tcp_input.c revision 1.222
1 /* $NetBSD: tcp_input.c,v 1.222 2005/02/28 16:20:59 jonathan 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 The NetBSD Foundation, Inc. 74 * All rights reserved. 75 * 76 * This code is derived from software contributed to The NetBSD Foundation 77 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 78 * Facility, NASA Ames Research Center. 79 * 80 * Redistribution and use in source and binary forms, with or without 81 * modification, are permitted provided that the following conditions 82 * are met: 83 * 1. Redistributions of source code must retain the above copyright 84 * notice, this list of conditions and the following disclaimer. 85 * 2. Redistributions in binary form must reproduce the above copyright 86 * notice, this list of conditions and the following disclaimer in the 87 * documentation and/or other materials provided with the distribution. 88 * 3. All advertising materials mentioning features or use of this software 89 * must display the following acknowledgement: 90 * This product includes software developed by the NetBSD 91 * Foundation, Inc. and its contributors. 92 * 4. Neither the name of The NetBSD Foundation nor the names of its 93 * contributors may be used to endorse or promote products derived 94 * from this software without specific prior written permission. 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 1.222 2005/02/28 16:20:59 jonathan Exp $"); 152 153 #include "opt_inet.h" 154 #include "opt_ipsec.h" 155 #include "opt_inet_csum.h" 156 #include "opt_tcp_debug.h" 157 158 #include <sys/param.h> 159 #include <sys/systm.h> 160 #include <sys/malloc.h> 161 #include <sys/mbuf.h> 162 #include <sys/protosw.h> 163 #include <sys/socket.h> 164 #include <sys/socketvar.h> 165 #include <sys/errno.h> 166 #include <sys/syslog.h> 167 #include <sys/pool.h> 168 #include <sys/domain.h> 169 #include <sys/kernel.h> 170 #ifdef TCP_SIGNATURE 171 #include <sys/md5.h> 172 #endif 173 174 #include <net/if.h> 175 #include <net/route.h> 176 #include <net/if_types.h> 177 178 #include <netinet/in.h> 179 #include <netinet/in_systm.h> 180 #include <netinet/ip.h> 181 #include <netinet/in_pcb.h> 182 #include <netinet/in_var.h> 183 #include <netinet/ip_var.h> 184 185 #ifdef INET6 186 #ifndef INET 187 #include <netinet/in.h> 188 #endif 189 #include <netinet/ip6.h> 190 #include <netinet6/ip6_var.h> 191 #include <netinet6/in6_pcb.h> 192 #include <netinet6/ip6_var.h> 193 #include <netinet6/in6_var.h> 194 #include <netinet/icmp6.h> 195 #include <netinet6/nd6.h> 196 #endif 197 198 #ifndef INET6 199 /* always need ip6.h for IP6_EXTHDR_GET */ 200 #include <netinet/ip6.h> 201 #endif 202 203 #include <netinet/tcp.h> 204 #include <netinet/tcp_fsm.h> 205 #include <netinet/tcp_seq.h> 206 #include <netinet/tcp_timer.h> 207 #include <netinet/tcp_var.h> 208 #include <netinet/tcpip.h> 209 #include <netinet/tcp_debug.h> 210 211 #include <machine/stdarg.h> 212 213 #ifdef IPSEC 214 #include <netinet6/ipsec.h> 215 #include <netkey/key.h> 216 #endif /*IPSEC*/ 217 #ifdef INET6 218 #include "faith.h" 219 #if defined(NFAITH) && NFAITH > 0 220 #include <net/if_faith.h> 221 #endif 222 #endif /* IPSEC */ 223 224 #ifdef FAST_IPSEC 225 #include <netipsec/ipsec.h> 226 #include <netipsec/ipsec_var.h> /* XXX ipsecstat namespace */ 227 #include <netipsec/key.h> 228 #ifdef INET6 229 #include <netipsec/ipsec6.h> 230 #endif 231 #endif /* FAST_IPSEC*/ 232 233 int tcprexmtthresh = 3; 234 int tcp_log_refused; 235 236 static int tcp_rst_ppslim_count = 0; 237 static struct timeval tcp_rst_ppslim_last; 238 static int tcp_ackdrop_ppslim_count = 0; 239 static struct timeval tcp_ackdrop_ppslim_last; 240 241 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 242 243 /* for modulo comparisons of timestamps */ 244 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 245 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 246 247 /* 248 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 249 */ 250 #ifdef INET6 251 #define ND6_HINT(tp) \ 252 do { \ 253 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 && \ 254 tp->t_in6pcb->in6p_route.ro_rt) { \ 255 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \ 256 } \ 257 } while (/*CONSTCOND*/ 0) 258 #else 259 #define ND6_HINT(tp) 260 #endif 261 262 /* 263 * Macro to compute ACK transmission behavior. Delay the ACK unless 264 * we have already delayed an ACK (must send an ACK every two segments). 265 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 266 * option is enabled. 267 */ 268 #define TCP_SETUP_ACK(tp, th) \ 269 do { \ 270 if ((tp)->t_flags & TF_DELACK || \ 271 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 272 tp->t_flags |= TF_ACKNOW; \ 273 else \ 274 TCP_SET_DELACK(tp); \ 275 } while (/*CONSTCOND*/ 0) 276 277 /* 278 * Convert TCP protocol fields to host order for easier processing. 279 */ 280 #define TCP_FIELDS_TO_HOST(th) \ 281 do { \ 282 NTOHL((th)->th_seq); \ 283 NTOHL((th)->th_ack); \ 284 NTOHS((th)->th_win); \ 285 NTOHS((th)->th_urp); \ 286 } while (/*CONSTCOND*/ 0) 287 288 /* 289 * ... and reverse the above. 290 */ 291 #define TCP_FIELDS_TO_NET(th) \ 292 do { \ 293 HTONL((th)->th_seq); \ 294 HTONL((th)->th_ack); \ 295 HTONS((th)->th_win); \ 296 HTONS((th)->th_urp); \ 297 } while (/*CONSTCOND*/ 0) 298 299 #ifdef TCP_CSUM_COUNTERS 300 #include <sys/device.h> 301 302 extern struct evcnt tcp_hwcsum_ok; 303 extern struct evcnt tcp_hwcsum_bad; 304 extern struct evcnt tcp_hwcsum_data; 305 extern struct evcnt tcp_swcsum; 306 307 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 308 309 #else 310 311 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 312 313 #endif /* TCP_CSUM_COUNTERS */ 314 315 #ifdef TCP_REASS_COUNTERS 316 #include <sys/device.h> 317 318 extern struct evcnt tcp_reass_; 319 extern struct evcnt tcp_reass_empty; 320 extern struct evcnt tcp_reass_iteration[8]; 321 extern struct evcnt tcp_reass_prependfirst; 322 extern struct evcnt tcp_reass_prepend; 323 extern struct evcnt tcp_reass_insert; 324 extern struct evcnt tcp_reass_inserttail; 325 extern struct evcnt tcp_reass_append; 326 extern struct evcnt tcp_reass_appendtail; 327 extern struct evcnt tcp_reass_overlaptail; 328 extern struct evcnt tcp_reass_overlapfront; 329 extern struct evcnt tcp_reass_segdup; 330 extern struct evcnt tcp_reass_fragdup; 331 332 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 333 334 #else 335 336 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 337 338 #endif /* TCP_REASS_COUNTERS */ 339 340 #ifdef INET 341 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 342 #endif 343 #ifdef INET6 344 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 345 #endif 346 347 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 348 349 POOL_INIT(tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", NULL); 350 351 int 352 tcp_reass(struct tcpcb *tp, struct tcphdr *th, struct mbuf *m, int *tlen) 353 { 354 struct ipqent *p, *q, *nq, *tiqe = NULL; 355 struct socket *so = NULL; 356 int pkt_flags; 357 tcp_seq pkt_seq; 358 unsigned pkt_len; 359 u_long rcvpartdupbyte = 0; 360 u_long rcvoobyte; 361 #ifdef TCP_REASS_COUNTERS 362 u_int count = 0; 363 #endif 364 365 if (tp->t_inpcb) 366 so = tp->t_inpcb->inp_socket; 367 #ifdef INET6 368 else if (tp->t_in6pcb) 369 so = tp->t_in6pcb->in6p_socket; 370 #endif 371 372 TCP_REASS_LOCK_CHECK(tp); 373 374 /* 375 * Call with th==0 after become established to 376 * force pre-ESTABLISHED data up to user socket. 377 */ 378 if (th == 0) 379 goto present; 380 381 rcvoobyte = *tlen; 382 /* 383 * Copy these to local variables because the tcpiphdr 384 * gets munged while we are collapsing mbufs. 385 */ 386 pkt_seq = th->th_seq; 387 pkt_len = *tlen; 388 pkt_flags = th->th_flags; 389 390 TCP_REASS_COUNTER_INCR(&tcp_reass_); 391 392 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 393 /* 394 * When we miss a packet, the vast majority of time we get 395 * packets that follow it in order. So optimize for that. 396 */ 397 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 398 p->ipqe_len += pkt_len; 399 p->ipqe_flags |= pkt_flags; 400 m_cat(p->ipre_mlast, m); 401 TRAVERSE(p->ipre_mlast); 402 m = NULL; 403 tiqe = p; 404 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 405 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 406 goto skip_replacement; 407 } 408 /* 409 * While we're here, if the pkt is completely beyond 410 * anything we have, just insert it at the tail. 411 */ 412 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 413 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 414 goto insert_it; 415 } 416 } 417 418 q = TAILQ_FIRST(&tp->segq); 419 420 if (q != NULL) { 421 /* 422 * If this segment immediately precedes the first out-of-order 423 * block, simply slap the segment in front of it and (mostly) 424 * skip the complicated logic. 425 */ 426 if (pkt_seq + pkt_len == q->ipqe_seq) { 427 q->ipqe_seq = pkt_seq; 428 q->ipqe_len += pkt_len; 429 q->ipqe_flags |= pkt_flags; 430 m_cat(m, q->ipqe_m); 431 q->ipqe_m = m; 432 q->ipre_mlast = m; /* last mbuf may have changed */ 433 TRAVERSE(q->ipre_mlast); 434 tiqe = q; 435 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 436 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 437 goto skip_replacement; 438 } 439 } else { 440 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 441 } 442 443 /* 444 * Find a segment which begins after this one does. 445 */ 446 for (p = NULL; q != NULL; q = nq) { 447 nq = TAILQ_NEXT(q, ipqe_q); 448 #ifdef TCP_REASS_COUNTERS 449 count++; 450 #endif 451 /* 452 * If the received segment is just right after this 453 * fragment, merge the two together and then check 454 * for further overlaps. 455 */ 456 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 457 #ifdef TCPREASS_DEBUG 458 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 459 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 460 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 461 #endif 462 pkt_len += q->ipqe_len; 463 pkt_flags |= q->ipqe_flags; 464 pkt_seq = q->ipqe_seq; 465 m_cat(q->ipre_mlast, m); 466 TRAVERSE(q->ipre_mlast); 467 m = q->ipqe_m; 468 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 469 goto free_ipqe; 470 } 471 /* 472 * If the received segment is completely past this 473 * fragment, we need to go the next fragment. 474 */ 475 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 476 p = q; 477 continue; 478 } 479 /* 480 * If the fragment is past the received segment, 481 * it (or any following) can't be concatenated. 482 */ 483 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 484 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 485 break; 486 } 487 488 /* 489 * We've received all the data in this segment before. 490 * mark it as a duplicate and return. 491 */ 492 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 493 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 494 tcpstat.tcps_rcvduppack++; 495 tcpstat.tcps_rcvdupbyte += pkt_len; 496 tcp_new_dsack(tp, pkt_seq, pkt_len); 497 m_freem(m); 498 if (tiqe != NULL) 499 pool_put(&tcpipqent_pool, tiqe); 500 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 501 return (0); 502 } 503 /* 504 * Received segment completely overlaps this fragment 505 * so we drop the fragment (this keeps the temporal 506 * ordering of segments correct). 507 */ 508 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 509 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 510 rcvpartdupbyte += q->ipqe_len; 511 m_freem(q->ipqe_m); 512 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 513 goto free_ipqe; 514 } 515 /* 516 * RX'ed segment extends past the end of the 517 * fragment. Drop the overlapping bytes. Then 518 * merge the fragment and segment then treat as 519 * a longer received packet. 520 */ 521 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 522 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 523 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 524 #ifdef TCPREASS_DEBUG 525 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 526 tp, overlap, 527 pkt_seq, pkt_seq + pkt_len, pkt_len); 528 #endif 529 m_adj(m, overlap); 530 rcvpartdupbyte += overlap; 531 m_cat(q->ipre_mlast, m); 532 TRAVERSE(q->ipre_mlast); 533 m = q->ipqe_m; 534 pkt_seq = q->ipqe_seq; 535 pkt_len += q->ipqe_len - overlap; 536 rcvoobyte -= overlap; 537 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 538 goto free_ipqe; 539 } 540 /* 541 * RX'ed segment extends past the front of the 542 * fragment. Drop the overlapping bytes on the 543 * received packet. The packet will then be 544 * contatentated with this fragment a bit later. 545 */ 546 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 547 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 548 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 549 #ifdef TCPREASS_DEBUG 550 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 551 tp, overlap, 552 pkt_seq, pkt_seq + pkt_len, pkt_len); 553 #endif 554 m_adj(m, -overlap); 555 pkt_len -= overlap; 556 rcvpartdupbyte += overlap; 557 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 558 rcvoobyte -= overlap; 559 } 560 /* 561 * If the received segment immediates precedes this 562 * fragment then tack the fragment onto this segment 563 * and reinsert the data. 564 */ 565 if (q->ipqe_seq == pkt_seq + pkt_len) { 566 #ifdef TCPREASS_DEBUG 567 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 568 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 569 pkt_seq, pkt_seq + pkt_len, pkt_len); 570 #endif 571 pkt_len += q->ipqe_len; 572 pkt_flags |= q->ipqe_flags; 573 m_cat(m, q->ipqe_m); 574 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 575 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 576 if (tiqe == NULL) 577 tiqe = q; 578 else 579 pool_put(&tcpipqent_pool, q); 580 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 581 break; 582 } 583 /* 584 * If the fragment is before the segment, remember it. 585 * When this loop is terminated, p will contain the 586 * pointer to fragment that is right before the received 587 * segment. 588 */ 589 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 590 p = q; 591 592 continue; 593 594 /* 595 * This is a common operation. It also will allow 596 * to save doing a malloc/free in most instances. 597 */ 598 free_ipqe: 599 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 600 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 601 if (tiqe == NULL) 602 tiqe = q; 603 else 604 pool_put(&tcpipqent_pool, q); 605 } 606 607 #ifdef TCP_REASS_COUNTERS 608 if (count > 7) 609 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 610 else if (count > 0) 611 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 612 #endif 613 614 insert_it: 615 616 /* 617 * Allocate a new queue entry since the received segment did not 618 * collapse onto any other out-of-order block; thus we are allocating 619 * a new block. If it had collapsed, tiqe would not be NULL and 620 * we would be reusing it. 621 * XXX If we can't, just drop the packet. XXX 622 */ 623 if (tiqe == NULL) { 624 tiqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 625 if (tiqe == NULL) { 626 tcpstat.tcps_rcvmemdrop++; 627 m_freem(m); 628 return (0); 629 } 630 } 631 632 /* 633 * Update the counters. 634 */ 635 tcpstat.tcps_rcvoopack++; 636 tcpstat.tcps_rcvoobyte += rcvoobyte; 637 if (rcvpartdupbyte) { 638 tcpstat.tcps_rcvpartduppack++; 639 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 640 } 641 642 /* 643 * Insert the new fragment queue entry into both queues. 644 */ 645 tiqe->ipqe_m = m; 646 tiqe->ipre_mlast = m; 647 tiqe->ipqe_seq = pkt_seq; 648 tiqe->ipqe_len = pkt_len; 649 tiqe->ipqe_flags = pkt_flags; 650 if (p == NULL) { 651 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 652 #ifdef TCPREASS_DEBUG 653 if (tiqe->ipqe_seq != tp->rcv_nxt) 654 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 655 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 656 #endif 657 } else { 658 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 659 #ifdef TCPREASS_DEBUG 660 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 661 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 662 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 663 #endif 664 } 665 666 skip_replacement: 667 668 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 669 670 present: 671 /* 672 * Present data to user, advancing rcv_nxt through 673 * completed sequence space. 674 */ 675 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 676 return (0); 677 q = TAILQ_FIRST(&tp->segq); 678 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 679 return (0); 680 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 681 return (0); 682 683 tp->rcv_nxt += q->ipqe_len; 684 pkt_flags = q->ipqe_flags & TH_FIN; 685 ND6_HINT(tp); 686 687 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 688 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 689 if (so->so_state & SS_CANTRCVMORE) 690 m_freem(q->ipqe_m); 691 else 692 sbappendstream(&so->so_rcv, q->ipqe_m); 693 pool_put(&tcpipqent_pool, q); 694 sorwakeup(so); 695 return (pkt_flags); 696 } 697 698 #ifdef INET6 699 int 700 tcp6_input(struct mbuf **mp, int *offp, int proto) 701 { 702 struct mbuf *m = *mp; 703 704 /* 705 * draft-itojun-ipv6-tcp-to-anycast 706 * better place to put this in? 707 */ 708 if (m->m_flags & M_ANYCAST6) { 709 struct ip6_hdr *ip6; 710 if (m->m_len < sizeof(struct ip6_hdr)) { 711 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 712 tcpstat.tcps_rcvshort++; 713 return IPPROTO_DONE; 714 } 715 } 716 ip6 = mtod(m, struct ip6_hdr *); 717 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 718 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 719 return IPPROTO_DONE; 720 } 721 722 tcp_input(m, *offp, proto); 723 return IPPROTO_DONE; 724 } 725 #endif 726 727 #ifdef INET 728 static void 729 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 730 { 731 char src[4*sizeof "123"]; 732 char dst[4*sizeof "123"]; 733 734 if (ip) { 735 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 736 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 737 } 738 else { 739 strlcpy(src, "(unknown)", sizeof(src)); 740 strlcpy(dst, "(unknown)", sizeof(dst)); 741 } 742 log(LOG_INFO, 743 "Connection attempt to TCP %s:%d from %s:%d\n", 744 dst, ntohs(th->th_dport), 745 src, ntohs(th->th_sport)); 746 } 747 #endif 748 749 #ifdef INET6 750 static void 751 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 752 { 753 char src[INET6_ADDRSTRLEN]; 754 char dst[INET6_ADDRSTRLEN]; 755 756 if (ip6) { 757 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 758 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 759 } 760 else { 761 strlcpy(src, "(unknown v6)", sizeof(src)); 762 strlcpy(dst, "(unknown v6)", sizeof(dst)); 763 } 764 log(LOG_INFO, 765 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 766 dst, ntohs(th->th_dport), 767 src, ntohs(th->th_sport)); 768 } 769 #endif 770 771 /* 772 * Checksum extended TCP header and data. 773 */ 774 int 775 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, int toff, 776 int off, int tlen) 777 { 778 779 /* 780 * XXX it's better to record and check if this mbuf is 781 * already checked. 782 */ 783 784 switch (af) { 785 #ifdef INET 786 case AF_INET: 787 switch (m->m_pkthdr.csum_flags & 788 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 789 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 790 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 791 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 792 goto badcsum; 793 794 case M_CSUM_TCPv4|M_CSUM_DATA: { 795 u_int32_t hw_csum = m->m_pkthdr.csum_data; 796 797 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 798 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 799 const struct ip *ip = 800 mtod(m, const struct ip *); 801 802 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 803 ip->ip_dst.s_addr, 804 htons(hw_csum + tlen + off + IPPROTO_TCP)); 805 } 806 if ((hw_csum ^ 0xffff) != 0) 807 goto badcsum; 808 break; 809 } 810 811 case M_CSUM_TCPv4: 812 /* Checksum was okay. */ 813 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 814 break; 815 816 default: 817 /* 818 * Must compute it ourselves. Maybe skip checksum 819 * on loopback interfaces. 820 */ 821 if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & 822 IFF_LOOPBACK) || 823 tcp_do_loopback_cksum)) { 824 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 825 if (in4_cksum(m, IPPROTO_TCP, toff, 826 tlen + off) != 0) 827 goto badcsum; 828 } 829 break; 830 } 831 break; 832 #endif /* INET4 */ 833 834 #ifdef INET6 835 case AF_INET6: 836 if (__predict_true((m->m_flags & M_LOOP) == 0 || 837 tcp_do_loopback_cksum)) { 838 if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 839 goto badcsum; 840 } 841 break; 842 #endif /* INET6 */ 843 } 844 845 return 0; 846 847 badcsum: 848 tcpstat.tcps_rcvbadsum++; 849 return -1; 850 } 851 852 /* 853 * TCP input routine, follows pages 65-76 of the 854 * protocol specification dated September, 1981 very closely. 855 */ 856 void 857 tcp_input(struct mbuf *m, ...) 858 { 859 struct tcphdr *th; 860 struct ip *ip; 861 struct inpcb *inp; 862 #ifdef INET6 863 struct ip6_hdr *ip6; 864 struct in6pcb *in6p; 865 #endif 866 u_int8_t *optp = NULL; 867 int optlen = 0; 868 int len, tlen, toff, hdroptlen = 0; 869 struct tcpcb *tp = 0; 870 int tiflags; 871 struct socket *so = NULL; 872 int todrop, dupseg, acked, ourfinisacked, needoutput = 0; 873 #ifdef TCP_DEBUG 874 short ostate = 0; 875 #endif 876 int iss = 0; 877 u_long tiwin; 878 struct tcp_opt_info opti; 879 int off, iphlen; 880 va_list ap; 881 int af; /* af on the wire */ 882 struct mbuf *tcp_saveti = NULL; 883 884 MCLAIM(m, &tcp_rx_mowner); 885 va_start(ap, m); 886 toff = va_arg(ap, int); 887 (void)va_arg(ap, int); /* ignore value, advance ap */ 888 va_end(ap); 889 890 tcpstat.tcps_rcvtotal++; 891 892 bzero(&opti, sizeof(opti)); 893 opti.ts_present = 0; 894 opti.maxseg = 0; 895 896 /* 897 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 898 * 899 * TCP is, by definition, unicast, so we reject all 900 * multicast outright. 901 * 902 * Note, there are additional src/dst address checks in 903 * the AF-specific code below. 904 */ 905 if (m->m_flags & (M_BCAST|M_MCAST)) { 906 /* XXX stat */ 907 goto drop; 908 } 909 #ifdef INET6 910 if (m->m_flags & M_ANYCAST6) { 911 /* XXX stat */ 912 goto drop; 913 } 914 #endif 915 916 /* 917 * Get IP and TCP header. 918 * Note: IP leaves IP header in first mbuf. 919 */ 920 ip = mtod(m, struct ip *); 921 #ifdef INET6 922 ip6 = NULL; 923 #endif 924 switch (ip->ip_v) { 925 #ifdef INET 926 case 4: 927 af = AF_INET; 928 iphlen = sizeof(struct ip); 929 ip = mtod(m, struct ip *); 930 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 931 sizeof(struct tcphdr)); 932 if (th == NULL) { 933 tcpstat.tcps_rcvshort++; 934 return; 935 } 936 /* We do the checksum after PCB lookup... */ 937 len = ntohs(ip->ip_len); 938 tlen = len - toff; 939 break; 940 #endif 941 #ifdef INET6 942 case 6: 943 ip = NULL; 944 iphlen = sizeof(struct ip6_hdr); 945 af = AF_INET6; 946 ip6 = mtod(m, struct ip6_hdr *); 947 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 948 sizeof(struct tcphdr)); 949 if (th == NULL) { 950 tcpstat.tcps_rcvshort++; 951 return; 952 } 953 954 /* Be proactive about malicious use of IPv4 mapped address */ 955 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 956 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 957 /* XXX stat */ 958 goto drop; 959 } 960 961 /* 962 * Be proactive about unspecified IPv6 address in source. 963 * As we use all-zero to indicate unbounded/unconnected pcb, 964 * unspecified IPv6 address can be used to confuse us. 965 * 966 * Note that packets with unspecified IPv6 destination is 967 * already dropped in ip6_input. 968 */ 969 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 970 /* XXX stat */ 971 goto drop; 972 } 973 974 /* 975 * Make sure destination address is not multicast. 976 * Source address checked in ip6_input(). 977 */ 978 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 979 /* XXX stat */ 980 goto drop; 981 } 982 983 /* We do the checksum after PCB lookup... */ 984 len = m->m_pkthdr.len; 985 tlen = len - toff; 986 break; 987 #endif 988 default: 989 m_freem(m); 990 return; 991 } 992 993 KASSERT(TCP_HDR_ALIGNED_P(th)); 994 995 /* 996 * Check that TCP offset makes sense, 997 * pull out TCP options and adjust length. XXX 998 */ 999 off = th->th_off << 2; 1000 if (off < sizeof (struct tcphdr) || off > tlen) { 1001 tcpstat.tcps_rcvbadoff++; 1002 goto drop; 1003 } 1004 tlen -= off; 1005 1006 /* 1007 * tcp_input() has been modified to use tlen to mean the TCP data 1008 * length throughout the function. Other functions can use 1009 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1010 * rja 1011 */ 1012 1013 if (off > sizeof (struct tcphdr)) { 1014 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1015 if (th == NULL) { 1016 tcpstat.tcps_rcvshort++; 1017 return; 1018 } 1019 /* 1020 * NOTE: ip/ip6 will not be affected by m_pulldown() 1021 * (as they're before toff) and we don't need to update those. 1022 */ 1023 KASSERT(TCP_HDR_ALIGNED_P(th)); 1024 optlen = off - sizeof (struct tcphdr); 1025 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1026 /* 1027 * Do quick retrieval of timestamp options ("options 1028 * prediction?"). If timestamp is the only option and it's 1029 * formatted as recommended in RFC 1323 appendix A, we 1030 * quickly get the values now and not bother calling 1031 * tcp_dooptions(), etc. 1032 */ 1033 if ((optlen == TCPOLEN_TSTAMP_APPA || 1034 (optlen > TCPOLEN_TSTAMP_APPA && 1035 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1036 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1037 (th->th_flags & TH_SYN) == 0) { 1038 opti.ts_present = 1; 1039 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1040 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1041 optp = NULL; /* we've parsed the options */ 1042 } 1043 } 1044 tiflags = th->th_flags; 1045 1046 /* 1047 * Locate pcb for segment. 1048 */ 1049 findpcb: 1050 inp = NULL; 1051 #ifdef INET6 1052 in6p = NULL; 1053 #endif 1054 switch (af) { 1055 #ifdef INET 1056 case AF_INET: 1057 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1058 ip->ip_dst, th->th_dport); 1059 if (inp == 0) { 1060 ++tcpstat.tcps_pcbhashmiss; 1061 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1062 } 1063 #ifdef INET6 1064 if (inp == 0) { 1065 struct in6_addr s, d; 1066 1067 /* mapped addr case */ 1068 bzero(&s, sizeof(s)); 1069 s.s6_addr16[5] = htons(0xffff); 1070 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1071 bzero(&d, sizeof(d)); 1072 d.s6_addr16[5] = htons(0xffff); 1073 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1074 in6p = in6_pcblookup_connect(&tcbtable, &s, 1075 th->th_sport, &d, th->th_dport, 0); 1076 if (in6p == 0) { 1077 ++tcpstat.tcps_pcbhashmiss; 1078 in6p = in6_pcblookup_bind(&tcbtable, &d, 1079 th->th_dport, 0); 1080 } 1081 } 1082 #endif 1083 #ifndef INET6 1084 if (inp == 0) 1085 #else 1086 if (inp == 0 && in6p == 0) 1087 #endif 1088 { 1089 ++tcpstat.tcps_noport; 1090 if (tcp_log_refused && 1091 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1092 tcp4_log_refused(ip, th); 1093 } 1094 TCP_FIELDS_TO_HOST(th); 1095 goto dropwithreset_ratelim; 1096 } 1097 #if defined(IPSEC) || defined(FAST_IPSEC) 1098 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1099 ipsec4_in_reject(m, inp)) { 1100 ipsecstat.in_polvio++; 1101 goto drop; 1102 } 1103 #ifdef INET6 1104 else if (in6p && 1105 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1106 ipsec4_in_reject_so(m, in6p->in6p_socket)) { 1107 ipsecstat.in_polvio++; 1108 goto drop; 1109 } 1110 #endif 1111 #endif /*IPSEC*/ 1112 break; 1113 #endif /*INET*/ 1114 #ifdef INET6 1115 case AF_INET6: 1116 { 1117 int faith; 1118 1119 #if defined(NFAITH) && NFAITH > 0 1120 faith = faithprefix(&ip6->ip6_dst); 1121 #else 1122 faith = 0; 1123 #endif 1124 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1125 th->th_sport, &ip6->ip6_dst, th->th_dport, faith); 1126 if (in6p == NULL) { 1127 ++tcpstat.tcps_pcbhashmiss; 1128 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1129 th->th_dport, faith); 1130 } 1131 if (in6p == NULL) { 1132 ++tcpstat.tcps_noport; 1133 if (tcp_log_refused && 1134 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1135 tcp6_log_refused(ip6, th); 1136 } 1137 TCP_FIELDS_TO_HOST(th); 1138 goto dropwithreset_ratelim; 1139 } 1140 #if defined(IPSEC) || defined(FAST_IPSEC) 1141 if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1142 ipsec6_in_reject(m, in6p)) { 1143 ipsec6stat.in_polvio++; 1144 goto drop; 1145 } 1146 #endif /*IPSEC*/ 1147 break; 1148 } 1149 #endif 1150 } 1151 1152 /* 1153 * If the state is CLOSED (i.e., TCB does not exist) then 1154 * all data in the incoming segment is discarded. 1155 * If the TCB exists but is in CLOSED state, it is embryonic, 1156 * but should either do a listen or a connect soon. 1157 */ 1158 tp = NULL; 1159 so = NULL; 1160 if (inp) { 1161 tp = intotcpcb(inp); 1162 so = inp->inp_socket; 1163 } 1164 #ifdef INET6 1165 else if (in6p) { 1166 tp = in6totcpcb(in6p); 1167 so = in6p->in6p_socket; 1168 } 1169 #endif 1170 if (tp == 0) { 1171 TCP_FIELDS_TO_HOST(th); 1172 goto dropwithreset_ratelim; 1173 } 1174 if (tp->t_state == TCPS_CLOSED) 1175 goto drop; 1176 1177 /* 1178 * Checksum extended TCP header and data. 1179 */ 1180 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1181 goto badcsum; 1182 1183 TCP_FIELDS_TO_HOST(th); 1184 1185 /* Unscale the window into a 32-bit value. */ 1186 if ((tiflags & TH_SYN) == 0) 1187 tiwin = th->th_win << tp->snd_scale; 1188 else 1189 tiwin = th->th_win; 1190 1191 #ifdef INET6 1192 /* save packet options if user wanted */ 1193 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1194 if (in6p->in6p_options) { 1195 m_freem(in6p->in6p_options); 1196 in6p->in6p_options = 0; 1197 } 1198 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1199 } 1200 #endif 1201 1202 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1203 union syn_cache_sa src; 1204 union syn_cache_sa dst; 1205 1206 bzero(&src, sizeof(src)); 1207 bzero(&dst, sizeof(dst)); 1208 switch (af) { 1209 #ifdef INET 1210 case AF_INET: 1211 src.sin.sin_len = sizeof(struct sockaddr_in); 1212 src.sin.sin_family = AF_INET; 1213 src.sin.sin_addr = ip->ip_src; 1214 src.sin.sin_port = th->th_sport; 1215 1216 dst.sin.sin_len = sizeof(struct sockaddr_in); 1217 dst.sin.sin_family = AF_INET; 1218 dst.sin.sin_addr = ip->ip_dst; 1219 dst.sin.sin_port = th->th_dport; 1220 break; 1221 #endif 1222 #ifdef INET6 1223 case AF_INET6: 1224 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1225 src.sin6.sin6_family = AF_INET6; 1226 src.sin6.sin6_addr = ip6->ip6_src; 1227 src.sin6.sin6_port = th->th_sport; 1228 1229 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1230 dst.sin6.sin6_family = AF_INET6; 1231 dst.sin6.sin6_addr = ip6->ip6_dst; 1232 dst.sin6.sin6_port = th->th_dport; 1233 break; 1234 #endif /* INET6 */ 1235 default: 1236 goto badsyn; /*sanity*/ 1237 } 1238 1239 if (so->so_options & SO_DEBUG) { 1240 #ifdef TCP_DEBUG 1241 ostate = tp->t_state; 1242 #endif 1243 1244 tcp_saveti = NULL; 1245 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1246 goto nosave; 1247 1248 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1249 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1250 if (!tcp_saveti) 1251 goto nosave; 1252 } else { 1253 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1254 if (!tcp_saveti) 1255 goto nosave; 1256 MCLAIM(m, &tcp_mowner); 1257 tcp_saveti->m_len = iphlen; 1258 m_copydata(m, 0, iphlen, 1259 mtod(tcp_saveti, caddr_t)); 1260 } 1261 1262 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1263 m_freem(tcp_saveti); 1264 tcp_saveti = NULL; 1265 } else { 1266 tcp_saveti->m_len += sizeof(struct tcphdr); 1267 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1268 sizeof(struct tcphdr)); 1269 } 1270 nosave:; 1271 } 1272 if (so->so_options & SO_ACCEPTCONN) { 1273 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1274 if (tiflags & TH_RST) { 1275 syn_cache_reset(&src.sa, &dst.sa, th); 1276 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1277 (TH_ACK|TH_SYN)) { 1278 /* 1279 * Received a SYN,ACK. This should 1280 * never happen while we are in 1281 * LISTEN. Send an RST. 1282 */ 1283 goto badsyn; 1284 } else if (tiflags & TH_ACK) { 1285 so = syn_cache_get(&src.sa, &dst.sa, 1286 th, toff, tlen, so, m); 1287 if (so == NULL) { 1288 /* 1289 * We don't have a SYN for 1290 * this ACK; send an RST. 1291 */ 1292 goto badsyn; 1293 } else if (so == 1294 (struct socket *)(-1)) { 1295 /* 1296 * We were unable to create 1297 * the connection. If the 1298 * 3-way handshake was 1299 * completed, and RST has 1300 * been sent to the peer. 1301 * Since the mbuf might be 1302 * in use for the reply, 1303 * do not free it. 1304 */ 1305 m = NULL; 1306 } else { 1307 /* 1308 * We have created a 1309 * full-blown connection. 1310 */ 1311 tp = NULL; 1312 inp = NULL; 1313 #ifdef INET6 1314 in6p = NULL; 1315 #endif 1316 switch (so->so_proto->pr_domain->dom_family) { 1317 #ifdef INET 1318 case AF_INET: 1319 inp = sotoinpcb(so); 1320 tp = intotcpcb(inp); 1321 break; 1322 #endif 1323 #ifdef INET6 1324 case AF_INET6: 1325 in6p = sotoin6pcb(so); 1326 tp = in6totcpcb(in6p); 1327 break; 1328 #endif 1329 } 1330 if (tp == NULL) 1331 goto badsyn; /*XXX*/ 1332 tiwin <<= tp->snd_scale; 1333 goto after_listen; 1334 } 1335 } else { 1336 /* 1337 * None of RST, SYN or ACK was set. 1338 * This is an invalid packet for a 1339 * TCB in LISTEN state. Send a RST. 1340 */ 1341 goto badsyn; 1342 } 1343 } else { 1344 /* 1345 * Received a SYN. 1346 */ 1347 1348 #ifdef INET6 1349 /* 1350 * If deprecated address is forbidden, we do 1351 * not accept SYN to deprecated interface 1352 * address to prevent any new inbound 1353 * connection from getting established. 1354 * When we do not accept SYN, we send a TCP 1355 * RST, with deprecated source address (instead 1356 * of dropping it). We compromise it as it is 1357 * much better for peer to send a RST, and 1358 * RST will be the final packet for the 1359 * exchange. 1360 * 1361 * If we do not forbid deprecated addresses, we 1362 * accept the SYN packet. RFC2462 does not 1363 * suggest dropping SYN in this case. 1364 * If we decipher RFC2462 5.5.4, it says like 1365 * this: 1366 * 1. use of deprecated addr with existing 1367 * communication is okay - "SHOULD continue 1368 * to be used" 1369 * 2. use of it with new communication: 1370 * (2a) "SHOULD NOT be used if alternate 1371 * address with sufficient scope is 1372 * available" 1373 * (2b) nothing mentioned otherwise. 1374 * Here we fall into (2b) case as we have no 1375 * choice in our source address selection - we 1376 * must obey the peer. 1377 * 1378 * The wording in RFC2462 is confusing, and 1379 * there are multiple description text for 1380 * deprecated address handling - worse, they 1381 * are not exactly the same. I believe 5.5.4 1382 * is the best one, so we follow 5.5.4. 1383 */ 1384 if (af == AF_INET6 && !ip6_use_deprecated) { 1385 struct in6_ifaddr *ia6; 1386 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1387 &ip6->ip6_dst)) && 1388 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1389 tp = NULL; 1390 goto dropwithreset; 1391 } 1392 } 1393 #endif 1394 1395 #ifdef IPSEC 1396 switch (af) { 1397 #ifdef INET 1398 case AF_INET: 1399 if (ipsec4_in_reject_so(m, so)) { 1400 ipsecstat.in_polvio++; 1401 tp = NULL; 1402 goto dropwithreset; 1403 } 1404 break; 1405 #endif 1406 #ifdef INET6 1407 case AF_INET6: 1408 if (ipsec6_in_reject_so(m, so)) { 1409 ipsec6stat.in_polvio++; 1410 tp = NULL; 1411 goto dropwithreset; 1412 } 1413 break; 1414 #endif 1415 } 1416 #endif 1417 1418 /* 1419 * LISTEN socket received a SYN 1420 * from itself? This can't possibly 1421 * be valid; drop the packet. 1422 */ 1423 if (th->th_sport == th->th_dport) { 1424 int i; 1425 1426 switch (af) { 1427 #ifdef INET 1428 case AF_INET: 1429 i = in_hosteq(ip->ip_src, ip->ip_dst); 1430 break; 1431 #endif 1432 #ifdef INET6 1433 case AF_INET6: 1434 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1435 break; 1436 #endif 1437 default: 1438 i = 1; 1439 } 1440 if (i) { 1441 tcpstat.tcps_badsyn++; 1442 goto drop; 1443 } 1444 } 1445 1446 /* 1447 * SYN looks ok; create compressed TCP 1448 * state for it. 1449 */ 1450 if (so->so_qlen <= so->so_qlimit && 1451 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1452 so, m, optp, optlen, &opti)) 1453 m = NULL; 1454 } 1455 goto drop; 1456 } 1457 } 1458 1459 after_listen: 1460 #ifdef DIAGNOSTIC 1461 /* 1462 * Should not happen now that all embryonic connections 1463 * are handled with compressed state. 1464 */ 1465 if (tp->t_state == TCPS_LISTEN) 1466 panic("tcp_input: TCPS_LISTEN"); 1467 #endif 1468 1469 /* 1470 * Segment received on connection. 1471 * Reset idle time and keep-alive timer. 1472 */ 1473 tp->t_rcvtime = tcp_now; 1474 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1475 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1476 1477 /* 1478 * Process options. 1479 */ 1480 #ifdef TCP_SIGNATURE 1481 if (optp || (tp->t_flags & TF_SIGNATURE)) 1482 #else 1483 if (optp) 1484 #endif 1485 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1486 goto drop; 1487 1488 if (TCP_SACK_ENABLED(tp)) { 1489 tcp_del_sackholes(tp, th); 1490 } 1491 1492 if (opti.ts_present && opti.ts_ecr) { 1493 /* 1494 * Calculate the RTT from the returned time stamp and the 1495 * connection's time base. If the time stamp is later than 1496 * the current time, or is extremely old, fall back to non-1323 1497 * RTT calculation. Since ts_ecr is unsigned, we can test both 1498 * at the same time. 1499 */ 1500 opti.ts_ecr = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1501 if (opti.ts_ecr > TCP_PAWS_IDLE) 1502 opti.ts_ecr = 0; 1503 } 1504 1505 /* 1506 * Header prediction: check for the two common cases 1507 * of a uni-directional data xfer. If the packet has 1508 * no control flags, is in-sequence, the window didn't 1509 * change and we're not retransmitting, it's a 1510 * candidate. If the length is zero and the ack moved 1511 * forward, we're the sender side of the xfer. Just 1512 * free the data acked & wake any higher level process 1513 * that was blocked waiting for space. If the length 1514 * is non-zero and the ack didn't move, we're the 1515 * receiver side. If we're getting packets in-order 1516 * (the reassembly queue is empty), add the data to 1517 * the socket buffer and note that we need a delayed ack. 1518 */ 1519 if (tp->t_state == TCPS_ESTABLISHED && 1520 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1521 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1522 th->th_seq == tp->rcv_nxt && 1523 tiwin && tiwin == tp->snd_wnd && 1524 tp->snd_nxt == tp->snd_max) { 1525 1526 /* 1527 * If last ACK falls within this segment's sequence numbers, 1528 * record the timestamp. 1529 */ 1530 if (opti.ts_present && 1531 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1532 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1533 tp->ts_recent_age = tcp_now; 1534 tp->ts_recent = opti.ts_val; 1535 } 1536 1537 if (tlen == 0) { 1538 /* Ack prediction. */ 1539 if (SEQ_GT(th->th_ack, tp->snd_una) && 1540 SEQ_LEQ(th->th_ack, tp->snd_max) && 1541 tp->snd_cwnd >= tp->snd_wnd && 1542 tp->t_partialacks < 0) { 1543 /* 1544 * this is a pure ack for outstanding data. 1545 */ 1546 ++tcpstat.tcps_predack; 1547 if (opti.ts_present && opti.ts_ecr) 1548 tcp_xmit_timer(tp, opti.ts_ecr); 1549 else if (tp->t_rtttime && 1550 SEQ_GT(th->th_ack, tp->t_rtseq)) 1551 tcp_xmit_timer(tp, 1552 tcp_now - tp->t_rtttime); 1553 acked = th->th_ack - tp->snd_una; 1554 tcpstat.tcps_rcvackpack++; 1555 tcpstat.tcps_rcvackbyte += acked; 1556 ND6_HINT(tp); 1557 1558 if (acked > (tp->t_lastoff - tp->t_inoff)) 1559 tp->t_lastm = NULL; 1560 sbdrop(&so->so_snd, acked); 1561 tp->t_lastoff -= acked; 1562 1563 tp->snd_una = th->th_ack; 1564 tp->snd_fack = tp->snd_una; 1565 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1566 tp->snd_high = tp->snd_una; 1567 m_freem(m); 1568 1569 /* 1570 * If all outstanding data are acked, stop 1571 * retransmit timer, otherwise restart timer 1572 * using current (possibly backed-off) value. 1573 * If process is waiting for space, 1574 * wakeup/selwakeup/signal. If data 1575 * are ready to send, let tcp_output 1576 * decide between more output or persist. 1577 */ 1578 if (tp->snd_una == tp->snd_max) 1579 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1580 else if (TCP_TIMER_ISARMED(tp, 1581 TCPT_PERSIST) == 0) 1582 TCP_TIMER_ARM(tp, TCPT_REXMT, 1583 tp->t_rxtcur); 1584 1585 sowwakeup(so); 1586 if (so->so_snd.sb_cc) 1587 (void) tcp_output(tp); 1588 if (tcp_saveti) 1589 m_freem(tcp_saveti); 1590 return; 1591 } 1592 } else if (th->th_ack == tp->snd_una && 1593 TAILQ_FIRST(&tp->segq) == NULL && 1594 tlen <= sbspace(&so->so_rcv)) { 1595 /* 1596 * this is a pure, in-sequence data packet 1597 * with nothing on the reassembly queue and 1598 * we have enough buffer space to take it. 1599 */ 1600 ++tcpstat.tcps_preddat; 1601 tp->rcv_sack_num = 0; 1602 tp->rcv_nxt += tlen; 1603 tcpstat.tcps_rcvpack++; 1604 tcpstat.tcps_rcvbyte += tlen; 1605 ND6_HINT(tp); 1606 /* 1607 * Drop TCP, IP headers and TCP options then add data 1608 * to socket buffer. 1609 */ 1610 if (so->so_state & SS_CANTRCVMORE) 1611 m_freem(m); 1612 else { 1613 m_adj(m, toff + off); 1614 sbappendstream(&so->so_rcv, m); 1615 } 1616 sorwakeup(so); 1617 TCP_SETUP_ACK(tp, th); 1618 if (tp->t_flags & TF_ACKNOW) 1619 (void) tcp_output(tp); 1620 if (tcp_saveti) 1621 m_freem(tcp_saveti); 1622 return; 1623 } 1624 } 1625 1626 /* 1627 * Compute mbuf offset to TCP data segment. 1628 */ 1629 hdroptlen = toff + off; 1630 1631 /* 1632 * Calculate amount of space in receive window, 1633 * and then do TCP input processing. 1634 * Receive window is amount of space in rcv queue, 1635 * but not less than advertised window. 1636 */ 1637 { int win; 1638 1639 win = sbspace(&so->so_rcv); 1640 if (win < 0) 1641 win = 0; 1642 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1643 } 1644 1645 switch (tp->t_state) { 1646 case TCPS_LISTEN: 1647 /* 1648 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1649 */ 1650 if (m->m_flags & (M_BCAST|M_MCAST)) 1651 goto drop; 1652 switch (af) { 1653 #ifdef INET6 1654 case AF_INET6: 1655 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1656 goto drop; 1657 break; 1658 #endif /* INET6 */ 1659 case AF_INET: 1660 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1661 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1662 goto drop; 1663 break; 1664 } 1665 break; 1666 1667 /* 1668 * If the state is SYN_SENT: 1669 * if seg contains an ACK, but not for our SYN, drop the input. 1670 * if seg contains a RST, then drop the connection. 1671 * if seg does not contain SYN, then drop it. 1672 * Otherwise this is an acceptable SYN segment 1673 * initialize tp->rcv_nxt and tp->irs 1674 * if seg contains ack then advance tp->snd_una 1675 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1676 * arrange for segment to be acked (eventually) 1677 * continue processing rest of data/controls, beginning with URG 1678 */ 1679 case TCPS_SYN_SENT: 1680 if ((tiflags & TH_ACK) && 1681 (SEQ_LEQ(th->th_ack, tp->iss) || 1682 SEQ_GT(th->th_ack, tp->snd_max))) 1683 goto dropwithreset; 1684 if (tiflags & TH_RST) { 1685 if (tiflags & TH_ACK) 1686 tp = tcp_drop(tp, ECONNREFUSED); 1687 goto drop; 1688 } 1689 if ((tiflags & TH_SYN) == 0) 1690 goto drop; 1691 if (tiflags & TH_ACK) { 1692 tp->snd_una = th->th_ack; 1693 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1694 tp->snd_nxt = tp->snd_una; 1695 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1696 tp->snd_high = tp->snd_una; 1697 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1698 } 1699 tp->irs = th->th_seq; 1700 tcp_rcvseqinit(tp); 1701 tp->t_flags |= TF_ACKNOW; 1702 tcp_mss_from_peer(tp, opti.maxseg); 1703 1704 /* 1705 * Initialize the initial congestion window. If we 1706 * had to retransmit the SYN, we must initialize cwnd 1707 * to 1 segment (i.e. the Loss Window). 1708 */ 1709 if (tp->t_flags & TF_SYN_REXMT) 1710 tp->snd_cwnd = tp->t_peermss; 1711 else { 1712 int ss = tcp_init_win; 1713 #ifdef INET 1714 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1715 ss = tcp_init_win_local; 1716 #endif 1717 #ifdef INET6 1718 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1719 ss = tcp_init_win_local; 1720 #endif 1721 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1722 } 1723 1724 tcp_rmx_rtt(tp); 1725 if (tiflags & TH_ACK) { 1726 tcpstat.tcps_connects++; 1727 soisconnected(so); 1728 tcp_established(tp); 1729 /* Do window scaling on this connection? */ 1730 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1731 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1732 tp->snd_scale = tp->requested_s_scale; 1733 tp->rcv_scale = tp->request_r_scale; 1734 } 1735 TCP_REASS_LOCK(tp); 1736 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1737 TCP_REASS_UNLOCK(tp); 1738 /* 1739 * if we didn't have to retransmit the SYN, 1740 * use its rtt as our initial srtt & rtt var. 1741 */ 1742 if (tp->t_rtttime) 1743 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1744 } else 1745 tp->t_state = TCPS_SYN_RECEIVED; 1746 1747 /* 1748 * Advance th->th_seq to correspond to first data byte. 1749 * If data, trim to stay within window, 1750 * dropping FIN if necessary. 1751 */ 1752 th->th_seq++; 1753 if (tlen > tp->rcv_wnd) { 1754 todrop = tlen - tp->rcv_wnd; 1755 m_adj(m, -todrop); 1756 tlen = tp->rcv_wnd; 1757 tiflags &= ~TH_FIN; 1758 tcpstat.tcps_rcvpackafterwin++; 1759 tcpstat.tcps_rcvbyteafterwin += todrop; 1760 } 1761 tp->snd_wl1 = th->th_seq - 1; 1762 tp->rcv_up = th->th_seq; 1763 goto step6; 1764 1765 /* 1766 * If the state is SYN_RECEIVED: 1767 * If seg contains an ACK, but not for our SYN, drop the input 1768 * and generate an RST. See page 36, rfc793 1769 */ 1770 case TCPS_SYN_RECEIVED: 1771 if ((tiflags & TH_ACK) && 1772 (SEQ_LEQ(th->th_ack, tp->iss) || 1773 SEQ_GT(th->th_ack, tp->snd_max))) 1774 goto dropwithreset; 1775 break; 1776 } 1777 1778 /* 1779 * States other than LISTEN or SYN_SENT. 1780 * First check timestamp, if present. 1781 * Then check that at least some bytes of segment are within 1782 * receive window. If segment begins before rcv_nxt, 1783 * drop leading data (and SYN); if nothing left, just ack. 1784 * 1785 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1786 * and it's less than ts_recent, drop it. 1787 */ 1788 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1789 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1790 1791 /* Check to see if ts_recent is over 24 days old. */ 1792 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 1793 /* 1794 * Invalidate ts_recent. If this segment updates 1795 * ts_recent, the age will be reset later and ts_recent 1796 * will get a valid value. If it does not, setting 1797 * ts_recent to zero will at least satisfy the 1798 * requirement that zero be placed in the timestamp 1799 * echo reply when ts_recent isn't valid. The 1800 * age isn't reset until we get a valid ts_recent 1801 * because we don't want out-of-order segments to be 1802 * dropped when ts_recent is old. 1803 */ 1804 tp->ts_recent = 0; 1805 } else { 1806 tcpstat.tcps_rcvduppack++; 1807 tcpstat.tcps_rcvdupbyte += tlen; 1808 tcpstat.tcps_pawsdrop++; 1809 tcp_new_dsack(tp, th->th_seq, tlen); 1810 goto dropafterack; 1811 } 1812 } 1813 1814 todrop = tp->rcv_nxt - th->th_seq; 1815 dupseg = FALSE; 1816 if (todrop > 0) { 1817 if (tiflags & TH_SYN) { 1818 tiflags &= ~TH_SYN; 1819 th->th_seq++; 1820 if (th->th_urp > 1) 1821 th->th_urp--; 1822 else { 1823 tiflags &= ~TH_URG; 1824 th->th_urp = 0; 1825 } 1826 todrop--; 1827 } 1828 if (todrop > tlen || 1829 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1830 /* 1831 * Any valid FIN or RST must be to the left of the 1832 * window. At this point the FIN or RST must be a 1833 * duplicate or out of sequence; drop it. 1834 */ 1835 if (tiflags & TH_RST) 1836 goto drop; 1837 tiflags &= ~(TH_FIN|TH_RST); 1838 /* 1839 * Send an ACK to resynchronize and drop any data. 1840 * But keep on processing for RST or ACK. 1841 */ 1842 tp->t_flags |= TF_ACKNOW; 1843 todrop = tlen; 1844 dupseg = TRUE; 1845 tcpstat.tcps_rcvdupbyte += todrop; 1846 tcpstat.tcps_rcvduppack++; 1847 } else if ((tiflags & TH_RST) && 1848 th->th_seq != tp->last_ack_sent) { 1849 /* 1850 * Test for reset before adjusting the sequence 1851 * number for overlapping data. 1852 */ 1853 goto dropafterack_ratelim; 1854 } else { 1855 tcpstat.tcps_rcvpartduppack++; 1856 tcpstat.tcps_rcvpartdupbyte += todrop; 1857 } 1858 tcp_new_dsack(tp, th->th_seq, todrop); 1859 hdroptlen += todrop; /*drop from head afterwards*/ 1860 th->th_seq += todrop; 1861 tlen -= todrop; 1862 if (th->th_urp > todrop) 1863 th->th_urp -= todrop; 1864 else { 1865 tiflags &= ~TH_URG; 1866 th->th_urp = 0; 1867 } 1868 } 1869 1870 /* 1871 * If new data are received on a connection after the 1872 * user processes are gone, then RST the other end. 1873 */ 1874 if ((so->so_state & SS_NOFDREF) && 1875 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1876 tp = tcp_close(tp); 1877 tcpstat.tcps_rcvafterclose++; 1878 goto dropwithreset; 1879 } 1880 1881 /* 1882 * If segment ends after window, drop trailing data 1883 * (and PUSH and FIN); if nothing left, just ACK. 1884 */ 1885 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1886 if (todrop > 0) { 1887 tcpstat.tcps_rcvpackafterwin++; 1888 if (todrop >= tlen) { 1889 /* 1890 * The segment actually starts after the window. 1891 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 1892 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 1893 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 1894 */ 1895 tcpstat.tcps_rcvbyteafterwin += tlen; 1896 /* 1897 * If a new connection request is received 1898 * while in TIME_WAIT, drop the old connection 1899 * and start over if the sequence numbers 1900 * are above the previous ones. 1901 * 1902 * NOTE: We will checksum the packet again, and 1903 * so we need to put the header fields back into 1904 * network order! 1905 * XXX This kind of sucks, but we don't expect 1906 * XXX this to happen very often, so maybe it 1907 * XXX doesn't matter so much. 1908 */ 1909 if (tiflags & TH_SYN && 1910 tp->t_state == TCPS_TIME_WAIT && 1911 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1912 iss = tcp_new_iss(tp, tp->snd_nxt); 1913 tp = tcp_close(tp); 1914 TCP_FIELDS_TO_NET(th); 1915 goto findpcb; 1916 } 1917 /* 1918 * If window is closed can only take segments at 1919 * window edge, and have to drop data and PUSH from 1920 * incoming segments. Continue processing, but 1921 * remember to ack. Otherwise, drop segment 1922 * and (if not RST) ack. 1923 */ 1924 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1925 tp->t_flags |= TF_ACKNOW; 1926 tcpstat.tcps_rcvwinprobe++; 1927 } else 1928 goto dropafterack; 1929 } else 1930 tcpstat.tcps_rcvbyteafterwin += todrop; 1931 m_adj(m, -todrop); 1932 tlen -= todrop; 1933 tiflags &= ~(TH_PUSH|TH_FIN); 1934 } 1935 1936 /* 1937 * If last ACK falls within this segment's sequence numbers, 1938 * and the timestamp is newer, record it. 1939 */ 1940 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1941 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1942 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1943 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1944 tp->ts_recent_age = tcp_now; 1945 tp->ts_recent = opti.ts_val; 1946 } 1947 1948 /* 1949 * If the RST bit is set examine the state: 1950 * SYN_RECEIVED STATE: 1951 * If passive open, return to LISTEN state. 1952 * If active open, inform user that connection was refused. 1953 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1954 * Inform user that connection was reset, and close tcb. 1955 * CLOSING, LAST_ACK, TIME_WAIT STATES 1956 * Close the tcb. 1957 */ 1958 if (tiflags & TH_RST) { 1959 if (th->th_seq != tp->last_ack_sent) 1960 goto dropafterack_ratelim; 1961 1962 switch (tp->t_state) { 1963 case TCPS_SYN_RECEIVED: 1964 so->so_error = ECONNREFUSED; 1965 goto close; 1966 1967 case TCPS_ESTABLISHED: 1968 case TCPS_FIN_WAIT_1: 1969 case TCPS_FIN_WAIT_2: 1970 case TCPS_CLOSE_WAIT: 1971 so->so_error = ECONNRESET; 1972 close: 1973 tp->t_state = TCPS_CLOSED; 1974 tcpstat.tcps_drops++; 1975 tp = tcp_close(tp); 1976 goto drop; 1977 1978 case TCPS_CLOSING: 1979 case TCPS_LAST_ACK: 1980 case TCPS_TIME_WAIT: 1981 tp = tcp_close(tp); 1982 goto drop; 1983 } 1984 } 1985 1986 /* 1987 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 1988 * we must be in a synchronized state. RFC791 states (under RST 1989 * generation) that any unacceptable segment (an out-of-order SYN 1990 * qualifies) received in a synchronized state must elicit only an 1991 * empty acknowledgment segment ... and the connection remains in 1992 * the same state. 1993 */ 1994 if (tiflags & TH_SYN) { 1995 if (tp->rcv_nxt == th->th_seq) { 1996 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 1997 TH_ACK); 1998 if (tcp_saveti) 1999 m_freem(tcp_saveti); 2000 return; 2001 } 2002 2003 goto dropafterack_ratelim; 2004 } 2005 2006 /* 2007 * If the ACK bit is off we drop the segment and return. 2008 */ 2009 if ((tiflags & TH_ACK) == 0) { 2010 if (tp->t_flags & TF_ACKNOW) 2011 goto dropafterack; 2012 else 2013 goto drop; 2014 } 2015 2016 /* 2017 * Ack processing. 2018 */ 2019 switch (tp->t_state) { 2020 2021 /* 2022 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2023 * ESTABLISHED state and continue processing, otherwise 2024 * send an RST. 2025 */ 2026 case TCPS_SYN_RECEIVED: 2027 if (SEQ_GT(tp->snd_una, th->th_ack) || 2028 SEQ_GT(th->th_ack, tp->snd_max)) 2029 goto dropwithreset; 2030 tcpstat.tcps_connects++; 2031 soisconnected(so); 2032 tcp_established(tp); 2033 /* Do window scaling? */ 2034 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2035 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2036 tp->snd_scale = tp->requested_s_scale; 2037 tp->rcv_scale = tp->request_r_scale; 2038 } 2039 TCP_REASS_LOCK(tp); 2040 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2041 TCP_REASS_UNLOCK(tp); 2042 tp->snd_wl1 = th->th_seq - 1; 2043 /* fall into ... */ 2044 2045 /* 2046 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2047 * ACKs. If the ack is in the range 2048 * tp->snd_una < th->th_ack <= tp->snd_max 2049 * then advance tp->snd_una to th->th_ack and drop 2050 * data from the retransmission queue. If this ACK reflects 2051 * more up to date window information we update our window information. 2052 */ 2053 case TCPS_ESTABLISHED: 2054 case TCPS_FIN_WAIT_1: 2055 case TCPS_FIN_WAIT_2: 2056 case TCPS_CLOSE_WAIT: 2057 case TCPS_CLOSING: 2058 case TCPS_LAST_ACK: 2059 case TCPS_TIME_WAIT: 2060 2061 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2062 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2063 tcpstat.tcps_rcvdupack++; 2064 /* 2065 * If we have outstanding data (other than 2066 * a window probe), this is a completely 2067 * duplicate ack (ie, window info didn't 2068 * change), the ack is the biggest we've 2069 * seen and we've seen exactly our rexmt 2070 * threshhold of them, assume a packet 2071 * has been dropped and retransmit it. 2072 * Kludge snd_nxt & the congestion 2073 * window so we send only this one 2074 * packet. 2075 * 2076 * We know we're losing at the current 2077 * window size so do congestion avoidance 2078 * (set ssthresh to half the current window 2079 * and pull our congestion window back to 2080 * the new ssthresh). 2081 * 2082 * Dup acks mean that packets have left the 2083 * network (they're now cached at the receiver) 2084 * so bump cwnd by the amount in the receiver 2085 * to keep a constant cwnd packets in the 2086 * network. 2087 * 2088 * If we are using TCP/SACK, then enter 2089 * Fast Recovery if the receiver SACKs 2090 * data that is tcprexmtthresh * MSS 2091 * bytes past the last ACKed segment, 2092 * irrespective of the number of DupAcks. 2093 */ 2094 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2095 th->th_ack != tp->snd_una) 2096 tp->t_dupacks = 0; 2097 else if (tp->t_partialacks < 0 && 2098 (++tp->t_dupacks == tcprexmtthresh || 2099 TCP_FACK_FASTRECOV(tp))) { 2100 tcp_seq onxt; 2101 u_int win; 2102 2103 if (tcp_do_newreno && 2104 SEQ_LT(th->th_ack, tp->snd_high)) { 2105 /* 2106 * False fast retransmit after 2107 * timeout. Do not enter fast 2108 * recovery. 2109 */ 2110 tp->t_dupacks = 0; 2111 break; 2112 } 2113 2114 onxt = tp->snd_nxt; 2115 win = min(tp->snd_wnd, tp->snd_cwnd) / 2116 2 / tp->t_segsz; 2117 if (win < 2) 2118 win = 2; 2119 tp->snd_ssthresh = win * tp->t_segsz; 2120 tp->snd_recover = tp->snd_max; 2121 tp->t_partialacks = 0; 2122 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2123 tp->t_rtttime = 0; 2124 if (TCP_SACK_ENABLED(tp)) { 2125 tp->t_dupacks = tcprexmtthresh; 2126 tp->sack_newdata = tp->snd_nxt; 2127 tp->snd_cwnd = tp->t_segsz; 2128 (void) tcp_output(tp); 2129 goto drop; 2130 } 2131 tp->snd_nxt = th->th_ack; 2132 tp->snd_cwnd = tp->t_segsz; 2133 (void) tcp_output(tp); 2134 tp->snd_cwnd = tp->snd_ssthresh + 2135 tp->t_segsz * tp->t_dupacks; 2136 if (SEQ_GT(onxt, tp->snd_nxt)) 2137 tp->snd_nxt = onxt; 2138 goto drop; 2139 } else if (tp->t_dupacks > tcprexmtthresh) { 2140 tp->snd_cwnd += tp->t_segsz; 2141 (void) tcp_output(tp); 2142 goto drop; 2143 } 2144 } else { 2145 /* 2146 * If the ack appears to be very old, only 2147 * allow data that is in-sequence. This 2148 * makes it somewhat more difficult to insert 2149 * forged data by guessing sequence numbers. 2150 * Sent an ack to try to update the send 2151 * sequence number on the other side. 2152 */ 2153 if (tlen && th->th_seq != tp->rcv_nxt && 2154 SEQ_LT(th->th_ack, 2155 tp->snd_una - tp->max_sndwnd)) 2156 goto dropafterack; 2157 } 2158 break; 2159 } 2160 /* 2161 * If the congestion window was inflated to account 2162 * for the other side's cached packets, retract it. 2163 */ 2164 if (TCP_SACK_ENABLED(tp)) 2165 tcp_sack_newack(tp, th); 2166 else if (tcp_do_newreno) 2167 tcp_newreno_newack(tp, th); 2168 else 2169 tcp_reno_newack(tp, th); 2170 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2171 tcpstat.tcps_rcvacktoomuch++; 2172 goto dropafterack; 2173 } 2174 acked = th->th_ack - tp->snd_una; 2175 tcpstat.tcps_rcvackpack++; 2176 tcpstat.tcps_rcvackbyte += acked; 2177 2178 /* 2179 * If we have a timestamp reply, update smoothed 2180 * round trip time. If no timestamp is present but 2181 * transmit timer is running and timed sequence 2182 * number was acked, update smoothed round trip time. 2183 * Since we now have an rtt measurement, cancel the 2184 * timer backoff (cf., Phil Karn's retransmit alg.). 2185 * Recompute the initial retransmit timer. 2186 */ 2187 if (opti.ts_present && opti.ts_ecr) 2188 tcp_xmit_timer(tp, opti.ts_ecr); 2189 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2190 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2191 2192 /* 2193 * If all outstanding data is acked, stop retransmit 2194 * timer and remember to restart (more output or persist). 2195 * If there is more data to be acked, restart retransmit 2196 * timer, using current (possibly backed-off) value. 2197 */ 2198 if (th->th_ack == tp->snd_max) { 2199 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2200 needoutput = 1; 2201 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2202 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2203 /* 2204 * When new data is acked, open the congestion window. 2205 * If the window gives us less than ssthresh packets 2206 * in flight, open exponentially (segsz per packet). 2207 * Otherwise open linearly: segsz per window 2208 * (segsz^2 / cwnd per packet), plus a constant 2209 * fraction of a packet (segsz/8) to help larger windows 2210 * open quickly enough. 2211 * 2212 * If we are still in fast recovery (meaning we are using 2213 * NewReno and we have only received partial acks), do not 2214 * inflate the window yet. 2215 */ 2216 if (tp->t_partialacks < 0) { 2217 u_int cw = tp->snd_cwnd; 2218 u_int incr = tp->t_segsz; 2219 2220 if (cw >= tp->snd_ssthresh) 2221 incr = incr * incr / cw; 2222 tp->snd_cwnd = min(cw + incr, 2223 TCP_MAXWIN << tp->snd_scale); 2224 } 2225 ND6_HINT(tp); 2226 if (acked > so->so_snd.sb_cc) { 2227 tp->snd_wnd -= so->so_snd.sb_cc; 2228 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2229 ourfinisacked = 1; 2230 } else { 2231 if (acked > (tp->t_lastoff - tp->t_inoff)) 2232 tp->t_lastm = NULL; 2233 sbdrop(&so->so_snd, acked); 2234 tp->t_lastoff -= acked; 2235 tp->snd_wnd -= acked; 2236 ourfinisacked = 0; 2237 } 2238 sowwakeup(so); 2239 tp->snd_una = th->th_ack; 2240 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2241 tp->snd_fack = tp->snd_una; 2242 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2243 tp->snd_nxt = tp->snd_una; 2244 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2245 tp->snd_high = tp->snd_una; 2246 2247 switch (tp->t_state) { 2248 2249 /* 2250 * In FIN_WAIT_1 STATE in addition to the processing 2251 * for the ESTABLISHED state if our FIN is now acknowledged 2252 * then enter FIN_WAIT_2. 2253 */ 2254 case TCPS_FIN_WAIT_1: 2255 if (ourfinisacked) { 2256 /* 2257 * If we can't receive any more 2258 * data, then closing user can proceed. 2259 * Starting the timer is contrary to the 2260 * specification, but if we don't get a FIN 2261 * we'll hang forever. 2262 */ 2263 if (so->so_state & SS_CANTRCVMORE) { 2264 soisdisconnected(so); 2265 if (tcp_maxidle > 0) 2266 TCP_TIMER_ARM(tp, TCPT_2MSL, 2267 tcp_maxidle); 2268 } 2269 tp->t_state = TCPS_FIN_WAIT_2; 2270 } 2271 break; 2272 2273 /* 2274 * In CLOSING STATE in addition to the processing for 2275 * the ESTABLISHED state if the ACK acknowledges our FIN 2276 * then enter the TIME-WAIT state, otherwise ignore 2277 * the segment. 2278 */ 2279 case TCPS_CLOSING: 2280 if (ourfinisacked) { 2281 tp->t_state = TCPS_TIME_WAIT; 2282 tcp_canceltimers(tp); 2283 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2284 soisdisconnected(so); 2285 } 2286 break; 2287 2288 /* 2289 * In LAST_ACK, we may still be waiting for data to drain 2290 * and/or to be acked, as well as for the ack of our FIN. 2291 * If our FIN is now acknowledged, delete the TCB, 2292 * enter the closed state and return. 2293 */ 2294 case TCPS_LAST_ACK: 2295 if (ourfinisacked) { 2296 tp = tcp_close(tp); 2297 goto drop; 2298 } 2299 break; 2300 2301 /* 2302 * In TIME_WAIT state the only thing that should arrive 2303 * is a retransmission of the remote FIN. Acknowledge 2304 * it and restart the finack timer. 2305 */ 2306 case TCPS_TIME_WAIT: 2307 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2308 goto dropafterack; 2309 } 2310 } 2311 2312 step6: 2313 /* 2314 * Update window information. 2315 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2316 */ 2317 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2318 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 2319 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 2320 /* keep track of pure window updates */ 2321 if (tlen == 0 && 2322 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2323 tcpstat.tcps_rcvwinupd++; 2324 tp->snd_wnd = tiwin; 2325 tp->snd_wl1 = th->th_seq; 2326 tp->snd_wl2 = th->th_ack; 2327 if (tp->snd_wnd > tp->max_sndwnd) 2328 tp->max_sndwnd = tp->snd_wnd; 2329 needoutput = 1; 2330 } 2331 2332 /* 2333 * Process segments with URG. 2334 */ 2335 if ((tiflags & TH_URG) && th->th_urp && 2336 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2337 /* 2338 * This is a kludge, but if we receive and accept 2339 * random urgent pointers, we'll crash in 2340 * soreceive. It's hard to imagine someone 2341 * actually wanting to send this much urgent data. 2342 */ 2343 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2344 th->th_urp = 0; /* XXX */ 2345 tiflags &= ~TH_URG; /* XXX */ 2346 goto dodata; /* XXX */ 2347 } 2348 /* 2349 * If this segment advances the known urgent pointer, 2350 * then mark the data stream. This should not happen 2351 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2352 * a FIN has been received from the remote side. 2353 * In these states we ignore the URG. 2354 * 2355 * According to RFC961 (Assigned Protocols), 2356 * the urgent pointer points to the last octet 2357 * of urgent data. We continue, however, 2358 * to consider it to indicate the first octet 2359 * of data past the urgent section as the original 2360 * spec states (in one of two places). 2361 */ 2362 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2363 tp->rcv_up = th->th_seq + th->th_urp; 2364 so->so_oobmark = so->so_rcv.sb_cc + 2365 (tp->rcv_up - tp->rcv_nxt) - 1; 2366 if (so->so_oobmark == 0) 2367 so->so_state |= SS_RCVATMARK; 2368 sohasoutofband(so); 2369 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2370 } 2371 /* 2372 * Remove out of band data so doesn't get presented to user. 2373 * This can happen independent of advancing the URG pointer, 2374 * but if two URG's are pending at once, some out-of-band 2375 * data may creep in... ick. 2376 */ 2377 if (th->th_urp <= (u_int16_t) tlen 2378 #ifdef SO_OOBINLINE 2379 && (so->so_options & SO_OOBINLINE) == 0 2380 #endif 2381 ) 2382 tcp_pulloutofband(so, th, m, hdroptlen); 2383 } else 2384 /* 2385 * If no out of band data is expected, 2386 * pull receive urgent pointer along 2387 * with the receive window. 2388 */ 2389 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2390 tp->rcv_up = tp->rcv_nxt; 2391 dodata: /* XXX */ 2392 2393 /* 2394 * Process the segment text, merging it into the TCP sequencing queue, 2395 * and arranging for acknowledgement of receipt if necessary. 2396 * This process logically involves adjusting tp->rcv_wnd as data 2397 * is presented to the user (this happens in tcp_usrreq.c, 2398 * case PRU_RCVD). If a FIN has already been received on this 2399 * connection then we just ignore the text. 2400 */ 2401 if ((tlen || (tiflags & TH_FIN)) && 2402 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2403 /* 2404 * Insert segment ti into reassembly queue of tcp with 2405 * control block tp. Return TH_FIN if reassembly now includes 2406 * a segment with FIN. The macro form does the common case 2407 * inline (segment is the next to be received on an 2408 * established connection, and the queue is empty), 2409 * avoiding linkage into and removal from the queue and 2410 * repetition of various conversions. 2411 * Set DELACK for segments received in order, but ack 2412 * immediately when segments are out of order 2413 * (so fast retransmit can work). 2414 */ 2415 /* NOTE: this was TCP_REASS() macro, but used only once */ 2416 TCP_REASS_LOCK(tp); 2417 if (th->th_seq == tp->rcv_nxt && 2418 TAILQ_FIRST(&tp->segq) == NULL && 2419 tp->t_state == TCPS_ESTABLISHED) { 2420 TCP_SETUP_ACK(tp, th); 2421 tp->rcv_nxt += tlen; 2422 tiflags = th->th_flags & TH_FIN; 2423 tcpstat.tcps_rcvpack++; 2424 tcpstat.tcps_rcvbyte += tlen; 2425 ND6_HINT(tp); 2426 if (so->so_state & SS_CANTRCVMORE) 2427 m_freem(m); 2428 else { 2429 m_adj(m, hdroptlen); 2430 sbappendstream(&(so)->so_rcv, m); 2431 } 2432 sorwakeup(so); 2433 } else { 2434 m_adj(m, hdroptlen); 2435 tiflags = tcp_reass(tp, th, m, &tlen); 2436 tcp_update_sack_list(tp); 2437 tp->t_flags |= TF_ACKNOW; 2438 } 2439 TCP_REASS_UNLOCK(tp); 2440 2441 /* 2442 * Note the amount of data that peer has sent into 2443 * our window, in order to estimate the sender's 2444 * buffer size. 2445 */ 2446 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2447 } else { 2448 m_freem(m); 2449 m = NULL; 2450 tiflags &= ~TH_FIN; 2451 } 2452 2453 /* 2454 * If FIN is received ACK the FIN and let the user know 2455 * that the connection is closing. Ignore a FIN received before 2456 * the connection is fully established. 2457 */ 2458 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2459 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2460 socantrcvmore(so); 2461 tp->t_flags |= TF_ACKNOW; 2462 tp->rcv_nxt++; 2463 } 2464 switch (tp->t_state) { 2465 2466 /* 2467 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2468 */ 2469 case TCPS_ESTABLISHED: 2470 tp->t_state = TCPS_CLOSE_WAIT; 2471 break; 2472 2473 /* 2474 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2475 * enter the CLOSING state. 2476 */ 2477 case TCPS_FIN_WAIT_1: 2478 tp->t_state = TCPS_CLOSING; 2479 break; 2480 2481 /* 2482 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2483 * starting the time-wait timer, turning off the other 2484 * standard timers. 2485 */ 2486 case TCPS_FIN_WAIT_2: 2487 tp->t_state = TCPS_TIME_WAIT; 2488 tcp_canceltimers(tp); 2489 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2490 soisdisconnected(so); 2491 break; 2492 2493 /* 2494 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2495 */ 2496 case TCPS_TIME_WAIT: 2497 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2498 break; 2499 } 2500 } 2501 #ifdef TCP_DEBUG 2502 if (so->so_options & SO_DEBUG) 2503 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2504 #endif 2505 2506 /* 2507 * Return any desired output. 2508 */ 2509 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2510 tcp_update_sack_list(tp); 2511 (void) tcp_output(tp); 2512 } 2513 if (tcp_saveti) 2514 m_freem(tcp_saveti); 2515 return; 2516 2517 badsyn: 2518 /* 2519 * Received a bad SYN. Increment counters and dropwithreset. 2520 */ 2521 tcpstat.tcps_badsyn++; 2522 tp = NULL; 2523 goto dropwithreset; 2524 2525 dropafterack: 2526 /* 2527 * Generate an ACK dropping incoming segment if it occupies 2528 * sequence space, where the ACK reflects our state. 2529 */ 2530 if (tiflags & TH_RST) 2531 goto drop; 2532 goto dropafterack2; 2533 2534 dropafterack_ratelim: 2535 /* 2536 * We may want to rate-limit ACKs against SYN/RST attack. 2537 */ 2538 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2539 tcp_ackdrop_ppslim) == 0) { 2540 /* XXX stat */ 2541 goto drop; 2542 } 2543 /* ...fall into dropafterack2... */ 2544 2545 dropafterack2: 2546 m_freem(m); 2547 tp->t_flags |= TF_ACKNOW; 2548 tcp_update_sack_list(tp); 2549 (void) tcp_output(tp); 2550 if (tcp_saveti) 2551 m_freem(tcp_saveti); 2552 return; 2553 2554 dropwithreset_ratelim: 2555 /* 2556 * We may want to rate-limit RSTs in certain situations, 2557 * particularly if we are sending an RST in response to 2558 * an attempt to connect to or otherwise communicate with 2559 * a port for which we have no socket. 2560 */ 2561 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2562 tcp_rst_ppslim) == 0) { 2563 /* XXX stat */ 2564 goto drop; 2565 } 2566 /* ...fall into dropwithreset... */ 2567 2568 dropwithreset: 2569 /* 2570 * Generate a RST, dropping incoming segment. 2571 * Make ACK acceptable to originator of segment. 2572 */ 2573 if (tiflags & TH_RST) 2574 goto drop; 2575 2576 switch (af) { 2577 #ifdef INET6 2578 case AF_INET6: 2579 /* For following calls to tcp_respond */ 2580 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2581 goto drop; 2582 break; 2583 #endif /* INET6 */ 2584 case AF_INET: 2585 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2586 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2587 goto drop; 2588 } 2589 2590 if (tiflags & TH_ACK) 2591 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2592 else { 2593 if (tiflags & TH_SYN) 2594 tlen++; 2595 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2596 TH_RST|TH_ACK); 2597 } 2598 if (tcp_saveti) 2599 m_freem(tcp_saveti); 2600 return; 2601 2602 badcsum: 2603 drop: 2604 /* 2605 * Drop space held by incoming segment and return. 2606 */ 2607 if (tp) { 2608 if (tp->t_inpcb) 2609 so = tp->t_inpcb->inp_socket; 2610 #ifdef INET6 2611 else if (tp->t_in6pcb) 2612 so = tp->t_in6pcb->in6p_socket; 2613 #endif 2614 else 2615 so = NULL; 2616 #ifdef TCP_DEBUG 2617 if (so && (so->so_options & SO_DEBUG) != 0) 2618 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2619 #endif 2620 } 2621 if (tcp_saveti) 2622 m_freem(tcp_saveti); 2623 m_freem(m); 2624 return; 2625 } 2626 2627 #ifdef TCP_SIGNATURE 2628 int 2629 tcp_signature_apply(void *fstate, caddr_t data, u_int len) 2630 { 2631 2632 MD5Update(fstate, (u_char *)data, len); 2633 return (0); 2634 } 2635 2636 struct secasvar * 2637 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 2638 { 2639 struct secasvar *sav; 2640 #ifdef FAST_IPSEC 2641 union sockaddr_union dst; 2642 #endif 2643 struct ip *ip; 2644 struct ip6_hdr *ip6; 2645 2646 ip = mtod(m, struct ip *); 2647 switch (ip->ip_v) { 2648 case 4: 2649 ip = mtod(m, struct ip *); 2650 ip6 = NULL; 2651 break; 2652 case 6: 2653 ip = NULL; 2654 ip6 = mtod(m, struct ip6_hdr *); 2655 break; 2656 default: 2657 return (NULL); 2658 } 2659 2660 #ifdef FAST_IPSEC 2661 /* Extract the destination from the IP header in the mbuf. */ 2662 bzero(&dst, sizeof(union sockaddr_union)); 2663 dst.sa.sa_len = sizeof(struct sockaddr_in); 2664 dst.sa.sa_family = AF_INET; 2665 dst.sin.sin_addr = ip->ip_dst; 2666 2667 /* 2668 * Look up an SADB entry which matches the address of the peer. 2669 */ 2670 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2671 #else 2672 if (ip) 2673 sav = key_allocsa(AF_INET, (caddr_t)&ip->ip_src, 2674 (caddr_t)&ip->ip_dst, IPPROTO_TCP, 2675 htonl(TCP_SIG_SPI)); 2676 else 2677 sav = key_allocsa(AF_INET6, (caddr_t)&ip6->ip6_src, 2678 (caddr_t)&ip6->ip6_dst, IPPROTO_TCP, 2679 htonl(TCP_SIG_SPI)); 2680 #endif 2681 2682 return (sav); /* freesav must be performed by caller */ 2683 } 2684 2685 int 2686 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 2687 struct secasvar *sav, char *sig) 2688 { 2689 MD5_CTX ctx; 2690 struct ip *ip; 2691 struct ipovly *ipovly; 2692 struct ip6_hdr *ip6; 2693 struct ippseudo ippseudo; 2694 struct ip6_hdr_pseudo ip6pseudo; 2695 struct tcphdr th0; 2696 int l, tcphdrlen; 2697 2698 if (sav == NULL) 2699 return (-1); 2700 2701 tcphdrlen = th->th_off * 4; 2702 2703 switch (mtod(m, struct ip *)->ip_v) { 2704 case 4: 2705 ip = mtod(m, struct ip *); 2706 ip6 = NULL; 2707 break; 2708 case 6: 2709 ip = NULL; 2710 ip6 = mtod(m, struct ip6_hdr *); 2711 break; 2712 default: 2713 return (-1); 2714 } 2715 2716 MD5Init(&ctx); 2717 2718 if (ip) { 2719 memset(&ippseudo, 0, sizeof(ippseudo)); 2720 ipovly = (struct ipovly *)ip; 2721 ippseudo.ippseudo_src = ipovly->ih_src; 2722 ippseudo.ippseudo_dst = ipovly->ih_dst; 2723 ippseudo.ippseudo_pad = 0; 2724 ippseudo.ippseudo_p = IPPROTO_TCP; 2725 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 2726 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 2727 } else { 2728 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 2729 ip6pseudo.ip6ph_src = ip6->ip6_src; 2730 in6_clearscope(&ip6pseudo.ip6ph_src); 2731 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 2732 in6_clearscope(&ip6pseudo.ip6ph_dst); 2733 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 2734 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 2735 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 2736 } 2737 2738 th0 = *th; 2739 th0.th_sum = 0; 2740 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 2741 2742 l = m->m_pkthdr.len - thoff - tcphdrlen; 2743 if (l > 0) 2744 m_apply(m, thoff + tcphdrlen, 2745 m->m_pkthdr.len - thoff - tcphdrlen, 2746 tcp_signature_apply, &ctx); 2747 2748 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2749 MD5Final(sig, &ctx); 2750 2751 return (0); 2752 } 2753 #endif 2754 2755 int 2756 tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcphdr *th, 2757 struct mbuf *m, int toff, struct tcp_opt_info *oi) 2758 { 2759 u_int16_t mss; 2760 int opt, optlen = 0; 2761 #ifdef TCP_SIGNATURE 2762 caddr_t sigp = NULL; 2763 char sigbuf[TCP_SIGLEN]; 2764 struct secasvar *sav = NULL; 2765 #endif 2766 2767 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 2768 opt = cp[0]; 2769 if (opt == TCPOPT_EOL) 2770 break; 2771 if (opt == TCPOPT_NOP) 2772 optlen = 1; 2773 else { 2774 if (cnt < 2) 2775 break; 2776 optlen = cp[1]; 2777 if (optlen < 2 || optlen > cnt) 2778 break; 2779 } 2780 switch (opt) { 2781 2782 default: 2783 continue; 2784 2785 case TCPOPT_MAXSEG: 2786 if (optlen != TCPOLEN_MAXSEG) 2787 continue; 2788 if (!(th->th_flags & TH_SYN)) 2789 continue; 2790 bcopy(cp + 2, &mss, sizeof(mss)); 2791 oi->maxseg = ntohs(mss); 2792 break; 2793 2794 case TCPOPT_WINDOW: 2795 if (optlen != TCPOLEN_WINDOW) 2796 continue; 2797 if (!(th->th_flags & TH_SYN)) 2798 continue; 2799 tp->t_flags |= TF_RCVD_SCALE; 2800 tp->requested_s_scale = cp[2]; 2801 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2802 #if 0 /*XXX*/ 2803 char *p; 2804 2805 if (ip) 2806 p = ntohl(ip->ip_src); 2807 #ifdef INET6 2808 else if (ip6) 2809 p = ip6_sprintf(&ip6->ip6_src); 2810 #endif 2811 else 2812 p = "(unknown)"; 2813 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2814 "assuming %d\n", 2815 tp->requested_s_scale, p, 2816 TCP_MAX_WINSHIFT); 2817 #else 2818 log(LOG_ERR, "TCP: invalid wscale %d, " 2819 "assuming %d\n", 2820 tp->requested_s_scale, 2821 TCP_MAX_WINSHIFT); 2822 #endif 2823 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2824 } 2825 break; 2826 2827 case TCPOPT_TIMESTAMP: 2828 if (optlen != TCPOLEN_TIMESTAMP) 2829 continue; 2830 oi->ts_present = 1; 2831 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2832 NTOHL(oi->ts_val); 2833 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2834 NTOHL(oi->ts_ecr); 2835 2836 /* 2837 * A timestamp received in a SYN makes 2838 * it ok to send timestamp requests and replies. 2839 */ 2840 if (th->th_flags & TH_SYN) { 2841 tp->t_flags |= TF_RCVD_TSTMP; 2842 tp->ts_recent = oi->ts_val; 2843 tp->ts_recent_age = tcp_now; 2844 } 2845 break; 2846 case TCPOPT_SACK_PERMITTED: 2847 if (optlen != TCPOLEN_SACK_PERMITTED) 2848 continue; 2849 if (!(th->th_flags & TH_SYN)) 2850 continue; 2851 if (tcp_do_sack) { 2852 tp->t_flags |= TF_SACK_PERMIT; 2853 tp->t_flags |= TF_WILL_SACK; 2854 } 2855 break; 2856 2857 case TCPOPT_SACK: 2858 tcp_sack_option(tp, th, cp, optlen); 2859 break; 2860 #ifdef TCP_SIGNATURE 2861 case TCPOPT_SIGNATURE: 2862 if (optlen != TCPOLEN_SIGNATURE) 2863 continue; 2864 if (sigp && bcmp(sigp, cp + 2, TCP_SIGLEN)) 2865 return (-1); 2866 2867 sigp = sigbuf; 2868 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 2869 memset(cp + 2, 0, TCP_SIGLEN); 2870 tp->t_flags |= TF_SIGNATURE; 2871 break; 2872 #endif 2873 } 2874 } 2875 2876 #ifdef TCP_SIGNATURE 2877 if (tp->t_flags & TF_SIGNATURE) { 2878 2879 sav = tcp_signature_getsav(m, th); 2880 2881 if (sav == NULL && tp->t_state == TCPS_LISTEN) 2882 return (-1); 2883 } 2884 2885 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 2886 if (sav == NULL) 2887 return (-1); 2888 #ifdef FAST_IPSEC 2889 KEY_FREESAV(&sav); 2890 #else 2891 key_freesav(sav); 2892 #endif 2893 return (-1); 2894 } 2895 2896 if (sigp) { 2897 char sig[TCP_SIGLEN]; 2898 2899 TCP_FIELDS_TO_NET(th); 2900 if (tcp_signature(m, th, toff, sav, sig) < 0) { 2901 TCP_FIELDS_TO_HOST(th); 2902 if (sav == NULL) 2903 return (-1); 2904 #ifdef FAST_IPSEC 2905 KEY_FREESAV(&sav); 2906 #else 2907 key_freesav(sav); 2908 #endif 2909 return (-1); 2910 } 2911 TCP_FIELDS_TO_HOST(th); 2912 2913 if (bcmp(sig, sigp, TCP_SIGLEN)) { 2914 tcpstat.tcps_badsig++; 2915 if (sav == NULL) 2916 return (-1); 2917 #ifdef FAST_IPSEC 2918 KEY_FREESAV(&sav); 2919 #else 2920 key_freesav(sav); 2921 #endif 2922 return (-1); 2923 } else 2924 tcpstat.tcps_goodsig++; 2925 2926 key_sa_recordxfer(sav, m); 2927 #ifdef FAST_IPSEC 2928 KEY_FREESAV(&sav); 2929 #else 2930 key_freesav(sav); 2931 #endif 2932 } 2933 #endif 2934 2935 return (0); 2936 } 2937 2938 /* 2939 * Pull out of band byte out of a segment so 2940 * it doesn't appear in the user's data queue. 2941 * It is still reflected in the segment length for 2942 * sequencing purposes. 2943 */ 2944 void 2945 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 2946 struct mbuf *m, int off) 2947 { 2948 int cnt = off + th->th_urp - 1; 2949 2950 while (cnt >= 0) { 2951 if (m->m_len > cnt) { 2952 char *cp = mtod(m, caddr_t) + cnt; 2953 struct tcpcb *tp = sototcpcb(so); 2954 2955 tp->t_iobc = *cp; 2956 tp->t_oobflags |= TCPOOB_HAVEDATA; 2957 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2958 m->m_len--; 2959 return; 2960 } 2961 cnt -= m->m_len; 2962 m = m->m_next; 2963 if (m == 0) 2964 break; 2965 } 2966 panic("tcp_pulloutofband"); 2967 } 2968 2969 /* 2970 * Collect new round-trip time estimate 2971 * and update averages and current timeout. 2972 */ 2973 void 2974 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 2975 { 2976 int32_t delta; 2977 2978 tcpstat.tcps_rttupdated++; 2979 if (tp->t_srtt != 0) { 2980 /* 2981 * srtt is stored as fixed point with 3 bits after the 2982 * binary point (i.e., scaled by 8). The following magic 2983 * is equivalent to the smoothing algorithm in rfc793 with 2984 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2985 * point). Adjust rtt to origin 0. 2986 */ 2987 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2988 if ((tp->t_srtt += delta) <= 0) 2989 tp->t_srtt = 1 << 2; 2990 /* 2991 * We accumulate a smoothed rtt variance (actually, a 2992 * smoothed mean difference), then set the retransmit 2993 * timer to smoothed rtt + 4 times the smoothed variance. 2994 * rttvar is stored as fixed point with 2 bits after the 2995 * binary point (scaled by 4). The following is 2996 * equivalent to rfc793 smoothing with an alpha of .75 2997 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2998 * rfc793's wired-in beta. 2999 */ 3000 if (delta < 0) 3001 delta = -delta; 3002 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3003 if ((tp->t_rttvar += delta) <= 0) 3004 tp->t_rttvar = 1 << 2; 3005 } else { 3006 /* 3007 * No rtt measurement yet - use the unsmoothed rtt. 3008 * Set the variance to half the rtt (so our first 3009 * retransmit happens at 3*rtt). 3010 */ 3011 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3012 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3013 } 3014 tp->t_rtttime = 0; 3015 tp->t_rxtshift = 0; 3016 3017 /* 3018 * the retransmit should happen at rtt + 4 * rttvar. 3019 * Because of the way we do the smoothing, srtt and rttvar 3020 * will each average +1/2 tick of bias. When we compute 3021 * the retransmit timer, we want 1/2 tick of rounding and 3022 * 1 extra tick because of +-1/2 tick uncertainty in the 3023 * firing of the timer. The bias will give us exactly the 3024 * 1.5 tick we need. But, because the bias is 3025 * statistical, we have to test that we don't drop below 3026 * the minimum feasible timer (which is 2 ticks). 3027 */ 3028 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3029 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3030 3031 /* 3032 * We received an ack for a packet that wasn't retransmitted; 3033 * it is probably safe to discard any error indications we've 3034 * received recently. This isn't quite right, but close enough 3035 * for now (a route might have failed after we sent a segment, 3036 * and the return path might not be symmetrical). 3037 */ 3038 tp->t_softerror = 0; 3039 } 3040 3041 void 3042 tcp_reno_newack(struct tcpcb *tp, struct tcphdr *th) 3043 { 3044 if (tp->t_partialacks < 0) { 3045 /* 3046 * We were not in fast recovery. Reset the duplicate ack 3047 * counter. 3048 */ 3049 tp->t_dupacks = 0; 3050 } else { 3051 /* 3052 * Clamp the congestion window to the crossover point and 3053 * exit fast recovery. 3054 */ 3055 if (tp->snd_cwnd > tp->snd_ssthresh) 3056 tp->snd_cwnd = tp->snd_ssthresh; 3057 tp->t_partialacks = -1; 3058 tp->t_dupacks = 0; 3059 } 3060 } 3061 3062 /* 3063 * Implement the NewReno response to a new ack, checking for partial acks in 3064 * fast recovery. 3065 */ 3066 void 3067 tcp_newreno_newack(struct tcpcb *tp, struct tcphdr *th) 3068 { 3069 if (tp->t_partialacks < 0) { 3070 /* 3071 * We were not in fast recovery. Reset the duplicate ack 3072 * counter. 3073 */ 3074 tp->t_dupacks = 0; 3075 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) { 3076 /* 3077 * This is a partial ack. Retransmit the first unacknowledged 3078 * segment and deflate the congestion window by the amount of 3079 * acknowledged data. Do not exit fast recovery. 3080 */ 3081 tcp_seq onxt = tp->snd_nxt; 3082 u_long ocwnd = tp->snd_cwnd; 3083 3084 /* 3085 * snd_una has not yet been updated and the socket's send 3086 * buffer has not yet drained off the ACK'd data, so we 3087 * have to leave snd_una as it was to get the correct data 3088 * offset in tcp_output(). 3089 */ 3090 if (++tp->t_partialacks == 1) 3091 TCP_TIMER_DISARM(tp, TCPT_REXMT); 3092 tp->t_rtttime = 0; 3093 tp->snd_nxt = th->th_ack; 3094 /* 3095 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 3096 * is not yet updated when we're called. 3097 */ 3098 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 3099 (void) tcp_output(tp); 3100 tp->snd_cwnd = ocwnd; 3101 if (SEQ_GT(onxt, tp->snd_nxt)) 3102 tp->snd_nxt = onxt; 3103 /* 3104 * Partial window deflation. Relies on fact that tp->snd_una 3105 * not updated yet. 3106 */ 3107 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 3108 } else { 3109 /* 3110 * Complete ack. Inflate the congestion window to ssthresh 3111 * and exit fast recovery. 3112 * 3113 * Window inflation should have left us with approx. 3114 * snd_ssthresh outstanding data. But in case we 3115 * would be inclined to send a burst, better to do 3116 * it via the slow start mechanism. 3117 */ 3118 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 3119 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 3120 + tp->t_segsz; 3121 else 3122 tp->snd_cwnd = tp->snd_ssthresh; 3123 tp->t_partialacks = -1; 3124 tp->t_dupacks = 0; 3125 } 3126 } 3127 3128 3129 /* 3130 * TCP compressed state engine. Currently used to hold compressed 3131 * state for SYN_RECEIVED. 3132 */ 3133 3134 u_long syn_cache_count; 3135 u_int32_t syn_hash1, syn_hash2; 3136 3137 #define SYN_HASH(sa, sp, dp) \ 3138 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3139 ((u_int32_t)(sp)))^syn_hash2))) 3140 #ifndef INET6 3141 #define SYN_HASHALL(hash, src, dst) \ 3142 do { \ 3143 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 3144 ((struct sockaddr_in *)(src))->sin_port, \ 3145 ((struct sockaddr_in *)(dst))->sin_port); \ 3146 } while (/*CONSTCOND*/ 0) 3147 #else 3148 #define SYN_HASH6(sa, sp, dp) \ 3149 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3150 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3151 & 0x7fffffff) 3152 3153 #define SYN_HASHALL(hash, src, dst) \ 3154 do { \ 3155 switch ((src)->sa_family) { \ 3156 case AF_INET: \ 3157 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 3158 ((struct sockaddr_in *)(src))->sin_port, \ 3159 ((struct sockaddr_in *)(dst))->sin_port); \ 3160 break; \ 3161 case AF_INET6: \ 3162 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 3163 ((struct sockaddr_in6 *)(src))->sin6_port, \ 3164 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 3165 break; \ 3166 default: \ 3167 hash = 0; \ 3168 } \ 3169 } while (/*CONSTCOND*/0) 3170 #endif /* INET6 */ 3171 3172 #define SYN_CACHE_RM(sc) \ 3173 do { \ 3174 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 3175 (sc), sc_bucketq); \ 3176 (sc)->sc_tp = NULL; \ 3177 LIST_REMOVE((sc), sc_tpq); \ 3178 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 3179 callout_stop(&(sc)->sc_timer); \ 3180 syn_cache_count--; \ 3181 } while (/*CONSTCOND*/0) 3182 3183 #define SYN_CACHE_PUT(sc) \ 3184 do { \ 3185 if ((sc)->sc_ipopts) \ 3186 (void) m_free((sc)->sc_ipopts); \ 3187 if ((sc)->sc_route4.ro_rt != NULL) \ 3188 RTFREE((sc)->sc_route4.ro_rt); \ 3189 if (callout_invoking(&(sc)->sc_timer)) \ 3190 (sc)->sc_flags |= SCF_DEAD; \ 3191 else \ 3192 pool_put(&syn_cache_pool, (sc)); \ 3193 } while (/*CONSTCOND*/0) 3194 3195 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL); 3196 3197 /* 3198 * We don't estimate RTT with SYNs, so each packet starts with the default 3199 * RTT and each timer step has a fixed timeout value. 3200 */ 3201 #define SYN_CACHE_TIMER_ARM(sc) \ 3202 do { \ 3203 TCPT_RANGESET((sc)->sc_rxtcur, \ 3204 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3205 TCPTV_REXMTMAX); \ 3206 callout_reset(&(sc)->sc_timer, \ 3207 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3208 } while (/*CONSTCOND*/0) 3209 3210 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3211 3212 void 3213 syn_cache_init(void) 3214 { 3215 int i; 3216 3217 /* Initialize the hash buckets. */ 3218 for (i = 0; i < tcp_syn_cache_size; i++) 3219 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3220 } 3221 3222 void 3223 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3224 { 3225 struct syn_cache_head *scp; 3226 struct syn_cache *sc2; 3227 int s; 3228 3229 /* 3230 * If there are no entries in the hash table, reinitialize 3231 * the hash secrets. 3232 */ 3233 if (syn_cache_count == 0) { 3234 syn_hash1 = arc4random(); 3235 syn_hash2 = arc4random(); 3236 } 3237 3238 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3239 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3240 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3241 3242 /* 3243 * Make sure that we don't overflow the per-bucket 3244 * limit or the total cache size limit. 3245 */ 3246 s = splsoftnet(); 3247 if (scp->sch_length >= tcp_syn_bucket_limit) { 3248 tcpstat.tcps_sc_bucketoverflow++; 3249 /* 3250 * The bucket is full. Toss the oldest element in the 3251 * bucket. This will be the first entry in the bucket. 3252 */ 3253 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3254 #ifdef DIAGNOSTIC 3255 /* 3256 * This should never happen; we should always find an 3257 * entry in our bucket. 3258 */ 3259 if (sc2 == NULL) 3260 panic("syn_cache_insert: bucketoverflow: impossible"); 3261 #endif 3262 SYN_CACHE_RM(sc2); 3263 SYN_CACHE_PUT(sc2); 3264 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3265 struct syn_cache_head *scp2, *sce; 3266 3267 tcpstat.tcps_sc_overflowed++; 3268 /* 3269 * The cache is full. Toss the oldest entry in the 3270 * first non-empty bucket we can find. 3271 * 3272 * XXX We would really like to toss the oldest 3273 * entry in the cache, but we hope that this 3274 * condition doesn't happen very often. 3275 */ 3276 scp2 = scp; 3277 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3278 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3279 for (++scp2; scp2 != scp; scp2++) { 3280 if (scp2 >= sce) 3281 scp2 = &tcp_syn_cache[0]; 3282 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3283 break; 3284 } 3285 #ifdef DIAGNOSTIC 3286 /* 3287 * This should never happen; we should always find a 3288 * non-empty bucket. 3289 */ 3290 if (scp2 == scp) 3291 panic("syn_cache_insert: cacheoverflow: " 3292 "impossible"); 3293 #endif 3294 } 3295 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3296 SYN_CACHE_RM(sc2); 3297 SYN_CACHE_PUT(sc2); 3298 } 3299 3300 /* 3301 * Initialize the entry's timer. 3302 */ 3303 sc->sc_rxttot = 0; 3304 sc->sc_rxtshift = 0; 3305 SYN_CACHE_TIMER_ARM(sc); 3306 3307 /* Link it from tcpcb entry */ 3308 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3309 3310 /* Put it into the bucket. */ 3311 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3312 scp->sch_length++; 3313 syn_cache_count++; 3314 3315 tcpstat.tcps_sc_added++; 3316 splx(s); 3317 } 3318 3319 /* 3320 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3321 * If we have retransmitted an entry the maximum number of times, expire 3322 * that entry. 3323 */ 3324 void 3325 syn_cache_timer(void *arg) 3326 { 3327 struct syn_cache *sc = arg; 3328 int s; 3329 3330 s = splsoftnet(); 3331 callout_ack(&sc->sc_timer); 3332 3333 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3334 tcpstat.tcps_sc_delayed_free++; 3335 pool_put(&syn_cache_pool, sc); 3336 splx(s); 3337 return; 3338 } 3339 3340 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3341 /* Drop it -- too many retransmissions. */ 3342 goto dropit; 3343 } 3344 3345 /* 3346 * Compute the total amount of time this entry has 3347 * been on a queue. If this entry has been on longer 3348 * than the keep alive timer would allow, expire it. 3349 */ 3350 sc->sc_rxttot += sc->sc_rxtcur; 3351 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) 3352 goto dropit; 3353 3354 tcpstat.tcps_sc_retransmitted++; 3355 (void) syn_cache_respond(sc, NULL); 3356 3357 /* Advance the timer back-off. */ 3358 sc->sc_rxtshift++; 3359 SYN_CACHE_TIMER_ARM(sc); 3360 3361 splx(s); 3362 return; 3363 3364 dropit: 3365 tcpstat.tcps_sc_timed_out++; 3366 SYN_CACHE_RM(sc); 3367 SYN_CACHE_PUT(sc); 3368 splx(s); 3369 } 3370 3371 /* 3372 * Remove syn cache created by the specified tcb entry, 3373 * because this does not make sense to keep them 3374 * (if there's no tcb entry, syn cache entry will never be used) 3375 */ 3376 void 3377 syn_cache_cleanup(struct tcpcb *tp) 3378 { 3379 struct syn_cache *sc, *nsc; 3380 int s; 3381 3382 s = splsoftnet(); 3383 3384 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3385 nsc = LIST_NEXT(sc, sc_tpq); 3386 3387 #ifdef DIAGNOSTIC 3388 if (sc->sc_tp != tp) 3389 panic("invalid sc_tp in syn_cache_cleanup"); 3390 #endif 3391 SYN_CACHE_RM(sc); 3392 SYN_CACHE_PUT(sc); 3393 } 3394 /* just for safety */ 3395 LIST_INIT(&tp->t_sc); 3396 3397 splx(s); 3398 } 3399 3400 /* 3401 * Find an entry in the syn cache. 3402 */ 3403 struct syn_cache * 3404 syn_cache_lookup(struct sockaddr *src, struct sockaddr *dst, 3405 struct syn_cache_head **headp) 3406 { 3407 struct syn_cache *sc; 3408 struct syn_cache_head *scp; 3409 u_int32_t hash; 3410 int s; 3411 3412 SYN_HASHALL(hash, src, dst); 3413 3414 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3415 *headp = scp; 3416 s = splsoftnet(); 3417 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3418 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3419 if (sc->sc_hash != hash) 3420 continue; 3421 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3422 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3423 splx(s); 3424 return (sc); 3425 } 3426 } 3427 splx(s); 3428 return (NULL); 3429 } 3430 3431 /* 3432 * This function gets called when we receive an ACK for a 3433 * socket in the LISTEN state. We look up the connection 3434 * in the syn cache, and if its there, we pull it out of 3435 * the cache and turn it into a full-blown connection in 3436 * the SYN-RECEIVED state. 3437 * 3438 * The return values may not be immediately obvious, and their effects 3439 * can be subtle, so here they are: 3440 * 3441 * NULL SYN was not found in cache; caller should drop the 3442 * packet and send an RST. 3443 * 3444 * -1 We were unable to create the new connection, and are 3445 * aborting it. An ACK,RST is being sent to the peer 3446 * (unless we got screwey sequence numbners; see below), 3447 * because the 3-way handshake has been completed. Caller 3448 * should not free the mbuf, since we may be using it. If 3449 * we are not, we will free it. 3450 * 3451 * Otherwise, the return value is a pointer to the new socket 3452 * associated with the connection. 3453 */ 3454 struct socket * 3455 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3456 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3457 struct socket *so, struct mbuf *m) 3458 { 3459 struct syn_cache *sc; 3460 struct syn_cache_head *scp; 3461 struct inpcb *inp = NULL; 3462 #ifdef INET6 3463 struct in6pcb *in6p = NULL; 3464 #endif 3465 struct tcpcb *tp = 0; 3466 struct mbuf *am; 3467 int s; 3468 struct socket *oso; 3469 3470 s = splsoftnet(); 3471 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3472 splx(s); 3473 return (NULL); 3474 } 3475 3476 /* 3477 * Verify the sequence and ack numbers. Try getting the correct 3478 * response again. 3479 */ 3480 if ((th->th_ack != sc->sc_iss + 1) || 3481 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3482 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3483 (void) syn_cache_respond(sc, m); 3484 splx(s); 3485 return ((struct socket *)(-1)); 3486 } 3487 3488 /* Remove this cache entry */ 3489 SYN_CACHE_RM(sc); 3490 splx(s); 3491 3492 /* 3493 * Ok, create the full blown connection, and set things up 3494 * as they would have been set up if we had created the 3495 * connection when the SYN arrived. If we can't create 3496 * the connection, abort it. 3497 */ 3498 /* 3499 * inp still has the OLD in_pcb stuff, set the 3500 * v6-related flags on the new guy, too. This is 3501 * done particularly for the case where an AF_INET6 3502 * socket is bound only to a port, and a v4 connection 3503 * comes in on that port. 3504 * we also copy the flowinfo from the original pcb 3505 * to the new one. 3506 */ 3507 oso = so; 3508 so = sonewconn(so, SS_ISCONNECTED); 3509 if (so == NULL) 3510 goto resetandabort; 3511 3512 switch (so->so_proto->pr_domain->dom_family) { 3513 #ifdef INET 3514 case AF_INET: 3515 inp = sotoinpcb(so); 3516 break; 3517 #endif 3518 #ifdef INET6 3519 case AF_INET6: 3520 in6p = sotoin6pcb(so); 3521 break; 3522 #endif 3523 } 3524 switch (src->sa_family) { 3525 #ifdef INET 3526 case AF_INET: 3527 if (inp) { 3528 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3529 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3530 inp->inp_options = ip_srcroute(); 3531 in_pcbstate(inp, INP_BOUND); 3532 if (inp->inp_options == NULL) { 3533 inp->inp_options = sc->sc_ipopts; 3534 sc->sc_ipopts = NULL; 3535 } 3536 } 3537 #ifdef INET6 3538 else if (in6p) { 3539 /* IPv4 packet to AF_INET6 socket */ 3540 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3541 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3542 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3543 &in6p->in6p_laddr.s6_addr32[3], 3544 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3545 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3546 in6totcpcb(in6p)->t_family = AF_INET; 3547 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3548 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3549 else 3550 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3551 in6_pcbstate(in6p, IN6P_BOUND); 3552 } 3553 #endif 3554 break; 3555 #endif 3556 #ifdef INET6 3557 case AF_INET6: 3558 if (in6p) { 3559 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3560 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3561 in6_pcbstate(in6p, IN6P_BOUND); 3562 } 3563 break; 3564 #endif 3565 } 3566 #ifdef INET6 3567 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3568 struct in6pcb *oin6p = sotoin6pcb(oso); 3569 /* inherit socket options from the listening socket */ 3570 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3571 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3572 m_freem(in6p->in6p_options); 3573 in6p->in6p_options = 0; 3574 } 3575 ip6_savecontrol(in6p, &in6p->in6p_options, 3576 mtod(m, struct ip6_hdr *), m); 3577 } 3578 #endif 3579 3580 #if defined(IPSEC) || defined(FAST_IPSEC) 3581 /* 3582 * we make a copy of policy, instead of sharing the policy, 3583 * for better behavior in terms of SA lookup and dead SA removal. 3584 */ 3585 if (inp) { 3586 /* copy old policy into new socket's */ 3587 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3588 printf("tcp_input: could not copy policy\n"); 3589 } 3590 #ifdef INET6 3591 else if (in6p) { 3592 /* copy old policy into new socket's */ 3593 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3594 in6p->in6p_sp)) 3595 printf("tcp_input: could not copy policy\n"); 3596 } 3597 #endif 3598 #endif 3599 3600 /* 3601 * Give the new socket our cached route reference. 3602 */ 3603 if (inp) 3604 inp->inp_route = sc->sc_route4; /* struct assignment */ 3605 #ifdef INET6 3606 else 3607 in6p->in6p_route = sc->sc_route6; 3608 #endif 3609 sc->sc_route4.ro_rt = NULL; 3610 3611 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3612 if (am == NULL) 3613 goto resetandabort; 3614 MCLAIM(am, &tcp_mowner); 3615 am->m_len = src->sa_len; 3616 bcopy(src, mtod(am, caddr_t), src->sa_len); 3617 if (inp) { 3618 if (in_pcbconnect(inp, am)) { 3619 (void) m_free(am); 3620 goto resetandabort; 3621 } 3622 } 3623 #ifdef INET6 3624 else if (in6p) { 3625 if (src->sa_family == AF_INET) { 3626 /* IPv4 packet to AF_INET6 socket */ 3627 struct sockaddr_in6 *sin6; 3628 sin6 = mtod(am, struct sockaddr_in6 *); 3629 am->m_len = sizeof(*sin6); 3630 bzero(sin6, sizeof(*sin6)); 3631 sin6->sin6_family = AF_INET6; 3632 sin6->sin6_len = sizeof(*sin6); 3633 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3634 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3635 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3636 &sin6->sin6_addr.s6_addr32[3], 3637 sizeof(sin6->sin6_addr.s6_addr32[3])); 3638 } 3639 if (in6_pcbconnect(in6p, am)) { 3640 (void) m_free(am); 3641 goto resetandabort; 3642 } 3643 } 3644 #endif 3645 else { 3646 (void) m_free(am); 3647 goto resetandabort; 3648 } 3649 (void) m_free(am); 3650 3651 if (inp) 3652 tp = intotcpcb(inp); 3653 #ifdef INET6 3654 else if (in6p) 3655 tp = in6totcpcb(in6p); 3656 #endif 3657 else 3658 tp = NULL; 3659 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3660 if (sc->sc_request_r_scale != 15) { 3661 tp->requested_s_scale = sc->sc_requested_s_scale; 3662 tp->request_r_scale = sc->sc_request_r_scale; 3663 tp->snd_scale = sc->sc_requested_s_scale; 3664 tp->rcv_scale = sc->sc_request_r_scale; 3665 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3666 } 3667 if (sc->sc_flags & SCF_TIMESTAMP) 3668 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3669 tp->ts_timebase = sc->sc_timebase; 3670 3671 tp->t_template = tcp_template(tp); 3672 if (tp->t_template == 0) { 3673 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3674 so = NULL; 3675 m_freem(m); 3676 goto abort; 3677 } 3678 3679 tp->iss = sc->sc_iss; 3680 tp->irs = sc->sc_irs; 3681 tcp_sendseqinit(tp); 3682 tcp_rcvseqinit(tp); 3683 tp->t_state = TCPS_SYN_RECEIVED; 3684 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3685 tcpstat.tcps_accepts++; 3686 3687 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 3688 tp->t_flags |= TF_WILL_SACK; 3689 3690 #ifdef TCP_SIGNATURE 3691 if (sc->sc_flags & SCF_SIGNATURE) 3692 tp->t_flags |= TF_SIGNATURE; 3693 #endif 3694 3695 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3696 tp->t_ourmss = sc->sc_ourmaxseg; 3697 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3698 3699 /* 3700 * Initialize the initial congestion window. If we 3701 * had to retransmit the SYN,ACK, we must initialize cwnd 3702 * to 1 segment (i.e. the Loss Window). 3703 */ 3704 if (sc->sc_rxtshift) 3705 tp->snd_cwnd = tp->t_peermss; 3706 else { 3707 int ss = tcp_init_win; 3708 #ifdef INET 3709 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3710 ss = tcp_init_win_local; 3711 #endif 3712 #ifdef INET6 3713 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3714 ss = tcp_init_win_local; 3715 #endif 3716 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3717 } 3718 3719 tcp_rmx_rtt(tp); 3720 tp->snd_wl1 = sc->sc_irs; 3721 tp->rcv_up = sc->sc_irs + 1; 3722 3723 /* 3724 * This is what whould have happened in tcp_output() when 3725 * the SYN,ACK was sent. 3726 */ 3727 tp->snd_up = tp->snd_una; 3728 tp->snd_max = tp->snd_nxt = tp->iss+1; 3729 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3730 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3731 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3732 tp->last_ack_sent = tp->rcv_nxt; 3733 tp->t_partialacks = -1; 3734 tp->t_dupacks = 0; 3735 3736 tcpstat.tcps_sc_completed++; 3737 SYN_CACHE_PUT(sc); 3738 return (so); 3739 3740 resetandabort: 3741 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3742 abort: 3743 if (so != NULL) 3744 (void) soabort(so); 3745 SYN_CACHE_PUT(sc); 3746 tcpstat.tcps_sc_aborted++; 3747 return ((struct socket *)(-1)); 3748 } 3749 3750 /* 3751 * This function is called when we get a RST for a 3752 * non-existent connection, so that we can see if the 3753 * connection is in the syn cache. If it is, zap it. 3754 */ 3755 3756 void 3757 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 3758 { 3759 struct syn_cache *sc; 3760 struct syn_cache_head *scp; 3761 int s = splsoftnet(); 3762 3763 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3764 splx(s); 3765 return; 3766 } 3767 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3768 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3769 splx(s); 3770 return; 3771 } 3772 SYN_CACHE_RM(sc); 3773 splx(s); 3774 tcpstat.tcps_sc_reset++; 3775 SYN_CACHE_PUT(sc); 3776 } 3777 3778 void 3779 syn_cache_unreach(struct sockaddr *src, struct sockaddr *dst, 3780 struct tcphdr *th) 3781 { 3782 struct syn_cache *sc; 3783 struct syn_cache_head *scp; 3784 int s; 3785 3786 s = splsoftnet(); 3787 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3788 splx(s); 3789 return; 3790 } 3791 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3792 if (ntohl (th->th_seq) != sc->sc_iss) { 3793 splx(s); 3794 return; 3795 } 3796 3797 /* 3798 * If we've retransmitted 3 times and this is our second error, 3799 * we remove the entry. Otherwise, we allow it to continue on. 3800 * This prevents us from incorrectly nuking an entry during a 3801 * spurious network outage. 3802 * 3803 * See tcp_notify(). 3804 */ 3805 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3806 sc->sc_flags |= SCF_UNREACH; 3807 splx(s); 3808 return; 3809 } 3810 3811 SYN_CACHE_RM(sc); 3812 splx(s); 3813 tcpstat.tcps_sc_unreach++; 3814 SYN_CACHE_PUT(sc); 3815 } 3816 3817 /* 3818 * Given a LISTEN socket and an inbound SYN request, add 3819 * this to the syn cache, and send back a segment: 3820 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3821 * to the source. 3822 * 3823 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3824 * Doing so would require that we hold onto the data and deliver it 3825 * to the application. However, if we are the target of a SYN-flood 3826 * DoS attack, an attacker could send data which would eventually 3827 * consume all available buffer space if it were ACKed. By not ACKing 3828 * the data, we avoid this DoS scenario. 3829 */ 3830 3831 int 3832 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3833 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 3834 int optlen, struct tcp_opt_info *oi) 3835 { 3836 struct tcpcb tb, *tp; 3837 long win; 3838 struct syn_cache *sc; 3839 struct syn_cache_head *scp; 3840 struct mbuf *ipopts; 3841 struct tcp_opt_info opti; 3842 3843 tp = sototcpcb(so); 3844 3845 bzero(&opti, sizeof(opti)); 3846 3847 /* 3848 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3849 * 3850 * Note this check is performed in tcp_input() very early on. 3851 */ 3852 3853 /* 3854 * Initialize some local state. 3855 */ 3856 win = sbspace(&so->so_rcv); 3857 if (win > TCP_MAXWIN) 3858 win = TCP_MAXWIN; 3859 3860 switch (src->sa_family) { 3861 #ifdef INET 3862 case AF_INET: 3863 /* 3864 * Remember the IP options, if any. 3865 */ 3866 ipopts = ip_srcroute(); 3867 break; 3868 #endif 3869 default: 3870 ipopts = NULL; 3871 } 3872 3873 #ifdef TCP_SIGNATURE 3874 if (optp || (tp->t_flags & TF_SIGNATURE)) 3875 #else 3876 if (optp) 3877 #endif 3878 { 3879 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3880 #ifdef TCP_SIGNATURE 3881 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 3882 #endif 3883 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 3884 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 3885 return (0); 3886 } else 3887 tb.t_flags = 0; 3888 3889 /* 3890 * See if we already have an entry for this connection. 3891 * If we do, resend the SYN,ACK. We do not count this 3892 * as a retransmission (XXX though maybe we should). 3893 */ 3894 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3895 tcpstat.tcps_sc_dupesyn++; 3896 if (ipopts) { 3897 /* 3898 * If we were remembering a previous source route, 3899 * forget it and use the new one we've been given. 3900 */ 3901 if (sc->sc_ipopts) 3902 (void) m_free(sc->sc_ipopts); 3903 sc->sc_ipopts = ipopts; 3904 } 3905 sc->sc_timestamp = tb.ts_recent; 3906 if (syn_cache_respond(sc, m) == 0) { 3907 tcpstat.tcps_sndacks++; 3908 tcpstat.tcps_sndtotal++; 3909 } 3910 return (1); 3911 } 3912 3913 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3914 if (sc == NULL) { 3915 if (ipopts) 3916 (void) m_free(ipopts); 3917 return (0); 3918 } 3919 3920 /* 3921 * Fill in the cache, and put the necessary IP and TCP 3922 * options into the reply. 3923 */ 3924 bzero(sc, sizeof(struct syn_cache)); 3925 callout_init(&sc->sc_timer); 3926 bcopy(src, &sc->sc_src, src->sa_len); 3927 bcopy(dst, &sc->sc_dst, dst->sa_len); 3928 sc->sc_flags = 0; 3929 sc->sc_ipopts = ipopts; 3930 sc->sc_irs = th->th_seq; 3931 switch (src->sa_family) { 3932 #ifdef INET 3933 case AF_INET: 3934 { 3935 struct sockaddr_in *srcin = (void *) src; 3936 struct sockaddr_in *dstin = (void *) dst; 3937 3938 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 3939 &srcin->sin_addr, dstin->sin_port, 3940 srcin->sin_port, sizeof(dstin->sin_addr), 0); 3941 break; 3942 } 3943 #endif /* INET */ 3944 #ifdef INET6 3945 case AF_INET6: 3946 { 3947 struct sockaddr_in6 *srcin6 = (void *) src; 3948 struct sockaddr_in6 *dstin6 = (void *) dst; 3949 3950 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 3951 &srcin6->sin6_addr, dstin6->sin6_port, 3952 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 3953 break; 3954 } 3955 #endif /* INET6 */ 3956 } 3957 sc->sc_peermaxseg = oi->maxseg; 3958 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3959 m->m_pkthdr.rcvif : NULL, 3960 sc->sc_src.sa.sa_family); 3961 sc->sc_win = win; 3962 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 3963 sc->sc_timestamp = tb.ts_recent; 3964 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 3965 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 3966 sc->sc_flags |= SCF_TIMESTAMP; 3967 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3968 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3969 sc->sc_requested_s_scale = tb.requested_s_scale; 3970 sc->sc_request_r_scale = 0; 3971 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3972 TCP_MAXWIN << sc->sc_request_r_scale < 3973 so->so_rcv.sb_hiwat) 3974 sc->sc_request_r_scale++; 3975 } else { 3976 sc->sc_requested_s_scale = 15; 3977 sc->sc_request_r_scale = 15; 3978 } 3979 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 3980 sc->sc_flags |= SCF_SACK_PERMIT; 3981 #ifdef TCP_SIGNATURE 3982 if (tb.t_flags & TF_SIGNATURE) 3983 sc->sc_flags |= SCF_SIGNATURE; 3984 #endif 3985 sc->sc_tp = tp; 3986 if (syn_cache_respond(sc, m) == 0) { 3987 syn_cache_insert(sc, tp); 3988 tcpstat.tcps_sndacks++; 3989 tcpstat.tcps_sndtotal++; 3990 } else { 3991 SYN_CACHE_PUT(sc); 3992 tcpstat.tcps_sc_dropped++; 3993 } 3994 return (1); 3995 } 3996 3997 int 3998 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 3999 { 4000 struct route *ro; 4001 u_int8_t *optp; 4002 int optlen, error; 4003 u_int16_t tlen; 4004 struct ip *ip = NULL; 4005 #ifdef INET6 4006 struct ip6_hdr *ip6 = NULL; 4007 #endif 4008 struct tcpcb *tp; 4009 struct tcphdr *th; 4010 u_int hlen; 4011 struct socket *so; 4012 4013 switch (sc->sc_src.sa.sa_family) { 4014 case AF_INET: 4015 hlen = sizeof(struct ip); 4016 ro = &sc->sc_route4; 4017 break; 4018 #ifdef INET6 4019 case AF_INET6: 4020 hlen = sizeof(struct ip6_hdr); 4021 ro = (struct route *)&sc->sc_route6; 4022 break; 4023 #endif 4024 default: 4025 if (m) 4026 m_freem(m); 4027 return (EAFNOSUPPORT); 4028 } 4029 4030 /* Compute the size of the TCP options. */ 4031 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4032 ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) + 4033 #ifdef TCP_SIGNATURE 4034 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4035 #endif 4036 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4037 4038 tlen = hlen + sizeof(struct tcphdr) + optlen; 4039 4040 /* 4041 * Create the IP+TCP header from scratch. 4042 */ 4043 if (m) 4044 m_freem(m); 4045 #ifdef DIAGNOSTIC 4046 if (max_linkhdr + tlen > MCLBYTES) 4047 return (ENOBUFS); 4048 #endif 4049 MGETHDR(m, M_DONTWAIT, MT_DATA); 4050 if (m && tlen > MHLEN) { 4051 MCLGET(m, M_DONTWAIT); 4052 if ((m->m_flags & M_EXT) == 0) { 4053 m_freem(m); 4054 m = NULL; 4055 } 4056 } 4057 if (m == NULL) 4058 return (ENOBUFS); 4059 MCLAIM(m, &tcp_tx_mowner); 4060 4061 /* Fixup the mbuf. */ 4062 m->m_data += max_linkhdr; 4063 m->m_len = m->m_pkthdr.len = tlen; 4064 if (sc->sc_tp) { 4065 tp = sc->sc_tp; 4066 if (tp->t_inpcb) 4067 so = tp->t_inpcb->inp_socket; 4068 #ifdef INET6 4069 else if (tp->t_in6pcb) 4070 so = tp->t_in6pcb->in6p_socket; 4071 #endif 4072 else 4073 so = NULL; 4074 } else 4075 so = NULL; 4076 m->m_pkthdr.rcvif = NULL; 4077 memset(mtod(m, u_char *), 0, tlen); 4078 4079 switch (sc->sc_src.sa.sa_family) { 4080 case AF_INET: 4081 ip = mtod(m, struct ip *); 4082 ip->ip_v = 4; 4083 ip->ip_dst = sc->sc_src.sin.sin_addr; 4084 ip->ip_src = sc->sc_dst.sin.sin_addr; 4085 ip->ip_p = IPPROTO_TCP; 4086 th = (struct tcphdr *)(ip + 1); 4087 th->th_dport = sc->sc_src.sin.sin_port; 4088 th->th_sport = sc->sc_dst.sin.sin_port; 4089 break; 4090 #ifdef INET6 4091 case AF_INET6: 4092 ip6 = mtod(m, struct ip6_hdr *); 4093 ip6->ip6_vfc = IPV6_VERSION; 4094 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4095 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4096 ip6->ip6_nxt = IPPROTO_TCP; 4097 /* ip6_plen will be updated in ip6_output() */ 4098 th = (struct tcphdr *)(ip6 + 1); 4099 th->th_dport = sc->sc_src.sin6.sin6_port; 4100 th->th_sport = sc->sc_dst.sin6.sin6_port; 4101 break; 4102 #endif 4103 default: 4104 th = NULL; 4105 } 4106 4107 th->th_seq = htonl(sc->sc_iss); 4108 th->th_ack = htonl(sc->sc_irs + 1); 4109 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4110 th->th_flags = TH_SYN|TH_ACK; 4111 th->th_win = htons(sc->sc_win); 4112 /* th_sum already 0 */ 4113 /* th_urp already 0 */ 4114 4115 /* Tack on the TCP options. */ 4116 optp = (u_int8_t *)(th + 1); 4117 *optp++ = TCPOPT_MAXSEG; 4118 *optp++ = 4; 4119 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4120 *optp++ = sc->sc_ourmaxseg & 0xff; 4121 4122 if (sc->sc_request_r_scale != 15) { 4123 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4124 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4125 sc->sc_request_r_scale); 4126 optp += 4; 4127 } 4128 4129 if (sc->sc_flags & SCF_TIMESTAMP) { 4130 u_int32_t *lp = (u_int32_t *)(optp); 4131 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4132 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4133 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4134 *lp = htonl(sc->sc_timestamp); 4135 optp += TCPOLEN_TSTAMP_APPA; 4136 } 4137 4138 if (sc->sc_flags & SCF_SACK_PERMIT) { 4139 u_int8_t *p = optp; 4140 4141 /* Let the peer know that we will SACK. */ 4142 p[0] = TCPOPT_SACK_PERMITTED; 4143 p[1] = 2; 4144 p[2] = TCPOPT_NOP; 4145 p[3] = TCPOPT_NOP; 4146 optp += 4; 4147 } 4148 4149 #ifdef TCP_SIGNATURE 4150 if (sc->sc_flags & SCF_SIGNATURE) { 4151 struct secasvar *sav; 4152 u_int8_t *sigp; 4153 4154 sav = tcp_signature_getsav(m, th); 4155 4156 if (sav == NULL) { 4157 if (m) 4158 m_freem(m); 4159 return (EPERM); 4160 } 4161 4162 *optp++ = TCPOPT_SIGNATURE; 4163 *optp++ = TCPOLEN_SIGNATURE; 4164 sigp = optp; 4165 bzero(optp, TCP_SIGLEN); 4166 optp += TCP_SIGLEN; 4167 *optp++ = TCPOPT_NOP; 4168 *optp++ = TCPOPT_EOL; 4169 4170 (void)tcp_signature(m, th, hlen, sav, sigp); 4171 4172 key_sa_recordxfer(sav, m); 4173 #ifdef FAST_IPSEC 4174 KEY_FREESAV(&sav); 4175 #else 4176 key_freesav(sav); 4177 #endif 4178 } 4179 #endif 4180 4181 /* Compute the packet's checksum. */ 4182 switch (sc->sc_src.sa.sa_family) { 4183 case AF_INET: 4184 ip->ip_len = htons(tlen - hlen); 4185 th->th_sum = 0; 4186 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4187 break; 4188 #ifdef INET6 4189 case AF_INET6: 4190 ip6->ip6_plen = htons(tlen - hlen); 4191 th->th_sum = 0; 4192 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4193 break; 4194 #endif 4195 } 4196 4197 /* 4198 * Fill in some straggling IP bits. Note the stack expects 4199 * ip_len to be in host order, for convenience. 4200 */ 4201 switch (sc->sc_src.sa.sa_family) { 4202 #ifdef INET 4203 case AF_INET: 4204 ip->ip_len = htons(tlen); 4205 ip->ip_ttl = ip_defttl; 4206 /* XXX tos? */ 4207 break; 4208 #endif 4209 #ifdef INET6 4210 case AF_INET6: 4211 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4212 ip6->ip6_vfc |= IPV6_VERSION; 4213 ip6->ip6_plen = htons(tlen - hlen); 4214 /* ip6_hlim will be initialized afterwards */ 4215 /* XXX flowlabel? */ 4216 break; 4217 #endif 4218 } 4219 4220 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4221 tp = sc->sc_tp; 4222 4223 switch (sc->sc_src.sa.sa_family) { 4224 #ifdef INET 4225 case AF_INET: 4226 error = ip_output(m, sc->sc_ipopts, ro, 4227 (ip_mtudisc ? IP_MTUDISC : 0), 4228 (struct ip_moptions *)NULL, so); 4229 break; 4230 #endif 4231 #ifdef INET6 4232 case AF_INET6: 4233 ip6->ip6_hlim = in6_selecthlim(NULL, 4234 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 4235 4236 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0, 4237 (struct ip6_moptions *)0, so, NULL); 4238 break; 4239 #endif 4240 default: 4241 error = EAFNOSUPPORT; 4242 break; 4243 } 4244 return (error); 4245 } 4246
Indexes created Sun Sep 21 20:09:37 GMT 2025