad TCP options to a 4 byte boundary. */ 4561 if (optlen % 4) { 4562 optlen += TCPOLEN_EOL; 4563 *optp++ = TCPOPT_EOL; 4564 } 4565 /* 4566 * According to RFC 793 (STD0007): 4567 * "The content of the header beyond the End-of-Option option 4568 * must be header padding (i.e., zero)." 4569 * and later: "The padding is composed of zeros." 4570 */ 4571 while (optlen % 4) { 4572 optlen += TCPOLEN_PAD; 4573 *optp++ = TCPOPT_PAD; 4574 } 4575 4576 /* compute the actual values now that we've added the options */ 4577 tlen = hlen + sizeof(struct tcphdr) + optlen; 4578 m->m_len = m->m_pkthdr.len = tlen; 4579 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4580 4581 #ifdef TCP_SIGNATURE 4582 if (sav) { 4583 (void)tcp_signature(m, th, hlen, sav, sigp); 4584 key_sa_recordxfer(sav, m); 4585 KEY_SA_UNREF(&sav); 4586 } 4587 #endif 4588 4589 /* 4590 * Send ECN SYN-ACK setup packet. 4591 * Routes can be asymetric, so, even if we receive a packet 4592 * with ECE and CWR set, we must not assume no one will block 4593 * the ECE packet we are about to send. 4594 */ 4595 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4596 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4597 th->th_flags |= TH_ECE; 4598 TCP_STATINC(TCP_STAT_ECN_SHS); 4599 4600 /* 4601 * draft-ietf-tcpm-ecnsyn-00.txt 4602 * 4603 * "[...] a TCP node MAY respond to an ECN-setup 4604 * SYN packet by setting ECT in the responding 4605 * ECN-setup SYN/ACK packet, indicating to routers 4606 * that the SYN/ACK packet is ECN-Capable. 4607 * This allows a congested router along the path 4608 * to mark the packet instead of dropping the 4609 * packet as an indication of congestion." 4610 * 4611 * "[...] There can be a great benefit in setting 4612 * an ECN-capable codepoint in SYN/ACK packets [...] 4613 * Congestion is most likely to occur in 4614 * the server-to-client direction. As a result, 4615 * setting an ECN-capable codepoint in SYN/ACK 4616 * packets can reduce the occurence of three-second 4617 * retransmit timeouts resulting from the drop 4618 * of SYN/ACK packets." 4619 * 4620 * Page 4 and 6, January 2006. 4621 */ 4622 4623 switch (sc->sc_src.sa.sa_family) { 4624 #ifdef INET 4625 case AF_INET: 4626 ip->ip_tos |= IPTOS_ECN_ECT0; 4627 break; 4628 #endif 4629 #ifdef INET6 4630 case AF_INET6: 4631 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4632 break; 4633 #endif 4634 } 4635 TCP_STATINC(TCP_STAT_ECN_ECT); 4636 } 4637 4638 4639 /* Compute the packet's checksum. */ 4640 switch (sc->sc_src.sa.sa_family) { 4641 case AF_INET: 4642 ip->ip_len = htons(tlen - hlen); 4643 th->th_sum = 0; 4644 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4645 break; 4646 #ifdef INET6 4647 case AF_INET6: 4648 ip6->ip6_plen = htons(tlen - hlen); 4649 th->th_sum = 0; 4650 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4651 break; 4652 #endif 4653 } 4654 4655 /* 4656 * Fill in some straggling IP bits. Note the stack expects 4657 * ip_len to be in host order, for convenience. 4658 */ 4659 switch (sc->sc_src.sa.sa_family) { 4660 #ifdef INET 4661 case AF_INET: 4662 ip->ip_len = htons(tlen); 4663 ip->ip_ttl = ip_defttl; 4664 /* XXX tos? */ 4665 break; 4666 #endif 4667 #ifdef INET6 4668 case AF_INET6: 4669 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4670 ip6->ip6_vfc |= IPV6_VERSION; 4671 ip6->ip6_plen = htons(tlen - hlen); 4672 /* ip6_hlim will be initialized afterwards */ 4673 /* XXX flowlabel? */ 4674 break; 4675 #endif 4676 } 4677 4678 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4679 tp = sc->sc_tp; 4680 4681 switch (sc->sc_src.sa.sa_family) { 4682 #ifdef INET 4683 case AF_INET: 4684 error = ip_output(m, sc->sc_ipopts, ro, 4685 (ip_mtudisc ? IP_MTUDISC : 0), 4686 NULL, tp ? tp->t_inpcb : NULL); 4687 break; 4688 #endif 4689 #ifdef INET6 4690 case AF_INET6: 4691 ip6->ip6_hlim = in6_selecthlim(NULL, 4692 (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp : NULL); 4693 rtcache_unref(rt, ro); 4694 4695 error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, 4696 tp ? tp->t_in6pcb : NULL, NULL); 4697 break; 4698 #endif 4699 default: 4700 error = EAFNOSUPPORT; 4701 break; 4702 } 4703 return (error); 4704 } 4705
Indexes created Tue Sep 23 23:09:58 GMT 2025