tcp_input.c revision 1.213
1 /* $NetBSD: tcp_input.c,v 1.213 2005/01/26 21:49:27 mycroft 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 1.213 2005/01/26 21:49:27 mycroft 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 (24 * 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 __P((const struct ip *, const struct tcphdr *)); 342 #endif 343 #ifdef INET6 344 static void tcp6_log_refused 345 __P((const struct ip6_hdr *, const struct tcphdr *)); 346 #endif 347 348 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 349 350 POOL_INIT(tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", NULL); 351 352 int 353 tcp_reass(tp, th, m, tlen) 354 struct tcpcb *tp; 355 struct tcphdr *th; 356 struct mbuf *m; 357 int *tlen; 358 { 359 struct ipqent *p, *q, *nq, *tiqe = NULL; 360 struct socket *so = NULL; 361 int pkt_flags; 362 tcp_seq pkt_seq; 363 unsigned pkt_len; 364 u_long rcvpartdupbyte = 0; 365 u_long rcvoobyte; 366 #ifdef TCP_REASS_COUNTERS 367 u_int count = 0; 368 #endif 369 370 if (tp->t_inpcb) 371 so = tp->t_inpcb->inp_socket; 372 #ifdef INET6 373 else if (tp->t_in6pcb) 374 so = tp->t_in6pcb->in6p_socket; 375 #endif 376 377 TCP_REASS_LOCK_CHECK(tp); 378 379 /* 380 * Call with th==0 after become established to 381 * force pre-ESTABLISHED data up to user socket. 382 */ 383 if (th == 0) 384 goto present; 385 386 rcvoobyte = *tlen; 387 /* 388 * Copy these to local variables because the tcpiphdr 389 * gets munged while we are collapsing mbufs. 390 */ 391 pkt_seq = th->th_seq; 392 pkt_len = *tlen; 393 pkt_flags = th->th_flags; 394 395 TCP_REASS_COUNTER_INCR(&tcp_reass_); 396 397 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 398 /* 399 * When we miss a packet, the vast majority of time we get 400 * packets that follow it in order. So optimize for that. 401 */ 402 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 403 p->ipqe_len += pkt_len; 404 p->ipqe_flags |= pkt_flags; 405 m_cat(p->ipre_mlast, m); 406 TRAVERSE(p->ipre_mlast); 407 m = NULL; 408 tiqe = p; 409 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 410 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 411 goto skip_replacement; 412 } 413 /* 414 * While we're here, if the pkt is completely beyond 415 * anything we have, just insert it at the tail. 416 */ 417 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 418 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 419 goto insert_it; 420 } 421 } 422 423 q = TAILQ_FIRST(&tp->segq); 424 425 if (q != NULL) { 426 /* 427 * If this segment immediately precedes the first out-of-order 428 * block, simply slap the segment in front of it and (mostly) 429 * skip the complicated logic. 430 */ 431 if (pkt_seq + pkt_len == q->ipqe_seq) { 432 q->ipqe_seq = pkt_seq; 433 q->ipqe_len += pkt_len; 434 q->ipqe_flags |= pkt_flags; 435 m_cat(m, q->ipqe_m); 436 q->ipqe_m = m; 437 q->ipre_mlast = m; /* last mbuf may have changed */ 438 TRAVERSE(q->ipre_mlast); 439 tiqe = q; 440 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 441 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 442 goto skip_replacement; 443 } 444 } else { 445 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 446 } 447 448 /* 449 * Find a segment which begins after this one does. 450 */ 451 for (p = NULL; q != NULL; q = nq) { 452 nq = TAILQ_NEXT(q, ipqe_q); 453 #ifdef TCP_REASS_COUNTERS 454 count++; 455 #endif 456 /* 457 * If the received segment is just right after this 458 * fragment, merge the two together and then check 459 * for further overlaps. 460 */ 461 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 462 #ifdef TCPREASS_DEBUG 463 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 464 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 465 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 466 #endif 467 pkt_len += q->ipqe_len; 468 pkt_flags |= q->ipqe_flags; 469 pkt_seq = q->ipqe_seq; 470 m_cat(q->ipre_mlast, m); 471 TRAVERSE(q->ipre_mlast); 472 m = q->ipqe_m; 473 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 474 goto free_ipqe; 475 } 476 /* 477 * If the received segment is completely past this 478 * fragment, we need to go the next fragment. 479 */ 480 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 481 p = q; 482 continue; 483 } 484 /* 485 * If the fragment is past the received segment, 486 * it (or any following) can't be concatenated. 487 */ 488 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 489 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 490 break; 491 } 492 493 /* 494 * We've received all the data in this segment before. 495 * mark it as a duplicate and return. 496 */ 497 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 498 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 499 tcpstat.tcps_rcvduppack++; 500 tcpstat.tcps_rcvdupbyte += pkt_len; 501 m_freem(m); 502 if (tiqe != NULL) 503 pool_put(&tcpipqent_pool, tiqe); 504 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 505 return (0); 506 } 507 /* 508 * Received segment completely overlaps this fragment 509 * so we drop the fragment (this keeps the temporal 510 * ordering of segments correct). 511 */ 512 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 513 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 514 rcvpartdupbyte += q->ipqe_len; 515 m_freem(q->ipqe_m); 516 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 517 goto free_ipqe; 518 } 519 /* 520 * RX'ed segment extends past the end of the 521 * fragment. Drop the overlapping bytes. Then 522 * merge the fragment and segment then treat as 523 * a longer received packet. 524 */ 525 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 526 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 527 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 528 #ifdef TCPREASS_DEBUG 529 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 530 tp, overlap, 531 pkt_seq, pkt_seq + pkt_len, pkt_len); 532 #endif 533 m_adj(m, overlap); 534 rcvpartdupbyte += overlap; 535 m_cat(q->ipre_mlast, m); 536 TRAVERSE(q->ipre_mlast); 537 m = q->ipqe_m; 538 pkt_seq = q->ipqe_seq; 539 pkt_len += q->ipqe_len - overlap; 540 rcvoobyte -= overlap; 541 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 542 goto free_ipqe; 543 } 544 /* 545 * RX'ed segment extends past the front of the 546 * fragment. Drop the overlapping bytes on the 547 * received packet. The packet will then be 548 * contatentated with this fragment a bit later. 549 */ 550 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 551 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 552 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 553 #ifdef TCPREASS_DEBUG 554 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 555 tp, overlap, 556 pkt_seq, pkt_seq + pkt_len, pkt_len); 557 #endif 558 m_adj(m, -overlap); 559 pkt_len -= overlap; 560 rcvpartdupbyte += overlap; 561 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 562 rcvoobyte -= overlap; 563 } 564 /* 565 * If the received segment immediates precedes this 566 * fragment then tack the fragment onto this segment 567 * and reinsert the data. 568 */ 569 if (q->ipqe_seq == pkt_seq + pkt_len) { 570 #ifdef TCPREASS_DEBUG 571 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 572 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 573 pkt_seq, pkt_seq + pkt_len, pkt_len); 574 #endif 575 pkt_len += q->ipqe_len; 576 pkt_flags |= q->ipqe_flags; 577 m_cat(m, q->ipqe_m); 578 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 579 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 580 if (tiqe == NULL) 581 tiqe = q; 582 else 583 pool_put(&tcpipqent_pool, q); 584 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 585 break; 586 } 587 /* 588 * If the fragment is before the segment, remember it. 589 * When this loop is terminated, p will contain the 590 * pointer to fragment that is right before the received 591 * segment. 592 */ 593 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 594 p = q; 595 596 continue; 597 598 /* 599 * This is a common operation. It also will allow 600 * to save doing a malloc/free in most instances. 601 */ 602 free_ipqe: 603 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 604 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 605 if (tiqe == NULL) 606 tiqe = q; 607 else 608 pool_put(&tcpipqent_pool, q); 609 } 610 611 #ifdef TCP_REASS_COUNTERS 612 if (count > 7) 613 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 614 else if (count > 0) 615 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 616 #endif 617 618 insert_it: 619 620 /* 621 * Allocate a new queue entry since the received segment did not 622 * collapse onto any other out-of-order block; thus we are allocating 623 * a new block. If it had collapsed, tiqe would not be NULL and 624 * we would be reusing it. 625 * XXX If we can't, just drop the packet. XXX 626 */ 627 if (tiqe == NULL) { 628 tiqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 629 if (tiqe == NULL) { 630 tcpstat.tcps_rcvmemdrop++; 631 m_freem(m); 632 return (0); 633 } 634 } 635 636 /* 637 * Update the counters. 638 */ 639 tcpstat.tcps_rcvoopack++; 640 tcpstat.tcps_rcvoobyte += rcvoobyte; 641 if (rcvpartdupbyte) { 642 tcpstat.tcps_rcvpartduppack++; 643 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 644 } 645 646 /* 647 * Insert the new fragment queue entry into both queues. 648 */ 649 tiqe->ipqe_m = m; 650 tiqe->ipre_mlast = m; 651 tiqe->ipqe_seq = pkt_seq; 652 tiqe->ipqe_len = pkt_len; 653 tiqe->ipqe_flags = pkt_flags; 654 if (p == NULL) { 655 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 656 #ifdef TCPREASS_DEBUG 657 if (tiqe->ipqe_seq != tp->rcv_nxt) 658 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 659 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 660 #endif 661 } else { 662 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 663 #ifdef TCPREASS_DEBUG 664 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 665 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 666 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 667 #endif 668 } 669 670 skip_replacement: 671 672 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 673 674 present: 675 /* 676 * Present data to user, advancing rcv_nxt through 677 * completed sequence space. 678 */ 679 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 680 return (0); 681 q = TAILQ_FIRST(&tp->segq); 682 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 683 return (0); 684 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 685 return (0); 686 687 tp->rcv_nxt += q->ipqe_len; 688 pkt_flags = q->ipqe_flags & TH_FIN; 689 ND6_HINT(tp); 690 691 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 692 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 693 if (so->so_state & SS_CANTRCVMORE) 694 m_freem(q->ipqe_m); 695 else 696 sbappendstream(&so->so_rcv, q->ipqe_m); 697 pool_put(&tcpipqent_pool, q); 698 sorwakeup(so); 699 return (pkt_flags); 700 } 701 702 #ifdef INET6 703 int 704 tcp6_input(mp, offp, proto) 705 struct mbuf **mp; 706 int *offp, proto; 707 { 708 struct mbuf *m = *mp; 709 710 /* 711 * draft-itojun-ipv6-tcp-to-anycast 712 * better place to put this in? 713 */ 714 if (m->m_flags & M_ANYCAST6) { 715 struct ip6_hdr *ip6; 716 if (m->m_len < sizeof(struct ip6_hdr)) { 717 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 718 tcpstat.tcps_rcvshort++; 719 return IPPROTO_DONE; 720 } 721 } 722 ip6 = mtod(m, struct ip6_hdr *); 723 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 724 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 725 return IPPROTO_DONE; 726 } 727 728 tcp_input(m, *offp, proto); 729 return IPPROTO_DONE; 730 } 731 #endif 732 733 #ifdef INET 734 static void 735 tcp4_log_refused(ip, th) 736 const struct ip *ip; 737 const struct tcphdr *th; 738 { 739 char src[4*sizeof "123"]; 740 char dst[4*sizeof "123"]; 741 742 if (ip) { 743 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 744 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 745 } 746 else { 747 strlcpy(src, "(unknown)", sizeof(src)); 748 strlcpy(dst, "(unknown)", sizeof(dst)); 749 } 750 log(LOG_INFO, 751 "Connection attempt to TCP %s:%d from %s:%d\n", 752 dst, ntohs(th->th_dport), 753 src, ntohs(th->th_sport)); 754 } 755 #endif 756 757 #ifdef INET6 758 static void 759 tcp6_log_refused(ip6, th) 760 const struct ip6_hdr *ip6; 761 const struct tcphdr *th; 762 { 763 char src[INET6_ADDRSTRLEN]; 764 char dst[INET6_ADDRSTRLEN]; 765 766 if (ip6) { 767 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 768 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 769 } 770 else { 771 strlcpy(src, "(unknown v6)", sizeof(src)); 772 strlcpy(dst, "(unknown v6)", sizeof(dst)); 773 } 774 log(LOG_INFO, 775 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 776 dst, ntohs(th->th_dport), 777 src, ntohs(th->th_sport)); 778 } 779 #endif 780 781 /* 782 * Checksum extended TCP header and data. 783 */ 784 int 785 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, int toff, 786 int off, int tlen) 787 { 788 789 /* 790 * XXX it's better to record and check if this mbuf is 791 * already checked. 792 */ 793 794 switch (af) { 795 #ifdef INET 796 case AF_INET: 797 switch (m->m_pkthdr.csum_flags & 798 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 799 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 800 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 801 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 802 goto badcsum; 803 804 case M_CSUM_TCPv4|M_CSUM_DATA: { 805 u_int32_t hw_csum = m->m_pkthdr.csum_data; 806 807 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 808 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 809 const struct ip *ip = 810 mtod(m, const struct ip *); 811 812 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 813 ip->ip_dst.s_addr, 814 htons(hw_csum + tlen + off + IPPROTO_TCP)); 815 } 816 if ((hw_csum ^ 0xffff) != 0) 817 goto badcsum; 818 break; 819 } 820 821 case M_CSUM_TCPv4: 822 /* Checksum was okay. */ 823 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 824 break; 825 826 default: 827 /* 828 * Must compute it ourselves. Maybe skip checksum 829 * on loopback interfaces. 830 */ 831 if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & 832 IFF_LOOPBACK) || 833 tcp_do_loopback_cksum)) { 834 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 835 if (in4_cksum(m, IPPROTO_TCP, toff, 836 tlen + off) != 0) 837 goto badcsum; 838 } 839 break; 840 } 841 break; 842 #endif /* INET4 */ 843 844 #ifdef INET6 845 case AF_INET6: 846 if (__predict_true((m->m_flags & M_LOOP) == 0 || 847 tcp_do_loopback_cksum)) { 848 if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 849 goto badcsum; 850 } 851 break; 852 #endif /* INET6 */ 853 } 854 855 return 0; 856 857 badcsum: 858 tcpstat.tcps_rcvbadsum++; 859 return -1; 860 } 861 862 /* 863 * TCP input routine, follows pages 65-76 of the 864 * protocol specification dated September, 1981 very closely. 865 */ 866 void 867 tcp_input(struct mbuf *m, ...) 868 { 869 struct tcphdr *th; 870 struct ip *ip; 871 struct inpcb *inp; 872 #ifdef INET6 873 struct ip6_hdr *ip6; 874 struct in6pcb *in6p; 875 #endif 876 u_int8_t *optp = NULL; 877 int optlen = 0; 878 int len, tlen, toff, hdroptlen = 0; 879 struct tcpcb *tp = 0; 880 int tiflags; 881 struct socket *so = NULL; 882 int todrop, acked, ourfinisacked, needoutput = 0; 883 #ifdef TCP_DEBUG 884 short ostate = 0; 885 #endif 886 int iss = 0; 887 u_long tiwin; 888 struct tcp_opt_info opti; 889 int off, iphlen; 890 va_list ap; 891 int af; /* af on the wire */ 892 struct mbuf *tcp_saveti = NULL; 893 894 MCLAIM(m, &tcp_rx_mowner); 895 va_start(ap, m); 896 toff = va_arg(ap, int); 897 (void)va_arg(ap, int); /* ignore value, advance ap */ 898 va_end(ap); 899 900 tcpstat.tcps_rcvtotal++; 901 902 bzero(&opti, sizeof(opti)); 903 opti.ts_present = 0; 904 opti.maxseg = 0; 905 906 /* 907 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 908 * 909 * TCP is, by definition, unicast, so we reject all 910 * multicast outright. 911 * 912 * Note, there are additional src/dst address checks in 913 * the AF-specific code below. 914 */ 915 if (m->m_flags & (M_BCAST|M_MCAST)) { 916 /* XXX stat */ 917 goto drop; 918 } 919 #ifdef INET6 920 if (m->m_flags & M_ANYCAST6) { 921 /* XXX stat */ 922 goto drop; 923 } 924 #endif 925 926 /* 927 * Get IP and TCP header. 928 * Note: IP leaves IP header in first mbuf. 929 */ 930 ip = mtod(m, struct ip *); 931 #ifdef INET6 932 ip6 = NULL; 933 #endif 934 switch (ip->ip_v) { 935 #ifdef INET 936 case 4: 937 af = AF_INET; 938 iphlen = sizeof(struct ip); 939 ip = mtod(m, struct ip *); 940 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 941 sizeof(struct tcphdr)); 942 if (th == NULL) { 943 tcpstat.tcps_rcvshort++; 944 return; 945 } 946 /* We do the checksum after PCB lookup... */ 947 len = ntohs(ip->ip_len); 948 tlen = len - toff; 949 break; 950 #endif 951 #ifdef INET6 952 case 6: 953 ip = NULL; 954 iphlen = sizeof(struct ip6_hdr); 955 af = AF_INET6; 956 ip6 = mtod(m, struct ip6_hdr *); 957 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 958 sizeof(struct tcphdr)); 959 if (th == NULL) { 960 tcpstat.tcps_rcvshort++; 961 return; 962 } 963 964 /* Be proactive about malicious use of IPv4 mapped address */ 965 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 966 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 967 /* XXX stat */ 968 goto drop; 969 } 970 971 /* 972 * Be proactive about unspecified IPv6 address in source. 973 * As we use all-zero to indicate unbounded/unconnected pcb, 974 * unspecified IPv6 address can be used to confuse us. 975 * 976 * Note that packets with unspecified IPv6 destination is 977 * already dropped in ip6_input. 978 */ 979 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 980 /* XXX stat */ 981 goto drop; 982 } 983 984 /* 985 * Make sure destination address is not multicast. 986 * Source address checked in ip6_input(). 987 */ 988 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 989 /* XXX stat */ 990 goto drop; 991 } 992 993 /* We do the checksum after PCB lookup... */ 994 len = m->m_pkthdr.len; 995 tlen = len - toff; 996 break; 997 #endif 998 default: 999 m_freem(m); 1000 return; 1001 } 1002 1003 KASSERT(TCP_HDR_ALIGNED_P(th)); 1004 1005 /* 1006 * Check that TCP offset makes sense, 1007 * pull out TCP options and adjust length. XXX 1008 */ 1009 off = th->th_off << 2; 1010 if (off < sizeof (struct tcphdr) || off > tlen) { 1011 tcpstat.tcps_rcvbadoff++; 1012 goto drop; 1013 } 1014 tlen -= off; 1015 1016 /* 1017 * tcp_input() has been modified to use tlen to mean the TCP data 1018 * length throughout the function. Other functions can use 1019 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1020 * rja 1021 */ 1022 1023 if (off > sizeof (struct tcphdr)) { 1024 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1025 if (th == NULL) { 1026 tcpstat.tcps_rcvshort++; 1027 return; 1028 } 1029 /* 1030 * NOTE: ip/ip6 will not be affected by m_pulldown() 1031 * (as they're before toff) and we don't need to update those. 1032 */ 1033 KASSERT(TCP_HDR_ALIGNED_P(th)); 1034 optlen = off - sizeof (struct tcphdr); 1035 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1036 /* 1037 * Do quick retrieval of timestamp options ("options 1038 * prediction?"). If timestamp is the only option and it's 1039 * formatted as recommended in RFC 1323 appendix A, we 1040 * quickly get the values now and not bother calling 1041 * tcp_dooptions(), etc. 1042 */ 1043 if ((optlen == TCPOLEN_TSTAMP_APPA || 1044 (optlen > TCPOLEN_TSTAMP_APPA && 1045 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1046 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1047 (th->th_flags & TH_SYN) == 0) { 1048 opti.ts_present = 1; 1049 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1050 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1051 optp = NULL; /* we've parsed the options */ 1052 } 1053 } 1054 tiflags = th->th_flags; 1055 1056 /* 1057 * Locate pcb for segment. 1058 */ 1059 findpcb: 1060 inp = NULL; 1061 #ifdef INET6 1062 in6p = NULL; 1063 #endif 1064 switch (af) { 1065 #ifdef INET 1066 case AF_INET: 1067 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1068 ip->ip_dst, th->th_dport); 1069 if (inp == 0) { 1070 ++tcpstat.tcps_pcbhashmiss; 1071 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1072 } 1073 #ifdef INET6 1074 if (inp == 0) { 1075 struct in6_addr s, d; 1076 1077 /* mapped addr case */ 1078 bzero(&s, sizeof(s)); 1079 s.s6_addr16[5] = htons(0xffff); 1080 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1081 bzero(&d, sizeof(d)); 1082 d.s6_addr16[5] = htons(0xffff); 1083 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1084 in6p = in6_pcblookup_connect(&tcbtable, &s, 1085 th->th_sport, &d, th->th_dport, 0); 1086 if (in6p == 0) { 1087 ++tcpstat.tcps_pcbhashmiss; 1088 in6p = in6_pcblookup_bind(&tcbtable, &d, 1089 th->th_dport, 0); 1090 } 1091 } 1092 #endif 1093 #ifndef INET6 1094 if (inp == 0) 1095 #else 1096 if (inp == 0 && in6p == 0) 1097 #endif 1098 { 1099 ++tcpstat.tcps_noport; 1100 if (tcp_log_refused && 1101 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1102 tcp4_log_refused(ip, th); 1103 } 1104 TCP_FIELDS_TO_HOST(th); 1105 goto dropwithreset_ratelim; 1106 } 1107 #if defined(IPSEC) || defined(FAST_IPSEC) 1108 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1109 ipsec4_in_reject(m, inp)) { 1110 ipsecstat.in_polvio++; 1111 goto drop; 1112 } 1113 #ifdef INET6 1114 else if (in6p && 1115 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1116 ipsec4_in_reject_so(m, in6p->in6p_socket)) { 1117 ipsecstat.in_polvio++; 1118 goto drop; 1119 } 1120 #endif 1121 #endif /*IPSEC*/ 1122 break; 1123 #endif /*INET*/ 1124 #ifdef INET6 1125 case AF_INET6: 1126 { 1127 int faith; 1128 1129 #if defined(NFAITH) && NFAITH > 0 1130 faith = faithprefix(&ip6->ip6_dst); 1131 #else 1132 faith = 0; 1133 #endif 1134 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1135 th->th_sport, &ip6->ip6_dst, th->th_dport, faith); 1136 if (in6p == NULL) { 1137 ++tcpstat.tcps_pcbhashmiss; 1138 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1139 th->th_dport, faith); 1140 } 1141 if (in6p == NULL) { 1142 ++tcpstat.tcps_noport; 1143 if (tcp_log_refused && 1144 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1145 tcp6_log_refused(ip6, th); 1146 } 1147 TCP_FIELDS_TO_HOST(th); 1148 goto dropwithreset_ratelim; 1149 } 1150 #if defined(IPSEC) || defined(FAST_IPSEC) 1151 if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1152 ipsec6_in_reject(m, in6p)) { 1153 ipsec6stat.in_polvio++; 1154 goto drop; 1155 } 1156 #endif /*IPSEC*/ 1157 break; 1158 } 1159 #endif 1160 } 1161 1162 /* 1163 * If the state is CLOSED (i.e., TCB does not exist) then 1164 * all data in the incoming segment is discarded. 1165 * If the TCB exists but is in CLOSED state, it is embryonic, 1166 * but should either do a listen or a connect soon. 1167 */ 1168 tp = NULL; 1169 so = NULL; 1170 if (inp) { 1171 tp = intotcpcb(inp); 1172 so = inp->inp_socket; 1173 } 1174 #ifdef INET6 1175 else if (in6p) { 1176 tp = in6totcpcb(in6p); 1177 so = in6p->in6p_socket; 1178 } 1179 #endif 1180 if (tp == 0) { 1181 TCP_FIELDS_TO_HOST(th); 1182 goto dropwithreset_ratelim; 1183 } 1184 if (tp->t_state == TCPS_CLOSED) 1185 goto drop; 1186 1187 /* 1188 * Checksum extended TCP header and data. 1189 */ 1190 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1191 goto badcsum; 1192 1193 TCP_FIELDS_TO_HOST(th); 1194 1195 /* Unscale the window into a 32-bit value. */ 1196 if ((tiflags & TH_SYN) == 0) 1197 tiwin = th->th_win << tp->snd_scale; 1198 else 1199 tiwin = th->th_win; 1200 1201 #ifdef INET6 1202 /* save packet options if user wanted */ 1203 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1204 if (in6p->in6p_options) { 1205 m_freem(in6p->in6p_options); 1206 in6p->in6p_options = 0; 1207 } 1208 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1209 } 1210 #endif 1211 1212 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1213 union syn_cache_sa src; 1214 union syn_cache_sa dst; 1215 1216 bzero(&src, sizeof(src)); 1217 bzero(&dst, sizeof(dst)); 1218 switch (af) { 1219 #ifdef INET 1220 case AF_INET: 1221 src.sin.sin_len = sizeof(struct sockaddr_in); 1222 src.sin.sin_family = AF_INET; 1223 src.sin.sin_addr = ip->ip_src; 1224 src.sin.sin_port = th->th_sport; 1225 1226 dst.sin.sin_len = sizeof(struct sockaddr_in); 1227 dst.sin.sin_family = AF_INET; 1228 dst.sin.sin_addr = ip->ip_dst; 1229 dst.sin.sin_port = th->th_dport; 1230 break; 1231 #endif 1232 #ifdef INET6 1233 case AF_INET6: 1234 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1235 src.sin6.sin6_family = AF_INET6; 1236 src.sin6.sin6_addr = ip6->ip6_src; 1237 src.sin6.sin6_port = th->th_sport; 1238 1239 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1240 dst.sin6.sin6_family = AF_INET6; 1241 dst.sin6.sin6_addr = ip6->ip6_dst; 1242 dst.sin6.sin6_port = th->th_dport; 1243 break; 1244 #endif /* INET6 */ 1245 default: 1246 goto badsyn; /*sanity*/ 1247 } 1248 1249 if (so->so_options & SO_DEBUG) { 1250 #ifdef TCP_DEBUG 1251 ostate = tp->t_state; 1252 #endif 1253 1254 tcp_saveti = NULL; 1255 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1256 goto nosave; 1257 1258 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1259 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1260 if (!tcp_saveti) 1261 goto nosave; 1262 } else { 1263 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1264 if (!tcp_saveti) 1265 goto nosave; 1266 MCLAIM(m, &tcp_mowner); 1267 tcp_saveti->m_len = iphlen; 1268 m_copydata(m, 0, iphlen, 1269 mtod(tcp_saveti, caddr_t)); 1270 } 1271 1272 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1273 m_freem(tcp_saveti); 1274 tcp_saveti = NULL; 1275 } else { 1276 tcp_saveti->m_len += sizeof(struct tcphdr); 1277 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1278 sizeof(struct tcphdr)); 1279 } 1280 nosave:; 1281 } 1282 if (so->so_options & SO_ACCEPTCONN) { 1283 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1284 if (tiflags & TH_RST) { 1285 syn_cache_reset(&src.sa, &dst.sa, th); 1286 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1287 (TH_ACK|TH_SYN)) { 1288 /* 1289 * Received a SYN,ACK. This should 1290 * never happen while we are in 1291 * LISTEN. Send an RST. 1292 */ 1293 goto badsyn; 1294 } else if (tiflags & TH_ACK) { 1295 so = syn_cache_get(&src.sa, &dst.sa, 1296 th, toff, tlen, so, m); 1297 if (so == NULL) { 1298 /* 1299 * We don't have a SYN for 1300 * this ACK; send an RST. 1301 */ 1302 goto badsyn; 1303 } else if (so == 1304 (struct socket *)(-1)) { 1305 /* 1306 * We were unable to create 1307 * the connection. If the 1308 * 3-way handshake was 1309 * completed, and RST has 1310 * been sent to the peer. 1311 * Since the mbuf might be 1312 * in use for the reply, 1313 * do not free it. 1314 */ 1315 m = NULL; 1316 } else { 1317 /* 1318 * We have created a 1319 * full-blown connection. 1320 */ 1321 tp = NULL; 1322 inp = NULL; 1323 #ifdef INET6 1324 in6p = NULL; 1325 #endif 1326 switch (so->so_proto->pr_domain->dom_family) { 1327 #ifdef INET 1328 case AF_INET: 1329 inp = sotoinpcb(so); 1330 tp = intotcpcb(inp); 1331 break; 1332 #endif 1333 #ifdef INET6 1334 case AF_INET6: 1335 in6p = sotoin6pcb(so); 1336 tp = in6totcpcb(in6p); 1337 break; 1338 #endif 1339 } 1340 if (tp == NULL) 1341 goto badsyn; /*XXX*/ 1342 tiwin <<= tp->snd_scale; 1343 goto after_listen; 1344 } 1345 } else { 1346 /* 1347 * None of RST, SYN or ACK was set. 1348 * This is an invalid packet for a 1349 * TCB in LISTEN state. Send a RST. 1350 */ 1351 goto badsyn; 1352 } 1353 } else { 1354 /* 1355 * Received a SYN. 1356 */ 1357 1358 #ifdef INET6 1359 /* 1360 * If deprecated address is forbidden, we do 1361 * not accept SYN to deprecated interface 1362 * address to prevent any new inbound 1363 * connection from getting established. 1364 * When we do not accept SYN, we send a TCP 1365 * RST, with deprecated source address (instead 1366 * of dropping it). We compromise it as it is 1367 * much better for peer to send a RST, and 1368 * RST will be the final packet for the 1369 * exchange. 1370 * 1371 * If we do not forbid deprecated addresses, we 1372 * accept the SYN packet. RFC2462 does not 1373 * suggest dropping SYN in this case. 1374 * If we decipher RFC2462 5.5.4, it says like 1375 * this: 1376 * 1. use of deprecated addr with existing 1377 * communication is okay - "SHOULD continue 1378 * to be used" 1379 * 2. use of it with new communication: 1380 * (2a) "SHOULD NOT be used if alternate 1381 * address with sufficient scope is 1382 * available" 1383 * (2b) nothing mentioned otherwise. 1384 * Here we fall into (2b) case as we have no 1385 * choice in our source address selection - we 1386 * must obey the peer. 1387 * 1388 * The wording in RFC2462 is confusing, and 1389 * there are multiple description text for 1390 * deprecated address handling - worse, they 1391 * are not exactly the same. I believe 5.5.4 1392 * is the best one, so we follow 5.5.4. 1393 */ 1394 if (af == AF_INET6 && !ip6_use_deprecated) { 1395 struct in6_ifaddr *ia6; 1396 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1397 &ip6->ip6_dst)) && 1398 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1399 tp = NULL; 1400 goto dropwithreset; 1401 } 1402 } 1403 #endif 1404 1405 #ifdef IPSEC 1406 switch (af) { 1407 #ifdef INET 1408 case AF_INET: 1409 if (ipsec4_in_reject_so(m, so)) { 1410 ipsecstat.in_polvio++; 1411 tp = NULL; 1412 goto dropwithreset; 1413 } 1414 break; 1415 #endif 1416 #ifdef INET6 1417 case AF_INET6: 1418 if (ipsec6_in_reject_so(m, so)) { 1419 ipsec6stat.in_polvio++; 1420 tp = NULL; 1421 goto dropwithreset; 1422 } 1423 break; 1424 #endif 1425 } 1426 #endif 1427 1428 /* 1429 * LISTEN socket received a SYN 1430 * from itself? This can't possibly 1431 * be valid; drop the packet. 1432 */ 1433 if (th->th_sport == th->th_dport) { 1434 int i; 1435 1436 switch (af) { 1437 #ifdef INET 1438 case AF_INET: 1439 i = in_hosteq(ip->ip_src, ip->ip_dst); 1440 break; 1441 #endif 1442 #ifdef INET6 1443 case AF_INET6: 1444 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1445 break; 1446 #endif 1447 default: 1448 i = 1; 1449 } 1450 if (i) { 1451 tcpstat.tcps_badsyn++; 1452 goto drop; 1453 } 1454 } 1455 1456 /* 1457 * SYN looks ok; create compressed TCP 1458 * state for it. 1459 */ 1460 if (so->so_qlen <= so->so_qlimit && 1461 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1462 so, m, optp, optlen, &opti)) 1463 m = NULL; 1464 } 1465 goto drop; 1466 } 1467 } 1468 1469 after_listen: 1470 #ifdef DIAGNOSTIC 1471 /* 1472 * Should not happen now that all embryonic connections 1473 * are handled with compressed state. 1474 */ 1475 if (tp->t_state == TCPS_LISTEN) 1476 panic("tcp_input: TCPS_LISTEN"); 1477 #endif 1478 1479 /* 1480 * Segment received on connection. 1481 * Reset idle time and keep-alive timer. 1482 */ 1483 tp->t_rcvtime = tcp_now; 1484 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1485 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1486 1487 /* 1488 * Process options. 1489 */ 1490 #ifdef TCP_SIGNATURE 1491 if (optp || (tp->t_flags & TF_SIGNATURE)) 1492 #else 1493 if (optp) 1494 #endif 1495 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1496 goto drop; 1497 1498 if (opti.ts_present && opti.ts_ecr) { 1499 u_int32_t now; 1500 1501 /* 1502 * Calculate the RTT from the returned time stamp and the 1503 * connection's time base. If the time stamp is later than 1504 * the current time, fall back to non-1323 RTT calculation. 1505 */ 1506 now = TCP_TIMESTAMP(tp); 1507 if (SEQ_GEQ(now, opti.ts_ecr)) 1508 opti.ts_ecr = now - opti.ts_ecr + 1; 1509 else 1510 opti.ts_ecr = 0; 1511 } 1512 1513 /* 1514 * Header prediction: check for the two common cases 1515 * of a uni-directional data xfer. If the packet has 1516 * no control flags, is in-sequence, the window didn't 1517 * change and we're not retransmitting, it's a 1518 * candidate. If the length is zero and the ack moved 1519 * forward, we're the sender side of the xfer. Just 1520 * free the data acked & wake any higher level process 1521 * that was blocked waiting for space. If the length 1522 * is non-zero and the ack didn't move, we're the 1523 * receiver side. If we're getting packets in-order 1524 * (the reassembly queue is empty), add the data to 1525 * the socket buffer and note that we need a delayed ack. 1526 */ 1527 if (tp->t_state == TCPS_ESTABLISHED && 1528 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1529 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1530 th->th_seq == tp->rcv_nxt && 1531 tiwin && tiwin == tp->snd_wnd && 1532 tp->snd_nxt == tp->snd_max) { 1533 1534 /* 1535 * If last ACK falls within this segment's sequence numbers, 1536 * record the timestamp. 1537 */ 1538 if (opti.ts_present && 1539 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1540 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1541 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1542 tp->ts_recent = opti.ts_val; 1543 } 1544 1545 if (tlen == 0) { 1546 /* Ack prediction. */ 1547 if (SEQ_GT(th->th_ack, tp->snd_una) && 1548 SEQ_LEQ(th->th_ack, tp->snd_max) && 1549 tp->snd_cwnd >= tp->snd_wnd && 1550 tp->t_dupacks < tcprexmtthresh) { 1551 /* 1552 * this is a pure ack for outstanding data. 1553 */ 1554 ++tcpstat.tcps_predack; 1555 if (opti.ts_present && opti.ts_ecr) 1556 tcp_xmit_timer(tp, opti.ts_ecr); 1557 else if (tp->t_rtttime && 1558 SEQ_GT(th->th_ack, tp->t_rtseq)) 1559 tcp_xmit_timer(tp, 1560 tcp_now - tp->t_rtttime); 1561 acked = th->th_ack - tp->snd_una; 1562 tcpstat.tcps_rcvackpack++; 1563 tcpstat.tcps_rcvackbyte += acked; 1564 ND6_HINT(tp); 1565 1566 if (acked > (tp->t_lastoff - tp->t_inoff)) 1567 tp->t_lastm = NULL; 1568 sbdrop(&so->so_snd, acked); 1569 tp->t_lastoff -= acked; 1570 1571 tp->snd_una = th->th_ack; 1572 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1573 tp->snd_high = tp->snd_una; 1574 m_freem(m); 1575 1576 /* 1577 * If all outstanding data are acked, stop 1578 * retransmit timer, otherwise restart timer 1579 * using current (possibly backed-off) value. 1580 * If process is waiting for space, 1581 * wakeup/selwakeup/signal. If data 1582 * are ready to send, let tcp_output 1583 * decide between more output or persist. 1584 */ 1585 if (tp->snd_una == tp->snd_max) 1586 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1587 else if (TCP_TIMER_ISARMED(tp, 1588 TCPT_PERSIST) == 0) 1589 TCP_TIMER_ARM(tp, TCPT_REXMT, 1590 tp->t_rxtcur); 1591 1592 sowwakeup(so); 1593 if (so->so_snd.sb_cc) 1594 (void) tcp_output(tp); 1595 if (tcp_saveti) 1596 m_freem(tcp_saveti); 1597 return; 1598 } 1599 } else if (th->th_ack == tp->snd_una && 1600 TAILQ_FIRST(&tp->segq) == NULL && 1601 tlen <= sbspace(&so->so_rcv)) { 1602 /* 1603 * this is a pure, in-sequence data packet 1604 * with nothing on the reassembly queue and 1605 * we have enough buffer space to take it. 1606 */ 1607 ++tcpstat.tcps_preddat; 1608 tp->rcv_nxt += tlen; 1609 tcpstat.tcps_rcvpack++; 1610 tcpstat.tcps_rcvbyte += tlen; 1611 ND6_HINT(tp); 1612 /* 1613 * Drop TCP, IP headers and TCP options then add data 1614 * to socket buffer. 1615 */ 1616 if (so->so_state & SS_CANTRCVMORE) 1617 m_freem(m); 1618 else { 1619 m_adj(m, toff + off); 1620 sbappendstream(&so->so_rcv, m); 1621 } 1622 sorwakeup(so); 1623 TCP_SETUP_ACK(tp, th); 1624 if (tp->t_flags & TF_ACKNOW) 1625 (void) tcp_output(tp); 1626 if (tcp_saveti) 1627 m_freem(tcp_saveti); 1628 return; 1629 } 1630 } 1631 1632 /* 1633 * Compute mbuf offset to TCP data segment. 1634 */ 1635 hdroptlen = toff + off; 1636 1637 /* 1638 * Calculate amount of space in receive window, 1639 * and then do TCP input processing. 1640 * Receive window is amount of space in rcv queue, 1641 * but not less than advertised window. 1642 */ 1643 { int win; 1644 1645 win = sbspace(&so->so_rcv); 1646 if (win < 0) 1647 win = 0; 1648 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1649 } 1650 1651 switch (tp->t_state) { 1652 case TCPS_LISTEN: 1653 /* 1654 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1655 */ 1656 if (m->m_flags & (M_BCAST|M_MCAST)) 1657 goto drop; 1658 switch (af) { 1659 #ifdef INET6 1660 case AF_INET6: 1661 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1662 goto drop; 1663 break; 1664 #endif /* INET6 */ 1665 case AF_INET: 1666 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1667 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1668 goto drop; 1669 break; 1670 } 1671 break; 1672 1673 /* 1674 * If the state is SYN_SENT: 1675 * if seg contains an ACK, but not for our SYN, drop the input. 1676 * if seg contains a RST, then drop the connection. 1677 * if seg does not contain SYN, then drop it. 1678 * Otherwise this is an acceptable SYN segment 1679 * initialize tp->rcv_nxt and tp->irs 1680 * if seg contains ack then advance tp->snd_una 1681 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1682 * arrange for segment to be acked (eventually) 1683 * continue processing rest of data/controls, beginning with URG 1684 */ 1685 case TCPS_SYN_SENT: 1686 if ((tiflags & TH_ACK) && 1687 (SEQ_LEQ(th->th_ack, tp->iss) || 1688 SEQ_GT(th->th_ack, tp->snd_max))) 1689 goto dropwithreset; 1690 if (tiflags & TH_RST) { 1691 if (tiflags & TH_ACK) 1692 tp = tcp_drop(tp, ECONNREFUSED); 1693 goto drop; 1694 } 1695 if ((tiflags & TH_SYN) == 0) 1696 goto drop; 1697 if (tiflags & TH_ACK) { 1698 tp->snd_una = th->th_ack; 1699 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1700 tp->snd_nxt = tp->snd_una; 1701 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1702 tp->snd_high = tp->snd_una; 1703 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1704 } 1705 tp->irs = th->th_seq; 1706 tcp_rcvseqinit(tp); 1707 tp->t_flags |= TF_ACKNOW; 1708 tcp_mss_from_peer(tp, opti.maxseg); 1709 1710 /* 1711 * Initialize the initial congestion window. If we 1712 * had to retransmit the SYN, we must initialize cwnd 1713 * to 1 segment (i.e. the Loss Window). 1714 */ 1715 if (tp->t_flags & TF_SYN_REXMT) 1716 tp->snd_cwnd = tp->t_peermss; 1717 else { 1718 int ss = tcp_init_win; 1719 #ifdef INET 1720 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1721 ss = tcp_init_win_local; 1722 #endif 1723 #ifdef INET6 1724 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1725 ss = tcp_init_win_local; 1726 #endif 1727 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1728 } 1729 1730 tcp_rmx_rtt(tp); 1731 if (tiflags & TH_ACK) { 1732 tcpstat.tcps_connects++; 1733 soisconnected(so); 1734 tcp_established(tp); 1735 /* Do window scaling on this connection? */ 1736 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1737 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1738 tp->snd_scale = tp->requested_s_scale; 1739 tp->rcv_scale = tp->request_r_scale; 1740 } 1741 TCP_REASS_LOCK(tp); 1742 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1743 TCP_REASS_UNLOCK(tp); 1744 /* 1745 * if we didn't have to retransmit the SYN, 1746 * use its rtt as our initial srtt & rtt var. 1747 */ 1748 if (tp->t_rtttime) 1749 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1750 } else 1751 tp->t_state = TCPS_SYN_RECEIVED; 1752 1753 /* 1754 * Advance th->th_seq to correspond to first data byte. 1755 * If data, trim to stay within window, 1756 * dropping FIN if necessary. 1757 */ 1758 th->th_seq++; 1759 if (tlen > tp->rcv_wnd) { 1760 todrop = tlen - tp->rcv_wnd; 1761 m_adj(m, -todrop); 1762 tlen = tp->rcv_wnd; 1763 tiflags &= ~TH_FIN; 1764 tcpstat.tcps_rcvpackafterwin++; 1765 tcpstat.tcps_rcvbyteafterwin += todrop; 1766 } 1767 tp->snd_wl1 = th->th_seq - 1; 1768 tp->rcv_up = th->th_seq; 1769 goto step6; 1770 1771 /* 1772 * If the state is SYN_RECEIVED: 1773 * If seg contains an ACK, but not for our SYN, drop the input 1774 * and generate an RST. See page 36, rfc793 1775 */ 1776 case TCPS_SYN_RECEIVED: 1777 if ((tiflags & TH_ACK) && 1778 (SEQ_LEQ(th->th_ack, tp->iss) || 1779 SEQ_GT(th->th_ack, tp->snd_max))) 1780 goto dropwithreset; 1781 break; 1782 } 1783 1784 /* 1785 * States other than LISTEN or SYN_SENT. 1786 * First check timestamp, if present. 1787 * Then check that at least some bytes of segment are within 1788 * receive window. If segment begins before rcv_nxt, 1789 * drop leading data (and SYN); if nothing left, just ack. 1790 * 1791 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1792 * and it's less than ts_recent, drop it. 1793 */ 1794 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1795 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1796 1797 /* Check to see if ts_recent is over 24 days old. */ 1798 if ((int)(TCP_TIMESTAMP(tp) - tp->ts_recent_age) > 1799 TCP_PAWS_IDLE) { 1800 /* 1801 * Invalidate ts_recent. If this segment updates 1802 * ts_recent, the age will be reset later and ts_recent 1803 * will get a valid value. If it does not, setting 1804 * ts_recent to zero will at least satisfy the 1805 * requirement that zero be placed in the timestamp 1806 * echo reply when ts_recent isn't valid. The 1807 * age isn't reset until we get a valid ts_recent 1808 * because we don't want out-of-order segments to be 1809 * dropped when ts_recent is old. 1810 */ 1811 tp->ts_recent = 0; 1812 } else { 1813 tcpstat.tcps_rcvduppack++; 1814 tcpstat.tcps_rcvdupbyte += tlen; 1815 tcpstat.tcps_pawsdrop++; 1816 goto dropafterack; 1817 } 1818 } 1819 1820 todrop = tp->rcv_nxt - th->th_seq; 1821 if (todrop > 0) { 1822 if (tiflags & TH_SYN) { 1823 tiflags &= ~TH_SYN; 1824 th->th_seq++; 1825 if (th->th_urp > 1) 1826 th->th_urp--; 1827 else { 1828 tiflags &= ~TH_URG; 1829 th->th_urp = 0; 1830 } 1831 todrop--; 1832 } 1833 if (todrop > tlen || 1834 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1835 /* 1836 * Any valid FIN or RST must be to the left of the 1837 * window. At this point the FIN or RST must be a 1838 * duplicate or out of sequence; drop it. 1839 */ 1840 if (tiflags & TH_RST) 1841 goto drop; 1842 tiflags &= ~(TH_FIN|TH_RST); 1843 /* 1844 * Send an ACK to resynchronize and drop any data. 1845 * But keep on processing for RST or ACK. 1846 */ 1847 tp->t_flags |= TF_ACKNOW; 1848 todrop = tlen; 1849 tcpstat.tcps_rcvdupbyte += todrop; 1850 tcpstat.tcps_rcvduppack++; 1851 } else if ((tiflags & TH_RST) && 1852 th->th_seq != tp->last_ack_sent) { 1853 /* 1854 * Test for reset before adjusting the sequence 1855 * number for overlapping data. 1856 */ 1857 goto dropafterack_ratelim; 1858 } else { 1859 tcpstat.tcps_rcvpartduppack++; 1860 tcpstat.tcps_rcvpartdupbyte += todrop; 1861 } 1862 hdroptlen += todrop; /*drop from head afterwards*/ 1863 th->th_seq += todrop; 1864 tlen -= todrop; 1865 if (th->th_urp > todrop) 1866 th->th_urp -= todrop; 1867 else { 1868 tiflags &= ~TH_URG; 1869 th->th_urp = 0; 1870 } 1871 } 1872 1873 /* 1874 * If new data are received on a connection after the 1875 * user processes are gone, then RST the other end. 1876 */ 1877 if ((so->so_state & SS_NOFDREF) && 1878 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1879 tp = tcp_close(tp); 1880 tcpstat.tcps_rcvafterclose++; 1881 goto dropwithreset; 1882 } 1883 1884 /* 1885 * If segment ends after window, drop trailing data 1886 * (and PUSH and FIN); if nothing left, just ACK. 1887 */ 1888 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1889 if (todrop > 0) { 1890 tcpstat.tcps_rcvpackafterwin++; 1891 if (todrop >= tlen) { 1892 /* 1893 * The segment actually starts after the window. 1894 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 1895 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 1896 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 1897 */ 1898 tcpstat.tcps_rcvbyteafterwin += tlen; 1899 /* 1900 * If a new connection request is received 1901 * while in TIME_WAIT, drop the old connection 1902 * and start over if the sequence numbers 1903 * are above the previous ones. 1904 * 1905 * NOTE: We will checksum the packet again, and 1906 * so we need to put the header fields back into 1907 * network order! 1908 * XXX This kind of sucks, but we don't expect 1909 * XXX this to happen very often, so maybe it 1910 * XXX doesn't matter so much. 1911 */ 1912 if (tiflags & TH_SYN && 1913 tp->t_state == TCPS_TIME_WAIT && 1914 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1915 iss = tcp_new_iss(tp, tp->snd_nxt); 1916 tp = tcp_close(tp); 1917 TCP_FIELDS_TO_NET(th); 1918 goto findpcb; 1919 } 1920 /* 1921 * If window is closed can only take segments at 1922 * window edge, and have to drop data and PUSH from 1923 * incoming segments. Continue processing, but 1924 * remember to ack. Otherwise, drop segment 1925 * and (if not RST) ack. 1926 */ 1927 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1928 tp->t_flags |= TF_ACKNOW; 1929 tcpstat.tcps_rcvwinprobe++; 1930 } else 1931 goto dropafterack; 1932 } else 1933 tcpstat.tcps_rcvbyteafterwin += todrop; 1934 m_adj(m, -todrop); 1935 tlen -= todrop; 1936 tiflags &= ~(TH_PUSH|TH_FIN); 1937 } 1938 1939 /* 1940 * If last ACK falls within this segment's sequence numbers, 1941 * and the timestamp is newer, record it. 1942 */ 1943 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1944 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1945 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1946 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1947 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1948 tp->ts_recent = opti.ts_val; 1949 } 1950 1951 /* 1952 * If the RST bit is set examine the state: 1953 * SYN_RECEIVED STATE: 1954 * If passive open, return to LISTEN state. 1955 * If active open, inform user that connection was refused. 1956 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1957 * Inform user that connection was reset, and close tcb. 1958 * CLOSING, LAST_ACK, TIME_WAIT STATES 1959 * Close the tcb. 1960 */ 1961 if (tiflags & TH_RST) { 1962 if (th->th_seq != tp->last_ack_sent) 1963 goto dropafterack_ratelim; 1964 1965 switch (tp->t_state) { 1966 case TCPS_SYN_RECEIVED: 1967 so->so_error = ECONNREFUSED; 1968 goto close; 1969 1970 case TCPS_ESTABLISHED: 1971 case TCPS_FIN_WAIT_1: 1972 case TCPS_FIN_WAIT_2: 1973 case TCPS_CLOSE_WAIT: 1974 so->so_error = ECONNRESET; 1975 close: 1976 tp->t_state = TCPS_CLOSED; 1977 tcpstat.tcps_drops++; 1978 tp = tcp_close(tp); 1979 goto drop; 1980 1981 case TCPS_CLOSING: 1982 case TCPS_LAST_ACK: 1983 case TCPS_TIME_WAIT: 1984 tp = tcp_close(tp); 1985 goto drop; 1986 } 1987 } 1988 1989 /* 1990 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 1991 * we must be in a synchronized state. RFC791 states (under RST 1992 * generation) that any unacceptable segment (an out-of-order SYN 1993 * qualifies) received in a synchronized state must elicit only an 1994 * empty acknowledgment segment ... and the connection remains in 1995 * the same state. 1996 */ 1997 if (tiflags & TH_SYN) { 1998 if (tp->rcv_nxt == th->th_seq) { 1999 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2000 TH_ACK); 2001 if (tcp_saveti) 2002 m_freem(tcp_saveti); 2003 return; 2004 } 2005 2006 goto dropafterack_ratelim; 2007 } 2008 2009 /* 2010 * If the ACK bit is off we drop the segment and return. 2011 */ 2012 if ((tiflags & TH_ACK) == 0) { 2013 if (tp->t_flags & TF_ACKNOW) 2014 goto dropafterack; 2015 else 2016 goto drop; 2017 } 2018 2019 /* 2020 * Ack processing. 2021 */ 2022 switch (tp->t_state) { 2023 2024 /* 2025 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2026 * ESTABLISHED state and continue processing, otherwise 2027 * send an RST. 2028 */ 2029 case TCPS_SYN_RECEIVED: 2030 if (SEQ_GT(tp->snd_una, th->th_ack) || 2031 SEQ_GT(th->th_ack, tp->snd_max)) 2032 goto dropwithreset; 2033 tcpstat.tcps_connects++; 2034 soisconnected(so); 2035 tcp_established(tp); 2036 /* Do window scaling? */ 2037 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2038 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2039 tp->snd_scale = tp->requested_s_scale; 2040 tp->rcv_scale = tp->request_r_scale; 2041 } 2042 TCP_REASS_LOCK(tp); 2043 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2044 TCP_REASS_UNLOCK(tp); 2045 tp->snd_wl1 = th->th_seq - 1; 2046 /* fall into ... */ 2047 2048 /* 2049 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2050 * ACKs. If the ack is in the range 2051 * tp->snd_una < th->th_ack <= tp->snd_max 2052 * then advance tp->snd_una to th->th_ack and drop 2053 * data from the retransmission queue. If this ACK reflects 2054 * more up to date window information we update our window information. 2055 */ 2056 case TCPS_ESTABLISHED: 2057 case TCPS_FIN_WAIT_1: 2058 case TCPS_FIN_WAIT_2: 2059 case TCPS_CLOSE_WAIT: 2060 case TCPS_CLOSING: 2061 case TCPS_LAST_ACK: 2062 case TCPS_TIME_WAIT: 2063 2064 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2065 if (tlen == 0 && tiwin == tp->snd_wnd) { 2066 tcpstat.tcps_rcvdupack++; 2067 /* 2068 * If we have outstanding data (other than 2069 * a window probe), this is a completely 2070 * duplicate ack (ie, window info didn't 2071 * change), the ack is the biggest we've 2072 * seen and we've seen exactly our rexmt 2073 * threshhold of them, assume a packet 2074 * has been dropped and retransmit it. 2075 * Kludge snd_nxt & the congestion 2076 * window so we send only this one 2077 * packet. 2078 * 2079 * We know we're losing at the current 2080 * window size so do congestion avoidance 2081 * (set ssthresh to half the current window 2082 * and pull our congestion window back to 2083 * the new ssthresh). 2084 * 2085 * Dup acks mean that packets have left the 2086 * network (they're now cached at the receiver) 2087 * so bump cwnd by the amount in the receiver 2088 * to keep a constant cwnd packets in the 2089 * network. 2090 */ 2091 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2092 th->th_ack != tp->snd_una) 2093 tp->t_dupacks = 0; 2094 else if (++tp->t_dupacks == tcprexmtthresh) { 2095 tcp_seq onxt; 2096 u_int win; 2097 2098 if (tcp_do_newreno && 2099 SEQ_LT(th->th_ack, tp->snd_high)) { 2100 /* 2101 * False fast retransmit after 2102 * timeout. Do not enter fast 2103 * recovery. 2104 */ 2105 tp->t_dupacks = 0; 2106 break; 2107 } 2108 2109 onxt = tp->snd_nxt; 2110 win = min(tp->snd_wnd, tp->snd_cwnd) / 2111 2 / tp->t_segsz; 2112 if (win < 2) 2113 win = 2; 2114 tp->snd_ssthresh = win * tp->t_segsz; 2115 tp->snd_recover = tp->snd_max; 2116 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2117 tp->t_rtttime = 0; 2118 tp->snd_nxt = th->th_ack; 2119 tp->snd_cwnd = tp->t_segsz; 2120 (void) tcp_output(tp); 2121 tp->snd_cwnd = tp->snd_ssthresh + 2122 tp->t_segsz * tp->t_dupacks; 2123 if (SEQ_GT(onxt, tp->snd_nxt)) 2124 tp->snd_nxt = onxt; 2125 goto drop; 2126 } else if (tp->t_dupacks > tcprexmtthresh) { 2127 tp->snd_cwnd += tp->t_segsz; 2128 (void) tcp_output(tp); 2129 goto drop; 2130 } 2131 } else if (tlen) { 2132 tp->t_dupacks = 0; /*XXX*/ 2133 /* drop very old ACKs unless th_seq matches */ 2134 if (th->th_seq != tp->rcv_nxt && 2135 SEQ_LT(th->th_ack, 2136 tp->snd_una - tp->max_sndwnd)) { 2137 goto drop; 2138 } 2139 break; 2140 } else 2141 tp->t_dupacks = 0; 2142 break; 2143 } 2144 /* 2145 * If the congestion window was inflated to account 2146 * for the other side's cached packets, retract it. 2147 */ 2148 if (!tcp_do_newreno) { 2149 if (tp->t_dupacks >= tcprexmtthresh && 2150 tp->snd_cwnd > tp->snd_ssthresh) 2151 tp->snd_cwnd = tp->snd_ssthresh; 2152 tp->t_dupacks = 0; 2153 } else { 2154 tcp_newreno(tp, th); 2155 } 2156 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2157 tcpstat.tcps_rcvacktoomuch++; 2158 goto dropafterack; 2159 } 2160 acked = th->th_ack - tp->snd_una; 2161 tcpstat.tcps_rcvackpack++; 2162 tcpstat.tcps_rcvackbyte += acked; 2163 2164 /* 2165 * If we have a timestamp reply, update smoothed 2166 * round trip time. If no timestamp is present but 2167 * transmit timer is running and timed sequence 2168 * number was acked, update smoothed round trip time. 2169 * Since we now have an rtt measurement, cancel the 2170 * timer backoff (cf., Phil Karn's retransmit alg.). 2171 * Recompute the initial retransmit timer. 2172 */ 2173 if (opti.ts_present && opti.ts_ecr) 2174 tcp_xmit_timer(tp, opti.ts_ecr); 2175 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2176 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2177 2178 /* 2179 * If all outstanding data is acked, stop retransmit 2180 * timer and remember to restart (more output or persist). 2181 * If there is more data to be acked, restart retransmit 2182 * timer, using current (possibly backed-off) value. 2183 */ 2184 if (th->th_ack == tp->snd_max) { 2185 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2186 needoutput = 1; 2187 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2188 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2189 /* 2190 * When new data is acked, open the congestion window. 2191 * If the window gives us less than ssthresh packets 2192 * in flight, open exponentially (segsz per packet). 2193 * Otherwise open linearly: segsz per window 2194 * (segsz^2 / cwnd per packet), plus a constant 2195 * fraction of a packet (segsz/8) to help larger windows 2196 * open quickly enough. 2197 * 2198 * If we are still in fast recovery (meaning we are using 2199 * NewReno and we have only received partial acks), do not 2200 * inflate the window yet. 2201 */ 2202 if (tp->t_dupacks == 0) { 2203 u_int cw = tp->snd_cwnd; 2204 u_int incr = tp->t_segsz; 2205 2206 if (cw > tp->snd_ssthresh) 2207 incr = incr * incr / cw; 2208 tp->snd_cwnd = min(cw + incr, 2209 TCP_MAXWIN << tp->snd_scale); 2210 } 2211 ND6_HINT(tp); 2212 if (acked > so->so_snd.sb_cc) { 2213 tp->snd_wnd -= so->so_snd.sb_cc; 2214 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2215 ourfinisacked = 1; 2216 } else { 2217 if (acked > (tp->t_lastoff - tp->t_inoff)) 2218 tp->t_lastm = NULL; 2219 sbdrop(&so->so_snd, acked); 2220 tp->t_lastoff -= acked; 2221 tp->snd_wnd -= acked; 2222 ourfinisacked = 0; 2223 } 2224 sowwakeup(so); 2225 tp->snd_una = th->th_ack; 2226 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2227 tp->snd_nxt = tp->snd_una; 2228 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2229 tp->snd_high = tp->snd_una; 2230 2231 switch (tp->t_state) { 2232 2233 /* 2234 * In FIN_WAIT_1 STATE in addition to the processing 2235 * for the ESTABLISHED state if our FIN is now acknowledged 2236 * then enter FIN_WAIT_2. 2237 */ 2238 case TCPS_FIN_WAIT_1: 2239 if (ourfinisacked) { 2240 /* 2241 * If we can't receive any more 2242 * data, then closing user can proceed. 2243 * Starting the timer is contrary to the 2244 * specification, but if we don't get a FIN 2245 * we'll hang forever. 2246 */ 2247 if (so->so_state & SS_CANTRCVMORE) { 2248 soisdisconnected(so); 2249 if (tcp_maxidle > 0) 2250 TCP_TIMER_ARM(tp, TCPT_2MSL, 2251 tcp_maxidle); 2252 } 2253 tp->t_state = TCPS_FIN_WAIT_2; 2254 } 2255 break; 2256 2257 /* 2258 * In CLOSING STATE in addition to the processing for 2259 * the ESTABLISHED state if the ACK acknowledges our FIN 2260 * then enter the TIME-WAIT state, otherwise ignore 2261 * the segment. 2262 */ 2263 case TCPS_CLOSING: 2264 if (ourfinisacked) { 2265 tp->t_state = TCPS_TIME_WAIT; 2266 tcp_canceltimers(tp); 2267 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2268 soisdisconnected(so); 2269 } 2270 break; 2271 2272 /* 2273 * In LAST_ACK, we may still be waiting for data to drain 2274 * and/or to be acked, as well as for the ack of our FIN. 2275 * If our FIN is now acknowledged, delete the TCB, 2276 * enter the closed state and return. 2277 */ 2278 case TCPS_LAST_ACK: 2279 if (ourfinisacked) { 2280 tp = tcp_close(tp); 2281 goto drop; 2282 } 2283 break; 2284 2285 /* 2286 * In TIME_WAIT state the only thing that should arrive 2287 * is a retransmission of the remote FIN. Acknowledge 2288 * it and restart the finack timer. 2289 */ 2290 case TCPS_TIME_WAIT: 2291 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2292 goto dropafterack; 2293 } 2294 } 2295 2296 step6: 2297 /* 2298 * Update window information. 2299 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2300 */ 2301 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2302 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 2303 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 2304 /* keep track of pure window updates */ 2305 if (tlen == 0 && 2306 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2307 tcpstat.tcps_rcvwinupd++; 2308 tp->snd_wnd = tiwin; 2309 tp->snd_wl1 = th->th_seq; 2310 tp->snd_wl2 = th->th_ack; 2311 if (tp->snd_wnd > tp->max_sndwnd) 2312 tp->max_sndwnd = tp->snd_wnd; 2313 needoutput = 1; 2314 } 2315 2316 /* 2317 * Process segments with URG. 2318 */ 2319 if ((tiflags & TH_URG) && th->th_urp && 2320 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2321 /* 2322 * This is a kludge, but if we receive and accept 2323 * random urgent pointers, we'll crash in 2324 * soreceive. It's hard to imagine someone 2325 * actually wanting to send this much urgent data. 2326 */ 2327 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2328 th->th_urp = 0; /* XXX */ 2329 tiflags &= ~TH_URG; /* XXX */ 2330 goto dodata; /* XXX */ 2331 } 2332 /* 2333 * If this segment advances the known urgent pointer, 2334 * then mark the data stream. This should not happen 2335 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2336 * a FIN has been received from the remote side. 2337 * In these states we ignore the URG. 2338 * 2339 * According to RFC961 (Assigned Protocols), 2340 * the urgent pointer points to the last octet 2341 * of urgent data. We continue, however, 2342 * to consider it to indicate the first octet 2343 * of data past the urgent section as the original 2344 * spec states (in one of two places). 2345 */ 2346 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2347 tp->rcv_up = th->th_seq + th->th_urp; 2348 so->so_oobmark = so->so_rcv.sb_cc + 2349 (tp->rcv_up - tp->rcv_nxt) - 1; 2350 if (so->so_oobmark == 0) 2351 so->so_state |= SS_RCVATMARK; 2352 sohasoutofband(so); 2353 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2354 } 2355 /* 2356 * Remove out of band data so doesn't get presented to user. 2357 * This can happen independent of advancing the URG pointer, 2358 * but if two URG's are pending at once, some out-of-band 2359 * data may creep in... ick. 2360 */ 2361 if (th->th_urp <= (u_int16_t) tlen 2362 #ifdef SO_OOBINLINE 2363 && (so->so_options & SO_OOBINLINE) == 0 2364 #endif 2365 ) 2366 tcp_pulloutofband(so, th, m, hdroptlen); 2367 } else 2368 /* 2369 * If no out of band data is expected, 2370 * pull receive urgent pointer along 2371 * with the receive window. 2372 */ 2373 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2374 tp->rcv_up = tp->rcv_nxt; 2375 dodata: /* XXX */ 2376 2377 /* 2378 * Process the segment text, merging it into the TCP sequencing queue, 2379 * and arranging for acknowledgement of receipt if necessary. 2380 * This process logically involves adjusting tp->rcv_wnd as data 2381 * is presented to the user (this happens in tcp_usrreq.c, 2382 * case PRU_RCVD). If a FIN has already been received on this 2383 * connection then we just ignore the text. 2384 */ 2385 if ((tlen || (tiflags & TH_FIN)) && 2386 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2387 /* 2388 * Insert segment ti into reassembly queue of tcp with 2389 * control block tp. Return TH_FIN if reassembly now includes 2390 * a segment with FIN. The macro form does the common case 2391 * inline (segment is the next to be received on an 2392 * established connection, and the queue is empty), 2393 * avoiding linkage into and removal from the queue and 2394 * repetition of various conversions. 2395 * Set DELACK for segments received in order, but ack 2396 * immediately when segments are out of order 2397 * (so fast retransmit can work). 2398 */ 2399 /* NOTE: this was TCP_REASS() macro, but used only once */ 2400 TCP_REASS_LOCK(tp); 2401 if (th->th_seq == tp->rcv_nxt && 2402 TAILQ_FIRST(&tp->segq) == NULL && 2403 tp->t_state == TCPS_ESTABLISHED) { 2404 TCP_SETUP_ACK(tp, th); 2405 tp->rcv_nxt += tlen; 2406 tiflags = th->th_flags & TH_FIN; 2407 tcpstat.tcps_rcvpack++; 2408 tcpstat.tcps_rcvbyte += tlen; 2409 ND6_HINT(tp); 2410 if (so->so_state & SS_CANTRCVMORE) 2411 m_freem(m); 2412 else { 2413 m_adj(m, hdroptlen); 2414 sbappendstream(&(so)->so_rcv, m); 2415 } 2416 sorwakeup(so); 2417 } else { 2418 m_adj(m, hdroptlen); 2419 tiflags = tcp_reass(tp, th, m, &tlen); 2420 tp->t_flags |= TF_ACKNOW; 2421 } 2422 TCP_REASS_UNLOCK(tp); 2423 2424 /* 2425 * Note the amount of data that peer has sent into 2426 * our window, in order to estimate the sender's 2427 * buffer size. 2428 */ 2429 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2430 } else { 2431 m_freem(m); 2432 m = NULL; 2433 tiflags &= ~TH_FIN; 2434 } 2435 2436 /* 2437 * If FIN is received ACK the FIN and let the user know 2438 * that the connection is closing. Ignore a FIN received before 2439 * the connection is fully established. 2440 */ 2441 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2442 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2443 socantrcvmore(so); 2444 tp->t_flags |= TF_ACKNOW; 2445 tp->rcv_nxt++; 2446 } 2447 switch (tp->t_state) { 2448 2449 /* 2450 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2451 */ 2452 case TCPS_ESTABLISHED: 2453 tp->t_state = TCPS_CLOSE_WAIT; 2454 break; 2455 2456 /* 2457 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2458 * enter the CLOSING state. 2459 */ 2460 case TCPS_FIN_WAIT_1: 2461 tp->t_state = TCPS_CLOSING; 2462 break; 2463 2464 /* 2465 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2466 * starting the time-wait timer, turning off the other 2467 * standard timers. 2468 */ 2469 case TCPS_FIN_WAIT_2: 2470 tp->t_state = TCPS_TIME_WAIT; 2471 tcp_canceltimers(tp); 2472 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2473 soisdisconnected(so); 2474 break; 2475 2476 /* 2477 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2478 */ 2479 case TCPS_TIME_WAIT: 2480 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2481 break; 2482 } 2483 } 2484 #ifdef TCP_DEBUG 2485 if (so->so_options & SO_DEBUG) 2486 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2487 #endif 2488 2489 /* 2490 * Return any desired output. 2491 */ 2492 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2493 (void) tcp_output(tp); 2494 if (tcp_saveti) 2495 m_freem(tcp_saveti); 2496 return; 2497 2498 badsyn: 2499 /* 2500 * Received a bad SYN. Increment counters and dropwithreset. 2501 */ 2502 tcpstat.tcps_badsyn++; 2503 tp = NULL; 2504 goto dropwithreset; 2505 2506 dropafterack: 2507 /* 2508 * Generate an ACK dropping incoming segment if it occupies 2509 * sequence space, where the ACK reflects our state. 2510 */ 2511 if (tiflags & TH_RST) 2512 goto drop; 2513 goto dropafterack2; 2514 2515 dropafterack_ratelim: 2516 /* 2517 * We may want to rate-limit ACKs against SYN/RST attack. 2518 */ 2519 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2520 tcp_ackdrop_ppslim) == 0) { 2521 /* XXX stat */ 2522 goto drop; 2523 } 2524 /* ...fall into dropafterack2... */ 2525 2526 dropafterack2: 2527 m_freem(m); 2528 tp->t_flags |= TF_ACKNOW; 2529 (void) tcp_output(tp); 2530 if (tcp_saveti) 2531 m_freem(tcp_saveti); 2532 return; 2533 2534 dropwithreset_ratelim: 2535 /* 2536 * We may want to rate-limit RSTs in certain situations, 2537 * particularly if we are sending an RST in response to 2538 * an attempt to connect to or otherwise communicate with 2539 * a port for which we have no socket. 2540 */ 2541 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2542 tcp_rst_ppslim) == 0) { 2543 /* XXX stat */ 2544 goto drop; 2545 } 2546 /* ...fall into dropwithreset... */ 2547 2548 dropwithreset: 2549 /* 2550 * Generate a RST, dropping incoming segment. 2551 * Make ACK acceptable to originator of segment. 2552 */ 2553 if (tiflags & TH_RST) 2554 goto drop; 2555 2556 switch (af) { 2557 #ifdef INET6 2558 case AF_INET6: 2559 /* For following calls to tcp_respond */ 2560 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2561 goto drop; 2562 break; 2563 #endif /* INET6 */ 2564 case AF_INET: 2565 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2566 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2567 goto drop; 2568 } 2569 2570 if (tiflags & TH_ACK) 2571 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2572 else { 2573 if (tiflags & TH_SYN) 2574 tlen++; 2575 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2576 TH_RST|TH_ACK); 2577 } 2578 if (tcp_saveti) 2579 m_freem(tcp_saveti); 2580 return; 2581 2582 badcsum: 2583 drop: 2584 /* 2585 * Drop space held by incoming segment and return. 2586 */ 2587 if (tp) { 2588 if (tp->t_inpcb) 2589 so = tp->t_inpcb->inp_socket; 2590 #ifdef INET6 2591 else if (tp->t_in6pcb) 2592 so = tp->t_in6pcb->in6p_socket; 2593 #endif 2594 else 2595 so = NULL; 2596 #ifdef TCP_DEBUG 2597 if (so && (so->so_options & SO_DEBUG) != 0) 2598 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2599 #endif 2600 } 2601 if (tcp_saveti) 2602 m_freem(tcp_saveti); 2603 m_freem(m); 2604 return; 2605 } 2606 2607 #ifdef TCP_SIGNATURE 2608 int 2609 tcp_signature_apply(void *fstate, caddr_t data, u_int len) 2610 { 2611 2612 MD5Update(fstate, (u_char *)data, len); 2613 return (0); 2614 } 2615 2616 struct secasvar * 2617 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 2618 { 2619 struct secasvar *sav; 2620 #ifdef FAST_IPSEC 2621 union sockaddr_union dst; 2622 #endif 2623 struct ip *ip; 2624 struct ip6_hdr *ip6; 2625 2626 ip = mtod(m, struct ip *); 2627 switch (ip->ip_v) { 2628 case 4: 2629 ip = mtod(m, struct ip *); 2630 ip6 = NULL; 2631 break; 2632 case 6: 2633 ip = NULL; 2634 ip6 = mtod(m, struct ip6_hdr *); 2635 break; 2636 default: 2637 return (NULL); 2638 } 2639 2640 #ifdef FAST_IPSEC 2641 /* Extract the destination from the IP header in the mbuf. */ 2642 bzero(&dst, sizeof(union sockaddr_union)); 2643 dst.sa.sa_len = sizeof(struct sockaddr_in); 2644 dst.sa.sa_family = AF_INET; 2645 dst.sin.sin_addr = ip->ip_dst; 2646 2647 /* 2648 * Look up an SADB entry which matches the address of the peer. 2649 */ 2650 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2651 #else 2652 if (ip) 2653 sav = key_allocsa(AF_INET, (caddr_t)&ip->ip_src, 2654 (caddr_t)&ip->ip_dst, IPPROTO_TCP, 2655 htonl(TCP_SIG_SPI)); 2656 else 2657 sav = key_allocsa(AF_INET6, (caddr_t)&ip6->ip6_src, 2658 (caddr_t)&ip6->ip6_dst, IPPROTO_TCP, 2659 htonl(TCP_SIG_SPI)); 2660 #endif 2661 2662 return (sav); /* freesav must be performed by caller */ 2663 } 2664 2665 int 2666 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 2667 struct secasvar *sav, char *sig) 2668 { 2669 MD5_CTX ctx; 2670 struct ip *ip; 2671 struct ipovly *ipovly; 2672 struct ip6_hdr *ip6; 2673 struct ippseudo ippseudo; 2674 struct ip6_hdr_pseudo ip6pseudo; 2675 struct tcphdr th0; 2676 int l, tcphdrlen; 2677 2678 if (sav == NULL) 2679 return (-1); 2680 2681 tcphdrlen = th->th_off * 4; 2682 2683 switch (mtod(m, struct ip *)->ip_v) { 2684 case 4: 2685 ip = mtod(m, struct ip *); 2686 ip6 = NULL; 2687 break; 2688 case 6: 2689 ip = NULL; 2690 ip6 = mtod(m, struct ip6_hdr *); 2691 break; 2692 default: 2693 return (-1); 2694 } 2695 2696 MD5Init(&ctx); 2697 2698 if (ip) { 2699 memset(&ippseudo, 0, sizeof(ippseudo)); 2700 ipovly = (struct ipovly *)ip; 2701 ippseudo.ippseudo_src = ipovly->ih_src; 2702 ippseudo.ippseudo_dst = ipovly->ih_dst; 2703 ippseudo.ippseudo_pad = 0; 2704 ippseudo.ippseudo_p = IPPROTO_TCP; 2705 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 2706 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 2707 } else { 2708 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 2709 ip6pseudo.ip6ph_src = ip6->ip6_src; 2710 in6_clearscope(&ip6pseudo.ip6ph_src); 2711 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 2712 in6_clearscope(&ip6pseudo.ip6ph_dst); 2713 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 2714 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 2715 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 2716 } 2717 2718 th0 = *th; 2719 th0.th_sum = 0; 2720 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 2721 2722 l = m->m_pkthdr.len - thoff - tcphdrlen; 2723 if (l > 0) 2724 m_apply(m, thoff + tcphdrlen, 2725 m->m_pkthdr.len - thoff - tcphdrlen, 2726 tcp_signature_apply, &ctx); 2727 2728 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2729 MD5Final(sig, &ctx); 2730 2731 return (0); 2732 } 2733 #endif 2734 2735 int 2736 tcp_dooptions(tp, cp, cnt, th, m, toff, oi) 2737 struct tcpcb *tp; 2738 u_char *cp; 2739 int cnt; 2740 struct tcphdr *th; 2741 struct mbuf *m; 2742 int toff; 2743 struct tcp_opt_info *oi; 2744 { 2745 u_int16_t mss; 2746 int opt, optlen = 0; 2747 #ifdef TCP_SIGNATURE 2748 caddr_t sigp = NULL; 2749 char sigbuf[TCP_SIGLEN]; 2750 struct secasvar *sav = NULL; 2751 #endif 2752 2753 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 2754 opt = cp[0]; 2755 if (opt == TCPOPT_EOL) 2756 break; 2757 if (opt == TCPOPT_NOP) 2758 optlen = 1; 2759 else { 2760 if (cnt < 2) 2761 break; 2762 optlen = cp[1]; 2763 if (optlen < 2 || optlen > cnt) 2764 break; 2765 } 2766 switch (opt) { 2767 2768 default: 2769 continue; 2770 2771 case TCPOPT_MAXSEG: 2772 if (optlen != TCPOLEN_MAXSEG) 2773 continue; 2774 if (!(th->th_flags & TH_SYN)) 2775 continue; 2776 bcopy(cp + 2, &mss, sizeof(mss)); 2777 oi->maxseg = ntohs(mss); 2778 break; 2779 2780 case TCPOPT_WINDOW: 2781 if (optlen != TCPOLEN_WINDOW) 2782 continue; 2783 if (!(th->th_flags & TH_SYN)) 2784 continue; 2785 tp->t_flags |= TF_RCVD_SCALE; 2786 tp->requested_s_scale = cp[2]; 2787 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2788 #if 0 /*XXX*/ 2789 char *p; 2790 2791 if (ip) 2792 p = ntohl(ip->ip_src); 2793 #ifdef INET6 2794 else if (ip6) 2795 p = ip6_sprintf(&ip6->ip6_src); 2796 #endif 2797 else 2798 p = "(unknown)"; 2799 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2800 "assuming %d\n", 2801 tp->requested_s_scale, p, 2802 TCP_MAX_WINSHIFT); 2803 #else 2804 log(LOG_ERR, "TCP: invalid wscale %d, " 2805 "assuming %d\n", 2806 tp->requested_s_scale, 2807 TCP_MAX_WINSHIFT); 2808 #endif 2809 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2810 } 2811 break; 2812 2813 case TCPOPT_TIMESTAMP: 2814 if (optlen != TCPOLEN_TIMESTAMP) 2815 continue; 2816 oi->ts_present = 1; 2817 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2818 NTOHL(oi->ts_val); 2819 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2820 NTOHL(oi->ts_ecr); 2821 2822 /* 2823 * A timestamp received in a SYN makes 2824 * it ok to send timestamp requests and replies. 2825 */ 2826 if (th->th_flags & TH_SYN) { 2827 tp->t_flags |= TF_RCVD_TSTMP; 2828 tp->ts_recent = oi->ts_val; 2829 tp->ts_recent_age = TCP_TIMESTAMP(tp); 2830 } 2831 break; 2832 case TCPOPT_SACK_PERMITTED: 2833 if (optlen != TCPOLEN_SACK_PERMITTED) 2834 continue; 2835 if (!(th->th_flags & TH_SYN)) 2836 continue; 2837 tp->t_flags &= ~TF_CANT_TXSACK; 2838 break; 2839 2840 case TCPOPT_SACK: 2841 if (tp->t_flags & TF_IGNR_RXSACK) 2842 continue; 2843 if (optlen % 8 != 2 || optlen < 10) 2844 continue; 2845 cp += 2; 2846 optlen -= 2; 2847 for (; optlen > 0; cp -= 8, optlen -= 8) { 2848 tcp_seq lwe, rwe; 2849 bcopy((char *)cp, (char *) &lwe, sizeof(lwe)); 2850 NTOHL(lwe); 2851 bcopy((char *)cp, (char *) &rwe, sizeof(rwe)); 2852 NTOHL(rwe); 2853 /* tcp_mark_sacked(tp, lwe, rwe); */ 2854 } 2855 break; 2856 #ifdef TCP_SIGNATURE 2857 case TCPOPT_SIGNATURE: 2858 if (optlen != TCPOLEN_SIGNATURE) 2859 continue; 2860 if (sigp && bcmp(sigp, cp + 2, TCP_SIGLEN)) 2861 return (-1); 2862 2863 sigp = sigbuf; 2864 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 2865 memset(cp + 2, 0, TCP_SIGLEN); 2866 tp->t_flags |= TF_SIGNATURE; 2867 break; 2868 #endif 2869 } 2870 } 2871 2872 #ifdef TCP_SIGNATURE 2873 if (tp->t_flags & TF_SIGNATURE) { 2874 2875 sav = tcp_signature_getsav(m, th); 2876 2877 if (sav == NULL && tp->t_state == TCPS_LISTEN) 2878 return (-1); 2879 } 2880 2881 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 2882 if (sav == NULL) 2883 return (-1); 2884 #ifdef FAST_IPSEC 2885 KEY_FREESAV(&sav); 2886 #else 2887 key_freesav(sav); 2888 #endif 2889 return (-1); 2890 } 2891 2892 if (sigp) { 2893 char sig[TCP_SIGLEN]; 2894 2895 TCP_FIELDS_TO_NET(th); 2896 if (tcp_signature(m, th, toff, sav, sig) < 0) { 2897 TCP_FIELDS_TO_HOST(th); 2898 if (sav == NULL) 2899 return (-1); 2900 #ifdef FAST_IPSEC 2901 KEY_FREESAV(&sav); 2902 #else 2903 key_freesav(sav); 2904 #endif 2905 return (-1); 2906 } 2907 TCP_FIELDS_TO_HOST(th); 2908 2909 if (bcmp(sig, sigp, TCP_SIGLEN)) { 2910 tcpstat.tcps_badsig++; 2911 if (sav == NULL) 2912 return (-1); 2913 #ifdef FAST_IPSEC 2914 KEY_FREESAV(&sav); 2915 #else 2916 key_freesav(sav); 2917 #endif 2918 return (-1); 2919 } else 2920 tcpstat.tcps_goodsig++; 2921 2922 key_sa_recordxfer(sav, m); 2923 #ifdef FAST_IPSEC 2924 KEY_FREESAV(&sav); 2925 #else 2926 key_freesav(sav); 2927 #endif 2928 } 2929 #endif 2930 2931 return (0); 2932 } 2933 2934 /* 2935 * Pull out of band byte out of a segment so 2936 * it doesn't appear in the user's data queue. 2937 * It is still reflected in the segment length for 2938 * sequencing purposes. 2939 */ 2940 void 2941 tcp_pulloutofband(so, th, m, off) 2942 struct socket *so; 2943 struct tcphdr *th; 2944 struct mbuf *m; 2945 int off; 2946 { 2947 int cnt = off + th->th_urp - 1; 2948 2949 while (cnt >= 0) { 2950 if (m->m_len > cnt) { 2951 char *cp = mtod(m, caddr_t) + cnt; 2952 struct tcpcb *tp = sototcpcb(so); 2953 2954 tp->t_iobc = *cp; 2955 tp->t_oobflags |= TCPOOB_HAVEDATA; 2956 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2957 m->m_len--; 2958 return; 2959 } 2960 cnt -= m->m_len; 2961 m = m->m_next; 2962 if (m == 0) 2963 break; 2964 } 2965 panic("tcp_pulloutofband"); 2966 } 2967 2968 /* 2969 * Collect new round-trip time estimate 2970 * and update averages and current timeout. 2971 */ 2972 void 2973 tcp_xmit_timer(tp, rtt) 2974 struct tcpcb *tp; 2975 uint32_t rtt; 2976 { 2977 int32_t delta; 2978 2979 tcpstat.tcps_rttupdated++; 2980 if (tp->t_srtt != 0) { 2981 /* 2982 * srtt is stored as fixed point with 3 bits after the 2983 * binary point (i.e., scaled by 8). The following magic 2984 * is equivalent to the smoothing algorithm in rfc793 with 2985 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2986 * point). Adjust rtt to origin 0. 2987 */ 2988 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2989 if ((tp->t_srtt += delta) <= 0) 2990 tp->t_srtt = 1 << 2; 2991 /* 2992 * We accumulate a smoothed rtt variance (actually, a 2993 * smoothed mean difference), then set the retransmit 2994 * timer to smoothed rtt + 4 times the smoothed variance. 2995 * rttvar is stored as fixed point with 2 bits after the 2996 * binary point (scaled by 4). The following is 2997 * equivalent to rfc793 smoothing with an alpha of .75 2998 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2999 * rfc793's wired-in beta. 3000 */ 3001 if (delta < 0) 3002 delta = -delta; 3003 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3004 if ((tp->t_rttvar += delta) <= 0) 3005 tp->t_rttvar = 1 << 2; 3006 } else { 3007 /* 3008 * No rtt measurement yet - use the unsmoothed rtt. 3009 * Set the variance to half the rtt (so our first 3010 * retransmit happens at 3*rtt). 3011 */ 3012 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3013 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3014 } 3015 tp->t_rtttime = 0; 3016 tp->t_rxtshift = 0; 3017 3018 /* 3019 * the retransmit should happen at rtt + 4 * rttvar. 3020 * Because of the way we do the smoothing, srtt and rttvar 3021 * will each average +1/2 tick of bias. When we compute 3022 * the retransmit timer, we want 1/2 tick of rounding and 3023 * 1 extra tick because of +-1/2 tick uncertainty in the 3024 * firing of the timer. The bias will give us exactly the 3025 * 1.5 tick we need. But, because the bias is 3026 * statistical, we have to test that we don't drop below 3027 * the minimum feasible timer (which is 2 ticks). 3028 */ 3029 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3030 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3031 3032 /* 3033 * We received an ack for a packet that wasn't retransmitted; 3034 * it is probably safe to discard any error indications we've 3035 * received recently. This isn't quite right, but close enough 3036 * for now (a route might have failed after we sent a segment, 3037 * and the return path might not be symmetrical). 3038 */ 3039 tp->t_softerror = 0; 3040 } 3041 3042 /* 3043 * Implement the NewReno response to a new ack, checking for partial acks in 3044 * fast recovery. 3045 */ 3046 void 3047 tcp_newreno(tp, th) 3048 struct tcpcb *tp; 3049 struct tcphdr *th; 3050 { 3051 if (tp->t_dupacks < tcprexmtthresh) { 3052 /* 3053 * We were not in fast recovery. Reset the duplicate ack 3054 * counter. 3055 */ 3056 tp->t_dupacks = 0; 3057 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) { 3058 /* 3059 * This is a partial ack. Retransmit the first unacknowledged 3060 * segment and deflate the congestion window by the amount of 3061 * acknowledged data. Do not exit fast recovery. 3062 */ 3063 tcp_seq onxt = tp->snd_nxt; 3064 u_long ocwnd = tp->snd_cwnd; 3065 3066 /* 3067 * snd_una has not yet been updated and the socket's send 3068 * buffer has not yet drained off the ACK'd data, so we 3069 * have to leave snd_una as it was to get the correct data 3070 * offset in tcp_output(). 3071 */ 3072 TCP_TIMER_DISARM(tp, TCPT_REXMT); 3073 tp->t_rtttime = 0; 3074 tp->snd_nxt = th->th_ack; 3075 /* 3076 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 3077 * is not yet updated when we're called. 3078 */ 3079 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 3080 (void) tcp_output(tp); 3081 tp->snd_cwnd = ocwnd; 3082 if (SEQ_GT(onxt, tp->snd_nxt)) 3083 tp->snd_nxt = onxt; 3084 /* 3085 * Partial window deflation. Relies on fact that tp->snd_una 3086 * not updated yet. 3087 */ 3088 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 3089 } else { 3090 /* 3091 * Complete ack. Inflate the congestion window to ssthresh 3092 * and exit fast recovery. 3093 * 3094 * Window inflation should have left us with approx. 3095 * snd_ssthresh outstanding data. But in case we 3096 * would be inclined to send a burst, better to do 3097 * it via the slow start mechanism. 3098 */ 3099 if (SEQ_LT(tp->snd_max, th->th_ack + tp->snd_ssthresh)) 3100 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 3101 + tp->t_segsz; 3102 else 3103 tp->snd_cwnd = tp->snd_ssthresh; 3104 tp->t_dupacks = 0; 3105 } 3106 } 3107 3108 3109 /* 3110 * TCP compressed state engine. Currently used to hold compressed 3111 * state for SYN_RECEIVED. 3112 */ 3113 3114 u_long syn_cache_count; 3115 u_int32_t syn_hash1, syn_hash2; 3116 3117 #define SYN_HASH(sa, sp, dp) \ 3118 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3119 ((u_int32_t)(sp)))^syn_hash2))) 3120 #ifndef INET6 3121 #define SYN_HASHALL(hash, src, dst) \ 3122 do { \ 3123 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 3124 ((struct sockaddr_in *)(src))->sin_port, \ 3125 ((struct sockaddr_in *)(dst))->sin_port); \ 3126 } while (/*CONSTCOND*/ 0) 3127 #else 3128 #define SYN_HASH6(sa, sp, dp) \ 3129 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3130 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3131 & 0x7fffffff) 3132 3133 #define SYN_HASHALL(hash, src, dst) \ 3134 do { \ 3135 switch ((src)->sa_family) { \ 3136 case AF_INET: \ 3137 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 3138 ((struct sockaddr_in *)(src))->sin_port, \ 3139 ((struct sockaddr_in *)(dst))->sin_port); \ 3140 break; \ 3141 case AF_INET6: \ 3142 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 3143 ((struct sockaddr_in6 *)(src))->sin6_port, \ 3144 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 3145 break; \ 3146 default: \ 3147 hash = 0; \ 3148 } \ 3149 } while (/*CONSTCOND*/0) 3150 #endif /* INET6 */ 3151 3152 #define SYN_CACHE_RM(sc) \ 3153 do { \ 3154 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 3155 (sc), sc_bucketq); \ 3156 (sc)->sc_tp = NULL; \ 3157 LIST_REMOVE((sc), sc_tpq); \ 3158 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 3159 callout_stop(&(sc)->sc_timer); \ 3160 syn_cache_count--; \ 3161 } while (/*CONSTCOND*/0) 3162 3163 #define SYN_CACHE_PUT(sc) \ 3164 do { \ 3165 if ((sc)->sc_ipopts) \ 3166 (void) m_free((sc)->sc_ipopts); \ 3167 if ((sc)->sc_route4.ro_rt != NULL) \ 3168 RTFREE((sc)->sc_route4.ro_rt); \ 3169 if (callout_invoking(&(sc)->sc_timer)) \ 3170 (sc)->sc_flags |= SCF_DEAD; \ 3171 else \ 3172 pool_put(&syn_cache_pool, (sc)); \ 3173 } while (/*CONSTCOND*/0) 3174 3175 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL); 3176 3177 /* 3178 * We don't estimate RTT with SYNs, so each packet starts with the default 3179 * RTT and each timer step has a fixed timeout value. 3180 */ 3181 #define SYN_CACHE_TIMER_ARM(sc) \ 3182 do { \ 3183 TCPT_RANGESET((sc)->sc_rxtcur, \ 3184 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3185 TCPTV_REXMTMAX); \ 3186 callout_reset(&(sc)->sc_timer, \ 3187 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3188 } while (/*CONSTCOND*/0) 3189 3190 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3191 3192 void 3193 syn_cache_init() 3194 { 3195 int i; 3196 3197 /* Initialize the hash buckets. */ 3198 for (i = 0; i < tcp_syn_cache_size; i++) 3199 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3200 } 3201 3202 void 3203 syn_cache_insert(sc, tp) 3204 struct syn_cache *sc; 3205 struct tcpcb *tp; 3206 { 3207 struct syn_cache_head *scp; 3208 struct syn_cache *sc2; 3209 int s; 3210 3211 /* 3212 * If there are no entries in the hash table, reinitialize 3213 * the hash secrets. 3214 */ 3215 if (syn_cache_count == 0) { 3216 syn_hash1 = arc4random(); 3217 syn_hash2 = arc4random(); 3218 } 3219 3220 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3221 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3222 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3223 3224 /* 3225 * Make sure that we don't overflow the per-bucket 3226 * limit or the total cache size limit. 3227 */ 3228 s = splsoftnet(); 3229 if (scp->sch_length >= tcp_syn_bucket_limit) { 3230 tcpstat.tcps_sc_bucketoverflow++; 3231 /* 3232 * The bucket is full. Toss the oldest element in the 3233 * bucket. This will be the first entry in the bucket. 3234 */ 3235 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3236 #ifdef DIAGNOSTIC 3237 /* 3238 * This should never happen; we should always find an 3239 * entry in our bucket. 3240 */ 3241 if (sc2 == NULL) 3242 panic("syn_cache_insert: bucketoverflow: impossible"); 3243 #endif 3244 SYN_CACHE_RM(sc2); 3245 SYN_CACHE_PUT(sc2); 3246 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3247 struct syn_cache_head *scp2, *sce; 3248 3249 tcpstat.tcps_sc_overflowed++; 3250 /* 3251 * The cache is full. Toss the oldest entry in the 3252 * first non-empty bucket we can find. 3253 * 3254 * XXX We would really like to toss the oldest 3255 * entry in the cache, but we hope that this 3256 * condition doesn't happen very often. 3257 */ 3258 scp2 = scp; 3259 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3260 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3261 for (++scp2; scp2 != scp; scp2++) { 3262 if (scp2 >= sce) 3263 scp2 = &tcp_syn_cache[0]; 3264 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3265 break; 3266 } 3267 #ifdef DIAGNOSTIC 3268 /* 3269 * This should never happen; we should always find a 3270 * non-empty bucket. 3271 */ 3272 if (scp2 == scp) 3273 panic("syn_cache_insert: cacheoverflow: " 3274 "impossible"); 3275 #endif 3276 } 3277 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3278 SYN_CACHE_RM(sc2); 3279 SYN_CACHE_PUT(sc2); 3280 } 3281 3282 /* 3283 * Initialize the entry's timer. 3284 */ 3285 sc->sc_rxttot = 0; 3286 sc->sc_rxtshift = 0; 3287 SYN_CACHE_TIMER_ARM(sc); 3288 3289 /* Link it from tcpcb entry */ 3290 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3291 3292 /* Put it into the bucket. */ 3293 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3294 scp->sch_length++; 3295 syn_cache_count++; 3296 3297 tcpstat.tcps_sc_added++; 3298 splx(s); 3299 } 3300 3301 /* 3302 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3303 * If we have retransmitted an entry the maximum number of times, expire 3304 * that entry. 3305 */ 3306 void 3307 syn_cache_timer(void *arg) 3308 { 3309 struct syn_cache *sc = arg; 3310 int s; 3311 3312 s = splsoftnet(); 3313 callout_ack(&sc->sc_timer); 3314 3315 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3316 tcpstat.tcps_sc_delayed_free++; 3317 pool_put(&syn_cache_pool, sc); 3318 splx(s); 3319 return; 3320 } 3321 3322 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3323 /* Drop it -- too many retransmissions. */ 3324 goto dropit; 3325 } 3326 3327 /* 3328 * Compute the total amount of time this entry has 3329 * been on a queue. If this entry has been on longer 3330 * than the keep alive timer would allow, expire it. 3331 */ 3332 sc->sc_rxttot += sc->sc_rxtcur; 3333 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) 3334 goto dropit; 3335 3336 tcpstat.tcps_sc_retransmitted++; 3337 (void) syn_cache_respond(sc, NULL); 3338 3339 /* Advance the timer back-off. */ 3340 sc->sc_rxtshift++; 3341 SYN_CACHE_TIMER_ARM(sc); 3342 3343 splx(s); 3344 return; 3345 3346 dropit: 3347 tcpstat.tcps_sc_timed_out++; 3348 SYN_CACHE_RM(sc); 3349 SYN_CACHE_PUT(sc); 3350 splx(s); 3351 } 3352 3353 /* 3354 * Remove syn cache created by the specified tcb entry, 3355 * because this does not make sense to keep them 3356 * (if there's no tcb entry, syn cache entry will never be used) 3357 */ 3358 void 3359 syn_cache_cleanup(tp) 3360 struct tcpcb *tp; 3361 { 3362 struct syn_cache *sc, *nsc; 3363 int s; 3364 3365 s = splsoftnet(); 3366 3367 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3368 nsc = LIST_NEXT(sc, sc_tpq); 3369 3370 #ifdef DIAGNOSTIC 3371 if (sc->sc_tp != tp) 3372 panic("invalid sc_tp in syn_cache_cleanup"); 3373 #endif 3374 SYN_CACHE_RM(sc); 3375 SYN_CACHE_PUT(sc); 3376 } 3377 /* just for safety */ 3378 LIST_INIT(&tp->t_sc); 3379 3380 splx(s); 3381 } 3382 3383 /* 3384 * Find an entry in the syn cache. 3385 */ 3386 struct syn_cache * 3387 syn_cache_lookup(src, dst, headp) 3388 struct sockaddr *src; 3389 struct sockaddr *dst; 3390 struct syn_cache_head **headp; 3391 { 3392 struct syn_cache *sc; 3393 struct syn_cache_head *scp; 3394 u_int32_t hash; 3395 int s; 3396 3397 SYN_HASHALL(hash, src, dst); 3398 3399 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3400 *headp = scp; 3401 s = splsoftnet(); 3402 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3403 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3404 if (sc->sc_hash != hash) 3405 continue; 3406 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3407 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3408 splx(s); 3409 return (sc); 3410 } 3411 } 3412 splx(s); 3413 return (NULL); 3414 } 3415 3416 /* 3417 * This function gets called when we receive an ACK for a 3418 * socket in the LISTEN state. We look up the connection 3419 * in the syn cache, and if its there, we pull it out of 3420 * the cache and turn it into a full-blown connection in 3421 * the SYN-RECEIVED state. 3422 * 3423 * The return values may not be immediately obvious, and their effects 3424 * can be subtle, so here they are: 3425 * 3426 * NULL SYN was not found in cache; caller should drop the 3427 * packet and send an RST. 3428 * 3429 * -1 We were unable to create the new connection, and are 3430 * aborting it. An ACK,RST is being sent to the peer 3431 * (unless we got screwey sequence numbners; see below), 3432 * because the 3-way handshake has been completed. Caller 3433 * should not free the mbuf, since we may be using it. If 3434 * we are not, we will free it. 3435 * 3436 * Otherwise, the return value is a pointer to the new socket 3437 * associated with the connection. 3438 */ 3439 struct socket * 3440 syn_cache_get(src, dst, th, hlen, tlen, so, m) 3441 struct sockaddr *src; 3442 struct sockaddr *dst; 3443 struct tcphdr *th; 3444 unsigned int hlen, tlen; 3445 struct socket *so; 3446 struct mbuf *m; 3447 { 3448 struct syn_cache *sc; 3449 struct syn_cache_head *scp; 3450 struct inpcb *inp = NULL; 3451 #ifdef INET6 3452 struct in6pcb *in6p = NULL; 3453 #endif 3454 struct tcpcb *tp = 0; 3455 struct mbuf *am; 3456 int s; 3457 struct socket *oso; 3458 3459 s = splsoftnet(); 3460 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3461 splx(s); 3462 return (NULL); 3463 } 3464 3465 /* 3466 * Verify the sequence and ack numbers. Try getting the correct 3467 * response again. 3468 */ 3469 if ((th->th_ack != sc->sc_iss + 1) || 3470 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3471 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3472 (void) syn_cache_respond(sc, m); 3473 splx(s); 3474 return ((struct socket *)(-1)); 3475 } 3476 3477 /* Remove this cache entry */ 3478 SYN_CACHE_RM(sc); 3479 splx(s); 3480 3481 /* 3482 * Ok, create the full blown connection, and set things up 3483 * as they would have been set up if we had created the 3484 * connection when the SYN arrived. If we can't create 3485 * the connection, abort it. 3486 */ 3487 /* 3488 * inp still has the OLD in_pcb stuff, set the 3489 * v6-related flags on the new guy, too. This is 3490 * done particularly for the case where an AF_INET6 3491 * socket is bound only to a port, and a v4 connection 3492 * comes in on that port. 3493 * we also copy the flowinfo from the original pcb 3494 * to the new one. 3495 */ 3496 oso = so; 3497 so = sonewconn(so, SS_ISCONNECTED); 3498 if (so == NULL) 3499 goto resetandabort; 3500 3501 switch (so->so_proto->pr_domain->dom_family) { 3502 #ifdef INET 3503 case AF_INET: 3504 inp = sotoinpcb(so); 3505 break; 3506 #endif 3507 #ifdef INET6 3508 case AF_INET6: 3509 in6p = sotoin6pcb(so); 3510 break; 3511 #endif 3512 } 3513 switch (src->sa_family) { 3514 #ifdef INET 3515 case AF_INET: 3516 if (inp) { 3517 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3518 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3519 inp->inp_options = ip_srcroute(); 3520 in_pcbstate(inp, INP_BOUND); 3521 if (inp->inp_options == NULL) { 3522 inp->inp_options = sc->sc_ipopts; 3523 sc->sc_ipopts = NULL; 3524 } 3525 } 3526 #ifdef INET6 3527 else if (in6p) { 3528 /* IPv4 packet to AF_INET6 socket */ 3529 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3530 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3531 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3532 &in6p->in6p_laddr.s6_addr32[3], 3533 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3534 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3535 in6totcpcb(in6p)->t_family = AF_INET; 3536 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3537 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3538 else 3539 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3540 in6_pcbstate(in6p, IN6P_BOUND); 3541 } 3542 #endif 3543 break; 3544 #endif 3545 #ifdef INET6 3546 case AF_INET6: 3547 if (in6p) { 3548 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3549 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3550 in6_pcbstate(in6p, IN6P_BOUND); 3551 } 3552 break; 3553 #endif 3554 } 3555 #ifdef INET6 3556 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3557 struct in6pcb *oin6p = sotoin6pcb(oso); 3558 /* inherit socket options from the listening socket */ 3559 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3560 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3561 m_freem(in6p->in6p_options); 3562 in6p->in6p_options = 0; 3563 } 3564 ip6_savecontrol(in6p, &in6p->in6p_options, 3565 mtod(m, struct ip6_hdr *), m); 3566 } 3567 #endif 3568 3569 #if defined(IPSEC) || defined(FAST_IPSEC) 3570 /* 3571 * we make a copy of policy, instead of sharing the policy, 3572 * for better behavior in terms of SA lookup and dead SA removal. 3573 */ 3574 if (inp) { 3575 /* copy old policy into new socket's */ 3576 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3577 printf("tcp_input: could not copy policy\n"); 3578 } 3579 #ifdef INET6 3580 else if (in6p) { 3581 /* copy old policy into new socket's */ 3582 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3583 in6p->in6p_sp)) 3584 printf("tcp_input: could not copy policy\n"); 3585 } 3586 #endif 3587 #endif 3588 3589 /* 3590 * Give the new socket our cached route reference. 3591 */ 3592 if (inp) 3593 inp->inp_route = sc->sc_route4; /* struct assignment */ 3594 #ifdef INET6 3595 else 3596 in6p->in6p_route = sc->sc_route6; 3597 #endif 3598 sc->sc_route4.ro_rt = NULL; 3599 3600 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3601 if (am == NULL) 3602 goto resetandabort; 3603 MCLAIM(am, &tcp_mowner); 3604 am->m_len = src->sa_len; 3605 bcopy(src, mtod(am, caddr_t), src->sa_len); 3606 if (inp) { 3607 if (in_pcbconnect(inp, am)) { 3608 (void) m_free(am); 3609 goto resetandabort; 3610 } 3611 } 3612 #ifdef INET6 3613 else if (in6p) { 3614 if (src->sa_family == AF_INET) { 3615 /* IPv4 packet to AF_INET6 socket */ 3616 struct sockaddr_in6 *sin6; 3617 sin6 = mtod(am, struct sockaddr_in6 *); 3618 am->m_len = sizeof(*sin6); 3619 bzero(sin6, sizeof(*sin6)); 3620 sin6->sin6_family = AF_INET6; 3621 sin6->sin6_len = sizeof(*sin6); 3622 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3623 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3624 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3625 &sin6->sin6_addr.s6_addr32[3], 3626 sizeof(sin6->sin6_addr.s6_addr32[3])); 3627 } 3628 if (in6_pcbconnect(in6p, am)) { 3629 (void) m_free(am); 3630 goto resetandabort; 3631 } 3632 } 3633 #endif 3634 else { 3635 (void) m_free(am); 3636 goto resetandabort; 3637 } 3638 (void) m_free(am); 3639 3640 if (inp) 3641 tp = intotcpcb(inp); 3642 #ifdef INET6 3643 else if (in6p) 3644 tp = in6totcpcb(in6p); 3645 #endif 3646 else 3647 tp = NULL; 3648 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3649 if (sc->sc_request_r_scale != 15) { 3650 tp->requested_s_scale = sc->sc_requested_s_scale; 3651 tp->request_r_scale = sc->sc_request_r_scale; 3652 tp->snd_scale = sc->sc_requested_s_scale; 3653 tp->rcv_scale = sc->sc_request_r_scale; 3654 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3655 } 3656 if (sc->sc_flags & SCF_TIMESTAMP) 3657 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3658 tp->ts_timebase = sc->sc_timebase; 3659 3660 tp->t_template = tcp_template(tp); 3661 if (tp->t_template == 0) { 3662 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3663 so = NULL; 3664 m_freem(m); 3665 goto abort; 3666 } 3667 3668 tp->iss = sc->sc_iss; 3669 tp->irs = sc->sc_irs; 3670 tcp_sendseqinit(tp); 3671 tcp_rcvseqinit(tp); 3672 tp->t_state = TCPS_SYN_RECEIVED; 3673 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3674 tcpstat.tcps_accepts++; 3675 3676 #ifdef TCP_SIGNATURE 3677 if (sc->sc_flags & SCF_SIGNATURE) 3678 tp->t_flags |= TF_SIGNATURE; 3679 #endif 3680 3681 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3682 tp->t_ourmss = sc->sc_ourmaxseg; 3683 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3684 3685 /* 3686 * Initialize the initial congestion window. If we 3687 * had to retransmit the SYN,ACK, we must initialize cwnd 3688 * to 1 segment (i.e. the Loss Window). 3689 */ 3690 if (sc->sc_rxtshift) 3691 tp->snd_cwnd = tp->t_peermss; 3692 else { 3693 int ss = tcp_init_win; 3694 #ifdef INET 3695 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3696 ss = tcp_init_win_local; 3697 #endif 3698 #ifdef INET6 3699 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3700 ss = tcp_init_win_local; 3701 #endif 3702 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3703 } 3704 3705 tcp_rmx_rtt(tp); 3706 tp->snd_wl1 = sc->sc_irs; 3707 tp->rcv_up = sc->sc_irs + 1; 3708 3709 /* 3710 * This is what whould have happened in tcp_output() when 3711 * the SYN,ACK was sent. 3712 */ 3713 tp->snd_up = tp->snd_una; 3714 tp->snd_max = tp->snd_nxt = tp->iss+1; 3715 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3716 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3717 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3718 tp->last_ack_sent = tp->rcv_nxt; 3719 3720 tcpstat.tcps_sc_completed++; 3721 SYN_CACHE_PUT(sc); 3722 return (so); 3723 3724 resetandabort: 3725 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3726 abort: 3727 if (so != NULL) 3728 (void) soabort(so); 3729 SYN_CACHE_PUT(sc); 3730 tcpstat.tcps_sc_aborted++; 3731 return ((struct socket *)(-1)); 3732 } 3733 3734 /* 3735 * This function is called when we get a RST for a 3736 * non-existent connection, so that we can see if the 3737 * connection is in the syn cache. If it is, zap it. 3738 */ 3739 3740 void 3741 syn_cache_reset(src, dst, th) 3742 struct sockaddr *src; 3743 struct sockaddr *dst; 3744 struct tcphdr *th; 3745 { 3746 struct syn_cache *sc; 3747 struct syn_cache_head *scp; 3748 int s = splsoftnet(); 3749 3750 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3751 splx(s); 3752 return; 3753 } 3754 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3755 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3756 splx(s); 3757 return; 3758 } 3759 SYN_CACHE_RM(sc); 3760 splx(s); 3761 tcpstat.tcps_sc_reset++; 3762 SYN_CACHE_PUT(sc); 3763 } 3764 3765 void 3766 syn_cache_unreach(src, dst, th) 3767 struct sockaddr *src; 3768 struct sockaddr *dst; 3769 struct tcphdr *th; 3770 { 3771 struct syn_cache *sc; 3772 struct syn_cache_head *scp; 3773 int s; 3774 3775 s = splsoftnet(); 3776 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3777 splx(s); 3778 return; 3779 } 3780 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3781 if (ntohl (th->th_seq) != sc->sc_iss) { 3782 splx(s); 3783 return; 3784 } 3785 3786 /* 3787 * If we've retransmitted 3 times and this is our second error, 3788 * we remove the entry. Otherwise, we allow it to continue on. 3789 * This prevents us from incorrectly nuking an entry during a 3790 * spurious network outage. 3791 * 3792 * See tcp_notify(). 3793 */ 3794 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3795 sc->sc_flags |= SCF_UNREACH; 3796 splx(s); 3797 return; 3798 } 3799 3800 SYN_CACHE_RM(sc); 3801 splx(s); 3802 tcpstat.tcps_sc_unreach++; 3803 SYN_CACHE_PUT(sc); 3804 } 3805 3806 /* 3807 * Given a LISTEN socket and an inbound SYN request, add 3808 * this to the syn cache, and send back a segment: 3809 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3810 * to the source. 3811 * 3812 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3813 * Doing so would require that we hold onto the data and deliver it 3814 * to the application. However, if we are the target of a SYN-flood 3815 * DoS attack, an attacker could send data which would eventually 3816 * consume all available buffer space if it were ACKed. By not ACKing 3817 * the data, we avoid this DoS scenario. 3818 */ 3819 3820 int 3821 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi) 3822 struct sockaddr *src; 3823 struct sockaddr *dst; 3824 struct tcphdr *th; 3825 unsigned int hlen; 3826 struct socket *so; 3827 struct mbuf *m; 3828 u_char *optp; 3829 int optlen; 3830 struct tcp_opt_info *oi; 3831 { 3832 struct tcpcb tb, *tp; 3833 long win; 3834 struct syn_cache *sc; 3835 struct syn_cache_head *scp; 3836 struct mbuf *ipopts; 3837 struct tcp_opt_info opti; 3838 3839 tp = sototcpcb(so); 3840 3841 bzero(&opti, sizeof(opti)); 3842 3843 /* 3844 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3845 * 3846 * Note this check is performed in tcp_input() very early on. 3847 */ 3848 3849 /* 3850 * Initialize some local state. 3851 */ 3852 win = sbspace(&so->so_rcv); 3853 if (win > TCP_MAXWIN) 3854 win = TCP_MAXWIN; 3855 3856 switch (src->sa_family) { 3857 #ifdef INET 3858 case AF_INET: 3859 /* 3860 * Remember the IP options, if any. 3861 */ 3862 ipopts = ip_srcroute(); 3863 break; 3864 #endif 3865 default: 3866 ipopts = NULL; 3867 } 3868 3869 #ifdef TCP_SIGNATURE 3870 if (optp || (tp->t_flags & TF_SIGNATURE)) 3871 #else 3872 if (optp) 3873 #endif 3874 { 3875 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3876 #ifdef TCP_SIGNATURE 3877 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 3878 #endif 3879 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 3880 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 3881 return (0); 3882 } else 3883 tb.t_flags = 0; 3884 3885 /* 3886 * See if we already have an entry for this connection. 3887 * If we do, resend the SYN,ACK. We do not count this 3888 * as a retransmission (XXX though maybe we should). 3889 */ 3890 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3891 tcpstat.tcps_sc_dupesyn++; 3892 if (ipopts) { 3893 /* 3894 * If we were remembering a previous source route, 3895 * forget it and use the new one we've been given. 3896 */ 3897 if (sc->sc_ipopts) 3898 (void) m_free(sc->sc_ipopts); 3899 sc->sc_ipopts = ipopts; 3900 } 3901 sc->sc_timestamp = tb.ts_recent; 3902 if (syn_cache_respond(sc, m) == 0) { 3903 tcpstat.tcps_sndacks++; 3904 tcpstat.tcps_sndtotal++; 3905 } 3906 return (1); 3907 } 3908 3909 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3910 if (sc == NULL) { 3911 if (ipopts) 3912 (void) m_free(ipopts); 3913 return (0); 3914 } 3915 3916 /* 3917 * Fill in the cache, and put the necessary IP and TCP 3918 * options into the reply. 3919 */ 3920 bzero(sc, sizeof(struct syn_cache)); 3921 callout_init(&sc->sc_timer); 3922 bcopy(src, &sc->sc_src, src->sa_len); 3923 bcopy(dst, &sc->sc_dst, dst->sa_len); 3924 sc->sc_flags = 0; 3925 sc->sc_ipopts = ipopts; 3926 sc->sc_irs = th->th_seq; 3927 switch (src->sa_family) { 3928 #ifdef INET 3929 case AF_INET: 3930 { 3931 struct sockaddr_in *srcin = (void *) src; 3932 struct sockaddr_in *dstin = (void *) dst; 3933 3934 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 3935 &srcin->sin_addr, dstin->sin_port, 3936 srcin->sin_port, sizeof(dstin->sin_addr), 0); 3937 break; 3938 } 3939 #endif /* INET */ 3940 #ifdef INET6 3941 case AF_INET6: 3942 { 3943 struct sockaddr_in6 *srcin6 = (void *) src; 3944 struct sockaddr_in6 *dstin6 = (void *) dst; 3945 3946 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 3947 &srcin6->sin6_addr, dstin6->sin6_port, 3948 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 3949 break; 3950 } 3951 #endif /* INET6 */ 3952 } 3953 sc->sc_peermaxseg = oi->maxseg; 3954 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3955 m->m_pkthdr.rcvif : NULL, 3956 sc->sc_src.sa.sa_family); 3957 sc->sc_win = win; 3958 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 3959 sc->sc_timestamp = tb.ts_recent; 3960 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 3961 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 3962 sc->sc_flags |= SCF_TIMESTAMP; 3963 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3964 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3965 sc->sc_requested_s_scale = tb.requested_s_scale; 3966 sc->sc_request_r_scale = 0; 3967 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3968 TCP_MAXWIN << sc->sc_request_r_scale < 3969 so->so_rcv.sb_hiwat) 3970 sc->sc_request_r_scale++; 3971 } else { 3972 sc->sc_requested_s_scale = 15; 3973 sc->sc_request_r_scale = 15; 3974 } 3975 #ifdef TCP_SIGNATURE 3976 if (tb.t_flags & TF_SIGNATURE) 3977 sc->sc_flags |= SCF_SIGNATURE; 3978 #endif 3979 sc->sc_tp = tp; 3980 if (syn_cache_respond(sc, m) == 0) { 3981 syn_cache_insert(sc, tp); 3982 tcpstat.tcps_sndacks++; 3983 tcpstat.tcps_sndtotal++; 3984 } else { 3985 SYN_CACHE_PUT(sc); 3986 tcpstat.tcps_sc_dropped++; 3987 } 3988 return (1); 3989 } 3990 3991 int 3992 syn_cache_respond(sc, m) 3993 struct syn_cache *sc; 3994 struct mbuf *m; 3995 { 3996 struct route *ro; 3997 u_int8_t *optp; 3998 int optlen, error; 3999 u_int16_t tlen; 4000 struct ip *ip = NULL; 4001 #ifdef INET6 4002 struct ip6_hdr *ip6 = NULL; 4003 #endif 4004 struct tcpcb *tp; 4005 struct tcphdr *th; 4006 u_int hlen; 4007 struct socket *so; 4008 4009 switch (sc->sc_src.sa.sa_family) { 4010 case AF_INET: 4011 hlen = sizeof(struct ip); 4012 ro = &sc->sc_route4; 4013 break; 4014 #ifdef INET6 4015 case AF_INET6: 4016 hlen = sizeof(struct ip6_hdr); 4017 ro = (struct route *)&sc->sc_route6; 4018 break; 4019 #endif 4020 default: 4021 if (m) 4022 m_freem(m); 4023 return (EAFNOSUPPORT); 4024 } 4025 4026 /* Compute the size of the TCP options. */ 4027 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4028 #ifdef TCP_SIGNATURE 4029 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4030 #endif 4031 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4032 4033 tlen = hlen + sizeof(struct tcphdr) + optlen; 4034 4035 /* 4036 * Create the IP+TCP header from scratch. 4037 */ 4038 if (m) 4039 m_freem(m); 4040 #ifdef DIAGNOSTIC 4041 if (max_linkhdr + tlen > MCLBYTES) 4042 return (ENOBUFS); 4043 #endif 4044 MGETHDR(m, M_DONTWAIT, MT_DATA); 4045 if (m && tlen > MHLEN) { 4046 MCLGET(m, M_DONTWAIT); 4047 if ((m->m_flags & M_EXT) == 0) { 4048 m_freem(m); 4049 m = NULL; 4050 } 4051 } 4052 if (m == NULL) 4053 return (ENOBUFS); 4054 MCLAIM(m, &tcp_tx_mowner); 4055 4056 /* Fixup the mbuf. */ 4057 m->m_data += max_linkhdr; 4058 m->m_len = m->m_pkthdr.len = tlen; 4059 if (sc->sc_tp) { 4060 tp = sc->sc_tp; 4061 if (tp->t_inpcb) 4062 so = tp->t_inpcb->inp_socket; 4063 #ifdef INET6 4064 else if (tp->t_in6pcb) 4065 so = tp->t_in6pcb->in6p_socket; 4066 #endif 4067 else 4068 so = NULL; 4069 } else 4070 so = NULL; 4071 m->m_pkthdr.rcvif = NULL; 4072 memset(mtod(m, u_char *), 0, tlen); 4073 4074 switch (sc->sc_src.sa.sa_family) { 4075 case AF_INET: 4076 ip = mtod(m, struct ip *); 4077 ip->ip_v = 4; 4078 ip->ip_dst = sc->sc_src.sin.sin_addr; 4079 ip->ip_src = sc->sc_dst.sin.sin_addr; 4080 ip->ip_p = IPPROTO_TCP; 4081 th = (struct tcphdr *)(ip + 1); 4082 th->th_dport = sc->sc_src.sin.sin_port; 4083 th->th_sport = sc->sc_dst.sin.sin_port; 4084 break; 4085 #ifdef INET6 4086 case AF_INET6: 4087 ip6 = mtod(m, struct ip6_hdr *); 4088 ip6->ip6_vfc = IPV6_VERSION; 4089 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4090 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4091 ip6->ip6_nxt = IPPROTO_TCP; 4092 /* ip6_plen will be updated in ip6_output() */ 4093 th = (struct tcphdr *)(ip6 + 1); 4094 th->th_dport = sc->sc_src.sin6.sin6_port; 4095 th->th_sport = sc->sc_dst.sin6.sin6_port; 4096 break; 4097 #endif 4098 default: 4099 th = NULL; 4100 } 4101 4102 th->th_seq = htonl(sc->sc_iss); 4103 th->th_ack = htonl(sc->sc_irs + 1); 4104 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4105 th->th_flags = TH_SYN|TH_ACK; 4106 th->th_win = htons(sc->sc_win); 4107 /* th_sum already 0 */ 4108 /* th_urp already 0 */ 4109 4110 /* Tack on the TCP options. */ 4111 optp = (u_int8_t *)(th + 1); 4112 *optp++ = TCPOPT_MAXSEG; 4113 *optp++ = 4; 4114 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4115 *optp++ = sc->sc_ourmaxseg & 0xff; 4116 4117 if (sc->sc_request_r_scale != 15) { 4118 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4119 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4120 sc->sc_request_r_scale); 4121 optp += 4; 4122 } 4123 4124 if (sc->sc_flags & SCF_TIMESTAMP) { 4125 u_int32_t *lp = (u_int32_t *)(optp); 4126 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4127 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4128 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4129 *lp = htonl(sc->sc_timestamp); 4130 optp += TCPOLEN_TSTAMP_APPA; 4131 } 4132 4133 #ifdef TCP_SIGNATURE 4134 if (sc->sc_flags & SCF_SIGNATURE) { 4135 struct secasvar *sav; 4136 u_int8_t *sigp; 4137 4138 sav = tcp_signature_getsav(m, th); 4139 4140 if (sav == NULL) { 4141 if (m) 4142 m_freem(m); 4143 return (EPERM); 4144 } 4145 4146 *optp++ = TCPOPT_SIGNATURE; 4147 *optp++ = TCPOLEN_SIGNATURE; 4148 sigp = optp; 4149 bzero(optp, TCP_SIGLEN); 4150 optp += TCP_SIGLEN; 4151 *optp++ = TCPOPT_NOP; 4152 *optp++ = TCPOPT_EOL; 4153 4154 (void)tcp_signature(m, th, hlen, sav, sigp); 4155 4156 key_sa_recordxfer(sav, m); 4157 #ifdef FAST_IPSEC 4158 KEY_FREESAV(&sav); 4159 #else 4160 key_freesav(sav); 4161 #endif 4162 } 4163 #endif 4164 4165 /* Compute the packet's checksum. */ 4166 switch (sc->sc_src.sa.sa_family) { 4167 case AF_INET: 4168 ip->ip_len = htons(tlen - hlen); 4169 th->th_sum = 0; 4170 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4171 break; 4172 #ifdef INET6 4173 case AF_INET6: 4174 ip6->ip6_plen = htons(tlen - hlen); 4175 th->th_sum = 0; 4176 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4177 break; 4178 #endif 4179 } 4180 4181 /* 4182 * Fill in some straggling IP bits. Note the stack expects 4183 * ip_len to be in host order, for convenience. 4184 */ 4185 switch (sc->sc_src.sa.sa_family) { 4186 #ifdef INET 4187 case AF_INET: 4188 ip->ip_len = htons(tlen); 4189 ip->ip_ttl = ip_defttl; 4190 /* XXX tos? */ 4191 break; 4192 #endif 4193 #ifdef INET6 4194 case AF_INET6: 4195 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4196 ip6->ip6_vfc |= IPV6_VERSION; 4197 ip6->ip6_plen = htons(tlen - hlen); 4198 /* ip6_hlim will be initialized afterwards */ 4199 /* XXX flowlabel? */ 4200 break; 4201 #endif 4202 } 4203 4204 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4205 tp = sc->sc_tp; 4206 4207 switch (sc->sc_src.sa.sa_family) { 4208 #ifdef INET 4209 case AF_INET: 4210 error = ip_output(m, sc->sc_ipopts, ro, 4211 (ip_mtudisc ? IP_MTUDISC : 0), 4212 (struct ip_moptions *)NULL, so); 4213 break; 4214 #endif 4215 #ifdef INET6 4216 case AF_INET6: 4217 ip6->ip6_hlim = in6_selecthlim(NULL, 4218 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 4219 4220 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0, 4221 (struct ip6_moptions *)0, so, NULL); 4222 break; 4223 #endif 4224 default: 4225 error = EAFNOSUPPORT; 4226 break; 4227 } 4228 return (error); 4229 } 4230
Indexes created Mon Sep 22 10:09:38 GMT 2025