tcp_input.c revision 1.387
1 /* $NetBSD: tcp_input.c,v 1.387 2018/03/23 08:57:40 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 1.387 2018/03/23 08:57:40 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 /* 1386 * Do quick retrieval of timestamp options. 1387 * 1388 * If timestamp is the only option and it's formatted as 1389 * recommended in RFC 1323 appendix A, we quickly get the 1390 * values now and don't bother calling tcp_dooptions(), 1391 * etc. 1392 */ 1393 if ((optlen == TCPOLEN_TSTAMP_APPA || 1394 (optlen > TCPOLEN_TSTAMP_APPA && 1395 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1396 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1397 (th->th_flags & TH_SYN) == 0) { 1398 opti.ts_present = 1; 1399 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1400 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1401 optp = NULL; /* we've parsed the options */ 1402 } 1403 } 1404 tiflags = th->th_flags; 1405 1406 /* 1407 * Checksum extended TCP header and data 1408 */ 1409 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1410 goto badcsum; 1411 1412 /* 1413 * Locate pcb for segment. 1414 */ 1415 findpcb: 1416 inp = NULL; 1417 #ifdef INET6 1418 in6p = NULL; 1419 #endif 1420 switch (af) { 1421 #ifdef INET 1422 case AF_INET: 1423 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1424 ip->ip_dst, th->th_dport, &vestige); 1425 if (inp == NULL && !vestige.valid) { 1426 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1427 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, 1428 th->th_dport); 1429 } 1430 #ifdef INET6 1431 if (inp == NULL && !vestige.valid) { 1432 struct in6_addr s, d; 1433 1434 /* mapped addr case */ 1435 in6_in_2_v4mapin6(&ip->ip_src, &s); 1436 in6_in_2_v4mapin6(&ip->ip_dst, &d); 1437 in6p = in6_pcblookup_connect(&tcbtable, &s, 1438 th->th_sport, &d, th->th_dport, 0, &vestige); 1439 if (in6p == 0 && !vestige.valid) { 1440 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1441 in6p = in6_pcblookup_bind(&tcbtable, &d, 1442 th->th_dport, 0); 1443 } 1444 } 1445 #endif 1446 #ifndef INET6 1447 if (inp == NULL && !vestige.valid) 1448 #else 1449 if (inp == NULL && in6p == NULL && !vestige.valid) 1450 #endif 1451 { 1452 TCP_STATINC(TCP_STAT_NOPORT); 1453 if (tcp_log_refused && 1454 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1455 tcp4_log_refused(ip, th); 1456 } 1457 tcp_fields_to_host(th); 1458 goto dropwithreset_ratelim; 1459 } 1460 #if defined(IPSEC) 1461 if (ipsec_used) { 1462 if (inp && ipsec_in_reject(m, inp)) { 1463 goto drop; 1464 } 1465 #ifdef INET6 1466 else if (in6p && ipsec_in_reject(m, in6p)) { 1467 goto drop; 1468 } 1469 #endif 1470 } 1471 #endif /*IPSEC*/ 1472 break; 1473 #endif /*INET*/ 1474 #ifdef INET6 1475 case AF_INET6: 1476 { 1477 int faith; 1478 1479 #if defined(NFAITH) && NFAITH > 0 1480 faith = faithprefix(&ip6->ip6_dst); 1481 #else 1482 faith = 0; 1483 #endif 1484 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1485 th->th_sport, &ip6->ip6_dst, th->th_dport, faith, &vestige); 1486 if (!in6p && !vestige.valid) { 1487 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1488 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1489 th->th_dport, faith); 1490 } 1491 if (!in6p && !vestige.valid) { 1492 TCP_STATINC(TCP_STAT_NOPORT); 1493 if (tcp_log_refused && 1494 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1495 tcp6_log_refused(ip6, th); 1496 } 1497 tcp_fields_to_host(th); 1498 goto dropwithreset_ratelim; 1499 } 1500 #if defined(IPSEC) 1501 if (ipsec_used && in6p && ipsec_in_reject(m, in6p)) { 1502 goto drop; 1503 } 1504 #endif 1505 break; 1506 } 1507 #endif 1508 } 1509 1510 tcp_fields_to_host(th); 1511 1512 /* 1513 * If the state is CLOSED (i.e., TCB does not exist) then 1514 * all data in the incoming segment is discarded. 1515 * If the TCB exists but is in CLOSED state, it is embryonic, 1516 * but should either do a listen or a connect soon. 1517 */ 1518 tp = NULL; 1519 so = NULL; 1520 if (inp) { 1521 /* Check the minimum TTL for socket. */ 1522 if (ip->ip_ttl < inp->inp_ip_minttl) 1523 goto drop; 1524 1525 tp = intotcpcb(inp); 1526 so = inp->inp_socket; 1527 } 1528 #ifdef INET6 1529 else if (in6p) { 1530 tp = in6totcpcb(in6p); 1531 so = in6p->in6p_socket; 1532 } 1533 #endif 1534 else if (vestige.valid) { 1535 /* We do not support the resurrection of vtw tcpcps. */ 1536 tcp_vtw_input(th, &vestige, m, tlen); 1537 m = NULL; 1538 goto drop; 1539 } 1540 1541 if (tp == NULL) 1542 goto dropwithreset_ratelim; 1543 if (tp->t_state == TCPS_CLOSED) 1544 goto drop; 1545 1546 KASSERT(so->so_lock == softnet_lock); 1547 KASSERT(solocked(so)); 1548 1549 /* Unscale the window into a 32-bit value. */ 1550 if ((tiflags & TH_SYN) == 0) 1551 tiwin = th->th_win << tp->snd_scale; 1552 else 1553 tiwin = th->th_win; 1554 1555 #ifdef INET6 1556 /* save packet options if user wanted */ 1557 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1558 if (in6p->in6p_options) { 1559 m_freem(in6p->in6p_options); 1560 in6p->in6p_options = NULL; 1561 } 1562 KASSERT(ip6 != NULL); 1563 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1564 } 1565 #endif 1566 1567 if (so->so_options & SO_DEBUG) { 1568 #ifdef TCP_DEBUG 1569 ostate = tp->t_state; 1570 #endif 1571 1572 tcp_saveti = NULL; 1573 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1574 goto nosave; 1575 1576 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1577 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1578 if (tcp_saveti == NULL) 1579 goto nosave; 1580 } else { 1581 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1582 if (tcp_saveti == NULL) 1583 goto nosave; 1584 MCLAIM(m, &tcp_mowner); 1585 tcp_saveti->m_len = iphlen; 1586 m_copydata(m, 0, iphlen, 1587 mtod(tcp_saveti, void *)); 1588 } 1589 1590 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1591 m_freem(tcp_saveti); 1592 tcp_saveti = NULL; 1593 } else { 1594 tcp_saveti->m_len += sizeof(struct tcphdr); 1595 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1596 sizeof(struct tcphdr)); 1597 } 1598 nosave:; 1599 } 1600 1601 if (so->so_options & SO_ACCEPTCONN) { 1602 union syn_cache_sa src; 1603 union syn_cache_sa dst; 1604 1605 KASSERT(tp->t_state == TCPS_LISTEN); 1606 1607 memset(&src, 0, sizeof(src)); 1608 memset(&dst, 0, sizeof(dst)); 1609 switch (af) { 1610 #ifdef INET 1611 case AF_INET: 1612 src.sin.sin_len = sizeof(struct sockaddr_in); 1613 src.sin.sin_family = AF_INET; 1614 src.sin.sin_addr = ip->ip_src; 1615 src.sin.sin_port = th->th_sport; 1616 1617 dst.sin.sin_len = sizeof(struct sockaddr_in); 1618 dst.sin.sin_family = AF_INET; 1619 dst.sin.sin_addr = ip->ip_dst; 1620 dst.sin.sin_port = th->th_dport; 1621 break; 1622 #endif 1623 #ifdef INET6 1624 case AF_INET6: 1625 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1626 src.sin6.sin6_family = AF_INET6; 1627 src.sin6.sin6_addr = ip6->ip6_src; 1628 src.sin6.sin6_port = th->th_sport; 1629 1630 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1631 dst.sin6.sin6_family = AF_INET6; 1632 dst.sin6.sin6_addr = ip6->ip6_dst; 1633 dst.sin6.sin6_port = th->th_dport; 1634 break; 1635 #endif 1636 } 1637 1638 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1639 if (tiflags & TH_RST) { 1640 syn_cache_reset(&src.sa, &dst.sa, th); 1641 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1642 (TH_ACK|TH_SYN)) { 1643 /* 1644 * Received a SYN,ACK. This should never 1645 * happen while we are in LISTEN. Send an RST. 1646 */ 1647 goto badsyn; 1648 } else if (tiflags & TH_ACK) { 1649 so = syn_cache_get(&src.sa, &dst.sa, 1650 th, toff, tlen, so, m); 1651 if (so == NULL) { 1652 /* 1653 * We don't have a SYN for this ACK; 1654 * send an RST. 1655 */ 1656 goto badsyn; 1657 } else if (so == (struct socket *)(-1)) { 1658 /* 1659 * We were unable to create the 1660 * connection. If the 3-way handshake 1661 * was completed, and RST has been 1662 * sent to the peer. Since the mbuf 1663 * might be in use for the reply, do 1664 * not free it. 1665 */ 1666 m = NULL; 1667 } else { 1668 /* 1669 * We have created a full-blown 1670 * connection. 1671 */ 1672 tp = NULL; 1673 inp = NULL; 1674 #ifdef INET6 1675 in6p = NULL; 1676 #endif 1677 switch (so->so_proto->pr_domain->dom_family) { 1678 #ifdef INET 1679 case AF_INET: 1680 inp = sotoinpcb(so); 1681 tp = intotcpcb(inp); 1682 break; 1683 #endif 1684 #ifdef INET6 1685 case AF_INET6: 1686 in6p = sotoin6pcb(so); 1687 tp = in6totcpcb(in6p); 1688 break; 1689 #endif 1690 } 1691 if (tp == NULL) 1692 goto badsyn; /*XXX*/ 1693 tiwin <<= tp->snd_scale; 1694 goto after_listen; 1695 } 1696 } else { 1697 /* 1698 * None of RST, SYN or ACK was set. 1699 * This is an invalid packet for a 1700 * TCB in LISTEN state. Send a RST. 1701 */ 1702 goto badsyn; 1703 } 1704 } else { 1705 /* 1706 * Received a SYN. 1707 */ 1708 1709 #ifdef INET6 1710 /* 1711 * If deprecated address is forbidden, we do 1712 * not accept SYN to deprecated interface 1713 * address to prevent any new inbound 1714 * connection from getting established. 1715 * When we do not accept SYN, we send a TCP 1716 * RST, with deprecated source address (instead 1717 * of dropping it). We compromise it as it is 1718 * much better for peer to send a RST, and 1719 * RST will be the final packet for the 1720 * exchange. 1721 * 1722 * If we do not forbid deprecated addresses, we 1723 * accept the SYN packet. RFC2462 does not 1724 * suggest dropping SYN in this case. 1725 * If we decipher RFC2462 5.5.4, it says like 1726 * this: 1727 * 1. use of deprecated addr with existing 1728 * communication is okay - "SHOULD continue 1729 * to be used" 1730 * 2. use of it with new communication: 1731 * (2a) "SHOULD NOT be used if alternate 1732 * address with sufficient scope is 1733 * available" 1734 * (2b) nothing mentioned otherwise. 1735 * Here we fall into (2b) case as we have no 1736 * choice in our source address selection - we 1737 * must obey the peer. 1738 * 1739 * The wording in RFC2462 is confusing, and 1740 * there are multiple description text for 1741 * deprecated address handling - worse, they 1742 * are not exactly the same. I believe 5.5.4 1743 * is the best one, so we follow 5.5.4. 1744 */ 1745 if (af == AF_INET6 && !ip6_use_deprecated) { 1746 struct in6_ifaddr *ia6; 1747 int s; 1748 struct ifnet *rcvif = m_get_rcvif(m, &s); 1749 if (rcvif == NULL) 1750 goto dropwithreset; /* XXX */ 1751 if ((ia6 = in6ifa_ifpwithaddr(rcvif, 1752 &ip6->ip6_dst)) && 1753 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1754 tp = NULL; 1755 m_put_rcvif(rcvif, &s); 1756 goto dropwithreset; 1757 } 1758 m_put_rcvif(rcvif, &s); 1759 } 1760 #endif 1761 1762 /* 1763 * LISTEN socket received a SYN from itself? This 1764 * can't possibly be valid; drop the packet. 1765 */ 1766 if (th->th_sport == th->th_dport) { 1767 int eq = 0; 1768 1769 switch (af) { 1770 #ifdef INET 1771 case AF_INET: 1772 eq = in_hosteq(ip->ip_src, ip->ip_dst); 1773 break; 1774 #endif 1775 #ifdef INET6 1776 case AF_INET6: 1777 eq = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, 1778 &ip6->ip6_dst); 1779 break; 1780 #endif 1781 } 1782 if (eq) { 1783 TCP_STATINC(TCP_STAT_BADSYN); 1784 goto drop; 1785 } 1786 } 1787 1788 /* 1789 * SYN looks ok; create compressed TCP 1790 * state for it. 1791 */ 1792 if (so->so_qlen <= so->so_qlimit && 1793 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1794 so, m, optp, optlen, &opti)) 1795 m = NULL; 1796 } 1797 1798 goto drop; 1799 } 1800 1801 after_listen: 1802 /* 1803 * From here on, we're dealing with !LISTEN. 1804 */ 1805 KASSERT(tp->t_state != TCPS_LISTEN); 1806 1807 /* 1808 * Segment received on connection. 1809 * Reset idle time and keep-alive timer. 1810 */ 1811 tp->t_rcvtime = tcp_now; 1812 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1813 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1814 1815 /* 1816 * Process options. 1817 */ 1818 #ifdef TCP_SIGNATURE 1819 if (optp || (tp->t_flags & TF_SIGNATURE)) 1820 #else 1821 if (optp) 1822 #endif 1823 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1824 goto drop; 1825 1826 if (TCP_SACK_ENABLED(tp)) { 1827 tcp_del_sackholes(tp, th); 1828 } 1829 1830 if (TCP_ECN_ALLOWED(tp)) { 1831 if (tiflags & TH_CWR) { 1832 tp->t_flags &= ~TF_ECN_SND_ECE; 1833 } 1834 switch (iptos & IPTOS_ECN_MASK) { 1835 case IPTOS_ECN_CE: 1836 tp->t_flags |= TF_ECN_SND_ECE; 1837 TCP_STATINC(TCP_STAT_ECN_CE); 1838 break; 1839 case IPTOS_ECN_ECT0: 1840 TCP_STATINC(TCP_STAT_ECN_ECT); 1841 break; 1842 case IPTOS_ECN_ECT1: 1843 /* XXX: ignore for now -- rpaulo */ 1844 break; 1845 } 1846 /* 1847 * Congestion experienced. 1848 * Ignore if we are already trying to recover. 1849 */ 1850 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1851 tp->t_congctl->cong_exp(tp); 1852 } 1853 1854 if (opti.ts_present && opti.ts_ecr) { 1855 /* 1856 * Calculate the RTT from the returned time stamp and the 1857 * connection's time base. If the time stamp is later than 1858 * the current time, or is extremely old, fall back to non-1323 1859 * RTT calculation. Since ts_rtt is unsigned, we can test both 1860 * at the same time. 1861 * 1862 * Note that ts_rtt is in units of slow ticks (500 1863 * ms). Since most earthbound RTTs are < 500 ms, 1864 * observed values will have large quantization noise. 1865 * Our smoothed RTT is then the fraction of observed 1866 * samples that are 1 tick instead of 0 (times 500 1867 * ms). 1868 * 1869 * ts_rtt is increased by 1 to denote a valid sample, 1870 * with 0 indicating an invalid measurement. This 1871 * extra 1 must be removed when ts_rtt is used, or 1872 * else an an erroneous extra 500 ms will result. 1873 */ 1874 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1875 if (ts_rtt > TCP_PAWS_IDLE) 1876 ts_rtt = 0; 1877 } else { 1878 ts_rtt = 0; 1879 } 1880 1881 /* 1882 * Header prediction: check for the two common cases 1883 * of a uni-directional data xfer. If the packet has 1884 * no control flags, is in-sequence, the window didn't 1885 * change and we're not retransmitting, it's a 1886 * candidate. If the length is zero and the ack moved 1887 * forward, we're the sender side of the xfer. Just 1888 * free the data acked & wake any higher level process 1889 * that was blocked waiting for space. If the length 1890 * is non-zero and the ack didn't move, we're the 1891 * receiver side. If we're getting packets in-order 1892 * (the reassembly queue is empty), add the data to 1893 * the socket buffer and note that we need a delayed ack. 1894 */ 1895 if (tp->t_state == TCPS_ESTABLISHED && 1896 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1897 == TH_ACK && 1898 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1899 th->th_seq == tp->rcv_nxt && 1900 tiwin && tiwin == tp->snd_wnd && 1901 tp->snd_nxt == tp->snd_max) { 1902 1903 /* 1904 * If last ACK falls within this segment's sequence numbers, 1905 * record the timestamp. 1906 * NOTE that the test is modified according to the latest 1907 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 1908 * 1909 * note that we already know 1910 * TSTMP_GEQ(opti.ts_val, tp->ts_recent) 1911 */ 1912 if (opti.ts_present && 1913 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1914 tp->ts_recent_age = tcp_now; 1915 tp->ts_recent = opti.ts_val; 1916 } 1917 1918 if (tlen == 0) { 1919 /* Ack prediction. */ 1920 if (SEQ_GT(th->th_ack, tp->snd_una) && 1921 SEQ_LEQ(th->th_ack, tp->snd_max) && 1922 tp->snd_cwnd >= tp->snd_wnd && 1923 tp->t_partialacks < 0) { 1924 /* 1925 * this is a pure ack for outstanding data. 1926 */ 1927 if (ts_rtt) 1928 tcp_xmit_timer(tp, ts_rtt - 1); 1929 else if (tp->t_rtttime && 1930 SEQ_GT(th->th_ack, tp->t_rtseq)) 1931 tcp_xmit_timer(tp, 1932 tcp_now - tp->t_rtttime); 1933 acked = th->th_ack - tp->snd_una; 1934 tcps = TCP_STAT_GETREF(); 1935 tcps[TCP_STAT_PREDACK]++; 1936 tcps[TCP_STAT_RCVACKPACK]++; 1937 tcps[TCP_STAT_RCVACKBYTE] += acked; 1938 TCP_STAT_PUTREF(); 1939 nd6_hint(tp); 1940 1941 if (acked > (tp->t_lastoff - tp->t_inoff)) 1942 tp->t_lastm = NULL; 1943 sbdrop(&so->so_snd, acked); 1944 tp->t_lastoff -= acked; 1945 1946 icmp_check(tp, th, acked); 1947 1948 tp->snd_una = th->th_ack; 1949 tp->snd_fack = tp->snd_una; 1950 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1951 tp->snd_high = tp->snd_una; 1952 m_freem(m); 1953 1954 /* 1955 * If all outstanding data are acked, stop 1956 * retransmit timer, otherwise restart timer 1957 * using current (possibly backed-off) value. 1958 * If process is waiting for space, 1959 * wakeup/selnotify/signal. If data 1960 * are ready to send, let tcp_output 1961 * decide between more output or persist. 1962 */ 1963 if (tp->snd_una == tp->snd_max) 1964 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1965 else if (TCP_TIMER_ISARMED(tp, 1966 TCPT_PERSIST) == 0) 1967 TCP_TIMER_ARM(tp, TCPT_REXMT, 1968 tp->t_rxtcur); 1969 1970 sowwakeup(so); 1971 if (so->so_snd.sb_cc) { 1972 KERNEL_LOCK(1, NULL); 1973 (void) tcp_output(tp); 1974 KERNEL_UNLOCK_ONE(NULL); 1975 } 1976 if (tcp_saveti) 1977 m_freem(tcp_saveti); 1978 return; 1979 } 1980 } else if (th->th_ack == tp->snd_una && 1981 TAILQ_FIRST(&tp->segq) == NULL && 1982 tlen <= sbspace(&so->so_rcv)) { 1983 int newsize = 0; /* automatic sockbuf scaling */ 1984 1985 /* 1986 * this is a pure, in-sequence data packet 1987 * with nothing on the reassembly queue and 1988 * we have enough buffer space to take it. 1989 */ 1990 tp->rcv_nxt += tlen; 1991 tcps = TCP_STAT_GETREF(); 1992 tcps[TCP_STAT_PREDDAT]++; 1993 tcps[TCP_STAT_RCVPACK]++; 1994 tcps[TCP_STAT_RCVBYTE] += tlen; 1995 TCP_STAT_PUTREF(); 1996 nd6_hint(tp); 1997 1998 /* 1999 * Automatic sizing enables the performance of large buffers 2000 * and most of the efficiency of small ones by only allocating 2001 * space when it is needed. 2002 * 2003 * On the receive side the socket buffer memory is only rarely 2004 * used to any significant extent. This allows us to be much 2005 * more aggressive in scaling the receive socket buffer. For 2006 * the case that the buffer space is actually used to a large 2007 * extent and we run out of kernel memory we can simply drop 2008 * the new segments; TCP on the sender will just retransmit it 2009 * later. Setting the buffer size too big may only consume too 2010 * much kernel memory if the application doesn't read() from 2011 * the socket or packet loss or reordering makes use of the 2012 * reassembly queue. 2013 * 2014 * The criteria to step up the receive buffer one notch are: 2015 * 1. the number of bytes received during the time it takes 2016 * one timestamp to be reflected back to us (the RTT); 2017 * 2. received bytes per RTT is within seven eighth of the 2018 * current socket buffer size; 2019 * 3. receive buffer size has not hit maximal automatic size; 2020 * 2021 * This algorithm does one step per RTT at most and only if 2022 * we receive a bulk stream w/o packet losses or reorderings. 2023 * Shrinking the buffer during idle times is not necessary as 2024 * it doesn't consume any memory when idle. 2025 * 2026 * TODO: Only step up if the application is actually serving 2027 * the buffer to better manage the socket buffer resources. 2028 */ 2029 if (tcp_do_autorcvbuf && 2030 opti.ts_ecr && 2031 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 2032 if (opti.ts_ecr > tp->rfbuf_ts && 2033 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 2034 if (tp->rfbuf_cnt > 2035 (so->so_rcv.sb_hiwat / 8 * 7) && 2036 so->so_rcv.sb_hiwat < 2037 tcp_autorcvbuf_max) { 2038 newsize = 2039 min(so->so_rcv.sb_hiwat + 2040 tcp_autorcvbuf_inc, 2041 tcp_autorcvbuf_max); 2042 } 2043 /* Start over with next RTT. */ 2044 tp->rfbuf_ts = 0; 2045 tp->rfbuf_cnt = 0; 2046 } else 2047 tp->rfbuf_cnt += tlen; /* add up */ 2048 } 2049 2050 /* 2051 * Drop TCP, IP headers and TCP options then add data 2052 * to socket buffer. 2053 */ 2054 if (so->so_state & SS_CANTRCVMORE) 2055 m_freem(m); 2056 else { 2057 /* 2058 * Set new socket buffer size. 2059 * Give up when limit is reached. 2060 */ 2061 if (newsize) 2062 if (!sbreserve(&so->so_rcv, 2063 newsize, so)) 2064 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 2065 m_adj(m, toff + off); 2066 sbappendstream(&so->so_rcv, m); 2067 } 2068 sorwakeup(so); 2069 tcp_setup_ack(tp, th); 2070 if (tp->t_flags & TF_ACKNOW) { 2071 KERNEL_LOCK(1, NULL); 2072 (void) tcp_output(tp); 2073 KERNEL_UNLOCK_ONE(NULL); 2074 } 2075 if (tcp_saveti) 2076 m_freem(tcp_saveti); 2077 return; 2078 } 2079 } 2080 2081 /* 2082 * Compute mbuf offset to TCP data segment. 2083 */ 2084 hdroptlen = toff + off; 2085 2086 /* 2087 * Calculate amount of space in receive window. Receive window is 2088 * amount of space in rcv queue, but not less than advertised 2089 * window. 2090 */ 2091 { 2092 int win; 2093 win = sbspace(&so->so_rcv); 2094 if (win < 0) 2095 win = 0; 2096 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 2097 } 2098 2099 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 2100 tp->rfbuf_ts = 0; 2101 tp->rfbuf_cnt = 0; 2102 2103 switch (tp->t_state) { 2104 /* 2105 * If the state is SYN_SENT: 2106 * if seg contains an ACK, but not for our SYN, drop the input. 2107 * if seg contains a RST, then drop the connection. 2108 * if seg does not contain SYN, then drop it. 2109 * Otherwise this is an acceptable SYN segment 2110 * initialize tp->rcv_nxt and tp->irs 2111 * if seg contains ack then advance tp->snd_una 2112 * if seg contains a ECE and ECN support is enabled, the stream 2113 * is ECN capable. 2114 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2115 * arrange for segment to be acked (eventually) 2116 * continue processing rest of data/controls, beginning with URG 2117 */ 2118 case TCPS_SYN_SENT: 2119 if ((tiflags & TH_ACK) && 2120 (SEQ_LEQ(th->th_ack, tp->iss) || 2121 SEQ_GT(th->th_ack, tp->snd_max))) 2122 goto dropwithreset; 2123 if (tiflags & TH_RST) { 2124 if (tiflags & TH_ACK) 2125 tp = tcp_drop(tp, ECONNREFUSED); 2126 goto drop; 2127 } 2128 if ((tiflags & TH_SYN) == 0) 2129 goto drop; 2130 if (tiflags & TH_ACK) { 2131 tp->snd_una = th->th_ack; 2132 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2133 tp->snd_nxt = tp->snd_una; 2134 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2135 tp->snd_high = tp->snd_una; 2136 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2137 2138 if ((tiflags & TH_ECE) && tcp_do_ecn) { 2139 tp->t_flags |= TF_ECN_PERMIT; 2140 TCP_STATINC(TCP_STAT_ECN_SHS); 2141 } 2142 2143 } 2144 tp->irs = th->th_seq; 2145 tcp_rcvseqinit(tp); 2146 tp->t_flags |= TF_ACKNOW; 2147 tcp_mss_from_peer(tp, opti.maxseg); 2148 2149 /* 2150 * Initialize the initial congestion window. If we 2151 * had to retransmit the SYN, we must initialize cwnd 2152 * to 1 segment (i.e. the Loss Window). 2153 */ 2154 if (tp->t_flags & TF_SYN_REXMT) 2155 tp->snd_cwnd = tp->t_peermss; 2156 else { 2157 int ss = tcp_init_win; 2158 #ifdef INET 2159 if (inp != NULL && in_localaddr(inp->inp_faddr)) 2160 ss = tcp_init_win_local; 2161 #endif 2162 #ifdef INET6 2163 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 2164 ss = tcp_init_win_local; 2165 #endif 2166 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 2167 } 2168 2169 tcp_rmx_rtt(tp); 2170 if (tiflags & TH_ACK) { 2171 TCP_STATINC(TCP_STAT_CONNECTS); 2172 /* 2173 * move tcp_established before soisconnected 2174 * because upcall handler can drive tcp_output 2175 * functionality. 2176 * XXX we might call soisconnected at the end of 2177 * all processing 2178 */ 2179 tcp_established(tp); 2180 soisconnected(so); 2181 /* Do window scaling on this connection? */ 2182 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2183 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2184 tp->snd_scale = tp->requested_s_scale; 2185 tp->rcv_scale = tp->request_r_scale; 2186 } 2187 TCP_REASS_LOCK(tp); 2188 (void)tcp_reass(tp, NULL, NULL, tlen); 2189 /* 2190 * if we didn't have to retransmit the SYN, 2191 * use its rtt as our initial srtt & rtt var. 2192 */ 2193 if (tp->t_rtttime) 2194 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2195 } else 2196 tp->t_state = TCPS_SYN_RECEIVED; 2197 2198 /* 2199 * Advance th->th_seq to correspond to first data byte. 2200 * If data, trim to stay within window, 2201 * dropping FIN if necessary. 2202 */ 2203 th->th_seq++; 2204 if (tlen > tp->rcv_wnd) { 2205 todrop = tlen - tp->rcv_wnd; 2206 m_adj(m, -todrop); 2207 tlen = tp->rcv_wnd; 2208 tiflags &= ~TH_FIN; 2209 tcps = TCP_STAT_GETREF(); 2210 tcps[TCP_STAT_RCVPACKAFTERWIN]++; 2211 tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop; 2212 TCP_STAT_PUTREF(); 2213 } 2214 tp->snd_wl1 = th->th_seq - 1; 2215 tp->rcv_up = th->th_seq; 2216 goto step6; 2217 2218 /* 2219 * If the state is SYN_RECEIVED: 2220 * If seg contains an ACK, but not for our SYN, drop the input 2221 * and generate an RST. See page 36, rfc793 2222 */ 2223 case TCPS_SYN_RECEIVED: 2224 if ((tiflags & TH_ACK) && 2225 (SEQ_LEQ(th->th_ack, tp->iss) || 2226 SEQ_GT(th->th_ack, tp->snd_max))) 2227 goto dropwithreset; 2228 break; 2229 } 2230 2231 /* 2232 * From here on, we're dealing with !LISTEN and !SYN_SENT. 2233 */ 2234 KASSERT(tp->t_state != TCPS_LISTEN && 2235 tp->t_state != TCPS_SYN_SENT); 2236 2237 /* 2238 * First check timestamp, if present. 2239 * Then check that at least some bytes of segment are within 2240 * receive window. If segment begins before rcv_nxt, 2241 * drop leading data (and SYN); if nothing left, just ack. 2242 * 2243 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2244 * and it's less than ts_recent, drop it. 2245 */ 2246 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2247 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2248 /* Check to see if ts_recent is over 24 days old. */ 2249 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2250 /* 2251 * Invalidate ts_recent. If this segment updates 2252 * ts_recent, the age will be reset later and ts_recent 2253 * will get a valid value. If it does not, setting 2254 * ts_recent to zero will at least satisfy the 2255 * requirement that zero be placed in the timestamp 2256 * echo reply when ts_recent isn't valid. The 2257 * age isn't reset until we get a valid ts_recent 2258 * because we don't want out-of-order segments to be 2259 * dropped when ts_recent is old. 2260 */ 2261 tp->ts_recent = 0; 2262 } else { 2263 tcps = TCP_STAT_GETREF(); 2264 tcps[TCP_STAT_RCVDUPPACK]++; 2265 tcps[TCP_STAT_RCVDUPBYTE] += tlen; 2266 tcps[TCP_STAT_PAWSDROP]++; 2267 TCP_STAT_PUTREF(); 2268 tcp_new_dsack(tp, th->th_seq, tlen); 2269 goto dropafterack; 2270 } 2271 } 2272 2273 todrop = tp->rcv_nxt - th->th_seq; 2274 dupseg = false; 2275 if (todrop > 0) { 2276 if (tiflags & TH_SYN) { 2277 tiflags &= ~TH_SYN; 2278 th->th_seq++; 2279 if (th->th_urp > 1) 2280 th->th_urp--; 2281 else { 2282 tiflags &= ~TH_URG; 2283 th->th_urp = 0; 2284 } 2285 todrop--; 2286 } 2287 if (todrop > tlen || 2288 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2289 /* 2290 * Any valid FIN or RST must be to the left of the 2291 * window. At this point the FIN or RST must be a 2292 * duplicate or out of sequence; drop it. 2293 */ 2294 if (tiflags & TH_RST) 2295 goto drop; 2296 tiflags &= ~(TH_FIN|TH_RST); 2297 /* 2298 * Send an ACK to resynchronize and drop any data. 2299 * But keep on processing for RST or ACK. 2300 */ 2301 tp->t_flags |= TF_ACKNOW; 2302 todrop = tlen; 2303 dupseg = true; 2304 tcps = TCP_STAT_GETREF(); 2305 tcps[TCP_STAT_RCVDUPPACK]++; 2306 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 2307 TCP_STAT_PUTREF(); 2308 } else if ((tiflags & TH_RST) && 2309 th->th_seq != tp->rcv_nxt) { 2310 /* 2311 * Test for reset before adjusting the sequence 2312 * number for overlapping data. 2313 */ 2314 goto dropafterack_ratelim; 2315 } else { 2316 tcps = TCP_STAT_GETREF(); 2317 tcps[TCP_STAT_RCVPARTDUPPACK]++; 2318 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 2319 TCP_STAT_PUTREF(); 2320 } 2321 tcp_new_dsack(tp, th->th_seq, todrop); 2322 hdroptlen += todrop; /*drop from head afterwards*/ 2323 th->th_seq += todrop; 2324 tlen -= todrop; 2325 if (th->th_urp > todrop) 2326 th->th_urp -= todrop; 2327 else { 2328 tiflags &= ~TH_URG; 2329 th->th_urp = 0; 2330 } 2331 } 2332 2333 /* 2334 * If new data are received on a connection after the 2335 * user processes are gone, then RST the other end. 2336 */ 2337 if ((so->so_state & SS_NOFDREF) && 2338 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2339 tp = tcp_close(tp); 2340 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 2341 goto dropwithreset; 2342 } 2343 2344 /* 2345 * If segment ends after window, drop trailing data 2346 * (and PUSH and FIN); if nothing left, just ACK. 2347 */ 2348 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2349 if (todrop > 0) { 2350 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 2351 if (todrop >= tlen) { 2352 /* 2353 * The segment actually starts after the window. 2354 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2355 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2356 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2357 */ 2358 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 2359 2360 /* 2361 * If a new connection request is received while in 2362 * TIME_WAIT, drop the old connection and start over 2363 * if the sequence numbers are above the previous 2364 * ones. 2365 * 2366 * NOTE: We need to put the header fields back into 2367 * network order. 2368 * 2369 * XXX This kind of sucks, but we don't expect 2370 * XXX this to happen very often, so maybe it 2371 * XXX doesn't matter so much. 2372 */ 2373 if ((tiflags & TH_SYN) && 2374 tp->t_state == TCPS_TIME_WAIT && 2375 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2376 tp = tcp_close(tp); 2377 tcp_fields_to_net(th); 2378 goto findpcb; 2379 } 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 } 2394 } else { 2395 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 2396 } 2397 m_adj(m, -todrop); 2398 tlen -= todrop; 2399 tiflags &= ~(TH_PUSH|TH_FIN); 2400 } 2401 2402 /* 2403 * If last ACK falls within this segment's sequence numbers, 2404 * record the timestamp. 2405 * NOTE: 2406 * 1) That the test incorporates suggestions from the latest 2407 * proposal of the tcplw (at) cray.com list (Braden 1993/04/26). 2408 * 2) That updating only on newer timestamps interferes with 2409 * our earlier PAWS tests, so this check should be solely 2410 * predicated on the sequence space of this segment. 2411 * 3) That we modify the segment boundary check to be 2412 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2413 * instead of RFC1323's 2414 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2415 * This modified check allows us to overcome RFC1323's 2416 * limitations as described in Stevens TCP/IP Illustrated 2417 * Vol. 2 p.869. In such cases, we can still calculate the 2418 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2419 */ 2420 if (opti.ts_present && 2421 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2422 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2423 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2424 tp->ts_recent_age = tcp_now; 2425 tp->ts_recent = opti.ts_val; 2426 } 2427 2428 /* 2429 * If the RST bit is set examine the state: 2430 * RECEIVED state: 2431 * If passive open, return to LISTEN state. 2432 * If active open, inform user that connection was refused. 2433 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT states: 2434 * Inform user that connection was reset, and close tcb. 2435 * CLOSING, LAST_ACK, TIME_WAIT states: 2436 * Close the tcb. 2437 */ 2438 if (tiflags & TH_RST) { 2439 if (th->th_seq != tp->rcv_nxt) 2440 goto dropafterack_ratelim; 2441 2442 switch (tp->t_state) { 2443 case TCPS_SYN_RECEIVED: 2444 so->so_error = ECONNREFUSED; 2445 goto close; 2446 2447 case TCPS_ESTABLISHED: 2448 case TCPS_FIN_WAIT_1: 2449 case TCPS_FIN_WAIT_2: 2450 case TCPS_CLOSE_WAIT: 2451 so->so_error = ECONNRESET; 2452 close: 2453 tp->t_state = TCPS_CLOSED; 2454 TCP_STATINC(TCP_STAT_DROPS); 2455 tp = tcp_close(tp); 2456 goto drop; 2457 2458 case TCPS_CLOSING: 2459 case TCPS_LAST_ACK: 2460 case TCPS_TIME_WAIT: 2461 tp = tcp_close(tp); 2462 goto drop; 2463 } 2464 } 2465 2466 /* 2467 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2468 * we must be in a synchronized state. RFC791 states (under RST 2469 * generation) that any unacceptable segment (an out-of-order SYN 2470 * qualifies) received in a synchronized state must elicit only an 2471 * empty acknowledgment segment ... and the connection remains in 2472 * the same state. 2473 */ 2474 if (tiflags & TH_SYN) { 2475 if (tp->rcv_nxt == th->th_seq) { 2476 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2477 TH_ACK); 2478 if (tcp_saveti) 2479 m_freem(tcp_saveti); 2480 return; 2481 } 2482 2483 goto dropafterack_ratelim; 2484 } 2485 2486 /* 2487 * If the ACK bit is off we drop the segment and return. 2488 */ 2489 if ((tiflags & TH_ACK) == 0) { 2490 if (tp->t_flags & TF_ACKNOW) 2491 goto dropafterack; 2492 else 2493 goto drop; 2494 } 2495 2496 /* 2497 * Ack processing. 2498 */ 2499 switch (tp->t_state) { 2500 2501 /* 2502 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2503 * ESTABLISHED state and continue processing, otherwise 2504 * send an RST. 2505 */ 2506 case TCPS_SYN_RECEIVED: 2507 if (SEQ_GT(tp->snd_una, th->th_ack) || 2508 SEQ_GT(th->th_ack, tp->snd_max)) 2509 goto dropwithreset; 2510 TCP_STATINC(TCP_STAT_CONNECTS); 2511 soisconnected(so); 2512 tcp_established(tp); 2513 /* Do window scaling? */ 2514 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2515 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2516 tp->snd_scale = tp->requested_s_scale; 2517 tp->rcv_scale = tp->request_r_scale; 2518 } 2519 TCP_REASS_LOCK(tp); 2520 (void)tcp_reass(tp, NULL, NULL, tlen); 2521 tp->snd_wl1 = th->th_seq - 1; 2522 /* fall into ... */ 2523 2524 /* 2525 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2526 * ACKs. If the ack is in the range 2527 * tp->snd_una < th->th_ack <= tp->snd_max 2528 * then advance tp->snd_una to th->th_ack and drop 2529 * data from the retransmission queue. If this ACK reflects 2530 * more up to date window information we update our window information. 2531 */ 2532 case TCPS_ESTABLISHED: 2533 case TCPS_FIN_WAIT_1: 2534 case TCPS_FIN_WAIT_2: 2535 case TCPS_CLOSE_WAIT: 2536 case TCPS_CLOSING: 2537 case TCPS_LAST_ACK: 2538 case TCPS_TIME_WAIT: 2539 2540 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2541 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2542 TCP_STATINC(TCP_STAT_RCVDUPACK); 2543 /* 2544 * If we have outstanding data (other than 2545 * a window probe), this is a completely 2546 * duplicate ack (ie, window info didn't 2547 * change), the ack is the biggest we've 2548 * seen and we've seen exactly our rexmt 2549 * threshhold of them, assume a packet 2550 * has been dropped and retransmit it. 2551 * Kludge snd_nxt & the congestion 2552 * window so we send only this one 2553 * packet. 2554 */ 2555 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2556 th->th_ack != tp->snd_una) 2557 tp->t_dupacks = 0; 2558 else if (tp->t_partialacks < 0 && 2559 (++tp->t_dupacks == tcprexmtthresh || 2560 TCP_FACK_FASTRECOV(tp))) { 2561 /* 2562 * Do the fast retransmit, and adjust 2563 * congestion control paramenters. 2564 */ 2565 if (tp->t_congctl->fast_retransmit(tp, th)) { 2566 /* False fast retransmit */ 2567 break; 2568 } else 2569 goto drop; 2570 } else if (tp->t_dupacks > tcprexmtthresh) { 2571 tp->snd_cwnd += tp->t_segsz; 2572 KERNEL_LOCK(1, NULL); 2573 (void) tcp_output(tp); 2574 KERNEL_UNLOCK_ONE(NULL); 2575 goto drop; 2576 } 2577 } else { 2578 /* 2579 * If the ack appears to be very old, only 2580 * allow data that is in-sequence. This 2581 * makes it somewhat more difficult to insert 2582 * forged data by guessing sequence numbers. 2583 * Sent an ack to try to update the send 2584 * sequence number on the other side. 2585 */ 2586 if (tlen && th->th_seq != tp->rcv_nxt && 2587 SEQ_LT(th->th_ack, 2588 tp->snd_una - tp->max_sndwnd)) 2589 goto dropafterack; 2590 } 2591 break; 2592 } 2593 /* 2594 * If the congestion window was inflated to account 2595 * for the other side's cached packets, retract it. 2596 */ 2597 tp->t_congctl->fast_retransmit_newack(tp, th); 2598 2599 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2600 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2601 goto dropafterack; 2602 } 2603 acked = th->th_ack - tp->snd_una; 2604 tcps = TCP_STAT_GETREF(); 2605 tcps[TCP_STAT_RCVACKPACK]++; 2606 tcps[TCP_STAT_RCVACKBYTE] += acked; 2607 TCP_STAT_PUTREF(); 2608 2609 /* 2610 * If we have a timestamp reply, update smoothed 2611 * round trip time. If no timestamp is present but 2612 * transmit timer is running and timed sequence 2613 * number was acked, update smoothed round trip time. 2614 * Since we now have an rtt measurement, cancel the 2615 * timer backoff (cf., Phil Karn's retransmit alg.). 2616 * Recompute the initial retransmit timer. 2617 */ 2618 if (ts_rtt) 2619 tcp_xmit_timer(tp, ts_rtt - 1); 2620 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2621 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2622 2623 /* 2624 * If all outstanding data is acked, stop retransmit 2625 * timer and remember to restart (more output or persist). 2626 * If there is more data to be acked, restart retransmit 2627 * timer, using current (possibly backed-off) value. 2628 */ 2629 if (th->th_ack == tp->snd_max) { 2630 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2631 needoutput = 1; 2632 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2633 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2634 2635 /* 2636 * New data has been acked, adjust the congestion window. 2637 */ 2638 tp->t_congctl->newack(tp, th); 2639 2640 nd6_hint(tp); 2641 if (acked > so->so_snd.sb_cc) { 2642 tp->snd_wnd -= so->so_snd.sb_cc; 2643 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2644 ourfinisacked = 1; 2645 } else { 2646 if (acked > (tp->t_lastoff - tp->t_inoff)) 2647 tp->t_lastm = NULL; 2648 sbdrop(&so->so_snd, acked); 2649 tp->t_lastoff -= acked; 2650 if (tp->snd_wnd > acked) 2651 tp->snd_wnd -= acked; 2652 else 2653 tp->snd_wnd = 0; 2654 ourfinisacked = 0; 2655 } 2656 sowwakeup(so); 2657 2658 icmp_check(tp, th, acked); 2659 2660 tp->snd_una = th->th_ack; 2661 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2662 tp->snd_fack = tp->snd_una; 2663 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2664 tp->snd_nxt = tp->snd_una; 2665 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2666 tp->snd_high = tp->snd_una; 2667 2668 switch (tp->t_state) { 2669 2670 /* 2671 * In FIN_WAIT_1 STATE in addition to the processing 2672 * for the ESTABLISHED state if our FIN is now acknowledged 2673 * then enter FIN_WAIT_2. 2674 */ 2675 case TCPS_FIN_WAIT_1: 2676 if (ourfinisacked) { 2677 /* 2678 * If we can't receive any more 2679 * data, then closing user can proceed. 2680 * Starting the timer is contrary to the 2681 * specification, but if we don't get a FIN 2682 * we'll hang forever. 2683 */ 2684 if (so->so_state & SS_CANTRCVMORE) { 2685 soisdisconnected(so); 2686 if (tp->t_maxidle > 0) 2687 TCP_TIMER_ARM(tp, TCPT_2MSL, 2688 tp->t_maxidle); 2689 } 2690 tp->t_state = TCPS_FIN_WAIT_2; 2691 } 2692 break; 2693 2694 /* 2695 * In CLOSING STATE in addition to the processing for 2696 * the ESTABLISHED state if the ACK acknowledges our FIN 2697 * then enter the TIME-WAIT state, otherwise ignore 2698 * the segment. 2699 */ 2700 case TCPS_CLOSING: 2701 if (ourfinisacked) { 2702 tp->t_state = TCPS_TIME_WAIT; 2703 tcp_canceltimers(tp); 2704 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2705 soisdisconnected(so); 2706 } 2707 break; 2708 2709 /* 2710 * In LAST_ACK, we may still be waiting for data to drain 2711 * and/or to be acked, as well as for the ack of our FIN. 2712 * If our FIN is now acknowledged, delete the TCB, 2713 * enter the closed state and return. 2714 */ 2715 case TCPS_LAST_ACK: 2716 if (ourfinisacked) { 2717 tp = tcp_close(tp); 2718 goto drop; 2719 } 2720 break; 2721 2722 /* 2723 * In TIME_WAIT state the only thing that should arrive 2724 * is a retransmission of the remote FIN. Acknowledge 2725 * it and restart the finack timer. 2726 */ 2727 case TCPS_TIME_WAIT: 2728 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2729 goto dropafterack; 2730 } 2731 } 2732 2733 step6: 2734 /* 2735 * Update window information. 2736 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2737 */ 2738 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2739 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2740 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2741 /* keep track of pure window updates */ 2742 if (tlen == 0 && 2743 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2744 TCP_STATINC(TCP_STAT_RCVWINUPD); 2745 tp->snd_wnd = tiwin; 2746 tp->snd_wl1 = th->th_seq; 2747 tp->snd_wl2 = th->th_ack; 2748 if (tp->snd_wnd > tp->max_sndwnd) 2749 tp->max_sndwnd = tp->snd_wnd; 2750 needoutput = 1; 2751 } 2752 2753 /* 2754 * Process segments with URG. 2755 */ 2756 if ((tiflags & TH_URG) && th->th_urp && 2757 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2758 /* 2759 * This is a kludge, but if we receive and accept 2760 * random urgent pointers, we'll crash in 2761 * soreceive. It's hard to imagine someone 2762 * actually wanting to send this much urgent data. 2763 */ 2764 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2765 th->th_urp = 0; /* XXX */ 2766 tiflags &= ~TH_URG; /* XXX */ 2767 goto dodata; /* XXX */ 2768 } 2769 /* 2770 * If this segment advances the known urgent pointer, 2771 * then mark the data stream. This should not happen 2772 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2773 * a FIN has been received from the remote side. 2774 * In these states we ignore the URG. 2775 * 2776 * According to RFC961 (Assigned Protocols), 2777 * the urgent pointer points to the last octet 2778 * of urgent data. We continue, however, 2779 * to consider it to indicate the first octet 2780 * of data past the urgent section as the original 2781 * spec states (in one of two places). 2782 */ 2783 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2784 tp->rcv_up = th->th_seq + th->th_urp; 2785 so->so_oobmark = so->so_rcv.sb_cc + 2786 (tp->rcv_up - tp->rcv_nxt) - 1; 2787 if (so->so_oobmark == 0) 2788 so->so_state |= SS_RCVATMARK; 2789 sohasoutofband(so); 2790 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2791 } 2792 /* 2793 * Remove out of band data so doesn't get presented to user. 2794 * This can happen independent of advancing the URG pointer, 2795 * but if two URG's are pending at once, some out-of-band 2796 * data may creep in... ick. 2797 */ 2798 if (th->th_urp <= (u_int16_t)tlen && 2799 (so->so_options & SO_OOBINLINE) == 0) 2800 tcp_pulloutofband(so, th, m, hdroptlen); 2801 } else 2802 /* 2803 * If no out of band data is expected, 2804 * pull receive urgent pointer along 2805 * with the receive window. 2806 */ 2807 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2808 tp->rcv_up = tp->rcv_nxt; 2809 dodata: /* XXX */ 2810 2811 /* 2812 * Process the segment text, merging it into the TCP sequencing queue, 2813 * and arranging for acknowledgement of receipt if necessary. 2814 * This process logically involves adjusting tp->rcv_wnd as data 2815 * is presented to the user (this happens in tcp_usrreq.c, 2816 * tcp_rcvd()). If a FIN has already been received on this 2817 * connection then we just ignore the text. 2818 */ 2819 if ((tlen || (tiflags & TH_FIN)) && 2820 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2821 /* 2822 * Insert segment ti into reassembly queue of tcp with 2823 * control block tp. Return TH_FIN if reassembly now includes 2824 * a segment with FIN. The macro form does the common case 2825 * inline (segment is the next to be received on an 2826 * established connection, and the queue is empty), 2827 * avoiding linkage into and removal from the queue and 2828 * repetition of various conversions. 2829 * Set DELACK for segments received in order, but ack 2830 * immediately when segments are out of order 2831 * (so fast retransmit can work). 2832 */ 2833 /* NOTE: this was TCP_REASS() macro, but used only once */ 2834 TCP_REASS_LOCK(tp); 2835 if (th->th_seq == tp->rcv_nxt && 2836 TAILQ_FIRST(&tp->segq) == NULL && 2837 tp->t_state == TCPS_ESTABLISHED) { 2838 tcp_setup_ack(tp, th); 2839 tp->rcv_nxt += tlen; 2840 tiflags = th->th_flags & TH_FIN; 2841 tcps = TCP_STAT_GETREF(); 2842 tcps[TCP_STAT_RCVPACK]++; 2843 tcps[TCP_STAT_RCVBYTE] += tlen; 2844 TCP_STAT_PUTREF(); 2845 nd6_hint(tp); 2846 if (so->so_state & SS_CANTRCVMORE) 2847 m_freem(m); 2848 else { 2849 m_adj(m, hdroptlen); 2850 sbappendstream(&(so)->so_rcv, m); 2851 } 2852 TCP_REASS_UNLOCK(tp); 2853 sorwakeup(so); 2854 } else { 2855 m_adj(m, hdroptlen); 2856 tiflags = tcp_reass(tp, th, m, tlen); 2857 tp->t_flags |= TF_ACKNOW; 2858 } 2859 2860 /* 2861 * Note the amount of data that peer has sent into 2862 * our window, in order to estimate the sender's 2863 * buffer size. 2864 */ 2865 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2866 } else { 2867 m_freem(m); 2868 m = NULL; 2869 tiflags &= ~TH_FIN; 2870 } 2871 2872 /* 2873 * If FIN is received ACK the FIN and let the user know 2874 * that the connection is closing. Ignore a FIN received before 2875 * the connection is fully established. 2876 */ 2877 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2878 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2879 socantrcvmore(so); 2880 tp->t_flags |= TF_ACKNOW; 2881 tp->rcv_nxt++; 2882 } 2883 switch (tp->t_state) { 2884 2885 /* 2886 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2887 */ 2888 case TCPS_ESTABLISHED: 2889 tp->t_state = TCPS_CLOSE_WAIT; 2890 break; 2891 2892 /* 2893 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2894 * enter the CLOSING state. 2895 */ 2896 case TCPS_FIN_WAIT_1: 2897 tp->t_state = TCPS_CLOSING; 2898 break; 2899 2900 /* 2901 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2902 * starting the time-wait timer, turning off the other 2903 * standard timers. 2904 */ 2905 case TCPS_FIN_WAIT_2: 2906 tp->t_state = TCPS_TIME_WAIT; 2907 tcp_canceltimers(tp); 2908 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2909 soisdisconnected(so); 2910 break; 2911 2912 /* 2913 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2914 */ 2915 case TCPS_TIME_WAIT: 2916 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * tp->t_msl); 2917 break; 2918 } 2919 } 2920 #ifdef TCP_DEBUG 2921 if (so->so_options & SO_DEBUG) 2922 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2923 #endif 2924 2925 /* 2926 * Return any desired output. 2927 */ 2928 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2929 KERNEL_LOCK(1, NULL); 2930 (void) tcp_output(tp); 2931 KERNEL_UNLOCK_ONE(NULL); 2932 } 2933 if (tcp_saveti) 2934 m_freem(tcp_saveti); 2935 2936 if (tp->t_state == TCPS_TIME_WAIT 2937 && (so->so_state & SS_NOFDREF) 2938 && (tp->t_inpcb || af != AF_INET) 2939 && (tp->t_in6pcb || af != AF_INET6) 2940 && ((af == AF_INET ? tcp4_vtw_enable : tcp6_vtw_enable) & 1) != 0 2941 && TAILQ_EMPTY(&tp->segq) 2942 && vtw_add(af, tp)) { 2943 ; 2944 } 2945 return; 2946 2947 badsyn: 2948 /* 2949 * Received a bad SYN. Increment counters and dropwithreset. 2950 */ 2951 TCP_STATINC(TCP_STAT_BADSYN); 2952 tp = NULL; 2953 goto dropwithreset; 2954 2955 dropafterack: 2956 /* 2957 * Generate an ACK dropping incoming segment if it occupies 2958 * sequence space, where the ACK reflects our state. 2959 */ 2960 if (tiflags & TH_RST) 2961 goto drop; 2962 goto dropafterack2; 2963 2964 dropafterack_ratelim: 2965 /* 2966 * We may want to rate-limit ACKs against SYN/RST attack. 2967 */ 2968 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2969 tcp_ackdrop_ppslim) == 0) { 2970 /* XXX stat */ 2971 goto drop; 2972 } 2973 /* ...fall into dropafterack2... */ 2974 2975 dropafterack2: 2976 m_freem(m); 2977 tp->t_flags |= TF_ACKNOW; 2978 KERNEL_LOCK(1, NULL); 2979 (void) tcp_output(tp); 2980 KERNEL_UNLOCK_ONE(NULL); 2981 if (tcp_saveti) 2982 m_freem(tcp_saveti); 2983 return; 2984 2985 dropwithreset_ratelim: 2986 /* 2987 * We may want to rate-limit RSTs in certain situations, 2988 * particularly if we are sending an RST in response to 2989 * an attempt to connect to or otherwise communicate with 2990 * a port for which we have no socket. 2991 */ 2992 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2993 tcp_rst_ppslim) == 0) { 2994 /* XXX stat */ 2995 goto drop; 2996 } 2997 /* ...fall into dropwithreset... */ 2998 2999 dropwithreset: 3000 /* 3001 * Generate a RST, dropping incoming segment. 3002 * Make ACK acceptable to originator of segment. 3003 */ 3004 if (tiflags & TH_RST) 3005 goto drop; 3006 if (tiflags & TH_ACK) { 3007 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3008 } else { 3009 if (tiflags & TH_SYN) 3010 tlen++; 3011 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 3012 TH_RST|TH_ACK); 3013 } 3014 if (tcp_saveti) 3015 m_freem(tcp_saveti); 3016 return; 3017 3018 badcsum: 3019 drop: 3020 /* 3021 * Drop space held by incoming segment and return. 3022 */ 3023 if (tp) { 3024 if (tp->t_inpcb) 3025 so = tp->t_inpcb->inp_socket; 3026 #ifdef INET6 3027 else if (tp->t_in6pcb) 3028 so = tp->t_in6pcb->in6p_socket; 3029 #endif 3030 else 3031 so = NULL; 3032 #ifdef TCP_DEBUG 3033 if (so && (so->so_options & SO_DEBUG) != 0) 3034 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 3035 #endif 3036 } 3037 if (tcp_saveti) 3038 m_freem(tcp_saveti); 3039 m_freem(m); 3040 return; 3041 } 3042 3043 #ifdef TCP_SIGNATURE 3044 int 3045 tcp_signature_apply(void *fstate, void *data, u_int len) 3046 { 3047 3048 MD5Update(fstate, (u_char *)data, len); 3049 return (0); 3050 } 3051 3052 struct secasvar * 3053 tcp_signature_getsav(struct mbuf *m) 3054 { 3055 struct ip *ip; 3056 struct ip6_hdr *ip6; 3057 3058 ip = mtod(m, struct ip *); 3059 switch (ip->ip_v) { 3060 case 4: 3061 ip = mtod(m, struct ip *); 3062 ip6 = NULL; 3063 break; 3064 case 6: 3065 ip = NULL; 3066 ip6 = mtod(m, struct ip6_hdr *); 3067 break; 3068 default: 3069 return (NULL); 3070 } 3071 3072 #ifdef IPSEC 3073 union sockaddr_union dst; 3074 3075 /* Extract the destination from the IP header in the mbuf. */ 3076 memset(&dst, 0, sizeof(union sockaddr_union)); 3077 if (ip != NULL) { 3078 dst.sa.sa_len = sizeof(struct sockaddr_in); 3079 dst.sa.sa_family = AF_INET; 3080 dst.sin.sin_addr = ip->ip_dst; 3081 } else { 3082 dst.sa.sa_len = sizeof(struct sockaddr_in6); 3083 dst.sa.sa_family = AF_INET6; 3084 dst.sin6.sin6_addr = ip6->ip6_dst; 3085 } 3086 3087 /* 3088 * Look up an SADB entry which matches the address of the peer. 3089 */ 3090 return KEY_LOOKUP_SA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI), 0, 0); 3091 #else 3092 return NULL; 3093 #endif 3094 } 3095 3096 int 3097 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 3098 struct secasvar *sav, char *sig) 3099 { 3100 MD5_CTX ctx; 3101 struct ip *ip; 3102 struct ipovly *ipovly; 3103 #ifdef INET6 3104 struct ip6_hdr *ip6; 3105 struct ip6_hdr_pseudo ip6pseudo; 3106 #endif /* INET6 */ 3107 struct ippseudo ippseudo; 3108 struct tcphdr th0; 3109 int l, tcphdrlen; 3110 3111 if (sav == NULL) 3112 return (-1); 3113 3114 tcphdrlen = th->th_off * 4; 3115 3116 switch (mtod(m, struct ip *)->ip_v) { 3117 case 4: 3118 MD5Init(&ctx); 3119 ip = mtod(m, struct ip *); 3120 memset(&ippseudo, 0, sizeof(ippseudo)); 3121 ipovly = (struct ipovly *)ip; 3122 ippseudo.ippseudo_src = ipovly->ih_src; 3123 ippseudo.ippseudo_dst = ipovly->ih_dst; 3124 ippseudo.ippseudo_pad = 0; 3125 ippseudo.ippseudo_p = IPPROTO_TCP; 3126 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 3127 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 3128 break; 3129 #if INET6 3130 case 6: 3131 MD5Init(&ctx); 3132 ip6 = mtod(m, struct ip6_hdr *); 3133 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 3134 ip6pseudo.ip6ph_src = ip6->ip6_src; 3135 in6_clearscope(&ip6pseudo.ip6ph_src); 3136 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 3137 in6_clearscope(&ip6pseudo.ip6ph_dst); 3138 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 3139 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 3140 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 3141 break; 3142 #endif /* INET6 */ 3143 default: 3144 return (-1); 3145 } 3146 3147 th0 = *th; 3148 th0.th_sum = 0; 3149 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 3150 3151 l = m->m_pkthdr.len - thoff - tcphdrlen; 3152 if (l > 0) 3153 m_apply(m, thoff + tcphdrlen, 3154 m->m_pkthdr.len - thoff - tcphdrlen, 3155 tcp_signature_apply, &ctx); 3156 3157 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 3158 MD5Final(sig, &ctx); 3159 3160 return (0); 3161 } 3162 #endif 3163 3164 /* 3165 * tcp_dooptions: parse and process tcp options. 3166 * 3167 * returns -1 if this segment should be dropped. (eg. wrong signature) 3168 * otherwise returns 0. 3169 */ 3170 3171 static int 3172 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, struct tcphdr *th, 3173 struct mbuf *m, int toff, struct tcp_opt_info *oi) 3174 { 3175 u_int16_t mss; 3176 int opt, optlen = 0; 3177 #ifdef TCP_SIGNATURE 3178 void *sigp = NULL; 3179 char sigbuf[TCP_SIGLEN]; 3180 struct secasvar *sav = NULL; 3181 #endif 3182 3183 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 3184 opt = cp[0]; 3185 if (opt == TCPOPT_EOL) 3186 break; 3187 if (opt == TCPOPT_NOP) 3188 optlen = 1; 3189 else { 3190 if (cnt < 2) 3191 break; 3192 optlen = cp[1]; 3193 if (optlen < 2 || optlen > cnt) 3194 break; 3195 } 3196 switch (opt) { 3197 3198 default: 3199 continue; 3200 3201 case TCPOPT_MAXSEG: 3202 if (optlen != TCPOLEN_MAXSEG) 3203 continue; 3204 if (!(th->th_flags & TH_SYN)) 3205 continue; 3206 if (TCPS_HAVERCVDSYN(tp->t_state)) 3207 continue; 3208 memcpy(&mss, cp + 2, sizeof(mss)); 3209 oi->maxseg = ntohs(mss); 3210 break; 3211 3212 case TCPOPT_WINDOW: 3213 if (optlen != TCPOLEN_WINDOW) 3214 continue; 3215 if (!(th->th_flags & TH_SYN)) 3216 continue; 3217 if (TCPS_HAVERCVDSYN(tp->t_state)) 3218 continue; 3219 tp->t_flags |= TF_RCVD_SCALE; 3220 tp->requested_s_scale = cp[2]; 3221 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3222 char buf[INET6_ADDRSTRLEN]; 3223 struct ip *ip = mtod(m, struct ip *); 3224 #ifdef INET6 3225 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 3226 #endif 3227 if (ip) 3228 in_print(buf, sizeof(buf), 3229 &ip->ip_src); 3230 #ifdef INET6 3231 else if (ip6) 3232 in6_print(buf, sizeof(buf), 3233 &ip6->ip6_src); 3234 #endif 3235 else 3236 strlcpy(buf, "(unknown)", sizeof(buf)); 3237 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3238 "assuming %d\n", 3239 tp->requested_s_scale, buf, 3240 TCP_MAX_WINSHIFT); 3241 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3242 } 3243 break; 3244 3245 case TCPOPT_TIMESTAMP: 3246 if (optlen != TCPOLEN_TIMESTAMP) 3247 continue; 3248 oi->ts_present = 1; 3249 memcpy(&oi->ts_val, cp + 2, sizeof(oi->ts_val)); 3250 NTOHL(oi->ts_val); 3251 memcpy(&oi->ts_ecr, cp + 6, sizeof(oi->ts_ecr)); 3252 NTOHL(oi->ts_ecr); 3253 3254 if (!(th->th_flags & TH_SYN)) 3255 continue; 3256 if (TCPS_HAVERCVDSYN(tp->t_state)) 3257 continue; 3258 /* 3259 * A timestamp received in a SYN makes 3260 * it ok to send timestamp requests and replies. 3261 */ 3262 tp->t_flags |= TF_RCVD_TSTMP; 3263 tp->ts_recent = oi->ts_val; 3264 tp->ts_recent_age = tcp_now; 3265 break; 3266 3267 case TCPOPT_SACK_PERMITTED: 3268 if (optlen != TCPOLEN_SACK_PERMITTED) 3269 continue; 3270 if (!(th->th_flags & TH_SYN)) 3271 continue; 3272 if (TCPS_HAVERCVDSYN(tp->t_state)) 3273 continue; 3274 if (tcp_do_sack) { 3275 tp->t_flags |= TF_SACK_PERMIT; 3276 tp->t_flags |= TF_WILL_SACK; 3277 } 3278 break; 3279 3280 case TCPOPT_SACK: 3281 tcp_sack_option(tp, th, cp, optlen); 3282 break; 3283 #ifdef TCP_SIGNATURE 3284 case TCPOPT_SIGNATURE: 3285 if (optlen != TCPOLEN_SIGNATURE) 3286 continue; 3287 if (sigp && memcmp(sigp, cp + 2, TCP_SIGLEN)) 3288 return (-1); 3289 3290 sigp = sigbuf; 3291 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3292 tp->t_flags |= TF_SIGNATURE; 3293 break; 3294 #endif 3295 } 3296 } 3297 3298 #ifndef TCP_SIGNATURE 3299 return 0; 3300 #else 3301 if (tp->t_flags & TF_SIGNATURE) { 3302 sav = tcp_signature_getsav(m); 3303 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3304 return (-1); 3305 } 3306 3307 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) 3308 goto out; 3309 3310 if (sigp) { 3311 char sig[TCP_SIGLEN]; 3312 3313 tcp_fields_to_net(th); 3314 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3315 tcp_fields_to_host(th); 3316 goto out; 3317 } 3318 tcp_fields_to_host(th); 3319 3320 if (memcmp(sig, sigp, TCP_SIGLEN)) { 3321 TCP_STATINC(TCP_STAT_BADSIG); 3322 goto out; 3323 } else 3324 TCP_STATINC(TCP_STAT_GOODSIG); 3325 3326 key_sa_recordxfer(sav, m); 3327 KEY_SA_UNREF(&sav); 3328 } 3329 return 0; 3330 out: 3331 if (sav != NULL) 3332 KEY_SA_UNREF(&sav); 3333 return -1; 3334 #endif 3335 } 3336 3337 /* 3338 * Pull out of band byte out of a segment so 3339 * it doesn't appear in the user's data queue. 3340 * It is still reflected in the segment length for 3341 * sequencing purposes. 3342 */ 3343 void 3344 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3345 struct mbuf *m, int off) 3346 { 3347 int cnt = off + th->th_urp - 1; 3348 3349 while (cnt >= 0) { 3350 if (m->m_len > cnt) { 3351 char *cp = mtod(m, char *) + cnt; 3352 struct tcpcb *tp = sototcpcb(so); 3353 3354 tp->t_iobc = *cp; 3355 tp->t_oobflags |= TCPOOB_HAVEDATA; 3356 memmove(cp, cp + 1, (unsigned)(m->m_len - cnt - 1)); 3357 m->m_len--; 3358 return; 3359 } 3360 cnt -= m->m_len; 3361 m = m->m_next; 3362 if (m == NULL) 3363 break; 3364 } 3365 panic("tcp_pulloutofband"); 3366 } 3367 3368 /* 3369 * Collect new round-trip time estimate 3370 * and update averages and current timeout. 3371 * 3372 * rtt is in units of slow ticks (typically 500 ms) -- essentially the 3373 * difference of two timestamps. 3374 */ 3375 void 3376 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3377 { 3378 int32_t delta; 3379 3380 TCP_STATINC(TCP_STAT_RTTUPDATED); 3381 if (tp->t_srtt != 0) { 3382 /* 3383 * Compute the amount to add to srtt for smoothing, 3384 * *alpha, or 2^(-TCP_RTT_SHIFT). Because 3385 * srtt is stored in 1/32 slow ticks, we conceptually 3386 * shift left 5 bits, subtract srtt to get the 3387 * diference, and then shift right by TCP_RTT_SHIFT 3388 * (3) to obtain 1/8 of the difference. 3389 */ 3390 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3391 /* 3392 * This can never happen, because delta's lowest 3393 * possible value is 1/8 of t_srtt. But if it does, 3394 * set srtt to some reasonable value, here chosen 3395 * as 1/8 tick. 3396 */ 3397 if ((tp->t_srtt += delta) <= 0) 3398 tp->t_srtt = 1 << 2; 3399 /* 3400 * RFC2988 requires that rttvar be updated first. 3401 * This code is compliant because "delta" is the old 3402 * srtt minus the new observation (scaled). 3403 * 3404 * RFC2988 says: 3405 * rttvar = (1-beta) * rttvar + beta * |srtt-observed| 3406 * 3407 * delta is in units of 1/32 ticks, and has then been 3408 * divided by 8. This is equivalent to being in 1/16s 3409 * units and divided by 4. Subtract from it 1/4 of 3410 * the existing rttvar to form the (signed) amount to 3411 * adjust. 3412 */ 3413 if (delta < 0) 3414 delta = -delta; 3415 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3416 /* 3417 * As with srtt, this should never happen. There is 3418 * no support in RFC2988 for this operation. But 1/4s 3419 * as rttvar when faced with something arguably wrong 3420 * is ok. 3421 */ 3422 if ((tp->t_rttvar += delta) <= 0) 3423 tp->t_rttvar = 1 << 2; 3424 3425 /* 3426 * If srtt exceeds .01 second, ensure we use the 'remote' MSL 3427 * Problem is: it doesn't work. Disabled by defaulting 3428 * tcp_rttlocal to 0; see corresponding code in 3429 * tcp_subr that selects local vs remote in a different way. 3430 * 3431 * The static branch prediction hint here should be removed 3432 * when the rtt estimator is fixed and the rtt_enable code 3433 * is turned back on. 3434 */ 3435 if (__predict_false(tcp_rttlocal) && tcp_msl_enable 3436 && tp->t_srtt > tcp_msl_remote_threshold 3437 && tp->t_msl < tcp_msl_remote) { 3438 tp->t_msl = tcp_msl_remote; 3439 } 3440 } else { 3441 /* 3442 * This is the first measurement. Per RFC2988, 2.2, 3443 * set rtt=R and srtt=R/2. 3444 * For srtt, storage representation is 1/32 ticks, 3445 * so shift left by 5. 3446 * For rttvar, storage representation is 1/16 ticks, 3447 * So shift left by 4, but then right by 1 to halve. 3448 */ 3449 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3450 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3451 } 3452 tp->t_rtttime = 0; 3453 tp->t_rxtshift = 0; 3454 3455 /* 3456 * the retransmit should happen at rtt + 4 * rttvar. 3457 * Because of the way we do the smoothing, srtt and rttvar 3458 * will each average +1/2 tick of bias. When we compute 3459 * the retransmit timer, we want 1/2 tick of rounding and 3460 * 1 extra tick because of +-1/2 tick uncertainty in the 3461 * firing of the timer. The bias will give us exactly the 3462 * 1.5 tick we need. But, because the bias is 3463 * statistical, we have to test that we don't drop below 3464 * the minimum feasible timer (which is 2 ticks). 3465 */ 3466 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3467 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3468 3469 /* 3470 * We received an ack for a packet that wasn't retransmitted; 3471 * it is probably safe to discard any error indications we've 3472 * received recently. This isn't quite right, but close enough 3473 * for now (a route might have failed after we sent a segment, 3474 * and the return path might not be symmetrical). 3475 */ 3476 tp->t_softerror = 0; 3477 } 3478 3479 3480 /* 3481 * TCP compressed state engine. Currently used to hold compressed 3482 * state for SYN_RECEIVED. 3483 */ 3484 3485 u_long syn_cache_count; 3486 u_int32_t syn_hash1, syn_hash2; 3487 3488 #define SYN_HASH(sa, sp, dp) \ 3489 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3490 ((u_int32_t)(sp)))^syn_hash2))) 3491 #ifndef INET6 3492 #define SYN_HASHALL(hash, src, dst) \ 3493 do { \ 3494 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3495 ((const struct sockaddr_in *)(src))->sin_port, \ 3496 ((const struct sockaddr_in *)(dst))->sin_port); \ 3497 } while (/*CONSTCOND*/ 0) 3498 #else 3499 #define SYN_HASH6(sa, sp, dp) \ 3500 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3501 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3502 & 0x7fffffff) 3503 3504 #define SYN_HASHALL(hash, src, dst) \ 3505 do { \ 3506 switch ((src)->sa_family) { \ 3507 case AF_INET: \ 3508 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3509 ((const struct sockaddr_in *)(src))->sin_port, \ 3510 ((const struct sockaddr_in *)(dst))->sin_port); \ 3511 break; \ 3512 case AF_INET6: \ 3513 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3514 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3515 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3516 break; \ 3517 default: \ 3518 hash = 0; \ 3519 } \ 3520 } while (/*CONSTCOND*/0) 3521 #endif /* INET6 */ 3522 3523 static struct pool syn_cache_pool; 3524 3525 /* 3526 * We don't estimate RTT with SYNs, so each packet starts with the default 3527 * RTT and each timer step has a fixed timeout value. 3528 */ 3529 static inline void 3530 syn_cache_timer_arm(struct syn_cache *sc) 3531 { 3532 3533 TCPT_RANGESET(sc->sc_rxtcur, 3534 TCPTV_SRTTDFLT * tcp_backoff[sc->sc_rxtshift], TCPTV_MIN, 3535 TCPTV_REXMTMAX); 3536 callout_reset(&sc->sc_timer, 3537 sc->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, sc); 3538 } 3539 3540 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3541 3542 static inline void 3543 syn_cache_rm(struct syn_cache *sc) 3544 { 3545 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3546 sc, sc_bucketq); 3547 sc->sc_tp = NULL; 3548 LIST_REMOVE(sc, sc_tpq); 3549 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3550 callout_stop(&sc->sc_timer); 3551 syn_cache_count--; 3552 } 3553 3554 static inline void 3555 syn_cache_put(struct syn_cache *sc) 3556 { 3557 if (sc->sc_ipopts) 3558 (void) m_free(sc->sc_ipopts); 3559 rtcache_free(&sc->sc_route); 3560 sc->sc_flags |= SCF_DEAD; 3561 if (!callout_invoking(&sc->sc_timer)) 3562 callout_schedule(&(sc)->sc_timer, 1); 3563 } 3564 3565 void 3566 syn_cache_init(void) 3567 { 3568 int i; 3569 3570 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 3571 "synpl", NULL, IPL_SOFTNET); 3572 3573 /* Initialize the hash buckets. */ 3574 for (i = 0; i < tcp_syn_cache_size; i++) 3575 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3576 } 3577 3578 void 3579 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3580 { 3581 struct syn_cache_head *scp; 3582 struct syn_cache *sc2; 3583 int s; 3584 3585 /* 3586 * If there are no entries in the hash table, reinitialize 3587 * the hash secrets. 3588 */ 3589 if (syn_cache_count == 0) { 3590 syn_hash1 = cprng_fast32(); 3591 syn_hash2 = cprng_fast32(); 3592 } 3593 3594 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3595 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3596 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3597 3598 /* 3599 * Make sure that we don't overflow the per-bucket 3600 * limit or the total cache size limit. 3601 */ 3602 s = splsoftnet(); 3603 if (scp->sch_length >= tcp_syn_bucket_limit) { 3604 TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW); 3605 /* 3606 * The bucket is full. Toss the oldest element in the 3607 * bucket. This will be the first entry in the bucket. 3608 */ 3609 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3610 #ifdef DIAGNOSTIC 3611 /* 3612 * This should never happen; we should always find an 3613 * entry in our bucket. 3614 */ 3615 if (sc2 == NULL) 3616 panic("syn_cache_insert: bucketoverflow: impossible"); 3617 #endif 3618 syn_cache_rm(sc2); 3619 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3620 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3621 struct syn_cache_head *scp2, *sce; 3622 3623 TCP_STATINC(TCP_STAT_SC_OVERFLOWED); 3624 /* 3625 * The cache is full. Toss the oldest entry in the 3626 * first non-empty bucket we can find. 3627 * 3628 * XXX We would really like to toss the oldest 3629 * entry in the cache, but we hope that this 3630 * condition doesn't happen very often. 3631 */ 3632 scp2 = scp; 3633 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3634 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3635 for (++scp2; scp2 != scp; scp2++) { 3636 if (scp2 >= sce) 3637 scp2 = &tcp_syn_cache[0]; 3638 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3639 break; 3640 } 3641 #ifdef DIAGNOSTIC 3642 /* 3643 * This should never happen; we should always find a 3644 * non-empty bucket. 3645 */ 3646 if (scp2 == scp) 3647 panic("syn_cache_insert: cacheoverflow: " 3648 "impossible"); 3649 #endif 3650 } 3651 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3652 syn_cache_rm(sc2); 3653 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3654 } 3655 3656 /* 3657 * Initialize the entry's timer. 3658 */ 3659 sc->sc_rxttot = 0; 3660 sc->sc_rxtshift = 0; 3661 syn_cache_timer_arm(sc); 3662 3663 /* Link it from tcpcb entry */ 3664 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3665 3666 /* Put it into the bucket. */ 3667 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3668 scp->sch_length++; 3669 syn_cache_count++; 3670 3671 TCP_STATINC(TCP_STAT_SC_ADDED); 3672 splx(s); 3673 } 3674 3675 /* 3676 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3677 * If we have retransmitted an entry the maximum number of times, expire 3678 * that entry. 3679 */ 3680 static void 3681 syn_cache_timer(void *arg) 3682 { 3683 struct syn_cache *sc = arg; 3684 3685 mutex_enter(softnet_lock); 3686 KERNEL_LOCK(1, NULL); 3687 3688 callout_ack(&sc->sc_timer); 3689 3690 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3691 TCP_STATINC(TCP_STAT_SC_DELAYED_FREE); 3692 goto free; 3693 } 3694 3695 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3696 /* Drop it -- too many retransmissions. */ 3697 goto dropit; 3698 } 3699 3700 /* 3701 * Compute the total amount of time this entry has 3702 * been on a queue. If this entry has been on longer 3703 * than the keep alive timer would allow, expire it. 3704 */ 3705 sc->sc_rxttot += sc->sc_rxtcur; 3706 if (sc->sc_rxttot >= tcp_keepinit) 3707 goto dropit; 3708 3709 TCP_STATINC(TCP_STAT_SC_RETRANSMITTED); 3710 (void)syn_cache_respond(sc); 3711 3712 /* Advance the timer back-off. */ 3713 sc->sc_rxtshift++; 3714 syn_cache_timer_arm(sc); 3715 3716 goto out; 3717 3718 dropit: 3719 TCP_STATINC(TCP_STAT_SC_TIMED_OUT); 3720 syn_cache_rm(sc); 3721 if (sc->sc_ipopts) 3722 (void) m_free(sc->sc_ipopts); 3723 rtcache_free(&sc->sc_route); 3724 3725 free: 3726 callout_destroy(&sc->sc_timer); 3727 pool_put(&syn_cache_pool, sc); 3728 3729 out: 3730 KERNEL_UNLOCK_ONE(NULL); 3731 mutex_exit(softnet_lock); 3732 } 3733 3734 /* 3735 * Remove syn cache created by the specified tcb entry, 3736 * because this does not make sense to keep them 3737 * (if there's no tcb entry, syn cache entry will never be used) 3738 */ 3739 void 3740 syn_cache_cleanup(struct tcpcb *tp) 3741 { 3742 struct syn_cache *sc, *nsc; 3743 int s; 3744 3745 s = splsoftnet(); 3746 3747 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3748 nsc = LIST_NEXT(sc, sc_tpq); 3749 3750 #ifdef DIAGNOSTIC 3751 if (sc->sc_tp != tp) 3752 panic("invalid sc_tp in syn_cache_cleanup"); 3753 #endif 3754 syn_cache_rm(sc); 3755 syn_cache_put(sc); /* calls pool_put but see spl above */ 3756 } 3757 /* just for safety */ 3758 LIST_INIT(&tp->t_sc); 3759 3760 splx(s); 3761 } 3762 3763 /* 3764 * Find an entry in the syn cache. 3765 */ 3766 struct syn_cache * 3767 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3768 struct syn_cache_head **headp) 3769 { 3770 struct syn_cache *sc; 3771 struct syn_cache_head *scp; 3772 u_int32_t hash; 3773 int s; 3774 3775 SYN_HASHALL(hash, src, dst); 3776 3777 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3778 *headp = scp; 3779 s = splsoftnet(); 3780 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3781 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3782 if (sc->sc_hash != hash) 3783 continue; 3784 if (!memcmp(&sc->sc_src, src, src->sa_len) && 3785 !memcmp(&sc->sc_dst, dst, dst->sa_len)) { 3786 splx(s); 3787 return (sc); 3788 } 3789 } 3790 splx(s); 3791 return (NULL); 3792 } 3793 3794 /* 3795 * This function gets called when we receive an ACK for a 3796 * socket in the LISTEN state. We look up the connection 3797 * in the syn cache, and if its there, we pull it out of 3798 * the cache and turn it into a full-blown connection in 3799 * the SYN-RECEIVED state. 3800 * 3801 * The return values may not be immediately obvious, and their effects 3802 * can be subtle, so here they are: 3803 * 3804 * NULL SYN was not found in cache; caller should drop the 3805 * packet and send an RST. 3806 * 3807 * -1 We were unable to create the new connection, and are 3808 * aborting it. An ACK,RST is being sent to the peer 3809 * (unless we got screwey sequence numbners; see below), 3810 * because the 3-way handshake has been completed. Caller 3811 * should not free the mbuf, since we may be using it. If 3812 * we are not, we will free it. 3813 * 3814 * Otherwise, the return value is a pointer to the new socket 3815 * associated with the connection. 3816 */ 3817 struct socket * 3818 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3819 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3820 struct socket *so, struct mbuf *m) 3821 { 3822 struct syn_cache *sc; 3823 struct syn_cache_head *scp; 3824 struct inpcb *inp = NULL; 3825 #ifdef INET6 3826 struct in6pcb *in6p = NULL; 3827 #endif 3828 struct tcpcb *tp = 0; 3829 int s; 3830 struct socket *oso; 3831 3832 s = splsoftnet(); 3833 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3834 splx(s); 3835 return (NULL); 3836 } 3837 3838 /* 3839 * Verify the sequence and ack numbers. Try getting the correct 3840 * response again. 3841 */ 3842 if ((th->th_ack != sc->sc_iss + 1) || 3843 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3844 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3845 m_freem(m); 3846 (void)syn_cache_respond(sc); 3847 splx(s); 3848 return ((struct socket *)(-1)); 3849 } 3850 3851 /* Remove this cache entry */ 3852 syn_cache_rm(sc); 3853 splx(s); 3854 3855 /* 3856 * Ok, create the full blown connection, and set things up 3857 * as they would have been set up if we had created the 3858 * connection when the SYN arrived. If we can't create 3859 * the connection, abort it. 3860 */ 3861 /* 3862 * inp still has the OLD in_pcb stuff, set the 3863 * v6-related flags on the new guy, too. This is 3864 * done particularly for the case where an AF_INET6 3865 * socket is bound only to a port, and a v4 connection 3866 * comes in on that port. 3867 * we also copy the flowinfo from the original pcb 3868 * to the new one. 3869 */ 3870 oso = so; 3871 so = sonewconn(so, true); 3872 if (so == NULL) 3873 goto resetandabort; 3874 3875 switch (so->so_proto->pr_domain->dom_family) { 3876 #ifdef INET 3877 case AF_INET: 3878 inp = sotoinpcb(so); 3879 break; 3880 #endif 3881 #ifdef INET6 3882 case AF_INET6: 3883 in6p = sotoin6pcb(so); 3884 break; 3885 #endif 3886 } 3887 switch (src->sa_family) { 3888 #ifdef INET 3889 case AF_INET: 3890 if (inp) { 3891 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3892 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3893 inp->inp_options = ip_srcroute(m); 3894 in_pcbstate(inp, INP_BOUND); 3895 if (inp->inp_options == NULL) { 3896 inp->inp_options = sc->sc_ipopts; 3897 sc->sc_ipopts = NULL; 3898 } 3899 } 3900 #ifdef INET6 3901 else if (in6p) { 3902 /* IPv4 packet to AF_INET6 socket */ 3903 memset(&in6p->in6p_laddr, 0, sizeof(in6p->in6p_laddr)); 3904 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3905 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3906 &in6p->in6p_laddr.s6_addr32[3], 3907 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3908 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3909 in6totcpcb(in6p)->t_family = AF_INET; 3910 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3911 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3912 else 3913 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3914 in6_pcbstate(in6p, IN6P_BOUND); 3915 } 3916 #endif 3917 break; 3918 #endif 3919 #ifdef INET6 3920 case AF_INET6: 3921 if (in6p) { 3922 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3923 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3924 in6_pcbstate(in6p, IN6P_BOUND); 3925 } 3926 break; 3927 #endif 3928 } 3929 #ifdef INET6 3930 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3931 struct in6pcb *oin6p = sotoin6pcb(oso); 3932 /* inherit socket options from the listening socket */ 3933 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3934 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3935 m_freem(in6p->in6p_options); 3936 in6p->in6p_options = 0; 3937 } 3938 ip6_savecontrol(in6p, &in6p->in6p_options, 3939 mtod(m, struct ip6_hdr *), m); 3940 } 3941 #endif 3942 3943 #if defined(IPSEC) 3944 if (ipsec_used) { 3945 /* 3946 * we make a copy of policy, instead of sharing the policy, for 3947 * better behavior in terms of SA lookup and dead SA removal. 3948 */ 3949 if (inp) { 3950 /* copy old policy into new socket's */ 3951 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, 3952 inp->inp_sp)) 3953 printf("tcp_input: could not copy policy\n"); 3954 } 3955 #ifdef INET6 3956 else if (in6p) { 3957 /* copy old policy into new socket's */ 3958 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3959 in6p->in6p_sp)) 3960 printf("tcp_input: could not copy policy\n"); 3961 } 3962 #endif 3963 } 3964 #endif 3965 3966 /* 3967 * Give the new socket our cached route reference. 3968 */ 3969 if (inp) { 3970 rtcache_copy(&inp->inp_route, &sc->sc_route); 3971 rtcache_free(&sc->sc_route); 3972 } 3973 #ifdef INET6 3974 else { 3975 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 3976 rtcache_free(&sc->sc_route); 3977 } 3978 #endif 3979 3980 if (inp) { 3981 struct sockaddr_in sin; 3982 memcpy(&sin, src, src->sa_len); 3983 if (in_pcbconnect(inp, &sin, &lwp0)) { 3984 goto resetandabort; 3985 } 3986 } 3987 #ifdef INET6 3988 else if (in6p) { 3989 struct sockaddr_in6 sin6; 3990 memcpy(&sin6, src, src->sa_len); 3991 if (src->sa_family == AF_INET) { 3992 /* IPv4 packet to AF_INET6 socket */ 3993 in6_sin_2_v4mapsin6((struct sockaddr_in *)src, &sin6); 3994 } 3995 if (in6_pcbconnect(in6p, &sin6, NULL)) { 3996 goto resetandabort; 3997 } 3998 } 3999 #endif 4000 else { 4001 goto resetandabort; 4002 } 4003 4004 if (inp) 4005 tp = intotcpcb(inp); 4006 #ifdef INET6 4007 else if (in6p) 4008 tp = in6totcpcb(in6p); 4009 #endif 4010 else 4011 tp = NULL; 4012 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 4013 if (sc->sc_request_r_scale != 15) { 4014 tp->requested_s_scale = sc->sc_requested_s_scale; 4015 tp->request_r_scale = sc->sc_request_r_scale; 4016 tp->snd_scale = sc->sc_requested_s_scale; 4017 tp->rcv_scale = sc->sc_request_r_scale; 4018 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 4019 } 4020 if (sc->sc_flags & SCF_TIMESTAMP) 4021 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 4022 tp->ts_timebase = sc->sc_timebase; 4023 4024 tp->t_template = tcp_template(tp); 4025 if (tp->t_template == 0) { 4026 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 4027 so = NULL; 4028 m_freem(m); 4029 goto abort; 4030 } 4031 4032 tp->iss = sc->sc_iss; 4033 tp->irs = sc->sc_irs; 4034 tcp_sendseqinit(tp); 4035 tcp_rcvseqinit(tp); 4036 tp->t_state = TCPS_SYN_RECEIVED; 4037 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 4038 TCP_STATINC(TCP_STAT_ACCEPTS); 4039 4040 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 4041 tp->t_flags |= TF_WILL_SACK; 4042 4043 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 4044 tp->t_flags |= TF_ECN_PERMIT; 4045 4046 #ifdef TCP_SIGNATURE 4047 if (sc->sc_flags & SCF_SIGNATURE) 4048 tp->t_flags |= TF_SIGNATURE; 4049 #endif 4050 4051 /* Initialize tp->t_ourmss before we deal with the peer's! */ 4052 tp->t_ourmss = sc->sc_ourmaxseg; 4053 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 4054 4055 /* 4056 * Initialize the initial congestion window. If we 4057 * had to retransmit the SYN,ACK, we must initialize cwnd 4058 * to 1 segment (i.e. the Loss Window). 4059 */ 4060 if (sc->sc_rxtshift) 4061 tp->snd_cwnd = tp->t_peermss; 4062 else { 4063 int ss = tcp_init_win; 4064 #ifdef INET 4065 if (inp != NULL && in_localaddr(inp->inp_faddr)) 4066 ss = tcp_init_win_local; 4067 #endif 4068 #ifdef INET6 4069 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 4070 ss = tcp_init_win_local; 4071 #endif 4072 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 4073 } 4074 4075 tcp_rmx_rtt(tp); 4076 tp->snd_wl1 = sc->sc_irs; 4077 tp->rcv_up = sc->sc_irs + 1; 4078 4079 /* 4080 * This is what whould have happened in tcp_output() when 4081 * the SYN,ACK was sent. 4082 */ 4083 tp->snd_up = tp->snd_una; 4084 tp->snd_max = tp->snd_nxt = tp->iss+1; 4085 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 4086 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 4087 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 4088 tp->last_ack_sent = tp->rcv_nxt; 4089 tp->t_partialacks = -1; 4090 tp->t_dupacks = 0; 4091 4092 TCP_STATINC(TCP_STAT_SC_COMPLETED); 4093 s = splsoftnet(); 4094 syn_cache_put(sc); 4095 splx(s); 4096 return (so); 4097 4098 resetandabort: 4099 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 4100 abort: 4101 if (so != NULL) { 4102 (void) soqremque(so, 1); 4103 (void) soabort(so); 4104 mutex_enter(softnet_lock); 4105 } 4106 s = splsoftnet(); 4107 syn_cache_put(sc); 4108 splx(s); 4109 TCP_STATINC(TCP_STAT_SC_ABORTED); 4110 return ((struct socket *)(-1)); 4111 } 4112 4113 /* 4114 * This function is called when we get a RST for a 4115 * non-existent connection, so that we can see if the 4116 * connection is in the syn cache. If it is, zap it. 4117 */ 4118 4119 void 4120 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 4121 { 4122 struct syn_cache *sc; 4123 struct syn_cache_head *scp; 4124 int s = splsoftnet(); 4125 4126 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4127 splx(s); 4128 return; 4129 } 4130 if (SEQ_LT(th->th_seq, sc->sc_irs) || 4131 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 4132 splx(s); 4133 return; 4134 } 4135 syn_cache_rm(sc); 4136 TCP_STATINC(TCP_STAT_SC_RESET); 4137 syn_cache_put(sc); /* calls pool_put but see spl above */ 4138 splx(s); 4139 } 4140 4141 void 4142 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 4143 struct tcphdr *th) 4144 { 4145 struct syn_cache *sc; 4146 struct syn_cache_head *scp; 4147 int s; 4148 4149 s = splsoftnet(); 4150 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 4151 splx(s); 4152 return; 4153 } 4154 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 4155 if (ntohl (th->th_seq) != sc->sc_iss) { 4156 splx(s); 4157 return; 4158 } 4159 4160 /* 4161 * If we've retransmitted 3 times and this is our second error, 4162 * we remove the entry. Otherwise, we allow it to continue on. 4163 * This prevents us from incorrectly nuking an entry during a 4164 * spurious network outage. 4165 * 4166 * See tcp_notify(). 4167 */ 4168 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 4169 sc->sc_flags |= SCF_UNREACH; 4170 splx(s); 4171 return; 4172 } 4173 4174 syn_cache_rm(sc); 4175 TCP_STATINC(TCP_STAT_SC_UNREACH); 4176 syn_cache_put(sc); /* calls pool_put but see spl above */ 4177 splx(s); 4178 } 4179 4180 /* 4181 * Given a LISTEN socket and an inbound SYN request, add 4182 * this to the syn cache, and send back a segment: 4183 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 4184 * to the source. 4185 * 4186 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 4187 * Doing so would require that we hold onto the data and deliver it 4188 * to the application. However, if we are the target of a SYN-flood 4189 * DoS attack, an attacker could send data which would eventually 4190 * consume all available buffer space if it were ACKed. By not ACKing 4191 * the data, we avoid this DoS scenario. 4192 */ 4193 int 4194 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 4195 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 4196 int optlen, struct tcp_opt_info *oi) 4197 { 4198 struct tcpcb tb, *tp; 4199 long win; 4200 struct syn_cache *sc; 4201 struct syn_cache_head *scp; 4202 struct mbuf *ipopts; 4203 struct tcp_opt_info opti; 4204 int s; 4205 4206 tp = sototcpcb(so); 4207 4208 memset(&opti, 0, sizeof(opti)); 4209 4210 /* 4211 * Initialize some local state. 4212 */ 4213 win = sbspace(&so->so_rcv); 4214 if (win > TCP_MAXWIN) 4215 win = TCP_MAXWIN; 4216 4217 #ifdef TCP_SIGNATURE 4218 if (optp || (tp->t_flags & TF_SIGNATURE)) 4219 #else 4220 if (optp) 4221 #endif 4222 { 4223 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4224 #ifdef TCP_SIGNATURE 4225 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4226 #endif 4227 tb.t_state = TCPS_LISTEN; 4228 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 4229 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 4230 return 0; 4231 } else 4232 tb.t_flags = 0; 4233 4234 switch (src->sa_family) { 4235 #ifdef INET 4236 case AF_INET: 4237 /* Remember the IP options, if any. */ 4238 ipopts = ip_srcroute(m); 4239 break; 4240 #endif 4241 default: 4242 ipopts = NULL; 4243 } 4244 4245 /* 4246 * See if we already have an entry for this connection. 4247 * If we do, resend the SYN,ACK. We do not count this 4248 * as a retransmission (XXX though maybe we should). 4249 */ 4250 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4251 TCP_STATINC(TCP_STAT_SC_DUPESYN); 4252 if (ipopts) { 4253 /* 4254 * If we were remembering a previous source route, 4255 * forget it and use the new one we've been given. 4256 */ 4257 if (sc->sc_ipopts) 4258 (void)m_free(sc->sc_ipopts); 4259 sc->sc_ipopts = ipopts; 4260 } 4261 sc->sc_timestamp = tb.ts_recent; 4262 m_freem(m); 4263 if (syn_cache_respond(sc) == 0) { 4264 uint64_t *tcps = TCP_STAT_GETREF(); 4265 tcps[TCP_STAT_SNDACKS]++; 4266 tcps[TCP_STAT_SNDTOTAL]++; 4267 TCP_STAT_PUTREF(); 4268 } 4269 return 1; 4270 } 4271 4272 s = splsoftnet(); 4273 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4274 splx(s); 4275 if (sc == NULL) { 4276 if (ipopts) 4277 (void)m_free(ipopts); 4278 return 0; 4279 } 4280 4281 /* 4282 * Fill in the cache, and put the necessary IP and TCP 4283 * options into the reply. 4284 */ 4285 memset(sc, 0, sizeof(struct syn_cache)); 4286 callout_init(&sc->sc_timer, CALLOUT_MPSAFE); 4287 memcpy(&sc->sc_src, src, src->sa_len); 4288 memcpy(&sc->sc_dst, dst, dst->sa_len); 4289 sc->sc_flags = 0; 4290 sc->sc_ipopts = ipopts; 4291 sc->sc_irs = th->th_seq; 4292 switch (src->sa_family) { 4293 #ifdef INET 4294 case AF_INET: 4295 { 4296 struct sockaddr_in *srcin = (void *)src; 4297 struct sockaddr_in *dstin = (void *)dst; 4298 4299 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4300 &srcin->sin_addr, dstin->sin_port, 4301 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4302 break; 4303 } 4304 #endif /* INET */ 4305 #ifdef INET6 4306 case AF_INET6: 4307 { 4308 struct sockaddr_in6 *srcin6 = (void *)src; 4309 struct sockaddr_in6 *dstin6 = (void *)dst; 4310 4311 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4312 &srcin6->sin6_addr, dstin6->sin6_port, 4313 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4314 break; 4315 } 4316 #endif /* INET6 */ 4317 } 4318 sc->sc_peermaxseg = oi->maxseg; 4319 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4320 m_get_rcvif_NOMPSAFE(m) : NULL, sc->sc_src.sa.sa_family); 4321 sc->sc_win = win; 4322 sc->sc_timebase = tcp_now - 1; /* see tcp_newtcpcb() */ 4323 sc->sc_timestamp = tb.ts_recent; 4324 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4325 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4326 sc->sc_flags |= SCF_TIMESTAMP; 4327 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4328 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4329 sc->sc_requested_s_scale = tb.requested_s_scale; 4330 sc->sc_request_r_scale = 0; 4331 /* 4332 * Pick the smallest possible scaling factor that 4333 * will still allow us to scale up to sb_max. 4334 * 4335 * We do this because there are broken firewalls that 4336 * will corrupt the window scale option, leading to 4337 * the other endpoint believing that our advertised 4338 * window is unscaled. At scale factors larger than 4339 * 5 the unscaled window will drop below 1500 bytes, 4340 * leading to serious problems when traversing these 4341 * broken firewalls. 4342 * 4343 * With the default sbmax of 256K, a scale factor 4344 * of 3 will be chosen by this algorithm. Those who 4345 * choose a larger sbmax should watch out 4346 * for the compatiblity problems mentioned above. 4347 * 4348 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4349 * or <SYN,ACK>) segment itself is never scaled. 4350 */ 4351 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4352 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4353 sc->sc_request_r_scale++; 4354 } else { 4355 sc->sc_requested_s_scale = 15; 4356 sc->sc_request_r_scale = 15; 4357 } 4358 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4359 sc->sc_flags |= SCF_SACK_PERMIT; 4360 4361 /* 4362 * ECN setup packet received. 4363 */ 4364 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4365 sc->sc_flags |= SCF_ECN_PERMIT; 4366 4367 #ifdef TCP_SIGNATURE 4368 if (tb.t_flags & TF_SIGNATURE) 4369 sc->sc_flags |= SCF_SIGNATURE; 4370 #endif 4371 sc->sc_tp = tp; 4372 m_freem(m); 4373 if (syn_cache_respond(sc) == 0) { 4374 uint64_t *tcps = TCP_STAT_GETREF(); 4375 tcps[TCP_STAT_SNDACKS]++; 4376 tcps[TCP_STAT_SNDTOTAL]++; 4377 TCP_STAT_PUTREF(); 4378 syn_cache_insert(sc, tp); 4379 } else { 4380 s = splsoftnet(); 4381 /* 4382 * syn_cache_put() will try to schedule the timer, so 4383 * we need to initialize it 4384 */ 4385 syn_cache_timer_arm(sc); 4386 syn_cache_put(sc); 4387 splx(s); 4388 TCP_STATINC(TCP_STAT_SC_DROPPED); 4389 } 4390 return 1; 4391 } 4392 4393 /* 4394 * syn_cache_respond: (re)send SYN+ACK. 4395 * 4396 * returns 0 on success. otherwise returns an errno, typically ENOBUFS. 4397 */ 4398 4399 int 4400 syn_cache_respond(struct syn_cache *sc) 4401 { 4402 #ifdef INET6 4403 struct rtentry *rt = NULL; 4404 #endif 4405 struct route *ro; 4406 u_int8_t *optp; 4407 int optlen, error; 4408 u_int16_t tlen; 4409 struct ip *ip = NULL; 4410 #ifdef INET6 4411 struct ip6_hdr *ip6 = NULL; 4412 #endif 4413 struct tcpcb *tp = NULL; 4414 struct tcphdr *th; 4415 struct mbuf *m; 4416 u_int hlen; 4417 #ifdef TCP_SIGNATURE 4418 struct secasvar *sav = NULL; 4419 u_int8_t *sigp = NULL; 4420 #endif 4421 4422 ro = &sc->sc_route; 4423 switch (sc->sc_src.sa.sa_family) { 4424 case AF_INET: 4425 hlen = sizeof(struct ip); 4426 break; 4427 #ifdef INET6 4428 case AF_INET6: 4429 hlen = sizeof(struct ip6_hdr); 4430 break; 4431 #endif 4432 default: 4433 return (EAFNOSUPPORT); 4434 } 4435 4436 /* worst case scanario, since we don't know the option size yet */ 4437 tlen = hlen + sizeof(struct tcphdr) + MAX_TCPOPTLEN; 4438 4439 /* 4440 * Create the IP+TCP header from scratch. 4441 */ 4442 #ifdef DIAGNOSTIC 4443 if (max_linkhdr + tlen > MCLBYTES) 4444 return ENOBUFS; 4445 #endif 4446 4447 MGETHDR(m, M_DONTWAIT, MT_DATA); 4448 if (m && (max_linkhdr + tlen) > MHLEN) { 4449 MCLGET(m, M_DONTWAIT); 4450 if ((m->m_flags & M_EXT) == 0) { 4451 m_freem(m); 4452 m = NULL; 4453 } 4454 } 4455 if (m == NULL) 4456 return ENOBUFS; 4457 MCLAIM(m, &tcp_tx_mowner); 4458 4459 /* Fixup the mbuf. */ 4460 m->m_data += max_linkhdr; 4461 if (sc->sc_tp) 4462 tp = sc->sc_tp; 4463 m_reset_rcvif(m); 4464 memset(mtod(m, u_char *), 0, tlen); 4465 4466 switch (sc->sc_src.sa.sa_family) { 4467 case AF_INET: 4468 ip = mtod(m, struct ip *); 4469 ip->ip_v = 4; 4470 ip->ip_dst = sc->sc_src.sin.sin_addr; 4471 ip->ip_src = sc->sc_dst.sin.sin_addr; 4472 ip->ip_p = IPPROTO_TCP; 4473 th = (struct tcphdr *)(ip + 1); 4474 th->th_dport = sc->sc_src.sin.sin_port; 4475 th->th_sport = sc->sc_dst.sin.sin_port; 4476 break; 4477 #ifdef INET6 4478 case AF_INET6: 4479 ip6 = mtod(m, struct ip6_hdr *); 4480 ip6->ip6_vfc = IPV6_VERSION; 4481 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4482 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4483 ip6->ip6_nxt = IPPROTO_TCP; 4484 /* ip6_plen will be updated in ip6_output() */ 4485 th = (struct tcphdr *)(ip6 + 1); 4486 th->th_dport = sc->sc_src.sin6.sin6_port; 4487 th->th_sport = sc->sc_dst.sin6.sin6_port; 4488 break; 4489 #endif 4490 default: 4491 return ENOBUFS; 4492 } 4493 4494 th->th_seq = htonl(sc->sc_iss); 4495 th->th_ack = htonl(sc->sc_irs + 1); 4496 th->th_flags = TH_SYN|TH_ACK; 4497 th->th_win = htons(sc->sc_win); 4498 /* th_x2, th_sum, th_urp already 0 from memset */ 4499 4500 /* Tack on the TCP options. */ 4501 optp = (u_int8_t *)(th + 1); 4502 optlen = 0; 4503 *optp++ = TCPOPT_MAXSEG; 4504 *optp++ = TCPOLEN_MAXSEG; 4505 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4506 *optp++ = sc->sc_ourmaxseg & 0xff; 4507 optlen += TCPOLEN_MAXSEG; 4508 4509 if (sc->sc_request_r_scale != 15) { 4510 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4511 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4512 sc->sc_request_r_scale); 4513 optp += TCPOLEN_WINDOW + TCPOLEN_NOP; 4514 optlen += TCPOLEN_WINDOW + TCPOLEN_NOP; 4515 } 4516 4517 if (sc->sc_flags & SCF_SACK_PERMIT) { 4518 /* Let the peer know that we will SACK. */ 4519 *optp++ = TCPOPT_SACK_PERMITTED; 4520 *optp++ = TCPOLEN_SACK_PERMITTED; 4521 optlen += TCPOLEN_SACK_PERMITTED; 4522 } 4523 4524 if (sc->sc_flags & SCF_TIMESTAMP) { 4525 while (optlen % 4 != 2) { 4526 optlen += TCPOLEN_NOP; 4527 *optp++ = TCPOPT_NOP; 4528 } 4529 *optp++ = TCPOPT_TIMESTAMP; 4530 *optp++ = TCPOLEN_TIMESTAMP; 4531 u_int32_t *lp = (u_int32_t *)(optp); 4532 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4533 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4534 *lp = htonl(sc->sc_timestamp); 4535 optp += TCPOLEN_TIMESTAMP - 2; 4536 optlen += TCPOLEN_TIMESTAMP; 4537 } 4538 4539 #ifdef TCP_SIGNATURE 4540 if (sc->sc_flags & SCF_SIGNATURE) { 4541 sav = tcp_signature_getsav(m); 4542 if (sav == NULL) { 4543 if (m) 4544 m_freem(m); 4545 return (EPERM); 4546 } 4547 4548 *optp++ = TCPOPT_SIGNATURE; 4549 *optp++ = TCPOLEN_SIGNATURE; 4550 sigp = optp; 4551 memset(optp, 0, TCP_SIGLEN); 4552 optp += TCP_SIGLEN; 4553 optlen += TCPOLEN_SIGNATURE; 4554 4555 } 4556 #endif 4557 /* Terminate and pad TCP options to a 4 byte boundary. */ 4558 if (optlen % 4) { 4559 optlen += TCPOLEN_EOL; 4560 *optp++ = TCPOPT_EOL; 4561 } 4562 /* 4563 * According to RFC 793 (STD0007): 4564 * "The content of the header beyond the End-of-Option option 4565 * must be header padding (i.e., zero)." 4566 * and later: "The padding is composed of zeros." 4567 */ 4568 while (optlen % 4) { 4569 optlen += TCPOLEN_PAD; 4570 *optp++ = TCPOPT_PAD; 4571 } 4572 4573 /* compute the actual values now that we've added the options */ 4574 tlen = hlen + sizeof(struct tcphdr) + optlen; 4575 m->m_len = m->m_pkthdr.len = tlen; 4576 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4577 4578 #ifdef TCP_SIGNATURE 4579 if (sav) { 4580 (void)tcp_signature(m, th, hlen, sav, sigp); 4581 key_sa_recordxfer(sav, m); 4582 KEY_SA_UNREF(&sav); 4583 } 4584 #endif 4585 4586 /* 4587 * Send ECN SYN-ACK setup packet. 4588 * Routes can be asymetric, so, even if we receive a packet 4589 * with ECE and CWR set, we must not assume no one will block 4590 * the ECE packet we are about to send. 4591 */ 4592 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4593 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4594 th->th_flags |= TH_ECE; 4595 TCP_STATINC(TCP_STAT_ECN_SHS); 4596 4597 /* 4598 * draft-ietf-tcpm-ecnsyn-00.txt 4599 * 4600 * "[...] a TCP node MAY respond to an ECN-setup 4601 * SYN packet by setting ECT in the responding 4602 * ECN-setup SYN/ACK packet, indicating to routers 4603 * that the SYN/ACK packet is ECN-Capable. 4604 * This allows a congested router along the path 4605 * to mark the packet instead of dropping the 4606 * packet as an indication of congestion." 4607 * 4608 * "[...] There can be a great benefit in setting 4609 * an ECN-capable codepoint in SYN/ACK packets [...] 4610 * Congestion is most likely to occur in 4611 * the server-to-client direction. As a result, 4612 * setting an ECN-capable codepoint in SYN/ACK 4613 * packets can reduce the occurence of three-second 4614 * retransmit timeouts resulting from the drop 4615 * of SYN/ACK packets." 4616 * 4617 * Page 4 and 6, January 2006. 4618 */ 4619 4620 switch (sc->sc_src.sa.sa_family) { 4621 #ifdef INET 4622 case AF_INET: 4623 ip->ip_tos |= IPTOS_ECN_ECT0; 4624 break; 4625 #endif 4626 #ifdef INET6 4627 case AF_INET6: 4628 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4629 break; 4630 #endif 4631 } 4632 TCP_STATINC(TCP_STAT_ECN_ECT); 4633 } 4634 4635 4636 /* Compute the packet's checksum. */ 4637 switch (sc->sc_src.sa.sa_family) { 4638 case AF_INET: 4639 ip->ip_len = htons(tlen - hlen); 4640 th->th_sum = 0; 4641 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4642 break; 4643 #ifdef INET6 4644 case AF_INET6: 4645 ip6->ip6_plen = htons(tlen - hlen); 4646 th->th_sum = 0; 4647 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4648 break; 4649 #endif 4650 } 4651 4652 /* 4653 * Fill in some straggling IP bits. Note the stack expects 4654 * ip_len to be in host order, for convenience. 4655 */ 4656 switch (sc->sc_src.sa.sa_family) { 4657 #ifdef INET 4658 case AF_INET: 4659 ip->ip_len = htons(tlen); 4660 ip->ip_ttl = ip_defttl; 4661 /* XXX tos? */ 4662 break; 4663 #endif 4664 #ifdef INET6 4665 case AF_INET6: 4666 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4667 ip6->ip6_vfc |= IPV6_VERSION; 4668 ip6->ip6_plen = htons(tlen - hlen); 4669 /* ip6_hlim will be initialized afterwards */ 4670 /* XXX flowlabel? */ 4671 break; 4672 #endif 4673 } 4674 4675 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4676 tp = sc->sc_tp; 4677 4678 switch (sc->sc_src.sa.sa_family) { 4679 #ifdef INET 4680 case AF_INET: 4681 error = ip_output(m, sc->sc_ipopts, ro, 4682 (ip_mtudisc ? IP_MTUDISC : 0), 4683 NULL, tp ? tp->t_inpcb : NULL); 4684 break; 4685 #endif 4686 #ifdef INET6 4687 case AF_INET6: 4688 ip6->ip6_hlim = in6_selecthlim(NULL, 4689 (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp : NULL); 4690 rtcache_unref(rt, ro); 4691 4692 error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, 4693 tp ? tp->t_in6pcb : NULL, NULL); 4694 break; 4695 #endif 4696 default: 4697 error = EAFNOSUPPORT; 4698 break; 4699 } 4700 return (error); 4701 } 4702
Indexes created Sun Sep 21 20:09:37 GMT 2025